UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Underspecification, parameters, and the acquisition of vowels Fee, Elizabeth Jane 1991

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

Item Metadata

Download

Media
831-UBC_1991_A1 F43.pdf [ 19.14MB ]
Metadata
JSON: 831-1.0100444.json
JSON-LD: 831-1.0100444-ld.json
RDF/XML (Pretty): 831-1.0100444-rdf.xml
RDF/JSON: 831-1.0100444-rdf.json
Turtle: 831-1.0100444-turtle.txt
N-Triples: 831-1.0100444-rdf-ntriples.txt
Original Record: 831-1.0100444-source.json
Full Text
831-1.0100444-fulltext.txt
Citation
831-1.0100444.ris

Full Text

UNDERSPECIFICATION, PARAMETERS, AND THE ACQUISITION OF VOWELS By ELIZABETH JANE FEE B.A., The University of B r i t i s h Columbia, 1978 M.A., The University of B r i t i s h Columbia, 1980 A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Department of Linguistics) We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH COLUMBIA October 1991 (c)Elizabeth Jane Fee, 1991 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of. ilrJ^U-t^Ti C C The University of British Columbia Vancouver, Canada Date DE-6 (2/88) ABSTRACT The goal of t h i s thesis i s to develop a parametric model of a c q u i s i t i o n which incorporates the idea that phonological systems are underlyingly unspecified for c e r t a i n feature values. I examine two variants of t h i s model: one based on the theory of Radical Underspecification (Archangeli and Pulleyblank 1986), and one based on the theory of Contrastive Underspecification (Steriade 1987). I assume the p r i n c i p l e s and parameters framework, where the i n i t i a l phonological system of the c h i l d i s assumed to be characterized by the unmarked parameter settings of UG. The two types of parameters that are examined i n d e t a i l are featural parameters and r u l e parameters. The unmarked settings of featural parameters are supplied by universal redundancy r u l e s . In most cases, the unmarked settings of rule parameters are assumed to be OFF, or non-application. I provide analyses of the v o c a l i c systems of Hungarian and Spanish, based on the parametric theories of Radical and Contrastive Underspecification, which demonstrate that c e r t a i n phonological parameters i n these languages must be reset to the marked option. The Hungarian analyses focus p a r t i c u l a r l y on spreading processes, while those i n Spanish focus on alternations that take place within verb conjugation classes. Given the differences between the i n i t i a l c h i l d state and the adult phonological systems of Hungarian and Spanish, the underspecification a c q u i s i t i o n models make c e r t a i n predictions i i i regarding acquisition in these languages. These predictions are then tested using data from children acquiring both Hungarian and Spanish. The early phonological systems of children acquiring Hungarian and Spanish are found to i n i t i a l l y be smaller than predicted by either acquisition model. To account for these results, and s t i l l maintain a parametric model, I propose a theory of feature a v a i l a b i l i t y , which specifies the order in which features may become part of a child's phonological system. In conjuction with this theory of feature av a i l a b i l i t y , the RU model is able to explain the development of children's early phonological inventories, as well as certain substitution patterns. The contrastive specifications required by the theory of CU cannot account for these aspects of the data. i v TABLE OF CONTENTS ABSTRACT i i ACKNOWLEDGEMENT v i i i Chapter 1 INTRODUCTION 1 1.1 Overview 4 1.1.1 An Acquisi t i o n Model 4 1.1.2 Components of the Acq u i s i t i o n Model 7 1.1.2.1 Underspecification Theory 7 1.1.2.2 Organization of UG 9 1.1.2.3 Constraints 12 1.1.2.4 The Nature of Phonological Development 13 1.1.3 Predictions 16 1.1.4 Some General Findings 17 1.2 Organization 19 Chapter 2 PARAMETRIC THEORIES OF UNDERSPECIFICATION 22 2.1 Aspects of Phonology 25 2.1.1 Lex i c a l Phonology 25 2.1.1.1 The Lexicon 26 2.1.1.2 The Cycle 27 2.1.1.3 P o s t - l e x i c a l Component 30 2.1.1.4 Structure Preservation 31 2.1.2 Representations 33 2.1.2.1 Non-linear Phonology 33 2.1.2.2 Feature Geometry 36 2.1.2.3 The OCP 37 2.1.2.4 Metrical Theory 39 2.1.3 Rules 47 2.2 Underspecification Theory 50 2.2.1 Radical Underspecification 55 2.2.1.1 Elimination of Redundancy 55 2.2.1.2 Rule Types 58 2.2.2 Contrastive Underspecification 62 2.2.2.1 Specifications 62 2.2.2.2 R-Rules 64 2.2.2.3 Phonological Rules / 67 2.2.3 Alternate Theories of Feature S p e c i f i c a t i o n 68 2.2.3.1 Clements (1987) 69 2.2.3.2 P r i v a t i v e Features 70 2.3 P r i n c i p l e s and Parameters 72 2.3.1 B i n a r i t y 74 2.3.2 Multiple Parameters 75 2.3.3 Non-parametric Acqui s i t i o n 77 2.4 L e a r n a b i l i t y 78 2.4.1 L e a r n a b i l i t y Condition 79 2.4.2 Continuity Assumption 80 2.4.3 No-negative Evidence Hypothesis 81 V 2.5 Parametric Approaches to Underspecification 83 2.5.1 Parameters and Radical Underspecification 85 2.5.1.1 UG i n RU 85 2.5.1.2 Input i n RU 101 2.5.2 Parameters and Contrastive Underspecification 103 2.5.2.1 UG i n CU 103 2.5.2.2 Input i n CU 114 Notes to Chapter 2 116 Chapter 3 THE VOCALIC SYSTEMS OF HUNGARIAN AND SPANISH 120 3.1 Harmony Systems i n Hungarian 123 3.1.1 Hungarian Vowels 123 3.1.2 Parametric RU Analysis of Hungarian Harmony 126 3.1.2.1 Back Harmony 131 3.1.2.2 Round Harmony 143 3.1.2.3 Low Front Vowel Formation 155 3.1.3 Parametric CU Analysis of Hungarian 156 3.1.3.1 Back Harmony 161 3.1.3.2 Round Harmony 169 3.1.3.3 Peripheral Rules 175 3.1.4 Summary and Comparisons 177 3.2 Spanish Vocalic Alternations 184 3.2.1 Spanish Vowels and Stress 185 3.2.2 Parametric RU Analysis of Spanish 188 Alternations 3.2.2.1 Spanish Verbal Classes 190 3.2.2.2 Alternating Vowel/Diphthongs 200 3.2.2.3 High/Mid Alternations 215 3.2.2.4 Peripheral Rules 226 3.2.3 Parametric CU Analysis of Spanish Alternations 229 3.2.3.1 High/Mid Alternations 231 3.2.3.2 Vowel/Diphthong Alternations 234 3.2.3.3 Peripheral Rules 241 3.2.4 Summary and Comparisons 243 Notes to Chapter 3 249 v i Chapter 4 A PARAMETRIC ACQUISITION THEORY 253 4.1 Aspects of Phonological Development 256 4.1.1 Speech Perception 256 4.1.2 Speech Production 258 4.1.2.1 Stages of Phonological Development 258 4.1.2.2 Consonantal Inventories 262 4.1.2.3 Natural Processes 265 4.1.2.4 Early S y l l a b l e Structures 267 4.1.3 Some General Issues 271 4.1.3.1 V a r i a b i l i t y 271 4.1.3.2 Levels of Organization 273 4.1.3.3 Units of Organization 277 4.1.3.4 Phonological Rules 279 4.1.4 Implications for the Ac q u i s i t i o n of Vocalic Systems 281 4.2 Assumptions for a Model of Phonological A c q u i s i t i o n 282 4.2.1 Perception 283 4.2.2 Phonological Development 284 4.2.2.1 Featural Parameters 284 4.2.2.2 Rule Parameters 285 4.2.2.3 Other Parameters 286 4.2.3 Substitution Processes 288 4.2.3.1 Paradigmatic Substitutions 290 4.2.3.2 Syntagmatic Substitutions 292 4.2.4 Summary 300 4.3 Predictions for the Acq u i s i t i o n of Hungarian and Spanish 302 4.3.1 RU Predictions 304 4.3.1.1 Hungarian 306 4.3.1.2 Spanish 313 4.3.2 CU Predictions 317 4.3.2.1 Hungarian 319 4.3.2.2 Spanish 324 4.3.3. Comparison of RU and CU Predictions 328 4.3.3.1 Hungarian 328 4.3.3.2 Spanish 332 4.3.3.3 Summary 336 Notes to Chapter 4 341 Chapter 5 VOCALIC ACQUISITION IN HUNGARIAN AND SPANISH 347 5.1 Phonetic Inventories and Phonological Contrasts Methodology 347 5.1.1 Phonetic Inventories 347 5.1.2 Substitutions 352 5.1.3 Phonological Inventories 353 5.2 Vocalic A c q u i s i t i o n i n Hungarian 355 5.2.1 Samples 356 5.2.2 Hungarian Vowels 357 5.2.3 Phonological Inventories 360 5.2.4 Mismatches 364 5.2.5 Substitution Patterns 372 5.2.5.1 Syntagmatic Substitutions 372 5.2.5.2 Paradigmatic Substitutions 375 v i i 5.3 Vocalic A c q u i s i t i o n i n Spanish 377 5.3.1 Samples 377 5.3.2 Spanish Vowels 379 5.3.3 Phonological Inventories 380 5.3.4 Mismatches 383 5.3.5 Substitution Patterns 389 5.3.5.1 Syntagmatic Substitutions 389 5.3.5.2 Paradigmatic Substitutions 392 5.4 Summary 392 Notes to Chapter 5 396 Chapter 6 TESTING THE PREDICTIONS 398 6.1 Predicted vs. Attested Phonological Inventories 399 6.1.1 The Inventories 400 6.1.2 Gradual A c q u i s i t i o n of Complex Vowels 403 6.1.3 Gradual A c q u i s i t i o n of the Inventory 405 6.1.3.1 Jakobson and Halle (1956) 407 6.1.3.2 Calabrese (1988) 411 6.1.3.3 A Theory of Feature A v a i l a b i l i t y 413 6.2 Inventories 416 6.2.1 Hungarian 416 6.2.1.1 RU Account 419 6.2.1.2 CU Account 425 6.2.2 Spanish 428 . 6.2.2.1 RU Account 428 6.2.2.2 CU Account 431 6.3 Substitution Patterns 433 6.3.1 Syntagmatic Substitutions 434 6.3.1.1 C r o s s - l i n g u i s t i c S i m i l a r i t i e s 434 6.3.1.2 C r o s s - l i n g u i s t i c Differences 437 6.3.2 Paradigmatic Substitutions 439 6.3.2.1 Hungarian 439 6.3.2.2 Spanish 441 6.4 Rules 442 6.4.1 Hungarian 443 6.4.2 Spanish 448 6.5 Summary 455 6.6 Implications 462 6.6.1 Implications for Phonological Theory 464 6.6.2 Implications for a Theory of Phonological Ac q u i s i t i o n 466 Notes to Chapter 6 473 REFERENCES 475 v i i i ACKNOWLEDGEMENT I would l i k e to thank the members of my thesis committee, David Ingram, P a t r i c i a Shaw and Ewa-Czaykowska-Higgins, for t h e i r comments, c r i t i c i s m s and encouragement throughout the writ i n g of t h i s t h e s i s . Writing a long-distance d i s s e r t a t i o n i s never easy, but the fact that the members of my committee were always avai l a b l e has made i t less d i f f i c u l t than i t could have been. David Ingram has been a friend, confidant and source of never-ending support since I f i r s t began my studies at U.B.C. David sparked my o r i g i n a l i n t e r e s t i n language a c q u i s i t i o n , and has kept the f i r e s going over the many years I have known him. He has helped to shape many of the ideas presented here and has read every word that has gone into t h i s t h e s i s , often three or four times. My i n t e r e s t i n non-linear phonology was i n i t i a t e d by Pat Shaw, and I have continued to p r o f i t from her insights over the years. Pat's comments and c r i t i c i s m s of the various drafts of t h i s thesis have always been exacting and constructive. She has been a valuable resource for someone who writes and thinks i n a rather impulsive manner. This thesis could not have been completed without the help of Ewa Czaykowska-Higgins. Ewa, perhaps because of a f a i r l y recently completed d i s s e r t a t i o n , always seemed to know when I needed a push and when I needed a kind word. She has been p a r t i c u l a r l y h e l p f u l i n pointing out where my assumptions were i n v a l i d or unsupported, and where the a c q u i s i t i o n sections needed to be expanded or revised so they could be understood by phonologists. I have p r o f i t e d greatly from the l i n g u i s t i c s community at U.B.C. Toni Borowsky was helpful i n much of the groundwork for t h i s t h e s i s . M.D. Kinkade has always been a v a i l a b l e for administrative help, and i n the early stages of my degree was a valuable source of information on anything to do with fieldwork. Bernard St-Jacques and Mike Rochemont provided useful advice on some of the more bureaucratic aspects of f i n i s h i n g a degree. The members of the student population at U.B.C. have provided help and encouragement over the years. Susan Blake, Kathy Hunt, Bruce Bagemihl, David M i l l a r d , Henry Davis and Nicola Bessell have provided useful discussion of many of the ideas presented here, as well as moral support when the going got rough. The phonologists at McGill, including Glyne Piggott, Jose T o u r v i l l e and Dominique Rodier, were supportive and f r i e n d l y to a stranger i n a strange land. Glyne, e s p e c i a l l y , helped me to integrate into the Montreal academic community, and provided me with a forum for some of work presented here. Perhaps the most f r u i t f u l period of my student career occurred when I attended the 1987 L.S.A. Summer I n s t i t u t e at Stanford University. The contacts and friends I made there, as well as the ambiance of Stanford i t s e l f , made for a ix productive and i n t e r e s t i n g summer. Courses I took from John McCarthy, John Goldsmith and Nick Clements helped me to decide the path of my d i s s e r t a t i o n research. My family (and more recently my extended family) have been of invaluable support over the years. My parents i n s t i l l e d i n me the b e l i e f that i t was possible to do anything, so long as one wanted i t badly enough. The encouragement and support of my parents, s i s t e r s , brothers, and in-laws i s both acknowledged and appreciated. My sons, Patrick and Christopher Lougheed, have watched t h i s thesis develop and evolve. They have grown up with the physical and mental debris of academia and have had to put up with a mother who was not always there (be i t mentally or p h y s i c a l l y ) . While there may have been times that they d i d not understand, t h e i r presence and love has always helped me to keep at l e a s t one foot i n the r e a l world. Last, but by no means lea s t , I would l i k e to thank Tom. He has been i n c r e d i b l y supportive and patient and has always been there when I needed someone to l i s t e n , someone to argue with, or simply a shoulder to cry on. This thesis i s dedicated, with love, to him. 1 CHAPTER 1 Introduction There i s an abundance of evidence i n c r o s s - l i n g u i s t i c phonological research that not a l l segments are f u l l y s p e c i f i e d for a l l feature values underlyingly. There i s l i t t l e agreement, however, as to exactly which features may be underspecified or under what conditions underspecification may occur. The theory of Radical Underspecification, outlined i n Archangeli and Pulleyblank (1986), maintains that only non-predictable feature values are included i n underlying representations. Archangeli and Pulleyblank (1989) argue that r a d i c a l underspecification of the feature [ATR] i s required i n Yoruba, while Abaglo and Archangeli (1989) argue that a r a d i c a l l y underspecified system i s necessary i n order to account for the v o c a l i c facts of Gengbe. An alternate theory of underspecification, that w i l l be refer r e d to here as Contrastive Underspecification, i s outlined i n Steriade (1987). Contrastive Underspecification assumes that underlying representations contain only those feature values that are necessary to d i s t i n g u i s h pairs of segments i n a given language. Mester and Ito (1989) argue that contrastive underspecification must be assumed i n order to account for the facts of p a l a t a l prosody i n Japanese mimetics. In a recent d i s s e r t a t i o n , Calabrese (1988) examines both types of underspecification and suggests that there may 2 be a parameter which allows languages to choose a phonological system that i s either r a d i c a l l y underspecified or c o n t r a s t i v e l y underspecified. Given a theory of Universal Grammar (UG), children's l i n g u i s t i c systems are possible adult grammars/ and are therefore constrained by the same sets of p r i n c i p l e s as adult grammars. Data from f i r s t language a c q u i s i t i o n should then be a revealing alternate source of evidence for the v a l i d i t y of l i n g u i s t i c theories. In t h i s thesis I propose to use data from phonological a c q u i s i t i o n to test the underlying premises of Radical and Contrastive Underspecification. I attempt to determine how these competing theories must be organized i n order to represent coherent models of a c q u i s i t i o n , and I w i l l examine the predictions these theories make for the a c q u i s i t i o n of the v o c a l i c systems of Hungarian and Spanish. These predictions are then tested using real-time a c q u i s i t i o n data. The task the c h i l d faces i n acquiring the grammar of a language can be shown as i n (1.1). (1.1) Input ;-> Grammar The c h i l d must determine, on the basis of the input received, the form of the grammar. When the input i s examined c l o s e l y , the schematization of a c q u i s i t i o n i n (1.1) can be shown to be incomplete. White (1989), summarizing research that has been c a r r i e d out over the past 20 years, i d e n t i f i e s three problems 3 that r e l a t e to input: underdetermination, degeneracy, and the No-negative Evidence Hypothesis. The problem of underdetermination i s that many aspects of grammar are not overtly v i s i b l e i n the input. Traces, for example, which are assumed to be ah i n t e g r a l part of the syntactic representation of sentences, are abstract e n t i t i e s which are not present i n the spoken language a c h i l d hears. The degeneracy problem i s one o r i g i n a l l y addressed i n Chomsky (1965), which points out that the language ch i l d r e n hear may contain ungrammatical or incomplete utterances. How i s i t possible for a c h i l d to determine which sentences are grammatical, and therefore generated by the grammar, and which are the r e s u l t of performance errors? The No-negative Evidence problem points out that our grammatical competence allows us to make judgments about the ungrammaticality of sentences as well as t h e i r grammaticality. If we assume (as i s generally done i n t h i s type of research) that c h i l d r e n learn only through p o s i t i v e evidence, how can they possibly learn that c e r t a i n construction types are not generated by the grammar? These problems demonstrate that children must a t t a i n the complex adult grammar using input that does not provide a l l the clues necessary for the language being acquired. This has been referred to as the projection problem or the l o g i c a l problem of language a c q u i s i t i o n . In an attempt to solve the projection problem, researchers such as Chomsky (1981a,b) have posited a mediating component to (1.1) c a l l e d Universal 4 Grammar or UG. (1.2) Input > UG > Grammar UG i s our s p e c i e s - s p e c i f i c endowment for language, which constrains the form of possible human grammars. The conception of a c q u i s i t i o n i n (1.2) has led to two consequences for l i n g u i s t s interested i n a c q u i s i t i o n research. F i r s t has been the creation of the f i e l d of l e a r n a b i l i t y , which investigates the l o g i c a l problems r e l a t i n g to language a c q u i s i t i o n . This f i e l d attempts to explain how the c h i l d , given an incomplete and imperfect set of input s t r i n g s , can achieve the adult grammar, i . e . to specify what form the input and UG components of (1.2) must take i n order that the grammar be attainable. Secondly, the schema i n (1.2) has led to the acceptance of a c q u i s i t i o n data as an important and useful source of evidence for the t e s t i n g of a l t e r n a t i v e l i n g u i s t i c theories. 1.1 Overview 1.1.1 An A c q u i s i t i o n Model In t h i s thesis I develop a model of a c q u i s i t i o n that i s based on the assumption that phonological systems are underlyingly unspecified for c e r t a i n feature values. I examine two variants of t h i s model: one that i s based on the assumption that underlying representations lack a l l predictable information, and one that i s based on the 5 assumption that underlying representations lack only non-contrastive information. I assume that t h i s a c q u i s i t i o n model i s organized according to the p r i n c i p l e s and parameters model of grammar o r i g i n a l l y proposed i n Chomsky (19813,13). The two variants of the a c q u i s i t i o n model make ce r t a i n predictions regarding the a c q u i s i t i o n of phonological systems, and I w i l l t e s t these predictions using a c q u i s i t i o n data from Hungarian, a Finno-Ugric language, and Spanish, a Romance language. In order to r e s t r i c t my task somewhat, I have chosen to focus on v o c a l i c systems, because t h i s i s where the majority of work i n underspecification has taken place. I also r e s t r i c t my investigations to the types of featural and rule parameters that are required given an RU or CU a c q u i s i t i o n model. While parameters have been proposed for many other aspects of phonological organization, these are the two sets of parameters that are the most c r u c i a l to a theory of underspecification. The underlying theme of t h i s thesis i s very s i m i l a r to that of Calabrese (1988). Calabrese develops a hierarchy of universal f i l t e r s to explain c e r t a i n facts of phonological a c q u i s i t i o n and foreign language transfer. Calabrese argues that UG consists of a set of f i l t e r s , or negative constraints, which are h i e r a r c h i c a l l y ordered. In the a c q u i s i t i o n of a language, children may learn a segment which v i o l a t e s a c e r t a i n f i l t e r i f segments v i o l a t i n g f i l t e r s further down the hierarchy have already been acquired. My objectives d i f f e r from Calabrese's i n a number of 6 c r u c i a l ways. F i r s t , I have chosen to use real-time a c q u i s i t i o n data to test the predictions of the parametric theories of RU and CU. This has been done so that cross-l i n g u i s t i c differences and relationships between adult phonological systems and a c q u i s i t i o n can be c l e a r l y demonstrated. Calabrese attempts only to account for some very general aspects of the Jakobsonian picture of phonological development, which sometimes s u f f e r from having been misinterpreted over the years (see Ingram 1989a). Secondly, while Calabrese attempts only to account for the development of phonological inventories, I attempt to account for the types of substitution processes and phonological rules that young children use, as well as the development of t h e i r i n i t i a l inventories. The a c q u i s i t i o n data that are analyzed i n t h i s thesis come from previously published studies of young childre n acquiring Hungarian and Spanish. The choice of languages studied was made by force rather than by choice. Although there i s a large inventory of c r o s s - l i n g u i s t i c a c q u i s i t i o n samples to choose from, I wanted to use only languages where data from several d i f f e r e n t children was av a i l a b l e , i n order to be c e r t a i n that the samples were t r u l y representative. I also wanted to choose languages for which previous underspecification analyses of the voc a l i c systems had been proposed. These two c r i t e r i a narrowed down the choice of languages considerably. 7 1.1.2 Components of the Acquisition Model I have dual objectives i n writing t h i s t h e s i s . F i r s t , I wish to examine the l e a r n a b i l i t y aspects of the theories of Radical and Contrastive Underspecification, looking at the innate mechanisms that must be attributed to the c h i l d and the learning procedures that are required to a t t a i n the adult language. Secondly, I wish to examine phonological cross-l i n g u i s t i c a c q u i s i t i o n data i n order to determine whether the underspecification analyses can c o r r e c t l y account for the patterns of development. In developing an a c q u i s i t i o n model which allows me to investigate these issues I have made ce r t a i n assumptions regarding 1) how underlying representations are structured, 2) the organization of UG, 3) the constraints that hold i n an a c q u i s i t i o n model such as t h i s , and 4) the nature of phonological development. Some of these issues are addressed i n the following sections. 1.1.2.1 Underspecification Theory The a c q u i s i t i o n model developed i n t h i s thesis adopts the basic premise that a l l features have binary values, but that underlying representations lack c e r t a i n feature values. Two variants of t h i s model are investigated: one based on the theory of Radical Underspecification, and the other based on the theory of Contrastive Underspecification. Radical Underspecification (RU) i s based on the notion of minimal redundancy developed i n Kiparsky (1982, 1985) and i s outlined i n d e t a i l i n Archangeli and Pulleyblank (1986). In t h i s 8 theory i t i s assumed that only non-redundant feature values e x i s t i n underlying representations. Redundant or predictable feature values are inserted by redundancy rules at some point during the l e x i c a l or p o s t - l e x i c a l phonology. Contrastive Underspecification (CU), sometimes referre d to as Restricted Underspecification (Mester and Ito 1989), i s developed from the theory of feature s p e c i f i c a t i o n s used i n Halle (1959) and i s outlined i n Steriade (1987). CU i s also discussed, i n a somewhat revised form, i n the works of Clements (1988), Christdas (1988) and Calabrese (1988). Steriade's theory assumes that only contrastive feature values are marked underlyingly, and that non-contrastive feature values are inserted l a t e i n the derivation by redundancy ru l e s . The theories of underspecification that are examined here make d i f f e r e n t claims about how children specify t h e i r phonological inventories, and about the ro l e that redundancy rules play i n phonological systems. An a l t e r n a t i v e conception of the s p e c i f i c a t i o n of underlying feature values assumes that features are p r i v a t i v e , or have only a single functional value. Den Dikken and van der Hulst (1990) argue that a l l features are p r i v a t i v e , while Steriade (1987), Mester and Ito (1989) and Piggott (1990, to appear) argue that c e r t a i n features are best viewed as being p r i v a t i v e . A p r i v a t i v e or unary feature can have only one possible marked value i n any language, and the unmarked value i s never s p e c i f i e d , even at some late point i n the phonology. The data presented i n Chapter 3 for Hungarian present a 9 challenge to a theory which assumes that a l l features are p r i v a t i v e , because there i t i s argued that [-round] must be the l e x i c a l l y s p e c i f i e d value of Hungarian. I believe that cases l i k e t h i s , where a un i v e r s a l l y redundant value can be shown to be marked underlyingly on a language-specific basis, w i l l help to demonstrate that some or a l l d i s t i n c t i v e features have binary values available for manipulation. 1.1.2.2 Organization of UG I assume that UG consists of a set of universal p r i n c i p l e s , and a set of parameters. P r i n c i p l e s are those aspects of UG that are held constant across a l l languages, while parameters are p r i n c i p l e s which have several s p e c i f i e d options. Syntactic parameters have been proposed to account for differences between languages i n the use of "empty" subjects (the Pro-drop parameter, see Hyams 1983 and Wexler and Manzini 1987), subject-aux inversion (Davis 1987), binding theory (Solan 1987) and subjacency ( R i z z i 1982). Phonological parameters have more recently been proposed to account for d i r e c t i o n a l i t y and maximal/minimal e f f e c t s of the feature hierarchy (Archangeli and Pulleyblank 1986, Piggott, to appear), feature hierarchitecture (Piggott, to appear), branching p o s s i b i l i t i e s of s y l l a b i c constituents (Kaye 1987) and stress placement (Hayes 1981, Dresher and Kaye 1988). While suggestions have been made for parameters with multiple settings (e.g. Wexler and Manzini 1987), I assume that the parameters that are required i n an a c q u i s i t i o n model 10 based on underspecification theory have only binary options. I show that universal redundancy rules can be viewed as the unmarked settings of featural parameters, although languages may choose to reset featural parameters to the marked option. The unmarked set t i n g of featural parameters i n i t i a l l y constrain the s p e c i f i c a t i o n of a l l children's phonological inventories, and consequently, i n languages which require the marked s e t t i n g of a featural parameter, the i n i t i a l system hypothesized by the c h i l d w i l l be d i f f e r e n t from the adult system. This parametric model can help to explain why many ac q u i s i t i o n researchers have noted that c r o s s - l i n g u i s t i c a l l y children's early phonological systems are very s i m i l a r , and only l a t e r take on language-particular q u a l i t i e s . In the p r i n c i p l e s and parameters model of grammar i t i s assumed that children's grammars may d i f f e r from adults only i n the parameter settings that characterize them. This model then places very severe r e s t r i c t i o n s on the types of phonological systems that can be attributed to c h i l d r e n , and on the developments that can occur i n the attainment of the adult grammar. I believe that t h i s i s an important and necessary constraint on a c q u i s i t i o n models, because for f a r too long a c q u i s i t i o n research has ignored advances i n l i n g u i s t i c theory and has attempted to e x i s t as an independent f i e l d . I believe that t h e o r e t i c a l l i n g u i s t i c s and a c q u i s i t i o n research can both benefit from closer t i e s between these f i e l d s . A c q u i s i t i o n research can help to determine which are the unmarked parameter settings, while t h e o r e t i c a l l i n g u i s t i c s 11 can provide a model against which we.can t e s t a c q u i s i t i o n data. In the course of t h i s thesis I examine parameters that r e l a t e to 1) redundancy rules, 2) phonological r u l e s , and 3) complex vowels. The f i r s t two types of parameters are discussed i n d e t a i l i n Chapter 2, as s p e c i f i c to theories of underspecification. The parameter r e l a t i n g to complex vowels i s not a p a r t i c u l a r feature of an underspecification model of a c q u i s i t i o n , but i s required to account for the a c q u i s i t i o n of long vowels i n Hungarian and diphthongs i n Spanish. The parameters that r e l a t e to phonological rules are i d e n t i c a l i n RU and CU. Based on work i n current non-linear phonology I assume that rule parameters allow for the spreading, deletion or i n s e r t i o n of phonological elements. The unmarked se t t i n g of a l l rule parameters i s OFF or non-a p p l i c a t i o n . In t h i s thesis I discuss how spreading rules can be used to describe what have been c a l l e d 'reduplications* i n c h i l d language. As stated above, the unmarked settings of redundancy rules are assumed to be provided by UG. In the theory of RU there are two possible types of redundancy r u l e s : context-free rules and context-sensitive rules. At the unmarked s e t t i n g , context-free rules i n s e r t u n i v e r s a l l y unmarked feature values redundantly, while at the marked set t i n g they provide the u n i v e r s a l l y marked values redundantly. At the unmarked se t t i n g context-sensitive rules provide universal feature co-occurrence r e s t r i c t i o n s , while at the marked s e t t i n g a 12 context-sensitive rule i s suppressed or eliminated i n the language-specific grammar, and the u n i v e r s a l l y redundant feature value i s marked underlyingly. In the theory of CU only context-sensitive redundancy rules are permitted. At the unmarked s e t t i n g these rules provide universal feature co-occurrence r e s t r i c t i o n s (as i n RU), while at the marked se t t i n g these redundancies do not apply and the contrastive feature values must be marked underlyingly. 1.1.2.3 Constraints The a c q u i s i t i o n model I develop i s constrained by the L e a r n a b i l i t y Condition (Pinker 1979, 1984), the Continuity Condition (Atkinson 1982, Pinker 1984), and the No-negative Evidence Hypothesis (Williams 1976, Baker 1979, Berwick 1985), which are the most commonly held assumptions i n the f i e l d of l e a r n a b i l i t y research. Both the L e a r n a b i l i t y and Continuity Conditions constrain the organization of UG. The L e a r n a b i l i t y Condition says that any developmental stage posited by an a c q u i s i t i o n theory must be attainable v i a an acquisiton mechanism that begins with UG and ends at the adult grammar. The Continuity Condition says that the p r i n c i p l e s and a c q u i s i t i o n mechanisms that are a v a i l a b l e to the c h i l d must be the same throughout the whole course of development. These two conditions are b u i l t d i r e c t l y into a p r i n c i p l e s and parameters model of phonology. Given the parameter se t t i n g account of a c q u i s i t i o n children's grammars can d i f f e r from adults' only i n the s e t t i n g of 13 parameters, and the only type of a c q u i s i t i o n that takes place i s the switching of parameters. The No-negative Evidence Hypothesis, which says that only p o s i t i v e evidence may be used i n a c q u i s i t i o n , i s also be assumed i n t h i s t h e s i s . In a parameter s e t t i n g model, t h i s constraint says that parameter r e s e t t i n g can only be triggered by p o s i t i v e evidence from the input. Given the No-negative Evidence Hypothesis, the ungrammatically of c e r t a i n representations or constructions can never be learned, but rather w i l l have to be provided for the c h i l d as a p r i n c i p l e of UG. 1.1.2.4 The Nature of Phonological Development In order to a t t a i n the phonological system of a language the c h i l d must c o r r e c t l y perceive the adult phonetic repertoire, determine the appropriate cues to attend to, and learn which sounds i n the language being acquired are used d i s t i n c t i v e l y . I make ce r t a i n assumptions about each of these steps, some of which r e s u l t from previous research into phonological development, and some of which follow d i r e c t l y from the a c q u i s i t i o n theory assumed. I assume, following work i n perception research, that chidren acquire the phonetic inventory of t h e i r language i n advance of phonological organization. In addition, following previous a c q u i s i t i o n evidence, I assume that ch i l d r e n begin to organize speech sounds into a phonological system sometime i n the f i r s t half of the second year. Given the parametric model 14 of a c q u i s i t i o n , UG w i l l provide the structure for the c h i l d ' s i n i t i a l phonological system i n the form of a set of universal p r i n c i p l e s and parameters. The ch i l d ' s f i r s t attempts at phonological organization w i l l show the e f f e c t s of the unmarked featural and rule parameters provided by UG, and po s i t i v e evidence w i l l be required to t r i g g e r the r e s e t t i n g of any one of these parameters. According to these assumptions, c r o s s - l i n g u i s t i c a l l y children's e a r l i e s t phonological systems w i l l be s i m i l a r i n that they are a l l constrained by the same set of universal p r i n c i p l e s and parameters, but may d i f f e r i n that these constraints may be applied to d i s t i n c t phonetic inventories. The parametric a c q u i s i t i o n theories based on the theories of RU and CU make very s i m i l a r claims about the types of sub s t i t u t i o n patterns that w i l l be found i n children's e a r l y speech. One type of pattern, that I c a l l paradigmatic substitutions, a r i s e when the unmarked parameter s e t t i n g provided by UG forces the c h i l d to represent d i s t i n c t i v e sounds i n a non-distinctive fashion. This type of sub s t i t u t i o n i s predicted to occur i n languages where the marked s e t t i n g of context-sensitive parameters i s c a l l e d f o r . The c h i l d ' s i n i t i a l phonological inventory w i l l be s p e c i f i e d according to the featural parameters (redundancy rules) of UG, while the language being acquired c a l l s f or a d i f f e r e n t pattern of s p e c i f i c a t i o n . In t h i s case the representations of two sounds may be collapsed and the unmarked sound w i l l be 15 produced i n place of a more marked one. Since both RU and CU allow for context-sensitive rules, and since I assume the same basic set of context-sensitive rules i n both theories, the types of paradigmatic substitutions they predict are i d e n t i c a l . D i f f erent substitution patterns are predicted for Hungarian and Spanish, however, because t h e i r phonetic inventories are quite d i f f e r e n t . I also assume, again following both previous research and the constraints of the parameter s e t t i n g model, that the structure of children's word forms i s i n i t i a l l y provided by a set of templates. These templates provide simple CV and CVCV s y l l a b l e structures to which featural information can be mapped. In c e r t a i n cases the adult target w i l l contain more information than the c h i l d can represent i n the template, and c e r t a i n elements i n the target w i l l be omitted i n the c h i l d ' s representation of that form. In other cases the f e a t u r a l information that the c h i l d can represent does not exhaustively f i l l the template, and portions of the template w i l l be underlyingly unspecified. In t h i s case I assume that the S a t i s f a c t i o n Condition (McCarthy and Prince 1986) forces the c h i l d to provide featural information for the unspecified s l o t s , e i t h e r by a paradigmatic s u b s t i t u t i o n provided by UG or through the spread of featural information from another segment or s y l l a b l e i n the word form. Substitutions that r e s u l t from a phonological r u l e f i l l i n g i n the featural information of an underlying empty s k e l e t a l s l o t are referred to as syntagmatic s u b s t i t u t i o n s . 16 Syntagmatic substitutions then occur across a word form, as opposed to paradigmatic substitutions, which are segment s p e c i f i c . I t i s predicted that both syntagmatic and paradigmatic substitutions w i l l occur when a sound i n the adult target i s not present i n the c h i l d ' s inventory, or when there i s some complexity i n the adult form that somehow int e r f e r e s with how much featural information the c h i l d can represent. 1.1.3 Predictions In Chapter 4 i t i s shown that the parametric a c q u i s i t i o n models based on the theories of RU and CU make remarkably s i m i l a r claims regarding the c h i l d ' s i n i t i a l phonological system. Both theories, for example, predict that c h i l d r e n acquiring Hungarian and Spanish w i l l i n i t i a l l y represent 5 d i s t i n c t vowels, which w i l l surface as [ i ] , [u], [e], [o] and [a], and that they w i l l represent these inventories using the features [high], [back] and [low]. Both theories p r e d i c t that c h i l d r e n acquiring Hungarian w i l l i n i t i a l l y substitute front unrounded vowels for front rounded vowels, a low back vowel for a low front vowel, and short or simple vowels f o r complex vowels. Both theories also predict s i m i l a r sets of phonological rule parameters for both Hungarian and Spanish, with the major difference being i n how these rules are assumed to operate. The two variants of the parametric a c q u i s i t i o n theory d i f f e r i n t h e i r predictions regarding the number and types of 17 featural parameters that must be reset i n order to achieve the adult systems of Hungarian and Spanish. RU predicts that 4 f e a t u r a l parameters must be reset i n the acquistion of Hungarian, while CU predicts only 2. In Spanish, RU predicts that only a single context-free parameter w i l l be reset, while CU predicts that no featural parameters w i l l be reset at a l l . When a featural parameter i s reset the c h i l d ' s phonological system w i l l be restructured to accommodate the new feature markings. RU therefore predicts that a greater number of r e s t r u c t u r i n g stages w i l l take place i n both Hungarian and Spanish than i s predicted by the theory of CU. 1.1.4 Some General Findings In both languages i t i s found that a large number of the mismatches that occur between a c h i l d ' s form and the adult target are the s u b s t i t u t i o n of a short or simple vowel for a complex one. The only predicted paradigmatic s u b s t i t u t i o n pattern that i n fact occurs i s the s u b s t i t u t i o n of [a] for /e/ i n Hungarian. In both languages a large number of mismatches are the r e s u l t of a rule of Spread f i l l i n g i n underlyingly unspecified feature values, so that both vowels i n a m u l t i s y l l a b i c c h i l d form are i d e n t i c a l on the surface. The fact that Spread i s found i n the speech of c h i l d r e n acquiring Hungarian and Spanish, even when Spanish does not appear to have a productive Spreading rule i s an i n t e r e s t i n g f i n d i n g . One of the main differences between the Hungarian and Spanish data relates to the conditions under which 18 substitutions occur. In Hungarian, i t i s found that substitutions generally occur when a sound i n the adult target i s not present i n the c h i l d ' s phonological inventory, although i n a few cases substitutions also occur when the adult form i s complex i n some way, such as containing a consonant c l u s t e r or being t r i s y l l a b i c . In Spanish, however, substitutions seem to be a function of the stress patterns of the language. Vowels whose features spread to another element i n the word form are almost e x c l u s i v e l y i n the stressed s y l l a b l e i n the adult target. It i s hypothesized that because of the complex stress system i n Spanish children mark stress i n every form. This adds to the complexity of the ch i l d ' s form, and as a r e s u l t only the fe a t u r a l information of the stressed vowel or s y l l a b l e i s mapped to the template. The early phonological systems of childre n acquiring Hungarian and Spanish are found to i n i t i a l l y be smaller than predicted by the parametric theories of underspecification. While both theories predict that children's e a r l i e s t systems w i l l contain 5 d i s t i n c t vowels, the data suggest that c h i l d r e n only gradually achieve a 5 vowel system. The f i r s t vowel i s /a/, followed by the mid vowels /e/ and /o/, then by / i / and f i n a l l y by /u/. In Hungarian the l a s t vowels added are the front rounded vowels /u7 and /o7. To account for these r e s u l t s , and s t i l l maintain the parametric model of a c q u i s i t i o n , I propose a theory of feature a v a i l a b i l i t y , which provides a basic order i n which features become av a i l a b l e to chil d r e n . This theory i s i n part based on the theory of 19 d i s t i n c t i v e features developed i n Jakobson and Halle (1956). I maintain a UG perspective of features and assume that child r e n have access to a l l d i s t i n c t i v e features innately, but the theory of feature a v a i l a b i l t y s p e c i f i e s the order i n which features can be used when the c h i l d begins to organize a phonological system. It i s found that given t h i s theory of feature a v a i l a b i l i t y , the RU model i s able to account for the development of children's early phonological inventories, while the CU model i s not. The types of s p e c i f i c a t i o n s required i n the CU model made i t impossible to predict which segments w i l l be added to the chi l d ' s inventory at a given time. I t i s also found that the type of underspecification required by the theory of RU i s able to explain the id i o s y n c r a t i c behaviour of [e] i n the Spanish data, while t h i s i s not possible given a contrastive underspecification system. I therefore conclude that an RU a c q u i s i t i o n model i s a better representation of UG than the CU model. 1.2 Organization This thesis i s organized i n the following fashion. In Chapter 2 I develop the parametric theories of Radical and Contrastive Underspecification, based on the theories outlined i n Archangeli and Pulleyblank (1986) and Steriade (1987), focussing on the type of information and feature values that are present i n underlying representations and the form of both redundancy and phonological rules. In doing so I f i r s t 20 outline some current aspects of non-linear phonology that are assumed in the course of this thesis, with a particular focus on those aspects of phonology that are assumed to be universal principles. I then discuss some of the basic assumptions of the principles and parameters model of grammar, and of the theory of learnability. Finally I examine the theories of RU and CU as parameter setting models. I describe how UG w i l l be organized given each theory, and the types of input that w i l l be required to trigger the resetting of parameters. In Chapter 3 I present certain facts regarding the vocalic systems of Hungarian and Spanish, which help to argue for a specific underspecification system. This type of argumentation is more necessary in the parametric theory of RU, since this theory allows for a greater number of language-specific choices than the parametric theory of CU. Each analysis concludes with a summary of how the grammar would be organized, given the proposed feature specifications and phonological rules. Chapter 4 presents a parametric theory of phonological acquisition. I f i r s t present some of the major findings of previous research on phonological acquisition, and then discuss the assumptions that I make regarding these findings. Finally the CU and RU predictions for the acquisition of the vocalic systems of Hungarian and Spanish are outlined, given the analyses developed in Chapter 3. Chapter 5 presents the acquisition data from Hungarian and Spanish. I present analyses of the phonological 21 inventories used by these children at two time periods, and the types of substitution patterns that occur i n the data. In Chapter 6 I look more c l o s e l y at the a c q u i s i t i o n data, with an eye to seeing how successful the parametric a c q u i s i t i o n theories of RU and CU are at capturing the a c q u i s i t i o n fa c t s . F i r s t I examine the discrepancies that e x i s t between the inventories used by the Spanish and Hungarian children and the inventories predicted by RU and CU, and show that a theory of feature a v a i l a b i l i t y can help explain the early patterns of development. I then attempt to account for the development of these early phonological inventories, the substitution patterns, and the phonological rules used by these two groups of children, given RU or CU. La s t l y , I discuss the implications these analyses have both for phonological theory and for a theory of phonological a c q u i s i t i o n . 22 CHAPTER 2 Parametric Theories of Underspecification In t h i s chapter I examine the theories of Radical and Contrastive Underspecification as representative of p r i n c i p l e s and parameters models of phonology. This w i l l be a wholly l o g i c a l enterprise; the development of a c q u i s i t i o n models capable of handling real-time a c q u i s i t i o n data w i l l be l e f t u n t i l Chapter 4. I f i r s t outline the two theories of underspecification, showing the types of features and feature values that are l e f t unspecified and how redundancy and phonological rules are dealt with. I w i l l then present some of the basic assumptions of the p r i n c i p l e s and parameters model of phonology and of l e a r n a b i l i t y theory, and demonstrate how Radical and Contrastive Underspecification may be revised to f i t i n with these assumptions. Many of these issues have already been faced i n Radical Underspecification (RU), since i t i s presented as a p r i n c i p l e s and parameters model of phonology i n Archangeli and Pulleyblank (1986). Contrastive Underspecification (CU), on the other hand, has not been developed as extensively as Radical Underspecification, and has not been discussed within the p r i n c i p l e s and parameters framework. In the 1960s and 1970s i t was assumed that language a c q u i s i t i o n took place as the c h i l d created rules compatible with the l i n g u i s t i c data. An "evaluation metric" would lead 23 the c h i l d to the appropriate adult grammar by allowing the c h i l d to choose between competing rules or rul e systems. The theory of UG (and i n p a r t i c u l a r the parametric model) gradually replaced t h i s rule-writing model, i n part because i t was found to be extremely d i f f i c u l t to understand or even characterize the evaluation m e t r i c 1 . In the theory of UG, the set of possible hypotheses that are necessary for a c q u i s i t i o n i s assumed to be innate. This type of theory requires more innate machinery than the rule-writing approach, but on the other hand i t severely l i m i t s the types of hypotheses that childr e n are assumed to make about language. The parametric model i s a s p e c i f i c theory of UG which assumes that the c h i l d comes equipped to the language learning process with a set of universal rules or p r i n c i p l e s and a set of parameters, which together make up UG. Parameters present the c h i l d with s p e c i f i c options of a p a r t i a l l y unspecified p r i n c i p l e 2 . Parameters are generally assumed to have binary settings — one achieves the unmarked option, the other the marked option. Thus i n addition to providing the c h i l d with a very l i m i t e d set of possible hypotheses, the parametric framework provides a theory of markedness that w i l l lead to predictions about order of a c q u i s i t i o n and complexity of phonological systems. F a i r l y recently, the p r i n c i p l e s and parameters model has become well-integrated into phonological theory. Phonological parameters have been proposed to account for d i r e c t i o n a l i t y of mapping or association (Archangeli and Pulleyblank 1989, 24 Piggott, to appear), feature hierarchitecture (Davis 1990; Piggott, to appear), branching p o s s i b i l i t i e s of s y l l a b i c constituents (Kaye 1987), stress assignment (Hayes 1981, Halle and Vergnaud 1987, Dresher and Kaye 1988) and the underlying s e l e c t i o n of feature values (e.g. Archangeli 1988; Archangeli and Pulleyblank 1986, 1989; Abaglo and Archangeli 1989). In t h i s chapter I argue that i f a parametric theory of underspecification i s adopted, UG w i l l include a set of feat u r a l parameters, which i n s e r t redundant feature values, and a set of rul e parameters, which provide the possible form of phonological r u l e s . RU assumes that UG supplies both context-free and context-sensitive featural parameters. When a context-free rule i s reset to the marked option a new rul e ( c a l l e d a complement rule) w i l l be created to i n s e r t the opposite feature value predicted by UG. When a context-s e n s i t i v e rule i s reset, a marked feature s p e c i f i c a t i o n w i l l be added to the underlying representation of a segment, but the r u l e i t s e l f w i l l remain i n the grammar. CU allows only for context-sensitive rules, and parameter switching w i l l occur and have the same eff e c t s as i n RU. This chapter w i l l be organized as follows. In 2.1 I discuss some general issues i n current phonological theory, looking s p e c i f i c a l l y at the theory of Lexical Phonology, aspects of phonological and metrical representations, and phonological rules. Many of the issues discussed i n 2.1 are both controversial and complex, but rather than o u t l i n i n g a l l approaches and controversies, I w i l l discuss only those issues 25 that are c r u c i a l for t h i s d i s s e r t a t i o n . In 2.2 I look at underspecification theory, focussing p a r t i c u l a r y on the theories of Radical and Contrastive Underspecification. In 2.3 I present some assumptions concerning the p r i n c i p l e s and parameters model of grammar, and i n 2.4 I discuss some of the basic assumptions of l e a r n a b i l i t y theory. In 2.5 the parametric theories of Radical and Contrastive Underspecification are outlined by superimposing the basic premises of these two theories upon the p r i n c i p l e s and parameters model of phonology. 2.1 Aspects of Phonological Theory 2.1.1 L e x i c a l Phonology In The Sound Pattern of English (henceforth SPE, Chomsky and Halle 1968) the syntactic component was assumed to feed d i r e c t l y into the phonological component of grammar. Word formation processes operated i n the syntax. Morphological junctures were represented by boundary symbols, and these remained v i s i b l e to the phonology. L e x i c a l Phonology developed out of the SPE framework, using insights from the work of Chomsky (1970) and Aronoff (1976) on word formation. In L e x i c a l Phonology boundaries are encoded through sets of bracketings and through l e v e l s , rather than through boundary symbols. There are two components to the lexicon — the l e x i c a l component and the p o s t - l e x i c a l component. In the l e x i c a l component rules apply only to words and/or morphemes, while i n the p o s t - l e x i c a l component rules apply to the output 26 of the syntax, i . e . to words, or to larger sets of str i n g s such as phrases or sentences. 2.1.1.1 The Lexicon The work of authors such as Siegel (1974), Aronoff (1976) and A l l e n (1978) demonstrated that there i s often a s i g n i f i c a n t i n t e r a c t i o n between phonological rules and morphological processes. Certain types of morphological operations appear to t r i g g e r c e r t a i n types of phonological r u l e s , while morphological operations often seem to c l u s t e r together. Pesetsky (1979) t r i e d to capture these facts by proposing that phonological rules can apply inside the lexicon, to the output of s p e c i f i c kinds of morphological operations. Groupings of morphological operations are c a l l e d l e v e l s or s t r a t a , and the output of each l e v e l i s subject to the rules of the phonology. Mohanan (1982) and Pulleyblank (1986) assume that there i s a single set of phonological rules i n the grammar of a language, but a s p e c i f i c r u l e may be constrained to apply only within the l e x i c a l component, only within the p o s t - l e x i c a l component, or within both 3. In t h i s model rules w i l l display d i f f e r e n t properties depending upon where they apply i n the phonology, because of the d i f f e r e n t constraints that hold of these two components. 27 (2.1) LEXICON MORPHOLOGY Underived l e x i c a l items Level 1: rule A * rule B... • r / Level n: rule X rule Y... PHONOLOGY p -rule i (domain L) p -rule i (domain P) > p -rule k. .(domain L&P) SYNTAX POSTLEXICAL PHONOLOGY 2.1.1.2 The Cycle Underived l e x i c a l items are fed into the f i r s t l e v e l of the morphology where they undergo morphological operations. Af t e r each morphological operation takes place these forms are fed back into the phonology, and then back into the morphology to undergo another morphological operation, and so on. Each feed through a morphological process and the set of phonological rules i s c a l l e d a cycle. Kiparsky (1982) notes that many phonological processes, such as T r i s y l l a b i c 28 Shortening (TSS), do not operate on underived forms. TSS applies to derived words such as serenity and i n v i t a t i o n , while i t does not apply to underived forms such as nightingale and ivory. The S t r i c t Cycle Condition was proposed to account for these facts (Mascar6 1976). The statement of the S t r i c t Cycle Condition given i n Kiparsky 1982: 41) i s reproduced i n (2.2). (2.2) The S t r i c t Cycle Condition (SCC) a. C y c l i c rules apply only to derived representations. b. A representation 4> i s derived with respect to r u l e R i n cycle j i f f 4> meets the s t r u c t u r a l analysis of R by v i r t u e of a combination of morphemes introduced i n cycle j or the a p p l i c a t i o n of a phonological rule i n cycle j . This condition w i l l p r o h i b i t forms such as ivory from undergoing c y c l i c phonological rules such as TSS, since they are underived forms i n the sense of (2.2). Kiparsky (1982) then argues that the SCC can be derived from the Elsewhere Condition (EC), a condition on the ordering of rules that has been adopted as a standard constraint on ru l e a p p l i c a t i o n (Kiparsky 1973, Koutsoudas, Sanders and N o l l 1974). 29 (2.3) The Elsewhere Condition (Kiparsky 1982: 136-7) Rules A and B i n the same component apply d i s j u n c t i v e l y to a form $ i f f : a. The s t r u c t u r a l description of a (the s p e c i f i c rule) properly includes the s t r u c t u r a l description of R (the general rule) b. The r e s u l t of applying a to $ i s d i s t i n c t from the r e s u l t of applying fi to $. In that case, a i s applied f i r s t , and i f i t takes e f f e c t , then 15 i s not applied. Kiparsky argues that every underived l e x i c a l item i s i n fact an i d e n t i t y r u l e , which by the EC w i l l block the a p p l i c a t i o n of any other phonological rule (since the i d e n t i t y r u l e w i l l always be more s p e c i f i c ) . With t h i s assumption, only the formulation of the EC i n (2.3) i s needed, obviating the s p e c i f i c statement of the SCC i n (2.2). While the SCC (or the EC) was successful i n capturing many aspects of the i n t e r a c t i o n of morphological and phonological processes, there i s some evidence that c e r t a i n types of rules do not obey the SCC. Kiparsky (1982) argued that structure-building rules, such as rules of s y l l a b i f i c a t i o n and stress assignment, and redundancy r u l e s , which f i l l i n feature values, may apply to non-derived forms (th i s argument i s l a t e r refuted i n Kiparsky 1985). Halle and Mohanan (1985) argue that the rule of f i n a l - n deletion i n English, which i s hot a structure-building r u l e , must also be 30 non-cyclic. This rule deletes the n i n forms l i k e columns, hymn-book and column, yet does not apply i n forms l i k e hymnal. The facts show that t h i s rule i s a l e x i c a l r u l e and must apply before compounding and i n f l e c t i o n s , yet i t must not apply on the same l e v e l as the - a l s u f f i x . In addition, t h i s r u l e applies to underived forms such as column and hymn. Halle and Mohanan suggest that f i n a l - n deletion i s part of a l e v e l of the morphology that i s non-cyclic. Kiparsky (1985) interprets these facts by saying that either there i s only a singl e l e x i c a l l e v e l i n English, or that brackets are not erased at the end of Level 1. 2.1.1.3 P o s t - l e x i c a l Component The most relevant c r i t e r i o n for determining whether a rul e i s l e x i c a l or p o s t - l e x i c a l i s the domain of a p p l i c a t i o n . Since p o s t - l e x i c a l rules may be fed by syntactic information, they often apply to segments across word boundaries as well as within words. Lexical rules, on the other hand, can apply only within words. The rule of Flapping i n English i s a t y p i c a l p o s t - l e x i c a l r u l e . Flaps occur inside words, as i n ladder ([laDar], but also occur across word boundaries, as i n h i t i t ! [hIDIt]. P o s t - l e x i c a l rules always apply i n an across-the-board fashion, since they cannot be s e n s i t i v e to l e x i c a l l y marked exceptions. It has also been suggested that p o s t - l e x i c a l rules may be optional and may be s e n s i t i v e to rate or s t y l e of speech (Kaisse and Shaw 1985). 31 2.1.1.4 Structure Preservation Kiparsky (1982,1985) argues that l e x i c a l rules are subject to Structure Preservation, while p o s t - l e x i c a l rules are not. Structure Preservation i s the constraint that features, feature combinations, s y l l a b l e types or any other structures that are not present underlyingly i n the language may not be referenced or derived. In Kiparsky (1985) i t i s shown that Structure Preservation w i l l r e s t r i c t the r u l e of Voicing Assimilation i n Russian to apply to obstruents, since voicing i s not s p e c i f i e d on sonorants anywhere i n the lexicon. Kiparsky assumes that the formal work of Structure Preservation i s c a r r i e d out by marking conditions operating i n the lexicon. In the case of Russian Voicing A s s i m i l a t i o n the relevant constraint i s given i n (2.4) (Kiparsky 1985: 108). (2.4) * [avoiced] [+son] This constraint says that neither [+voice] nor [-voice] i s marked underlyingly on sonorants i n Russian, and therefore by Structure Preservation neither feature value may be added to the representation of a sonorant anywhere i n the l e x i c a l phonology. This prevents sonorants from being affected by the rule of Voicing Assimilation, and allows Voicing A s s i m i l a t i o n to be stated without a target condition excluding the sonorant c l a s s . Phonetic implementation rules, such as those that derive aspirated stops i n English, are p o s t - l e x i c a l rules that are 32 non-structure preserving since they derive segments that are not underlyingly d i s t i n c t i v e . In Russian the constraint i n (2.4) must be turned o f f p o s t - l e x i c a l l y i n order that sonorants be phonetically r e a l i z e d as voiced segments. There have been a number of arguments given i n the l i t e r a t u r e that Structure Preservation as stated by Kiparsky i s too strong. Borowsky (1986) claims that Structure Preservation must i n fact be turned off before the end of the lexicon i n English i n order to account for the a s s i m i l a t i o n of velar nasals. Mohanan and Mohanan (1984) argue that c e r t a i n places of a r t i c u l a t i o n that are not underlyingly d i s t i n c t i v e i n Malayalam must be derived i n the course of the l e x i c a l phonology, and therefore that Structure Preservation does not play a r o l e at a l l i n the lexicon of t h i s language. Sproat (1985) also argues that Catalan v i o l a t e s Structure Preservation. Archangeli and Pulleyblank (1986) employ marking conditions such as (2.4), but argue that i t must be s t i p u l a t e d for each constraint whether i t holds i n the l e x i c a l component, the p o s t - l e x i c a l component or both. In the following chapters I assume the o r i g i n a l version of Structure Preservation (given by Kiparsky 1982, 1985), as there i s no evidence eit h e r i n Hungarian or Spanish that an alternate version i s required. It w i l l be shown i n 3.3.1 that Structure Preservation can explain neutral vowel behaviour i n the Hungarian Back Harmony system. 33 2.1.2 Representations 2.1.2.1 Non-linear Phonology The focus of generative phonology, as presented i n works such as SPE, was on rule writing, rule ordering and the derivation of l e x i c a l items. In the l a t e 1970s, due to such works as Goldsmith's Autoseqmental Phonology (1976), t h i s focus changed to show a greater i n t e r e s t i n representations. Goldsmith proposed that i n d i v i d u a l features belong to separate t i e r s , where they can be linked by association l i n e s to one or more than one segment. Autosegmental representations were o r i g i n a l l y proposed to account for tone patterns, which t r a d i t i o n a l l y posed problems for standard generative approaches, since tones often spread over a domain larger than a segment, or s h i f t from one segment to another by phonological or morphological operations. I t has since been recognized that many melodic features, including n a s a l i t y , voicing, g l o t t a l i z a t i o n and places of a r t i c u l a t i o n can be autosegmentalized, or generally act independently of other features. An example i s a feature that may be a morphological property of a root and underlyingly may ' f l o a t ' or be unassociated to any p a r t i c u l a r segment i n that root (e.g. Archangeli and Pulleyblank's 1986 discussion of [-ATR] i n Yoruba). A l t e r n a t i v e l y , a segment may be l e x i c a l l y linked to a single segment but a phonological r u l e may operate to add association l i n e s between t h i s feature and other segments i n a root (e.g. the Type A languages given i n Piggott (to appear)). Both these cases are examples of 34 autosegmentalized features; features which behave independently of the remainder of the melodic representation. Goldsmith (1976), i n his treatment of tone, proposed that tones are always underlyingly unassociated, and that they become anchored segmentally by means of the WeiIformedness Condition. (2.5) Well-formedness Condition (Goldsmith 1976: 27) a. A l l vowels are associated with at least one tone. b. A l l tones are associated with at lea s t one vowel. c. Association l i n e s do not cross. These conventions allowed more than one tone to be linked to a single vowel to create contour tones. Work by Clements and Ford (1979) and Halle and Vergnaud (1982) demonstrated that (2.5) i s too weak as i t predicts that when there i s a mismatch between the number of tones and the number of vowels, several options are possible. In fact, i n a given language, there i s generally only one method of resolving the mismatch. Clements and Ford argue that (2.5) i s also too strong because i t predicts that tones may be multiply linked, which i s not the case i n a l l languages. Halle and Vergnaud (1982) then propose that a set of association conventions apply only to free or f l o a t i n g tones. These conventions as adopted by Pulleyblank (1986: 11) are given i n (2.6). 35 (2.6) Association Conventions Map a sequence of tones onto a sequence of tone-bearing units, a. from l e f t to r i g h t b. i n a one-to-one r e l a t i o n Well-formedness Condition: Association l i n e s do not cross. These conventions do not allow for the mapping of more than one tone to a given segment. Multiple linkings of a sing l e tone to more than one tone-bearing unit or of one tone-bearing unit to more than a single tone are accomplished by language-s p e c i f i c r u l e , according to Pulleyblank (1986). Autosegmentalized features of any sort are assumed to be linked to representations by a set of Association Conventions such as those i n (2.6) (e.g. see Piggott, to appear), although i n Archangeli and Pulleyblank (1989) and Lieber (1987) i t i s claimed that the d i r e c t i o n of i n i t i a l association i s parameterizable. In the unmarked case association proceeds from l e f t to r i g h t , but i n marked cases, such as the association of Yoruba [ATR], association operates from r i g h t to l e f t . In Chapters 3 and 5 I w i l l adopt the version of the Association Conventions given i n (2.6), with the added assumption that the i n i t i a l d i r e c t i o n of mapping i s parameterizable. 36 2.1.2.2 Feature Geometry It was r e a l i z e d as early as Jakobson and Halle (1956) that segments do not consist of bundles of t o t a l l y unorganized features, but i t was not u n t i l Mohanan (1983) and Clements (1985) that the f i r s t models of feature geometry were proposed. Feature geometries attempt to account for r e s t r i c t i o n s on how features i n t e r a c t i n human languages. The geometry i n (2.7) i s one possible model of feature hierarchitecture that i s consistent with recent work i n t h i s area (cf. Clements 1985, Sagey 1986) 4. (2.7) Feature Geometry x Skeleton Root Node [nasal] Laryngeal Node [voice; [spr.gl [CG] v Supralaryngeal Node [cont] [cons] [lateral] Labial Node [round] [high] Dorsal Node Tback] [distr] 37 There i s a d i s t i n c t i o n i n (2.7) between nodes and features. Nodes (given as " o " ) dominate features or nodes, while features (given i n [ ] brackets) are the terminal elements i n the tree. Three a r t i c u l a t o r nodes -- L a b i a l , Coronal and Dorsal — are immediately dominated by the Place node. The vowel features of [back], [high] and [low] are dominated by the Dorsal node, which i s the node also used to represent ve l a r and uvular consonants. The Coronal node dominates [anterior] and [distributed] while the L a b i a l node dominates the feature [round], a fourth feature used i n the d e s c r i p t i o n of v o c a l i c systems. In Archangeli and Pulleyblank (1986) a feature geometry i s proposed i n which a separate node, the Secondary Place Node, dominates a l l vowel features. This node was meant to capture the fact that harmony rules are almost e x c l u s i v e l y triggered by vowels and have vowels as t h e i r targets. I t has since been argued that other aspects of representations and ru l e types can account for these facts, while incorporating the model i n (2 .7) 3. 2.1.2.3 The OCP The Obligatory Contour P r i n c i p l e (OCP) was o r i g i n a l l y proposed by Leben (1973) to account for the d i s t r i b u t i o n of surface tones i n Mende, a Niger-Congo language. In t r i s y l l a b i c nouns i n Mende the tonal patterns LLH and HHL do not occur. Leben used the OCP to p r o h i b i t i d e n t i c a l adjacent 38 tones, so that i f a t r i s y l l a b i c noun has three tonal autosegments, only HLH and LHL patterns w i l l be permitted. Other patterns, such as HLL and LHH w i l l be derived from underlying sequences of HL or LH tones, with spreading of the second underlying tone to the f i n a l s y l l a b l e . In t h i s way the OCP and p r i n c i p l e s of spreading are able to account for the systematic gap i n the tonal patterns of t h i s language. In McCarthy (1979) the OCP was applied to segmental patterns i n Arabic, to account for the systematic lack of roots of the form /CAC^C-,/. McCarthy's generalized statement of the OCP i s given i n (2.8). (2.8) OCP (McCarthy 1986: 208) At the melodic l e v e l , adjacent i d e n t i c a l elements are prohibited. McCarthy (1986) systematically examines evidence for and counterexamples against the OCP i n many d i f f e r e n t languages. In Afar, for example, a rule of syncope applies to forms i n (2.9a) but f a i l s to apply i n the forms i n (2.9b) where the a p p l i c a t i o n of syncope would r e s u l t i n two i d e n t i c a l adjacent consonants. 39 (2.9) Afar (McCarthy 1986: 220) digib-t-e digb-e 'she/I married' me*i er-ta meSr-a •you/he k i l l s a c a l f wager-n-e wagr-e 'we reconciled/he reconciled adad-e *add-e 'I/he trembled' danan-e *dann-e 'I/he was hurt' xarar-e *xarr-e 'he burned' McCarthy concludes that i n the vast majority of languages, the OCP appears to hold of segments at the l e x i c a l l e v e l , but does not necessarily hold at the l e v e l of phonetic implementation. He suggests that i f convincing examples of the OCP not holding at the l e x i c a l l e v e l can be found, then the OCP could be regarded as a parameterizable p r i n c i p l e of UG, but one that i n the unmarked case holds of a l l non-prosodic information i n the lexicon (but see Paradis and Prunet 1990 for a convincing argument that perhaps the OCP does not have parametric options). Recently i t has been suggested that the OCP holds not just of enti r e segments, but of in d i v i d u a l features (e.g. Archangeli and Pulleyblank 1986, Mester to appear). 2.1.2.4 Metrical Theory In generative phonology s y l l a b l e s were not considered to be primitives of the theory, and could only be referenced as some sequence of consonants and vowels. In Kahn (1976) i t i s argued that s y l l a b l e s are constituents of phonological representations, represented as a separate t i e r dominating 40 s k e l e t a l s l o t s . There have been many subsequent elaborations of the non-linear approach to s y l l a b l e structure, the most common of which are the rule-based approach (Kahn 1976, Steriade 1982, Levin 1985), the templatic-approach (Selkirk 1978, Halle and Vergnaud 1978) and the government approach (Kaye and Lowenstamm 1984). Here I outline the rule-based approach of Levin (1985), and t h i s framework w i l l be employed i n Chapter 3 i n the analysis of Spanish s y l l a b l e structure. Levin's framework focusses p a r t i c u l a r l y on the univeral and language-specific aspects of a theory of s y l l a b i c i t y . S y l l a b i c constituents are generated by a version of X-bar theory, where each s y l l a b l e contains one and only one head. The universal components of t h i s theory are given i n (2.10) (Levin 1985: 12). (2.10) A. X-bar theory i . Categorial Component a. N-Placement b. Complex-N i i . Projection a. Project N" b. Project N' i i i . Incorporation a. Incorporate into N" b. Incorporate into N' i v . Adjunction (to N") B. Condition on Structure-Dependent Rules C. Sonority Hierarchy 41 N-Placement i s the process which determines the s y l l a b l e head. N-Placement may be marked i n the lexicon or may be determined by a redundancy or phonological r u l e . Levin assumes that [ s y l l a b i c ] i s not an operative feature, but rather that i n many languages the category N i s erected by rul e over segments s p e c i f i e d as [-consonantal] and/or [-high]. Complex Ns may e x i s t , c o n t r o l l e d by a parameter associated with the Complex-N process of the categorial component. Although Levin does not s p e c i f i c a l l y address the issue of markedness i n complex Ns, I assume that Universal Grammar i n i t i a l l y t e l l s the c h i l d that Ns may not branch, but that based on p o s i t i v e evidence from the input the c h i l d may switch t h i s parameter to allow a single N to dominate two sk e l e t a l s l o t s (this aspect of Levin's theory w i l l be discussed i n Chapter 5). The rules i n (2.10) w i l l apply c y c l i c a l l y , so that i f an epenthetic vowel i s inserted a f t e r the i n i t i a l a p p l i c a t i o n of N-Placement, i t w i l l receive a designation as a nucleus on a subsequent cycle. Afte r N i s found N" i s projected by picking up segments immediately to the l e f t of N. N" i s the maximal projection, and therefore the s y l l a b l e node i t s e l f . The f i n a l p r o j e c t i o n i s N', and t h i s projection i s p a r t i c u l a r l y influenced by language-specific information. Project N' w i l l pick up any remaining post-nuclear segments. Incorporation i s a process which then allows additional s k e l e t a l s l o t s to be incorporated under N" or N*. This process w i l l allow for both complex onsets and codas on a language-particular basis. The 42 operation of N-placement, Project N" and N' and N" Incorporation are shown i n (2.11) for the English word t r i p . (2.11) a. N-Placement N I X X X X Redundancy | rul e inserts [-cons] [-cons] t r i p b. Project N" X X X X I [-cons] t r i p 43 c. Project N' d. Incorporation [-cons] t r i p Each of the operations of N-Placement/ Project N" and Project N' must operate within the constraints of the Sonority Hierarchy, as shown i n (2.IOC). Levin posits a universal hierarchy of features which determine sonority ranking, which serves as the base from which language-particular sonority 44 scales are developed. If a given s t r i n g cannot be exhaustively s y l l a b i f i e d i n adherence with the language-s p e c i f i c sonority scale, then several possible options are ava i l a b l e to resolve t h i s problem. The most common option, i s that a s k e l e t a l s l o t , which can then undergo the r u l e of N-Placement, i s inserted or epenthesized into the representation. This new segment w i l l then serve as the head of a new s y l l a b l e to which the previously u n s y l l a b i f i a b l e segment may be associated. Zee (1988) discusses such a case of Epenthesis i n Bulgarian. In Bulgarian l i q u i d s are not able to act as n u c l e i , and therefore a form such as /grk/ 'Greek' w i l l undergo epenthesis as shown i n (2.12). (2.12) Underlying /grk/ Epenthesis x x x x g r k S y l l a b i f i c a t i o n N" X X X X [-cons] g r k [grak] 'Greek' 45 Within t h i s view Epenthesis i s seen as an automatic r e s u l t of the s y l l a b i f i c a t i o n algorithms i n a language, and does not require that such rules be language-specific phonological rules such as rules of spreading or delinking (to be discussed i n 2.1.3). The past decade has also seen much research into a theory of metrical s t r e s s . Liberman and Prince (1975) propose a metrical theory i n which stress i s determined by the r e l a t i v e prominence of s y l l a b l e s , determined through binary branching tree structures erected over s y l l a b l e s . Each node i n the tree i s l a b e l l e d as either w (weak) or s (strong), and each binary branching tree i s c a l l e d a foot. Both terminal and non-terminal constituents are l a b e l l e d for prominence, allowing stress to be assigned to words of more than two s y l l a b l e s . Hayes (1981) revises and expands on t h i s theory of s t r e s s , arguing that a theory of metrical tree structure and a small number of parameters can account for stress assignment i n the world's languages. Hayes' p r i n c i p l e s of tree construction are given i n (2.13) (Hayes 1981: 48). 46 Tree Construction Project rimes. Optionally form a subprojection of [+syllabic] segments within the rime. Select either r i g h t or l e f t nodes as dominant. Form the largest possible binary branching tree, such that recessive nodes do not branch. Optionally, i t may be s p e c i f i e d that i . A l l terminal nodes are counted as non-branching. i i . Dominant nodes must be terminal. i i i . Dominant nodes must branch. The f i r s t p r i n c i p l e i n (2.13a) says that rhymes form the i n i t i a l p rojection on which trees are erected, or o p t i o n a l l y that t h i s projection i s the nucleus. Since Hayes (1981) predates Levin (1985) I assume that i n a Levin-style analysis (2.13a) would say 'project N', o p t i o n a l l y project N. ' (2.13b) provides the choice of whether the foot i s l e f t or right-dominant. By (2.13c) foot construction continues, erecting feet over rhymes or s y l l a b l e s i n accordance with (2.13b), where nodes marked w never branch. The parameters i n (2.13c) i n the unmarked case are OFF, and i n the marked case are i n e f f e c t . If Parameter (2.13ci) i s chosen foot construction w i l l be quantity s e n s i t i v e , ignoring branching within the rhyme or nucleus. If (2.13cii) i s chosen, then foot construction i s maximally binary rather than unbounded, and i f ( 2 . 1 3 c i i i ) i s chosen then nodes are l a b e l l e d s i f and only i f they branch. Following (2.13) stress w i l l be assigned (2.13) a. b. c. 47 to the form merepet from Maranungku using the parameters given i n (2.14a and b)(Hayes 1981: 51). (2.14) a. Going from l e f t to r i g h t , construct binary, quantity i n s e n s i t i v e , l e f t dominant feet. b. Group the feet into a l e f t dominant word tree. c. merepet The most prevalent competing theory of stress assignment i s g r i d theory, where the branching structures of Hayesian tree theory are not recognized (Prince 1983). In the analysis of stress assignment i n Spanish given i n Chapter 3 I w i l l adopt the Hayesian framework, although I believe nothing hinges on t h i s p a r t i c u l a r choice of theory. 2.1.3 Rules Non-linear phonology and theories of feature geometry have led to a much clearer view of how phonological representations are structured. This, along with the move i n syntax to l i m i t rule types, has led phonologists to a constrained view of the number and types of phonological rules that may occur i n any language. Piggott (to appear), for example, allows for the rules shown i n (2.15). 48 (2.15) Phonological Rules a. Spread a A B a b. Delink a A B A B I I " > I I a A a., a 1 / : ) d. Insert a A I a* B -> A B I I Each of these rules i s a simple operation which adds or deletes some part of the phonological representation. Spread a i s the rule involved i n most assim i l a t i o n processes, adding an association l i n e between a feature already anchored and a new segment. Delink a i s the d i s s i m i l a t i o n rule which deletes some element due to the presence of a second element. Fusion i s a process that i s not unique to Piggott's work, although i t i s f i r s t discussed as a rule of phonology i n Piggott (to appear). Fusion creates a double l i n k i n g between an element (feature, s k e l e t a l s l o t , etc.) on one t i e r and two elements on 49 another t i e r . Insert a i s the process which accounts f o r the i n s e r t i o n of a feature (or features) into the melodic component. Although Piggott views Fusion as a separate and d i s t i n c t r u l e from those i n (2.15a/ b and d), i t i s possible to view Fusion as the i n s e r t i o n of an association l i n e . In the analysis of Spanish and Hungarian presented i n Chapter 3 I f i n d no evidence for a separate rule of Fusion and consequently I assume that t h i s rule i s simply one possible i n s t a n t i a t i o n of the rule Insert a. Piggott (to appear) assumes that the operation of any one of the rules i n (2.15) i s regulated by a D i r e c t i o n a l i t y Parameter which says that rules may spread leftward, rightward or i n a b i d i r e c t i o n a l fashion. I assume that the unmarked s e t t i n g of a D i r e c t i o n a l i t y parameter i s always the same d i r e c t i o n as the i n i t i a l mapping regulated by the Association Conventions. A more precise statement of how these rules operate w i l l depend upon the theory of feature s p e c i f i c a t i o n assumed (see 2.2.1 - 2.2.3). For example, Spread a w i l l behave very d i f f e r e n t l y i n a theory of Contrastive Underspecification than i n Radical Underspecification. In the contrastive model, both feature values may be underlyingly s p e c i f i e d , i n which case i f a spreads to ft i t w i l l also have to delink a from ft. In the r a d i c a l model, only a single feature value w i l l be present underlyingly, so a w i l l only spread to ft i f ft i s not s p e c i f i e d for a. A theory which assumes that features are p r i v a t i v e 50 ( i . e . have only a single functional value) w i l l assume that Spread a functions as i n Radical Underspecification. 2.2 Underspecification Theory Underspecification has been a part of phonological theory since the days of the Prague School l i n g u i s t s . In Trubetzkoy (1969) archiphonemes are elements represented by the set of features that are common to a group of phonemes i f those phonemes are neutralized i n some context. Archiphonemes lack those feature values that are involved i n the n e u t r a l i z a t i o n process. Trubetzkoy discusses the case of word-final devoicing i n German (Trubetzkoy 1958/69: 79): In German the b i l a t e r a l opposition d-t i s neutralized i n f i n a l p o s i t i o n . The opposition member, which occurs i n the p o s i t i o n of n e u t r a l i z a t i o n , from a phonological point of view i s neither a voiced stop nor a voiceless stop but "the nonnasal dental occlusive i n general". Archiphonemes are only p o s i t i o n a l elements -- i n contexts where segment contrasts are not neutralized segments are f u l l y s p e c i f i e d . Halle (1959), adopted the Praguian notion of archiphoneme, but also assumed that features and feature values may be unspecified due to d i s t r i b u t i o n a l constraints, combinatorial p o s s i b i l i t i e s , grammatical context, or the nature of the segment i t s e l f . Unspecified feature values are f i l l e d i n during the course of the derivation by morpheme structure constraints or by phonological r u l e s . Specified features have one of three possible values: "+", or "0". In the t r a d i t i o n of Halle (1959) underlying 51 representations i n SPE were assumed to be redundancy free (Chomsky and Halle 1968: 381): Languages d i f f e r with respect to the sounds they use and the sound sequences they permit i n words. Thus each language places c e r t a i n conditions on the form of phonetic matrices and hence of the configurations of pluses and minuses (indicating membership i n one of a p a i r of complementary categories) that may appear as entries i n the c l a s s i f i c a t o r y matrices of the lexicon. These constraints make i t possible to predict, i n a given language, the s p e c i f i c a t i o n of features i n p a r t i c u l a r segments. Such p r e d i c t a b i l i t y applies to segments i n i s o l a t i o n (e.g. i n Finnish, a l l obstruents are voiceless) as well as to segments i n p a r t i c u l a r contexts (e.g. i n English, /s/ i s the only true consonant admissible before a true consonant i n w o r d - i n i t i a l p o s i t i o n ) . Rules describing these constraints can r e a d i l y be formulated within our framework, and can be interpreted as specifying the c o e f f i c i e n t s of p a r t i c u l a r features i n p a r t i c u l a r environments. It i s therefore natural to propose that such rules be incorporated i n the grammar and that the features that are predictable be l e f t unspecified i n l e x i c a l e n tries. The d e s i r a b i l i t y of underspecification had been questioned by Lightner (1963) and Stanley (1967), however, who argued that the underlying omission of feature values could lead to a system of unwanted ternary contrasts. Chomsky and Halle accepted the Lightner and Stanley arguments and dealt with the problem by ordering a l l redundancy rules i n a block to f i l l i n l e x i c a l l y unspecified feature values before the operation of phonological rules. In t h i s way the underlying s p e c i f i c a t i o n of features did not inte r a c t with the phonological rules of the language. Unspecified feature values were to be interpreted as the unmarked values of features, allowing markedness values to be computed for each l e x i c a l item. Kiparsky (1982) also adopted the premise that underlying representations are free from a l l redundancy, but dealt with 52 the Lightner-Stanley objections i n a d i f f e r e n t fashion from Chomsky and Halle (1968). Kiparsky proposed that within a given environment, only a single feature value may be l e x i c a l l y s p e c i f i e d , so that i f a phonological r u l e applies to f i l l i n the alternate value a three-way contrast w i l l never be possible within that p a r t i c u l a r environment. Chomsky and Halle (1968) believed that ordering redundancy rules before a l l phonological rules avoided a ternary system and at the same time allowed s i g n i f i c a n t l e x i c a l generalizations to be captured. They f e l t the e a r l y a p p l i c a t i o n of redundancy rules was warranted because "no good examples have been discovered of empirically s i g n i f i c a n t generalizations that r e s u l t from ordering these r u l e s " (Chomsky and Halle 1968: 386). Since that time a number of empirical arguments for the underspecification of segmental features at the time that phonological rules apply have been found. One example i s the voicing of obstruents i n Japanese (Ito and Mester 1986, Steriade 1987, Mester and Ito 1989). The two phenomena involved are Rendaku and Lyman's Law. Rendaku i s a r u l e which voices i n i t i a l obstruents i n the second element of a compound: (2.16) Rendaku a. o r i + kami —> origami 'paper f o l d i n g ' b. yo + sakura —> yozakura 'blossoms at night' c. yama + tera --> yamadera 'mountain temple' d. kake + futon --> kakebuton 'top futon' 6 53 Rendaku i s a productive phonological rule applying to compounds made up of native Japanese morphemes whose components stand i n a modifier-head r e l a t i o n s h i p . I t i s formulated i n Ito and Mester (1986) as a rule associating the feature [+voice] before the second member of a Rendaku compound. This feature w i l l be associated to the f i r s t e l i g i b l e segment i n the second member of the compound. Lyman's Law i s a constraint i n Japanese which says that Rendaku i s blocked i f any segment i n the second compound member i s s p e c i f i e d for voicing. In (2.17) the i n i t i a l consonants of the second member of a compound do not undergo Rendaku because they are followed by a voiced segment. (2.17) Lyman's Law a. s i r o + t a b i --> s i r o t a b i 'white t a b i ' b. maru + hadaka —> maruhadaka 'completely naked' c. taikutsu + sinogi --> taikutsusinogi ' t i m e - k i l l i n g ' d. doku + tokage —> dokutokage 'poisonous l i z a r d ' e. onna + kotoba —> onnakotoba 'feminine speech' Ito and Mester (1986) formalize Lyman's Law as a r u l e which deletes the [+voice] Rendaku feature when a [+voice] feature i s already part of the representation of the second compound member. (2.17d and e) show that Lyman's Law applies even when a voiceless obstruent surfaces between the i n i t i a l voiceless obstruent of the second compound member and a voiced obstruent. If voiceless obstruents i n Japanese are s p e c i f i e d as [-voice] they would be expected to block the operation of 54 Lyman's Law, since they would make the voicing of the f i n a l voiced obstruent i n v i s i b l e to the r u l e . Ito and Mester suggest that t h i s evidence shows that voiceless obstruents must be underlyingly unspecified for the feature [voice] at the point i n the derivation where Lyman's Law applies. This gives the representation i n (2.18) for the form 'poisonous l i z a r d ' i n (2.17e). (2.18) [+voice] [+voice] £ I [o n n a] RENDAKU [k o t o b a] 'feminine speech' This analysis demonstrates that i n Japanese obstruents are unspecified for voicing at the point i n the phonology when ce r t a i n phonological rules apply - 7. Pulleyblank (1985) presents arguments that i n T i v tone i s underlingly unspecified and c e r t a i n tone bearing units remain unspecified u n t i l the p o s t l e x i c a l component. Ito (1984) provides evidence that /a/ i n Ainu must be unspecified for backness at the point i n the derivation where a r u l e of d i s s i m i l a t i o n applies. Evidence of t h i s sort has led phonologists to attempt to develop theories of underspecification that w i l l predict when and where underspecified representations e x i s t . Currently, there are two major theories of underspecification: Radical Underspecification, outlined i n Archangeli and Pulleyblank (1986) and Contrastive 55 Underspecification, outlined i n Steriade (1987). These theories w i l l be discussed i n 2.2.1 and 2.2.2, and i n 2.2.3 I w i l l present some al t e r n a t i v e views of feature s p e c i f i c a t i o n . 2.2.1 Radical Underspecification 2.2.1.1 Elimination of Redundancy The theory of RU (Archangeli 1984, 1988; Archangeli and Pulleyblank 1986, 1989; Pulleyblank 1986, 1988) i s an exploration of the nature of phonological rules and representations which adopts Kiparsky's (1982) view that a l l redundant feature values are omitted from underlying representations and may remain unspecified throughout the lexicon. Only non-predictable feature values are present i n the underlying system of a language, and only a s i n g l e value of any feature i s permitted. Two types of redundancy rules are distinguished. Default rules i n s e r t u n i v e r s a l l y predictable feature values, while complement rules i n s e r t feature values that are language s p e c i f i c . Both types of rules may be context-free, i n s e r t i n g feature values i n general, or context-sensitive 8, i n s e r t i n g feature values that are based on co-occurrence r e s t r i c t i o n s . Redundancy rules are ordered among the phonological rules of a language by the set of ordering constraints given i n (2.19). 56 (2.19) Ordering Constraints (Archangeli and Pulleyblank 1986) a. Redundancy rules apply as l a t e as possible i n the grammar. b. A redundancy rule must apply at the l e v e l at which reference i s made to the feature value being inserted (The Redundancy Rule Ordering Constraint or RROC). c. A redundancy rule applies as early as possible at the l e v e l dictated by the RROC. The i n t e r a c t i o n of these constraints prohibits the p r a c t i c a l use of a ternary system of feature values. At some point i n the derivation, before a redundancy rule f i l l s i n a value of [F], there w i l l be a contrast between the underlying value of [F] ([ctF]) and [OF], At a l a t e r stage i n the d e r i v a t i o n , i f a phonological rule applies to [-aF], the unmarked feature value, then by (2.19b) and (2.19c) the redundancy ru l e w i l l apply immediately before the phonological r u l e and there w i l l be a contrast between [a] and [-aF]. If there are no phonological rules i n the language that manipulate the unmarked feature value, then the redundancy ru l e i n s e r t i n g [-aF] does not apply u n t i l the p o s t l e x i c a l component, by (2.19a). RU also assumes that redundancy rules are subject to the EC (Kiparsky 1973, 1982; Koutsoudas, Sanders and N o l l 1974) given i n (2.3). The EC w i l l force a context-sensitive redundancy rule to apply before a context-free rule i f they reference the same feature value, since the environment for 57 the a p p l i c a t i o n of a context-sensitive rule i s more s p e c i f i c . The a p p l i c a t i o n of the context-sensitive rule w i l l block the app l i c a t i o n of the context-free rule i n the p a r t i c u l a r environment s p e c i f i e d by the context-sensitive r u l e . The EC also constrains the ordering of phonological rules with respect to redundancy rules, as i n the case of Yokuts Echo Vowel Formation, which i s ordered before the more general default rule of [high] i n s e r t i o n (Archangeli and Pulleyblank 1986). In a hypothetical language which has the 9 consonants /p,t,k,b,d,g,m,n,1/ i n i t s phonological inventory, RU predicts the s p e c i f i c a t i o n of the features [sonorant], [voice], [ l a t e r a l ] , and [nasal] i n (2.20), given the accompanying set of default r u l e s . (2.20) p t k b d g m n 1 [sonorant] [voice] + + + [ l a t e r a l ] + [nasal] + + Default rules: [ ] —> [-son] [+nasal] —> [+son] [ ] --> [-lateral] [+lateral] —> [+son] [ ] —> [-nasal] [+son] —> [+voice] [-son] --> [-voice] These consonants w i l l be further s p e c i f i e d for place of 58 a r t i c u l a t i o n , which w i l l then d i s t i n g u i s h the l a b i a l s , alveolars and velars from one another. The feature [sonorant] i s completely redundant i n t h i s hypothetical system, since the sonorancy values of a l l segments i n (2.20) are given by the context-free rule i n s e r t i n g [-son] and the two context-s e n s i t i v e rules predicting [+son] on segments marked [+lateral] or [+nasal]. 2.2.1.2 Rule Types Archangeli and Pulleyblank (1986) provide a typology of possible r u l e types with accompanying markedness statements. Each and every phonological rule has a function and an argument, and may also have a target and/or t r i g g e r condition. (2.21) gives the possible s p e c i f i c a t i o n s of rul e s , where " I " designates functions or operations, " I I " arguments, and " I I I " t r i g g e r / t a r g e t conditions. (2.21) I. a. ( i n s e r t ) d < 2 > t Q U l t / d e l e t e b. ( r a a x i m a l ) d e f a u l t / m i n i m a l c. (content) d« £ a ux t/structure d. (same d i r e c t i o n ) d e f a u i t / o p p o s i t e d i r e c t i o n / b i d i r e c t i o n a l I I . node or feature(s) I I I . t r i g g e r condition target condition There are four function parameters, each of which has a default or unmarked set t i n g . "Insert/delete" indicates that a feature or association l i n e i s added to or taken away from the 59 representation, with " i n s e r t " being the unmarked option. The "maximal/minimal" function describes the adjacency of rul e targets. At the unmarked "maximal" set t i n g a rul e which targets a node or feature scans the highest l e v e l of s y l l a b i c structure providing access to that target, while at the "minimal" s e t t i n g a rule scans the t i e r containing the target. Rules targeting vowels can therefore apply to segments that are not immediately adjacent at the s k e l e t a l l e v e l , but must be adjacent or l o c a l at the l e v e l of the nucleus. This constraint on adjacency i s termed the L o c a l i t y Condition i n Archangeli and Pulleyblank (1987). "Content/structure" indicates that i t i s a feature or an association l i n e which i s added to the representation by the rule, with "content" being the default or unmarked option. The d i r e c t i o n a l i t y parameter describes the d i r e c t i o n of spreading, with "same d i r e c t i o n " i n d i c a t i n g that the unmarked set t i n g for t h i s function i s i n the same d i r e c t i o n as the i n i t i a l association of s k e l e t a l s l o t s to melodic structure (as given by a set of Association Conventions such as those i n (2.6)). This typology of rule functions or operations predicts that there are two unmarked types of phonological rules a v a i l a b l e to children -- default rules and rules of epenthesis. Default rules use only default settings of function parameters and inse r t arguments that are provided by UG. A context-free default rule i n s e r t i n g the feature [-low] w i l l be stated as i n (2.22), where the parentheses indicate the default settings of parameters. 60 (2.22) Default [-low] I. a. (insert) c. (content) d. (same direction) I I . [-low] (2.22) says that [-low] w i l l be inserted anywhere (a set of Configuration Constraints w i l l d i c t a t e where i t can attach and where i t cannot), and i n s e r t i o n w i l l operate i n l e f t - t o - r i g h t fashion, assuming that t h i s i s the unmarked d i r e c t i o n of i n i t i a l association. The maximal/minimal parameter i s ir r e l e v a n t to t h i s rule, since there i s no s p e c i f i e d t a r g e t 9 . A context-sensitive default rule w i l l use the same function parameters as i n (2.22) but w i l l require the addition of a target condition. The second type of maximally unmarked ru l e i s an epenthesis r u l e , where the function parameters are again set to the defaults, with the argument s p e c i f i e d as a s k e l e t a l s l o t 1 0 . (2.23) Epenthesis I. a. (insert) c. (content) d. (same direction) I I . s k e l e t a l s l o t Such a rule w i l l only be posited when s y l l a b i f i c a t i o n or other l i c e n s i n g requirements make i t necessary. The inserted 61 s k e l e t a l s l o t w i l l have no featural content; however, since (2.22) i s not a redundancy rule, i t w i l l not be subject to the Ordering Constraints i n (2.18) and w i l l apply i n the l e x i c a l phonology before the redundancy r u l e s 1 1 . In t h i s way the feature s p e c i f i c a t i o n s of the epenthetic segment w i l l be provided by the redundancy rules of the language. If we follow current assumptions about Epenthesis (as discussed i n 2.1.2.4) then the unmarked status of (2.23) i s explained by the fact that i t i s simply an automatic r e s u l t of the s y l l a b i f i c a t i o n algorithms i n a p a r t i c u l a r language. The rule i n (2.23) w i l l not have the same status as other phonological rules (such as those i n (2.15), which do not i n general have a bearing on prosodic representations. Rules that w i l l involve s e t t i n g at lea s t one of the function parameters i n (2.22) to the marked value w i l l be those that delete features or structure; spreading or ass i m i l a t i o n rules which use the marked "structure" parameter; rules whose targets are s p e c i f i e d as "minimal"; and rules that operate i n some d i r e c t i o n other than the canonical d i r e c t i o n of i n i t i a l association. Thus RU recognizes three of the rul e types given i n (2.15): Deletion, Spread and Insertion, without allowing for Fusion. Both the type of underspecification assumed by RU and the rule typology i n (2.21) have implications for how phonological rules operate i n t h i s theory. Because only a single feature value i s present i n any environment, Spreading rules w i l l be able to manipulate only a single feature value. The typology 62 i n (2.21) formalizes Kiparsky's claim (Kiparsky 1985) that i n the unmarked case features spread only to segments which are •free' or unspecified for that feature. The fact that a target i s free i s stated i n (2.21) by the lack of a target condition (2.21III). In the analysis of Khalka Mongolian harmony given i n Archangeli and Pulleyblank (1986), i t i s assumed that Round Harmony, which spreads [+round], lacks a target condition, explaining why the rule i s blocked by / l i / . I t i s rare that two values of any feature are s p e c i f i e d at once (although i t could happen: i f a voiceless l a t e r a l were present i n the inventory shown i n (2.20) t h i s l a t e r a l would presumably be s p e c i f i e d as [-voice]), but i f a r u l e spreading [+F] encounters a segment s p e c i f i e d as [-F] spreading should presumably also be blocked). 2.2.2 Contrastive Underspecification 2.2.2.1 Specifications The theory of CU was f i r s t outlined i n Steriade (1987), and has been taken up i n works such as Clements (1987), Calabrese (1988) and Mester and Ito (1989). Steriade's work was not meant as a t h e o r e t i c a l t r e a t i s e , but rather as an a l t e r n a t i v e to the theory of RU. As a r e s u l t , many aspects of the theory are not f u l l y a r t i c u l a t e d , although most c r u c i a l aspects of the theory may be drawn from s p e c i f i c analyses given i n Steriade (1987). Steriade (1987) makes an i n i t i a l d i s t i n c t i o n between t r i v i a l and n o n - t r i v i a l underspecification. T r i v i a l 63 underspecification occurs when a segment lacks s p e c i f i c a t i o n for a p a r t i c u l a r feature at a l l stages i n a derivation. It arises i n the case of monovalent features or a r t i c u l a t o r nodes/ where only a single value of the feature or node i s operative, and i n the case of features which never acquire a value for a second feature because of r e s t r i c t i o n s i n the feature hierarchy. An example of the l a t t e r type of t r i v i a l underspecification i s a Dorsal consonant which w i l l never receive a s p e c i f i c a t i o n for [an t e r i o r ] , since [anterior] i s a feature that i s dependent on the Coronal a r t i c u l a t o r node (see (2.7)). N o n - t r i v i a l underspecification occurs when a segment lacks a s p e c i f i c a t i o n for a feature at some stage i n the derivation, but eventually becomes s p e c i f i e d for that feature value. The important d i s t i n c t i o n within n o n - t r i v i a l underspecification i s made between R-values of features, which are redundant feature values, and D-values, which are those values of a feature, that contrast within a s p e c i f i c c lass of segments. R-values are absent from underlying representations, while D-values must both be present. A contrastive s p e c i f i c a t i o n of the hypothetical consonantal system shown i n (2.20) i s given i n (2.24). 64 (2.24) p t k b d g m n l [sonorant] [voice] - - - + + + a [ l a t e r a l ] - + [nasal] - - + + Redundancy rules: [+nasal] —> [+son] [+lateral] —> [+son] [+son] —> [+voice] [+cons] —> [-son] [+cons] --> [-lateral] [+cons] —> [-nasal] The voiced and voiceless obstruents contrast with regard to the feature [voice] i n t h i s system, so voiceless obstruents are underlying marked [-voice] and the voiced ones as [+voice]. / I / contrasts with /d/ i n terms of l a t e r a l i t y , so both are s p e c i f i e d for a value of the feature [ l a t e r a l ] . The two nasals contrast with the corresponding voiced obstruents, and therefore /b/ and /d/ are s p e c i f i e d as [-nasal] while /m/ and /n/ are s p e c i f i e d as [+nasal]. In t h i s system [sonorant] i s completely redundant. 2.2.2.2 R-Rules There i s no formal contrast i n CU between universal and language-specific redundancy (R-^rules). R-rules i n s e r t R-65 values. They are always context-sensitive, since they i n s e r t redundant values based on co-occurrence r e s t r i c t i o n s , and they may have one of two functions (Steriade 1987: 359): (2.25) a. the enhancement of perceptual salience, e.g. [+back] —> [+round], or b. the demonstration of the r e s t r i c t e d d i s t r i b u t i o n of a content feature within the realm of a s t r i c t u r e feature ( s t r i c t u r e features are [sonorant], [cons], [continuant], [high], [low]) e.g. [+son] —> [+voice]. Steriade points out that assuming that R-rules may have only the functions i n (2.24) rules out the p o s s i b i l i t y of an R-rule such as [+round] —> [+high], which i s used by Archangeli and Pulleyblank (1986) i n the analysis of Nyangumarta. This r u l e i s not excluded on the grounds that i t i s language-specific, but on the grounds that i t does not have one of the functions i n (2.25). The R-rules for vo c a l i c segments used i n Steriade (1987) are given i n (2.26) 1 2. 66 R-rules 1. [+low] ---> [-high] 2. [+low] — > [+back] 3. [+back] ---> [+round] [-low] 4. [-low] ---> [-back] [-round] 5. [+high] ---> [+round] [+ATR] In Steriade's analysis of Hungarian Back Harmony an R-rule i s e x t r i n s i c a l l y ordered before a phonological r u l e . Two R-rules and a r u l e of Back Harmony are posited for Hungarian (the CU analysis of Hungarian BH w i l l be discussed i n d e t a i l i n 3.1.5.2): (2.27) R-rule 1: [+low] —> [+back] R-rule 2: [-low, -round] —> [-back] Back Harmony ( i t e r a t i v e , feature-changing) [aback] [Aback] V . . . V The rule of Back Harmony i s ordered between R-rules 1 and 2 i n order that the low long vowel /a:/ can act as a t r i g g e r of Harmony. R-rule 2 cannot be ordered before Harmony, because i t f i l l s i n the backness s p e c i f i c a t i o n s of / i / , / i : / and /e:/ and these vowels are transparent to the Harmony process. The 67 EC, which orders s p e c i f i c rules before more general ones, cannot force the ordering of R-rule 1 before Back Harmony, since the s t r u c t u r a l description of R-rule 1 i s not part of the s t r u c t u r a l description of Back Harmony. Steriade (1987) also posits an e x t r i n s i c ordering r e l a t i o n s h i p between the rule of Round Harmony i n Mongolian and the redundancy rule i n s e r t i n g the predictable value of [round]. She argues, however, that language-specific ordering between phonological rules and redundancy rules i s unwanted, and instead assumes that i n Mongolian [round] acts as a monovalent feature. In t h i s language then the redundancy r u l e i n s e r t i n g [-round] i s not required. Unfortunately, an alternate analysis of the Hungarian facts i s not provided. 2.2.2.3 Phonological Rules Because CU assumes that both values of a contrastive feature are underlyingly s p e c i f i e d , the theory must also assume that a s s i m i l a t i o n rules may be e i t h e r f e a t u r e - f i l l i n g or feature-changing. Steriade (1987: 339) states: nothing, i n my view, prevents a r u l e of vowel harmony from operating sequentially, by successive spreading and delinking steps, as shown i n (1): (1) F -F -F -F F -F -F V C V C V C V --> V C V C V C V --> F -F F V C V C V C V --> V C V C V C V A rule propogating the feature [F] w i l l therefore not be blocked when a segment i s encountered that i s s p e c i f i e d for 68 [F], as i s assumed i n the unmarked case by Radical Underspecification. Steriade provides examples of feature-changing processes i n the Height Harmony system of Pasiego, following McCarthy (1984) and i n the Back Harmony system of Hungarian, following the analysis of Farkas and Beddor (1987), shown i n (2.27). In Pasiego either [+high] or [-high] i n i t i a t e s Height Harmony, spreading to a vowel s p e c i f i e d as e i t h e r [+high]. In Hungarian, Steriade assumes that e i t h e r [+back] or [-back] i n i t i a t e s the harmony process and that targets may be s p e c i f i e d as either [+back]. If r u l e t r i g g e r s may be s p e c i f i e d for either value of a feature, then these rules w i l l have to be stated using alpha notation, as shown i n (2.27) . 2.2.3 Alternate Theories of Feature S p e c i f i c a t i o n The majority of current analyses dealing with underspecified values adopt either the Radical or Contrastive framework. Calabrese (1988), Clements (1987) and Christdas (1988) argue for s l i g h t l y modified versions of CU. Calabrese's revisions involve deriving the pattern of underspecification i n a given language from the set of universal f i l t e r s that are v i o l a t e d language-specifically (this i s e s s e n t i a l since Calabrese uses f i l t e r s or negative constraints, whereas Steriade uses universal redundancy or R-r u l e s ) . Calabrese's claims regarding the hierachy of universal f i l t e r s w i l l be discussed i n 6.1.2.2. In 2.2.3.1 I w i l l b r i e f l y outline Clements' theory. A number of other 69 works> such as Den Dikken and van der Hulst (1990) and Piggott (1990, to appear) assume that features are p r i v a t i v e rather than binary, and therefore do not adopt a theory of underspecification at a l l . 2.2.3.1 Clements (1987) Clements' (1987) position on underspecification i s that both values of a contrastive d i s t i n c t i v e feature are always f u l l y s p e c i f i e d i n URs, as are the so-called 'primary' or major class features such as [sonorant] and [consonantal]. Incomplete underlying s p e c i f i c a t i o n of values occurs when features are not used contrastively, or when they represent a r t i c u l a t o r nodes ( i . e . L a b i a l , Coronal and Dorsal i n (2.6)). There are no univeral default rules to f i l l i n missing values, but language-particular rules may add missing feature values. The hypothetical 9 consonant language s p e c i f i e d i n (2.20) and (2.24) i s s p e c i f i e d according to Clements' system i n (2.28). (2.28) p t k b d g m n l [sonorant] - - - - - - + + + [voice] - - - + + + [l a t e r a l ] - + [nasal] - - + + The feature [sonorant] i s f u l l y s p e c i f i e d since i t i s a major class feature, while the features [voice], [ l a t e r a l ] and [nasal] are contrastive. A r t i c u l a t o r nodes would behave i n a monovalent fashion ( i . e . are either present or absent), with 70 Labia l present for /p,b, and m/ and absent for a l l other features, etc. In t h i s thesis I w i l l t reat the Clements (1987) (and Christdas 1988) view of underspecification as another form of CU, since vowel features, which are neither major class features nor a r t i c u l a t o r nodes, w i l l be c o n t r a s t i v e l y s p e c i f i e d . Clements* treatment of vowels w i l l therefore be equivalent to a contrastive one. 2.2.3.2 P r i v a t i v e Features Each of the theories of underspecification discussed assume that features are binary, as was f i r s t proposed i n Jakobson and Halle (1956). Trubetzkoy (1939/69), on the other hand, proposes that featural oppositions may be p r i v a t i v e , where features are either present or absent; gradual, where features have degrees; or equipollent, where features have polar d i s t i n c t i o n s that are each l o g i c a l l y possible. In the feature geometry i n (2.7) nodes are assumed to behave i n a p r i v a t i v e (or monovalent) fashion, even by proponents of underspecification theory. More relevant to a discussion of underspecification i s the issue of whether terminal features are binary or p r i v a t i v e . Piggott (1990) proposes that the feature [nasal] i s u n i v e r s a l l y a p r i v a t i v e feature, while Mester and Ito (1989) and Steriade (1987) propose that [voice] and [round] are p r i v a t i v e features i n Japanese and Mongolian resp e c t i v e l y . Den Dikken and van der Hulst (1990) assume that a l l features 71 are p r i v a t i v e , and Goldsmith (1987) proposes that a l l features are e i t h e r p r i v a t i v e or equipollent. In the majority of cases i t w i l l be impossible to determine whether a feature i s pr i v a t i v e or whether i t i s binary but has only a single feature value present underlyingly. This i s because underspecification theory assumes that underlyingly unspecified feature values are f i l l e d i n by redundancy rules l a t e i n the derivation, and therefore redundant values w i l l generally have no phonological e f f e c t s . Mester and Ito (1989: 283) warn that t h i s may make comparisons of RU and a theory of p r i v a t i v e features d i f f i c u l t : If not a l l features are p r i v a t i v e (see den Dikken & van der Hulst 1988 and references c i t e d there for a s t r i c t l y single-valued approach), a p r i n c i p l e d typology of features must be developed. Otherwise private features could be invoked ad libitum whenever i t seems that RU i s needed. It i s only i n the case of a binary feature whose predictable value has phonological e f f e c t s that we w i l l be able to d i s t i n g u i s h i t from a p r i v a t i v e feature. In RU, the phonological use of predictable features i s accomplished by the RROC, given i n (2.19b). Some examples of the use of the RROC are presented i n Archangeli and Pulleyblank (1986, 1989) for Yoruba and i n Archangeli and Pulleyblank (1986) for Yokuts. I believe that i t i s contingent upon those who assume that features may be pr i v a t i v e to account for these cases. In t h i s thesis I assume that terminal features are binary, and always have two possible values. Based on the h i s t o r i c a l development of d i s t i n c t i v e features as binary units, I believe that the binary approach i s more conservative than the 72 p r i v a t i v e one. 2.3 P r i n c i p l e s and Parameters The p r i n c i p l e s and parameters theory of grammar was f i r s t outlined i n Chomsky (1981a,b) and has been widely used to account for a v a r i e t y of facts i n both syntax and phonology. Parameters are used to explain c r o s s - l i n g u i s t i c differences, and at the same time make predictions about the course of language a c q u i s i t i o n . For these reasons, i n t h i s thesis I w i l l develop a theory of phonological a c q u i s i t i o n that assumes that UG contains both p r i n c i p l e s and parameters. In the following sections I outline how the p r i n c i p l e s and parameters model of grammar i s organized, and show how some previously proposed parameters are assumed to operate. The p r i n c i p l e s and parameters theory emphasizes the u n i v e r s a l i t y of the t h e o r e t i c a l constructs used i n a l l languages, while predicting that languages d i f f e r i n c e r t a i n constrained ways: What we expect to f i n d , then, i s a highly structured theory of UG based on a number of fundamental p r i n c i p l e s that sharply r e s t r i c t the class of attainable grammars and narrowly constrain t h e i r form, but with parameters that have to be fixed by experience. If these parameters are embedded i n a theory of UG that i s s u f f i c i e n t l y r i c h i n structure then the languages that are determined by f i x i n g t h e i r values one way or another w i l l appear to be quite diverse, since the consequences of one set of choices may be very d i f f e r e n t from the consequences of another set; yet at the same time, l i m i t e d evidence, just s u f f i c i e n t to f i x the parameters of UG, w i l l determine a grammar that may be very i n t r i c a t e and w i l l i n general lack grounding i n experience i n the sense of an inductive basis. (Chomsky 1981a: 3-4) According to t h i s view a c q u i s i t i o n becomes a r e l a t i v e l y simple 73 endeavour. A p a r t i c u l a r hypothesis i s supplied by UG, and evidence from the language being acquired w i l l confirm that t h i s i n i t i a l hypothesis i s correct, or suggest that an alternate hypothesis should be chosen. In the l a t t e r case, the c h i l d need not create a new hypothesis, but simply "switch" a parameter, and a new hypothesis w i l l be given. The p r i n c i p l e s and parameters approach provides a b u i l t -i n theory of markedness. The parameter settings provided by UG are the default or unmarked settings, while the other s e t t i n g ( s ) , which can only be achieved v i a evidence from the input, are marked. A language that makes use of the default s e t t i n g of a parameter w i l l be less complex or marked than a language that uses the marked set t i n g . In a c q u i s i t i o n , then, t h i s model predicts that the c h i l d w i l l f i r s t assume the unmarked parameter set t i n g , and only switch to a marked parameter s e t t i n g i f there i s evidence that t h i s i s necessary. Developmentally, the chi l d ' s speech should f i r s t be characterized by the unmarked parameter s e t t i n g , and only l a t e r be characterized by the marked se t t i n g . In Hyams (1983), for example, the ac q u i s i t i o n of the "pro-drop" or AGR/PRO parameter (Chomsky 1981a, R i z z i 1982) i s discussed. This parameter was proposed to explain differences between languages i n the a b i l i t y to have null-subjects i n tensed c l a u s e s 1 3 . Hyams argues that the unmarked parameter s e t t i n g allows null-subjects, as i n I t a l i a n , while the marked s e t t i n g requires l e x i c a l subjects, as i n English. The low incidence of l e x i c a l i z e d subjects i n the speech of young childre n 74 acquiring English i s due to the i n i t i a l unmarked parameter f o r AGR/PRO (this analysis has sparked much controversy -- see Gu i l f o y l e 1984 and Lebeaux 1987 for two reanalyses of these f a c t s ) . 2.3.1 B i n a r i t y The majority of parameters that have been discussed i n the l i t e r a t u r e , i n both syntax and phonology, have binary settings. For example, Bach (1965) claims that languages have the choice of having a rule of Wh-movement or not (Williams 1987 c a l l s the Wh-movement parameter an "existence" parameter). A phonological example i s given i n Archangeli and Pulleyblank (1989) where the i n i t i a l association of autosegmentalized features i s assumed to operate i n l e f t - t o -r i g h t fashion i n the unmarked case and i n r i g h t - t o - l e f t fashion i n the marked case (see 2.1.2.1). Bach's Wh-Movement parameter i s a binary parameter that has a choice between ON and OFF, while Archangeli and Pulleyblank's i n i t i a l a ssociation parameter i s binary but has two s p e c i f i e d choices: LEFT-TO-RIGHT or RIGHT-TO-LEFT. Suggestions for parameters with multiple settings have been given i n Wexler and Manzini (1987) for the governing category of anaphors, or for the d i r e c t i o n of spreading i n Archangeli and Pulleyblank (1986). Fodor (1989) argues that non-binary parameters can and should be analyzed as sets of parameters, each with a binary choice. Only binary parameters w i l l provide a cle a r statement of the markedness of parametric 75 systems. Archangeli and Pulleyblank*s d i r e c t i o n of spreading parameter (1986) can be stated as two separate parameters, as i s done i n Piggott (to appear): (2.29) D i r e c t i o n a l i t y Parameter(s) i . Multiple Settings (Archangeli and Pulleyblank 1986) Same dir e c t i o n / o p p o s i t e / b i d i r e c t i o n a l i i . Binary Settings (Piggott (to appear)) a. Spread leftward (yes/no) b. Spread rightward (yes/no) Archangeli and Pulleyblank's parameter i s t i e d to another parameter which regulates the d i r e c t i o n of i n i t i a l a ssociation (see, for example, (2.6)), but Piggott's parameter does not make t h i s connection. The b i d i r e c t i o n a l s e t t i n g i n (2.29i) i s achieved i f the YES option i s chosen for both parameters i n (2 . 2 9 i i ) . The featural and rul e parameters investigated i n 2.5 w i l l a l l be binary, c e r t a i n of them providing ON-OFF choices, and others two s p e c i f i e d choices. 2.3.2 Multiple Parameters Dresher and Kaye (1988) encounter cross-parameter dependencies i n t h e i r account of parameterized stress systems. An example i s given involving the three parameters i n (2.30). (2.30) PI The word-tree i s strong on the (LEFT/RIGHT) P2 Feet are (BINARY/UNBOUNDED) P5 Feet are quantity sensitive (QS) (YES/NO) 76 In order for a learner to determine which s e t t i n g of PI i s required, a "window" at the l e f t or r i g h t edge of a word must be sampled. P2 w i l l determine the size of the window, but the s e t t i n g of P2 i s dependent upon the s e t t i n g of P5. If the NO option of P5 i s chosen, then the BINARY option of P2 must also be chosen, and i f the YES option of P5 i s chosen, then P2 must be UNBOUNDED. Thus the default s e t t i n g of P2 w i l l depend upon the appropriate s e t t i n g of P5 i n the language. P2 and P5 must somehow be t i e d together, or the learner may make a choice that i t i s not possible to retreat from. Williams (1987: xi) says of such interconnections: The question i s , how complicated are such contingencies — i n the worst case, one can imagine the parameters were so paralyzingly interconnected that they a l l had to be set "at one time" and the evidence was the union-of a l l the evidence relevant for any of them. Dresher and Kaye do not provide evidence as to how these dependencies are dealt with, but simply note that they e x i s t . Such interdependencies predict that children w i l l remain with the unmarked parameter setting u n t i l a l l the evidence i s i n , then a l l parameters w i l l be reset at once. With regard to the interdependence of default values, i t remains an empirical question how these problems w i l l be sorted out. Wexler and Manzini (1987) i n t h e i r discussion of the parameter associated with anaphors, adopt the L e x i c a l Parameterization Hypothesis (LPH) f i r s t discussed i n Borer (1984). The LPH assumes that parameters are set for i n d i v i d u a l l e x i c a l items, rather than for a construct as a whole. There w i l l be d i f f e r e n t parametric p o s s i b i l i t i e s for 77 each l e x i c a l anaphor. Davis (1987) argues that the adoption of the LPH w i l l lead to "undergeneralization" problems, i . e . random scattering of values of a parameter throughout the lexicon. Part of the reasoning behind the development of the parametric model was to capture generalizations that spread through the syntax and the lexicon, and the adoption of the LPH ignores these generalizations. I assume that the fe a t u r a l and rule parameters discussed i n t h i s and subsequent chapters are parameters which once set, apply to a l l e l i g i b l e l e x i c a l and syntactic categories. 2.3.3 Non-parametric Acquis i t i o n While parameters are assumed to account for many aspects of a c q u i s i t i o n , they obviously cannot account for everything. Chomsky (1981a: 7-8) makes a d i s t i n c t i o n between "core grammar" and the "periphery". Experience -- i n part, a construct based on an i n t e r n a l state given or already attained — serves to f i x the parameters of UG, providing a core grammar, guided perhaps by a structure of preferences and im p l i c a t i o n a l r e l a t i o n s among the parameters of the core theory... But i t i s hardly to be expected that what are c a l l e d "languages" or " d i a l e c t s " or even "ideolects" w i l l conform p r e c i s e l y or perhaps even very c l o s e l y to the systems determined by f i x i n g the parameters of UG. This could only happen under i d e a l i z e d conditions that are never r e a l i z e d i n fact i n the r e a l world of heterogeneous speech communities. Furthermore, each actual "language" w i l l incorporate a periphery of borrowings, h i s t o r i c a l residues, inventions, and so on, which we can hardly expect to — and indeed would not want to — incorporate within a p r i n c i p l e d theory of UG. For such reasons as these, i t i s reasonable to suppose that UG determines a set of core grammars and that what i s a c t u a l l y ' represented i n the mind of an i n d i v i d u a l even under the i d e a l i z a t i o n to a homogeneous speech community would be a core grammar with a periphery of marked elements and constructions. 78 L e x i c a l a c q u i s i t i o n , word meaning, subcategorization requirements and exceptions w i l l a l l be aspects of the periphery. Children w i l l have to acquire the form and meaning of l e x i c a l items and af f i x e s by rote, although UG w i l l c e r t a i n l y supply structure and some content to the l e x i c a l e n t r i e s . While t h i s separation of core and periphery i s l o g i c a l and probably necessary, i t has sparked some concern. Williams (1987) worries that t h i s d i v i s i o n may cause non-parametric aspects of a c q u i s i t i o n to be neglected. Fodor (1989) warns that we must be careful i n assuming d i f f e r e n t a c q u i s i t i o n mechanisms for core and peripheral facts since t h i s may r e s u l t i n p o s i t i n g rule-writing mechanisms a l l over again. She suggests that the n u l l hypothesis, that the same devices that account for core grammar also hold i n the periphery, should be maintained as long as possible. In Chapter 4 I w i l l discuss some aspects of phonological a c q u i s i t i o n that I believe to be peripheral to a core system of p r i n c i p l e s and parameters. I t remains an empirical issue whether or not UG w i l l be able to account for a l l remaining aspects of the a c q u i s i t i o n data that w i l l be investigated i n Chapter 5. 2.4 L e a r n a b i l i t y L e a r n a b i l i t y theory i s the f i e l d of inquiry that developed with the conception of a c q u i s i t i o n given i n (1.2) (e.g. Gold 1967, Wexler and Hamburger 1973, Baker 1979, Berwick 1985). L e a r n a b i l i t y attempts to account for how the 79 c h i l d constructs, the adult grammar based on l i m i t e d input. L e a r n a b i l i t y research has focussed on two areas: the p r i n c i p l e s of UG and the type of input available to the c h i l d . While t h i s i s not a homogeneous f i e l d , there are c e r t a i n working hypotheses that are held to by the majority of researchers. In the following sections I discuss three such hypotheses that I assume i n developing the underspecification a c q u i s i t i o n models: the L e a r n a b i l i t y Condition, the Continuity Assumption and the No-negative Evidence Hypothesis. 2.4.1 The L e a r n a b i l i t y Condition The L e a r n a b i l i t y Condition (Pinker 1979, 1984) i s i n s p i r e d by the fact that a l l children are ultimately successful at language a c q u i s i t i o n . It says that any developmental stage must be attainable v i a an a c q u i s i t i o n mechanism that begins with Universal Grammar and ends up with the adult grammar. Furthermore i t must be possible to convert any intermediate rule system into the adult state of grammar by means of the a c q u i s i t i o n process. This condition has been instrumental i n the development of the p r i n c i p l e s and parameters model of grammar, since i f we assume that both a c q u i s i t i o n a l stages and language-specific differences are the r e s u l t of the r e s e t t i n g of parameters, then by d e f i n i t i o n any a c q u i s i t i o n a l stage w i l l be a possible human grammar. While the L e a r n a b i l i t y Condition appears to be an obvious constraint on a c q u i s i t i o n research, only i n the l a s t ten or f i f t e e n years has i t been generally accepted. The 80 L e a r n a b i l i t y Condition would pro h i b i t many of the grammars written for children's speech during the 1960s and 1970s, which were composed of sets of rules that bore absolutely no resemblance to any type of adult grammar 1 4. The L e a r n a b i l i t y Condition t e l l s us that intermediate stages of a c q u i s i t i o n cannot be studied i n a vacuum — we must be able to account for how they are achieved, and how they develop into the adult system. 2.4.2 The Continuity Condition The Continuity Assumption (Atkinson 1982, Pinker 1984) states that the p r i n c i p l e s and a c q u i s i t i o n mechanisms that are a v a i l a b l e to a c h i l d must be the same over the whole course of development. Within the p r i n c i p l e s and parameters framework the Continuity Assumption says that UG contains a l l the p r i n c i p l e s that are both necessary for a c q u i s i t i o n and for a characterization of the adult grammar. C r o s s - l i n g u i s t i c differences are the r e s u l t of d i f f e r e n t parametric choices. Developmental changes i n the c h i l d ' s grammar are also assumed to be the r e s u l t of parameter switching, and not the r e s u l t of changes i n the c h i l d ' s cognitive makeup. An a l t e r n a t i v e to the Continuity Assumption exists i n the form of the Maturational Hypothesis (Borer and Wexler 1987). This hypothesis assumes that language i s a b i o l o g i c a l mechanism that i s not f u l l y s p e c i f i e d at b i r t h . Certain aspects of UG w i l l "mature" at s p e c i f i e d times, and w i l l i n i t i a t e changes i n the c h i l d ' s grammar. Even i f the 81 appropriate evidence i s available to a c h i l d at some e a r l i e r point, a construction w i l l not be "acquired" u n t i l the p r i n c i p l e regulating i t appears i n UG. This view of a c q u i s i t i o n i s a much less constrained one than the Continuity Assumption, since i t allows the postulation of new j u s t -matured p r i n c i p l e s to account for more d i f f i c u l t aspects of a c q u i s i t i o n . In t h i s thesis I w i l l assume that the phonological a c q u i s i t i o n i s constrained by the Continuity Assumption, and that maturation does not play a part i n the developmental of a phonological system. 2.4.3 The No-negative Evidence Hypothesis The primary l i n g u i s t i c data (PLD) or input a c h i l d receives i s c r u c i a l to the a c q u i s i t i o n of the grammar of a s p e c i f i c language. Without evidence about the language being acquired the c h i l d would never move on from the set of unmarked parameters provided by UG. The No-negative Evidence Hypothesis (Williams 1976, Baker 1979, Berwick 1985) says that only p o s i t i v e evidence i s available to language l e a r n e r s 1 5 . Research by Brown and Hanlon (1970) has demonstrated that d i r e c t negative evidence i n the form of corrections are rare and Braine (1971) has shown that children do not make use of corrections even i f they are provided. The No-Negative Evidence hypothesis also rules out the a v a i l a b i l i t y of i n d i r e c t negative evidence — the determination of ungrammaticality based on computations of the p r o b a b i l i t y of occurrence i n a given amount of input. The a c q u i s i t i o n 82 l i t e r a t u r e generally assumes that neither type of negative evidence i s ava i l a b l e , although there have been t h e o r e t i c a l suggestions (e.g. Chomsky 1981, Lasnik 1989) that c h i l d r e n may make use of i n d i r e c t negative evidence. The No-negative Evidence Hypothesis applied to a parameter s e t t i n g theory of grammar makes predictions about markedness and l i n g u i s t i c systems, i n the form of the Subset P r i n c i p l e (Berwick 1985, Wexler and Manzini 1987). If the language being acquired i s a proper subset of a language generated by the i n i t i a l s e tting of a parameter, the c h i l d w i l l never learn on the basis of po s i t i v e evidence that the incorrect grammar has been chosen. Only i f there i s some markedness condition associated to parameters s t a t i n g that the i n i t i a l parameter set t i n g must generate the more s p e c i f i c language w i l l p o s i t i v e evidence from the PLD be enough to t e l l the c h i l d that the marked parameter set t i n g i s required, i f the language being acquired i s the more general one. Negative evidence has been discussed i n phonological development with regard to negative constraints. Negative constraints have been used i n a wide range of work i n phonology (e.g. Paradis 1988, Calabrese 1988) to block c e r t a i n types of structures or feature combinations. In some cases the constraints t r i g g e r repair rules which fix-up the disallowed representation. The question i s , how can the non-occurrence of some phonological item be learned when i t does not occur? Ingram (1990) discusses t h i s l e a r n a b i l i t y issue, arguing that negative constraints can be learned on the basis 83 of p o s i t i v e evidence. He hypothesizes that the child, may f i r s t acquire the r e s t r i c t i o n i n the form of an i f - t h e n condition, which at some point becomes restructured as a negative condition. Ingram suggests that t h i s r e s t r u c t i n g may take place because the negative condition i s less complex than the p o s i t i v e condition. In Chapter 3 two types of negative constraints w i l l be posited for Hungarian. The f i r s t i s a marking condition which i s a formalization of Structure Preservation (see 2.1.1.4), and I show that i n fact t h i s condition cannot be stated i n a p o s i t i v e fashion, given the bounds of a theory of Radical Underspecification. The second i s a surface phonetic constraint which appears to t r i g g e r a repair strategy that fixes-up the disallowed segment so that i t can eventually be r e a l i z e d . 2.5 Parametric Approaches to Underspecification In t h i s section I w i l l combine the p r i n c i p l e s and parameters model of grammar with underspecification theory, i n order to develop a phonological theory which makes predictions about the course of a c q u i s i t i o n . To my knowledge t h i s has been previously attempted only once i n Ingram (1990) 1 S. Ingram's work examines the l o g i c a l l e a r n a b i l i t y of the Wikchamni v o c a l i c system as given i n Archangeli (1985). Ingram concludes that i t i s possible to view the a c q u i s i t i o n of t h i s system as a series of developmental stages leading to a r a d i c a l l y underspecified adult system. My work d i f f e r s from Ingram's i n two important respects. F i r s t , the predictions 84 for the a c q u i s i t i o n of Hungarian and Spanish that w i l l be developed i n Chapter 4 are based e n t i r e l y on the primitives of the theories of Radical and Contrastive Underspecification, without making reference to s p e c i f i c p r i n c i p l e s of a c q u i s i t i o n . Secondly, I w i l l use real-time language a c q u i s i t i o n data to test the parametric underspecification predictions, where Ingram uses hypothetical data to argue his case. The picture of phonological a c q u i s i t i o n that I assume i s the following. UG supplies each c h i l d with the innate aspects of l i n g u i s t i c structure i n the form of a set of p r i n c i p l e s and parameters. Resetting of a parameter can only be accomplished given p o s i t i v e evidence from the PLD that the marked option i s c a l l e d for (No-Negative Evidence Hypothesis, see 2.4.3). Since the UG system forms a template for the l i n g u i s t i c system u n t i l such time as the parameters are reset, at the very e a r l i e s t stages children's phonological systems cross-l i n g u i s t i c a l l y w i l l be characterized by the rules and parameter settings of UG. Given exposure to the input and time to change parameters, children's phonological systems w i l l more c l o s e l y resemble the adult systems being acquired and w i l l look less l i k e the systems supplied by UG. The c h i l d ' s system at any one time w i l l contain a l l aspects of UG that are invariable (the Continuity Assumption, see 2.4.2), and a c q u i s i t i o n can proceed given UG and some exposure to the input (the L e a r n a b i l i t y Condition, see 2.4.1). In the following sections, the theories of Radical and 85 Contrastive Underspecification w i l l be discussed as p r i n c i p l e s and parameters models of phonology. I w i l l focus on the same two problems that have been the centre of research i n l e a r n a b i l i t y theory -- the p r i n c i p l e s of UG that are necessary with the adoption of either CU or RU, and the input that must be a v a i l a b l e to the c h i l d i n order that the underspecification system of a language be determinable. There w i l l be no major changes to the theories as outlined i n 2.2.1 and 2.2.2, although there w i l l be c e r t a i n additions and elaborations that allow these theories to be viewed as parametric models. 2.5.1 Parameters and Radical Underspecification 2.5.1.1 UG i n RU In Archangeli (1984) RU i s investigated as a theory of ac q u i s i t i o n , although t h i s research was c a r r i e d out before the p r i n c i p l e s and parameters approach to grammar was f u l l y developed, and before c e r t a i n c r u c i a l aspects of phonological theory, such as the theory of feature geometry, came to l i g h t . Archangeli's a c q u i s i t i o n model i s given i n (2.31). 86 (2.31) A c q u i s i t i o n - Archangeli (1984) UG Isolating featural oppositions ALPHABET FORMATION ALPHABET Default rules > Feature Min. Principle Matrix (context-free and Feature Value Min. Complement context-sensitive) Principle rules c Input Archangeli assumes that UG supplies the language learner with the knowledge of how to i s o l a t e sounds based on feature oppositions and a set of universal default r u l e s 1 " 7 . The components of UG, combined with language-specific phonological information serve as input to a procedure c a l l e d Alphabet Formation, which creates the Alphabet or underlying phonological system of a language. Alphabet Formation selects one value of a feature as the underlyingly marked value, with an accompanying redundancy rule f i l l i n g i n the unmarked value. 87 (2.32) Alphabet Formation 1 8 1. Given an opposition [aF] -- [-aF] i n environment Q i n underlying representation, one value "a" (where "-a" i s the u n i v e r s a l l y marked value) i s selected as the matrix value for F i n Q and the other value i s s p e c i f i e d by an automatically formed complement value: [ ] —> [-aF] / Q 2. In the absence of language i n t e r n a l motivation f o r s e l e c t i n g "a" as the matrix value for a feature F, the value "-a" i s selected as the matrix value where: [ ] —> [aF] / Q i s a member of the set of default r u l e s . The f i r s t clause of (2.32) says that i f there i s p o s i t i v e evidence i n the language that the u n i v e r s a l l y predictable value of a feature must be present underlyingly, then a complement r u l e w i l l be created to redundantly specify the opposite value supplied by UG. The second clause of (2.32) says that i f there i s no l i n g u i s t i c evidence that a complement value i s required, then the default r u l e holds. Archangeli assumes that Alphabet Formation includes the Feature Minimization P r i n c i p l e : 88 (2.33) Feature Minimization P r i n c i p l e (Archangeli 1984: 50) A grammar i s most highly valued when underlying representations include the minimal number of features necessary to make d i f f e r e n t the phonemes of the language. and a p r i n c i p l e which says minimize the number of feature values s p e c i f i e d i n the matrix component. The Alphabet of a language consists of the set of complement rules arrived at a f t e r Complement Rule Formation and a matrix component. The matrix i s the array of underlyingly marked feature values needed i n a language. The matrix includes only feature markings that are not provided by e i t h e r the default rules of UG or the complement rules of the language. Several additions must be made to the conception of UG given i n (2.31) i f RU i s to be seriously considered as a p r i n c i p l e s and parameters model. The f i r s t i s the addition of the P r i n c i p l e s of Systematization and Minimal Redundancy to UG. Since RU i s based on the notion that minimal redundancy i s an organizing p r i n c i p l e of the grammar, I believe the p r o h i b i t i o n on redundancy i s better stated as a p r i n c i p l e of UG than as s p e c i f i c r e s t r i c t i o n s on the i n c l u s i o n of redundancy i n the learning process as i n (2.34). 89 (2.34) The Minimal Redundancy Condition (MRC) 1 9 a. Underlying representations do not contain redundant information. b. The most highly valued system contains the minimal number of features and feature values needed to d i s t i n g u i s h the inventory of a language. The MRC w i l l t e l l the c h i l d that underlying representations are set up using only non-redundant featural information. The MRC can also be used as a gauge of markedness, where the l e a s t marked segments w i l l be those with the fewest feature markings. In addition to the MRC, we must also take the a c q u i s i t i o n model back one step and assume that there i s a p r i n c i p l e which t e l l s the c h i l d to create a phonological system for the language, i n which feature values have possible binary settings, and c e r t a i n feature values are underlying. This can be stated as the P r i n c i p l e of Systematization. (2.35) The RU P r i n c i p l e of Systematization Feature values that are not predictable are marked. This set of markings forms the underlying matrix of a language. The MRC w i l l then ensure that redundancy rules e x i s t to f i l l i n predictable feature values and that features whose values are wholly predictable w i l l not be a part of the underlying system. The MRC w i l l also lead the c h i l d to r e a l i z e that a 90 segment that i s s p e c i f i e d only by predictable feature values w i l l not take part i n the l e x i c a l phonological rules of the language. ' Archangeli and Pulleyblank (1986) and Archangeli (1988) assume that default rules are the i n i t i a l unmarked parameters of UG, but they do not look at what t h i s assumption means for a c q u i s i t i o n . The set of default rules for vowels that I w i l l use i n t h i s thesis are given i n (2.36). (2.36) Universal Default Rules for Vowels 2 0 Context-free rules FCRs 1. [ ] —> [-low] 5. [+low] —> [-high] 2. [ ] —> [+high] 6. [+low] —> [+back] 3. [ ] --> [-back] 7. [+low] --> [-round] 4« [ 1 — > [-round] 8. [+back] —> [+round] [-low] 9. [-back] —> [-round] [-low] Rules 1-3 and 5-6 are taken from Archangeli (1988), r u l e 4 i s taken from Archangeli and Pulleyblank (1986) and r u l e 8 from Archangeli (1984) 2 1. Rules 8 and 9 are used i n Archangeli (1984). This p a r t i c u l a r set of rules i s used because 1) they appear to define the majority of the feature redundancies that r e l a t e to v o c a l i c systems (with the exclusion of the features that r e l a t e to [ATR])/ and 2) they allow us to use a comparable set of universal rules i n both theories of underspecification investigated i n t h i s and subsequent 91 chapters. The context-sensitive rules w i l l be c a l l e d feature co-occurrence redundancies (FCRs) since they define redundant values based on the co-occurrence of p a r t i c u l a r feature values. FCRs are often stated i n the format [ ] —> [+F]/[+G] (Archangeli 1984) but Kiparsky (1985)/ Archangeli (1988) and Steriade (1987) adopt the format [+G] —> [+F]. I assume these two formats are notational equivalents, and state the rules using the l a t t e r form. The MRC, the RU P r i n c i p l e of Systematization and the default rules i n (2.36) w i l l t e l l the c h i l d to represent a 3 vowel system /i , a , u / and a 5 vowel system /i,e,a,o,u/ as i n (2.37). The RU P r i n c i p l e of Systematization and the MRC t e l l the c h i l d that feature values that are not predictable are marked underlyingly. If the default rules provide the predictable values, then only those values that are not given by the rules i n (2.36) w i l l be marked underlyingly. (2.37) a. Universal 3 Vowel System b. Universal 5 Vowel System i a u i e a o u high low + back + + round / i / i s the t o t a l l y unspecified vowel i n each system, and the le a s t marked vowel i n the system by the MRC22. The features [high] and [round] are redundant i n the three vowel system and high low + back + round 92 the feature [round] i s redundant i n the f i v e vowel system, since a l l values of these features are t o t a l l y predictable, given the rules i n (2.36). To achieve a f u l l y s p e c i f i e d matrix for the vowel systems i n (2.37) the default rules i n (2.36) apply to.the matrix when they can, governed by the Ordering Constraints i n (2.19) and the EC i n (2.3). I assume that these constraints are also part of UG. Archangeli and Pulleyblank (1986) argue that the default rules as stated i n (2.36) are derivative from the typology of rules given i n (2.21), and therefore are e a s i l y stated i n a parametric framework. I adopt t h i s conception of default r u l e s , although below I w i l l expand on how they w i l l function as parameters. The typology given i n (2.21) i s repeated here as (2.38). (2.38) I. a. ( i n s e r t j d e f a u i t / d e l e t e b. (maximal)c«fa U i t/minimal c. ( c o n t e n t ) d S f a u i t / s t r u c t u r e d. (same d i r e c t i o n ) d e f a u l t / o p p o s i t e d i r e c t i o n / b i d i r e c t i o n a l I I . node or feature(s) I I I . t r i g g e r condition target condition A default rule i s the i n s t a n t i a t i o n of the default settings of the function parameters (2.381), with the argument and/or target condition s p e c i f i e d by UG ((2.38II) and (2.38III)). (2.22), repeated here as (2.39) i s an example of a context-93 free default rule stated i n terms of the parameters i n t h i s typology. (2.39) inserts [-low] as the redundant s p e c i f i c a t i o n of the feature [low]. (2.39) Default [-low] I. a. (insert) c. (content) d. (same direction) I I . [-low] I assume that the target/trigger condition ( i . e . (2.38III)) i s not parameterizable, since i t refers to a feature value that must be present i n order for the r u l e to apply. If p o s i t i v e evidence shows the c h i l d that a context-free default rule does not hold i n the language, then several options are a v a i l a b l e . Any of the 4 function parameters i n (2.381) could be switched to the marked option, with the argument remaining the same. Resetting any one of these parameters would have no e f f e c t s i n the phonology, however, since the argument of the rule i s not an underlying feature that can be deleted or spread, d i r e c t i o n a l i t y i s i r r e l e v a n t , and there i s no target condition to which to r e l a t e the maximal/minimal. The only workable option i s to switch the argument parameter ( i . e . (2.38II) to the opposite value, which w i l l then force the u n i v e r s a l l y predictable value to be used underlyingly. In t h i s view context-free default rules represent binary parameters with two s p e c i f i e d options. The unmarked option i s given by UG, the other w i l l be a r r i v e d at 94 by switching the argument parameter to the opposite value. An implementation of the res e t t i n g of a context-free f e a t u r a l parameter i s given i n Archangeli and Pulleyblank (1989). There i t i s argued that Yoruba underlyingly makes use of [-ATR], the value of [ATR] which i s generally assumed to be predictable. In the parametric model t h i s means that i n Yoruba the parameter for [ATR] has been switched to the marked value. A c h i l d learning Yoruba w i l l i n i t i a l l y assume the default rules of UG where [-ATR] i s predictable. Experience w i l l t e l l the c h i l d the parameter requires r e s e t t i n g , and once t h i s i s accomplished [+ATR] w i l l become predictable. The c h i l d ' s underlying v o c a l i c system w i l l be restructured when the parameter for [ATR] i s switched 2 3, as shown i n (2.40). (2.40a) represents the i n i t i a l system supplied by UG, while (2.40b) represents the system a f t e r the parameter regulating [ATR] has been reset at the marked option (for s i m p l i c i t y ' s sake only the context-free default rules are given). 95 (2.40) a I n i t i a l Default rules: [ ] —> [+high] [ ] --> [-low] [ ] —> [-back] [ ] —> [-ATR] i e E a o 0 u high low back ATR + + + + + + + + b. Yoruba Default r u l e s : [ ] —> [+high] [ ] —> [-low] [ ] --> [-back] Complement r u l e : [ ] --> [+ATR] i e E a o 0 u high - -low + back + + + ATR The c h i l d ' s i n i t i a l hypothesis i s that a l l the default rules of UG hold i n the language being acquired, and the c h i l d acquiring Yoruba w i l l then i n i t i a l l y assume that [+ATR] i s the marked feature value. When the c h i l d learns that [-ATR] must be the l e x i c a l l y marked feature value (how the c h i l d learns t h i s w i l l be discussed i n 2.5.1.2), the context-free f e a t u r a l parameter for [ATR] w i l l be switched to the marked option. The e f f e c t of t h i s parameter switching w i l l be to replace the default r u l e for [ATR] with a complement rule which ins e r t s [+ATR] redundantly, and to restructure the underlying feature markings so that [-ATR] i s the l e x i c a l l y marked value. Archangeli and Pulleyblank (1986) do not discuss how FCRs would be dealt with i n a parametric model, however; Archangeli (1988) b r i e f l y discusses t h i s issue with regard to the v o c a l i c 96 system of Auca. The inventory of Auca has two low vowels /a/ and /*/, and does not have a high back vowel. Archangeli argues that the grammar of Auca must formally p r o h i b i t a p p l i c a t i o n of the default rule [+low] --> [+back] (rule 6 i n (2.36)), because i n t h i s language the rule does not supply the predictable values of both low vowels. This i s equivalent to saying that the unmarked parameter set t i n g of an FCR i s ON or APPLICATION and the marked set t i n g i s SUPPRESSION. If an FCR i s suppressed language-specifically, then the feature value that i s no longer predictable must be given underlyingly by the p r i n c i p l e s i n (2.34) and (2.35). Given these assumptions, the c h i l d ' s i n i t i a l and l a t e r representations of the vowels of Auca would be as i n (2.41) 2 4. (2.41) a. I n i t i a l system b. Auca system a/ae e i o a a e e i o high - - -low + + + back + + + In the i n i t i a l representation, given i n (2.41a), the c h i l d assumes that the default rules ( i . e . those i n (2.36)) hold absolutely. Since rule 6 i n (2.36) t e l l s the c h i l d that [+low] vowels are redundantly [+back], the c h i l d w i l l assume that t h i s holds for both [+low] vowels. Both /a/ and / a e / w i l l be marked s o l e l y as [+low] and there w i l l be d i s t i n c t representations for only 4 vowels 2 5. When the c h i l d learns that there are i n fact two phonologically d i s t i n c t low vowels 97 i n the language (how t h i s i s learned i s discussed i n 2.5.1.2), the context-sensitive featural parameter (FCR) w i l l be reset to the marked option. The e f f e c t of r e s e t t i n g t h i s parameter w i l l be to underlyingly mark the feature value that was made redundant by the FCR, and to eliminate the FCR from the language-specific grammar. Thus i n Auca [+back] w i l l be l e x i c a l l y marked for /a/, since there i s no longer a context-s e n s i t i v e rule to in s e r t [+back] redundantly. The l e x i c a l marking of [+back] i s required by the RU P r i n c i p l e of Systematization i n (2.35), which says that feature values that are not redundant are marked underlyingly. The r e s t r u c t u r i n g that takes place when a context-sensitive parameter i s reset to the marked option requires only the addition of feature values to the phonological system, rather than the removal and replacement of feature values as occurs when a context-free parameter i s reset. An a l t e r n a t i v e conception of the parameterization of FCRs i s that r e s e t t i n g does not eliminate the FCR from the language-specific grammar, but rather forces the s p e c i f i c a t i o n of the marked feature value l a n g u a g e - s p e c i f i c a l l y 2 S . In t h i s view the e f f e c t of resetting a context-sensitive parameter i s not SUPPRESSION of the r u l e , but rather ADD A MARKED FEATURE VALUE UNDERLYINGLY. Given that I assume rule 6 i n (2.36) i s n o t a t i o n a l l y equivalent to a rule which says [ ] —> [+back] /[+low], both rules say 'the unmarked value of [back] i s [+back] for a segment that i s also [+low]'. If t h i s i s the unmarked s e t t i n g of the parameter, then the marked s e t t i n g 98 w i l l be 'the marked value of [back] i s [-back] for a segment that i s also [+low]'. In the case of Auca, when the c h i l d encounters evidence that [+back] i s not redundant for a l l [+low] vowels, the parameter for rule 6 i n (2.36) w i l l be reset to the marked option, which w i l l force [-back] to be marked underlyingly for / a e /. In t h i s conception of parameterization, the default rule w i l l have to remain i n the language-specific grammar, i n order to supply the u n i v e r s a l l y redundant value. Rule 6 i n (2.36) w i l l remain i n the grammar of Auca to provide the redundant value of /a/. Given t h i s conception of the marked option of context-sensitive parameters, the c h i l d ' s i n i t i a l and restructured representations of the vowels of Auca w i l l be as given i n (2.42). (2.42) a. I n i t i a l system b. Auca system a/ae e i o a ae e i o high - - -low + • + + back + + The problem with t h i s second conception of the parameterization of context-sensitive featural parameters (FCRs) can be demonstrated i f we assume that Auca has an as s i m i l a t i o n rule which spreads the feature [+back]. Given the conception of parameterization where the marked option of a context-sensitive rule i s ADD A MARKED FEATURE VALUE UNDERLYINGLY, rul e 6 w i l l be a part of the grammar of Auca and 99 w i l l be ordered p r i o r to the rule spreading [+back] by the Redundancy Rule Ordering Constraint i n (2.19b). /a/ and /o/ w i l l then be triggers of the rule, but /a?/, which we would expect to be a target, w i l l not be, since i t i s s p e c i f i e d as [-back]. This i s exactly the case that arises i n Hungarian, as w i l l be discussed i n 3.1.3. In Hungarian i t can be shown that the low front vowel must be a possible target of the a s s i m i l a t i o n r u l e , and therefore cannot be marked underlyingly as [-back]. For t h i s reason I w i l l adopt Archangeli's (1988) conception of context-sensitive parameter r e s e t t i n g , where the marked option i s SUPPRESSION of the rule i t s e l f , with concomitant marking of the u n i v e r s a l l y redundant feature value. In the parametric model of RU outlined above there are two ways that the grammar of a s p e c i f i c language can d i f f e r from the grammar supplied by UG: 1) the opposite value of a universal context-free feature may be marked underlyingly i n an across-the-board fashion, as the r e s u l t of the r e s e t t i n g of a context-free default rule to the marked option; or 2) a u n i v e r s a l l y redundant feature value may be marked i n a s p e c i f i c context, as the r e s u l t of the r e s e t t i n g of an FCR to the marked option. The device which resets a context-free featural parameter i s i d e n t i c a l to Alphabet Formation i n (2.27). Alphabet Formation cannot, however, account for the r e s e t t i n g of an FCR parameter, so I w i l l abandon Alphabet Formation altogether and simply assume that i n a parametric RU theory the learning procedure includes a parameter switching 100 mechanism. The r u l e typology i n (2.38), i n addition to supplying the formal statement of default rules, also supplies the possible phonological rule types for RU, as discussed i n 2.2.1.2. Default rules and rules of Epenthesis are the unmarked rules supplied by UG, since these are the rules that require only the default function parameter settings. Deletion, Insertion of an association l i n e and Spread are s l i g h t l y more marked ru l e s , since they require a function parameter to be reset. In addition, the d i r e c t i o n a l i t y parameter i s assumed i n the default case to be t i e d i n to the d i r e c t i o n a l i t y of the i n i t i a l association of anchors and autosegments shown i n (2.6). If t h i s i n i t i a l association proceeds from L—>R then that w i l l be unmarked d i r e c t i o n of spreading for a phonological r u l e . If i n i t i a l association i s R—>L, then that again w i l l be the unmarked d i r e c t i o n of association for a phonological r u l e . Only i n the marked case w i l l phonological rules and i n i t i a l association operate i n d i f f e r e n t d i r e c t i o n s . Once the RU ac q u i s i t i o n model i n (2.31) i s revised so that UG includes the MRC and Systematization P r i n c i p l e s , the possible parameterization of rule operations and arguments, the Ordering Constraints, the p r i n c i p l e s of Autosegmental and Lex i c a l Phonology discussed i n 2.1 (the OCP (2.8), the EC (2.3), Structure Preservation (in 2.1.1.4), and the Association Conventions (2.6)), i t w i l l resemble (2.43). 101 (2.43) Revised Model of RU Acquisition 2" 7 UG Features Principles Principles of Autoseg. and Lex. Phonology Ordering Principles Ordering Constraints Elsewhere Condition MRC Systematization Parameters 3" Rule Functions insert maximal content same direction Rule Arguments 3 9 context-free [+high] [-low] [-back] [-round] FCRs [-high]/[+low] [+back]/[+low] [+rnd]/[+back] [-low] [-rnd]/[+low] [-rnd]/[-back] [-low] j Input] 2.5.1.2 Input i n RU If used to constrain a p r i n c i p l e s and parameters account of grammar, the No-Negative Evidence Hypothesis i n 2.4.3 says that only p o s i t i v e evidence can be used as evidence for the re s e t t i n g of a parameter. In the parametric theory of RU p o s i t i v e evidence w i l l be needed to tr i g g e r the r e s e t t i n g of context-free and context-sensitive featural parameters and f o r the r e s e t t i n g of function parameters c o n t r o l l i n g the possible types of phonological rules. In order to reset a context-free parameter, the input w i l l have to contain alternations which LEARNING MECHANISM Parameter Switching Device ALPHABET Matrix Complement rules 102 demonstrate that a un i v e r s a l l y predictable feature value must be marked underlyingly. In Yoruba, for example (described i n Archangeli and Pulleyblank 1989), the context-free parameter for [ATR] w i l l be reset when the c h i l d encounters evidence that [-ATR] must be the underlyingly marked feature value. Such evidence w i l l come from alternations i n the input showing that high vowels do not have [ATR] r e s t r i c t i o n s , that only /a/, /O/ and /E/ are triggers of [ATR] harmony 3 0, and that / i / acts as a t o t a l l y unspecified vowel i n the language (see Pulleyblank 1988). Evidence that a vowel has no underlying s p e c i f i c a t i o n s w i l l come from i d i o s y n c r a t i c behaviour such as transparency i n harmony processes or the fact that i t regu l a r l y disappears i n ce r t a i n environments. The alternations involved i n [ATR] Spread w i l l lead the c h i l d towards the correct underlying s p e c i f i c a t i o n of the vowels of Yoruba, and to the correct statement of the rule of [ATR] Spread. Thus there i s a great deal of i n t e r a c t i o n i n RU between the determination of how phonological rules operate and the r e s e t t i n g of featural parameters 3 1. In the case of Auca, the FCR which says that [+back] i s a predictable feature of low vowels must be reset before the c h i l d can represent both low vowels d i s t i n c t l y i n the phonological inventory. Evidence t r i g g e r i n g the r e s e t t i n g of t h i s context-sensitive parameter w i l l come from alternations i n the language which demonstrate that /a/ and /ae/ are phonologically d i s t i n c t segments. Evidence for t h i s w i l l come from minimal pairs or from alternations that r e s u l t from the 103 a p p l i c a t i o n of phonological rules. 2.5.2 Parameters and Contrastive Underspecification 2.5.2.1 UG i n CU We can envisage a model/ s i m i l a r i n structure to that i n (2.43) , as a CU parameter se t t i n g model of a c q u i s i t i o n . In order to function as an a c q u i s i t i o n theory t h i s 1 model must have a p r i n c i p l e which t e l l s the c h i l d that only contrastive feature values are present underlyingly, and i t must have a theory of how redundancy rules function as featural parameters. These are additions to the theory as proposed i n Steriade (1987). As a f i r s t step I assume that UG contains a p r i n c i p l e t e l l i n g the c h i l d to create a CU phonological system for the language, just as there was a p r i n c i p l e regulating the establishment of a phonological system i n RU (see (2.35). I state t h i s as the CU P r i n c i p l e of Systematization. (2.44) The CU P r i n c i p l e of Systematization Feature values that are used to contrast segments are marked underlyingly. This set of markings forms the underlying matrix of a language. A p r i n c i p l e p r o h i b i t i n g non-contrastive redundancy i n the system i s also required, since CU assumes that only contrastive n o n - t r i v i a l feature markings e x i s t underlyingly (see 2.2.2.1). This p r i n c i p l e w i l l ensure that only the minimal number of contrastive markings are included i n the 104 matrix, and w i l l lead to a method for determining the markedness of phonological systems. (2.45) The R e s t r i c t i v e Redundancy Condition (RRC) a. Underlying representations do not contain feature values that are not used c o n t r a s t i v e l y . b. The most highly valued system contains the fewest number of features and feature values needed to c o n t r a s t i v e l y d i s t i n g u i s h the inventory of a language. (2.45a) w i l l t e l l the c h i l d to mark only those feature values that are required for contrastive s p e c i f i c a t i o n or that are not given by universal rules, while the second clause says that a segment (or system) with n feature markings w i l l be less marked than a segment (or system) with n+1 markings. In 2.2.2.2 the R-rules of CU are discussed. R-rules are the redundancy rules which f i l l i n feature co-occurrence r e s t r i c t i o n s . Steriade wishes to eliminate language-specific ordering relationships between phonological rules and R-rules, even though i n practice she does not always do t h i s (see the analysis of Hungarian Back Harmony discussed i n 2.2.2.2). She says (Steriade 1987: 357) By deciding that [round] i s single-valued we can also avoid the assumption of language-specific orderings between redundancy rules introducing R-values and D-values. It i s c l e a r l y Steriade's intent that such language-specific orderings (permitted i n RU by the RROC) be prohibited. In addition, i n the theory of CU transparency e f f e c t s can only 105 e x i s t i f a given class of segments lacks redundant feature values i n the phonology, and these transparency e f f e c t s would would be destroyed i f R-rules were permitted to apply p r i o r to the a p p l i c a t i o n of phonological rules. I therefore assume that ordering relationships between phonological rules and redundancy rules are prohibited i n the CU parametric model and consequently that R-rules w i l l always apply l a t e i n the de r i v a t i o n i I state t h i s ordering r e s t r i c t i o n as the CU Ordering P r i n c i p l e . (2.46) CU Ordering P r i n c i p l e R-rules apply as l a t e as possible i n the grammar. Several obstacles are encountered i n attempting to tr a n s l a t e the s p e c i f i c s of CU into a p r i n c i p l e s and parameters model of a c q u i s i t i o n . These problems stem from the fact that i n each analysis i n Steriade (1987) the contrastive s p e c i f i c a t i o n s of only one or two features are given. As a r e s u l t i t i s not always clear which R-rules are needed or what t h e i r exact status i s . I w i l l attempt to resolve some of these problems by examining the f i v e vowel systems of Ainu (described i n Steriade 1987) and Auca (described i n Archangeli 1988) . The symmetrical 5 vowel system of Ainu w i l l be c o n t r a s t i v e l y s p e c i f i e d as i n (2.47) (Steriade 1987 discusses only the s p e c i f i c a t i o n s of [back] for Ainu -- I have supplied the other values). 106 (2.47) Ainu i e a o u high + + low + back + + The s p e c i f i c a t i o n s i n (2.47) are arrived at by determining which segments contrast within the realm of a singl e feature. The pairs / i / and /e/ and /u/ and /o/ contrast with each other i n terms of the feature [high], so each receives a s p e c i f i c a t i o n for that feature, /a/ and to/ are both back non-high vowels which contrast i n lowness, so /a/ i s marked [+low] and /o/ i s marked [-low]. The four non-low vowels contrast i n backness, so the front vowels are marked [-back] and the back vowels [+back]. None of the vowels contrast with regard to the feature [round]. The set of R-rules given i n (2.48) w i l l provide the f u l l y s p e c i f i e d system of Ainu. (2.48) R-rules 1. [+low] > [-high] 2. [+low] > [+back] 3. [+low] > [-round] 4. [+back] > [+round] [-low] 5. [-back] > [-round] [-low] 6. [+high] > [-low] 7. [-high] > [-low] [-back] 107 This set of rules i s made, up of 3 rules given i n Steriade (1987) ( i . e . rules 1, 2 and 5 i n (2.48)), supplemented by rules 3 and 4 which ins e r t the redundant values of [round] and rules 6 and 7 which ins e r t the redundant s p e c i f i c a t i o n s for [ l o w ] 3 2 . Steriade (1987) argues that [round] may be a monovalent feature, but because i t i s argued i n 3.1 that [-round] must be an accessible feature value i n Hungarian, I w i l l assume that [round], l i k e [back], [high] and [low], has binary options. I assume that the rules i n (2.48) are the set of R-rules provided by UG, which w i l l constrain children's e a r l i e s t phonological systems. The f i r s t 5 rules i n (2.48) are i d e n t i c a l to set of universal FCRs given i n (2.36) for RU. In RU the redundant s p e c i f i c a t i o n s for [low] are r e a l i z e d through context-free r u l e 4 i n (2.36), but since CU does not allow for context-free rules, these redundancies must be stated i n a context-sensitive manner, as rules 6 and 7 i n (2.48) . Auca has a f i v e vowel system which i s c o n t r a s t i v e l y s p e c i f i e d i n (2.49). (2.49) Auca i e o a ae high + low - - + + back - + + -round The underlying s p e c i f i c a t i o n s are arrived at as follows, / i / and /e/ contrast i n terms of height, so / i / i s marked [+high] 108 and /e/ [-high], /e/, /o/, /a/ and / a e / contrast i n lowness and backness, and so each receives a s p e c i f i c a t i o n for those features. The feature [round] i s t o t a l l y redundant i n t h i s system. In applying the R-rules i n (2.48) to t h i s system, i t becomes apparent that two additional R-rules must be used to achieve f u l l s p e c i f i c a t i o n of /o/ and / i / . These rules are required because of the asymmetrical system of Auca, so I w i l l assume that they are language-specific r u l e s . The R-rules needed to f i l l i n a l l unspecified values i n (2.49) are those i n (2.50) . (2.50) R-rules of Auca Universal: 1. [+low] -> [-high] --> [+back] •-> [-round] •-> [+round] 2. [+low] 3. [+low] 4. [+back] [-low] 5. [-back] > [-round] [-low] 6. [+high] > [-low] 7. [-high] > [-low] [-back] Language-specific: 8. [+back] > [-high] [-low] 9. [+high] > [-back] 109 Steriade (1987) argues that the universal/language-specific d i s t i n c t i o n i s not an important one i n underspecification theory. An asymmetrical system such as that of Auca, however, suggests such a d i s t i n c t i o n must exi s t insofar as R-rules are concerned. Rules 8 and 9 i n (2.50) are e s s e n t i a l to achieve f u l l s p e c i f i c a t i o n of the Auca system i n (2.49), yet i t i s not obvious that we would want to include rules such as these i n a universal core set of R-rules. In applying the universal R-rules to achieve f u l l s p e c i f i c a t i o n of the Auca system, we f i n d that r u l e 2 does not serve a r o l e . Rule 2 says that [+low] vowels are redundantly [+back], and while t h i s holds for /a/, i t i s not true of /ae/. This r u l e i s not used to f i l l i n a the redundant value of [back] f o r e i t h e r /a/ or /ae/, since these two vowels contrast with regard to t h i s feature, and therefore by the CU P r i n c i p l e of Systematization i n (2.44) both segments are underlyingly s p e c i f i e d for backness. As was discussed e a r l i e r , i n a parametric account of underspecification theory the universal redundancy rules are assumed to be the unmarked parameters supplied by UG. In CU t h i s means that the R-rules supplied by UG ( i . e . those i n (2.48)) w i l l apply i n the c h i l d ' s i n i t i a l phonological system, and values not supplied by these rules w i l l be marked underlyingly. When the c h i l d discovers through p o s i t i v e evidence that two segments contrast with regard to a given feature value (when UG predicts that these segments do not contrast), the c h i l d w i l l have to reset an R-rule parameter. 110 The e f f e c t s of the r e s e t t i n g of an R-rule are s l i g h t l y d i f f e r e n t than the e f f e c t s of r e s e t t i n g a context-sensitive f e a t u r a l parameter i n RU. In CU r e s e t t i n g means that by (2.44) the contrastive feature s p e c i f i c a t i o n s w i l l be added underlyingly, and the R-rule may remain i n the language-s p e c i f i c grammar, although i t w i l l no longer have any p a r t i c u l a r function. Thus i n CU we can say that the unmarked set t i n g of an R-rule i s ON or application, while the marked se t t i n g i s simply OFF, with the contrastive values s p e c i f i e d underlyingly by the CU P r i n c i p l e of Systematization i n (2.44). According to t h i s parametric account of CU, i n the i n i t i a l phonological system of the c h i l d acquiring Auca the R-rules supplied by UG w i l l hold absolutely. Feature values that are redundantly supplied by these rules w i l l not be a part of the underlying representation, giving the s p e c i f i c a t i o n s i n (2.51). (2.51) I n i t i a l S p e c i f i c a t i o n of Auca i e o a/ae high + low - + back - + round Since R-rule 2 i n (2.50) gives [+back] as a redundant s p e c i f i c a t i o n for [+low] vowels, neither /a/ nor / a e / w i l l be marked for the feature [back] and the representations of these I l l vowels w i l l be nondistinct. By rule 2 the low vowel w i l l surface as [a]. The contrastive specification of these 4 vowels contrasts /a -ae/ and /o/ for lowness, / i / and /e/ for height and /e/ and /o/ for backness. Positive evidence w i l l be required to t e l l the child that in fact /a/ and / a e / are phonologically distinct vowels. Once such evidence is encountered the parameter giving [+back] as a redundant value on low vowels w i l l be reset, and the contrastive values of the feature [back] w i l l be added to the representations of / a e / and /a/. This in turn triggers the addition of the contrastive specification [-low] for /e/, which now contrasts with / a e / with regard to this feature. We can assume that R-rule 2 may remain in the language-particular grammar of Auca, although i t w i l l have no particular effect. This distinction between what happens to a context-sensitive rule in CU and RU occurs because the underlyingly specifications in CU are determined by which features contrast, whereas in RU they are determined by what is is non-redundant. Once the child has marked the contrastive values of [back] for /a/ and / a e / , and [low] for /e/, the adult specified system given in (2.49) w i l l have been achieved. The child acquiring Ainu w i l l also i n i t i a l l y assume that the R-rules of UG hold absolutely, and as discussed above, the R-rules that we are assuming are the universal core rules, are the set of rules required for the specification of the vowels of Auca. There w i l l therefore be no positive evidence in the input t e l l i n g the child that new feature markings must be 112 s p e c i f i e d , and no parameters w i l l be reset. The c h i l d ' s i n i t i a l contrastive system w i l l be i d e n t i c a l to the adult system ( i . e . the system i n (2.47)). Because of the fact that both values of features are sometimes present underlyingly i n CU, i t i s c r u c i a l that t h i s theory allow for phonological rules that are feature-changing, and for phonological rules to be i n i t i a t e d by both values of a feature. Steriade's r u l e of Hungarian Back Harmony shown i n (2.27) exemplifies both of these c h a r a c t e r i s t i c s . The t r i g g e r of BH i s eithe r [+back] or [-back], and the spreading of one of these feature values i s not blocked by a segment s p e c i f i e d by the alternate feature value. In fact Steriade assumes that i t i s just those vowels that are unspecified for anyvalue of [back] which are the transparent or neutral vowels i n t h i s system 3 3. Thus the notion of a possible phonological r u l e i s very d i f f e r e n t i n CU than i n RU, where only a single value of any feature i s ever present within a given context. Since CU does not have a b u i l t - i n theory of rule types, I assume that i n CU phonological rules may spread, delink or i n s e r t as discussed i n 2.1.3. I also assume that phonological rules operate from L-->R, R—>L or b i d i r e c t i o n a l l y and that the unmarked d i r e c t i o n for a p a r t i c u l a r language i s i n the same d i r e c t i o n as i n i t i a l association, as i n (2.6). A parameterized model of CU ac q u i s i t i o n i s given i n (2.52) (I again include the OCP, the EC, the SCC, Structure Preservation and the Association Conventions). 113 (2.52) Parameterized CU Model of A c q u i s i t i o n UG Features Principles Principles of Autoseg. and Lex. Phonology Constraints on Rules Ordering Principle EC RRC LEARNING MECHANISM ALPHABET Systematization > Parameter Switching > Matrix Parameters Device R-rules Input UG i n (2.52) has fewer components than the the RU model i n (2.43), since i t does not allow for the l e x i c a l s p e c i f i c a t i o n of context-free language-particular feature values. The learning mechanisms for both models are e s s e n t i a l l y i d e n t i c a l : both are parameter switching devices. The Alphabet of CU has fewer components than the Alphabet of RU since again i t does not allow for the presence of language-particular context-free r u l e s . 114 2.5.2.2 Input i n CU In CU evidence from the PLD w i l l be necessary to t r i g g e r the r e s e t t i n g of R-rule parameters and for the a c q u i s i t i o n of phonological rules. Just as i n RU, alternations i n the input demonstrating that c e r t a i n feature redundancies do not operate i n the language w i l l be needed i f a parameter i s to be switched to the marked option. In Auca, for example, the ch i l d ' s i n i t i a l representation w i l l not d i s t i n g u i s h the two low vowels of the language. Evidence that /a/ and / a e / are phonologically d i s t i n c t w i l l be required to reset the R-rule parameter [+low] —> [+back], the subsequent markings of [-back] for / a e / , [+back] for /a/ and [-low] for /e/. This type of evidence w i l l come p r i n c i p a l l y from minimal pairs demonstrating that /a/ and / a e / are d i s t i n c t and also from alternations that are the r e s u l t of the ap p l i c a t i o n of phonological r u l e s . There w i l l be less i n t e r a c t i o n i n CU between the c h i l d ' s discovery of how a phonological r u l e operates and the re s e t t i n g of parameters than i n RU, since CU does not allow for language-specific values of underlying features. Languages w i l l vary only i n the classes of segments that are redundant for a s p e c i f i c feature, and thus phonological rules can only be affected to the extent that they manipulate d i f f e r e n t sets of possible segments. In CU i t w i l l never be the case that a rule of the adult phonology cannot be used at a l l by a very young c h i l d because a given feature value i s not yet present underlyingly. It may, however, be the case that a 115 rule used by a c h i l d may apply to an overgeneral or undergeneral set of segments. Because the theory of CU must assume that phonological rules may operate i n a feature-changing fashion and may be i n i t i a t e d by either value of a feature, the c h i l d w i l l have to pay p a r t i c u l a r attention to how a given rule operates i n the language. This theory then predicts that childr e n may have some d i f f i c u l t y i n determining the exact form of a phonological r u l e . This contrasts with the RU theory i n which the mechanisms of rule operation are l a r g e l y given by the theory i t s e l f , and consequently predicts that c h i l d r e n w i l l have l i t t l e d i f f i c u l t y i n determining the exact form of phonological r u l e s . 116 Notes to Chapter 2 1 Shortness or s i m p l i c i t y was assumed to be involved i n the metric, and theories such as the Derivational Theory of Complexity ( M i l l e r 1962) grew out of t h i s assumption. Research, however, demonstrated that "simpler" sentences, as defined by the theory of Transformational Grammar (Chomsky 1957) were not necessarily the ones f i r s t understood by childr e n (Fodor and Garrett 1967, Fodor Bever and Garrett 1974). 2 This was the o r i g i n a l view of parameters, as p r i n c i p l e s with chunks l e f t out. The parameters then f i l l e d i n the holes. The more recent view i s that the p r i n c i p l e i t s e l f i s f u l l y s p e c i f i e d innately and represents the unmarked parameter s e t t i n g . The marked parameter se t t i n g i s a d i f f e r e n t but c l o s e l y r e l a t e d version of the unmarked s e t t i n g . 3 Others believe that p a r t i c u l a r phonological rules are assigned to s p e c i f i c s t r a t a , and not to the phonology i n general (Kaisse and Shaw 1985). 4 The feature geometry i n (2.7) i s the model that w i l l be assumed i n t h i s t h e s i s , although I w i l l not be arguing for t h i s p a r t i c u l a r structure. 5 The recent work of E. Pulleyblank (1989) attempts to revise how v o c a l i c features are represented i n the geometry. Archangeli and Pulleyblank (1989) propose a geometry where the feature [ATR] i s dependent upon i t s own a r t i c u l a t o r node, the Tongue Root Node. s Ito and Mester (1986) state that the [f] i n futon i s underlyingly /h/. /h/ becomes [f] before [u] by a r u l e of L a b i a l i z a t i o n and becomes [c*] before [ i ] by P a l a t a l i z a t i o n . 7 Mester and Ito (1989) reanalyze t h i s as support for the fact that [voice] i s a p r i v a t i v e feature i n Japanese. e Redundancy rules of the type [+G] —> [-F] are not i n fact context-sensitive from a l o g i c a l or syntactic perspective (see Chomsky 1968). However, i n phonological research t h i s type of r u l e i s often referred to as a context-sensitive redundancy r u l e , and I w i l l maintain t h i s t r a d i t i o n . 9 Archangeli and Pulleyblank (1986: Appendix A: 352) use the maximal function s e t t i n g i n the statement of Epenthesis; however, since the maximal/minimal function i s supposed to r e l a t e to s p e c i f i e d targets only, I assume i t i s i r r e l e v a n t . 117 1 0 Archangeli and Pulleyblank (1986) do not discuss markedness of feature arguments. If Epenthesis i s the le a s t marked i n s e r t i o n process then a hierarchy of possible arguments i s a necessary addition to the rule typology. 1 1 This i s true as long as one of the redundancy rules i s not ordered p r i o r to Epenthesis by the RROC. 1 2 Rules (2.26-1 and -2) and (2.26-4 and -5) have the function given i n (2.25b), while rule (2.26-3) has function (2.25a). 1 3 This parameter was supposed to account for a c l u s t e r i n g of properties, such as n u l l subjects and a u x i l i a r y behaviour. 1 4 Braine (1963), for example, describes children's e a r l y word combinations as consisting of Pivots (frequently occurring words) and X-class (infrequently occurring) words. Bloom (1970) l i s t s one sentence type for Kathryn as S —> Pivot + N, where the sentence lacks a VP altogether. 1 5 See C a r r o l l and Roberge (1988) for some i n t e r e s t i n g suggestions on differences between LI and L2 language learners i n the u t i l i z a t i o n of negative evidence. 1 S There have been several recent attempts to use the tools of current phonological theory to describe a c q u i s i t i o n data (e.g. Spenser 1986, Sandanandan 1987, and Iverson and Wheeler 1988). 1 7 Later works i n the RU framework (e.g. Archangeli 1986, Pulleyblank 1988, Archangeli and Pulleyblank 1986) assume other p r i n c i p l e s of UG, and these are discussed i n 2.1. 1 8 In Archangeli (1984) a complement rule i s a redundancy r u l e i n s e r t i n g a predictable value of a feature created on the basis of language-specific evidence, and default rules are a l l other redundancy rules. Complement rules may therefore i n s e r t the same predictable value as a default r u l e . Archangeli (1984:65) states Alphabet Formation as: 1. Given an opposition [a F] — [ - a F] i n environment Q i n underlying representation, one value "b" i s selected as the matrix value for F i n Q and the other value i s sp e c i f i e d by an automatically formed complement value: [ ] -> [-b F] / Q 2. In the absence of language in t e r n a l motivation for sel e c t i n g a value as the matrix value for a feature F, the value "b" i s selected as the matrix value where: [ ] —> [-b F] / Q i s a member of the set of default rules. 118 The statement of Alphabet Formation I give i n (2.32) i s a r e v i s i o n of Alphabet Formation which corresponds to the Archangeli and Pulleyblank (1986) d e f i n i t i o n of complement and default r u l e s . In Archangeli and Pulleyblank (1986) and i n subsequent works complement rules are default rules that i n s e r t the opposite predictable value from the default r u l e , rather than rules that are learned from language-specific evidence ( i f the only property of a complement ru l e i s that i t be learned then i t i s possible that i t w i l l be i d e n t i c a l to a default r u l e ; while i n Archangeli and Pulleyblank (1986) a complement ru l e must redundantly i n s e r t the opposite value from a default r u l e ) . I prefer the l a t e r version of complement rules because i t allows us to d i s t i n g u i s h between u n i v e r s a l l y supplied and language-particular feature values. 1 9 I have taken t h i s name from Pulleyblank (1988). 2 0 I do not include a default rule for [ATR] since i t i s not required for the analyses i n Section 3. FCR 5 i s r e a l l y a d i f f e r e n t type of rule from FCRs 6 and 7 since i t represents an i n v i o l a b l e dependency, whereas 6 and 7 represent unmarked dependencies. As pointed out i n Spenser (1986) the r e s t r i c t i o n given by rule 5 might be better captured by a r e v i s i o n of feature geometry. In Chapter 6 I w i l l show that from an a c q u i s i t i o n perspective, i t appears that the context-free r u l e for [high] (rule 2 i n (2.36)) should be [ ] —> [-high]. 2 1 Actual statements of default rules are rare i n the l i t e r a t u r e , probably because no one wishes to commit themselves to a set of rules that may prove to be wrong. Archangeli and Pulleyblank (1986) give a set of context-free rules i n Appendix A, but the context-sensitive rules are conspicuously absent. I have had to put together a set of rules i n order to perform complete analyses of the v o c a l i c systems of Hungarian and Spanish and to develop predictions that w i l l follow from these analyses. If one or some of these rules prove to be incorrect (as I argue may be the case for the context-free rule for [high] i n Chapter 5) t h i s does not argue against the theory i t s e l f . 2 2 This does not by i t s e l f imply that / i / w i l l be the f i r s t vowel acquired because the parametric theory of RU predicts that a l l 5 vowels should be present given the default rules of UG. However, I w i l l show i n Chapter 5 that t h i s i s not i n fact what happens i n a c q u i s i t i o n . 2 3 Unconstrained restructuring means that a c q u i s i t i o n data w i l l have l i t t l e import on issues r e l a t i n g to l i n g u i s t i c theory. The parametric theories of RU and CU discussed i n Chapter 2 predict that only c e r t a i n r e s t r i c t e d types of res t r u c t u r i n g based on parameter switching w i l l occur (see White 1981 and Ingram 1989 for views on the r e s t r u c t u r i n g debate i n a c q u i s i t i o n ) . 119 2 4 This i s a s i m i l a r to the representation of Auca given i n Archangeli (1988), except that I have not marked [-high] on the [+low] vowels. Archangeli also shows that a language-s p e c i f i c redundancy rule i s needed i n Auca to eliminate /u/ from the system. I w i l l not be discuss rules such as t h i s that eliminate segments from inventories, but w i l l focus on redundancy rules that achieve f u l l s p e c i f i c a t i o n of a given set of segments. 2 5 The d e f i n i t i o n of "distinctness" needed for underspecification theory i s d i f f e r e n t from the d e f i n i t i o n given i n Chomsky and Halle (1968), where two segments are d i s t i n c t only i f they have opposite values for a given feature. If one adopts RU then the lack of a s p e c i f i c a t i o n for [F] must be enough to make that segment underlyingly d i s t i n c t from a segment s p e c i f i e d as [+F] or [-F]. 2 6 This alternate hypothesis was suggested to me by P.A. Shaw and i s perhaps more feasi b l e from a markedness point of view, since / a e / w i l l have a greater number of feature markings that /a/. However, markedness and feature markings are not unequivocally related i n the theory of RU, as w i l l be shown i n Chapter 3, where I argue that /B/ i s the t o t a l l y unspecified vowel i n the Hungarian v o c a l i c system i n the RU framework. 2 7 I include only v o c a l i c featural parameters i n (2.42) and l a t e r i n (2.50) since these are the parameters required for the analyses of the v o c a l i c systems i n Chapter 3. 2 8 Several of the p r i n c i p l e s of Autosegmental and L e x i c a l Phonology ( e s p e c i a l l y the OCP and the Association Conventions) may be parameterizable, but I ignore t h i s hypothesis here and l i s t only featural parameters under the category "parameters". 2 9 When combined with the rule functions these w i l l produce redundancy rules such as [ ] —> [+high], [+low] —> [+back], etc. 3 0 Archangeli and Pulleyblank (1989) argue that [-ATR] i s inserted on /a/ by a context-sensitive redundancy ru l e [+low] —> [-ATR] before the rule of ATR Spread. 3 1 Calabrese (1988) discusses t h i s i n t e r a c t i o n i n some d e t a i l . 3 2 I again assume these rules are notationally equivalent to [ ] —> [+F]/[+G]. 3 3 For these reasons CU can never contain a p r i n c i p l e such as the RROC i n (2.19b) because i t would completely negate these transparency f a c t s . 120 CHAPTER 3 The Vocalic Systems of Hungarian and Spanish In t h i s chapter I w i l l discuss a number of phonological processes which operate i n the v o c a l i c systems of Hungarian, a Finno-Ugric language and Spanish, a Romance language. These facts w i l l be analyzed according to the parametric theories of Radical Underspecification (RU) and Contrastive Underspecification (CU) described i n 2.5.1 and 2.5.2, i n most cases drawing from previous analyses of these f a c t s . In Chapter 4 I outline the predictions that these analyses make for the a c q u i s i t i o n of Hungarian and Spanish, and l a t e r i n Chapter 6 these predictions w i l l be examined i n l i g h t of a c q u i s i t i o n data from both Hungarian and Spanish. The analyses of the v o c a l i c systems of Hungarian and Spanish presented here are useful for two reasons. F i r s t , they help to demonstrate how the theories of RU and CU d i f f e r . Hungarian has a large vowel system, with front rounded and unrounded vowels, with both a long and short vowel s e r i e s . The analyses i n t h i s chapter focus on harmony processes which spread values of [back] and [round] throughout a word. Both the RU and CU analyses of Hungarian use the p r i n c i p l e of Structure Preservation (Kiparsky 1982, 1985) to explain why cert a i n segments are transparent to Harmony, although the generality and s i g n i f i c a n c e of t h i s constraint d i f f e r s considerably i n the two analyses. 121 The Spanish vowel system i s a symmetrical 5 vowel system, which also permits c e r t a i n types of diphthongs. The Spanish data presented here focus mainly on alternations that take place within verbs. One set of alternations involves a vowel which surfaces as high i n some contexts and mid i n others, and the second involves a vowel which surfaces i n some contexts as a monophthong and i n other contexts as a diphthong. The RU and CU analyses of Spanish do not d i f f e r as much as the Hungarian analyses, since the RU account of Spanish does not force the early application of redundancy rules within the l e x i c a l component. [high] proves to be a p a r t i c u l a r l y important feature i n Spanish i n both analyses, and i t s marked values are manipulated by several phonological r u l e s . The majority of rules posited for Spanish by both the RU and CU analyses are sensi t i v e to stress. The second reason for the existence of the analyses i n t h i s chapter i s to argue for s p e c i f i c underspecified vowel inventories. This i s p a r t i c u l a r l y important i n the theory of Radical Underspecification, where either value of a feature may be present i n underlying representations. On the basis of the analyses of Hungarian and Spanish I argue that c e r t a i n f e a t u r a l and rule parameters must be reset by a c h i l d acquiring these languages. When a featural parameter i s reset from the unmarked option provided by Universal Grammar, the underlyingly marked values i n that language w i l l also d i f f e r from those provided by UG. The unmarked option of phonological rules i s generally assumed to be OFF, although a 122 language may choose to make use of a p a r t i c u l a r rule and reset i t to ON. In the analyses which follow, I argue that c e r t a i n language-particular feature markings, and c e r t a i n phonological rules are required to explain c e r t a i n types of data i n Hungarian and Spanish. One of the most important difference between the RU and CU treatments of Hungarian harmony i s i n the ordering of redundancy rules and other phonological processes. The theory of RU assumes the Redundancy Rule Ordering Constraint (RROC), which orders a redundancy rule before a phonological ru l e which makes reference to that redundant value. The e f f e c t of t h i s constraint i n Hungarian i s to interweave redundancy rules and harmony rules. The theory of CU that i s being developed i n t h i s thesis does not adopt an ordering p r i n c i p l e such as the RROC (although see the discussion of t h i s question i n 2.5.2), and a l l redundancy rules apply l a t e i n the phonology, a f t e r harmony rules. This chapter w i l l be organized as follows. In 3.1 I present and analyze the Hungarian data within the parametric theories of RU and CU. In 3.2 the same w i l l be done for the Spanish data. Each section concludes with a summary of the language-specific aspects of the phonology ( i . e . those aspects of the phonology that must be learned) given each underspecification account, a schema of how the l e x i c a l and p o s t l e x i c a l components must be organized according to these analyses, and a comparison of the two analyses. In each case my goal i s to determine which analysis presents a more 123 coherent account of the data, assuming that the l e a r n a b i l i t y of a p a r t i c u l a r theory should i n part be re l a t e d to i t s a b i l i t y to account for a set of data i n an elegant and non-s t i p u l a t i v e fashion. 3.1 Harmony Systems i n Hungarian In t h i s section I examine how Back and Round Harmony operate i n Hungarian, and develop analyses of these facts based on the parametric theories of RU and CU. Although both types of Harmony do operate within loanword vocabulary, there i s l i t t l e agreement about the facts ( p a r t i c u l a r l y with regard to Back Harmony), and for t h i s reason I r e s t r i c t myself to native words. The expectation i s that the loanword facts may be analyzeable using the same mechanisms once the facts are better known. In 3.1.1 the v o c a l i c system of Hungarian i s discussed. I pay p a r t i c u l a r attention to the asymmetries that e x i s t between the short and long vowel system, and show how configuration constraints can be used to describe these asymmetries. In 3.1.2 the RU analysis of the Hungarian data are presented, and i n 3.1.3 the CU analysis of these facts i s given. In 3.1.4 the analyses are summarized, and a b r i e f comparison of the RU and CU analyses i s made. 3.1.1 Hungarian Vowels The Hungarian vowel system (Standard Budapest d i a l e c t ) as described i n Ringen (1988) i s given i n (3.1). The vowels are 124 given both i n t h e i r orthographic and phonetic forms. (3.1) Short Long i [ i ] tl[y] u[u] i [ i : ] <i[y:} u[u:] o[<*>] o[o] e[e:] #[<£:] 6[o: ] e[c] a[D] a[a:] Phonetically, the short low back vowel i s [o] and the short low front vowel [e]. The rounding of [a] i s assumed i n Vago (1980), Kornai (1987) and Ringen (1988) to be a purely phonetic r u l e , and the l i n k i n g vowel ef f e c t s discussed i n 3.1.2.2 provide evidence that t h i s i s true. H i s t o r i c a l l y the language contained both mid and low front vowels, but these have been merged i n most d i a l e c t s , including the Standard. The facts of Back Harmony and Round Harmony to be presented i n (3.9) and (3.13) demonstrate that the short low front vowel and the long mid front vowel have d i s t i n c t vowel heights. The low vowel [e] alternates with [a] i n Back Harmony, while the mid vowel [e] i s transparent to t h i s process. The long mid vowel alternates with the mid back and mid front rounded vowels i n Round Harmony, while the low vowel does not p a r t i c i p a t e i n t h i s process. Stress i n Hungarian i s rule-governed and always f a l l s on the i n i t i a l s y l l a b l e i n a word (Kontra and Ringen 1986). In standard autosegmental theory long vowels d i f f e r from short vowels i n that long vowels are associated to two s k e l e t a l s l o t s and short vowels to only a single s l o t 1 . In the analyses which are to follow I assume there i s a s i n g l e 125 set of feature s p e c i f i c a t i o n s for the 8 vowels / i / , l u l , l e i , l o l , l e i , /a/, /u7 and IQI, and that Hungarian has chosen a marked s y l l a b i f i c a t i o n option which allows for complex or branching nuclei (see (2.10Aib)). Following Ringen (1988) and Jensen and Stong-Jensen (1989) I assume there are two d i s t i n c t non-high unrounded front vowels i n Hungarian: /e/ and l e i . lei alternates with /a/ i n the Back Harmony system, while /e/ alternates with /o/ and 161 i n the Round Harmony system, /e/ surfaces as [e] only when i t i s associated with two s k e l e t a l s l o t s dominated by a branching nucleus, which I assume, following Levin (1985), i s the representation of a long vowel (see discussion i n 2.1.2.4). l e i , on the other hand, can surface only when associated to a single s k e l e t a l s l o t dominated by a nucleus, i . e . when i t i s a short vowel. These p a r t i c u l a r constraints are not a function of Structure Preservation, since s y l l a b i c structure i n Hungarian w i l l be derived during the course of the l e x i c a l phonology (see 2.1.2.4) and since there are underlying feature sets that correspond to these vowels. I therefore assume that the constraints on /e/ and lei hold at surface structure a f t e r redundancy rules have applied, and that they can be stated as i n (3.2) . 126 (3.2) * N * N / \ I X X X \ / I [-high] [-high] [+low] [-low] [-back] [-back] [-round] [-round] *[£:] *[e] Domain: p o s t - l e x i c a l In t h i s and following chapters, Hungarian data w i l l be given i n orthographic form, with the phonetic form of a vowel also provided i f i t i s long or [£]. 3.1.2 Parametric RU Analysis of Hungarian The components of UG that I assume must be part of a parametric RU theory of a c q u i s i t i o n are discussed i n 2.5.1, and a schematized model i s given i n (2.43). The most basic p r i n c i p l e of UG i s the Minimal Redundancy Condition, discussed i n 2.5.1.1. This condition, repeated here as (3.3), t e l l s the c h i l d that only non-redundant information i s present i n underlying representations. 127 (3.3) The Minimal Redundancy Condition (MRC) a. Underlying representations do not contain redundant information. b. The most highly valued system contains the minimal number of features and feature values needed to d i s t i n g u i s h the inventory of a language. Redundant information i s provided by a set of universal redundancy ru l e s , or default rules, which may be e i t h e r context-free or context-sensitive. In 2.5.1.1 i t i s argued that these rules function as featural parameters. UG supplies the unmarked form of the r u l e , but a p a r t i c u l a r language may choose to reset a featural parameter to the marked option. If a context-free parameter i s reset at the marked option, a complement rule w i l l be created to i n s e r t the opposite feature value supplied by UG. The default r u l e w i l l be l o s t i n the language-particular grammar, and the complement ru l e w i l l take i t s place. If a context-sensitive parameter (FCR) i s reset at the marked option, the r u l e w i l l again be eliminated from the language-particular system, although i n t h i s case i t w i l l not be replaced. This means that the feature value supplied by that r u l e w i l l no longer be redundant and by the MRC i t must be added as a marked feature value. In RU i t i s assumed that because redundancy rules are one type of phonological rule, they are subject to the Elsewhere Condition (EC, see (2.3)), which orders s p e c i f i c rules before 128 more general ones. RU also assumes that redundancy rules are subject to the ordering constraints i n (2.19), repeated here as (3.4). (3.4) Ordering Constraints (Archangeli and Pulleyblank 1986) a. Redundancy rules apply as la t e as possible i n the grammar. b. A redundancy rule must apply at the l e v e l at which reference i s made to the feature value being inserted (the Redundancy Rule Ordering Constraint or RROC) c. A redundancy rule applies as early as possible at the l e v e l dictated by the RROC. By these constraints redundancy rules w i l l apply as l a t e i n the grammar as possible ( i . e . p o s t - l e x i c a l l y ) unless a phonological rule makes reference to a p a r t i c u l a r redundant feature value, i n which case the redundancy rule w i l l be ordered before the phonological ru l e . These ordering constraints are c r u c i a l to the analysis of the Hungarian v o c a l i c system. The set of universal default rules that I assume are given i n (3.5) (repeated from (2.36). 129 (3.5) Universal Default Rules Context-free rules FCRs 1. [ ] " --> [-low] 5. [+low] - - > [-high] 2. [ ] ---> [+high] 6. [+low] — > [+back] 3. [ ] ---> [-back] 7. [+low] - - > [-round] 4. [ ] ---> [-round] 8. [+back] --> [+round] [-low] 9. [-back] --> [-round] [-low] In the following sections I argue, on the basis of facts of Hungarian Back and Round harmony, that the following redundancy rules from (3.5) have been reset to the marked option i n Hungarian: (3.6) Marked Hungarian Featural Parameters Context-free rules FCRs 2. [ ] --> [+high] 6. [+low] --> [+back] 4« [ ] — > [-round] 9. [-back] —> [-round] [-low] The fact that 4 default rules of UG must be reset i n Hungarian i s not p a r t i c u l a r l y s u r p r i s i n g , since the system i s marked, containing front rounded vowels and a front low vowel. According to the parametric view of redundancy rules developed i n 2.5.1, the rules i n (3.6) w i l l not be a part of the language-particular grammar of Hungarian. The context-s e n s i t i v e r u l e parameters w i l l not be replaced, but the 130 context-free parameters w i l l be replaced by complement rules i n s e r t i n g [-high] and [+round]. This w i l l produce the system of redundancy rules for Hungarian shown i n (3.7). (3.7) Hungarian Redundancy Rules Default Rules 1. [ ] --> [-low] 5. [+low] —> [-high] 3. [ ] --> [-back] 7. [+low] --> [-round] 8. [+back] —> [+round] [-low] Complement Rules 2a. [ ] —> [-high] 4a. [ ] —> [+round] Applied to the set of rules given i n (3.7) the EC w i l l order rul e 8 before 4a and rule 5 before 2a. If the rules i n (3.7) are used to i n s e r t the redundant values of Hungarian vowels, the marked ( i . e . non-redundant) feature values w i l l be those i n (3.8). (3.8) Parametric RU Vocalic System of Hungarian 2 i e e u o a t l O high + + + low + + back + + + round This system i s i d e n t i c a l to the r a d i c a l l y underspecified system argued for by Jensen and Stong-Jensen (1989). 131 The underspecified vowel array i n (3.8) i s not the system of s p e c i f i c a t i o n s that w i l l be provided for the c h i l d by Universal Grammar, and a discussion of how these two systems d i f f e r i s provided i n Chapter 4. In the following sections I use arguments from several aspects of the vo c a l i c system of Hungarian to show that the system i n (3.8) i s the r a d i c a l l y underspecified system that i s required for t h i s language. These arguments focus on demonstrating that the feature values [+back] and [-round] are phonologically active feature values i n Hungarian, and more s p e c i f i c a l l y , that these feature values must be l e x i c a l l y s p e c i f i e d . In addition, I argue that /oY, represented as the t o t a l l y unspecified vowel i n (3.8), i s the epenthetic vowel i n Hungarian, and the vowel that w i l l surface i f neither Back nor Round Harmony applies i n that context. 3.1.2.1 Back Harmony (BH) Hungarian BH operates both within roots and across c e r t a i n s u f f i x e s . Native Hungarian roots may contain a l l back vowels, a l l front vowels, a mixed set of back vowels co-occurring with the neutral vowels / i / or /e/, or neutral vowels only. The majority of neutral-vowel-only roots are followed by suffixes containing front vowels, but c e r t a i n roots l i k e hid and eel take back vowel s u f f i x e s 3 . 132 (3.9) Back Harmony4 a. Back vowels u [ u: ] t 'road• ha[a:]z 'house' va[a:]ros ' c i t y ' b. Non-neutral front vowels Ot ' f i v e ' be[e]tU ' l e t t e r ' Orom 'joy' c. Back or front non-neutral r a d l [ l : ] r 'eraser' kavics 'pebble', ta[a;]nye[e:]r 'plate' id6 /[o: ] 'time' d. Neutral vowels only i . f i l l e [ e : ] r 'penny' s z l [ i : ] n 'colour' i i . hi.[i:]d 'bridge' c e [ e : ] l 'goal' Dative Ablative utna[o]k u t t o [ o : ] l haznak hazt61 varosnak varostol Otne[e]k ott«[o:]l bettlnek betut01 OrOmnek orflmto'l vowel and neutral vowel radirnak r a d f r t d l kavicsnak k a v i c s t d l tanyernak tanyert61 ido'nek iddto^l f i l l e r n e k f i l l e r t 6 l szlnnek szinto 7! hidnak h i d t d l celnak c e l t o l BH i s also found i n loanword phonology, although I w i l l concentrate excl u s i v e l y on the facts as they pertain to native words 5. There i s a lack of agreement i n the l i t e r a t u r e over the status of [e] i n the BH system. [e] i s sometimes found i n roots with back vowels (cefeIruza 'pencil' and krape\e1k 'chap') and therefore appears to be a neutral vowel, yet i t 133 alternates with [a] i n the dative s u f f i x nak/nek. Vago (1980) assumes [e] i s a neutral vowel, receiving i d e n t i c a l treatment to the vowels [ i ] , [1] and [e]. Ringen (1988) and Kontra and Ringen (1986) treat [e] as a harmonic vowel, belonging to the same class as the back vowels and the front rounded vowels. I w i l l show that the chameleon-like behaviour of [e] can be explained by assuming that [e] has two d i s t i n c t sources — /e/ and l e i . The facts i n (3.9) show cle a r evidence that there are two phonologically d i s t i n c t low vowels i n Hungarian: /a/ and l e i . The declensions of 'house' i n (3.9a) demonstrate that r o o t - i n i t i a l /a/ i s followed by a back vowel, while those of 'pebble' show that r o o t - i n i t i a l lal may also be followed by a neutral vowel ( [ i ] , [1] or [6]). R o o t - i n i t i a l l e i , as shown by the declensions of ' l e t t e r ' i n (3.9b) may be followed by a front non-neutral vowel, and l i k e a l l vowels may also be followed by a neutral vowel. The fact that the nak variant of the Dative s u f f i x always follows back non-neutral vowels while the nek variant always follows front non-neutral vowels also demonstrates that phonologically lal and lei are d i s t i n c t segments. If lal and lei are to have d i s t i n c t underlying representations then the universal context-sensitive rule [+low] —> [+back] (rule 6 i n (3.5)) cannot hold i n Hungarian. Only when t h i s rule i s suppressed from the language-particular grammar can the two vowels lal and lei have d i s t i n c t s p e c i f i c a t i o n s for [back]. This conceptualization of how context-sensitive redundancy 134 rules (FCRs) are parameterized i n the theory of RU i s discussed i n d e t a i l i n 2.5.1.1s. The BH data i n (3.9) also demonstrate that the front unrounded vowels / i / and /e/ must be phonologically d i s t i n c t from the rounded vowels /u7 and 16/. / i / and /e/ can co-occur with eit h e r front or back vowels, while /u/ and /O/ co-occur only with other front rounded vowels or with neutral vowels. This demonstrates that these two pairs of vowels must be distinguished using the feature [round], and therefore that context-sensitive rule 9 i n (3.5) must also be eliminated from the grammar of Hungarian (see 2.5.1.1). Once t h i s r u l e i s eliminated from the grammar, one member of each p a i r w i l l be s p e c i f i e d for a value of [round], while the other member w i l l be unspecified. The marked value of [round] w i l l depend upon the context-sensitive r u l e for t h i s feature that the language employs. I w i l l return to the marked s p e c i f i c a t i o n for [round] i n 3.1.2.2. The basic RU analysis of BH proposed here i s taken from Jensen and Stong-Jensen (1989). The default rules of UG, given i n (3.5), suggest that i n the unmarked case [-back] w i l l be a redundant feature feature value i n Hungarian and [+back] the l e x i c a l l y marked, phonologically active value. There i s no evidence i n the harmony systems of t h i s language to suggest that t h i s UG s p e c i f i c a t i o n should be overturned, and i n fa c t , there i s evidence from the BH system to support the assumption that [+back] must be the l e x i c a l l y s p e c i f i e d value. This evidence w i l l be discussed i n conjunction with 135 the marking condition given i n (3.11). Following Jensen and Stong-Jensen I assume that [+back] i s a property of the root, rather than of i n d i v i d u a l segments, and exists i n underlying representations as a f l o a t i n g autosegment. Root and s u f f i x a l vowels (at l e a s t those s u f f i x e s which take part i n BH) w i l l be underlyingly unspecified for backness. Association of f l o a t i n g [+back] i accomplished by the Association Conventions of UG (see (2.6)), which map t h i s f l o a t i n g feature onto the leftmost e l i g i b l e target. The Association Conventions apply c y c l i c a l l y , and therefore can reapply once morphological material i s added to the stem. BH i s a rule which spreads [+back] from l e f t to r i g h t , the same d i r e c t i o n as i n i t i a l association" 7. (3.10) Back Harmony Spread [+back] L —> R Domain: l e x i c a l ( c y c l i c ) The tr i g g e r s of BH w i l l be /u/, /of and /a/ — those vowels that are l e x i c a l l y marked as [+back]. How then do we block the application of BH to the neutral vowels [ i ] , [ i : ] and [e:]? Again following Jensen and Stong-Jensen I assume that the vowels given as / i / and /e/ i n (3.8) are prohibited from undergoing BH by Structure Preservation. The underlying system of Hungarian does not contain non-low back unrounded vowels, so by Structure Preservation these vowels cannot be derived. As discussed 136 i n 2.1.1.4, and following Kiparsky (1985), the formal work of Structure Preservation i s performed by marking conditions which hold i n the lexicon. The formal p r o h i b i t i o n against back non-low unrounded vowels i n Hungarian can be stated using the language-specific marking condition i n (3.11). (3.11) *[+back] [-low] [-round] (3.11) does not require any added learning for the c h i l d (once these p a r t i c u l a r feature s p e c i f i c a t i o n s are acquired), since i t i s simply a formalization of Structure Preservation, a p r i n c i p l e of UG. (3.11) i s a checking mechanism which constrains the type of feature combinations that are permitted i n underlying representations and throughout the l e x i c a l phonology, and i t w i l l block association or spread of [+back] to vowels which are s p e c i f i e d at that point i n the d e r i v a t i o n as [-low] and [-round]. We would not expect the neutral vowels to be underlyingly s p e c i f i e d as [-low] i f , as shown i n (3.8), [+low] i s the l e x i c a l l y marked feature value. Following Jensen and Stong-Jensen, however, I assume that the marking condition i n (3.11) i s an active part of the phonology, and therefore w i l l be subject to the RROC (see (2.19). Although (3.11) holds of URs, the MRC, given i n (3.3), prohibits redundant s p e c i f i c a t i o n s at t h i s l e v e l , and therefore (3.11) w i l l have no active role at t h i s point. During the l e x i c a l 137 phonology, however, (3.11) w i l l come into play to check the outputs of rules such as BH or RH. The e f f e c t of t h i s condition w i l l be to force the redundant values mentioned i n t h i s condition to be inserted p r i o r to the operation of the r u l e . The condition i n (3.11), the RROC, and the phonological rules of Hungarian which manipulate the feature [+back] are re l a t e d i n the following fashion. (3.11) acts as a condition on the ouput of BH, given i n (3.10), and the Association Conventions, which provide the i n i t i a l association of f l o a t i n g [+back] features. (3.11) w i l l t r i g g e r the e a r l y a p p l i c a t i o n of the context-free rule i n s e r t i n g [-low] (rule 1 i n (3.5)) and the context-sensitive rule p r e d i c t i n g [-round] for [+low] vowels (rule 7 i n (3.5)), before BH and before Association of [+back], since these values are s p e c i f i c a l l y mentioned i n the statement of the condition. Once these redundancy rules have applied, the vowels / i / and /e/ w i l l be s p e c i f i e d as [-low] and [-round] and by (3.11) [+back] w i l l not be permitted to spread or l i n k to them. Floa t i n g [+back] w i l l , however, s t i l l be available to associate or spread to another vowel i n the stem or s u f f i x . Vowels that are not s p e c i f i e d as [+back] either through the i n i t i a l association of [+back] or through BH w i l l receive t h e i r s p e c i f i c a t i o n s for [-back] by rule 3 i n (3.5). The fact that (3.11) contains the s p e c i f i c a t i o n [+back] supports our assumption that the l e x i c a l l y s p e c i f i e d value of [back] i n Hungarian i s that provided by UG. (3.11) i s a 138 condition which blocks the derivation of back non-low unrounded vowels, and consequently requires the use of the feature [+back]. If we were to suppose that [-back] was the phonologically active value of [back], [+back] would s t i l l have to be present i n the statement of (3.11), suggesting that the redundancy rules i n s e r t i n g [+back] are ordered p r i o r to (3.11) and the rule of BH. These [+back] s p e c i f i c a t i o n s would then block a l l applications of BH, and there would be no a l t e r n a t i n g vowels, such as those shown i n (3.9). I therefore take t h i s as confirming evidence that [+back] i s the l e x i c a l l y s p e c i f i e d value i n t h i s language. BH, as shown i n (3.9) i s a rule which applies both within roots and a f t e r morpheme concatenation. The r u l e i s obviously l e x i c a l , since i t has exceptions and applies only within words. Jensen and Stong-Jensen argue that BH i s a p o s t c y c l i c r u l e because i t must apply a f t e r Epenthesis (which they also argue i s postcyclic) and because i t applies within roots. They argue that there i s no s p e c i f i c evidence that BH must apply both c y c l i c a l l y and p o s t c y c l i c a l l y , so by the P r i n c i p l e s of Domain Assignment (Halle and Mohanan 1985) i t w i l l be p o s t c y c l i c . I depart from Jensen and Stong-Jensen's analysis on the issue of domain assignment. Following Levergood (1984), Pulleyblank (1986) and Archangeli and Pulleyblank (1989) I assume that the applicatio n of the Association Conventions to l i n k [+back] to a root vowel w i l l create a derived environment r o o t - i n t e r n a l l y so that the SCC w i l l not block 139 the a p p l i c a t i o n of BH. Thus BH can apply c y c l i c a l l y , and w i l l follow the Association of [+back] and the redundancy ru l e s , which are a l l structure-building r u l e s . In t h i s way the SCC i s maintained and BH can be l e x i c a l and s t i l l apply within roots. Neutral-vowel-only roots which surface with back s u f f i x a l vowels ( i . e . the forms i n (3.9dii)) can be analyzed as having a f l o a t i n g [+back] autosegment which i s prohibited from attaching to the root vowel by (3.11). This autosegment w i l l remain f l o a t i n g u n t i l a s u f f i x i s added, and [+back] can then be associated to the s u f f i x a l vowel. The derivation of the native words bettinek, radirt61 and celnak w i l l take place as shown i n (3.12). For the purposes of these derivations I assume the underlying feature s p e c i f i c a t i o n s given i n (3.8) although I have not yet argued for the l e x i c a l l y marked values of [round] and [high]. I ignore a l l Round Harmony ef f e c t s u n t i l 3.1.1.2. 140 (3.12) BH i n native words' Underlying [ ] - > [ - l o ] [ + l o ] — >[-rd] b c t tl X X X X [+lo] [+hi] b £ t tl X X X X / 1 [+lo] [-lo] [+hi] [-rd] [-rd] r £ d i r X X X X X X [+bk] [+lo] [-lo] [+hi] I [-rd] [-rd] e e l X X X X [+bk] [-rd] e e l X X X X [+bk] [-lo] [-rd] Association of [+back] BH n/a n/a r a d i r X X X X X X [+bk] [+lo] [-lo] [ j h i ] [-rd] [-rd] Prohibited by (3.11) prohibited by (3.11) n/a 141 Morphology, betil + nek XXXX XXX [+lo][-lo][+lo] I [+hi] [-rd] radi r + t6 1 XXXXXX XXXX [+bi] I [+lo][-lo] I [+hi] I [-rd][-rd] ce 1 + nek XXXX XXX [+bk; [-lo] [+lo] [-rd] [ ] — > [ - l o ] [+lo]—>[-rd] betu + nek XXXX XXX [+lo][-lo][+lo] [+hi] [-rd] [-rd] radi r + t& 1 XXXXXX XXXX / [+bk] [+lo][-lo] [-lo] [+hi] [-rd][-rd] ce 1 + nek XXXX XXX [+bJ [-lo] [+lo] [-rd] [-rd] Association of [+back] n/a n/a BH n/a ce 1 + nak XXXX XXX [+bk^" [-lo] [+lo] [-rd] [-rd] n/a 142 None of the vowels i n these forms have underlying linked values for the feature [back], although r a d l r and eel have a f l o a t i n g [+back] feature. In r a d l r the f l o a t i n g feature w i l l associate to the i n i t i a l vowel by the Association Conventions, but i n eel association w i l l be blocked by the marking condition i n (3.11). (3.11) w i l l t r i g g e r e a r l y a p p l i c a t i o n of the redundancy rules [ ] --> [-low] and [+low] —> [-round], shown as rules 1 and 7 i n (3.5). Once [+back] i s associated BH can operate, although i t i s inapplicable i n betti and eel and i t i s blocked i n r a d l r by (3.11). When the suffixes are added Association of [+back] can reapply, and w i l l act to associate the f l o a t i n g feature to the s u f f i x a l vowel i n the form celnak, with redundancy rules 1 and 7 i n (3.5) applying f i r s t . BH can then reapply, and i t w i l l spread the [+back] feature linked to the i n i t i a l root vowel i n r a d l r to the s u f f i x a l vowel. Although i t could be argued that BH should not apply from the i n i t i a l root vowel i n r a d l r to the s u f f i x a l vowel because t h i s operation does not respect the adjacency of target and t r i g g e r , Ringen (1988) argues that adjacency (or the L o c a l i t y Condition as i t i s c a l l e d i n Archangeli and Pulleyblank 1987) i s not relevant i n t h i s case. By the marking condition i n (3.11) the second root vowel i n r a d l r i s not a possible target of BH, and we can then assume that the i n i t i a l root vowel and the s u f f i x a l vowel are adjacent for the purposes of t h i s r u l e . BH i s inapplicable i n betllnek because there i s no [+back] 143 feature to spread, and also i n celnak, since the [+back] feature l i n k s to the s u f f i x . 3.1.2.2 Round Harmony (RH) Suffixes such as the Dative and Ablative have one alternant containing a back vowel, and one containing a front vowel, but those with short mid vowels generally have three a l t e r n a n t s 9 . In these ternary suffixes [o] follows a back vowel, [0] follows a front rounded vowel, and [e] follows a front unrounded vowel. The forms i n (3.13) show the agreement i n rounding between root and s u f f i x a l vowels i n the 2nd person p l u r a l , a l l a t i v e and p l u r a l forms. (3.13) Round Harmony a. hoz ftf[o:]z ne[e:]z f e [ e ] j s z f [ i : ] n b. f o l d f e [ £ ] j c. gerezd Orom 'bring' 'cook' ' see * 'head' •colour' 'earth' 'head' ' s l i c e ' 'joy' 2PL hoztok fcYztok nezte[e]k fejtek szintek A l l a t i v e ('toward') foldhoz fejhez P l u r a l gerezdek oromok The forms i n (3.14) demonstrate that low root vowels do 144 not t r i g g e r RH. (3.14) va[a:]ros ' c i t y ' la[a:]nchoz 'chain' haz 'house' A l l a t i v e varoshoz lanchoz hazhoz If [+low] vowels were triggers of RH we would expect the root and s u f f i x a l vowels following the i n i t i a l root vowels i n the forms i n (3.14) to surface as [-round], but t h i s i s not the case. Neither short [a], nor long [a:] triggers RH. RH also appears to be more lim i t e d than BH i n that i t a f f e c t s only short vowels. As noted by Kornai (1987) and Jensen and Stong-Jensen (1989) suffixes which demonstrate the ternary a l t e r n a t i o n between o/o/e always contain short vowels. Suffixes such as the Ablative, which have long mid (3.15) have only the alternants [o] and [6]. Ablative a. e[e]mbe[e]r ' man' embert6[o:]1 s z i [ i : ] n ' colour' s z i n t o l f i l l e [ e : ] r 'penny' f i l l e r t t f l b. bab 'bean' babtd[o:]1 hu[u: ] r 'chord * hurtol ce[e: ]1 •goal' celtc-l If RH does not operate i n these forms, and we wish to maintain that the value of roundness of s u f f i x a l vowels i s supplied either by the root vowels or by redundancy r u l e s , 145 t h i s suggests that Hungarian must have a redundancy ru l e supplying [+round] for the Ablative s u f f i x a l vowels. The rounding of the s u f f i x a l vowels i n the forms for 'bean* and 'chord' i s possibly supplied by a redundancy ru l e which rounds back vowels (these forms do p a r t i c i p a t e i n BH); however, t h i s r u l e cannot supply the roundness values for the other forms. It i s then possible that the rounding of these vowels i s supplied by a general r u l e [ ] —> [+round]. If such a ru l e i s to e x i s t i n the grammar of Hungarian, then the universal default rule 4 i n (3.5) must have been suppressed, and [-round] w i l l be the l e x i c a l l y marked value. If we assume that [-round] i s the l e x i c a l l y s p e c i f i e d value i n Hungarian, a form such as szl n t f f l can be represented as i n (3.16). (3.16) N N / \ / \ x x x x + x x x x \ / [+hi] [-rd] sz i n t V 1 Both the vowel of the root and the s u f f i x are long, and I assume that the s u f f i x a l vowel i s underlyingly unspecified for the features [round] and [back]. The root vowel i s s p e c i f i e d as [-round] underlyingly, and i s therefore a possible t r i g g e r of RH. RH does not appply i n t h i s form, 146 however, because the s u f f i x a l vowel i s l o n g 1 0 . Jensen and Stong-Jensen (1988) account for the fact that long vowels do not p a r t i c i p a t e i n RH by constraining t h i s r u l e to apply to adjacent moras. They argue that s p e c i f y i n g adjacency at the l e v e l of the nucleus does not d i s t i n g u i s h long from short vowels. Given that I am adopting the s y l l a b i c framework of Levin (1985) which assumes that nuclei (and not moras) are primitives, and given that metrical theory recognizes the notion of branching (see 2.1.2.4), I assume that the constraint on the operation of RH can be stated as holding of branching n u c l e i . Returning to the question of the l e x i c a l s p e c i f i c a t i o n of [round], a second argument that [-round] may be the s p e c i f i e d value i n Hungarian comes from the marking condition i n (3.11). This condition i s used to block the d e r i v a t i o n of back unrounded non-low vowels by BH, and as we w i l l see, by RH. I t i s not possible to state (3.11) using [+round], since the p r o h i b i t i o n i s against the presence of the back counterparts of / i / and /e/, i . e . the set of back unrounded vowels. Given that (3.11) must include the feature value [-round], there are two t h e o r e t i c a l p o s s i b i l i t i e s within the theory of RU that could account for the presence of t h i s feature value. The f i r s t i s that [-round] i s the l e x i c a l l y s p e c i f i e d value i n Hungarian, which i n turn means that the context-free default rule for [round], given i n (3.5), must be changed to a rule i n s e r t i n g [+round] redundantly. 147 The second p o s s i b i l i t y i n the theory of RU i s that [-round] i s not a l e x i c a l l y s p e c i f i e d value, but rather that [+round] i s the l e x i c a l l y s p e c i f i e d value and [-round] i s a redundant value inserted p r i o r to (3.11) by the RROC. I believe that there are two reasons to r e j e c t t h i s option. The f i r s t i s that (3.11) i s a condition which holds both of URs and throughout the l e x i c a l phonology, and i f [-round] were a redundant feature value inserted p r i o r to (3.11) t h i s value would have to be inserted p r i o r to the UR. The MRC, given i n (3.3), s p e c i f i c a l l y rules out t h i s p o s s i b i l i t y . The second reason for r e j e c t i n g the option that the [-round] value used i n (3.11) i s a redundant value r e l a t e s to the blocking e f f e c t s of s p e c i f i e d feature values i n the theory of RU. The only possible redundancy r u l e i n (3.5) which could f i l l i n redundant [round] i s r u l e 4 i n (3.5) (rule 9 i n (3.5) inserts [-round] only on non-back non-low vowels). If rule 4 were ordered p r i o r to (3.11), a l l vowels underlyingly unspecified for [+round] would become [-round] p r i o r to (3.11) and consequently p r i o r to RH, and RH would have no e f f e c t i n the language. I therefore assume that the fact that the marking condition i n (3.11) requires the use of [-round] i s one further piece of evidence that t h i s i s the l e x i c a l l y marked feature value i n Hungarian. At t h i s point i t i s necessary to b r i e f l y examine the epenthesis facts of Hungarian. MacWhinney (1974) notes the presence of a 'linking vowel' which surfaces as [o], [e] or [0], exactly l i k e the vowels of the ternary s u f f i x e s i n 148 (3.13). Vago (1980) argues that t h i s l i n k i n g vowel does not e x i s t underlyingly, but rather i s inserted before consonant i n i t i a l s u f f i x e s , such as the accusative H b or the p l u r a l -k, and i n the f i n a l s y l l a b l e of ce r t a i n r o o t s 1 1 . (3.17) Linking Vowel P l u r a l a. v i r a f a : ] g * flower' viragok bo[o:]r 'skin' borok ke[e]nye[e:]r 'bread' kenyere[e]k bab * bean' babok Accusative b. te[£]le[£]k 'plot' teleke[£]t z i r a [ a : ] f ' g i r a f f e ' z i r a f o t te[£]he[e:]n 'cow' tehene[£]t x c. /bokr/ * shrub' bokor /tehr/ 'load' te[£]he[£]r The forms i n (3.17a) show the addition of the l i n k i n g vowel before the p l u r a l s u f f i x H£ and those i n (3.17b) show the addition of t h i s vowel before the accusative s u f f i x ^ t . The forms i n (3.17c) demonstrate the presence of t h i s l i n k i n g vowel r o o t - i n t e r n a l l y . Vago (1980) argues that the best analysis of the i n f l e c t i o n a l paradigms of nouns stems such as 'shrub' and 'load' assumes that underlyingly these roots contain only a single vowel and that the surface forms are derived through epenthesis. The q u a l i t y of the epenthetic vowel w i l l i n part be determined by the rules of BH and RH. 149 In RU i t i s claimed that the simplest account of an epenthetic vowel i s the i n s e r t i o n of a bare s k e l e t a l s l o t (see 2.2.1.2), with a l l surface feature s p e c i f i c a t i o n s provided by the redundancy rules of the language. In 2.5.1.1 I discuss Epenthesis i n the context of RU, arguing that t h i s r u l e can be assumed to be an automatic r e s u l t of s y l l a b i f i c a t i o n p r i n c i p l e s . An empty s k e l e t a l s l o t w i l l be added to the representation i f c e r t a i n segments remain u n s y l l a b i f i a b l e , and the empty s k e l e t a l s l o t w i l l be interpreted as a s y l l a b l e head ( i . e . a vowel). Since 2 of the 3 vowels that surface i n the epenthetic vowel p o s i t i o n i n (3.18) are mid, I assume, following Jensen and Stong-Jensen (1989) and Vago (1980), that the vowel that underlies t h i s a l t e r n a t i o n i s both [-low] and [-high]. If the i n i t i a l feature s p e c i f i c a t i o n s of t h i s vowel are to be supplied by the redundancy rules of the language, then we must assume that i n Hungarian the context-free parameter for the feature [high] has been reset, and the language possesses a complement rul e which inserts the feature value [-high]. The q u a l i t y of the epenthetic vowels i n (3.18) provide the strongest source of evidence for the l e x i c a l s p e c i f i c a t i o n of [-round] i n Hungarian. The epenthetic vowel and the mid front unrounded vowel cannot be the same vowel since the epenthetic vowel i s a target of both BH and RH, while the mid front unrounded vowel i s neutral to the BH process. While i t could be suggested that the rounding of the l i n k i n g vowel following a root such as bab i n (3.18) i s 150 supplied by the surface rounding of /a/ (see (3.1)), i t i s just as true that the l i n k i n g vowel following a root such as v i r a g , where the long low vowel i s never rounded, also surfaces as a round vowel. These facts then suggest that the epenthetic vowel must underlyingly be eit h e r /o/ or /o7, both of which are [+round]. Given the assumption that the feature s p e c i f i c a t i o n s of epenthetic segments are supplied by redundancy r u l e s , one such r u l e w i l l have to be [ ] --> [+round]. Following the analysis of RH i n Jensen and Stong-Jensen (1989), RH can be stated as i n (3.18). (3.18) Round Harmony Spread [-round] L —> R Target condition: [-high] Nucleus may not branch Domain: l e x i c a l ( c y c l i c ) Like the feature [+back], [-round] w i l l be l e x i c a l l y unassociated i n Hungarian, and w i l l be linked by the Association Conventions p r i o r to the operation of RH. The [-high] target condition i s necessary so that only mid and low vowels w i l l be affected by RH. The marking condition given i n (3.11) w i l l operate to check the outputs of (3.18), to make c e r t a i n the back non-low unrounded vowels are not derived. Redundancy rules 1 and 7 i n (3.7) must be ordered p r i o r to RH, since these rules i n s e r t redundant values that 151 are mentioned i n the marking condition (see the discussion i n 3.1.2.1 regarding the rel a t i o n s h i p of (3.11) and the rul e of BH) . If [-high] i s a redundant value i n Hungarian, as the l i n k i n g vowel facts suggest, and RH mentions [-high] as a target condition, then the RROC w i l l order a l l redundancy rules which make reference to [-high] before the operation of (3.17). The triggers of RH w i l l be only those vowels s p e c i f i e d as [-round], / i / and /e/ w i l l be l e x i c a l l y s p e c i f i e d as [-round], while the low vowels /a/ and /e/ w i l l become s p e c i f i e d as [-round] by redundancy ru l e 7 i n (3.7), which w i l l also be ordered p r i o r to (3.18) by the RROC. Low vowels need not be s p e c i f i c a l l y excluded from undergoing RH, since both /a/ and /e/ are redundantly [-round], and therefore would not be changed by the r u l e . The operation of RH i s shown i n (3.19) i n the derivations of nezhez and fejhez from (3.13). I assume the s u f f i x a l vowel i s underlyingly unspecified, as the l i n k i n g vowel would be. The derivation of these forms begins with the addition of the A l l a t i v e s u f f i x , a f t e r redundancy rules 2a, 5 and 7 from (3.7) have applied within the root. 152 (3.19) Morphology n e z + h O z f e j + h O z X X X X X X X X X X X X X V i [-rd] [-rd] I I [-hi] [-hi] [+lo] BH n/a n/a [ ]—>[-hi] n e z + h 0 z X X X X X X X V [-rd] [-hi] [-hi] f e j X X X I [-rd] [-hi] I [+lo] h o z X X X [-hi] X X X X V— [-rd] [-hi] RH n e z + h e z f e j + h e z X X X X X X X X X [-rd~f I [-hi] [-hi] [-hi] [+lo] BH does not apply i n either of these forms, since neither root has a [+back] autosegment. In both forms the [-round] s p e c i f i c a t i o n of the root vowel spreads to the s u f f i x a l vowel by RH, a f t e r the redundancy rule [ ] —> [-high] has applied ([+low] —> [-high] i s inapplicable here). The [back] and [low] s p e c i f i c a t i o n s w i l l be f i l l e d i n la t e i n the der i v a t i o n 153 by redundancy rules 1 and 3 i n (3.7). Following Jensen and Stong-Jensen (1989), I assume that RH must be e x t r i n s i c a l l y ordered a f t e r BH, i n order to achieve the correct r e s u l t with forms such as lanchoz i n (3.14). The i n i t i a l root vowel i n t h i s form i s both a t r i g g e r of BH and a t r i g g e r of RH (although /a/ i s not underlyingly s p e c i f i e d as [-round] i t receives t h i s redundant s p e c i f i c a t i o n by rule 7 i n (3.5) before the operation of the r u l e ) . The correct derivation of lanchoz i s shown i n (3.2). (3.20) 1 £ n c + h o z X X X X X X X X [+bk] [+low] [ ] --> [-lo] l £ n c + h o z [+low]—>[-rd] x x x x x x x x [+bk] \j | [+low] [-lo] [-rd] Association of [+bk] 1 a n c + h o z x x x x x x x x V [+bk] I [+low] I [-rd] [-lo] 154 BH 1 a n c + h o z x x x x x x x x [+bk] [+low] [-lo] [-rd] RH blocked by (3.11) lanchoz In t h i s derivation RH i s ordered a f t e r BH (and therefore by default a f t e r the Association of [+back]. Redundancy rules 1 and 7 i n (3.7) are ordered p r i o r to the Association of [+back] (they are triggered by the marking condition i n (3.11) which serves as a f i l t e r on the output of t h i s a s s o c i a t i o n ) . [+back] i s linked to the root vowel and then BH operates to spread [+back] to the s u f f i x a l vowel. RH i s consequently blocked by (3.11), for i t s a p p l i c a t i o n would derive an unrounded non-low back vowel. If RH were to apply before BH and the Association Conventions (after the application of redundancy rules 1 and 7 from (3.7)), the s u f f i x a l vowel would become [-round] and BH would be blocked from applying by (3.11). The output would then be the ungrammatical lanchez. Although e x t r i n s i c ordering among phonological rules i s not desirable, i t appears to be required i n t h i s p a r t i c u l a r instance. 155 3.1.2.3 Low Front Vowel Formation As discussed i n 3.1 I have assumed that Hungarian contains the configuration constraints shown i n (3.2). A mid short unrounded vowel can be derived through the operation of BH or RH. One such vowel arises i n the course of the derivation of nezhez given i n (3.19). We must assume that there i s a ru l e of Low Front Vowel Formation, as given i n (3.21), which w i l l change a short mid front unrounded vowel into a low front unrounded vowel. (3.21) Low Front Vowel Formation Insert: [+low] Target condition: [-low] [-high] [-back] [-round] Non-branching nucleus Domain: p o s t l e x i c a l (3.21) changes the value of [low] of a non-low, non-high front unrounded vowel. It i s a purely phonetic r u l e which applies a f t e r a l l redundant s p e c i f i c a t i o n s have been assigned and may apply to the output of both BH and RH. (3.21) appears to be a d i r e c t consequence of the surface constraint given i n (3.2) on *[e] and (3.2) may then be an example of a dynamic constraint, as has been discussed by Pulleyblank and Archangeli (1990) and LaCharite (1990). 156 3.1.3 Parametric CU Analysis of Hungarian The components of UG that I assume are part of a parametric CU theory of a c q u i s i t i o n are discussed i n 2.5.2 and a schematized model i s given i n (2.52). The most basic p r i n c i p l e of t h i s theory i s the R e s t r i c t i v e Redundancy Condition or RRC, given i n (2.45) and repeated here i n (3.22). (3.22) The R e s t r i c t i v e Redundancy Condition (RRC) a. Underlying representations do not contain feature values that are not used c o n t r a s t i v e l y . b. The most highly valued system contains the fewest number of features and feature values needed to co n t r a s t i v e l y d i s t i n g u i s h the inventory of a language. In CU D-values of features are those values that contrast within a s p e c i f i c class of segments. D-values are always present underlyingly, while R-values or redundant feature values are provided by R-rules. Contrastively s p e c i f y i n g the v o c a l i c system of Hungarian produces the system i n ( 3 . 2 3 ) 1 3 . (3.23) Parametric CU Vowel System of Hungarian i e s u o a t l B high + - + - + -low - + - + back - + + + - -round - - + + 157 The contrastive s p e c i f i c a t i o n s are a r r i v e d at as follows, / i / , /il/ and /u/ contrast with /e/, /oV and /of with regard to the feature [high], and so are each s p e c i f i e d for one value of t h i s feature, /e/ and /e/ and /a/ and /o/ contrast with regard to the feature [low] and are therefore each s p e c i f i e d for a value of [low], /u/, /o/ and /a/ are each paired for backness with /il/, /oY and /e/, and so are s p e c i f i e d as [+back] or [-back], / i / and /e/ contrast i n roundness with /il/ and /0/, and so each one of these segments i s s p e c i f i e d as either [+round] or [-round]. In the system i n (3.23) /a/ and /o/ also contrast with regard to the feature [round] at a phonological l e v e l , even though [a] i s [+round] phonetically. I have chosen to represent these segments as contrasting only with regard to the feature [low] for two reasons. F i r s t , /a/ does not p a r t i c i p a t e i n the RH process, as shown by the forms i n (3.14), which we would expect i t to i f i t were s p e c i f i e d as [-round]. Secondly, the low vowels /a/ and /e/ p a r t i c i p a t e i n a process of Low Vowel Lengthening (MacWhinney 1974, Vago 1980), which does not a f f e c t lo/. I w i l l not discuss t h i s process here, but simply assume that i t i s required i n the grammar of Hungarian, and that i t requires that /a/ be s p e c i f i e d as [+low]. The fact that there i s some indeterminacy i n the types of contrastive s p e c i f i c a t i o n s that can be posited for a given language suggests that there may be some l e a r n a b i l i t y problems associated with t h i s type of theory. This issue 158 w i l l be addressed at length i n Chapter 6. As with the RU analysis presented i n 3.1.3, the system i n (3.23) assumes that /e/ and /e/ have d i f f e r e n t sets of feature s p e c i f i c a t i o n s , /e/ only surfaces when i t i s associated with two sk e l e t a l s l o t s dominated by a branching nucleus, while fcl can only surface when associated to a single s k e l e t a l s l o t dominated by a nucleus. These r e s t r i c t i o n s are formalized i n (3.2), as phonetic constraints which do not block the derivation of [e] or [e:] during the l e x i c a l phonology. As i s discussed i n 2.5.2.1, i n a parametric theory of CU R-rules are the unmarked featural parameters supplied by UG. These rules i n s e r t feature values that are not l e x i c a l l y s p e c i f i e d . The set of universal R-rules given i n (2.48) i s repeated here i n (3.24). (3.24) Universal R-rules: 1. [+low] — > [-high] 2. [+low] — > [+back] 3. [+low] - - > [-round] 4. [+back] - - > [+round] [-low] 5. [-back] - - > [-round] [-low] 6. [+high] - - > [-low] 7. [-high] — > [-low] [-back] 159 Resetting a featural parameter to the marked s e t t i n g i s allowed f o r , and means that c e r t a i n s p e c i f i c a t i o n s that are provided u n i v e r s a l l y by UG must be marked underlyingly because they are contrastive. In RU, when a context-sensitive r u l e parameter i s reset, by the MRC that rule must be eliminated from the language-specific grammar (see discussion i n 3.1.3). In CU, however, the R-rule need not be eliminated from the language-particular grammar, because there i s not the same re l a t i o n s h i p between redundant information and redundancy rul e s , and because redundancy rules play no part i n the active phonology (this issue i s discussed i n 2.2.2.2 and 2.5.2.1). Every language w i l l therefore have the same core set of R-rules i n (3.23), regardless of the underlying markings i n the language. As shown i n 2.5.2.1, i t i s possible that language-s p e c i f i c R-rules w i l l have to be added to the grammar of the language i n order to achieve f u l l s p e c i f i c a t i o n of the system. The data showing BH i n (3.9) and RH i n (3.13) demonstrate that Hungarian has d i s t i n c t front and back low vowels and front round and unround vowels. This means that R-rules 2 and 5 i n (3.24) have been reset to the marked option i n t h i s language. (3.25) Marked Featural Parameters of Hungarian *2. [+low] --> [+back] *5. [-back] —> [-round] [-low] The r e s u l t of the resetting of these featural parameters 160 w i l l be that the c h i l d c ontrastively s p e c i f i e s the low vowels for the feature [back] and the front non-low vowels for the feature [round]. When t h i s i s done the system shown i n (3.23) w i l l be achieved. In (3.23) / i / and /e/ are not s p e c i f i e d for backness since they do not contrast with back vowels, and the R-rules i n (3.24) do not supply the [-back] s p e c i f i c a t i o n s for these vowels. Consequently, the language-specific r u l e i n (3.26) must be added to the core set of rules to provide these s p e c i f i c a t i o n s . (3.26) Hungarian Language-Specific R-rule: 8. [-low] —> [-back] [-round] Given that CU assumes that both values of a feature are l e x i c a l l y marked where these values are used c o n t r a s t i v e l y , there i s no requirement i n t h i s theory for a set of context-free redundancy rules as i n the theory of RU. The c h i l d acquiring Hungarian must learn that the featural parameters i n (3.25) must be reset, with the r e s u l t that the contrastive s p e c i f i c a t i o n s of [back] for low vowels and the contrastive s p e c i f i c a t i o n s of [round] for front vowels are marked underlyingly. In Steriade (1987) i t i s shown that i n CU phonological rules may be feature-changing, may be blocked when a segment i s encountered that i s s p e c i f i e d for some value of the spreading feature, and may be i n i t i a t e d by both values of a 161 feature. In order to further constrain the number and types of phonological rules allowed by t h i s theory, i n 2.5.2.1 I assume that the parameteric theory of CU allows only for rules of spreading, delinking, fusion or i n s e r t i o n . These can be viewed as rule parameters, with the unmarked value set at OFF. If there i s p o s i t i v e evidence i n a language that a ru l e i s operative, then the value of that rule parameter w i l l be set to ON. In the statement of phonological rules i n the theory of CU I state the function of the rule ( i . e . spread, delete), the argument (i.e.. [+back]) and the t r i g g e r / t a r g e t conditions, as well as the domain of the rule ( i . e . l e x i c a l ( c y c l i c / p o s t c y c l i c ) , p o s t l e x i c a l ) . In the following sections I present analyses of Hungarian Back and Round Harmony, and also of several more peripheral rules which manipulate vowels i n t h i s language. Unlike i n the RU analyses, I do not attempt to argue for p a r t i c u l a r l e x i c a l values of features. Rather, I focus on the feature-changing aspects of these rules, and on demonstrating how these rules can account for the data using a set of contrastive s p e c i f i c a t i o n s . 3.1.3.1 Back Harmony The parametric CU analysis of BH presented here follows that i n Steriade (1987) 1 4. The major problem associated with Steriade's (1987) CU analysis of Hungarian, discussed i n 2.2.2.2, can be resolved given the parametric theory of CU 162 and the assumption that short and long vowels i n Hungarian have i d e n t i c a l feature matrices. In Steriade's analysis the redundancy rule [+low] —> [+back] i s ordered p r i o r to the r u l e of BH, so that /a/ [a:] can t r i g g e r BH. This ordering r e l a t i o n s h i p i s not necessary i n the parametric theory, given the system i n (3.23) i n which i t i s assumed that /a/ contrasts with /£/ i n backness and i s underlyingly marked as [+back]. Steriade's analysis focusses p a r t i c u l a r l y on the operation of BH i n loanwords, while mine deals almost exc l u s i v e l y with native words. BH i s an i t e r a t i v e , feature-changing rule which spreads [+back]. (3.27) Back Harmony Spread [+back] L —> R Feature-changing Domain: l e x i c a l ( c y c l i c ) (3.27) i s a c y c l i c l e x i c a l r u l e , and may therefore have exceptions. In her analysis, Steriade claims that Structure Preservation prohibits BH from spreading [back] to the vowels / i / and /e/, since they are underlyingly unspecified for the feature [back], and therefore may not l e x i c a l l y be associated with t h i s feature. I assume, as i n the RU analysis, that the conditions encompassed by Structure Preservation are formally given by a set of marking conditions. The p r o h i b i t i o n on back non-low rounded vowels can be stated as i n (3.28). 163 (3.28) * [+back] [-round] (3.28), l i k e the RU condition given i n (3.11), holds throughout the lexicon, and w i l l therefore be i n e f f e c t when BH applies. This condition must be turned o f f p o s t l e x i c a l l y , i n order for [a] to receive a phonetic s p e c i f i c a t i o n for [-round]. The marking condition i n (3.28) d i f f e r s from the RU condition i n (3.11) i n that the RU condition also includes the feature value [-low]. This difference arises because of the feature s p e c i f i c a t i o n s required by these two theories. (3.28), the CU condition, does not include the feature [-low] because then only /e/, and not / i / , would be subject to t h i s condition (see (3.23)). (3.11), the RU condition, requires the presence of [-low] so that /e/ i s not included as one of the segments that does not have a back counterpart (see (3.8)). Steriade argues, following Ringen (1980), that i t i s the SCC (given i n (2.2)) which stops BH from operating root-i n t e r n a l l y i n loanwords. Ringen (1980) argues that only the i n i t i a l root vowel of harmonic roots i s s p e c i f i e d for the feature [back], while i n disharmonic roots (e.g. loanwords), a l l root vowels are underlyingly s p e c i f i e d for a value of [back]. The SCC (or the Revised Alternation Condition as discussed i n Ringen 1980) w i l l prevent BH from applying within disharmonic roots, since i t could only operate i n a 164 n e u t r a l i z i n g fashion, while i t would permit BH to apply i n harmonic roots because these applications would not be n e u t r a l i z i n g . I therefore assume that i n native roots (the a p p l i c a t i o n of BH i n loanwords w i l l not be dealt with here) [back] i s s p e c i f i e d only for the i n i t i a l v owel 1 3. The d i f f i c u l t y that arises i f we assume that only i n i t i a l vowels i n native words are s p e c i f i e d for [back] i s that n o n - i n i t i a l root vowels may underlyingly have d i f f e r e n t feature s p e c i f i c a t i o n s than those provided i n (3.23). For example, a n o n - i n i t i a l vowel which surfaces as [il] a f t e r BH, w i l l be underlyingly s p e c i f i e d only as [+high] and [+round]. There i s no vowel given i n (3.23) which corresponds to t h i s p a r t i c u l a r set of feature combinations. We w i l l then have to assume that i n CU Structure Preservation prohibits only those feature combinations that have associated marking conditions such as (3.28), and that there i s no condition associated with a feature combination such as [+high] and [+round]. This p a r t i c u l a r i n t e r p r e t a t i o n of Structure Preservation contrasts sharply with that given for RU. As discussed i n 3.1.2.1, Structure Preservation i n RU permits only the sets of underlying feature s p e c i f i c a t i o n s to e x i s t throughout the L e x i c a l Phonology, and marking conditions (such as (3.11)) w i l l e x i s t to s p e c i f i c a l l y rule out a l l other feature combinations. In the CU analyses which follow I specify vowels that do not correspond to an underlying feature matrix i n (3.23) as 'V. Steriade claims that i n the unmarked case, s u f f i x a l 165 vowels i n Hungarian are s p e c i f i e d as [-back]. This i s c r u c i a l i n order to account for the neutral-vowel-only words given i n (3.9di). In these forms the root vowel i s one of the two neutral vowels, which are unspecified for the feature [back], and therefore do not act as triggers of BH; and yet the s u f f i x a l vowel surfaces as [-back]. It i s then c r u c i a l to assume that s u f f i x a l vowels are l e x i c a l l y s p e c i f i e d as [-back]. The fact that s u f f i x a l vowels must be s p e c i f i e d as [-back] then rules out the p o s s i b i l i t y of p o s i t i n g [back] as a f l o a t i n g autosegment which associates to the i n i t i a l root vowel. In order to account for neutral-vowel-only forms which take back vowel suffixes (shown i n ( 3 . 9 d i i ) , we would then have to assume that the Association Conventions, which would l i n k a f l o a t i n g autosegment to a s u f f i x a l vowel, operate i n a feature-changing manner. This i s e n t i r e l y at odds with our conception of the Association Conventions as simple l i n k i n g mechanisms. I therefore assume that i n native roots the i n i t i a l vowel i s l e x i c a l l y linked to some s p e c i f i c a t i o n for the feature [back]. Steriade d i d not attempt to account for the neutral-vowel-only roots i n (3.9dii) which take back vowel s u f f i x e s . I assume that these roots t r i g g e r a rule which adds [+back] to the r e p r e s e n t a t i o n 1 6 . This rule, l i k e BH, w i l l have to be feature-changing, and w i l l be triggered by a small f i x e d -class set of native roots. 166 (3.29) Insert [+back] Insert: [+back] Trigger Condition: only in roots marked * Domain: lexical (cyclic) Feature-changing The [+back] feature cannot link to the root vowel (by (3.28)), but w i l l be able to link to a suffixal vowel. We now have a l l the components necessary to describe the functioning of BH in the forms in (3.9). Derivations of native words betunek, r a d l r t 6 l and celnak are shown in (3.30) (ignoring for the moment the effects of RH). In this derivation I have ordered the rule Insert [+back] after BH, although the rules may apply in either order. (3.30) BH in native words Underlying b c t V r a d i r * c e l X X X X X X X X X X X X X X [-bk] [+bk] [+lo] [+lo] [-lo] [+hi] [+hi] [-hi] [+rd] [-rd] [-rd] 167 BH b e t i i X X X X [-bk] [+lo] [+hi] [+rd] Prohibited by (3.28) n/a Morphology beta XXXX [-bk] nek [+16] [+hi] t+rd] XXX I I [-bk] [+bk] I I [+lo] [+lo] radi r XXXXXX to 1 XXXX V [-bk] [-hi] BH betti XXXX \A [-bk] [+lo] nek XXX radi r + to 1 [+hi] [+rd] XXXXXX "\ L [-bk] [+bk] I I [+lo] [+lo] XXXX [-bk] [-hi] n/a Insert [+back] n/a n/a *ce 1 XXXX + nak XXX [+bk][-bk] [-lo] [+lo] [-hi] [-rd] 168 BH can apply i n a structure-building fashion i n betti to spread [-back] to the second root vowel from the i n i t i a l vowel. BH i s blocked i n r a d l r by (3.28) since / i / i s underlyingly s p e c i f i e d as [-round]. BH i s not applicable i n e e l . The s u f f i x e s nek and t f l l are then added, i n which the s u f f i x vowels are both s p e c i f i e d as [-back]. BH then reapplies to spread [-back] i n bettl and [+back] i n r a d l r to the s u f f i x . In bettlnek I assume that BH applies vacuously, while i n radlrnak i t applies i n a feature-changing fashion. The f i n a l step i s for [+back] to be added to the s u f f i x a l vowel i n celnak by (3.29), delinking the o r i g i n a l [-back] s p e c i f i c a t i o n . In the derivation of radlrnak i n (3.30) the [+back] feature of the i n i t i a l vowel i s prohibited from associating to the neutral vowel of the root by (3.28). BH can reapply once the Dative s u f f i x i s added, and the [+back] feature of the i n i t i a l vowel w i l l spread over the neutral vowel to l i n k to the s u f f i x . This spreading w i l l not v i o l a t e the L o c a l i t y Condition (Archangeli and Pulleyblank 1987) i f fx/ and /e/ are assumed to be i n e l i g i b l e targets as a r e s u l t of (3.28). The i n i t i a l root vowel and the target s u f f i x a l vowel i n radlrnak can be adjacent, i f the intervening nucleus node i s ignored for the purposes of the rule of BH. In a neutral-vowel-only root which does not belong to the exceptional class ( i . e . those forms i n (3.9di)), the s u f f i x a l vowel surfaces as [-back]. This i s e a s i l y accounted 169 for i f the s u f f i x a l vowel i s underlyingly s p e c i f i e d as [-back] and the i n i t i a l vowel has no s p e c i f i c a t i o n for backness. (3.31) Underlying sz i n X X X X V [+hi] [-rd] BH n/a Morphology sz i n + n e k X x x x [-bk] [+lo] BH n/a 3.1.3.2 Round Harmony Steriade (1987) does not attempt to account for the RH facts of Hungarian, but i t i s possible to develop an analysis using the parametric theory of CU. The three vowels that alternate through RH are [e], [o] and [0], which are a l l [-high] and [-low]. The forms i n (3.14) demonstrate that the 170 s u f f i x a l vowel i n the '2nd person p l u r a l ' , ' a l l a t i v e ' and ' p l u r a l ' must be unspecified for roundness, or BH to the s u f f i x a l vowel w i l l be blocked by the marking condition i n (3.28). And f i n a l l y , the neutral-vowel-only form szlntek i n (3.13) demonstrates that the s u f f i x a l vowel i n these forms must be unspecified for backness, or (3.28) would block the operation of RH to the s u f f i x a l vowel i n these forms. Given that none of the three vowels e/o/tt i s unspecified for both [round] and [back], we must assume, as was also necessary i n the discussion of BH i n 3.1.3.1, that the s u f f i x a l vowel underlying the RH alternations i s not represented by one of the feature sets shown i n (3.23). Again i t i s necessary to assume that t h i s vowel can e x i s t because i t i s not s p e c i f i c a l l y ruled out by a marking condition such as (3.28). We could also entertain the proposal that the s u f f i x a l vowels which p a r t i c i p a t e i n RH are completely unspecified, as i n the RU analysis. Given that we must assume that sets of feature s p e c i f i c a t i o n s can e x i s t which are not s p e c i f i c a l l y l i s t e d i n (3.23), i t could also follow that one possible vowel representation i s the absence of any features. This proposal, however, can be seen to be untenable, i n l i g h t of a form such as foldhOz, which would undergo both BH and RH. Assuming that both [-back] and [+round] spread from the root vowel to a t o t a l l y unspecified s u f f i x a l vowel, we then f i n d that there are no redundancy rules given i n (3.24) or (3.26) which could add the [-high] and [-low] s p e c i f i c a t i o n s of the 171 s u f f i x a l vowel. A rule i s given i n (3.36) which adds p r e c i s e l y these feature s p e c i f i c a t i o n s ; however, t h i s p a r t i c u l a r r u l e contains a target condition r e s t r i c t i n g i t s a p p l i c a t i o n to empty s k e l e t a l s l o t s ; This r u l e w i l l then not be able to target a vowel previously s p e c i f i e d as [-back] and [+round]. The suffixes which undergo ternary alternations between e, o and 0 w i l l then d i f f e r from those that undergo only binary alternations involving BH i n that the ternary s u f f i x e s are unspecified for both backness and roundness, while the binary s u f f i x e s are unspecified only for roundness. RH i s a l e x i c a l rule which can have exceptions. I assume that i n general [round] i s linked only to the i n i t i a l root vowel (as [back] i s ) . The representation of vowels i n (3.23) shows that the possible triggers of RH are / i / , /e/, / t i / and /0/, with the f i r s t two s p e c i f i e d as [-round] and the second two [+round]. According to t h i s set of s p e c i f i c a t i o n s the back non-low vowels /u/ and /of w i l l not be t r i g g e r s of RH. Being s p e c i f i e d as [+back], these two vowels w i l l t r i g g e r only BH, spreading [+back] to a s u f f i x a l vowel, which w i l l then become [+round] by R-rule 4 i n (3.24). [-round] w i l l be prohibited from associating to a vowel s p e c i f i e d as [+back] by the marking condition i n (3.28), and since we know that only mid vowels p a r t i c i p a t e i n RH, a target condition must indicate that only [-high] vowels p a r t i c i p a t e ([a] w i l l be prohibited from p a r t i c i p a t i n g because of (3.28)). As i n the RU analysis, RH w i l l also be 172 blocked from applying to long vowels. The rule of RH i s given i n (3.32). (3.32) Round Harmony Spread [+round] L --> R Target condition: [-high] Nucleus may not branch Domain: l e x i c a l ( c y c l i c ) Feature-changing Derivations for hazhoz, fffztflk and fejtek from (3.13) are given i n (3.33). (3.33) Round Harmony Morphology h a z + h V z BH X X X X V [+bk] [+lo] [+lo] X X X [-lo] I [-hi] h o z X X X X X X X [+bk] [-lo] [-hi] f 0 z + t V k X X X X X X X V [-bk] [-hi] [+rd]' f o [-hi] [+rd] [-lo] [-hi] t o k X X X X X X X [-bk] [-lo] I [-hi] f e j + t V k X X X X X X [-bk] [+lo] [-lo] I [-hi] f e j t V k X X X X X X [-bk] [+lo] [ j l o ] [-hi] 173 RH n/a [-hi/ [-hi] 1/ [+rd] n/a R-rules 4 and 5 i n (3.24) z + h o z x x x [-lo] [-hi] I [+rd] n/a The vowels i n the 2nd person p l u r a l and A l l a t i v e s u f f i x e s are underlyingly s p e c i f i e d as [-high] and [-low]. BH and RH may apply i n any order. BH applies i n a l l three forms to give the s u f f i x a l vowel a s p e c i f i c a t i o n for backness. RH operates only i n fcYztok to specify the s u f f i x a l vowel as [+round]. The s u f f i x a l vowels i n hazhoz and fejhez receive t h e i r surface s p e c i f i c a t i o n s for roundness l a t e i n the der i v a t i o n by R-rules 4 and 5 from (3.24). The f i n a l step i n the deri v a t i o n of fejhez i s the lowering of the s u f f i x a l vowel to [e]. This rule w i l l be discussed i n 3.1.3.3. In a neutral-vowel-only root such as szfntek, shown i n (3.13a)/ RH w i l l operate as shown i n (3.34). 174 (3.34) Morphology sz i n + t V k x V x [+hi] [-rd] x x x I [-hi] [-lo] BH n/a RH sz i n + t V k X X X X V [+hi] X X X ' [-hi] [-lo] [-rd] ru l e 9 i n (3.26) BH i s inapplicable i n t h i s form. RH w i l l spread [-round] to the s u f f i x a l vowel, and the language-particular R-rule i n (3.26) w i l l supply the [-back] surface value for t h i s vowel. There i s no reason to posit any ordering r e l a t i o n s h i p between BH and RH as i s required i n the RU analysis. Each 175 ru l e may apply whenever possible, because of the fact that we have assumed that the s u f f i x a l vowels are unspecified for eit h e r [round] or [back]. 3.1.3.3 Peripheral Rules As discussed i n 3.1.2 I assume that there are two d i s t i n c t non-high unrounded front vowels i n Hungarian: /e/ and /e/. /e/ surfaces as [e] only when i t i s associated with two s k e l e t a l s l o t s dominated by a branching nucleus, which I assume, following Levin (1985), i s the representation of a long vowel (see discussion i n 2.1.2.4). /c/, on the other hand, can surface only when associated to a single s k e l e t a l s l o t dominated by a nucleus, i . e . when i t i s a short vowel. These constraints are given i n (3.2). When a short front mid vowel i s derived through the operation of BH or RH (e.g. fejtek i n (3.33)) the constraint on [e] w i l l t r i g g e r a rule of Low Front Vowel Formation, given i n (3.35). (3.35) Low Front Vowel Formation Insert: [+low] Target condition: [-low] [-high] [-back] [-round] Non-branching nucleus Domain: p o s t l e x i c a l 176 (3.35) changes the value of [low] of a mid front unrounded vowel. It i s a purely phonetic rule which applies a f t e r a l l redundant s p e c i f i c a t i o n s have been assigned, and can apply to the outputs of both BH and RH. The f i n a l rule that i s necessary for t h i s CU analysis of Hungarian harmony i s a rule i n s e r t i n g the features for an epenthetic vowel s l o t . If we are to maintain that Epenthesis ins e r t s a vowel s l o t automatically when s y l l a b i f i c a t i o n requires i t , i n the CU analysis i t w i l l also be necessary to have a rule providing some feature s p e c i f i c a t i o n s for t h i s empty s l o t . Such a rul e i s not necessary i n the RU analysis, since the redundancy rules of the language i n s e r t the feature s p e c i f i c a t i o n s of the epenthetic vowel. In CU, however, redundancy rules are only context-sensitive, and i t i s therefore impossible to add features to an empty s k e l e t a l s l o t . Since the l i n k i n g vowel facts i n (3.18) show that the epenthetic vowel undergoes the same e/o/B al t e r n a t i o n as vowels affected by RH, I assume that the features that must be inserted are just those required for the s u f f i x a l vowels shown i n (3.33) and (3.34). This r u l e can then be stated as i n (3.36). (3.36) Epenthetic Vowel S p e c i f i c a t i o n Insert: [-high] [-low] Target Condition: an empty s k e l e t a l s l o t Domain: l e x i c a l ( c y c l i c ) 177 This rule w i l l i n s e r t the features [-high] and [-low] onto a s k e l e t a l s l o t which has no other feature s p e c i f i c a t i o n s . A vowel which has undergone (3.36) w i l l then be e l i g i b l e to p a r t i c i p a t e i n RH and BH. Since a l l three of these rules are c y c l i c l e x i c a l rules, there i s no s p e c i f i c ordering required between them. 3.1.4 Summary and Comparisons Following the analyses of BH and RH i n 3.1.2 the language-specific information that i s required i n the grammar of Hungarian, given the RU analysis, i s as follows: (3.37) Marked parameter settings: FCRs: *[+low] —> [+back] *[-back] —> [-round] [-low] Context-free: [ ] —> [-high] [ ] —> [+round] S y l l a b i f i c a t i o n : Complex N - ON Phonological rules: Back Harmony Round Harmony Low Front Vowel Formation Ordering: BH before RH Surface Constraints: */e/ */e:] 178 There are two marked context-free featural parameters required i n Hungarian, and two marked context-sensitive featural parameters. The language also requires that the s y l l a b i f i c a t i o n option allowing for complex nuclei be set to the marked option, so that long vowels can e x i s t . Three phonological rules are also present i n Hungarian. It i s not necessary to state the Association of [+back] as a s p e c i f i c r u l e of Hungarian, since i t i s an a p p l i c a t i o n of the Association Conventions, which are themselves a part of UG. Back Harmony i s a r e l a t i v e l y simple rule i n that i t does not involve the statement of any target conditions, while RH requires several. Low Vowel Formation i s an unmarked ru l e i n the sense that i t inserts content l i k e a redundancy r u l e ; however, i t has a complex target condition. The grammar of Hungarian requires a statement that BH must be ordered p r i o r to RH. The grammar of Hungarian also contains a marking condition, which i s a formalization of Structure Preservation, p r o h i b i t i n g the derivation of a back unrounded non-low vowel. This condition w i l l hold throughout the lexicon, and i n fact there i s no reason to believe that i t i s ever turned o f f . I have not s p e c i f i c a l l y stated t h i s condition since i t i s a d i r e c t r e s u l t of a p r i n c i p l e of UG. The two constraints stated i n (3.36) p r o h i b i t the feature s p e c i f i c a t i o n s of /e/ attached to a single non-branching nucleus, and the feature s p e c i f i c a t i o n s of /e/ attached to a 179 branching nucleus. While these p a r t i c u l a r feature combinations do e x i s t , they do not e x i s t i n combination with a p a r t i c u l a r s y l l a b i c structure. These constraints cannot hold of the lexicon as shown by the fact that they do not block the derivation of [e]. The fact that these two configuration constraints e x i s t i n Hungarian helps to explain the i d i o s y n c r a t i c behaviour of orthographic e. While e alternates with [a] i n the Dative s u f f i x through BH, i t sometimes appears to v i o l a t e BH constraints i n forms such as ceruza and krapek. If we assume that e i s underlying /e/ i n these words (allowed by the fact that the constraint on *[e] i s phonetic), and that a f l o a t i n g [+back] autosegment i s prohibited from associating to /e/ by the constraint i n (3.28), then i n fact these forms do not v i o l a t e BH r e s t r i c t i o n s . The i n t e r a c t i o n of the redundancy rules i n (3.7) and the phonological rules and constraints of the language that are dictate d by the RROC produce produce the following picture of the phonology of Hungarian: 180 (3.38) Parametric RU Grammar of Hungarian LEXICON Rules 1, 2a, 5 and 7 i n (3.7) Association of [+bk] Marking BH Condition RH J (3.11) POSTLEXICON Rules 3, 4a and 8 i n (3.7) Constraints on /e/ and lex/ Low Front Vowel Formation The only e x t r i n s i c ordering r e l a t i o n s h i p required i s that between BH and RH. A l l other rules apply when and where they can i n conjunction with the Ordering Constraints (see 2.5.1.1) and Universal Grammar. In the CU analysis of BH and RH the following language-s p e c i f i c information i s required: 181 (3.39) Marked parameter settings: *[+low] —> [+back] *[-back] —> [-round] [-low] S y l l a b i f i c a t i o n : complex N - ON Language-specific R-rule: [-low] —> [-back] [-round] Phonological Rules: Back Harmony Round Harmony Epenthetic Vowel S p e c i f i c a t i o n Insert [+back] Low Front Vowel Formation Lex i c a l Markings: Certain native roots are marked * to undergo Insert [+back] *[+back] - Holds only i n lexicon [-round] Constraints: *[e] *[£•] Holds at surface structure A l l the phonological rules require r e s e t t i n g of a r u l e parameter to ON. RH i s more complex than BH because i t requires a target condition, as does Low Front Vowel Formation. Insert [+back] i s probably the most complex ru l e because i t applies only to a small r e s t r i c t e d class of words. 182 In t h i s analysis i t i s not necessary that RH and BH be e x t r i n s i c a l l y ordered, as i s required i n the RU analysis of Hungarian. The constraints on [e] and [e:] prevent the surfacing of [e] and [ c : ] . In addition Hungarian also has the constraint given i n (3.28) which i s supplied by Structure Preservation, but there must be a statement i n the phonology of Hungarian that t h i s constraint turns o f f p o s t l e x i c a l l y . The behaviour of the vowel which surfaces as [e] can be explained by the in t e r a c t i o n of these two types of constraints. Orthographic e may be underlyingly e i t h e r /e/ or /e/. If i t i s derived from /e/, the marking condition i n (3.28) w i l l prevent i t from p a r t i c i p a t i n g i n the l e x i c a l rules of BH and RH, while i f i t i s derived from /e/, i t w i l l alternate with /a/ through the application of BH. Because t h i s theory does not adopt the RROC redundancy rules cannot be ordered among the phonological rules of a language. The only rule of Hungarian which applies a f t e r (or at the same time as) the R-rules i s the phonetic r u l e of Low Front Vowel Formation, giving the following picture of the phonology of Hungarian. 183 (3.40) Parametric CU Phonology of Hungarian LEXICON Back Harmony Round Harmony Marking Insert [+back] Condition Epenthetic Vowel S p e c i f i c a t i o n (3.28) POSTLEXICON R-rules Low Front Vowel Formation *[e] *[£:] Comparing the language-specific information required by the RU and CU analyses of Hungarian harmony, we see that the RU analysis posits that 4 featural parameters require r e s e t t i n g i n t h i s language, while the CU analysis posits that only 2 featural parameters must be reset. I f , however, we compare the phonological r u l e systems required by these two analyses, I believe the RU analysis proves to be the superior one. The CU analysis requires a larger number of phonological rules than the RU analysis, since i t requires a rule of Insert [+back] i n order to account for the small class of neutral-vowel-only roots which take back s u f f i x a l vowels, and a rule providing the feature s p e c i f i c a t i o n s of epenthetic vowels. The exceptional behaviour of the subclass of neutral-vowel-only roots i s 184 accounted for i n the RU analysis by assuming that [+back] i s a f l o a t i n g autosegment, and that Structure Preservation p r o h i b i t s t h i s f l o a t i n g feature from associating to neutral vowels. The feature s p e c i f i c a t i o n s of epenthetic vowels are provided automatically by the redundancy rules of the language. The CU analysis of Hungarian harmony has one added problem i n that these analyses simply do not work unless we assume that c e r t a i n vowel combinations not present i n the basic underlying inventory of the language are permitted to e x i s t . I have hypothesized that c e r t a i n feature combinations, not s p e c i f i c a l l y outlawed by marking constraints, may be allowed, and supplement the basic contrastive system. Undoubtedly t h i s w i l l add to the load of a language learner. 3.2 Spanish Vocalic Alternations In the following sections I present analyses of several processes which operate i n the verbal system of Spanish. One set of alternations appears i n a subclass of 3rd conjugation class verbs, where a root vowel alternates between a mid and high front vowel. The other alternation, which takes place across a l l 3 conjugation classes, shows v a r i a t i o n between a f a l l i n g diphthong and a simple vowel. These alternations are found i n the C a s t i l i a n d i a l e c t of Spanish, the prevalent d i a l e c t spoken i n Spain. In 3.2.1 I discuss the Spanish vocalic inventory and 185 present an analysis of stress. In 3.2.2 I present the RU analysis of the Spanish v o c a l i c alternations, arguing that [+high] must be a l e x i c a l l y s p e c i f i e d feature value i n t h i s language. These analyses demonstrate that stress and s y l l a b i f i c a t i o n are i n t e g r a l i n the statement of the phonological rules which control vowel/diphthong and high/mid vowel alternations. In 3.2.3 I present the CU analysis of these f a c t s , and here the focus again i s on the importance of the feature [high] i n t h i s system, and the i n t e r a c t i o n of s y l l a b i f i c a t i o n and stress with the other phonological r u l e s . As i n the analyses of Hungarian, the CU analysis does not require the same types of argumentation for underlying feature values as i n the RU analysis, since both values of the feature [high] are assumed to be marked underlyingly. Spanish data w i l l i n general be given i n orthographic form, which, p a r t i c u l a r l y for vowels, corresponds quite c l o s e l y to the phonological system i t s e l f . Phonetic representations w i l l sometimes be used for the representation of diphthongs. 3.2.1 Spanish Vowels and Stress The Spanish vowel system ( C a s t i l i a n d i a l e c t ) i s given i n (3.41). (3.41) Spanish Vowels i u e o a 186 Phonetically, /e/ varies between [e] and [e] and /o/ between [o] and [3] (Macpherson (1985). There are no long vowels i n the language, although both r i s i n g and f a l l i n g diphthongs are f r e q u e n t 1 7 . Harris (1983) notes that i n Spanish rhymes only f a l l i n g diphthongs may be followed by a l i q u i d , nasal, g l i d e or s, which comprises the set of segments permitted i n coda p o s i t i o n i n Spanish (e.g. siempre, muerte). Rising diphthongs are not permitted to be followed by consonants. I take t h i s as evidence that only f a l l i n g diphthongs are representative of complex nuclei and that r i s i n g diphthongs must be analyzed as nucleus-coda sequences 1 8. The component elements of a diphthong must come from the set of vowels shown i n (3.41). In order to permit complex nuclei Spanish has chosen the marked parameter se t t i n g of the Complex N parameter, as discussed i n 2.1.2.4. The s y l l a b i f i c a t i o n mechanisms of Spanish w i l l be discussed i n 3.2.2.2. Stress i n Spanish f a l l s on one of the l a s t 3 s y l l a b l e s . Penultimate stress i s unmarked on vowel-final words, while f i n a l stress i s the unmarked pattern i n consonant-final forms. Halle, Harris and Vergnaud (1991) demonstrate that the r e l a t i v e l y complex stress patterns of Spanish can be dealt with by assuming that word stress i n Spanish i s assigned both c y c l i c a l l y and p o s t c y c l i c a l l y , and that the Stress Erasure Convention (Halle 1990) applies. The Stress Erasure Convention (SEC) i s a universal p r i n c i p l e which says 187 that a l l metrical structure i s erased at the beginning of each l e v e l . Halle, Harris and Vergnaud (1991) argue that the language-particular aspects of stress assignment i n Spanish are as given i n (3.42). (3.42) a. word-final vowels are extrametrical b. the rightmost v i s i b l e s y l l a b l e receives an accent mark c. Left-headed feet are constructed i n binary fashion from r i g h t to l e f t d. Main stress i s assigned to the rightmost foot by constructing an unbounded right-headed constituent. Using the rules i n (3.41) and the SEC, Halle, Harris and Vergnaud demonstrate that stress i n Spanish i s rule-governed. Forms that demonstrate exceptional antepenultimate stress (for vowel-final words) or penultimate stress (for consonant-f i n a l forms) rather than the more unmarked penultimate or ultimate stress patterns are treated as being exceptional with regard to the rule i n (3.42b). In the following analysis of Spanish I adopt the stress account of Halle, Harris and Vergnaud (1991), with some elaborations for the verbal forms that I discuss. Stress i s p a r t i c u l a r l y important i n the analyses of the imperfective forms discussed i n conjunction with the high/mid a l t e r n a t i o n f a c t s . 188 3.2.2 Parametric RU Analyses of Spanish Vocalic Alternations The components of UG that are assumed i n the parametric RU model are given i n 3.1.2. These include the MRC, Ordering Constraints, and Universal Default Rules. In the following sections I demonstrate that [+high] must be the l e x i c a l l y marked value of [high] i n Spanish, contrary to the universal markedness considerations that have been assumed i n e a r l i e r chapters. If [+high] i s l e x i c a l l y marked i n Spanish, then the context-free parameter for [high] (rule 2 i n (3.5)) has been reset at the marked option, and a complement ru l e has been created to f i l l i n [-high] as the redundant value. Assuming that the other redundancy rules are unchanged from UG, the redundancy rules necessary for Spanish are those i n (3.43). 189 (3.43) Spanish Redundancy Rules Default Rules 1. [ ] —> [-low] 5. [+low] - - > [-high] 3. [ ] —> [-back] 6. [+low] --> [+back] 4. [ ] —> [-round] 7. [+low] - - > [-round] 8. [+back] [-low] — > [+round] 9. [-back] [-low] - - > [-round] Complement Rule 2a.[ ] --> [-high] If the rules i n (3.43) supply the redundant feature values i n Spanish the marked values w i l l be those i n (3.44) 1 9. (3.44) Parametric RU Spanish Vowel System i e . a o u high + + low + back + + [round] i s a t o t a l l y redundant feature i n t h i s system, and /e/ i s t o t a l l y unspecified. This system i s s i m i l a r to the universal 5 vowel system given i n (2.37), except for the l e x i c a l l y marked values of [high]. If Spanish requires only one featural parameter to be reset at the marked option i t i s a much less complex system than that of Hungarian (discussed i n 3.1), since Hungarian 190 requires that 4 such parameters be reset. The l e x i c a l marking of [+high] i n Spanish i s demonstrated by c e r t a i n facts r e l a t i n g to verb paradigms, and therefore some general issues regarding the formation of Spanish verbs are f i r s t discussed. F i r s t person present i n d i c a t i v e amo como pienso pierdo avanzo muevo compro valgo 3rd conjugation theme vowel - / i / 3.2.2.1 Spanish Verbal Classes There are three verb conjugation classes i n Spanish. Representative paradigms are given for each class i n (3.45). (3.45) Spanish verb classes 1st conjugation 2nd conjugation theme vowel - /a/ theme vowel - /e/ I n f i n i t i v e s amar comer pensar perder avanzar mover comprar valer Past P a r t i c i p l e amado comido pensado perdido avanzado movido comprado valido v i v i r pedir concebir p a r t i r v i v i d o pedido concebido partido vivo pido concibo parto 191 Spanish i n f i n i t i v e s are composed of a root, a theme vowel (/a/ i n 1st conjugation, /e/ i n 2nd conjugation, and / i / i n 3rd conjugation), and the i n f i n i t i v e marker - r . Harris (1969) argues that a theme vowel i s underlyingly present i n the verb paradigms for a l l tenses and aspects, but i f the tense/aspect marker which follows i s v o w e l - i n i t i a l , the theme vowel i s deleted, t h i s assumption produces the following underlying forms for the present i n d i c a t i v e and imperfective of the verbs amar, temer and v i v i r , of the 1st, 2nd and 3rd conjugation classes, respectively (stressed vowels are shown i n c a p i t a l s ) . 192 (3.46) Underlying Representations of Verbal Conjugations Present Indicative Surface Imperfect Surface a. 1st conjugation amar 'to l i k e ' yo am + a + o Amo am + a + a + amAba tu am + a + s Amas am + a + a + s amAbas e l am + a + <i> Ama am + a + a + amAba nosotros am + a + mos amAmos am + a + a + mos amAbamos e l l o s am + a + n Aman am + a + a + n amAban b. 2nd conjugation temer 'to fear' yo tem + e + o tEmo tem + e + a + temla til tern + e + s tEmes tem + e + a + s temlas e l tem + e + 0 tEme tem + e + a + <t> temla nosotros tem + e + mos temEmos tem + e + a + mos temlamos e l l o s tem + e + n tEmen tem + e + a + n temlan c. 3rd conjugation v i v i r 'to l i v e ' yo v i v + i + o vivo viv + i + a + 0 v i v l a tu v i v + i + s vlves v i v + i + a + s v i v l a s e l v i v + i + </> vlve viv + i + a + <t> v i v l a nosotros v i v + i + mos vivlmos viv + i + a + mos vivlamos e l l o s v i v + i + n vlven v i v + i + a + n v i v l a n The person endings for the present i n d i c a t i v e of a l l three conjugation classes are underlyingly i d e n t i c a l , however, we must assume that a /b/ i s inserted between the theme vowel 193 and the person/number marker i n 1st conjugation forms. I w i l l not attempt to motivate t h i s i n s e r t i o n rule here. I adopt t h i s p a r t i c u l a r analysis of verbs, and assume, following Harris (1969), that a rule of Vowel Deletion accounts for the deletion of the theme vowel i n the '1st person singular present i n d i c a t i v e ' forms. The theme vowel i s l e f t i n t a c t i n a l l other persons shown i n (3.46). Vowel Deletion w i l l delink the leftmost of two adjacent vowels brought together by morphological concatenation, i f that vowel i s not stressed. The statement of t h i s r u l e i s c r u c i a l l y concerned with stress, as the stressed theme vowel i n the imperfective forms i s not deleted, while the unstressed theme vowel i n the 1st person singular i n d i c a t i v e i s . In analyzing the stress patterns of Spanish verbs, I assume, as do Halle, Harris and Vergnaud (1991), that the theme vowel i s a c y c l i c a f f i x , while the imperfective and number/person markers shown i n (3.46) are non-cyclic a f f i x e s . In addition, I assume that the number/person markers are a f f i x e s which are exceptions with regard to (3.42b). Verb stems which contain consonant-final number/person markers w i l l then generally demonstrate penultimate stress, rather than the more unmarked ultimate pattern, since the f i n a l v i s i b l e s y l l a b l e w i l l not be marked with an accent and stress w i l l be assigned v i a (3.42c). Stress w i l l then be assigned i n the 1st person singular i n d i c a t i v e form tEmo as shown i n (3.47). 194 (3.47) a. x x x + x + x t e m e o root theme 1st pers vowel sing ind. Stress rules: b. x x x + x + x t e r n e (o) c. / \ / \ s w X X X + X + X t e r n e (o) d. Vowel Deletion: s I X X X X t e m o This form i s made up of the root and the c y c l i c theme vowel a f f i x /e/, followed by the non-cyclic number/person marker. The f i n a l vowel i s extrametrical by (3.42a), and since the number/person marker i s an exception to (3.42b), a l e f t -headed foot w i l l be erected supplying stress to the root vowel. If Vowel Deletion applies a f t e r stress assignment to 195 delete the unstressed theme vowel, then the surface form tEmo Is explained. Stress i s assigned i n the 1st person p l u r a l i n d i c a t i v e form temEmos as shown i n (3.48). . (3.48) a. Morphology x x x + x + X X X t e m e m o s root theme 1st pers vowel p i . ind. Stress rules b. / \ / s X X X + X + t e m e w x x x m o s This word i s consonant-final, so (3.42a) w i l l not apply. Since the number/person marker i s exceptional with regard to (3.42b), a left-headed foot w i l l be erected over the f i n a l two vowels, giving the theme vowel primary stress. If a rul e of Vowel Deletion i s prohibited from applying to stressed vowels, t h i s w i l l explain why the theme vowel remains i n t h i s form. In the imperfective forms shown i n (3.46) we see that 196 stress i s always fixed on the theme vowel. Halle, Harris and Vergnaud (1991) discuss only the 3rd person singular imperfective forms of 1st conjugation verbs, and they assume that the f i n a l vowel ( f i n a l [a] i n amAba) i s extrametrical (by (3.42a)), accounting for the penultimate stress pattern. In contrast, I assume that these forms can be explained by reference to the fact that the number/person markers i n general are exceptional with regard to (3.42b). If we assume that a l l imperfect forms also have a number/person a f f i x following the imperfective marker (even i f the number/person marker i s phonologically empty, as i t i s i n the 1st and 3rd person s i n g u l a r ) , then we can explain the stress patterns i n the majority of imperfective forms. This analysis can then account for imperfective verb forms i n a l l three conjugation classes, whereas the Halle, Harris and Vergnaud account i s only able to account for c e r t a i n forms i n the 1st conjugation c l a s s . My account i s more consistent with the facts of Spanish, i n that i t assumes that imperfectives, l i k e other verbal paradigms, always contain person/number markers. The 1st person singular imperfective v i v l a can be explained as shown i n (3.49). 197 (3.49) a. Morphology: X X X + X + X + 0 v i v i a root theme imp. 1st pers vowel sg. ind. b. Stress rules: A / \ s w X X X + X + X + <j> v i v i a This form i s made up of a root, a theme vowel ( / i / ) , the imperfective marker (/a/) and the empty 1st person singular marker 2 0. The number/person marker i s exceptional to (3.42b), so the f i n a l vowel i n the form i s not accented, and a l e f t -headed foot i s erected over the f i n a l two vowels, s t r e s s i n g the theme vowel. The theme vowel w i l l then not be e l i g i b l e for Vowel Deletion. The stress pattern of the 1st person p l u r a l imperfectives (e.g. vivlamos) i s s t i l l r e c a l c i t r a n t given t h i s analysis, as i t i s i n the Halle, Harris and Vergnaud anal y s i s . I w i l l not attempt to develop an analysis of these forms here, although i t i s expected that one can be provided. Vowel Deletion (which I c a l l Vowel Delinking) can now be 198 stated as i n (3.50). (3.50) Vowel Delinking Delink: s k e l e t a l s l o t from the nucleus Target Condition: the leftmost unstressed N of two adjacent N*s Domain: l e x i c a l ( c y c l i c ) This rule delinks the s k e l e t a l s l o t of the leftmost vowel from a nucleus when two vowels become immediately adjacent due to morphological concatenation. It must be a c y c l i c l e x i c a l r u l e because i t never deletes a vowel i n t r a -morphemically. The operation of (3.50) i n the present i n d i c a t i v e and imperfective forms vivo and v i v l a are compared i n (3.51). 199 (3.51) a. N X X X N I X I I [+hi] [+hi] N x [+bk] o ] [ [ [ v i v ] i ] root theme 1st pers vowel sg. ind. Present Indicative [vivo] b. s N N I I x I I [+hi] [+hi] N X X X I [+bk] [[[ v i v ] i ] a ] <p] root theme imp. 1st pers vowel sg. Imperfect [v i v l a ] 200 In (3.51a) both the theme vowel and '1st person i n d i c a t i v e ' have been added to the root v i v . In (3.51b) the root i s followed by the theme vowel, the imperfective marker and the 1st person singular marker (which i s <f>). S y l l a b i f i c a t i o n i n Spanish w i l l be ordered p r i o r to the other phonological rules that are to be discussed, since these rules a l l have target conditions which specify s y l l a b i f i c a t i o n or stress requirements. Stress i s assigned to the root vowel i n (3.51a) (see (3.47)) and to the theme vowel i n (3.51b) (see (3.49)). Only the theme vowel i n (3.51a) w i l l therefore be a target of Vowel Delinking, since the leftmost of the two adjacent nuclei i n (3.51b) i s stressed. Vowel Delinking w i l l be discussed further with regard to the high/mid vo c a l i c alternations examined i n 3<~2«2o3» 3.2.2.2 Alternating Vowel/Diphthongs In c e r t a i n 1st, 2nd or 3rd conjugation verbs an unstressed mid or high vowel ( [ i ] , [ e ] , [ o ] or [u]) varies with a stressed f a l l i n g diphthong (shown here as phonetic [ye] for orthographic i e and [we] for orthographic u e ) 2 1 . The second member of such as diphthong i s always [e]. 201 (3.52) Alternating vowel/diphthongs a. Front vowels t[e]ndEmos t[yE]nden adqu[i]riO n[e]gO b. Back vowels c[o]ntO c[wE]nto j[u]gAmos j[wE]ga 'we/they tend' - 2nd conj. adqu[yE]re *he acquired/he acquires' - 3rd conj n[y E]°J a 'he denied/he denies* - 1st conj. •he told/I t e l l * - 1st conj. 'we play/he plays' - 1st conj. The most common alternations are those found i n forms such as tendemos/tienden and conto/cuento, where a f a l l i n g diphthong alternates with a mid front or back vowel. Alternations between a f a l l i n g diphthong and a high front or back vowel, such as those found i n juqamos/jueqa and adquirio/adquiere, are r e l a t i v e l y rare. In some 3rd conjugation verbs there i s a ternary a l t e r n a t i o n pattern between a high vowel, a mid vowel, and a f a l l i n g diphthong. This pattern i s shown i n (3.53) with the verb mentir. (3.53) "to l i e " I n f i n i t i v e mentir Past P a r t i c i p l e mentldo Gerund mintiEndo Present Indicative Present Subjunctive miEnto mentlmos miEnta mintAmos miEntes mentis miEntas mintAis miEnte miEnten miEnta miEntan 202 The root diphthong always appears under stress, but e i t h e r the high or mid vowel may appear i n unstressed positions — the high vowel when followed by a diphthong or by a morpheme beginning with [a], and the mid vowel when followed by [ i ] . These alternations have been discussed at great length i n the Spanish l i t e r a t u r e . The vowel underlying t h i s a l t e r n a t i o n has been described as a simple vowel /e/ which undergoes diphthongization under c e r t a i n conditions (Harris 1969, Brame and Bordelois 1973), as an underlying diphthong / i e / (or /ue/) that i s sometimes changed to a monophthong (Norman and Sanders 1977), or as a complex l e x i c a l entry which includes both a simple vowel and a diphthong (Hooper 1976). I assume; following Harris (1985), that a l t e r n a t i n g vowel/diphthongs are underlyingly represented as two s k e l e t a l s l o t s , the f i r s t of which may or may not have associated v o c a l i c features, and the second of which i s t o t a l l y unspecified. The underlying representation of the roots 'deny' and 'play' from (3.52) are compared to the representations of the roots peg 'beat' and mont 'climb' of the non-alternating class i n (3.54). 203 (3.54) a. Alternating vowel/diphthong forms X X X X X X X X I I I I I I n e g j u g b. Non-alternating forms X X X X X X X I I I I I I I p e g m o n t In both neg/nieq and juq/jueq the root vowel i s represented as two s k e l e t a l s l o t s , while the root vowels i n the non-al t e r n a t i n g forms are represented as single s k e l e t a l s l o t s . In o u t l i n i n g how s y l l a b i f i c a t i o n of the underlying forms i n (3.54) w i l l proceed, I adopt the s y l l a b i f i c a t i o n algorithm of Levin (1985) discussed e a r l i e r i n 2.1.2.4. The universal components of t h i s theory are repeated here as (3.55). 204 X-bar theory i . Categorial Component a. N-Placement b. Complex-N i i . Projection a. Project N" b. Project N' i i i . Incorporation a. Incorporate into N" b. Incorporate into N' i v . Adjunction (to N") Condition on Structure-Dependent Rules Sonority Hierarchy Vowels project N (by N-placement); N" (onset position) i s projected by picking up the segment immediately to the l e f t of N (by Project N"); and N' i s projected to pick up remaining post-nuclear elements (by Project N'). In Levin's framework s y l l a b i c i t y i s derived through the operation of redundancy or phonological rules, or i t may be s p e c i f i e d l e x i c a l l y . In Spanish, the only segments that may be s y l l a b i c are the vowels shown i n (3.40) and diphthongs composed of combinations of these vowels, and therefore s y l l a b i c i t y i s a redundant phenomenon i n the language. I assume that the redundancy rule projecting N i n Spanish i s as shown i n (3.56). (3.55) A. B. C, 205 (3.56) N-Placement - Spanish N I x > x I I [-cons] [-cons] If consonants are underlyingly marked [+cons], then vowels w i l l be underlyingly unspecified for the feature [cons]. Since (3.56) makes reference to the feature value [-cons], the redundancy rule i n s e r t i n g [-cons] on vowels w i l l be inserted p r i o r to application of (3.56) by the RROC. The fact that Spanish permits diphthongs demonstrates that i t has chosen the marked parameter s e t t i n g permitting complex nuclei (Complex-N i n (3.55)). As discussed i n 3.2.1 I assume that the only type of complex N allowed i n Spanish i s a f a l l i n g diphthong, and t h i s p a r t i c u l a r type of diphthong only appears morpheme-internally. A l l other sequences of vowels w i l l be analyzed as complex rhymes or two adjacent s y l l a b l e heads. Spanish requires the language-specific statement i n (3.57) to indicate when a complex N can be formed. 206 (3.57) Complex N Formation When two sk e l e t a l s l o t s (X) marked as [-cons] are adjacent: If X 2 i s equal to or more sonorous than X x then X i X 2 i s s y l l a b i f i e d as a branching N. The rightmost member i s designated as the head of N 2 2. Since t h i s rule again makes reference to the feature [-cons], the redundancy rule i n s e r t i n g that feature value w i l l be ordered p r i o r to t h i s process, by the RROC. (3.57) w i l l j o i n two adjacent s k e l e t a l s l o t s into a branching N, provided that the second s l o t i s at least as sonorous as the f i r s t . N Placement and Complex N Formation cannot destroy previously erected s y l l a b l e structures i n Spanish, which accounts for the fact that complex nuclei a r i s e only within morphemes i n t h i s language. A theme vowel followed by a v o w e l - i n i t i a l s u f f i x i s always r e a l i z e d as two d i s t i n c t s y l l a b l e s , suggesting that N-Placement and not Complex-N Formation i s responsible for the placement of the n u c l e i . If Complex N Formation were able to create new structure, then we would expect that these vowel sequences could also become diphthongs -- a r e s u l t that i s not borne out i n the data. (3.55C) i s the sonority hierarchy, which helps to determine how s y l l a b i f i c a t i o n proceeds. Levin, following Steriade (1982), assumes that the hierarchy of features which p a r t i c i p a t e i n sonority are fixed u n i v e r s a l l y , as i s the sonority difference between the two values of any feature. 207 The sonority properties of languages d i f f e r only i n the i n c l u s i o n of features i n the language-particular sonority scale, and i n the Minimal Sonority Distance required between segments i n c l u s t e r s . Levin (1985: 77-78) discusses how a theory of underspecification, where feature values are not f u l l y s p e c i f i e d , can be reconciled with a sonority scale such as that proposed i n Steriade (1982). Levin suggests that feature combinations be matched to positions i n the sonority hierarchy only i f they have i d e n t i c a l feature s p e c i f i c a t i o n s . If features or feature values change during the course of the derivation then they w i l l automatically be reassessed against the sonority scale. Following Levin's p r i n c i p l e s , a sonority scale for Spanish vowels i s given i n (3.58). (3.58) Spanish Sonority Scale - Vowels 1. [-cons], [+high] 2. [-cons] 3. [-cons], [+low] The l e a s t sonorous vowels are those s p e c i f i e d as [-cons] and [+high] and the most sonorous are those s p e c i f i e d as [-cons] and [+low]. A vowel s p e c i f i e d only as [-cons] (by the context-free redundancy rule) w i l l be more sonorous than a [+high] vowel, but less sonorous than a [+low] vowel. Only the features [cons], [low] and [high] are relevant to t h i s scale — other feature s p e c i f i c a t i o n s w i l l be ignored i n the 208 matching algorithm. Given two morpheme-internal vowel sequences / i e / and / a i / we can see the e f f e c t s of (3.57) and (3.58) as follows: (3.59) a. x x b. x x I I [-cons] [-cons] I [+lo] [+hi] /a/ / i / [-cons] [-cons] [+hi] / i / /e/ Pos i t i o n i n sonority scale 1 2 3 1 S y l l a b i f i c a t i o n N N N / \ I I X X X X I I I I [-cons] [-cons] [-cons] [-cons] [+lo] [+hi] [+hi] I assume that underlyingly vowels are not s p e c i f i e d for the feature [cons]; however, [-cons] i s added to the representations by a context-free redundancy ru l e immediately p r i o r to N-Placement and Complex N Formation. Each vowel w i l l be matched against the sonority scale i n (3.55), which 209 w i l l specify that / i / i s less sonorous than /e/, while /a/ i s more sonorous than / i / . Given the r e s t r i c t i o n s on sonority i n (3.58) only the vowel sequence i n (3.59a) w i l l be s y l l a b i f i e d as a complex N. In (3.59b) both vowels w i l l be s y l l a b i f i e d separately, and w i l l most l i k e l y surface as two d i s t i n c t s y l l a b l e s . The derivation of the forms juqamos and nieqa from (3.51) u n t i l the assignment of stress are shown i n (3.60) 2 3; (3.60) [ 1 — > [-cons] n g a X X X X + X I I I [-C][-C] [-C] 3 u g X X X X I I [-CH-C] I [+hi] [+bk] a • x + I [-C] m o s x x x I [-c] [+bk] [+lo] [+lo] Project N Complex N Formation N N / I I n g a x x x x x I I I [-C][-C] [-C] x N / I u X X I I g X N I a x I m x [ -c][-c] [-c] [+hi] i N I o x I [-C] [+bk] [+bk] [+lo] [+lo] 210 Project N" N" N" N" N" N" N N / I I n g a x x x x x I I I [-CJ[-C] [-C] N N N / I I I j u g a m o s X X X x x x x x I I I [-C][-C] [-C] I [+hi] [+bk] I [+bk] [+lo] [+lo] Project N' N" N" N" n/a / I I I N I N' I N j u g a m o s x x x x x x x x I I I [-CH-C] [-C] [+hi] I [+bk] [+lo] [-C] [+bk] 211 Stress Assignment s N" N" I I N N / I I n g a X X X X X I I I [-C][-C] [-C] N" s I N' I I N / I N I N" I N' I N I [+lo] j u g a m o s x x x x x x x x I I I I [-C][-C] [-C] [-C] I [+hi] I [+bk] [+bk] [+lo] The root vowels are underlyingly represented with two sk e l e t a l s l o t s , i n d i c a t i n g that these vowels are of the alte r n a t i n g type. In my analysis of these facts the i n i t i a l s l o t i n the underlying diphthong i n juqamos w i l l be s p e c i f i e d as [+high] and [+back], while the second s l o t w i l l be t o t a l l y unspecified. The [+high] s p e c i f i c a t i o n of the i n i t i a l member of the diphthong i s required since t h i s vowel surfaces as a high vowel when i t i s reduced to a monophthong (juqamos). The i n i t i a l s l o t i n the diphthong i n nieqa w i l l not be s p e c i f i e d as [+high] since t h i s vowel i s r e a l i z e d as a mid vowel when i t i s reduced to a monophthong (nego). The second 212 member of t h i s diphthong w i l l also have no underlying s p e c i f i c a t i o n s . Complex N Formation w i l l apply to create a branching N over these vowels, while a single N w i l l be placed over the remaining vowels. Project N" and Project N' w i l l then apply to s y l l a b i f y the remaining segments. Stress i s assigned to the penultimate s y l l a b l e i n both forms: i n both cases because the number/person markers do not tri g g e r (3.42b). Harris (1985) posits three rules to account for the surface r e a l i z a t i o n of the alternating vowel/diphthongs shown i n (3.54). Diphthongization i s a rule which adds the feature [-cons] to a s k e l e t a l s l o t with no associated f e a t u r a l material i n a branching rhyme of a stressed s y l l a b l e . Default i s a rule (or rules) which adds the default features to a vowel s p e c i f i e d as [-cons] through Diphthongization. The f i n a l r u l e , High-Glide Formation, inserts [+high] on the non-head member of a branching rhyme, making the i n i t i a l member a g l i d e . I adopt Harris' analysis i n s p i r i t , although my formulation of these rules i s s l i g h t l y d i f f e r e n t . Given that I have already assumed that a default rule f i l l s i n the feature [-consonantal] before N-Placement, there i s no reason to p o s i t an additional rule of Diphthongization. The rule of High-Glide Formation w i l l supply the [+high] feature value for the i n i t i a l member of a stressed diphthong, such as that i n nieqa. 213 (3.61) High-Glide Formation Insert [+high] Target condition: non-head member of a branching stressed N Domain: p o s t c y c l i c High-Glide Formation must apply a f t e r stress i s assigned and since stress i s both c y c l i c and p o s t c y c l i c I assume (3.61) belongs to the po s t c y c l i c component. Continuing the derivation from (3.60), I show the operation of High-Glide Formation i n the forms nieqa and juqamos (nuclear s l o t s are underl