UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Epistemic modality and evidentiality in Gitksan at the semantics-pragmatics interface Peterson, Tyler Roy Gösta 2010

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2010_spring_peterson_tyler.pdf [ 2.75MB ]
Metadata
JSON: 24-1.0069730.json
JSON-LD: 24-1.0069730-ld.json
RDF/XML (Pretty): 24-1.0069730-rdf.xml
RDF/JSON: 24-1.0069730-rdf.json
Turtle: 24-1.0069730-turtle.txt
N-Triples: 24-1.0069730-rdf-ntriples.txt
Original Record: 24-1.0069730-source.json
Full Text
24-1.0069730-fulltext.txt
Citation
24-1.0069730.ris

Full Text

Epistemic Modality and Evidentiality in Gitksan at the Semantics-Pragmatics Interface by Tyler Roy Gösta Peterson B.Mus., University of British Columbia, 1999 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in The Faculty of Graduate Studies (Linguistics)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) April, 2010 c Tyler Roy Gösta Peterson 2010  Abstract The aim of this dissertation is to provide an empirically driven, theoretically informed investigation of how speakers of Gitksan, a Tsimshianic language spoken in the northwest coast of Canada, express knowledge about the world around them. There are three main goals that motivate this investigation, summarized below: (1)  (i.) To provide the first detailed description of the evidential and modal system in Gitksan. (ii.) To provide a formal semantic and pragmatic account of this system that adequately explains the meanings of the modals and evidentials, as well as how they are used in discourse. (iii.) To identify and examine the specific properties the Gitksan evidential/modal system brings to bear on current theories of semantics and pragmatics, as well as the consequences this analysis has on the study of modality and evidentiality cross-linguistically. In addition to documenting the evidential and modal meanings in Gitksan, I test and  work through a variety of theoretical tools from the literature designed to investigate evidentiality and modality in a language. This begins by determining what level of meaning the individual evidentials in Gitksan operate on. The current state of research into the connection between evidentiality and epistemic modality has identified two different types of evidentials defined by the level of meaning they operate on: propositional and illocutionary evidentials. These two types correspond to a distinction between modal evidentials and non-modal evidentials respectively. I show that Gitksan has both modal ii  Abstract and non-evidentials. This leads to an analysis where the Gitksan modal evidentials are treated as a specialized type of epistemic modals, and the non-modal evidentials are sentential force specifiers. I also identify various features of the evidential system that bring specific issues to bear upon current theories of the semantics and pragmatics of modality. This has four outcomes: first, I present a novel analysis of variable modal force in modals with fixed quantification: variable modal force in Gitksan modal evidentials is determined by the ordering source. Secondly, I discuss Conjectural Questions: when a modal evidential is added to a question it reduces the interrogative force of the question. This follows from the modal semantics of evidentials. Thirdly, I introduce the notion of Pragmatic blocking: modal and non-modal evidentials interact in discourse contexts, and implicate a speaker’s attitude towards the evidence they have for a proposition. And fourthly, I develop the first formal analysis of mirativity and non-literal uses of evidentials, analyzing them both as cases of conversational implicature.  iii  Table of Contents Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  ii  Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  iv  List of Tables  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  x  List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  xi  List of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv 1 Introduction  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  1  1.1  Goals of the Investigation . . . . . . . . . . . . . . . . . . . . . . . . . .  1  1.2  Empirical Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  4  1.2.1  Gitksan and the Tsimshianic Languages . . . . . . . . . . . . . .  5  1.2.2  Gitksan Orthography  . . . . . . . . . . . . . . . . . . . . . . . .  9  1.2.3  The Basic Features of Gitksan Morphology and Syntax . . . . . .  9  1.2.3.1  Morphology  . . . . . . . . . . . . . . . . . . . . . . . .  9  1.2.3.2  Syntax and Clause Types . . . . . . . . . . . . . . . . .  10  1.3  1.2.4  Evidentiality and Epistemic Modality in Gitksan  . . . . . . . . .  11  1.2.5  Studies on Evidentiality . . . . . . . . . . . . . . . . . . . . . . .  14  Theoretical Context  . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  15  1.3.1  The Relations Between Evidentiality and Epistemic Modality  . .  17  1.3.2  The Theoretical Tools . . . . . . . . . . . . . . . . . . . . . . . .  19 iv  Table of Contents 1.3.2.1  Determining Levels of Meaning . . . . . . . . . . . . . .  20  1.3.2.2  Static Semantics: Evidentials as Epistemic Modals . . .  21  1.3.2.3  Dynamic Semantics: Evidentials as Sentential Force Specifiers  1.4  1.5  . . . . . . . . . . . . . . . . . . . . . . . . . . . .  23  . . . . . . . . . . . . . . . . . . . . . . . . . . .  25  Structure of the Thesis 1.4.1  A Description of Evidentials in Gitksan  . . . . . . . . . . . . . .  25  1.4.2  Modal Evidentials  . . . . . . . . . . . . . . . . . . . . . . . . . .  26  1.4.2.1  A Challenge: Variable Modal Force  . . . . . . . . . . .  27  1.4.2.2  Conjectural Questions . . . . . . . . . . . . . . . . . . .  28  1.4.3  Non-Modal Evidentials  . . . . . . . . . . . . . . . . . . . . . . .  29  1.4.4  Evidentiality and Modality at the Semantics-Pragmatics Interface  30  1.4.4.1  Pragmatic Blocking . . . . . . . . . . . . . . . . . . . .  31  1.4.4.2  Mirativity and Metaphor  . . . . . . . . . . . . . . . . .  32  Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  34  2 Epistemic Modality and Evidentiality in Gitksan  . . . . . . . . . . . .  37  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  37  2.1  The Issues  2.2  The Plan  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  38  2.3  Describing Evidentiality . . . . . . . . . . . . . . . . . . . . . . . . . . .  39  2.4  Describing Evidentiality in Gitksan . . . . . . . . . . . . . . . . . . . . .  41  2.5  The Reportative =kat  . . . . . . . . . . . . . . . . . . . . . . . . . . . .  47  2.5.1  The Evidential Meanings of =kat . . . . . . . . . . . . . . . . . .  48  2.5.2  A Note on the Social Uses of =kat  . . . . . . . . . . . . . . . . .  55  . . . . . . . . . . . . . . . . . . . . . . . . . . . .  57  2.6  The Inferential =ima 2.6.1  The Evidential Meanings of =ima  . . . . . . . . . . . . . . . . .  58  2.6.2  The Modal Interpretations of =ima  . . . . . . . . . . . . . . . .  62  2.7  The Morphosyntactic Distribution of =kat and =ima . . . . . . . . . . .  64  2.8  ’ The Sensory inferential nakw  72  2.8.1  . . . . . . . . . . . . . . . . . . . . . . . .  ’ The Evidential Meanings of nakw  . . . . . . . . . . . . . . . . .  73 v  Table of Contents 2.8.2  ’ The Pragmatic Uses of nakw : Mirativity and Metaphor  2.8.3  ’ The Morphosyntactic Distribution of nakw  . . . . .  84  . . . . . . . . . . . .  86  Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  87  2.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  92  2.9  3 Evidentiality and Levels of Meaning  . . . . . . . . . . . . . . . . . . . .  93  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  93  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .  95  3.1  The Issues  3.2  The Plan  3.3  Evidentials as Epistemic Modals  3.4  Formal Pragmatic Approaches to Evidentiality  3.5  96  . . . . . . . . . . . . . . 103  3.4.1  Speech Act Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 103  3.4.2  Evidential Hierarchies and Context Dependence . . . . . . . . . . 107  3.4.3  Evidentials as Sentential-Force Specifiers . . . . . . . . . . . . . . 108  Determining Levels of Meaning: The Tests . . . . . . . . . . . . . . . . . 109 3.5.1  3.5.2  3.5.3 3.6  . . . . . . . . . . . . . . . . . . . . . .  Tests Regarding Truth . . . . . . . . . . . . . . . . . . . . . . . . 111 3.5.1.1  Known Truth/Falsity  . . . . . . . . . . . . . . . . . . . 111  3.5.1.2  Assent/Dissent . . . . . . . . . . . . . . . . . . . . . . . 114  3.5.1.3  Cancellability of Evidence Type Requirement . . . . . . 118  Embeddability and Scope . . . . . . . . . . . . . . . . . . . . . . 118 3.5.2.1  Embeddability . . . . . . . . . . . . . . . . . . . . . . . 118  3.5.2.2  Scope with Respect to Interrogatives . . . . . . . . . . . 121  3.5.2.3  Interaction with Negation . . . . . . . . . . . . . . . . . 123  Interim Summary  . . . . . . . . . . . . . . . . . . . . . . . . . . 123  Gitksan Evidentials: Propositional or Illocutionary Operators? 3.6.1  3.6.2  . . . . . 125  Tests Regarding Truth . . . . . . . . . . . . . . . . . . . . . . . . 125 3.6.1.1  Known Truth/Falsity  . . . . . . . . . . . . . . . . . . . 126  3.6.1.2  Assent/Dissent . . . . . . . . . . . . . . . . . . . . . . . 132  3.6.1.3  Cancellability of Evidence Type Requirement . . . . . . 136  Scope and Embeddability . . . . . . . . . . . . . . . . . . . . . . 137 vi  Table of Contents  3.7  3.6.2.1  Embeddability . . . . . . . . . . . . . . . . . . . . . . . 137  3.6.2.2  Scope with Respect to Interrogatives . . . . . . . . . . . 145  3.6.2.3  Interaction with Negation . . . . . . . . . . . . . . . . . 149  Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150  4 The Semantics of the Modal Evidentials =ima and =kat  . . . . . . . 152  4.1  The Issues  4.2  The Plan  4.3  A Modal Analysis of =ima and =kat . . . . . . . . . . . . . . . . . . . . 156  4.4  The Variable Modal Force of =ima and =kat 4.4.1  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155  . . . . . . . . . . . . . . . 160  The Variability of =ima . . . . . . . . . . . . . . . . . . . . . . . 161 4.4.1.1  Context-Conditioned Modal Force . . . . . . . . . . . . 161  4.4.1.2  Modals in Coordinated Sentences  4.4.1.3  Evidence Type and Modal Force . . . . . . . . . . . . . 166  . . . . . . . . . . . . 164  4.4.2  The Variability of =kat  4.4.3  Default Modal Force . . . . . . . . . . . . . . . . . . . . . . . . . 171  4.4.4  The Ordering Source in Deriving Variable Modal Force . . . . . . 172  . . . . . . . . . . . . . . . . . . . . . . . 169  4.4.4.1  Ordering Sources with Fixed Quantification . . . . . . . 175  4.4.4.2  An Emerging Theoretical Typology: Variable Force in St’át’imcets Modals . . . . . . . . . . . . . . . . . . . . 183  4.4.4.3 4.4.5 4.5  Strengthening and Weakening in Paradigms . . . . . . . 188  Interim Summary  . . . . . . . . . . . . . . . . . . . . . . . . . . 191  Modal =ima and Conjectural Questions . . . . . . . . . . . . . . . . . . 192 4.5.1  The Properties of Conjectural Questions . . . . . . . . . . . . . . 194  4.5.2  An Analysis of Conjectural Questions  4.5.3  Conjectural Questions in Other Languages, and with Other Evi-  . . . . . . . . . . . . . . . 198  dentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202 4.6  Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205  vii  Table of Contents 5 The Pragmatics of Evidentiality in Gitksan . . . . . . . . . . . . . . . . 207 5.1  The Issues  5.2  The Plan  5.3  ’ The Dynamic Semantics of nakw  5.4  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210 . . . . . . . . . . . . . . . . . . . . . . 212  5.3.1  Pragmatic Presupposition . . . . . . . . . . . . . . . . . . . . . . 213  5.3.2  Context Change Potential . . . . . . . . . . . . . . . . . . . . . . 215  5.3.3  Presupposing Sensory Evidence and the Common Ground . . . . 216  5.3.4  ’ The Context Change Potential of nakw (p) . . . . . . . . . . . . . 219  ’ nakw as a Sentential Force Specifier  . . . . . . . . . . . . . . . . . . . . 220  5.4.1  A Case for Evidential Sentential Force . . . . . . . . . . . . . . . 221  5.4.2  Evidentials as Sentential Force Specifiers (Portner 2006) . . . . . 224 5.4.2.1  The Model of Discourse . . . . . . . . . . . . . . . . . . 226  5.4.2.2  Meanings of the Evidentials as Update Functions . . . . 227  5.4.2.3  ’ nakw as an Evidential Sentential Force Specifier  5.4.2.4  Interaction Between the Evidential Modals =ima and =kat  5.4.3  . . . . 228  . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229  ’ nakw is Not an Illocutionary Force Modifier . . . . . . . . . . . . 232  5.5  ’ Explaining the Interaction Between =ima and nakw : Pragmatic Blocking 234  5.6  ’ The Extended Pragmatics of nakw : Mirativity and Metaphor 5.6.1  5.7  Approaching the Category of Mirativity . . . . . . . . . . . . . . 243 5.6.1.1  Evidentiality and Mirativity  . . . . . . . . . . . . . . . 244  5.6.1.2  Mirativity and Epistemic Modality . . . . . . . . . . . . 248  5.6.2  An Analysis of Mirativity as Conversational Implicature . . . . . 251  5.6.3  Nonliteral Uses of Evidentials . . . . . . . . . . . . . . . . . . . . 256  Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263  6 Conclusion 6.1  . . . . . . 238  . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265  Revisiting the Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . 269 6.1.1  The Conceptual Relations . . . . . . . . . . . . . . . . . . . . . . 269 viii  Table of Contents 6.1.2  The Encoding Relations . . . . . . . . . . . . . . . . . . . . . . . 269  6.1.3  The Formal Semantic Relations . . . . . . . . . . . . . . . . . . . 270  6.1.4  The Formal Pragmatic Relations . . . . . . . . . . . . . . . . . . 272  Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274  Appendices A Gitksan Orthography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285  ix  List of Tables 1.1  Lexically vs. contextually determined modal meaning and force . . . . .  28  2.1  A sample typology in Aikhenvald’s categorization of evidential meanings  40  2.2  The evidential system in Nisgha’a (Tarpent 1987) . . . . . . . . . . . . .  42  2.3  A description of the evidential system in Gitksan . . . . . . . . . . . . .  42  3.1  Test results for St’át’imcets and Quechua . . . . . . . . . . . . . . . . . . 124  3.2  A Propositional/Illocutionary typology based on the levels of meaning tests. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124  3.3  The Propositional vs. Illocutionary status of the evidential System in Gitksan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125  3.4  ’ The embeddability of =ima, =kat, and nakw. . . . . . . . . . . . . . . . . 145  3.5  ’ . . . . . . . . . . . . . . . . . . . 151 Test results for =ima, =kat, and nakw  4.1  Lexically vs. contextually determined modal base and force (cf. Rullmann et al. 2008) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160  4.2  Possible worlds that can be distinguished in terms of three propositions. . 174  4.3  Empty vs. Non-empty Ordering Sources in Strong/Weak Modals . . . . . 188  4.4  Speaker/Addressee knowledge of the answer across question types . . . . 194  4.5  Knowing and requiring an answer across question types . . . . . . . . . . 198  5.1  The modal vs. illocutionary status of the evidential system in Gitksan . . 207  x  List of Figures 1.1  Language Families of the Pacific Northwest Coast (source: maps.fphlcc.ca).  1.2  Map of the Three Tsimshianic Territories (and neighbouring languages): ’ Coast Tsimshian (Smalgyax), Nisgha’a, and Gitksan (source: maps.fphlcc).  6  7  xi  List of Abbreviations The abbreviations used in this dissertation are adapted from Rigsby (1986) and Tarpent (1987). Those marked with an asterisk are introduced in this dissertation: -  suffic boundary  =  clitic boundary  1  first person  2  second person  3  third person  caus  causative  cnd  common noun determiner*  comp  complementizer  cond  conditional  conj  conjunction  dem  demonstrative  dist  distal  distal.loc distal locative evid  evidential  geo.loc  fixed geographic location  xii  List of Abbreviations fut  future  imperf  imperfective  incept  inceptive  interrog interrogative ints  intensifier  irr  irrealis  loc  locative  mod  modal  neg  negation  obl  oblique  obl.pro oblique pronoun pl  plural  poss  possessive  prog  progressive  pnd  proper noun determiner*  redup  reduplication  refl  reflexive  rel  relativizer  rep  reportative  sg  singular  s.rel  subject relativizer  tr  transitivizer  xiii  Acknowledgements I’m very fortunate to have met many Gitksan people over the past five years, spending hours upon hours absorbing their knowledge of the language, way of life, culture, and hospitality. I begin with Doreen Jensen, my first Gitksan language consultant: Doreen’s dedication, generosity, and knowledge sets a true example, not just as a language consultant, but as a friend. Sadly, she passed away just as I was preparing the final draft of this dissertation, and I wish she could’ve seen yet another one of the many things she contributed to in her dedication to her language and culture. I also spent many hours over the years with Barbara Sennott, whose sense of humour and patience with my various adventures and follies in Gitksan were crucial to this dissertation. I also spent a lot of time in the Gitksan speaking communities as their guest. In Kispiox I greatly benefited from the knowledge, and generosity of Margaret Heit, Fern Weget, Gwen Simms, Holly Weget, Leiwa and Alvin Weget, Sheila Campbell, Clyde and Marlene Skulsh, Louise Wilson, Bob Wilson, Roy Wilson, David Harris, and Jeffery Harris. From Gitanmaax (Hazelton): Thelma Blackstock, Fanny Smith, and Jane Smith. From Kitwanga: Chief Gary Williams. From Kitsegukla: Ray Jones. From Sigit’ox (Glen Vowell): Frances Sampson. There are also other people in the Gitksan communities who extended a hand of friendship and helped me along the way, including Pete, and Chuck Heit, Shirley Muldo, and Laurel Mould. I walked into Lisa Matthewson’s office one day and asked her to be my supervisor. I had almost no training in semantics, and a handful of half-baked ideas – she could’ve easily declined. Thankfully for me, she didn’t. Lisa tirelessly and enthusiastically supported everything I did, from the first elicitation plan, to preparing for the defence. I’ve decided the best way to show my gratitude to her is to follow her example: giving xiv  Acknowledgements nothing less than 110% to my students. I have the same respect and gratitude for the other two members of my committee, Hotze Rullmann and Henry Davis. I really enjoyed sitting in Hotze’s office, where we freely riffed ideas together on semantics and other topics – a great way to learn. Henry’s ‘tough love’ quite simply made me a better linguist. His uncompromising ideals of getting the data right made this a better dissertation. Rose-Marie Déchaine and Ori Simchen were my university examiners, and I especially appreciate the time they took to work through this disseration and give me their thoughtful questions and comments. Rose-Marie and I go back to my undergrad days, and I can trace my interest in fieldwork to her. Peggy Speas also gave a lot of her time as external examiner; her attention to detail made me think of issues I hadn’t considered before. Doug Pulleyblank and Eric Vatikiotis-Bateson were instrumental people in my life at UBC. Doug’s encouragement got me into grad school, and Eric’s kept me there. The faculty at UBC are an amazing group of people: their dedication to their students – regardless of their area of speciality – makes a lasting impression. I was priviledged to work with Gunnar Hansson, Martina Wiltschko, Bryan Gick, Joseph Stemberger and Patricia Shaw, who was particularily supportive of my endeavors. And then there is Edna Dharmaratne: she sets the standard because she is everything: an administrator, counselor, provider of food, sanity, and most importantly friendship. I enjoyed the friendship of many people in the linguistics department. Wow, it’s been 7 years, and I hope I can name everyone (in totally random order): Solveiga Armoskaite, Leszek Barczak, Mario Chavez-Peon, Yunhee Chung, Carrie Gillon, Peter Jacobs, Masaru Kiyota, Karsten Koch, John Lyon, Calisto Mudzingwa, Ian Wilson, Ryan Waldie, Mark Scott, Anita Szakay, Beth Rogers, Martin Oberg, Sonja Thoma, Murray Schellenberg, Scott Mackie, Patrick Littell, Christiana Christodoulou, Donald Derrick, Joel Dunham, and Raphael Girard, and Analia Gutierrez. Among these, three people stand out  xv  Acknowledgements to me: Jason Brown has been an unconditional friend since the beginning; He was always there for the tough moments. Susie Jones helped keep things real, and I will always think of Fusheini Hudu and the those 3 a.m. discussions we had working in our office together for almost three years. My research on Gitksan was greatly facilitated by Bruce Rigsby, whose collegiality and generosity was invaluable. I hope I’ve upheld his high standards of research on the language. Jay Powell and Marie-Lucie Tarpent were also very supportive. I benefitted from discussions with other linguists and scholars in related fields, including Irene Heim, Kai von Fintel, Eric McCready, Paul Portner, Denny Moore, Amelia Reis Silva, John Wynne, Andrés Salanova, and Charles Menzies. I am also greatful to Johan Rooryck and Willem Adelaar at Leiden University for letting me take up my post-doc position there while in the final stages of this dissertation. I look forward to working with them. My non-linguist friends and family, and they helped me keep things in perspective. I can only name a few of the many: Svea Thompson, Divona Leitner, Oscar Pederson, Anna Peterson, Derek Franson, Nancy and Nadine Dinshaw, Leighton Matthews, Lara Tarantini, Mary McKinney, and Lars, Anne and Laura Kaario. Finally, there are two special people that were a part of this journey: Gessiane, Ailime (I hope to read your dissertation some day!), and the Picanço family in Brazil. This research relied on the support of several funding agencies, including the SSHRC Doctoral Fellowship program, a Jacobs Fund Grant (The Whatcom Museum Society) and a Phillips Fund for Native Research Grant (The American Philosophical Society), and the Endangered Languages Documentation Programme (SOAS). Completing the PhD program – and in fact many things in my life – would’ve been considerably more difficult if it were not for the unconditional support of my parents, Betty and Victor Peterson. It’s a small token compared to what they’ve done for me, but I dedicate this dissertation to them.  xvi  Chapter 1 Introduction 1.1  Goals of the Investigation  The aim of this dissertation is to provide an empirically driven, theoretically informed investigation of how speakers of Gitksan, a Tsimshianic language spoken in the northwest coast of Canada, express epistemic knowledge about the world around them. There are three main goals that motivate this investigation, summarized in (1.1): (1.1) (i.) To provide the first detailed description of the evidential and modal system in Gitksan. (ii.) To provide a formal semantic and pragmatic account of this system that adequately explains the meanings of the modals and evidentials, as well as how they are used in discourse. (iii.) To identify and examine the specific properties the Gitksan evidential/modal system that bear upon current theories of semantics and pragmatics, as well as the consequences this analysis has on the study of modality and evidentiality cross-linguistically. The central assumption that drives this research is that a universal property of all languages is the ability to express notions of epistemic possibility and probability, and that language users have a variety of devices for encoding both how they acquire knowledge about the world around them, as well as their attitude towards that knowledge. One of these ways is through linguistic modality. The primary function of linguistic modality is to enable us to talk about possibilities and necessities in terms of our attitude towards a proposition. Thus, we can use a modal expression to talk about the 1  1.1. Goals of the Investigation ways the world could be – or a possible world – given what we know about a situation at the moment we utter it. For example, in the Gitksan example in (1.2), a speaker is claiming that, in some possible world consistent with what they know about August, or their experience with picking berries, the berries are ripe: (1.2) mugwimahl maay’ mukw=ima=hl maay’ ripe=ima=cnd berries “The berries might be ripe.” This is an example of epistemic modality, and it gives us the ability to go beyond directly observable facts and events in the actual world we live in. However, some languages take into account directly or indirectly observable evidence a speaker has for making an assertion by lexically encoding the source of evidence a speaker has for the epistemic statements they make. This characterizes an evidential system. Gitksan ’ also possesses an evidential system. In example (1.3), the morpheme nakw encodes the indirect visual evidence a speaker has for asserting that the berries are ripe; they see people returning from the picking grounds with buckets full of berries, and they infer from this visual evidence that the berries are ripe:. ’ (1.3) nakwhl mukwhl maay’ ’nakw =hl mukw=hl maay’ ’ nakw =cnd ripe=cnd berries “Looks like the berries are ripe.” This sets the empirical stage for this dissertation, which is an investigation of the semantics and pragmatics of epistemic knowledge in Gitksan in terms of how the language encodes specific kinds of evidence, and how these line up with the modal notions of necessity and possibility. I present evidence that Gitksan not only encodes different types of evidence, but that these evidentials are capable of expressing modal meanings such as necessity and possibility. I also show that the discourse context governs which and how these evidentials are used, distributing evidential and modal meaning across the 2  1.1. Goals of the Investigation semantics and pragmatics of the language. Thus, evidentiality and modality in Gitksan is indeed a semantics-pragmatics interface phenomenon. There are a variety of theoretical tools that we can apply in investigating evidentiality and modality in a language. These tools are introduced in §1.3 below. In addition to determining the kinds of evidence types that are lexically encoded in Gitksan, one of the first analytical tasks is determining what level of meaning the specific evidentials operate on. This invokes a line of empirical and theoretical research which studies the connection between evidentiality and epistemic modality. Currently, we ca distinguish two different types of evidentials, defined by the level of meaning they operate on: propositional and illocutionary evidentials.1 I take these two types to correspond to a distinction between modal evidentials and non-modal evidentials, respectively. Determining the status of an evidential in these terms will then indicate the kind of formal analysis that evidential will be amenable to. In (1.4) I outline the status of the evidentials in Gitksan in terms of these two types, and also the type of analysis that is used to explain their meanings in this dissertation: (1.4) Gitksan has both propositional and illocutionary evidentials. This leads to the following two main claims: (i.) Modal evidentials are treated as a specialized type of epistemic modals (Izvorski 1997; Matthewson et al. 2004). (ii.) Non-modal evidentials are sentential force specifiers (Portner 2006). There are four specific theoretical and analytical outcomes that follow from the claims in (1.4), outlined in (1.5): (1.5) (i.) A novel analysis of variable modal force in modals with fixed quantification: variable modal force in Gitksan modal evidentials is determined by the ordering source. 1  Terminology adapted from Faller (2002).  3  1.2. Empirical Context (ii.) The introduction of Conjectural Questions: when a modal evidential is added to a question it reduces the interrogative force of the question. This follows from its modal semantics. (iii.) The introduction of Pragmatic blocking: modal and non-modal evidentials interact in discourse contexts, and implicate a speaker’s attitude (modal force) towards the evidence they have for a proposition. (iv.) The first formal analysis of mirativity and non-literal uses of evidentials, which are treated as conversational implicature. This chapter introduces the specific empirical and theoretical issues involved in (1.4) and (1.5), which are addressed in detail in this dissertation. In §1.2 I place this study within an empirical context by briefly reviewing the typological/functional study of evidentiality. I also introduce the Gitksan language and the relevant features of its evidential system that are the focus of this dissertation. In §1.3 I turn to the theoretical context, outlining the specific kinds of theoretical tools that are necessary for undertaking a study of this kind. §1.4 provides an outline of the dissertation, including a discussion of the specific theoretical and analytical outcomes in (1.4) and (1.5). Finally, §1.5 discusses the methodology I followed in investigating this data.  1.2  Empirical Context  This investigation into epistemic modality and evidentiality in Gitksan can be placed into two empirical contexts. The first context is local: this investigation contributes to both Gitksan language scholarship, and to scholarship on the Tsimshianic language family, by documenting an area of the grammar that has not been previously studied. Additionally, this dissertation is the first semantic and/or pragmatic study of a Tsimshianic language. The other context is the study of evidentiality and epistemic modality cross-linguistically. Thus, it is hoped that this study can show how Gitksan fits in, and what it can poten4  1.2. Empirical Context tially tell us about evidential meanings from a typological perspective. In subsections 1.2.1 and 1.2.3 I introduce Gitksan and the Tsimshianic language, as well as the basic morphological and syntactic features of Gitksan. In §1.2.4 I present the empirical scope of this dissertation, and highlight the central features of the evidential system in Gitksan which are analyzed in the subsequent chapters. In §1.2.5 I discuss the typological research on evidentiality and epistemic modality.  1.2.1  Gitksan and the Tsimshianic Languages  The Tsimshianic languages are spoken on the northwest coast of Canada, almost entirely within the province of British Columbia, adjacent areas of the interior, and the southern tip of the Alaska panhandle (see Figures 1.1 and 1.2 on the following pages).2 There are four linguistic and socio-cultural divisions that make up the Tsimshianic family, given in (1.6):3 (1.6) The Tsimshianic Languages (Rigsby 1986, p. 25) Coast Tsimshianic ’ Coast Tsimshian (Smalgyax) South Tsimshian (Sgüüxs) Interior Tsimshianic Nisgha’a Gitksan ’ The Coast Tsimshian (Smalgyax) reside to the north and south of the Skeena River delta, and the South Tsimshian (Sgüüxs) were reported to live to the south of this area, primarily in the villages of Klemtu and Hartley Bay. The Nisgha’a reside in the Nass River Valley and along Observatory Inlet, and the Gitksan reside in the easterly adjacent upper Skeena and Kispiox valleys, and the Skeena watershed. 2 3  All areas are approximate. Also note the alternate spellings/pronunciations of Gitxsan, and Nisga’a. South Tsimshian (Sgüüxs) – now considered likely to be extinct – is not shown on this map.  5  1.2. Empirical Context  Figure 1.1: Language Families of the Pacific Northwest Coast (source: maps.fphlcc.ca).  6  1.2. Empirical Context  Figure 1.2: Map of the Three Tsimshianic Territories (and neighbouring languages): ’ Coast Tsimshian (Smalgyax), Nisgha’a, and Gitksan (source: maps.fphlcc).  7  1.2. Empirical Context The word Gitksan is morphologically complex, meaning ‘people of the Skeena River’ (git- ‘people of’, xsan ‘(to) gamble’; ‘Skeena River’). The Gitksan often refer to their lan’ ’ guage as simalgax, which means ‘the real or true language’ (sim-algax ‘true-language’). The language has been referred to as Gitxsan or Gitksan by scholars, or Gitxsanimx or Gitxsanimax by native speakers when distinguishing it from Nisgha’a (Nisga’amx) or Coast Tsimshian (Ts’imsanimx). However, the Nisgha’a and Coast Tsimshian people ’ This has created some confusion, since also refer to their languages using simalgax. 4 ’ many published sources on Coast Tsimshian simply refer to the language as Smalgyax.  Gitksan comprises of two major varieties, or dialect areas: Western Gitksan, Geets’imx (‘downstream language’), and Eastern Gitksan, Gaanimx (‘upstream language’). Western Gitksan is spoken in the settlements along the Skeena river, including the villages of Kitwanga (Gitwingax ‘people of the place of rabbits’), and Kitseguecla (Gijigyukwhla ‘people of Jigyukwhla’, also the name of a nearby mountain), and also in the village of Kitwancool (Gitwinhlguu’l ‘people of the narrow place’), north of Kitwanga.5 Eastern Gitksan is spoken primarily in the settlements along the Kispiox river, in villages of Kispiox (Ansbahyaxw, ‘hiding place’), Glen Vowell (Sigit’ox, also the name of a nearby mountain), and Hazelton (Gitan’maaxs, ‘people of the place of torchlight fishing’), at the junction of the Skeen and Kispiox rivers. There are also two other main villages in northern Gitksan territory (upper Kispiox river watershed) that are no longer inhabited: Kisgegas (Gisgaga’as ‘people of the place of small white gulls’) and Kuldo (or Gitgaldoo’o, ‘people of the wilderness or backwoods’). Variation between the two dialects is minimal, and consists mostly in phonological shifts and lexical differences (Brown 2008). While there are no exact or official figures, it is currently estimated that there are approximately 400 speakers of Gitksan, most of whom are over the age of 50. Although there are some teaching materials (Powell and Stevens 1977), as well as recent efforts 4  Further discussion of the use of these terms can be found in Rigsby (1986), and see Brown (2008)  for a detailed discussion of Gitksan and Tsimshianic relations and language scholarship. 5 Some Gitksan speakers claim that the Gitksan spoken in Kitwancool is a third dialect.  8  1.2. Empirical Context to introduce the Gitksan language into the public school system, children are no longer acquiring the language. These facts place Gitksan on the list of the world’s many endangered languages.  1.2.2  Gitksan Orthography  The practical orthography developed for Gitksan by Hindle and Rigsby (1973) is used throughout this dissertation to ensure that the data here is accesible to the speakers and teachers of Gitksan. The Hindle and Rigsby orthography, along with the phonetic equivalents, is given in Appendix A.  1.2.3  The Basic Features of Gitksan Morphology and Syntax  This subsection introduces the basic features of Gitksan morphology and syntax that are relevant to this study.6 1.2.3.1  Morphology  Verb morphology in Gitksan is less complex than in other languages in the northwest coast linguistic area, such as in the neighbouring Wakashan and Athapaskan languages (Rigsby 1986, p. 84). For example, a verb in Gitksan may be morphologically simple, as a typical intransitive clause in (1.7) shows: (1.7) limxt Margaret limx=t Margaret sing=pnd Margaret “Margaret sings/sang.” However, a Gitksan verb frequently hosts a range of morphology, such as transitivity, voice, number and aspect, as well as enclitics, which include the determiners =hl, =t, 6  For more detailed descriptions of Gitksan clause structure and agreement see Rigsby (1986) and  Hunt (1993).  9  1.2. Empirical Context dip, and =s.7 A template for verbal morphology is given in (1.8): (1.8) [[verb [verb theme preverbs[verb stem [verb root verb((-connector)-compound)] -derivational suffixes]]-pronominal suffixes]=enclitics)] Example (1.9) is a transitive sentence which exemplifies this template, showing the typical morphological complexity of a Gitksan verb: (1.9) gunsixpts’axwis Billt Clara ’as Gwen gwin-si-xpts’axw-i-(t)=s Bill=t Clara ’a=s Gwen caus-caus-fear-tr-3=pnd Bill=pnd Clara obl=pnd Gwen “Bill had Gwen frighten Clara.” 1.2.3.2  Syntax and Clause Types  There are two distinct clause types in Gitksan (and also in the other Tsimshianic languages) that Rigsby (1986) calls the Independent, and the Dependent clauses. Generally speaking, Independent clauses have a canonical VSO word order: the verb is the first constituent in the clause followed by the subject and object, as example (1.10) shows. Dependent clauses are clauses where another constituent heads the clause, followed then by the verb, subject and object. (1.10)  a. Independent jabis Sheilahl hon tsap-i-(t)=s Sheila=hl hon cook-tr-3=pnd Sheila=pnd fish “Sheila cooked the fish.”  7  However, the status of the =s has not been completely determined. It has also been analyzed as  a case marker. See Hunt (1993) for a synchronic analysis of =s, and Peterson (2006) for a diachronic analysis.  10  1.2. Empirical Context b. Dependent yukwt tsaps Sheilahl hon yukw=t tsap-(t)=s Sheila=hl hon prog-3 cook-3=pnd Sheila=cnd fish “She is cooking the fish” Independent and Dependent clauses are distinguished by a variety of morphological differences, but one of the key distinguishing features concerns the form, and distribution of the agreement markers on the verb and the consituent before the verb. In Independent clauses such as (1.10)a., the verb only bears one agreement marker (which is often silent), a suffix indexed to the ergative argument, Sheila. Independent clauses are used only in main clauses, and occur only when the verb is not preceded by certain material, which include certain intransitive verbs, temporal/aspectual particles, sentential conjunction, and subordinators.8 In Dependent clauses such as (1.10)b. the verb is preceded by another constituent, in this case the the progressive marker (yukw ). In Independent clause, the agreement patterns become more complex, as two agreement markers surface: the enclitic =t attached to the preverbal material yukw, which is indexed to the ergative argument (Sheila), and a suffix -t on the verb indexed to the absolutive argument (the fish).9 A more detailed discussion of the syntax and clause types in Gitksan is given in chapters 3 and 5.  1.2.4  Evidentiality and Epistemic Modality in Gitksan  Examples (1.2) and (1.3) above gave a glimpse of how Gitksan expresses epistemic and evidential knowledge. In this subsection I highlight in more detail some of the core lexical items and constructions that are examined in this dissertation. 8 9  See Hunt (1993) for more details regarding the syntax and agreement in Gitksan. See Peterson (2006) for more details on the morphosyntax of ergative agreement in Gitksan and  the other Tsimshianic languages.  11  1.2. Empirical Context Consider a scenario where two friends are sitting at a restaurant having coffee. Your friend just got back from John’s place, where she was helping him set out the fish for smoking. You ask her what John’s doing today, and she answers in (1.11): (1.11) sihont si-hon=t caus-fish=pnd “John is doing up  John John John fish.” (processing, cleaning, smoking, canning)  She knows John is smoking fish because she just returned from helping him. In a slightly different version of this scenario, you’re thinking of asking John for a favour, but your friend suggests he might be busy. This time, your friend wasn’t at John’s place. She says (1.12): John (1.12) sihongatit John si-hon=kat=t caus-fish=kat=pnd John “[I heard] John is doing up fish.” By inserting =kat into the sentence in (1.12), your friend is telling you that she doesn’t know directly whether or not John is smoking fish, but that she overheard it, for example, at the gas station. In other words, she has indirect evidence in the form of a report for the information that John is doing up fish. It’s now August, and you and a friend need to ask John for another favour. You haven’t seen John in a while, but because it’s the end of summer – the time of year for smoking fish – your friend thinks he might be too busy to help because of (1.13): (1.13) sihonimat John si-hon=ima=t John caus-fish=ima=pnd John “John might/must be doing up fish.” The insertion of =ima into (1.13) also encodes a kind of indirect evidence, but this time it comes from knowledge of previous experiences with the fishing season. Note 12  1.2. Empirical Context the translations of =ima in (1.13): if a speaker is simply speculating that because it’s August, it’s only a possibility John is doing up fish, then =ima can be translated as might. However, a speaker may feel more certain that John is doing up fish because they know that every year he does, and that there is so much fish this year others will need his help, and that he received a large quantity of fish from his brother yesterday. This is a stronger base of reasoning for the speculation, making it very probable John is doing up fish. In this scenario, =ima can be translated as must. Now you decide to find out for yourself what John is up to, so you drive by his place: you see smoke coming out of the wilp sihon (smokehouse), and his truck in the driveway. Your friend exclaims (1.14): ’ (1.14) nakwhl sihons John ’ nakw =hl si-hon-(t)=s John ’nakw =cnd caus-fish-3sg=pnd John “John must be processing fish.” “It looks like John’s doing up fish.” ’ The use of nakw in this sentence means that your friend is inferring from something they actually see – the smoke and the truck – that John is doing up fish. In addition to being translated as ‘It looks like...’, this kind of visual evidence corresponds with a ’ stronger modal translation using must. nakw is different from =ima because there is ’ some kind of sensory evidence for the inference. This means that nakw is infelicitous in the previous context where you and your friend are simply speculating. ’ The morphemes =kat, =ima and nakw are used by Gitksan speakers to encode the kinds of knowledge they have for making inferences and claims about the world around ’ them. Also note that the translations of =ima and nakw given in (1.13) and (1.14) give us glimpse into how these evidentials interact in different contexts to implicate the expression of modal force. 10  10  There are other expressions of epistemic modality in Gitksan, such as the attitude verbs ha’nigoodi  ‘think/believe’, and wilaa ‘know’. In this dissertation I will limit myself to examining the morphology specialized for evidentiality and epistemic modality, setting aside for another study the propositional attitude verbs and other adverbials expressing epistemic modality.  13  1.2. Empirical Context Previous research on modality and/or evidentiality in Gitksan or any of the Tsimshianic languages is limited: Rigsby in his grammar of Gitksan (1986) does not discuss modality nor evidentiality. Tarpent in her dissertation (1987) on Nisgha’a has some brief description of modals and evidentials in that language. In §1.4 I map out how I plan to fill this empirical gap in Gitksan by examining and analyzing in detail the complete range of meanings and uses of these morphemes.  1.2.5  Studies on Evidentiality  Studies on evidentiality from a typological or functional perspective have been largely shaped by the seminal collection of papers in the Chafe and Nichols (1986) volume Evidentiality: The Linguistic Coding of Epistemology, and Aikhenvald’s sizeable (2006) study of evidentials in Evidentiality. Whereas the Chafe et al. volume presents a variety of papers looking at the functions of evidentials in a few languages (including descriptions of the extended meanigs of evidentials, such as mirativity), Aikhenvald was the first to present an encompassing typology of evidential systems based on a large number of languages.11 From the perspective of epistemic modality, a central work is Palmer (2006), who treats evidentiality as a feature of epistemic modality crosslinguistically. The data presented in this dissertation bears upon both perspectives of evidential and modal meaning cross-linguistically, by offering a detailed study of how these meanings are manifested in a single language. This kind of focused research is needed in order to further our understanding of the range of variation in the kinds of evidence sources evidentials may encode, and the network of relationships they may form with one another, and how they are used in discourse. 11  See also Aikhenvald and Dixon (2003) for a collection of language specific studies on evidentiality.  For a useful overview of the state of the art in evidentials research, see Speas (2008) and Rooryck (2001a;b).  14  1.3. Theoretical Context  1.3  Theoretical Context  One of the challenges facing typological studies of evidentiality is that, because there are so many descriptive studies for evidentials in various languages, the descriptions and terminology of different studies is not always consistent or rigorously verifiable (cf. Matthewson 2004; von Fintel and Matthewson 2008). There is an emerging branch of research which approaches this problem by developing a more theoretically informed and testable methodology for investigating evidential categories (cf. Faller 2002; de Haan 1998; 1999). A result of this is that we are now better equipped with a range of tools derived in contemporary semantic and pragmatic theory that allow us to develop the kinds of field methodologies we need for both investigating and explaining evidential meaning. This dissertation contributes to this line of research, as an empirically driven, theoretically informed investigation of modal and evidential meaning in Gitksan. One of the outcomes of this kind of undertaking is that we can see how there is a dynamic relationship between data and theory. One one hand, we can use theoretical models to make predictions about the kinds of data we are investigating. On the other hand, the kinds of data that emerge from this in turn may invoke different kinds of theories and analyses, and also require us to adjust or refine these theories in order to better explain the data. This two-way relationship can be observed when we search for the answer to the kinds of research questions such as these: what does evidentiality look like in language X, and how can we explain it? In pursuing answers to these questions in Gitksan, I show that in order to explain evidentiality in Gitksan we need a theory of static semantics (the fixed meanings of an evidential), a theory of dynamic semantics (how the felicity of certain evidentials is constrained by and affects the discourse context). §1.3.2 introduces the kinds of theoretical tools needed to adequately describe and explain evidential meaning in Gitksan: in §1.3.2.1 I review the specific theoretical diagnostics that are used to determine the level of meaning a particular evidential operates on. The question here is, does a particular 15  1.3. Theoretical Context evidential operate on a propositional level (i.e. it contributes to the truth conditions of a proposition), or an illocutionary level (i.e. it does not contribute to the truth conditions of a proposition)? Once we determine what level an evidential operates on, the next task is to decide on the appropriate theoretical analysis, the outcome of which will allow us to test specific predictions as to the semantics and pragmatics of a specific evidential. This directly taps into a current issue in the formal study of evidential meaning: §1.3.2.2 introduces the current approach to evidentials, which are analyzed as a specialized type of epistemic modal. This represents a static semantic analysis of evidentials. At this point we can now begin to distinguish modal evidentiality (evidentials as epistemic modals) from non-modal evidentiality (illocutionary evidentials). Non-modal evidentials require a different kind of semantics, since their meanings operate above the propositional level, at the sentence level. There are currently several competing theoretical approaches to non-modal evidentials. §1.3.2.3 introduces a theory of dynamic semantics, which analyzes evidential meaning in terms of the way in which an evidential changes/updates a discourse context. Evidentials, whether modal or non-modal, are used in discourse. They not only enter into relationships with each other, but also acquire and implicate other meanings when used by speakers in certain contexts. We saw this in the examples involving =ima ’ and nakw in (1.13) and (1.14) and how they implicate modal force. However there are ’ also meanings associated with nakw that are directly tied to the context it’s used in, ’ specifically in how nakw can be used to express surprise (mirativity) or metaphor. Thus, we need a theory of conversational meaning that can handle the kinds of meanings that can be implicated by both modal and non-modal evidentials. These different analytic components are not mutually exclusive: evidentials which have a static semantics (e.g. they are epistemic modals) also have a dynamic semantics and can implicate other kinds of meaning in conversation. However, this does not necessarily work the other way: an evidential that operates on an illocutionary level does  16  1.3. Theoretical Context not necessarily have a static meaning. Thus, we begin to see how evidential systems are not uniformly semantic or pragmatic, but rather truly an interface phenomenon. As a point of entry into these topics, it is useful to introduce one of the central issues in investigating the concept of evidentiality, which is done in §1.3.1: to what extent are epistemic modality and evidentiality conceptually related to one another.  1.3.1  The Relations Between Evidentiality and Epistemic Modality  Probing the conceptual relationship between epistemic modality and evidentiality can be facilitated by identifying four specific relations between the two, outlined in (1.15) (adapted from Schenner 2010): (1.15) (i.) Conceptual relation: What is the relation between the concept of evidentiality and the concept of epistemic modality? (ii.) Encoding relation: What is the relation between the encoding of evidential concepts and the encoding of modal concepts in a language? (iii.) Formal semantic relation: Can evidentials in language be analyzed parallel to epistemic modals in formal semantics (i.e. quantification over possible worlds) or do evidentials require a different, independent kind of analysis? (iv.) Formal pragmatic relation: What kinds of meanings to evidentials acquire or implicate in conversation? How are these implicated meanings related to their core evidential or modal meanings? Two overlapping conceptions of evidentiality can be identified: under a narrow conception, evidentials are linguistic markers that encode the speaker’s type of evidence or source of information for an assertion (de Haan 2001; Lazard 2001; Faller 2002; Aikhenvald 2006; and others). As such, evidentiality and epistemic modality are treated (in 17  1.3. Theoretical Context principle) as independent categories. A wider conception views evidentials as linguistic markers that indicate the speaker’s type of evidence for her claim and/or the degree of its reliability, probability or certainty (Palmer 2006; Ifantidou 2001; Rooryck 2001a;b; and others). Under this conception, evidentials are essentially a sub-species of epistemic modals. At the conceptual level evidentiality (understood as a person’s type of source of information) and epistemic modality (understood as the person’s judgment regarding the necessity or possibility of a proposition expressing the relevant piece of information) are different but related: The type of source influences the strength of belief in the truth of the proposition (see Davis et al. 2007; Speas 2008). However, the two concepts are not always entirely interchangeable because other contextual factors are at play. If a person asserts a proposition necessary, this by itself does not overtly communicate the type of evidence the person has for making that assertion. For example, the fact that a person has reportative evidence for the truth of a proposition does not determine a certain strength of belief in this proposition. This link between evidentiality and epistemic modality has been the focus of much recent research, where many researchers claim that some evidentials in some languages are a specialized subtype of epistemic modality (Willett 1988, p. 52; Izvorski 1997; Palmer 2006; Matthewson et al. 2004; Rullmann et al. 2008; Peterson 2008). Within epistemic modality itself there are two basic construals: according to the first, epistemic modals indicate the degree of commitment of the speaker to the embedded proposition, or the prejacent. Palmer (2006, p. 54) argues that evidentials are epistemic modals, because “their whole purpose is to provide an indication of the degree of commitment of the speaker: he offers a piece of information, but qualifies its validity for him in terms of the type of evidence he has”. According to the second, epistemic modals mark the necessity or possibility of the prejacent relative to some body of evidence or knowledge (von Fintel and Gillies 2007; von Fintel and Iatridou 2009).  18  1.3. Theoretical Context Indeed, cross-linguistic research has shown that languages differ in this respect. On the one hand there are languages that possess clearly distinct systems for evidentiality and epistemic modality. For example Pawnee, Wintu and Makah have distinct sets of morphemes for coding modal and evidential meanings (cf. Mithun 1999; Aikhenvald and Dixon 2003). On the other hand there are languages that use a single set of markers to denote evidential source and epistemic strength, as we will see in Gitksan. In general, epistemic modals often acquire evidential meanings (see von Fintel and Gillies (2007) for a discussion of the evidential properties of epistemic modals in English), and evidentials may be extended to epistemic modal meanings (cf. Faller 2002). What this suggests is that evidentials and epistemic modals combine expressions in various ways encoding evidential and epistemic meaning components. In §1.3.2 I introduce the specific tools for exploring these two relations.  1.3.2  The Theoretical Tools  A recently emerging generalization in the theoretical literature on evidentials is that evidential interpretations can arise on different levels of meaning. This has led to the discovery that in some languages, evidentials are a specialized type of epistemic modal: they are semantic operators that contribute to truth conditions; while in other languages evidential meanings are not a semantic phenomenon (i.e. they are not propositional operators), rather, they operate at the pragmatic level, and thus are characterized as illocutionary operators ((Izvorski 1997; Faller 2002; 2003; Davis, Potts, and Speas 2007; Matthewson, Rullmann, and Davis 2004; McCready and Ogata 2007; Davis, Matthewson, and Rullmann 2009; Rullmann, Matthewson, and Davis 2008)). The evidential system of Gitksan provides evidence that this theoretical generalization is not only cross-linguistically viable, but that the semantics-pragmatics ‘split’ in the distribution of evidentials can occur within the same language (see also Faller 2002).  19  1.3. Theoretical Context 1.3.2.1  Determining Levels of Meaning  Much recent literature has been devoted to developing various tests and diagnostics to determine whether a given evidential marker operates at the illocutionary or propositional level (Lyons 1977; Papafragou 2000; 2006; Garrett 2001; Faller 2002; 2003; Matthewson et al. 2004). Based on these studies, Waldie et al. (2009) classify these diagnostics into two broad categories: tests which are sensitive to the truth value of the prejacent in (1.16), and tests which are sensitive to scope and embeddability in (1.17): (1.16) Tests involving Truth Values (i.) Known Truth/Falsity: Is the sentence felicitous if the prejacent is known to be true or false? (ii.) Assent/Dissent: Can the contribution of the evidential be agreed or disagreed with? (iii.) Cancellability of type of evidence requirement: Can the evidence type requirement be cancelled? (1.17) Tests involving Scope and Embedability (i.) Embeddability: Can the evidential be understood as part of the propositional content of an embedded clause (i.e. the antecedent of a conditional, under a factive attitude verb, under a verb of saying)? (ii.) Scope with respect to interrogatives: Can the evidential take scope over a speech act? (iii.) Interaction with negation: Is the evidence type requirement affected by negation? It is important to note that these tests are only the foundations of a theoretical explanation of the behaviour of evidentials. Another way to view them is by drawing an analogy between these tests and the kinds of tests one applies in determining syntactic 20  1.3. Theoretical Context constituency. For example, it is well known that before one can develop a theoretical analysis of the syntax of nominals in a language (i.e. wh-movement), it is necessary to determine the constituency of potential noun phrases (i.e. DPs) in that language. Tests for constituency include, movement, deletion, substitution, and coordination etc. The results of these tests are only part of the analysis, and provide a foundation for the next step in the analysis. Thus, the diagnostic tests in (1.16) and (1.17) provide a theoretical foundation for determining whether the evidentials in a language should be formally treated as illocutionary operators or semantic operators. In chapter 3 I work through the definition of these tests in detail as they are applied in two unrelated languages, St’át’imcets and Cuzco Quechua. Following this, I evaluate their applicability to the Gitksan data. 1.3.2.2  Static Semantics: Evidentials as Epistemic Modals  If these tests determine that an evidential is a semantic operator, we are in a position to approach the question in (1.15) (iii.): Can evidentials in a language be analyzed parallel to epistemic modals in formal semantics, or do evidentials require a different kind of analysis? There is a growing amount of research confirming the former option: propositional evidentials are analyzed as a type of epistemic modal, and do not require their own formal analysis. The most prominent analysis of natural language modality involves the application of possible world semantics, which treats modal expressions as quantifiers over possible worlds (Kratzer 1981; 1991; and others). Quantification encodes what we interpret as might and must: necessity modals are treated as universal quantifiers, and possibility modals as existential. The actual interpretations of a modal (i.e. epistemic, circumstantial and deontic) are determined by the kinds of possible worlds that are quantified over. Two independent, contextually-determined functions, or conversational backgrounds – a modal base and an ordering source – determine a set of possible worlds. Different conversational backgrounds correspond to the kinds of things we know  21  1.3. Theoretical Context about the actual world, be it the particular facts of some circumstance (circumstantial), the evidence available to us (i.e. epistemic), or the laws that are in place (i.e. deontic) etc. Thus, in English (and many other languages) we interpret the meaning of a modal by the context it’s used in. An epistemic conversational background is the set of premises it represents in view of the facts known in the actual world. For example, in view of the available evidence to us, such as the fact that his livingroom lights are on, a sentence as “John must be at home” is true iff John is home in all the possible worlds w that are compatible with the available evidence in w, such as those worlds in which his livingroom lights are on. There are languages which lexically encode specific kinds of evidential meaning, but in which the evidentials have a modal semantics. Matthewson et al. (2004), Davis et al. (2009), and Rullmann et al. (2008) claim that the individual evidentials in St’át’imcets are in fact epistemic modals. Consider the St’át’imcets inferential evidential k’a in (1.18), which means that the speaker came to believe the sentence by means of inference based on perceived evidence, or general facts about the world: (1.18) plan k’a tu7 wa7 tsu7c na máq7-a already infer then impf melt(inch) det snow-det “The snow must’ve already melted.” For Matthewson et al, k’a has a modal semantics, and its evidential meaning is encoded by presupposition, which restricts the modal base to only those worlds that are compatible with some actual-world indirect evidence. In the inferential case in (1.18), this means that the modal k’a quantifies over worlds in which (for example) people walk in the house with muddy boots instead of snowy boots. The sentence in (1.18) then asserts that in all worlds in which the proposition “people walk in the house with muddy boots instead of snowy boots” is true, the snow already melted. Because the actual world is presupposed to be a world in which there are people walking in with muddy boots instead of snowy boots, the sentence makes a claim about the actual world: the snow has melted in the actual world. 22  1.3. Theoretical Context Under this approach, evidentials have basically the same denotation as an epistemic modal: their meaning operates at the level of the proposition, and determine the truth conditions of that proposition. 1.3.2.3  Dynamic Semantics: Evidentials as Sentential Force Specifiers  A static semantics equates the meaning of a sentence with its truth conditions: the conditions in which a sentence is true or false. This is the kind of semantics that is used to account for the meaning of modal evidentials such as k’a in St’át’imcets above. However, if the tests in (1.17) show that an evidential is an illocutionary operator, a different analysis is required. Whereas propositional evidentials are amenable to a modal analysis, there are (at least) two competing analyses of non-modal evidentials: (i.) an illocutionary analysis, which treats evidentials as speech act operators (Faller 2002), and (ii.) an analysis of evidentials within a dynamic semantics (Portner 2006; Davis et al. 2007).12 Faller (2002) argues that some evidentials in Quechua do not contribute to the propositional content of an utterance. Rather, they modify the sincerity conditions of a speech act. In essence, illocutionary evidentials have the effect of introducing additional content into the set of preconditions of an assertion. Within a dynamic semantics, the meaning of a sentence is its context change potential, or the way in which it changes/updates a (discourse) context (Heim 1982; 1990). In other words, the interpretation of an utterance (with or without an evidential) is both dependent on context, and changes the context. I argue in chapter 5 that the theory of dynamic semantics provides an effective way of explaining the use of non-modal evidentials, where an evidential is defined in terms of its context change potential. Just as in a static semantics, presupposition is also crucial in determining the felicity of a non-modal evidential in a discourse context. Whereas modal 12  There is another recent theoretical approach whereby evidentials are analyzed as contributors of  not-at-issue assertions (Murray to appear a). I won’t be discussing this approach here as it was not available at the time of writing.  23  1.3. Theoretical Context evidentials have semantic presuppositions (definedness conditions on a proposition) that restrict the modal base worlds, non-modal evidentials carry a pragmatic presupposition which places a condition on their use in a conversation. For example, the felicitous use ’ of a nakw (p) sentence in Gitksan, given in (1.19), requires that the utterance context contains sensory evidence for p: (1.19) Context: You drive by John’s place and you see smoke coming out of the wilp sihon. ’ nakwhl sihons John ’nakw =hl si-hon-(t)=s John ’ nakw =cnd caus-fish-3sg=pnd John “John must be processing fish.” “It looks like John’s doing up fish.” This condition is satisfied by the context, and so uttering (1.19) is felicitous in that context. Portner (2006) adapts the theory of conversational update and introduces two modifications to the classical model: first is that the common ground is only one of potentially numerous sets of propositions in the context. These other sets represent propositions that encode cognitively or communicatively important categories, such as the different subtypes of evidentiality. Evidentials under this approach are sentential force operators that specify a kind of conversational update: they can function to conversationally update the common ground, the set of mutually believed propositions, or other specialized sets of propositions. The evidential meanings are the result of the different lexical evidentials that specify which specialized set of propositions to update in the discourse. Under Portner’s (2006) analysis, evidentials are not really illocutionary modifiers, but rather ‘sentential force specifiers’: they are grammaticized elements which specify precisely which kind of conversation update is to be performed. However, we will see in ’ chapter 5, that nakw (p) statements are not added to the common ground – they are not asserted. This is based on Faller’s insight, where there is a type of speech act with fewer 24  1.4. Structure of the Thesis commitments than assertion described as putting which can be modified in various ways to represent the different kinds of evidentials meanings in a language. Davis et al. (2007) develop a different approach using dynamic semantics, where the meanings of evidentials shift a particular contextual parameter, specifically, the degree of certainty a speaker must have in a proposition before they can utter it. Under this approach, evidentials form a hierarchy that is context dependent. Built into the theory is the notion that the effect an evidential sentence has on the context depends upon how reliable that evidential sentence is in the context.  1.4  Structure of the Thesis  This dissertation is structured around the application of these empirical and theoretical tools, the outcome of which is five proposals concerning the semantics and pragmatics of evidentiality in Gitksan. This structure is outlined in the following subsections.  1.4.1  A Description of Evidentials in Gitksan  Chapter 2 describes the inventory of evidentials in Gitksan, where I show that Gitksan has three morphemes that uniquely encode information about the speaker’s evidence for a sentence. These meanings are summarized in (1.20): (1.20) (i.) Reportative =kat The speaker came to believe a sentence by means of a report (whether that report is first-hand, second-hand, or folklore). (ii.) Inferential =ima The speaker came to believe a sentence by means of inference (whether that inference is based on perceived evidence, general facts, or previous experience with similar situations).  25  1.4. Structure of the Thesis ’ (iii.) Sensory (Direct physical) nakw The speaker came to believe a sentence by means of inference from sensory evidence (whether that sensory evidence is visual, tactile, auditory, or olfactory). This is framed within Aikhenvald’s (2006) study of evidentials, and Gitksan is placed within this typology. Along with these descriptions, the morphosyntactic properties of the evidentials in (1.20) are also discussed, as are the co-occurrence restrictions between them.  1.4.2  Modal Evidentials  Having established the kinds of evidence the evidentials in Gitksan encode, Chapter 3 applies the diagnostics in (1.16) and (1.17) to test what level of meaning the individual evidentials operate on. The outcome of these tests show that the evidentials =ima and ’ ’ =kat are propositional operators. The results for nakw are exactly the opposite: nakw cannot be an epistemic modal, but is a non-modal evidential. . In Chapter 4 I argue that both =ima and =kat semantically belong to the category of modal evidentials. In analyzing =ima and =kat as epistemic modals, I generally follow the modal analysis of evidentials in Bulgarian and St’át’imcets (Izvorski 1997; Rullmann et al. 2008): Both =ima and =kat introduce quantification over possible worlds; their evidential interpretations are the result of a presupposition that restricts the modal base to epistemically accessible worlds where evidence of some specific type holds. In the case of modal =ima, this presupposition places a condition on the modal base such that it contain worlds in which some general kind of inferential evidence holds. The presupposition attached to =kat restricts the modal base to worlds in which ’ a report was made. Despite the fact that the tests show that nakw is not a propositional ’ operator unlike =ima and =kat, the evidential meaning of nakw is also achieved by the same mechanism as for =ima and =kat: there is a sensory evidence presupposition ’ associated with nakw 26  1.4. Structure of the Thesis Using a modal analysis of =ima and =kat as a base, in chapter 4 I also identify and analyze two properties these modals have that bear on both empirical and theoretical treatments of epistemic modals: the variable modal force of =ima and =kat, and the effect =ima has when inserted into a question in creating a conjectural question. 1.4.2.1  A Challenge: Variable Modal Force  In Chapter 4 I also analyze in detail a challenge modal =ima and =kat present to the standard analysis of modality. Both =ima and =kat differ from modal auxiliaries might and must in English in two respects: first, whereas the modal base of a modal in English is determined by the context (e.g. epistemic, deontic, or circumstantial), =ima and =kat lexically encode their epistemic meanings (e.g. =ima cannot have a circumstantial or deontic meaning). Secondly, unlike modals in English, =ima and =kat do not lexically encode modal force, rather, it is determined by the context. This feature can be observed in the translations in (1.21): (1.21) yugwimahl dim iixwt yukw=ima=hl tim iixw-t prog=mod=cnd fut fish-3 “He might be going fishing.” “He must be going fishing.” “He’s probably going fishing.” “He’s likely going fishing.” “He could be going fishing.” “Maybe/perhaps he’s going fishing.” Gitksan shares this property with St’át’imcets, where the evidential modals in that language also have variable force. This ‘reversed’ arrangement between English, Gitksan and St’át’imcets is summarized in Table 1.1 below (see also Rullmann et al. 2008): This presents a challenge to the standard denotations of modals, where must and might lexically encode quantificational force. The modals =ima and =kat differ from English modal auxiliaries in that they do not lexically encode modal force. In chapter 4 I claim that these various degrees of modal force correspond to (at least) two different 27  1.4. Structure of the Thesis Modal Meaning  Modal force  English  Context (epistemic, deontic etc.)  Lexical (might, must)  Gitksan  Lexical  Context  St’át’imcets Lexical  Context  Table 1.1: Lexically vs. contextually determined modal meaning and force  types of ordering sources in Gitksan (cf. §1.3.2.2). The ordering source is a contextuallydetermined function which imposes a particular evaluative ordering of the worlds in the modal base (Kratzer 1981). Whereas the modal base contains propositions representing facts or knowledge about the world as assessed by the speaker in a given scenario, the ordering source contains propositions representing beliefs, ideals, norms, intentions, and universally-held assumptions about normal courses of events in the world. These two conversational backgrounds interact: the propositions that comprise the ordering source impose an ideal ordering on the modal base worlds. Specifically, I will argue that the weak/strong interpretations of =ima and =kat correspond to empty/non-empty ordering sources which order an existentially quantified epistemic modal base. This analysis predicts the opposite effect in languages that have a universally quantified modal base. I show how this prediction is borne out in the evidential modals in St’át’imcets. 1.4.2.2  Conjectural Questions  Continuing in Chapter 4, I focus on an additional property of =ima: the insertion of the modal =ima into a question creates a non-interrogative utterance, roughly translatable using ‘I wonder.’ Examples of =ima added to both a yes-no and wh-question are given (1.22)a. and b., respectively. Both are translated as a statement of uncertainty:  28  1.4. Structure of the Thesis (1.22)  a. neeyimahl sdinhl xbiist tusta nee=ima=hl xbiist tust=a stin-(t)=hl neg=mod=cnd be.heavy-3sg=cnd box dem=interrog “I wonder if that box isn’t heavy.” ’ ’ b. naayima ’ant ginamhl xhlawsxw ’as John ’ ’ naa=ima ’an-t kinam-(t)=hl xhlawsxw ’a=s John who=mod s.rel-3sg give-3sg=cnd shirt obl=pnd John “I wonder who gave this shirt to John.”  Following Littell et al. (2010), these constructions are called conjectural questions. A central featue of conjectural questions is that they are statements that can invite, but don’t require an answer in order to be felicitous in conversation. Additionally, an answer can be offered by either speaker. The central claim pursued at the end of chapter 4 is that conjectural questions have the semantics of ordinary questions: they denote sets of propositions. The presupposition analysis of =ima is then applied to the question: the presuppositions carried by each proposition in the question denotation conjoin, so that the conjectural question as a whole presupposes everything presupposed by each of its members. The resulting conjoined presupposition entails that there is mixed evidence, and therefore it implicates that the speaker does not expect the hearer to be able to provide an answer to the question. The outcome is a reduced interrogative force for conjectural questions: not only is the hearer not required to answer, the speaker is encoding that the hearer is probably not able to answer.  1.4.3  Non-Modal Evidentials  ’ As mentioned above, the tests from (1.16) show that nakw has very different characteris’ is a non-modal evidential. tics from =ima and =kat: what these tests show is that nakw ’ Following Portner (2006), in chapter 5 I present the claim that nakw clauses have their own sentential force, or what I call an ‘evidential sentential force’. The function of evidential sentential force is to add the prejacent to the set of discourse propositions that a ’ speaker has inferential evidence for. nakw sentences do not have the illocutionary force 29  1.4. Structure of the Thesis of assertion – they do not add a proposition to the common ground – but are similar to other sentential forces: their illocutionary force is determined by the context of use. This ’ is reflected in the variety of translations of nakw , as something question-like, rhetorical statements, and occasionally warnings. I proceed in two steps: First, we can use a dynamic semantics to examine the ’ context change potential of sentences of the form nakw (p) in Gitksan. Here I claim that ’ by making a statement nakw (p), a speaker pragmatically presupposes that he or she has sensory evidence for p. More specifically, because of the sensory evidence lexically ’ ’ encoded by nakw, the addition of nakw (p) to a Context Set C requires that C contain only worlds where the sensory evidence proposition is true, which happens because the proposition is in the common ground. ’ Secondly, while this explains the use of nakw in conversation in terms of its context ’ change potential, we still require an account of nakw in terms of its sentential force. ’ If nakw (p) sentences are not assertions, then what exactly do they contribute to the ’ ’ conversation? In §5.3 I develop a dynamic semantics of nakw by proposing that nakw is ’ a novel sentential force operator. Under this analysis, the sentential force of nakw is to add a proposition to the inferential set: the set of propositions a speaker has inferential evidence for.  1.4.4  Evidentiality and Modality at the Semantics-Pragmatics Interface  The picture that emerges at this point is that evidentiality in Gitksan is an interface phenomenon par excellence. Not only is the evidential system in Gitksan split between modal and non-modal evidentials, these modals often interact with each other in certain ’ contexts. Specifically, both =ima and nakw are felicitous in contexts where there is ’ sensory evidence for an inference. This means that =ima and nakw compete at some level in these contexts, and they express different attitudes of the speaker towards the ’ sensory evidence: =ima expresses a weaker confidence in the proposition than nakw . 30  1.4. Structure of the Thesis ’ Because =ima and nakw operate on different levels of meaning, in chapter 5 I claim this requires a semantics-pragmatics interface principle which I call pragmatic blocking. ’ The other robust features of nakw, its mirative and non-literal meanings, are also ’ ’ rooted in the use of nakw in conversation: when a speaker uses nakw (p) statement when they know or believe the proposition embedded under it to be true, a mirative interpretation is conveyed; when they believe that proposition to be false, a metaphorical interpretation arises. Chapter 5 also presents a novel analysis of the mirative and ’ metaphorical uses of nakw as conversational implicature. 1.4.4.1  Pragmatic Blocking  ’ Example (1.23) shows that when there is sensory evidence available in the context, nakw takes over the interpretation of must, and =ima is relegated to meaning only might: (1.23) Context: You see scratch marks on the outside wall of your friend’s house that weren’t there last night. a. hlakhlakstiidimahl as’oshl haahlxan hlak-hlaks-tiit=ima=hl haahlxan as-’os=hl pl-scratch-3pl=mod=cnd pl-dog=cnd wall “The dogs might’ve scratched the wall.” ’ b. nakwt ’ nakw =t evid=3pl “The dogs  hlakhlakstiithl as’oshl haahlxan hlak-hlaks-tiit=hl as-’os=hl haahlxan pl-scratch-3pl=cnd pl-dog=cnd wall must’ve scratched the wall.”  Chapter 5 provides an analysis of the modal force alternations found in sensory evidence contexts of the kind exemplified in (1.23): the evidence presuppositions at’ tached to =ima and nakw stand in a blocking relation to one another mediated by the application of Maximize Presupposition, which requires that the strongest possible presupposition be used in any given context (Heim 1991; Sauerland 2003; Schlenker 2006). ’ nakw has a more specified presupposition – one that presupposes the speaker has sensory 31  1.4. Structure of the Thesis evidence for an epistemic claim – and therefore blocks =ima, which lacks this specification, in sensory evidence contexts. The effect this has when translated into English is ’ that nakw will typically be translated as must. However, =ima is also felicitous in these contexts, but it indicates that a speaker does not believe the sensory evidence to be suitable for a strong, must-like claim. In formal terms, =ima ‘implicates’ the negated ’ sensory evidence presupposition of nakw, thus conveying what is interpreted as might (Sauerland 2003). 1.4.4.2  Mirativity and Metaphor  ’ Chapter 5 concludes by analyzing two more properties of nakw : its mirative and nonliteral (metaphorical) uses in discourse. Mirativity refers to the grammatical marking of a proposition as representing information which is surprising to the speaker (DeLancey ’ 1997, 2001). A mirative interpretation is associated with the evidential nakw, as the translations in (1.24) show: (1.24)  ’ a. bagw nidiit ’ pakw nidiit arrive.pl 3pl “They’ve arrived.” ’ b. nakwhl bagwdiit ’nakw =hl pakw=tiit evid=cnd arrive.pl=3pl “They must’ve arrived!” “Looks like they’ve arrived!”  ’ Under its evidential reading, the use of nakw means the speaker has indirect sensory evidence for a proposition, such as a truck parked in the driveway, or noise in the hallway. ’ When a speaker witnesses an event, nakw can be used to express surprise at a situation, such as the unexpected arrival of guests at a party. This basic analysis presented in chapter 5 is as follows: with a statement of the form evid(p), a speaker cannot know for certain p is in fact true. If a speaker knows p is 32  1.4. Structure of the Thesis true, then we expect Gricean considerations to ensure that a speaker assert p, and not evid(p). A mirative statement results when a speaker knows p is in fact true. Under this view, a mirative statement doesn’t assert something new because p is already a part of the common ground, and this is what results in implicature. ’ There is another pragmatic feature associated with nakw : in addition to its evidential ’ and mirative uses, nakw has a metaphorical use. Consider a context where the speaker is watching a baseball game. The star batter on the speaker’s favourite team keeps missing the ball and striking out, jeopardizing the outcome of the game. Out of exasperation, the speaker exclaims (1.25): ’ (1.25) nakwhl sinst ’nakw =hl sins-t evid=cnd blind-3sg “He must be blind!” “Is he blind or something?” “Looks like he’s blind!” The flipside of mirativity with regards to the truth value of p is the use of an evidential in a metaphorical statement, which arises when a speaker knows p is false. For example, upon uttering (1.25), the speaker is saying that ‘he must be blind’, something he knows is is false. Thus, the speaker is flouting the maxim of Quality (“do not say what you believe to be false”). What the speaker is doing is asserting (1.25) in order to implicate that the batter is performing counter to expectations, or that the batter has the attributes of blindness. This forms a three-way formal system for the pragmatic use of an evidential, as given in (1.26): (1.26) (i.) In asserting ev(p), the Speaker does not know it’s part of the common ground that either p or not p: Evidential without any implicated meaning (ii.) In asserting ev(p), the Speaker knows it’s part of the common ground that p: mirativity as Quantity implicature (iii.) In asserting ev(p), the Speaker knows it’s part of the common ground that not p: metaphor as Quality implicature 33  1.5. Methodology This bears directly on the status of mirativity as a natural linguistic class, and the debate within the literature as to whether or not mirativity is a separate semantic category, or simply an extension of evidentiality (cf. DeLancey 1997; 2001). There are two outcomes of this analysis. First, this gives us a way to derive mirativity from other components of the grammar in a predictable way through implicature. This analysis also predicts a relation between mirativity and metaphor based on the speaker’s knowledge of the truth or falsity of p.  1.5  Methodology  There are many unique challenges in the elicitation of meaning, and how those meanings are used in context. This is particularly the case when eliciting evidential and modal meanings in languages with evidentials, where there is no obvious correspondence with how a metalanguage such as English encodes these meanings. Throughout the data elicitation phase of this research, I followed the semantic fieldwork methodology of Matthewson (2004), where I aim to establish facts about the meaning of utterances that contain evidentials. I also develop new ways to explore specialized meanings involving evidentials, such as mirativity and metaphor. These methodological principles are highlighted below. Coverage. Data was collected from 16 speakers, representing each of the six major Gitksan speaking communities in northern BC (excluding Kitwancool), plus two urban speakers in Vancouver, and across the two main dialects of Gitksan, Western and Eastern Gitksan. There were no discernible differences relevant to evidential or modal meaning between these dialects or communities. Presenting contexts. The primary elicitation method was the presentation of contexts that target specific kinds of evidential meanings, and then asking speakers what kind of sentence in Gitksan would adequately describe that scenario. When it was pos-  34  1.5. Methodology sible, culturally relevant objects and situations were used in constructing a scenario, for example, smoking fish or berry picking. Presenting sentences. This technique complements the above one. Here, I would present an evidential sentence to a speaker and then ask them to create a story or scenario where that evidential sentence would be appropriate. Sometimes this was done out-of-the blue in isolation, and other times sentences with different kinds of evidentials were presented in a sequence, and the speaker was asked to make the appropriate adjustment to the scenario in order to isolate the evidential meaning. Conversation analysis. Fortunately, there are still households in the Gitksan communities where one can hear the language used on a daily basis between members of a family. This afforded me the opportunity to observe the language and how it’s used without any interference from the researcher. Not surprisingly, evidentials were abundant in natural conversation. Thus, with the permission of the speakers, I made note of these overheard conversation fragments. Often I would re-elicit them from the same speakers later. Additionally, throughout the course of fieldwork, many personal anecdotes and traditional stories were shared by the speakers. Perhaps not surprisingly, evidentials were less used in these contexts, but there were some relevant data, especially involving the difference between the reportative and quotative, and how the former cannot co-occur with the other evidentials, but the latter can. Speaker introspection. A crucial aspect of this kind of data collection is including a speaker’s intuitions about the meaning of the individual evidentials, not just in terms of when they are (in)felicitous in a context, but as a way to really define what an evidential can mean and how it can be used in a conversation. This last aspect is especially important, as natural conversation cannot always be easily constructed in a field situation.  35  1.5. Methodology Testing infelicity and ungrammaticality. In addition to testing felicity, it is also important to test for ungrammaticality. This is especially relevant when examining embeddability, which involves the syntax, semantics and pragmatics of evidentials. Given that syntactic research on Gitksan is still in its early stages, one goal of this research was to shed some light into the morphosyntax of constructions involving evidentials, particularly second position clitics. This is necessary in order to check for the co-occurrence restrictions of evidentials and whether their syntactic position affected their interpretation. Technical notes. Throughout this dissertation I follow the tradition in Tsimshianic linguistic study by using four lines in each language example. The top example is the language data using the Gitksan practical orthography (cf. Appendix A), and usually represents the typical spellings used by speakers. The second line contains the morphological analysis, including the underlying forms of the morphemes before obstruent voicing applies. The third line is the morpheme gloss, and the fourth line is the translation. Immediately above a language example is a paraphrase of the context that was presented to the speaker. All examples are from fieldwork, unless otherwise cited.  36  Chapter 2 Epistemic Modality and Evidentiality in Gitksan In chapter 1 I suggested that all languages have ways of linguistically encoding the source and/or reliability of knowledge speakers have for making claims and observations about the world around them. This chapter applies that assumption to Gitksan, and the goal is to provide a systematic and detailed semantic and morphosyntactic description of how speakers of Gitksan linguistically encode their epistemic notions. As mentioned in the previous chapter, there are no descriptions of epistemic modality nor evidentiality in the Gitksan literature. Thus, in addition to providing the first description of this system, the data presented in this chapter will allow us to see how Gitksan can enrich our understanding of evidentiality from a typological perspective, as well as a first glimpse at the issues that the Gitksan system brings to bear on current theories of modality and evidentiality.  2.1  The Issues  Given the assumption that all languages have the means to express the source and/or reliability of knowledge, we cannot assume that languages encode these notions in the same way. Chapter 1 (cf. §1.3.1) discussed how languages have the possibility of dividing the labour of expressing epistemic knowledge either through a system of modals (usually encoding reliability of some knowledge), or evidentials (encoding the source of some knowledge), or both. Thus, an investigation of this in Gitksan bears on the first two relations between epistemic modals and evidentials developed in (1.15), and reformulated 37  2.2. The Plan for Gitksan in (2.1): (2.1) (i.) Conceptual relation: To what extent does Gitksan manifest a relation between the concept of evidentiality and the concept of epistemic modality? (ii.) Encoding relation: To what extent is there a relation between the encoding of evidential concepts and the encoding of modal concepts in Gitksan? The data from Gitksan presented in this chapter and analyzed in chapters 3 and 4 show that not only are epistemic modality and evidentiality not mutually exclusive, but that Gitksan possesses both modal and non-modal evidentials.  2.2  The Plan  In order to investigate the status of evidentiality in Gitksan or any language, it is necessary to have in place a set of categorized evidential meanings. In §2.3 I introduce Aikhenvald’s (2006) system of categorizing evidential meanings, which I use in §2.4 to give a preliminary description of the evidential system in Gitksan. I also look across to evidentiality in Nisgha’a, a mutually intelligible language, for some indication of how evidentials might manifest themselves in Gitksan. §2.4 looks at the cognates of the evidentials in Gitksan, and uses Tarpent’s (1987) description of Nisgha’a evidentials as a point of departure in examining the three morphemes in Gitksan that encode modal and evidential meaning. Sections 2.5 and 2.6 investigation the semantic properties of the reportative =kat and the inferential evidential =ima, while §2.7 looks at their morphosyntactic properties. The semantic and morphosyntactic properties of the sensory ’ evidential nakw are examined in §2.8, and §2.9 looks at how all three of these morphemes interact. §2.10 summarizes.  38  2.3. Describing Evidentiality  2.3  Describing Evidentiality  Under a typical interpretation, a claim that a language possesses an ‘evidential system’ means that that language has a paradigm of linguistic markers, where each member of that paradigm uniquely encodes something about the speaker’s type of evidence or source of information for their claim. More specifically, the members of this evidential paradigm would correspond to a systematic arrangement of evidential meanings, such as those found in the systems developed by, for example, Palmer (2006), Aikhenvald (2006), Willett (1988) or Speas (2008). Aikhenvald’s (2006) typology is arranged around the source and type of verification that evidential markers provide in making an evidential statement. The source indicates who the statement is attributed to (first hand or second hand), the type indicates how the speaker obtained the information (visually, by inference, through belief, by hearsay, by quoting). Aikhenvald creates a ranked list of evidential systems based on the number of different markers in the language. These kinds of evidential meanings are divided by Aikhenvald into the categories given in (2.2): (2.2) (i.) Witness vs. Nonwitness (ii.) Firsthand vs. Secondhand vs. Thirdhand (iii.) Sensory a. Visual b. Nonvisual (i.e. auditory, olfactory, etc.) (iv.) Inferential a. Direct physical b. General knowledge c. Experience d. Past deferred realization  39  2.3. Describing Evidentiality (v.) Reportative a. Hearsay b. Quotative (vi.) Assumed These categorizations form contrasts with each other, while implicating or including the meanings of others. For example, Aikhenvald generalizes that a language which has a witness evidential usually contrasts it with a nonwitness evidential which indicates that the information was not witnessed personally, but was obtained through a secondhand source or was inferred by the speaker. However, a witness evidential can also include finer ranges of meaning, perhaps indicating that the information source was obtained through direct observation by the speaker. Usually this is from visual observation (eyewitness), but some languages also mark information directly heard together with information directly seen. Another cross-linguistically important evidential category is the reportative. Even within this category grammatical distinctions can be made between a general reportative evidential (i.e. hearsay), and the quotative. A sample typology of how the different evidential categories can combine and cross cut each other is given in Table 2.1 (adapted from Schenner 2010): Evidential distinctions  Languages  Firsthand, Non-firsthand  Jarawara, Yukaghir  Direct/Visual, Inferred, Reported  Quechua, Qiang, Shasta  Visual, Non-visual sensory, Inferred  Washo, Siona  Visual, Non-visual sensory, Inferred, Reported  Tucano, Eastern Pomo  Direct/Visual, Inferred, Assumed, Reported  Tsafiki, Shipibo-Konibo  Table 2.1: A sample typology in Aikhenvald’s categorization of evidential meanings  40  2.4. Describing Evidentiality in Gitksan Although there are alternatives to Aikhenvald’s evidential categories (i.e. Willett 1988; Palmer 2006), the organization of evidential meanings in (2.2) represents a robust range of the evidential meanings which are found cross-linguistically. I adopt Aikhenvald’s typology of evidentials in (2.2) as a useful heuristic move at this point in terms of organizing the detailed descriptions of the three morphemes in Gitksan dedicated to expressing epistemic and evidential meanings. Additionally, these evidential features can be used to build a more finely-grained definition of the properties of the Gitksan evidential system, and will allow us to identity how Gitksan fits into a typology of evidential systems cross-linguistically.  2.4  Describing Evidentiality in Gitksan  A first, practical step in examining evidentiality in Gitksan is to begin with Tarpent’s (1987) description of evidentials in the mutually intelligible language, Nisgha’a, and then test plausible cognates of them in Gitksan. Tarpent identifies three morphemes which encode epistemic and evidential meanings in Nisgha’a, each of which turn out to have identical cognates in Gitksan. These are what Tarpent characterizes as the ‘reportative’ 13 ’ =kat, the ‘dubitative’ =ima, and the modal/evidential nakw. Both =ima and =kat are  ’ described by Tarpent as verbal enclitics; however, the modal/evidential nakw has the syntactic distribution of a full verb which takes a complement clause. Table 2.2 below summarizes Tarpent’s original labels and evidence types for the Nisgha’a evidential system. The following subsections show that Tarpent’s descriptions of the cognates of =kat, ’ =ima, and nakw in Nisgha’a are generally accurate, but we can further sharpen their meanings considerably. For example, what is immediately noticeable about Tarpent’s descriptions in Table 2.2 is how they only broadly correspond to the categories found 13  ’ Tarpent alternates between glossing nakw as a modal and evidential. Additionally, Tarpent includes  =ima and =kat as part of a system of ‘evidential postclitics’ (1987, p. 489). The other postclitics in the Nisgha’a system could not be verified in Gitksan.  41  2.4. Describing Evidentiality in Gitksan Original gloss  Evidence type  =kat  reportative  report  =ima  dubitative  indirect/direct  ’ nakw  modal/evidential direct  Table 2.2: The evidential system in Nisgha’a (Tarpent 1987)  in Aikhenvald’s typology. Also, there is an overlap in the evidential meanings between ’ =ima and nakw (which was also originally defined as a modal by Tarpent). Thus, a ’ finer-grained characterization is needed in order to distinguish =ima from nakw, and to verify the empirical difference between what she described as an evidential and a modal. In anticipation of doing this, I have replaced Tarpent’s original glosses by adopting Aikhenvald’s typology in (2.2): the ‘dubitative’=ima is reglossed as ‘mod’ (modal), ’ and the modal/evidential gloss for nakw is now ‘evid’ (evidential). However, I’ve maintained Tarpent’s original gloss for the reportative, =kat. Thus, the glosses used in the remainder of this dissertation, and the Aikhenvald evidence type that is eventually ascribed to them, are given in Table 2.3 below: Gloss  Aikhenvald Evidence type  =kat  reportative (rep)  Reportative – Hearsay  =ima  modal (mod)  Inferential (all senses a. – d.)  ’ nakw  evidential (evid)  Inferential – Direct physical  Table 2.3: A description of the evidential system in Gitksan  By way of introduction to the meanings of these evidentials, example (2.3) is a minimal set using a typical stative/intransitive verb. (2.3) also gives a snapshot of the typical translations of these evidentials given by consultants:  42  2.4. Describing Evidentiality in Gitksan (2.3)  a. mukwhl maay’ mukw=hl maay’ ripe=cnd berries “The berries are ripe.” maay’ b. mugwigathl mukw=kat=hl maay’ ripe=rep=cnd berries “[I heard] the berries are ripe.” c. mugwimahl maay’ mukw=ima=hl maay’ ripe=mod=cnd berries “The berries might/must be ripe.” “Maybe the berries are ripe.” ’ d. nakwhl mukwhl maay’ ’ nakw =hl mukw=hl maay’ evid=cnd ripe=cnd berries “The berries must be ripe.” “Looks like the berries are ripe.”  The meaning of =kat supports the standard definition of a reportative: in example (2.3)b., a speaker has come to believe that the berries are ripe through a report from another source. That source could be anonymous or unknown (perhaps over the radio scanner), or from someone who was actually out berry picking (e.g. a family member who was there earlier) (cf. Aikhenvald’s hearsay evidence). ’ From the consultants’ translations and Tarpent’s descriptions, =ima and nakw appear at face value to make epistemic claims of varying force about the ripeness of the berries rather than tell us anything about the type of evidence a speaker has for asserting that the berries are ripe. However, there are five semantic and pragmatic features ’ that distinguish =ima and nakw. The first involves what can be interpreted as modal force: =ima can express either epistemic possibility or necessity (commonly translated ’ by consultants as must or might), while nakw can only express necessity (often translated as must and very rarely as might). The second difference regards evidence: =ima  43  2.4. Describing Evidentiality in Gitksan is compatible with any type of indirect evidence, including a simple speculative judgment about the present or future with little or no observable evidence, or a recollection ’ of a past event where the details are no longer clear. The distribution of nakw is more restricted: its interpretation requires the speaker to have some type of perceptual evidence for the assertion. The context in (2.4) brings out this difference: (2.4)  Context: You’re sitting at home talking about going berry-picking. It’s August, and the berries are usually ripe this time of year on the Suskwa (a traditional picking grounds).  This context involves inferential evidence from speculation, general knowledge or experience. Example (2.3)c. using =ima is felicitous with this type of evidence, while ’ nakw in (2.3)d. is infelicitous – it cannot be used when the speaker is only speculating or inferring from general knowledge. In a related context, given in (2.5), where there is some kind of physical evidence available to the speaker – in this case, purple hands – ’ both nakw and =ima are felicitous: (2.5)  Context: People are arriving home after a day of berrypicking up in the Suskwa. They’re carrying buckets of berries, and their hands are all purple.  ’ This physical evidence interpretation is supported by the frequent translation of nakw as “It looks like . . . ”, and corresponds to Aikhenvald’s inferential evidence from some type ’ of direct, perceptible, physical results. Thus, nakw is felicitous with a specific subset of the inferential evidence types that =ima is. Or in other words, =ima is felicitous in ’ both inferring from physical and reasoning contexts, while nakw is felicitous only when inference is from physical evidence. ’ The third difference between =ima and nakw involves a pragmatic distinction: =ima can appear in any type of clause, including assertions, questions, negation, or condition’ als. However, nakw cannot be used in questions, nor in the antecedent of a conditional, ’ nor can nakw be negated or otherwise embedded in any way. This is discussed in more 44  2.4. Describing Evidentiality in Gitksan detail in §2.8, but an example illustrating this difference using negation is given in (2.6):14 (2.6)  a. neeyimahl mukwhl maay’ nee=ima=hl mukw=hl maay’ neg=mod=cnd ripe=cnd berries “The berries might not be ripe.” ’ b. *nee=nakw =hl mukw=hl maay’  ’ The fourth distinguishing feature between =ima and nakw also involves pragmatics: ’ ’ nakw can be used to express mirativity. Example (2.7)a. shows how nakw can be used to mark a proposition that represents information which is surprising to the speaker ’ (DeLancey 1997; 2001). This effect is restricted to nakw : =ima does not have this mirative effect in b.: (2.7) Context: You’re sitting around with friends playing cards, and a couple more friends unexpectedly walk through the door wanting to join the game. ’ a. nakwhl bagw=diit ’ nakw =hl pakw=tiit evid=cnd arrive.pl=3pl “They’re here!” “Looks like they made it!” b. # bagwima bakw=ima arrive.pl “They might be  ’ nidiit ’ nidiit 3pl here.”  ’ Finally, nakw has a non-literal/metaphorical use, rendering an expression similar to a rhetorical question in English in example (2.8):  14  ’ This is also characterized as a (morpho)syntactic difference: nakw is a full-fledged verb, while =ima  is a predicate enclitic. This is also discussed in more detail in §2.8.  45  2.4. Describing Evidentiality in Gitksan (2.8) Context: You’re watching a baseball game. The star batter on the speaker’s favourite team keeps missing the ball and striking out, jeopardizing the outcome of the game. ’ nakwhl sinst ’ nakw =hl sins-t evid=cnd blind-3 “He must be blind!” “Is he blind or something?” “Looks like he’s blind!” ’ In both its mirative and non-literal uses, nakw maintains its evidential function: the speaker is making an assertion based on what they infer from the physical results they perceive – in (2.8) the fact that the batter keeps missing the ball, and in (2.7) the fact that they can see people coming through their front door. The mirative and ’ metaphorical uses of nakw are examined in detail in Chapter 5. In contrast, =ima in example (2.9) below is also felicitous in the baseball context in (2.8), but it cannot have this non-literal effect: =ima must express that the batter may be literally blind, as suggested in the accompanying context: (2.9) Context: You’re walking down the street and notice a man walking along with a white cane and dark sunglasses (Felicitous context offered by consultant). ’ sinsima nit ’ sins=ima nit blind=mod 3sg “He might/must be blind.” In the following sections, I examine in detail the meanings and morphosyntax of the three Gitksan evidentials outlined above.  46  2.5. The Reportative =kat  2.5  The Reportative =kat  Reportative evidentials indicate that the information was reported to the speaker by another sentient source. In Aikhenvald’s categorization of evidential meanings in (2.2), languages can distinguish between hearsay evidentials and quotative evidentials. Hearsay indicates reported information that may or may not reflect literally the statement that was made by the source, while a quotative sets off a direct quotation of another person’s speech. Hearsay evidentials can also be broken down further: according to Willett (1988, p. 57, 96), there are three kinds of reported evidence, defined in (2.10): (2.10) (i.) Second-hand evidence: The speaker claims to have heard of the situation described from someone who was a direct witness. (ii.) Third-hand evidence: The speaker claims to have heard of the situation described by someone who was not a direct witness. (iii.) Evidence from folklore: The speaker claims to have heard of the situation described as part of established oral history. In Nisgha’a, Tarpent (1987, p.499) states that “by using =kat the speaker disclaims responsibility for the truth of the utterance because he is only reporting information originating with others.” Example (2.11) is a typical example of =kat in Nisgha’a, with its cognate in Gitksan given in (2.12): (2.11) Nisgha’a (Tarpent 1987: 499) siipkwgat t siipkw=kat t sick=rep pnd “[I heard] Mary is  Mary Mary Mary sick.”  47  2.5. The Reportative =kat (2.12) Gitksan siipxwkatit Mary siipxw=kat=t Mary sick=rep=pnd Mary “[I heard] Mary is sick.”  2.5.1  The Evidential Meanings of =kat  Reportative =kat is felicitous in all three of Willett’s reportative types in (2.10). Example (2.13) presents contexts that involve second-hand information: in all of these contexts the speaker is reporting a proposition made by someone who was present and witnessed the relevant eventuality:15 (2.13) Second-hand Evidence: a. Context: The speaker is talking about a time during her childhood when she took a boat from Vancouver Island to Prince Rupert. The speaker does not remember herself exactly where the boat arrived, but was told about it by her older sister, who was there. ’ m’ go’ohl Prince Rupert bagwgat nuu ’ m ’ ko’=hl Prince Rupert nuu pakw=kat arrive.pl=rep 1pl loc=cnd Prince Rupert “[I heard] We got to Prince Rupert.”  15  The edges of =kat are subject to the global phonological rules in Gitksan of obstruent voicing and  deletion. This results in various allomorphs of =kat such as =gat, =ka, and =ga.  48  2.5. The Reportative =kat b. Context: The speaker is telling her friends at the coffee shop that Mary had her long hair cut recently. She hasn’t seen Mary’s hair herself yet, but knows because the speaker’s sister is the hairdresser who did it. gungojigas Mary-hl gest Mary=hl kwin-kots-i-(t)=kat=s kes-t caus-cut-tr-3sg=rep=pnd Mary=cnd hair-3sg “[I heard] Mary had her hair cut.”  c. Context: The speaker is reassuring a mutual friend that Mark’s sister will repay a debt because Mark gave her some money, and the speaker talked to her about it already. ’ ginamigas Mark=hl daala ’ Mark=hl taala kinam-i-(t)=kat=s give-tr-3=rep=pnd Mark=cnd money “[I heard] Mark gave money to his (Mark’s)  ’ahl gimxtit ’a=hl kimxt-t obl=cnd sister-3sg sister.”  The third-hand evidence in (2.14) is from a source who then reports it to another source, who in turn reports it to the speaker:16 (2.14) Third-hand Evidence: a. Context: John didn’t go to work today. You ask your co-workers where he is. None of them have seen John sick, but their boss told them earlier in the morning. siipxwgatit John siipxw=kat=t John sick=rep=pnd John “[I heard] John is sick.” 16  There is a class of verbs called ’T’-class verbs in Nisgha’a and Gitksan. The meaning or function  of the morpheme -t- has not been determined (although see Tarpent (1987) for details on its morphosyntactic distribution); thus, I follow the convention in the Gitksan/Nisgha’a literature and leave it unglossed.  49  2.5. The Reportative =kat b. Context: All of the children in the neighbourhood are excited about a new dog named Sammy. A parent is talking to a neighbour about the new dog, after overhearing his name is Sammy. siwatdigathl gyathl ’os ’ahl Sammy gyat=hl ’os ’a=hl Sammy si-wat-t-i-(t)=kat=hl caus-name-T-tr-3=rep=cnd man=cnd dog obl=cnd Sammy “[I heard] The man named his dog Sammy.” (adapted from Rigsby 1986: 291)  c. Context: People are discussing the various contributions for a feast. Someone heard from the person who did the accounting that Walter also put in money, but the accountant didn’t actually witness Walter doing this (as it’s done anonymously). Walterhl daala lumagdigas Walter=hl taala lumak-t-i-(t)=kat=s donate-t-tr-3=rep=pnd Walter=cnd money “Walter donated/contributed/put in money.”  Example (2.15) shows the use of =kat in folklore, involving a typical description of the legend character, Weget: (2.15) Folklore laxmo’ongat wil lax-mo’on=kat wil loc-salt=rep comp “Weget was born in the  sgyats Weget skyat=s Weget born=pnd Weget sea.”  =kat is also felicitous in contexts where the source is entirely anonymous, or can’t be determined, as in reporting something overheard from another conversation, as in (2.16), or it can also be used to report predictions about future states or events, as in (2.17): 50  2.5. The Reportative =kat (2.16) Context: You’re telling your friend that you overheard at the coffee shop earlier that John won at bingo last night. xstagas John go’ohl bingo gaxxw John ko’=hl bingo kaxxw xsta=kat=s win=rep=pnd John loc=cnd bingo last.night “John won at bingo last night.” (2.17) Context: A friend of yours heard you were getting married and congratulates you when you meet her in the store. ’ niin dim naksgat ’ niin tim naks=kat fut marry=rep 2sg “[I heard] you’re getting married!” In more complex sentences, such as those that contain complement clauses, the interpretation of =kat is sensitive to the predicate it attaches to. In example (2.18) which does not contain =kat, the speaker was present when Mark made the statement, and the speaker is directly reporting what Mark said, that John would leave for the coast: (2.18) mahldis Mark ’ahl gimxdit dim mahl-T-i-(t)=s Mark ’a=hl kimxt-t tim tell-t-tr-3=pnd Mark obl=cnd sister-3 fut John go’ohl laxmo’on John ko’=hl lax-mo’n John loc=cnd geo.loc-coast “Mark told/said to his sister that John is leaving for 324)  ’ wil saa daawhls ’ wil saa taawhl=s comp away leave=pnd  the coast.” (Rigsby 1986:  In (2.19), =kat attaches to the verb within the matrix clause, and the speaker is now reporting that she heard about Mark telling his sister that John would leave for the coast. In this case, the reportative evidence is oriented towards the speaker: she heard from Mark’s co-worker that Mark told his sister that John would leave for the coast. 51  2.5. The Reportative =kat (2.19) Report: The speaker is asserting based on a report that Mark told his sister John would leave for the coast. Context: The speaker heard from Mark’s co-worker that John was going to be away for the weekend, and the co-worker overheard Mark talking to his sister on the phone about John going to the coast. mahldigas Mark ’ahl gimxdit dim wil saa Mark ’a=hl gimxt-t tim wil saa mahl-t-i-(t)=kat=s tell-t-tr-3sg=rep=pnd Mark obl=cnd sister-3sg fut comp away daa’whls John go’ohl laxmo’on lax-mu’n taa’whl=s John ko’=hl leave=pnd John loc=cnd geo.loc-coast “Mark told/said to his sister that John is leaving for the coast.” (adapted from Rigsby 1986: 324) However, in (2.20) =kat is attached to the verb in the complement clause. The reportative evidence is now re-oriented to the matrix subject, Mark, and crucially not to the speaker: it is Mark who has reported evidence that John will leave for the coast. Similar to (2.18), the speaker is simply reporting what Mark said, which includes Mark’s reportative evidence: (2.20) Report: John is leaving for the coast (as heard by Mark). Context: You had lunch with Mark. While at lunch his sister came up and Mark told her that he heard John would leave for the coast. mahldis Mark ’ahl gimxdit dim wil saa mahl-T-i-(t)=s Mark ’a=hl kimxt-t tim wil saa tell-t-tr-3sg-pnd Mark obl=cnd sister-3sg fut comp away ’ daawhltgatit John go’ohl laxmo’on ’ John ko’=hl lax-mo’n taawhl=t=kat=t leave=3sg=rep=pnd John loc=cnd geo.loc-coast “Mark told/said to his sister that John is leaving for the coast.” A similar example is given in (2.21): in a., which contains =kat on the matrix verb, the speaker is reporting from another source that Gwen told her husband that Bill 52  2.5. The Reportative =kat showed his house to John. The reportative in example b., where =kat is attached to the embedded verb, is oriented towards the matrix subject Gwen, who is reporting from another source that Bill showed his house to John: (2.21)  a. Report: The speaker has reportative evidence for asserting that Gwen told her husband that Bill showed his (Bill’s) house to John (in order to sell it). Context: You overheard a conversation where people were talking about Bill’s house being for sale. Everyone knows that both Gwen’s husband and John have always been interested in the house. Gwen was also in the room, and someone thought to remind her but you say that you heard she already told her husband. mahldigas Gwen ’ahl nakst wil Gwen ’a=hl naks-t wil mahl-T-i-(t)=kat=s Gwen obl=cnd husband-3sg comp tell-t-tr-3sg=rep=pnd gungya’adis Billhl wilpt ’as John kwin=kya’-T-i-(t)=s Bill=hl wilp-t ’a=s John caus-see-t-tr-3sg=pnd Bill=cnd house-3sg obl=pnd John “Gwen told/said to her husband that Bill showed his house to John.”  b. Report: The speaker is asserting that Gwen has reported evidence that Bill showed his house to John. Context: John has always been interested in buying Bill’s house. Gwen told her husband that she heard from Bill’s wife that Bill showed his house to John. mahldis Gwen ’ahl nakst mahl-t-i-(t)=s Gwen ’a=hl naks-t tell-T-tr-3sg=pnd Gwen obl=cnd husband-3sg gungya’adigas Billhl wilpt Bill=hl wilp-t kwin=kya’a-T-i-(t)=kat=s caus-see-t-tr-3sg=rep=pnd Bill=cnd house-3sg “Gwen told/said to her husband that Bill showed his  wil wil comp ’as John ’a=s John obl=pnd John house to John.” 53  2.5. The Reportative =kat The morphosyntactic distribution of =kat is discussed in more detail in §2.7; there it is shown that the distribution of =kat is somewhat flexible, although it is typically found encliticized to the first syntactic constituent of a sentence. However, what the examples in (2.21) show is that the interpretation of =kat is sensitive to complex clauses involving verbs of saying. This is summarized in the following schema in (2.22): (2.22) (i.) rep(say(Gwen,p)) = (2.21)a. (ii.) say(Gwen(rep(p))) = (2.21)b. Gitksan also has what can be characterized as a ‘quotative’ tiya, which is frequently translated as the verb say. Quotative tiya can be distinguished from reportative =kat in two ways. First, tiya functions much as ‘say’ does in English: it directly reports the speech of another person (or a close paraphrase), as in an excerpt from a story in (2.23): (2.23)  a. dixdahlxw tixtahlxw distal.loc “ ‘Up there he  se’et, diya. se’-t, tiya. find-3 say.3 found it,’ he/she/it said.”  ’ b. nidiit yatshl ’ nitiit yats=hl 3pl sing=cnd “ ‘It was for them  gyat, diyahl hanakgi. hanak-ki. kyat, tiya=hl man say=cnd woman-dist that the man sang,’ the woman said.”  On the other hand, =kat does not typically fulfil this kind of quotative function: when a speaker makes a statement of the form =kat(p), they are expressing that the content of proposition is obtained from another source, not an exact report of what that source said. The second way the quotative can be distinguished from =kat is in the fact that they can co-occur. In example (2.24), a speaker is asking about what it was that John reportedly said:  54  2.5. The Reportative =kat (2.24) Context: A and B are talking about a discussion B overheard at bingo last night. Someone was asking to borrow money to play. John advised this person against lending the money. B talked to C who was there, and overheard John’s part in the exchange. A. gwigat kwi=kat what=rep “What was “What was  diyat John tiya=t John say=pnd John it that John was supposed to have said?” it that John was said to have said?”  ’ John B. ham ji ginamhl daala loodiit, diyagat ’ ha-m tsi kinam=hl taala loo-tiit, tiya=kat John neg-2sg irr give=cnd money obl.pro-3pl say=rep John “(It was said that) John said not to give them (any) money.” Under Aikhenvald’s evidential typology, the reportative is divided into two subtypes, the hearsay and the quotative. Given this, tiya may be considered as an ideal candidate for the role of a quotative evidential, while =kat fulfills the function of a hearsay evidential. I will maintain the label of reportative for =kat, and set aside further investigation of quotative tiya.  2.5.2  A Note on the Social Uses of =kat  A speaker can utilize the meaning of =kat when they wish to either make a polite request, show respect, or to distance themselves. This was noted in the Boas (1912) texts of Nisgha’a, where =kat is frequently attributed to the speech of people of lower social status, as in (2.25).  55  2.5. The Reportative =kat (2.25) Nisgha’a (Tarpent 1987, p. 500) Context: Someone is preparing food for the chief, when one of his sons notices the dish. He wants to prevent a potentially embarrassing situation, but also not offend the person who prepared the food. niigatgidiit gwiix gip-(t)=hl sim’oogit t kunhl kwiix kip-(t)=hl sim’ookit t kun=hl nii=kat-ki-tiit pnd this=cnd neg=rep-ints-3pl always eat-3=cnd chief ’anwinin anwin-n what.one.has-2sg “(I am told) the chief is not at all keen on eating that stuff.” This politeness effect is reproduced in the Gitksan example in (2.26): (2.26) Context: Someone wants a group of guests who have been camped out at the edge of the village to move. ’ m’ ji ji lugwgat nisi t’aahlakw ’ m ’ tsi t’aahlakw nisi tsi lukw=kat irr tomorrow irr relocate=rep 2pl “(I’m supposed to tell you to) you should move tomorrow.” (Gitksan example adapted from Boas 1912 and Tarpent 1987, p. 500) The meaning of a reportative evidential is ideally suited for achieving a politeness effect: the speaker can attribute any commands or potentially awkward statements to an unidentified secondary source. That source may not actually exist, as in (2.26), but given the meaning of the reportative, the addressee must assume that the source does exist. In a related use, =kat can be used when the speaker wishes to use the reported nature of the proposition to his or her advantage, typically by reporting a proposition they know is false. In (2.27), the speaker is trying to trick his brother into doing his work by ‘misreporting’ something their parents said:  56  2.6. The Inferential =ima (2.27) Context: Your brother knows he has to work tomorrow cleaning fish, but he tries to trick you into doing it, implying that this direction came from your parents. dim sixsaksiningathl hon ji t’aahlakw tim six-saks-in=n=kat=hl hon tsi t’aahlakw fut redup-sharp-caus=2sg=rep=cnd fish irr tomorrow “(I heard) you’re supposed to clean the fish tomorrow.”  2.6  The Inferential =ima  Tarpent (1987, p. 498) describes what she calls the ‘dubitative’ enclitic =ima in Nisgha’a as an evidential which expresses that “the speaker thinks that what he says could be true on the basis of what he knows or can infer, but does not want to commit himself as he could be proved wrong.” At first blush, the form and function of the =ima described by Tarpent in Nisgha’a corresponds directly to its cognate =ima in Gitksan, judging by the translations given to it by Gitksan speakers. Compare (2.28) with (2.29): (2.28) Nisgha’a (Tarpent 1987: 497) yugwima haywis gingulx yukw=ima haywis kinkulx prog=mod rain Kincolith “It might be raining in Kincolith.” “It’s probably raining in Kincolith.” (2.29) Gitksan yugwima wis gingulx yukw=ima wis kinkulx prog=mod rain Kincolith “It might be raining in Kincolith.” “It must be raining in Kincolith.” This section takes Tarpent’s description of =ima as an evidential as a point of departure, and probes further its evidential meanings. 57  2.6. The Inferential =ima  2.6.1  The Evidential Meanings of =ima  The uncertain or non-committal nature of =ima that Tarpent describes can be observed when someone is speculating about future events, as in (2.30), or in the spontaneous exchange between family members in (2.31): (2.30) Context: You’re thinking about going to bingo tonight. You feel lucky. ’ y’ xstayima nii ’ y’ xsta=ima nii win=mod 1sg “I might win.” “Maybe I’ll win.” witxws Alvin? (2.31) GS: gaxguhl Alvin? kaxkwi=hl witxw=s when=cnd arrive=pnd Alvin “When is Alvin arriving?” ’ LW: witxwima nit silkwsax ’ witxw=ima nit silkwsax arrive=mod 3sg noon.time “He might arrive around noon.” However, what exactly are the reasons or evidence a speaker adduces for making =ima-statements? As a heuristic step, we can approach the evidential meanings of =ima using Aikhenvald’s typology of inferential evidentials, where inferential evidentiality is divided into the subtypes in (2.32):17 (2.32) (i.) Information inferred by observable physical evidence (ii.) Information inferred by general knowledge (iii.) Information inferred/assumed because of speaker’s experience with similar situations 17  The fourth Aikhenvald type, evidence from past deferred realization, has not yet been tested for.  58  2.6. The Inferential =ima Given the speculative nature of (2.31) and (2.30), these examples show how a speaker is likely relying on general knowledge (that winning money is possible if you play bingo), or LW’s experience with similar situations (the fact that Alvin is usually home by noon when he goes to Smithers). We can further tease these meanings apart by constructing contexts for statements such as in (2.33), where the speaker is responding with uncertainty to a question as to whether John is at home. In this context, a speaker can infer from her experience with John’s schedule, that he might or must be at home. Example (2.33) is also felicitous in contexts where a speaker makes a statement based on inferences from physical evidence, in this case, visual: (2.33) Inference from a speaker’s experience with similar situations: You need to ask John for a favour. You’re sitting at John’s friend’s place and you ask her if she knows if John is back from work yet. She says that he is always back from work by 5pm, so John’ll be home by now. Inference from observable evidence: You need to ask John for a favour. You drive by his place with a friend and notice the lights are on and his truck is in the driveway. t’ayimat John t’a=ima=t John at.home=mod=pnd John “John may/must be at home.” “John’s probably at home.” “Maybe John’s at home.” The statement in (2.34) is felicitous in contexts which allow a speaker to infer from observable evidence, or evidence from general knowledge. Note also how both epistemic possibility and necessity can be expressed using =ima, translated as both might and must, as well as degrees in between:  59  2.6. The Inferential =ima (2.34) Inference from observable evidence: People are arriving home after a day of berry picking up in the Suskwa. They’re carrying buckets of berries, and their hands are all purple. Inference from general knowledge: You’re sitting at home talking about going berry-picking. It’s August, and the berries are usually ripe this time of year on the Suskwa. maay’ mugwimahl mukw=ima=hl maay’ ripe=mod=cnd berries “The berries might be ripe.” “The berries must be ripe.” “The berries are probably ripe.” “It seems the berries are ripe.” “ Maybe the berries are ripe.” A crucial feature of these contexts is that speaker has not witnessed the ripe berries firsthand, rather, there is some kind of evidence that the speaker can use to infer a claim about ripe berries in the forest. This restriction is illustrated in the contexts in (2.35) – (2.37): (2.35) Context: You’re looking out the window during a storm. # yukwimahl dim wis, ii gya’ay’ yukw=ima=hl tim wis, ii kya’-y’ prog=mod=cnd fut rain conj see-1sg “It might/must be raining, and I see it (outside).” (2.36) Context: Your friend is slicing up bait for fishing and you see them cut their hand. #kotsinimahl ’on-n kots-i-n=ima=hl ’on-n cut-tr-2sg=mod=cnd hand-2sg “You might’ve cut your hand.” 60  2.6. The Inferential =ima (2.37) #ye’eyimathl wan ye’e=ima=hl wan walk=mod=cnd deer spagaytgan spakaytkan forest “A deer might be around  asun, ii gya’ay’ loot ahl a-sun, ii kya’a-y’ loo-t a=hl loc-here conj see-1sg obl.pro-3 loc=cnd  here, and I see it in the forest.”  Comment: “There’s no point saying it might be around here if you see it yourself.”  Other examples in (2.38) and (2.39) contrast inference from observable evidence with inference from a speaker’s experience with similar situations: (2.38) Inference from observable evidence: You had five pieces of hoxs (half-smoked salmon) left when you checked yesterday. Today, you go to get some to make hagwiljam (a kind of soup) and you notice it’s gone. It’s not that you only think it’s Fern, you know it’s her because you see the hoxs skins in her room.18 Inference from a speaker’s experience with similar situations: You had five pieces of hoxs left when you checked yesterday. Today, you go to get some hoxs to make hagwiljam and you notice they’re gone. You’re not sure who took them, but you know Fern is the person in your household who really likes hoxs, and usually eats a lot whenever she gets the chance. gubimas Fern=hl hoxs kup=ima=s Fern=hl hoxs eat=mod=pnd Fern=cnd hoxs “Maybe Fern ate the hoxs.” “Fern must’ve eaten the hoxs.”  18  Contexts adapted from Matthewson et al. (2004).  61  2.6. The Inferential =ima (2.39) Inference from observable evidence: Visual: You sneak into the bedroom and see that she’s lying down with her eyes closed. Auditory: You can hear snoring. Inference from a speaker’s experience with similar situations: It’s 5 o’clock. Grandma is in her room and always has a nap at this time of day. wogimat naa’a naa’a wok=ima=t sleep=mod=pnd grandmother “Grandmother might/must be sleeping.” “Maybe Grandmother is sleeping.” The above data show that =ima does not distinguish between the different inference subtypes in (2.32). Thus, =ima could be characterized as a ‘general’ inferential evidential. This designation has a place within Aikhenvald’s classification of evidential systems crosslinguistically: It is fairly common for a language to have what Aikhenvald calls a ‘3-term system’, or an evidential paradigm made up of three evidentials: visual ’ sensory, inferential, and reportative (where the ‘visual sensory’ evidential is nakw, which is discussed in the next section). Languages that have a 3-term system such as this include Aymara, Shastan languages, Qiang languages, Maidu, Quechuan languages, and Northern Embera languages.  2.6.2  The Modal Interpretations of =ima  Cross-linguistically, Palmer (2006, p. 25) suggests that languages which morphologically distinguish a contrast between inference from observation and inference from experience usually indicate an evidential system. So far, this appears not to be the case with =ima: the previous subsection shows that =ima does not distinguish between evidence from observable result, general knowledge, or inference from past experience. Palmer also claims that typology supports the notion that when there is a morphological contrast between simple speculation (without evidence) and inference from evidence, this typically indicates a possibility-necessity contrast. This also appears to not be the case 62  2.6. The Inferential =ima with =ima. Nonetheless, the most common translations of =ima offered by consultants involve the use of a variety of epistemic modals of varying degrees. For example, the translations given by a variety of speakers for the sentence in example (2.40) show just how variable the modal force interpretations of =ima are, ranging from might on one end to must, and a variety of expressions of modal force in between: (2.40) Context: You’re wondering what your friend is doing. You notice his rod and tackle box are not in their usual place. yugwimahl dim iixwt yukw=ima=hl tim iixw-t prog=mod=cnd fut fish-3 “He might be going fishing.” “He must be going fishing.” “He’s probably going fishing.” “He’s likely going fishing.” “He could be going fishing.” “Maybe/perhaps he’s going fishing.” The type of information available to a speaker and what they can infer from it naturally influences the strength of belief in the truth of the proposition. However, the concepts underpinning evidentiality and epistemic modality are not interchangeable because other contextual factors intervene. If a person considers a proposition a possibility, this by itself does not give any indication as to the type of evidence the person has. Similarly, the fact that a person has reportative evidence for the truth of a proposition does not determine a certain strength of belief in this proposition. In an example such as (2.40), the context is simple enough that both must and might are felicitous translations in English: depending on a speaker’s previous experiences with John and his rod and tackle box, John might be fishing, or he must be fishing. The modal =ima is felicitous where a speaker considers only this fact and no others. Or, this fact in combination with other facts, could lead a speaker of English to use a ‘stronger’ modal (such as the fact his truck isn’t in the driveway, it’s August so that means the sockeye are running,  63  2.7. The Morphosyntactic Distribution of =kat and =ima his gumboots are gone etc.) The issue of variable modal force is analyzed in detail in chapter 4.  2.7  The Morphosyntactic Distribution of =kat and =ima  Both reportative =kat and modal =ima belong to the class of second position clitics in Gitksan, which include various discourse particles, pronominal markers (the ‘Series I’ pronouns) and determiners. Chapter 1 gave a brief introduction on the morphosyntax of Gitksan, which I will elaborate on here. (2.41) is a schema of a clause in the Independent order: the V is the first constituent in the clause, to which both =kat and =ima attach. Additionally, if there is more than one enclitic, =kat and =ima are closest to the verb: (2.41) Independent Clause (i.) V=kat(=cl) YP ZP ... (ii.) V=ima(=cl) YP ZP ... In Dependent clauses, schematized in (2.42), a phrase precedes the V, and, for most speakers, both =kat and =ima shift to encliticize to this phrase to maintain the second position: (2.42) Dependent Clause (i.) XP=kat(=cl) V YP ZP ... (ii.) XP=ima(=cl) V YP ZP ... However, there is some speaker variation in Dependent clauses, where =kat and =ima may remain on the V, as (2.43) shows:  64  2.7. The Morphosyntactic Distribution of =kat and =ima (2.43) (i.) XP(=cl) V=kat YP ZP ... (ii.) XP(=cl) V=ima YP ZP ... In matrix clauses, the surface position of =kat is variable in three ways. First, the most common is the second position, as can be seen in the majority of the examples above. In example (2.44)a., =kat surfaces in the second position, attached to the first element in the clause, the progressive yukw. However example (2.44)b. shows that, with some speakers, =kat remains on the verb despite the available second position: (2.44)  a. yukwgat yukw=kat prog=rep “I hear you’re b. yukw yukw prog “I hear  dim wotdinhl tim wot-T-i-n=hl fut sell-t-tr-1sg=cnd gonna be selling your car  kartxwn loot kartxw-n loo-t car-1sg obl.pro-3sg to him.”  kartxwn loot dim wotdingathl kartxw-n loo-t tim wot-T-i-n=kat=hl obl.pro-3sg fut sell-t-tr-1sg=rep=cnd car-1sg you’re gonna be selling your car to him.”  Secondly, there is a less common pattern, shown in (2.45), where =kat attaches clause-finally: (2.45)  ’ a. hlabisxwgat niin ’ hlabisxw=kat niin 2sg tired=rep “I hear you’re tired.” ’ b. hlabisxw niingat ’ hlapisxw niin=kat tired 2sg=rep “I hear you’re tired.”  The second position distribution of =kat is sensitive to the syntax: in (2.46)a., =kat attaches to the edge of the clefted constituent and not the first word in the sentence, as in the ungrammatical (2.46)b.: 65  2.7. The Morphosyntactic Distribution of =kat and =ima (2.46)  a. go’ohl [ko’=hl loc=cnd Clara Clara Clara “(They say)  wilps wildi na’agat sgyats wilp=s wil-t skyat=s na’a]=kat house=pnd grandmother=rep comp-3sg be.born=pnd  it was at her grandmother’s house that Clara was born.”  b. *[go’ohl=kat wilps na’a] wildi sgyats Clara A similar example is given in (2.47), where =kat attaches to the clefted DP, the first phrase in the sentence, and not the first position adjective that modifies tk’ihlxw : (2.47)  sdils a. k’uba tk’ihlxw=gat ’ant Bill stil-(t)=s [k’uba tk’ihlxw]=kat ’ant Bill s.rel accompany-3sg=pnd Bill small child=rep “It was the child who went with Bill.” b. *[k’uba=kat tk’ihlxw] ’ant sdils Bill  Thirdly, there is also some variation when the first word also has a clitic attached. Negation in Gitksan is typically formed by the complex nee, which is a negation marker, plus the enclitic =tii which is a contrastive marker (Rigsby 1986; Tarpent 1987). As mentioned above =kat competes with other clitics for the second position, the result of which is the kind of variation in (2.48): (2.48)  sdilis Leiwat Fern a. neegatdiit stil-i-(t)=s Leiwa=t Fern nee=kat=tii=t neg=rep=contr=3sg go.with-tr-3sg=pnd Leiwa=pnd Fern “Leiwa didn’t go with Fern.” a. neediigatit sdilis Leiwat Fern stil-i-(t)=s Leiwa=t Fern nee=tii=kat=t neg=contr=rep=3sg go.with-tr-3sg=pnd Leiwa=pnd Fern “Leiwa didn’t go with Fern.”  66  2.7. The Morphosyntactic Distribution of =kat and =ima c. neediitgat sdilis Leiwat Fern stil-i-(t)=s Leiwa=t Fern nee=tii=t=kat neg=contr=3sg=rep go.with-tr-3sg=pnd Leiwa=pnd Fern “Leiwa didn’t go with Fern.” In discourse contexts it is common for =kat to attach to every predicate, as in this story fragment about a hunter in (2.49): ’ (2.49) wogigat nit laxjehl lakw, neegatdiihl yookxwt, ’ nit lax-tsehl lakw, yookxw=t, wok=kat nee=kat=tii=hl sleep=rep 3sg loc-edge fireplace neg=rep=contr=cnd eat=3sg ii saa ye’etgat ii saa ye’e-t=kat conj away go=3sg=rep “He slept around the fireplace, he didn’t eat, and after that he left.” Also common in discourse, =kat can attach to whatever phrase the speaker is reporting, as in the response to the following question in (2.50), or as part of an ongoing discourse, as in (2.51): (2.50)  a. na dimt na dim-t who fut-3 “Who’ll come  ’an suudy’ ’an suut-y’ rel go.get-1sg and get me?”  b. Gwengat (dimt Gwen=kat dim-t Gwen=rep fut-3 “(I heard) it’s Gwen  ’an suudin) ’an suut-n rel go.get-2sg that’ll come and get you”  (2.51) Context: People are sitting around talking about the new restaurant that just opened up in Hazelton. amagat ama=kat good=rep “(I heard) it’s good.” 67  2.7. The Morphosyntactic Distribution of =kat and =ima The key generalization here is that the variable position of =kat does not change its interpretation. This can be observed in another possible variation of (2.48) given in (2.52): in a. =kat is in the second position, whereas in b. it remains on the verb. This variation does not correspond to a change in the scopal relations between negation and =kat. In other words, neither a. nor b. in (2.52) can be interpreted as ‘I didn’t hear Leiwa went with Fern.’ (2.52)  a. neegatdii sdilis Leiwat Fern stil-i-(t)=s Leiwa=t Fern nee=kat=tii neg=rep=contr go.with-tr-3=pnd Leiwa=pnd Fern “(I heard) Leiwa didn’t go with Fern.” = “I didn’t hear Leiwa went with Fern.” Leiwat Fern a. needii sdiligas Leiwa=t Fern nee=tii stil-i-(t)=kat=s neg=contr go.with-tr-3=rep=pnd Leiwa=pnd Fern “(I heard) Leiwa didn’t go with Fern.” = “I didn’t hear Leiwa went with Fern.”  The issue of scope and the position of both =kat and =ima is discussed in more detail in Chapter 3. The morphosyntactic distribution of =ima is identical to that of =kat. First, it is common in discourse for =ima to encliticize to whatever phrase is ‘under speculation’, as can be seen in the following conversation fragments where =ima attaches to the kinds of elliptical responses commonly found in discourse. In (2.53) LW1, =ima attaches to an adjunct, in (2.54) a pronoun, and in (2.55) a nominal: (2.53) GS: gaxguhl dim ye’en go’ohl Smithers? kaxkwi=hl tim ye’-n ko’=hl Smithers? when=cnd fut go-2sg loc=cnd Smithers “When are you going to Smithers?”  68  2.7. The Morphosyntactic Distribution of =kat and =ima LW1: t’aahlagwima t’aahlakw=ima tomorrow=mod “Maybe tomorrow.” ’ y’ go’ohl Smithers t’ahlakw LW2: dim ye’eyima nii ’ ’ Smithers t’ahlakw tim ye’=ima niiy ko’=hl fut go=mod 1sg loc=cnd Smithers tomorrow “I might go to Smithers tomorrow.” “Maybe I’ll go to Smithers tomorrow.” ’ (2.54) GS: nida? ’ nit=a? 3sg=interrog “Is it him/her?” (upon hearing someone you’ve been expecting pull up into the driveway) ’ LW: nidima ’ nit=ima 3sg=mod “Maybe (it’s him).” (2.55) DH: W’agyt luu galiiyuuwithl galant, wyit wil witxw’t dimil yeet, gasgoohlit ixsta’danhl hla lats’ Gwisgwoos. “He followed the trail until he got to where he was going because he liked the taste of the Gwisgwoos (a mythical bird).” ’ BH: ii hindahl wadiit? ’ ii hinta=hl wa-tiit? conj arrive.at=cnd place-3pl “And where did they get to?” DH: ansiilinaasxwdiidima an-siilinaasxw-tiit=ima geo.loc-hunting.grounds-3pl=mod “Maybe their hunting grounds.”  69  2.7. The Morphosyntactic Distribution of =kat and =ima DH: (elicited) bakwdiid=ima pakw-tiit=ima arrive.pl-3pl=mod “Maybe they got to  ansiilinaasxwdiit an-siilinaasxw-tiit geo.loc-hunting.grounds-3pl their hunting grounds.”  As with =kat, the most common surface position for =ima is as an enclitic to the first syntactic phrase in a clause. Because Gitksan is VSO, this is typically the verb, as in (2.56): ’ (2.56) t’aahldiidimahl iis ’a=hl lipnidiit ’ [t’aahl-tiit]vp =ima=hl iis ’a=hl lip-nitiit pick.berries-3pl=mod=cnd soapberries obl=cnd refl-3pl “They might be picking soapberries for themselves.” Example (2.57) is a Dependent clause: the first constituent in the clause is not the verb, but a preverbal phrase, in this case the progressive yukw. As with =kat, =ima encliticizes to the first phrase in the clause: (2.57) yukwima [yukw]xp =ima prog=mod “They might be  ’ dim t’aahldiithl iis ’a=hl lipnidiit ’ tim [t’aahl-tiit]vp =hl iis ’a=hl lip-nitiit fut pick.berries-3pl=cnd soapberries obl=cnd refl-3pl picking soapberries for themselves.”  However, example (2.58) shows that, with some speakers, =ima remains on the verb despite the available second position on yukw : (2.58) yukwhl dim t’aahldiidimahl iis ’a=hl [yukw]xp =hl tim [t’aahl-tiit]vp =ima=hl iis ’a=hl prog=cnd fut pick.berries-3pl=mod=cnd soapberries obl=cnd ’ lipnidiit ’ lip-nitiit refl-3pl “They might be picking soapberries for themselves.” There is a fourth, less robust pattern, as shown in (2.59), where =ima encliticizes to the clause. This pattern appears to be restricted to clauses with sentence-final pronouns: 70  2.7. The Morphosyntactic Distribution of =kat and =ima (2.59)  ’ a. siipxwima niin ’ S [siipxw=ima niin] sick=mod 2sg “You might be sick.” “Maybe you’re sick.” ’ b. siipxw niinima ’ S [siipxw niin]=ima sick 2sg=mod “You might be sick.” “Maybe you’re sick.”  As mentioned above, the surface variability of =kat and =ima in single clauses does not reflect a difference of meaning of any kind. This generalization contrasts with languages such as Plains Cree, where the morphosyntactic attachment site of an evidential correlates with different evidential meanings (Blain and Déchaine 2007; Déchaine 2008). Additionally, the clause types in Gitksan also do not reflect any evidential meaning, as the syntax and semantics of =ima and =kat are identical in independent and dependent clauses. This observation is relevant for the class of languages where clause-typing corresponds to different evidential meanings. For example, Plains Cree has two clause types that occur as matrix clauses, the Independent and Conjunct modes, which correspond to different kinds of evidential meaning. Déchaine (2008) frames this in terms of mediated knowledge: with the Independent mode, the speaker is present, and thus has unmediated knowledge of the event (direct evidence). In the Conjunct mode, the speaker is not present, and thus has mediated knowledge (indirect evidence) (see Blain and Déchaine (2007), Déchaine (2008), and Cook (2008) for details). The syntactic analysis of =ima and =kat must await further research, but what is clear is that position does not affect meaning.  71  ’ 2.8. The Sensory inferential nakw  2.8  ’ The Sensory inferential nakw  ’ Tarpent (1987, p. 354) claims that nakw in Nisgha’a is both a modal and evidential which ‘introduces a highly probable statement based on direct evidence’. (2.60) Nisgha’a (Tarpent 1987, p. 354) ’ nakwhl ’ nakw =hl evid=cnd “Baby must  beebii woks wok-(t)=s peepii sleep-3sg=pnd Baby be sleeping!” 19  ’ Tarpent translates nakw as the epistemic modal must, but does not include with her examples contexts which illustrate its evidential meaning encoding ‘direct evidence’. ’ Gitksan consultants corroborate this translation of nakw in Gitksan, and have also provided a basic evidential context for its use, given in (2.61): (2.61) Gitksan Context: You see Baby lying on her tummy, not moving. ’ nakwhl ’ nakw =hl evid=cnd “Baby must  woks beebii wok-(t)=s peepii sleep-3sg=pnd Baby be sleeping.”  This subsection describes in detail the evidential meanings, and the semantic, prag’ matic and morphosyntactic distribution of nakw. It is shown that Tarpent’s original ’ description of nakw in Nisgha’a is generally accurate, and I have imported from Nis’ gha’a Tarpent’s gloss of nakw as an evidential.20 However, as with =kat and =ima ’ above, we can further sharpen the meanings and distribution of nakw. In anticipation 19  ’ Tarpent includes the exclamation mark (!) in all of her translations of nakw. I have capitalized ’Baby’ on the assumption that it is being used as a proper noun, as it receives the proper noun enclitic determiner =s and not the expected common noun determiner =hl. My Gitksan consultants agree with this usage. 20 ’ In some parts of her grammar, Tarpent sometimes glosses nakw as a modal.  72  ’ 2.8. The Sensory inferential nakw ’ of this, this subsection claims that nakw corresponds to Aikhenvald’s description of an inferential evidential that specifies that a speaker has physical evidence. More specif’ ically, a sentence containing nakw indicates that the eventuality was not personally experienced but was inferred from indirect evidence, and that this indirect evidence is of a sensory nature. It is the physical nature of this evidence that likely led Tarpent to ’ describe nakw in Nisgha’a as a ‘direct’ evidential.  2.8.1  ’ The Evidential Meanings of nakw  Based on Tarpent’s observations (i.e. ‘a highly probable statement based on direct ’ evidence’), contexts were provided to consultants where nakw is expected to emerge. ’ For example, one would expect nakw to be felicitous in contexts such as the following, where a speaker can see that John’s lights are on: Context: You need to ask John for a favour. You drive by his place with a friend and notice the lights are on and his truck is in the driveway. Your friend says to you “Why don’t you stop by and ask; John must be at home.” However, =ima can be still be used in this context, as section §2.6.1 above showed that =ima is felicitous in contexts that provide observable evidence such as this. Thus, ’ an alternative elicitation strategy was used: =ima was directly contrasted with nakw by constructing minimal pairs between them, and asking the consultants to differentiate between (2.62) and (2.63) through constructing the appropriate contexts:21  21  ’ There is also a homophonous spatial/distal marker nakw. Tarpent (1987, p. 354) notes that there  ’ ’ ’ is a stress difference between evidential and distal nakw : evidential nakw is unstressed [nakw], while ’ ’ distal nakw is stressed [nákw]. This is confirmed by two consultants (BS and LW), who describe distal ’ ’ nakw as ‘longer’ than evidential nakw.  73  ’ 2.8. The Sensory inferential nakw (2.62) mugwimahl mukw=ima=hl ripe=mod=cnd “The berries might  maay’ maay’ berries be ripe.”  Comments: “When you say mugwimahl maay’ to someone it’s like you’re sitting at home talking about it, trying to decide if you go picking or not.”  ’ (2.63) nakwhl mukwhl maay’ ’ nakw =hl mukw=hl maay’ evid=cnd ripe=cnd berries “The berries must be ripe.” =“The berries might be ripe.” ’ Comments: “When you say nakwhl mukwhl maay’ you see people running through the forest with buckets all happy, or people coming home from the Suskwa with buckets full of berries. Not really good when you’re just thinking about it.” (BH)  What is crucial to note in this methodology is that the consultants were asked to ’ directly contrast =ima with nakw. These comments and their translations provide an ’ important clue as to the evidential meaning of nakw : it is felicitous only in contexts where a speaker can make an inference from the physical evidence available provided by ’ the circumstances. Given this, nakw would be expected to be infelicitous in a context where a speaker can only infer from experience, general knowledge, or simply conjecture. This prediction is borne out in the infelicity of example (2.64), a context where there is no physical evidence available to the speaker: (2.64) Context: You’re sitting at home talking about going berry-picking. It’s August, and the berries are usually ripe this time of year on the Suskwa. ’ # nakwhl mukwhl maay’ ’nakw =hl mukw=hl maay’ evid=cnd ripe=cnd berries “The berries must be ripe.”  74  ’ 2.8. The Sensory inferential nakw ’ The second important feature differentiating =ima from nakw is found in their trans’ lations. Consultants consistently comment that this physical evidence makes nakw carry more force, which is why it is frequently translated as must. Thus, the second translation ’ in (2.63),“The berries might be ripe”, is not typically an acceptable translation of nakw. ’ The evidential nature of nakw is more transparent in one of its frequent translations as ‘I can see that ...’, as in (2.65): (2.65) Context: Your friend laughs at you when you tell her you won $1000 at bingo last night. ’ nagwimi xsingy’ ’ nakw =mi xsink-y’ evid=2sg disbelieve-1sg “I can see that you don’t believe me.” Examples (2.63) and (2.65) involve a variety of indirect, physical evidence that was perceived visually by the speaker (happy people running through the forest, ripe berries, ’ laughing etc.). More specifically, evidential interpretations of nakw cover senses other than visual, such as tactile (2.66), auditory (2.67), and olfactory senses (2.68): (2.66) Context: You touch your daughter’s forehead and it’s very hot. ’ nakwhl siipxwin ’ nakw =hl siipxw-n evid=cnd sick-2sg “You must be sick!” (2.67) Context: You hear your friend’s stomach start to grumble loudly. ’ nakwhl xdaxwin ’nakw =hl xtaxw-n evid=cnd hungry-2sg “You must be hungry!”  75  ’ 2.8. The Sensory inferential nakw (2.68) Context: You’re chopping wood out by the smokehouse, and you can smell smoke and fish. ’ nakwhl sihons Bob ’nakw =hl si-hon-(t)=s Bob evid=cnd caus-fish-3sg=cnd Bob “Bob must be smoking/preparing/doing up fish.” ’ Another frequent translation of nakw offered by consultants is one that resembles a rhetorical question in English, as in (2.69): (2.69) Context: A car comes tearing down the road in front of your house. The road is covered in ice and you’re watching him blow by at 100km/hr. ’ nakwhl maalut ’nakw =hl maalu-t evid=cnd crazy-3sg “Is he crazy? He must be crazy!” These same examples modified with =ima are still compatible with the context given above. However, =ima in physical evidence contexts will always convey a more speculative, less forceful interpretation, as in (2.70): (2.70) Context: You see or hear of a bear wandering around the village. The bear is actually very friendly – a little unusual for a bear. ’ maaluyima nit ’ maalu=ima nit crazy=mod 3 “He/she/it might be crazy.” In contrasting (2.69) with (2.70) there are two things happening: first, (2.70) lacks the rhetorical effect of (2.69): a speaker is simply asserting that the bear might be crazy. Secondly, in (2.70) the speaker is concluding that it’s not necessarily the case that a bear 76  ’ 2.8. The Sensory inferential nakw seen wandering around the village is crazy; it may be sick. However, (2.69) expresses that a speaker believes the observable evidence they perceive unambiguously warrants ’ a necessity-like modal interpretation, and this function is fulfilled by nakw. Similarly, in (2.71)a., a speaker is expressing that it’s not necessary that the blood on the rocks is from your friend’s hand – it could be blood from the bait you were cutting up, whereas in b. the speaker is committing to the claim that blood they observe on the rocks is indeed from your hand: (2.71) Context: You and a friend are going fishing. You notice blood on the rocks ahead of you where your friend is walking. ’ a. k’ojinimahl ’onin ’ k’ots-i-n=ima=hl ’on-n cut-tr-2sg=mod=cnd hand-2sg “You may’ve cut your hand.” ’ ’ ’onin b. nagwimi g’otshl ’ ’nakw =mi k’ots=hl ’on-n evid=2sg cut=cnd hand-2sg “You must’ve cut your hand.” ’ Comments (paraphrased): k’otsinimahl ’onin: You might’ve cut your hand, or, I think ’ ’ you cut your hand. nagwimi g’otshl ’onin: It looks like you cut your hand, you must’ve because there’s blood on the rocks.  ’ Both =ima and nakw are felicitous in this context, where a speaker has access to ’ physical evidence in making a judgment about a situation. However, nakw requires this evidence to be observable, whereas =ima does not. This distinction can be further brought out in cases of simple speculation. Recall (2.31) above, where a speaker is ’ speculating on the arrival of Alvin. Given the lack of observable evidence, nakw is expected to be infelicitous, as example (2.72) shows: (2.72)  witxws Alvin? Q. gaxguhl kaxwi=hl witxw=s Alvin? when=cnd arrive=pnd Alvin “When is Alvin arriving?” 77  ’ 2.8. The Sensory inferential nakw ’ t’aahlakw A1. witxwima nit ’ witxw=ima nit t’aahlakw arrive=mod 3 tomorrow “He might arrive tomorrow” “I think he’ll arrive tomorrow.” ’ A2. nakwhl witxwt t’aahlakw ’nakw =hl witxw-t t’aahlakw evid=cnd arrive-3 tomorrow = “He might arrive tomorrow.” Observable evidence is not necessary in making speculative judgments, which often ’ predict future events. Therefore nakw is predicted to not be possible in these contexts, ’ as the second response in (2.72) shows. However, nakw can be used if there is at least some accessible physical evidence which can be inferred from the context: (2.73) Context: I can see his truck in the driveway. ’ nakwhl ’ nakw =hl evid=cnd “He must’ve  hlihl witxwt hlihl witxw-t already arrived-3 arrived already.”  An important consideration in evidence contexts is the orientation of the evidence. For example, (2.74) has two different contexts: the first involves a speaker’s experience ’ with similar situations. In this context we expect nakw to be infelicitous. However, ’ the second context involves observable physical evidence, yet nakw is still infelicitous. This infelicity is attributed to the fact that the evidence under consideration does not support the inference that the horse must’ve run away. Thus, as expected, =ima retains its flexibility in expressing varying degrees of modal force:  78  ’ 2.8. The Sensory inferential nakw (2.74) Inference from a speaker’s experience with similar situations: There was a terrible storm earlier in the day, which can spook the horse. The horse has been known to escape from the field when it gets scared. Information inferred by observable physical evidence: People are sitting around talking about a fence around a field that’s in disrepair. a. guxwimahl gyuwatan kuxw=ima=hl kyuwatan run.away=mod=cnd horse “The horse might’ve/must’ve run away.” ’ b. #nakwhl guxwhl gyuwadan ’ nakw =hl kuxw-(t)=hl kyuwatan evid=cnd run.away-3sg=cnd horse “The horse must’ve ran away.” By adjusting the context to include physical observable evidence that supports an ’ inference that the horse must’ve run away, as in (2.75), nakw is felicitous. In these observable evidence contexts, the modal strength interpretations are split between =ima ’ ’ and nakw, where =ima expresses might, and nakw expresses must: (2.75) Inference from observable evidence: You see there are tracks in the field that lead through a hole in the fence. a. guxwimahl gyuwatan kuxw=ima=hl kyuwatan run.away=mod=cnd horse “The horse might’ve run away.” ’ b. nakwhl guxwhl gyuwatan ’ nakw =hl kuxw-(t)=hl kyuwatan evid=cnd run.away-3sg=cnd horse “The horse must’ve ran away.” A similar alternation can be observed in (2.76) and (2.77): 79  ’ 2.8. The Sensory inferential nakw (2.76) Inference from a speaker’s experience with similar situations: It’s November – a typically rainy month. a. yugwimahl dim yukw=ima=hl tim prog=mod=cnd fut “It might/must be going ’ b. #nakwhl ’ nakw =hl evid=cnd “It must be  wis wis rain to rain.”  yukw dim wis yukw tim wis prog fut rain going to to rain.”  (2.77) Inference from observable evidence: You hear some raindrops on the roof. There is a certain scent to the air that is associated with imminent rain. a. yugwimahl tim wis yukw=ima=hl tim wis prog=mod=cnd fut rain “It might be going to rain.” ’ b. nakwhl ’ nakw =hl evid=cnd “It must be  yukw dim wis yukw tim wis prog fut rain going to to rain.”  ’ nakw is characterized as an indirect evidential where a speaker draws an inference from some type of physical evidence. The above cases were visual, and below are exam’ ’ ples of tactile nakw in (2.78), and auditory nakw in (2.79). In both of the cases below, a speaker is directly ‘experiencing’ the evidence they have for making either a might-like ’ =ima-assertion, or a must-like nakw -assertion; in (2.78), it is touching a child’s hot forehead, and in (2.79), it is listening to a rumbling stomach:  80  ’ 2.8. The Sensory inferential nakw (2.78) Context: You touch your daughter’s forehead and it’s very hot. a. siipxw siipxw sick “You’re  ’ niin ’niin 2sg sick.”  ’ b. siipxwima niin ’ siipxw=ima niin sick=mod 2sg “You might be sick.” ’ c. nakwhl siipxwin ’ nakw =hl siipxw-n evid=cnd sick-2sg “You must be sick.” “Looks like you’re sick.” (2.79) Context: Your hear your friend’s stomach start to grumble loudly. a. xdaxw xtaxw hungry “You’re  ’ niin ’ niin 2sg hungry.”  ’ b. xdaxwima niin ’ xtaxw=ima niin hungry 2sg “You might be hungry.” ’ c. nakwhl xdaxwin ’nakw =hl xtaxw-n evid=cnd hungry-2sg “You must be hungry!” The predicates in (2.78) and (2.79) involve an internal mental or physical state. ’ These are ideal cases for drawing out the modal-like interpretations of =ima and nakw . It was suggested above (with more argumentation to come in chapter 3) that =ima is 81  ’ 2.8. The Sensory inferential nakw better characterized as an epistemic modal, as it does not distinguish any specific kind ’ of evidence. The same question could be asked of nakw, as it also appears to carry a ’ modal quality, at least if we consider consultants’ translations of nakw. The fact that evidentiality is conceptually distinct from epistemic modality does however not preclude the possibility that languages have specific linguistic markers that may combine both. This appears to be borne out in Gitksan: the fact that =ima conveys both possibility and necessity could be correlated with the fact that it can be used speculatively (pos’ sibility), and within the presence of evidence (possibility and necessity). nakw is more specialized, but the correlation still holds: because it is only felicitous in contexts that have observable physical evidence, it expresses necessity. But the internal states represent contexts where a speaker must rely on what they believe is suitable evidence for an assertion. Of course, a speaker may make a simple assertion if they wish to completely commit to the truth value of a proposition, as in the a. examples of (2.78) and (2.79). ’ However, their use of =ima and nakw provides the speaker with options for modulating their claim. Contrast these internal state verbs with example (2.80): (2.80) Context: You see people walking through the door. a. bagw pakw arrive.pl “They’ve  ’ nidiit ’ nidiit 3pl arrived.”  ’ b. nakwhl bagwidiit ’ nakw =hl pakw=tiit evid=cnd arrive.pl=3pl “They must’ve arrived.” “Looks like they’ve arrived.” Given this context, a speaker is also directly experiencing the evidence they have for ’ a nakw -assertion. However, with in a context with a verb such as arrive, the truth or falsity of the event described by the verb is easily verifiable. This is what gives rise to ’ a mirative use of nakw, which is discussed in chapter 5.  82  ’ 2.8. The Sensory inferential nakw ’ The must-like quality of nakw is supported by the fact that multiple occurrences ’ of nakw in coordinated sentences that contain contradictory propositions is infelicitous. For example, the translations of (2.81) and (2.82) lead to a contradiction ( φ ∧ ¬φ). ’ (2.81) *nakwhl kuxwhl gyuwatan, ii needii ’nakw =hl kuxw-(t)=hl kyuwatan, ii nee=tii evid=cnd run.away-3sg=cnd horse conj neg=contr guxwimahl gyuwatan kuxw-(t)=ima=hl kyuwatan run.away-3sg=mod=cnd horse # “The horse must’ve ran away, and it must not have.” # “Looks like the horse ran away, and it looks like it didn’t.” ’ (2.82) *hla nakwhl ’ hla nakw =hl incept evid=cnd wis wis rain # “It must be starting  dim wis, ii needii hla tim wis, ii nee=tii hla fut rain, conj neg=contr incept  ’ nakwhl ’ nakw =hl evid=cnd  ’ nakw dim ’ nakw tim evid fut  to rain, and it must not be.”  Constructions of the form  φ∧  ψ, where φ is incompatible with ψ, are also infe-  licitous and ungrammatical, as in (2.83) and (2.84): ’ (2.83) *nakwhl guxwhl ’ nakw =hl kuxw-(t)=hl evid=pnd run.away-3sg=cnd ’ guxwindiithl (kuba) ’ kuxw-’en-tiit=hl (kuba) run.away-caus-3pl=cnd small  ’ kyuwatan, ii nakwhl ’ kyuwatan, ii nakw =hl horse conj evid=pnd ’tihlxwhl gyuwatan ’tihlxw=hl kyuwatan child=cnd horse  #“The horse must’ve ran away, and the kids must’ve chased it away.” ’ ’ ’ (2.84) *nakwhl yukw dim gahahlalsdiit, ii nakwhl yukw dim ’ ’ nakw =hl yukw tim nakw =hl yukw tim kahahla’lst-tiit, ii conj evid=cnd prog fut evid=cnd prog fut work.pl-3pl iixwdiit iixw-tiit fish-3pl #“They must be working (today), and they must’ve gone fishing.” 83  ’ 2.8. The Sensory inferential nakw However, despite these observations, there is independent evidence from the mor’ phosyntactic, semantic, and pragmatic distribution of nakw that suggests against a ’ modal treatment of nakw, and these are reviewed in the following subsections. ’ ’ Given the predicate status of nakw, a logical option would be to approach nakw as type of auxiliary verb. For example, in (2.85)a., the progressive yukw functions as a ’ predicate that nominalizes the lower verb, taking it as its argument (cf. (2.42)). nakw morphosyntactically is exactly the same, as in b.: (2.85)  a. yukwt jabs Sheilahl hon [yukw]pred [-t tsap-(t)=s Sheila=hl hon]arg prog-3sg cook-3sg=pnd Sheila=cnd fish “Sheila is cooking the fish” ’ b. nakwt jabs Sheilahl hon ’ [nakw ]pred [-t tsap-(t)=s Sheila=hl hon]arg evid-3sg cook-3sg=pnd Sheila=cnd fish “Sheila must’ve cooked the fish!” “I see that Sheila cooked the fish.”  ’ It would then be a natural step to claim nakw is an evidential verb, similar to see or ’ hear in English. This accords with the second translation of nakw in (2.85)b. However, ’ under this analysis, nakw has propositional status, and in chapter 3 I will argue that ’ nakw cannot undergo any of the operations we expect of propositional objects. Unlike an evidential verb in English, (2.85)b. cannot be questioned, negated or otherwise challenged.  2.8.2  ’ The Pragmatic Uses of nakw : Mirativity and Metaphor  Mirativity refers to the grammatical marking of a proposition as representing information which is surprising to the speaker (DeLancey 1997; 2001). A mirative interpretation ’ is also associated with nakw, which can also be used to express surprise at a situation, such as the unexpected arrival of guests at a party in (2.86):  84  ’ 2.8. The Sensory inferential nakw (2.86)  ’ a. bagw nidiit ’ pakw nidiit arrive.pl 3pl “They’ve arrived.” ’ b. nakwhl bagwidiit ’ nakw =hl pakw=tiit evid=cnd arrive.pl=3pl “They must’ve arrived.” “Looks like they’ve arrived.”  What is notable about (2.86)b. is that there is a certain threshold that is crossed ’ with respect to the evidential requirements of nakw. In all of the contexts examined so far, a speaker is making an assertion inferring from indirect evidence of a physical nature. However, in (2.86), a speaker is indeed witnessing the actual event: if there are people standing in the doorway, and the speaker sees them, then one would expect a simple assertion would be felicitous and adequate – there is no need for an evidentialbased assertion. These are the key ingredients for a mirative interpretation, and this is presented and analyzed in detail in chapter 5: when a speaker chooses to make a ’ nakw -assertion over a regular assertion, they are expressing that the circumstances they are faced with are surprising and perhaps unexpected. Note that this is actually not ’ incompatible with the evidential requirements of nakw as they’ve been defined so far: ’ a speaker is still basing their nakw -assertion on observable evidence – the fact they see people standing in the doorway. ’ In addition to its evidential and mirative uses, nakw has a metaphorical use, rendering an expression similar to a rhetorical question/statement in English. Consider a context where the speaker is watching a baseball game. The star batter on the speaker’s favourite team keeps missing the ball and striking out, jeopardizing the outcome of the game. Out of exasperation, the speaker comments:  85  ’ 2.8. The Sensory inferential nakw (2.87) Context: A friend is at bat in a baseball game. A couple of really easy pitches were thrown his way, but he missed them. His frustrated teammates yell out ‘Are you blind?!’ (Taken from an anecdote given by Jane Smith, and both translations volunteered by her.) ’ nakwhl sinst ’ nakw =hl sins-t evid=cnd blind-3sg “He must be blind!” “Is he blind or something?” “Looks like he’s blind!” ’ The mirative and metaphorical uses of nakw are examined in detail in chapter 5.  2.8.3  ’ The Morphosyntactic Distribution of nakw  ’ The morphosyntactic behaviour of nakw is markedly different from =ima or =kat. In ’ subsection 2.8.1, nakw was described as a kind of auxiliary verb (cf. example (2.85)). ’ ’ This description of nakw is supported by the observation that nakw patterns identically to other auxiliary verbs such as the progressive yukw and the imperfective hliskw, as shown in (2.88): (2.88)  a. yukwhl [yukw]pred [=hl prog=cnd “The women are  kahahlal’stdiithl haanak kahahlal’st-tiit=hl haanak]arg redup.pl-work-3pl=cnd women.pl working.”  b. hliskwhl kahahlal’stdiithl haanak [hliskw]pred [=hl kahahlal’st-tiit=hl haanak]arg imperf.=cnd redup.pl-work-3pl=cnd women.pl “The women finished working.” ’ c. nakwhl kahahlal’stdiithl haanak ’ [nakw ]pred [=hl kahahlal’st-tiit=hl haanak]arg evid=cnd redup.pl-work-3pl=cnd women.pl “The women must be working.” 86  2.9. Interactions More specifically, these three auxiliary verbs function as intransitive predicates, creating Dependent clauses (cf. (2.42)): they nominalize the following VP, marking it with the common noun determiner =hl, as schematized in (2.89): ’ (2.89) {yukw, hliskw, nakw }=hl VP YP ’ However, there is one feature that sets nakw apart from the other auxiliary verbs: ’ no other element can precede nakw in a clause. For example, the inceptive marker, ’ hla, precedes yukw in (2.90)a. However, this is ungrammatical with nakw, as shown in (2.90)b.: (2.90)  a. hla yukwhl hla yukw=hl incept prog=cnd “The women are about b. *hla hla incept  kahahlal’stdiithl haanak kahahlal’st-tiit=hl haanak redup.pl-work-3pl=cnd women.pl to start working.”  ’ haanak nakwhl kahahlal’stdiithl ’nakw =hl haanak kahahlal’st-tiit=hl evid=cnd redup.pl-work-3pl=cnd women.pl  ’ This covers some of the basic syntactic features of nakw. The syntactic behaviour ’ of nakw, including the restriction that no other element may precede it, is examined in more detail in chapter 3, §3.6.2.1, and chapter 5.  2.9  Interactions  In English it is possible to express combinations of modal and reported meanings through sentences such as (2.91): (2.91)  a. I heard the berries must be ripe. b. I might’ve heard the berries must be ripe.  87  2.9. Interactions Given that these meanings potentially correspond to the evidential-modal system of Gitksan examined so far, a logical step would be to see to what extent =kat, =ima, and ’ nakw combine to express meanings such as those in (2.91). However, this subsection ’ shows that combinations of =kat, =ima, and nakw are in general not possible. As a starting point, to my knowledge there is no Gitksan-internal restriction on the stacking of enclitics, thus it is expected that =kat and =ima could appear simultaneously stacked on a predicate in order to achieve the kinds of readings that correspond to those in (2.91). This is a plausible move, as we might expect the relative order between =kat and =ima would reflect a difference in scope. However, the stacking of =kat and =ima is not possible, as (2.92) shows: (2.92)  a. *mukw=ima=kat=hl maay’ ripe=mod=rep=cnd berries =“(I heard) the berries might/must be ripe.” =“I might’ve heard the berries are ripe.” maay’ b. *mukw=kat=ima=hl ripe=rep=mod=cnd berries =“(I heard) the berries might/must be ripe.” =“I might’ve heard the berries are ripe.”  A potential alternative strategy is to force the co-occurrence of =ima and =kat. This can be done using the syntax of Dependent clauses. Recall that Dependent clauses consist of a sentence initial predicate which nominalizes the VP (cf. (2.42)). Given their second position behaviour, both =kat and =ima encliticize to the sentence initial predicate. An example of this is given in (2.93)a., where modal =ima encliticizes to the progressive yukw. However, it was also shown that there is a certain amount of speaker variation (cf. (2.43)), where the second position clitics remain on the verb rather than assuming the second position, as shown in (2.93)b.:  88  2.9. Interactions (2.93)  a. yukwima [yukw]xp =ima prog=mod ’ lipnidiit ’ lip-nitiit refl-3pl “They might be  dim t’aahltiithl iis ahl tim [t’aahl-tiit]vp =hl iis a=hl fut pick.berries-3pl=cnd soapberries obl=cnd  picking soapberries for themselves.”  b. yukw dim t’aahltiidimahl iis ahl [yukw]xp tim [t’aahl-tiit]vp =ima=hl iis a=hl prog fut pick.berries-3pl=mod=cnd soapberries obl=cnd ’ lipnidiit ’ lip-nitiit refl-3pl “They might be picking soapberries for themselves.” Given the potential for these two positions (XP and VP) to host =kat and =ima, nothing in the morphosyntax prevents both =kat and =ima from appearing in the same sentence, each occupying one of these positions. This would also be a reasonable way of controlling the scope between =kat and =ima, thus obtaining the different readings in (2.91). However, because these positions are subject to free variation, the exact opposite is predicted, and these readings do not obtain: there are no scopal interactions because these positions are not semantically significant. Example (2.94)a. shows that =ima > =kat does not achieve a “might’ve heard” reading, and (2.94)b. shows that =kat > =ima does not achieve a “heard that X might’ve” reading: (2.94)  a. *yukw=ima tim t’aahl-tiit=kat=hl iis ’a=hl prog=mod fut pick.berries-3pl=rep=cnd soapberries obl=cnd ’ lip-nitiit refl-3pl =“I might’ve heard they’re picking soapberries for themselves.” b. *yukw=kat=hl tim t’aahl-tiit=ima=hl iis prog=rep=cnd fut pick.berries-3pl=mod=cnd soapberries ’ ’a=hl lip-nitiit obl=cnd refl-3pl =“(I heard) they might be picking soapberries for themselves.” 89  2.9. Interactions ’ ‘to hear’ (2.95), and a Gitksan has a verb which expresses reported meaning, laxni ’ propositional attitude verb haniigoot ’to think/believe’ (2.96), which can express modal meaning: ’ y’ (2.95) laxni wil xstas bingo gaxxw John go’ohl ’ y’ wil laxnixsta=s bingo kaxxw John ko’=hl hear-1sg comp win=pnd John loc=cnd bingo last.night “I heard that John won at bingo last night.” ’ siipxwin (2.96) haniigoodi y’ wil ’ ’ wil siipxw-n haniigoot-y comp sick-2sg think-1sg “I think you’re sick.” Rather than combining =kat and =ima simultaneously within the same sentence, ’ and haniigoot. ’ the meanings in (2.91) are achieved in combination with laxni Example (2.97)a. shows how a report > modal reading obtains by embedding the modal =ima ’ The reverse modal > report reading is obtained in the complement clause of laxni. ’ itself, as in (2.97)b.: by encliticizing modal =ima to laxni (2.97)  ’ y’ a. laxni wil xstayimas ’ y’ wil laxnixsta=ima=s hear-1sg comp win=mod=pnd “I heard that John might’ve won at  John John John bingo  bingo gaxxw go’ohl bingo kaxxw ko’o=hl loc=cnd bingo last.night last night.”  ’ yima ’ b. laxni wil xstas John go’ohl bingo gaxxw ’ y=ima ’ John ko’o=hl laxniwil bingo kaxxw xsta=s hear-1sg=mod comp win=pnd John loc=cnd bingo last.night “I might’ve heard that John won at bingo last night.” ’ Embedding modal =ima in the complement clause of the attitude verb haniigoot, as in (2.98), has the effect of further modulating a modal statement (i.e. ‘harmonic’ readings; cf. Palmer 2006): ’ (2.98) haniigoodi y’ wil siipxwinima ’ siipxw-n=ima haniigooty’ wil comp sick-2sg=mod think-1sg “I think you might be sick.” 90  2.9. Interactions The same combinatorial restriction found with =kat and =ima extends to evidential ’ ’ nakw. Recall that evidential nakw requires the speaker to have access to some type of physical evidence. Even though there is no independent reason in the morphosyntax to ’ exclude combinations of nakw with =kat or =ima, the physical evidence requirement of ’ nakw would naturally exclude these combinations, as (2.99) shows: (2.99)  ’ a. *nakw=kat=hl pakw=tiit evid=rep=cnd arrive.pl=3pl =“(I heard) it looks like they’ve arrived.” = “(I heard) they must’ve arrived.” ’ b. *nakw=ima=hl pakw=tiit evid=mod=cnd arrive.pl=3pl =“It might look like they’ve arrived.” = “Maybe they must’ve arrived.”  ’ The incompatibility of nakw and =kat in (2.99)a. can be explained given their ’ respective evidential meanings: whereas nakw and =kat are both indirect evidentials, ’ nakw requires this indirect evidence to be of a sensory nature, while =kat requires the evidence to come in the form of a report from another source. ’ The nature of the incompatibility found with nakw and =ima in (2.99)b. is somewhat ’ ’ different. Recall that nakw and =ima stand in a specific kind of relationship: nakw is felicitous in a subset of contexts that =ima is felicitous in, specifically those in which ’ there is physical evidence. The function of nakw in these contexts is to express a ’ speaker’s confidence in the physical evidence in supporting a nakw -assertion, giving it a must-like interpretation. On the other hand, the function of =ima in a physical evidence ’ context is to stand in contrast to nakw in expressing a speaker’s doubt in using that physical evidence to make an inference. Thus, given this contrasting relationship, it ’ would seem odd to combine both nakw and =ima within the same sentence.  91  2.10. Summary  2.10  Summary  In this chapter I gave a semantic and morphosyntactic description of the three evidentials in Gitksan, the reportative =kat, the inferential =ima (glossed as mod) and the sensory ’ evidential nakw. It was shown that all three of these evidentials encode different kinds of indirect evidence. The reportative =kat indicates that the information was reported to the speaker by another person. The more general =ima is felicitous in any kind of context that has indirect evidence, whether through the physical senses, or from general ’ knowledge or speculation. The evidential nakw is more specific: it can only be used in contexts where the speaker has sensory evidence for an inference. There are also ’ additional pragmatic properties associated with nakw, where it can have a mirative and metaphorical use. These uses do not extend to inferential =ima. The reportative =kat and modal =ima share the same morphosyntactic class: they are classic second position clitics. Their surface position is somewhat variable (likely determined by phonological factors), but this has no effect on their interpretation. This generalization bears upon the hypothesis that the morphosyntactic placement of an evidential can reflect a difference in meaning (see Blain and Déchaine (2007); Déchaine (2008); Cook (2008) for details), and that in some languages, clause-typing corresponds to different evidential meanings. This is also not borne out in Gitksan, where the Independent vs. Dependent clause types do not appear to reflect any evidential meanings. ’ The evidential nakw, however, has very different syntactic properties: it has the ’ syntactic characteristics of a predicate, or auxiliary verb. nakw is also quite restricted in its surface position: it can only occur at the front of a clause. Finally, there is a restriction on more than one evidential from co-occuring in the same clause. ’ Of particular interest is the translations of =ima and nakw : both are translated by speakers as epistemic modals. However, it is shown in the following chapters, that in ’ the case of nakw, this is misleading: various tests are applied in the next chapter that ’ show nakw cannot be an epistemic modal, despite its translation as one. On the other hand, these same tests confirm that both =ima and =kat are indeed modal evidentials. 92  Chapter 3 Evidentiality and Levels of Meaning Chapter 2 described the core meanings and basic morphosyntax of the three morphemes in Gitksan that encode evidentiality. In this chapter I take the next step in probing their meanings by applying a set of theoretical tests that will determine what level of meaning the individual evidentials operate on. I assume that there are two levels of meaning, discussed in detail below: a propositional level (i.e. an evidential contributes to the truth conditions), or an illocutionary level (i.e. an evidential does not contribute to the truth conditions). Given the outcome of these tests, we can then ascertain the appropriate semantic or pragmatic analysis of them, which is undertaken in chapters 4 and 5.  3.1  The Issues  I frame the task of testing the individual Gitksan evidentials by revisiting the third of the four relations between evidentiality and epistemic modality that were identified in Chapter 1, in (3.1): (3.1) Formal semantic relation:  (cf. 1.15(iii.))  Can the individual evidentials in Gitksan be analyzed parallel to epistemic modals in formal semantics (i.e. quantification over possible worlds) or do they require a different, independent kind of analysis? ’ One prominent characteristic of both evidential =ima and sensory evidential nakw is their translations into English as modal verbs might and must:  93  3.1. The Issues (3.2) t’ayimat John t’a=ima=t John at.home=mod=pnd John “John might/must be at home.” ’ (3.3) nakwhl ’ nakw =hl evid=cnd “The berries  mukwhl maay’ mukw=hl maay’ ripe=cnd berries must be ripe.”  This feature bears directly on the debate within the literature on the encoding of evidential and epistemic notions. There are two theoretical sides to this debate. One the one hand, de Haan (1999) claims that evidentiality and epistemic modality are mutually exclusive: evidentiality encodes the source of the information contained in the utterance, while epistemic modality encodes the degree of commitment on the part of the speaker to the truth of the information. De Haan therefore proposes that an evidential which distinguishes only an information source is not a modal (see also Lazard 2001; Aikhenvald 2006). Matthewson et al. (2004) show that the St’át’imcets evidentials encode the source of evidence, and do not encode distinctions of certainty, thus they fall into de Haan’s definition of an evidential category. However, they also show that the individual evidentials in St’át’imcets must be analyzed as epistemic modals in the sense of being operators which quantify over epistemically accessible worlds. This analysis bears upon the question in (3.1): are the same formal tools adequate or appropriate for analyzing both epistemic modals and evidentials? Are there evidentials that are not amenable to a modal analysis? And if so, what kind of analysis is appropriate for evidentials that are not epistemic modals? A formal analysis of evidentiality as a type of epistemic modality was first utilized by Izvorski (1997) to analyze the evidential nature of the present perfect in Bulgarian. This inspired similar analyses of evidentials in other languages (e.g. Garrett 2001 on Tibetan, McCready 2005, McCready and Asher 2006 and McCready and Ogata 2007 on Japanese, Chung 2005 on Korean, and Matthewson et al. 2004 on St’át’imcets). However, Faller (2002) argues convincingly using data from the Cuzco Quechua evidential 94  3.2. The Plan system that evidentials are not a homogeneous class cross-linguistically. She shows that some evidentials in Quechua cannot be analyzed as epistemic modals and require a different kind of analysis. According to the illocutionary-level analysis developed in Faller (2002), an evidential modifies the illocutionary force and/or the sincerity conditions of a speech act. Quechua also shows that a language may have both modal and illocutionary evidentials. As highlighted in chapter 1, we can use this debate to distinguish between modal (i.e. St’át’imcets) and non-modal (i.e. Quechua) evidentials. Modal evidentials are propositional operators. Thus, we would predict them to behave as other propositional operators. For example, a modal evidential should contribute to the truth conditions of a sentence, be embeddable, and be sensitive to scope with respect to other operators. We predict the opposite with non-modal evidentials: they should not contribute to truth conditions, nor should they be embeddable or interact with any propositional operators.  3.2  The Plan  The goal of this chapter is to determine the modal vs. non-modal status of the individual evidentials in Gitksan. Before I undertake this task, in section 3.3 I review the basic semantics of modality and how it has been used to analyze modal evidentials. In §3.4 I turn to the formal approaches to non-modal evidentials: as outlined in chapter 1, there are competing theories, each of which are introduced in §3.4: the speech-act/illocutionary operators approach (Faller 2002), and the Dynamic Semantics approach (Portner 2006; Davis et al. 2007). Then in §3.5 I present the set of theoretical tests that have been used in the literature for determining the level of meaning an evidential operates on. This is done by examining the predictions these tests make for the individual evidentials in St’át’imcets, a language with modal evidentials (Matthewson et al. 2004), and Quechua, a language with nonmodal evidentials.  95  3.3. Evidentials as Epistemic Modals §3.6 is then devoted to testing these predictions on the individual Gitksan evidentials ’ =ima, =kat, and nakw. The outcome of these tests reveal that the evidential system in Gitksan confirms Faller’s hypothesis: Gitksan represents a language which shows a language-internal ‘split’ system, including both modal and non-modal evidentials. Both =ima and =kat are in fact epistemic modals. The results are the opposite for ’ nakw : despite its typical translation as must, it is not a modal evidential, but rather ’ a non-modal evidential. A non-modal analysis of nakw is given in chapter 5. In §3.7 I summarize these results.  3.3  Evidentials as Epistemic Modals  There are three interacting components to a modal: quantification, which determines what we interpret as modal force, and two conversational backgrounds, which work in tandem to determine the meanings of a modal in a given context (Kratzer 1981; 1991; among others). The first component is quantification: modals are quantifiers over possible worlds. Quantification encodes what we interpret as might and must: necessity modals are treated as universal quantifiers, and possibility modals as existential. (3.4)  a. “John must be at home.” = must(John be at home) b. “John might be at home.” = might(John be at home)  The interpretations of a modal are further determined by two independent, contextuallydetermined functions, or conversational backgrounds: a modal base and an ordering source (Kratzer 1991). We intuitively understand conversational backgrounds as the kinds of things we know about the actual world, be it the particular facts of some circumstance, the evidence available to us, the laws in place, the normal course of events etc. Using example (3.4), we could be taking the fact that John’s lights are on in the actual world as evidence. Another way to understand conversational backgrounds is the set of premises they represent in view of the facts known in the actual 96  3.3. Evidentials as Epistemic Modals world. For example, in view of the available evidence to us, such as the fact that his livingroom lights are on, “John must be at home” is true iff John is home in all the worlds w that are compatible with the available evidence in w, such as those worlds in which his livingroom lights are on. More technically, a conversational background picks out for the actual world w the set of worlds w which are accessible from w, via an accessibility relation. This forms the modal base worlds B(w) over which the modal quantifies. Thus a sentence of the form must(p), as in (3.4)a., states that the prejacent p ‘John be at home’ is true in all epistemically accessible worlds, while in (3.4)b. might(p), p is true in at least one epistemically accessible world. Thus, we can give the sentences in (3.4) the denotations in (3.5): (3.5)  a. “John must be at home.” must(B)(w)(John be at home)  c  = 1 iff ∀w ∈ B(w) : John be at  home c (w ) = 1 b. “John might be at home.” might(B)(w)(John be at home)  c  = 1 iff ∃w ∈ B(w) : John be at  home c (w ) = 1 Assuming an epistemic modal base B, (3.5)a. reads “In all the possible worlds w compatible with the speaker’s knowledge in the actual world w (i.e. where his livingroom lights are on), John is at home”, while b. reads “In at least one possible world w compatible with the speaker’s knowledge in the actual world w, John is at home.” Modal bases in English are determined by the context of utterance. The denotations in (3.5) have an epistemic modal base, provided by the context c. However, the modal sentences in (3.4) can have more than one meaning, depending on the context, as shown in (3.6):  97  3.3. Evidentials as Epistemic Modals (3.6) “John must be at home.” a. His lights are on, and his car is in the driveway. b. His parents imposed a 9pm curfew, and it’s now 10pm.  epistemic deontic  With an epistemic conversational background, the modal base will determine a set of worlds w which are epistemically accessible from w, or those worlds compatible with everything we know in w. With a deontic conversational background, the modal base contains the accessible worlds where the laws are the same as in w. The different kinds of modal base worlds picked out by a conversational background are represented by the kinds of meanings that are familiar in (3.7): (3.7) (i.) In B(w) worlds where what we know holds true (ii.) In B(w) worlds where laws holds (iii.) In B(w) worlds where certain physical laws hold (iv.) In B(w) worlds where desires are the same as in w  epistemic deontic circumstantial bouletic  Thus, the denotations for modals in English are given in (3.9):22 (3.9)  a. must(B)(w)(p) b. might(B)(w)(p)  22  c  = 1 iff ∀w ∈ B(w) : p c (w ) = 1 c  = 1 iff ∃w ∈ B(w) : p c (w ) = 1  If f is a conversational background, then the set of worlds that are accessible from w is the set  of worlds in which all the propositions of f (w) are true, or ∩f (w). We can have the conversational background as a parameter of evaluation (Kratzer 1991), or we can treat it as an argument of the verb, as in the denotations in (3.8) (following von Fintel and Heim 2007): (3.8)  a. must = λf.λp.λw.∀w [w ∈ ∩f (w) → p(w ) = 1] b. might = λf.λp.λw.∃w [w ∈ ∩f (w) ∧ p(w ) = 1]  For convenience, I will use the notation B(w) instead of ∩f (w).  98  3.3. Evidentials as Epistemic Modals It should be noted that the modals in (3.9) have just one meaning – they are not lexically ambiguous – but this single meaning contains a contextual parameter that determines how they are interpreted. The context c provides the modal base worlds, thus, depending on the context, they can be used to express any of the modal interpretations given in (3.7). I return to the semantics of modals in more detail in chapter 4. The denotations in (3.9) do not include a component for encoding a speaker’s relation to the evidence that makes up an epistemic conversational background. This is probably rooted in the empirical generalization that modals in languages such as English encode a speaker’s degree of commitment (i.e. modal force), but do not carry any explicit evidential information (although see von Fintel and Gillies (2007), who claim that English ‘must’ does carry an evidential restriction). However, there is a non-trivial link between modality and evidential meaning. This link can be viewed as a continuum, where at one end there are non-evidential morphemes such as modal verbs or tense/aspect morphemes that are felicitous only in specific kinds of evidential contexts, but without lexically encoding a specific type of evidence. We see this in languages such as Georgian and Bulgarian. At the other end of the continuum are the kinds of morphemes that lexically encode different kinds of evidential sources, as in St’át’imcets and Gitksan. For example, in both Georgian (3.10), and Chechen (3.11), the use of the perfect aspect conveys an evidential meaning, typically of the reported or inferred variety: (3.10) Georgian (Topadze 2007) Context: Someone told me about it; or I inferred it from the many cars parked outside. teat’r-ši bevri xalx-i q’opil-a theatre-in many people-nom be.perf-3sg “There were many people in the theatre.” “As it seems, there were many people in the theatre.  99  3.3. Evidentials as Epistemic Modals (3.11) Chechen (Molochieva 2007) Context: I didn’t see her and I was told about her visit; or I saw the cookies she had brought. Zaara so c’a quuch-lie dwa-j-ax-na xilliera Zara.nom 1s.nom home arrive-conv.post away-J-con.ant cop.plu “Zara had left before I arrived.” Comrie (1976) notes that “the semantic similarity . . . between perfect and inferential lies in the fact that both categories present an event not in itself, but via its results.” Thus, the perfect describes a completed event in the past relative to the moment of utterance, but which has lasting consequences perceptible at the time of speech. Izvorski (1997) shows that in Bulgarian, the ‘perfect of evidentiality’ (glossed as ‘pe’) has an indirect evidential interpretation in addition to its aspectual one: (3.12) Bulgarian (Izvorski 1997, p. 228) Maria Maria “Maria “Maria  celunala Ivan kissed.pe Ivan has kissed Ivan.” apparently kissed Ivan.”  perfect perfect of evidentiality (pe)  Izvorski claims that the evidential interpretation of pe in Bulgarian requires a more specialized type of proposition than an ordinary epistemic modal base: it is not sufficient for a proposition to be simply known for it to be considered (indirect) evidence for the core proposition. Because this evidential interpretation is not lexicalized, but parasitic on the perfect aspect, Izvorski posits an evidential operator ‘ev’ to handle the modal/evidential semantics, which is analyzed as a universal epistemic modal. The indirect evidence requirement is added in the form of a presupposition, which requires that a speaker have indirect evidence for an assertion using the pe. This is summarized in (3.13):  100  3.3. Evidentials as Epistemic Modals (3.13) The Interpretation of evp: a. Assertion:  p in view of the speaker’s knowledge state  b. Presupposition: Speaker has indirect evidence for p In other words, a modal base B(w) is further restricted by the presupposition to those worlds in which the evidence a speaker has for the assertion they make holds. This can be more precisely formulated in (3.14) (adopted from Izvorski 1997): (3.14) B(w) = {u ∈ W : ∀p[(p is the indirect evidence in w) → u ∈ p]} The perfect of evidentiality in Bulgarian does not lexically encode a specific evidence source, and can have other indirect evidence readings, including a reportative one. On the other end of the continuum are languages which specifically encode different kinds of evidential sources, but have a modal semantics. In a series of articles, Matthewson et al. (2004), Rullmann et al. (2008), and Davis et al. (2009) claim that the individual evidentials in St’át’imcets are in fact epistemic modals. The data in St’át’imcets presents a clear case of modals having specialized evidential meanings: the reportative ku7 in (3.15)a., the inferential k’a in b., and the perceived evidence -an’ in c.: (3.15)  a. Reportative ku7 : The speaker came to believe the sentence by means of a report. wa7 ku7 ku sts’éts’qwaz’ l-ta stswáw’cw-a be rep det trout in-det creek-det “[I heard] There are trout in the creek.” b. Inferential k’a: The speaker came to believe the sentence by means of inference based on perceived evidence, or general facts about the world. plan k’a tu7 wa7 tsu7c na máq7-a already infer then impf melt(inch) det snow-det “The snow must’ve already melted.” 101  3.3. Evidentials as Epistemic Modals c. Perceived Evidence -an’ : The speaker came to believe the sentence by means of inference from visual evidence. pel’p-s-ácw-an’ nelh neklíh-sw-a lost-caus-2sg.conj-perc.evid det.pl key-2sg.poss-det “It looks like you’ve lost your keys.” Rullmann et al. adapt the presupposition approach of Izvorski for evidential modals in St’át’imcets. First, they claim that the various evidential modals in (3.15) place an evidence presupposition on the modal base, and secondly, that this evidence presupposition can be specialized to include different subtypes of evidence. Example (3.16) gives the semantics of the inferential evidential k’a: (3.16) Semantics of k’a (inferential) k’a  c,w  (Matthewson et al. 2004, p. 245)  is only defined if c provides a modal base B such that for all worlds  w ∈ B(w), the inferential evidence in w holds in w , and f is a choice function such that f (B(w)) ⊆ B(w). If defined, k’a  c,w  = λf.λp.∀w [w ∈ f (B(w)) → p(w ) = 1].  The evidential modal suffix -an’ requires a more specialized sub-type of evidence than k’a. This is written into the presupposition in example (3.17): (3.17) Semantics of -an’ (inferential - perceived) -an’  c,w  is only defined if c provides a modal base B such that for all worlds w ,  w ∈ B(w), the perceived evidence in w holds in w , and f is a choice function such that f (B(w)) ⊆ B(w). If defined, -an’  c,w  = λf.λp.∀w [w ∈ f (B(w)) → p(w ) = 1].  The presupposition associated with the inferential evidential -an’ requires that the modal base contain all those worlds in which the perceived evidence in w holds. However, their analysis involves another contextually determined parameter, a choice function f 102  3.4. Formal Pragmatic Approaches to Evidentiality which picks out a subset of B(w). This is the locus of the variable modal force of the modal evidentials in St’át’imcets: the larger the subset of the modal base selected by f , the stronger the modal force expressed. f may simply be the identity function, which results in a universal must-like reading. If f selects a proper subset of the modal base, the resulting reading is weaker, although that subset is still universally quantified over. The issue of variable modal force, which is also found in the Gitksan modal =ima, is analyzed in chapter 4.  3.4  Formal Pragmatic Approaches to Evidentiality  Faller (2002) shows that most (but not all) evidentials in Quechua are not amenable to a modal analysis, and develops an illocutionary analysis utilizing classical speech act theory. Portner (2006) develops an alternative analysis within a dynamic semantics which treats evidentials as sentential force specifiers. Davis et al. (2007) also develop a kind of dynamic semantics approach, analyzing the meanings of evidentials in terms of their ability to ‘shift’ a particular contextual parameter, specifically, the degree of certainty that one must have in a proposition before one can utter it.  3.4.1  Speech Act Theory  Under classical speech act theory, utterances are treated as consisting of an illocutionary force, F , and a separate level of propositional content, p, such that F and p together, or F (p), form a complete utterance used to accomplish a speech act (Searle 1969). The illocutionary force F of an utterance also has a logical structure comprising five components, given in (3.18):  103  3.4. Formal Pragmatic Approaches to Evidentiality (3.18) (i.) The illocutionary point of an utterance (ii.) Strength of the illocutionary point (iii.) Preparatory conditions (iv) Mode of achievement (v.) Sincerity conditions Of these components, the sincerity conditions are the most intuitively accessible. For example, we understand particles such as damn and frankly as expressing something of the attitude or sincerity of a speaker towards the sentence they accompany. This kind of meaning is the most transparent with interjections such as damn, as in (3.19): (3.19) Damn(p) illocutionary force = assertion “p” sincerity condition = The speaker is upset that p Similarly, utterance-level adverbs such as frankly and alas are also amenable to this kind of illocutionary analysis, as in (3.20): (3.20) Frankly(p) illocutionary force = assertion “p” sincerity condition = The speaker is being frank in expressing p The intuition behind sincerity conditions is that you can’t utter a sentence such as (3.19) sincerely unless you are upset that p, and you can’t sincerely utter (3.20) without speaking frankly about p. There are two tests for determining whether a morpheme’s contribution is to the illocutionary force of an utterance or its propositional content. First, if a morpheme contributes only to the illocutionary force of an utterance, then in an indirect speech context, that morpheme cannot be understood as part of the propositional content of the indirectly described speech act. In other words, we do not expect illocutionary 104  3.4. Formal Pragmatic Approaches to Evidentiality operators to be embeddable. This effect can be observed with illocutionary adverbials such as frankly, honestly, and with attitudinal adverbials such as unfortunately, sadly (Ifantidou-Trouki 1993; Faller 2002). For example, (3.21) shows that frankly is not semantically embeddable, while reportedly and obviously are: (3.21) (i.) If John’s book has frankly sold very little, you shouldn’t be surprised. (ii.) If the ball was reportedly over the line, the matter should be investigated further. (iii.) If the cook obviously won’t poison the soup, we can eat the meal without worrying. (Faller 2002, p. 216, data from Ifantidou-Trouki 1993) Likewise, if a morpheme contributes only to the illocutionary force of an utterance, then it is expected that one cannot challenge or disagree with the meaning of that morpheme, as in (3.22): (3.22) (i.) A: Damn! Barbara cut my hair too short again!. B: No! That’s not true. (= you are not upset.) (ii.) A: Frankly, my opinion is that Bruce should do it. B: No! Not true (= you are not being frank.) Faller’s work on the evidential system of Cuzco Quechua likely represents the most prominent approach to examining evidentials and their properties as illocutionary operators (Faller 2002). Quechua has several enclitic suffixes that mark evidentiality or the nature of the speaker’s justification for making the claim. As an illustration, the reportative marker -si, as in (3.23), indicates that the speaker heard the information expressed in the claim from someone else: (3.23) para-sha-n-si rain-prog-3-si “It’s raining.” ev: speaker was told that it is raining 105  3.4. Formal Pragmatic Approaches to Evidentiality -mi indicates that the speaker has direct (usually perceptual) evidence for the claim, as in (3.24): (3.24) para-sha-n-mi rain-prog-3-mi “It’s raining.” ev: speaker sees that it is raining -chá indicates that the speaker’s background knowledge leads him to believe the information in the claim to be true, as in (3.25). (3.25) para-sha-n-chá rain-prog-3-chá “It may/must be raining.” ev: speaker conjectures that it is raining based on some type of inferential evidence Faller (2002) argues that evidentials in Quechua (except chá) do not contribute to the propositional content of an utterance. Rather, they modify the sincerity conditions of the speech act. For example, the reportative evidential -si modifies the commitment that is usually associated with an assertive operator, and changes it from one of asserting to one of ‘presenting’. Thus, illocutionary evidentials have the effect of introducing additional content into the set of preconditions of an assertion. With the reportative -si the propositional content p is not asserted, and Faller posits a special speech act present for this situation, on which the speaker simply presents a proposition without making claims about its truth. This is formally implemented in (3.26), where a condition is added to the set of sincerity conditions where a source other than the speaker asserts the truth of the proposition: (3.26) para-sha-n-si rain-prog-3-si p = “It is raining.” ill = present(p) sinc = ∃s2 [Assert(s2 , p) ∧ s2 ∈ / h, s]  (Faller 2002, 199) 106  3.4. Formal Pragmatic Approaches to Evidentiality In words, the sincerity condition in (3.26) says that there is another speaker, s2 , other than the speaker of the sentence, who asserts p. Thus, most Quechua evidentials only mark the source of information: they do not contribute a modal semantics to the sentence. No matter which evidential is used, the proposition expressed by a declarative remains the same as that of the phrase with which the evidential combines. Faller also claims that Quechua evidentials do not directly mark degree of certainty, though there are correlations: -mi normally establishes a high degree of certainty; -chá normally establishes a minimal degree of certainty (meaning the proposition is at least possible), and reportative -si has no uniform correlation with a degree of certainty. Additionally, under Faller’s approach, a declarative sentence does not necessarily fully assert the proposition expressed: -mi results in full assertion, while -chá results in a weakened assertion.  3.4.2  Evidential Hierarchies and Context Dependence  Davis et al. (2007) claim that when a speaker has more than one source of evidence for a proposition in a given context, they will use the evidential that represents the best or most reliable kind of evidence they have for that proposition. Thus we can differentiate between reliable and unreliable hearsay evidence, as in (3.27): (3.27) Speaker concludes from the evidence that John was at the feast last night. a. Speaker heard from a reliable source that John was at the feast last night. b. Speaker heard from a local drunk that John was at the feast last night. Whether a particular kind of evidence is better or more reliable than another can vary with the context. In a reliable evidence context such as (3.27)a., the speaker will use a hearsay/reportative evidential rather than a more general inference evidential, if the language possesses both. On the other hand, if the evidence comes from what is perceived to be an unreliable source in that context, the speaker will use the more general inference evidential rather than the more specific hearsay/reportative one. 107  3.4. Formal Pragmatic Approaches to Evidentiality This is formalized in the following way: every context c has a ‘subjectivity probability threshold’ cT that falls between 1 (absolute certainty) and 0 (absolute doubt). In any particular context c, a speaker S can only assert a proposition p if speaker S assigns to p a subjective probability threshold that is higher than cT . In other words, a speaker S can only assert p if the degree to which S believes p, or is certain that p is above the ‘certainty threshold’ for that context. The meaning of an evidential can be characterized in terms of the effects it has on the subsequent context. For example, when a speaker asserts ev(p) in context c, the speaker is claiming they have inferential evidence for p. The subjective probability threshold cT for c is ‘shifted’, so that it becomes the probability cT that a proposition with inferential evidence is true. The proposition p is asserted in the new context (c+ ev(p)), which has the new probability threshold cT .  3.4.3  Evidentials as Sentential-Force Specifiers  Portner (2006) develops an alternative to the classical speech act theory analysis presented in Faller (2002) in accounting for the evidential meanings of the individual evidentials in Quechua. Under Portner’s analysis, the Quechua evidentials are not really illocutionary modifiers, but rather ‘sentential force specifiers’: they are grammaticized elements which specify precisely which kind of conversational update is to be performed. This is based on Faller’s insight, where there is a type of speech act with fewer commitments than assertion described as putting (cf. Faller’s ‘presenting’) which can be modified in various ways to represent the different kinds of evidentials meanings in a language (Faller 2002; 2003). This is an adaptation of the theory of dynamic semantics, where the meaning of a sentence, or its context change potential, is the instructions it gives as to how to update the common ground, the aspects of knowledge, or set of propositions, that participants in a conversation assume to share (Heim 1990). Portner introduces a modification to this theory: first is that the common ground is only one of potentially numerous sets  108  3.5. Determining Levels of Meaning: The Tests of propositions. These other sets represent propositions that encode other categories, such as the different kinds of evidentiality. Evidentials under this approach are sentential force operators that specify a kind of conversational update: they can function to conversationally update the common ground, the set of mutually believed propositions, or other specialized sets of propositions. I take up a dynamic semantics of analysis of ’ nakw in chapter 5, where Portner’s innovations are presented in more detail.  3.5  Determining Levels of Meaning: The Tests  In the previous subsections we looked at a modal (propositional) and several non-modal (illocutionary) analyses of evidentials. We do not expect a particular type of meaning (e.g. an evidential meaning) to be constrained either only to the propositional level or to the illocutionary level (Faller 2002; 2003). However, modal and a non-modal analyses of evidentiality make different predictions about the semantic characteristics of an evidential. These centre on the kinds of characteristics we associate with any propositional operator. For example, we expect a propositional operator to affect the truth conditions of a sentence, and it should be able to fall within the scope of other operators such as if. Conversely, evidentials which are not propositional operators should do neither. In chapter 1 I introduced the set of diagnostics which can aid us in determining whether a given evidential marker operates at the illocutionary or propositional level. These tests were collected from different lines of research investigating the semantics of epistemic modality, and its connection to evidentiality (Lyons 1977; Papafragou 2000; 2006; Garrett 2001; Faller 2002; 2003; Matthewson et al. 2004; Waldie et al. 2009). These tests can be classified into two categories: tests which are sensitive to the truth value of the prejacent in (3.28), and tests which are sensitive to scope and embeddability in (3.29):  109  3.5. Determining Levels of Meaning: The Tests (3.28) Tests involving Truth Values (i.) Known Truth/Falsity: Is the sentence felicitous if the prejacent is known to be true or false? (ii.) Assent/Dissent: Can the contribution of the evidential be agreed or disagreed with? (iii.) Cancellability of type of evidence requirement: Can the evidence type requirement be cancelled? (3.29) Tests involving Scope and Embedability (i.) Embeddability: Can the evidential be understood as part of the propositional content of an embedded clause (i.e. the antecedent of a conditional, under a factive attitude verb, under a verb of saying)? (ii.) Scope with respect to interrogatives: Can the evidential take scope over a speech act? (iii.) Interaction with negation: Is the evidence type requirement affected by negation? There are additional considerations with regards to some of these tests. For example, in a particular language it may not be possible to use a certain evidential in an embedded clause for independent syntactic reasons (see the discussion of Nuu-chah-nulth in Waldie et al. 2009). Thus, if an evidential can be embedded, it is a propositional operator. But if it can’t be embedded, we can’t necessarily conclude that the evidential is an illocutionary operator. Additionally, following Matthewson et al. (2004), Waldie et al. show that both a modal and non-modal analysis make the same predictions with regards to interaction with negation, and with regards to cancellability of evidence. Keeping these considerations in mind, the tests in (3.28) and (3.29) provide a methodological foundation for determining whether the evidentials in a language should be formally treated as propositional or illocutionary operators which correspond to a modal or 110  3.5. Determining Levels of Meaning: The Tests non-modal analysis of evidentials respectively. In this section I work through in detail the definition of these tests as they were applied in St’át’imcets and Quechua, and then I evaluate their applicability to the Gitksan data.  3.5.1  Tests Regarding Truth  3.5.1.1  Known Truth/Falsity  We predict certain effects when a speaker modulates their assertion using a modal or evidential when they already know the prejacent p to be true or false. This is captured by the test in (3.30): (3.30) The Known Truth/Falsity Test: If the use of the evidential is felicitous when the speaker knows the prejacent is true or false, the evidential cannot be a modal. In Quechua, the reportative evidential -si modifies the sincerity condition of an utterance to one of ‘presenting’ rather than ‘asserting’. This was illustrated in example (3.26), repeated here as (3.31): (3.31) para-sha-n-si rain-prog-3-si “It’s raining.” ill = presenting sinc = The speaker heard from someone that p This illocutionary analysis of -si predicts that the truth value of p is opaque to the reportative evidential. Thus, it is possible to utter a sentence of the form p-si even when p is known by the speaker to be false, as in (3.32) and (3.33): (3.32) para-sha-n-si ichaqa mana creinichu rain-prog-3-si but not I.believe “[I heard] It’s raining, but I don’t believe it.”  (Faller 2002: 160)  111  3.5. Determining Levels of Meaning: The Tests (3.33) Pay-kuna-s ñoqa-man-qa qulqi-ta muntu-ntin-pi saqiy-wa-n, (s)-he-pl-rep I-illa-top money-acc lot-incl-loc leave-1O-3 mana-má riki riku-sqa-yui ni un sol-ta centavo-ta-pis not-surp right see-PP-2 not one sol-acc cent-acc-add saqi-sha-wa-n-chu leave-prog-1O-3-neg “They [reportedly] left me a lot of money, but, as you have seen, they didn’t leave me one sol, not one cent.” (Faller 2002: 191) The opposite effect obtains under a modal analysis: a speaker is asserting that a sentence S is either possibly or necessarily true (given the evidence). This effect is observable in the infelicity of (3.34), where the modal claim is inconsistent with the falsity of the embedded proposition: (3.34) # It may be raining, but I don’t believe it. # It’s not raining, but it may be. Thus, a modal analysis of evidentials predicts the same outcome: they are infelicitous when the prejacent is known to be false.23 Matthewson et al. show that this makes the correct prediction for the set of evidentials in St’át’imcets, given in (3.35): (3.35)  a. Inferential k’a #wa7 k’a kwis, t’u7 aoz t’u7 k-wa-s kwis impf infer rain but neg just det-impf-3.poss rain “It may/must be raining, but it’s not raining.” b. Perceived evidence -an’ #wá7-as-an’ kwis, t’u7 aoz t’u7 k-wa-s kwis impf-3.conj-perc.evid rain but neg just det-impf-3.poss rain “It’s apparently raining, but it’s not raining.”  23  This observation only holds true with epistemic modals. Sentences such as (3.34) are felicitous with  root modals: “It should be raining, but it isn’t.”  112  3.5. Determining Levels of Meaning: The Tests c. Reportative ku7 #um’-en-tsal-itas ku7 i án’was-a xetspqíqen’kst táola, give-dir-1sg.obj-3pl.erg rep det.pl two-det hundred dollar t’u7 aoz kw s-7um’-en-tsál-itas ku stam’ but neg det nom-give-dir-1sg.obj-3pl.erg det what “They gave me $200 [I was told], but they didn’t give me anything.” A similar result obtains when the speaker knows the prejacent to be true.24 In Quechua, sentences of the form S-mi have the modified sincerity condition which requires that the speaker directly witness and believe the truth of S: (3.36) para-sha-n-mi rain-prog-3-mi p = “It’s raining.” ill = assertions p sinc = {Bel(s, p), ev = See(s, ep )} strength = +1  (Faller 2002: 164)  Again we see the same opacity effects regarding the truth value of the embedded proposition: under an illocutionary operator analysis, the assertion made by a sentence containing -mi is simply the proposition itself. This predicts that it should be possible to use an illocutionary evidential such as -mi when the embedded proposition is already known to be true. On the other hand, under a modal analysis, we expect a quality implicature in using an epistemic modal when the embedded proposition is known to be true, as in (3.37)(i.). This implicature can be cancelled, as in (ii.): (3.37) Context: Spoken on November 6th, 2008. (i.) ? Barack Obama must be the next president.  (S. Cable class h/o)  (ii.) Yes, Barack Obama must be the next president. In fact, we know he is! 24  However, there are certain cases involving logical inference where epistemic modals can be used  when the speaker knows the precajent is true; von Fintel and Gillies (2007) claim that this is an evidential use of an epistemic modal such as must. This is discussed in more detail in §3.6.1.1.  113  3.5. Determining Levels of Meaning: The Tests Example (3.37) is infelicitous (or at least very marginal) in this context because it is very unlikely that the speaker would be unaware on the day after the presidential election who the next president will be. This is typically analyzed as a quality implicature: a speaker is using a weaker claim when the stronger, non-modal claim would be felicitous (see also von Fintel and Gillies (2007), p. 38 for discussion). This is the same effect observed by Matthewson et al. in St’át’imcets: the use of the evidentials is infelicitous if the speaker knows that the embedded proposition is true, as in (3.38): (3.38)  a. Inferential k’a #ts’um’-qs-án’-as k’a kw s-Lémya7 kw s-Roger; suck-nose-dir-3.erg infer det nom-Lémya7 det nom-Roger ats’x-en-lhkán wi7 zam’ see-dir1sg.subj emph after.all “Lémya7 must’ve kissed Roger; actually, I saw it.” b. Perceived evidence -an’ #ts’um’-qs-án’-as-an’ kw s-Lémya7 kw s-Roger; suck-nose-dir-3.erg-perc.evid det nom-Lémya7 det nom-Roger ats’x-en-lhkán wi7 zam’ see-dir1sg.subj emph after.all “Lémya7 apparently kissed Roger; actually, I saw it.” c. Reportative ku7 #ts’um’-qs-án’-as ku7 kw s-Lémya7 kw s-Roger; suck-nose-dir-3.erg rep det nom-Lémya7 det nom-Roger ats’x-en-lhkán wi7 zam’ see-dir1sg.subj emph after.all “[I was told] Lémya7 kissed Roger; actually, I saw it.”  3.5.1.2  Assent/Dissent  Assuming that meaning which is at the illocutionary level is not truth conditional, we expect that it cannot be targeted for assent or dissent (Papafragou 2000; 2006; Faller 114  3.5. Determining Levels of Meaning: The Tests 2002; to appear; Matthewson et al. 2004). This is formulated in the following test: (3.39) The Assent/Dissent Test: One cannot disagree with the content contributed by an illocutionary operator because a speech act does not have a truth value. We saw this effect with English illocutionary adverbs in (3.22): if a morpheme’s function is to modify the illocutionary force of an utterance, then one cannot object to the contribution of a speech act operator by saying something like ‘That’s false’. If modals are propositional operators, we expect their content to be agreed or disagreed with: In example (3.40), B’s reply does not deny the prejacent (that Jo is the thief). Instead, B denies the modal claim that Jo must be the thief: (3.40)  A: Jo must be the thief. B: That’s not true. There are some other plausible suspects. Jo may be entirely innocent. (Matthewson et al. 2004, adapted from a similar example in Faller 2002, p. 113)  In order to apply this test we need to ensure that the assent or dissent take the form of an explicit agreement with, or denial of, the truth of the relevant aspect of meaning. This is motivated by the fact that there are always strategies for challenging or rejecting different levels of meaning, whether presupposition or sincerity conditions of speech acts. Thus, only the requirement that the challenge take the form of (the relevant language’s equivalent of) “That is (not) true” ensures that the test distinguishes presuppositional material from material which contributes to the truth conditions of the utterance. This can be seen in example (3.41), where B” challenges the propositional content of A’s assertion:  115  3.5. Determining Levels of Meaning: The Tests (3.41)  A: Harriet likes the sociolinguistics professor. B: What sociolinguistics professor? I didn’t know we had one! B’: ?? That’s not true. We don’t have a sociolinguistics professor. B”: That’s not true. She hates him.  . Matthewson et al. (2004) demonstrate this with the evidentials in St’át’imcets, which do allow dissent with the modal claim, using the explicit “That is (not) true” response in (3.42)b. to the claim in a.: (3.42) Inferential k’a Context: A is driving past John’s house with B and sees John’s lights are on. a. wá7 k’a l-ta tsítcw-s-a s-John; tákem i be infer in-det house-3.poss-exis nom-John all det.pl sts’ák’w-s-a wa7 s-gwel light-3.poss-exis impf stat-burn “John must be home; all his lights are on.” b. aoz kw-a-s wenácw; papt wa7 lháp-en-as neg det-impf-3.poss true always impf forget-dir-3.erg kw-a-s lháp-an’-as i sts’ák’w-s-a det-impf-3.poss put.out-dir-3.erg det.pl light-3.poss-exis lh-as úts’qa7 when-3.conj go.out “That’s not true. He always forgets to turn his lights off when he goes out.” B’s statement = “John is not home.” B’s statement = “It’s not true that John must be home.” Matthewson et al. (2004) show the same effect with the perceived evidence evidential -an’ in (3.43):  116  3.5. Determining Levels of Meaning: The Tests (3.43) Perceived evidence -an’ Context: A is driving past John’s house with B and sees John’s lights are on. a. wá7-as-an’ l-ta tsítcw-s-a s-John; tákem be-3.conj-perc.evid in-det house-3.poss-exis nom-John all i sts’ák’w-s-a wa7 s-gwel det.pl light-3.poss-exis impf stat-burn “Looks like John is home; all his lights are on.” b. Same answer as in (3.42)b. B’s statement = “John is not home.” B’s statement = “It’s not true that John must be home.” Faller’s work does not contain the right kind of example to show that Quechua evidentials pattern as speech act operators for this test. Faller does give the following example in (3.44), but as Matthewson et al. point out it illustrates disagreement with the evidence type rather than dissent with the prejacent itself, or a hypothesized modal claim: (3.44)  a. Inés-qa qaynunchaw ñaña-n-ta-s watuku-sqa ines-top yesterday sister-3-acc-rep visit-pst.2 p = “Inés visited her sister yesterday.’ Ev = speaker was told that p b. Mana-n chiqaq-chu. # Mana-n chay-ta willa-rqa-sunki-chu not-bpg true-neg not-bpg this-acc tell-pst.1-3s2o-neg “That’s not true. You were not told this.”  Both a modal and speech act analyses predict Faller’s data, i.e. that you cannot say ‘That’s not true. You didn’t hear this.’25 Only a modal evidential would predict the kinds of data given in examples (3.42) and (3.43) because only a modal evidential gives a modal claim which can then be denied. 25  Although presuppositions are challengeable, you don’t usually challenge them using ’That’s not  true’.  117  3.5. Determining Levels of Meaning: The Tests 3.5.1.3  Cancellability of Evidence Type Requirement  The literature contains one other test involving the truth value of the prejacent, called the Cancellability Test, given in (3.45): (3.45) The Cancellability Test: Can the evidence type requirement be cancelled? However, this test is not very useful for distinguishing between the modal analysis and the speech operator analysis, because both analyses predict that the evidence type requirement cannot be cancelled, albeit for different reasons (see also Waldie et al. 2009). In the modal analysis, the evidence type requirement is a presupposition, so it cannot be cancelled (Izvorski 1997). In the speech act operator analysis, the evidence type requirement cannot be cancelled either, because it is a sincerity condition (Faller 2002; to appear).  3.5.2  Embeddability and Scope  A propositional and an illocutionary analysis of evidentials make different predictions regarding embeddability and scope. 3.5.2.1  Embeddability  Another test for whether evidentials contribute to the truth conditions of a sentence involves embedding. The idea is that an element which can be semantically embedded within the antecedent of a conditional or under a factive attitude verb or a verb of saying must be contributing to the propositional content, and is therefore not an illocutionary operator. Likewise, if a morpheme contributes only to the illocutionary force of an utterance, if we put it in an indirect speech context, that morpheme cannot be understood as part of the propositional content of the indirectly described speech act. This test is formulated as (3.46):  118  3.5. Determining Levels of Meaning: The Tests (3.46) The Embeddability Test: An illocutionary operator cannot be understood as part of the propositional content of an embedded clause, but a modal can. However, in order to properly assess the facts surrounding embedding and its applicability as a test, it is necessary to examine more closely what embedding actually entails. In doing this we find that there is a need to distinguish between semantic embedding, and structural or (morpho-)syntactic embedding (i.e. subordination). Schenner (2010) characterizes this difference in (3.47): (3.47)  a. An expression is syntactically embedded if it occurs in a clause distinct from the root clause (i.e. in an adverbial, relative or complement clause). b. An expression is semantically embedded if it is interpreted in the scope of some other operator.  The pair of related sentences in example (3.48) illustrates the difference between syntactic and semantic embedding: (3.48)  a. If the dog barks, the postman might run away. b. The dog might bark. The postman might run away.  The conditional in example a. is a case of both syntactic and semantic embedding. However, the same message is conveyed in the sequence of sentences in b., but the embedded interpretation is provided by the discourse understanding – the semantics and pragmatics – and not the syntax. However, there are several classes of expressions that are semantically unembedded, even if they occur in syntactically embedded positions (cf. Potts 2005). Thus syntactic embedding does not entail semantic embedding. To see the difference, consider syntactically embedded appositives, parentheticals in English, as in the appositive relative clause who I met yesterday in example (3.49). Crucially, it is not interpreted in the scope of the belief operator thinks, even though it occurs in its complement clause, i.e. in its syntactic scope. 119  3.5. Determining Levels of Meaning: The Tests (3.49) Lara thinks that her mother, who I met last night, will enjoy the gardens. Neither is there an entailment in the other direction, i.e. semantic embedding does not entail syntactic embedding. This was shown in example (3.48), but another example of semantic subordination in the absence of syntactic embedding are cases of modal subordination (Roberts 1989; McCready and Ogata 2007), as in (3.50), where the second sentence is interpreted as semantically subordinate to the proposition expressed by the first sentence: (3.50) A thief might break into the house. He would take the silver. The English modal might can be semantically embedded, and indicates that it is the matrix subject making the inference, rather than the speaker, as in (3.51): (3.51) John said that he might’ve won = John said “I might’ve won!” We find these kinds of embedded environments within the antecedent of a conditional, or under a factive attitude verb or a verb of saying. If an evidential can embed within any of these environments, it must be contributing to the propositional content, and cannot be an illocutionary operator. Matthewson et al. (2004) have found that evidential modals in St’át’imcets can embed under a verb of saying, as in the examples in (3.52): (3.52) St’át’imcets a. tsut kw s-Lémya7 kw s-melyíh ku7 ta say det nom-Lémya7 det nom-marry rep det í7mats-s-a s-Rose grandchild-3.poss-exis nom-Rose “Lémya7 said that [she was told that] Rose’s grandchild got married.” [Lémya7 was told; Lémya7 did not witness it; ku7 relates to the report given to Lémya7] b. tsut say ta det  s-Lémya7 kw s-tup-un’-ás k’a s-Maria nom-Lémya7 det nom-punch-dir-3.erg infer nom-Maria sésq’wez’-s-a younger.sibling-3.poss-det 120  3.5. Determining Levels of Meaning: The Tests “Lémya7 said that Maria must have hit her younger brother.” Only Lémya7 in this context is making a deduction from evidence. c. tsut s-Lémya7 kw s-tup-un’-ás-an’ s-Maria say nom-Lémya7 det nom-punch-dir-3.erg-perc.evid nom-Maria ta sésq’wez’-s-a det younger.sibling-3.poss-det “Lémya7 said that Maria must have hit her younger brother.” Only Lémya7 in this context is making a deduction from perceived evidence. In these examples, the reported, inferential or perceived evidence is oriented towards the subject of the matrix clause, and not the speaker of the sentence. This is the crucial interpretation that indicates semantic embedding, and that evidentials in St’át’imcets are propositional operators and not illocutionary ones. Faller shows that the reportative -si can syntactically embed under a verb of saying. However, the relevant interpretation does not obtain: the report remains anchored to the speaker as in (i.) and (ii.), and not the subject of the matrix clause (iii.): (3.53) Marya ni-wa-rqa-n Pilar-(*si) chayamu-sqa-n-ta-s Marya say-1O-past1-3 Pilar arrive-pp-3-acc-si “Marya told me that Pilar arrived.” (i.) speaker was told by someone else that Marya told the speaker that Pilar arrived. (ii.) speaker was told by Marya that Pilar arrived. (iii.) = Marya was told that Pilar arrived. 3.5.2.2  (Faller 2002: 222)  Scope with Respect to Interrogatives  A general assumption is that pragmatic operators take wide scope over propositional operators. Thus, we expect a difference in behaviour between an illocutionary evidential and a modal evidential. A modal evidential should be within the scope of an interrogative operator, whereas a speech act operator could in principle take scope over the interrogative. This forms the test in (3.54): 121  3.5. Determining Levels of Meaning: The Tests (3.54) The Interrogative Scope Test: Epistemic modals cannot take scope over an illocutionary act, such as performing a request/asking a question. Faller (2002a) shows that the reportative -si can be used to ask a question on someone else’s behalf. The context in example (3.55) is that the investigator’s question to the mother-in-law is not heard, so the consultant repeats the question on the investigator’s behalf. (3.55) Imayna-ta-s ka-sha-nki how-acc-rep be-prog-2 “How are you?”  (Faller 2002: 251)  Matthewson et al. (2007) report the opposite effect in St’át’imcets, where the reportative ku7 in (3.56) cannot be used to indicate that the speaker is asking the question on someone else’s behalf: (3.56) St’át’imcets (Matthewson et al. 2007: 232) swat ku7 k-wa táns-ts-an who rep det-impf dance-caus-1sg.erg “Who did they say I was dancing with?” = “(She says) Who was I dancing with?” However, in St’át’imcets, the insertion of the inferential modal k’a into a question has a different effect, creating a non-interrogative utterance, roughly translatable using ‘I wonder...’, as (3.57) shows: (3.57) St’át’imcets a. lán=ha kwanen-s-as already=ynq take.redup-caus-3.erg ni=n-s-mets-cál=a det.abs=1sg.poss-nom=write-act=exis “Has she already got my letter?” 122  3.5. Determining Levels of Meaning: The Tests b. lan=as=há=k’a kwanen-s-as already=3.sbjn=ynq=infer take.redup-caus-3.erg ni=n-s-mets-cál=a det.abs=1sg.poss-nom=write-act=exis “I wonder if she’s already got my letter.” “I don’t know if she got my letter or not.” (Matthewson 2008) The semantics of evidential questions will be examined in detail in chapter 4. 3.5.2.3  Interaction with Negation  Another test involving scope concerns the interaction of evidentials with negation: (3.58) The Interaction with Negation Test: Can the evidence type requirement take narrow scope with respect to negation? As with the Cancellability Test, both analyses predict the same results. In the modal analysis, the evidence type requirement is a presupposition, and will therefore project through negation. In the speech act operator analysis, the evidence type requirement is a sincerity condition, which is not affected by negation either. Moreover, speech act operators should not be able to occur in the scope of negation in the first place. The theories therefore make very similar predictions for this test, and it is not very useful for our purposes.26  3.5.3  Interim Summary  In all of the tests, the predictions for a modal analysis of the evidentials in St’át’imcets are borne out, while the predictions of the illocutionary operator analysis hold for the Quechua examples presented above. However, two of the tests, The Cancellability Test 26  The asserted content of the modal, i.e., the quantifier, should in principle be able to take wide  or narrow scope with respect to negation. However, it is difficult to test this because of the variable quantificational force of the modal, as well as the fact that scopal interaction is often restricted for independent reasons. Variable modal force is discussed in detail in the following chapter, and also see Matthewson et al. (2007).  123  3.5. Determining Levels of Meaning: The Tests and the Interaction with Negation Test, do not distinguish between the two analyses. The results are summarized in Table 3.1 (adapted from Matthewson et al. 2007), and based on this, a theoretical typology can be organized as in Table 3.2 (cf. Waldie et al. 2009): St’át’imcets  CQ  Felicitous if p is known to be true or false?  No  Yes  Pass assent/dissent test?  Yes  No  Evidence type cancellable?  No  No  Semantically embeddable?  Yes  No  Able to scope outside interrogatives?  No  Yes  Evidence requirement affected by negation  No  No  Table 3.1: Test results for St’át’imcets and Quechua  Yes  No  Felicitous if p is known to be true or false?  Illocutionary  Propositional  Pass assent/dissent test?  Propositional  Illocutionary  Evidence type cancellable?  –  –  Semantically embeddable?  Propositional  Illocutionary  Able to scope outside interrogatives?  Illocutionary  Propositional  –  –  Evidence requirement affected by negation?  Table 3.2: A Propositional/Illocutionary typology based on the levels of meaning tests.  124  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? The next section applies these tests to the Gitksan data, where I show that not only do languages vary in what level of meaning evidentiality is expressed, but that the same language can possess both illocutionary and propositional (modal) evidentials.  3.6  Gitksan Evidentials: Propositional or Illocutionary Operators?  The previous section showed that evidential interpretations can arise on different levels of meaning. This section shows that the evidential system of Gitksan provides evidence that a propositional-illocutionary ‘split’ in the distribution of evidentials can occur within the same language. Specifically, by applying the tests in the previous section, in this section I claim that both modal =ima and reportative =kat are epistemic ’ modals, and that nakw is an illocutionary operator, as summarized in Table 3.3. Gloss  Evidence type  Level of Meaning  =kat  reportative (rep)  Reportative  Propositional  =ima  modal (mod)  Inferential  Propositional  ’ nakw  evidential (evid)  Inferential – Sensory  Illocutionary  Table 3.3: The Propositional vs. Illocutionary status of the evidential System in Gitksan  The outcome of these tests as applied to the Gitksan data will help us determine the appropriate modal and non-modal analysis of the individual evidentials. This is undertaken in chapters 4 and 5.  3.6.1  Tests Regarding Truth  Recall that the use of a propositional evidential when the embedded proposition is known to be true or false results in infelicity. A speaker avoids the use of an evidential if they  125  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? have direct evidence or knowledge, as the use of an evidential implicates the lack of knowledge. 3.6.1.1  Known Truth/Falsity  The claim that =ima is a modal predicts that a sentence containing =ima will be infelicitous if the proposition embedded under it is known to be true or false. The examples in (3.59) and (3.60) show that =ima cannot be used if the embedded proposition is known to be false (or extremely unlikely to be true), either based on some kind of prior relevant knowledge or experience: (3.59) Context: You wake up and see the sun shining on the bedroom wall. #yugwimahl dim wis yukw=ima=hl tim wis prog=mod=cnd fut rain “It might/must be raining.” (3.60) Context: It’s August. #yugwimahl dim maadim yukw=ima=hl tim maatim prog=mod=cnd fut snow “It might/must be snowing.” The same results obtain with the reportative =kat: the examples in (3.61) – (3.63) show the infelicity of =kat when the speaker knows that the embedded proposition is false:  126  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? (3.61) Context: You know John was at work yesterday. #sihongatit John k’yoots John k’yoots si-hon=kat=t caus-fish=rep=pnd John yesterday “[I heard] John canned fish yesterday.” Comments (paraphrased): ‘Why say you heard it from someone else when you know it’s not true yourself?’  (3.62) Context: You know that John always insists in sitting at a different position in the feast hall, contrary to the customary seating arrangement by wilp (house). #aluugat ’aluu=kat visible=rep “[I heard] He  t’aahl ’as ayuuk t’aa=hl ’ayuuk ’a=s sit=cnd traditional.law obl-pnd understands the (traditional) law.”  ’ nit ’nit 3sg  (3.63) Context: You hear from your best friend (a reliable source) that John won at bingo last night. You were at bingo last night, and you remember seeing his brother Bill win the jackpot instead. ’ yhl ’ #laxni xstat John ’ y=hl ’ laxnixsta=t John hear-1sg=cnd win=pnd John wilaay’ wil needii ’ wilaa-y wil nee=tii know-1sg comp neg=contr “I heard that John won at bingo was there too).”  bingo gaxxw ii go’ohl ’ap bingo kaxxw ii ’ap ko’=hl loc=cnd bingo last.night conj assert xstat xsta-t win-3sg last night, but I know he didn’t win (because I  The same infelicity results when the speaker knows that the embedded proposition is true. Neither =ima nor =kat can be used in this case. However, it should be pointed out that there are two different issues to consider here: (i) directly observing the event itself, which gives rise to infelicity, and (ii) knowing in some other way that the prejacent is 127  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? true/false through logical inference (von Fintel and Gillies 2007; von Fintel and Iatridou 2009), which does not give rise to infelicity. I will discuss each of these in turn, and how =ima and =kat behave in contexts that reflect these two issues. Directness. The first issue involves a speaker directly witnessing the event expressed by the prejacent, thereby believing it is true. In example (3.64) with =ima, and (3.65) with =kat, cannot be used: (3.64)  a. Context: Your friend is showing you how to cook something, and while watching them you see them accidentally cut themself. #gojinimahl ’onin ’on-n kots-i-n=ima=hl cut-tr-2=mod=cnd hand-2sg “You might’ve/must’ve cut your hand.” b. Context: You’re looking out the window during a storm. wis, #yugwimahl #yukw=ima=hl wis, prog=mod=cnd rain “It might be raining, and  ii gya’ay’ ii gya’-y’ conj see-1sg I see it (outside).”  wan asun, ii (3.65) #ye’egathl wan a-sun, ii ye’e=kat=hl walk=rep=cnd deer loc-here conj sbagaytgan spagaytgan forest “I heard a deer walked around here, and I  gya’ay’ loot ahl kya’-y’ loo-t a=hl see-1sg obl.pro-3sg loc=cnd  see it in the forest.”  Comment: “There’s no point saying someone else sees it if you do yourself."  Logical Inference. von Fintel and Gillies (2007) discuss a class of cases where the prejacent is known to be true based on logical inference, but the use of the modal must is still felicitous, as in (3.66): 128  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? (3.66)  The ball is in A or in B or in C. It is not in A. It is not in B. So, it must be in C.  They adduce cases such as these to support the claim that epistemic modals incorporate an evidential meaning component (von Fintel and Gillies 2007, p. 39). The reason ‘must’ is bad in the normal cases where we know p is true is because in these cases we have too direct evidence. In the logical inference case, we have indirect evidence, so the use of ‘must’ is felicitous. ’ However, neither =ima nor nakw are felicitous in cases involving logical inference such as (3.66): a speaker will always make the strongest, unmodalized claim, if they ’ know the truth of the prejacent. With nakw, this restriction follows from the fact that the context lacks sensory evidence. However, it is less clear why =ima is infelicitous. I do not have a concrete explanation for this, except to suggest that in non-sensory evidence contexts, such as those involving logical inference, =ima can also express a weak, might-like reading. This is less infelicitous in English, as the variation on (3.66) shows in (3.67): (3.67)  The ball is in A or in B or in C. It is not in A. It is not in B. ? So, it might be in C.  This strengthens the claim that =ima is an epistemic modal, but weakens somewhat the effectiveness of the Known Truth/Falsity Test, making it a one-way test: if an evidential can be used when the truth or falsity of the prejacent is known, then it may be either an illocutionary or modal evidential. If an evidential can’t be used, this is evidence that it is a modal. ’ Turning now to the behaviour of nakw with respect to the Known Truth/Falsity ’ test, in its normal use, nakw behaves the same as =ima and =kat: it is infelicitous if the speaker knows the embedded proposition is true, as in (3.68)(a.) or (b.): 129  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? (3.68)  a. Context: You’re looking out the window during a storm. ’ #nakwhl ’ nakw =hl evid=cnd “It must be  yukw dim wis yukw tim wis prog fut rain raining.”  b. Context: Upon pulling clothes off the line that have been in the sun all day. ’ #nakwhl ’nakw =hl evid=cnd “It must be  gwalkw kwalkw dry dry.”  Comment: ‘If you’re touching it and it’s dry, you would just say ‘it’s dry’.’  ’ However, nakw is also quite different than =ima or =kat. First, when a speaker ’ uses nakw knowing the embedded proposition is false, a non-literal/metaphorical use is triggered, rendering an expression similar to a must-type rhetorical question/statement ’ in English, as in (3.69)a. Conversely, when a speaker uses nakw knowing a proposition is true a mirative meaning is expressed, as in (3.69)b. Mirativity is the marking of a proposition that represents information which is new and possibly surprising to the speaker (DeLancey 1997, 2001): (3.69)  ’ a. nakwhl sinst ’ nakw =hl sins-t evid=cnd blind-3 “He must be blind!” “Is he blind or something?” “Looks like he’s blind!” Context A – Sensory evidence: You see a man walking down the street with a white cane. Context B – The proposition is known to be false (metaphorical): You’re watching a baseball game. The star batter on the speaker’s favourite team keeps missing the ball and striking out, jeopardizing the outcome of the game.  130  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? ’ b. nakwhl bagwdiit ’ nakw =hl pakw=tiit evid=cnd arrive.pl=3pl “They’re here!” “Looks like they made it!” Context A – Sensory evidence: You see a pickup in the driveway. Context B – The proposition is known to be true (mirative): You see your friends standing in the doorway.  In contrast, =ima in example (3.70) is also felicitous in the context in (3.69), but it cannot have this pragmatic effect: =ima must express that the batter is literally blind, or inferential evidence of the arrival of people: (3.70)  ’ a. sinsima nit ’ sins=ima nit blind=mod 3 “He might/must be blind.” (literal comment) b. bagwima bakw=ima arrive.pl=mod “They might be  ’ nidiit ’ ni-tiit det-3pl here.” (no surprise)  ’ In both its metaphorical and mirative uses, nakw maintains its evidential function: the speaker is making a type of assertion based on what they perceive as sensory evidence – in (3.69)a. the fact that the batter keeps missing the ball, and in (3.69)b. the fact that they can see people coming through their front door. It should be pointed out that the infelicity that arises when the prejacent is known to be false does not in and of itself show that =ima and =kat are modals, just as ’ nakw is also infelicitous in these contexts; we could expect the same infelicity if they were illocutionary operators, perhaps stemming from a conflict in the relevant kind ’ of evidence. However, we do see different behaviour emerge between =ima and nakw ’ when comparing the examples in (3.69) with those in (3.70): nakw is felicitous when the prejacent is known to be true or false, although it must express an extra meaning, but 131  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? neither =ima nor =kat can do this. This effect, and its bearing on the Truth/Falsity test, is examined in chapter 5. 3.6.1.2  Assent/Dissent  The application of the Assent/Dissent Test shows the propositional status of both =ima and =kat. For example, suppose someone looks out of their kitchen window in Kispiox and makes the following claim with =ima in the matrix clause of the conditional in (3.71): (3.71) ji Kispiox ii da yukwhl hoti wis go’ohl tsi Kispiox ii ta yukw=hl hoti wis ko’=hl irrealis cond prog=cnd rain loc=cnd Kispiox conj comp gitwangak yugwimahl wis go’ohl kitwangak yukw=ima=hl wis ko’=hl prog=mod=cnd rain loc=cnd Kitwanga “If it’s raining in Kispiox, then it might/must be raining in Kitwanga.” (cf. Faller 2002: 130-133) Recall the requirement that the challenge take the form of “That is (not) true” in order to ensure that the test distinguishes presuppositional or illocutionary material from material which contributes to the truth conditions of the utterance. In Gitksan a listener may agree with the modal claim as in (3.72)a., or disagree as in b. using =ima, or actively challenge it as in c., which is the Gitksan equivalent to a “That is (not) true” dissent, or “I don’t think you’re right”: (3.72)  ’ a. nidima ’ nit=ima 3sg=mod “Maybe.” Comment: True, it’s possibly raining because those are the usual weather patterns.  132  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? b. neeyima nee=ima neg=mod “Maybe not.” Comment: You don’t really know for sure - I was there once, and while it was raining in Kispiox it wasn’t raining in Kitwanga.  c. neediihl ha’nigoody’ ji nee=tii=hl ha’nigood-y’ tsi neg=contr=cnd think-1sg irr “I don’t think you’re right.” = “It’s not true that it must/might be = “It’s not raining in Kitwanga.”  hugwaxn hugwax-n correct-2sg raining in Kitwanga.”  (3.72)c. is an explicit dissent: you can’t assume that just because it’s raining in Kispiox, that it must be raining in Kitwanga, regardless of your evidence. Further evidence that =ima passes the Assent/Dissent test comes from a test applied to epistemic modals in English and the modal evidentials in St’át’imcets (Matthewson et al. 2004). von Fintel and Gillies (2007) suggest that sentences containing epistemic modals, such as (3.73), perform two speech acts simultaneously: (3.73) There might’ve been a mistake in the calculation. The first speech act involves the assertion that it is compatible with the evidence that there has been a mistake, and the second involves suggesting (with a lack of conviction) that there has been a mistake, or advice not to overlook the possibility that there has been a mistake. This allows hearers to respond to an epistemic modal claim by targeting either the epistemic claim or the embedded proposition. They show this using the following scenario in (3.74) (von Fintel and Gillies 2007; von Fintel and Iatridou 2009, adapted from an example in von Fintel 2005. See also Matthewson et al. 2004, p. 223–225 for an application of this test in St’át’imcets):  133  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? (3.74) Context: Pascal and Mordecai are playing Mastermind. After some rounds where Mordecai gives Pascal some hints about the solution, Pascal says: ‘There might be some reds.’ Mordecai, knowing the solution, has a range of possible responses: (i.) That’s right. There might be. (ii.) That’s right. There are. (iii.) That’s wrong. There can’t be. (iv.) That’s wrong. There aren’t. In the response using the modal in (3.74)(i.) Mordecai assents to the modal claim that Pascal’s evidence allows that there are reds. In the response in (iii.), Mordecai denies the modal claim that Pascal’s evidence allows that there are reds. In other words, these two modal responses agree with or deny the modal claim, not the prejacent. Matthewson et al. (2004) applied the same test in St’át’imcets which confirms a modal analysis of the evidentials in that language. We find the same results with =ima when the Mastermind scenario in (3.74) is reproduced in Gitksan in (3.75):27 (3.75) neeyimahl nee=ima=hl neg=mod=cnd “Is it maybe red?”  27  ihleetxwda ihleetxwt=a be.red=interrog (lit. “Is it not possibly red?”)  (i.) ee’e, ee’e, yes, “Yes,  ’ nidima ’ nit=ima 3sg=mod maybe.”  (ii.) ee’e, ee’e, yes, “Yes,  ihleetxwt ihleetxwt be.red it is red.”  The context in (3.75) is adjusted slightly from (3.74): the first speaker in (3.75) is simply guessing  the colour of a particular peg.  134  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? (iii.) neediihl hugwaxit, nee=tii=hl hugwax-t, neg=contr=cnd correct=3sg, “It’s not correct. It is not possibly  neeyimahl ihleetxwt nee=ima=hl ihleetxwt neg=mod=cnd be.red red.”  neediihl ihleetxwt (iv.) neediihl hugwaxit, hugwax-t, nee=tii=hl ihleetxwt nee=tii=hl neg=contr=cnd correct=3sg, neg=contrast=cnd be.red “It’s not correct. It is not red.” There are two considerations when examining the application of the assent/dissent ’ ’ test to nakw : first, statements involving any kind of negation and nakw are judged by speakers to sound odd and unnatural. Secondly, in example (3.76), a speaker is making ’ a nakw statement based on the visual and auditory evidence of someone sneezing. While ’ not technically ungrammatical, the response in (3.76)a. with nakw cannot be used to assent to the claim in (3.76). A stronger effect is observed in another response in (3.76)b. using negation. The negative response used in (3.76)c. cannot be used to dissent from ’ the meaning of nakw : ’ (3.76) nakw=hl siipxw-t evid=cnd sick-3sg “He must be sick.” ’ a. # ee’e, nakw=hl ap wil-t No, evid=cnd ? do.something-3sg = “Yes, this must be what’s happening.” (I agree because his face is all red.) ’ b. *nee=tii nakw=hl siipxw-t neg=contr evid=cnd sick-3sg = “No, he can’t be sick.” (I saw him at work today and he looked fine.) c. # nee=tii=hl ha’nigood-y’ tsi hugwax-n irr correct-2sg neg=contr=cnd think-1sg = “I don’t think you’re right.” (cf. (3.72)c.) ’ It will be shown in the next subsection that nakw cannot be syntactically embedded, ’ which accounts for the ungrammaticality of (3.76)b., where nakw is embedded under 135  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? ’ the negation predicate. However, in (3.76)a. I assume that nakw is not syntactically embedded as ee’e is a general expression of agreement or positive response in conversation, and not a part of the syntax of the clause it precedes. Nonetheless, cannot be used to assent to (3.76). Recall from §3.5.1.2 that in order to apply the Assent/Dissent test we need to ensure that the dissent take the form of an explicit denial of the truth of the relevant aspect of meaning. Thus, the challenge must take the form of “That is (not) true” to ensure that the test distinguishes presuppositional material from material which contributes to the truth conditions of the utterance (cf. (3.41)). We see this in (3.76)c., which is the relevant form in Gitksan for denying the truth of a statement, and how it cannot be used to deny (3.76) (cf (3.72)c.). 3.6.1.3  Cancellability of Evidence Type Requirement  §3.5.1.3 discussed the inapplicability of the Cancellability test as a way of distinguishing a modal from an illocutionary. However, it is still important to test whether the predictions common to both theories are upheld. Under a modal analysis of evidentials, the evidential meaning is encoded as a presupposition and not implicature. Thus, we expect that the evidential meaning it encodes cannot be cancelled or negated. The Gitksan data confirms this in examples 3.64 and 3.65, repeated here: (3.77) #yugwimahl #yukw=ima=hl prog=mod=cnd “It might be raining,  wis, ii wis, ii rain conj and I see it  gya’ay’ gya’-y’ see-1sg (outside).”  (3.78) #ye’egathl wan asun, ii wan a-sun, ii ye’e=kat=hl walk=rep=cnd deer loc-here conj spagaytgan spagaytgan forest “I heard a deer walked around here, and I  gya’ay’ loot ahl kya’-y’ loo-t a=hl see-1sg obl.pro-3sg loc=cnd  see it in the forest.” 136  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? The inapplicability of the cancellability test is for theoretical reasons, and not a matter for empirical confirmation. The predictions of both a modal and illocutionary analysis cannot distinguish them in principle.  3.6.2  Scope and Embeddability  Because speech act operators do not contribute to the propositional content they should, in principle, not be embeddable under expressions that take propositions as their arguments, such as propositional attitude verbs and conditional operators like ‘if’. Modals on the other hand are propositional and therefore should be embeddable. 3.6.2.1  Embeddability  In investigating the embeddability of evidentials, it is necessary to determine whether they can occur syntactically and/or semantically embedded in complement, adverbial and/or relative clauses. The difference we are looking for in determining semantic scope is the orientation of the modal/evidential claim to either to the subject of the matrix clause, or to the speaker of the sentence. Both modal =ima and reportative =kat can be both syntactically and semantically embedded. In (3.79)a., =kat attaches to the matrix clause, and the speaker is reporting that she heard about Mark telling his sister that he would leave for the coast. The reported evidence is oriented towards the speaker: they are presupposing that the report was reliable and at least possibly true. However, in (3.79)b. =kat is attached to the verb in the complement clause. The reportative evidence is now re-oriented to the matrix subject, Mark, and not crucially not to the speaker: it is Mark who has reported evidence that John will leave for the coast. The speaker is asserting this because they witnessed this event first-hand:  137  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? (3.79)  a. Report: The speaker is reporting that Mark told his sister that John would leave for the coast. Context: You heard from Mark’s co-worker that John was going to be away for the weekend, and the co-worker overheard Mark talking to his sister on the phone about John going to the coast. mahldigas Mark ’ahl gimxdit Mark ’a=hl kimxt-t mahl-T-i-(t)=kat=s tell-t-tr-3=rep=pnd Mark obl=cnd sister-3 ’ daawhlt John go’ohl laxmo’on ’ lax-mo’n taawhl=t John ko’=hl leave=pnd John loc=cnd geo.loc-coast “Mark told/said to his sister that John would leave  dim wil saa tim wil saa fut comp away  for the coast.”  b. Report: Mark told his sister that he heard that John would leave for the coast. Context: You had lunch with Mark. While at lunch his sister came up and Mark told her that he heard John would leave for the coast. mahldis Mark ’ahl gimxdit dim wil saa mahl-T-i-(t)=s Mark ’a=hl gimxt-t tim wil saa tell-t-tr-3-pnd Mark obl=cnd sister-3 fut comp away laxmo’on John go’ohl daa’whltkatit taa’whl=t=kat=t John ko’=hl lax-mu’n leave=3=rep=pnd John loc=cnd geo.loc-coast “Mark told/said to his sister that he would leave for the coast.” The same results obtain with =ima when it is attached to the matrix verb, mahl : the inferential evidence is now oriented towards the speaker, and not Granny. In (3.80) you were learning how to can berries with Granny, and in this context you can infer from the fact you had this learning experience, that it’s possible Granny told you that a certain berry will taste better once it’s left until autumn:  138  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? (3.80) Context: You learned from your aunt how to can berries last autumn. Several people were also there, including Granny, who also has experience in canning berries. mahliyimas nits’iits’ looy’ mahl-i-(t)=ima=s nits’iits’ loo-y’ say-tr-3sg=mod=pnd grandmother obl.pro-1sg xwsit xwsit autumn “Granny might’ve told me that it will taste better in  ’ ji dim ixsta hla ’ tim ixsta tsi hla fut taste irr incept  the autumn.”  By contrast, when =ima is embedded in the complement of a verb, it has the same effect as it does with =kat: the evidence is related to the matrix subject, and not to the speaker. In example (3.81), a speaker is asserting that Granny has inferential evidence, based on her experience in canning berries, that the berries might taste good in the autumn: (3.81) Context: You’re learning how to can berries, and you’re telling a friend that Granny suggested that the particular berry you were canning might taste better the longer it’s left to sit, maybe by the autumn. ’ mahlis nits’iits’ looy’ dim ixstayima ji hla ’ mahl-i-(t)=s nits’iits’ loo-y’ tim ixsta=ima tsi hla say-tr-3sg=pnd grandmother obl.pro-1sg fut taste=mod irr incept xwsit xwsit autumn “Granny told me it might taste better in the autumn.” Similar minimal pairs of this alternation are given in examples (3.82) and (3.83): in (3.82)a., the speaker is talking about who scored the winning goal, even though she wasn’t there herself. She is inferring from the fact that Louise was the only one at the soccer game that she might or must be the source of this information. In (3.82)b., the 139  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? speaker is simply asserting that Louise has inferential evidence that John assisted Tony in scoring the winning goal by kicking the ball to him. Louise’s claim is based on the evidence that John is the striker on the team, and this is typically the job of the striker in soccer: (3.82)  a. Context: You’re talking with your friends about the soccer game that morning. You weren’t there yourself, but you know that Tony scored the winning goal assisted by John’s kick. But you can’t remember how you know this. You likely heard it from Louise earlier, who was there because her cousin was playing. mahliyimas Louise looy’ mahl-i-(t)=ima=s Louise loo-y’ say-tr-3sg=mod=pnd Louise obl.pro-1sg John-hl hlit ’as Tony John=hl hlit ’a=s Tony John=cnd ball obl=pnd Tony “Louise might’ve/must’ve told me John kicked  wilt hlo’oxsis wil=t hlo’oxs-(t)=s comp=3sg kick-3sg=cnd  the ball to Tony.”  b. Context: You’re talking with your friends about the soccer game that morning. You weren’t there yourself, but you were talking earlier with Louise, who was there. Louise knew that Tony made the winning goal, but she wasn’t sure if he was assisted by John – who is the striker on the team – or another player. mahlis Louise looy’ wilt mahl-i-(t)=s Louise loo-y’ wil=t say-tr-3sg=pnd Louise obl.pro-1sg comp=3sg hlo’oxsiyimas John-hl hlit ’as hlo’oxs-i-(t)=ima=s John=hl hlit ’a=s kick-tr-3sg=mod=cnd John=cnd ball obl=pnd “Louise told me John might’ve/must’ve kicked the ball  Tony Tony Tony to Tony.”  The embedded evidentials in the examples above involve inference from experience and general knowledge: Granny’s experience in canning berries, and general knowledge 140  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? about the role of a striker in a soccer game (the assist in goal scoring). In the next minimal pair in (3.83), the embedded evidential involves inference from observation: in b., Barbara infers from the fact that there are big footprints around the car that Tyler (who has big feet) pushed the car there after it broke down: (3.83)  a. Context: You’re talking about a mutual friend’s car that you know broke down. It was sitting at Barbara’s place, so she might’ve been the one who told you how it got there. mahliyimas Barbara looy’ wilt mahl-i-(t)=ima=s Barbara loo-y’ wil=t say-tr-3sg=mod=pnd Barbara obl.pro-1sg comp=3sg ’ tisimas Tylerhl kartxw ’tis-(t)=ima=s Tyler=hl kartxw Tyler=cnd car push-3=mod=cnd “Barbara might’ve told me Tyler pushed the car.” b. Context: You’re telling a mechanic friend that Barbara’s car is sitting in the yard. Barbara thinks Tyler pushed it because of the big footprints in the mud around the car. mahlis Barbara looy’ wilt mahl-i-(t)=s Barbara loo-y’ wil=t say-tr-3sg=pnd Barbara obl.pro-1sg comp=3sg Tylerhl kartxw Tyler=hl kartxw Tyler=cnd car “Barbara told me Tyler might’ve pushed the car.”  ’ tisimas ’ tis-(t)=ima=s push-3=mod=cnd  Finally, in (3.84)a., Gwen is expressing the epistemic claim that the dog might bite her. In (3.84)b, the matrix verb is the predicate xpts’axw ‘to be afraid’, which takes the same complement clause marked by =ima. As with the verb of saying, it is Gwen herself who is expressing the possibility that the dog might bite her:  141  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? (3.84) Context: Gwen keeps complaining about the dogs in front of Fern’s house. She has mentioned her fear of being bitten before, and even though the dogs haven’t bitten her, it’s possible they might given their aggressive nature. a. mahlis Gwen wilt hats’imahl ’os mahl-i-(t)=s Gwen wil=t hats’-i-(t)=ima=hl ’os say=pnd Gwen comp=3sg bite-tr-3sg=mod=cnd dog “Gwen said the dog might bite her.” Gwen hoti b. simxpts’axwt Gwen hoti sim-xpts’axw=t true-be.afraid=pnd Gwen comp “Gwen’s really afraid the dog might  hats’imahl ’os hats’-i-(t)=ima=hl ’os bite-tr-3sg=mod=cnd dog bite her.”  ’ The embedding facts of nakw are markedly different from the enclitics =kat and ’ =ima. This is mostly rooted in the status of nakw as an intransitive verb, which takes ’ as its argument a nominalized verb. This places nakw in the same syntactic class of nominalizing intransitive verbs as the progressive yukw, and the imperfective hliskw (see ’ also Tarpent 1987, p. 350, who describes nakw as an auxiliary verb, along with yukw and hliskw ). These constructions are compared in (3.85): (3.85)  a. yukwhl gahahlal’stdiithl haanak yukw =hl haanak kahahlal’st-tiit=hl prog=cnd redup.pl-work-3pl=cnd women.pl “The women are working.” b. hliskwhl gahahlal’stdiithl haanak hliskw =hl haanak kahahlal’st-tiit=hl imperf=cnd redup.pl-work-3pl=cnd women.pl “The women finished working.” ’ c. nakwhl gahahlal’stdiithl haanak ’ nakw =hl haanak kahahlal’st-tiit=hl evid=cnd redup.pl-work-3pl=cnd women.pl “The women must be working.”  142  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? ’ In probing further the syntactic features of nakw, consider a typical intransitive verb construction example (3.86). These are the simplest constructions in Gitksan, and minimally consist of a verbal predicate and its DP argument.28 The DP is marked with either a common noun determiner =hl (cnd), or a proper noun determiner =t/=s (pnd), which encliticize to the verb: (3.86)  a. [V]pred [=det NP]arg haanak b. gahahlal’sthl [kahahlal’st]pred [=hl haanak]arg redup.pl.work=cnd women.pl “The women work.”  When intransitive predicates such as yukw and hliskw in (3.85) are applied to an intransitive (or transitive) sentence such as (3.86), they nominalize the intransitive predicate, appearing before it and marking it with the common noun determiner =hl. This is sketched out in (3.87): (3.87)  a. yukwhl [yukw]pred [=hl prog=cnd “The women are  gahahlal’stdiithl haanak haanak]arg kahahlal’st-tiit=hl women.pl redup.pl-work-3pl=cnd working.”  b. hliskwhl gahahlal’stdiithl haanak haanak]arg [hliskw]pred [=hl kahahlal’st-tiit=hl imperf.=cnd redup.pl-work-3pl=cnd women.pl “The women finished working.” ’ c. nakwhl gahahlal’stdiithl haanak ’ [nakw ]pred [=hl kahahlal’st-tiit=hl haanak]arg evid=cnd redup.pl-work-3pl=cnd women.pl “The women must be working.” ’ These data show so far that nakw behaves syntactically like an auxiliary verb in ’ Gitksan. However, there are several unique features that set nakw apart from the 28  This is a preliminary syntactic analysis; more investigation is required.  143  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? ’ other auxiliary verbs in (3.87). First, unlike the predicates yukw and hliskw, nakw cannot appear under negation, which occupies the first position in a clause, as shown in ’ (3.88)c. Example d. shows that placing nakw before negation also does not rescue its grammaticality. (3.88)  a. needii haanak yukwhl gahahlal’stdiithl nee=tii haanak yukw=hl kahahlal’st-tiit=hl neg=contr prog=cnd redup.pl-work-3pl=cnd woman.pl “The women are not working.” b. needii hliskwhl gahahlal’stdiithl haanak haanak nee=tii hliskw=hl kahahlal’st-tiit=hl neg=contr imperf.=cnd redup.pl-work-3pl=cnd woman.pl “The women are not finished working.” ’ c. *needii nakwhl gahahlal’stdiithl haanak ’ nee=tii nakw =hl haanak] kahahlal’st-tiit=hl neg=contr evid=cnd redup.pl-work-3pl=cnd woman.pl “The women must not be working.” ’ d. *nakwhl needii gahahlal’stdiithl haanak ’ haanak] nakw =hl nee=tii kahahlal’st-tiit=hl evid=cnd neg=contr redup.pl-work-3pl=cnd woman.pl “The women must not be working.”  ’ Secondly, nakw cannot appear in a conditional. Example (3.89)a. shows the basic structure of a conditional in Gitksan, and the embeddability of the progressive predicate yukw. (3.89)b. shows that the modal =ima can also be embedded in the consequent, ’ while (3.89)c. shows that nakw cannot appear in this same embedded position: (3.89)  a. ji da yukwhl wis go’ohl ansbayaxw ii hodi tsi da yukw=hl wis ko’=hl ansbayaxw ii hoti irr cond prog=cnd rain loc=cnd Kispiox conj comp gitwangak yukwhl wis go’ohl yukw=hl wis ko’=hl kitwangak prog=cnd rain loc=cnd Kitwanga “If it’s raining in Kispiox, then it’s raining in Kitwanga.” 144  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? b. ji ansbayaxw ii da yukwhl hodi wis go’ohl ji anspayaxw ii da yukw=hl hoti wis ko’=hl irr cond prog=cnd rain loc=cnd Kispiox conj comp gitwangak yugwimahl wis go’ohl kitwangak yukw=ima=hl wis ko’=hl prog=mod=cnd rain loc=cnd Kitwanga “If it’s raining in Kispiox, then it might/must be raining in Kitwanga.” ’ c. *... ii hoti nakw yukw=hl wis ko’=hl kitwangak ... conj comp evid prog=cnd rain loc=cnd Kitwanga “If it’s raining in Kispiox, then it must be raining in Kitwanga.” ’ If nakw belongs to the same morphosyntactic class of items that can both embed ’ under negation and the conditional, then we would expect nakw to occur in these po’ sitions as well. However, the examples above show that nakw cannot be syntactically ’ embedded. Given these facts, I claim that the reason why nakw cannot be syntactically embedded is because is cannot be semantically embedded. This suggests the ’ illocutionary status of nakw. ’ Table 3.4 summarizes the embedding results of =ima, =kat, and nakw. Syntactic  Semantic  –  –  =kat =ima ’ nakw  ’ Table 3.4: The embeddability of =ima, =kat, and nakw.  3.6.2.2  Scope with Respect to Interrogatives  Both =ima and the reportative =kat have a productive use in questions. When =kat is used in a question, a speaker is not reporting a question, but is asking the addressee what she knows about something on the basis of reported evidence. In other words, a speaker asking a question with a reportative is targeting an answer that the addressee 145  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? may know, or may only have reportative evidence for. This can be observed in example (3.90), where a speaker is enquiring about when the bus will arrive in Prince George. By using =kat in the question, the speaker implies that the answer to this question, given in c., is going to be second hand, since they know their companion is not the one who determines the bus schedule: (3.90) Context: You and a friend are taking the overnight bus to Prince George. You can’t remember what time you arrive, but your friend who was the one who booked the tickets and she might know. a. gaxgwi dim bagwim’ ’ kaxgwi tim pakw-m when fut arrive.pl-1pl “When is it we’ll get there?” b. gaxgwigat dim bagwim’ ’ kaxgwi=kat tim pakw-m when=rep fut arrive.pl-1pl “When is it (did they say) we’ll get there?” “When is it (did you hear) we’ll get there?” c. silkwsax t’aahlagwigat silkwsax t’aahlakw=kat noon tomorrow=rep “(I heard/They said) at noon tomorrow.” In questions, =kat is oriented towards the addressee’s knowledge: the speaker is enquiring about the reported evidence the speaker assumes the addressee has for an answer (i.e. from the ticket agent). The targeting of the addressee’s knowledge is implied in the alternate translations offered by consultants and given in (3.90) “When is it did you hear...”. In both (3.90) and (3.91), the speaker has a reasonable expectation that the addressee would have some kind of reported evidence to support an answer to their question: in (3.90) it would be because the addressee booked the bus tickets; in (3.91) it would be because her sister just got off the phone with a reliable source for an answer to her question: 146  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? (3.91) Context: You’re asking your sister about what your niece bought at the store yesterday. Your sister just got off the phone with your mother, who was with your niece. a. gwigathl giigwit k’yoots kiikw-t k’yoots kwi=kat=hl what=rep=cnd buy-3sg yesterday “What is it (did they say) she bought yesterday?” “What is it did you hear she bought yesterday.” b. giigwigathl kiikw-i-(t)=kat=hl buy-tr-3=rep=cnd “(I heard) she bought  andadaala k’yoots anta-daala k’yoots container-money yesterday a wallet/purse yesterday.”  The insertion of the modal =ima into a question has a different effect from that of =kat: it creates a non-interrogative utterance, roughly translatable using ‘I wonder...’, as in (3.92) – (3.94): (3.92) Context: You’re sitting around with friends discussing life. You know that you need to find another job, but you also have the possibility of going back to college. ’ gwiyimahl dim jaby’ joxkuuhl ’ kwi=ima=hl tim tsap-y’ tsoxkuuhl what=mod=cnd fut do/make-1sg next.year “I’m wondering what to do next year.” (3.93) Context: Someone unfamiliar pulls into the driveway to talk to your uncle. a. naa tun naa t=xwin who pnd=this.one “Who is s/he?” “Who is this person?”  147  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? b. naayima tun naa=ima t=xwin who=mod pnd=this.one “I wonder who this s/he/this person is.” “Who might this s/he/this person be?” (3.94)  a. naayima ’ant sdils John naa=ima ’an=t stil-(t)=s John who=mod textscs.rel=3 accompany-3=pnd John “I wonder who went with John.” “Who might’ve gone with John?” b. Clarayima ’ant sdils John Clara=ima ’an=t stil-(t)=s John Clara=mod s.rel=3 accompany-3=pnd John “It’s Clara who might’ve gone with John.” “Maybe it was Clara who went with John.” “It must’ve been Clara who went with John.”  The use of modal =ima in questions and the effect it has in reducing the interrogative force of a question is examined in detail in the next chapter. ’ As we’ve seen with the other tests, nakw diverges significantly from =ima and =kat: ’ nakw cannot participate in any kind of interrogative statement. Example (3.95) is a yes/no question, formed by adding the interrogative enclitic =a to the sentence. Despite ’ occupying its required clause-initial position (cf. (3.87)) nakw is ungrammatical: ’ ’ (3.95) *nakwhl xmiyeenis Jasona ’ ’ nakw =hl x-miyeen-(t)=s Jason=a evid=cnd consume-smoke-3sg=pnd Jason=interrog = “Must Jason be smoking?” Comparing (3.96)a. with b. shows how the progressive yukw can occur within a ’ question, although nakw, being of the same syntactic category as yukw, cannot. Ex’ ample (3.96)c. shows how moving nakw to the first position of a wh-question is also ungrammatical:  148  3.6. Gitksan Evidentials: Propositional or Illocutionary Operators? (3.96)  a. nayukw ’ant sdils Clara na=yukw ’an-t sdil=s Clara who=prog s.rel-3sg go.with=pnd Clara “Who is going with Clara?” ’ b. *na=nakw ’an-t sdil=s Clara who=evid s.rel-3sg go.with=pnd Clara “Who must’ve gone with Clara?” ’ c. *nakw=na ’an-t sdil=s Clara evid=who s.rel-3sg go.with=pnd Clara “Who must’ve gone with Clara?”  This test suggests that both =ima and =kat are propositional operators. However, ’ the results are less clear with nakw, as it cannot participate in any kind of interrogative construction. 3.6.2.3  Interaction with Negation  As with cancelling the evidence requirement test, negation does not distinguish between a propositional and a modal analysis of evidentials. In a modal analysis, the evidence requirement is a presupposition, and will therefore project through negation. In a speech act operator analysis, the evidence requirement is a sincerity condition, which is not affected by negation either. In (3.97) the reported evidence projects through negation, as it is not a part of the assertion. The same effect can be observed in (3.98) with modal =ima: (3.97) neegatdiit sdilis Leiwat Fern stil-i-(t)=s Leiwa=t Fern nee=kat=tii=t neg=rep=contr=3sg go.with-tr-3sg=pnd Leiwa=pnd Fern “[I have reported evidence that] It wasn’t Leiwa who went with Fern.” = “[It’s not the case that I have reported evidence that] Leiwa who went with Fern.”  149  3.7. Summary (3.98) neeyimatiihl txookxwhl smax nee=ima=tii=hl txookxw=hl smax neg=mod=contr=cnd eat(pl)=cnd bears “[I have inferential evidence that] The bears might not have eaten.” = “[It’s not the case that I have inferential evidence that] The bears might have eaten.” ’ As shown in (3.76) and in the previous chapter, nakw and negation cannot occur within the same clause, thus the ungrammaticality of (3.99): ’ (3.99) *nee=tii nakw txookxw=hl smax neg=contr evid eat(pl)=cnd bears = “[I have sensory evidence that] The bears might not have eaten.” = “[It’s not the case that I have sensory evidence that] The bears might have eaten.” This is consistent with the with the fact that speech act operators in general cannot occur in the scope of negation.  3.7  Summary  In sum, what these six tests show is that =ima and =kat operate on the propositional level: they cannot be used when the proposition is known to be true or false, but a speaker may dissent from a statement made with either =ima or =kat. Moreover, both ’ patterns differently: =ima and =kat can be semantically embedded. Evidential nakw ’ nakw can be used when the speaker knows the prejacent is true or false. This has a pragmatic effect, expressing a mirative meaning when the proposition is true, and a ’ metaphorical meaning when the proposition is false. It has also been shown that nakw cannot be assented to, dissented from, or embedded in any way. These results indicate a split: while =ima and =kat are propositional, similar to the evidential modals in ’ St’át’imcets in the previous section, nakw operates on an illocutionary level, much like the Quechua evidentials. It was also shown that tests based on cancellation of the  150  3.7. Summary evidence-type and interaction with negation do not distinguish between =ima, =kat, ’ and nakw. These results are summarized in Table 3.5 below. =ima, =kat  ’ nakw  Felicitous if p is known to be true or false?  No  Yes  Pass assent/dissent test?  Yes  No  Semantically embeddable?  Yes  No  Able to scope outside interrogatives?  No  –  ’ Table 3.5: Test results for =ima, =kat, and nakw  151  Chapter 4 The Semantics of the Modal Evidentials =ima and =kat The tests in the previous chapter show that both =ima and =kat operate at the level of the proposition. In contrast, these tests show the opposite results for the evidential ’ nakw : it cannot be a propositional operator. This chapter takes the first part of these ’ results and presents a modal analysis of =ima and =kat, while nakw is analyzed in the following chapter.  4.1  The Issues  There are three main issues in adequately explaining the meaning of =ima and =kat. The first involves being able to account for both the evidence source restrictions and semantic properties of =ima and =kat. The previous chapter reviewed a semantic approach to evidentials that analyzes them as specialized types of epistemic modals: the evidence source restrictions are encoded by presuppositions, which restrict the modal base to those accessible worlds where evidence of that type holds. This chapter shows that both =ima and =kat are amenable to this kind of analysis. However, this leads directly to the second issue: a Gitksan speaker faced with the task of translating a sentence of the form =ima(p) into English, will use a variety of sentences that express varying degrees of modal strength from might to must, and other forms expressing modal force. This can be observed in (4.1):  152  4.1. The Issues (4.1) yugwimahl dim iixwt yukw=ima=hl tim iixw-t prog=mod=cnd fut fish-3 “He might be going fishing.” “He must be going fishing.” “He’s probably going fishing.” “He’s likely going fishing.” “He could be going fishing.” “Maybe/perhaps he’s going fishing.” “It seems he’s fishing.” There is similar variability in translations of the reportative =kat. Sentences of the form =kat(p) are typically translated into English as “I hear/heard that p”. This translation is often given to =kat(p) sentences when the speaker considers the source of the report to be reliable. However, there is another translation given to =kat(p) sentences when a speaker is neutral or perhaps considers the report to be less reliable, as shown in (4.2): (4.2) xstagas John go’ohl bingo gaxxw bingo kaxxw xsta=kat John ko’=hl win=rep John loc=cnd bingo last.night Reliable report: “I heard that John won at bingo last night.” Neutral reliability: “Apparently, John won at bingo last night.” These translations can also be considered to represent a kind of variable modal force, where “I heard that p” is akin to must, and “Apparently p” is somewhat weaker, resembling might. If =kat is also analyzed as an epistemic modal, we face the same problem as with =ima in formally representing this variability. Example (4.3) shows that within a possible worlds semantics, modal force is encoded by quantification over a modal base. In languages such as English, quantificational force is lexically encoded by modal verbs, such as might and must (Kratzer 1991), and the conversational background (e.g. epistemic, circumstantial, deontic etc. cf. (3.7)) is provided by the context:  153  4.1. The Issues (4.3) “John might/must be at home.” c  a. might(B)(w)(John be at home)  = 1 iff ∃w ∈ B(w) : John be at  home c (w ) = 1 b. must(B)(w)(John be at home)  c  = 1 iff ∀w ∈ B(w) : John be at  home c (w ) = 1 Thus, the leading questions are: first, if the conversational background of a modal is provided by the context, why are the meanings of =ima and =kat fixed with respect to the type of evidence? Secondly, if modal force is encoded by quantification, how does a variable force modal such as =ima fit into the universal-existential dual? How do the variable translations of =kat fit in? The third issue involves another effect associated with =ima: when =ima is inserted into a question it creates a non-interrogative utterance, roughly translatable using “I wonder.” An example of the effect =ima has in questions is given in (4.4): example a. is an ordinary wh-question in Gitksan, while example b. is a question that contains =ima, which is translated as a statement of uncertainty: (4.4)  ’ a. naa ’ant ginamhl xhlaw’sxw ’as Johna ’ naa ’an-t kinam-(t)=hl xhlaw’sxw ’a=s John who s.rel-3sg give-3sg=cnd shirt obl=pnd John “Who gave this shirt to John?” ’ b. naayima ’an-t ginamhl xhlaw’sxw ’as John ’ naa=ima ’an-t kinam-(t)=hl xhlaw’sxw ’a=s John who=mod s.rel-3sg give-3sg=cnd shirt obl=pnd John “I wonder who gave this shirt to John.”  The variable modal force of =ima, and the effect it has in questions, is also found in other unrelated, northwest coast languages such as St’át’imcets (Lillooet Salish) and NìePkepmxcín (Thompson Salish). Can a modal analysis account for this, or is some other mechanism required to turn a question into a statement of uncertainty? What level of meaning does this effect operate on? 154  4.2. The Plan  4.2  The Plan  The main claim pursued in this chapter is rooted in the generalizations regarding =ima and =kat discussed in the previous chapter: =ima and =kat semantically belong to the category of epistemic modals. Both =ima and =kat introduce quantification over possible worlds. In §4.3 it is argued that their evidential interpretations are the result of a presupposition that restricts the modal base to epistemically accessible worlds where evidence of some specific type holds. In the case of modal =ima, this presupposition places a condition on the modal base such that it contain worlds in which some general kind of inferential evidence holds. The presupposition attached to =kat restricts the modal base to worlds in which a report was made. The modals =ima and =kat differ from English modal auxiliaries in that they do not lexically encode modal force. In §4.4 I approach this question by relating the Gitksan data to two other separate, yet similar phenomena involving the expression of variable modal force in two unrelated languages: Bulgarian, and St’át’imcets. Research on variable modal force in these languages has resulted in two different technical refinements that can be applied to the denotations in (4.3): the ordering source (Kratzer 1991; Izvorski 1997), and the choice function (Rullmann et al. 2008). Both of these approaches attribute variable modal force not to the choice of the quantifier, which is uniformly universal, but to a parameter of interpretation that determines the value of a second function – an ordering source or choice function – operating on the modal base. Both of these contextually determined functions restrict the modal base in different ways, the effect of which determines the interpretation of modal force. Both of these analyses can be applied to the epistemic modal =ima in sentences such as (4.1), and the reportative =kat sentences in (4.2). However, in §4.4 I depart from all previous analyses of variable modal force evidentials in claiming that the various degrees of modal force correspond to (at least) two different types of ordering sources in Gitksan, and that the choice function analysis can be reduced to an ordering source one. Specifically, the weak/strong interpretations of =ima 155  4.3. A Modal Analysis of =ima and =kat correspond to empty/non-empty ordering sources which order an existentially quantified epistemic modal base. This analysis is then extended to St’át’imcets, where the weak/strong interpretations of modals in that language also correspond to empty/nonempty ordering sources, but over a universally quantified modal base. What this gives us is a unified account and a theoretical typology of languages in which modal forces vary under a fixed quantifier. §4.5 turns to the the effect =ima has when added to a question in what are called Conjectural Questions (CQs). Following Littell et al. (2010), my central claim is that CQs have the semantics of ordinary questions: they denote sets of propositions. The presupposition analysis of =ima is then applied to the question: the presuppositions carried by each proposition in the question denotation conjoin, so that the CQ as a whole presupposes everything presupposed by each of its members. The resulting conjoined presupposition entails that there is mixed evidence, and therefore that the speaker does not expect the hearer to be able to provide an answer to the question. The outcome is a reduced interrogative force for CQs: not only is the hearer not required to answer, the speaker is encoding that the hearer is probably not able to answer.  4.3  A Modal Analysis of =ima and =kat  In this section I support the line of argumentation emerging in the literature, specifically Izvorski (1997), Matthewson et al. (2007), Rullmann et al. (2008), and McCready and Ogata (2007), that evidentials in some languages are epistemic modals. To recap the results of chapter 3, both =ima and =kat pass the tests for being propositional operators: they contribute to the truth-conditional content of an assertion, they can be embedded, and the modal claim they are a part of can be assented to and dissented from. This places them in the same semantic class as the individual evidentials in St’át’imcets, for which Rullmann et al. give a modal analysis, a general denotation for which is given in (4.5):  156  4.3. A Modal Analysis of =ima and =kat (4.5) modal  c,w  is only defined if c provides a modal base B.  If defined, modal  c,w  = λf.λp.∀w [w ∈ f (B(w)) → p(w )].  Recall the Rullmann et al. claim that the modal force of the St’át’imcets evidentials is uniformly universal. Whether the modal evidential is understood to quantify over the entire modal base (strong) or a sub-part (weak) of the modal base depends on the choice of the function f . The individual evidential meanings of the modal evidentials reflect different types of definedness conditions on the modal base. For example, with the reportative ku7, the modal base is restricted to worlds where the reported evidence in w holds. Thus, reportative ku7 has the denotation in (4.6): (4.6) Semantics of ku7 (reportative) ku7  c,w  is only defined if c provides a modal base B such that for all worlds  w ∈ B(w), the reported evidence in w holds in w , and f is a choice function such that f (B(w)) ⊆ B(w). If defined, ku7  c,w  = λf.λp.∀w [w ∈ f (B(w)) → p(w ) = 1].  To see how this works, consider the example in (4.7): (4.7) wa7 ku7 ku sts’éts’qwaz’ l-ta stswáw’cw-a be rep det trout in-det creek-det “[I heard] There are trout in the creek.” In (4.7), the speaker’s evidence for p is the fact that a report was made. Therefore, the accessible worlds in the reportative case are all those worlds in which (for example) the speaker overheard John, an experienced fisherman in the village, talking in the coffee shop. The sentence in (4.7) then asserts that in all worlds of this type in which a report from John about fishing is reliable, there are trout in the creek. Since the actual world is presupposed to be a world in which John said there are trout in the creek, the sentence  157  4.3. A Modal Analysis of =ima and =kat makes a strong claim about the actual world: unless John is not reliable, there are trout in the creek in the actual world.29 The analyses of -an’ and k’a are exactly parallel; the only difference is located in the definedness condition, which for -an’ requires that the modal base contain all those worlds in which the perceived evidence in w holds, and k’a requires that it contain all worlds in which the inferential evidence in w holds. A modal analysis, coupled with the presupposition which encodes the individual evidential meanings, makes the right predictions regarding the semantic properties of =ima and =kat. Modal =ima is similar to the inferential k’a in St’át’imcets: =ima requires that the modal base contain only those worlds in which the inferential evidence in w holds. Evidence for this comes from the fact that =ima is infelicitous when used in other kinds of conversational backgrounds. For example, the verb da’akhlxw is a circumstantial modal, and is typically translated as might, can, or able to in English (Davis et al. 2009). =ima is infelicitous in a circumstantial context such as (4.8) (adapted from Kratzer 1991): (4.8) Context: You’re up in the Suskwa and notice a burnt patch of forest. You know that huckleberries typically take seed in burnt alpine areas. a. #limxsimahl limxs=ima=hl grow=mod=cnd “Berries might be  maay’ maay’ berries growing  go’osun go’osun loc.here here.”  dim limxshl maay’ go’osun b. da’akhlxwhl da’akhlxw =hl tim limxs=hl maay’ go’osun circ=cnd fut grow=cnd berries loc.here “Berries might/can/are able to grow here.” 29  Matthewson et al. (2007: 212) note that a presuppositional analysis of reportatives does not mean  the same thing as “John said that p.” Under the modal analysis of reportatives, the sentence presupposes the existence of a report which constitutes evidence for p, and asserts that p must be true, given that report. In a sentence containing a verb of saying, the sentence asserts that a report was made, and does not commit the speaker to any claim about the truth or otherwise of p.  158  4.3. A Modal Analysis of =ima and =kat Example (4.8)a. with modal =ima would only be felicitous in a context where the speaker can infer that there might be berries growing there at that moment, such as people coming back to town with purple hands. This puts us in a position to develop a preliminary semantics for =ima and =kat, given in (4.9) and (4.10) (adapted from Matthewson et al. (2004)): (4.9) A Preliminary Semantics for =kat (to be revised) =kat  c,w  is only defined if c provides a modal base B such that for all worlds  w ∈ B(w), the reported evidence in w holds in w , and f is a choice function such that f (B(w)) ⊆ B(w). If defined, =kat  c,w  = λf.λp.∀w [w ∈ f (B(w)) → p(w ) = 1].  (4.10) A Preliminary Semantics for =ima (to be revised) =ima  c,w  is only defined if c provides a modal base B such that for all worlds  w ∈ B(w), the inferential evidence in w holds in w , and f is a choice function such that f (B(w)) ⊆ B(w). If defined, =ima  c,w  = λf.λp.∀w [w ∈ f (B(w)) → p(w ) = 1].  Both =ima and =kat share another feature with the evidential modals in St’át’imcets: variable quantificational force. The next section shows that this is where Gitksan and St’át’imcets diverge empirically: whereas the modals in St’át’imcets have a universal interpretation by default, =ima and =kat have a might-like, existential reading by default. Rullmann et al. locate this variability in the choice function. However, in §4.4.4 I propose that another conversational background, the ordering source, can replace this function and make the right predictions for the variability found in both Gitksan and St’át’imcets evidential modals. Thus, the denotations in (4.9) and (4.10) are ultimately not appropriate for =kat and =ima.  159  4.4. The Variable Modal Force of =ima and =kat  4.4  The Variable Modal Force of =ima and =kat  As argued in the previous section, =ima and =kat semantically belong to the category of epistemic modals, such as might and must in English: they introduce quantification over possible worlds. However, both =ima and =kat differ from English modal auxiliaries in two respects: first, whereas the conversational background of a modal in English is determined by the context, both =ima and =kat lexically encode an epistemic conversational background through a presupposition restricting the modal base worlds. Secondly, unlike modals in English, neither =ima nor =kat lexically encode modal force, rather, it is determined by the context.30 This ‘reversed’ arrangement is summarized in Table 4.1:  English  Modal Base  Modal force  Context  Lexical  Gitksan Lexical  Context  Table 4.1: Lexically vs. contextually determined modal base and force (cf. Rullmann et al. 2008)  However, it is important to clarify what it means for a modal to have ‘variable force’. We can observe the variable modal force effect of =ima and =kat by refining the contexts in such a way that restricts the use of either might as possibility or must as necessity in any given language. We can then test this against the distribution of =ima and =kat in these contexts, where we find that both =ima and =kat are felicitous both in contexts where possibility and necessity is required. 30  The effects of variable modal force are somewhat more difficult to observe with the reportative.  This is described in detail in §4.4.2 below.  160  4.4. The Variable Modal Force of =ima and =kat  4.4.1  The Variability of =ima  The variable modal force of =ima can be conditioned by both the context it occurs in, and the type of evidence a speaker has for an =ima-assertion. 4.4.1.1  Context-Conditioned Modal Force  The type of information available to a speaker and what they can infer from it naturally influences the strength of belief in the truth of the proposition. However, the fact that a person has inferential evidence for the truth of a proposition does not determine a certain strength of belief in this proposition. In (4.11) the context is simple enough that both must and might are felicitous translations in English: depending on a speaker’s previous experiences with John and his rod and tackle box, John might be fishing, or he must be fishing: (4.11) Context: You’re wondering where your friend is. You notice his rod and tackle box are not in their usual place. yugwimahl dim iixwt yukw=ima=hl tim iixw-t prog=mod=cnd fut fish-3 “He might be going fishing.” “He must be going fishing.” “He’s probably going fishing.” “He’s likely going fishing.” “He could be going fishing.” “Maybe/perhaps he’s going fishing.” In (4.12), a speaker concludes that, based on the fact that her father was frequently away when she was a child, it must’ve been her mother who fed her:31 (4.12) Context: The speaker’s father was away frequently when she was a child. 31  For any two (or more) sentences S1 and S2 , I define the relation  to mean that a consultant  judges a translation S1 to be more felicitous than S2 given the context, without S2 being necessarily infelicitous.  161  4.4. The Variable Modal Force of =ima and =kat naa’ayima ’an yookxwiny’ naa’a=ima yookxw-in-y’ ’an mother(informal)=mod s.rel eat-caus-1sg “It must’ve been mother who fed/cooked for me.” “It might’ve been mother who fed/cooked for me.” The above context relies on a logical deduction a speaker is making in order to reconstruct a past event she trying to recall. This same kind of deduction is used in (4.13) in order to retrace a recent sequence of events and explain an outcome of those events. In (4.13), deduction leads to the speaker to conclude that, given that everyone else who ate the bad fish got sick, Gwen must also be sick because of the fish. The necessity modal must or some equivalent phrase is more felicitous in English than the weaker might, and =ima is also felicitous: (4.13) Context: There was a bad can of fish: everyone at the dinner got sick, and Gwen was there too. Gwen xsidinhl hont xsit-in-(t)=hl fish=t Gwen vomit-caus-3sg=cnd fish=pnd Gwen “The fish must’ve made her sick.” “The fish might’ve made her sick.” In another case, the speaker concludes that, upon hearing that Alvin’s new truck broke down recently, (4.14) is an appropriate response to the suggestion that the fuel pump is responsible for the breakdown: (4.14) Context: You hear that Alvin’s truck broke down on the way up to the Suskwa. It’s a very reliable truck, but someone suggests that the problems he’s having starting it indicate a problem with the fuel pump. ’ nidimahl gan wilt ’ nit=ima=hl wil-t kan 3sg=mod=cnd coord do.something-3sg “That must’ve been why it happened.” “That might’ve been why it happened.” 162  4.4. The Variable Modal Force of =ima and =kat We can apply the same methodology and examine the felicity of =ima in contexts where the expression of necessity is less felicitous than possibility. In the context given in (4.15), there is no reason why the speaker’s uncle must know the people he’s talking to; they could just be strangers asking for directions: (4.15) Context: You see your uncle stopped at the intersection talking to some people through the window of his pickup. You and your friends don’t recognize the people. ’ wilaayimas nibib-y’ (nidiit) ’ ’ wilaa-i-(t)=ima=s nipip-y (nitiit) know-tr-3=mod=pnd mother’s.brother-1sg 3pl “My uncle might know them.” “My uncle must know them.” ’ Likewise, in a story told by DH in (4.16), the destination of the group in this passage is not a central part of the story and is likely not determined, thus it is not necessarily the case that they went to their hunting grounds, only a logical possibility: (4.16) BH: ii ndahl ii nda=hl conj where=cnd “And where did they  w’adiit? w’a-tiit? arrive-3pl get to?”  DH: bakwdiid=ima ansilinasxwdiit pakw-tiit=ima an-si-lin-sxw-tiit arrive.pl-3pl=mod geo.loc-caus-trap-antipass-3pl “Maybe they got to their hunting grounds.” “They must’ve gotten to their hunting grounds.” In (4.17), two people are speculating on why someone in the news (like a politician or movie star) passed away. In the absence of any concrete information, there is no reason why the person must’ve been sick: it could’ve been a heart attack, or an accident:  163  4.4. The Variable Modal Force of =ima and =kat (4.17) Context: You see on TV that someone in the news passed away. ’ nagwimahl siipxwt ’ nakw=ima=hl siipxw-t dist=mod=cnd sick-3sg “S/he might’ve been sick.” “S/he must’ve been sick.” In another conversation fragment in (4.18), a speaker weakens the potential universal modal force of =ima by explicitly stating that they forget. This causes the second translation of (4.18) to sound infelicitous: (4.18) Context: Alvin’s wife asks when Alvin is getting back from shopping in Smithers. Alvin told you the night before, but you forget. witxws Alvin? GS: gaxguhl kaxkwi=hl witxw=s Alvin? when=cnd arrive=pnd Alvin “When is Alvin arriving?” ’ y, ’ ’ LW: tag witxwima nit silkwsax ’tak-y, ’ silkwsax witxw=ima nit ’ forget-1sg arrive=mod 3sg at.noon “I forget, he might arrive around noon.” “I forget, he must arrive around noon.” 4.4.1.2  Modals in Coordinated Sentences  The cases above show that in contexts where the expression of epistemic necessity is less felicitous than the expression of possibility, and vice versa, =ima is felicitous in both types of contexts. We can expand on this test of modal strength: Rullmann et al. (2008) note that when sentences containing modals in English are coordinated, certain combinations of modal strength are less felicitous than others. Example (4.19) and (4.20) show how a sentence of the form ‘must p and might/must not p’ is a contradiction in English, yet the use of =ima in these coordinated constructions is acceptable to consultants: 164  4.4. The Variable Modal Force of =ima and =kat (4.19) guxwimahl gyuwadan, ii neeyimahl kuxw=ima=hl kyuwatan, ii nee=ima=hl run.away=mod=cnd horse conj neg=mod=cnd guxwimahl gyuwadan kuxw-(t)=ima=hl kyuwatan run.away-3sg=mod=cnd horse (a.) “Maybe the horse ran away, and maybe it hasn’t.” (b.) #“The horse must’ve ran away, and it might/must not have.” (4.20) hla yugwimahl dim wis, ii neeyima hla yukw hla yukw=ima=hl tim wis, ii nee=ima hla yukw incept prog=mod=cnd fut rain, conj neg=mod incept prog dim wis tim wis fut rain (a.) “It might start raining, and it might not.” (b.) #“It must start raining, and it might/must not have.” Modal =ima can also be used in coordinated sentences where the speaker is listing a number of possibilities to explain some circumstance. Here again, =ima shows its possibility reading, as in (4.21) and (4.22): (4.21) guxwimahl gyuwadan, ii guxwindiidimahl kuxw=ima=hl kyuwatan, ii kuxw-’en-tiit=ima=hl run.away=mod=cnd horse conj run.away-caus-3pl=mod=cnd ’ ’ tihlxwhl gyuwadan (kuba) ’ ’ (kuba) tihlxw=hl kyuwatan small child=cnd horse “Maybe the horse ran away, and maybe the kids chased it away.” (4.22) Context: All of the trucks are out of the driveway. It’s possible that everyone has gone to work, but since it’s a holiday, it’s possible they went fishing instead. ’ ii yugwimahl dim yugwimahl dim gahahlalsdiit, yukw=ima=hl tim yukw=ima=hl tim gahahla’lst-tiit, ii conj prog=mod=cnd fut prog=mod=cnd fut work.pl-3pl iixwdiit iixw-tiit fish-3pl “Maybe they’re working (today), and maybe they went fishing.” 165  4.4. The Variable Modal Force of =ima and =kat As predicted by the idea that =ima has variable modal force, sentences of the form ‘might p and not p’ are bad in Gitksan, just as they are in English. In example (4.24), a speaker is already aware of the truth or falsity of a proposition, thus the use of =ima on the first conjunct is infelicitous: (4.23) #hla yugwimahl dim hla yukw=ima=hl tim incept prog=mod=cnd fut tim wis tim wis fut rain #“It might/must start raining, but  wis, ii needii hla yukw wis, ii nee=tii hla yukw rain conj neg=contr incept prog  it won’t rain.”  Also as in English, the reverse arrangement is felicitous if the conjunct marked by =ima is used to speculate about a future event: (4.24) neediihl wis, ii nee=tii=hl wis, ii neg=contr=cnd rain, conj “It’s not raining, but it might start  dim wis hla yugwimahl hla yukw=ima=hl tim wis incept prog=mod=cnd fut rain raining.”  All of the examples presented in this subsection confirm that =ima can be interpreted as might or must. I showed this by showing that =ima is felicitous when the contexts are set up to control for the felicity of might over must as in English, and vice versa. 4.4.1.3  Evidence Type and Modal Force  There is one way the must vs. might modal force readings of =ima can be teased apart: this involves examining contexts where there is some kind of observable physical evidence available to the speaker. Recall example (2.3), repeated in (4.25) for convenience. In a context where a speaker is making an inference based on general knowledge or similar previous experiences, =ima displays its variable force. On the other hand, the evidential ’ nakw is infelicitous: it requires the speaker to have observable evidence:  166  4.4. The Variable Modal Force of =ima and =kat (4.25) Context: You’re sitting at home talking about going berry-picking. It’s August, and the berries are usually ripe this time of year on the Suskwa. a. mugwimahl maay’ mukw=ima=hl maay’ ripe=mod=cnd berries “The berries might/must be ripe.” “Maybe the berries are ripe.” ’ b. #nakwhl mukwhl maay’ ’nakw =hl mukw=hl maay’ evid=cnd ripe=cnd berries “The berries must be ripe.” “Looks like the berries are ripe.”  When the context is adjusted to include observable evidence, as in in (4.26), the use ’ of nakw is felicitous, which is typically translated by consultants as a necessity modal must, and very rarely the weaker might. (4.26) Context: People are arriving home after a day of berrypicking up in the Suskwa. They’re carrying buckets of berries, and their hands are all purple. a. mugwimahl maay’ mukw=ima=hl maay’ ripe=mod=cnd berries “The berries might be ripe.” ’ b. nakwhl mukwhl maay’ ’nakw =hl mukw=hl maay’ evid=cnd ripe=cnd berries “The berries must be ripe.” “Looks like the berries are ripe.”  In chapter 2 I showed that =ima is also felicitous in observable evidence contexts as ’ in (4.26); however, in these observable evidence contexts where both =ima and nakw are ’ felicitous, =ima is interpreted as the weaker might, while nakw takes over the stronger 167  4.4. The Variable Modal Force of =ima and =kat must interpretation. Looking at it from another angle, given an observable evidence context such as in (4.27), a speaker is faced with a choice of making an epistemic claim ’ using =ima or nakw. Upon observing some money on the floor, by using =ima, a speaker is claiming that it’s possible but not necessary that the money was dropped by ’ the person standing ahead in line. By using the evidential nakw, a speaker is making a stronger inference from this same physical evidence: the money must’ve been dropped by the person: (4.27) Context: You and a friend are at the store and you see a dollar coin on the ground. It could be yours, your friend’s, or the person in front of you in line may’ve dropped it. ’ a. kwoodindimahl daalat ’ kwoo-T-in-t=ima=hl daala-t drop-t-caus-3sg=mod=cnd money-3sg “S/he might’ve dropped his/her money.” ’ ’ b. nakwt kwoodindimahl daalat ’ ’ nakw =t kwoo-T-in-t=ima=hl daala-t evid=3sg drop-t-caus-3sg=mod=cnd money-3sg “S/he must’ve dropped his/her money.” This is where the modal properties of =ima clearly emerge. Assuming the standard notions of epistemic modality, the use of must over might is rooted in the speaker’s certainty level about the proposition expressed. Thus, a speaker who uses a possibility modal is less certain about the truth of the embedded proposition than a speaker who uses a necessity modal. This distinction is mapped on to the distribution of =ima and ’ nakw in the observable evidence contexts they are both felicitous in: by asserting (4.27)a. when b. is also felicitous, a speaker is conveying that s/he only might’ve dropped their money. This is because b. can unambiguously mean that they must’ve dropped their money. ’ A note on the translation of nakw must be reiterated here from chapter 2: speakers ’ commonly translate nakw into the modal auxiliary must in English. However, we also 168  4.4. The Variable Modal Force of =ima and =kat ’ saw how nakw clearly does not pass the test for being a propositional operator. Chapter ’ 5 presents a non-modal analysis of nakw as sentential force specifier, and also analyzes ’ ’ the interaction between =ima and nakw as a ‘pragmatic blocking’ effect: nakw blocks =ima from expressing its strong, necessity-like reading.  4.4.2  The Variability of =kat  Unlike =ima, sentences with =kat are not translated using epistemic modals such as must and might, rather, =kat(p) sentences are typically translated as “I hear/heard p”, as in (4.28): (4.28) Context: The speaker is telling her friends at the coffee shop that Mary had her long hair cut recently. She hasn’t seen Mary’s hair herself yet, but knows because the speaker’s sister is the hairdresser who did it. gungojigas kwin-kots-i-(t)=kat=s caus-cut-tr-3=rep=pnd “[I hear] Mary had her hair  Mary-hl gest Mary=hl kes-t Mary=cnd hair-3 cut.”  However, =kat is frequently translated with modal adverb apparently, as in (4.29) (see also Hunt 1993 for several examples): ’ ’ ’ haniiguy ’ahl lo’op (4.29) majigathl pax ’ ’ ’ ’a=hl lo’op ha-nii-kuy pax mats-i-(t)=kat=hl loc=cnd rock hit-tr-3=rep=cnd instr-in-light (i.) “I hear he hit the window with a rock (and broke it).” (ii.) “Apparently, he hit the window with a rock.” With translation (i.), the speaker is transmitting the report of an adult who happened to be working across the street in their yard when they saw the window of the speaker’s house being broken. The speaker judges this to be a reliable source, and this sentence receives a “I hear/heard p” translation. However, in translation (ii.), the speaker either 169  4.4. The Variable Modal Force of =ima and =kat holds a neutral attitude towards the report, or has less confidence in the report. This would be the case if the speaker receives the report from one of the children who were there and wanted to avoid punishment or blame. We see the same contrast in (4.30): (4.30) lumakdigas John=hl daala John=hl daala lumakt-i-(t)=kat=s donate-tr-3=rep=pnd John=cnd money (i.) “I heard John put in money (for the feast).” (ii.) “Apparently, John put in money.” A group of people are counting up the contributions after a feast, and speculating about the different contributions people made that night. A speaker may translate (4.30) as (i.) if they overheard the information from one of the people who are responsible for the final accounting, thus normally a reliable source. On the other hand, if someone simply overheard from an unknown source in a crowded room that John also contributed, the translation in (ii.) is felicitous. However, it is important to note that this is not necessarily an unreliable source: by using apparently a speaker is conveying a neutral attitude towards the proposition – maybe the report is reliable, maybe it isn’t. We can generalize the kinds of contexts and interpretations of (i.) and (ii.) in the examples above in the following way: in both (4.29) and (4.30) speaker judgments vary in the use of the translation “I hear/heard p”, which can mean that a speaker either believes the source of the report to be reliable, or is neutral towards the source. Translations with apparently can also mean a neutral attitude towards the source, but more often used only when a speaker views the source as less reliable. The variable modal force of =ima is transparent in its translations as must or might. However, examples such as (4.29) and (4.30) illustrate that the modal force of =kat reveals itself in the difference between universal and existential force, which would correspond to a difference between the paraphrases of (i.) and (ii.) above in (4.31):  170  4.4. The Variable Modal Force of =ima and =kat (4.31) (i.) “[Given what I’ve heard], p must be true” (ii.) “[Given what I’ve heard], p may be true” Because the existence of the report is presupposed and crucially not asserted, the truth conditions of reportative sentences depend not on whether there was a report, but on whether the report was true in the speaker’s view. The speaker asserts that it is at least possible that the report was true (Izvorski 1997; Faller 2002; Matthewson et al. 2007). More specifically, under a modal semantics, the modal force of a reportative evidential equates to how reliable the source of the report is. If we assumed that a reportative evidential necessarily has universal force, then we would predict that it could only be used in cases where the report comes from a reliable source. A reportative evidential with existential force would be used in cases where the report does not come from a perfectly reliable source, since the embedded proposition would only be true in some of the worlds where the report is made.  4.4.3  Default Modal Force  The previous two subsections showed how both =ima and =kat display variable modal force. Under a possible worlds analysis this corresponds to variable quantificational force. In some contexts =ima and =kat have existential quantification, and in other contexts they have universal quantification. However, there is a preference for existential quantificational force. There is suggestive evidence for this claim based on speakers’ judgments of both =ima and =kat when asked specifically about the strength in meaning they convey, and in the preference for might-like translations out-of-the-blue. There may be a paradigmatic explanation for the default existential force of =ima: there is a way to disambiguate in favour of the universal translation, especially in con’ texts where the evidential nakw is felicitous. This was shown in the translations in (4.26) ’ and (4.27) above: even though the diagnostics in chapter 3 showed that evidential nakw cannot be a modal, it is consistently translated as one, and as only having must-like force. Thus, Gitksan has a strategy for clearly disambiguating =ima as existential. 171  4.4. The Variable Modal Force of =ima and =kat With =kat it is less clear: speakers mostly translate =kat as “I hear/heard p”. However, given that this translation can cover the spectrum of reliability from reliable to neutral to unreliable, a default interpretation is more difficult to identify. Unlike =ima, there is also nothing to lexically disambiguate =kat. In the next section it is argued that the source of variable modal force in =ima and =kat is not located in the quantifier, which is uniformly existential for both =ima and =kat, but in the ordering source.  4.4.4  The Ordering Source in Deriving Variable Modal Force  It is claimed here that we already have a tool on hand for handling variable modal force: the ordering source. Kratzer (1991) developed the ordering source to have wide application to a number of phenomena such as unwanted entailments (i.e.  φ → φ),  the weak readings of necessity modals, different interpretations involving circumstantial possibilities, counterfactual conditionals, and graded modality. It is the last of these that relates to the issue facing us with the variable modal force of =ima and =kat. The general idea is as follows: in English, the modal base B is contextually determined by a conversational background. However, there is a second conversational background, the ordering source O, which imposes a particular evaluative ordering of the B-worlds. The general idea is that the modal base B contains propositions representing facts or knowledge about the world as assessed by the speaker in a given scenario, while the ordering source contains propositions representing beliefs, ideals, norms, intentions, and universally-held assumptions about normal courses of events in the world. These two conversational backgrounds interact: the propositions that comprise the ordering source impose an ideal ordering on those that comprise the modal base. For example, a modal statement such as may(p) is interpreted as meaning that p is the case in some B-worlds (for some contextually-determined B) which are ranked as best by some salient O. Likewise, must(p) is interpreted as meaning that p is the case in all B-worlds ranked as best by O.  172  4.4. The Variable Modal Force of =ima and =kat More formally, the ordering source g, like the modal base B, is also a function from worlds to sets of propositions. If g is applied to a world w, the resulting set of propositions g(w) induces a partial order ≤g(w) on the modal base worlds. This is defined as follows in (4.32): (4.32) (i.) ∀w1 , w2 ∈ X : w1 ≤g(w) w2 iff {p ∈ g(w) : w2 ∈ p} ⊆ {p ∈ g(w) : w1 ∈ p} (ii.) For a given partial order ≤g(w) on worlds, define the selection function Og(w) that selects the set of ≤g(w) -best worlds from any set X of worlds: ∀X ⊆ W : Og(w) (X) = {w ∈ X : ¬∃w ∈ X : w ≤g(w) w} adapted from von Fintel and Heim 200732 Given the partial order as defined above, w1 ≤g(w) w2 means that w1 verifies all the propositions in g(w) that w2 does, that is, w1 is equally close to or closer to the ‘ideal’ than w2 is with respect to g(w), iff among the propositions in g(w), those that are satisfied in w2 are a subset of those that are satisfied in w1 . Thus, the meaning of w1 ≤g(w) w2 with an epistemic modal base can be paraphrased as ‘w1 is at least as desirable or in accordance with what we know at w than w2 ’. It may also be noted that ordering does not technically give us degrees of conformity with the ordering source. In English, the sets of propositions that induce the function Og(w) are contextually determined, just as they are with the modal base. Thus denotations of the modals must and might including the ordering source would be as in (4.33): (4.33) must might  c,w c,w  = λp.∀w [w ∈ Og(w) (B(w)) → p(w ) = 1]. = λp.∃w [w ∈ Og(w) (B(w)) ∧ p(w ) = 1].  The effects of the ordering source are intuitively the clearest with root modals. With deontic modals, the propositions that make up the ordering source typically express rules to be followed by anyone who respects the regulations for driving a car: the deontic 32  In von Fintel and Heim’s notation, the ordering source is notated as Max. I have used O simply  for mnemonic reasons, following Portner (1997).  173  4.4. The Variable Modal Force of =ima and =kat conversational background “what the driving regulations provide”. This conversational background functions as the ordering source and thus orders the modal base worlds according to their compliance with its propositions. As an illustration, let g(w) contain the following propositions: p1 = “you have a driving license”, p2 = “your car is insured” and p3 = “you’re wearing a seatbelt”, g(w) = {p1 , p2 , p3 }.33 We can distinguish eight different sets of worlds in terms of these three propositions, given in Table 4.2. For instance, world w4 is the world where you don’t have a driving license but your car is insured and you are wearing a seatbelt. p1  p2  p3  w0  1  1  1  w1  1  1  0  w2  1  0  1  w3  1  0  0  w4  0  1  1  w5  0  1  0  w6  0  0  1  w7  0  0  0  Table 4.2: Possible worlds that can be distinguished in terms of three propositions.  We have for instance that w2 ≤ w6 , i.e. a world such as w2 , where the only deviance from the norm g(w) is that your car is not insured, is closer to g(w) than a world where your car is not insured and you don’t have a driving license.34 (4.34) {p : p ∈ g(w) ∧ w6 ∈ p} ⊆ {p : p ∈ g(w) ∧ w2 ∈ p} To get a feel for how this works with epistemic modality, consider an epistemic reading of (4.35): this sentence has an epistemic modal base, containing a set of propositions 33 34  Example adapted from Nauze (2008). Notice that the ordering is partial: the worlds w1 and w2 cannot be ordered by Og(w) : driving  without a seatbelt does not comply more with g(w) than driving without insurance, nor vice versa.  174  4.4. The Variable Modal Force of =ima and =kat that make up our knowledge in the actual world, and a stereotypical ordering source, which represents a normal course of events in this context. (4.35) John must be fishing. Suppose we know that John’s rubber boots are missing, and his truck is gone. The prejacent of (4.35) is not true in all the worlds compatible with what we know: given this modal base it’s entirely plausible he went berry picking instead of fishing. But in the ordering source is a proposition which says ‘rubber boots are used for fishing’. Using this ordering source proposition, all the worlds in which the missing boots are used for fishing are going to count as better than worlds in which they are not. The next step is applying an ordering source analysis to account for the variable modal force of =ima and =kat. It is shown in this subsection that the interaction between the two conversational backgrounds, the modal base and ordering source, provides us with the technical resources for accounting for this variation in modal force readings. The basic thrust of the analysis is that the quantification of =ima is fixed as existential, while the ordering source is what modulates what is translated as modal force in English. More specifically, the interpretation of strong/weak modals in Gitksan is not a function of the universal/existential dual, but whether the ordering source is empty (weak) or non-empty (strong). It is also shown here how this analysis can be extended to St’át’imcets, another language that displays the same type of variable force modality, and how we can also capture default readings in both languages in a straightforward way using the ordering source. 4.4.4.1  Ordering Sources with Fixed Quantification  I will start with the assumption that the value of a quantifier is not determined by a parameter of interpretation: there is no contextually determined function that fixes the value of quantification over the modal base as existential in one context and universal in another. The quantification associated with =ima is fixed, just as it is with lexical modals in English. This leaves two other formal options within the system: both the 175  4.4. The Variable Modal Force of =ima and =kat modal bases and ordering sources are independent conversational backgrounds; either can be determined lexically, or determined by the context. As mentioned above (cf. (4.33)), in English both the modal base and ordering source are contextually determined. In contrast, in Gitksan the modal base is restricted to indirect evidence (through the presupposition lexically associated with =ima). However, the value of the ordering source in Gitksan is entirely contextually determined. I claim that this gives us a formal way for explaining why context plays a role in determining the force of a modal statement in Gitksan. More specifically, I begin by grounding the following analysis in the basic notion that the belief state of the speaker, and what a speaker considers a normal course of events, play a formal role in modulating what we interpret as modal force (Kratzer 2002). As shown above, a stereotypical ordering source involves propositions representing the normal course of events, or relatively fixed ideas about the uses of things like rubber boots. The interpretation of an epistemic modal can also be conditioned by a speaker’s beliefs. This characterizes a doxastic ordering source. Thus, a doxastic modal statement is one in which the epistemic modal quantifies over the worlds of a modal base ordered by an ideal determined by the belief state of a speaker. The effects of doxastic and stereotypical ordering sources, and the contrast they draw out between Gitksan and English, can be observed in (4.36): (4.36) yugwima=hl iixwis yukw=ima=hl iixw-(t)=s prog=mod=cnd fish-3=cnd “John must be fishing.” “John  John John John might be fishing.”  B(w) epistemic: {John’s rubber boots are missing; his truck is not in the driveway; it’s fishing season} g(w) stereotypical: {Rubber boots are used for fishing; rubber boots are not ideal for hunting, or berry picking} g(w) doxastic: {Knowing how much John likes fishing} 176  4.4. The Variable Modal Force of =ima and =kat Given the fact that John’s rubber boots are missing, and that rubber boots are used for fishing, “John must be fishing” is more felicitous than “John might be fishing.” This strong reading is derived from the stereotypical ordering source: the modal base B(w) contains the proposition that John’s rubber boots are missing and the ordering source g(w) concerns the the typical use of rubber boots. If the world of evaluation w is such that rubber boots are typically used for fishing, the ordering source Og(w) is such that worlds in which John uses his boots for fishing are ranked more highly than worlds in which he uses them for some other purpose (other things being equal). (4.36) will assert that the worlds in which John uses his boots for fishing, are worlds in which ‘John is fishing’ is true. Since the speaker believes w to be such a world, it follows that the speaker believes ‘John is fishing’ is true in w. Because of this belief in the typical use of rubber boots, the interpreted modal force is strong, translated as must. If, however, the speaker does not believe that rubber boots are used solely for fishing – the boots could be used for berry picking – the proportion of accessible worlds where it is true that John is fishing because his rubber boots are missing, will be smaller. This is because the set of worlds in g(w) also contains worlds where he didn’t go fishing (beside the ones where he did). This leads to the resulting interpretation that ‘John is fishing’ is only a slight possibility in w. The doxastic reading functions in the same way, only instead of evaluating the normal course of events or uses of items such as boots, the speaker is evaluating a modal statement based on their beliefs, such as their prior experience with John’s likes or activities. This analysis is based on Izvorski (1997), who shows that this variable modal force effect is present in the perfect in Bulgarian, which has a modal (and evidential) interpretation in addition to its aspectual one: (4.37) az sâm došâl 1sg be-1sg.pres come.p.part “I have come.” “I have apparently come.” (Izvorski 1997: 228) Izvorski claims there is a covert evidential operator (Ev) in sentences such as (4.37) that has a modal semantics. The force of this modal in Evp sentences is determined 177  4.4. The Variable Modal Force of =ima and =kat by the speaker’s belief or trust in the evidence. In its report reading, (4.37) can mean “I may have come”, “I probably came”, or “I must have come”, given what a person X says. In other words, the more trustworthy X is, the closer to a universal interpretation the modal has. Under an inferential reading, Evp sentences like (4.37) can interpreted along the lines of “I must have come” because in stereotypical contexts, the speaker bases her reasoning on a highly reliable source of (indirect) evidence. These effects are captured by the ordering source, as the actual domain of quantification is restricted by the ordering source. A sketch of Izvorski’s analysis is given in (4.38): (4.38) B = {p: a speaker considers p indirect evidence in w} B(w) = {u ∈ W : ∀p[(p is indirect evidence in w) → u ∈ p]} g(w) = {p: a speaker believes p with respect to the indirect evidence in w} In words, an Evp statement is true in a world w with respect to the conversational backgrounds provided by B(w) and g(w), iff p is true in all worlds accessible from w which come closest to the ideal represented by the speaker’s beliefs regarding the available indirect evidence in w (Izvorski 1997: 9). Thus a denotation for modal/evidential sentences of the form Ev(p) in Bulgarian is given in (4.39): (4.39) Ev  c,w  is only defined if c provides a modal base B such that for all worlds  w ∈ B(w), the inferential evidence in w holds in w Ev(p)  c,w  = 1 iff for ∀w ∈ Og(w) (B(w)) : p(w )  c,w  = 1.  The modal base and ordering source are the propositions which narrow down or order the set of accessible worlds under either an inferential or reportative interpretation of pe, as in (4.40): (4.40) Ivan izpil vsičkoto vino včera Ivan drunk-pe all.the wine yesterday “Ivan apparently drank all the wine yesterday.” (Izvorski 1997: 228)  178  4.4. The Variable Modal Force of =ima and =kat (i.) Inferential interpretation: B(w) = {There are empty wine bottles in Ivan’s office} g(w) = {If there are empty wine bottles in someone’s office, that person has drunk the wine} (ii.) Reportative interpretation: B(w) = {Mary said that Ivan drank the wine} g(w) = {Normally, Mary is reliable as a source of information} The reason for universal quantification in (4.39) comes from the observation that these kinds of modal/evidential statements in Bulgarian tend towards a default strong reading. A universal modal statement is then pragmatically weakened by the ordering source, the effect being a might-like translation. However, in Gitksan, =ima tends towards a weak might-like reading by default. This is built into the denotation of =ima, given in (4.41), where quantification is existential: (4.41) The Semantics of =ima (Final) =ima  c,w  (revised from (4.10))  is only defined if c provides a modal base B such that for all worlds  w ∈ B(w), the inferential evidence in w holds in w . If defined, =ima  c,w  = λp.∃w [w ∈ Og(w) (B(w)) ∧ p(w ) = 1].  At face value, it would seem counterintuitive to attribute a strong, must-like reading to a modal with existential quantification. However, we can find a similar effect of semantic strengthening in the nominal domain in the entailment patterns with existentially quantified DPs. For example, restricting an existential statement leads to strengthening: an expression like “some male students” is semantically stronger than the expression “some students”, since a sentence of the form “some male students smoke” is true in a subset of situations in which “some students smoke” is true (i.e., the former asymmetrically entails the latter). We can derive the same strengthening effect from the ordering of an existentially quantified modal base: in example (4.36), a set of worlds 179  4.4. The Variable Modal Force of =ima and =kat where John’s boots are missing, and boots are used for fishing is a more restricted set of worlds than one in which only John’s boots are missing. If you assert that in some possible world where John’s boots are missing, he’s fishing, that is a weaker claim than asserting that in some possible world where John’s boots are missing and boots are used for fishing, he’s fishing. However, we need the connection to the actual world to make the latter statement stronger: assuming that the actual world is a world in which the ordering source propositions are true. Therefore, asserting that in some of the smaller set of worlds, John is fishing, is a stronger claim than asserting that in some of the larger set of worlds, John is fishing. Note that the opposite parallel can be drawn with Izvorki’s universal analysis in (4.39): restricting a universal quantifier leads to weakening, as “all male students smoke” is weaker than “all students smoke”. Both the strengthened existential and the weakened universal are achieved by a non-empty ordering source. This analysis accounts for the strong, must-like reading of =ima, and also allows us to derive its default weak reading. Whereas a non-empty ordering source restricts the modal base to a subset of O-ideal words, an empty ordering source would logically remove this restriction by mapping every possible world to the empty set. Kratzer (1991: 645) characterizes this as alethic modality. Whereas epistemic modality has an epistemic modal base with an ordering based on doxastic reasoning or stereotypicality, alethic modality is a kind of purely logical modality: it does not relativize the modal to any particular kind of facts, rather, our epistemic reasoning is based solely on the facts that comprise the modal base. Thus a weak reading of (4.36) is obtained in (4.42), where the unordered modal base is simply existentially quantified over: (4.42) yukw=ima=hl iixw-(t)=s John prog=mod=cnd fish-3=cnd John “John might be fishing.” “John must be fishing.” B(w) epistemic: {John’s rubber boots are missing; his truck is not in the driveway; it’s fishing season} g(w) empty: ∅ 180  4.4. The Variable Modal Force of =ima and =kat This basically resembles a standard might-as-existential modal: a speaker of (4.42) is asserting that in some world where John’s rubber boots are missing, his truck is not in the driveway, and it’s fishing season, John is fishing in that world. This is the locus of the might  must meaning: the difference between saying that in some world where  his boots are missing, he’s fishing (the empty ordering source), and saying that in some world where his boots are missing and boots are used for fishing, he’s fishing (the nonempty ordering source, and assuming that the actual world is such a world), is claimed to be a big enough difference so that the latter gives you a stronger, must-like reading. In sum, the indirect evidence presupposition placed on the modal base ensures the epistemic interpretation of =ima, while the value of the ordering source is contextually determined. By default, the ordering source is empty, but other contextual factors can intervene and provide the ordering source with propositions that order the worlds of the modal base according to some doxastic or stereotypical ideal. Thus, the modal force interpretations of =ima can be schematized as in (4.43): (4.43) The modal force interpretations of =ima: B(w) = {u ∈ W : ∀p[(p is indirect evidence in w) → u ∈ p]} g(w) = (i.) Strong: {p: a speaker believes p with respect to the indirect evidence in w} (ii.) Weak: (default): ∅ Turning to the variable modal force of =kat, we can apply an ordering source analysis to the reliability of the reported evidence. As with =ima, we can build into the denotation the default weak interpretation of =kat in (4.44):  181  4.4. The Variable Modal Force of =ima and =kat (4.44) The Semantics of =kat (Final) =kat  c,w  (revised from (4.9))  is only defined if c provides a modal base B such that for all worlds  w ∈ B(w), the reported evidence in w holds in w . If defined, =kat  c,w  = λp.∃w [w ∈ Og(w) (B(w)) ∧ p(w ) = 1].  Consider again the case of the feast contributions in (4.45), repeated from the previous section: (4.45) lumaktigas John=hl daala John=hl daala lumakt-i-(t)=kat=s donate-tr-3=rep=pnd John=cnd money (i.) “I heard John put in money (for the feast).” (ii.) “Apparently, John put in money.” Faller observes in Quechua that if the reliability of the source is unknown, only an existential analysis predicts a reportative sentence to be true (2002: 109): if the reliability of a source is unknown, then the set of worlds in which that report is heard will include both worlds where the report is true, and worlds where it is false. Both translations in (4.45) can reflect this kind of neutrality, and I claim that this is derived from an empty ordering source, as in (4.46): (4.46) Modal base and Ordering source for a weak interpretation of (4.45): B(w) reportative: {a conversation was overheard in the feast hall} g(w) empty: ∅ However, if a speaker views the source of the report as reliable, then the ordering source reflects this and is non-empty, as in (4.47): (4.47) Modal Base and Ordering Source for a strong interpretation of (4.45): B(w) reportative: {The accountant was loudly discussing John’s contribution} 182  4.4. The Variable Modal Force of =ima and =kat g(w) non-empty: {Normally, accountants are reliable sources for accounting matters} The ordering source analysis for =ima applies mutatis mutandis to =kat: an existentially quantified modal base is given a stronger interpretation through a non-empty ordering source. This is schematized for =kat in (4.48): (4.48) The modal force interpretations of =kat: B(w) = {u ∈ W : ∀p[(p is reported evidence in w) → u ∈ p]} g(w) = (i.) Strong: {p: a speaker believes p with respect to the reported evidence in w} (ii.) Weak: (default): ∅ 4.4.4.2  An Emerging Theoretical Typology: Variable Force in St’át’imcets Modals  An ordering source analysis can be applied to the St’át’imcets inferential evidential k’a, as in (4.49):35 (4.49)  a. t’ak k’a tu7 kents7á ku míxalh go.along infer then dietic det bear “A bear must’ve gone around here.” (Rullmann et al. 2008: 5) b. wa7 k’a séna7 qwenúxw impf infer counter sick “He may be sick.” (Rullmann et al. 2008: 5)  35  Portner (2009) also discusses the use of the ordering source instead of the choice function in  St’át’imcets. See also Kratzer (2009) for a discussion of the effects of an empty ordering source on an epistemic modal base. These publications were not available until just prior to the filing of this dissertation.  183  4.4. The Variable Modal Force of =ima and =kat However, St’át’imcets modals differ from Gitksan modals in one key respect: whereas the default modal force of =ima is weak, the default interpretation of modals in St’át’imcets is strong. Recall from the previous section that Rullmann et al. (2008) locate the variability in modal force in a contextually-determined choice function f . The choice function picks out a subset of the modal base, which is universally quantified over. The larger the subset of the modal base selected by f , the stronger the modal force expressed. f may simply be the identity function, which results in a universal must-like reading. If f selects a proper subset of the modal base, the resulting reading is weaker, although that subset is still universally quantified over. The denotation of inferential k’a is repeated below in (4.50): (4.50) Semantics of k’a (inferential) k’a  c,w  is only defined if c provides a modal base  B such that for all worlds w ∈ B(w), the inferential evidence in w holds in w , and f is a choice function such that f (B(w)) ⊆ B(w). If defined, k’a  c,w  = λf.λp.∀w [w ∈ f (B(w)) → p(w ) = 1].  However, given the difference in default quantification between Gitksan and St’át’imcets, we can extend the ordering source analysis to the St’át’imcets modals in a straightforward way: in the previous subsection it was shown that =ima has fixed existential quantification over a presupposed epistemic modal base; the default weak interpretation comes from the ordering source, which is assumed to be empty by default (cf. (4.41)). The revised denotation of St’át’imcets modals such as k’a given in (4.51) can involve the exact same components as in Gitksan, except that it has universal quantification: (4.51) k’a  c,w  is only defined if c provides a modal base B such that for all worlds  w ∈ B(w), the inferential evidence in w holds in w . If defined, k’a  c,w  = λg.λp.∀w [w ∈ Og(w) (B(w)) → p(w ) = 1].  This analysis gives us a unified account of variable modal force found in the evidential modals of both Gitksan and St’át’imcets: modal evidentials in these languages 184  4.4. The Variable Modal Force of =ima and =kat involve a presupposed epistemic modal base, and an ordering source, which is empty by default in both languages. This empty ordering source is what derives the default weak interpretation of existential =ima, and the default strong interpretation of universal k’a, as schematized in (4.52): (4.52) The interpretations of k’a B(w) = {u ∈ W : ∀p[(p is indirect evidence in w) → u ∈ p]} Strong (default): g(w) = ∅ Weak: g(w) = {p: the speaker believes p with respect to the indirect evidence in w} With an empty ordering source, modal k’a is universally quantifying over literally every possible world in the modal base (the worlds in which the relevant evidence holds), representing alethic modality, just as =ima does under a default reading. However, a non-empty ordering source has the opposite effect with a universally quantified modal base: whereas a non-empty ordering source has the effect of pragmatically strengthening an existential modal, a non-empty ordering source essentially weakens a universal modal claim. This approach has its roots in Kratzer’s original characterizations of the modal base and ordering source. Whereas the modal base will always contain a consistent set of facts, other sources of information that can make up a potential ordering source may be inconsistent, or inconsistent with these facts (Kratzer 1991). For example, in (4.49)a., a speaker may be faced with a variety of facts which can include overturned garbage cans, tracks in the mud, apples missing off the tree, stories overheard in the coffee shop etc. This is a set of consistent propositions that comprise the modal base. Given the abundant evidence that a bear was present, it would be true in all stereotypical worlds consistent with the evidence that a bear did in fact go around there. However, a universal modal claim can be weakened if the speaker believes the modal base evidence to be less reliable, or that there are other plausible courses of events. For example, a 185  4.4. The Variable Modal Force of =ima and =kat modal base for (4.49)b. could be the symptoms or evidence typical of having a cold (i.e. a red face, runny nose etc.), and the doxastic ordering source would concern the speaker’s belief in the applicability of this kind of evidence. If the world of evaluation w is such that this kind of evidence is normally right in indicating that someone is sick, we will consider those accessible worlds where this evidence holds. Thus, (4.49)b. will assert that all those worlds are p-worlds, and the interpreted modal force is strong. However, if these symptoms are considered unreliable or inconclusive as evidence for being sick – coming in from a cold winter day would produce the same symptoms – the set of accessible worlds where this evidence holds, and that this indicates an illness, will be very restricted; hence, the resulting interpretation is that p is only slightly possible in w. Doxastic ordering sources contain information characterizing a speaker’s belief state. An empty ordering source clearly differs in this respect, involving reasoning purely from accepted, speaker-external facts rather than considering the belief state of the speaker. To get a feel for the difference between the doxastic and empty ordering sources in St’át’imcets, consider example (4.53) uttered by a speaker in isolation upon hearing a knock at the door: (4.53) nilh k’a kw s-Henry wa7 pegwpeg’wtsám’ foc infer det nom-Henry impf knock.repeatedly “That’ll be Henry knocking.” (Rullmann et al. 2008: 7) This can be understood as simply saying something about the speaker’s beliefs, specifically that it’s compatible with her beliefs that the person at the door is Henry. In contrast to this doxastic reading, consider (4.53) again, but this time uttered to a hearer in the following context in (4.54): (4.54) Henry said he would come tonight, so if he isn’t here yet it follows that (4.53). Here the speaker is not simply commenting on her belief state. She is rather making a statement of general fact, specifically that the evidence provided by the modal base 186  4.4. The Variable Modal Force of =ima and =kat leads to the following conclusion: that it is compatible with all known facts that the person at the door is Henry. She is not claiming that her belief state follows from the evidence (see Tancredi 2007 for a similar discussion and Lakoff 1972, p. 233 for similar examples and explanations). This kind of ‘pure’ modality does not involve identification of an ideal in any sense, and so does not involve an ordering source either. Thus, under its default reading – which is (4.54) – modal k’a only involves quantification over a modal base. Could a choice function analysis be extended to Gitksan =ima? There are a variety of issues that suggest not. The first consideration is theoretical: what distinguishes a choice function from the ordering source in deriving the effect of variable modal force? Both the choice function and ordering source rely on context to determine their value. An ordering source analysis provides a truth-conditional way of picking out a set of ideal modal base worlds, which is then quantified over. However, exactly what kinds of modal base worlds are picked out by the choice function? In other words, the ordering source is principled because it orders worlds according to propositions which are believed or stereotypical, whereas f could be totally random in the set of worlds which it picks out. Additionally, if we maintain a dual conversational background treatment of modality, how does the ordering source interact with f ? Another consideration is empirical: St’át’imcets modals such as k’a have a strong, must-like reading by default, which under a choice function analysis means that f is by default the identity function. We could apply a choice function to the existentially quantified modal base of =ima. This faces the same theoretical problem: how do we know or assess which worlds are picked out by f that would give a strong reading of =ima? Another option would be to have universal quantification for =ima, giving it the same denotation as St’át’imcets k’a. However, given the default weak reading of =ima, we would then need some way of explaining why some languages have identity (St’át’imcets) vs. non-identity (Gitksan) function readings by default.  187  4.4. The Variable Modal Force of =ima and =kat By attributing variability of modal force to an effect of empty vs. non-empty ordering sources, a typology of modality emerges that captures not only the variable modal force readings of modals in Gitksan and St’át’imcets, but also the default readings these modals have, which are uniformly treated as the effect of an empty ordering source, as Table 4.3 shows: Gitksan (∃)  St’át’imcets (∀)  Strong  non-empty  empty (default)  Weak  empty (default)  non-empty  Table 4.3: Empty vs. Non-empty Ordering Sources in Strong/Weak Modals  4.4.4.3  Strengthening and Weakening in Paradigms  There are certain empirical challenges in claiming that a language has a default interpretation in terms of modal force. Rullmann et al. (2008) note that in both textual examples and elicitation contexts, the universal is preferred. I have found the opposite tendency in the translations of observed conversations in Gitksan, where there is a preponderance of the weaker existential interpretations (cf. (2.55)). In this subsection I look at additional supporting evidence for the default quantificational force in both Gitksan and St’át’imcets, as summarized in Table 4.3. To begin with, we can view variable modal force and default values in terms of strengthening and weakening a quantifier. The two possibilities for this are given in (4.55)(Rullmann et al. 2008):36  36  Rullmann et al. identify two other possibilities. One is Ambiguity: Modals can be lexically am-  biguous between existential and universal readings. The other is Underspecification: Quantificational force is underspecified and requires setting through some mechanism. They also note that it is unclear what kind of theoretical mechanism could fix the value of a quantifier in these two options.  188  4.4. The Variable Modal Force of =ima and =kat (4.55) (i.) ∃ plus strengthening: Modals (in some languages) are uniformly existential and are strengthened though some mechanism. (ii.) ∀ plus weakening: Modals (in some languages) are uniformly universal and are weakened though some mechanism. Options (i.) and (ii.) appear to be borne out by Gitksan and St’át’imcets, respectively. Secondly, we have a theory-internal mechanism for strengthening or weakening the modal interpretation of a quantifier by adding another function over its domain: the ordering source orders the domain of the quantifier according to some ideal. There are also pragmatic and paradigmatic considerations: both Gitksan and St’át’imcets have other modal-type words that are not ambiguous with regards to modal force. In example (4.56) St’át’imcets has the adverb sxek which is translated as ‘maybe’. (4.56) St’át’imcets a. lh-7ámh-as kw s-qwal’út-s-al’ap sxek hyp-good-3.conj det nom-talk-caus-2pl.erg maybe um’-en-tumulh-ás kelh give-dir-2pl.obj-3.erg fut “If you talk to him nicely, he might give you some.” b. nihl ku cw7aoy-s kw s-k’úl’-em múta7 ku pála7 k’a foc det neg-3.poss det nom-make-mid again det one infer sxek xetspásq’et maybe week “So she wouldn’t have to make more for about a week maybe.” This can be considered as a disambiguation strategy: Whereas k’a has a default universal reading, there is a morpheme dedicated to an unambiguously weaker modal force reading. The opposite arrangement of this occurs in Gitksan: whereas =ima has a default ’ existential reading, the evidential nakw can only be translated as a strong, must-like modal: 189  4.4. The Variable Modal Force of =ima and =kat (4.57) Context: You and a friend are at the store and you see a dollar coin on the ground. It could be yours, your friend’s, or the person in front of you in line may’ve dropped it. ’ a. kwoodindimahl daalat ’ kwoo-T-in-t=ima=hl daala-t drop-t-caus-3sg=mod=cnd money-3sg “She might’ve dropped her money.” ’ ’ b. nakwt kwoodintdima=hl daala-t ’nakw =t ’ kwoo-T-in-t=ima=hl daala-t evid=3sg drop-t-caus-3sg=mod=cnd money-3sg “She must’ve/#might’ve dropped her money.” (4.58) Context: You and a friend are going fishing. You notice blood on the rocks ahead of you where your friend is walking. ’ ’onin a. k’ojinimahl ’ k’ots-i-n=ima=hl ’on-n cut-tr-2sg=mod=cnd hand-2sg “You may’ve cut your hand.” ’ ’ ’onn b. nagwimi k’otshl ’ ’ nakw =mi k’ots=hl ’on-n 2sg cut=det evid “You must’ve/#might’ve cut your hand.” As discussed earlier in this chapter and in chapter 2, there are restrictions on the ’ distribution of nakw : it is only felicitous in contexts in which the speaker can make an inference from observable evidence. When =ima appears in these contexts, it is almost always translated as a possibility modal. In the next chapter this arrangement is ’ analyzed as a case of pragmatic blocking: nakw is more specialized for the strong reading, and blocks =ima from that reading. This is relevant because when a speaker chooses ’ to use =ima in a context where nakw is felicitous, they are essentially expressing that they don’t believe the evidence they have warrants a stronger statement. This would again amount to existential quantification over a set of worlds that is restricted by the 190  4.4. The Variable Modal Force of =ima and =kat ordering source: There is a proposition that is a part of the ordering source that would say that, normally, when you see blood at a person’s feet (and they are using a knife), that person cut themself. Additionally, there is no ‘weaker’ epistemic modal in Gitksan other than =ima. The ’ Gitksan equivalent of the St’át’imcets adverb sxek is the word nidima, which is common in Gitksan discourse, and is frequently translated as ‘maybe’. It is morphologically ’ and modal =ima. This was complex, composed of the 3sg independent pronoun, nit, observed in the Mastermind example in Chapter 3 (3.75), repeated in (4.59): (4.59)  a. neeyimahl nee=ima=hl neg=mod=cnd “Is it maybe red?” b. ee’e, ee’e, yes, “Yes,  ihleetxwda ihleetxwt=a be.red=interrog (lit. “Is it not possibly red?”)  ’ nidima ’ nit=ima 3sg=mod maybe.”  I take this paradigmatic relationship found in both St’át’imcets and Gitksan to be supporting evidence for default quantificational force in these languages, as summarized in Table 4.3.  4.4.5  Interim Summary  Gitksan represents a class of languages in which an epistemic modal can be translated as having variable modal force, while lexically encoding an epistemic conversational background. Assuming that modals are quantifiers over possible worlds, it was argued that the modal =ima is uniformly existential, and shown that variability of modal force can be derived from the ordering source. The strong/weak interpretations of =ima correspond to non-empty vs. empty ordering sources. This in turn predicted a theoretical typology. In St’át’imcets, the arrangement is reversed: quantification is  191  4.5. Modal =ima and Conjectural Questions fixed as universal, but the strong/weak interpretations of the epistemic modal in that language correspond to empty vs. non-empty ordering sources. This analysis is rooted in Kratzer’s original (1991) analysis of graded modality, and how the modal paraphrases in (4.60) which encode finer distinctions of possibility and necessity, correspond to different ordering sources: (4.60)  a. Es kann gut sein, dass...  Human Possibility  “There is a good possibility that...” b. Es besteht eine geringe Möglichkeit, dass...  Slight Possibility  “There is a slight possibility that...” c. Es ist wahrscheinlich, dass...  Human Necessity  “It is probable that...” Gitksan and St’át’imcets simply represent languages where the grades of modality as expressed in the various paraphrases in (4.60) are admitted by a single lexical item, or a many-to-one relation between ordering sources and an epistemic modal. The translation of the epistemic modals =ima and =kat into English reveal that a speaker’s certainty level about the proposition must still be underlyingly there. This is not about distinguishing different sources of evidence, but how a speaker’s world knowledge, beliefs and experiences condition their attitude towards the propositions in the modal base. This is invoked by context, and the ordering source is a contextually determined function which provides the formal means to truth-conditionally capture the effects this knowledge and/or beliefs has on the modal base.  4.5  Modal =ima and Conjectural Questions  In chapter 3, §3.6.2.2 it was shown that modal =ima has a peculiar effect in interrogatives. The insertion of =ima into a question creates a non-interrogative utterance, translated by speakers using ‘I wonder....’ Examples of this effect are given below: example (4.61)a. is an evidential assertion, (4.61)b. is an ordinary yes-no question, and 192  4.5. Modal =ima and Conjectural Questions (4.61)c. contains both the evidential and the yes-no question marker and is translated as a statement of uncertainty:37 (4.61)  a. sdinimahl xbiist tust stin=ima=hl xbiist tust be.heavy=mod=cnd box dem “That box might be heavy.” b. neehl sdinhl xbiist tusta xbiist tust=a nee=hl stin=hl dem=interrog neg=cnd be.heavy=cnd box “Is that box not heavy?” c. neeyima=hl sdinhl xbiist tust=a nee=ima=hl stin=hl xbiist tust=a neg=mod=cnd be.heavy=cnd box dem=interrog “I wonder if that box isn’t heavy.”  The same effect is shown in (4.62) with wh-questions: (4.62)  ’ a. ginamiyimas ’ kinam-i-(t)=ima=s give-tr-3sg=mod=pnd “Gwen might’ve given the b. naa naa who “Who  ’ Gwenhl xhlawsxw ’as John ’ Gwen=hl xhlawsxw ’a=s John Gwen=cnd shirt obl=pnd John shirt to John.”  ’ ’ ’an-t ginamhl xhlawsxw ’as John ’ ’ ’an-t kinam-(t)=hl xhlawsxw ’as John s.rel-3sg give-3sg=cnd shirt obl=pnd John gave this shirt to John?”  c. naayima ’an-t naa=ima ’an-t who=mod s.rel-3sg “I wonder who gave this  ’ ’ ginamhl xhlawsxw ’as John ’ ’ kinam-(t)=hl xhlawsxw ’as John give-3sg=cnd shirt obl=pnd John shirt to John.”  Littell et al. call these kinds of questions containing inferential evidentials such as =ima Conjectural Questions (CQs): they are syntactically and semantically questions, 37  This subsection is largely adapted from Littell, Matthewson & Peterson 2009.  193  4.5. Modal =ima and Conjectural Questions but pragmatically CQs have the force of assertions.38 The apparent reduced interrogative force of the CQs might suggest that they are some kind of rhetorical question. However, Littell et al. argue that CQs are distinct from rhetorical questions, and form part of a three-way typology, given in Table 4.4, of question-types based on expectations of Speaker/Addressee knowledge of the answer:39 Speaker  Addressee  Ordinary Questions  No  Yes  Rhetorical Questions  Yes  Yes  Conjectural Questions  No  No  Table 4.4: Speaker/Addressee knowledge of the answer across question types  Given these properties, how we can derive the right semantics and pragmatics for CQs? Ideally, we want to derive the meaning compositionally, using only the independentlyneeded semantics for the elements contained within CQs. Littell et al. show that this is attainable, given the independently motivated modal analysis of evidentials presented in §4.3: =ima has a modal semantics but carries a presupposition that there is evidence of a certain type for the prejacent.  4.5.1  The Properties of Conjectural Questions  Modal =ima appears to affect the illocutionary force of a question, lessening it; thus, are the b. utterances in (4.61) and (4.62) actual questions or assertions? In approaching this question, it is necessary to start with the basics and review the three different but 38  Recent work in Inquisitive Semantics suggests a fourth property: whether or not an utterance is in-  quisitive, a property shared by questions and some kinds of assertions, such as disjunctions (Groenendijk 2006). CQs do appear to be inquisitive, in that they raise the issue of which of a set of alternatives holds. 39 ’ Rhetorical questions in Gitksan can only be formed with the evidential nakw. These are examined in chapter 5.  194  4.5. Modal =ima and Conjectural Questions interrelated notions of question (Higginbotham 1996): (4.63)  Syntactic: An instance of a certain sort of linguistic structure. Semantic: An utterance with a certain type of denotation. Pragmatic: A particular sort of speech act.  CQs have the syntax and semantics associated with questions, but do not have the same pragmatic properties as questions. The data from Gitksan support this: CQs in Gitksan take the characteristic syntactic form of questions, with either a wh-element taking a particular sort of complement, or the usual yes-no question particle, as examples (4.61) and (4.62) above showed. Furthermore, results show that CQs syntactically embed in the same manner as ordinary questions, as in (4.64): (4.64) neediihl wilaaxs Henry ji ixstadinimas nee-tii=hl wilaax-(t)=s Henry tsi ixsta-t-in-(t)=ima=s neg-contr=cnd know-3=pnd Henry irr taste-T-caus-3sg=mod=pnd Lisahl xdii x-tii Lisa=hl Lisa=pnd consume-tea ‘Henry doesn’t know if Lisa might like tea drinking.’ Not only are CQs syntactically questions, they denote the same sorts of things that questions denote. That CQs embed under predicates like know, ask, etc. in an identical manner to ordinary questions is prima facie evidence that they are of the same type. This can be treated under a fairly standard approach (Hamblin 1973; see Groenendijk and Stokhof 1982; 1984 for an alternative view): a question denotes a set of propositions, each of which is a (partial) answer to the question. The question set contains the set of possible answers (as in Hamblin 1973, but unlike in Karttunen 1977), as in a yes-no question in (4.65), or in the wh-question in (4.66): (4.65) is that box heavy  w  = {that box is heavy, that box is not heavy}  195  4.5. Modal =ima and Conjectural Questions (4.66) who gave this shirt to John  w  = {that Gwen gave this shirt to John, that  Leiwa gave this shirt to John, that Holly gave this shirt to John,...} = {p : ∃x[p = that x gave this shirt to John]} The semantics of CQs can be fairly straightforwardly handled by a Hamblin-set analysis: in the denotation of a question containing a modal, the modal is distributed across the set of possible answers, as shown in (4.67),: (4.67) who ♦ gave this shirt to John  w  = {that Gwen ♦ gave this shirt to John, that  Leiwa ♦ gave this shirt to John, that Holly ♦ gave this shirt to John,...} = {p : ∃x[p = that x ♦ gave this shirt to John]} The insertion of a modal intuitively has the effect of ‘weakening’ of the interrogative force of the question: the speaker is asking only who could have possibly given this shirt to John, rather than who did give this shirt to John. It is shown below that the evidence presuppositions of the epistemic modal =ima, when inserted into a question, are actually what is responsible for a further weakening of interrogative force. In terms of pragmatics and discourse, Ordinary Questions (OQs) such as (4.65) and (4.66) have three features: first, an OQ is a request by the speaker for information from the addressee. Secondly, its answer is not known to the Speaker, but the Speaker thinks the Addressee may know it. Thirdly, an OQ requires an answer in order for the dialogue to be felicitous (Caponigro and Sprouse 2007). However, not everything that is a syntactic or semantic question is, by this definition, a pragmatic question. Consider an Ordinary Question vs. a Rhetorical Question (RQ), given in (4.68) (examples adapted from Caponigro and Sprouse 2007): (4.68)  a. ‘John looks like an interesting syntactician.’ OQ: ‘What does he know about semantics?’ [Possible answers: He knows a lot about semantics; He doesn’t know a lot about semantics; etc.]  196  4.5. Modal =ima and Conjectural Questions b. ‘I don’t think we should have John on our short list.’ RQ: ‘(After all,) what does he know about semantics?’ [Implicates he knows nothing about semantics.] RQs and OQs are syntactically and semantically the same, but pragmatically different (Sadock 1971; Han 2002; Sprouse 2007; Caponigro and Sprouse 2007): an RQ differs from an OQ in that the answer is known to the Speaker and the Addressee, and they both also know that the other knows the answer as well. In terms of the requirement for an answer, RQs also differ from OQs in that they can have, but do not require an answer. CQs are similar to RQs in these respects. They have same syntactic form and alternative semantics as OQs, but the sentential force of a declarative. CQs can have, but do not require an answer. For the CQ in (4.69)a., either the Speaker or the Addressee can respond with (4.69)b.: (4.69)  a. nayima ’ant sdils John na=ima ’an-t stil-(t)=s John who=mod s.rel-3sg accompany-3=pnd John ‘I wonder who went with John.’ b. Billyima (’ant sdils John) Bill=ima (’an-t stil-(t)=s John) Bill=mod s.rel-3sg accompany-3=pnd John ‘Maybe it was Bill (who went with John.).’  However, CQs are not acceptable in RQ situations, as shown in (4.70): (4.70) Context: The speaker is watching a baseball game. The star batter on the speaker’s favourite team keeps missing the ball and striking out, jeopardizing the outcome of the game. Out of exasperation, the speaker exclaims (4.70). ’ sinsima nida ’ sins=ima nit=a blind=mod 3sg=interrog = “Is he blind or something?” 197  4.5. Modal =ima and Conjectural Questions CQs also differ from RQs in terms of Addressee knowledge. In an RQ, typically both the Speaker and Addressee know the answer. CQs, in contrast, are typically bad in situations in which the Addressee can be assumed to know the answer (cf. also Rocci 2007, p. 147). This is shown not only in (4.70), but in other cases of Addressee knowledge such as (4.71). (4.71) #neeyimahl nee=ima=hl ynq=mod=cnd ‘I wonder if you’re  xwdaxina xwdax-n=a hungry-2sg=interrog hungry.’  In sum, a CQ differs from an OQ and RQ in that it is a statement expressing uncertainty or wondering. An CQ is unlike both an OQ and an RQ in that its answer is not known to the Speaker or the Addressee, and they both also think that the other does not know the answer. A CQ invites, but does not require, an answer from the Addressee, and may be answered by either the Speaker or the Addressee, similar to an RQ. These claims are summarized in table 4.5: S. knows answer  A. knows answer  Answer req’d  OQ  No  Yes  Yes  CQ  No  No  No  RQ  Yes  Yes  No  Table 4.5: Knowing and requiring an answer across question types  4.5.2  An Analysis of Conjectural Questions  The goal is to derive the reduced interrogative force of CQs from the semantics of CQs, rather than by positing the presence or absence of an invisible speech-act-operator for which there is no syntactic or semantic evidence. This means that we want to use only independently-needed aspects of the meanings of evidentials and questions to derive the 198  4.5. Modal =ima and Conjectural Questions right semantics and pragmatics for CQs. Following Littell et al. (2009), the central claim here is that CQs in Gitksan have the semantics of ordinary questions, but exhibit a reduced interrogative force in the pragmatics due to their evidential presuppositions. The semantics of questions outlined in (4.65) and (4.66) will serve as a base for the claim that the presupposition introduced by a question is the conjunction of the presuppositions introduced by the statements in its Hamblin set.40 This conjunction of presuppositions is usually not detectable, as each proposition in the question set introduces exactly the same presupposition. This is illustrated in (4.72) and (4.73) using the presupposition associated with too: (4.72) Is that box heavy too? {that that box is heavy too, that that box isn’t heavy too} (4.73) Has Jason stopped smoking? {that Jason has stopped smoking, that Jason has not stopped smoking} In (4.72) all propositions in the question set presuppose that some salient box is heavy other than the one currently in question, and in (4.73) all the propositions in the question set presuppose that Jason used to smoke. However, the interesting cases are where each member of the Hamblin set introduces a different presupposition. (4.74) Who here doesn’t drink anymore? {that Tyler doesn’t drink any more, that Lisa doesn’t drink anymore, ...} (4.75) Who went to Paris again? {that Scott went to Paris again, that Edna went to Paris again, ...} The question in (4.74) presupposes of each x in the contextually salient group that x used to drink, and (4.75) presupposes of each x in the contextually salient group that x has been to Paris. 40  For a similar idea, namely that a question presupposes all the presuppositions of its sub-constituents,  see Guerzoni (2003).  199  4.5. Modal =ima and Conjectural Questions Evidence that the combined presupposition exists is found in the interpretations in (4.76)a,b. The exclusive particle only presupposes that its embedded proposition is true. The conjoined presupposition of (4.76)a. is therefore that each country has two cities. While this is not true for strictly every country in the world (e.g. Vatican City or Tuvalu), the assumption is nevertheless fairly commonly held, and therefore the question is felicitous. However, (4.76)b. is odd: although some countries do have two capital cities (e.g., Bolivia, Swaziland) it is definitely infelicitous to presuppose this of each country. (4.76)  a. Which countries have only two cities? {that Canada has only two cities, that Iceland has only two cities, ...} (presupposes of each country x that x has two cities.) b. #Which countries have only two capital cities? {that Canada has only two capital cities, that Iceland has only two capital cities, ...} (presupposes of each country x that x has two capitals.)  The reduction of interrogative force is predicted from two assumptions. First, questions presuppose the conjunction of the presuppositions of their partial answers; secondly, evidentials such as =ima introduce presuppositions of evidence. The denotations and presuppositions of a yes-no question and a wh-question in Gitksan are illustrated in (4.77) and (4.78) respectively: ’ (4.77) xmiyeenimat Jasona ’ x-miyeen-(t)=ima=t Jason=a consume-smoke-3sg=mod=pnd Jason=interrog ‘I wonder if Jason smokes.’ = {that Jason might smoke [presupposing there is inferential evidence that Jason smokes], that Jason might not smoke [presupposing there is inferential evidence that Jason doesn’t smoke]}  200  4.5. Modal =ima and Conjectural Questions ’ ’ (4.78) naayima ’ant ginamhl xhlawsxw ’as John ’ ’ naa=ima ’an-t kinam-(t)=hl xhlawsxw ’a=s John who=mod s.rel-3 give-3=cnd shirt obl=pnd John “I wonder who gave this shirt to John.” = {that Gwen might have given this shirt to John [presupposing there is inferential evidence that Gwen gave this shirt to John], that Leiwa might have given this shirt to John [presupposing there is inferential evidence that Leiwa gave this shirt to John], that Holly might have given this shirt to John [presupposing there is inferential evidence that Holly gave this shirt to John], ...} = {p : ∃x[p = that x might have given this shirt to John [presupposing there is inferential evidence that x gave this shirt to John]]} The conjoined presupposition of (4.78) is that there is inferential evidence that Gwen gave this shirt to John, and there is inferential evidence that Leiwa gave this shirt to John, and there is inferential evidence that Holly gave this shirt to John, and so on. We can now connect this to the claims in table 4.5 about speaker and addressee knowledge: a speaker who utters a question (not knowing the answer) but at the same time makes explicit that she believes the evidence is utterly mixed (even contradictory), is indicating her belief that the hearer is not in a position to answer the question, and nor is an answer required of the addressee. However, there is an additional pragmatic angle worth considering in CQ sentences such as (4.78). Consider a slightly different context where the speaker requires an answer. In this case, it would be simpler and more succinct for the speaker to simply utter a regular OQ, which requires an answer from the addressee in order for the discourse to be felicitous. CQs are more complex constructions than OQs, and by using an evidential in a question, a speaker is implicating that the speaker was not in a position to utter an OQ, and thus that the hearer is assumed to lack an answer to the question.  201  4.5. Modal =ima and Conjectural Questions  4.5.3  Conjectural Questions in Other Languages, and with Other Evidentials  The analysis presented in this section predicts that CQs should be a wider phenomenon cross-linguistically. Littell et al. (2010) report the same effects in both St’át’imcets in (4.79) and NìePkepmxcín (Salish) in (4.80): when an evidential is inserted into a question it creates an evidential question:41 (4.79) St’át’imcets a. lán=k’a kwanen-s-as already=infer take.redup-caus-3.erg ni=n-s-mets-cál=a det.abs=1sg.poss-nom=write-act=exis ‘She must have already got my letter.’ b. lán=ha kwanen-s-as already=ynq take.redup-caus-3.erg ni=n-s-mets-cál=a det.abs=1sg.poss-nom=write-act=exis ‘Has she already got my letter?’ c. lan=as=há=k’a kwanen-s-as already=3.sbjn=ynq=infer take.redup-caus-3.erg ni=n-s-mets-cál=a det.abs=1sg.poss-nom=write-act=exis ‘I wonder if she’s already got my letter.’ ‘I don’t know if she got my letter or not.’ (4.80) NìePkepmxcín a. y’e-mín-s=nke e=Meagan e=ti good-rel-3.sub=infer det=Meagan det=tea ‘Meagan must like the tea. / Apparently, Meagan likes tea.’ 41  See also Littell (2010) for a detailed discussion of other evidentials in questions, and NìePkepmxcín.  202  4.5. Modal =ima and Conjectural Questions b. kéP k=s=y’e-mín=s e=Meagan e=ti whether irl=nom=good-rel=3.poss det=Meagan det=tea ‘Does Meagan like the tea?’ c. kéP=ws=nke k=s-y’e-mín-s e=Meagan whether=sbjn=infer irl=nom-good-rel=3.erg det=Meagan e=ti det=tea ‘I wonder whether Meagan likes the tea.’ The conjectural/inferential evidential is not the only evidential to appear in questions, but it is the only one to have this ‘I wonder’ effect. Both =ima and =kat have the same semantics, differing only in the formulation of their presuppositions, yet =kat does not have the same effect in questions as =ima does. Reportative questions, for example, are straightforward questions meaning something like ‘Have you heard ...?’. An example of this is found in (4.81): (4.81) Context: You and a friend are taking the overnight bus to Prince George. You can’t remember what time you arrive, but your friend booked the tickets and she might know. a. gaxgu dim bagwim’ ’ kaxgwi tim pakw-m when fut arrive.pl-1pl ‘When is it we’ll get there?’ ’ dim bakw-m’ b. gaxgugat ’ kaxgwi=kat tim pakw-m when=rep fut arrive.pl-1pl ‘When is it (did they say/did you hear) we’ll get there?’ c. silkwsax t’aahlakw=kat silkwsax t’aahlakw=kat noon tomorrow=rep ‘(I heard/They said) at noon tomorrow.’  203  4.5. Modal =ima and Conjectural Questions The analysis above predicts that these questions would introduce conjoined presuppositions, too, to the effect that there is mixed or contradictory reportative evidence, in the same way that conjectural questions introduce a conjoined presupposition that there is mixed or contradictory conjectural evidence. However, in none of these reportative questions does there appear to be any meaning akin to ‘reports are mixed’, nor does there appear to be any resulting signal that the speaker does not expect the addressee to be able to answer. However, Murray (2009b) reports that in Cheyenne it appears to be the reportative, rather than the conjectural, that has this effect. Why this same effect would be caused by different evidentials in different languages is another issue for future research. The account above will thus need to be refined and expanded to properly account for evidentials other than the conjectural evidential, with the eventual goal of accounting for Conjectural Questions as a unified phenomenon. One possible direction to pursue is focusing on the meanings of individual evidentials and taking into account their paradigmatic relation to one another in terms of specific kinds of inferential evidence they encode. For example, in Gitksan both the modal =ima and reportative =kat are inferential evidentials, but =kat encodes a more specific kind of inferential evidentiality, specifically that the inferential evidence must be a report. The suggestion here is that =kat cannot convey a ‘wonder’ interpretation when put into a question because the kind of evidential information encoded by =kat is too specific to allow for any kind of controversy. In other words, only the ‘weakest’ evidential can be used in a CQ, and this is the more general =ima. ’ Evidential nakw is also more specific than =ima in that it encodes sensory evidence. ’ It is also therefore worth observing that nakw does not have this effect in questions. The ’ analysis of nakw in the next chapter shows that its evidential meaning is also encoded ’ by presupposition. Yet, as with =kat, questions with nakw cannot have a ‘wonder’-type translation, as (4.82) shows:  204  4.6. Summary (4.82) Context: You can smell cigarette smoke coming from outside. You can see Jason standing outside (but not the actual cigarette). ’ *nakwhl ’nakw =hl evid=cnd “Jason must = “I wonder  ’ xmiyeenis Jasona ’ x-miyeen-(t)=s Jason=a consume-smoke-3sg=pnd Jason=interrog be smoking.” if Jason smokes.”  ’ However, (4.82) is ungrammatical as a result of the restriction on nakw from being a part of any kind of interrogative construction, thus we are unable to gain any further ’ insight from nakw into this issue.  4.6  Summary  Three issues concerning the semantics of evidentials were addressed in this chapter. The first involved the semantics of the evidentials =ima and =kat. The tests in chapter 3 showed that both =ima and =kat are propositional operators. I showed that both =ima and =kat are modal evidentials: they are quantifiers over possible worlds. The analyses of =ima and =kat are exactly parallel; the only difference resides in the definedness condition, which for =ima requires that the modal base contain those worlds in which the inferential evidence in w holds, and for =kat requires that the modal base contain those worlds in which the reported evidence in w holds. A modal analysis of =ima and =kat led to the second issue addressed here: the variable modal force of =ima and =kat. I attribute the variable modal force not to variable quantification, which is fixed as existential, but in the ordering source. The weak/strong interpretations of =ima and =kat correspond to empty/non-empty ordering sources respectively. I extended this to St’át’imcets, where the weak/strong interpretations of modals also correspond to non-empty/empty ordering sources, but over a universally quantified modal base. The choice function analysis of Rullmann et al. can be reduced  205  4.6. Summary to an ordering source one, and what this gives us is a unified account and a theoretical typology of languages in which modal forces vary under a fixed quantifier. Thirdly, I examined the effect =ima has when added to a question, in what are called Conjectural Questions. It was shown that we can attribute the lessened interrogative force that characterizes a CQ to the independently needed presuppositions associated with =ima. The presuppositions introduced by =ima are carried by each proposition in the question denotation, and conjoin with each other. The resulting conjoined presupposition entails that there is mixed evidence about the question at hand. Thus, the presupposition of mixed evidence functions to indicate reduced confidence on the speaker’s part that the hearer is in a position to know the true answer. The outcome is a reduced interrogative force for CQs: not only is the hearer not required to answer, the speaker is encoding that the hearer is probably not able to answer.  206  Chapter 5 The Pragmatics of Evidentiality in Gitksan In chapter 3 a battery of four diagnostics was applied that tested which level of meaning an evidential operates on. These tests revealed that both =ima and =kat operate at the propositional level, and in chapter 4 both =ima and =kat were analyzed as epistemic modals whose evidential meaning is encoded through presupposition. However, ’ these same tests show that the sensory evidential nakw diverges from the modals =ima ’ and =kat in every respect: nakw cannot be a propositional operator. This indicates ’ that nakw operates above the proposition at the illocutionary level. These results are summarized in table 5.1, repeated from chapter 3 for convenience: Gloss  Evidence type  Level of Meaning  =kat  reportative (rep)  Reportative  Propositional  =ima  modal (mod)  Inferential  Propositional  ’ nakw  evidential (evid)  Inferential – Sensory  Illocutionary  Table 5.1: The modal vs. illocutionary status of the evidential system in Gitksan  The goal of this chapter is to give a formal account of the range of meanings and ’ uses of nakw, and how it fits into the system of evidentiality and epistemic modality in Gitksan.  207  5.1. The Issues  5.1  The Issues  ’ In order to adequately explain the meaning of nakw, there are four specific issues that ’ have to be addressed. The first issue involves accounting for the core meaning of nakw in ’ terms of its felicitous use in discourse. The evidential meanings of nakw were presented ’ in chapter 2: sentences of the form nakw (p) require that a speaker have indirect sensory ’ evidence for inferring p. In other words, nakw is infelicitous in contexts that involve conjecture, or a speaker’s inference for p based on past experiences, as example (5.1) shows: ’ (5.1) nakwt ’ nakw =t evid=3pl “The dogs  hlakhlakshl hlak-hlaks-(t)=hl pl-scratch-3sg=cnd must’ve scratched the  as’oshl haahlxan haahlxan as-os=hl pl-dog=cnd wall wall.”  Felicitous (sensory evidence): You see scratch marks on the outside wall of your friend’s house that weren’t there last night. Infelicitous (past experience): Your friend tells you about the scratches on the wall outside her house. You know that her dogs get really restless when they’re tied up for too long, and maybe they did this because they were anxious.  ’ Secondly, evidence was presented in chapter 3 that showed nakw patterns morphosyntactically in a similar way to an auxiliary verb that takes a complement clause. ’ However, unlike verbs, nakw cannot undergo any of the usual operations expected of verbs: it cannot be embedded or questioned. Further to this, evidence is presented ’ below that shows nakw is in complementary distribution with other sentential forces, ’ such as assertions, questions and imperatives. Does this mean that nakw signals a novel ’ sentence type? If nakw (p) sentences are not asserted, what is their illocutionary status? ’ Thirdly, despite the diverging characteristics of nakw from modal =ima and repor’ is nonetheless an indirect evidential. Thus, we expect it to interact tative =kat, nakw with the other indirect evidentials in Gitksan in some systematic way. This is indeed the ’ case: it was shown in chapter 2 that =ima and nakw stand in relation to one another: in 208  5.1. The Issues contexts where a speaker makes an inference based on speculation or experience, =ima may be interpreted as either must or might, as (5.2) (5.2) Context: Your friend tells you about the scratches on the wall outside her house. You know that her dogs get really restless when they’re tied up for too long, and maybe they did this because they were anxious. as’oshl hlakhlakstiidimahl as-os=hl hlak-hlaks-tiit=ima=hl pl-scratch-3pl=mod=cnd pl-dog=cnd “The dogs might’ve/must’ve scratched  haahlxan haahlxan wall the wall.”  However, example (5.3) shows that when there is sensory evidence available in the ’ context, nakw takes over the interpretation of must, and =ima is relegated to meaning only might: (5.3) Context: You see scratch marks on the outside wall of your friend’s house that weren’t there last night. haahlxan a. hlakhlakstiithl as’oshl hlak-hlaks-tiit=hl as-os=hl haahlxan pl-scratch-3pl=mod=cnd pl-dog=cnd wall “The dogs might’ve scratched the wall.” ’ b. nakwt ’ nakw =t evid=3pl “The dogs  hlakhlakshl as’oshl haahlxan hlak-hlaks-(t)=hl as-os=hl haahlxan pl-scratch-3sg=cnd pl-dog=cnd wall must’ve scratched the wall.”  ’ Modal =ima and non-modal nakw are both inferential evidentials, however, they operate on different levels of meaning. Thus, the leading question is how do linguistic objects that operate on different levels of meaning interact in this way? ’ Lastly, nakw has two additional interpretations that draw upon its evidential status. In chapter 3 §3.6.1.1, the Known Truth/Falsity Test showed that modal =ima is infelicitous if the speaker knows the truth or falsity of the proposition embedded under the 209  5.2. The Plan ’ evidential. Under normal conversational use, when sentences of the form nakw (p) are used in a conversation, the speaker is claiming that they have indirect sensory evidence ’ for p. However, when nakw (p) is used in contexts where the speaker knows p is true, ’ usually through the direct observation of the event of p, nakw has a mirative meaning. There is also another effect associated with the opposite arrangement: when a speaker ’ ’ knows that p is false, nakw has a non-literal meaning. Why does nakw have this meaning when other sentences, including regular assertions and =ima(p) sentences, cannot convey these meanings?  5.2  The Plan  Because all of these issues are interconnected and directly involve the pragmatic meaning ’ ’ of nakw, I approach them incrementally in the following way: first, in §5.3, nakw is examined in terms of how it functions in conversation. All of the evidence presented so ’ far indicates that nakw is highly dependent on the local context of its use. Dynamic ’ semantics provides an effective way of explaining the use of nakw in terms of its context ’ change potential, or the ‘instructions’ nakw gives as to how to update the Common Ground. Just as we saw in the denotations of modal =ima and =kat in the previous chapter, this chapter shows that presupposition is also crucial in determining the felicity ’ of nakw in a conversation. Whereas =ima and =kat have semantic presuppositions ’ (definedness conditions on a proposition) that restrict the modal base worlds, nakw carries a pragmatic presupposition which places a condition on its use in a conversation: ’ the felicitous use of a nakw (p) sentence requires that sensory evidence is entailed by the Common Ground. Whereas semantic presuppositions are treated formally as conditions on the definedness of a proposition, and are characterized as constraints on the discourse context, pragmatic presuppositions are beliefs about the context that must be attributed to a speaker. ’ While this explains the use of nakw in conversation in terms of its context change ’ ’ potential, we still require an account of nakw in terms of its sentential force. If nakw (p) 210  5.2. The Plan sentences are not assertions, then what exactly do they contribute to the conversation? ’ ’ §5.4 refines the Dynamic Semantics of nakw by proposing that nakw is a novel sentential force operator. Portner (2006) suggests an enrichment of Dynamic Semantics to include other sets of propositions distinct from the common ground. For example, the sentential force of a declarative is assertion: a proposition is added to the common ground, the set of propositions representing the shared assumptions of the discourse participants. ’ Under this analysis, the sentential force of nakw adds a proposition to the inferential set: the set of propositions a speaker has inferential evidence for. ’ A Dynamic semantics analysis of nakw in terms of pragmatic presupposition allows us ’ to place the sensory evidence presupposition associated with nakw into a relation with the semantic presupposition associated with =ima. This is facilitated by the notion that semantic presuppositions give rise to pragmatic presuppositions within a discourse context (von Fintel 2005; Simons 2006). We then have a way of explaining why, in ’ sensory evidence contexts, nakw assumes that strong must-like translation, while =ima ’ takes the weaker might-like reading: the sensory evidence presupposition of nakw is ’ more specialized than the indirect evidence presupposition of =ima, therefore nakw blocks =ima from the strong reading. ’ The other robust features of nakw, its mirative and non-literal meanings, are also ’ rooted in the use of nakw in conversation. §5.6 presents an analysis of the mirative ’ as conversational implicature. More specifically, a speaker making a nakw (p) statement when they know p to be true is flouting the Maxim of Quantity: a speaker is contributing more to the discourse than is required for the purposes of conversation. This flout implicates that a speaker is surprised or otherwise unprepared for the event that p ’ denotes. The flipside is metaphor: a speaker making a nakw (p) statement when they know p to be false is flouting the Maxim of Quality.  211  ’ 5.3. The Dynamic Semantics of nakw  5.3  ’ The Dynamic Semantics of nakw  A static semantics equates the meaning of a sentence with its truth conditions: the conditions in which a sentence is true or false. This is the kind of semantics that was presented in analyzing modal =ima and reportative =kat. Dynamic semantics involves the meanings of sentences in contexts. More specifically, the meaning of a sentence is its context change potential, or the way in which it changes a discourse context (Heim 1982, 1990). Thus, the interpretation of a sentence is dependent on context. For example, (5.4) contains two different sequences of two sentences: (5.4)  a. A boy walks on to the basketball court. He is dribbling a basketball. b. ? He is dribbling a basketball. A boy walks on to the basketball court.  The sequencing of these sentences affects their felicity. We can understand the referent of the pronoun he in (5.4)a. if it is preceded by a noun phrase, such as a boy, which may serve as its antecedent. Reversing these sentences produces an odd discourse as in (5.4)b., or at least one in which the pronoun has to be resolved differently. This phenomenon reflects a sensitivity sentences have to discourse; informally this would be that the indefinite noun phrase may set up a discourse referent which can be referred back to by a following pronoun. Context can be built into a static semantics in terms of parameters of interpretation. Indeed, the context is necessary for the interpretation of a modal in English, as we saw in the previous chapter how the context provides modal bases and ordering sources of different varieties that determine the interpretation of a modal. However, a dynamic semantics systematically investigates how the interpretation of an expression changes the context, creating a new context out of the old one, and thus affects how subsequent expressions in a discourse are interpreted. There are a variety of linguistic phenomena which require a dynamic account of interpretation that is provided by the theory of dynamic semantics. These include anaphoric pronouns such as in example (5.4), temporal sequencing, and presuppositions 212  ’ 5.3. The Dynamic Semantics of nakw of various kinds. It is the last of these that is relevant to the pragmatic analysis of ’ nakw and its function in discourse. We saw the use of semantic presupposition in the denotations of modal =ima and reportative =kat in the previous chapter. As this section ’ shows, it is pragmatic presupposition that is relevant in determining the felicity of nakw in discourse.  5.3.1  Pragmatic Presupposition  One of the main drivers of dynamic semantics is presupposition. There are two main views on presupposition in the literature, and how presupposition fits into general theories of meaning (Soames 1989): (i.) Semantic Presupposition: Sentences have two components: their propositional content, and their presuppositional content. A semantic presupposition is a necessary condition for the proposition to have a truth value: q is a semantic presupposition for p iff q is true whenever p is true or false. (ii.) Pragmatic Presupposition: Presuppositions are admittance conditions for sentences into a context (Stalnaker 2002; Heim 1982, 1990). One important assumption is that semantic presuppositions become the pragmatic presuppositions of speakers, as speakers should believe that contexts satisfy the conditions required to allow their utterances to have a truth value (Simons 2006). In other words, the semantic presuppositions of the propositions that are expressed by the sentence that is uttered also have to be pragmatic presuppositions of the utterance. This ’ will become an important assumption when the interaction between nakw and =ima is examined in §5.5. The notion of pragmatic presupposition is rooted in the theory of presupposition of Stalnaker (2002; 1974):  213  ’ 5.3. The Dynamic Semantics of nakw A proposition P is a pragmatic presupposition of a speaker in a given context just in case the speaker assumes or believes that P , assumes or believes that his addressee assumes or believes that P , and assumes or believes that his addressee recognizes that he is making these assumptions, or has these beliefs (Stalnaker 1974, 573). In other words, a speaker presupposes P just in case she believes that P is in the common ground. This places a constraint on possible discourse contexts in which sentences may be felicitously uttered. Two core components are defined in (5.5) (Stalnaker 1974; Heim 1990): (5.5) (i.) Common Ground (CG): The set of propositions which the participants in a conversation agree to be uncontroversial for the purposes of conversation. (ii.) Context Set (C): The set of worlds in which every proposition in the Common Ground is true. The notions of the Common Ground and Context Set provide us with the formal tools talk about the state of a conversation at a given point in time. These may include previous shared history of the participants, world knowledge, the purpose of the discourse and so on. For example, suppose in a particular conversation the participants have agreed on three facts: it’s raining, John’s truck is in the driveway, and Bill is hungry. We assume they know these things because they can see that they are true, or perhaps because someone may have previously brought them to the discourse by asserting them. Thus a sample CG would be as in (5.6): (5.6) CG= {the proposition that it’s raining; the proposition that John’s truck is in the driveway; the proposition that Bill is hungry} The Context Set C is then the set of possible worlds in which it’s raining and John’s truck is in the driveway and Bill is hungry: 214  ’ 5.3. The Dynamic Semantics of nakw (5.7) C = the set of worlds in which it’s raining and John’s truck is in the driveway and Bill is hungry. Using this notion we can explain the idea that presuppositions are conditions on whether a sentence is admissible into a conversation. In general, the theory of pragmatic presuppositions states that a presupposition is a statement of the form “Sentence S can only be used in C if ...”. For example, sentences with the particle too attached to them require that there is some previous discourse-relevant sentence of the same meaning denoting the same state or activity etc. This can be readily captured by presupposition: (5.8) For any Context Set C, “Alvin’s truck is in the driveway too” can be used in C iff C entails that some other person’s truck is in the driveway. Given the Common Ground in (5.6), the sentence “Alvin’s truck is in the driveway too” is felicitous, since the Context Set illustrated in (5.7) entails that someone else’s truck, namely John’s, is in the driveway.  5.3.2  Context Change Potential  What this gives is a way of examining the use of a sentence based on its context change potential. This is incrementally driven by the rule that allows us to add a sentence into an evolving conversation. In a nutshell when you update a Context Set C with S, all those worlds in which S is not true are discarded from C. This is captured by the rule in (5.9): (5.9) For any Context Set C and sentence S, C + S = C ∩ S . (adapted from Heim 1990) C ∩ S is the intersection of C with the proposition expressed by S. Since the Context Set is defined as the intersection of the propositions in the Common Ground, this is equivalent to saying that the proposition expressed by S is added to the common  215  ’ 5.3. The Dynamic Semantics of nakw ground. In other words, it becomes a proposition which the participants in a conversation take as uncontroversial. Using the toy conversation from above, we can imagine an earlier stage in the conversation where the proposition that John’s truck is in the driveway was added to the Context Set by following the rule in (5.9). The effect of this is that all the worlds in which John’s truck is not in the driveway are discarded from C, giving us an updated context C∗. The sentence “Alvin’s truck is in the driveway too” is next to be added to C∗ through the application of (5.9). However, this sentence has a presupposition attached to it, lexically represented by the particle too. This presupposition requires that the sentence can only be added to C∗ if C∗ entails that some other person’s truck is also in the driveway. C∗ in fact does entail this, and the sentence “Alvin’s truck is in the driveway too” can be added to C∗.  5.3.3  Presupposing Sensory Evidence and the Common Ground  ’ Before we can apply this kind of analysis to evidential nakw, we need to make more precise how sensory evidence can be pragmatically presupposed. When a dog walks into a backyard where a conversation is taking place, he becomes salient in the context. The consequence of the fact that this dog walked in is that it becomes part of the common ground, simply because the discourse participants can see that it’s true. The immediate consequences of this depend on the context – for example, if it already was common ground beforehand that everything in the vicinity (i.e. house, yard etc.) belongs to Alvin, now it is common ground that Alvin has a dog. This can apply to other senses: Imagine sitting in Alvin’s living room and you hear a child crying in the next room. In this same context, it is now common ground that Alvin has a child. One prediction made by modeling this sensory evidence requirement as a presupposition, is that contexts which lack sensory evidence should result in a presupposition failure. However, as is commonly the case, these requirements can be accommodated 216  ’ 5.3. The Dynamic Semantics of nakw within certain limits. Accommodation is the most flexible when dealing with subjective judgments, as in the case in (5.10): (5.10) Context: Your friend laughs at you when you tell her you won $1000 at bingo last night. ’ nawimi xsingy’ ’nakw =mi xsink-y’ evid=2sg disbelieve-1sg “I can see that you don’t believe me.” ’ However, more objective statements using nakw are less easily accommodated, with the potential result of presupposition failure. This presupposition failure effect can be inferred from the consultant’s comments in example (5.11): (5.11) Context: You’re chopping wood out by the smokehouse. ’ nakwhl sehons Bob ’nakw =hl se-hon-(t)=s Bob evid=cnd caus-fish-3=cnd Bob “Bob must be smoking/preparing/doing up fish.” Comment: Really? Can you smell something or see smoke by coming out of his wilpsehon (smokehouse)?  The sentence in (5.11) presupposes that the context entails a proposition about some physical evidence in the discourse situation. Since we imagine (5.11) to be uttered in a context which does not entail such a proposition, and since contexts are transparent (and hence, if p is not entailed by the context then the proposition that the context doesn’t entail p is), accommodation is not possible. ’ The notion of accommodation places certain requirements on the hearer: nakw always has the effect of drawing a hearer’s attention to any possible source of sensory evidence, be it visual, tactile, or auditory. We can contrast this effect with the evidential modal =ima. Recall from chapter 4 that =ima presupposes that the modal base 217  ’ 5.3. The Dynamic Semantics of nakw contain only worlds where there is inferential evidence for p. This is a weaker presuppo’ sition than the one associated with nakw. This would suggest that it is accommodated more easily. For example, someone may utter (5.12) in a conversation about berry ’ picking: nakw is infelicitous in the context, while =ima is fine: (5.12) Inference from general knowledge: You’re sitting at home talking about going berry-picking. It’s August, and the berries are usually ripe this time of year on the Suskwa. ’ a. #nakwhl mukwhl maay’ ’nakw =hl mukw=hl maay’ evid=cnd ripe=cnd berries “The berries must be ripe.” b. mugwimahl maay’ mukw=ima=hl maay’ ripe=mod=cnd berries “The berries might/must be ripe.” ’ The presupposition of nakw is more specific, requiring sensory evidence in the context that must be in the common ground and hence available to speaker and hearer. This cannot be as easily accommodated. Modal =ima merely presupposes that there is a certain kind of evidence in the modal base. Since this is about the belief-state of the speaker, it can more easily be accommodated. Intuitively speaking, the hearer may not have the evidence themselves, but is willing to take the speaker’s word for it. The theoretical issue here concerns how to tie these facts to the difference between pragmatic and semantic presupposition. I suggest that a pragmatic presupposition is a presupposition on the utterance context, and therefore tied to the present state of the conversation, or the “here and now”.42 It is a presupposition on sensory evidence that is in the utterance context and available to all the interlocutors. 42  See Cook (2008) for a related phenomenon with Cree indexicals.  218  ’ 5.3. The Dynamic Semantics of nakw  5.3.4  ’ The Context Change Potential of nakw (p)  We can use this theory to examine the Context Change Potential of sentences of the form ’ ’ nakw (p) in Gitksan. Here it is claimed that by making a statement nakw (p), a speaker pragmatically presupposes that he or she has sensory evidence for p at the utterance ’ time. More specifically, the addition of nakw (p) to a Context Set C requires that C contain only worlds where the proposition that the speaker has sensory evidence for p is true. For example, when a speaker utters (5.13), there must be some sensory evidence in the common ground that John is home, and a statement of the sentence in example (5.14) presupposes the speaker has some sensory evidence for claiming the hearer has a cut hand: (5.13) Context: You need to ask John a favour, so you and a friend drive by John’s place to see if he’s home. ’ nakwhl ta’as John ’ nakw =hl ta’a-(t)=s John evid=cnd at.home-3=pnd John “John must be home.” “Looks like John’s home.” presupposition: There is visual evidence publicly available in the utterance context (John’s truck is in the driveway). p: John is at home.  (5.14) Context: You and a friend are fishing. You’re sitting on the rocks, cutting up bait. You notice blood on the rocks at your friend’s feet. ’ ’onin nakwhl gotshl ’ nakw =mi kots-(t)=hl ’on-n evid=2sg cut-3=cnd hand-2sg “You must’ve cut your hand.” “I see you cut your hand.” presupposition: The speaker has visual evidence (blood on the rocks). p: You cut your hand.  219  ’ 5.4. nakw as a Sentential Force Specifier An important claim made here is that the common ground must provide sensory ’ evidence that is both interpretable by the speaker making a nakw -sentence, and available ’ to the hearer in assessing the sensory evidence presupposition attached to nakw. The infelicity of (5.11) can be attributed to the fact that the speaker does not have access to the sensory evidence – which is specifically targeted in their response – that would ’ license the presupposition attached to the nakw -sentence, nor does a hearer have access to any sensory evidence that may lead them to accommodate such a presupposition. ’ A dynamic semantics analysis of nakw also correctly accounts for the temporal con’ straints placed on nakw statements such as (5.13) – (5.14). For example, imagine extending the context in (5.13), where you and a friend continue driving past John’s house without stopping, and return home. Even though you saw earlier (sensory evidence) that John’s lights were on, uttering (5.13) would result in a presupposition failure: the common ground at that point no longer has any sensory evidence present in the discourse context.  5.4  ’ nakw as a Sentential Force Specifier  As outlined in §3.4.1, within classical speech act theory utterances are treated as consisting of an illocutionary force, F , and a separate level of propositional content, p, such that F and p together, or F (p), form a complete utterance used to accomplish a speech act (Searle 1969: 49-50). For example, a sentence with the illocutionary force of assertion adds a proposition to the common ground. However, we also need to take into account a sentence’s form independently of illocutionary force. Some of the more common sentence forms or types typically include declaratives, imperatives, and interrogatives. For example, both sentences in (5.15) are declaratives: (5.15)  a. Sammy the dog is cute. b. I’d like for you to sit down now.  220  ’ 5.4. nakw as a Sentential Force Specifier However, these sentences differ in illocutionary force: (5.15)a. is a declarative sentence with the illocutionary force of an assertion. While the sentence in (5.15)b. has a similar form to a., its illocutionary force is that of requesting, or perhaps demanding. Likewise, the sentential force of interrogatives is asking, and the sentential force of imperatives is requiring (Portner 1997). In many languages there are also other sentence types, including exclamatives and promissives. When a declarative sentence is uttered with assertive illocutionary force, the common ground is updated by applying its context change potential (a function from contexts to contexts) to the current common ground. Context change potential is a property of the sentence, whereas illocutionary force is a property of an utterance of a sentence. The context change potential of a sentence is only “realized” if the sentence is uttered assertively (i.e. with assertive illocutionary force).  5.4.1  A Case for Evidential Sentential Force  Sentential force is often encoded by a sentence’s form, and this is partially the case in Gitksan: the sentences in (5.16) have the same predicates, but there are three different sentential forces which also correspond to three different syntactic and morphological clause types: Declarative, Interrogative, and Imperative: (5.16)  a. jabis tsap-i-(t)=s cook-tr-3=pnd “Sheila cooked the  Sheilahl hon Sheila=hl hon Sheila=pnd fish fish.”  Declarative[Sheila cooked the fish] c. naa ’ant japhl hon naa ’an-t tsap-(t)=hl hon who s.rel-3 cook-3=cnd fish “Who cooked the fish?” Interrogative[who cooked the fish]  221  ’ 5.4. nakw as a Sentential Force Specifier b. jabnhl hon tsap-n=hl hon cook-2=cnd fish “Cook the fish!” Imperative[(you) cook the fish ] There are two distinct clause types in the Tsimshianic languages: what Rigsby (1986) calls the Independent, and the Dependent. Generally speaking, Independent clauses have a canonical VSO word order. Declaratives such as (5.16)a. are Independent clauses. Dependent clauses are clauses where a constituent other than the verb, such as a whelement, is at the front of the clause. This element is followed then by the verb, subject and object. Interrogative sentences such as (5.16)b. are always Dependent clauses. The typing of a clause as Independent or Dependent is accompanied by a variety of other morphological alternations. The two notable features of Dependent clauses are the second position ergative clitics (replaced by a common noun determiner in intransitive sentences), and object or intransitive subject independent pronouns replaced by agreement on the verb. There is a third clause type that I call the Imperative, which corresponds to the imperative sentential force of (5.16)c. Imperative clause types can be characterized as ‘half’ Independent (they have the Independent, verb-initial, VSO word order), and half Dependent (they have ergative morphology, which is only found in Dependent clauses). Independent clauses are always declarative, and typically have the illocutionary force of an assertion. However, Independent declarative clauses can also have other illocutionary forces, such as promising and warning. On the other hand, Dependent clauses are not exclusive to any sentential or illocutionary force. Example (5.16)b. is a Dependent clause with interrogative sentential force. However, the progressive and negation are both Dependent clauses, as in (5.17), yet their illocutionary force is assertion:  222  ’ 5.4. nakw as a Sentential Force Specifier (5.17)  a. yukwt jabs Sheilahl hon yukw-t tsap-(t)=s Sheila=hl hon prog-3sg cook-3sg=pnd Sheila=cnd fish “Sheila is cooking the fish.” b. neediit jabs Sheilahl hon nee=tii-t tsap-(t)=s Sheila=hl hon neg=contr-3sg cook-3sg=pnd Sheila=cnd fish “Sheila didn’t cook the fish.”  ’ As with interrogatives, the progressive and negation, sentences with nakw are always Dependent: ’ (5.18) nakwt jabs Sheilahl hon ’nakw -t tsap-(t)=s Sheila=hl hon evid-3sg cook-3sg=pnd Sheila=cnd fish “Sheila must’ve cooked the fish!” “I see Sheila is cooking fish!” “Looks like Sheila is cooking fish.” ’ The fact that nakw introduces a Dependent clause reveals nothing specific about its sentential force. Thus, the main task of this subsection is to identify the sentential force ’ ’ of nakw sentences. I claim that nakw clauses have their own sentential force, or what I call an ‘evidential sentential force’. The function of evidential sentential force is to add ’ to the set of discourse propositions for which a speaker has inferential evidence. nakw sentences do not have the illocutionary force of assertion – they do not add a proposition to the common ground – but are similar to other sentential forces: their illocutionary force is largely determined by the context of use. ’ Supporting evidence that nakw sentences are not assertions comes from its characterization by consultants. Twelve Gitksan consultants were asked to characterize a ’ nakw sentence in a typical sensory evidence context. For example, in (5.18) this context is one where a speaker sees the empty cans of ts’al (dried sockeye salmon) and smells fish cooking in a pot in the kitchen. Aside from its evidential meaning, which ’ speakers agree on unanimously, speakers variously describe nakw as sounding similar to 223  ’ 5.4. nakw as a Sentential Force Specifier asking a yes/no question, emphasizing something, surprise, negating (in the sense of a negative tag question: “Sheila didn’t actually cook the fish, did she?”) or conveying a sense that something they thought couldn’t actually happen actually did happen (this ’ characterization relates to a mirative meaning of nakw, discussed in §5.6). ’ Recall from §3.6.2.2 that nakw cannot participate in any kind of interrogative state’ ment, including a yes/no question. This indicates that nakw sentences cannot have the sentential force of an interrogative, as example (5.19) shows: (5.19)  ’ Jasona a. yukhl xmiyeenis ’ yukw=hl x-miyeen-(t)=s Jason=a prog=cnd consume-smoke-3sg=pnd Jason=interrog “Is Jason smoking?” ’ ’ b. *nakwhl xmiyeenis Jasona ’ ’ nakw =hl x-miyeen-(t)=s Jason=a evid=cnd consume-smoke-3sg=pnd Jason=interrog = “Jason must be smoking!?”  ’ Finally, there are also two other robust characterizations of nakw made by consul’ tants: while nakw expresses a fairly high degree of certainty as to the truth of the embedded predicate, the speaker is not certain of it.  5.4.2  Evidentials as Sentential Force Specifiers (Portner 2006)  ’ Section 5.3 sketched out a dynamic semantics for nakw, which presupposes that a speaker ’ have sensory evidence in the utterance context for a nakw -statement. Portner (2006) adapts the theory of conversational update and introduces two modifications to the classical model: the first is that the common ground is only one of potentially numerous sets of propositions. These other sets represent propositions that encode cognitively or communicatively important categories, such as the different subtypes of evidentiality. Evidentials under this approach are sentential force operators that specify a kind of conversational update: they can function to conversationally update the common ground, the set of mutually believed propositions, or to update specialized sets of propositions. 224  ’ 5.4. nakw as a Sentential Force Specifier Portner’s second modification is the introduction of a wider set which contains the common ground set and all of the other specialized sets: the ‘presented’ set ps. Portner characterizes the ps as the set of propositions of which the participants are mutually aware. The common ground is a crucial subset of ps, but so are other subsets. We must keep track of subsets of ps representing propositions which belong to different categories. For example, we keep track of the propositions to which we have made additional commitments (e.g., the common ground) and about which we have additional kinds of information (e.g., those for which the speaker had indirect evidence). Under this analysis, simple assertion is conversational update where a speaker adds p to the common ground, or the specialized subset cg(ps) of the presented set ps. This offers an alternative to the classical speech act theory analysis presented by Faller (2002a) in accounting for the evidential meanings of the individual evidentials in Quechua. Faller uses utterance modifiers such as alas to motivate the addition of a class of illocutionary modifiers into classical speech act theory. This is done largely to account for the fact that sentences of the form ev(p) are not assertions in Quechua. For example, the reportative suffix -si has the illocutionary force of ‘presenting’ a proposition. (5.20) para-sha-n-si rain-prog-3-si p = “It is raining.” ill = present(p) sinc = ∃s2 [Assert(s2 , p) ∧ s2 ∈ / h, s]  (Faller 2002:199)  Under Portner’s analysis, the Quechua evidentials are not really illocutionary modifiers, but rather ‘sentential force specifiers’: they are grammaticized elements which specify precisely which kind of conversation update is to be performed. This is based on Faller’s insight, where there is a type of speech act with fewer commitments than assertion which can be modified in various ways to represent the different kinds of evidential meanings in a language (Faller 2002, 2003, to appear; von Fintel 2005).43 43  See Déchaine (2008) for a related analysis of evidentiality in Plains Cree, where sentences, without  any evidentials, have the default force of ‘presenting’ rather than asserting. Whereas with assertion a  225  ’ 5.4. nakw as a Sentential Force Specifier 5.4.2.1  The Model of Discourse  In Portner’s model of discourse, instead of the common ground as the basic construct, there is a wider set called the presented set: the set of propositions which the participants in a conversation are mutually aware of. Portner (2006) defines a model of discourse ds as a pair ps, F of a set of propositions ps and a tuple of selection functions F from ps to subsets of ps: F = f1 , ..., fn . Each f represents a cognitively and linguistically real way of categorizing a proposition which has been presented and accepted. One or more f ’s may be modified so that p ∈ f (ps). Every language has cg as an available f and in every language cg is the default f for propositions that are presented.44 However, the linguistic system of a language may have one or more grammaticized ways of indicating other types of f to update with a given sentence, and evidentials are one such way. In its simplest form, when a proposition is put into ps, F may also be modified as follows: (5.21)  1. Simple put: For any discourse structure ds = ps, F , put(ds, p) = ps , F , where  (Portner 2006: 9–10)  (a.) ps = ps ∪ {p}, and (b.) F is that sequence such that, for each f which is a component of F , f is replaced by f , where Dom(f ) = {ps ∪ {p}} and f (ps ∪ {p}) = f (ps). 2. put which changes a member of F : For any discourse structure ds = ps, F , and any f which is a component of F , there is a variety of put, putf , such that putf (ds, p) = ds as follows: Let α be the discourse component in ds corresponding to α in ds. Then ds is the minimal expansion of ds such that all of the following are satisfied: a. p ∈ ps b. for all selection functions g which are components of F , Dom(g) = {ps } speaker presents p as true, with the presentative a speaker simply presents p and source of information. 44 However, Déchaine (2008) claims that the cg is not the universal default.  226  ’ 5.4. nakw as a Sentential Force Specifier c. p ∈ f (ps ) d. Any further constraints imposed on discourse structures are satisfied. Given this technical implementation, evidentials can then correspond to varieties of put: for language L, a valid discourse structure has the form in (5.22): (5.22) dsL = psL , cgL , evid1L , ..., evidnL . 5.4.2.2  Meanings of the Evidentials as Update Functions  Portner models the evidential system of Quechua by treating the different evidentials as corresponding to varieties of putf , which are essentially the formal representations of individual evidential sentential forces. Given the definition in (5.22), a valid discourse structure for Quechua has the form in (5.23): (5.23) ds = ps, cg, bpg, report, conj  (cf. (5.22))  The sincerity conditions which Faller used to encode the evidential meanings of the different evidentials are defined by Portner as constraints on ds, given in (5.24): (5.24)  mi : bpgL (ps) ⊆ cgL (ps) si : {{w : ∃x[Say(x, p, w)]} : p ∈ reportL (ps)} ⊆ cgL (ps) chá: {♦p : p ∈ conjL (ps)} ⊆ cgL (ps)  However, the actual meanings of the individual evidentials are defined as update functions in (5.25) that correspond to the different varieties of put, the uses of which are subject to the constraints in (5.24): (5.25)  mi = λp.λds.putbpg (ds, p) si = λp.λds.putreport (ds, p) chá = λp.λds.putconj (ds, p)  227  ’ 5.4. nakw as a Sentential Force Specifier The absence of an evidential gives simple assertion: If S is a root node which denotes a proposition p, shift p to λds.putcg (ds, p). For example, a sentence of the form p-si must satisfy the constraint on the discourse that there is some speaker who asserted p, and that this speaker is neither the hearer nor the current speaker. There is no condition that the speaker believes p, and as such p is not added to the common ground, but to report(ps) (cf. (5.25)). This shows how the common ground is affected as a “side effect” of the speaker using a reportative evidential: for every proposition p that is in the report(ps), the proposition that “someone said that p” is added to the common ground. 5.4.2.3  ’ nakw as an Evidential Sentential Force Specifier  Portner’s analysis of evidentials as sentential force modifiers can be straightforwardly ’ ’ applied to nakw. This accomplishes two things: first, we can account for nakw as encoding a kind of sentential force, which is not assertion. Secondly, under Portner’s model, evidentials are sentential force modifiers, not illocutionary ones. This leaves us ’ open to account for the varying illocutionary forces of nakw purely through its context of use. ’ ’ If a speaker utters a sentence with nakw, this sentence expresses nakw (p). Because ’ ’ nakw (p) has been expressed, nakw (p) ∈ ps. The evidence shown above suggests that ’ we don’t want nakw (p) to automatically become a member of the common ground, cg(ps), as it is not asserted, yet we want to keep track of its inferential status in the conversation. Let infer(ps) be the subset of ps which was put forth on the basis of some ’ sensory evidence the speaker has. Thus, to put forth nakw (p) successfully as evidence for p, means p is added to infer(ps). ’ (5.26) The definition of nakw as an update function ’ nakw = λp.λds.putinf er (ds, p) ’ ’ A nakw sentence performs a double conversational update: in uttering nakw (p), p is added to ps and infer(ps). An evidential prejacent is not added to the common 228  ’ 5.4. nakw as a Sentential Force Specifier ground. Rather, what is added to the common ground is the fact that an evidential claim was made: “there is sensory evidence that p”. There is one minor difference made between the analysis presented here and Portner’s analysis of Quechua: Portner models the sincerity conditions of the different Quechua evidentials as general constraints on the discourse as in (5.24). However, I maintain ’ that the sensory evidence requirement of nakw is not a constraint on the discourse, but rather a pragmatic presupposition. The reasons for this involve the interaction between ’ ’ nakw and modal =ima in sensory evidence contexts, where nakw blocks =ima from the must-like reading. The analysis in the next section attributes this effect as a result of their respective presuppositions. Under this analysis, an evidential does not affect the propositional content of a sentence in the following sense: The highest phrase (the sister of the evidential) which denotes a proposition doesn’t incorporate the evidential, and this proposition is the one presented. Additionally, Portner shows that this analysis can also explain the assent/dissent facts: the challenge “That’s not true” does not have access to the speech act level of meaning, and therefore targets the denotation of the highest propositiondenoting constituent. 5.4.2.4  Interaction Between the Evidential Modals =ima and =kat  The evidential modals =ima and =kat can appear in any clause type. The only cooccurrence restriction that =ima and =kat have is in imperatives, with each other, ’ ’ and with nakw. Example (5.27) shows that nakw sentences with =ima or =kat are uninterpretable: (5.27)  ’ a. #nakwimahl siipxwin ’nakw =ima=hl siipxw-n evid=mod=cnd sick-2sg ’ a. #nagwgathl siipxwin ’ siipxw-n nakw =kat=hl evid=rep=cnd sick-2sg 229  ’ 5.4. nakw as a Sentential Force Specifier This co-occurrence restriction on evidentials can be explained by extending the sentential update theory to =ima and =kat. If a speaker has inferential evidence for a proposition, that proposition is added to the inferential set inf er(ps). This is achieved ’ by the use of nakw, because it has the sentential force of inferring. However, there is no ’ theoretical reason why inf er(ps) should be exclusive to nakw. We can use this assumption to account for the co-occurrence restriction in (5.27) in the following way: =ima(p) makes a double conversational contribution: =ima(p) is added to cg(ps) – it asserts its modal content – and p is added to inf er(ps) (following Portner’s 2006 analysis of Quechua chá). Reportative =kat(p) also makes a double conversational contribution: =kat(p) is added to cg(ps), and p is also added to inf er(ps). This gives a formal ’ way to rule these combinations out: nakw, =ima, and =kat ‘compete’ for inf er(ps). There are two additional predictions this pragmatic analysis of =ima and =kat makes. First, =ima and =kat should also compete against each other for inf er(ps), and this is what we find as sentences with co-occurring =ima and =kat are infelicitous: (5.28) #yugwgatit jabimas Sheilahl hon Sheila=hl hon yukw=kat-t tsap-(t)=ima=s prog=rep-3 cook-3=mod=pnd Sheila=cnd fish = “[I heard] Sheila might be cooking the fish.” The second prediction is that there should be some kind of interchangeability between ’ nakw, =ima, and =kat, so long as their presuppositions are satisfied by the context. It is not uncommon for sentences with =kat, as in (5.29), to be translated not as reportative (i.e. “I hear...”), but with a modal-like word such as apparently (this effect was also discussed in the previous chapter): (5.29) naksxwitgathl naks-xw-i-t=kat=hl marry-pass-tr-3sg=rep=cnd “Apparently Kathy married Bill.” “[I heard] Kathy married Bill.”  hanak tust tBill hanak tust t=Bill woman dem pnd=Bill (adapted from Hunt 1994: 117)  230  ’ 5.4. nakw as a Sentential Force Specifier The use of =kat presupposes the speaker has inferential evidence of the reported variety for the assertion p. This presupposition is satisfied in this context, thus, =kat(Kathy married Bill) is added to cg(ps) – it asserts the possibility according to the reportative evidence that Kathy married Bill – and Kathy married Bill is also added to inf er(ps). Portner’s analysis of Quechua showed that there is a specific set, report(ps), for those propositions that a speaker has reportative evidence for. This is not the case in Gitksan for two reasons: first, =kat is not a sentential force operator – it is an epistemic modal – therefore we predict there is no specialized report(ps) set. Secondly, this analysis allows the kinds of different translations of =kat in discourse, as (5.29) shows – =kat need not be translated only as “I hear...”. This same kind of interchangeability between =ima and =kat can be observed in discourse. Imagine a conversation fragment in (5.30), where several people are discussing the whereabouts of Alvin.45 GS asks: (5.30) GS: gaxguhl witxws Alvin? Alvin? kaxkwi=hl witxw=s when=cnd arrive=pnd Alvin “When is Alvin coming back?” LW: silkwsaxgat silkwsax=kat noon=rep “[I heard] noon.” In this context, LW heard from her sister earlier in the morning that Alvin would come back at noon. HW enters the discourse later and asks the same question: witxws Alvin? (5.31) HW: gaxguhl kaxkwi=hl witxw=s Alvin? when=cnd arrive=pnd Alvin “When is Alvin coming back?”  45  This artificial conversation was based on actual events and recreated with consultants.  231  ’ 5.4. nakw as a Sentential Force Specifier GS1: silkwsax silkwsax noon “noon.” GS2: silkwsaxima silkwsax=ima noon=mod “Maybe noon.” Because =kat(p) has updated both cg(ps) and inf er(ps), GS may reply with either an elliptical statement in (5.31) GS1, or with the inferential =ima in GS2. Portner predicted that discourse relations like anaphora and temporal sequencing should be able to hold within inf er(ps), just as they commonly do within cg(ps). This follows from the fact that all of the subsets of ps are formally the same kind of thing. A dynamic semantics provides the participants in a conversation a way to track the commitments made to both the inferential status of a proposition (p ∈ inf er(ps)), and its possible or probable truth (=kat(p) ∈ cg(ps)) as the discourse proceeds, just as we’ve seen in (5.30) and (5.31). In looking at the other two possible interchanges between the evidentials in Gitksan, ’ we don’t expect to find the same kind of interchangeability between nakw and =kat. This is due to the specific nature of the evidence they specify: reportative evidence can ’ in no natural way be sensory, and vice versa. However, we do expect nakw and =ima to interact given the fact that the broad inferential nature of =ima includes the kinds ’ of sensory evidence nakw specifies. This is examined separately in §5.5 below.  5.4.3  ’ nakw is Not an Illocutionary Force Modifier  ’ One advantage of treating nakw as a sentential force modifier instead of a Faller-type illocutionary modifier, is the ability to account for its illocutionary functions. Although ’ this was not tested explicitly during fieldwork, there is suggestive evidence that nakw, like many other sentential forces, can be associated with a variety of illocutionary acts. These different illocutionary acts reveal themselves in the varieties of discourse contexts 232  ’ 5.4. nakw as a Sentential Force Specifier ’ and translations of nakw. As mentioned above, many Gitksan speakers remark on the ’ exclamatory quality of nakw, as well as giving the impression of asking a question or negative attitude. All three of these qualities can be found in (5.32) and (5.33): (5.32) Context: At the party you notice that John’s jacket is gone. ’ ’ nakwhl tawhls John ’nakw =hl ’ tawhl-(t)=s John evid=cnd leave-3=pnd John “John must’ve left!” Comments: You expect someone in the room might confirm or deny this; you’re surprised that he would leave without saying goodbye; you don’t believe it’s true.  ’ (5.33) nakwhl ’ nakw =hl evid=cnd “Is that bear  maaluhl smax maalu-(t)=hl smax crazy-3=cnd bear crazy or something?”  tust tust that “That bear must be crazy!”  presupposition: The speaker has visual evidence (watching a bear wandering around the village). assertion: The bear is crazy. implicature: This is unusual behaviour for a bear; it could be dangerous (to the people and bear). Context: You’re watching a bear wandering around the streets in the village during broad daylight.  ’ nakw is also fairly common in warnings of various degrees, as the following examples show: (5.34) Context: You can smell smoke from the bar-b-que. ’ nakwhl mihlmihlt ’ nakw =hl mihl-mihl-t evid=cnd redup-burn-3sg “It’s burning!” 233  ’ 5.5. Explaining the Interaction Between =ima and nakw: Pragmatic Blocking (5.35) Context: A misbehaving child is defying her parents’ order to go home because it’s past her curfew. ’ nagwimi xsingy’ ’nakw =mi xsink-y’ evid=2sg disbelieve-1sg “I can see you don’t believe me.” (5.36) Context: You notice a wallet on the floor at the feet of person ahead of you in line. ’ nagwimi galithl andadaalan ’nakw =mi kali-t=hl anda-daala-n evid=2sg drop-3sg=cnd container-money-2sg “You must’ve dropped your wallet.” ’ Under a sentential force analysis of nakw, the illocutionary force of sentences of the ’ form nakw (p) can be determined by the context of use. In other words, the illocutionary ’ force of nakw is not lexically specified as it would be under a speech act analysis. .  5.5  Explaining the Interaction Between =ima and ’ nakw : Pragmatic Blocking  ’ We are now in a position to examine the interpretations of =ima and nakw when they are felicitous in the same contexts, ones where there is sensory evidence for an epistemic ’ claim. This is exemplified in (5.37). In these contexts, nakw is typically translated as must while =ima is translated as might: (5.37) Context: You and a friend are fishing. You’re sitting on the rocks, cutting up bait. You notice blood on the rocks at your friend’s feet.  234  ’ 5.5. Explaining the Interaction Between =ima and nakw: Pragmatic Blocking ’ a. gotsinimahl ’onin ’ ’on-n kots-i-n=ima=hl cut-tr-2sg=mod=cnd hand-2sg “You might’ve cut your hand.” ’ b. nagwimi gotsihl ’ nakw =mi kots-(t)=hl evid=2sg cut-3=cnd “You must’ve cut your  ’ ’onin ’ ’on-n hand-2sg hand.”  ’ you’re trying to say ‘You Comments (paraphrased): When you say k’otsinimahl ’onn, might’ve cut your hand’, or ‘Maybe you cut your hand’. You’re not totally sure because ’ ’ you’re saying ‘It looks like you it could be fish blood. When you say nagwmi g’otshl ’onn cut your hand . . . you must’ve because there’s blood on the rocks.’  This context provides the speaker with what could be construed as visual evidence that the hearer cut himself while preparing bait at the river’s edge. Recall from example (4.1) in the previous chapter that =ima can be translated as either must or might. However, consultants often comment that when you have the appropriate context to ’ ’ use either nakw or =ima – one that has sensory evidence – nakw is somehow ‘stronger’ than =ima. I take this intuition as a starting point in working towards the claim ’ that nakw takes over the must-type interpretation in these sensory evidence contexts, blocking =ima from a universal interpretation. In a nutshell, this can be attributed to ’ the principle of blocking: nakw is more specialized for the ‘strong’ (i.e. must) reading than =ima, and thus blocks =ima from that reading. The formal implementation of this blocking relationship is achieved by the application of Maximize Presupposition: use the most informative presupposition that is satisfied in a context (Heim 1991; Sauerland 2003; Schlenker 2006), defined in (5.38): (5.38) Maximize Presupposition (MP): If a sentence S1 with the presupposition p1 entails S2 with the presupposition p2 , and p1 is a scalar alternative of p2 , the assertion of S2 entails that the speaker doesn’t believe p1 to be entailed by the common ground. 235  ’ 5.5. Explaining the Interaction Between =ima and nakw: Pragmatic Blocking The use of =ima carries the semantic presupposition of inferential evidence, while ’ nakw carries a pragmatic presupposition of sensory inferential evidence in the utterance ’ context. The sensory evidence presupposition associated with nakw prevents it from being felicitous in contexts which lack sensory evidence. The weaker presupposition of =ima is less specific, and is satisfied in contexts with any type of inferential evidence, sensory or not. These evidence presuppositions can be placed on a scale, schematized in (5.39):46 (5.39) [[visual, auditory, other sensory]nakw ’  reasoning, assumption]=ima  ’ The modal force readings of =ima and nakw map to this continuum: the more indirect the evidence, the ‘weaker’ the reading; the more direct the evidence, the ‘stronger’ the reading is. Recall the assumption made earlier in this chapter that semantic presuppositions become the pragmatic presuppositions of speakers. Once this is in place, it should now be straightforward to apply MP to this scale in the following steps, assuming the following conditions are met: following Schlenker (2006), I assume that the application of MP is triggered by certain lexical items which have a pre-determined presuppositional ’ ’ scale, such as the evidential scale mapped to =ima and nakw in (5.39). As such, nakw ’ and =ima are scalar alternatives as the presupposition of nakw in a sentence such as (5.37)b. entails the presupposition of =ima in (5.37)a.: having sensory evidence entails that you have evidence. Another requirement of MP is that it only compares utterances whose assertive components are contextually equivalent. This condition is met by the fact that both (5.37)a. and b. are felicitous in the same context. MP now selects among these the assertion that carries the strongest presupposition compatible with the common ground without yielding a presupposition failure. ’ In (5.37), both nakw -assertions and =ima-assertions are felicitous (satisfying contextual equivalency), and the assertion of b. entails a. However, when the evidence is 46  In (5.39) I am using bracket notation simply for the purposes of grouping the types of evidence.  236  ’ 5.5. Explaining the Interaction Between =ima and nakw: Pragmatic Blocking ’ being assessed within a certain context, the use of the nakw -assertion blocks the =imaassertion in that context if the speaker believes they have sensory evidence for making that epistemic assertion. Now, if a speaker uses an =ima-assertion associated with the ’ weaker presupposition but in the same context where the use of nakw is potentially felicitous, such as in (5.37)a., the use of =ima with its non-specific evidence presupposition implies that you don’t believe your direct sensory evidence is adequate to make a stronger claim, and thus implicates the negation of the sensory evidence presupposition (Sauerland 2003). In other words, (5.37)a. implicates that ‘It is not the case that (I believe) I have sensory evidence that you cut your hand’. The outcome is that the =ima-assertion is translated as might. Thus, there are two interleaving pragmatic properties to evidential assertion in Gitksan: (i.) the scalar presuppositions lexically ’ encoded by nakw and =ima, and (ii.) the scalar implicature that is triggered by either ’ a nakw - or =ima-assertion. Maximize Presupposition has been effectively applied to a variety of phenomena involving the scalar distribution of presuppositions. This analysis contributes to this line of research by applying MP to an evidential system: the modal =ima carries the presupposition that a speaker has some kind of evidence for an assertion, while the ’ evidential nakw carries the presupposition that a speaker has sensory evidence for an assertion. When the common ground provides sensory evidence for an assertion, MP selects among these the assertion that carries the strongest presupposition compatible with the common ground without yielding a presupposition failure. ’ In the next section I turn to a different aspect of nakw, and how it is used in conversation to convey a mirative and metaphorical meaning.  237  ’ 5.6. The Extended Pragmatics of nakw: Mirativity and Metaphor  5.6  ’ The Extended Pragmatics of nakw : Mirativity and Metaphor  ’ This section presents an analysis of how nakw contributes to both the mirative and metaphorical interpretations of sentences.47 The connection between evidentiality and mirativity has received some attention in the literature, particularly in various language grammars and typological studies, yet the category of mirativity has still not found a place within any theory of meaning. In a nutshell, mirativity refers to the grammatical marking of a proposition as representing information which is surprising to the speaker (DeLancey 1997, 2001). The translations in example (5.40) show the mirative ’ interpretation is associated with the evidential nakw : (5.40)  ’ a. bagw nidiit ’ pakw nidiit arrive.pl 3pl “They’ve arrived.” ’ b. nakwhl bagw=diit ’nakw =hl pakw=tiit evid=cnd arrive.pl=3pl “They must’ve arrived!” “Looks like they’ve arrived!”  ’ Under its evidential reading, the use of nakw means the speaker has indirect sensory evidence for a proposition, such as a truck parked in the driveway, or noise in the hallway. ’ When a speaker witnesses an event, nakw can be used to express surprise at a situation, such as the unexpected arrival of guests at a party. ’ There is another pragmatic feature associated with nakw : in addition to its evidential ’ and mirative uses, nakw has a metaphorical use. Consider a context where the speaker is watching a baseball game. The star batter on the speaker’s favourite team keeps missing the ball and striking out, jeopardizing the outcome of the game. Out of exasperation, 47  This section is largely adapted from Peterson (2010).  238  ’ 5.6. The Extended Pragmatics of nakw: Mirativity and Metaphor the speaker sarcastically exclaims:48 ’ (5.41) nakwhl sinst ’nakw =hl sins-t evid=cnd blind-3sg “He must be blind!” “Is he blind or something?” “Looks like he’s blind!” ’ This is a nonliteral use of nakw : the speaker is not asserting that the batter is literally blind, rather, they are drawing attention to the poor performance of the batter by attributing his missing the ball as a result of blindness. Whereas there is an established tradition of research on metaphor in literary studies, philosophy, and linguistics, its connection to evidentiality has not been previously explored in much detail. There is suggestive evidence from a variety of languages that there is a connection between the nonliteral uses of miratives and evidentials. This can be observed even in the translations of the Gitksan example in (5.41), which would also be appropriate nonliteral statements in English in this context. Also in English we see the link between evidentiality and nonliteral interpretations in how evidential verbs such as see can be used in the following context in (5.42): (5.42) Context: Your daughter is only allowed to use the computer on the weekends. However, there is an assignment due at school, and she asks to use the computer on a weeknight to finish it. You give her permission, but when you come home, you see her playing computer games instead of working on her project (adapted from Gilmour et al. (2010)). “I see you’re working on your project.” (nonliteral/evidential) 48  It is possible to explicitly distinguish between two different kinds of nonliteral interpretation: (i.)  sarcastic (e.g., intending the opposite of what is literally said); and (ii.) metaphorical (e.g., involving some sort of parallelism or correspondence between what is meant and what is said, for instance between being literally blind, and blind in an extended metaphorical sense). These are in fact different cases, but I assume both involve “nonliteral” meaning.  239  ’ 5.6. The Extended Pragmatics of nakw: Mirativity and Metaphor Even language-internally, a quick survey of how mirativity is conveyed in English reveals a wide variety of ways of how one can express surprise when a friend unexpectedly shows up at a party: (5.43) You made it! I don’t believe you made it! Looks like you made it! That must be you! Wow, you’re here! Is that really you?! That can’t be who I think it is! etc. There are a number of questions that come out of the observations above. The first ’ involves examining the notion of mirativity as a natural linguistic class. Why is nakw used in conveying mirativity in Gitksan? How is mirativity related to modality, as in English in (5.43)? Is there any systematic connection between evidentiality and and the non-literal use of an evidential? Are there any empirical generalizations that can draw these features of mirativity and metaphor together, and can this be approached in a compositional way? This section addresses these questions by examining evidentiality as the semantic and pragmatic source of mirativity, the constructions and morphemes mirativity is associated with, and its source in the psychological orientation of a speaker to evidence and events. In all of its manifestations, mirativity is shown to be linked to the semantics and pragmatics of evidentiality. From here, steps are taken towards a formal account of mirativity as a pragmatic phenomenon: mirativity operates at the speech act level, and does not contribute to the truth conditional meaning of a sentence. In a nutshell, what distinguishes a mirative statement from a non-mirative statement in an example such as (5.40) is conversational  240  ’ 5.6. The Extended Pragmatics of nakw: Mirativity and Metaphor implicature: evidential expressions have a mirative interpretation as the result of a Quantity implicature. Recall from chapter 3 that in a statement of the form ev(p), where p is the proposition associated with the evidential (ev), a speaker cannot know for certain that p is in fact true. This is the Known Truth/Falsity test, repeated here from chapter 3: (5.44) The Known Truth/Falsity Test: If the use of the evidential is felicitous when the speaker knows the prejacent is true or false, the evidential cannot be a modal. This test says that if an evidential can be used when the prejacent is true or false, it is not a modal. For example, For example, in Cuzco Quechua a speaker may use the direct evidential -mi if they know p is true (Faller 2002). Additionally, the test also suggests that a non-modal evidential would be straightforwardly felicitous when the prejacent is known to be true or false without implicature. This contrasts with a modal evidential: if a speaker knows p is true, we expect either infelicity or Gricean considerations to ensure that a speaker assert p, and not ev(p) (cf. (3.37)). However, I suggest it may not be as simple as this. The tests applied in chapter 3 ’ ’ showed that nakw is a non-modal evidential (i.e. nakw cannot be embedded, assented to/dissented from etc.), yet in a way it does obey the Truth/Falsity test: in its normal ’ usage, nakw (p) statements are used in discourse to convey that the speaker has sensory evidence for p, not knowing whether p is true or false. A mirative statement results when ’ a speaker uses nakw (p) knowing (p) is in fact true, and a metaphorical statement results ’ when a speaker nakw (p) knowing that p is false. These facts may tell us something more about the applicability of the Known Truth/Falsity test: in languages such as Quechua with non-modal evidentials, we do not see (to my knowledge) any extra meaning when ’ the prejacent is known to be true or false. However, with nakw there is extra meaning, and they correspond to the truth or falsity of p. I claim that these observations form a three-way formal system for the pragmatic use of an evidential, as given in (5.45): 241  ’ 5.6. The Extended Pragmatics of nakw: Mirativity and Metaphor (5.45) (i.) In asserting ev(p), the Speaker does not know it’s part of the common ground that either p or not p: Evidential without any implicated meaning (ii.) In asserting ev(p), the Speaker knows it’s part of the common ground that p: mirativity as Quantity implicature (iii.) In asserting ev(p), the Speaker knows it’s part of the common ground that not p: metaphor as Quality implicature This bears directly on the status of mirativity as a natural linguistic class, and the debate within the literature as to whether or not mirativity is a separate semantic category, or simply an extension of evidentiality (cf. DeLancey 1997; 2001). One of the outcomes of this analysis is a unified treatment of mirativity: its effects are derived from other components of the grammar in a predictable way through implicature. This analysis also predicts a relation between mirativity and metaphor based on the speaker’s knowledge of the truth or falsity of p. I take a fairly standard approach to analyzing the nonliteral uses of evidentials, such as the Gitksan example in (5.41) (Grice 1989).49 For example, upon uttering (5.41), the speaker literally says that ‘he must be blind’, something he knows is is false. Thus, the speaker is flouting the maxim of Quality (“do not say what you believe to be false”). What the speaker is doing is asserting (5.41) in order to implicate that the batter is performing counter to expectations, or that the batter has the attributes of blindness. The next section examines in detail the meanings and sources of mirativity and its systematic relation to evidentiality. In §5.6.2 a formal pragmatic analysis is presented of how mirativity is conversationally implicated. §5.6.3 turns to the nonliteral uses of evidentials in examining the effect of an evidential statement when the speaker knows the embedded proposition is false. 49  The features of metaphor and their study are numerous and complex. My intention here is not to  offer an account of metaphor in general or argue for a particular approach to metaphor, but only to explore the link between evidentials and metaphorical interpretations.  242  ’ 5.6. The Extended Pragmatics of nakw: Mirativity and Metaphor  5.6.1  Approaching the Category of Mirativity  Although descriptions of the mirative have appeared in various language grammars and in the typological literature, discussions of mirativity as a cross-linguistic phenomenon usually begin with DeLancey (1997; 2001), who defines mirativity as marking information which is ‘new to the speaker’, or more specifically: [Mirativity] marks both statements based on inference and statements based on direct experience for which the speaker had no psychological preparation, and in some languages hearsay data as well. What these apparently disparate data sources have in common ... is that the proposition is one which is new to the speaker, not yet integrated into his overall picture of the world. (DeLancey 1997: 35-36) Mirativity covers semantic dimensions variously described as ‘non-expected’ information (Egerod and Hansson 1974), information for which the speaker is ‘not prepared’ (Slobin and Aksu 1982), ‘immediate meaning’ (Nichols 1986), and ‘new knowledge’ (DeLancey 2001: 369 for other references). Dickinson (2000, p. 379) refines the definition of mirativity to include the speaker’s immediate experience of an event: if the event does not correlate well with a speaker’s expectations, the proposition coding the event receives special marking.50 However the ‘mirative’ (and the related ‘admirative’) include not only expressions of newly emerged evidence, but often also inferences based on such evidence (see also Friedman 2003; Aikhenvald 2006, p. 195-215 for an overview). What these descriptions from various languages and studies suggest is that mirativity, as a conceptual category at least, may be universal: it is a plausible claim that all languages have the means to encode an event or state as occurring outside normal 50  Dickenson (2000: 379) also notes other construals of mirativity based on the speaker’s past ex-  periences of similar situations and his general knowledge, physical interactions or cultural and social norms. I won’t be discussing these occurrences of the mirative, assuming for my purposes that these construals still reduce to a speaker’s unprepared state of mind at the time of utterance.  243  ’ 5.6. The Extended Pragmatics of nakw: Mirativity and Metaphor expectations. In order to deepen our understanding of mirativity, and to draw these descriptions together into a more cohesive and focused picture, it is useful to examine the systematic relationship mirativity has with the better understood categories of evidentiality and epistemic modality. Mirativity forms a conceptual natural class with evidentiality and epistemic modality as these three categories express something about a speaker’s physical, psychological and temporal orientation to events and states (cf. Dickenson 2000; DeLancey 2001: 379). The summary in (5.46) outlines this connection: (5.46) (i.) Epistemic modality marking: encodes the speaker’s attitude towards the proposition in terms of certainty or probability. (ii.) Evidential marking: encodes the source of the speaker’s knowledge. (iii.) Mirative marking: encodes the relationship between the proposition and the speaker’s overall expectations and assumptions in a given context. Chapters 3 and 4 examined the link between modality and evidentiality: the evidentials =ima and =kat are epistemic modals. However, the link between evidentiality and mirativity, and epistemic modality and mirativity are examined more closely in this subsection. 5.6.1.1  Evidentiality and Mirativity  In building a picture of mirativity from the ground up, recall so far that the primary function of an evidential is to give a speaker a way of talking about events they haven’t personally seen, heard, or otherwise taken part in. In example (5.47), the evidential ’ nakw is used to encode that a speaker has sensory evidence for an event that they have not witnessed directly:  244  ’ 5.6. The Extended Pragmatics of nakw: Mirativity and Metaphor (5.47) Context (sensory evidence): You get to Bob’s place and you can smell or see smoke. ’ nakwhl sehons Bob ’nakw =hl se-hon-(t)=s Bob evid=cnd caus-fish-3=cnd Bob “Bob must be smoking fish” “Looks like Bob is smoking fish” At an intuitive level, an event that is witnessed is more certain than one that occurs sight unseen, and an event that is witnessed from beginning to end is less surprising than one that is only inferred or deduced from its results (Dickenson 2000). If we adjust the context slightly to include not only the sensory evidence, but the speaker actually witnessing the event of Bob smoking fish, (5.47) is still felicitous. However, (5.47) carries an additional meaning: the speaker is surprised or otherwise unprepared for the fact that Bob is smoking fish. This additional meaning characterizes the mirative use of an evidential, and illustrates the notion of the ‘unprepared mind’ (DeLancey 1997): an event may be perceived to be out of one’s control, unexpected, and thus surprising to the speaker, or when they come into contact with the results of the event. However, the distinction between witnessing the event and witnessing the results of ’ the event can be subtle. In example (5.48), both nakw and -mIş in Turkish have an evidential meaning when the speaker infers they cut themselves upon observing blood at their feet (the results of the hand-cutting event). When they observe their cut hand (the hand-cutting event itself), the mirative meaning emerges: the speaker didn’t actually witness the event of cutting, and the results of the event are surprising to the speaker: (5.48)  a. Gitksan ’ nagwin kotshl ’onin ’nakw =n kots-(t)=hl ’on-n evid=1sg cut-3sg=cnd hand-1sg “I must’ve cut my hand.” “I see I cut my hand.” 245  ’ 5.6. The Extended Pragmatics of nakw: Mirativity and Metaphor b. Turkish el-im-i kes-miş-im hand-1sg.poss-acc cut-mir/evid-1sg “I must’ve cut my hand.” (A. Tekant p.c.) Inferential: There is blood at your feet. Mirative: You see the cut on your hand.  In Gitksan, if a speaker witnesses the actual event of cutting, they can use a nonevidential statement which lacks a mirative effect. It is only in the context where the ’ speaker uses nakw when a plain assertion would also be felicitous, that the mirative meaning emerges. Another way to view this is in terms of knowledge of p: in its purely ’ evidential use, nakw is signals that the speaker is not sure if p is true; under a mirative reading, the speaker knows p is true.. There is also another angle of meaning. The event(s) leading to the cut hand in (5.48) were likely inadvertent. This implies a lack of involvement or control on the part of the speaker, thus they react with surprise at the outcome. Example (5.49) also shows this, where a speaker could comment to a mother at the conclusion of her daughter’s piano recital: (5.49) Turkish (Aksu-Koç and Slobin 1986, 162) kiz-iniz çok iyi piyano çal-iyor-muş daughter-2pl.poss very good piano play-pres-mir/evid “Your daughter plays the piano very well.” The speaker directly witnessed the entire event of piano playing, but indicates using -mIş that he was not psychologically prepared for the high quality of the performance. Slobin and Aksu (1982: 196) also describe -mIş as representing an experience for which the speaker had no ‘premonitory awareness’ . When -mIş occurs with a first person subject, it indicates lack of conscious awareness on the part of the speaker, not simply lack of speaker involvement. 246  ’ 5.6. The Extended Pragmatics of nakw: Mirativity and Metaphor The extended meaning of an evidential to convey a sense of surprise also presents ’ us with a potential contradiction: the use of nakw when the speaker actually witnesses the event they have evidence for in (5.47), would appear to undermine its evidential meaning: Gricean considerations would compel a speaker to simply assert “Bob is smoking fish” if the speaker did indeed witness the event of Bob smoking fish. However, ’ we can draw these two interpretations of nakw together if we view this in terms of distancing: whereas evidentiality indicates physical distancing from an event, mirativity indicates psychological distancing (Dickenson 2000). In some languages these are marked separately but in Gitksan and Turkish and many other languages, evidentiality and mirativity are encoded by the evidential markers of the language.51 Chapter 3, §3.3, discussed languages that do not have lexical evidentials, but where evidential meanings arise through the use of the perfect aspect. It is not surprising that we find a mirative use of the perfect as well. In Bagvalal, the aspectual auxiliary ek◦ ’a carries a mirative meaning: (5.50) Bagvalal (Tatevosov 2001) Context: The speaker looks into his desk and finds 100 rubles there; he had completely forgotten about this money being there. di-č’ as b-uk’a-b-o ek◦ ’a! 1.sg.obl-cont money n-be-n-conv aux.prs “(I see) I have money!” (5.51) Context: The speaker watches Ali trying to put on the hat. At last Ali succeeds. ali-r butuna ˜e¯sa-m-o ek◦ ’a! Ali-erg hat put.on-n-conv aux.prs “Ali has put on the hat!” 51  In other languages such as Hare, Dargwa and Chechen, mirative meaning is formally detached from  evidentiality, although it is still dependent on it. Because these languages have morphology dedicated to mirative meaning, mirativity is conventionally implicated. See DeLancey (2001); Peterson (2010) for details.  247  ’ 5.6. The Extended Pragmatics of nakw: Mirativity and Metaphor English also lacks lexical evidentials, although a mirative meaning can be attributed to evidential verbs when these are used in the context of witnessing the actual event. ’ Example (5.52) uses the same context as the Gitksan example in (5.47) with nakw : evidential verbs such as look like and see are felicitous when the speaker observes the event embedded under the evidential verb. This expresses the mirative:52 (5.52) “Looks like Bob is smoking fish!” “I see Bob is smoking fish!” 5.6.1.2  Mirativity and Epistemic Modality  There is also a relation between epistemic modality marking and mirativity. As with evidential-marked miratives, a mirative reading of an epistemic modal in English is mostly clearly obtained where a speaker is surprised at the results of a previous event. In the context given in example (5.53), a mirative interpretation can be expressed using either the strong epistemic modal must in (i.), or a plain assertion with a ‘surprised intonation’ in (ii.) (see footnote on this page). A mirative interpretation is less felicitous with the weak epistemic modal might, as in (iii.): (5.53) Context: Said upon awakening over one’s books after a long night studying (context adapted from Aksu-Koç & Slobin 1986: 160) (i.) “I must’ve fallen asleep!” (ii.) ? “I fell asleep!” (iii.) #“I might’ve fallen asleep!” 52  Intonation is another way to express mirativity in English, and may overlay the evidential state-  ments in (5.52). A ‘surprise’ intonation is how a plain assertion such as “Bob is smoking fish!” can register mirativity. Nonetheless, the sentences in (5.52) can still express the unexpected or unprepared psychological state of the speaker at witnessing Bob smoking fish, although usually with the support of intonation.  248  ’ 5.6. The Extended Pragmatics of nakw: Mirativity and Metaphor It is beyond the scope of this dissertation to examine in more detail the mirative use of epistemic modals in languages such as English, as I will be limiting myself to the relationship between evidentiality and mirativity, as described in the previous subsection. However, there are two points worth making. The first point regards the use of modal force: to convey the surprised or unprepared state of the speaker in (5.53), the strong modal must is used over both the weaker modal might and a plain assertion. Because modals don’t overtly encode an evidence source/type, they may reveal something different of the nature of mirativity than we find with evidentials. It seems natural that, in encoding a speaker’s state of surprise, the ‘strongest’ lexical item would be used. However, in example (5.54), the weaker modal might is used to convey a speaker’s unprepared state, not upon encountering any kind of evidence as in (5.53), but at the possibility of winning: (5.54) Context: Your husband tells you that he thinks your lucky numbers came up on the weekly lotto. (i.) “I might’ve won!” (ii.) # “I must’ve won!” (iii.) # “I won!” This is expected, as when a speaker is surprised at a possibility, a possibility modal is natural. However, intonation is likely the actual carrier of the mirative contribution in (5.54), as the possibility is actually part of the proposition a speaker is surprised at. This is different from example (5.53), where the proposition a speaker is surprised at is the plain one without any modal. It is in those cases where the strong modal must be used for the mirative. Secondly, a mirative use of an epistemic modal in English is infelicitous in a context where the speaker actually witnesses the event, as in (5.55):  249  ’ 5.6. The Extended Pragmatics of nakw: Mirativity and Metaphor (5.55) Context: A friend unexpectedly shows up a party. (i.) “You’re here!” (ii.) # “You must be here!” (iii.) # “You might be here!” This restriction likely follows from the fact that epistemic modals are propositional operators. This would also predict that evidential modals cannot be used miratively, which is confirmed with the modal evidentials in St’át’imcets (Matthewson p.c.). This is also the case in Gitksan with the modal evidential =ima in example (5.56), which cannot be used if the speaker witnesses the event embedded under it: (5.56) Context: Your friend is showing you how to cook something, and while watching them you see them accidentally cut themself. ’onin #kotsinima=hl kots-i-n=ima=hl ’on-n cut-tr-2=modal=cnd hand-2sg “You might’ve/must’ve cut your hand.” The lack of mirativity in the modals in (5.55) and (5.56) is derived from their status as propositional operators. However, in (5.54) the proposition a speaker is surprised at is the plain one with a weak modal. This shows that if one is really surprised at a necessity statement, must would be felicitous in a mirative. As in (5.54), if you see some evidence that you’ve won, for example, if there is a person walking towards you holding out the trophy, then ‘I must’ve won!’ would be felicitous. However, it is not the modal that’s conveying mirativity, but the intonation. These observations relates to the claim of von Fintel and Gillies (2007) that epistemic ‘must’ in English