UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Syntactic features in agrammatic production Sanchez, Monica Eszter 1996

You don't seem to have a PDF reader installed, try download the pdf

Item Metadata

Download

Media
831-ubc_1997-19650X.pdf [ 14.36MB ]
Metadata
JSON: 831-1.0088003.json
JSON-LD: 831-1.0088003-ld.json
RDF/XML (Pretty): 831-1.0088003-rdf.xml
RDF/JSON: 831-1.0088003-rdf.json
Turtle: 831-1.0088003-turtle.txt
N-Triples: 831-1.0088003-rdf-ntriples.txt
Original Record: 831-1.0088003-source.json
Full Text
831-1.0088003-fulltext.txt
Citation
831-1.0088003.ris

Full Text

S Y N T A C T I C F E A T U R E S IN A G R A M M A T I C PRODUCTION by M O N I C A ESZTER S A N C H E Z B.A., The University of Toronto, 1988 M . A . , The University of Toronto, 1989  A THESIS S U B M I T T E D IN P A R T I A L F U L F I L L M E N T OF T H E R E Q U I R E M E N T S F O R T H E D E G R E E OF DOCTOR OF PHILOSOPHY in T H E F A C U L T Y OF G R A D U A T E STUDIES Department of Linguistics  We accept this thesis as conforming to the required standard  T H E U N I V E R S I T Y OF BRITISH C O L U M B I A December 1996 © Monica Eszter Sanchez, 1996  In  presenting this  degree at the  thesis  in  University of  partial  fulfilment  of  the  requirements  British Columbia, I agree that the  for  an advanced  Library shall make  it  freely available for reference and study. I further agree that permission for extensive copying  of  this thesis for  department  or  by  his  or  scholarly purposes may be granted her  representatives.  It  is  by the  understood  that  head of copying  my or  publication of this thesis for financial gain shall not be allowed without my written permission.  Department  of  L\ M 6 Ul STl Cl$  The University of British Columbia Vancouver, Canada  Date  DE-6 (2/88)  IW.  j  %  ABSTRACT  This thesis examines the nature of the language deficit called agrammatism, the linguistic syndrome usually associated with Broca's aphasia.  I focus on the narratives produced by  agrammatics of five different languages, English, Dutch, German, French, and Italian, the transcripts of which are collected in Menn & Obler (1990).  M y goal is to account for the  omission and substitution errors that characterize agrammatic production. Agrammatic utterances with omissions display appropriate word order. Among these are structures that include adverbs, negation, and verb-second matrix clauses. These structures are derived by movement to functional projections.  I argue that agrammatic clauses include  minimally three functional projections above V P . I propose that the full array of functional projections is present in agrammatic speech. I argue that the most concise account of agrammatic production is one in which Universal Grammar governs agrammatic speech. Although any syntactic category may be omitted, not all categories are omitted with the same frequency. Lexical categories are better retained than functional categories; and nominal categories are better retained than verbal categories. I propose a Principle of Robustness whereby the more Formal features a category is specified for, the more Robust it is. The net result is that the more features a syntactic category is specified for, the more likely it is retrieved. This results in the following Retrieval Hierarchy: N > V , A , D > P, T, K > C, where ">" means "better retained than". In addition to omissions, agrammatic speech includes substitutions:  Syntactic  substitutions display two striking characteristics. First, substitutions are not cross-categorial. Second, substitutions are subject to the Single Feature Constraint: only one optional Formal feature from agreement (person, number and gender), Case and tense is altered. To derive these characteristics, I argue that the structure of the Lexicon is paradigmatic. Both omissions and substitutions lead me to a discussion of Lexical Insertion, the process by which words are inserted into syntactic structures. I conclude that the agrammatic deficit lies  ii  outside the phonological, syntactic and semantic components proper. Instead, omissions and substitutions result from an impairment to the interface mechanisms between the Lexicon, the Syntax and the Phonology.  iii  TABLE OF CONTENTS Abstract Table of Contents  iv  List of Tables  viii  List of Figures  x  Glossary  xi  Acknowledgement  xiii  Chapter 1 1.1  1.2  2.3 2.4  3.2  3 4 5 6 9  AGRAMMATISM  11  A General Description of Agrammatism Agrammatism Is a Cohesive Syndrome 2.2.1 Variability 2.2.2 Comprehension-Production Dissociation The Data 2.3.1 The Subjects 2.3.2 Description of the Data Issues in Agrammatism  Chapter 3 3.1  1  Theoretical Framework 1.1.1 General Model 1.1.2 Phrase Structure 1.1.3 Syntactic Features Outline of the Thesis  Chapter 2 2.1 2.2  INTRODUCTION  A G R A M M A T I S M : OMISSION ERRORS  The Role and Significance of Universal Grammar 3.1.1 Universal Grammar and Agrammatic Grammar Are Not Independent 3.1.1.1 Major Constituent Order 3.1.1.2 Word Order Within Constituents 3.1.2 Agrammatic Grammar Is Constrained by Universal Grammar Functional Projections in Agrammatic Production 3.2.1 Agrammatic Clauses Are More than Lexical Projections 3.2.1.1 Nominal Functional Projections 3.2.1.1.1 Missing Articles 3.2.1.1.2 Missing Possessor 3.2.1.2 Verbal Functional Proj ections 3.2.2 Agrammatic Clauses Have at Least Two Functional Projections 3.2.2.1 How Much Structure in Agrammatic Clauses? 3.2.2.1.1 Movement Requiring Only NegP 3.2.2.1.2 Movement Requiring One Functional Projection 3.2.2.1.3 Movement Requiring Two Functional Projections  iv  12 14 15 18 19 20 21 22 26 29 30 33 39 47 48 52 53 53 56 58 73 76 79 81 86  3.3  3.4  3.5  3.2.2.2 (More) Arguments Against the Truncation Hypothesis 3.2.2.2.1 Transition Clause-Final Tensed Verbs 3.2.2.2.2 Wh-Elements Do Not Co-Occur with Infinitive Verbs 3.2.2.2.3 Subject Clitics Do Not Co-Occur with Untensed Verbs 3.2.2.2.4 No Infinitival Auxiliaries or Modals 3.2.2.2.5 Few Root Infinitives in Italian 3.2.2.3 Arguments Against the Optional Tense Hypothesis 3.2.3 Agrammatic Clauses Have the Full Array of Syntactic Projections 3.2.3.1 Movement to C P 3.2.3.2 Movement to T P . 3.2.3.3 Movement to a Functional Projection between T P and V P Lexical Projections in Agrammatic Production 3.3.1 Missing Lexical Categories 3.3.1.1 Missing Verb 3.3.1.2 Missing Noun 3.3.2 Missing Arguments 3.3.2.1 Missing Subjects and Objects 3.3.2.2 Subject-Verb Agreement 3.3.2.3 The Nature of Null Arguments The Distribution of Omissions 3.4.1 Are Rates of Omission a Measure of Severity? 3.4.2 Which Syntactic Categories Get Omitted? 3.4.2.1 The General Retention Hierarchy 3.4.2.2 Lexical versus Functional 3.4.2.3 Nominal versus Verbal Conclusion A G R A M M A T I S M : SUBSTITUTION ERRORS  153  The Nature of Substitutions 4.1.1 Substitution Is Not Chaotic 4.1.2 Substitution Involves Agreement, Case and Tense Features 4.1.2.1 Phonologically Constrained Substitutions 4.1.2.2 Semantically Constrained Substitutions 4.1.2.3 Syntactically Constrained Substitutions 4.1.3 Subtitution Involves a Single Feature Shift 4.1.3.1 Substitution Is Not Random Feature Swapping 4.1.3.2 Verbal Substitution 4.1.3.2.1 Agreement 4.1.3.2.2 Tense 4.1.3.3 Nominal Substitution 4.1.3.3.1 Agreement 4.1.3.3.2 Case The Distribution of Substitutions 4.2.1 Verbal Substitution 4.2.1.1 Agreement 4.2.1.1.1 Person 4.2.1.1.2 Number 4.2.1.1.3 Gender 4.2.1.2 Tense 4.2.2 Nominal Substitution 4.2.2.1 Agreement 4.2.2.1.1 Person  154 154 159 159 160 163 166 166 169 169 171 173 173 177 180 181 182 182 183 183 184 185 186 186  Chapter 4 4.1  4.2  87 88 90 90 92 93 97 101 102 106 110 113 113 114 115 118 118 125 127 137 137 145 146 147 149 151  v  4.3  4.2.2.1.2 Number 4.2.2.1.3 Gender 4.2.2.2 Case 4.2.3 Robustness and Substitution Conclusion A G R A M M A T I S M : W H E R E IS T H E DEFICIT?  Chapter 5 5.1  5.2  5.3 5.4 5.5  Whatan Agrammatic Deficitlsn't 5.1.1 Not a Purely Phonological Deficit 5.1.2 Not a Purely Morphological Deficit 5.1.3 Not a Purely Semantic Deficit Agrammatic Production: A Deficit in Lexical Insertion 5.2.1 The Lexical Item 5.2.2 Lexical Insertion 5.2.2.1 Full Lexical Insertion 5.2.2.2 SplitLexicallnsertion 5.2.3 Accounting for Omissions 5.2.3.1 The Principle of Robustness 5.2.3.2 Deriving Omissions 5.2.4 Accounting for Substitutions 5.2.4.1 Paradigms in the Lexicon 5.2.4.2 Deriving Substitutions The Relationship between Omission and Substitution 5.3.1 Substitution Rates Do Not Reflect Severity of Deficit 5.3.2 There Is No Substitution Hierarchy Constraints on a Model of Language Production 5.4.1 Competence and Performance 5.4.2 A Processing Account of Agrammatic Production Conclusion  Chapter 6 6.1 6.2 6.3  188 189 189 190 192  CONCLUDING REMARKS  194 196 196 200 203 205 206 208 209 215 220 220 223 230 231 232 239 239 240 244 245 248 252 255  The Competence-Performance Debate Summary of the Thesis Further Research  255 257 260  Bibliography  261  Appendix  277  (1) Agrammatic Clauses without Errors: Negation (2) Agrammatic Clauses without Errors: Adverb (3) Agrammatic Clause Order: Subject Precedes Untensed Verb (4) Agrammatic Clause Order: Subject Precedes Tensed Verb/Auxiliary/Modal (5) Agrammatic Clauses: Subject Precedes Adjective (6) Agrammatic Dutch and German: Non-subject in Initial Position (7) Agrammatic German: Embedded Clauses with Final Tensed Verb/Auxiliary (8) Agrammatic French and Italian: Clitics (9) Agrammatic Italian: SVO Order  277 278 279 281 283 284 286 287 288  vi  [ 10) Agrammatic Clauses: Adjective and Noun ^11) Agrammatic Clause: Verb and Adverb [ 12) Agrammatic Clause: Verb and Negation ^13) Agrammatic Clauses: Replacement Sequences for Determiners ^14) Agrammatic Clauses: Null Determiner with Possessive Interpretation [15) English Agrammatic Clauses: Null Possessive Morpheme [ 16) Agrammatic Untensed Clauses: Subject-Neg and Subject-Adv ; 17) Agrammatic Italian Subject-Initial Clauses: S-V, S-Adj, S-N and S-PP ^18) Agrammatic Clauses: Missing Verb [ 19) Agrammatic Clauses: Replacement Sequences for Verbs 20) Agrammatic Omissions: Missing Subject Noun ^21) AgrammaticOmissions: Missing ObjectNoun [22) Agrammatic Clauses: Replacement Sequences for Nouns [23) AgrammaticOmissions: Missing Subject - l S g [24) AgrammaticOmissions: Missing Subject- 2Sg [25) AgrammaticOmissions: Missing Subject DP - 3Sg Referential [26) AgrammaticOmissions: Missing Subject DP - 3Sg ExpletiveQuasi-Argument [27) Agrammatic Omissions: Missing Subject DP - 3 Sg Expletive Non-Argument 28) AgrammaticOmissions: Missing Subject DP - 1P1 [29) AgrammaticOmissions: Missing Subject DP - 3P1 30) Agrammatic Omissions: Missing Object DP - l S g 31) AgrammaticOmissions: Missing Object DP - 2Sg 32) AgrammaticOmissions: Missing Object DP - 3Sg 33) AgrammaticOmissions: Missing Object DP - 3 PI 34) Agrammatic Clauses: Replacement Sequences for Arguments 35) Agrammatic Clauses: Agreement Errors on Verb with Overt Subject 36) Agrammatic Omissions: Null Subject with Agreement on Verb 37) Agrammatic Clauses: Null Arguments with no Antencedents in the Discourse and/or That Introduce New Discourse Topic 38) Agrammatic Person Substitution: Verb, Auxiliary, Copula 39) Agrammatic Number Substitution: Verb, Auxiliary, Modal ^40) Agrammatic Tense Substitution: Tensed Verb, Auxiliary, and Modal 41) Agrammatic Verbal Substitution: Untensed Forms [42) Agrammatic Substitution: Number on Nouns 43) Agrammatic Substitution: Number on Determiners ^44) Agrammatic Substitution: Gender on Pronouns 45) Agrammatic Substitution: Gender on Adjectives '46) Agrammatic Substitution: Gender on Determiners 47) Agrammatic Substitution: Case on Determiners  vii  289 291 292 293 295 296 297 299 300 303 304 305 307 308 312 313 317 318 320 321 322 323 324 327 328 329 331 333 336 338 340 341 342 343 344 345 346 350  LIST OF TABLES Table 1: Evidence for the Lexical Projection Hypothesis (LPH)  72  Table 2: Evidence for the Truncation Hypothesis (TH)  96  Table 3: Evidence for the Optional Tense Hypothesis (OTH)  100  Table 4: Agrammatic English Mean Length of Utterance and Mean Rate of Production  137  Table 5: Mean Length of Utterance (MLU) and Mean Rate of Production (MRP) - Ranges for Agrammatics and Controls  138  Table 6: Agrammatic English Rates of Omission  140  Table 7: Summary of Agrammatic M L U , M R P and Average Omission  141  Table 8: Agrammatic French Rates of Omission  142  Table 9: Agrammatic Italian Rates of Omission  143  Table 10: Agrammatic Dutch Rates of Omission  144  Table 11: Agrammatic German Rates of Omission  144  Table 12: Agrammatic Omission Rates of Functional versus Lexical Categories  148  Table 13: Agrammatic Omission Rates of Verbal and Nominal Categories  150  Table 14: Attested Agrammatic Verbal Substitutions  173  Table 15: Attested Agrammatic Nominal Substitutions  179  Table 16: Agrammatic Percentage of Verbal Feature Substitutions  181  Table 17: Agrammatic Person Substitution Pattern  182  Table 18: Agrammatic Number Substitution Pattern  183  Table 19: Agrammatic Gender Substitution Pattern  184  Table 20: Agrammatic Tense Substitution Pattern  184  Table 21: Agrammatic Substitution Pattern for Untensed Verbs  185  Table 22: Percentage of Nominal Substitutions in the Agrammatic Data  186  Table 23: Agrammatic Person Substitution Pattern  188  Table 24: Agrammatic Number Substitution Pattern  188  Table 25: Agrammatic Gender Substitution Pattern  189  viii  Table 26: Agrammatic Case Substitution Pattern Table 27: Agrammatic Determiner Error Rate Table 28: Summary of Evidence from Omissions and Substitutions Table 29: English Pronoun Paradigm Table 30: English Present Tense Copula Paradigm Table31: Agrammatic Substitution & Omission Rates Table 32: English Agrammatic Substitution Rates Table 33: Dutch Agrammatic Substitution Rates Table 34: German Agrammatic Substitution Rates Table 35: French Agrammatic Substitution Rates Table 36: Italian Agrammatic Substitution Rates  ix  LIST O F  FIGURES  Figure 1: General Model of Universal Grammar  4  Figure 2: Differing Comprehension and Production Modalities  18  Figure 3: Mean Length of Utterance (MLU) Ranges for Agrammatics(A) and Controls (C) 139 Figure 4: Mean Rate of Production (MRP) Ranges for Agrammatics (A) and Controls (C)  139  Figure 5: Range of Rates of Omission (in %) for Functional and Lexical Categories  149  Figure 6: Range of Rates of Omission (in %) for Verbal and Nominal Categories  151  Figure 7: General Schema of Full Lexical Insertion  210  Figure 8: General Schema of Split Lexical Insertion  216  Figure 9: Model of (a Split Lexical Insertion) Grammar  218  Figure 10: Interaction of Competence-Performance-Motor Systems  245  Figure 11: Internal Structure of the Motor Coding System  246  Figure 12: A Model of Grammatical Competence  248  Figure 13: Model of (a Split Lexical Insertion) Grammar  258  GLOSSARY The following abbreviations are used in examples, tables and figures A Acc Adj Adv AG AgrSP AgrOP AP Aux CI Comp Conj CP CR Dal Def Det DP F F-category FP Gen Indef Inf IO IP K L-category  U Loc LP LPH LRRH M MLU Mod MRP N N N-category Neg NegP NG Nom NP 0 OTH P Part  adjective accusative (predicate) adjective adverb agrammatic grammar subject agreement phrase object agreement phrase adjective phrase auxiliary clitic complementizer conjunction Complementizer phrase Capuceto Rojo dative definite determiner determiner phrase feminine functional category functional projection genitive indefinite infinitive indirect object inflectional phrase kase lexical category lexical item locative lexical projection Lexical Projection Hypothesis Little Red Riding Hood masculine mean length of utterance modal mean rate of production neuter (context different from "noun") noun (context different from "neuter") nominal category negation negation phrase normal grammar nominative noun phrase object Optional Tense Hypothesis preposition participle xi  PCR PI Prs Pron Pst RH RRH S Sg t t AP  IO  t^-p t T TH TP UG V V-category Vinf Vpart Vtns 1 2 3 wh  Petit Chaperon Rouge plural present pronoun past Riding Hood Red Riding Hood subject singular trace of an adjective phrase trace of an indirect object trace of an object trace of a subject trace of a verb phrase trace of a Wh-phrase tense Truncation Hypothesis tense phrase Universal Grammar verb verbal category infinitival verb participial verb tensed verb first person second person third person  xii  ACKNOWLEDGEMENT The most difficult task in finishing any project is making sure that all those deserving of it receive appropriate recognition. My first thanks must go to Elizabeth Cowper. She introduced me to linguistics and is the reason I went into the field. Thanks Liz. You are one of the best teachers I have ever had. Your insight, energy and enthusiasm are a constant drive for me. Among the many other influences on me during my years at U of T are Yves Roberge, who supervised my M A forum paper, and Dianne Massam, who encouraged me at every turn. I would like to thank all the members of my committee for their guidance and patience. Henry Davis, my supervisor, thanks for your insight, for listening to all my complaints, and for all the haggling you did on my behalf. Rose-Marie Dechaine, thanks for your inspiration, for your perceptive comments,! and for getting excited about the data. Hamida Demirdache, thanks for the many hours of challenging discussion, for pushing my arguments as far as possible, and for never giving up on me. And John Gilbert, thanks for the many conversations about Weigl & Bierwisch and for your unwavering support. My influences at U B C go beyond my committee. Thank you Guy Carden, David Ingram, Carolyn Johnson, and Michael Rochemont, for your help, your guidance and your clarity and insight in answering questions. And thanks to you Pat Shaw, for, among other things, getting me through my first year at U B C ; your words "finish the year well" are a source of inspiration to me. Although academic guidance and encouragement are crucial, I could not have completed this task without the support of my peers and friends in the Linguistics Department. Thank you Eleanor Blain, Helmi Braches, Elizabeth Currie, Ping Jiang-King, Myles Leitch, Lisa Matthewson, Nike Ola, Taylor Roberts, Kimary Shahin, Takeru Suzuki, Aki Uechi, Yan Feng Qu, for always stopping to chat, for the many hours of commiserating, for your sympathetic and supportive shoulders, and for more than periodically checking up on me. Thank you Juta Kitching, for your encouragement and for always having time to talk. And thanks Susan Blake, for your very dear friendship. My close friends outside the world of UBC helped me maintain my sanity. Thank you Cathy Simon and Joe Moulins, for reminding me of what is truly important in life; Cory Hackett, for all the delightful philosophical banter; Deirdre Domegan, for the Sunday brunch distractions from linguistics; Carmina Gaite, for being a sister; Nicola McDonald, for the long talks and your wise advice; Stephanie Trenciansky, for your hugs, smiles and empathy; William Burchill and Michael Kyba, for the many walks and talks; M C Busque, Lea K i v i , Monica Kowal, Ursula Mueller, Debbie Sabadash, and Leslie Seifert, for your friendship and all your long-distance support. Work on this thesis has spanned many years. However, the bulk of what follows was written the summer of 1996, during which time I lived with dear friends. The Gaites are my second family. Thank you Carmina, David, Carlene and Carlos for your home, your hospitality and most of all your laughter. I would like to thank my brothers, Federico and Bernardo Sanchez, and my sister-in-law Kim Montgomeriefor their constant words of encouragement and for putting up with a flood of e-mail complaints. Most importantly, I wish to thank my parents, Eszter and Adolfo Sanchez. Without their love and support, none of this would have been possible. Mom, for always listening with pride to my alien words of linguistics, and Dad, for always asking if I was happy, I dedicate this thesis to you.  xiii  Chapter 1 INTRODUCTION  This thesis examines the nature of the language deficit called agrammatism. M y goal is to account for the omission and substitution errors that characterize agrammatic production. For omission I focus on the nature of the syntactic categories and projections that underlie agrammatic clauses. For substitution I discuss the identity of the syntactic features that are mis-selected arid their organization into paradigms. Both types of errors lead me to a discussion of Lexical Insertion, the process by which words are inserted into syntactic structures (sentences).  I will  show that omissions and substitutions result from two distinct deficits within the mechanisms governing Lexical Insertion. Linguists have used "internal" data to formulate hypotheses about the structure of language; that is they relied solely on data provided by the intuitions of non-pathological adult native speakers.  "External" data have also been included (Jakobson, 19??).  External data  consists of data from language acquisition (first and second), slips of the tongue in nonpathological speakers, and language disorders, in other words, language data other than grammaticality judgments from native speakers.  Many have turned to language acquisition in  order to test their hypotheses about the structure of language and to provide a better understanding of language development (Bloom, 1970; Hyams, 1986a; Davis, 1987; O'Grady, Peters & Masterson, 1989; Radford, 1990; Guilfoyle & Noonan, 1992; Deprez & Pierce, 1993; Poeppel & Wexler, 1993; among many others).  Others have addressed the issues raised by  language breakdown. More specifically, quite a few attempts have been made to develop a syntactic account for what is called agrammatic speech (Caplan, 1987; Grodzinsky, 1984b, 1990; to namejustafew). Agrammatism is the linguistic syndrome that is usually associated with Broca's aphasia. Broca's aphasia results from damage to the anterior portion of the left hemisphere of the brain (Broca's area). The most striking characteristics of agrammatic production are the omissions and  1  substitutions of various parts of speech.  This thesis investigates agrammatic production with  special attention to the following syntactic categories: Noun, Verb, Determiner, Auxiliary/Modal. It focuses on the omission and substitution errors produced by agrammatics of five different languages, English, Dutch, German, French, and Italian. The utterances in (1) are a sample of agrammatic omission errors.  Examples of omissions in English are from Menn (1990) (la),  Dutch from Kolk etal. (1990) (lb) German from Stark & Dressier (1990) (lc), French from Nespoulous et al. (1990) (Id), and Italian from Miceli & Mazzucchi (1990) (le). Omissions are indicated by square brackets' []'. Items that are omitted are underlined in the translation. (1) Agrammatic clauses with omissions a. the wife was showing [DP] the clock "the wife was showing him the clock"  English  b. zondagochtend [Aux] [DP] gewerkt Sunday-morning [Aux] [DP] woricPart "Sunday morning I had worked"  Dutch  c. spater begutachtet [DP] seinen Kukuruz later examines [DP] his corn "later he examines his corn"  German  d. et peu apres [DP] contemple [Det] recolte and little after [DP] gazes at [Det] harvest "and a little later he considers the harvest"  French  e. i (gli) uccelli non mangiano [DP] the:M-Pl (the:M-Pl) birds not eat:3Pl [DP] "the birds don't eat the seeds"  Italian  The utterances in (2) illustrate agrammatic substitution errors. Examples of substitution errors in English are from Menn (1990) (2a), Dutch from Kolk et al. (1990) (2b) German from Stark & Dressier (1990) (2c), French from Nespoulous etal. (1990) (2d), and Italian from Miceli & Mazzucchi (1990) (2e). Italics indicate the item that is substituted; the content of the following rounded brackets '()' represents the speaker's target. Items that are substituted are underlined in the translation.  2  (2) Agrammatic Substitutions a. I [Cop] flat on the (my) back "I was flat on my. back"  English  b. de dief kwam binnen door de (het) raam the thief come:Pst in through the:MIF (the:N) window "the thief came in through the window"  Dutch  c. [DP] hatte einem (einen) Korb gepackt [DP] have:Pst-3Sg cuDat (a:Acc) basket pack:Part "she had packed abasket"  German  d. je marchais avec le (la) grande canne anglaise I walked with the:M (the:F) long stick English "I was walking with die long English walking stick"  French  e. il contadino mangia i (il) granone the farmer eats the:Pl (the:Sg) grain: Sg "the farmer eats die grain"  Italian  Agrammatic data serves the same purpose as internal data: as a source of empirical evidence for building a model of grammar. As such, I use agrammatic data to test and constrain linguistic theory. However, the interplay between linguistic theory and external data is a twoway street, wherein the Principles of Universal Grammar can explain the mechanisms that produce agrammatic utterances. Thus this thesis serves not only as a bridge between theoretical linguistics and the clinical disciplines but also as an example of how the two disciplines can mutually inform one another.  1.1  THEORETICAL F R A M E W O R K  In this section I introduce the general constructs and assumptions that will be used throughout the thesis. Much of the analysis of the agrammatic data requires more specific theoretical proposals. These will be developed as they become relevant. I begin with the overall framework.  3  1.1.1  General Model  The theoretical constructs that I adopt hail from Generative Grammar. Each of us is endowed with the ability to acquire and ultimately manipulate language. Following Chomsky (1965, 1981, 1986, 1993), I assume thatthis linguistic knowledge is governed by the principles of Universal Grammar (henceforth UG). There has been much debate in the linguistic literature about the nature of U G . At the very least, however, U G is composed of a dictionary (Lexicon), a syntactic component (Syntax), a semantic component (Semantics) and a phonological component (Phonology). The Lexicon is a structured list of words.  Each word is a Lexical Item (LI). The  information contained in the Lexicon includes the meaning, the syntactic category (noun, verb etc.), other syntactic features (such as agreement, etc.), and the pronunciation of the LI. Each of these four sets of information is discussed in detail in chapter 5. The Lexicon feeds into the Syntax. The process by which L i s are inserted into the Syntax is called Lexical Insertion. I discuss the nature and locus of Lexical Insertion in chapter 5. The structure that is produced by the Syntax is in turn input to the Phonology and the Semantics. The model is schematized below. Figure 1: General Model of Universal Grammar LEXICON SYNTAX SEMANTICS  \  PHONOLOGY  The Syntax is responsible for sentence structure. It may be either derivational or representational; in other words, a series of operations may apply to one structure deriving another, or else syntactic representations may be subject to a set of constraints that eliminate illicit structures. I treat the two as equivalent. Any implications for the analysis of the agrammatic data will be discussed when relevant.  4  The Phonology produces the string of sounds associated with the intended utterance. It governs word and sentential stress, syllable structure, and allophonic variation, among other things. The Semantics specifies the semantic features of a morpheme and the relations between the morphological items. The semantic component may also be the locus of discourse and pragmatic constraints. Each component is relevant to agrammatic production. considering substitution errors.  This is in evidence when  There are three types of substitution errors: phonologically  related, semantically related and syntactically related. Although I discuss each in turn, I focus on the errors that are syntactically constrained. In other words, I concentrate on the syntax as the central locus of the deficit.  1.1.2 Phrase Structure  I assume that every syntactic category (noun, verb, and so on) projects syntactic structure. I assume that these structural representations are constrained by X-bar theory, which licenses phrase structures of the following basic type (Chomsky, 1986): (3) X P Projection of a Head X XP  Specifier  X' X  Complement  In (3), X is the head. Sister to X is its complement. Sister to X ' is the specifier position, [Spec XP].  The position of specifiers and complements relative to their sisters is subject to  parametric variation: they may be either to the left or to the right of their sisters. Some syntactic heads, like noun (N) and verb (V), project this structure. Other heads are adjuncts, like adverb (Adv). Adjuncts are adjoined to syntactic projections in the following manner.  5  (4) Adjunction to X P XP Y  " " ^ ^ ^ X P  Specifier  X' X  Complement  The adjunct Y is adjoined to X P . Within a given language, all phrasal projections have the same hierarchical structure. For example, in English a noun phrase (NP) always has the following structure (the rounded brackets indicate that the specifier and complement positions may not be filled, or may simply be absent). (5) Structure of N P  I also adopt the stronger hypothesis that in all languages X P has the same hierarchical structure. That is, for all languages that have a head N , each of these languages projects the structure in (5) (taking parametric variation with respect to order into consideration). However, not all languages have the full inventory of heads . 1  1.1.3 Syntactic Features  Following Chomsky (1970), Jackendoff (1977) and many others I assume that syntactic heads are bundles of categorial features. For example, Chomsky (1970) proposes that Noun is [+Nominal, -Verbal] and Verb is [-Nominal, +Verbal]. This set of features distinguishes between the four categories N(oun), V(erb), P(reposition) and A(djective), as shown in (6). Kayne (19??) presents a model whereby Specifiers are always to. the left of the head and Complements are always to the right. 1  6  (6)  [+ Nominal]  [- Nominal]  A N  V P  [+ Verbal] [- Verbal]  The category names (N, V , etc.) are simply abbreviations for the feature bundles. It is the features which project phrases. Although the consensus is that syntactic categories are made up of features, a debate revolves around which features are relevant. Fukui (1986), Abney (1987) and Dechaine (1993) (among others) propose more comprehensive categorial models that include categories beyond N , V , P and A . Fukui (1986), Fukui & Speas (1987) and many others propose a distinction between lexical and functional categories. Lexical categories consist of the traditional categories that denote entities (e.g. wolf), actions (e.g. eat), locations (e.g. behind) and states (e.g. beautiful) i.e. N , V , P and A . Functional categories have a more abstract interpretation. In general terms, they locate in space/time/etc. the entities, actions, locations and states. Consider the phrase this wolf. Among other functions, the determiner this (category D) locates in space the entity wolf. Consider a second phrase this wolf ate. The tense marking past (category T) locates in time the action eat.  For Fukui (1986), every category has precise featural  specifications, as shown in (7). [-Functional]  (7)  [-Kase]  [+Kase]  [-Nominal] [-Verbal]  [+Verbal]  [+Nominal]  N  [-Nominal]  V unacc  V trans/unerg  [+Nominal]  7  [+Functional] [-Kase]  [+Kase]  C that  C +WH  D the  D 's  I to  I Tns/Agr  Abney (1987), on the other hand, proposes the smaller set of features in (8). (8)  [-Functional] [-Nominal] [+Nominal]  [+Functional]  V , Aux, P N , A , Q, Adv  I, C D , Deg  Following Chomsky (1970), both Fukui and Abney capture the nominal-verbal distinction. Fukui includes both [+/- Nominal] and [+/- Verbal] whereas Abney settles for [+/Nominal]. Their feature systems also distinguish between lexical and functional categories by having lexical categories defined as [-Functional] and functional categories defined as [+Functional]. However, several problems arise with respect to these models. The abundance of features proposed by Fukui leaves us with 'empty slots': we expect to find the [+Kase] counterparts to A and N , for example. With Abney's model, a problem opposite to Fukui's arises: too many distinctions are collapsed. For languages like English, A and N are different and need to be identified as such. In this thesis, I adopt Dechaine's (1993) feature system.  Dechaine's (1993) feature  system is shown in (9). Like Fukui and Abney, Dechaine captures both the lexical-functional and the nominal-verbal distinctions: functional elements are [+Functional]; nominal elements are [+Nominal]; lexical and verbal elements are unspecified for [Functional] and [Nominal], respectively.  Nominal Referential Functional  C  T  V  +  + +  +  K  D  N  +  + + +  + +  +  However, Dechaine differs from Fukui and Abney in two ways.  P  A +  First, she adopts the  three features, [Referential], [Functional], and [Nominal] . This results in fewer features (and 2  Dechaine (1993) includes the feature [Referential]. She argues that it is required to distinguish between categories that have privileged relationships and categories that do not, in other words to account for the selectional properties of the various syntactic categories. 2  8  consequently fewer distinctions) than Fukui, but more than Abney. Dechaine's system is more accurate and, consequently, more desirable. Second, Dechaine differs from Fukui and Abney is in her use of privative features. Having both [+] and [-] values of each feature forces us to stipulate which value is to be selected for a given syntactic process. If, on the other hand, features are privative, then necessarily it is the feature that is present that is selected for a given syntactic process. No stipulation is required. Once again, Dechaine's is the preferred model; as such, it is the model that I adopt as the basis of my analysis of categorial disturbances in agrammatism.  To account for the full range of  agrammatic omission errors I modify Dechaine's feature inventory (chapter 5).  1.2  OUTLINE OF T H E THESIS  The thesis is organized into the following sections. In chapter 2, I provide a discussion of agrammatism. I begin with a general description of the agrammatic deficit, and defend the thesis that agrammatism is a cohesive syndrome. I then turn to the data that is analyzed in this thesis. The data consists of agrammatic production narratives from five languages: English, Dutch, German, French and Italian. I review the criteriafor the selection of subjects. In addition I describe the production tasks as well as the procedures for reconstructing agrammatic utterances. The chapter ends with a discussion of the questions that are relevant to the study of agrammatism and a brief overview of the issues that are addressed in the following chapters . In chapter 3,1 focus on omission errors. I show that Agrammatic Grammar is governed by the principles of Universal Grammar. I continue with an investigation of the number of (i)  DP D'  TP T'  D ^ ^ ^  NP T VP N' V N V N and D have a privileged relationship. V and T have a privileged relationship. P and A are not involved in privileged relationships. Dechaine (1993) also includes K(ase) and C(omplementizer)  9  functional projections in agrammatic clauses, concluding that (at least) CP, T P and a functional projection between TP and V P (possibly AgrOP) are required. The rate and distribution of omissions is then addressed. It is shown that lexical categories are produced more frequently than functional categories, and nominal categories more frequently than verbal categories, with the following retention hierarchy: N > V > D > Aux/Modal, where ">" means "better retained than". In chapter 4, I address the nature and distribution of substitution errors.  I show that  substitutions involve agreement, Case and tense features, but any substitution involves only a single feature.  I argue that this Single Feature Constraint on substitution results from the  paradigmatic structure of the Lexicon. In chapter 5, I develop an account of agrammatic omission and substitution. I argue that the deficit is not located in the phonological, morphological, syntactic or semantic components. I conclude that the agrammatic deficit is located instead at the interface between the Lexicon, Syntax and Phonology. Agrammatic omissions and substitutions result from damage to the mechanisms responsible for Lexical Insertion. Chapter 6 concludes the thesis.  Here, I summarize my findings and conclude with  suggestions for further research. Following chapter 6, I include an appendix of the agrammatic data.  as categories that are not in privileged relationships. See Dechaine (1993) for arguments supporting the choice of this and the other two features. 10  Chapter 2 AGRAMMATISM  Agrammatism in Broca's aphasia results from damage to the anterior portion of the left hemisphere of the brain. Historically, agrammatism was referred to as a production deficit, with little attention paid to comprehension (Pick, 1913; Goodglass, 1968; Tissot, Mounin & Lhermitte, 1973). As research progressed, discussion about agrammatism focused on reduced abilities in both production and comprehension (Zurif etal., 1972; Zurif etal., 1974; Miceli etal., 1983). The disorder was shown to involve differing retention of lexical categories (noun, verb, adjective, adverb, and preposition) and functional categories (determiners, tense markings, and other grammatical morphemes). The deficiency may be characterized by an ability to retain lexical categories, accompanied by widespread omission of functional categories. However, of the lexical categories retained, there is a tendency for agrammatics to use nouns more often than other categories. Moreover, it is not the case that all function words are always omitted; rather, there is an increased probability of omission correlated with the severity of the deficit. This chapter begins with a description of agrammatic production.  Because of the  variability between patients and the dissociation between comprehension and production, it has been argued that agrammatism cannot be considered a cohesive syndrome (Badecker & Caramazza, 1985). However, I argue that neither variability nor dissociation are valid arguments against agrammatism being a single syndrome and conclude that agrammatism is a unified deficit. Before delving into the linguistic analysis of agrammatism I present the data to be analyzed. I describe the subjects from Menn & Obler (1990), focusing on the severity of the deficit and the method of reconstructing agrammatic utterances. I end the chapter with a summary of the hypotheses to be tested.  /  11  2.1  A G E N E R A L DESCRIPTION OF A G R A M M A T I S M  There is a wide range of language disorders associated with brain injury. Lesser (1978) estimates that a total of 78 different subclassifications have been proposed. Of the 78, only two are assigned relatively firm status: Broca's and Wernicke's aphasia.  Broca's aphasia, or  nonfluent aphasia, results from a lesion to the anterior portion of the left hemisphere of the brain. Agrammatism is the linguistic syndrome that is usually associated with Broca's aphasia. Wernicke's aphasia, or fluent aphasia, is associated with a lesion to the posterior regions of the left hemisphere. Paragrammatism is the linguistic syndrome usually associated with Wernicke's aphasia. These two are frequently considered opposing syndromes. In very general terms, agrammatic speech is slow and labored, and is marked by omissions; paragrammatic speech is fluent, but displays numerous substitutions. A n on-going debate among some aphasiologists is whether or not there is a distinction between agrammatism and paragrammatism (Pick, 1913; von Stockert & Bader, 1976; Friederici, 1982; Shapiro & Levine, 1990; Blumstein et al., 1991; Kolk & Heeschen, 1992, among numerous others).  Wernicke himself viewed the two as separate syndromes, using this  .difference to argue for distinct anatomical sites for language: "The great variability of the clinical picture of aphasia moves between the two extremes of pure motor aphasia [agrammatism] and the pure sensory form [paragrammatism]. The demonstration of these two types must be regarded as conclusive proof of the existence of two anatomically separate language centers" (Eling, 1994: 68).  Although the evidence favors two syndromes, this debate does not affect the issue under  discussion in this thesis. The production data remain the same regardless of whether they are part of a larger or smaller syndrome. I therefore leave the debate for others to continue and focus on the specifics of the agrammatic deficit.  12  Agrammatic production is frequently described as resembling telegraphic speech . 1  The  term "telegraphic speech", however, is a misnomer. The linguistic features of agrammatism are not the same as those found in telegrams (Tesak & Dittman, 1991). Telegrams show the following linguistic characteristics: complete omission of functional categories, virtually no omission of lexical categories, and no substitution errors. While agrammatics omit functional categories, they do not omit all functional categories all the time: there is a gradient loss of functional categories in agrammatic speech. This is not found in telegraphese. Further, agrammatic omission errors include lexical categories, albeit to a lesser degree. This too contrasts with telegraphese. Lastly, agrammatics produce substitution errors, which are completely absent from telegraphic style. These errors involve producing the wrong form of a particular category. Substitutions may be semantic in that the wrong word altogether is selected (producing boy instead of girl, for example), or they may be (morpho)syntactic (selecting tell instead of told) . Clearly, agrammatic speech is not telegraphese. 2  Although most studies have focused on omission and substitution errors, these are not the only features of agrammatic speech. Agrammatics suffer from slow and labored speech, and produce fewer words per utterance. Moreover, agrammatics produce utterances that have reduced structural complexity (Saffran et al., 1989), thus lacking embedded clauses. In addition to the linguistic deficit, agrammatics may have articulatory and phonatory difficulties. These difficulties result in segmental and voicing errors . 3  Agrammatics may also suffer from impaired comprehension, as well as disturbed reading and writing. However, each modality suffers independently of the others. This means that while production is impaired, comprehension may be intact(Kolk etal., 1982; Micelietal., 1983).  Kolk & Heeschen (1992) offer an alternative explanation of agrammatism. They argue that agrammatic production is the result of overuse of ellipsis. Notice however, that ellipsis cannot address the substitution errors also found in agrammatism. Substitutions may also be phonological. This is discussed in chapter 4. See Sanchez (1992a) for a detailed discussion of a case study involving a non-fluent speaker with a phonological deficit. 1  2 3  13  A final characteristic of agrammatic speech is variability. There is variation both within an individual and between speakers. There are many different influences on variability, which I address below.  2,2  A G R A M M A T I S M S A COHESIVE SYNDROME  A further debate in the field of aphasiology is whether or not agrammatism is a single syndrome (Badecker & Caramazza, 1985; Miceli et al., 1989; Caplan, 1991, 1987; Grodzinsky, 1991). The crux of the debate lies with two issues: variability and a comprehension-production dichotomy. Badecker & Caramazza (B&C) (1985) argue that agrammatism ought to be viewed as several distinct syndromes. As evidence, they cite variability between patients and the fact that difficulties in comprehension and production do not always co-occur . 4  Because agrammatic  patients may omit main verbs, incorrectly nominalize these same verbs, display semantic illformedness, and show different patterns of omission of free and bound functional morphemes, B & C conclude that either 1) deficits exist in a variety of processing components or 2) the impairment is of a higher order, with the expression of the disturbance being affected by other linguistic features. There are two issues which must be addressed in the single-syndrome debate: variability and the comprehension-production dissociation. First, I argue that variability cannot be an argument against viewing agrammatism as a single syndrome since it is predicted to be a characteristic of language breakdown.  Second, I argue that the comprehension-production  dissociation is illusory. The two modalities appear to be affected independently because the disruption to comprehension can be easily masked. I conclude that agrammatic production results from a single cohesive syndrome.  Others have used this dissociation to argue for a distinction between the processing mechanisms for comprehension and those for production (Nespoulous et al., 1988; Caramazza& Hillis, 1989; Druks& Marshall, 1991). 4  14  2.2.1 Variability  There is variation both between agrammatics (inter-agrarnmatic variability) and within a single individual (intra-agrammatic variability). For variation within an individual, there are two types: variation within a session (intra-sessional variability), and variation from session to session (inter-sessional variability). Intra-sessional variability may be due to task variation (see Heeschen & Kolk (1988) on German-speaking aphasics). Within a single session, frequency of omission and substitution may change depending on the task.  With a more constrained  production task (picture description as opposed to general conversation) agrammatics produce fewer omissions but more substitutions (Hofstede & Kolk, 1994). Inter-sessional variability is due to performance factors such as tiredness, time of day, and so on. Agrammatics produce fewer errors when they are rested (as do normals). Moreover, agrammatics produce fewer errors in morning sessions than in afternoon sessions. Inter-agrarnmatic variability is of two types: variation within the same language (intralinguistic variability) and variation between languages (inter-linguistic variability). Intra-linguistic variability results from several factors. The first factor is degree of impairment. Generally the more impaired the individual, the more omissions are produced (I show this in chapter 3). The second influence is attentional factors and compensatory strategies.  No two individuals adopt  exactly the same strategies to compensate for their deficit, neither do they have the same attention rate for any given task. While some agrammatics prefer avoidance, others adopt the over-use of familiar words, or the use of semantic opposites or the use of circumlocutions (describing the meaning of the intended word), and so on. Moreover, the interaction between the actual deficit and the compensatory mechanisms further affects variation (Nespoulous & Dordain, 1991) . The 5  third factor relates to the intra-agrammatic influences stated above. If agrammatic A performs  Nespoulous & Dordain (1991) argue that (French-speaking) agrammatics do not use automatic processing for functional morphemes and that agrammatic variability is due to an interaction between the actual deficit and compensatory mechanisms.  5  15  task (a) in the morning and agrammatic B performs task (b) in the afternoon, they are sure to show variation in omission and substitution. Inter-linguistic variation results from several factors. First, variation is due to the interagrammatic factors described above. Differing degrees of impairment and differing compensatory strategies affect levels of omission and substitution across languages, as they do within languages.  Second, intra-agrammatic factors also contribute to variability across languages.  And third, language-specific structural differences influence inter-linguistic variation (Caplan, 1987, 1991; Grodzinsky, 1984ab, 1991). Agrammatism may appear quite different from one language to another, with language-specific constraints governing agrammatic errors (Peuser & Fittschen, 1977; Grodzinsky, 1984ab; Caplan, 1987) . Bates & Wulfeck state that "language 6  differences account for more variance than patient group difference in all our experiments to date" (1989a:329).  To give one example, agrammatics produce more substitution errors in non-  concatenative languages where omissions would violate morphological restrictions. There are many different influences on variability. What is important to notice is the cumulative effects of the various factors. One set of factors (set 1) influences intra-agrammatic variability. This set, along with an added series of factors (set 2) influence inter-agrammatic intra-linguistic variability.  Language-specific structural differences make up a third set of  influencing factors (set 3). Set 1, 2 and 3 affect inter-linguistic variability. It is not surprising, therefore, to find variability in agrammatic production. The predictions from syntactic theory with respect to variability are clear: variability is predicted to occur. There are two areas of interest: variation within a language (intra-linguistic variability) and variation between languages (inter-linguistic variability). Intra-linguistic variability is an expected characteristic. Any given language includes a set of syntactic categories, say N(oun), V(erb), A(djective), and P(reposition). Each of these four categories consists of a There is a substantial body of more recent research that provides evidence for such languagespecific constraints. Among these we find the following: Chen, 1989, 1993; Hagiwara & Caplan, 1990; Kehayia et al., 1990; Lorch, 1990; Kehayia & Jarema, 1991; MacWhinney & Osman-Sagi, 1991; MacWhinney etal., 1991; Safi-Stagni, 1991; Vaid & Pandit, 1991; Wulfeck 6  16  different bundle of categorial features. Minimally, N is [+Nominal, -Verbal]; V is [-Nominal, +Verbal]; A is [+Nominal, +Verbal]; and P is [-Nominal, -Verbal]. At this point, the identity of the features is not relevant. What is important is that the feature bundles are distinct from each other. Syntactic categories are distinguished by their categorial features. A l l else being equal, a linguistic disorder should reflect this contrast.  We would, therefore, expect category-specific  rates of omission and thus intra-linguistic variability. Indeed, it would be surprising if we didn't see any variability. Variability is the reflection of the linguistic system itself. A similar argument can be made for inter-linguistic variability. First, different languages employ different sets of syntactic categories. For example, English has N , V , A and P, whereas Walpirihas N / A , V , and P. Omissions errors in Walpiri would necessarily look different from omission errors in English by virtue of the fact that their categorial inventories are different. Second, a given syntactic category does not necessarily encode the same series of agreement, Case and tense features for each language that employs it. This is illustrated by contrasting the definite article in English and French. The English definite article the is unspecified for gender or number.  French on the other hand has three definite articles: the masculine singular le, the  feminine singular la, and the plural les. The expectation is that these two languages should display different omissions and substitution patterns. Third, languages display surface variation for identical relational structures. Consider abstract Case, for instance. English Case morphology is impoverished (only pronouns show Case distinctions: I (nominative Case) versus me (objective Case) versus my/mine (genitive Case)). Hungarian Case morphology, on the other hand, is quite substantial. English relies heavily on word order for subject and object interpretation whereas Hungarian relies on overt Case morphology. linguistic disorders to reflect each of these distinctions.  Once again, we expect  Syntactic theory therefore predicts a  rather high level of variability in agrammatic production. Consequently, variability is not an argument against agrammatism being a single syndrome.  etal., 1991; Hagiwara, 1993; Haverkort, 1993. These findings parallel those of Fromkin (1968, 1971) where phonological errors in normals obey the phonological rules of the language. 17  2.2.2 Comprehension-Production Dissociation  I now address the comprehension-production dissociation. B & C argue that because the two modalities do not suffer equally within a Broca's aphasic, agrammatism cannot be considered a cohesive syndrome. First, B & C assume that agrammatism must affect these two modalities equally. This assumption, however, is false. Second, B & C maintain that there is in fact a dissociation between the two modalities. I argue that the dissociation is illusory. To begin, a certain amount of independent processing for each of the modalities is required since comprehension involves hearing and production involves speaking.  The link  between the utterance and the motor system involves auditory mechanisms for comprehension and articulatory mechanisms for production.  The link between the motor system and the  phonological processes involves the auditory pathway for comprehension and the oral pathway for production. Thus, the differences in modality (hearing versus speaking) forces two distinct pathways at the outset. I schematize the pathways below. Figure 2: Differing Comprehension and Production Modalities UG Ears  Comprehension  Production  Mouth  The principles of U G underlie both modalities. However, comprehension and production display independent pathways beyond U G . The fact that there are two distinct pathways allows for the possibility that the two modalities are independently affected. If the deficit lies outside U G , production may be disrupted independently of comprehension. Kolk et al. (1982), Miceli et al. (1983) and Nespoulous et al. (1989) argue that there is indeed a dissociation between production and comprehension but that this dissociation reflects two distinct impairments: expressive (production) and receptive (comprehension) agrammatism.  Although Broca's  aphasics usually display a disruption to both production and comprehension, they claim this need  18  not be so: expressive and receptive agrammatism can occur independently of each other. Broca's aphasics with only a production deficit have expressive agrammatism, those with only a comprehension deficit have receptive agrammatism . 7  However, many have argued that 'intact' comprehension is in fact illusory. Numerous studies have shown that agrammatics do have comprehension difficulties (Heilman & Scholes, 1976; von Stockert & Bader, 1976; Zurif & Caramazza, 1976; Caramazza & Berndt, 1978, 1985; Goodglass etal., 1975; Bradley etal., 1980; Caramazza etal., 1981; Bates et at., 1987b; among others) . Moreover, Zurif et al. (1972) and Zurif et al. (1976) argue that agrammatism involves 8  all language modalities: speaking, hearing, reading and writing. Because these difficulties lie with functional morphemes rather than lexical categories, the comprehension disturbance can be masked. The comprehension deficits in agrammatism are not as obvious as the production deficits and require specialized testing. Although work on the production-comprehension dissociation is still in flux, I maintain that agrammatics do indeed have both a production and comprehension disturbance, contrary to B & C ' s claim. I now proceed with the assumption that agrammatic production is a cohesive syndrome and leave the debate for others to continue.  2.3  T H E DATA  The empirical data for this thesis comes from the production of ten agrammatic speakers, the transcripts of which are collected in Menn & Obler (1990). agrammatic production data is dependent on several factors.  The interpretation of the  First, since I am comparing ten  different speakers, I must establish that each of the ten speakers has the same syndrome, namely agrammatism. To this end, I describe the criteria for selection. Second, the nature of the For more discussion on the production-comprehension dissociation, see the following authors: Hittmair-Delazer et al., 1994; Friederici et al., 1992; MacWhinney et al., 1991; Lukatela et al., 1988; Grossman etal., 1986. M y personal experience with agrammatics parallels these studies. I have never observed agammatics that display intact comprehension. 7  8  19  production tasks allows us to determine the intent of the speaker, and thus their linguistic target. This leads me to the methods of reconstruction. Agrammatic production displays omission and substitution errors. The agrammatic utterances therefore need to be reconstructed. Knowledge of the linguistic target is therefore paramount.  2.3.1 The Subjects  The agrammatic data under study are transcripts collected in Menn and Obler (1990) (M&O). The data was produced by ten agrammatic speakers, two speakers of each of the following languages: English, Dutch, German, French, and Italian. Since the personal and educational background for each agrammatic is different, each subject was paired with a control. Non-pathological native speakers were used as controls. The controls were matched for age, sex, bilingualism, education, background, and handedness. Four criteria were used to select the subjects. These are the following. (1) Criteria for Selecting Agrammatic Subjects a. Demographic b. Neurolinguistic c. Neurological d. Neuropsychological The demographic criteria ensured that subjects were native speakers of the language under study, were preferably monolingual, used the native language most of the time (if not monolingual), were between 18 and 65, and were literate. The neurolinguistic criteria ensured that the subjects were all agrammatic by clinical standards. M & O define "agrammatic by clinical standards" "as being moderately fluent, having slow and halted speech, with three or four words being the usual maximum uninterrupted string" (1990:14). Further, these criteria ensured that the subjects could speak well enough to produce at least 250 words, could produce recognizable clauses and phrases of at least three words, had  20  adequate comprehension to perform the required tasks, had minimal articulatory impairment (dysarthria), and only mild word-finding difficulties. The neurological criteria ensured that the subjects all had a unilateral left-sided lesion (preferably anterior to the rolandic fissure) and were physically stable (i.e. at least 3 to 4 months post onset of the lesion). The neuropsychological criteria ensured that subjects were able to concentrate, could understand the production tasks, were free of general cognitive damage, and had minimal visual and auditory problems. Each of the subjects under study in this thesis meets the criteria described above.  2.3.2 Description of the Data  Each of the agrammatics and controls were required to produce four narratives. These narratives consist of the following: a history of their state of health, a description of a complex picture (the 'Cookie Theft' picture), a description of four picture sequences (a farmer planting and harvesting crops; a thief caught in the midst of stealing; a couple on a picnic with their meal stolen by a dog; a man who oversleeps, is woken and then rushed by bis wife, and falls asleep at work), and the telling of 'LittleRed Riding Hood'. The speech of the agrammatics is divided into utterances. string in which a tensed verb is or should have been used. omissions and substitutions.  Utterances are defined as a Agrammatic speech includes  This means that missing and incorrect elements must be  reconstructed. There are several factors that influence the choice of the reconstruction. First, only the addition of syntactically obligatory elements is permitted (Brown, 1973) - give a more detailed set of criteria for defining 'obligatory'. Second, production task, context, discourse sequence and intonation patterns are used to determine the linguistic target.  Third, stereotypes and  'metacomments' (e.g. "yeah, right") are not analyzed. Lastly, utterances are considered to  21  include a substitution when the syntactic and discourse contexts force the use of a word form other than that being produced. A l l the data analyzed in this thesis comes from the Menn & Obler (1990) collection. The transcripts from each language was collected and discussed by different researchers. The English data is from Menn (1990); the Dutch is from Kolk, H e l i n g & K e y ser (1990); the German is from Stark & Dressier (1990); the French is from Nespoulous, Dordain, Perron, Jarema & Chazal (1990); and the Italian is from Miceli & Mazzucchi (1990). The reader should keep in mind that all data comes from these respective authors, unless otherwise indicated. As mentioned above, the transcripts have been segmented into utterances. For omission errors, I have adopted Menn & Obler's (1990) notation, placing the omitted elements in square brackets, "[]".  For substitution errors, I have created my own notation: substitutions are  italicized, with the target provided within the following parentheses. A final comment about the data deals with the number of tokens.  The agrammatic  production data that I analyze involves a small number of tokens. Despite this small sample I must emphasize the importance of any aberrations from normal speech. Any production distinct from normal production must be accounted for. Any patterns governing disturbed production must be predicted by Universal Grammar. If patterns generalize across languages, as I show that they do, it behooves U G all the more to account for them.  2.4  ISSUES IN A G R A M M A T I S M  The current consensus is that agrammatics tend to omit functional categories in general (Caplan, 1987; Grodzinsky, 1984b, 1990). This consensus has led to a line of research that leaves five issues unresolved. First, since omission of functional categories is the most striking aspect of agrammatic production, attention has centered on the behavior of functional categories rather than that of lexical categories. categories exist.  A few studies on the production of specific lexical  Friederici & Saddy (1991) discuss agrammatic processing of the various  22  syntactic categories. Fradis et al. (1992) compare Romanian-speaking aphasic production of nouns, verbs, adjectives and adverbs. However, there have been few attempts to compare the relative omission (and substitution) rates of the various lexical and functional categories . 9  Second, despite the consensus that functional categories are impaired, systematic studies of individual functional categories are scant (Grodzinsky, 1988, 1991; Hofstede & Kolk, 1994). There have been studies which touch on prepositions and particles (Jackendoff, 1977; Friederici, 1982; Grodzinsky, 1988), pronouns (Linebarger et al., 1983; Jarema et al., 1987; Friederici et al., 1991; Jarema & Friederici, 1994) and determiners (Goodglass et al., 1972; Zurif et al., 10  1972; Gleason et al., 1975; Goodenough et al., 1977; Schwartz et al., 1980; Bernstein, 1994; Jarema & Friederici, 1994). However, these studies focus only on comprehension , which 11  brings us to the third issue. Most studies focus on comprehension rather than production. What research there is on agrammatic production of functional categories focuses on prepositions (Friederici, 1982; Grodzinsky, 1988; Reyes, 1989; Tesak& Hummer, 1994) and inflectional morphology (Miceli 12  The cross-linguistic collections of agrammatic narratives in Menn & Obler (1988, 1990) certainly attempt to tabulate the omission and substitution rates of the different categories (the languages include English, German, Swedish, Icelandic; French, Italian; Polish, Serbo-Croatian; Hindi; Finnish; Hebrew; Chinese; Japanese). They argue that a well-defined hierarchy of omission cannot be established but that the following tendency is exhibited: Aux & empty V > pre/post-position & pronouns > Det > N & V . Moreover they argue that the agrammatic deficit cannot be reduced to a purely phonological, morphological or syntactic deficit. For many of the languages not enough data was collected to be able to make any generalizations. However, in this thesis I reanalyze their data and find that, for the 5 languages that I focus on, generalizations are indeed possible. Friederici et al. (1991) report on aphasic comprehension of pronouns in German, Dutch, concluding that the errors depend on the semantic and syntactic information that is encoded rather than their phonological status, syntactic category or grammatical relation. Jarema & Friederici (1994) investigate agrammatic comprehension of articles and pronouns in French, with similar conclusions. Jarema & Kehayia (1992) are a notable exception. They investigate the comprehension and production of number and tense in French-speaking agrammatics. Friederici (1982) focuses on the production and grammaticality judgment of aphasics with regards to prepositions. She finds that there is a distinction between Wernicke's and Broca's aphasics. She finds that Wernicke's aphasics perform better with syntactically determined prepositions for both tasks. Broca's aphasics, on the other hand, perform well on the judgment task regardless of the role of the preposition. However, with the production task, agrammatics have difficulties producing syntactically conditioned prepositions. Grodzinsky (1988) investigates agrammatic treatment of prepositions and argues that the deficit involves government. 9  10  11  2  23  etal., 1983; Lorch, 1990; MacWhinney& Osman-Sagi, 1991; De Bleser & Luzzatti, 1994) . 13  Virtually no attention has been paid to the production of functional categories within the noun phrase, namely determiners and pronouns. Although some argue that similar aspects are affected in comprehension and production (Bradley, Garret & Zurif, 1980; Friederici & Saddy, 1991), there is no consensus with respect to the parallel between the two modalities (Grodzinsky, 1990). Moreover, even if we were to assert a parallel between comprehension and production deficits, independent studies of production are still required. Fourth, much of the focus has remained on omission errors. Substitution errors are just as striking. Grodzinsky (1984) argues against characterizing agrammatism as an omission deficit. He shows that while free functional morphemes are omitted, substitution errors are found with bound functional morphemes. Grodzinsky makes the first important steps in the direction of noticing and explaining substitution errors.  However, several problems arise. Grodzinsky's  claim implies that omission and substitution only involve functional categories. I argue that this is not the case: lexical categories are also affected by omission and substitution.  As well,  Grodzinsky's claim implies that free morphemes do not undergo substitution. This too I show to be false: both free and bound morphemes undergo substitution. Lastly, where substitution errors do exist, Grodzinsky claims that the shift is to a base form (a stem containing no affixes) or the result of guessing (which, presumably is random). I argue that neither of these approaches is in evidence: substitution patterns are not random and do not always select the base form, even when one is available. Moreover, Grodzinsky's explanation assumes that all categories that undergo substitution do so to the same extent. In other words, the assumption is that different categories suffer the same rate of substitution. Once again, I show this to be false. Lastly, research on agrammatism focuses on omission and substitution errors, as aooposed to misarticulations and rate of speech. These errors have been explained in terms of Tesak & Hummer (1994) argue against Grodzinsky, finding that there is no difference between governed and ungoverned prepositions. Bebout (1993) discusses aphasic treatment of morphological and syntactic negation, with morphological un- being more intact. She also discusses the complexities involved with adverbs, prepositions and derivational affixes. 3  24  phonology (Kean, 1977, 1979, 1980), morphology (Jarema & Kehayia, 1992), semantics (Caramazza&Hillis, 1989) and syntax (Grodzinsky, 1984a, 1984b, 1990; Caplan, 1987, 1992). However, the two following questions have rarely been asked: what exactly is omission? and what exactly is substitution?  In other words, what exactly are the structures that underlie  agrammatic utterances? When omissions are in evidence, what is 'missing'? The sounds? The syntactic word? The syntactic structure? When substitutions are produced, what features are altered? Segmental features? Syntactic features? Semantic features? In the following chapters I address these five issues.  The focus of this thesis is  production. The bulk of my analysis centers on the structures that underlie the agrammatic utterance, concluding that the structures themselves are produced as in normal speakers.  I  investigate the omission and substitution patterns of both lexical and functional categories. Specifically, I examine the behavior of the following categories: noun, verb, determiner, pronoun, clitic, auxiliary and modal. I conclude that the agrammatic deficit lies outside the phonological, syntactic and semantic components proper. Instead, omissions and substitutions result from an impairment to the interface mechanisms between the Lexicon, the Syntax and the Phonology.  25  Chapter 3 A G R A M M A T I S M : OMISSION ERRORS  Agrammatic production has traditionally been characterized as involving errors of omission and errors of substitution. This chapter addresses errors of omission. The utterances in (1) illustrate omission errors in agrammatic speech. Examples are from English (la), Dutch (lb), German (lc), French (Id), and Italian(le). Omissions are indicated by square brackets'[]'. Items which are omitted are underlined in the translation. (1) Agrammatic clauses with omissions a. the wife was showing [DP] the clock "the wife was showing him the clock"  English  b. zondagochtend [Aux] [DP] gewerkt Sunday-morning [Aux] [DP] work: Part "Sunday morning I had worked"  Dutch  c. spater begutachtet [DP] seinen Kukuruz later examines [DP] his corn "later he examines his corn"  German  d. et peu apres [DP] contemple [Det] recolte and little after [DP] gazes at [Det] harvest "and a little later he considers the harvest"  French  e. / (gli) uccelli non mangiano [DP] the:M-Pl (the:M-Pl) birds not eat:3Pl [DP] "the birds don't eat the seeds"  Italian  There are two basic issues which arise when investigating errors of omission: the nature of the omissions and their distribution. I address the nature of omissions first. What is omission? In other words, when an utterance includes an omission, what exactly is missing? The sound pattern? The syntactic category?  The syntactic projections associated with that  category? If structures are projected when agrammatics produce utterances with omissions, which ones are projected and what principles govern these structures? To determine what omissions are we must establish (i) what principles govern the words that are produced and (ii) the syntactic structures that underlie agrammatic utterances with omi ssions. 26  I begin by addressing the role of U G in agrammatic production.  I show that the  agrammatics under study produce utterances that are indistinguishable from normal production. Normal intact utterances are by definition governed by the principles of Universal Grammar (UG). I argue that intact agrammatic utterances are governed by the same U G principles found in the non-pathological population. Moreover, I maintain that U G principles govern agrammatic utterances that show omissions. Specifically, I argue that agrammatics do not produce clauses that violate the principles of U G . There are two possible interpretations of this claim. First, the agrammatic grammar and the normal grammar may be identical. Thus, the grammar of an English-speaking agrammatic would be the same as the grammar of a non-pathological English speaker. The second possibility is that the agrammatic speaker has a grammar different from the non-pathological speaker, but one that is nevertheless governed by U G . For example, an English-speaking agrammatic may now possess the grammar of say, Italian. I will defend the position that agrammatics retain the grammar of their own native language. I will then address the issue of what syntactic structures underlie agrammatic utterances that contain omissions. The logical possibilities are as follows: a complete absence of syntactic structures, a minimal set of (lexical) projections, a more substantial but incomplete set of projections, the full array of syntactic projections. I discuss each in turn, beginning with the most spartan hypothesis, that there is no syntactic structure. If there is a complete absence of structure, a possible result is chaos. If agrammatic utterances are chaotic, we expect random ordering of constituents. The fact that we get basic word order argues against a chaotic approach to agrammatic production, To derive basic word order minimally we must have a string of lexical categories (N, V , etc.), with no phrasal projections. Mechanisms independent of grammatical principles may govern agrammatic utterances, thus producing basic word order. The second possibility is that there is a minimal amount of phrasal structure. Minimally we must have the lexical projections NP, V P and S. To account for early language acquisition, Radford (1990) proposes the Lexical Projection Hypothesis, whereby only lexical projections are available. I test this hypothesis on agrammatic production. The Lexical Projection Hypothesis  27  derives subject-predicate order as well as verb-object order. However, it cannot account for the presence of functional projections nor can it derive any word orders requiring movement to functional projections.  I argue that both of these are in evidence in agrammatic speech.  Functional projections must be present. The Lexical Projection Hypothesis must be discarded for agrammatic production. The third possibility is that a fairly substantial but incomplete array of syntactic structures are projected in agrammatic clauses. Ultimately I argue that agrammatic clauses with omissions require the presence of at least two functional projections above the verb phrase (VP).  Once  again I draw from the acquisition literature. Around the age of 2, children frequently produce main clauses with infinitival verbs (henceforth root infinitives), where this option is disallowed in the target languages. This is often referred to as the optional infinitive stage. There are two proposals that I consider. Rizzi's (1994a,b) Truncation Hypothesis and Poeppel & Wexler's (1993) Optional Tense Hypothesis (also see Wexler, 1993, 1994). The Truncation Hypothesis states that all functional projections above and including TP are absent from root infinitives. The result is that only one functional projection is present in these structures. Hypothesis states that only T P is absent from root infinitives.  The Optional Tense  I test both proposals using  agrammatic production and show that neither can account for the data. Although the agrammatic data supports the existence of only two functional projections, I argue for the fourth and last possible set of syntactic structures: the full array of projections. My argument is based on the fact that agrammatic clauses display the full assortment of syntactic movement. If only two functional projections are present (assuming a model that requires more than two projections), we expect a restricted pattern of movement. This does not appear to be the case. I therefore conclude that agrammatic clauses, regardless of how many omissions, contain the full set of syntactic projections. Having argued that the agrammatic clause includes the complete array of syntactic projections, I discuss the rate and distribution of omissions. Although any syntactic category may be omitted, not all categories are omitted with the same frequency. I show that the severity of the  28  deficit affects rates of omission. The more severe the deficit (as indicated by Mean Length of Utterance (MLU) and Mean Rate of Production (MRP) in words/minute), the higher the omission rates. Furthermore, I show that while absolute rates of omissions do vary from agrammatic to agrammatic, the relative rates of omissions are constant across agrammatics and across languages. The hierarchy of retention is as follows: N > V > D > Aux/Modal, where ">" means "better retained than". Moreover, two categorial splits are apparent.  Lexical categories are  produced more frequently than functional categories. Nominal categories are produced more frequently than verbal categories. I end this chapter with a summary of the findings and return to the explanantion of omission errors to chapter 5.  3.1  T H E R O L E A N D SIGNIFICANCE OF U N I V E R S A L G R A M M A R  The single most important question to ask about agrammatic production is the following: What principles govern agrammatic utterances? A follow-up question is: What is the role of Universal Grammar? First, let me define the various domains that this discussion assumes. Universal Grammar (UG) governs the languages of non-pathological speakers. Each language is an instantiation of U G , which may result in a different language-specific grammar. The nonpathological speaker of any given language (or dialect) works with the Normal Grammar of that language (or dialect). I refer to this Normal Grammar as N G .  N G is licensed by U G since it is  one of the possible languages that is governed by U G principles. Assume the language of agrammatic speakers is governed by some set of principles (even if that set is simply the principle that there are no principles). I refer to this set as the Agrammatic Grammar, or A G . We can now discuss the relationship between A G and U G . There are four logical possibilities: U G and A G are completely separate; U G and A G intersect; A G is governed by U G , but is distinct from N G ; A G is constrained by U G only indirectly, via N G . The firjst possibility is that A G and U G are independent of each other. The second is that some aspects of A G are constrained by U G but  29  others are not. The third possibility is that A G is directly constrained by U G , and is a grammar distinct from N G . The last possibility is that U G constrains A G via NG; in other words, A G is not reanalyzed as a grammar other than the original N G .  3.1.1 Universal Grammar and Agrammatic Grammar Are Not Independent  The first possibility is that A G and U G are independent of each other.  If the grammar  governing agrammatic production is completely removed from U G , there is no reason to expect agrammatic utterances to be similar to normal utterances. However, in addition to utterances with omissions, agrammatic speakers produce utterances with no omission or substitution errors. The data in (2) are examples for English (2a), Dutch (2b), German (2c), French (2d) and Italian 2(e). The examples in (1) and (2) are from the same speakers: E F for English, D B for Dutch, G B for German, FC for French and IV for Italian. Each of the other agrammatic speakers (EE, D H , G M , F A and IR) can produce utterances with and without errors as well. For other examples of agrammatic clauses without errors see Appendix (1-2). (2) Agrammatic Clauses without Errors a. the dog steals the chicken  English  (EE)  b. de oude vrouw is ook woedend the old woman is also furious "the old woman is also furious"  Dutch  (DH)  c. der Wolf war tot the wold was dead "the wolf was dead"  German  (GM)  d. j ' a i subi une attaque la nuit I have suffenPart an attack the night "I had an attack at night"  French  (FA)  e. e' un po' difficile spiegare la situazione be:Prs-3Sg a bit difficult explain:Inf the situation "it's a bit difficult to explain the situation"  Italian  (IV)  30  The data in (2) are identical to clauses produced by normal speakers. Recall that normal clauses are by definition governed by U G (either via N G or not). That means one of two things. First, if A G is independent of U G , A G must be able to produce strings indentical to those produced by U G by means of principles that differ from those of U G . It is conceivable that these strings are produced using a set of semantic heuristics completely independent of U G .  Such is  the contention of Safran et al. (1980). They argue that constituent ordering is dependent on factors like animacy and potency. Safran et al. test the production abilities of five Englishspeaking agrammatic subjects. The agrammatics performed picture descriptions as well as sentence ordering tasks, which required them to piece together written sentence fragments. For the first experiment, the subjects were asked to describe pictures that depicted action relations (3) and locative relations (4). Examples of the target sentences are presented below. (3) Some Target Sentences for Experiment 1 in Safran et al. (1980): Action Relations a. The girl runs to the man. b. The man washes the baby. c. The girl runs to the house. d. The man washes the car. (4) Some Target Sentences for Experiment 1 in Safran et al. (1980): Locative Relations a. The pencil is in the sink. b. The ring is in the cup. c. The bird is in the sink. d. The mouse is in the cup. The (a-b) sentences have like animacy for both arguments. For (3ab) both arguments are animate. For (4ab) both arguments are inanimate. The (c-d) sentences have opposing animacy for the two arguments in each of the sentences. Safran et al. found that agrammatic "[w]ord order was incorrect on more than one third of the trials in like-animacy conditions [...] In contrast, few errors occurred in the A/I [animate/inanimate] conditions" (1980:268)'. This effect was also It should be noted that Safran et al. (1980) scored any deviation from canonical word order as errors. Thus the string "the wagon is pull the boy" would be scored as a word order error rather than as a correct passive sentence (cf. the wagon is pulled by the boy). Since correct non1  31  found in the sentence ordering task. They conclude that "agrammatic speech is generated without underlying structures that represent logical relations" (1980:278). Instead, word ordering reflects factors like animacy . 2  There are three problems with a semantic approach. First, semantic heuristics are general and apply across languages. If semantic heuritics govern word order, we expect agrammatics to use the same word order strategies across different languages.  However, language-specific  canonical word order is preserved in sentence production cross-linguistically (Pick, 1913; Smith & Mimica, 1984; Bates et al., 1988; Nespoulous etal., 1988; Ulatowska, 1988; Chen, 1989, 1993; Niemi etal., 1990; MacWhinney & Osman-Sagi, 1991; MacWhinney et al., 1991; Slobin, 1991) . 3  Since canonical word order varies from language to language, the set of semantic  heuristics would have to vary accordingly, incidentally matching the orders derived by U G / N G . Second, if word order is semantically driven, then it should be fairly rigidly constrained. In other words, for any given language, an agrammatic is expected to produce minimal variation in word order, if any. However, agrammatics use the same variety of complex relations and structures in a clause as normals do, though at a much lower rate (Armstrong, 1992). Moreover, when comparing agrammatic and normal production, Bates et al. (1988) find that English-, French- and Italian-speaking agrammatics produce both canonical and non-canonical word orders. Although there is a tendency for agrammatics to overuse the canonical word order, they do produce correct non-canonical word order in appropriate discourse situations. Such noncanonical word orders might be semantically driven; if so, however, the semantic heuristics would have to derive all the orders that incidenatlly match those derived by the principles of U G / N G . Further, variation in word order is more restricted for English-speaking agrammatics, where word order is more rigid, than for Italian-speaking agrammatics, where a freer word order canonical word orders are part of the inventory of agrammatic production, they cannot be simply dismissed as errors. Schwartz et al. (1980) report similar findings with a picture-pointing comprehension test. Word order is also preserved in comprehension. Studies on grammaticality judgments (Wulfeck etal., 1991) have shown that English speaking agrammatics are sensitive to word order. These findings reveal that agrammatics retain the ability to construct representations. As well, Japanese 2  3  32  is the norm. This reveals that agrammatics must have access to language-specific syntactic constraints governing word order. Bates etal. (1988) conclude that word order in agrammatism is intact. Third, if A G is independent from U G , it is completely coincidental that agrammatic strings have the same ordering as the strings produced by non-pathological speakers. For the production experiment in Safran et al. (1980), the agrammatics all produce an N - V - N order (when they manage to produce all three components). This is exactly what U G / N G would derive. I show below that constituent order in agrammatic utterances is identical to that of nonpathological speakers. There are two basic types of constituent ordering: major constituent ordering (subject-predicate), and ordering within a constituent (determiner-noun, adjective-noun, verb-objedct, auxiliary-verb, adverb-verb). I consider each in turn below.  3.1.1.1 Maj or Constituent Order  Major constituent order is preserved . Subjects are appropriately ordered with respect to 4  their predicates, and objects are appropriately ordered with respect to their verbs (Goodglass et al., 1972). The word orders for each of the five languages are not identical. I summarize the canonical word orders below (S=subject; Aux=auxiliary; V=verb; Adj=adjective; 0=object), with predicates placed in square brackets "[ ]". (5)  a. English:  S - [ (Aux) - V - O ]  b. Dutch:  S - [ Aux - O - V ] S-[V-0]  c. German:  S - [ Aux - O - V ] S-[V-0]  aphasics rely on (language-specific) canonical word order to interpret sentences (Hagiwara & Caplan, 1990). Phrasal integrity is also observed in comprehension tasks. Blumstein et al. (1991) show that while agrammatics do not show facilitory effects to word pairs forming constituents, they do show inhibitory effects with word pairs that cross syntactic boundaries.  4  33  d. French:  S - [ (Aux) - V - O ]  c. Italian:  [ (Aux) V - O ] - S  In (6), subjects are appropriately ordered with respect to their verbs. The last line of each example indicates the word order . For more examples of S-V order see Appendix (3-4). 5  The English and Italian data each display a single apparent counter-example to this expected word order. I begin with the English example, in (i). (i) open the window "the window is open" In (i) the string is interpreted in Menn (1990) as "the window is open". If we accept this intepretation, it appears as though the verb open has raised to T but that the subject the window has not raised to [Spec TP]. However, there is another possible interpretation of this utterance: "she opened the window". Here, an action is described, with the resulting omission of the subject pronoun and the past tense marker. Pronoun and tense substitution are unambiguously attested elsewhere, which lends support to such an analysis. Moreover, the narrative which contains this utterance is a sequence of action descriptions. The sequence is presented below in (ii), with the 'translation' in (iii). a. the girl asked the boy [P] [Det] cookie [P] [Det] jar b. [DP] leans [P] c. [DP] trips on the floor a'. [Det] mother washes the dishes b'. she absently pours water c'. [DPI open the window a". [Pleo] [Cop] [Det] nice day b". [Pleo] [Cop] a view shrubs, trees, grass a. the girl asked the boy for a cookie from the jar b. the stool leans over c. it trips on the floor a'. the mother washes the dishes b'. she absently lets water pour (over the sink) c'. she opened the window a". it's a nice day b". there's a view of shrubs, trees, and grass This sequence of statements is made up of three cohesive passages (the first sequence begins with (iia), the second with (ha'), the third with (iia")). The first makes reference to the activity that the boy and girl are involved in. The second describes the mother's actions. The third shifts to a description of the view outside the window, and thus outside the main picture. The most likely interpretation is "she opened the window", where a referential pronoun is omitted. Consequently, there is no word order violation. Rather, we have omission of the subject pronoun and a tense shift. Consider now the Italian example in (iv) (iv) il paniere prende con dentro, qua, con dentro tante cose the basket takes with inside, here, with inside many things "the basket has many things inside" The order is interpreted to be prende il paniere con dentro tante cose, with the verb in initial posision. However, the sentence in (ii) simply displays a subject-shift construction. This is achieved by left dislocation. This process is usually accompanied by a resumptive clitic, which is absent in the agrammatic utterance. Whether or not the resumptive clitic is required, no word 5  34  (6) Agrammatic Clause Order: Subject-Predicate Order a. [Det] nurse shaved me "the nurse shaved me" S-V-0  English  b. [Det] agent pakt de dief . [Det] policeman catch:3Sg the thief "a/the policeman catches the thief S -V-0  Dutch  c. ein Einbrecher nutzt die [N] a thief uses the [N] "a thief uses the opportunity" S - V - [O]  r  German  d. ma femme [Aux] appeleVer un ami et les pompiers my wife [Aux] call:Part/Inf a friend and the firemen "my wife called a friend and the firemen" S-V-0  French  e. [CL] [Aux] mangiato il cane [CL] [Aux] eat:PastPart-M-Sg the dog "the dog has/had eaten it" [CI] - [Aux] - V - S  Italian  For each of the examples in (6), the word order is appropriate: subject-predicate (S-V) for English, French, Dutch and German; and predicate-subject (V-S) for Italian. In (7), subjects appropriately precede their predicate adjectives . The required order for each of the languages is: 6  S-Cop-Adj (Cop = copula; Adj = predicate adjective). The German example includes negation (Neg), which is appropriately ordered but which can be ignored for our purposes (since we are concerned with subject-predicate ordering). For more examples of subject-predicate adjective order, see Appendix (5).  order violation has been produced. If the clitic is not required, the utterances is fine as is. If the clitic is required, then the agrammatic speaker has omitted the clitic, but the order remains licit. This example cannot constitute a word order violation. Italian displays an apparent counter-example to this word order. Consider the following, (i) teso molto teso mi sentivo io tense very tense refl felt: lSg I "I felt tense, very tense" The required order is apparently io mi sentivo teso molto teso. However, the order in (i) is actually grammatical since word order in Italian is fairly liberal. The sentence in (i) displays fronting of the complement of sentivo "felt" (i.e. teso molto teso) and right dislocation of the subject io. 6  35  (7) Agrammatic Clause Order: Subject Precedes Predicate Adjective a. [Det] chicken [Cop] gone "the chicken is gone" S - [Cop] - Adj  English  b. [Det] kippebout [Cop] weg [Det] drumstick [Cop] gone "the drumstick is gone" S - [Cop] - Adj  Dutch  c. die Abgase [Cop] nicht desund the exhaust-fumes [Cop] not healthy "the exhaust fumes were not healthy" S - [Cop] - Neg - Adj  German  d. les the "my S-  French  (mes) souvenirs [Cop] flous (my) memories [Cop] vague memories were vague" [Cop] - Adj  e. la nonna e Cappuccetto Rosso [Cop] salve the grandmother and L R R H [Cop] safe:F-Pl "the grandmother and L R R H were safe" S - [Cop] - Adj  Italian  In (8), subjects correctly precede predicate nominals. The required order for each of the languages is: S-Cop-NP (NP=noun phrase). (8) Agrammatic Clause Order: Subject Precedes Predicate Nominal a. R R H [Cop] [Det] pretty girl " L R R H is a pretty girl" S - [Cop] - N P  English  b. ik [Cop] automonteur I [Cop] motor-mechanic "I was an automechanic" S - [Cop] - N P  Dutch  c. der (das) erste Wort [Cop] Scheisse the:Nom-M-Sg (the:N) first word:N [Cop] shit "the first word was shit" S - [Cop] - NP  German  d. c'etaitune petite fille it be:Impf-3Sg a little girl "it was a little girl" S - Cop - NP  French  36  e. io [Cop] assessore pure I [Cop] member too "I am a member too" S - [Cop] - N P  Italian  Agrammatics also correctly produce non-canonical word orders.  First, agrammatic  production includes Wh-questions, as illustrated below. (9) Agrammatic Clause Order: Wh-word in Initial Position a. no examples in the database  English  b. hoe heet dat how call: lSg that "how do you call that?" Wh-V-0  Dutch  c. warum hast du [Adv] [Adj] Ohren why have:Prs-2Sg you [Adv] [Adj] ears "why do you have such big ears?" Wh - V - S - [Adv] - O  German  d. ou [DP] [Modal] voir [Det] cardiologue where [DP] [Modal] see:Inf [Det] cardiologist "where I would see a cardiologist" Wh - [S] - [Mod] - V - 0  French  e. dove [Cop] lafoccacia where [Cop] the cake "where is the cake?" Wh - [Cop] - S  Italian  7  For each sentence in (9), the Wh-phrase is in the expected initial position. Moreover, for Dutch and German, the verb is in the required second position. The next set of non-canonical word orders are specific to Dutch and German. Both languages allow non-subjects in initial position, with tensed verbs in second position.  This order is illustrated with the following  examples. For more examples of non-subjects in initial position, see Appendix (6).  7  This is an example of an embedded question. 37  (10) Agrammatic Dutch and German: Non-Subjects in Initial Position a. [Det] kippebout [Aux] [Det] hond gepakt [Det] drumstick [Aux] [Det] dog take:Part "the dog takes the drumstick" O - [Aux] - S - V  Dutch  b. spater begutachtet [DP] seinen Kukuruz later examines [DP] his corn "later he examines his corn" Adv - V - [S] - O  German  The Dutch sentence has the object in initial position. The German sentence has an adverb in initial position. Both examples are licit structures. Dutch and German also display an alternate word order in embedded clauses: tensed verbs and tensed auxiliaries must be in final position . 8  This order is also attested in agrammatic production. For more examples of embedded clauses, see Appendix (7). (11) Agrammatic Dutch and German: Embedded Clauses with Final Tensed Verb/Auxiliary a. no examples in the database  Dutch  b. damit ich besser horen kann so that I better heanlnf can:Prs-3Sg "so that I can hear better" Comp - S - Mod - V - Aux  German  French and Italian also display word orders that differ from those presented in (5). Both languages make use of object clitics (CI). The required word order is: S-Cl-V. This contrasts with the canonical order where the object follows the verb. The clitic word order is attested in German displays a single apparent counter-example to correct Vtns placement, in (i). (i) der Wolf fragt sie ah wo geht sie hin the wolf asks her ah where goes she to "the wolf asks her where she's going" The apparent correct word order is der Wolf fragt sie ah wo sie hin-geht. The sequence wo geht sie hin in (i) is interpreted as an indirect question. The word order of indirect questions is like that of embedded clauses, with Vtns in final position. But Vtns in the sequence is in second position. However, this utterance can be analyzed as a direct question, with the formal Sie : (ii) der Wolf fragt sie ah: "Wo geht Sie hin" the wolf asks her ah: "Where goes you to" "the wolf asks her: "Where are you going?"" With the sequence interpreted as a direct question, there is no word order violation. 8  38  agrammatic speech, as illustrated below. For more examples of agrammatic clauses with clitics, see Appendix (8). (12) Agrammatic French and Italian: Clitics a. elle h (le) reveille she her (him) wakes "she wakes him up" S - CI - V  French  b. [Det] ragazzo li prende [Det] boy them gets "the boy gets them" S - CI - V  Italian  As stated above, the canonical word order for Italian is V-O-S. However, S-V-O is also fairly common. The latter is attested in agrammatic speech as well. For more examples of SVO order in Italian agrammatic clauses, see Appendix (9). (13) Agrammatic Italian: SVO Order io ho visto [Det] sala da operazione I have:lSg see:PastPart [Det] room of operation "I saw the operating room" S-V-0  Italian  Thus both canonical and non-canonical word orders are preserved in agrammatic speech. These orders are identical to those produced by normals, whose production is governed U G . If the principles governing agrammatic utterances (AG) are independent of U G , they must mimic every structure produced by U G .  3.1.1.2  Word Order Within Constituents  Word order internal to constituents is preserved. Within major phrasal constituents such as DP and V P , the syntactic categories that are produced are appropriately ordered with respect to each other. Within nominal constituents, determiners are appropriately ordered with respect to  39  nouns (Friederici & Saddy, 1991) , as are adjectives (Sanchez, 1992b; Grodzinsky, 1984ab). 9  Within verbal constituents, objects are appropriately ordered with respect to verbs (Goodglass et al., 1972) and auxiliaries are appropriately ordered with respect to verbs. I begin with DP. The DP-internal order is {Det N} for all languages. This is the order that is produced, as illustrated in (14). In each case, the determiner precedes the noun that it specifies. For the sentences that follow in this section I underline the category under discussion (here it is the determiner) and place the relevant categories within curly brackets "{}". The last line of each example displays the categories contained within the curly brackets. (14) Agrammatic Nominal Constituents: {Det N} - {the girl} asked {the boy} [P] [Det] cookie [P] [Det] jar "the girl asked the boy for a cookie from the jar" {DetN} ... {DetN}  English  b. en dan [Det] agent pakt {de dief} and then [Det] policeman catch:3Sg the thief "and then a/the policeman catches the thief {DetN}  Dutch  c. {die Frau} weckt [DP] wieder the woman wakes [DP] again "the woman wakes him again" {DetN}  German  d. {les enfants} cherchent {les gateaux} the children look:Prs-3Pl the cakes "the children are looking for the cakes" {DetN} ... {DetN}  French  e. {il ragazzo} prende [Det] biscotti the boy gets [Det] cookies "the boy is getting some cookies" {DetN}  Italian  a  Adjectives and nouns are appropriately ordered with respect to each other . The required 10  order for adjectives and the nouns they modify are as follows: {Adj N} for English, Dutch, and Further, agrammatics use determiners to identify the specific category of nominal elements (Grossman etal., 1986); Kolk (1978) studies agrammatic comprehension of adjectives and concludes that they "fail to integrate modifying adjectives in sentences like Old sailors tell sad stories". However, this does not contradict the claim about production that agrammatics retain appropriate Adj-N word order. 9  10  40  German; {Adj N} or {N Adj} for French (depending on the subclass of adjective); {N Adj} for Italian. These are the orders that are produced, as illustrated in (15) . For every sentence I have 11  underlined the adjective and placed the relevant categories Within curly brackets "{}". Notice that for French both orders are attested, with the correct subclasses of adjectives. For more examples of Adjectives modifying nouns in agrammatic clauses, see Appendix (10). (15) Agrammatic Nominal Constituents: Adjective and Noun a. my {right side} was limp {Adj N}  English  b. ja de {oude vrouw} is ook woedend yes the old woman is also furious "the old woman is also furious" {Adj N}  Dutch  c. das Paar steht voll vor [Det] {leeren Korb} the couple stands full in-front-of [Det] empty basket "the couple stands in front of the/ an empty basket" {Adj N}  German  d. [Det] {grands yeux}, les {yeux percants} [Det] big eyes, the eyes piercing "big eyes, piercing eyes" {Adj N} ... {NAdj}  French  e. solanto r{occhio sinistra} [CL] [Aux] aperto only the eye left [CL] [Aux] open:PastPart "only my left eye was torn open" {NAdj}  Italian  Thus the order within nominal constituents is preserved in agrammatic production. I now turn to V P . Objects are appropriately ordered with respect to their verbs , as shown in (16). 12  French displays a single apparent counter-example with respect to correct adjective placement. The utterance in (ia) should have the order in (ib). (i) a. et Grand-mere comme vous les avez euh les grands (grandes) dents and Grandmother how you them have um the big:bare (big:F) teeth:F "and Grandmother, what big teeth you have" b. et Grand-mere comme vous les avez grandes les dents and Grandmother how you them have big:F the teeth:F There are two possible interpretations of this example. First, there is a hesitation ("euh") between the verb avez "have" and the determiner les "the". This may indicate an omitted adjective. Second, the first instance of les is interpreted as the clitic "them". It may be an anticipitory error: the anticipation of the determiner les "the". English displays an apparent word order problem with respect to direct and indirect objects. Consider the example in (i). 11  12  41  (16) Agrammatic V P Order: Object is Adjacent to Verb a. [DP] spread [Det] cloth, napkin (napkins), dish (dishes) "they spread a cloth, napkins, dishes" [S] - V - O  English  b. [DP] [Aux] [Det] gas uit-gedaan [DP] [Aux] [Det] gas off-turned:Part "she turned off the gas" [S] - [Aux] - 0 - V  Dutch  c. [DP] hatte einem (einen) Korb gepackt German [DP] have:Pst-3Sg a:Dat-M-Sg (a:Acc-M-Sg) basket pack:Part "she had packed a basket" [S] - Aux - O - V d. je [Aux] perdu la parole I [Aux] lose:Part the speech "I lost my speech" S - [Aux] - V - 0  French  e. il ragazzo prende [Det] biscotti the boy gets [Det] cookies "the boy is getting some/the cookies' S-V-0  Italian  Objects follow their verbs in English, French and Italian. For Dutch and German, objects precede the main verbs, but follow the auxiliaries. Thus canonical word order is preserved in agrammatic speech. Next, auxiliaries and modals are appropriately ordered with respect to verbs . For matrix 13  clauses the required order for each of the languages is {Aux/Mod V } . This order is attested for each language, as illustrated in (17).  I have underlined the auxiliary and placed the relevant  categories within curly brackets "{}". For each example except German, the auxiliary and the verb are neighboring categories. For German, the object intervenes, resulting in the following order: {Aux O V}. Although this is the expected order, what is relevant is that the auxiliary precedes the verb. (i)  the boy give (gives) to the girl a cookie "the boy gives a cookie to the girl" The appropriate word order is the boy gives a cookie to the girl. However, with focus N P shift, (i) is a legitimate structure (note: pauses are present preceding and following the girl).  42  (17) Agrammatic Verbal Constituents: {Aux Vpart} or {Mod Vinf} a. the wife {was showing} [DP] the clock "the wife was showing him the clock" {Aux Vpart}  English  b. [DP] {mogtlopen} [DP] must:Prs-lSg walk:Inf "I must walk" {Mod Vinf}  Dutch  c. [DP] {hatte einem (einen) Korb gepackt} [DP] have:Pst-3Sg a:Dat-M-Sg (a:Acc-M-Sg) basket pack:Part "she had packed a basket for grandmother" {Aux O Vpart}  German  d. je [CI] {suis occupe} a faire a manger I [CI] am occupy:PastPart to make:Inf to eat:Inf "I busied myself with getting smething to eat" {Aux Vpart}  French  e. Giovedi {sono andato} sotto [per] [V] la T A C Thursday am go:PastPart downstairs [to] [V] the CT-scan "Thursday I went down for a CT-scan" {Aux Vpart}  Italian  There is a different order for Dutch and German embedded clauses: {V Aux/Mod}. For more examples of embedded clauses in agrammatic production, see Appendix (7). ./  (18) Agrammatic Dutch and German: {V Aux} in Embedded Clause a. no examples in database  Dutch  b. weil sie (es) wa' [zu] {essen haben (hat)} German because they (it) something [to] eat:Inf have:3PUInf (have:3Sg) "because they have something to eat" {Vinf Aux} Further evidence for intact word order lies with the correct ordering of adverbs and negation with respect to verbs.  Adverbs follow tensed verbs, copulas and auxiliaries, but  precede untensed verbs (like participials Vpart and infinitivals Vinf) - I illustrate the {Vtns Adv} 14  1 recognize that auxiliaries and modals are not necessarily in V P . What is relevant here is that auxiliaries and modals are appropriately ordered with respect to main verbs. German and Dutch each display an apparent counter-example to correct word order. I begin with Dutch Vpart-Vinf structures, in (i). 13  14  43  order in (19) and the {Adv Vpart/inf} order in (20).  For more examples of adverbs in  agrammatic clauses, see Appendix (11). (19) Agrammatic Verbal Constituents: {Vtns Adv}, {Cop Adv} or {Aux Adv} a. it {'s just} foolish {Cop Adv}  English  b. aber die Oma {liegt noch immer} im Bett but the grandma lie:Prs-3Sg still in+the bed "but the grandma is still lying in bed" {Vtns Adv}  Dutch  c. und [DP] {vergisst ganz} and [DP] forget:Pres completely "and she completely forgets" {Vtns Adv}  German  d. j'{ai encore} etouffe et tout I have:Prs-lSg still suffocate:Part and all "I suffocated still and all" {Aux Adv}  French  (i)  a. inbreker binnen-stappen Dutch burglar inside-step:Inf "the burglar stepped inside" b. thuis-kommen Dutch home-come:Inf "they came home" These examples supposedly violate verb compounding. The appropriate word order are as follows. (ii) a. staptbihnen Dutch step:3Sg inside b. komen thuis Dutch come:3Pl home Compound verbs are composed of a particle and a verb stem. If the V is in the infinitive or past participial form, the particle is prefixed to the stem: Parti cle+Vinf. Otherwise, the particle follows the verb as an independent word: Vtns Particle. The Particle+Vinf is exactly what we have in (i). These examples are not instances of word order violations. Rather, they are instances of agrammatic untensed clauses. Now consider the German apparent counter example to correct adverb placement, in (iii). (iii) en eten vlug German and eat:Inf quickly "and (he) ate quickly" The correct word order is en vlug eten. Thus (iii) consists of a misplaced Adv. This interpretation is based on the assumption that eten is in fact the infinitival form. However, dialectal variation allows for eten to be interpreted as 3P1 as well as infinitival variation (pc from various native German speakers). If eten is 3P1, then the order in (133a) is the correct order. The tensed 3PI verb raises to C, past the Adv. This utterance is not counter-evidence to intact word order production. 44  Itahan  e. io {parlo assai} I talk:Prs-lSg much "I talk a lot" {Vtns Adv} (20) Agrammatic Verbal Constituents: {Adv Vpart} or {Adv Vinf} a. no examples in database  English  b. en [DP] [Modal] gauw eten and [DP] [Modal] quickly eatlnf "and he ate quickly " {Adv Vinf}  Dutch  c. no examples in database  German  d. [c'est] pour mieux t'entendre / ' (mon) enfant [it's] to better you hear the.Sg (my:Sg) child "Better to hear you with, my child" {Adv Vinf}  French  e. no examples in database  Italian  Lastly, I consider negation (Neg). For all but Italian, negation follows tensed verbs, auxiliaries and modals, but precedes untensed verbs (like participials Vpart and infinitivals Vinf). The orders are schematized as follows: {Vtns Neg}, {Aux Neg}, {Mod Neg}, {Neg Vpart} or {Neg Vinf}. Italian requires that negation precede the verb, whether tensed or untensed, with the resulting {Neg V } order.  These are the orders that are attested in agrammatic speech, as  illustrated below. For both German and Italian, other categories intervene between the verb and negation. However, these are appropriate orders. For more examples of negation in agrammatic clauses, see Appendix (12). (21) Agrammatic Verbal Constituents: {Vtns Neg}, {Aux Neg} or {Mod Neg} a. I {can't} speak {Mod Neg}  English  b. nou ik {weet met} hoor maarhet {is niet} goed well I know:Prs-lSg not you-know but is not right "well, I don't know but it's not right" {Vtns Neg} and {Aux Neg}  Dutch  45  c. das {weiss ich nicht} that know :Prs I not "that I do not know" {Vtns S Neg}  German  d. je {pouvais plus} me lever I could no-more myself get-up:Inf "I could no longer get up" {Vtns Neg}  French  e. io {non lo so} I not itknow:Prs-lSg "I don't understand" {Neg CI Vtns}  Italian  (22) Agrammatic Verbal Constituents: {Neg Vpart} or {Neg Vinf} a. the woman is washing the dishes English and didln't pay} attention/or (to) the sink "the woman is washing the dishes and didn't pay attention to the sink" {Neg Vinf} b. en nou twee maanden [Mod] [DP] {niet lopen} and well two months not walk-Inf "and after two months I still couldn't walk" {Neg Vinf}  Dutch  c. und die Mutter konnte {nicht zu [DP] begleiten} and the mother can:3Sg not to[DP] accompany:Part "and the mother can't take it to grandmother" {Neg IO Vpart}  German  d. elle {a pas oublie} she has not forget:Part "she hasn't forgotten" {Aux Neg Vpart}  French  e. {non s'e accorta} not self is aware:Part "she doesn't notice" {Neg CI Aux Vpart}  Italian  Thus the order within verbal constituents is preserved in agrammatic production. Agrammatic speakers can produce clauses that are intact. Moreover, all constituents (and individual words) that are produced are in the order appropriate to the speaker's language. If A G consists of a series of semantic rules independent of the principles of U G , these rules must be extremely sophisticated and must very suspiciously mimic all of the language-specific ordering contraints, which are governed directly by U G (or indirectly, via NG). 46  Not only must they  account for all the correct word orders, they must also be able to match up non-canonical word orders to appropriate discourse situations. Moreover, if purely semantic rules govern agrammatic production, it is coincidental that agrammatic utterances match consistently strings that are normally derived by syntactic projection and movement, governed by U G .  Such systematic  coincidence is unlikely. It must thus be the case that A G and U G share some principles.  3.1.2 Agrammatic Grammar is Constrained by Universal Grammar  The second possible relationship between A G and U G is that they intersect, sharing some principles but not others. The implication is that some of the agrammatic utterances are governed by U G while others are governed by principles independent of U G .  We then expect to find  agrammatic utterances that do not pattern like those governed by.UG.  However, I have argued  that all agrammatic word orders conform to normal word orders. Thus whatever is produced by an agrammatic falls within the confines of U G . That being the case, it seems odd to assume that agrammatic speakers, who have difficulty with language production, must control two language systems (UG and an independent set of semantic rules), while non-pathological speakers, who do not have these difficulties, simply use one grammar. Agrammatic speakers are suffering from reduced language ability yet they would be expected to control an added grammar with principles independent of the normal grammar. Also, it is unclear how the two systems would combine when agrammatic utterances start off as non-pathological but end up with missing elements.  Following Occams's razor, I  assume that the system underlying language production for both the agrammatic and the normal speaker is the same grammatical system. Both are governed by the principles of U G . In other words, A G is constrained by U G . Two alternatives remain. Either A G is identical to the normal grammar N G , or A G is different from N G .  The question to answer is the following: Are  agrammatics working with a grammar that is identical to that of their native language or not? A  47  first step is determining what and how much structure is projected in agrammatic clauses. This is the issue to which I turn in the following sections.  3.2  F U N C T I O N A L PROJECTIONS IN A G R A M M A T I C PRODUCTION  As illustrated in (23), most agrammatic utterances are syntactically impoverished in that not all the words that are required of a normal utterance are present. For each of the following sentences, at least one lexical item is missing. (23) Agrammatic Clauses with Omissions a. the wife was showing [DP] the clock "the wife was showing him the clock"  English  b. zondagochtend [Aux] [DP] gewerkt Sunday-morning [Aux] [DP] work: Part "Sunday morning I had worked"  Dutch  c. spater begutachtet [DP] seinen Kukuruz later examines [DP] his corn "later he examines his corn"  German  d. et peu apres [DP] contemple [Det] recolte and little after [DP] gazes at [Det] harvest "and a little later he considers the harvest"  French  e. i (gli) uccelli non mangiano [DP] the:M-Pl (the:M-Pl) birds not eat:3Pl [DP] "the birds don't eat the seeds"  Italian  The agrammatic data is quite similar to data from language acquisition, examples of which are presented below. Example (24a) is from Weverink (1990), (24b) from Wagner (1985), (24c) from Pierce (1989), and (24d) from Guasti (1992). (24) Acquisition clauses a. ikpaf't[DP]op Ipack:Psr-lSg [DP] pit "I pack it"  Dutch  48  b. Caesar tieg(t) e(r) [DP] nicht Caesar get:Psr-3Sg he not "Caesar did not get it"  German  c. [DP] est pas mort [DP] be:Prs-3Sg not dead "it/he is not dead"  French  d. vuole dormendo want:Psr-3Sg sleep:PrsPart "he wants to sleep"  Italian  Both agrammatic and acquisition data include strings that have 'missing' lexical items compared to the non-pathological adult. Researchers in both fields wish to determine what structures underlie these strings and whether they are the same structures as adults (for acquisition) and normals (for agrammatism). Despite the similarities, there are some differences between acquisition and agrammatism. With acquisition the system under study is one that is developing, whereas the agrammatic system is a fully developed adult system that has 'gone wrong'.  This distinction becomes  relevant when we consider the evidence that runs through the acquisition literature. One of the goals of acquisition research is to establish which projections are present in early language production. One type of evidence lies in establishing whether or not the child can distinguish between tensed and untensed verbs. The assumption made in the acquisition literature is that if children can distinguish between tensed and untensed clauses, U G will force the required movement(s). Thus, if children can distinguish between tensed and untensed verbs, headmovement to a functional projection (FP) is available in early acquisition (specifically V-to-I movement). That being the case, the relevant FPs must also be present . 15  However, I argued  above that agrammatics produce normal clauses which by definition include FPs. Thus, simply In addition to lexical categories projecting structures in the syntax (i.e. noun and verb projecting NP and V P , respectively), functional categories are argued to project syntactic structures as well. Thus verbal inflection projects an IP; complementizers project a CP; and determiners project a DP. There is much debate about how many functional projections are available to the syntax. Some have proposed splitting IP into an agreement phrase (AgrP) and a tense phrase (IP or TP, depending on one's theoretical bent) (Pollock, 1989; Chomsky, 1991; among others). And the numbers grow from there, with aspect phrases (AspP), agreement 1 5  49  the presence of a tensed V or Aux (in the correct position) is arguably evidence for an FP. We want to establish what FPs are present when there is no lexical evidence for them. In other words, if the clause in question is untensed, is there an IP, a C P or any other FP?  Since the  presence of a tensed verbal element (in the correct position) is itself evidence for an FP, the relevant data for our current purposes are restricted to clauses that do not contain items that display tense (present, past or future), examples of which I present below. I refer to these types of clauses as untensed clauses. Untensed verb forms therefore include participial and infinitival verbs.  For my present purposes I ignore the distinction between inflected and uninflected  untensed verb forms, where English bare verbs are an example of uninflected forms and English participials illustrate inflected forms. (25) Agrammatic Untensed Clauses a. the thief [Aux] arrested "the thief is arrested"  English  b. net boven [Aux] [DP] gehaald just upstairs [Aux] [DP] reach:Part "I had j ust reached upstairs"  Dutch  c. mir [Aux] auch passiert me:Dat[Aux] also happen:Part "that has also happened to me"  German  d. je [Aux] perdu la parole [P] trois minutes I [Aux] lose:Part the speech [P] three minutes "I lost my speech within three minutes"  French  e. [CI] [Aux] mangiato il cane [CI] [Aux] eat:Part-M-Sgthe dog "the dog has/had eaten it"  Italian  These agrammatic untensed clauses find their counterparts in early acquisition. Around the age of 2, children frequently produce main clauses with infinitival verbs (henceforth root infinitives), where this option is disallowed in the target languages. This is often referred to as the optional infinitive stage. Examples are shown in (26). Example (a) is from Klima & Bellugi phrases specific to subjects (AgrSP) and objects (AgrOP), and so on. As much as possible, I will use the neutral FP Afunctional projection) to refer to functional projections in general. 50  (1966) in Guasti (1992), (b) from Weverink (1990), (c) from Wagner (1985), (d) from Pierce (1989), and (e) from Guasti (1992). (26) Acquisition Clauses with Infinitival Verb a. he no bite you  English  b. pappa schoenen wassen daddy shoes wash:Inf  Dutch  c. Zahne pussen teeth brush:Inf  German  d. pas manger la poupee not eat:Inf the doll  French  e. lavare i piatti wash:Inf the dishes  Italian  Recent work in syntactic acquisition has focused on the grammatical principles and structures that govern the production of root infinitives . 16  The debate is about how much  structure is actually present in these acquisition data. A t one end of the debate, it is claimed that only lexical categories (and their projections) are available in child language production (Parodi, 1990; Radford, 1990; Ouhalla, 1991; Meisel, 1994). At the other end, it is claimed that the same projections exist in both adult and child production (Weissenborn, 1990; Boser, Lust, Santelmann & Whitman, 1992; Poeppel & Wexler, 1993, 1994). Similar issues are found in the study of agrammatic production data.  Agrammatic  production includes clauses with missing tensed verbs (or auxiliaries and modals). How much structure is present in these clauses? There are four possibilities. These are as follows. (27)  a. There is no structure. b. There is a minimal amount of structure. c. There is a substantial amount of structure, but not the full array of projections. d. The structure is complete.  For detailed discussions and analyses, please see the following references: Lebeau, 1988; Clahsen, 1990; Platzack, 1990; Radford, 1990; Weissenborn, 1990; Guilfoyle & Noonan, 1992; Poeppel & Wexler, 1993; Wexler, 1993; Rizzi, 1994ab; Ingram & Thompson, 1996; and many others. 16  51  As argued above, the first possibility can be easily dismissed. The rest of this section is spent determining how much structure is present in agrammatic untensed clauses.  To this end, I draw from the acquisition literature, focusing on three  proposals: The Lexical Projection Hypothesis, The Truncation Hypothesis, and the Optional Tense Hypothesis. Radford (1990) proposes that only lexical projections are present (also see Lebeaux, 1988). I label this the Lexical Projection Hypothesis.  Rizzi (1994a,b) offers the  Trucation Hypothesis. He proposes that children in the optional infinitive stage produce a truncated syntactic structure.  Specifically, children strip off the outer clausal layers (for his  structure that means CP, AgrSP and TP).  In other words, they retain everything below T P (for  our purposes, that means they retain a single FP). Poeppel & Wexler (1993) propose the Optional Tense Hypothesis (also see Wexler, 1994) whereby these same children retain everything but T P for untensed clauses (for our purposes, that means they are missing only one FP). Although there is a gradation of FPs from Radford through Rizzi to Wexler, all three propose that untensed clauses are incomplete structures, missing all to one FP.  I work through each proposal in turn,  comparing the early acquisition and the agrammatic data.  3.2.1  Agrammatic Clauses Are More than Lexical Projections  I now consider the Lexical Projection Hypothesis. To derive basic word order we must have minimally the lexical projections NP and V P . Radford (1990) argues that only lexical projections are available in early production (the Lexical Projection Hypothesis or LPH).  The  implication of such an approach is that early acquisition will neither contain functional projections nor display movement that requires the presence of functional projections (FP). Extending a Lexical Projection Hypothesis to agrammatic production would predict that agrammatics retain lexical projections (LPs), but not FPs. I argued above that the data required to test for the presence of FPs must display no lexical evidence of such projections. For verbal FPs, the data is reduced to untensed clauses. For nominal FPs, we must consider determiner  52  phrases (DPs) without determiners. It is difficult to show any evidence for nominal functional projections . 17  Consequently, the strength of my argument is determined by verbal functional  projections. I begin the discussion of FPs with the nominal functional projection DP.  3.2.1.1 Nominal Functional Projections  I now address the existence of nominal FPs, in particular DP. The constructions under scrutiny are those that are missing the head D. If these missing items are in fact absent from the syntax, projecting no syntactic structure, then they cannot enter into any relations, whether syntactic or semantic. I will discuss two missing Ds: articles and English possessor's.  3.2.1.1.1 Missing Articles In the languages under study, determiners encode the novel-familiar distinction. Indefinite articles generally introduce new discourse referents, while definite articles indicate familiar discourse referents.  If determiners are missing, we do not expect to retrieve this  distinction. However, the novel-familiar distinction is available even when determiners are missing. In each of the following sentences, the referent is novel, being introduced into the discourse. The missing determiner is thus interpreted as the indefinite article. I have underlined the relevant determiner.  Below each sentence I give the relevant context, showing that the  ref erent i s indeed novel. ) (28) Agrammatic clauses: Null Determiner with Novel Interpretation a. Red Riding Hood [Cop] [Det] pretty girl " L R R H was a pretty girl"  English  Context: This is the introductory sentence to the story of L R R H .  Evidence for nominal FPs relies on movement within the DP (Giorgi & Longobardi, 1991). However, such strings are not available in the agrammatic database under study. 17  53  b. [Pet] boer [Cop] [P] [Det] zaaien [Det] farmer [Cop] [P] [Det] sow " A farmer is sowing"  Dutch  Context: This is the introductory sentence to a picture description about a farmer planting, and harvesting corn. c. das Madchen hatte sie immer [Det] rote Kappe auf'm Kopf the girl have:Prs-3Sg she always [Det] red cap on the head "the girl alwasy has a red cap on her head"  German  Context: The story is L R R H . The subject has introduced L R R H and is explaining why her name is L Red RR. This is the first mention of the cap. d. y fouille les meubles he search:Prs-3Sg the furniture  French  et prend un (une) montre et [P] [Det] argent and take:Prs-3Sg cuM (a:F) watch and [P] [Det] money "he searches the furniture and takes a watch and some money" Context: The description is of a thief stealing. The subject has described the thief entering the apartment. This is the first mention of what the thief is actually taking. e. [Det] assessore sanitario dice: [Det] assessor sanitary say:Prs-3Sg:  Italian  "subito, subito, subito, [V] [Det] ambulanza!" "quick, quick, quick,[V] [Det] ambulance!" "The health inspector sasys: "Quick, get an ambulance!"" Context: The subject is describing the scene of his stroke. He was at a council meeting, had had a headache and then had fainted. This is the first description of the people around him calling for an ambulance. For each of the following sentences, the referent is familiar, having been introduced earlier in the discourse. The missing determiner is thus interpreted as the definite article. I have underlined the relevant determiner. Below each sentence I give the relevant discourse context, showing that the referent is indeed familiar. (29) Agrammatic Clauses: Null Determiner with Familiar Interpretation a. and [Det] wolf says: "Better smell" "and die wolf says: "The better to smell you with"" Context: The story is L R R H . familiar character.  English  The wolf has already been introduced and is now a  54  b. rDetl meisje [Mod] lachen  Dutch  [Det] girl [Mod] laugh:Inf "the girl must laugh"  Context: The description is of the Cookie Theft Picture. The girl is introduced two sentences earlier. c. no (clear cut) examples in database  German  d. peu apres [DP] contemple [Pet] recolte little after [DP] gaze:Prs-3Sg [Det] harvest "a while later he gazes at die harvest"  French  Context: The description is of a farmer planting and harvesting. The subject has described the fanner planting and the corn growing. The harvest is thus a familiar referent, even if the word itself has not been used before in the discourse. e. [Det] ragazzo li prende [Det] boy them take:Prs-3Sg "the boy is taking them (the cookies)"  Italian  Context: The description is of the Cookie Theft Picture. Both the boy and the cookies hae been introduced two sentences earlier. The subject is clarifying that it is the boy who is taking the cookies and the girl who is eating them. Since sentences exhibit the novel-familiar distinction, some (null) determiner must be present. Moreover, the behavior of null determiners is indentical to that of overt determiners. There is then no reason to assume that the Agrammatic Grammar is deviant with respect to determiners. These data argue for a DP projection with a phonologically null determiner in D . 18  Further evidence for the presence of D P are replacement sequences. Agrammatics occasionally repeat a failed sequence. The second attempt frequently displays words that were missing in the first attempt. Examples are presented below. I have underlined in the second attempt of the word that is missing from the original attempt. For more examples of replacement sequences for determiners, see Appendix (13).  Please note that I am not implying that null determiners in child language acquisition follow the pattern described here. I maintain that acquisition and agrammmatism are not mirror images of each other. 18  55  (30) Agrammatic Clauses: Replacement Sequences for Determiners a. the dog sniffs [Det] chicken "the dog sniffs the chicken"  English  a', the dog steals the chicken  English  b. [Det] boer [V] [P] [Det] vogelverschrikker [Det] farmer [V] [P] [Det] scarecrow "the farmer is looking at a scarecrow"  Dutch  b'. de boer [V] maithe farmer [V] cor"the farmer wants corn"  Dutch  c. [Det] Mutter [Cop] verliebt [Det] mother [Cop] in+love:Prt "the mother is in love"  German  c'. die Mutter i st verliebt the mother is in+love:Prt "the mother is in love"  German  d. un chien suit un couple pendant [Det] trajet a dog follow:Prs a couple during [Det] trip "a dog follows a couple during a trip"  French  d'. un trajet pour pi que-niquer a trip to picnic: Inf "a picnic trip"  French  e. [Det] marito sta [V] [Det] sveglia [Det] husband be:Prs-3Sg [V] [Det] alarm+clock "the husband putting the alarm clock"  Italian  e'. l'uomo sta posando noprendere (prendendo) la sveglia Italian the man be:Prs-3Sg putting no takeilnf (taking) the alarm clock "the man is putting the alarm clock" These repetition sequences are evidence that the determiner is present underlyingly. A DP must therefore be projected.  3.2.1.1.2  Missing Possessor  A second construction that requires the presence of D is possession.  There are two  different types of constructions to consider: possessive determiner (my, your, and the like) and the English possessor marker's. Following a substantial body of research (Abney, 1987; Ritter,  56  1988, 1989; Drijkoningen, 1990; Tang, 1990; Giorgi & Longobardi, 1991; Vergnaud & Zubizaretta, 1992; and many others) I assume a D P structure for arguments. The structures for possessives are not uncontroversial. What is consistently argued, however, is that possessives are nominal FPs of some sort. So even if they are not DPs, some nominal FP must be present. I adopt the DP strcuture for expository purposes (and to avoid confusion with the use of FP in later discussions). Consider the strings and structures presented below. (31)  a. the cookie b. her cookie c. Carmina's cookie  (32)  a.  DP  b.  DP  D' D the  c. D'  NP cookie  D her  DP  C a r m i n a ^ ^ D' NP cookie  D 's  NP cookie  The definite article the occupies the head D (32a), as does the possessive her (32b). In (32c), the possessor Carmina is in [Spec DP], with the possessor marker 's in D. If possessive Ds are absent from the syntax, we do not expect a possessive reading of the arguments with the missing determiners. However, examples of this sort exist. These are illustrated in (33-34). For more examples of null possessive determiners in agrammatic clauses, see Appendix (14). For more examples of null possessive morphemes, see Appendix (15). (33) Agrammatic Clauses: Null Determiner with Possessive Interpretation a. the thief lifts [Poss] body "the thief lifts his body"  English  b. no examples in database  Dutch  c. ihr Schpuck nein Schmuck und [Poss] Geld versteckt German theinN 'dewelry' no jewelry and money hide:Part "(the family has hidden) their jewelry and their money in the oven" 19  This possessive structure has the following replacement sequence. a. the thief lifts his INI "the thief lifts his body" a'. the thief lifts IDetl body "the thief lifts his body" 57  English  19  d. apres, j ' [Aux] attendu [Poss] sortie after, I [Aux] wait:Part [Poss] release "afterwards, I waited for my discharge"  French  e. no examples in database  Italian  (34) Agrammatic Clauses: Null Possessive Morpheme a. L R R H go (goes) to see grandma['s] house " L R R H goes to grandma's house"  English  b. the wolf is running to LRRH['s] granma "the wolf is running to L R R H ' s granma"  English  Like sentences with their overt counterparts, these sentences exhibit a possessive interpretation. There is then no reason to assume that the Agrammatic Grammar is deviant with respect to possession. Thus, some phonologically null possessive determiner in D (or at least some nominal FP) must be projected.  3.2.1.2  Verbal Functional Projections  For the domain that I am considering, all movement is to a verbal FP. If verbal FPs are not projected the only agrammatic word orders that will be produced are the underlying word orders i.e. those not requiring movement. I will refer to the structures underlying these orders as underlying structures (and where appropriate, D-Structure). The underlying structure of English and French is shown in (35), that for Dutch and German in (36), and that for Italian in (37) . 20  NoteihatI have adopted VP-internal subjects (Fukui & Speas, 1986; Kitagawa, 1986; Koopman & Sportiche, 1988; Kuroda, 1988). The other option is to base-generate subjects in specifier position of an FP, [Spec FP]. If subjects are base-generated in [Spec FP], the presence of a subject is itself evidence for an FP. Thus any early acquisition or agrammatic clause with a subject would have to project (at least) one F P .  For discussions of word order, see Haegeman (1992, 1994) for German and Dutch SOV. 58  (35) English and French Underlying Structure: SVO VP SUBJECT  V  V E R B ^ ^ ^ OBJECT (36) German and Dutch Underlying Structure: SOV VP SUBJECT  V  OBJECT  VERB  (37) Italian Underlying Structure: VOS VP V VERB  ,  <  /  S  X  SUBJECT  OBJECT  English and French are underlyingly SVO, Dutch and German are underlyingly SOV, and Italian is underlyingly V O S . This is summarised in the following list. (38) Underlying Word Orders a. b. c. d. e.  English Dutch German French Italian  SVO SOV SOV SVO VOS  If early acquisition and agrammatic untensed clauses involve no movement, we expect the following underlying orders: SVO for English and French, SOV for Dutch and German, and VOS for Italian, These orders are in fact attested for both early acquisition (39) and agrammatism (40), as the following examples illustrate ((39a) from Klima & Bellugi (1966) in Guasti, 1992; (39b) from Weverink, 1990; (39c) from Wagner, 1985; (39d) from Pierce, 1989; (39e) from Guasti, 1994).  59  (39) Early Acquisition Clauses: Underlying Word Order a. he [Aux] no bite you "he doesn't bite you" S - [Aux] - Neg - V - O b. pappa [Mod] [Det] schoen wassen daddy [Mod] [Det] shoes wash:Inf "Daddy is washing his shoes" S-[Mod]-0-V c. Thorsten das haben Thorsten this have:Inf "Thorsten has this" S-O-V d. [DP] [Aux] pas attrape/erune fleur [DP] [Aux] not catch:Part/Inf a flower "I/she hasn't caught a flower" [S] - [Aux] - Neg - V - O e. per lavalei piatti in-order-to wash:Inf the dishes "so that she can wash the dishes" Comp-V-0 (40) Agrammatic Clauses: Underlying Word Order a. he open the doors "he opens the doors" S-V-0 b. [Det] moeder [Modal] [Det] afwas doen [Det] mother [Modal] [Det] dishes do:Inf "the mother is doing the dishes" S - [Mod] - O - V c. und dann Garag(e) gearbeit[et] and then garage work:Part "and then I worked in a garage" Conj - Temp - S - [Mod] - O - V d. je [Aux] perdu la parole [P] trois minutes I [Aux] lose:Part the speech [P] three minutes "I lost my speech within three minutes" S - [Aux]-V-O-TimePP e. [CI] [Aux] mangiatoil cane [CI] [Aux] eat:PastPart-M-Sg the dog "the dog has/had eaten if' [CI] - [Aux] - V - S  60  The constituent order for English and French is S V O . The constituent order for Dutch 21  and German is SOV.  The constituent order for Italian is VOS. The specific orders illustrated  above for the acquisition clauses are the following: SVO for English; SOV for Dutch; SOV for German; V O for French; and V O for Italian. The specific orders illustrated above for the agrammatic clauses are the following: SVO for English; SOV for Dutch; OV for German; SVO for French; and SV for Italian. Each of these orders are predicted by L P H since none involves movement to an F P . However, these word orders are not the only ones produced in early acquisition and agrammatic utterances. The other word orders are derived via movement to FPs, and thus require the presence of those FPs. For the relevant constructions I turn to the distribution of negation. There has been some debate about the structure and position of Neg (Dechaine, 1995; Kayne, 1989; Laka, 1990; Ouhalla, 1991; Zanutinni, 1990). With respect to the identity of Neg, there are two main schools of thought. Either Neg is a head (Pollock, 1989; Rizzi, 1990) or it is an adjunct (Webelhuth & den Besten, 1987; Clahsen, 1988a) . If it is a head, it projects its own 22  phasal structure, as shown in (41a). If it is an adjunct, it adjoins to V P . The resulting structure is illustrated in (41b).  I have presented French examples where the verb is unambiguously in its participial form, The participial is not always morphologially distinct from the infinitival. The following examples illustrate this ambiguity. (i) Tenseless clauses: French participial or infinitival Verb a. ma femme appeleVer un ami et les pompiers my wife call:Part/Inf a friend and the firemen "my wife called a friend and the firemen to drive me to the clinic" b. la femme reveille/er l'hornme entre temps the woman wake:Part/Inf the man in the meantime "the woman woke the man up in the meantime" In (a) the phonological form of the verb is [aepdle]. This can be interpreted as the participial appele or the infinitival appeler. The same holds for the other two examples. Putting aside such examples, the point remains: the agrammatic word order is the same as the D-Structure word order. Bayer (1990a) argues that Neg left-adjoins to V in German. 2 2  61  (41)  a.  NegP  b.  N e g ^ ^  ^ ^ N e g ' Neg  VP VP  VP  Notice that if Neg is a functional head that projects its own phrasal structure, NegP is an FP. The Lexical Projection Hypothesis would then predict that Neg is not produced. The only way thatLPH can accommodate negation is to assume that Neg is a lexical category that adjoins to the V P structure, like the structure in (41b). I therefore assume the structure in (41b) (for this section). If we include subjects and objects, the following structures result. The structure for English and French is (42a), for Dutch and German is (42b), for Italian is (42c). (42)  a.  VP  b.  N e g ' ^ ^ VP SUBJECT  N e g ^ ^ V  VP  SUBJECT OBJECT  VERB c.  VP  V  OBJECT  VERB  VP Neg^^^VP V ' ^ ^  SUBJECT  VERB^^^OBJECT The structures in (42) are those allowed by L P H . Assuming that Neg (and Adv) are anchors that indicate which constituents have moved (Pollock, 1989), any orders deviating from those in (42) require movement . Thus the word orders are predicted to be the following: Neg23  S-V-0 for English and French; Neg-S-O-V for Dutch and German; and Neg-V-O-S for Italian. The Lexical Projection Hypothesis predicts that no other orders will be attested. Because the word orders in the following list require movement to an FP, they are predicted not to arise. I repeat in (43') the underlying word orders for the languages under study. 23  See Iatridou (1994) and Williams (1994) for arguments against this assumption. 62  (43) Word Orders Predicted to Be Unattested by the Lexical Projection Hypothesis a. b. c.  S - Neg V - Neg 0 - Neg  d. e. f. gh. i.  V- S s- V 0- V V- 0 0- s s -0  (for (for (for (for (for (for  SVO VOS SVO SOV SVO VOS  and SOV languages) languages) and VOS languages) languages) and SOV languages) languages)  (43') Underlying Word Orders a. b. c. d. e.  English Dutch German French Italian  SVO SOV SOV SVO VOS  Early acquisition and agrammatic clauses display many of the word orders listed in (43). However, not all appear in untensed clauses. This is the case for word orders that require the verb to move from its underlying position. A verb moves from its underlying position in order to check its tense features (among other features). Consequently, if a verb moves, it will be tensed. The orders that involve a tensed verb (Vtns) are the following. (44) Word orders that Require a Tensed Verb a. b. c.  V-S V-O V - Neg  (for SVO and SOV languages) (for SOV languages)  These orders are attested in both early acquisition and agrammatism. We find subjects in postverbal position for acquisition (45) and agrammatism (46) , as shown below ((45a) from 24  Pierce, 1992; (45b) from Weissenborn, 1990; (45c) from Pierce, 1989).  We also find one instance of the V-S order for English shaved me, nurse "she shaved me, the nurse" This is arguably a dislocated argument rather than the D-Structure subject, given the pause between me and nurse. 2 4  63  (45) Early acquisition: {V-S} a. came a man "a man came" V-S  English  b. kauft Angela buy:3Sg Angela "Angela buys" V-S  German  c. litmaman read:Prs-3Sg mommy "Mommy reads" V-S  French  (46) Agrammatic clauses: {V-S} a. horloge en andere spullen {pakt hij} clock and other things take:3Sg he "he takes a clock and other things" 0-{V-S}  Dutch  b. und wegen der Vogel steht Kraus (Vogelscheuche) and because of+the birds stands crow (scarecrow) "and because of the birds, there's a scarecrow" Conj - AdvP - {V - S}  German  The V-S order requires that V move from its underlying position. This movement is possible only if an FP is present to offer a landing site. This movement is schematized in the following diagrams, (47a) for SVO languages (for acquisition, in our case) and (47b) for SOV languages). (47)  a.  FP F' F  VP SUBJECT  V  V E R B ^ ^ OBJECT  64  FP F' ^  VP  SUBJECT  V  OBJECT  VERB  Next we find verbs preceding the object (in language that are otherwise O-V) in data from acquisition (48) and agrammatism (49) as shown below ((48a) from Weverink (1990); (48b) f  from Wagner (1985)) (48) Early acquisition clauses: {V-O} a. ik {pakt't} op I pack it up "I am packing it up" S - {V - 0} - Pit  Dutch  b. mein Ffubsaube {had Tiere} din my helicopter has animals in+it "my helicopter has animals in it" S - {V - 0 } - PP  German  (49) Agrammatic Dutch and German: {V- 0} a. en dan agent {pakt de dief} and then policeman catch:3Sg the thief "and then a/the policeman catches the thief Conj - Adv - S - {V - 0 }  Dutch  b. spater {begutachtet [DP] seinen Kukuruz} later examines his corn "later he examines his corn" Adv - {V - [S] - 0}  German  The V - 0 order requires that V move from its underlying position (in an O-V system). This movement is possible only if an FP is present to offer a landing site. This movement is schematized in the following diagram (this is relevant to SOV languages only).  65  (50)  FP F' F  VP SUBJECT  V  OBJECT  VERB  Last with respect to tensed verbs, we find verbs preceding Neg for acquisition (51) and agrammatism (52), as shown below ((51a) from Vainikka (1993); (51b) from Wagner (1985) in Wexler (1994); (51c) from Deprez & Pierce (1993)) . 25  (51) Early Acquisition: {Vtns Neg} or {Mod Neg} a. I {can't} wear it S - {Mod -.Neg} - V - O  English  b. Caesar {tieg e nich} Caesar gets he not "Caesar doesn't get it" Top - {Vtns - S - Neg}  German  c. [DP] {veux pas} lolo [DP] want not milk "I don't want milk" S - {Vtns - Neg} - O  French  (52) Agrammatic Clauses: {Vtns Neg}, {Aux Neg} or {Mod Neg} a. I •[can't} speak S - {Mod - Neg} - V  English  b. ik {weet niet} hoor maar het {is niet} goed I know:Prs-lSg not you-know but is not right "I don't know but it's not right" S - {Vtns - Neg} - "filler" - Conj - {Aux - Neg} - Adj  Dutch  Italian negation always precedes V . Therefore Italian does not provide any evidence for movement to an FP. The French data displays an interesting property. French negation consists of ne pas. In the agrammatic and acquisition data, the ne drops out. This is also true of certain dialects of French. One possible explanation is that ne is an unstressed morpheme that must cliticizeto the verb, and consequently drops out easily. 66  c. das {weiss ich nicht} that know:Prs I not "that I do not know" O-{Vtns-S-Neg}  German  d. je {pouvais plus} me lever I could no-more myself get-up:Inf "I could no longer get up" S - {Mod - Neg} - CI - V  French  To derive the V-Neg order, the V must move from its underlying position. This movement is possible only if an FP is present to offer a landing site.  This movement is  schematized in the following diagrams, (53a) for SVO language and (53b) for SOV languages. (53)  a.  OBJECT  FP F' VP Neg  / /  "  X x  VP  SUBJECT OBJECT^  V VERB  Recall however, that these 3 sets (V-S, V - 0 and V-Neg) involve tensed verbs.  The  presence and correct placement of a tensed verb (or Aux/Mod) is itself evidence for an FP (most  67  likely IP/TP). I now turn to the untensed constructions that the Lexical Projection Hypothesis predicts not to arise. They are listed below. (54) Word Orders Predicted to Be Unattested by the Lexical Projection Hypothesis a. b.  S - Neg O-Neg  c. d. e. f.  S-V S-O O-V O-S  (for (for (for (for  VOS languages) VOS languages) SVO and VOS languages) SVO and SOV languages)  Each of these orders requires an argument (subject or object) to move to the specifier position of an FP, [Spec FP].  Some of these orders are attested in early acquisition and  agrammatism. First, we find subjects preceding Neg for acquisition (55) and agrammatism (56), as shown below ((55a) from Bellugi (1967) in Deprez & Pierce (1993); (55b) from Clahsen et al. (1994)). The German example has an adverb instead of a negation. The same principle holds, with adverbs adjoining to V P and the subect having to move out of the V P . For more examples of Subject-Neg/Adv in agrammatic clauses, see Appendix (16). (55) Early Acquisition Clauses: S-Neg a. {he no} bite you 'he isn't biting you' {S - Neg} - V - O  English  b. {du nich} kochen you not cook:Inf "you are not cooking' {S-Neg}-V  German  (56) Agrammatic Clauses: S-Neg or S-Adv a. no examples in database  English  b. {[Poss] vrouw [Mod] toevallig} eventjes thuis-kommen [Poss] wife [Mod] accidentally for-a-while home-come:Inf "my wife accidentally came home for a while" {S - [Mod] - Adv} - PP -V  Dutch  68  c. drei Monate [Mod] ich uberhaupt nicht reden three months I at-all not speak:Inf "for three months I couldn't speak at all" PP - [Mod] - {S - Neg} - V  German  d- {je [CI] [V] peu (pas) du tout} le voyage I [CI] [V] little (not) at all the trip "I don't remember the trip" {S-[C1]-[V]-Neg}0  French  e. no examples in database  Italian  To derive the S-Neg order, the S must move from its underlying position to a specifier position higher than the V P . This movement is possible only if an FP is present to offer a landing site. This movement is schematized in the following diagrams, (57a) for SVO languages and (57b) for SOV languages. (57)  FP  a.  Specifier F  VP Neg^^VP SUBJECT^^^ V VERB^^OBJECT  Specifier  The next set of word orders are S-V and S-O, for Italian. I present the examples below. For more examples of subject-initial Italian agrammatic clauses, see Appendix (17).  69  (58) Agrammatic Italian: {S-V} and {S-0} a. {il signore e la signora [Aux] meravigliati} the gentleman and the lady [Aux] astonish: Part-Pi "the gentleman and the lady are astonished" {S - [Aux] - V}  Italian  b. {io [Cop] assessore} pure I [Cop] member too "I am a member too" {S - [Cop] - 0} - Adv  Italian  To derive both the S-V and the S-0 orders, the S must move from its underlying postverbal position to a specifier position higher than the V P . This movement is possible only if an FP is present to offer a landing site. This movement is schematized in the following diagram. (59)  We are left with the following 3 word orders: O-Neg, O-V and O-S. A l l 3 word orders involve moving the object. Other than topicalization (and the like), object fronting is common only in Dutch and German. The O-V order is not relevant to the Dutch and German data since they are SOV (so OV is the Underlying-Structure order).  However, Dutch and German do  display the O-S and O-Neg orders, which cannot be underlying orders. O-S order is illustrated in (60). O-Neg order is illustrated in (61) and (62) ((61a) from Clahsen et al. (1994)). Once again, the Dutch and German agrammatic sentences in (62) have adverbs instead of negation, with the structures being the same.  70  (60) Agrammatic Clauses: {O-S} a. {[Det] kippebout [Aux] [Det] hond} gepakt [Det] drumstick [Aux] [Det] dog take:Part "the dog has taken the drumstick" {O - [Aux] - S} - V  Dutch  b. no examples in database  German  (61) Early Acquisition Clauses: {O-Neg} da {Auto nicht} fahren there car not drive Adv - {O - Neg} - V  German  (62) Agrammatic untensed clauses: O-Adv a. en {kantoor [Mod] verder} slapen and office [Mod] further sleep:Inf "and (he) can sleep some more at the office" Conj - {O - [Mod] - Adv} - V  Dutch  b. {mir [Aux] auch} passiert me:Dat [Aux] also happen:Part "that has also happened to me" {O - [Aux] - Adv} - V  German  To derive both the O-S and the O-Neg/Adv orders, the S must move from its underlying position to a specifier position higher than the V P . This movement is possible only if an FP is present to offer a landing site. This movement is schematized in the following diagram. (63)  FP  Specifier  F' F  VP  Neg/Adv  VP V ' ^ ^  OBJECT  SUBJECT  VERB  The Lexical Projection Hypothesis (LPH) cannot account for the presence of FPs nor can it derive any word orders requiring movement of a head (eg. V°) to a higher F° position, or of a  71  phrase (eg. DP) to [Spec FP]. Since both of these are in evidence in agrammatic speech, FPs must be present. The Lexical Hypothesis therefore does not hold for agrammatic speech. I have summarized the (lack of) evidence for L P H in the chart below. The chart should be read as follows. The leftmost column lists various word orders. The second column indicates if the word orders are attested in early acquisition. The third column indictes if the word orders are attested in agrammatic production. And the last column indicates whether or not these word orders are predicted by the Lexical Projection Hypothesis (LPH). orders are indicated by "V".  Predicted or attested word  Word orders that are predicted not to occur and those that are  unattested are indicated by "*". For example, consider the second row (excluding the title row). The order in question is S-Neg/Adv. It is predicted by Lexical Projection Hypothesis not to occur, hence "*" under LPH Prediction . However, this word order is attested in both early acquisition and agrammatic production, hence "V ' under their respective columns. 4  Table 1: Evidence for the Lexical Projection Hypothesis (LPH) Acquisition Data  Agrammatism Data  LPH Prediction  V-Subject (all but I)  V  .V  *  V-Object ( D & G )  V  V  *  V - Neg/Adv  V  V  *  Subject - Neg/Adv  V  V  *  Subject - V (I)  ?  V  *  Subject - Object (I)  ?  V  *  Object-Subject ( D & G )  *  V  *  Object - Neg/Adv (D & G)  V  V  *  72  Note that for the table above E=English, D=Dutch, G=German, F=French and I=Italian. The Lexical Projection Hypothesis states that there are no FPs. If no FPs are present, all the syntax is done within the confines of lexical projections. We then would not expect to find any word orders that must be derived by movement to FPs. In other words, we expect to find word orders identical to the D-Structure word order, excluding the possibility of movement by adjunction to a lexical projection.  However, both early acquisition and agrammatic clauses  display orders other than D-Structure. These orders require movement of various constituents. It cannot be the case that only lexical projections make up the structure of the untensed acquisition and agrammatic clause. The Lexical Projection Hypothesis therefore does not hold. Functional projections are indeed present. But how many, and which ones?  3.2.2 Agrammatic Clauses Have at Least Two Functional Projections  I have argued that untensed agrammatic clauses involve more than lexical projections. I now turn to the next logical possibility: agrammatic clauses have a substantial amount of structure, but not the full array of projections. Within the acquisition literature there are two such proposals: Rizzi's (1994a,b) Truncation Hypothesis (TH) and Poeppel & Wexler's (1993) Optional Tense Hypothesis (OTH). Rizzi argues that root infinitives in early acquisition arise from truncating the adult structure: children project everything below TP. Poeppel & Wexler argue that TP is missing in root infinitives, but that everything else is present. The implications for these two hypotheses depend on the structures that one adopts. Consider the following two structures. The general phrase structure #1 in (65) is that assumed by Poeppel & Wexler (1993) . The general phrase structure #2 in (66) is that assumed by Rizzi (1994a,b). 26  Where Poeppel & Wexler (1993) use IP, I substitute TP for consistency. 73  (65) General Phrase Structure #1 CP ^  TP ^  VP  (66) General Phrase Structure #2 CP ^ AgrSP ^ TP AgrOP VP  The Truncation Hypothesis predicts that everything lower than TP is projected.  The  resulting structures are as follows. The structure in (67) assumes the general phrase structure #1 in (65). The structure in (68) assumes the general phrase structure #2 in (66). (67) Phrase Structure #1 Predicted by the Truncation Hypothesis VP (68) Phrase Structure #2 Predicted by the Truncation Hypothesis AgrOP ^ VP  If the only functional projections higher than V P are C P and TP, the Truncation Hypothesis predicts that no FPs are present in truncated structures.  This is equivalent to the  Lexical Projection Hypothesis, for tenseless clauses. A s argued above, there is indeed movement to FPs, even in clauses with no functional inflection: consequently, the structure in (67) is not viable. If, however, the general phrase structure is expanded, the Truncation Hypothesis predicts  74  the presence of one FP higher than VP, as proposed by Rizzi himself . 27  I now consider the predicted structures for the Optional Tense Hypothesis, which predicts that everything except TP is projected. The resulting structures are as follows. The structure in (69) assumes the general phrase structure #1 in (65). The structure in (70) assumes the general phrase structure #2 in (66). (69) Phrase Structure #1 Predicted by the Optional Tense Hypothesis CP ^  VP  (70) Phrase Structure #2 Predicted by the Optional Tense Hypothesis CP ^  AgrSP ^ AgrOP " VP  If the only functional projections higher than V P are C P and TP, the Optional Tense Hypothesis predicts the presence of one FP higher than V P . This is the same prediction as the Truncation Hypothesis #2.  If the general phrase structure is expanded, the Optional Tense  Hypothesis predicts the presence of three FPs higher than V P . The former corresponds to Poeppel & Wexler's original version of the Optional Tense Hypothesis.  Thus both the  Truncation Hypothesis and the Optional Tense Hypothesis predict the presence of a single F P higher than V P . However, allowing for a broader interpretation (i.e. the use of general phrase structure #2 for both hypotheses), the two hypotheses contrast: the Truncation Hypothesis predicts a maximum of one FP higher than V P , whereas the Optional Tense Hypothesis allows for more than one FP higher than VP.  The two hypotheses contrast further: the Truncation  Neither of the structures in (67-68) allows for object shift in German and Dutch, since object shift requires the presence of [Spec CP]. However, object shift is attested in agrammatic Genral  27  75  Hypothesis predicts CP to be absent; the Optional Tense Hypothesis predicts CP to be present. I begin with a general investigation of the syntactic structures underlying agrammatism by determining the number of FPs required to account for the agrammatic data. I then consider more specifically the two hypotheses discussed above, the Truncation Hypothesis and the Optional Tense Hypothesis, comparing acquisition and agrammatic data.  I conclude that neither  hypothesis, as formulated by their authors, can account for the agrammatic data.  3.2.2.1 How Much Structure Is in Agrammatic Clauses?  The two hypotheses predict a different number of FPs in untensed clauses: the Truncation Hypothesis predicts the existence of only one FP higher than V P whereas the Optional Tense Hypothesis predicts more than one. I begin by revisiting the underlying structures of the languages in question. The underlying structure for English and French presented in (71), that for Dutch and German in (72), and thatfor Italian in (73). (71) English and French Underlying Structure: SVO VP V  SUBJECT VERB  OBJECT  (72) German and Dutch Underlying Structure: SOV VP SUBJECT  V  OBJECT  VERB  (73) Italian Underlying Structure: VOS VP V VERB  SUBJECT OBJECT  and Dutch, as discussed below. 76  Movement to an FP constitutes evidence for that FP.  Any order that differs from the  underlying orders presented above requires movement to an FP.  Moreover, if any of the  constituents move past a 'fixed structure', that too constitutes evidence for FPs. The 'fixed structures' I consider are adverbs and negation (Pollock, 1989). Some adverbs adjoin to V P . Adding adverbs to the structures above results in the following. (74) English and French Underlying Structure: Adv-S-V-0 VP A d v ^ ^ V P SUBJECT  V  VERB  OBJECT  (75) German and Dutch Underlying Structure: Adv-S-O-V VP Adv  VP  SUBJECT OBJECT  VERB  (76) Italian Underlying Structure: Adv-V-O-S VP Adv  VP V ' ^ ^ ^ SUBJECT  VERB^^OBJECT The underlying orders with Adv is the following: Adv-S-V-0 for English and French, Adv-S-O-V for Dutch and German, and Adv-V-O-S for Italian. Adverbs are adjuncts. They do not provide landing sites for moving constituents. As such, for every constituent that moves past  77  Adv there must be one FP. If two constituents move past Adv, there must be (at least) two FPs to provide the appropriate docking sites. I now turn to the second construction: negation. When discussing negation in the section 3.2.1.2 (when arguing against the Lexical Projection Hypothesis), I assumed that negation was adjoined to VP, like adverbs are here. The alternative is to have negation as a functional head that projects NegP. I repeat the two structures below, (77a) illustrates negation as a functional head, and (77b) negation as an adjunct. (77)  a.  If negation is an adjunct, it provides no landing site for moving constituents. If negation is a functional head, it does. Although the debate between adjuct and functional head allows me to choose either alternative, for purposes of illustration I assume negation to be a functional head. Negation therefore offers a landing site for constituent movement; the site is [Spec NegP]. In order to show that there is at least one FP above NegP, two constituents must move; one docking in [Spec NegP] and one landing in [Spec FP]. This choice thus places a heavier burden on the data. Adding NegP to the underlying structures in (71)-(73) results in the following. (78) English and French Underlying Structure: Neg-S-V-0 NegP "^^Neg' Neg  VP  SUBJECT  V  VERB^^OBJECT  78  (79) German and Dutch Underlying Structure: Neg-S-O-V NegP Neg' Neg  VP  SUBJECT OBJECT  VERB  (80) Italian Underlying Structure: Neg-V-O-S NegP ^ ^ N e g ' Neg  VP V'^^SUBJECT  V E R B ^ ^ ^ OBJECT The word orders that can only be accounted for by the presence of FPs require movement out of the V P . Recall that the relevant data are untensed clauses. The absence of tense can be achieved in two ways. First, if a verb is present it must be untensed (infinitival or participial) (note that no tensed auxiliary or modal can be present either). Second, a clause may simply be lacking any verbal element. Contra Pollock (1989), I assume that untensed verbs (infinitivals as well as participials) are licit in their underlying position and so do not (need to) move overtly . I 28  now turn to the various word orders that require movement and their implications for FPs. I consider three types of movement: those requiring only NegP, those requiring one FP other than NegP, and those requiring two FPs other than NegP.  3.2.2.1.1  Movement Requiring Only NegP  Consider the following list of word orders. I have restricted the list to orders that can actually be produced in the respective languages. I indicate word orders that are attested in the 2 8  I assume that participials do not project a participial phrase (PartP). 79  database with a check mark 'V' to the left of the string. Unattested word orders I leave bare. This presentation is adopted throughout the rest of this section (3.2.2.1).  More studies are  required to determine whether or not these unattested orders are accidental gaps.  This proviso  holds for all unattested word orders under discussion. (81) Word Orders Requiring Movement but No Functional Projections other than NegP a. b. c.  V  S - Neg - (V) - O S - Neg - O - (V) O-Neg-S-(V)  for SVO and VOS languages for SOV languages for SOV languages  Subjects raise past Neg, as illustrated in (82). The last line of each example displays the word order. I include in curly brackets the portion of the utterance that is relevant to movement. This presentation is maintained throughout section 3.2.2.1. For more examples see Appendix (16). (82) Agrammatic Clauses: {S-Neg} a. no examples in database  English  b. {ik [Cop] niet} bij I [Cop] not with "I was not conscious" {S - [Cop] - Neg} - P  Dutch  c. und {die Abgase [Cop] nicht} desund and the exhaust-fumes [Cop] not healthy "and the exhaust fumes were not healthy"  German  Conj-{S-[Cop]-Neg}-Adj French  d. no examples in database  Italian  e. no examples in database  For each of the examples, the subject moves past Neg and lands in [Spec NegP]. This is schematized below.  80  (83)  NegP SUBJECT  Neg' N e g V P  Subject movement past Neg does not require any FP other than NegP  3.2.2.1.2  29  Movement Requiring One Functional Projection  The following sets of word orders each require one FP other than NegP. There is a large range of possible word orders; however, only a few are attested in the database. I suggest that paucity of evidence is a function of the nature of the data, not the grammar. First, the required data consists of untensed clauses, which immediately reduces the selection. Second, agrammatics omit all categories; however, omissions are only counted for items that are missing from obligatory contexts. This hides the fact that optional categories are omitted, probably much more frequently; adverbs and negation are two such optional categories (given the production tasks at hand). Since discerning many of the word orders is dependent on the presence of these optional elements, it is not surprising that these orders are rare or unattested on the basis of surface distribution.  Third, while agrammatics do produce correct non-canonical word orders in  appropriate discourse contexts, they prefer canonical word orders. Many of the word orders that display movement are non-canonical. Once again, it is not surprising that the evidence is limited. The clauses that constitute evidence for movement thus require the following combination of  Note that if the debate favors negation as adjunct, subject/object movement past Neg is evidence for one FP higher than Neg, as schematized in (i). If two constituents move past Neg, then two FPs are required, and so on. (i) FP 2 9  SUBJECT or OBJECT  VP  81  characteristics: the presence of the right obligatory constituents, the presence of optional categories, the absence of tense, and non-canonical word orders. The production tasks that the agrammatic subjects are required to perform for this database are not particularly rich. For example, a picture description does not necessarily encourage the production of negation or adverbs . 30  It is surprising that any clauses with the appropriate characteristics are produced.  Therefore, even if the tokens are few, they must be taken seriously and be accounted for. (84) Word Orders Requiring One Functional Projection Higher than NegP a. V  O - S - (V)  for SOV languages  b. V  Subject-initial  for VOS languages  (85) Word Orders Requiring One Functional Projection Higher than NegP a. b. V c. V d. V e. V c.  S - Adv - (V) - O S - Adv - O - (V) O - Adv PPloc - Adv Predicate -Adv S-O-Neg-(V)  for for for for for for  SVO SOV SOV SOV SOV SOV  and SOV languages languages languages languages languages languages  d.  O - S - Neg - (V)  for SOV languages  (86) Word Orders Requiring One Functional Category Lower than NegP S - Neg - O - (V)  for VOS languages  I now discuss the derivation of each attested word order.  In non-pathological Dutch,  objects may raise past subjects. For each of the examples, the object moves past the subject and thus out of the V P . Since NegP is not present, another FP must supply its specifier position as a landing site. The examples are illustrated in (87). The movement is schematized in (88). (87) Agrammatic Dutch and German: {O-S} a. {[Det] kippebout [Aux] [Det] hond} gepakt [Det] drumstick [Aux] [Det] dog take: Part "the dog takes the drumstick" {O - [Aux] - S} - Vpart  Dutch  What is required is a battery of production tasks that focus on adverbs, prepositions, questions, and any other construction that requires movement.  3 0  82  b. {[Det] man [Modal] [Det] wekker} roepen [Det] man [Modal] [Det] alarm call:Inf "the alarm clock is calling/waking the man" {O - [Mod] - S} - Vinf  Dutch  c. no examples in database  German  (88) Derived Syntactic Structure for Sentence (87a) FP OBJECT kippebout ([Det] drumstick)  F' VP SUBJECT hond ([Det] dog)  VERB gepakt (take:Part)  Next, Italian subjects surface in clause-initial position. This ordering requires the subject to move out of the V P and into the specifier of a dominating FP. The construction is illustrated in (89). The derivation is schematized in (90). (89) Agrammatic Italian Subject-Initial Clauses: a.{S-V}, b.{S-Adj}, c.{S-N} andd.{S-PP} a. {la signora [Aux] scocciata} the lady [Aux] annoy:Part "the lady is annoyed" {S - [Aux] - Vpart}  Italian  b. e {la nonna e Cappuccetto Rosso [Cop] salve} and the grandmother and L R R H [Cop] safe:F-Pl "and the grandmother and L R R H were safe" Conj-{S-[Cop]-Adj}  Italian  c. e {io [Cop] assessore} pure and I [Cop] member too "and I am a member too" Conj - {S - [Cop] - N} - Adv  Italian  d. {[Det] acqua [V] sotto (sul) [Det] pavimenti (pavimento)} [Det] water [V] under (on) [Det] floor:Pl (floor:Sg) "the water goes on the floor" {S - [V] - PP}  83  Italian  (90) Derived Syntactic Structure for Sentence (89a) FP SUBJECT la signora (the lady)  ^ F" F ^ ^  VP  VERB scocciata (annoyed)  Each of the following four sets of sentences involves movement past Adv. The four constituents that move are as follows: subjects (91), objects (92), locative prepositional phrases (Pploc) (93), and predicates (adjective phrases, AP) (94).  These word orders require that the  relevant constituent move out of the V P , past Adv and into the specifier of a dominating FP. The constructions are illustrated below. I present only one derivation since the same movement out of the V P is involved. The structure in (95) includes two adverbs. The movement still holds for those sentences with only one adverb. (91) Agrammatic Clauses: {S-Adv} a. no examples in database  English  b. [Adv] [Modal] [Det] {vrouw toevallig} eventjes thuis-kommen Dutch [Adv] [Modal] [Det] wife accidentally for-a-while home-come:Inf "then my wife accidentally came home for a while" [Adv] - [Mod] - {S - Adv} - Adv - Vinf c. und {der Wolf [V] inzwischen} die Grossmutter ah im Bauch and the wolf [V] meanwhile the grandmother uh in+the belly "and in the meantime the wolf has the grandmother in its belly" Conj - {S - [V] - Adv} - O - PP  German  d. no examples in database  French  e. no examples in database  Italian  84  (92) Agrammatic Dutch and German: {O-Adv} a. no examples in database  Dutch  b. [DP] [Aux] {mir auch} passiert [DP] [Aux] me:Dat also happen:Part "that has also happened to me" [S]-[Aux]-{O-Adv}-Vpart  German  (93) Agrammatic Dutch and German: {PPloc-Adv} a. en [DP] [Modal] hard lopen en {[P] kantoor verder} slapen and [DP] [Modal] fast run:Inf and [P] office further sleep:Inf "and he must run fast and can sleep some more at the office" Conj - [S] - [Mod] - Adv - Vinf - Conj - {PPloc - Adv} - Vinf  Dutch  b. {[P] kantoor [V] [DP] eindelijk} rust [P] office [V] [DP] finally peace "at the office he finally has some peace" {PPloc - [V] - [DP] - Adv} - Adj  Dutch  c. no examples in database  German  (94) Agrammatic Clauses: {AP-Adv}or {DP-Adv} (Predicate movement) a. nou ja {moeilijk [Modal] [Pleo] altijd} blijven well yes difficult [Modal] [Pleo] always remain: Inf "well yes, it will always remain difficult" Fill - {AP - [Mod] - [Pleo] - Adv} - Vinf  Dutch  b. {verdwenen [Cop] de kip} disappeared [Cop] the chicken "the chicken has disappeared" {AP-[Cop]-S}  Dutch  c. {acht uur [Cop] pas (al)} eight o'clock [Cop] only (already) "it's already eight o'clock" { D P - [Cop]-Adv}  Dutch  d. no examples in database  German  85  (95) Derived Syntactic Structure for Sentence (91b) FP SUBJECT vrouw ([Det] wife)  F' F  VP ADV VP toevallig (accidentally) A D V eventjes (for-a-while) L  VP  VERB thuis-kommen (home-come:Inf) A l l of the agrammatic clauses presented above have required the presence of an FP other than NegP. Thus agrammatic untensed clauses require at least one FP above NegP.  3.2.2.1.3  Movement Requiring Two Functional Projections  The word orders reflecting two FPs other than NegP are few. I present them below. (96) Word Orders Requiring Two Functional Projections above NegP a. V b.  S-O-Adv-(V) O - S - Adv-(V)  for SOV languages for SOV languages  The following sentence involves movement past Adv of both the subject and the object. Two FPs are required. The sentence is presented in (97). Its derivation is schematized in (98). (97) Agrammatic Dutch: {S-O-Adv}  31  maar {ik [V] [Det] auto graag} but I [V] [Det] car happily "but I happily drive the car" Conj - {S - [V] - O - Adv}  Dutch  Note that the adverb graag "happily" cannot be a main predicate adjective since cross-categorial substitution does not occur (see section 4.1.1).  3 1  86  (98) Derived Structure for Sentence (97) FP1 SUBJECT ik (I)  F' F OBJECT auto ([Det] car)  FP2 F' F  VP ADV graag (happily)  VP L  V VERB 0  Both the subject and the object move from their VP-internal positions past Adv to specifier positions of functional projections higher than the V P . Since each argument requires its own landing site, two functional projections must be present.  Thus agrammatic (untensed)  clauses must include at least two functional projections (other than NegP).  Neither the  Truncation Hypothesis nor the Optional Tense Hypothesis accounts for this agrammatic production data. I now return to the specific analyses entailed by the Truncation Hypothesis and the Optional Tense Hypothesis. I begin with Rizzi's (1994b) Truncation Hypothesis.  3.2.2.2  (More) Arguments Against the Truncation Hypothesis  I have argued that two FPs higher than V P are required to account for the agrammatic data. This alone is counter-evidence for the Truncation Hypothesis. However, Rizzi provides 5 other arguments in support of the Truncation Hypothesis. The Truncation Hypothesis predicts the following.  87  (i) Vtns appears in clause-final position in early Dutch and German (for matrix clauses) (ii) Wh-elements do not co-occur with Vinf (iii) Subject pronouns rarely co-occur with Vinf (iv) Root infinitives do not involve infinitival Aux or Modal (v) Root infinitives are rare in early Italian. I present each of the five predictions for the acquisition data and in turn address each one to the agrammatic data, showing that only one is borne out.  3.2.2.2.1  Clause-Final Tensed Verbs  Although the tensed-untensed distinction is fairly well established in earlier acquisition (Deprez & Pierce, 1993; Poeppel & Wexler, 1993; Guasti, 1992; Wexler, 1994), with Vtns in second position and untensed V in clause-final position, Rizzi argues that Vtns can occasionally be found in clause-final position. Consider the early German clauses Rizzi cites from Park (1981, in Deprez & Pierce (1993)) (99a) and Clahsen et al. (1994) (99b-c). (99) Early German: Vtns in clause-final position a. da Bela Kuche-backe macht there Bela cake-baking does "Bela is baking a cake" Adv - S - O - Vtns b. der Hahn nich macht the rooster not does "the rooster is not doing (anything)" S - Neg - Vtns c. Julia Schere nicht darf Julia scissors not may "Julia may not have the scissors" S - O - Neg - Vtns Rizzi proposes a transition stage where TP is operational, but CP is not. The tensed V raises toT, but since TP is right-headed (in Dutch and German), Vtns remains clause-final. Let us run through the derivation of each sentence in (99). following clause structure from Belletti (1990).  88  First, Rizzi (1994b) assumes the  (100) General phrase structure (CP) AgrSP NegP Neg  T  SUBJECT  Note that sentence (99a) has the word order Adv-S-O-V. This is the underlying order and so does not require any movement other than string-vacuous V-to-T. Both (99b) and (99c) include negation. They have the orders S-Neg-Vtns and S-O-NegVtns, respectively. Both of these sentences involve movement of the subject past (or to) NegP. If this is a transitional stage, then C P is absent. Given the structure in (100), these orders can only be derived if S and O each move into one of [Spec NegP] or [Spec AgrSP]. As long as both of these are present, then these orders are derivable . 32  Although early acquisition may go through a transition period that reflects a missing C P , this data is unattested in agrammatic German (and Dutch). Vtns appears in clause-final position  Friedeman(1993) argues that the few examples of negated untensed sentences in early French (7 out of 137) are instances of constituent negation. If NegP is higher than TP, then the truncation account predicts the rarity of Neg in untensed clauses in French acquisition. However, there seems to be much debate about (i) the placement of NegP (see Pollock (1989) for below TP; Belletti(1990) for above TP; among others), and (ii) the importance of Neg in early acquisition (see Deprez & Pierce (1993) for arguments using NegP in early acquisition). Regardless, the following predictions result from the Truncation Hypothesis: 1. If NegP is higher than TP, NegP should be omitted; negated root infinitives being rare. 2. If NegP is lower than TP, NegP should be available; negated root infinitives being frequent. 3. If the position of NegP is parametrized, then the child data will be dependent on the particular setting. 3 2  89  in only two contexts: embedded clauses or V2. When in embedded clauses, Vtns is supposed to appear clause-finally. When clause-final and V2 positions are one and the same, no argument can be made. Crucially, clause-final Vtns in utterances of more than two constituents are unattested in agrammatic German and Dutch. Thus this particlar prediction does not hold for agrammatism.  3.2.2.2.2  Wh-Elements Do Not Co-Occur with Infinitive Verbs  Rizzi's second argument for the Truncation Hypothesis is that Wh-elements do not cooccur with infinitival verbs (Vinf) (also see Crisma (1992) for French; Weissenborn (1992a) for German). If root infinitives lack CP (as is predicted by the Truncation Hypothesis), we would not expect to find Vinf with Wh-elements, since Wh-elements are in [Spec CP]. As predicted by the Truncation Hypothesis, we do not find the following in early acquisition. (101) Unattested child utterances: Wh-elements with Vinf *was Hans essen what Hans eat:Inf Agrammatic speech, on the other hand, does have Wh-phrases in untensed clauses. These are shown in (102). (102) Agrammatic clause: Root infinitive with Wh-element ou [DP] [Mod] voir cardiologue where [DP] [Mod] see: Inf cardiologist "where I had to see a cardiologist"  French  The Truncation Hypothesis does not hold for agrammatic data.  3.2.2.2.3  Subject Clitics Do Not Co-Occur with Untensed Verbs  Rizzi's third argument for the Truncation Hypothesis is that, in early French, subject clitics rarely co-occur with untensed verbs, such as Vinf and Vpart (Rizzi cites Pierce (1989:45) for French and Haegeman (1994) for Dutch). The argument is the following. The Truncation Hypothesis states that root infinitives only have projections lower than TP. Subject clitics (in 90  French, at least) require AgrSP ; but AgrSP is higher than TP. Thus root infinitives are 33  predicted not to include subject pronouns. This prediction is borne out in the early acquisition data. However, this is not the case with the agrammatic data. Although agrammatic French does not display subject clitics with Vinf, we do find subject clitics with Vpart, as illustrated in (103). (103) Agrammatic French: Subject Clitics in Untensed Clauses a. je [Aux] perdu la parole [P] trois minutes I [Aux] lose:Part the speech [P] three minutes "I lost my speech within three minutes" b. apres, j ' [Aux] attendu [Poss] sortie after, I [Aux] waitPart [Poss] release "afterwards, I waited to be discharged" Rizzi argues that the presence of a subject clitic entails the projection of AgrSP. These untensed clauses have subject pronouns. Their structure must therefore include AgrSP. AgrSP is above TP. This is contrary to the Truncation Hypothesis prediction that untensed clauses do not co-occur with subject clitics since all projections above and including T P are absent. The Truncation Hypothesis therefore does not hold for agrammatics. Further evidence is provided by the following example. (104) Agrammatic French: Subject Clitics in Verbless clause je [CI] [V] pen (pas) du tout le voyage I [CI] [V] little (not) at all the trip "I don't remember the trip at all" In (104) the verb is missing. For all intents and purposes, this ought to be a truncated structure, in the sense of the Truncation Hypothesis. As I will argue below, a missing lexical category does not entail a missing lexical projection. Both subject and object are present and Rizzi(1986) argue that subject pronouns cliticizeonto AgrS; Pierce(1989) reasons that subject pronouns are AgrS markers; and Cardinaletti & Starke (1993) maintain that subject pronouns are  3 3  91  interpretable. However, these arguments could be within the confines of a V P if it were not for two other facts.  First, the clause includes a subject clitic. Following Rizzi, this entails the  projection of AgrSP. Moreover, the clause in question includes Neg.  Since the subject pronoun  is to the left of Neg, and assuming the subject is underlyingly VP-internal, minimally it moves into [Spec NegP]. The Truncation Hypothesis assumes that only projections lower than TP are present. However, this untensed clause has both AgrSP and NegP, both of which are higher than TP. The Truncation Hypothesis therefore does not hold.  3.2.2.2.4  No Infinitival Auxiliaries or Modals  Rizzi's fourth argument for the Truncation Hypothesis is that infinitival Aux or Modal is unattested. Untensed clauses involve lexical verbs and not modals or auxiliaries. We do not find the following (egs from Wexler (1994)). (105) Unattested structures: Infinitival auxiliaries a. *avoir mange have:Inf eatPart "to have eaten"  French  b. *etre venu be:Inf come:Part "to have come"  French  c. * gefaukt haben buy:Part have:Inf "to have bought"  German  Rizzi (1994a) argues that Aux is either generated in T, or heads its own V P but is inherently specified for T features, which it must check by moving to T. In either case, the presence of Aux implies the presence of T, and thus a tensed Aux. The prediction that results from the Truncation Hypothesis is the following. Since root infinitives have no T P or higher projections, root infintives will never include Aux or Modals. Specifically, we ought never to  weak and must be licensed by AgrS. In each case, the relationship is such that the presence of a subject pronoun requires that AgrSP be projected. 92  find untensed Aux or Modals. This prediction is borne out in both early acquisition and agrammatic speech.  3.2.2.2.5  Few Root Infinitives in Italian  Rizzi'sfinal argumentfor the Truncation Hypothesis is the virtual non-occurence of root infinitives in Italian (Guasti, 1992; Cipriani et al., 1993). Root infinitives are frequent in early French (Pierce, 1989) and early German, but not in early Italian. Rizzi attributes this to the differing role of TP for Vinf in the various languages. Compare the following two paradigms. (106) French a. Jean n'a pas b. *Jean ne pas a  lu le livre lu le livre  c. Jean ne lit pas le livre d. *Jean ne pas lit le livre e. n'avoir f. ne  pas pas avoir  lu le livre lu le livre  g. *ne lire pas le livre h. ne pas lire le livre (107) Italian a. Gianni non ha piu letto i l libro b. *Gianni non piu ha letto i l libro c. Gianni non legge piu i l libro d. *Gianni non piu legge i l libro e. non aver piu letto i l libro f. *non piu aver letto i l libro g. non leggere piu il libro h. *non piu leggere i l libro (108) English translation of the paradigms above a. John has not read the book b. John not has read the book c. John reads not the book d. John not reads the book 93  e. to have not read the book f. not to have read the book g. to read not the book h. not to read the book In French both Aux and V must raise to T when they are tensed, as shown in (106a-d). This is true for Italian as well, as shown in (107a-d). The derivations for (106c) and (107c) are schematized below. (109) Derivationfor French sentence (106c) TP  tV 1  DP le livre  (110) Derivationfor Italian sentence (107c) TP D P ^ ^ T ' Gianni T non legge  VP  tV J  DP i l libro  [Spec VP] is to the right in Italian according to previous discussion. 94  The behavior of infinitivals is where the two languages differ. In French, infinitival Aux may but need not raise to T (106e-f), and infinitival V must not raise to T (106g-h). In Italian, however, both infinitival Aux and V must raise to T (107e-h), shown in (111) (111) Derivation for French sentence (106g) TP  T ^ ^  VP  Neg ne pas  VP V lire 1  -X  DP le livre  (112) Derivation for Italian sentence (107g) TP  T non leggere  VP Neg piu  VP ^/"X^ V tV J  DP il libro  The Truncation Hypothesis states that TP is absent from root infinitives. This is of no concern to French, since Vinf must not raise to T. However, Italian Vinf moves to T. So if no TP is available, Vinf is not licensed and will not be produced. Thus the absence of root infinitives in early Italian argues against the presence of TP in such clauses. In agrammatic Italian, root infinitives are rare.  This parallels the early Italian data.  However, root infinitives are also rare in agrammatic French. This is in contrast to the frequent production of root infinitives in early French. The languages that display the largest number of untensed clauses are Dutch and German. Thus, in the agrammatic data the contrast is between  95  French and Italian, on the one hand, and Dutch and German, on the other. This contrast is not predicted by the Truncation Hypothesis. I summarize the evidence for the Truncation Hypothesis in the chart below. Table 2: Evidence for the Truncation Hypothesis (TH) Acquisition Data  Agrammatism Data  TH Prediction  Transition Clause-final Vtns  V?  *  V  Wh&Vinf/part  *  . V  *  Subject clitic & Vinf/part  *  V  *  Infinitival Aux or Modal  *  *  *  Root infinitives in Italian & French  *  V  *  The Truncation Hypothesis states that root infinitives only project structures lower than TP. The implication for this proposal is that untensed clauses only exhibit movement to positions below TP. Moreover, we do not expect to find untensed verbs (whether Vinf or Vpart) cooccuring with elements that require these outer-layer F-projections.  Although the Truncation  Hypothesis appears to be consistent with the acquisition data, these co-occurence restrictions cannot be extended to agrammatism. While it is true that auxiliaries and modals do not appear in infinitival form, no other predictions from the Truncation Hypothesis are borne out. The Truncation Hypothesis therefore does pot hold for agrammatic production.  96  3.2.2.3 Arguments Against the Optional Tense Hypothesis  Poeppel & Wexler argue for the existence of TP and CP in early acquisition based on the following distributionarfacts.  First, early German displays the V 2 construction. Regardless of  what appears in first position, German-speaking children place Vtns in second position and Vinf in clause-final position. Second, early French displays correct Neg distribution: Neg follows Vtns but precedes Vinf. Further, early Italian displays appropriate clitic placement: clitics precede Vtns and follow Vinf (Guasti, 1992). These three facts argue that children can distinguish between tensed and untensed verbs.  Moreover, Vtns bears appropriate agreement.  The  conclusion is that Vtns must move to T, for French and Italian, and then into C for German. Therefore, the functional projections T P and CP are present in tensed clauses in early acquisition. However, each of these arguments applies to tensed verbs.  As argued previously for  agrammatic utterances, the presence of a correctly positioned tensed element itself implies the presence of TP. I am concerned with the presence of FPs in untensed clauses. Poeppel & Wexler do address the issue of untensed clauses in early acquisition. This is the purpose of the Optional Tense Hypothesis.  The Optional Tense Hypothesis states that  untensed clauses project all functional projections except TP. This means that the only FP in untensed clauses is CP.  Consider the relevant structures.  Poeppel & Wexler assume the  following: Neg is an adjunct (like Adv), and only two FPs dominate V P (and Neg), these being CP and T P (my arguments against the Optional Tense Hypothesis assumes their structures). The structure is presented below (with subject, object and verb included in their underlying positions) . German is underlyingly SOV, with both V P and TP being right-headed (Vikner & 35  Schwartz, 1992; Vikner, 1990; Haegeman, 1992).  Where Poeppel & Wexler (1993) use IP, I have TP for consistency. 97  (113) German (and Dutch): Structure Assumed by Poeppel and Wexler CP  C ^ ^ ^ T P  V P ^ ^ T Neg/Adv  VP  SUBJECT  V  OBJECT  VERB  If TP is absent from untensed clauses, the structure is the following. (114) Structure Predicted by the Optional Tense Hypothesis (for German and Dutch) CP  C ^ ^ ^ V P Neg/Adv  VP  SUBJECT OBJECT  V VERB  With the Optional Tense Hypothesis structure, there is only one landing site for moving arguments, [Spec CP]. Therefore only one constituent can move past Neg or Adv. Specifically, we do not expect to find O-S-Neg/Adv-V. This construction requires two landing sites, one for the subject and one for the object, as schematized below.  98  (115) Derivation of O-S-Neg/Adv-V (for German and Dutch) CP OBJECT C '^^^^^^  ^ T P  S U B J E C T ' ^ ^ ^ ^ ^ T ' V P ^ ^ ^ T Neg/Adv  VP  5r The Optional Tense Hypothesis therefore predicts the order O-S-Neg/Adv to be unattested. This order is indeed unattested in both early German and agrammatic German and Dutch (see previous section for why this might be an accidental gap).  However, the Optional  Tense Hypothesis makes a further prediction: the order S-O-Neg/Adv should be unattested as well. I am unaware if this order appears in early German, but agrammatic Dutch displays such a structure (116). (116) Agrammatic Dutch: S-O-Adv maarik auto graag but I car happily "but I like to drive the car" Conj - S - O - A d v  Dutch  The underlying order for (116) is Adv-S-O: graag ik auto . However, we have S-O-Adv. This means that both the subject and the object have moved past Adv. As such two FPs above V P are required to accommodate the movement. The Optional Tense Hypothesis cannot account for such a sentence. Furthermore, even with T P present, the structure that Poeppel & Wexler adopt cannot accommodate such a string.  Following Poeppel & Wexler, the functional  projections are TP and CP. To derive the sentence in (116), the subject would have to move to  99  [Spec CP] and the object to [Spec TP]. The derivation is schematized below. (117) Derivation of O-S-Neg/Adv-V (for German and Dutch) CP SUBJECT ik (I)  C ' ^ ^ ^ ^ ^ T P OBJECT ^ ~ ^ ^ ^ ^ T ' auto (car) V P ^ ^ T A D V ^ " ^ " VP graag X \ (happily) t. V  This derivation is not licit. The subject must move to [Spec TP] to be assigned Case (it can then move to [Spec CP]). The CP-TP structure cannot account for this order. The Optional Tense Hypothesis, with its missing TP, cannot account for the full array of agrammatic clauses; moreover, neither can a model that is restricted to the functional projections C P and T P . The evidence for the Optional Tense Hypothesis is summarized in the table below.  Table 3: Evidence for the Optional Tense Hypothesis (OTH) Acquisition Data  Agrammatism Data  OTH Prediction  0 - S - Adv  *  *  *  S - 0 - Adv  *  V  *  Extending the Optional Tense Hypothesis to agrammatic production results in the following: only one functional projection is available in agrammatic untensed clauses.  100  The  prediction from such a structure is that only one constituent can move out of V P , and higher than Adv. Agrammatic Dutch displays the order S-O-Adv, where two constituents move past Adv. Two functional projections higher than V P are therefore required.  The Optional Tense  Hypothesis cannot account for this construction.  3.2.3 Agrammatic Clauses Have the Full Array of Syntactic Projections  I have argued for the existence of two FPs based on the number of constituents that move out of V P .  I now turn my argument in a different direction. First, I adopt an analysis that  motivates movement to specific syntactic positions: V-to-T, subject-to-[Spec TP], objectto[AgrOP], and so forth. Second, I assume that each FP in agrammatic clauses behaves as it would in a normal clause. Under such assumptions, we will not find subjects perched in [Spec AgrOP], or T P hosting objects, and so on. In other words, movement of an object is evidence for AgrOP (or CP, depending what else is present in the sentence), movement of a subject is evidence for TP, initial Wh-phrases are evidence for CP, and so forth. If only two FPs are present, we expect only two types of movement. I argue that agrammatic clauses display a larger inventory of movement and, consequently, require a larger set of FPs.  I propose that the full  array of projections is present in agrammatic clauses. Consider the following general phrase structure . 36  I recognize that AgrSP is another FP that many would include in such a structure. Both AgrSP and TP host subject. I collapse the two into T P for expository purposes. The data display movement to only a single FP that hosts a subject, so evidence for one or the other is difficult to tease apart. Why favor TP? AgrSP exists to provide the Spec-head relationship for subject-verb agreement. TP also affords this relationship and is required for nominative Case-assignment (to the subject) by T. Moreover, the most recent trend (Chomsky, 1993, 1995) is to do away with AgrPs altogether. While I am not prepared to rid myself of AgrOP, I do my part by eliminating AgrSP. 3 6  101  This structure displays three FPs other than NegP: CP, T P and an FP between TP and V P . I consider movement to each of these projections in turn, beginning with movement to C P .  3.2.3.1  Movement to C P  The first set of constructions are those that require CP. In normal German and Dutch, any constituent can move into sentence initial position (Haegeman, 1994). The position they move to is [Spec CP]. Both subjects and objects may move to [Spec CP]. However, in the agrammatic data under scrutiny this movement is indistinguishable from their respective movements to [Spec TP] and [Spec AgrOP].  To indentify movement to [Spec CP], other  constituents must be considered. The agrammatic data offers four such cases: fronting of locative prepositional phrases (PPloc), predicate-fronting, object-fronting and Wh-phrases in initial position. The sentences in (119) are examples of locative prepositional phrases in initial position. (119) Agrammatic Clauses: PP1 oc-Adv a. [P] kantoor [Mod] verder slapen [P] office [Mod] further sleep:Inf "(he) can sleep some more at the office" (S) - PPloc - [Mod] - Adv - Vinf  Dutch  b. [P] kantoor [V] [DP] eindelijk rust [P] office [V] [DP] finally peace "at the office he finally has some peace" PPloc - [V] - [DP] - Adv - Adj  Dutch  102  In (119), the locative PP ("at the office',') precedes Adv.  The derivation of the surface  order is the following. ThePP is underlyingly VP-internal, which means that it starts off to the right of Adv. The PP must then move to an FP dominating V P . This cannot be AgrOP, which hosts the object.  Nor can it be TP, which host the subject.  The landing site is [Spec C P ] .  Agrammatic clauses require CP. The derivation is schematized in (120). (120) Derivation of Sentence (119a): PP-Fronting CP PP ^ kantoor ([P] office)  ^  ^ C  ^  C ^  ADV verder (further)  VP ^ VP V V V slapen (sleep: Inf)  Next, the sentences in (121) illustrate predicate fronting. The predicates in question are the adjective phrases (AP) "difficult" (121a) and "disappeared" (121b) as well as the DP "eight o'clock" (121c). (121) Agrammatic Dutch and German: Predicate-Adv a. moeilijk [Modal] [Pleo] altijd blijven difficult [Modal] [Pleo] always remain:Inf "it will always remain difficult" A P - [Mod] - [Pleo] - Adv - Vinf  Dutch  b. verdwenen [Cop] de kip disappeared [Cop] the chicken "the chicken has disappeared" A P - [Cop] - S  Dutch  103  c. acht uur [Cop] pas (al) eight o'clock [Cop] only (already) "It's already eight o'clock" DP - [Cop] - Adv  Dutch  In (121), the predicates precedes Adv. The derivation matches that of the locative PPs. The predicates are underlyingly VP-internal, starting off to the right of Adv. They then move to an FP dominating VP. Once again, the landing site is [Spec CP]. Agrammatic clauses require CP. The derivation is schematized in (122). (122) Derivation of Sentence (121a): Predicate Fronting CP A P ^ ^ ^ ^ ^ C moeilijk (difficult)  C  ^ VP A D V ^ ^ ^ ^ V P altijd (always)  w  V V blijven (remain :Inf)  An alternate intrepretation for (121bc) is the following: the whole V P moves to [Spec CP] rather than just the AP/DP. The derivation would result in the following structure. (123) Derivation of Sentence (121c): VP-Fronting CP V P ^ ^ ^ ^ ^ C D P ^ ^ ^ VV 0 acht uur (eight o'clock)  C ' ^  X  V P  Adv pas (only)  104  Next, we find object-shift in Dutch, where the object precedes the subject. Examples are presented below. (124) Agrammatic Dutch and German: Object-Subject a. [Det] kippebout [Aux] [Det] hond gepakt [Det] drumstick [Aux] [Det] dog take: Part "the dog takes the drumstick" O - [Aux] - S - V  Dutch  b. [Det] man [Modal] [Det] wekker roepen [Det] man [Modal] [Det] alarm calhlnf "the alarm clock is calling/waking the man" O - [Mod] - S - V  Dutch  Although there is no overt evidence for subject movement, I have assumed in this derivation that the subject does indeed raise to [Spec TP] (as it should to receive nominative Case), the surface position of the object is [Spec CP]. The derivation is shown in (125). (125) Derivationfor Sentence (124a): Object-Subject-V CP OBJECT kippebout ([Det] drumstick)  C  /  /  X  "TP  SUBJECT hond ([Det] dog)  VP  T  V -to J  V gepakt (take:Part)  The object first raises to [Spec CP]. This derivation requires T P and CP (even if the subject does not raise, the object still moves to [Spec CP] since that is the focus position). Next, Wh-phrases surface clause-initially. This ordering requires the Wh-phrase to move out of the V P and into the specifier Of a dominating FP. The construction is illustrated in (126). The derivationis schematized in (127).  105  (126) Agrammatic Clause: Wh-elementin Clause-Initial Position a. ou [DP] [Mod] voir cardiologue where [DP] [Mod] see: Inf cardiologist "where I had to see a cardiologist"  French  b. dove [Cop] la foccacia where [Cop] the cake "where is the cake?"  Italian  (127) Derivation of (126b): Wh-Phrases CP Wh-Phrase dove (where)  C  VP V  DP la foccacia (the cake)  V 0  Raised Wh-phrases are always in [Spec CP].  In (127) the Wh-phrase is underlyingly  internal to VP, then raises to [Spec CP]. Once again, agrammatic clauses require C P . Evidence for movement to [Spec CP] is fairly robust. The evidence includes PP-fronting in Dutch, predicate-fronting in Dutch and German, and Wh-phrases.  3.2.3.2  Movement to TP  The second set of constructions are those that require TP: subject-initial clauses in Italian and subjects preceding Adv and Neg in Dutch and German. In Italian, subjects underlyingly follow their predicate. Thus, any sentence that is subject-initial requires that the subject move out of the V P .  The position they move to is [Spec TP]. Subject-initial clauses are presented in  (128).  106  (128) Agrammatic Italian Subject-Initial Clauses a. la signora [Aux] scocciata the lady [Aux] annoy:Part "the lady is annoyed" S - [Aux] - Vpart  Italian  b. la nonna e Cappuccetto Rosso [Cop] salve the grandmother and L R R H [Cop] safe:F-Pl "the grandmother and L R R H were safe" S - [Cop] - Adj  Italian  c. io [Cop] assessore pure and I [Cop] member too "and I am a member too" S - [Cop] - N - Adv  Italian  d. [Det] acqua [V] sotto (sul) [Det] pavimenti (pavimento) [Det] water [V] under (on) [Det] floor:Pl (floor:Sg) "the water goes on the floor" S - [V] - PP  Italian  In (128), the subjects precede their predicates, whether they are verbs (128a), adjectives (128b), nouns (128c), or PPs (128d). The derivation is as follows. The subject is underlyingly VP-internal, starting off to the right of the predicate.  The subject then moves to an F P  dominating V P . It lands is [Spec TP] to receive nominative Case (and check agreement with the verb/copula). Agrammatic clauses therefore require TP. The derivation is schematized in (129). (129) Derived Syntactic Structure for Sentence (128a) TP SUBJECT la signora (the lady)  T' T ^ ^ ^  VP  VERB scocciata (annoyed)  107  The next set of constructions I consider are Dutch and German clauses where the subject precedes Adv. Examples are presented in (130). (130) Agrammatic Clauses: Subject-Adv a. [Adv] [Modal] [Det] vrouw toevallig eventjesthuis-kommen Dutch [Adv] [Modal] [Det] wife accidentally for-a-while home-come: Inf "then my wife accidentally came home for a while" [Adv] - [Mod] - S - Adv - Adv - Vinf b. ik [Aux] nou 21 jaar gewerkt I [Aux] now 21 year work:Part "I have worked for 21 years" S - [Aux] - Adv - Vpart  Dutch  c. der Wolf [V] inzwischen die Grossmutter ah im Bauch the wolf [V] meanwhile the grandmother uh in+the belly "in the meantime the wolf has the grandmother in its belly" S - [V] - Adv - O - PP  German  d. das Kind und ihre (seine) Grossmutter [Cop] wieder normal the child and her:F (her:N) grandmother [Cop] again normal "the child and her grandmother were normal again" S - [Cop] - Adv - Adj  German  In (130), the subjects precede Adv. The derivation, schematized in (131), parallels that of clause-initial subjects in Italian. Subjects are underlyingly VP-internal, starting off to the right of Adv. The subjects then move to an FP dominating V P . They land in [Spec TP] to receive nominative Case (and check agreement with the verb/copula). Agrammatic clauses require TP. (131) Derivation for Sentence (130d): Subject-Adv TP DP das Kind und ihre Grossmutter T (the child and her grandmother) ADV wieder (again)  VP VP V  t.  A P ^ ^ V normal (normal)  108  The last set of constructions exemplifying movement to [Spec TP] are Dutch and German clauses where subjects precede Neg, as in (132). Earlier I argued that subjects land in [Spec NegP]. The purpose then was to determine how many FPs are required in agrammatic clauses. Having established the number of FPs, I am now investigating the nature of these FPs. Whether or not these subjects move to [Spec NegP], they must move to [Spec TP] to receive Case. (132) Agrammatic Clauses: Subject-Neg a. ik [Cop] niet bij I [Cop] not with "I was not conscious" S - [Cop] - Neg - P  Dutch  b. das [Cop] nicht gut that [Cop] not good "that's not good" S - [Cop] - Neg - Adj  German  c. die Abgase [Cop] nicht desund the exhaust-fumes [Cop] not healthy "the exhaust fumes were not healthy" S - [Cop] - Neg - Adj  German  In (132), the subjects precede Neg. The derivation matches that of the previous set. Subjects are underlyingly VP-internal, starting off to the right of Neg. The subjects then move to [Spec TP] to receive their nominative Case (and check agreement with the verb/copula). Once again, agrammatic clauses require TP. The derivation is schematized in (133). (133) Derivation of Sentence (132b): Subject-Neg TP SUBJECT das (that)  T ^ "  """"" V P NEG nicht (not)  VP L  V AP gut (good)  109  V 0  A l l three sets have displayed subject movement.  Subjects move into [Spec TP].  Agrammatic clauses thus require T P . Evidence for movement to [Spec TP] is substantial. The evidence includes subject-initial sentences in Italian, and clauses that display subjects preceding negation and adverbs.  3.2.3.3 Movement to a Functional Projection between TP and V P  The following two sets of constructions are those that require a functional projection between T P and V P , possibly AgrOP : the Subject-Object-Adv construction and objects 37  preceding Adv in German. Evidence for such an F P is thus based on object-shift (Homberg, 1986) . The Subject-Object-Adv construction is illustrated in (134). 38  The derivation is  schematized in (13 5). (134) Agrammatic Clause: Subject-Object-Adv ik [V] [Det] auto graag I [V] [Det] car happily "I happily drive the car" S-[V]-0-Adv  Dutch  Chomsky's (1995) Minimalist framework does away with Agr projections, including AgrOP. Instead, movement to the specifier of a projection headed by a "small v", [Spec vP], takes place. Homberg's generalization (1986) states that if the object shifts, the verb also moves. Thus object shift is a diagnostic for V-movementand hence evidence for whatever projection V moves to.  3 8  110  (135) Derivationfor Sentence (134): Subject-Object-Adv TP SUBJECT ik (I)  FP^^^T  OBJECT auto ([Det] car)  F' F ^ ^ " VP ADV graag (happily) L  VP  V 0 Both the subject and the object precede Adv. But the subject precedes the object. The object must be in [Spec FP] (possibly [Spec AgrOP]), with the subject in [Spec TP] (and possibly having subsequently moved to [Spec CP]). Agrammatic clauses thus require require a functional projection between TP and V P , possibly AgrOP. Next, agrammatic German displays object shifting. This is illustrated in (136) and (137). (136) Agrammatic German: Object-Adv [DP] [Aux] mir auch passiert [DP] [Aux] me:Dat also happen: Part "that has also happened to me" [S] - [Aux] - O - Adv - V  German  (137) Derived Syntactic Structure for Sentence (136) FP OBJECT mir (me)  VP ADV auch (also)  t,  111  VERB passiert (happen:Part)  The object precedes Adv. The object is underlyingly VP-internal, starting off to the right of Adv. It then moves to [Spec FP], passing Adv. Again, agrammatic clauses require an F P which dominates VP.  I propose that the FP is AgrOP since the movement involves object  shifting. However, this utterance has an alternative interpretation, presented in (138). (138) Agrammatic German: Object-Adv mir [Aux] [DP] auch passiert me:Dat [Aux] [DP] also happen:Part "that has also happened to me" O - [Aux] - [S] - Adv - V  German  Here, the object is focused and moves to [Spec CP]. The derivation is schematized in (139) . (139) Derived Syntactic Structure for Sentence (138)  (also)  u 1  VERB passiert (happen: Part)  Object fronting in (139) consists of focus NP shift to [Spec CP]. The object thus raises past Adv to [Spec CP]. Given this interpretation of the movement, agrammatic clauses require CP. I have argued that agrammatic clauses display movement to [Spec AgrOP], [Spec TP] and [Spec CP]. Consequently, AgrOP, T P and CP are required in agrammatic clauses. Evidence for movement to [Spec TP] and [Spec CP] is substantial. movement to [Spec AgrOP].  112  However, evidence is weaker for  There is much debate about how many FPs are projected in the syntax of normals. The debate has lead to an ever increasing inventory of FPs, ranging from the now familiar C P and TP, to AgrOP and AgrSP, to Asp(ect)P, Num(ber)P, K(ase)P and on from there. Arguing for the full array of FPs in agrammatic clauses is a rather difficult, if not impossible task. Nevertheless, I propose exactly that: agrammatic clauses consist of the full array of FPs available in the speaker's language i.e. agrammatism does not target any specific syntactic structure.  The  prediction is that whatever word orders result from these FPs will be in evidence in agrammatic clauses (tensed and untensed). In other words, no possible word order should be systematically missing. The tasks that make up the database for this thesis consists of fairly unconstrained production tasks: personal history (of illness), picture descriptions, and story telling. A database resulting from a battery of specialized production tasks is now needed to test for the various constructions and predicted word orders. I leave this for further research.  3.3  L E X I C A L PROJECTIONS IN A G R A M M A T I C PRODUCTION  I have focused on whether or not functional projections are present in the syntactic structure of agrammatic clauses. I have yet to address the issue of missing lexical projections. I now turn to two sets of data: missing lexical categories (specifically verbs and nouns) and missing arguments.  With cumulative arguments based on plausibility, I maintain that lexical  categories proj ect phrasal structure in agrammatic clauses.  3.3.1  Missing Lexical Categories  What occurs when lexical categories, i.e. nouns and verbs, are omitted? Do they have projections without phonetic content (eg. [  vp  e] and [  NP  e])? Or are they simply absent, not  projecting any syntactic structure in the syntax? Let us assume for the moment that Ns and Vs that are omitted are absent from the syntax. That being the case, these categories are not able to  113  function within the syntax (i.e. * [  v p  e] and * [  NP  e]) . Specifically they cannot be involved in  theta-assignment or Case-assignment . I will consider the two projections of lexical categories: 39  verb and noun.  3.3.1.1 Missing Verb  If V is absent from the syntax, theta-roles and (objective) Case cannot be assigned. We therefore do not expect to find utterances that have verbs missing but where the thematic role of the arguments are interpreted or where objective Case is assigned.  Consider the following  agrammatic clauses with missing verbs. For more examples of missing verbs in agrammatic clauses, see Appendix (18). (140) Agrammatic Clauses: Missing Verb a. the wife is [Vj: "Hurry up!" "the wife is saying: "Hurry up!"  English  b. de man en een vrouw [V] in de mand the man and a woman [V] in the basket "the man and a woman look in the basket"  Dutch  c. und der Wolf [V] inzwischen die Grossmutter ah im Bauch and the wolf [V] meanwhile the grandmother uh in+the belly "and in the meantime the wolf has the grandmother in its belly"  German  d. la grand-mere [V]: "Ouvre, la chevillettecherra" the grandmother [V]: "Open, the latch will lift" "the grandmother says: "Open, the latch will lift""  French  e. marito sta [V] [Det] sveglia husband is [V] [Det] alarm-clock "the husband is turning off the alarm clock"  Italian  In each of the above examples, the arguments and their relationships are clear given the context. If the arguments are receiving their theta-roles, then some form of the V must be present to assign them. Thus V and its projections, V and VP, are present.  The issue of agreement relationships was addressed in the preceding section. 114  Further evidence for the presence of V are replacement sequences, as described in section 3.2.1.1.1. In the second attempt I have underlined the word that is missing from the original attempt. For more examples of replacement sequences for verbs in agrammatic production, see Appendix (19). (141) Agrammatic Clauses: Replacement Sequences for Verbs a. no examples in database  English  b. no examples in database  Dutch  c. einmal [V] die Mutter... once [V] the mother... "once the mother wanted . . . "  German  c'. einmal wollte [DP] . . . once want:Pst [ D P ] . . . "once the motehr wanted . . . "  German  d. no example in database  French  e. [Det] marito sta [V] [Det] sveglia [Det] husband be:Prs-3Sg [V] [Det] alarm+clock "the husband putting the alarm clock"  Italian  e'. l'uomo sta posando no prendere (prendendo) la sveglia Italian the man be:Prs-3Sg putting no take:Inf (taking) the alarm clock "the man is putting the alarm clock" These replacement sequences are indirect evidence that the verb is present underlyingly. A V P must therefore also be projected.  3.3.1.2  Missing Noun  I now turn to missing nouns. If a noun is missing, and consequently not projected, then there is no argument to receive a theta-role. However, thematic roles are interpreted even when the noun is absent. These omissions show up as missing subject nouns (142) and missing object nouns (143). For more examples of missing subject nouns, see Appendix (20). examples of missing object nouns, see Appendix (21).  115  For more  (142) Agrammatic Clauses: Missing Subject Noun a. the [N] [V]: "[Det] better [to] eat you all up" "the wolf says: 'The better to eat you all up"" * the someone says: "The better to eat you all up"  English  b. en dan [Modal] de [N] opstaan and then [Modal] the [N] up-get:Inf "and then the man must get up" ^ the someone must get up  Dutch  c. so [Cop] die [N] so gut so [Cop] the [N] so good "thus the (parts of the) body is/are good" the something is good  German  (143) Agrammatic Clauses: Missing Object Noun a. the thief lifts his [N] "the thief lifts his body." ^ the thief lifts his something/someone  English  b. [DP] [Aux] de 't [N] vergeten [DP] [Aux] the:M/Ffhe:N [N] forgotten:Part "she had forgotten the gas " she had forgotten the something  Dutch  c. ein Einbrecher nutzt die [N] a thief uses the [N] "a thief uses the opportunity" * a thief uses the something/someone  German  d. puis les meubles modernes garnissent les [N] and the furniture modern decorate the [N] "and the modern furniture decorates the rooms" the modern furniture decorates the something  French  e. io sono andato [P+Det] portanino [P+Det] terza [N] sotto Italian I be:Prs-3Sg go:Part [P+Det]] chairman [P+Det] third [N] downstairs "I went with the orderly to the third floor downstairs" I went with the orderly the third something downstairs Each of the utterances in (142) and (143) is missing the N of the DP, leaving a stranded Determiner. These missing Ns receive specific interpretations. In other words they do not have an indefinite meaning of "something" or "someone". In the case of missing subjects, if there is no NP projection, the constructions are not licit.  116  If however, N is syntactically active but  phonologically null, V assigns its theta-role, DP receives a theta-role, and the specific interpretation is accounted for. In the case of missing objects, if no NP is present conceivably the verbs in question have an intransitive variant in the Agrammatic Grammar (AG). However, with an intransitive form of the verb, only an activity reading should result. As argued above, the interpretations are not general activity readings.  Further, if nouns are absent from the syntax, then A G contains  intransitive determiners that are homophonous with the regular determiners in both N G and in other utterances of A G . This is unlikely. Thus the missing noun must be syntactically active, with an NP projected in the syntax. Further evidence for the presence of N are replacement sequences, as described in section 3.2.1.1.1. I have underlined in the second attempt the word that is missing from the original attempt. For more examples of replacement sequences for nouns, see Appendix (22). (144) Agrammatic Clauses: Replacement Sequences for Nouns a. the thief lifts his [N] 'the thief lifts his body  English  a', the thief lifts [Det] body "the thief lifts his body"  English  b. en dan [Mod] de [N] opstaan and then [Mod] the [N] up+get "and then the man must get up"  Dutch  b'. de man begint met opstaan the man begin: Prs-3Sg with up+get "the man starts to get up"  Dutch  c. no examples in database  German  d. dans son jardi pour faire [N] in his garden to give [N] "in his garden to frighten"  French  d'. pour [V] peur aux moineaux to [V] fright to+the birds "to frighten the birds"  French  e. no examples in database  Italian  117  These repetition sequences are evidence that the noun is present underlyingly. A NP must therefore also be projected.  3.3.2 Missing Arguments  As we saw above Ns and Vs can be phonologically null.  So too can full subject and  object arguments. There are two issues that must be addressed with respect to null arguments: is an argument projected and if so, what is the nature of it? First, I establish the syntactic presence of null arguments by addressing the issues of theta-roles and subject-verb agreement. I then turn to the nature of null arguments: are these arguments instances of pro or are they non-pronominal DPs with no phonetic content (DP-0)? I address the licensing and identification conditions of these null arguments to see if they match those of pro. I conclude that null arguments are DP-0s rather than pro.  3.3.2.1  Missing Subjects and Objects  Do missing arguments  project syntactic structures in agrammatic utterances?  First,  consider missing subjects. If subjects are radically missing i.e. no syntactic projection, then the Extended Projection Principle (EPP) is violated. I have argued that the Agrammatic Grammar is governed by the principles of U G .  One of those principles is the EPP, which requires the  presence of a subject position (the subject need not be phonologically overt, as in the case of pro). A G must therefore project a subject position. If a noun is produced in argument position then a DP must be present in the syntax to receive the theta-role that a predicate assigns. But what if the whole D P is omitted? Is an argument projected? A s with the missing Ns above, these missing DPs are interpretable and receive appropriate theta-roles. If the argument is absent, and consequently not projected, then there would be no argument to receive a theta-role. Flow ever, thematic roles are interpreted even  118  when the argument is absent. These omissions show up as missing subjects (145) and missing objects (146). For more examples of missing subject arguments, see Appendix (23-29). For more examples of missing object arguments, see Appendix (30-33). (145) AgrammaticOmissions: Missing Subject DP a. [DP] pass (passed) out "I passed out"  English  b. en [DP] kom later and [DP] come:Prs-lSg later "and I will come later"  Dutch  c. Grossmutter warum hast [DP] [Det] [Adv] grossen Mund grandmother why have:Psr-2Sg [DP] [Det] [Adv] big mouth "grandmother, why do you have such a big mouth?"  German  d. [DP] [Aux] mis un epouvantail pour chasser les oiseaux [DP] [Aux] put:PastPart a scarecrow to chase the birds "he put in a scarecrow to chase the birds away"  French  Italian e. [DP] sta per scivolare sulla (dallo) sgabello [DP] is about slip:Inf on+the:F-Sg (from+the:M-Sg) stool:M-Sg "the boy is about to slip off the stool" (146) Agrammatic Omissions: Missing Object DP a. [Poss] wife, Rosa, take (took) [DP] [P] Loveladies "my wife Rosa took me to Loveladies"  English  b. midden in het bos pakte de heks nee [Det] wolf [DP] middle in the forest caught the witch no [Det] wolf [DP] "in the middle of the forest the witch, no the wolf caught her"  Dutch  c. die Frau weckt [DP] wieder the woman wakes [DP] again "the woman wakes him again"  German  d. ma femme [Aux] appele/er un ami et les pompiers my wife [Aux] call:Part/Inf a friend and the firemen  French  pour [CL] conduire a la clinique to [CL] drive: Inf to the clinic "my wife called a friend and the firemen to drive me to the clinic" e. un poliziotto [CL] aspetta a policeman [CL] waits "a policeman waits for him"  Italian  119  In each case V is present and the thematic roles it assigns to subject and object are interpretable, given the context. V must assign its theta-roles. Consequently, some argument must be present to receive these theta-roles. A possible alternative is that the theta-roles are somehow assigned internally. In other words, the theta-roles could be satisfied within the lexicon i.e. verbs are reclassified to optionally assign theta-roles.  In such a situation the  arguments would not be required to be projected in the syntax. However, since no DPs with a specified set of features is assigned the theta-roles, we expect the internalized arguments to receive a generic/non-specific interpretation. This is is not the case. Both subjects and objects receive specific interpretations. I begin with the objects in (146). For the English (146a) and French (146d) sentences, the speaker (lSg) is the object. For the Dutch (146b) clause, the object her indicates Little Red Riding Hood (LRRH). For the German 146(c) sentence, the object him refers to the man in the picture. And lastly, for the Italian utterance in (146e) the object him is interpreted as the burglar. With internalized objects we expect a generic reading of 'someone', rather than the specific readings described above. The only alternative is to have an argument in the syntax i.e. the predicates are not reclassified as intransitive. A similar argument holds for null subjects.  There are two possible approaches to  internalized subjects. The first parallels that of objects: the subjects are interpreted as an indefinite 'someone'. Clearly this is not the case, as can be seen by the specific interpretation of each of the subjects in (145): the English (145a) and Dutch (145b) subjects are both lSg; the German (145c) is 2Sg; the French (145d) is the farmer in the picture; and the Italian (145e) is the boy in the picture. There is a second possibility, however. Semantic principles independent of the normal grammar may allow subjects (and objects) to receive default interpretations: that of the speaker, lSg, or that of the topic, 3Sg. While semantic principles form part of the interpretation of arguments, I assume that these principles are governed by U G . The alternative is a set of semantic heuristics that are independent of U G .  There are three reasons to reject a U G -  independent semantic solution to null subjects.  First, this strategy once again involves  120  manipulating two sets of principles for what is a single structure: semantic principles and syntactic principles. In pro-drop languages such as Italian, subjects are standardly phonologically absent. The production/interpretation of these null subjects is done through syntactic means (namely the projection of 'little pro') . In other words, non-pathological speakers of languages 40  like Italian use syntactic principles to produce/interpret null arguments. These syntactic principles result from U G and are thus independently available to agrammatic speakers. If agrammatics are adopting a semantic set of heuristics when producing null arguments, they are in fact manipulating two sets of principles. They use the normal syntactic principles when producing overt arguments, and the semantic principles when producing null arguments. This account sets up a rather odd system. Normal speakers use a single (syntactic) system to produce overt and null arguments, whereas agrammatic speakers manipulate two systems to produce these same structures. As argued above, if one system can do the job, why posit the existence of a second? The second argument against a semantic account of null arguments lies in the fact that agrammatics do not simply omit l S g and 3Sg persons. Although l S g and 3Sg are the majority of null arguments, we see omission of a variety of person specifications. These are illustrated below. For more examples of missing lSg subjects, see Appendix (23); for missing 2Sg subject, Appendix (24); for 3Sg referential subjects, Appendix (25); for expletive subjects, Appendix (26-27); for 1P1 subjects, Appendix (28); and for 3P1 subjects, Appendix (29). (147) AgrammaticOmissions: Missing Subject - l S g a. [DP] pass (passed) out "I passed out"  English  b. maar [DP] [Aux] begonnen van [met] boodschappen doen but [DP] [Aux] start:PrsPart well [to] errands do:Inf "but I have started to do errands"  Dutch  c. [DP] war nicht mehr da [DP] be:Pst-lSg not more there "I wasn't there anymore"  German  1 assume a model of grammar that incorporates empty categories. The argument above would be different for models without empty categories, such as L F G , HPSG, categorial grammar.  40  121  d. no examples in database  French  e. structure not relevant since Italian is pro-drop  Italian  (148) AgrammaticOmissions: Missing Subject-2Sg a. no examples in database  English  b. enja [DP] moetpraten [P] [Det] maatschappij and yes [DP] mustPrs talk:Inf [P] [Det] society "you must talk in company"  Dutch  c. Grossmutter warum hast [DP] [Det] [Adv] grossen Mund grandmother why have:Psr-2Sg [DP] [Det] [Adv] big mouth "grandmother, why do you have such a big mouth?"  German  d. no examples in database  French  e. structure not relevant since Italian is pro-drop  Italian  (149) AgrammaticOmissions: Missing Subject- 3Sg Referential a. [DP] forgot the wash, the dishes "she forgot the wash, the dishes"  English  b. of [DP] [Aux] [Det] gas uit-gedaan or [DP] [Aux] [Det] gas off-turned:Part "or she turned off the gas"  Dutch  c. spater begutachtet [DP] seinen Kukuruz later examines [DP] his corn "later he examines his corn"  German  d. peu apres [DP] contemple [Det] recolte little after [DP] consider:Prs-3Sg [Det] harvest "a little later he considers the harvest"  French  e. [DP] trova un poliziotto e [DP] l'arresta [DP] finds a policeman and [DP] him arrests "the thief finds a policeman and the policeman arrests him"  Italian  41  (150) AgrammaticOmissions: Missing Subject - Expletive a. [Pleo] [Cop] a view [P] shrubs, trees [Conj] grass "there is a view of shrubs, trees and grass"  English  b. [Pleo] [Cop] duidelijktezien [Pleo] [Cop] clear to see:Inf "it is clear to see"  Dutch  Normally Italian does not require overt subjects, since it is pro-drop. However, overt subjects are required in this utterance to clarify the agents of each verb.  4 1  122  c. [Pleo] fehlt was [Pleo] miss:Prs-3Sg something "there is something missing"  German  d. [Pleo] [Cop] pour mieux t'entendre / ' (mon) enfant [Pleo] [Cop] to better you hear the.Sg (my:Sg) child "The better to hear you with, my child"  French  e. structure not relevant since Italian is pro-drop  Italian  (151) Agrammatic Omissions: Missing Subject - 1P1 English  a. no examples in database  b. [DP] [V] deze kamer en [Det] douche en [Adv] miks meer Dutch [DP] [V] this room and [Det] shower-bath and [Adv] nothing more "we have this room and a shower and then nothing more. c. no examples in database  German  d. no examples in database  French  e. structure not relevant since Italian is pro-drop  Italian  (152) Agrammatic Omissions: Missing Subject - 2P1 No examples found in the data. (153) Agrammatic Omissions: Missing Subject-3P1 a. [DP] spread [Det] cloth, napkin[s], dishfes], forks, spoons "they spread a cloth, riapkins, dishes, forks, spoons"  English  b. [DP] [Modal] thuis-kommen [DP] [Modal] home-come:Inf "they had come home"  Dutch  c. [DP] [Cop] normal (normaux) [DP] [Cop] normal:Sg (normal:PI) "they were normal"  French  d. no examples in database  French  e. structure not relevant since Italian is pro-drop  Italian  It is true that the majority of the subjects are l S g or 3Sg. However, I argue that this results from the production tasks.  Recall that the agrammatics' production tasks involve  conversing about their illness, picture descriptions and the telling of L R R H . The nature of the production tasks is such that the most frequently used subjects are l S g (recounting of illness) and  123  3Sg (picture description and story telling). Consequently the most frequent null subjects are l S g and 3Sg. Nevertheless, other person interpretations are attested. Subjects receive a variety of interpretations: lSg, 2Sg, 3Sg referential, 3Sg expletive, 1 PI and3Pl (2P1 being unattested in the data I have analyzed). A semantic account must be discarded. Further evidence for the presence of DP are replacement sequences, as described in section 3.2.1.1.1.  I have underlined in the second attempt the word that is missing from the  original attempt. For more replacement sequences for arguments, see Appendix (34). (154) Agrammatic Clauses: Replacement Sequences for Arguments a. [DP] shaved [DPI "the nurse shaved me"  English  a'. [Detl nurse shaved me "the nurse shaved me"  English  b. no examples in database  Dutch  c. no examples in database  German  d. y fait la sieste [P] rattraper [DPI perdu he take:Prs the nap [P] catch [DP] lost "he takes a nap the make up for wasted time  French  d'. le temps perdu de sommeil the time lost of sleep "the lost sleeping time"  French  e. no examples in database Italian These repetition sequences are indirect evidence that the argument is present underlyingly. A DP must therefore be projected. The third argument against a purely semantic account of null subjects involves subjectverb agreement, which I discuss in more detail below. If null subjects are absent from the syntax and receive their interpretation from a set of semantic heuristics, we do not expect them to have syntactic reflexes.  However, they trigger subject-verb agreement, a canonically syntactic  relation. A semantic solution must therefore be discarded.  124  3.3.2.2  Subject-Verb Agreement  Further evidence for the syntactic presence of a (phonologically null) argument comes from agreement markings on the verb. If omission entails the absence of projections, then when arguments are omitted, they should project no structure. This means that any other categories that are produced and that are dependent upon the null argument for their featural identity will suffer. Verbal agreement is a case in point. Verbal agreement is realized through a local relation with the subject argument, the Spec-head relation (Chomsky, 1993, 1995). This structural relation is schematized below. I adopt FP as the generic Functional Projection. (155) Structure for Subject-Verb Agreement FP DP " ^ ^ ^ ^  F  F  ) The subject DP is in [Spec FP], and the V is in the head position of FP.  Subject-verb  agreement is achieved through this Spec-head relation. If the DP is absent, then agreement is predicted to go awry. If the DP is present, agreement should be fine. While agreement does suffer from substitution errors, these errors are not the result of D P omission. D P omission is correlated neither negatively nor positively with the presence of agreement. First, agreement errors occur when DP is present, as illustrated in (156). This is the opposite of what is predicted. For more examples of agreement errors on verbs with overt subjects, see Appendix (35). (156) Agrammatic Errors on Verb with Overt Subject a. R R H says: "My, your ears is (are) larger" " R R H says: "My, you ears are larger"  English  b. van [Det] hond ruiken (ruikt) well [Det] dog smell:3Pl (smell:3Sg) "a/the dog smells something"  Dutch  125  c. (damit) [Det] Paar nicht gemerkt haben (hat) (so that) [Det] couple not notice:Part have:Pl (have:Sg) "so that the couple doesn't notice it"  German  d. un couple qui vont (va) manger sur l'herbe a couple that go:3Pl (go:3Sg) eatrlnf on the grass "it's a couple who are eating on the grass"  French  e. l'uomo no insomma eh uccido (uccide) la il lupo Italian the man no right hmm kilhlSg (kill:3Sg) the:F-Sg the:M-Sg wolf "the man kills the w o l f Second, absence of DP is not (necessarily) accompanied by substitution errors on the verb, as shown in (157). When the subject argument is missing, agreement on the verb is not random. Agrammatic speakers generally produce the correct agreement even if the Lexical Item is not phonologically present in the utterance. For more examples of agreement on verbs with null subjects, see Appendix (36). (157) Agrammatic Clauses: Agreement on Verb with Null Subject a. [DP] likes grandmother's voice "she likes grandmother's voice"  English  b. ja [DP] dacht van yes [DP] think:Pst-1 Sg well "I thought so"  Dutch  c. [DP] war nicht mehr da [DP] be:Pst-lSg not more there "I wasn't there anymore"  German  d. [DP] trova un poliziotto e[DP] 1 'arresta [DP] find:Prs-3Sg a policeman and [DP] him arrest:Prs-3Sg "the thief finds a policeman and the policeman arrests him"  Italian  42  Even if the subject is phonologically null, the appropriate projection must be available for the Spec-head agreement relation of subject and verb. Moreover, the agreement features that are assigned during the derivation must be present, even if the phonological matrix of the Lexical Item is not. Thus the grammatical structure must include the projections of the null subject.  * Italian is a pro-drop language which means that subjects need not be overt unless the discourse requires it. This example is one such case. 126  3.3.2.3  The Nature of Null Arguments  Some languages like Italian allow subjects to be phonologically null, where languages like English do not allow this option. Consider the following sentences. (158) a. lui ha telefonato he have:3Sg telephone:Part "he has telephoned" b. ha telefonato have:3Sg telephone: Part "he has telephoned" (159) a. he has telephoned b.*has telephoned Italian is referred to as pro-drop language, where the subject (pronoun) can drop. English is a non-pro-drop language. In order to interpret the phonologically null subject, it must be recoverable or identifiable (Taraldsen, 1978; Chomsky, 1981, 1982; Jaeggli, 1982; Huang, 1984, 1987, 1989). This distinction becomes relevant when we consider agrammatic sentences with null arguments. Are null subjects in agrammatic production pro, or are they non-pronominal DPs with no phonetic content, which I label DP-0? In other words, have non-pro-drop languages reverted to being pro-drop, or are null subjects full DPs that are missing their phonological specifications? The empirical distinction between the two may seem negligible, but they afford different predictions. If null arguments are in fact pro, they must be licensed and identified as such. If null arguments are full DPs, then their licensing conditions will match that of DPs. What licenses pro! And how is pro identified? There are two types of languages that license pro, languages with rich agreement (eg. Italian) and languages with no agreement (eg. Chinese).  I begin with the licensing and identification conditions of pro for Italian-type  languages and then address those for Chinese-type languages. If we compare Italian and English  127  once again, we notice a marked difference in the agreement morphology of the verb. English has an impoverished verbal paradigm whereas Italian verbal agreement is rich. The following two paradims for the verb eat illustrate my point. (160) English and Italian Paradigms for the Verb Eat English  Italian  I  eat  mangio  you  eat  mangi  she/he  eats  mangia  we  eat  mangi amo  you  eat  mangiate  they  eat  mangi ano  Whereas Italian has a different agreement morpheme for each person {-o for "I", - 0 for "you", -a for "she/he", and so forth), English only identifies 3Sg with -s. It therefore looks as though rich agreement systems license the use of pro, whereas impoverished systems do not. While this characterization may work for subject pro, it does not hold for object pro. Italian and English contrast: Italian has object pro, whereas English does not.  Again  Consider the  following sentences discussed in Rizzi (1986a). (161) a. questo conduce la gente a concludere quanto segue this leads the people to conclude what follows "this leads people to conclude what follows" b. questo conduce pro a concludere quanto segue this leads pro to conclude what follows "this leads (some arbitrary being) to conclude what follows" (162) a. this leads people to conclude what follows b.*this leads to conclude what follows Although Italian has pro, as in (161b), there is no object agreement on the verb that parallels the subject agreement discussed above. It cannot be the case that pro is licensed by rich  128  agreement. Moreover, the interpretation of the two pros is different.  Subject pro is specific  whereas objectpro is arbitrary. In (161b) there is no specific individual that is being referred to. Rather, an arbitrary 'someone' is the object. Perhaps objectpro is different from subject prol If we compare null objects in Chichewa to those of Italian, we see this is not so. following sentences from Baker (1988). SAgr indicates 'subject agreement'.  Consider the  OAgr indicates  ' obj ect agreement'. (163) a. mikango yanu i-na-thamangits-a mbuzi zathu lions your SAgr-Past-chase-Aspect goats our "your lions chased our goats" b. mikango yanu i-na-zi-thamangits-a mbuzi zathu lions your SAgr-Past-OAgr-chase-Aspect goats our "your lions chased our goats" c. mikango yanu i-na-zi-thamangits-a pro lions your SAgr-Past-OAgr-chase-Aspect "your lions chased them (our goats)" In (163a) no object agreement is marked on the verb, but the object DP mbuzi zathu "our goats" is overt. In (163b) both object agreement -zi- and the object DP mbuzi zathu "our goats" are present. In (163c) only object agreement -zi- is overt, with an object pro as object argument. In Chichewa, object pro (as in (163c)) receives a specific interpretation. This differs from the Italian objectpro that receives an arbitrary interpretation. But note the morphological difference: Chichewa has object agreement marked on the verb, Italian does not. Agreement morphology certainly plays a role with respect to pro.  However, we must conclude that agreement  morphology identifies pro, but does not license it. What then licenses prol  Rizzi (1986a)  separates licensing from identification and proposes the following. (164) Pro-drop Parameter pro is governed by a designated X°. (165) Identification Convention pro has the feature complex specified on X°; otherwise, it has arbitary features.  129  The class of heads that may license pro under government vary from language to language. In Italian, V and T license pro. Hence Italian has both subject and objectpro. Subject pro is identified in the following manner: it has the feature complex that is specified on the verbal head. This feature complex includes person and number specifications. Consequently, subject pro receives a specific interpretation. However, neither V nor T bear object agreement markings. Consequently, object pro must have arbitrary features and hence an arbitrary interpretation. In English, no head licenses pro , so neither subject nor object pro can appear. 43  If the agrammatic speakers of non-pro-drop languages have adopted a pro-drop strategy for null arguments, the licensing and identification conditions of these elements should be the same as those discussed above (or as those discussed for Chinese, below). If not, we either have anew type of pro with its own set of conditions or we are witnessing D P - 0 . Four of the five languages under study are non-pro-drop , Italian being the one pro-drop language. 44  If each  instance of null arguments in English, Dutch, German and French agrammatic utterances is in fact an instance of pro, we should find that subject pro is licensed by a governing Functional head and object pro is licensed by V . Further, if no indentification features on either of these licensers are present, the null arguments should receive an arbitrary interpretation. I begin with null objects. Each null object receives a specific interpretation, as illustrated in the following examples (recall that I also argued for specific interpretation of null objects above). (166) AgrammaticOmissions: Null Objects a. the wife was showing [DP] the clock "the wife was showing him the clock"  English  b. en [Det] hond [Modal] [DP] op-eten and [Det] dog [Modal] [DP] up-eat:Inf "and the dog has eaten the chicken up"  Dutch  This is not quite accurate. English does display pro in a very restricted environment: we find pro in impersonal constructions (Roberge &Vinet, 1989). Roberge & Vinet (1989) argue that French is a pro-drop language, with subject pro being identified by the subject clitic and objectpro identified by the object clitic. 4 4  130  c. und die Mutter konnte [DP] nicht zu [DP] begleiten and the mother could:3Sg [DP] not to [DP] accompany:PastPart "and the mother could not accompany L R R H to grandmother"  German  d. le chaperon rouge [CL] donne le panier avec les friandises the R R H [CL] gives the basket with the goodies " L R R H gives him (the wolf) the basket with the goodies"  French  If the null objects are in fact object pro, the only way they can receive a specific identification is if the licensing head bears the appropriate features. However, none of the four languages in question displays object agreement. This means that each instance of a null object should have an arbitrary interpretation. This prediction is not borne out.  Thus null objects  cannot be pro. A similar argument can be made for null subjects. Each null subject receives a specific interpretation, as illustrated in the following examples (recall that I also argued for specific interpretation of null subjects above). For more examples of null subjects, see Appendix (2329). (167) Agrammatic Omissions: Null Subjects a. [DP] pass (passed) out "I passed out"  English  b. zaterdag [Modal] [DP] slecht praten Saturday [Modal] [DP] badly talklnf "Saturday I could only talk badly"  Dutch  c. also [DP] [Modal] Buchstaben aus-lassen thus [DP] [Modal] letter:Plout-leave:Inf "so I leave out letters"  German  d. [DP] [Aux] mis un epouvantail pour chasser les oiseaux [DP] [Aux] put:PastPart a scarecrow to chase the birds "he put in a scarecrow to chase the birds away"  French  If the null subjects are in fact subject pro, the only way they can receive a specific identification is if the licensing head bears the appropriate features. question does indeed have subject-verb agreement.  Each of the languages in  Thus, they should allow a specific  interpretation of subject pro. However, as the above examples show, the agreement markings are 131  not always present. In (167), the clauses do not have any tense or agreement morphemes. With no agreement features present we expect the null subjects to receive an arbitrary interpretation. This prediction is not borne out. Thus null subjects cannot be pro. In summary, null arguments in agrammatic production cannot be instances of pro that are identified as inrich-agreementlanguages like Italian. I now consider the identification conditions of pro in Chinese-type languages. Unlike Italian, Chinese does not display rich agreement. It nevertheless licenses pro.  Huang (1989) argues that Chinese has subject pro but not object pro.  Both subjects and objects drop in Chinese, as illustrated in the following examples (all Chinese data is from Huang (1989)). Asp indicates 'aspect marker', Perf indicates 'perfective' and Q indicates'question morpheme'. (168) a. Zhangsan kanjian Lisi le ma? Zhangsan see Lisi Asp Q "Did Zhangsan see Lisi?" b. (ta) kanjian (ta) le(he) see (he) Perf "(he) saw (him)" c. wo xiang (ta) kanjian (ta) le I think (he) see (he) Perf "I think (he) saw (him)" In (168a) both arguments are overt.  In (168b-c) both subject and object may be  phonologically null. However, null subjects and null objects are not identified in the same way. For example, an embedded object cannot be A-bound by the matrix subject, whereas an embedded subject can. This is illustrated below. The null arguments are represented as je]. (169) a. Zhangsan shuo [ [e] hen xihuan Lisi] Zhangsan say [e] very like Lisi "Zhangsan said that [he] liked Lisi" b. Zhangsan shuo [ Lisi hen xihuan [e] ] Zhangsan say Lisi very like [e] "Zhangsan said that Lisi liked [him]"  132  In (169a) the null subject [e] may either refer to Zhangsan or to someone else who is the discourse topic. In (169b) the null object [e] can only refer to the discourse topic, and not to Zhangsan. Huang (1984) proposes that null objects in Chinese are variables that are A-bar bound by an empty operator (Op). Since variables must be A-free, they cannot be A-bound by the matrix subject Zhangsan. Instead they receive their interpretation via the operator Op from an antecedentin the discourse, specifically the discourse topic . Null subjects, on the other hand, 45  are identified via Huang' s (1989) Generalized Control Rule, presented below. (170) Generalized Control Rule (GCR) A n empty pronominal is controlled in its control domain (if it has one). Huang defines a control domain as the following (also see Manzini, 1983; Nishigauchi, 1984). (171) a is the control domain for (3 iff it is the minimal category that satisfies both (a) and (b): (a)  a is the lowest S or NP that contains (i) p\ or (ii) the minimal maximal category containing (3 (henceforth MMC(|3)).  (b)  a contains a SUBJECT accessible to p\  A ' S U B J E C T ' can either be an actual subject argument or (rich) Agr(eement). If pro has a control domain, then it must have a "local, unique, non-arbitrary antecedent" (Huang, 1989:194). This is the controller, which gives pro its reference. If pro has no control domain, then its antecedent may be in the discourse. There are two possible control domains: the lowest NP/S (or DP/TP) containing pro, or the lowest NP/S (or DP/TP) containing MMCipro). Combining variation in accessible SUBJECTS and variation in control domains leaves us with the following six scenarios.  For further evidence see Huang (1984). Similar accounts have been proposed for other languages. See Hasegawa (1984/85) for Japanese, Huang (1984) for German, Raposo (1986) for Portuguese, Lillo-Martin (1986) for America Sign Language, and Authier (1988) for KiNande. 4 5  133  (172) Possible Environments for Subject pro a. b. c. d. e. f.  pro pro pro pro pro pro  is matrix subject of a tensed clause with rich Agreement is matrix subject of a tensed clause with poor Agreement is matrix subject of an untensed clause is embedded subject of a tensed clause with rich Agreement is embedded subject of a tensed clause with poor Agreement is embedded subject of an untensed clause  Let us consider each of the possibilities in (172). If pro is the matrix subject of a clause with Agr rich enough to control it (as in (172a)), its control domain is the matrix clause, with Agr as the controller. If however, pro is the matrix subject of a clause with poor Agr (as in (172b)), it does not have a control domain and may receive its interpretation from an antecedent in the discourse. The same holds for pro as the matrix subject of an untensed clause (as in (172c)): pro receives its interpretation from an antecedent in the discourse. If pro is an embedded subject, there are three possible situations. If the embedded clause is tensed and has rich Agr (as in (172d)), then the embedded clause itself is the control domain, with Agr being the controller (this parallels the situation in (172a)). If the embedded clause is tensed but has poor Agr (as in (172e)), then the matrix clause is the control domain, with the matrix subject as the controller. Lastly, if the embedded clause is untensed (as in (172f)), the matrix clause is the control domain, once again with the matrix subject acting as controller. I argued above that null arguments in agrammatic production are not identified like pro in Italian. I now consider if they are identified like pro in Chinese: do agrammatic null arguments follow the Generalized Control Rule? To begin, the only language under study that has rich agreement is Italian. Of the five languages, Italian is the only pro-drop language. I focus on the four non-pro-drop languages for the moment. These all have poor Agr. The inventory of environments for pro, as presented in (172), is therefore reduced: (172a) and (172d) are excluded, since they involve clauses with rich Agr. Further, with the exception of one example from German, there are no instances of null arguments in embedded clauses. I therefore set aside (172e) and (172f). The environments that remain involve a null argument in tensed clauses with poor Agr or a null argument in untensed clauses. The result is the same: if the null argument is 134  pro, it must receive its reference from an antecedent in the preceding discourse. Specifically, pro must receive its reference from the preceding discourse topic. If agrammatic null arguments are indeed pro, we do not expect to find them introducing new topics. However, this is exactly the case, as illustrated below. More examples are presented in Appendix (37). (173) Agrammatic Clauses: Null Arguments with no Antencedents in the Discourse and/or That Introduces New Discourse Topic a. [DP] shaved [DP] "the nurse shaved me"  English  Context: The preceding discourse topic revolves around patients in general and the speaker's position (flat on his back). b. [DPI [V]: "Ga maar maar ik kom later" [DP] [V]: "go:Imp just but I come:Prs-lSg later" "I said: "You can go but I will come later on""  Dutch  Context: The preceding discourse topic revolves around the activity of the speaker's wife and children (and St. Nicholas). c. einmal wollte [DP] [Comp] [DP] der Grossmutter rDP] [V] once want:Pst [DP] [Comp] [DP] the grandmother [DP] [V]  German  weil die Grossmutter krank gewesen ist because the grandmother sick be: Part be:Prs-3Sg "once the mother wanted L R R H to take a basket to grandmother because grandmother was sick" Context: The preceding discourse topic was L R R H and why her name was L Red R H . The was no previous mention of a basket. d. et apres [DP] fini and after [DP] ended "and once the movie was over"  French  Context: The preceding discourse sets up the activity (watching television) but does not mention movie explicitly. Even Italian, a pro-drop language, displays 'missing' arguments, as shown in (174).  135  (174) Agrammatic Italian: Null Arguments That Introduce New Discourse Topic a. [DP] sta per scivolare [DP] be:Prs-3Sg about slip:Inf "the boy is about to slip"  Italian  Context: The preceding discourse topic was the girl eating cookies. b. [DP] cresce [DP] grow:Prs-3Sg "the corn grows"  Italian  Context: The preceding discourse topic was the farmer waiting. c. [DP] guarda che guarda il punto di maturazione [DP] watch:Prs-3Sg that watch:Prs-3Sg the point of maturation "the farmer watches for the stage of ripeness"  Italian  Context: The preceding discourse topic was the corn. d. [DP] trova un poliziotto e [DP] l'arresta [DP] find:Prs-3Sg a policeman and [DP] him arrest:Prs-3Sg "the thief finds a policeman and the policeman arrests him"  Italian  Context: The discourse topic shifts so the subjects must be expressed.unclear e. [DP] corre [DP] run:Prs-3Sg "the man runs"  Italian  Context: The preceding discourse topic involves the wife's activities. focus shift and requires an overt subject. In each case, the subject should be overt since it introduces a new topic.  This is a  The null  arguments in (173) and (174) do not receive their interpretation from the preceding discourse topics. They do not pattern with Chinese-type pro. I conclude therefore that agrammatic null arguments are not pro.  Rather, agrammatic null arguments are full DPs that are lacking their  phonological features: D P ^ 0 . I now return to a question left unresolved. relationship between the Agrammatic Grammar and U G .  The first issue I addressed was the I concluded that A G is constrained by  U G , leaving undecided the issue of whether A G is identical to the Normal Grammar or different from it. I argue that A G and N G are identical based on missing arguments.  136  3.4  T H E DISTRIBUTION OF OMISSIONS  I will now discuss the rate and distribution of omissions. Although any syntactic category may be omitted, not all categories are omitted with the same frequency. What determines rate of omission?  First I must distinguish between absolute rate of omission and relative rate of  omission . I will establish that two factors affect these rates: severity of deficit and categorial 46  identity. Severity of deficit influences the absolute rate of omission: the more severe the deficit, the higher the rates of omission. Categorial identity influences the relative rates of omissions. In other words, the identity of the syntactic category, whether N or V or D and so on, has consequences for omission. I develop a Principle of Robustness that accounts for the relative rates of omission.  3.4.1  Are Rates of Omission a Measure of Severity?  I begin with the role that the severity of the deficit plays in determining rates of omission. Severity of deficit can be estimated by Mean Length of Utterance (MLU) and Mean Rate of Production (MRP).  M L U is measured in words/phrase; M R P is measured in words/minute.  Lower M L U and M R P indicates fewer words per phrase and fewer words per minute, respectively. Hence lower M L U and M R P indicate a more severe impairment. Consider the following measure of severity for the two English-speaking agrammatics. The following table indicates the M L U and M R P for each agrammatic speaker and for their respective controls (each agrammatic requires a control matched for age, handedness, education, etc.).  Absolute rates of omission are the rates of omission of a specific category (across the five languages and/or across the various agrammatic speakers). Relative rates of omission are the rates of omission of the various categories within a given language and/or speaker.  4  137  Table 4: Agrammatic English Mean Length of Utterance and Mean Rate of Production EF  Control  EE  Control  M L U (word/phrase)  2.8  10.0  2.3  15.0  M R P (words/min)  18.8  176.4  14.4  151.2  First, each measurement ( M L U and MRP) is dramatically lower for the agrammatic as compared to the control. This is to be expected considering that agrammatics have a language deficit, whereas the controls do not.  This difference is observable for each of the languages  under study. The range for M L U of the agrammatic speakers is 2.3 to 6.0. The range for M L U of the normal speakers is 7.8 to 15.0. The range for M R P of the agrammatic speakers is 10.6 to 70.2. The range for M R P of normal speakers is 110.9 to 176.4. These ranges are displayed in the table and figures below. Table 5: Mean Length of Utterance (MLU) and Mean Rate of Production (MRP) - Ranges for Agrammatics and Controls Agrammatic  Control  M L U (word/phrase)  2.3 - 6.0  7.8 - 15.0  M R P (words/min)  10.6 - 70.2  110.9- 176.4  138  Figure 3: Mean Length of Utterance (MLU) Ranges for Agrammatics (A) and Controls (C) 15 14 13 12 11 10 9 8 7 6 5 43 2 1 0-, n  n  Figure 4: Mean Rate of Production (MRP) Ranges for Agrammatics (A) and Controls (C) 180 170 160 150^ 140130 120 110 100 9080 70 60^ 504030 H 20^ 1 0^  1 0  A  C  Second, there is a distinction between the two English agrammatics: EE's deficit is more severe than EF's. Consider now their rates of omissions. The table (and all subsequent tables) 139  must be read as follows. The EF's omission figure for N is 1/92 - 1.1%.  "92" indicates the  number of obligatory contexts for N . "1" indicates the number of omissions.  And "1.1%"  indicates the percentage of omission for N . Table 6: Agrammatic English Rates of Omission EE  EF Tokens  Context  %  Tokens  Context  %  N  2  92  2.2%  3  139  2.2%  V  3  69  4.3%  6  51  11.8%  D  7  99  7.0%  58  105  55.2%  Aux/Modal  8  55  14.6%  21  23  91.3%  Average omission rate Severity of deficit  40.1%  7.0%  deficit E F < deficit E E  EE's omission rates are higher than EF's. This is true for almost syntactic category: V 11.8% versus 4.3%, D - 55.2% versus 7.0%, and Aux/Modal - 91.3% versus 14.6%. This is reflected in the average omission rates: for E E it is 40.1%, and for E F it is 7.0%. From this we can conclude that severity of deficit (as established above and in chapter 2) influences rate of omission: the more severe the deficit, the higher the rates of omission. However, equating severity of deficit with a specific average omission rate is not possible. First, as discussed in chapter 2, individuals adopt varying compensatory strategies. These diverse strategies influence rates of omission both across languages and within a given language. Second, other factors, such as agreement and Case features, influence omission rates. Since languages encode these features differently, rates of omission vary accordingly. Thus, the  140  severity-omission match is not stable across languages.  What we must therefore focus on is  inter-language variation. Although this severity-omission correlation does not consistently carry through each of the languages under study, the pattern holds generally. The following table displays the Mean Length of Utterance, the Mean Rate of Production and the average percentage of omission. The first initial for each of the agrammatics indicates the speaker's language: E = English, D = Dutch, G - German, F = French, and I - Italian. Table 7: Summary of Agrammatic M L U , M R P and Average Omission MLU  MRP  Omission  EF  2.8  18.8  7.0%  EE  2.3  14.4  40.1%  DB  2.8  41.1  47.0%  DH  2.8  38.6  38.8%  GB  3.8  28.3  9.3%  GM  3.9  31.6  17.2%  FC  6.0  43.0  19.1%  FA  5.8  70.2  4.1%  IR  2.4  19.3  23.7%  TV  3.5  31.2  9.4%  The English agrammatics follow the severity-omission correlation. The same holds for the French and Italian agrammatics. In each of these three languages, the higher the severity of  141  the deficit (as measured by shorter M L U and slower MRP), the higher the average omission rate The French rates of omission are presented in Table 8, the Italian in Table 9. Table 8: Agrammatic French Rates of Omission FA  FC Tokens  Context  %  Tokens  Context  %  N  5  196  2.6%  1  104  1.0%  V  7  131  5.3%  1  96  1.0%  D  52  284  18.3%  9  217  4.2%  Aux/Modal  16  32  50.0%  6  60  10.0%  Average omission rate Severity of deficit  19.1%  4.1% deficit F C > deficit F A  142  Table 9: Agrammatic Italian Rates of Omission IR  TV  Tokens  Context  %  Tokens  Context  %  N  7  93  7.5%  4  184  2.2%  V  20  101  19.8%  4  96  4.2%  D  44  175  25.1%  15  105  14.3%  Aux/Modal  25  59  42.4%  5  30  16.7%  Average omission rate Severity of deficit  9.4%  23.7% deficit IR > deficit IV  This pattern holds for each syntactic category. For French, F C is more severely impaired than F A , based on M L U and M R P . The omission rates reflect this: N - 2.6% versus 1.0%, V 5.3% versus 1.0%, D - 18.3% versus 4.2%, and Aux/Modal - 50.0% versus 10.0%.  This is  further reflected in the average omission rates: for F C it is 19.1%, and for F A it is 4.1%. For Italian, IR is more severely impaired than IV. The omission rates reflect this: N - 7.5% versus 2.2%, V - 19.8% versus 4.2%, D - 25.1% versus 14.3%, and Aux/Modal - 42.4% versus 16.7%. This is further reflected in the average omission rates: for IR it is 23.7%, and for IV it is 9.4%. The higher the severity of deficit, the higher the omission rates. For the Dutch and German agrammatics, the average omission rates appear to indicate the opposite, namely that their average omission rate is higher for the less severely impaired agrammatic. Let us look at the specific omission rates for these two languages. The Dutch rates of omission are present in Table 10, the German rates of omission in Table 11.  143  Table 10: Agrammatic Dutch Rates of Omission DH  DB Tokens  Context  %  Tokens  Context  %  N  2  104  1.9%  4  148  2.7%  V  16  91  17.6%  31  81  38.3%  D  132  162  81.5%  85  154  55.2%  Aux/Modal  66  76  86.8%  39  66  59.1%  47.0%  Average omission rate Severity of deficit  38.8%  deficit D B < deficit D H  Table 11: Agrammatic German Rates of Omission GB  GM  Tokens  Context  %  Tokens  Context  %  N  2  170  1.2%  4  138  2.9%  V  6  103  5.8%  8  97  8.3%  D  39  211  18.5%  30  194  15.5%  Aux/Modal  5  43  11.6%  16  38  42.1%  Average omission rate Severity of deficit  9.3%  17.2% deficit G B > deficit G M  144  For Dutch, based on M L U and M R P , D H is more severely impaired than DB. However, DH's omission rates are not consistently higher than DB's. This pattern only holds for N and V: N - 2.7% versus 1.9% and V - 38.3% versus 17.6%. The categories D and Aux/Non-lex show the opposite pattern: D - 55.2% versus 81.5% and Aux/Modal - 59.1% versus 86.8%. Notice ;  that (i) DH's and D B ' s M L U ' s are identical (2.8), and (ii) their MRP's are close (41.1 versus 38.6, respectively). It is therefore possible that their deficits are of approximately equal severity and that the unexpected rates of omissions are due to differing compensatory mechanisms. For German, again based on M L U and LRP, G B is more severely impaired than G M . However, the only category that follows the established pattern is D: 18.5% versus 15.5%. Each of the other categories shows the opposite pattern: N - 1.2% versus 2.9%, V - 5.8% versus 8.3%, and Aux/Modal - 11.6% versus 42.1%. This is further reflected in the average omission rates: for GB it is 9.3%, and for G M it is 17.2%. As with the Dutch agrammatics, G B and G M have similar M L U s (3.8 and 3.9, respectively) and MRPs (28.3 and 31.6, respectively). Moreover, the only truly dramatic distinction in rates of omission is for Aux/Modal (albeit in the unpredicted direction). Once again, the unexpected rates of omissions could be due to disparate compensatory mechanisms. I have shown that for three of the five languages the correlation is the following: the more severe the deficit, as measured by M L U and M R P , the higher the rates of omission for each syntactic category under study. The omission-deficit correlation is indirect rather than direct. Although the pattern is not absolute, other factors impinge on omission rates, thus altering the pattern. These factors include (i) compensatory mechanisms, (ii) language-specific constraints on omission, and (iii) consistency in testing conditions (see section 2.2.1 for a discussion on factors affecting variability).  3.4.2  Which Syntactic Categories Get Omitted?  A l l syntactic categories get omitted.  What is of interest, however, is that syntactic  145  categories do not all have the same rates of omission. I showed above that a specific level of deficit (if indeed that can be accurately determined) cannot be associated with an absolute rate of omission. This is also true for individual syntactic categories. Any given,category cannot be associated with a specific rate of omission. In other words, the category D cannot be equated with a specific omission rate either across languages or across speakers within a language. Several factors affect the omission rate of a given category: severity of deficit (as discussed above), compensatory mechanisms (as discussed in chapter 2), and encoding of agreement and Case features (as discussed in chapter 4). The severity of deficit and compensatory mechanisms alter the omission rate of a category both across languages and within a language.  Feature  encoding affects the omission rate of a category across languages. We must therefore focus on relative rates of omission. The questions I address in this section are the following. Is there a general hierarchy of omission (or retention)? Does this hierarchy hold across speakers of a given language? Does it hold across languages? What factors determine or influence the hierarchy?  3.4.2.1  The General Retention Flierarchy  Before delving into specific distinctions, I briefly discuss the relative rates of omission in general terms. Four syntactic categories are included in the omission charts above: N , V , D , Aux/Modal. No absolute omission rate can be associated with any given category. Moreover, factors like compensatory mechanisms can affect omission rates such that they go opposite what the severity of the deficit would predict. But what about the omission rate of one category with respect to the omission rate of another for the same agrammatic speaker? In other words, do Ds have a consistently higher/lower rate of omission than say Vs? Compiling the omission rates from all the above tables results in the following Retention Hierarchy, where ">" means "better retained than". (175) Retention Hierarchy N > V > D > Aux/Modal 146  This Retention Hierarchy holds for each individual under study, with the one exception of G B . G B ' s retention is as follows. (176) G B ' s Retention Hierarchy N > V > Aux/Modal> D . The only differnce between G B ' s retention hierarchy and the general Retention Hierarchy is that GB retains fewer Ds than Aux/Modals. G B maintains the N/V distinction, where N is better retained than V . G B also mainatins the N-V/D-Aux(Modal) distinction: both N and V are better retained than either D or Aux/Modal. I now turn to specific oppositions within the hierarchy.  3.4.2.2  Lexical versus Functional Categories  Agrammatic speech has traditionally been characterized as omitting closed class items and grammatical morphemes, which make up the syntactic set called functional categories. In this section, I discuss agrammatic production of functional categories (F-categories) and lexical categories (L-categories). Let us consider the agrammatic production of these categories.  The table shows the  omission rates of all the agrammatic speakers. For each individual, F-categories have a higher rate of omission than L-categories.  147  Table 12: Agrammatic Omission Rates of Functional versus Lexical Categories Functional categories  Lexical categories  Tokens  Context  %  Tokens.  Context  %  EF  28  260  10.8%  4  178  2.3%  EE  133  231  57.6%  9  201  4.4%  DB  246  421  58.4%  69  338  20.4%  DH  156  411  38.0%  87  477  18.2%  GB  44  356  12.4%  25  284  8.8%  GM  54  385  14.0%  19  329  5.8%  FC  86  445  19.3%  25  369  5.8%  FA  14  343  4.1%  4  273  1.5%  IR  81  301  26.9%  44  246  17.9%  IV  19  183  10.4%  10  317  3.2%  The range of omission rates is quite broad (see section 2.2.1 for discussion of variability). The range of omission rates for F-categories is 4.1% to 58.4%.  The range of  omission rates for L-categories is 1.5% to 20.4%. These are schematized in figure 5 below.  148  Figure 5: Range of Rates of Omission (in %) for Functional and Lexical Categories 100 + 90—j— 80 + 70 + 60—j—50—{— 40 + 30 + 20 + 10 + 5 + 0 + functional categories  lexical categories  As is reflected in the bar graph, F-categories have a higher rate of omission than L categories. What also becomes apparent is that F-categories have a much broader range of omission rates than L-categories. Any model of syntactic categories must be consistent with these facts.  In order to account for the agrammatic treatment of F-categories, a syntactic  distinction between L-categories and F-categories is required. I argue below that the agreement, Case and tense features that a category encodes affects the rates of omissions (and substitution; see chapter 4). Since functional categories encode more of these features, it comes as no surprise that their omission rates vary more than those of lexical categories.  3.4.2.3  Nominal versus Verbal Categories  Several studies have shown that agrammatics have more difficulties with Verbal categories than with Nominal categories (for English, Jakobson, 1964; for Dutch, Kolk et al., 1982; for Italian, Miceliet al., 1983; Zingeser & Berndt, 1990). In this section I investigate the division between Nominal categories (N-categories) and Verbal categories (V-categories). Let us consider the agrammatic production of these categories. The table in (166) shows the omission rates of all the agrammatic speakers. For each individual, N-categories are better retained than Vcategories, (i.e. omission rates are higher with verbs than with nouns) the exception being G B , 149  where V-category omission is 8.8% and N-category omission is 10.8% (also notice that GB is the only agrammatic whose omission rates do not follow the retention hierarchy). Table 13: Agrammatic Omission Rates of Verbal and Nominal Categories Verbal categories  Nominal categories  Tokens  Context  %  Tokens  Context  %  EF  13  151  8.6%  8  191  4.2%  EE  35  117  29.9%  61  244  25.0%  DB  112  216  51.9%  134  266  50.4%  DH  11  204  54.4%  89  302  29.5%  GB  19  215  8.8%  41  381  10.8%  GM  34  211  16.1%  33  332  9.9%  FC  29  206  14.1%  57  480  11.9%  FA  7  200  3.5%  10  321  3.1%  IR  62  233  26.6%  51  268  19.0%  TV  14  199  7.0%  19  289  6.6%  The range of omission rates is quite broad (see section 2.2.1 for discussion of variability). The range of omission rates for verbal categories is 3.5% to 54.4%. The range of omission rates for nominal categories is 3.1% to 29.5%. below.  150  These are schematized in figure 6  Figure 6: Range of Rates of Omission (in %) Verbal and Nominal Categories 10090^ 80 - | 70 - \ 60-] 40-| 3020 - | 10—1  nominal categories  verbal categories  As is reflected in the bar graph, verbal categories (tend to) have a higher rate of omission than nominal categories. Verbal categories also have a much broader range of omission rates than nominal categories. This parallels the F-category/L-category distinction. Since verbs are more like functional categories than nouns, they are expected to behave more like F-categories than nouns. Any model of syntactic categories must be consistent with these facts.  In order to  account for the agrammatic treatment of verbal categories, a syntactic distinction between verbal categories and nominal categories is required. Once again, agreement, Case and tense features affect the rates of omission. Verbal categories can encode more of these features. They are therefore expected to display a wider range of omission rates than nominal categories.  3.5  CONCLUSION  In this chapter I addressed the issue of omission errors in agrammatic production, focusing on the nature and distribution of omissions. In order to determine what mechanisms are impaired I established what structures are intact.  I argued against the Lexical Projection  Hypothesis, the Truncation Hypothesis and the Optional Tense Hypothesis, and argued for the presence of (at least) CP, T P and AgrOP. Moreover, I proposed that agrammatic clauses require the complete array of syntactic projections.  This array includes not only verbal functional 151  projections (such as CP and TP), but also nominal functional projections (like DP) and lexical projections (NP and V P in particular). I then discussed the rate and distribution of omissions. Although any syntactic category may be omitted, not all categories are omitted with the same frequency.  I showed that the more  severe the deficit (as indicated by M L U and MRP), the higher the overall omission rates. Furthermore, I demonstrated that while absolute rates of omissions do vary from agrammatic to agrammatic, the relative rates of omissions are constant across agrammatics and across languages. The hierarchy is as follows: N > V > D > Aux/Modal, where ">" means "better retained or more often produced than". A more detailed survey of the various categories then followed.  I concluded that two  distinctions are relevant to agrammatism: the lexical-functional and nominal-verbal distinctions. Lexical categories are produced more frequently than functional categories, and nominal categories are produced more frequently than verbal categories.  152  Chapter 4 A G R A M M A T I S M : SUBSTITUTION ERRORS  Agrammatic speech is characterized by omissions and substitutions.  I addressed  omission errors in the preceding chapter. I now turn to substitution errors. The utterances in (1) illustrate substitution errors in agrammatic speech. Italics indicate the item that is produced; the content of the following rounded brackets '()' represents the speaker's target. (1)  a. My, your ears is (are) larger "My, your ears are larger"  English  b. de man and een vrouw pakt (pakken) nee de hond pakt the kip Dutch the man and a woman take:3Sg (take:3Pl) no the dog take:3Sg the chicken "the man and a woman take, no the dog takes the chicken" c. weil er nicht noch nicht aus schlafen (geschlafen) bin ist German because he not yet not Particle sleep.Tnf (sleep:Part) be:Prs-lSg be:3Sg "because he has not had enough sleep" d. le couple y sont (est) arrive pour manger the couple they be:Prs-Pl (be:Sg) arrive:Part to eat:Inf "the couple arrive to eat" e. i genitori stava (stavano) in pensiero the:M-Pl parents was:3Sg (was:3Pl) in thought "the parents were thinking"  French  Italian  Agrammatic utterances with substitutions are different from those with omissions in one obvious way: there are no missing words, as shown in (1). There are utterances that include both omissions and substitutions. However, what is relevant here is that the two need not cooccur. The only deviation from non-pathological speech is the production of an inappropriate lexical item. Otherwise, the utterance is normal: it displays appropriate word order, and the full array of structures is projected. What then is the nature of substitution? In what way is the lexical item 'inappropriate'? First, I argue that substitutions are not completely random. Substitutions are of the same syntactic category. A noun substitutes for a noun, a verb for a verb, and so on. Second, I argue that substitutions are further constrained. There are three types of substitutions: phonological, semantic and syntactic. Phonological substitutions involve 153  mis-selecting a word with the same initial segment: pick for picnic , for example. Semantic substitutions involve mis-selecting a word along semantic lines: wife for man, trip for fall, and the like. And syntactic substitutions involve mis-selecting a word with different agreement (person, number and gender), Case or tense features: say, dogs for dog. I then argue that syntactic substitutions are even more restricted. A given substitution error may only involve a single feature (the Single Feature Constraint). Having established the nature of substitution errors, I investigate their distribution. What are the most frequent feature shifts? What direction do shifts follow? In some cases, feature substitutions are bi-directional. When we compare the feature systems encoded in the different syntactic categories, we notice that agrammatics of differing languages do not treat a given category in the same way. For instance, determiners are frequently omitted in English and Dutch, whereas determiners suffer from substitution rather than omission in German. Further, the number of agreement and Case features encoded in a given category influences the substitution rate.  4.1  T H E N A T U R E OF SUBSTITUTION  In this section, I establish that substitutions are restricted to the same syntactic category. I then show that the features involved in substitution are agreement, Case and tense features. Next, I argue that these features are not randomly manipulated. Rather, for any given substitution a single feature is changed.  4.1.1 Substitution Is Not Chaotic  What is substitution? The first possible explanation is that substitution errors are completely random, with no constraints governing them. If substitutions are random, then any word can substitute for any other word. For example, the singular noun cat could substitute for  154  the 3Sg verb eats in the target utterance the dog eats the chicken. The resulting utterance would then be: the dog cat the chicken. However, this cross-category substitution is unattested. In agrammatism we do not find substitution errors that replace one syntactic category with another (Garrett, 1988). This parallels the categorial restriction for substitution errors in normals (Nooteboom, 1969; Fromkin, 1971; Stemberger, 1985). This is illustrated below. First, verbs substitute for verbs. (2) Agrammatic Substitution: Verb for Verb a. the boy give (gives) to the girl a cookie "the boy gives to a/the girl a cookie"  English  b. maar verder [DP] kom (komt) vanself but further [DP] come:lSg (come:3Sg) by-itself "but further such a thing comes by itself  Dutch  c. im Biiro hat [DP] nicht gearbeit[et] in+the office has [DP] not work: Part  German  und (sondern) schlafen (geschlafen) and (rather) sleep:Inf (sleep:Part) "in the office he didn't work, rather he sleeps" d. no examples in database  French  e. [Det] due sposini prepara (preparano) da mangiare in terra Italian [Det] two newlyweds prepare:3Sg (prepare:3Pl) to eat:Inf in ground "the two newlyweds get ready to eat on the ground" In the English example, the bare verb give substitutes for the verb gives. In Dutch kom "come:Prs-lSg" substitutes for komt "come:Prs-3Sg". The German example involves the verb sleep: the infinitive schlafen substitutes for the participial geschlafen. The Italian substitution involves prepara for preparano, both being instances of the verb "prepare". Modals and auxiliaries substitute for modals and auxiliaries, respectfully. This is illustrated in (3). (3) Agrammatic Substitution: Modal/Auxiliary for Modal/Auxiliary a. I can't (couldn't) speak "I couldn't speak"  English  155  b. [DP] moet (moest) [DP] uitsprechen [DP] must:Prs (must:Pst) [DP] pronounce:Inf "I must pronounce everything"  Dutch  c. damit... [Det] Paar [DP] nicht gemerkt haben (hat) so that... [Det] couple [DP] not notice:Part have:Pl (have:Sg) "so that the couple doesn't notice it"  German  d. [Pleo] [Cop] un couple qui vont (va) manger sur l'herbe French [Pleo] [Cop] a couple that go.-3PI (go:3Sg) eat:Inf on the grass "it's a couple who are eating on the grass" e. la bambina sono e andata a ... the:F-Sg girl be:3Pl/lSg be:3Sg go:Part to ... "the girl went to..."  Italian  In the English example, the modal can substitutes for the modal could. In Dutch moet "must:Prs" substitutes for moest "must:Pst". The German example involves the auxiliary have: the plural form haben substitutes for the singular form hat. In French the plural vont replaces the singular va. The Italian substitution involves sono for e, both being instances of the auxiliary "be". Next, nouns substitute for nouns, as shown below. (4) Agrammatic Substitution: Noun for Noun a. the men (man) was eating "the man was eatins"  English  b. [V] [DP] twee kamer (kamers) [V] [DP] two room.Sg (room:Pl) "we have two rooms"  Dutch  c. [DP] ist in, ist in Gedanke (Gedanken) wo gewesen [DP] is in, is in thought:Sg (fhought:Pl) somewhere be:Part "she is somewhere else in her thoughts"  German  d. no examples in database  French  e. i l contadino prepara [Det] seme (semi) the farmer prepares [Det] seed:Sg (seed:Pl) "the farmer prepares the seeds"  Italian  In the English example, the noun men substitutes for the noun man. In Dutch kamer 'room" substitutes for kamers "rooms". The German example involves the noun "thought": the 156  singular form Gedanke substitutes for the plural form Gedanken. The Italian substitution involves seme for semi, both being instances of the noun "seed". Next, determiners substitute for determiners, as shown in (5). (5) Agrammatic Substitution: Determiner for Determiner a. I [Cop] flat on the (my) back "I was flat on my back"  English  b. de dief kwam binnen door de (het) raam the thief come:Past in through the:Sg-M/F (the:N) window "the thief came in through the window"  Dutch  c. [DP] hatte einem (einen) Korb gepackt [DP] have:Pst-3Sg a:Dat-M-Sg (a:Acc) basket pack:Part "she had packed a basket"  German  d. je marchais avec le (la) grande canne anglaise I walked with the:M (the:F) long stick English "I was walking with the long English walking stick"  French  e. il contadino mangia i (il) granone the farmer eats the:M-Pl (the:M-Sg) grain:Sg "the farmer eats the grain"  Italian  In the English example, the determiner the substitutes for the determiner my. In Dutch de "the:Sg-M/F" substitutes for het "the:Sg-N". The German example involves the indefinite article "a": the dative form einem substitutes for the accusative form einen. In French the masculine definite article le replaces the feminine la. The Italian substitution involves the plural i for the sigular li, both being instances of the determiner "the". The descriptive generalization is the following: mis-selected words are of the same syntactic category. Is it possible that this restriction is semantic rather than syntactic? This category constraint cannot be a semantic constraint, since mapping between semantic function and syntactic category is not consistent (Maratsos & Chalkley, 1980). For example, verbs (6a), adjectives (6b) and nouns (6c) can behave like semantic predicates, as shown below.  157  (6) Possible Predicates a. Carmina danced. b. Carmina is pretty. c. Carmina is a woman. If substitutions were simply semantically constrained, we would expect to find verbs, adjectives and nouns substituting for each other as predicates. It is difficult to tell if crosscategorial substitution has taken place in an utterance with omissions. Consider the following. (6')  a. Carmina dancer b. Carmina [is] [a] dancer c. Carmina dancer (danced) (6'a) is a possible agrammatic utterance. It can be interpreted in one of two ways.  Either the copula is and the determiner a are omitted, or the noun dancer is substituting for the verb danced. The first interpretation involves no cross-categorial shift whereas the second does. How can we determine which is correct? I have favored the non-cross-categorial interpretation for the following reason. If agrammatic substitutions can be cross-categorial, we expect to find examples of the following. (6")  Carmina is a danced (dancer) In (6"), the verb danced substitutes for the noun dancer. This is a clear case of cross-  categorial substitution. This type of substitution is not attested in agrammatic production. Substitution errors are syntactically constrained: they must be of the same syntactic category. I reiterate the generalization in the following constraint. (7) Categorial Constraint on Substitution X substitutes for Y , iff category X = category Y Since substitutions are syntactically constrained, substitutions are not random.  158  4.1.2 Substitution Involves Agreement, Case and Tense Features  I have established that substitution errors are constrained by syntactic category. Are they governed by any other constraints? If not, we would expect that any word from a given category could substitute for a target of that same category. For example, any noun, say bohemians, could substitute for the noun dog in the utterance a dog eats the chicken, resulting in the following: a bohemians eats the chicken. However, this is not the case. Substitutions are further constrained in one of three ways: they are phonologically constrained, semantically constrained or syntactically constrained.  I discuss each in turn below, beginning with  phonological substitutions.  4.1.2.1 Phonologically Constrained Substitutions  The first set of substitution errors are phonologically constrained substitutions. Phonological substitutions involve words with the same initial segment, porridge for poorhouse, for example. I found only three examples of phonological substitutions in the corpus under study, all three from German. These are presented below. (8) Agrammatic Substitutions: Phonologically Related a. und das Madchen dankt[e] dachte der Wolf ist nicht gut and the girl thanks thought the wolf is not good "and the girl thought the wolf isn't good"  German  b. das Paar geht pickenpickeln ... picknick (picknicken) the couple goes pick.Tnfpiss.Tnf... picnic (picnic:Inf) "the couple goes on a picnic"  German  c. ... geht der Bauer eine (einen) Kukuruz proben probieren ... goes the farmer a:F (a:M) corn tesf.Inf taste:Inf "(still later) the farmer goes to taste the corn"  German  Although there are few such errors in the agrammatic corpus, they match the phonological substitutions in normal speakers, which have been studied extensively (Fromkin,  159  1968; MacKay, 1969, 1970a; Nooteboom, 1969; Fay, 1980ab; Cutler & Fay, 1982; Garrett, 1988). Each of the three examples involves repetition sequences, where the first item produced is incorrect, but is followed by the correct word. In each case, the mis-selection shares the same initial segment or syllable as the correct item. In (8a) dankt "thanks" is produced for dachte "thought", both beginning with [da]. In (8b) first picken "pick", and then pickeln "piss" are selected instead of picknick "picnic". Each item begins with the three segments [pik]. And last, in (8c) proben "test" is produced for probieren "taste". Both start with the segments [prob]. Thus, phonological substitutions are evidence that substitutions can be constrained by phonological shape in addition to syntactic category.  4.1.2.2 Semantically Constrained Substitutions  The second set of substitution errors are semantically constrained substitutions. Semantic substitutions involve words with similar semantic features, say boy for girl, where both noun have the features [+human] and [+young]. Semantic substitutions are in evidence in agrammatic production. They involve any syntactic category. I present examples of semantic substitutions of verbs in (9). The only examples in the corpus of semantic substitutions for verbs are from English, (9) Agrammatic Semantic Substitutions: Verb a. [DP] trips (is falling) on the floor "he is falling on the floor"  English  b. R H looks [at] (listens to) [Poss] voice, " R H listens to his voice"  English  In (9a) trip substitutes for fall, where both verbs involve some descending action. In (9b) look substitutes for listen. These two verbs involve the senses; look involves vision, listen  160  involves audition. The next example involves a German modal. Here sollt "should" substitutes for harm "can", both involving mood. (10) Agrammatic Semantic Substitutions: Modal damit ich sie, dich ah hare (hbren)... sollt (kann) German in-order-that I her, you uh hear:Prs-lSg (heanlnf)... should (can) "so that I can hear you" Next we find semantic substitutions of nouns. These are illustrated below. (11) Agrammatic Semantic Substitutions: Noun a. [Det] wife (man) eats breakfast "the man eats breakfast"  English  b. no examples in database  Dutch  c. und es rinnt der Hahn (das Wasser) iiber and it flows the faucet (the water) over "and the water is overflowing"  German  d. no examples in database  French  e. e va sul motore (camion) and go:3Sg on+the engine (truck) "and he goes on the truck"  Italian  In the English example, the noun wife substitutes for the noun man. These nouns share the semantic features [+human] and [+adiilt]. The German example involves the noun Hahn "faucet" substitutes for the noun Wasser "water". "Faucet" and "water" are semantically related in that water comes out of the faucet. The Italian substitution involves motore "engine" for camion "truck". Again, the semantic link is one of inclusion: an engine is contained within a truck. This last example has its literary parallel in synecdoche, where a part (of an object, say) is used to indicate the whole. Next we find semantic substitutions of prepositions, illustrated in (12).  161  (12) Agrammatic Semantic Substitution: Preposition a. the woman is washing the dishes English and didn't pay (doesn't pay) attention for (to) the sink "the woman is washing the dishes and isn't paying attention to the sink" b. no examples in database  Dutch  c. dann die worte hinten (nach) und vor then the:Acc-Pl words behind (after) and before "then the words after and before"  German  d. [DP] [V] visite en (a) [Poss] grandmere [DP] [V] visite by (to) [Poss] grandmother "she's paying a visit to her grandmother"  French  e. va con (a) vendere il granturco goes with (to) sell:Inf the corn "he goes to sell the corn"  Italian  For preposition substitutions, the semantic link is difficult to tease apart from the grammatical function.  In the English example, the preposition for substitutes for the  preposition to. Here, some element of directionality is shared. In German hinten "behind" substitutes for nach "after". Both prepositions have an element of "distance from the speaker, where in space for "behind" or time for "after". In the French example, the preposition en "by" substitutes for the preposition a "to", again involving direction. The Italian substitution involves con "with" for a "to", a plausible link being "with intent". The last set of semantic substitutions are adverbs. There are only two such examples, one from Dutch and one from French. (13) Agrammatic Semantic Substitution: Adverb a. [Pleo] [Cop] acht uur pas (al) [V] [Det] mevrouw [Pleo] [Cop] eight o'clock only (already) [V] [Det] woman ""It's already eight o'clock" says the woman"  Dutch  b. je [me] [V] peu (pas) du tout le voyage I [myself] [V] little (not) at all the trip "I don't remember the trip"  French  In Dutch the adverb pas "only" substitutes for the adverb al "already". Both adverbs indicate a form of brevity, here with respect to time. The French substitution involves peu 162  "little" for pas "not", both involving some negative semantic feature (these are not adverbs but do involve a modifying category). These examples are evidence that substitutions can be constrained by semantic features, as well as by syntactic category.  4.1.2.3 Syntactically Constrained Substitutions  The third set of substitutions errors are syntactically constrained substitutions. Syntactic substitutions involve words with similar syntactic features, specifically agreement, Case and tense features. Syntactic substitutions show a shift in these syntactic features. This form of substitution is the most common of the three types in this corpus. Examples of agreement shift, Case shift and tense shift are presented below. I begin with agreement substitutions, as in (14-15). (14) Agrammatic Nominal Substitution: Agreement a. the men (man) was eating "the man was eating"  English  b. [V] [DP] twee kamer (kamers) [V] [DP] two room.Sg (room:Pl) "we have two rooms"  Dutch  c. [DP] ist in, ist in Gedanke (Gedanken) wo gewesen [DP] is in, is in thought:Sg (thought:Pl) somewhere be:Part "she is somewhere else in her thoughts"  German  d. les (le) tabouret bascule the:Pl (the:Sg) stool tips "the stool tips over"  French  e. [Det] acqua [V] sotto (sul) [Det] pavimenti (pavimento) [Det] water [V] under (on) [Det] floor:Pl (floor: Sg) "the water goes on the floor"  Italian  Each of the examples in (14) involves a nominal agreement shift, specifically a shift in number. In (14a) men is produced for man, the plural noun for the singular. In Dutch (14b), the singular kamer "room" substitutes for the plural kamers "rooms". A similar shift occurs in 163  German (14c), with Gedanke "thought" substituting for Gedanken "thoughts". In French (14d), the number shift involves a determiner: les "the:Pl" substitutes for le "the:Sg". And last, the Italian plural pavimenti "floors" is produced instead of the singular pavimento "floor". (15) Agrammatic Verbal Substitution: Agreement a. your ears is (are) larger "your ears are larger"  English  b. van [Det] hond ruiken (ruikt) [DP] well [Det] dog smell:3Pl (smell:3Sg) [DP] "the dog smells something"  Dutch  c. ... [Det] Paar [DP] nicht gemerkt haben (hat) ... [Det] couple [DP] not notice:Part have:Pl (have:Sg) "(so that) the couple doesn't notice it"  German  d. [Pleo] [Cop] un couple qui vont (va) manger sur l'herbe [Pleo] [Cop] a couple that go:3Pl (go:3Sg) eat:Inf on the grass "it's a couple who are eating on the grass"  French  e. i genitori stava (stavano) in pensiero the parents be:Pst-3Sg (be:Pst-3Pl) in thought "the parents were worried"  Italian  As with the nominal shifts, each of the verbal agreement shifts in (15) involves number. In (15a) is is produced for are, the singular copula for the plural. In Dutch (15b), the plural ruiken "smell" substitutes for the singular ruikt "smells". A similar shift occurs in German (15c), with plural haben "have" substituting for singular hat "has". The French example (15d) follows the same pattern, with plural vont "go" substituting for singular va "goes". And last, the Italian singular stava "was" is produced instead of the plural stavano "were". The second type of feature shift involves Case, as in (16) (16) Agrammatic Nominal Substitution: Case a. I [Cop] flat on the (my) back "I was flat on my back"  English  b. no examples in database  Dutch  c. [DP] hatte einem (einen) Korb gepackt [DP] have:Pst-3Sg a:Dat-M-Sg (a:Acc) basket pack:Part "she had packed a basket" 164  German  d. pour mieux t'entendre / ' (mon) enfant to better you hear the (my) child  French  e. no examples in database  Italian  I have included under Case substitutions two different types of shifts. The first is illustrated by the English and French examples, in (16a) and (16d), respectively. In both the non-possessive determiner substitutes for the possessive determiner: English the for my and French / ' "the" for mon "my". The second type involves the Case of the determiner. In German, determiners have four Case forms: nominative, accusative, dative and genitive. Substitution errors with German determiners involve shifting between these four Cases (among other shifts). In (16c) the dative indefinite article einem is produced instead of the accusative  einen. I now turn to the last type of substitution, tense shift. (17) Agrammatic Verbal Substitution: Tense a. I can't (couldn't) speak "I couldn't speak"  English  b. [DP] moet (moest) [DP] uitsprechen [DP] must:Prs (must:Pst) [DP] pronounce:Inf "I had to pronounce everything" c. im Biiro hat [DP] nicht gearbeit[et] in+the office has [DP] not work:Part  Dutch  German  und (sondern) schlafen (geschlafen) and (rather) sleepdnf (sleep:Part) "in the office he didn't work, rather he sleeps" d. no examples in database  French  e. " E ' tardi" disse (dice) "It's late" say:Pst-3Sg (say:Prs) ""It's late" she says"  Italian  Again, I have included under tense substitutions two different types of shifts. The first is illustrated by the English (17a), Dutch (17b) and Italian (17e) examples. In both English and  165  Dutch, present substitutes for past: can for could and moet "must:Prs" for moest "must:Pst". The Italian displays the reverse, past for present: disse "said" for dice "says". The second type involves verbs bearing no tense features, in other words, infinitival and participial forms. The shift remains within the untensed paradigm (no tensed verbs substitute for untensed, or vice versa). In German, the infinitival form of "sleep" schlafen substitutes for the participial form  geschlafen. Syntactic substitutions are evidence that substitutions can be constrained by syntactic features as well as syntactic category.  4.1.3 Substitution Involves a Single Feature Shift  I have established that substitution errors are constrained by more than just syntactic category. They are governed by either phonological shape, semantic features or syntactic features.  I will focus on syntactic substitutions. Are there any constraints governing the  substitution of syntactic features? First, I argue against syntactic features randomly shifting with each other i.e. Case for number, number for gender, and so on. Rather, substitutions occur within a very restricted domain. The same features substitute for each other, Case for Case, number for number, etc. I then take a closer look at verbal and nominal substitutions and show that substitutions are even further constrained: any given substitution involves a single syntactic feature.  4.1.3.1 Substitution Is Not Random Feature Swapping  Can any syntactic feature substitute for any other syntactic feature? If so, we would expect to find gender substituting for Case, number for person, Case for number, and so on. I call this Random Feature Swapping. Random Feature Swapping predicts that two Case markings can appear on a word, where Case and number would be required, for example. This  166  cross-feature substitution is unattested in agrammatic production. Why should this be the case? I propose that cross-feature substitution is not possible. To begin, what would such a substitution entail? For concreteness, let us consider the nominative pronoun him. Him is specified for person (3rd), number (singular), gender (masculine) and Case (nominative i.e. 'subject'). Him can undergo substitution involving these syntactic features. Let us attempt the feature swap suggested above: replace number with (another) Case. This requires replacing singular with accusative, say. The resulting feature bundle is the following: 3rd, feminine, nominative and accusative. There is no word that corresponds to this collection of features. Moreover such a combination of features is not licit: there is a conflict of Case features (an element cannot be both Nominative and Accusative). Perhaps, Random Feature Swapping can be saved. A possible scenario is the following. Features do swap indiscriminately. However, when no word matches the set of features, the substitution crashes and therefore is not produced. Thus, the Case-for-number swap attempted above can occur: it simply crashes. This account is unappealing for three reasons. First, it allows for a large array of vacuous substitutions. Any time the substitution crosses features, it crashes. This may seem sensible when all the features appear in a single morpheme, like the him example. But what of a syntactic feature that is instantiated by an independent morpheme? We could expect to find the possessive marker 's being added to the pronoun him. The result would be him's. These errors are unattested in agrammatism. Notice that the him's error violates the morphological restrictions of English. This brings us to the second argument against Random Feature Swapping: it runs contrary to other types of production errors. Production errors are constrained by the grammar. Fromkin (1968, 1971) argues that phonological errors obey the phonological rules of the language. Languagespecific constraints also govern agrammatic errors . Specifically, production errors respect the 1  See any of the following for such language-specific constraints on agrammatic behavior: Peuser & Fittschen, 1977; Grodzinsky, 1984; Caplan, 1987; Slobin & Talay, 1988; Tzeng & Chen, 1988; Chen, 1989, 1993; Wulfeck et al., 1989; Hagiwara & Caplan, 1990; Kehayia et al., 1  167  morphological constraints of the language (Kehayia & Jarema, 1991; MacWhinney & OsmanSagi, 199.1; Kehayia, 1992). Random Feature Swapping states that the mechanisms underlying substitutions may violate the constraints of the language, even though the phonetic output does not. This is an unattractive situation. More appealing is the proposal that substitutions occur within a very restricted domain. Like the categorial constraint, where nouns substitute for nouns (verbs for verbs, etc.), features of one kind substitute for features of the same kind: Case for Case, number for number, person for person, and so on. Third, Random Feature Swapping makes the wrong predictions: it predicts that any feature can substitute for any other feature, and that any number of features can be substituted. If we restrict ourselves to licit combinations of syntactic features, we could logically expect to find Case substituting for Case and number substituting for number at the same time. For example, in French the possessive plural determiner mes "my" could substitute for the singular feminine definite determiner la "the". This particular substitution would involve a shift in both number and Case (I exclude gender here since mes is unspecified for gender). However, these types of substitution errors are unattested. Rather, there is a pattern within the set of syntactic features: any given substitution involves a single syntactic feature. I propose the following constraint. (18) Single Feature Constraint on Substitution X substitutes for Y , iff category X and category Y differ by feature a where a = one of {person, number, gender, Case, tense} I show below that substitution errors follow the Single Feature Constraint.  1990; Lorch, 1990; MacWhinney et al., 1991; Safi-Stagni, 1991, 1993; Vaid & Pandit, 1991; Wulfeck et al., 1991; Hagiwara, 1993; Haverkort, 1993; and many others. 168  4.1.3.2 Verbal Substitution  In verbal substitutions we find substitutions of agreement and tense. Agreement substitutions involve person (1st, 2nd, and 3rd), number (plural and singular) and gender (feminine, masculine and neuter). Tense substitutions are split between tensed (past and present) and untensed (participials and infinitives) verbs.  4.1.3.2.1 Agreement Verbal agreement substitution divides into person (1st, 2nd, and 3rd), number (plural and singular), and gender (feminine, masculine and neuter). These three sets are illustrated below. I begin with Person substitution, as in (19). More examples of person substitutions on verbs are presented in Appendix (38). (19) Agrammatic Substitution: Person on Verbs a. he open (opens) the doors "he open the doors"  English  b. maar verder [DP] kom (komt) vanself but further [DP] come:Prs-lSg (come:3Sg) by-itself "but further such a thing comes by itself  Dutch  c. no examples in database  German  d. no examples in database  French  e. il lupo scruto (scruta) la bambina the wolf watch:Prs-lSg (watch:3Sg) the girl "the wolf watches the girl"  Italian  In (19a) open is produced for opens, a non-3Sg form of the verb for the 3Sg form . In 2  Dutch (19b), the l S g kom "come" substitutes for the 3Sg ruikt "comes". A similar shift occurs in Italian (19e), with scruto "watch: l S g " substituting for scruta "watch:3Sg".  I am aware that the English verb in question is a bare form. However, I am assuming an agreement shift by analogy with the unambiguous Italian examples. Notice that these English examples can also be interpreted as number shifts from 3Sg to 3P1. 2  169  I now turn to number substitution, as in (20). For more examples of verbal number substitutions, see Appendix (39). (20) Agrammatic Substitution: Number on Verbs a. your ears is (are) larger "your ears are larger"  English  b. van [Det] hond ruiken (ruikt) [DP] well [Det] dog smell:Prs-3Pl (smell:3Sg) [DP] "the dog smells something"  Dutch  c. ... [Det] Paar [DP] nicht gemerkt haben (hat) ... [Det] couple [DP] not notice:Part have:Prs-3Pl (have:3Sg) "(so that) the couple doesn't notice it"  German  d. [Pleo] [Cop] un couple qui vont (va) manger sur l'herbe French [Pleo] [Cop] a couple that go:Prs-3Pl (go:3Sg) eat:Inf on the grass "it's a couple who are eating on the grass" e. i genitori stava (stavano) in pensiero the parents be:Pst-3Sg (be:3Pl) in thought "the parents were worried"  Italian  The examples in (20) match those presented above: each of the verbal agreement shifts involves number. In (20a) is is produced for are, the singular copula for the plural. In Dutch (20b), the plural ruiken "smell" substitutes for the singular ruikt "smells". A similar shift occurs in German (20c), with haben "have" substituting for hat "has". The French example (20d) follows the same pattern, with vont "go" substituting for va "goes". And last in (20e), the Italian singular stava "was" is produced instead of the plural stavano "were". The last type of agreement shift involves gender, as in (21). (21) Agrammatic Substitution: Gender on Verbs e andato (andata) nel bosco is gone:M-Sg (gone:F) in+the wood " she went into the woods"  Italian  The only example of verbal gender shift is in Italian. That participial "gone" is produced in the masculine andato instead of the feminine andata.  170  Crucially, what is not attested is a substitution that combines these three features. In other words, we do not find a 'mixed' shift in person, number and gender, or any combination of the three features. If person shifts, it shifts alone. If number shifts, no other feature shifts. This is true for gender as well. Only a single feature ever shifts. Agreement shifts follow the Single Feature Constraint.  4.1.3.2.2 Tense Both tensed and untensed verbs undergo substitutions that are not shifts in agreement features. Substitutions with tensed verbs involve the various tense markings: present and past, specifically. These are illustrated below. More examples of tense substitutions are presented in Appendix (40). (22) Agrammatic Verbal Substitution: Tense a. I can't (couldn't) speak "I couldn't speak"  English  b. [DP] moet (moest) [DP] uitsprechen [DP] must:Prs (must:Pst) [DP] pronounce:Inf  Dutch  "I had to pronounce everything" c. no examples in database  German  d. no examples in database e. " E ' tardi" disse (dice) "It's late" say:Past-3Sg (say:Prs) ""It's late" she says"  French Italian  In the English (22a) and Dutch (22b), present substitutes for past: can for could and moet "mustPrs" for moest "must:Pst". The Italian example (22e) displays the reverse, past for present: disse "said" for dice "says". I include a second type of substitution under tense shift: substitutions involving untensed verbs, in other words, verbs bearing no tense features. Substitutions with untensed verbs involve shifting between participials and infinitivals. These are illustrated below. For more examples of substitutions involving untensed verbs, see Appendix (41). 171  (23) Agrammatic Verbal Substitution: Non-finite Forms a. the couple was surprise (surprised) to, was surprise (surprised) that the basket was open "the couple was surprised that the basket was open"  English  b. no examples in database  Dutch  c. im Biiro hat [DP] nicht gearbeit[et] in+the office has [DP] not work:Part  German  und (sondern) schlafen (geschlafen) and (rather) sleepilnf (sleep:Part) "in the office he didn't work, rather he sleeps" d. no examples in database  French  e. l'uomo sta posando noprendere (prendendo) la sveglia the man is putting no take:Inf (taking) the alarm-clock "the man is putting/taking the alarm clock"  Italian  In (23a), the bare verb surprise substitutes for the participial surprised?. In German (23c), the infinitival form of "sleep" schlafen substitutes for the participial form geschlafen. In (23e) the Italian infinitival prendere "take" substitutes for the participial prendendo "taking". As with the agreement substitutions, one feature shifts at a time. Moreover, the shift remains within the untensed paradigm: untensed verbs do not substitute for tensed verbs, nor vice versa. I summarize the verbal substitution patterns in the table below.  The general problem lies with the nature of bare verbs in English. Bare verbs function as both the infinitival form and the non-3Sg 'present tense' forms. The infinitival and non-3Sg present forms of main verbs are, therefore, morphologically indistinguishable. However I base the analysis on the languages where the forms are unambiguous. Since the English data is consistent with the proposed analysis, I assume number/person shifting and infinitival shifting wherever appropriate. 3  172  Table 14: Attested Agrammatic Verbal Substitutions Person Person  Number  Gender  Tense  Untensed  V  Number  V V  Gender  V  Tense  V  Untensed  A given feature shift does not co-occur with any other feature shift. When the feature person shifts to any setting of that person feature, we do not find a simultaneous shift in either number, gender, tense or untensed settings. This restriction carries through the entire selection of verbal features. Substitution thus adheres to the Single Feature Constraint.  4.1.3.3 Nominal Substitutions  In nominal substitutions we find substitutions of agreement and Case.  4.1.3.3.1 Agreement Nominal agreement substitution divides into person (1st, 2nd, and 3rd), number (plural and singular), and gender (feminine, masculine and neuter). Not every nominal category can or does undergo substitution for each agreement feature. In this database, the only examples of nominal person shift are found with object clitics. If this restriction turns out to be true for a larger database, it is significant because is gives us a diagnostic for clitic status. Since of the five languages only French and Italian have clitics, these are the only two languages that display person shift. 173  (24) Agrammatic Substitution: Person on Clitics a. et avant me (se) [Aux] deguise/er avec (en) chaperon and before selftlSg (self:3Sg) disguise:Part/Infwith (as) R H "and before (he) disguised himself as L R R H "  French  b. e la donna si (16) sveglia and the woman self (him) wakes "and the woman wakes him"  Italian  In the French example, the l S g me "self:lSg" substitutes for se "self:3Sg". In the Italian example, both clitics are 3Sg. The clitic shifts to a reflexive form si "self:3g" from the non-reflexive lo "him" target. Unlike person, number substitution is attested in nouns (25), adjectives (26) and determiners (27). More examples of number substitutions with nouns are presented in Appendix (42). (25) Agrammatic Substitution: Number on Nouns a. the men (man) was eating "the man was eating"  English  b. [V] [DP] twee kamer (kamers) [V] [DP] two room.Sg (room:Pl) "we have two rooms"  Dutch  c. [DP] ist in, ist in Gedanke (Gedanken) wo gewesen [DP] is in, is in thought.Sg (thought:Pl) somewhere be:Part "she is somewhere else in her thoughts"  German  d. no examples in database  French  e. [Det] acqua [V] sotto (sul) [Det] pavimenti (pavimento) [Det] water [V] under (on) [Det] floor:Pl (floor: Sg) "the water goes on the floor"  Italian  In (25a) men is produced for man, the plural noun for the singular. In Dutch (25b), the singular kamer "room" substitutes for the plural kamers "rooms". A similar shift occurs in German (25c), with Gedanke "thought" substituting for Gedanken "thoughts". And last, the Italian plural pavimenti "floors" is produced instead of the singular pavimento "floor".  174  Only one example of number substitution on adjectives is attested. In the French example below, the singular "normal" normal substitutes for the plural normaux. (26) Agrammatic Substitution: Number on Adjectives d. [DP] [Cop] normal (normaux) [DP] [Cop] normal.Sg (normahPl) "they were normal"  French  Number substitution is also attested with determiners, as in (27). For more examples of number substitution with determiners, see Appendix (43). (27) Agrammatic Substitution: Number on Determiners a. no examples in database  English  b. no examples in database  Dutch  c. no examples in database  German  d. les (le) tabouret bascule the:Pl (the:Sg) stool tips "the stool tips over"  French  e. il contadino mangia i (il) granone the farmer eats the:M-Pl (the:Sg) grain:Sg "the farmer eats the grain"  Italian  In (27d) the French plural determiner les "the" substitutes for the singular le. In the Italian example, the plural / "the" is produced instead of the singular il. Gender substitution also affects a wide range of nominal categories. We find gender substitution in pronouns (28), clitics (29), adjectives (30) and determiners (31). For more examples of gender substitutions involving pronouns, see Appendix (44). (28) Agrammatic Substitution: Gender on Pronouns a. no examples in database  English  b. no examples in database  Dutch  c. sie (es) wollte die kranke Oma besuchen she:F (she:N) wanted the sick grandma visit "she wanted to visit her sick grandma"  German  175  d. le loup demande au [N] ou // (elle) va the wolf asks to+the [N] where he (she) goes "the wolf asks L R R H where she is going"  French  e. no examples in database  Italian  In (28c) the feminine pronoun sie "she" is produced instead of the required neuter es. In the French example, the masculine il "he" substitutes for the feminine elle "she". The only gender substitution with clitics involves a shifts from the masculine le "him" to the feminine la "her". (29) Agrammatic Substitution: Gender on Clitics elle la (le) reveille she her (him) wake up  French  Gender shifts on adjectives are also attested, as in (30). More examples of gender substitutions with adjectives are presented in Appendix (45). (30) Agrammatic Substitution: Gender on Adjectives a. distinction not available  English  b. no examples in database '  Dutch  c. no examples in database  German  d. et Grand-mere comme vous les avez les grands (grandes) dents French and Grandmother how you them have the big:M (big:F) teeth:F "and Grandmother, what big teeth you have" e. sta zitto (zitta) be:Prs-3Sg quiet:M (quiet:F) "Be quiet!"  Italian  Both examples in (30) involve substituting the masculine form for the feminine: French grands "big:M" for grandes "big:F' and Italian zitto "quiet:M" for zitta "quiet:F\ Gender shift shows up most frequently on determiners, when the distinction is available. Examples are presented in (31). More examples of gender substitution involving determiners are presented in Appendix (46).  176  (31) Agrammatic Nominal Substitution: Gender on Determiners a. distinction not available  English  b. de dief kwam binnen door de (het) raam the thief come:Past in through the:Sg-M/F (the:N) window "the thief came in through the window"  Dutch  c. das kleine Madchen streckt ihre (seine) Hand emport the little girl stretches her:F (her:N) hand up "the little girl stretches her hand out"  German  d. la scene montre le (la) maman the scene shows the:M (the:F) mother "the scene is shows the mother"  French  e. il la finestra the:M-Sg the:F-Sg window "the window"  Italian  In Dutch (31b) the feminine/masculine form for "the" de substitutes for the neuter form het.  The German example in (31c) displays a similar shift: the feminine ihre "her:F"  substitutes for the neuter seine "her:N". In French (31c) the masculine "the" ie is produced instead of the required feminine la. A n d last, in the Italian example, the first attempt at the definite article is the masculine il rather than the feminine la. As with the verbal substitutions, a single agreement feature shifts at a time. Mixed substitutions are not attested. Once again, substitutions follow the Single Feature Constraint.  4.1.3.3.2 Case Case substitution involves the various Case markings: nominative, accusative, dative, etc. I address two types of Case substitution. The first involves shifting from [+Case] to [Case]. This is exemplified in English and French. The shift we observe is from a possessive determiner to a definite determiner. For our purposes we can describe this shift as possessive to non-possessive. These are illustrated below. For more examples of Case substitution with determiners, see Appendix (47).  177  (32) Agrammatic Substitution: Case on Determiners a. I [Cop] flat on the (my) back "I was flat on my back"  English  b. pour mieux t'entendre/' (mon) enfant to better you hear the (my) child  French  The only feature that is shifting in both the examples in (32) is Case. Although the possessive determiners are each specified for person (here 1st), no person shift can be taking place since the definite articles are unspecified for person. For English, both the possessive my and the definite the are unspecified for number. But for French this is not the case. Both the possessive mon and the definite / ' are singular. No number shift has occurred. This mixed agreement and Case shift is unattested in the agrammatic corpus. Let us now turn to the second type of Case shift. Unlike English and French, Dutch and German mark their nominal categories for a number of Cases. For example, determiners are marked for nominative, accusative, dative and genitive. Thus the Case shifts that occur in Dutch and German are not simply shifts between possessive and non-possessive categories. We are witness to shifts between accusative and nominative, dative and accusative, and so on. As with gender substitution, Case substitution is most often attested with determiners. In (33b), the dative indefinite article einem substitutes for the accusative einen. More examples of Case substitutions involving determiners are presented in Appendix (47). (33) Agrammatic Substitution: Case on Determiners a. no examples in database  Dutch  b. [DP] hatte einem (einen) Korb gepackt [DP] have:Pst-3Sg a:Dat-M-Sg (a:Acc) basket pack:Part "she had packed a basket"  German  Case substitutions also affect adjectives (34), nouns (35) and pronouns (36). However, there is only one example in the corpus of each type of substitution, all three being from German.  178  (34) Agrammatic Substitution: Case on Adjectives [P] lauter guten gute (guten) Dinge (Dingen) German [P] nothing-but good good:Acc-Pl (good:Dat) thing:Acc-Pl (thing:Dat) "with nothing but good things" (35) Agrammatic Substitution: Case on Nouns bei den Kinder (Kindern) bin ich zu Hause geblieben German by the:Dat-Pl children:Nom/Acc (children:Dat) was: lSg I at home stay:Part "I stayed at home with the children" (36) Agrammatic Substitution: Case on Pronouns die Oma spent ihn (ihm) auf the grandma unlocks him:Acc-M-Sg (him: Dat) PRT "the grandma unlocks (it) for him"  German  Each of the examples above display a shift from a dative form to an accusative (or accusative/nominative) form. In all the examples of Case shifts, only Case is substituted. No other feature changes. Once again, substitutions only involve a single feature. I summarize the substitution pattern in the table below. Table 15: Attested Agrammatic Nominal Substitutions Person Person Number  Number  Gender  Case  V V V  Gender  V  Case  179  A given feature shift does not co-occur with any other feature shift . When the feature 4  Case shifts to any setting of that Case feature, we do not find a simultaneous shift in either person, number or gender settings. This restriction carries through the entire selection of nominal features. Substitution thus adheres to the Single Feature Constraint.  4.2  T H E DISTRIBUTION OF SUBSTITUTIONS  I have established that substitution errors involve a single feature shift among agreement, Case and tense features. But is every possible feature sTuft equally instantiated? Put differently, what are the most frequent feature shifts? Although some feature display a favored direction, the general conclusion is that markedness is not a governing principle for substitutions. Using the substitution rate of determiners as a case study, I argue that number of features influences substitution: the more features a category encodes, the higher the substitution rate (and the lower omission rate).  There are two putative exceptions to the Single Feature Constraint. Both involve German determiners. They are displayed below. (i) da begegnet sie ein sie eine Wolf German there meet:Prs-3Sg she a:Nom-M-Sg she a:Nom/Acc-F-Sg wolf "there she meets a w o l f (ii) in der in das Biiro German in the:Nom-M-Sg in the:Acc-N-Sg office "in the office" In (i), the required form of the determiner is indicated as einem "a:Dat-M-Sg". The first attempt ein is Nom-M-Sg; this involves a single feature shift, Case. The second attempt eine is Nom/Acc-F-Sg; this involves a double shift, Case and gender. However, the choice of einem as target is not uncontroversial. Another possible target is einen "a:Acc-M-Sg" (p.c. from native speakers). The difference is between a dative object for "meet" or an accusative object. If the target is einen "a:Acc-M-Sg", both attempts shift by a single feature: ein is nominative instead of the required accusative; eine is feminine instead of the required masculine. In (ii) the target is das "the:Acc-N-Sg". The first attempt is der "Nom-M-Sg". This attempt displays an apparent double shift, Case and gender. However, the Acc-N form of the (das) is homophonous with the Nom-M form. It is possible that the shift only involves the following: der " N o m - M " to das "Nom-N", with the target being das " A c c - M " . In other words, the actual target was never uttered (but is phonologically identical to the second attempt). We are then left with 2 single feature shifts: one of gender, der to das; and one of Case, das to das. 4  180  4.2.1 Verbal Substitution  Verbal substitutions include agreement and tense shifts. I summarize the substitution rates for person, number, gender, tense and untensed shifts in the table below. Table 16: Agrammatic Percentage of Verbal Feature Substitutions Person  Number  Gender  Tense  Untensed  EF  6/14-42.9%  1/14-7.1%  n/a  4/14-28.6%  3/14-21.4%  EE  2/8 - 25%  0%  n/a  4/8 - 50%  2/8 - 25%  DB  1/3 - 33.3%  1/3 - 33.3%  n/a  1/3 - 33.3%  0%  DH  0%  2/2 - 100%  n/a  0%  0%  GB  0%  1/2 - 50%  n/a  0%  1/2 - 50%  GM  1/3 - 33.3%  0%  n/a  0%  2/3 - 66.7%  FC  0/0  0/0  0/0  0/0  0/0  FA  0%  2/2 - 100%  0%  0%  0%  IR  9/15-60%  5/15-33.3%  1/15-6.7%  0%  1/15-6.7%  TV  0%  4/6 - 66.7%  0%  2/6-33.3%  0%  There are two observations to make. First, there is no single feature that consistently undergoes substitution more often than the other features.  Second, the substitutions are  minuscule in number, with the one exception of Italian (where the numbers are simply small). Since there are so few tokens of any given error in a single speaker or language, it is impossible  181  to determine in what direction a feature shifts. I therefore group together all the shifts for each verbal feature.  4.2.1.1 Agreement  I consider the three agreement features: person, number and gender. Most of the person shifts involve 3rd to 1st. Number is almost evenly split between plural and singular, the shift being bi-directional i.e. plural shifts to singular and singular shifts to plural. Gender involves a single token, shifting from feminine to masculine.  4.2.1.1.1 Person Of the 11 person shifts, the majority (10) involve 3rd to 1st person, specifically 3Sg to lSg.  Table 17: Agrammatic Person Substitution Pattern TO 1st  2nd  1st FROM  3rd 1  2nd 3rd  9  1  Most of these are found in one Italian subject (IR). The confusion lies with describing the actions of an individual in the story or picture description (the wolf, for example) as done by the speaker himself.  182  4.2.1.1.2 Number Number displays a bi-directional shift between singular and plural, with the shift to singular being very slightly higher, 9 of the 16 shifts being from plural to singular. This is displayed in the table below. Table 18: Agrammatic Number Substitution Pattern TO Plural FROM  Plural  Singular 9  Singular  7  With such an even distribution between the two shifts, number substitution cannot be governed by markedness, which would presumably favor one direction over the other.  4.2.1.1.3  Gender  Only one verbal gender shift is displayed. The token is an Italian past participial. The shift is from feminine to masculine. It is presented below. (37) Agrammatic Verbal Substitution Pattern: Gender Errors e andato (andata) nel bosco is gone:M-Sg (gone:F) in+the wood "she went into the woods"  183  Italian  Table 19: Agrammatic Gender Substitution Pattern TO Masculine FROM  Feminine  Masculine Feminine  1  4.2.1.2 Tense  The majority of tense shifts are from past to present. Table 20: Agrammatic Tense Substitution Pattern TO Past FROM  Past  Present 9  Present  2  Substitutions involving untensed verbs shift between infinitival and participial forms. The majority of these errors shift to the infinitive form, where the infinitive is not a bare form of the verb but rather includes an infinitival suffix (for all but English).  184  Table 21: Agrammatic Substitution Pattern for Untensed Verbs TO Infinitive FROM  Infinitive  Participle 3  Participle  6  4.2.2 Nominal Substitution  Nominal substitutions include agreement and Case shifts. I summarize the substitution rates for person, number, gender and Case shifts in the table below.  185  Table 22: Percentage of Nominal Substitutions in the Agrammatic Data Person  Number  Gender  Case  EF  0%  1/1- 100%  0%  0%  EE  0%  3/5 - 60%  0%  2/5 - 40%  DB  0%  0%  1/1 - 100%  0%  DH  0%  1/3 - 33.3%  2/3 - 66.6%  0%  GB  0%  5/24 - 20.8%  10/24-41.7%  9/24 - 37.5%  GM  0%  0%  21/31 -67.7%  10/31-32.3%  FC  2/21 - 9.5%  2/21 -9.5%  13/21-61.9%  4/21 - 19.1%  FA  0%  0%  3/3 - 100%  0%  IR  0%  4/12 - 33.3%  8/12 - 66.7%  0%  IV  1/5 - 20%  2/5 - 40%  2/5 - 40%  0%  Total # of tokens  3  18  50  25  The feature with the overall highest levels of substitution is gender, followed by Case. Since there are so few tokens of any given error in a single speaker or language, it is difficult to determine in what direction a feature shifts. I therefore group together all the shifts for each nominal feature.  186  4.2.2.1 Agreement  I consider the three agreement features: person, number and gender. A l l three person shifts involve clitics, two of which shift to a reflexive form . Number is evenly split between plural and singular, displaying a bi-directional shift. The largest number of feature shifts involve gender, most of which shift to the feminine.  4.2.2.1.1 Person There are very few tokens of nominal person shift. We find two tokens in French and one in Italian. I present them below. (38) Agrammatic Nominal Substitution: Person Errors a. avant me (se) [Aux] deguise/er avec (en) chaperon French before selftlSg (self:3Sg) [Aux] disguise:Part7lnfvw'm (as) R H "before disguised himself as L R R H " b. a ce moment-la un policier s' (1') attend at that moment a policeman self (him) wait:Prs-3Sg "at that moment a policeman waits for him"  French  c. e la donna si (lo) sveglia and the woman self (him) wake:Prs-3Sg "and the woman wakes him"  Italian  A l l three person shifts involves object clitics. The French examples involve two types of shifts. First, person shifts from 3rd to 1st. Second, person remains'within 3rd person, but shifts from 3rd person to a reflexive 3rd person. The Italian person also shifts from 3rd person to a reflexive 3rd person. The shifts are presented in the table below.  187  Table 23: Agrammatic Person Substitution Pattern TO 1st  2nd  3rd  1st FROM  2nd 3rd  1  2 (Refl)  4.2.2.1.2 Number There is a larger pool of number errors. No preferred direction is indicated, however. Instead we find bi-directional shifting between singular and plural. Table 24: Agrammatic Number Substitution Pattern TO Plural FROM  Plural  Singular 9  Singular  9  As before, number substitution shows no influence from markedness, displaying a bidirectional pattern.  188  4.2.2.1.3 Gender When it comes to gender errors, all or most of them are syntactic and not semantic. In other words, a shift in gender is not in the direction of the biological sex of the argument in question. The only language which exhibits a (relatively) substantial number of semantic gender errors is German. A l l the gender shifts are displayed in the table below. Table 25: Agrammatic Gender Substitution Pattern TO  #  FROM  Masculine  Feminine  Neuter  M/F  M/N  16  37  5  2  0  21  3  Masculine  24  Feminine  12  11  Neuter  22  5  M/F  1  M/N  1  1 15  2 1  1  Although gender shifts seem to display almost all patterns, over half of the shifts are to the feminine form (37/60), followed by a shift to the masculine form (16/60).  4.2.2.2 Case  Case substitutions favor shifts to nominative and accusative forms. The Case shifts are displayed in the table below.  189  Table 26: Agrammatic Case Substitution Pattern TO  #  FROM  Nom  Acc  Dat  Gen  Nom/Acc  5  8  1  1  8  1  3  Nom  2  Acc  4  3  Dat  12  2  Gen  5  Nom/Acc  0  2 1 6  5  Once again the shifts display a variety of patterns. However, two tendencies are observed. First, about half of the Case shifts involve shifting from dative to some other Case (12/23). Second, with two exceptions, shifts are to nominative or accusative. Since one of the Cases is a combined nominative/accusative, it is difficult to determine which of the two is favored. Keeping to the tokens that can distinguish between the two, eight shift to accusative while five shift to nominative. Thus there is a slight tendency to favor the accusative. *This needs to be tied in to verb class - do the accusative shifts have accusative semantics?*  4.2.3 Robustness and Substitution  I showed above that substitutions do not pattern in a specific direction, although a few features seem to favor somewhat one direction over another. Are there any other factors that influence substitution? A given syntactic category does not display the same substitution (or  190  omission) rate across languages. What syntactic properties could be contributing to the variability (setting aside the factors discussed in chapter 2)? I take as a case study the category of determiner. Consider the following Table, which indicates the percentage of determiners that are omitted and that are substituted for each agrammatic. Table 27: Agrammatic Determiner Error Rate Omission  Substitution  EF  7.0%  1.0%  EE  55.2%  1.9%  DB  81.5%  0.6%  DH  55.2%  1.3%  GB  18.5%  8.1%  GM  15.5%  16.0%  FC  18.3%  63%  FA  4.2%  1.4%  IR  21.5%  1.7%  IV  14.3%  7.6% •  I focus on Dutch and German since they display the sharpest contrast with respect to omission and substitution rates. German determiners encode more agreement and Case features than any of the other languages. German determiners encode 2 number features (singular and  191  plural), 3 gender features (feminine, masculine and neuter) and 4 Case features (nominative, accusative, dative and genitive). In contrast, Dutch determiners encode 2 number features (singular and plural), two gender features (M/F and N), and no Case features. The substitution rates for Dutch determiners are 0.6% and 1.3%. The substitution rates for German determiners are 8.1% and 16.0%. Thus, German determiners, which encode more features than Dutch determiners, display a higher substitution rate than Dutch determiners. Moreover, German determiners have lower omission rates than Dutch determiners: 18.5% and 15.5% for German; 81.5% and 55.2% for English. Thus, German determiners encode more agreement and Case features than Dutch determiners and show a lower omission rate. I propose that the difference in omission and substitution results from Robustness. The more agreement and Case features a category encodes, the more Robust it is. The more Robust a syntactic category, the better retained it is. However, it is then more likely to suffer from substitution errors. German determiners, which encode more features than Dutch determiners, are more Robust. Consequently, they show a lower rate of omission but a higher rate of substitution. I develop a more explicit Principle of Robustness in chapter 5.  4.3  CONCLUSION  I discussed three types of substitutions: phonologically constrained, semantically constrained and syntactically constrained. I concentrated on syntactic substitutions. These substitution errors display four characteristics. First, substitutions are of the same syntactic category. A noun substitutes for a noun, a verb for a verb, and so on. Substitutions are therefore subject to the Categorial Constraint, repeated below. (39) Categorial Constraint on Substitution X substitutes for Y , iff category X = category Y  192  Second, substitutions involve the mis-selection of agreement, Case and tense features. These mis-selections are governed by the Single Feature Constraint, whereby only a single agreement, Case or tense feature can be mis-selected. The constraint is presented in (40). (40) Single Feature Constraint on Substitution X substitutes for Y , iff category X and category Y differ by feature a where a = one of {person, number, gender, Case, tense} Third, substitution involves markedness only minimally. In other words, substitution does not involve a shifting to a less marked feature specification. In fact, substitutions are frequently bi-directional. Last, substitutions are influenced by the number of features encoded in a given category. The more features that are encoded in a category (i.e. the more Robust a category), the higher the substitution rate (and the lower the omission rate). This Principle of Robustness is articulated more explicitly in the following chapter.  193  Chapter 5 A G R A M M A T I S M : W H E R E IS T H E DEFICIT?  In the two previous chapters I discussed the nature of omissions and substitutions. I proposed that agrammatic clauses include the full array of syntactic projections and that omissions are not the result of syntactic structures that are 'missing'. With respect to substitutions, I showed that agreement, Case and tense features are manipulated and that only one feature shifts at a time. In this chapter I address the following question: Where in the Grammar does the agrammatic deficit take place? In other words, what mechanisms are impaired such that they result in omission and/or substitution errors? I investigate the phonological, morphological and semantic levels, and conclude that none of these linguistic levels accounts for agrammatic omission and substitution errors.  The deficit must lie outside the linguistic levels of  representation per se. I propose that the agrammatic deficit lies with the interface between the Lexicon and the linguistic levels of representation, specifically the process of Lexical Insertion. Lexical Insertion is the process by which Lexical Items are introduced into the structures that are manipulated by the grammar, be it the semantic, syntactic, morphological or phonological component. Lexical Items are the elements contained within the Lexicon.  In order to fully  understand the process of Lexical Insertion, I address the nature of Lexical Items.  Any given  Lexical Item is made up of a bundle of features. This bundle is composed of 4 subarrays: a phonological matrix, optional formal features, intrinsic formal features and semantic features. The phonological matrix consists of all the relevant phonological features making up the Lexical Item: the feature geometry, syllabic structure, prosodic structure, etc. The optional formal features consist of agreement, Case and tense features. The intrinsic formal features consist of categorial features and Case-assigning features (the latter are only relevant to V , P and any other category that assigns Case). The semantic features consist of idiosyncratic semantic information ([canine] for wolf, [feline] for cat, and so on) and for V theta-grids (or their equivalent).  194  Subsequent to establishing what a Lexical Item is, I turn to the general process of Lexical Insertion. 1 discuss two proposals for Lexical Insertion: Full Lexical Insertion (Chomsky, 1965, 1975, 1981, 1986, 1993, 1995; and many others), Split Lexical Insertion (Otero, 1976, 1983; den Besten, 1977; Fiengo, 1980; Koster, 1987; Anderson, 1992; Birring, 1993; Halle & Marantz, 1993). I argue for a version of Split Lexical Insertion, the detailed mechanisms of which I make explicit. Split Lexical Insertion has the following three mechanisms: Feature Insertion, Feature Matching and Lexical Retrieval. Feature Insertion involves the following: it takes from the Lexicon a feature bundle consisting of the semantic features and the intrinsic formal features of a Lexical Item and inserts them into the syntactic component in the course of the derivation. Once the derivation is complete, Feature Matching scans the relevant paradigm and pairs the feature bundle with the appropriate phonological matrix. Lexical Retrieval, the final mechanism, inserts the fully specified feature bundle into the structure.  I argue that Lexical  Insertion is impaired for both omission and substitution. However, the specific mechanisms that are impaired are different.  For omission, Lexical Retrieval is damaged: the complete feature  bundle is not retrieved, leaving the utterance without the phonological matrix. For substitution, Feature Matching is damaged: the Base semantico-syntactic feature bundle is paired with the wrong phonological matrix, resulting in a word with the wrong optional formal features. Further, I argue that a Principle of Robustness governs omissions and that the paradigmatic organization of the Lexicon accounts for the characteristics of substitutions. Thus omission and substitution result from different impairments. Lastly, I consider the interaction between the principles of U G and processing mechanisms.  I define competence and performance, equating competence with U G and  performance with processing mechanisms. I then discuss the constraints that are placed on a model of language production.  195  5.1  W H A T A N A G R A M M A T I C DEFICIT ISN'T  I have argued that the structure of agrammatic clauses include the full array of syntactic projections. Moreover, I claimed that the deficit lies outside the syntax proper. In this section, I argue against a purely phonological, morphological, or semantic account of omission and substitution errors. In other words, neither omissions nor syntactic substitutions can be reduced to phonological, morphological or semantic errors. The evidence I present is of two types. First, both omission and substitution display certain characteristics. I show that none of these accounts can explain all these characteristics. Second, each account predicts certain behavior. I show that these predictions are not borne out. I conclude that the agrammatic deficit lies in the interface between the Lexicon and the linguistic levels of representation. The precise nature of the deficit is addressed in 5.2.  5.1.1  Not a Purely Phonological Deficit  The phonological component (henceforth PF) manipulates the phonological matrices of morphemes.  A phonological matrix includes featural, segmental, syllabic, and prosodic  information. When combined with syntactic information, PF also contributes the stress patterns of sentences. If PF is the locus of the agrammatic deficit, we expect production characteristics that involve only features, segments, syllables, prosody and sentential stress.  However,  agrammatic production is typically characterized by omission and substitution errors that display syntactic characteristics. Agrammatics may display deviant segmental articulation (Blumstein et al., 1978).  However, this results from articulatory and phonatory difficulties that are  independent of the agrammatic deficit per se (cf. dysarthria). I begin with omission errors. The characteristics of omission errors are summarized below.  196  (1) Characteristics of omission errors a. b. c. d. e.  Omissions involve all syntactic categories Lexical categories are better retained than functional categories The various functional categories display different retention patterns Nominal categories are better retained than verbal categories 'Parts of words' are not omitted (where bare forms are not licit in the language)  Kean (1977, 1980)' argues that the phonological word is "the domain over which the assignment of stress takes place" . 2  She claims that agrammatic speech tends to reduce the  structure of the sentence to the minimal string of phonological words.  Since Kean defines  'phonological word' in terms of stress, her account only addresses omissions in languages where words are either stressed or unstressed. English is such a language.  However, French and  Italian, while displaying reduced stress, do not display unstressed words. Kean's account must at the outset be modified to accommodate these types of languages. I now turn to the five characteristics in (1). A l l syntactic categories (potentially) can be omitted. A phonological deficit splits omission along phonological and non-phonological word status. The prediction is that not all categories are omitted. A phonological deficit does not account for characteristic (la). F(unctional) categories, being unstressed, are not considered phonological words, and, therefore, tend to be omitted . L(exical) categories, being stressed, are phonological words, and, 3  Nespoulous & Dordain (1990) also argue that agrammatic speech results from a phonological deficit. Blumstein & Goodglass (1972) and Goodglass et al. (1975) argue that English-speaking agrammatics retain stress perception. This sensitivity to stress is apparent in their production: agrammatics produce appropriate word stress. Notice that word stress being intact is yet another argument against a phonological deficit. However, I restrict myself to omissions and substitutions since these are the foci of the thesis. Schwartz, Saffran and Marin (SS&M) (1980) argue that Kean's claim is too strong. If agrammatics are insensitive to F-categories and, consequently, to the purely syntactic aspects of grammatical morphemes, they ought to be able to recover underlying structures where grammatical morphemes are not essential. In other words, they ought to be able to map an N-VN sequence onto the canonical (English) SVO relationship. S S & M argue that agrammatics do not exhibit an SVO overgeneralization with respect to comprehension. They conclude that agrammatics suffer from a word order defect. However, I must repeat the following two points: S S & M study comprehension, and I focus on production; with respect to production, I argued that word order is unimpaired. 1  2  3  197  therefore, tend not to be omitted . The prediction is that lexical categories are better retained than 4  functional categories.  This is indeed the case.  A phonological deficit therefore potentially  accounts for the lexical-functional distinction in (lb). Although a phonological deficit explains the lexical-functional distinction, it predicts all Fcategories to have the same rates of omission, all other (phonological) things being equal. In other words, a phonological account does not predict the differing omission rates between the various F-categories. In section3.4.2.2 I showed that F-categories do not have the same rate of omission. Specifically, D is better retained than T. A phonological deficit cannot account for characteristic (lc). L-categories all bear stress.  According to Kean's definition, L-categories are  phonological words and should be unaffected. Although omission rates of L-categories are lower than those of F-categories, L-categories still are omitted. Consequently, it is not the case that only non-phonological words are omitted, as predicted by a phonological account. Even were we to allow for L-category omissions, a phonological account predicts all L-categories to have the same rate of omission. Specifically, nouns and verbs ought to have the same rate of omission. This prediction is not borne out: nouns are better retained than verbs, as shown in section 3.4.2.3. A phonological deficit cannot account for characteristic (Id). The last characteristic in (1) is that 'parts' of words are not omitted.  For example,  agreement markings on verbs are not omitted, leaving a bare verb, where bare forms are not licit in the language. A phonological deficit predicts only that non-phonological words, not parts of words, tend to be omitted. A phonological deficit therefore accounts for characteristic (le). I now turn to substitution errors.  To begin, a phonological account does not predict  substitution errors. The prediction is that phonological words are retained, and non-phonological words are omitted. Thus the existence of substitution errors themselves are evidence against a Brown (1980) cites work done by Kellar (1978) which reveals that the abnormal sorting pattern, which makes agrammatics (who have a lesion in the anterior portion of the brain) sensitive to stressed elements and correspondingly insensitive to unstressed elements, is also present in aphasics who have a lesion in the posterior section of the brain. He, therefore, argues that it is "premature to link this [stress] deficitto agrammatism" (1980: 291).  4  198  phonological account.  However, let us assume that substitutions are predicted.  The  characteristics of substitution errors are summarized below. (2) Characteristics of Substitution Errors a. A l l syntactic categories are substituted b. Substitutions involve a single syntactic feature c. Substitutions are governed by the Principle of Robustness As with omissions, all syntactic categories are (potentially) substituted.  Once again, a  phonological deficit splits substitution along phonological and non-phonological word status. The prediction is that not all categories are substituted. A phonological deficit does not account for characteristic (2a). A phonological account makes two further predictions with respect to substitutions. Substitutions are either phonologically less-marked or they are random, involving any number or type of features. This prediction is not borne out.  First, markedness is not involved, neither  syntactic nor phonological, since shifts are (for the most part) bi-directional (see section 4.2). Second, substitution errors are not random. They are governed by the Single Feature Constraint, where only a single agreement, Case or tense feature is altered (see section 4.1.3) . 5  A  phonological deficit fails to account for characteristic (2b). Lastly, a phonological deficit cannot predict which categories suffer substitution errors and which suffer omission errors. Under phonological rule, a given category should suffer the same omission/substitution rate regardless of the number of syntactic features it encodes. Once again, this is not the case. Categories that encode more syntactic F-features are more likely to be substituted and less likely to be omitted.  This requires access to syntactic features.  A  phonological deficitfails to accountfor characteristic (2c).  This single feature shift parallels the single feature change in phonological slips of the tongue (see Fromkin, 1968, 1971).  5  199  Since a phonological deficit cannot account most of the production characteristics of agrammatism, the phonological component must be rejected as the unique locus of the deficit . 6  5.1.2 Not a Purely Morphological Deficit  I argue that a purely morphological deficit, with no reference to syntactic features, is not a possible explanation for agrammatism .  Let us begin by considering a (purely) morphological  7  component.  First, this component is responsible for derivational, inflectional and compound  morphology. A morphological deficit predicts agrammatics to have difficulties with all three types of morphology. However, agrammatics have problems with neither derivational  (De  Bleser & Bayer, 1985; Eling, 1986; Miceli & Caramazza, 1988; Ulatowska, 1988; Bebout, 1993) nor compound (Hittmair-Delazer et al., 1994)  morphology.  Of the three, only inflectional  morphology is affected. Thus only the aspect of morphology that makes reference to syntactic features is affected. Allowing for this focus on inflectional morphology, if the morphological component is the locus of the agrammatic deficit, we expect production characteristics that involve only morphological units.  In other words, a morphological deficit predicts omissions and  substitutions to display characteristics along purely morphological lines. The distinctions are between roots (bare forms), stems, affixes and (morphological) clitics, with a more general split between free and bound morphemes. We therefore expect to find omissions and substitutions dividing along these types of morphemes. I begin with omission errors.  The characteristics of omission errors are summarized  below. In addition to not accounting for the agrammatic characteristics, a phonological deficit makes certain predictions. In a phonological deficit we expect to find a disruption of the phonological rules of the language. This prediction is unattested. Hungarian aphasics, for example, retain vowel harmony (MacWhinney & Osman-Sagi, 1991). This runs contrary to some who argue for a morphological deficit (Bradley et al., 1980; Lapointe, 1983; Kehayia et al., 1990; Kehayia, 1992). However, they too run into opposition  6  7  200  (3) Characteristics of omission errors a. Omissions involve all syntactic categories b. Lexical categories are better retained than functional categories c. The various functional categories display different retention patterns d. Nominal categories are better retained than verbal categories e. 'Parts of words' are not omitted (where bare forms are not licit in the language) Although all syntactic categories can be omitted, the tendency is for free morphemes to be omitted and bound morphemes to be substituted. A morphological deficit can account for this splits by adopting a free-bound distinction. A morphological deficit is compatible with characteristic (3a). L-categories are, for the most part, stems.  F-categories are, for the most part  (morphological) clitics (cf. determiners, pronouns, reduced auxiliaries and modals).  A  morphological deficit can account for the lexical-functional distinction by adopting an omission contrast between stems and clitic categories. Stems are retained, and clitics are omitted. A morphological account can therefore accountfor characteristic (3b). However, variation within these stem and clitic divisions is more difficult to account for. Not all F-categories share the same omission rate. The distinction within F-categories cannot be one of clitic versus affix, since affixes on the whole are substituted rather than omitted (see discussion below). A morphological deficit does not accountfor characteristic (3c). Although a morphological deficit predicts the stem-clitic distinction, it does not account for the omission contrast between nouns and verbs (Zingeser & Berndt, 1990). Both nouns and verbs are stems. Both ought to suffer the same omission rate. However, they do not: verbs are omitted more frequently than nouns. A morphological deficit does not account for characteristic (3d). The last characteristic in (3) is that 'parts' of words are not omitted. This brings us to the realm of substitution. A morphological deficit predicts that bare forms may be substituted for  (De Bleser & Bayer, 1985; Dressier, 1986; Jarema & Kadzielawa, 1987; Dressier & Denes, 1990; Bebout, 1993; De Bleser & Luzzatti, 1994). 201  inflected forms. However, this prediction is not borne out . Bare forms are not produced when 8  they are not licit in the language. Moreover, where these so called bare forms are produced, it is only when they are options with the relevant paradigms. In other words, if a given paradigm does not include the bare form, it will not be produced. A morphological deficit therefore does not account for characteristic (3e). I now turn to substitution errors. The characteristics of substitution errors are summarized below. (4) Characteristics of Substitution Errors a. A l l syntactic categories are substituted b. Substitutions involve a single syntactic feature c. Substitutions are governed by the Principle of Robustness As mentioned above, free morphemes tend to be omitted and bound morphemes tend to be substituted. A morphological deficit accounts for characteristic (4a). Since inflectional morphology has access to agreement, Case and tense features, a morphological deficit can accountfor characteristics that are determined by these specific features. A morphological deficit then accounts for characteristic (4b-c). A morphological deficit makes one last prediction: any given morphological category displays omission or substitution.  For example, free morphemes are omitted, and bound  morphemes are substituted. However, some morphological classes display split behavior: the omission of a free morpheme and the substitution of a bound morpheme.  This is true of  determiners, pronominal clitics (for French and Italian) (see chapters 3 and 4) and Chinese nominal classifiers (Tzeng et al., 1991). Morphology is left to explain why free morphemes display the behavior of both free and bound morphemes. Since a morphological deficit does not account for all the production characteristics of agrammatism, I reject it as the unique locus of the deficit. Lapointe (1985) argues that English speaking agrammatics show a preference for the least marked verbal form, and claims that this tendency is universal. Bastiane et al. (1991) show that  8  202  5.1.3  Not a Purely Semantic Deficit  The semantic component specifies the semantic features of a morpheme and the relations between the morphological items. The semantic component is also the locus of discourse and pragmatic constraints.  If agrammatic production results from a semantic deficit, we expect  agrammatics to have disturbed discourse and pragmatic structures. Although research in this areas is only at the preliminary stages, these predictions are not borne out. Pragmatic factors governing reference (such as the use of novel/familiar information) are found to be intact in agrammatics (Wulfeck et al., 1989). As well, agrammatics have no difficulty with contextual appropriateness of adverbs (Hupet etal., 1986; Frederix, 1985). I now turn to omission errors. The characteristics of omission errors are summarized below. (5) Characteristics of omission errors a. Omissions involve all syntactic categories b. Lexical categories are better retained than functional categories c. The various functional categories display different retention patterns d. Nominal categories are better retained than verbal categories e. 'Parts of words' are not omitted (where bare forms are not licit in the language) Conceivably, a semantic deficit affects anything with semantic content, thus all syntactic categories. A semantic deficit therefore accounts for characteristic (5a). A semantic deficit separates categories along semantic lines.  If syntactic categories  display distinctions along these semantic lines, then they are predicted to behave differently with respect to omission. Semantic features distinguish F-categories from L-categories and nouns from verbs. A semantic deficit therefore accounts for characteristics (5b) and (5d). However, variation within F-categories is syntactically defined rather than semantically defined. A semantic deficit does not accountfor characteristic (5c).  for Dutch agrammatics, this is not the case. In chapter 4 I argue that morphological markedness is not the governing principle for substitution errors. 203  I now turn to substitution errors. The characteristics of substitution errors are summarized below. (6) Characteristics of Substitution Errors a. A l l syntactic categories are substituted b. Substitutions involve a single syntactic feature c. Substitutions are governed by the Principle of Robustness Once again, a semantic deficit affects anything with semantic content, thus all syntactic categories. A semantic deficit therefore accounts for characteristic (6a). However, if the semantic component is the locus of the agrammatic deficit, we expect production characteristics to involve features that influence semantic interpretation.  Since  agreement, Case and tense features are not crucial for semantic interpretation, a semantic deficit would not predict these features to 'go wrong'. A semantic deficit does not account for any characteristics that involve these features, namely (6b-c). Since a semantic deficit does not predict all of the production characteristics of agrammatism, the semantic component must be rejected as the unique locus of the deficit. I summarize the evidence from omissions and substitutions against phonological, morphological and semantic accounts of agrammatic production in the following table.  204  Table 28: Summary of Evidence from Omissions and Substitutions  5.2  Agrammatic Data  Phonological Account  Morphological Account  Semantic Account  Omissions involve all categories  V  *  V  V  L-categories have fewer omissions  V  V  V  V  Omissions rates vary between F-categories  V  *  *  *  Nouns have fewer omissions than verbs  V  *  *  V  Bare forms are not produced  V  V  *  *  Substitutions involve all categories  V  *  V  Single syntactic features shift  V  *  V  *  Robustness effect for substitutions  v  *  V  *  •  A G R A M M A T I C PRODUCTION: A DEFICIT IN L E X I C A L INSERTION  The omission and substitution errors under study cannot be accounted for by a purely phonological, morphological or semantic deficit. Moreover, I argued that agrammatic clauses include the full array of syntactic projections, leaving syntactic movement unimpaired. In other words, omissions and substitutions are syntactic errors that do not result from a deficit within the syntactic component. Then what exactly is affected when omissions and substitutions are in 205  evidence? I propose that the deficit is one that involves the process of Lexical Insertion. Lexical Insertion is the process by which Lexical Items are introduced into the structures that are manipulated by the grammar, be it the semantic, syntactic, morphological or phonological component. Lexical Items are the elements contained within the Lexicon.  In order to fully  understand the process of Lexical Insertion, I address the nature of Lexical Items.  5.2.1 The Lexical Item  The Lexicon contains a list of all the words in the speaker's language. Each word is a Lexical Item (henceforth LI). Following Chomsky (1993), an LI is made up of a feature array, which includes a phonological matrix, formal features (F-features) and semantic features. There are two types of F-features, intrinsic and optional . The feature array consists of 4 sub-arrays: a 9  phonological matrix, a bundle of optional F-features, a bundle of intrinsic F-features (which consists of categorial features, for the most part), and a semantic feature array. The phonological matrix consists of all the relevant phonological features making up the Lexical Item: segmental, syllabic and prosodic structures, etc. The optional F-features consist of agreement, Case and tense features. The intrinsic F-features consist of categorial features and Case-assigning features (the latter are only relevant to V , P and any other category that assigns Case) . The semantic 10  features consist of idiosyncratic semantic information ([canine] for wolf, [feline] for cat, and so on) and for V theta-grids. The LI for wolf'is presented in (7).  This split between intrinsic and optional F-features seems to match very closely Gazdar et al.'s (1985) split between Major (intrinsic, more or less) and Minor (optional, more or less) features. There is variation in what constitutes optional and intrinsic F-features. The 3rd person feature is an intrinsic F-feature for all nouns. However, pronouns, determiners and agreement on verbs may vary in person specifications. For example, we have the pronouns /, you, she for 1st, 2nd and 3rd person, respectively. In these cases, the person feature is an optional F-feature rather than and intrinsic one. Note also that Case features are specific to nominal categories and tense features are specific to verbal categories. 9  10  206  (7)  LI for wolf WOLF:  [wojlfj  [Singular] [0 Gender] [0 Case]  [Lexical] [Nominal] [Referential] [3rd Person]  [Canine] [Wild]  I use upper case to indicate the abstractform of the item under scrutiny. In other words, the term ' W O L F ' refers to the abstract entity which denotes the English noun wolf. The LI in (7) shows 4 columns. Each column represents one of the 4 feature sub-arrays.  I describe each  column from left to right. The first column within the LI is the phonological matrix. I represent the matrix by the phonemic entry of the morpheme, side-stepping the phonological details. The second column consists of the optional F-features (in this case wolf is unspecified for gender and Case features, which I indicate with '0') . The third column are the intrinsic F-features, here 11  the categorial and person features. The last column are the semantic features. 1 have displayed the feature sub-arrays from most to least language-specific. Each syntactic category is a bundle of categorial features. The use of N(oun) or V(erb) is thus an abbreviation for these bundles. This is illustrated in (8). (8)  N Lexical Referential Nominal  Lexical Referential  The termN refers to the bundle of privatives features [Lexical Referential Nominal], and the term V refers to the bundle of features [Lexical Referential].  I adopt the use of this  abbreviation for the specifications of LI, when the precise categorial features are not relevant. The abbreviated LI for wo If is presented in (9).  Although I include in the array the optional F-features that are unspecified, it is also possible that these features are absent from the array altogether. Either approach is consistent with my analysis. 11  207  (9)  LI for wolf (English) WOLF:  [woolf]  [Singular] [0 Gender] [0 Case]  N [3rd Person]  [Canine] [Wild]  Let us now consider the LI for the German noun Wolf. (10)  LI for Wolf (German) WOLF:  [volfj  [Singular] [Masculine] [0 Case]  N [3rd Person]  [Canine] [Wild]  The intrinsic F-features and the semantic features are identical to that of English W O L F . Where the differences lie are with respect to the phonological matrix and the optional F-features. This is exactly the locus of omission and substitution. In summary, an LI is a feature bundle consisting of four sub-arrays of features: phonological features, optional F-features, intrinsic F-features and semantic features.  5.2.2  Lexical Insertion  The Lexicon is the basis from which syntactic structures are projected. As mentioned in the preceding section, the elements contained within the Lexicon are Lexical Items (Lis). Lexical Insertion is then the process by which Lis are introduced into the syntax . The L i s project the 12  syntactic structures. More specifically, the categorial features (the intrinsic F-features) that make up the LI project the syntactic structure. These syntactic structures are then subject to the various  There is a further debate about Lexical Insertion, independent of where it takes place: is Lexical Insertion rule-based or representation-based? Jackendoff (1995) proposes that the Lexicon is an interface level that matches three independent sets of representations: phonological, syntactic and semantic. Lexical Insertion then is not insertion at all. Rather it consists of a matching condition on representations. A similar debate runs through the processing literature, with production models being rule-based (Schlesinger, 1977) or representation-based (McClelland & Rumelhart, 1981; Stemberger, 1985; Pritchett, 1992). What is at issue for my purposes is the location of Lexical Insertion, not the nature of the process. I therefore remain neutral with respect to which variant I prefer. 12  208  syntactic operations and constraints within the syntactic component. There are two general approaches to Lexical Insertion. First, Chomsky's (1965, 1975, 1981, 1986, 1993) Lexical Insertion is a process which introduces the fully specified LI at the start of the derivation . I label this Full Lexical Insertion. In an alternative version of Lexical 13  Insertion the phonological matrix is not introduced when syntactic structures are projected. Instead, the phonological matrix is inserted after all syntactic operations have been performed, but before the string enters the phonological component (Otero, 1976, 1983; den Besten, 1977; Fiengo, 1980; Koster, 1987; Anderson, 1992; Biiring, 1993; Halle & Marantz, 1993). I call this Split Lexical Insertion. In this section I describe the two approaches to Lexical Insertion. I begin with Full Lexical Insertion and show that problems arise when attempting to account for omissions and substitutions. I then turn to Split Lexical Insertion.  5.2.2.1 Full Lexical Insertion  Full Lexical Insertion (FLI) introduces all four feature subarrays into the syntactic component at the start of the derivation (Chomsky, 1965, 1975, 1981, 1986, 1993). I refer to this mechanism as Lexical Insertion. A general diagram illustrating this mechanism is provided below.  I use the term "derivation" to denote the process of constructing a grammatical sentence, whether the process consists of the application of a collection of rules or the use of constraints on representations. 13  209  Figure 7: General Schema of Full Lexical Insertion semantic features intrinsic F-features optional F-features phonological features  LEXICON  SYNTAX SEMANTICS  PHONOLOGY  The process of Lexical Insertion takes from the Lexicon the complete feature bundle (with all four subarrays) and introduces it into the syntactic component at the start of the derivation. I argued that agrammatic clauses have the full set of projections. If the intrinsic F-features project the syntactic structures, these features must be available to the agrammatic speakers. Moreover, the mechanism that introduces these features must be fully functional in agrammatics. Thus, the process of Lexical Insertion (which introduces intrinsic F-features into the syntactic component) cannot be impaired. But how is this feature bundle created? What are the processes involveD in combining the various feature subarrays? Following Chomsky (1993, 1995) I assume that the semantic features and the intrinsic Ffeatures are inherent to a Lexical Item. This bundle of two subarrays makes up the Base (feature bundle) of an LI in the Lexicon. The Base (feature bundle) is schematized as follows. (11) Base Feature Bundle for a Lexical Item in the Lexicon semantic features intrinsic F-features  The optional F-features are added to this Base via a mechanism I call Optional Feature Insertion. The resulting tripartite feature bundle is presented below.  210  (12) Tripartite Feature Bundle for a Lexical Item in the Lexicon semantic features intrinsic F-features optional F-features  The Lexical Item must now receive its phonological matrix. The phonological matrix is dependent on the optional F-features that are present in the tripartite feature bundle (which is why the phonological matrix is not inserted into the matrix before the optional features are).  To  illustrate, the pronouns he and she have identical semantic and intrinsic features. They differ in their optional features: he is [Masculine] and she is [Feminine]. I argue below that the Lexicon is organized in paradigms. To retrieve the appropriate phonological matrix, a mechanism first scans the relevant paradigm and pairs the tripartite feature bundle with the appropriate phonological matrix. I label this mechanism Feature Matching. A second mechanism retrieves and inserts this matched phonological matrix into the tripartite feature bundle, producing a complete feature bundle. I label this mechanism Phonological Matrix Insertion. The resulting complete feature bundle is schematized below. (13) Complete Feature Bundle for a Lexical Item in the Lexicon semantic features intrinsic F-features optional F-features phonological features  The process of Lexical Insertion inserts this complete feature bundle into the syntactic component. Within the syntactic component the process of Feature Checking, among other functions, ensures that words agree in their feature specifications (subjects agreeing with verbs, for example) (Chomsky, 1993, 1995) . 14  Chomsky (1995) mentions that optional F-features may be assigned during the derivation rather than being inserted in the Lexicon and then checked in the derivation. If the phonological 14  211  Having described the various mechanisms contributing to Lexical Insertion, I now consider the following question: how are substitutions and omissions produced? I begin with substitutions.  Substitutions are in effect mismatches between sets of optional features and  phonological matrices. I propose that substitutions result from an impaired Feature Matching mechanism and/or an impaired Phonological Matrix Insertion mechanism. When no substitution occurs, the following processes take place: optional features are assigned to the Base feature bundle (via Optional Feature Insertion); the correct phonological matrix is paired with the tripartite feature bundle (via Feature Matching); that same phonological matrix is retrieved and inserted into the tripartite feature bundle (via Phonological Matrix Insertion); the complete feature bundle is inserted into the syntactic component (via Lexical Insertion); and the optional features are checked in the syntax (via Feature Checking). When a substitution occurs, Optional Feature Insertion still assigns optional F-features to the Base feature bundle, as such Feature Checking does not force the derivation to crash since the optional features are correct (it is the phonological matrix that is incorrectly matched). Beyond that, processes go awry. There are three possibilities. First, the wrong phonological matrix is paired with the tripartite bundle, this particular wrong matrix is (correctly) retrieved and inserted into the tripartite bundle, and the complete feature bundle (with the wrong phonological features) is inserted into the syntactic component. This sequence results from an impaired Feature Matching mechanism (FM). Second, the correct phonological matrix is paired with the tripartite bundle, the wrong matrix is retrieved and inserted into the tripartite bundle, and the complete feature bundle (with the wrong phonological features) is inserted into the syntactic component. This sequence results from an impaired Phonological Matrix Insertion mechanism. Third, the wrong phonological matrix is paired with the tripartite bundle, a different phonological wrong matrix is retrieved and inserted into the tripartite bundle, and the complete matrix is specified at the start of the derivation, having optional F-features assigned during the derivation requires the following. The phonological matrix itself must be specified for the optional features, which must be checked with the optional features that are assigned during the derivation, otherwise there is no way of ensuring that the phonological matrix is matched to the correct optional features. If the two sets of optional features match, the derivation is licit. If they do not match, the derivation crashes. This is unduly complicated since it requires the feature  212  feature bundle (with the wrong phonological features) is inserted into the syntactic component. This sequence results from impairment to both Feature Matching and Phonological Matrix Insertion. A l l three possible sequences result in substitutions. How to decide between them? To begin, substitutions are subject to the Single Feature Constraint. If both Feature Matching and Phonological Matrix Insertion are impaired, more than a single feature is likely to shift. Moreover, if impairment to a single mechanism can account for substitution, why posit a deficit to two mechanisms? Thus either Feature Matching or Phonological Matrix Insertion is impaired. Which one is impaired? Once again I return to the Single Feature Constraint on substitutions. Assume Feature Matching is impaired. To derive a single feature shift, the impaired Feature Matching must (i) choose the correct paradigm and (ii) pick a phonological matrix close to the target rather than anywhere in that paradigm. This requires unimpaired selection of the paradigm as well as some awareness of the actual target. If Feature Matching is indeed impaired, neither bit of information should be accessible. Now assume Phonological Matrix Insertion is impaired. A fully functional Feature Matching has lead us to the correct paradigm and ultimately the correct target. Since Phonological Matrix Insertion is already located at the target, when it is impaired, it copies a neighboring matrix rather than one in another paradigm or far off from the target. In other words, an impaired Phonological Matrix Insertion readily lends itself to the Single Feature Constraint. I therefore propose that substitutions result from an impaired Phonological Matrix Insertion. I now turn to omissions. What mechanism is impaired such that omissions result? There are two possibilities: Optional Feature Insertion and Phonological Matrix Insertion. I begin with the former.  As mentioned above, optional F-features and phonological features are closely  linked. The specific identity of the phonological matrix is determined by what optional features are present. Conceivably, if no optional features are present then no phonological matrix can be retrieved. It is possible then that omissions result from an impaired Optional Feature Insertion. checking mechanism that feature assignment is to replace. 213  I therefore assume that optional  When Optional Feature Insertion is functional, the following takes place: optional F-features are assigned; a phonological matrix is paired with the tripartite feature bundle (via Feature Matching); the phonological matrix is inserted into the feature bundle (via Phonological Matrix Insertion); and the complete feature bundle is inserted into the syntactic component (via Lexical insertion). When Optional Feature Insertion is non-functional, optional features are not assigned, no phonological matrix is paired or retrieved, and an incomplete feature bundle (with oniy semantic and intrinsic features) is inserted into the syntactic component . 15  There are two problems with such a proposal. First, one of the functions of Feature Checking is to motivate movement. In other words, syntactic categories move in order to check their optional features. If no optional features are present, then movement should not take place. However, as I argued in chapter3, movement takes place even when omissions are in evidence. Second, Optional Feature Insertion manipulates optional features but not intrinsic features.  I  argue below that (the number of) intrinsic features influence omission rates. It is unclear how a process that does not handle intrinsic features can account for a phenomenon that is influenced by intrinsic features. These problems disappear if the impairment lies with Phonological Matrix Insertion. When Phonological Matrix Insertion is functional, the following takes place: optional F-features are assigned (via Optional Feature Insertion); a phonological matrix is paired with the tripartite feature bundle (via Feature Matching); the phonological matrix is inserted into the feature bundle; and the complete feature bundle is inserted into the syntactic component (via Lexical Insertion). When Phonological Matrix Insertion is non-functional, the following sequence occurs: optional features are assigned (via Optional Feature Insertion); a phonological matrix is paired to the tripartite feature bundle (via Feature Matching); no phonological matrix is retrieved and/or inserted into the tripartite bundle; and an incomplete feature bundle (with only semantic, intrinsic and optional features) is inserted into the syntactic component. features are assigned in the Lexicon for Full Lexical Insertion.  214  I now return to the problems with the Optional Feature Insertion account mentioned above. Optional F-features are required if movement is to take place in the syntax. The Optional Feature Insertion account eliminates optional features from the feature bundle that enters the syntactic component. With the Phonological Matrix Insertion account, the feature bundle that is inserted into the syntactic component includes optional F-features. This eliminates the first of the two problems. The second problem revolves around the Retention Hierarchy. I argue below that syntactic categories that are specified for more categorial features (intrinsic F-features) are more likely to be retained than those specified for fewer categorial features. In other words, items with fewer intrinsic features are more likely to be omitted. This means that the (impaired) mechanism that results in omission should have access to the intrinsic features of that item. Optional Feature Insertion simply manipulates optional F-features,  with no access to intrinsic features.  Phonological Matrix Insertion, on the other hand, handles the entire tripartite feature bundle, which includes intrinsic F-features.  A Phonological Matrix Insertion account eliminates the  second problem as well. I therefore propose that omissions result from an impaired Phonological Matrix Insertion. For Full Lexical Insertion, both substitutions and omissions result from an impaired Phonological Matrix Insertion mechanism. With substitutions, the wrong matrix is retrieved and inserted into the feature bundle. With omissions, no phonological matrix is retrieved. The locus for both types of errors is the same; however, the process is different.  5.2.2.2  Split Lexical Insertion  Split Lexical Insertion introduces the semantic features, intrinsic F-features and optional F-features into the syntactic component at the start of the derivation and inserts the phonological matrix after all syntactic operations have been performed, but before the string enters the Recall that intrinsic F-features must be inserted into the syntactic component since they project the syntactic structures that are present in agrammatic clauses, even when omissions are in evidence. 15  215  phonological component (Otero, 1976, 1983; den Besten, 1977; Fiengo, 1980; Koster, 1987; Anderson, 1992; Biiring, 1993; Halle & Marantz, 1993). A general diagram illustrating these mechanisms is provided below. Figure 8: General Schema of Split Lexical Insertion semantic features intrinsic F-features optional F-features  LEXICON  SYNTAX phonological features SEMANTICS  /  PHONOLOGY  Split Lexical Insertion has two mechanisms that connect the Lexicon to the syntactic component: Initial Insertion and Lexical Retrieval. Initial Insertion involves the following: it takes from the Lexicon a feature bundle of an LI consisting of the semantic features, the intrinsic F-features and the optional F-Features and inserts the bundle into the syntactic component at the beginning of the derivation. Lexical Retrieval is the process that retrieves the appropriate phonological matrix once the syntactic derivation is complete and before the string enters P F  1 6  (the precise nature of this mechanism is described below). I argued that agrammatic clauses have the full set of projections. If the intrinsic F-features project the syntactic structures, these features must be available to the agrammatic speakers. Moreover, the mechanism that introduces these features must be fully functional in agrammatics.  The process must precede the phonological component since strings that include omissions are subject to post-Lexical phonological rules. Consider the following utterance from French, (i) je [Aux] arrive I [Aux] arrive:Part "I arrived" The pronoun would normally be je if the auxiliary were present: je suis arrive. However, what is actually produced is the following: j' arrive. The (phonologically determined) j' variant is required since the following word begins with a vowel. See Sanchez (1992a) for other such examples. 16  216  Thus, the process of Initial Insertion (which introduces intrinsic F-features into the syntactic component) cannot be impaired. The agrammatic deficit must lie with other mechanisms. Once again, following Chomsky (1993, 1995) I assume that the semantic features and the intrinsic F-features are inherent to a Lexical Item. This bundle of two subarrays makes up the Base (feature bundle) of an LI in the Lexicon. It is schematized below. (14) Base Feature Bundle for a Lexical Item in the Lexicon semantic features intrinsic F-features  As with Full Lexical Insertion, the optional F-features are added to this Base via Optional Feature Insertion. The resulting tripartite feature bundle for an LI in the Lexicon is presented below. (15) Tripartite Feature Bundle for a Lexical Item in the Lexicon semantic features intrinsic F-features optional F-features  It is at this point that Split Lexical Insertion differs from Full Lexical Insertion. The process of Initial Insertion inserts this tripartite feature bundle into the syntactic component. Within the syntactic component the process of Feature Checking, among other functions, ensures that words agree in their feature specifications (subjects agreeing with verbs, for example) (Chomsky, 1993, 1995) . 17  Before entering the phonological component, the Lexical Item must receive its phonological matrix. I argue below that the Lexicon is organized in paradigms. To retrieve the Chomsky (1995) mentions that optional F-features may be assigned during the derivation rather than being inserted in the Lexicon and then checked in the derivation. Either approach is 17  217  appropriate phonological matrix, the Feature Matching mechanism scans the relevant paradigm and pairs the tripartite feature bundle with the appropriate phonological matrix. As stated above, Phonological Matrix Insertion retrieves and inserts the phonological matrix into the tripartite feature bundle. This results in a complete feature bundle, schematized below. (16) Complete Feature Bundle for a Lexical Item in the Lexicon semantic features intrinsic F-features optional F-features phonological features  The syntactic component interfaces with PF and L F . The complete feature bundle is introduced into this interface level. I label this process Lexical Retrieval, as schematized below. Figure 9: Model of (a Split Lexical Insertion) Grammar  18  semantic features intrinsic F-features optional F-features  LEXICON  SYNTAX phonological features SEMANTICS  /  PHONOLOGY  Having described the mechanisms contributing to the Split Lexical Insertion, I now consider the following question: how are substitutions and omissions produced? I begin with substitutions. To repeat, a substitution is a mismatch between a set of optional features and a phonological matrix. As with Full Lexical Insertion, I propose that substitutions result from an consistent with the account of agrammatic omission and substitution presented for Split Lexical Insertion.  218  impaired Phonological Matrix Insertion mechanism. The process is identical to that described for Full Lexical Insertion. I now turn to omissions. There are two possible mechanisms that may be impaired: Phonological Matrix Insertion or Lexical Retrieval. The Phonological Matrix Insertion alternative is as described for Full Lexical Insertion. I now consider the Lexical Retrieval alternative. When Lexical Retrieval is functional, the following occurs: the tripartite feature bundle is introduced into the Lexicon; the bundle is paired to a phonological matrix; the phonological matrix is retrieved and inserted into the tripartite bundle; and the complete bundle is then inserted into the syntactic interface. When Lexical Retrieval is non-functional, the sequence is as follows: the tripartite feature bundle is introduced into the Lexicon; the bundle is paired to a phonological matrix; the matrix is retrieved and inserted into the tripartite bundle; but the complete feature bundle is not introduced into the syntax interface. There are two possible loci for omission errors: Phonological Matrix Insertion and Lexical Retrieval. There are no empirical differences between the two.  Choosing the Phonological  Matrix Insertion account has in two implications. First, the same processes are impaired whether we have Full Lexical Insertion or Split Lexical Insertion. Second, omissions and substitutions result from impairment to the same mechanism, namely Phonological Matrix Insertion. Choosing an Lexical Retrieval account implies that (i) there is a distinction between Full Lexical Insertion and Split Lexical Insertion, and (ii) omissions and substitutions result from impairments to different mechanisms, substitutions from an impairment to Phonological Matrix Insertion and omissions from an impairment to Lexical Retrieval. There are two reasons to prefer a Lexical Retrieval account. First, if an impaired Lexical Retrieval results in omissions, omissions are automatically distinct from substitutions, which result from an impaired Phonological Matrix Insertion. I argue below that substitutions and omissions are in fact distinct error types. Second, choosing the Lexical Retrieval account means It is also possible that Lexical Retrieval separates the complete feature bundle into two components. The phonological matrix is introduced into the Syntax-PF interface, and the semantic, intrinsic and optional features are introduced into the Syntax-LF interface. 18  219  that Lexical Insertion is split. While my focus is syntactic substitutions, there are a variety of possible substitutions errors (semantic, for instance).  Although Full Lexical Insertion can  account for these different types of substitutions, we would not necessarily expect different types. With Split Lexical Insertion, we expect a variety of substitutions. I therefore maintain that omissions result from an impaired Lexical Retrieval mechanism and that substitution results from an impaired Phonological Matrix Insertion mechanism.  5.2.3 Accounting for Omissions  I argued that omissions result from impairment to Lexical Retrieval. Now I turn to the factors that influence the mechanism of Lexical Retrieval. I propose a Principle of Robustness whereby the quantity of features determines the likelihood of omission or retrieval; specifically, the quantity of syntactic features.  5.2.3.1 The Principle of Robustness  Recall that agrammatics (more often) omit F-categories and V-categories. In other words, they (more often) retrieve L-categories and N-categories. For a unified featural account of agrammatic behavior we must have the features [Functional] and [Verbal] OR [Lexical] and [Nominal]. I adopt Dechaine's (1993) model of categorial features, which I repeat below (see section 1.1.3 for a discussion of syntactic features).  (18)  Dechaine's (1993) model of categorial features  Nominal Referential Functional  C  T  V +  +  + +  K  D  N  +  + + •+  + +  +  220  P  A +  This model has the features [Functional] and [Nominal]. These features do not allow a unified account for the agrammatic deficit. A revised set of features is required: one feature must be changed. Agrammatism can be characterized in one of two ways: either agrammatics have difficulty accessing specified categories, or agrammatics can better retrieve specified categories. Let us consider the first option whereby agrammatics have difficulties accessing specified categories. Then, the more features a category has, the more specified it is. The more specified it is, the more difficulties it creates. Since F-categories and V-categories are more problematic for the agrammatic speaker, the required features for this approach to the deficit are [Functional] and [Verbal].  F-categories are [Functional], and L-categories are unspecified for the feature  [Functional].  V-categories are [Verbal], and N-categories are unspecified for the feature  [Verbal]. The resulting categorial table is shown in (19). (19) Functional Verbal Referential  C  T  V  V V  V V V  V V  K  D  V  V V  N  V  P  A  V  This model predicts that the fewer features a category has, the better retained it will be (by an agrammatic). If we compare nouns and verbs, we see that nouns have fewer features than verbs, and are, therefore, predicted to be better retained than verbs. As argued above, this is indeed the case. This model, in fact, predicts the following hierarchy of retention, where ">"  means  "better retained than". (20)  Incorrectly Predicted Retrieval Hierarchy A > N , P, K > V , D, C > T This hierarchy predicts that adjectives will show the best retention. This prediction is not  borne out (Kolk, 1978; Rizzi, 1985; and many other references asserting that nouns are the category best retained). Let us now consider the second alternative, whereby agrammatics can better access  221  specified categories. Once again, the more features a category has, the more specified it is. But now, the more specified a category is, the better retained it is.  In other words, specified  categories are more robust and 'easier' to retrieve than unspecified categories. Since L-categories and N-categories are better retained in agrammatic speech, the required features for this approach to the deficit are [Lexical] and [Nominal]. unspecified for the feature [Lexical].  L-categories are [Lexical], and F-categories are  N-categories are [Nominal], and V-categories are  unspecified for the feature [Nominal]. The resulting categorial table is shown in (21). V  K  D  V V V  V  V V V V V  T  (21) Lexical Nominal Referential  N  p  V  V V  This model predicts that the more features a category has, the better retained it will be (by an agrammatic) . If we compare nouns and verbs, we see that nouns have more features than 19  verbs, and are, therefore, predicted to be better retained than verbs. As argued above, this is indeed the case. The hierarchy of retention that this particular model predicts is presented in (22). (22)  Predicted Retrieval Hierarchy N  > V , A , D > P, T, K > C  This hierarchy predicts that nouns are the most specified category and, therefore, will show the best retrieval. Nouns are indeed the category best retrieved in agrammatism. The fact that this most basic prediction is supported, whereas that of the hierarchy in (13) is not, argues in favor of the feature model in (21), where [Lexical] and [Nominal] (and [Referential]) are the necessary features. Moreover, consider the Hierarchy of Retention evidenced by the agrammatic If the features are reversed such that N is the least specified syntactic category, the hierarchy still holds. In other words, adopting the features [Functional], [Verbal] and [-Referential] results in the same hierarchy but with the opposite specification: the less specified, the better retained. However, there is no evidence indicating that N must be the least specified. Moreover, when considering substitution errors, the more optional F-feature a category encodes, the more substitutions (and fewer omission) it undergoes. In other words, Robustness is at work in substitution. To match this, I favor Robustness rather than lack of specification for omission. 19  222  speakers in the previous section.  (23)  Retention Hierarchy N > V > D > Aux/Modal The hierarchy in (23) closely resembles that presented in (22). The only distinction lies in  the division between V and D for (23) and the lack thereof for (22) How does this hierarchy connect to Lexical Insertion? Specifically, how do categorial features fit into Lexical Retrieval? I propose a Principle of Robustness. (24)  Principle of Robustness Category A is more ROBUST than category B if it is specified for more F-features than category B . The process of Lexical Retrieval is then subject to the Principle of Robustness in that it  can better manipulate syntactic categories that are more ROBUST.  I argue below that this  Principle extends to all syntactic features, including agreement, Case and tense features. The net result of the interaction between Lexical Retrieval and the Principle of Robustness is that categories that are more robust are more often retrieved and consequently show up more frequently in agrammatic production.  5.2.3.2  Deriving Omissions  I now trace the process of omission from beginning to end. I argued that omissions result from impairment to the process of Lexical Retrieval present in a Split Lexical Insertion model. The derivation I present below thus assumes such a model (as described in the section 5.2.2.2). Consider the sentences in (25). The sentence in (25a) contains an omission, the determiner . 20  The sentence in (25b) does not. 20  The original sentence from the database is the following. 223  (25) Sentences for omission derivation a. [Det] wolf eats grandmother b. the wolf eats grandmother Both sentences require the following L i s . (26) Lis for the Items that Project the Sentences in (25) THE:  [63]  [0 Person] [0 Number] [0 Gender] [0 Case]  [Nominal] [Referential]  [Definite referent]  WOLF:  [woulf]  [Singular] [0 Gender] [0 Case]  [Lexical] [Nominal] [Referential] [3rd Person]  [Canine] [Wild]  EATS:  [its]  [3rd Person] [Singular] [0 Gender] [Present]  [Lexical] [Referential]  [Consumption of food]  TENSE:  [0]  [0 Person] [0 Number] [0 Gender]  [Referential] [+Finite] [Assign Nom] [Present]  GRANDMOTHER:  [graendmASdr] [Singular] [Feminine]  21  [Lexical] [Nominal] [Referential] [3rd Person]  [Female] [Elder] [Relation]  The derivation of the sentences in (25) begins with the mechanisms within the Lexicon. In the Lexicon, the inherent specifications of the L i s include the semantic features and the intrinsic F-features. The Base feature bundle for each of the L i s in (27) are displayed below. The left-hand set of features are the intrinsic F-features, and the right-hand set are (an approximation of) the semantic features. (i) [Det] wolf eat (eats) grandmother "the wolf eats grandmother" The sentence contains a substitution as well as an omission. For our present purposes, I ignore the substitution error and assume the correct verbal form.  224  (27) Base Feature Bundle for Lexical Items in (26)  THE:  WOLF:  Nominal Referential  Definite referent  Lexical Nominal Referential 3rd Person  Canine Wild  Consumption of food  EATS:  Lexical Referential  TENSE:  Referential Assign Nom  +Finite Present  Lexical Nominal Referential 3rd Person  Female Bder Relation  GRANDMOTHER:  Still within the Lexicon, the process of Optional Feature Insertion adds to this Base the optional F-features. The resulting tripartite feature bundles for all the Lis are presented below.  A n alternative is that grandmother is unspecified for gender, [0 Gender]. The use of she in reference to grandmother would derive from the semantic feature [Female].  21  225  (28) Tripartite Feature Bundle for Lexical Items in (26) THE Nominal Referential  Definite referent  WOLF Singular  Lexical Nominal Referential 3rd Person  Canine Wild  EATS 3rd Person Singular Present  Lexical Referential  Consumption of food  TENSE Referential Assign Nom  +Finite Present  GRANDMOTHER Singular Feminine  Lexical Nominal Referential 3rd Person  Female Bder Relation  The process of Initial Insertion inserts this tripartite feature bundle into the syntactic component. The categorial features project their X-bar structure. The resulting structure is the  226  following. The only verbal functional categories I include is TP. The structure may be expanded to include AgrSP, AgrOP, CP, etc. (29) Projected Syntactic Structure for Sentences in (25) TP T' T [Present]  VP V  DP D'  NP D THE  NP  EATS GRANDMOTHER  WOLF Within the syntactic component movement takes place to check (optional) features and for Case assignment. The subject DP [THE WOLF] raises to [Spec TP] to receive nominative Case (from T). This derivation is schematized in the tree below. Note that in English the main verb does not raise to check its features. However, verbs in other languages (such as French, Dutch and German) do raise to T. (30) Derived Structure for the Sentences in (25) TP D P " ^ ^ ~ ^ ^ ~ T' THE^OLF  T ' ^ ^ " VP t D P ^ ^ ^ "  V  GRANDMOTHER The words in upper case indicate the tripartite feature bundles schematized in (28). What is missing are the phonological matrices. To retrieve the appropriate phonological matrix for each LI, the Feature Matching mechanism picks and scans the relevant paradigm and pairs the tripartite  227  feature bundle with the appropriate phonological matrix. The process of Phonological Matrix Insertion then retrieves and inserts the appropriate phonological matrix into the tripartite feature bundle. If the derivation is free of omission, as in (25b), Lexical Retrieval takes the complete feature bundles and introduces them into the Syntax interface with PF/LF (inserting them into their appropriate slots). This is schematized by the following tree. (31) Structure with Phonological Matrices for Sentence (25b) TP D P ' ^ ^ ~ ^ ^  T'  [63] [wcolf] (the wolf)  (eats)  [graendmA6dr] (grandmother)  If however omission is in evidence, as in (25a), the process of Lexical Retrieval is nonfunctional for that LI. Lexical Retrieval handles the complete feature bundle and therefore has access to the (number of) F-features within each LI. The complete feature bundles are presented below. (32) Complete Feature Bundle for Lexical Items in (25a) THE Nominal Referential [6a]  228  Definite referent  WOLF Singular  Lexical Nominal Referential 3rd Person  Canine Wild  [wcolf] EATS 3rd Person Singular Present  Lexical Referential  Consumption of food  [its] TENSE Referential Assign Nom  +Finite Present  [0] GRANDMOTHER Singular Feminine  Lexical Nominal Referential 3rd Person  Female Elder Relation  [graendmAddr]  In section 5.2.3.1 I argued that omissions are governed by the Principle of Robustness, which is repeated below. (33)  Principle of Robustness Category A is more ROBUST than category B if it is specified for more F-features than category B .  229  Robustness governs Lexical Retrieval. Thus Lis that are specified for more F-features are more likely to be retrieved than those with fewer specified F-features. Of all the L i s in (32), the determiner has the fewest F-features, It is therefore most likely not to be retrieved and inserted into the Syntax interface with PF/LF. The resulting structure for the utterance with' the omission is schematized by the following tree. (34) Structure with Phonological Matrices for Sentence (25a)  (eats)  [graendmA6dr] (grandmother)  The determiner is omitted while the other more Robust categories are retrieved.  5.2.4 Accounting for Substitutions  The Principle of Robustness determines the likelihood of retrieval. But what determines substitutions? In the preceding chapter I established that substitutions involved manipulating agreement (person, number, gender), Case and tense features.  Moreover, I argued that these  features are not randomly manipulated. Rather, for any given substitution a single feature is changed. This restricted pattern can only be accounted for by a model that incorporates paradigm structures for agreement, Case and tense. To account for these facts, I propose that Lexical Items within the Lexicon are organized in inflectional paradigms. The process of Feature Matching scans these paradigms in order to pair the tripartite (semantic, intrinsic and optional) feature  230  bundle with a phonological matrix.  Then Phonological Matrix Insertion copies the selected  matrix and inserts it into the feature bundle, which Lexical Retrieval introduces into the Syntax interface with PF/LF.  Agrammatic substitutions then result from an impaired Phonological  Matrix Insertion mechanism.  5.2.4.1 Paradigms in the Lexicon  What is a paradigm? A paradigm is multi-dimensional list of linguistically related forms. Within Generative Grammar paradigms have, for the most part, been reduced to a display method, with no theoretical status. Contrary to this trend, several linguists have proposed that linguistic structures are organized in paradigms (Halle, 1973; Williams, 1981; Bybee, 1985, 1988, 1991; Carstairs, 1987, 1991; 1994; Wurzel, 1987, 1990; Matthews, 1991; Postma, 1992; Rohrbacher, 1994; Thompson, 1995).  Specifically, paradigmatic structures govern the  architecture of the Lexicon. As stated in the earlier definition, a paradigm consists of several dimensions. Each dimension corresponds to a syntactic feature: Person, Number, Gender, Case, Finite and Non-Finite. Each dimension has a range of corresponding values. A given language can chose from the range. The dimension of Person has the values 1, 2 and 3 (for the languages under study). The dimension of Number has the values Singular (Sg) and Plural (PI). Gender has the values M(asculine), F(eminine), N(euter), or a combination of these three. The range of values available to the dimension of Case is Nom(inative), Acc(usative), Dat(iye) and Gen(itive). The dimension Finite has the values Present (Prs) and Past (Pst). And Non-Finite has the values Part(iciple) and Inf(initive).  A paradigm for a given syntactic category combines these  dimensions. For example, pronouns display a Person-Number paradigm. I illustrate this for English in the table below.  231  Table 29: English Pronoun Paradigm Sg  PI  I  we  1 2 3  you s/he  they  For English pronouns, l S g , 1P1, 3Sg and 3P1 each has its own cell. Second Person has but a single cell containing you for both plural and singular. Which dimensions combine for which categories is determined on a language-bylanguage basis. For example, for the 5 languages under study, the dimension Person and Finite interact, but Person and Non-Finite do not. We find verbal agreement in tensed verbs, but not in untensed verbs. In contrast, Portuguese displays agreement in infinitival verbs, revealing that the dimensions Person and Non-Finite can interact. There are two important characteristics of substitution to consider. First, substitutions are not cross-categorial. Second, substitutions are subject to the Single Feature Constraint. If the Lexicon is organized in paradigms, these characteristics of substitution are accounted for.  5.2.4.2  Deriving Substitutions  I now trace the process of substitution from beginning to end. I argued that substitutions result from impairment to the process of Phonological Matrix Insertion within a Split Lexical Insertion model. The derivation I present below thus assumes such a model (as described in the section 5.2.2.2). Consider the sentences in (35). The sentence in (35a) contains a substitution, the copula. The sentence in (35b) does not.  232  (35) Sentences for substitution derivation a. your ears is larger b. your ears are larger Both sentences require the following L i s . (36) Lis for the Items that Project the Sentences in (35) [Possession] [Definite referent]  YOUR:  tjor]  [2nd Person] D [0 Number] [0 Gender]  EARS:  [ir]  [Plural] [0 Case]  IS:  [iz]  [3rd Person] [Singular] [0 Gender] [Present]  [Existence] [+Finite]  ARE:  [ar]  [3rd Person] [Plural] [0 Gender] [Present]  [Existence] [+Finite]  TENSE:  [0]  [0 Person] [0 Number] [0 Gender]  LARGER:  [larjdr]  [0 Person] [0 Number] [0 Gender] [0 Case]  N [3rd Person]  [Auditory] [Body part]  [+Finite] T [Assign Nom] [Present] [Size] [Big] [Comparative]  The derivation of the sentences in (35) begins with the mechanisms within the Lexicon. In the Lexicon, the inherent specifications of the L i s include the semantic features and the intrinsic F-features. The Base feature bundle for each of the L i s in (36) are displayed below. The left-hand set of features are the intrinsic F-features, and the right-hand set are (an approximation of) the semantic features.  233  (37) Base Feature Bundle for Lexical Items in (36)  D  Possession Definite referent  N 3rd Person  Auditory Body part  YOUR  EARS:  IS:  Existence +Finite  ARE:  Existence +Finite  TENSE:  T Assign Nom  +Finite Present  Size Big Comparative  LARGER:  Still within the Lexicon, the process of Optional Feature Insertion adds to this Base the optional F-features. The resulting tripartite feature bundles for all the relevant Lis are presented below.  234  (38) Tripartite Feature Bundle for Lexical Items in (36) YOUR  2nd Person  D  Possession Definite referent  EARS Plural  N 3rd Person  Auditory Body part  IS Existence +Finite  3rd Person Singular Present  ARE Existence +Hnite  3rd Person Plural Present TENSE Referential Assign Nom  +Finite Present  LARGER Size Big Comparative  235  As with the omission derivation in the previous section, the process of Initial Insertion inserts this tripartite feature bundle into the syntactic component. The categorial features project their X-bar structure.  The resulting structure is the following.  The only verbal functional  categories I include is TP. The structure may be expanded to include AgrSP, AgrOP, CP, etc. (39) Projected Syntactic Structure for Sentences in (35) TP  T [Present]  VP V V 'BE'  AP DP D'  D YOUR  A' A LARGER NP N' N EARS  Within the syntactic component movement takes place to check (optional) features and for Case assignment. The subject DP Y O U R EARS raises to [Spec TP] in order to be assigned nominative Case (from T). The copula raises to pick up the T affix and to check its (strong) features. This derivation is schematized in the tree below. (40) Derived Structure for the Sentences in (35) TP D P ' " ^ ^ ^ ^ Y O U R EARS  [V+T] 'BE'  T' VP V AP A' A LARGER  236  The words in upper case indicate the tripartite feature bundles schematized in (38). What is missing are the phonological matrices. To retrieve the appropriate phonological matrix for each LI, the Feature Matching mechanism picks and scans the relevant paradigm and pairs the tripartite feature bundle with the appropriate phonological matrix. If the derivation is free of omission, as in (35b), Phonological Matrix Insertion copies and inserts the appropriate phonological matrix into the tripartite feature bundle. Lexical Retrieval then inserts the complete feature bundles into the Syntax interface with PF/LF. The resulting structure is schematized below. (41) Structure with Phonological Matricesfor Sentence (35b) TP D P ^ ^ ^ ~ ^ ^ ^ [jor][irs] (you ears)  T  [V+T] [ar] (are)  VP V AP A' A [larjdr] (larger)  If however substitution is in evidence, as in (35a), the process of Phonological Matrix Insertion copies and inserts the wrong phonological matrix into the tripartite feature bundle. Consider the paradigm in question. Table 30: English Present Tense Copula Paradigm  Sg  PI  1  am  are  2  are  are  3  is  are  237  The target is are. Phonological Matrix Insertion copies the phonological matrix from the wrong cell. PMI can copy either am or is. In this example, is is copied and inserted into the tripartite feature bundle. Lexical Retrieval then inserts the complete feature bundles into the Syntax interface with the phonological component.  The result is a substitution error, as  schematized below. (42) Structure with Phonological Matrices for Sentence (35a) TP DP [jor] [irs] (you ears)  [V+T] [IZ] (is)  Substitutions display the following two characteristics. Substitutions shift by a single feature, and they do not cross categories. The process of Feature Matching is not impaired for substitution. This process picks the appropriate paradigm, which ensures that the category that is produced is the same as that of the target, in fact within the same paradigm as the target. Feature Matching also scans the paradigms for the target. This means that the copying and insertion mechanism that is impaired (Phonological Matrix Insertion) is located at the target.  PMI is  impaired so copies and inserts the wrong phonological matrix. However, since it is located at the target, the cell it copies is a neighbor to the target. Since the phonological matrices are organized in paradigms, the neighboring cells differ by a single feature. The result is a single feature shift.  238  5.3  T H E RELATIONSHIP B E T W E E N OMISSION A N D SUBSTITUTION  Agrammatic production includes both omission and substitution errors.  Are these two  types of errors distinct? Or do they result from the same deficit? There are three reasons to conclude that omission and substitution are distinct. First, there are the obvious production characteristics: omission involves the lack of a word, whereas substitution involves the presence of the wrong form of a word. Second, omission rates (tend to) reflect the level of severity of the agrammatic deficit. Substitution rates do not. Third, omission rates follow a set hierarchy: nouns are best retained, then verbs, then determiners, with auxiliaries and modals trailing last. No such hierarchy is present in substitution rates.  5.3.1 Substitution Rates Do Not Reflect Severity of Deficit  The first observation is that substitution rates, unlike omission rates, do not remotely reflect the level of agrammatic severity. The following table compares average substitution rate and average omission rate.  239  Table 31: Agrammatic Substitution & Omission Rates Substitution  Omission  EF  6.3%  6.8%  EE  6.1%  40.1%  DB  0.9%  47.0%  DH  1.1%  38.8%  GB  6.1%  9.3%  GM  6.4%  17.0%  FC  5.1%  19.1%  FA  2.4%  4.1%  IR  3.6%  23.7%  TV  6.4%  9.4%  Recall that higher omission rates (tend to) match a more severe level of impairment. The substitution rates do not match this trend.  First, substitution rates are for the most part  considerably lower than omission rates. Second, there is not as much of a difference between the two agrammatic speakers for each language. Third, any difference between speaker does not necessarily match the direction of the severity.  5.3.2 There Is No Hierarchy of Substitution  Omission rates pattern into a general retention hierarchy. I repeat the hierarchy below.  240  (43) Retention Hierarchy N > V > D > Aux/Modal Substitution rates do not display this pattern. In fact, they do not readily display any pattern. Table 32: English Agrammatic Substitution Rates EF  EE  Tokens  Context  %  Tokens  Context  %  N  1  92  1.1%  4  139  2.9%  V  11  69  15.9%  10  51  19.6%  D  1  99  1.0%  2  105  1