🔙 Quay lại trang tải sách pdf ebook An Introduction to English Sentence Structure Ebooks Nhóm Zalo AN INTRODUCTION TO ENGLISH SENTENCE STRUCTURE ANDREW RADFORD CAMBRIDGE CAMBRIDGE www.cambridge.org/9780521516936 This page intentionally left blank An Introduction to English Sentence Structure This outstanding resource for students offers a step-by-step, practical introduction to English syntax and syntactic principles, as developed by Chomsky over the past 15 years. Assuming little or no prior background in syntax, Andrew Radford outlines the core concepts and how they can be used to describe various aspects of English sentence structure. This is an abridged version of Radford’s major new textbook Analysing English Sentences (also published by Cambridge University Press), and will be welcomed as a handy introduction to current syntactic theory. ANDREw RADFORD is Professor & Head of the Department of Language and Linguistics at the University of Essex. His recent publications include Minimalist Syntax: Exploring the Structure of English (Cambridge, 2004) and English Syntax: An Introduction (Cambridge, 2004). An Introduction to English Sentence Structure ●ANDREW RADFORD University of Essex CAMBRIDGE UNIVERSITY PRESS Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo Cambridge University Press The Edinburgh Building, Cambridge CB2 8RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521516938 © Andrew Radford 2009 This publication is in copyright. Subject to statutory exception and to the provision of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published in print format 2009 ISBN-13 978-0-511-50666-6 eBook (EBL) ISBN-13 978-0-521-51693-8 hardback ISBN-13 978-0-521-73190-4 paperback Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate. Contents Preface page viii 1 Grammar 1 1.1 Overview 1 1.2 Traditional grammar: Categories and functions 1 1.3 Universal Grammar 11 1.4 The Language Faculty 15 1.5 Principles of Universal Grammar 19 1.6 Parameters 22 1.7 Parameter-setting 26 1.8 Summary 30 1.9 Bibliographical background 32 Workbook section 33 2 Structure 39 2.1 Overview 39 2.2 Phrases 39 2.3 Clauses 44 2.4 Clauses containing complementisers 49 2.5 Testing structure 51 2.6 Structural relations and the syntax of polarity items 58 2.7 The c-command condition on binding 62 2.8 Bare phrase structure 64 2.9 Summary 66 2.10 Bibliographical background 69 Workbook section 70 3 Null constituents 81 3.1 Overview 81 3.2 Null subjects 81 3.3 Null auxiliaries 86 3.4 Null T in finite clauses 89 3.5 Null T in infinitive clauses 94 3.6 Null C in finite clauses 96 3.7 Null C in infinitive clauses 101 3.8 Defective clauses 105 3.9 Null determiners and quantifiers 108 3.10 Summary 111 v vi CONTENTS 3.11 Bibliographical background 113 Workbook section 114 4 Head movement 120 4.1 Overview 120 4.2 T-to-C movement 120 4.3 Movement as copying and deletion 123 4.4 V-to-T movement 128 4.5 Head movement 132 4.6 Auxiliary Raising 134 4.7 Another look at negation 137 4.8 DO-support 140 4.9 Summary 144 4.10 Bibliographical background 146 Workbook section 147 5 Wh-movement 152 5.1 Overview 152 5.2 Wh-questions 152 5.3 Wh-movement as copying and deletion 155 5.4 Driving wh-movement and auxiliary inversion 161 5.5 Pied-piping of material in the domain of a wh-word 165 5.6 Pied-piping of a superordinate preposition 171 5.7 Long-distance wh-movement 174 5.8 Multiple wh-questions 182 5.9 Summary 185 5.10 Bibliographical background 188 Workbook section 189 6 A-movement 196 6.1 Overview 196 6.2 Subjects in Belfast English 196 6.3 Idioms 199 6.4 Argument structure and theta-roles 201 6.5 Unaccusative predicates 205 6.6 Passive predicates 211 6.7 Long-distance passivisation 215 6.8 Raising 219 6.9 Comparing raising and control predicates 221 6.10 Summary 227 6.11 Bibliographical background 229 Workbook section 230 7 Agreement, case and A-movement 237 7.1 Overview 237 7.2 Agreement 237 7.3 Feature Valuation 240 7.4 Uninterpretable features and Feature Deletion 242 CONTENTS vii 7.5 Expletive it subjects 246 7.6 Expletive there subjects 251 7.7 Agreement and A-movement 258 7.8 EPP and agreement in control infinitives 261 7.9 EPP and person agreement in defective clauses 262 7.10 Defective clauses with expletive subjects 267 7.11 Summary 272 7.12 Bibliographical background 274 Workbook section 275 8 Split projections 279 8.1 Overview 279 8.2 Split CP: Force, Topic and Focus projections 279 8.3 Split TP: Aspect and Mood projections 287 8.4 Split VP: Transitive ergative structures 292 8.5 Split VP: Other transitive structures 298 8.6 Split VP: Unaccusative structures 304 8.7 Split VP: Passive and raising structures 310 8.8 Summary 313 8.9 Bibliographical background 316 Workbook section 317 9 Phases 323 9.1 Overview 323 9.2 Phases 323 9.3 Intransitive and defective clauses 327 9.4 Phases and A-bar movement 330 9.5 A-bar movement in transitive clauses 334 9.6 Uninterpretable features and feature inheritance 340 9.7 Independent probes 346 9.8 Subject questions 355 9.9 More on subextraction 359 9.10 Summary 362 9.11 Bibliographical background 363 Workbook section 364 Glossary and list of abbreviations 370 References 410 Index 435 Preface Aims This book supercedes my English Syntax book, published in 2004. Although there is much in common between the two books, it should be noted that this book contains new material and new analyses (particularly in later chapters). It has two main aims. The first is to provide an intensive introduction to recent work in syntactic theory (more particularly to how the syntactic component operates within the model of grammar assumed in recent work within the framework of Chomsky’s Minimalist Program). The second is to provide a description of a range of phenomena in English syntax, making use of Minimalist concepts and assumptions wherever possible. Key features The book is intended to be suitable both for people with only minimal grammatical knowledge, and for those who have already done quite a bit of syntax but want to know something (more) about Minimalism. It is not historicist or comparative in orientation, and does not presuppose knowledge of earlier or alternative models of grammar. It is written in an approachable style, avoiding unnecessary complexity and unexplained jargon. Each chapter contains: r a core text (divided up into eight sections or so) focusing on a specific topic r a summary recapitulating the main points in the chapter r a list of key concepts/principles introduced in the chapter r a bibliographical section providing extensive references to original source material r a workbook section containing two different kinds of exercise r a set of model answers accompanying the exercises, together with extensive helpful hints designed to eliminate common errors students make and to help students whose native language is not English r an extensive glossary and integral list of abbreviations The bibliographical background section often contains references to primary research works which are highly technical in nature, and so it would not be viii PREFACE ix appropriate for students to tackle them until they have read the whole book: they are intended to provide a useful source of bibliographical information for extended essays or research projects in particular areas, rather than being essential back-up reading: indeed, the exercises in the book are designed in such a way that they can be tackled on the basis of the coursebook material alone. The glossary at the end of the book provides simple illustrations of how key technical terms are used (both theory-specific terms like EPP and traditional terms like subject): technical terms are written in bold print when they are mentioned for the first time in the main text (italics being used for highlighting particular expressions – e.g. a key word appearing in an example sentence). The glossary also contains an integrated list of abbreviations. The book is intensive and progressive in nature, which means that it starts at an elementary level but gets progressively harder as you delve further into the book. A group of students I taught an earlier version of the book to gave the following degree-of-difficulty score to each chapter on a 5-point scale ranging from 1 = very easy to 5 = very hard: ch.1 = 1.7; ch.2 = 2.2; ch.3 = 2.7; ch.4 = 2.9; ch.5 = 3.2; ch.6 = 3.4; ch.7 = 3.7; ch.8 = 4.2; ch.9 = 4.4. Successive chapters become cumulatively more complex, in that each chapter presupposes material covered in previous chapters as well as introducing new material: hence it is helpful to go back and read material from earlier chapters every so often. In some cases, analyses presented in earlier chapters are subse- quently refined or revised in the light of new assumptions made in later chapters. Teaching materials For teachers adopting the book, I have developed a series of web materials (in the form of Powerpoint transparencies) designed to provide two hours’ worth of teaching material for each chapter. The relevant materials present detailed step- by-step analyses of those exercise examples which have the symbol (w) after them in the coursebook. They can be accessed at www.cambridge.org/radford Companion volume This book is being produced in parallel with a longer version entitled Analysing English Sentences: A Minimalist Approach. In this shorter version, the main text (particularly in the later chapters) is generally about a third shorter than the main text in the longer version (with the exception of chapters 1 and 6). This shorter version is aimed primarily at students whose native language is not English, and who are taking (English) syntax as a minor rather than a major course. The two books have an essentially parallel organisation into chapters and sections (though additional sections, technical discussion and bibliographial references are included in the longer version), and contain much the same exercise material. x PREFACE In keeping the two books parallel in structure and organisation as far as possible, I am mindful of the comment made in a review of two earlier books which I produced in parallel longer and shorter versions (Radford 1997a and Radford 1997b) that some readers may wish to read the short version of a given chapter first, and then look at the longer version afterwards, and that this is ‘not facilitated’ if there is ‘an annoyingly large number of non-correspondences’ between the two (Ten Hacken 2001, p. 2). Accordingly, I have tried to maximise correspondence between the ‘long’ and ‘short’ versions of these two new books. Acknowledgments I am grateful to Neil Smith (of University College London) for his forebearance in patiently wading through an earlier draft of the manuscript and pointing out some of the imperfections in it, while managing to make his comments challenging and good-humoured at the same time. Thanks also go to my Essex colleague Bob Borsley for helpful comments, and to Mich`ele Vincent for preparing the index. Dedication This book is dedicated to my long-suffering wife Khadija (who has had to put up with extended periods of authorial autism) and to her family, who have always spoiled me shamefully (and done their best to indulge my every whim) whenever we visit Morocco. 1 Grammar ● ● 1.1 Overview In broad terms, this book is concerned with aspects of grammar. Gram- mar is traditionally subdivided into two different but interrelated areas of study – morphology and syntax. Morphology is the study of how words are formed out of smaller units (called morphemes), and so addresses questions such as ‘What are the component morphemes of a word like antidisestablishmentarianism, and what is the nature of the morphological operations by which they are combined together to form the overall word?’ Syntax is the study of the way in which phrases and sentences are structured out of words, and so addresses questions like ‘What is the structure of a sentence like What’s the president doing? and what is the nature of the grammatical operations by which its component words are combined together to form the overall sentence structure?’ In this chapter, we begin (in §1.2) by taking a brief look at the approach to the study of syntax taken in traditional grammar: this also provides an opportunity to introduce some useful grammatical terminology. In the remainder of the chapter, we look at the approach to syntax adopted within the theory of Universal Grammar developed by Chomsky. ● ● 1.2 Traditional grammar: Categories and functions Within traditional grammar, the syntax of a language is described in terms of a taxonomy (i.e. classificatory list) of the range of different types of syntactic structures found in the language. The central assumption underpinning syntactic analysis in traditional grammar is that phrases and sentences are built up of a series of constituents (i.e. syntactic units), each of which belongs to a specific grammatical category and serves a specific grammatical function. Given this assumption, the task of the linguist in analysing the syntactic struc- ture of any given type of sentence is to identify each of the constituents in the sentence, and (for each constituent) to say what category it belongs to and what function it serves. For example, in relation to the syntax of a simple sentence like: (1) Students protested 1 2 i GRAMMAR it would traditionally be said that the sentence consists of two constituents (the word students and the word protested), that each of these constituents belongs to a specific grammatical category (students being a plural noun and protested a past tense verb) and that each serves a specific grammatical function (students being the subject of the sentence, and protested being the predicate). The overall sentence Students protested has the categorial status of a clause which is finite in nature (by virtue of denoting an event taking place at a specific time), and has the semantic function of expressing a proposition which is declarative in force (in that it is used to make a statement rather than e.g. ask a question). Accordingly, a traditional grammar of English would tell us that the simplest type of finite declarative clause found in English is a sentence like (1), in which a nominal subject is followed by a verbal predicate. Let’s briefly look at some of the terminology used here. In traditional grammar, words are assigned to grammatical categories (called parts of speech) on the basis of their semantic properties (i.e. meaning), mor- phological properties (i.e. the range of different forms they have) and syn- tactic properties (i.e. word-order properties relating to the positions they can occupy within sentences): a set of words which belong to the same category thus have a number of semantic, morphological and syntactic properties in common. There are traditionally said to be two different types of word, namely content words/contentives (= words which have substantive lexical content) on the one hand, and function words/functors (= words which essentially serve to mark grammatical properties) on the other. The differences between the two can be illustrated by comparing a contentive like car with a functor like they. A noun like car has substantive lexical content in that it denotes an object which typically has four wheels and an engine, and it would be easy enough to draw a picture of a typical car; by contrast, a pronoun such as they has no descriptive content (e.g. you can’t draw a picture of they), but rather is a functor which simply marks grammatical (more specifically, person, number and case) properties in that it is a third person plural nominative pronoun. Because they have lexical semantic content, content words often (though not always) have antonyms (i.e. ‘oppo- sites’) – e.g. the adjective tall has the antonym short, the verb increase has the antonym decrease, and the preposition inside has the antonym outside: by con- trast, a typical function word like e.g. the pronoun me has no obvious antonym. Corresponding to these two different types of (content and function) word are two different kinds of grammatical category – namely lexical/substantive cat- egories (= categories whose members are content words) on the one hand, and functional categories (= categories whose members are function words) on the other. Let’s begin by looking at the main lexical/substantive categories found in English – namelynoun, verb, adjective, adverbandpreposition(conventionally abbreviated to N, V, A, ADV and P in order to save space). Nouns (= N) are traditionally said to have the semantic property that they denote entities: so, bottle is a noun (since it denotes a type of object used to contain liquids), 1.2 Traditional grammar: Categories and functions 3 water is a noun (since it denotes a type of liquid) and John is a noun (since it denotes a specific person). There are a number of distinct subtypes of noun: for example, a noun like chair is a count noun in that it can be counted (cf. one chair, two chairs . . .), whereas a noun like furniture is a mass noun in that it denotes an uncountable mass (hence the ungrammaticality of ∗one furniture, ∗two furnitures – a prefixed star/asterisk being used to indicate that an expression is ungrammatical). Likewise, a distinction is traditionally drawn between acommon noun like boy (which can be modified by a determiner like the – as in The boy is lying) and a proper noun like Andrew (which cannot be used in the same way in English, as we see from the ungrammaticality of ∗The Andrew is lying). Count nouns generally have the morphological property that they have two different forms: a singular form (like horse in one horse) used to denote a single entity, and a plural form (like horses in two horses) used to denote more than one entity. Common nouns have the syntactic property that only (an appropriate kind of) noun can be used to end a sentence such as They have no . . . In place of the dots here we could insert a singular count noun like car, or a plural count noun like friends or a mass noun like money, but not other types of word (e.g. not see or slowly or up, as these are not nouns). A second lexical/substantive category is that of verb (= V). These are tradi- tionally said to have the semantic property that they denote actions or events: so, eat, sing, pull and resign are all (action-denoting) verbs. From a syntactic point of view, verbs have the property that only an appropriate kind of verb (in its uninflected infinitive form) can be used to complete a sentence such as They/It can . . . So, words like stay, leave, hide, die, starve and cry are all verbs and hence can be used in place of the dots here (but words like apple, under, pink and if aren’t). From a morphological point of view, regular verbs like cry in English have the property that they have four distinct forms: e.g. alongside the bare (i.e. uninflected) form cry we find the present tense form cries, the past tense/perfect participle/passive participle form cried and the progressive participle form crying. (See the Glossary of terminology at the end of this book if you are not familiar with these terms.) A third lexical/substantive category is that of adjective (= A). These are traditionally said to have the semantic property of denoting states or attributes (cf. ill, happy, tired, conscientious, red, cruel, old etc.). They have the syntactic property that they can occur after be to complete a sentence like They may be . . . (as with They may be tired/ill/happy etc.), and the further syntactic property that (if they denote a gradable property which can exist in varying degrees) they can be modified by a degree word like very/rather/somewhat (cf. She is very happy). Many (but not all) adjectives have the morphological property that they have comparative forms ending in -er and superlative forms ending in -est (cf. big/bigger/biggest). A fourth lexical/substantive category is that of adverb (= ADV). These often have the semantic property that they denote the manner in which an action is per- formed (as with well in She sings well). Regular adverbs have the morphological 4 i GRAMMAR property that they are formed from adjectives by the addition of the suffix -ly (so that corresponding to the adjective sad we have the adverb sadly). A syntactic property of adverbs is that an adverb (like e.g. badly) is the only kind of word which could be used to end sentences such as She behaved . . ., He treats her ... or He worded the statement . . . The fifth and final lexical/substantive category found in English is that of preposition (= P). Many of these have the semantic property of marking location (cf. in/on/off/inside/outside/under/above/below). They have the syntactic prop- erty that a preposition (with the appropriate kind of meaning) can be modified by right in the sense of ‘completely’, or by straight in the sense of ‘directly’ (as with the preposition down in He fell right down the stairs and the preposition to in He went straight to bed). Prepositions have the morphological property that they are invariable/uninflected forms (e.g. the preposition off has no past tense form ∗offed, no superlative form ∗offest and so on). In addition to the five lexical/substantive categories identified above, English also has a number of functional categories. One such functional category is that of determiner (= D) – a category whose members are traditionally said to include the definite article the and the demonstrative determiners this/that/these/those. They are called determiners because they have the semantic property that they determine specific semantic properties of the noun expression that they introduce, marking it as a definite referring expression: for example, an expression like the car in a sentence such as Shall we take the car? is a definite referring expression in the sense that it refers to a definite (specific) car which is assumed to be familiar to the hearer/addressee. A related class of words are those which belong to the functional category quantifier (= Q), denoting expressions of quantity, such as some/all/no/any/each/every/most/much/many. (We shall also take the indefinite article a to be a quantifier – one which quantifies over a single entity.) A further type of functional category found in English is that of pronoun (= PRN). Pronouns are items which are said to ‘stand in place of’ (the meaning of the prefix pro-) or ‘refer back to’ noun expressions. However, there are reasons to think that there are a number of different types of pronoun found in English and other languages. For example, in sentences such asJohn has a red car and Jim has a blue one, the word one is traditionally said to be a pronoun because it has no lex- ical semantic content of its own, but rather takes its content from its antecedent (i.e. one refers back to the noun car and so one is interpreted as having the same meaning as car). However, from a morphological perspective, the pronoun one behaves like a regular count noun in that it has a plural form ending in -s (as in I’ll take the green apples if you haven’t got any red ones). So, more accurately, we could say that one is an N-pronoun (or pronominal noun). By contrast, in a sentence like Many miners were rescued, but some died, the word some seems to function as a Q-pronoun (i.e. a pronominal quantifier). And in a sentence like These apples are ripe, but those aren’t, the word those seems to be a D-pronoun (i.e. a pronominal determiner). Indeed, some linguists have argued that so-called personal pronouns like I/me/we/us/you/he/him/she/her/it/they/them are also 1.2 Traditional grammar: Categories and functions 5 D-pronouns: the rationale for this is that some such pronouns can be used as determiners which modify a following noun (as in We republicans don’t trust you democrats, where we could be argued to be a determiner modifying the noun republicans, and you could be seen as a determiner modifying the noun democrats). While, as noted here, pronouns can be argued to belong to a number of distinct types of category, in order to simplify discussion I shall simply refer to them as belonging to the category PRN throughout this book. (Because there are a number of different types of pronoun, some linguists prefer to refer to them by using the more general term proform.) Another type of functional category found in English is that of auxiliary (verb). They have the semantic property of marking grammatical properties such as tense, aspect, voice or mood (see the Glossary of terminology at the end of the book if you are not sure what these terms mean). Auxiliaries have the syntactic property that (unlike lexical/main verbs) they can be inverted with their subject in questions (so that corresponding to a statement like It is raining we have the question Is it raining? where the auxiliary is has moved in front of the subject it and is said to have been inverted). The items italicised in (2) below (in the use illustrated there) are traditionally categorised as auxiliaries taking a [bracketed] complement containing a bold-printed verb: (2) (a) He has/had [gone] (b) She is/was [staying at home] (c) They are/were [taken away for questioning] (d) He really does/did [say a lot] (e) You can/could [help us] (f) They may/might [come back] (g) He will/would [get upset] (h) I shall/should [return] In the uses illustrated here, have/be in (2a,b) are (perfect/progressive) aspect auxiliaries, be in (2c) is a (passive) voice auxiliary, do in (2d) is an expletive or dummy auxiliary (i.e. one with no intrinsic lexical semantic content), and can/could/may/might/will/would/shall/should in (2e–h) are modal auxiliaries. What auxiliaries in sentences like those above have in common is the fact that they inflect for present/past tense. Hence, in work in syntax over the past ten years or so, they have been said to belong to the category T (= tense-marked auxiliary). An interesting word which has been argued to be related to tense-marking auxiliaries in work over the past thirty years or so is the infinitive particle to, in sentences such as: (3) They are now expecting the president to be impeached tomorrow In a sentence like (3), infinitival to seems to have future time-reference (in that the act of impeachment will take place at some time in the future), and this is why we can use the word tomorrow in the to-clause. In this respect, infinitival to seems 6 i GRAMMAR to have much the same function as the auxiliary will in They are now expecting that the president will be impeached tomorrow, suggesting that infinitival to is an infinitival tense marker, and so belongs to the same category T as present/past tense auxiliaries such as is/was. The difference between auxiliaries and infinitival to is that most auxiliaries overtly inflect for present/past tense (though this is not true of the invariable auxiliaries must and ought), whereas infinitival to is invariable in form. We can thus say that an auxiliary like will is a finite T constituent, whereas infinitival to is a nonfinite T. The last type of functional category which we will look at is a kind of word (like each of the words italicised in the examples below) which is traditionally termed a (subordinating) conjunction: (4) (a) I think [that you may be right] (b) I doubt [if you can help me] (c) I’m anxious [for you to receive the best treatment possible] Each of the bracketed clauses in (4) is a complement clause, in that it is the complement of the word immediately preceding it (think/doubt/anxious); for this reason, the italicised word which introduces each clause is known in work since the 1960s as a complementiser (= C), and this is the terminology which will be adopted throughout this book. Complementisers are functors in the sense that they encode particular sets of grammatical properties. For example, complemen- tisers encode (non)finiteness by virtue of the fact that they are intrinsically finite or nonfinite. More specifically, the complementisers that and if are inherently finite in the sense that they can only be used to introduce a finite clause (i.e. a clause containing a present or past tense auxiliary or verb, like the present tense auxiliaries may and can in 4a and 4b); by contrast, for is an inherently infinitival complementiser, and so can be used to introduce a clause containing infinitival to (as in 4c). Moreover, that introduces a declarative clause (i.e. one which has the force of a statement), if introduces an interrogative clause (i.e. one which has the force of a question) and for introduces an irrealis clause (i.e. one relating to a hypothetical event which hasn’t yet taken place and may or may not take place at some stage in the future). Hence, we can say that is a finite declarative complementiser, if is a finite interrogative complementiser and for is an infinitival irrealis complementiser. Using the set of syntactic categories outlined above, we can employ the tra- ditional labelled bracketing technique to categorise words (i.e. assign them to grammatical categories) in a way which describes how they are being used in a particular sentence. Using this technique, the words in sentence (5a) below can be categorised as in (5b): (5) (a) The president is clearly feeling angry that Congress has refused to negotiate with him (b) [D The] [N president] [T is] [ADV clearly] [V feeling] [A angry] [C that] [N Congress] [T has] [V refused] [T to] [V negotiate] [P with] [PRN him] 1.2 Traditional grammar: Categories and functions 7 The labelled bracketing in (5b) tells us that the is a D/determiner, president a N/noun, is a T/present tense auxiliary, clearly an ADV/adverb, feeling a V/verb, angry an A/adjective, that a C/complementiser, Congress a N/noun, has a T/present tense auxiliary, refused a V/verb, to a T/infinitival tense particle, negotiate a V/verb, with a P/preposition and him a PRN/pronoun. The discussion of grammatical categories presented above is merely a brief sketch: however, it suffices to illustrate the point that when traditional grammar- ians analyse the syntax of sentences, they begin by assigning each of the words in the sentence to a grammatical category which describes how it is being used in the sentence concerned. Grammatical differences between individual words belonging to the same category are traditionally described in terms of sets of grammatical features, and these features (by convention) are enclosed in square brackets. For example, both she and us are pronouns, but they differ in that she is a third person pronoun which is feminine in gender, singular in number and nominative in case, whereas us is a first person pronoun which is plural in number and accusative in case. Accordingly, we can describe the differences between these two pronouns by saying that the pronoun she carries the features [third-person, singular-number, feminine-gender, nominative-case], whereas us carries the features [first-person, plural-number, accusative-case]. As noted at the beginning of this section, traditional grammarians are also con- cerned to describe thegrammatical functionswhich words and other expressions fulfil within the sentences containing them. We can illustrate this point in terms of the following set of sentences: (6) (a) John smokes (b) The president smokes (c) The president of Utopia smokes (d) The former president of the island paradise of Utopia smokes Sentence (6a) comprises the noun John which serves the function of being the subject of the sentence (and denotes the person performing the act of smoking), and the verb smokes which serves the function of being the predicate of the sentence (and describes the act being performed). In (6a), the subject is the single noun John; but as the examples in (6b,c,d) show, the subject of a sentence can also be an (italicised) phrase like the president, or the president of Utopia or the former president of the island paradise of Utopia. Now consider the following set of sentences: (7) (a) John smokes cigars (b) John smokes Cuban cigars (c) John smokes Cuban cigars imported from Havana (d) John smokes a specific brand of Cuban cigars imported by a friend of his from Havana Sentence (7a) comprises the subject John, the predicate smokes and the comple- ment (or direct object) cigars. (The complement cigars describes the entity on 8 i GRAMMAR which the act of smoking is being performed; as this example illustrates, subjects normally precede the verb with which they are associated in English, whereas complements typically follow the verb.) The complement in (7a) is the single noun cigars; but a complement can also be a phrase: in (7b), the complement of smokes is the phrase Cuban cigars; in (7c) the complement is the phrase Cuban cigars imported from Havana; and in (7d) the complement is the phrase a specific brand of Cuban cigars imported by a friend of his from Havana. A verb which has a noun or pronoun expression as its direct object complement is traditionally said to be transitive. From a semantic perspective, subjects and complements share in common the fact that they generally represent entities directly involved in the particular action or event described by the predicate: to use the relevant semantic terminology, we can say that subjects and complements are arguments of the predicate with which they are associated. Predicates may have one or more arguments, as we see from sentences such as (8) below, where each of the bracketed nouns is a different argument of the italicised predicate: (8) (a) [John] resigned (b) [John] felt [remorse] (c) [John] sent [Mary] [flowers] A predicate like resign in (8a) which has a single argument is said to function as a one-place predicate (in the relevant use); one like feel in (8b) which has two arguments is a two-place predicate; and one like send in (8c) which has three arguments is a three-place predicate. In addition to predicates and arguments, sentences can also contain adjuncts, as we can illustrate in relation to (9) below: (9) (a) The president smokes a cigar after dinner (b) The president smokes a cigar in his office In both sentences in (9), smokes functions as a two-place predicate whose two arguments are its subject the president and its complement a cigar. But what is the function of the phrase after dinner which also occurs in (9a)? Since after dinner isn’t one of the entities directly involved in the act of smoking (i.e. it isn’t consuming or being consumed), it isn’t an argument of the predicate smoke. On the contrary, after dinner simply serves to provide additional information about the time when the smoking activity takes place. In much the same way, the italicised expression in his office in (9b) provides additional information about the location of the smoking activity. An expression which serves to provide (optional) additional information about the time or place (or manner, or purpose etc.) of an activity or event is said to serve as an adjunct. So, after dinner and in his office in (9a,b) are both adjuncts. So far, all the sentences we have looked at in (6–9) have been simple sentences which contain a single clause. However, alongside these we also find complex sentences which contain more than one clause, like (10) below: 1.2 Traditional grammar: Categories and functions 9 (10) Mary knows John smokes If we take the traditional definition of a clause as a predication structure (more precisely, a structure containing a predicate which has a subject, and which may or may not also contain one or more complements and adjuncts), it follows that since there are two predicates (knows and smokes) in (10), there are correspondingly two clauses – the smokes clause on the one hand, and the knows clause on the other. The smokesclause comprises the subject Johnand the predicate smokes; the knows clause comprises the subject Mary, the predicate knows and the comple- ment John smokes. So, the complement of knows here is itself a clause – namely the clause John smokes. More precisely, the smokes clause is a complement clause (because it serves as the complement of knows), while the knows clause is the main clause (or principal clause or independent clause or root clause). The overall sentence (10) Mary knows John smokes is a complex sentence because it contains more than one clause. In much the same way, (11) below is also a complex sentence: (11) The press clearly think the president deliberately lied to Congress Once again, it comprises two clauses – one containing the predicate think, the other containing the predicate lie. The main clause comprises the subject the press, the adjunct clearly, the predicate think and the complement clause the president deliberately lied to Congress. The complement clause in turn com- prises the subject the president, the adjunct deliberately, the predicate lie and the complement to Congress. As was implicit in our earlier classification of (1) as a finite clause, tra- ditional grammars draw a distinction between finite and nonfinite clauses. In this connection, consider the contrast between the italicised clauses below (all of which function as the complement of an underlined adjective or verb): (12) (a) She was glad that he apologised (b) She demanded that he apologise (c) I can’t imagine him apologising (d) It would be sensible for him to apologise (e) It’s important to know when to apologise The italicised clauses in (12a,b) are finite, and it is characteristic of finite clauses in English that they contain an (auxiliary or main) verb marked for tense/mood, and can have a nominative pronoun like he as their subject. In (12a), the verb apologised is finite by virtue of being inflected for past tense and indicative mood, and by virtue of having a nominative subject (he); in (12b), the verb apol- ogise is finite by virtue of being inflected for subjunctive mood (and perhaps present tense, though this is far from clear), and by virtue of having a nomina- tive subject (he). A clause containing a verb in the indicative mood denotes a real (or realis, to use the relevant grammatical term) event or state occurring at 10 i GRAMMAR a specific point in time; a subjunctive clause by contrast denotes a hypotheti- cal or unreal (= irrealis) event or state which has not yet occurred and which may never occur. In contrast to the italicised clauses in (12a,b), the clauses ital- icised in (12c–e) are nonfinite, in that they contain no verb marked for tense or mood, and do not allow a nominative subject. For example, the verb apologis- ing in (12c) is nonfinite because it is a tenseless and moodless gerund form, and has an accusative subject him. Likewise, the verb apologise in (12d,e) is a tenseless and moodless infinitive form (as we see from the fact that it fol- lows the infinitive particle to), and has an accusative subject him in (12d), and a ‘silent’ (implicit) subject in (12e). (Excluded from our discussion here are gerund structures with genitive subjects like the italicised in ‘I can’t stand his perpetual(ly) whining about syntax’, since these are more nominal than clausal in nature.) As the examples in (12) illustrate, whether or not a clause is finite in turn determines the kind of subject it can have, in that finite clauses can have a nomi- native pronoun like he as their subject, but nonfinite clauses cannot. Accordingly, one way of telling whether a particular clause is finite or not is to see whether it can have a nominative pronoun (like I/we/he/she/they) as its subject. In this connection, consider whether the italicised clauses in the dialogues in (13a,b) below are finite or nonfinite: (13) (a) SPEAkER A: I know you cheat on me SPEAkER B: OK, I admit it. I cheat on you. But not with any of your friends (b) SPEAkER A: I know you cheat on me SPEAkER B: Me cheat on you? No way! I never would! The fact that the italicised clause in speaker B’s reply in (13a) has the nominative subject I suggests that it is finite, and hence that the verb cheat (as used in the italicised sentence in 13a) is a first person singular present tense form. By contrast, the fact that the italicised clause in speaker B’s reply (13b) has the accusative subject me suggests that it is nonfinite, and that the verb cheat (as used in the italicised sentence in 13b) is an infinitive form (and indeed this is clear from sentences like Me be a cheat? No way! where we find the infinitive form be).In addition to being finite or nonfinite, each clause within a sentence has a specific force. In this connection, consider the following simple (single-clause) sentences: (14) (a) He went home (b) Are you feeling OK? (c) You be quiet! (d) What a great idea that is! A sentence like (14a) is traditionally said to be declarative in force, in that it is used to make a statement. (14b) is interrogative in force in that it is used to ask a question. (14c) is imperative in force, by virtue of being used to issue an order 1.3 Universal Grammar 11 or command. (14d) is exclamative in force, in that it is used to exclaim surprise or delight. In complex sentences, each clause has its own force, as we can see in relation to (15) below: (15) (a) He asked where she had gone (b) Did you know that he has retired? (c) Tell her what a great time we had! In (15a), the main (asked) clause is declarative, whereas the complement (gone) clause is interrogative; in (15b) the main (know) clause is interrogative, whereas the complement (retired) clause is declarative; and in (15c), the main (tell) clause is imperative, whereas the complement (had) clause is exclamative. We can summarise this section as follows. From the perspective of tradi- tional grammar, the syntax of a language is described in terms of a taxonomy (i.e. a classificatory list) of the range of different phrase-, clause- and sentence- types found in the language. So, for example, a typical traditional grammar of (say) English will include chapters on the syntax of negatives, interroga- tives, exclamatives, imperatives and so on. The chapter on interrogatives will note (e.g.) that in main-clause questions in English like ‘Is he winning?’ the present tense auxiliary is inverts with (i.e. moves in front of) the subject he, but not in complement clause questions like the if-clause in ‘I wonder if he is winning’, and will typically not be concerned with trying to explain why auxil- iary inversion applies in main clauses but not complement clauses: this reflects the fact that the primary goal of traditional grammar is description rather than explanation. ● ● 1.3 Universal Grammar In contrast to thetaxonomicapproach adopted in traditional grammar, Chomsky takes a cognitive approach to the study of grammar. For Chomsky, the goal of the linguist is to determine what it is that native speakers know about their native language which enables them to speak and understand the language, and how this linguistic knowledge might be represented in the mind/brain: hence, in studying language, we are studying a specific kind of cognition (i.e. human knowledge). In a fairly obvious sense, any native speaker of a language can be said to know the grammar of his or her native language. For example, any native speaker of English can tell you that the negative counterpart of I like syntax is I don’t like syntax, and not e.g. ∗I no like syntax: in other words, native speakers know how to combine words together to form expressions (e.g. negative sentences) in their language. Likewise, any native speaker of English can tell you that a sentence like She loves me more than you is ambiguous and has two interpretations which can be paraphrased as ‘She loves me more than she loves you’ and ‘She loves me more than you love me’: in other words, native speakers also know how to interpret (i.e. assign meaning to) expressions in their language. 12 i GRAMMAR However, it is important to emphasise that this grammatical knowledge of how to form and interpret expressions in your native language is tacit (i.e. subconscious) rather than explicit (i.e. conscious): so, it’s no good asking a native speaker of English a question such as ‘How do you form negative sentences in English?’ since human beings have no conscious awareness of the processes involved in speaking and understanding their native language. To introduce a technical term devised by Chomsky, we can say that native speakers have grammatical competence in their native language: by this, we mean that they have tacit knowledge of the grammar of their language – i.e. of how to form and interpret words, phrases and sentences in the language. In work in the 1960s, Chomsky drew a distinction between competence (the native speaker’s tacit knowledge of his or her language) and performance (what people actually say or understand by what someone else says on a given occa- sion). Competence is ‘the speaker-hearer’s knowledge of his language’, while performance is ‘the actual use of language in concrete situations’ (Chomsky 1965, p. 4). Very often, performance is an imperfect reflection of competence: we all make occasional slips of the tongue, or occasionally misinterpret something which someone else says to us. However, this doesn’t mean that we don’t know our native language or that we don’t have competence in it. Misproductions and misinterpretations are performance errors, attributable to a variety of perfor- mance factors like tiredness, boredom, drunkenness, drugs, external distractions and so forth. A grammar of a language tells you what you need to know in order to have native-like competence in the language (i.e. to be able to speak the lan- guage like a fluent native speaker): hence, it is clear that grammar is concerned with competence rather than performance. This is not to deny the interest of performance as a field of study, but merely to assert that performance is more properly studied within the different – though related – discipline of psycholin- guistics, which studies the psychological processes underlying speech production and comprehension. Thus, when we study the grammatical competence of a native speaker of a language like English we’re studying a cognitive system internalised within the brain/mind of native speakers of English which is the product of a ‘cognitive organ’ which is ‘shared among human beings and in crucial respects unique to them’ (Chomsky 2006, p. 1). In the terminology adopted by Chomsky (1986a, pp.19–56), our ultimate goal in studying competence is to characterise the nature of the internalised linguistic system (or I-language, as Chomsky terms it) which makes native speakers proficient in English. Such an approach has obvious impli- cations for the descriptive linguist who is concerned to develop a grammar of a particular language like English. According to Chomsky (1986a, p. 22) a gram- mar of a language is ‘a theory of the I-language . . . under investigation’. This means that in devising a grammar of English, we are attempting to uncover the internalised linguistic system (= I-language) possessed by native speakers of English – i.e. we are attempting to characterise a mental state (a state of competence, and thus linguistic knowledge). 1.3 Universal Grammar 13 Chomsky’s ultimate goal is to devise a theory of Universal Grammar/UG which generalises from the grammars of particular I-languages to the grammars of all possible natural (i.e. human) I-languages. He defines UG (1986a, p. 23) as ‘the theory of human I-languages . . . that identifies the I-languages that are humanly accessible under normal conditions’. (The expression ‘are humanly accessible’ means ‘can be acquired by human beings’.) In other words, UG is a theory about the nature of possible grammars of human languages: hence, a theory of Universal Grammar answers the question: ‘What are the defining characteristics of the grammars of human I-languages?’ There are a number of criteria of adequacy which a theory of Universal Grammar must satisfy. One such criterion (which is implicit in the use of the term Universal Grammar) is universality, in the sense that a theory of UG must provide us with the tools needed to provide a descriptively adequate grammar for any and every human I-language (i.e. a grammar which correctly describes how to form and interpret expressions in the relevant language). After all, a theory of UG would be of little interest if it enabled us to describe the grammar of English and French, but not that of Swahili or Chinese. However, since the ultimate goal of any theory is explanation, it is not enough for a theory of Universal Grammar simply to list sets of universal properties of natural language grammars; on the contrary, a theory of UG must seek to explain the relevant properties. So, a key question for any adequate theory of UG to answer is: ‘Why do grammars of human I-languages have the properties they do?’ The requirement that a theory should explain why grammars have the properties they do is conventionally referred to as the criterion of explanatory adequacy. Since the theory of Universal Grammar is concerned with characterising the properties of natural (i.e. human) I-language grammars, an important question which we want our theory of UG to answer is: ‘What are the defining character- istics of human I-languages which differentiate them from, for example, artificial languages like those used in mathematics and computing (e.g. Java, Prolog, C etc.), or from animal communication systems (e.g. the tail-wagging dance per- formed by bees to communicate the location of a food source to other bees)?’ It therefore follows that the descriptive apparatus which our theory of Universal Grammar allows us to make use of in devising natural language grammars must not be so powerful that it can be used to describe not only natural languages, but also computer languages or animal communication systems (since any such excessively powerful theory wouldn’t be able to pinpoint the criterial properties of natural languages which differentiate them from other types of communication system). In other words, a third condition which we have to impose on our theory of language is that it be maximally constrained: that is, we want our theory to pro- vide us with technical devices which are so limited in their expressive power that they can only be used to describe natural languages, and are not appropriate for the description of other communication systems. A theory which is constrained in appropriate ways should enable us to provide a principled explanation for why 14 i GRAMMAR certain types of syntactic structure and syntactic operation simply aren’t found in natural languages. One way of constraining grammars is to suppose that gram- matical operations obey certain linguistic principles, and that any operation which violates the relevant principles leads to ungrammaticality: see the discussion in §1.5 below for a concrete example. A related requirement is that linguistic theory should provide grammars which make use of the minimal theoretical apparatus required: in other words, gram- mars should be as simple as possible. Some earlier work in syntax involved the postulation of complex structures and principles: as a reaction to the excessive complexity of this kind of work, Chomsky in work over the past two decades has made the requirement to minimise the theoretical and descriptive apparatus used to describe language the cornerstone of the Minimalist Program for Linguistic Theory which he has been developing. He has suggested that language is a per- fect system of optimal design in the sense that natural language grammars create structures which are designed to interface perfectly with other components of the mind – more specifically with speech and thought systems, so that (in the words of Chomsky 2005b, p. 2) ‘Language is an optimal way to link sound and meaning.’ To make this discussion rather more concrete, let’s look at the internal organi- sation of the grammar of a language. One component of a grammar is a lexicon (= dictionary = list of all the lexical items/words in the language and their linguistic properties), and in forming a given sentence out of a set of words, we first have to take the relevant words out of the lexicon. Our chosen words are then combined together by a series of syntactic computations in the syntax (i.e. in the syntac- tic/computational component of the grammar), thereby forming a syntactic structure. This syntactic structure serves as input into two other components of the grammar. One is the semantic component which maps (i.e. ‘converts’) the syntactic structure into a corresponding semantic representation (i.e. into a representation of linguistic aspects of its meaning): the other is a PF component, so called because it maps the syntactic structure into a PF representation (i.e. a representation of its Phonetic Form, giving us a phonetic spellout for each word, telling us how it is pronounced). The semantic representation interfaces with systems of thought, and the PF representation with systems of speech – as shown in diagrammatic form below: (16) semantic semantic THOUGHT component representation SYSTEMS Lexicon syntactic Syntax structure PF PF SPEECH component representation SYSTEMS Chomsky (2005b, p. 3) refers to the interface with thought systems as the ‘conceptual-intentional interface (CI)’, and to the interface with speech sys- tems as the ‘sensory-motor interface (SM)’. In terms of the model in (16), an 1.4 The Language Faculty 15 important consideration is that the (semantic and PF) representations which are ‘handed over’ to the (thought and speech) interface systems should contain only elements which are legible by the appropriate interface system – so that the semantic representations handed over to thought systems contain only elements contributing to meaning, and the PF representations handed over to speech sys- tems contain only elements which contribute to phonetic form (i.e. to determining how the sentence is pronounced). The neurophysiological mechanisms which underlie linguistic competence make it possible for young children to acquire language in a remarkably short period of time. Accordingly, a fourth condition which a linguistic theory must meet is that of learnability: it must provide grammars which are learnable by young children in a short period of time. The desire to maximise the learnability of natural language grammars provides an additional argument for minimising the theoretical apparatus used to describe languages, in the sense that the simpler grammars are, the simpler it is for children to acquire them. ● ● 1.4 The Language Faculty Mention of learnability leads us to consider the related goal of devel- oping a theory of language acquisition. An acquisition theory is concerned with the question of how children acquire grammars of their native languages. Children generally produce their first recognisable word (e.g. Mama or Dada) by around the age of twelve months (with considerable variation between indi- vidual children, however). For the next six months or so, there is little apparent evidence of grammatical development in their speech production, although the child’s productive vocabulary typically increases by about five words a month until it reaches around thirty words at age eighteen months. Throughout this single-word stage, children’s utterances comprise single words spoken in isola- tion: e.g. a child may say Apple when reaching for an apple, or Up when wanting to climb up onto someone’s knee. During the single-word stage, it is difficult to find any immediately visible evidence of the acquisition of grammar, in that children do not make productive use of inflections (e.g. they don’t productively add the plural -s ending to nouns, or the past tense -d ending to verbs), and don’t productively combine words together to form two- and three-word utterances. (However, it should be noted that perception experiments have suggested that infants may acquire some syntactic knowledge even before one year of age.) At around the age of eighteen months (though with considerable variation from one child to another), we find the first visible signs of the acquisition of grammar: children start to make productive use of inflections (e.g. using plural nouns like doggies alongside the singular form doggy, and inflected verb forms like going/gone alongside the uninflected verb form go), and similarly start to produce elementary two- and three-word utterances such as Want Teddy, Eating cookie, Daddy gone office etc. From this point on, there is a rapid expansion in 16 i GRAMMAR their grammatical development, until by the age of around thirty months they have typically acquired a wide variety of the inflections and core grammatical constructions used in English, and are able to produce adult-like sentences such as Where’s Mummy gone? What’s Daddy doing? Can we go to the zoo, Daddy? etc. (though occasional morphological and syntactic errors persist until the age of four years or so – e.g. We goed there with Daddy, What we can do? etc.). So, the central phenomenon which any theory of language acquisition must seek to explain is this: how is it that after a long-drawn-out period of many months in which there is no obvious sign of grammatical development, at around the age of eighteen months there is a sudden spurt as multiword speech starts to emerge, and a phenomenal growth in grammatical development then takes place over the next twelve months? This uniformity and (once the spurt has started) rapidity in the pattern of children’s linguistic development are the cen- tral facts which a theory of language acquisition must seek to explain. But how?Chomsky maintains that the most plausible explanation for the uniformity and rapidity of first language acquisition is to posit that the course of acquisition is determined by a biologically endowed innate Faculty of Language/FL (or language acquisition program, to borrow a computer software metaphor) within the brain, which provides children with a genetically transmitted algorithm (i.e. set of procedures) for developing a grammar, on the basis of their linguistic experience (i.e. on the basis of the speech input they receive). The way in which Chomsky visualises the acquisition process can be represented schematically as in (17) below (where L is the language being acquired): (17) E xperience of L → Faculty of Language/FL → Grammar of L Children acquiring a language will observe people around them using the lan- guage, and the set of expressions in the language which a child hears (and the contexts in which they are used) in the course of acquiring the language constitute the child’s linguistic experience of the language. This experience serves as input to the child’s Faculty of Language/FL, which incorporates a set of UG principles (i.e. principles of Universal Grammar) which enable the child to use the experi- ence to devise a grammar of the language being acquired. Thus, the input to the language faculty is the child’s experience, and the output of the language faculty is a grammar of the language being acquired. The claim that the course of language acquisition is determined by an innate language faculty is known popularly as the innateness hypothesis. Chomsky maintains that the ability to speak and acquire languages is unique to human beings, and that natural languages incorporate principles which are also unique to humans and which reflect the nature of the human mind: 1.4 The Language Faculty 17 Whatever evidence we do have seems to me to support the view that the ability to acquire and use language is a species-specific human capacity, that there are very deep and restrictive principles that determine the nature of human language and are rooted in the specific character of the human mind. (Chomsky 1972, p. 102) Moreover, he notes, language acquisition is an ability which all humans possess, entirely independently of their general intelligence: Even at low levels of intelligence, at pathological levels, we find a command of language that is totally unattainable by an ape that may, in other respects, surpass a human imbecile in problem-solving activity and other adaptive behaviour. (Chomsky 1972, p. 10) In addition, the apparent uniformity in the types of grammars developed by dif- ferent speakers of the same language suggests that children have genetic guidance in the task of constructing a grammar of their native language: We know that the grammars that are in fact constructed vary only slightly among speakers of the same language, despite wide variations not only in intelligence but also in the conditions under which language is acquired. (Chomsky 1972, p. 79) Furthermore, the rapidity of acquisition (once the grammar spurt has started) also points to genetic guidance in grammar construction: Otherwise it is impossible to explain how children come to construct gram- mars . . . under the given conditions of time and access to data. (Chomsky 1972, p. 113) (The sequence ‘under . . . data’ means simply ‘in so short a time, and on the basis of such limited linguistic experience’.) What makes the uniformity and rapidity of acquisition even more remarkable is the fact that the child’s linguistic experience is often degenerate (i.e. imperfect), since it is based on the linguistic performance of adult speakers, and this may be a poor reflection of their competence: A good deal of normal speech consists of false starts, disconnected phrases, and other deviations from idealised competence. (Chomsky 1972, p. 158) If much of the speech input which children receive is ungrammatical (because of performance errors), how is it that they can use this degenerate experience to develop a (competence) grammar which specifies how to form grammatical sentences? Chomsky’s answer is to draw the following analogy: Descartes asks: how is it when we see a sort of irregular figure drawn in front of us we see it as a triangle? He observes, quite correctly, that there’s a dispar- ity between the data presented to us and the percept that we construct. And he argues, I think quite plausibly, that we see the figure as a triangle because there’s something about the nature of our minds which makes the image of a triangle easily constructible by the mind. (Chomsky 1968, p. 687) 18 i GRAMMAR The obvious implication is that in much the same way as we are genetically predisposed to analyse shapes (however irregular) as having specific geometric properties, so too we are genetically predisposed to analyse sentences (however ungrammatical) as having specific grammatical properties. A further argument Chomsky uses in support of the innateness hypothesis relates to the fact that language acquisition is an entirely subconscious and invol- untary activity (in the sense that you can’t consciously choose whether or not to acquire your native language – though you can choose whether or not you wish to learn chess); it is also an activity which is largely unguided (in the sense that parents don’t teach children to talk): Children acquire . . . languages quite successfully even though no spe- cial care is taken to teach them and no special attention is given to their progress. (Chomsky 1965, pp. 200–1) The implication is that we don’t learn to have a native language, any more than we learn to have arms or legs; the ability to acquire a native language is part of our genetic endowment – just like the ability to learn to walk. Studies of language acquisition lend empirical support to the innateness hypothesis. Research has suggested that there is a critical period for the acquisition of syntax, in the sense that children who learn a given language before puberty generally achieve native competence in it, whereas those who acquire a (first or second) language after the age of nine or ten years rarely manage to achieve native-like syntactic competence. A particularly poignant example of this is a child called Genie, who was deprived of speech input and kept locked up on her own in a room until age thirteen. When eventu- ally taken into care and exposed to intensive language input, her vocabulary grew enormously, but her syntax never developed. This suggests that the acqui- sition of syntax is determined by an innate ‘language acquisition program’ which is in effect switched off (or gradually atrophies) around the onset of puberty. Further support for the key claim in the innateness hypothesis that the human Language Faculty comprises a modular cognitive system autonomous of non- linguistic cognitive systems such as vision, hearing, reasoning or memory comes from the study of language disorders. Some disorders (such as Specific Lan- guage Impairment) involve impairment of linguistic abilities without concomi- tant impairment of other cognitive systems. By contrast, other types of disorder (such as Williams Syndrome) involve impairment of cognitive abilities in the absence of any major impairment of linguistic abilities. This double dissoci- ation between linguistic and cognitive abilities lends additional plausibility to the claim that linguistic competence is the product of an autonomous Language Faculty. Given the assumption that human beings are endowed with an innate language faculty, the overall goal of linguistic theory is to attempt to uncover: 1.5 Principles of Universal Grammar 19 the properties that are specific to human language, that is, to the ‘faculty of language’ FL. To borrow Jespersen’s formulation eighty years ago, the goal is to unearth ‘the great principles underlying the grammars of all languages’ with the goal of ‘gaining a deeper insight into the innermost nature of human language and of human thought.’ The biolinguistic perspective views FL as an ‘organ of the body,’ one of many subcomponents of an organism that interact in its normal life. (Chomsky 2005b, p. 1) However, Chomsky (2006, p. 1) notes that some properties of human language may reflect ‘principles of biology more generally, and perhaps even more funda- mental principles about the natural world’. Accordingly: development of language in the individual must involve three factors: (1) genetic endowment, which sets limits on the attainable languages, thereby making language acquisition possible; (2) external data, converted to the experience that selects one or another language within a narrow range; (3) principles not specific to FL. (Chomsky 2006, p. 2: FL = Faculty of Language) The ‘third factor principles’ referred to under (3) ‘enter into all facets of growth and evolution’ and include ‘principles of efficient computation’ (Chomsky 2006, p. 2) and – more generally – ‘properties of the human brain that deter- mine what cognitive systems can exist, though too little is yet known about these to draw specific conclusions about the design of FL’ (Chomsky 2006, fn. 6)● ● 1.5 Principles of Universal Grammar If (as Chomsky claims) human beings are biologically endowed with an innate language faculty, an obvious question to ask is what the nature of the language faculty is. An important point to note in this regard is that children can in principle acquire any natural language as their native language (e.g. Afghan orphans brought up by English-speaking foster parents in an English-speaking community acquire English as their first language). It therefore follows that the language faculty must incorporate a theory of Universal Grammar/UG which enables the child to develop a grammar of any natural language on the basis of suitable linguistic experience of the language (i.e. sufficient speech input). Experience of a particular language L (examples of words, phrases and sentences in L which the child hears produced by native speakers of L in particular contexts) serves as input to the child’s language faculty which incorporates a theory of Universal Grammar providing the child with a procedure for developing a grammar of L. If the acquisition of grammatical competence is indeed controlled by a genet- ically endowed language faculty incorporating a theory of UG, then it fol- lows that certain aspects of child (and adult) competence are known without 20 i GRAMMAR experience, and hence must be part of the genetic information about language with which we are biologically endowed at birth. Such aspects of language would not have to be learned, precisely because they form part of the child’s genetic inheritance. If we make the (plausible) assumption that the language faculty does not vary significantly from one (normal) human being to another, those aspects of language which are innately determined will also be universal. Thus, in seek- ing to determine the nature of the language faculty, we are in effect looking for UG principles (i.e. principles of Universal Grammar) which determine the very nature of language. But how can we uncover such principles? The answer is that since the relevant principles are posited to be universal, it follows that they will affect the application of every relevant type of grammatical operation in every language. Thus, detailed analysis of one grammatical construction in one language could reveal evidence of the operation of principles of Universal Grammar. By way of illustration, let’s look at question-formation in English. In this connection, consider the following dialogue: (18) SPEAkER A: He had said someone would do something SPEAkER B: He had said who would do what? In (18), speaker B largely echoes what speaker A says, except for replacing someonebywhoandsomethingbywhat.For obvious reasons, the type of question produced by speaker B in (18) is called an echo question. However, speaker B could alternatively have replied with a non-echo question like that below: (19) Who had he said would do what? If we compare the echo question He had said who would do what? in (18) with the corresponding non-echo question Who had he said would do what? in (19), we find that (19) involves two movement operations which are not found in (18). One is an auxiliary inversion operation by which the past tense auxiliary had is moved in front of its subject he. The other is a wh-movement operation by which the wh-word who is moved to the front of the overall sentence, and positioned in front of had. (A wh-word is a question word like who/what/where/when etc. beginning with wh.) A closer look at questions like (19) provides evidence that there are UG principles which constrain the way in which movement operations may apply. An interesting property of the questions in (18B, 19) is that they contain two auxiliaries (had and would) and two wh-words (who and what). Now, if we compare (19) with the corresponding echo-question in (18), we find that the first of the two auxiliaries (had) and the first of the wh-words (who) is moved to the front of the sentence in (19). If we try inverting the second auxiliary (would) and fronting the second wh-word (what), we end up with ungrammatical sentences, as we see from (20c–e) below (key items are bold-printed/italicised, and the corresponding echo question is given in parentheses; 20a is repeated from the echo question in 18B, and 20b from 19): 1.5 Principles of Universal Grammar 21 (20) (a) He had said who would do what? (= echo question) (b) Who had he said would do what? (cf. He had said who would do what?) (c) ∗Who would he had said do what? (cf. He had said who would do what?) (d) ∗What had he said who would do? (cf. He had said who would do what?) (e) ∗What would he had said who do? (cf. He had said who would do what?) If we compare (20b) with its echo-question counterpart (20a) He had said who would do what? we see that (20b) involves preposing the first wh-word who and the first auxiliary had, and that this results in a grammatical sentence. By contrast, (20c) involves preposing the first wh-word who and the second auxiliary would; (20d) involves preposing the second wh-word what and the first auxiliary had; and (20e) involves preposing the second wh-word what and the second auxiliary would. The generalisation which emerges from the data in (20) is that auxiliary inversion preposes the closest auxiliary had (i.e. the one nearest the beginning of the sentence in (20a) above) and likewise wh-fronting preposes the closest wh- expression who. The fact that two quite distinct movement operations (auxiliary inversion and wh-movement) are subject to the same locality condition (which requires preposing of the most local – i.e. closest – expression of the relevant type) suggests that one of the principles of Universal Grammar incorporated into the language faculty is a Locality Principle which can be outlined informally as: (21) Locality Principle Grammatical operations are local In consequence of (21), auxiliary inversion preposes the closest auxiliary, and wh-movement preposes the closest wh-expression. It seems reasonable to sup- pose that (21) is a principle of Universal Grammar (rather than an idiosyncratic property of question-formation in English). In fact, the strongest possible hypoth- esis we could put forward is that (21) holds of all grammatical operations in all natural languages, not just of movement operations; and indeed we shall see in later chapters that other types of grammatical operation (including agreement and case assignment) are subject to a similar locality condition. If so, and if we assume that abstract grammatical principles which are universal are part of our biological endowment, then the natural conclusion to reach is that (21) is a principle which is biologically wired into the language faculty, and which thus forms part of our genetic make-up. A theory of grammar which posits that grammatical operations are constrained by innate principles of UG offers the important advantage that it minimises the burden of grammatical learning imposed on the child (in the sense that children do not have to learn e.g. that auxiliary inversion affects the first auxiliary in a sentence, or that wh-movement likewise affects the first wh-expression). This is an important consideration, since we saw earlier that learnability is a criterion of adequacy for any theory of grammar – i.e. any adequate theory of grammar must be able to explain how children come to learn the grammar of their native language(s) in such a rapid and uniform fashion. The UG theory developed by Chomsky provides a straightforward account of the rapidity of the child’s 22 i GRAMMAR grammatical development, since it posits that there is a universal set of innately endowed grammatical principles which determine how grammatical operations apply in natural language grammars. Since UG principles which are innately endowed are wired into the language faculty and so do not have to be learned by the child, this minimises the learning load placed on the child, and thereby maximises the learnability of natural language grammars. It also (correctly) predicts that there are certain types of error which children will not make – e.g. producing sentences such as (20c–e). ● ● 1.6 Parameters Thus far, we have argued that the language faculty incorporates a set of universal principles which guide the child in acquiring a grammar. However, it clearly cannot be the case that all aspects of the grammar of languages are uni- versal; if this were so, all natural languages would have the same grammar and there would be no grammatical learning involved in language acquisition (i.e. no need for children to learn anything about the grammar of the language they are acquiring), only lexical learning (viz. learning the lexical items/words in the language and their idiosyncratic linguistic properties, e.g. whether a given item has an irregular plural or past tense form). But although there are universal prin- ciples which determine the broad outlines of the grammar of natural languages, there also seem to be language-particular aspects of grammar which children have to learn as part of the task of acquiring their native language. Thus, lan- guage acquisition involves not only lexical learning but also some grammatical learning. Let’s take a closer look at the grammatical learning involved, and what it tells us about the language acquisition process. Clearly, grammatical learning is not going to involve learning those aspects of grammar which are determined by universal (hence innate) grammatical opera- tions and principles. Rather, grammatical learning will be limited to those param- eters (i.e. dimensions or aspects) of grammar which are subject to language- particular variation (and hence vary from one language to another). In other words, grammatical learning will be limited to parametrised aspects of grammar (i.e. those aspects of grammar which are subject to parametric variation from one language to another). The obvious way to determine just what aspects of the grammar of their native language children have to learn is to examine the range of parametric variation found in the grammars of different (adult) natural languages. We can illustrate one type of parametric variation across languages in terms of the following contrast between the English example in (22a) below and its Italian counterpart in (22b): (22) (a) Maria thinks that ∗(they) speak French (b) Maria pensa che parlano francese ‘Maria thinks that speak French’ 1.6 Parameters 23 (The notation ∗(they) in 22a means that the sentence is ungrammatical if they is omitted – i.e. that the sentence ∗Maria thinks that speak Frenchis ungrammatical.) The finite (present tense) verb speak in the English sentence (22a) requires an overt subject like they, but its Italian counterpart parlanospeak in (22b) has no overt subject. However, there are two pieces of evidence suggesting that the Italian verb parlanospeak must have a ‘silent’ subject of some kind. One is semantic in nature, in that the verb parlanospeak is understood as having a third person plural subject, and this understood subject is translated into English as they; in more technical terms, this amounts to saying that in the relevant use, the verb parlanospeak is a two-place predicate which requires both a subject argument and an object argument, and so it must have an ‘understood’ silent subject of some kind in (22b). The second piece of evidence is grammatical in nature. Finite verbs agree with their subjects in Italian: hence, in order to account for the fact that the verb parlanospeak is in the third person plural form in (22b), we need to posit that it has a third person plural subject to agree with. Since the verb parlanospeak has no overt subject, it must have a null subject which can be thought of as a silent or invisible counterpart of the pronoun they which appears in the corresponding English sentence (22a). This null subject is conventionally designated as pro, so that (22b) has the fuller structure Maria pensa che pro parlano francese ‘Maria thinks that pro speak French,’ where pro is a null subject pronoun. The more general conclusion to be drawn from our discussion here is that in languages like Italian, any finite verb can have either an overt subject like Maria or a null pro subject. But things are very different in English. Although finite verbs can have an overt subject like Maria in English, they cannot normally have a null pro subject – hence the ungrammaticality of ∗Maria thinks that speak French (where the verb speak has a null subject). So, finite verbs in a language like Italian can have either overt or null subjects, but in a language like English, finite verbs can generally have only overt subjects, not null subjects. We can describe the differences between the two types of language by saying that Italian is a null-subject language, whereas English is a non-null-subject language. More generally, there appears to be parametric variation between languages as to whether or not they allow finite verbs to have null subjects. The relevant parameter (termed the Null Subject Parameter) would appear to be a binary one, with only two possible settings for any given language L, viz. L either does or doesn’t allow any finite verb to have a null subject. There appears to be no language which allows the subjects of some finite verbs to be null, but not others – e.g. no language in which it is OK to say Drinks wine (meaning ‘He/she drinks wine’) but not OK to say Eats pasta (meaning ‘He/she eats pasta’). The range of grammatical variation found across languages appears to be strictly limited to just two possibilities – languages either do or don’t systematically allow finite verbs to have null subjects. A more familiar aspect of grammar which appears to be parametrised relates to word order, in that different types of language have different word orders in specific types of construction. One type of word-order variation can be 24 i GRAMMAR illustrated in relation to the following contrast between English and Chinese questions: (23) (a) What do you think he will say? (b) Ni xiang ta hui shuo shenme You think he will say what? In simple wh-questions in English (i.e. questions containing a single word begin- ning with wh- like what/where/when/why) the wh-expression is moved to the beginning of the sentence, as is the case with what in (23a). By contrast, in Chinese, the wh-word does not move to the front of the sentence, but rather remains in situ (i.e. in the same place as would be occupied by a corresponding non-interrogative expression), so that shenme ‘what’ is positioned after the verb shuo ‘say’ because it is the (direct object) complement of the verb, and comple- ments of the relevant type are normally positioned after their verbs in Chinese. Thus, another parameter of variation between languages is the Wh-Parameter – a parameter which determines whether wh-expressions are fronted (i.e. moved to the front of the overall interrogative structure containing them) or not. Signif- icantly, this parameter again appears to be one which is binary in nature, in that it allows for only two possibilities – viz. a language either does or doesn’t allow wh-movement (i.e. movement of wh-expressions to the front of the sentence). Many other possibilities for wh-movement just don’t seem to occur in natural language: for example, there is no language in which the counterpart of who undergoes wh-fronting but not the counterpart of what (e.g. no language in which it is OK to say Who did you see? but not What did you see?). Likewise, there is no language in which wh-complements of some verbs can undergo fronting, but not wh-complements of other verbs (e.g. no language in which it is OK to say What did he drink? but not What did he eat?). It would seem that the range of parametric variation found with respect to wh-fronting is limited to just two possibilities: viz. a language either does or doesn’t allow wh-expressions to be systematically fronted. Let’s now turn to look at a rather different type of word-order variation, concerning the relative position of heads and complements within phrases. It is a general (indeed, universal) property of phrases that every phrase has a head word which determines the nature of the overall phrase. For example, an expression such as students of philosophy is a plural Noun Phrase because its head word (i.e. the key word in the phrase whose nature determines the properties of the overall phrase) is the plural noun students: the noun students (and not the noun philosophy) is the head word because the phrase students of philosophy denotes kinds of student, not kinds of philosophy. The following expression of philosophy which combines with the head noun students to form the Noun Phrase students of philosophy functions as the complement of the noun students. In much the same way, an expression such as in the kitchen is a Prepositional Phrase which comprises the head preposition in and its complement the kitchen. Likewise, an expression such as stay with me is a Verb Phrase which comprises the head verb 1.6 Parameters 25 stay and its complement with me. And similarly, an expression such as fond of fast food is an Adjectival Phrase formed by combining the head adjective fond with its complement of fast food. In English all heads (whether nouns, verbs, prepositions or adjectives etc.) immediately precede their complements; however, there are also languages like Korean in which all heads immediately follow their complements. In informal terms, we can say that English is a head-first language, whereas Korean is a head-last language. The differences between the two languages can be illustrated by comparing the English examples in (24) below with their Korean counterparts in (25): (24) (a) Close the door (b) desire for change (25) (a) Muneul dadara (b) byunhwa-edaehan galmang Door close change-for desire In the English Verb Phraseclose the doorin (24a), the head verbcloseimmediately precedes its complement the door; if we suppose that the door is a Determiner Phrase, then the head of the phrase (= the determiner the) immediately precedes its complement (= the noun door). Likewise, in the English Noun Phrase desire for change in (24b), the head noun desire immediately precedes its complement for change; the complement for change is in turn a Prepositional Phrase in which the head preposition for likewise immediately precedes its complement change. Since English consistently positions heads before complements, it is a head-first language. By contrast, we find precisely the opposite ordering in Korean. In the Verb Phrase muneul dadara (literally ‘door close’) in (25a), the head verb dadara ‘close’ immediately follows its complement muneul ‘door’; likewise, in the Noun Phrase byunhwa-edaehan galmang (literally ‘change-for desire’) in (25b) the head noun galmang ‘desire’ immediately follows its complement byunhwa-edaehan ‘change-for’; the expression byunhwa-edaehan ‘change-for’ is in turn a Prepositional Phrase whose head preposition edaehan ‘for/about’ immediately follows its complement byunhwa ‘change’ (so that edaehan might more appropriately be called a postposition; prepositions and postpositions are differents kinds of adposition). Since Korean consistently positions heads imme- diately after their complements, it is a head-last language. Given that English is head-first and Korean head-last, it is clear that the relative positioning of heads with respect to their complements is one word-order parameter along which lan- guages differ; the relevant parameter is termed the Head Position Parameter. It should be noted, however, that word-order variation in respect of the relative positioning of heads and complements falls within narrowly circumscribed limits. There are many logically possible types of word-order variation which just don’t seem to occur in natural languages. For example, we might imagine that in a given language some verbs would precede and others follow their complements, so that (e.g.) if two new hypothetical verbs like scrunge and plurg were coined in English, then scrunge might take a following complement, and plurg a preceding 26 i GRAMMAR complement. And yet, this doesn’t ever seem to happen: rather, all verbs typically occupy the same position in a given language with respect to a given type of complement. What this suggests is that there are universal constraints (i.e. restrictions) on the range of parametric variation found across languages in respect of the relative ordering of heads and complements. It would seem that there are only two different possibilities which the theory of Universal Grammar allows for: a given type of structure in a given language must either be head-first (with the relevant heads positioned immediately before their complements), or head-last (with the relevant heads positioned immediately after their complements). Many other logically possible orderings of heads with respect to complements appear not to be found in natural language grammars. The obvious question to ask is why this should be. The answer given by the theory of parameters is that the language faculty imposes genetic constraints on the range of parametric variation permitted in natural language grammars. In the case of the Head Position Parameter (i.e. the parameter which determines the relative positioning of heads with respect to their complements), the language faculty allows only a binary set of possibilities – namely that a given kind of structure in a given language is either consistently head-first or consistently head-last. We can generalise our discussion in this section in the following terms. If the Head Position Parameter reduces to a simple binary choice, and if the Wh-Parameter and the Null Subject Parameter also involve binary choices, it seems implausible thatbinaritycould be an accidental property of these particular parameters. Rather, it seems much more likely that it is an inherent property of parameters that they constrain the range of structural variation between languages, and limit it to a simple binary choice. Generalising still further, it seems possible that all grammatical variation between languages can be characterised in terms of a set of parameters, and that for each parameter, the language faculty specifies a binary choice of possible values for the parameter. ● ● 1.7 Parameter-setting The theory of parameters outlined in the previous section has impor- tant implications for a theory of language acquisition. If all grammatical variation can be characterised in terms of a series of parameters with binary settings, it follows that the only grammatical learning which children have to undertake in relation to the syntactic properties of the relevant class of constructions is to determine (on the basis of their linguistic experience) which of the two alter- native settings for each parameter is the appropriate one for the language being acquired. So, for example, children have to learn whether the native language they are acquiring is a null subject language or not, whether it is a wh-movement language or not, and whether it is a head-first language or not . . . and so on for all the other parameters along which languages vary. Of course, children also face 1.7 Parameter-setting 27 the formidable task of lexical learning – i.e. building up their vocabulary in the relevant language, learning what words mean and what range of forms they have (e.g. whether they are regular or irregular in respect of their morphology), what kinds of structures they can be used in and so on. On this view, the acquisition of grammar involves the twin tasks of lexical learning and structural learning (with the latter involving parameter-setting). This leads us to the following view of the language acquisition process. The central task which the child faces in acquiring a language is to construct a grammar of the language. The innate Language Faculty incorporates (i) a set of universal grammatical principles, and (ii) a set of grammatical parameters which impose severe constraints on the range of grammatical variation permitted in natural languages (perhaps limiting variation to binary choices). Since universal principles don’t have to be learned, the child’s syntactic learning task is limited to that of parameter-setting (i.e. determining an appropriate setting for each of the relevant grammatical parameters). For obvious reasons, the theory outlined here (developed by Chomsky at the beginning of the 1980s) is known as Principles- and-Parameters Theory/PPT. The PPT model clearly has important implications for the nature of the lan- guage acquisition process, since it vastly reduces the complexity of the acquisition task which children face. PPT hypothesises that grammatical properties which are universal will not have to be learned by the child, since they are wired into the language faculty and hence part of the child’s genetic endowment: on the con- trary, all the child has to learn are those grammatical properties which are subject to parametric variation across languages. Moreover, the child’s learning task will be further simplified if it turns out (as research since 1980 has suggested) that the values which a parameter can have fall within a narrowly specified range, perhaps characterisable in terms of a series of binary choices. This simplified parameter- setting model of the acquisition of grammar has given rise to a metaphorical acquisition model in which the child is visualised as having to set a series of switches in one of two positions (up/down) – each such switch representing a different parameter. In the case of the Head Position Parameter, we can imag- ine that if the switch is set in the up position (for particular types of head), the language will show head-first word order in relevant kinds of structure, whereas if it is set in the down position, the order will be head-last. Of course, an obvious implication of the switch metaphor is that the switch must be set in either one position or the other, and cannot be set in both positions. (This would preclude e.g. the possibility of a language having both head-first and head-last word order in a given type of structure.) The assumption that acquiring the grammar of a language involves the rel- atively simple task of setting a number of grammatical parameters provides a natural way of accounting for the fact that the acquisition of specific parameters appears to be a remarkably rapid and error-free process in young children. For example, young children acquiring English as their native language seem to set the Head Position Parameter at its appropriate head-first setting from the very 28 i GRAMMAR earliest multiword utterances they produce (at around eighteen months of age), and seem to know (tacitly, not explicitly, of course) that English is a head-first lan- guage. Accordingly, the earliest verb phrases and Prepositional Phrases produced by young children acquiring English consistently show verbs and prepositions positioned before their complements, as structures such as the following indicate (produced by a young boy called Jem/James at age twenty months; head verbs are italicised in (26a) and head prepositions in (26b), and their complements are in non-italic print): (26) (a) Touch heads. Cuddle book. Want crayons. Want malteser. Open door. Want biscuit. Bang bottom. See cats. Sit down (b) On Mummy. To lady. Without shoe. With potty. In keyhole. In school. On carpet. On box. With crayons. To mummy The obvious conclusion to be drawn from structures like (26) is that children like Jem consistently position heads before their complements from the very earliest multiword utterances they produce. They do not use different orders for different words of the same type (e.g. they don’t position the verb see after its complement but the verb want before its complement), or for different types of words (e.g. they don’t position verbs before and prepositions after their complements). A natural question to ask at this point is how we can provide a principled expla- nation for the fact that from the very onset of multiword speech we find English children correctly positioning heads before their complements. The Principles- and-Parameters model enables us to provide an explanation for why children manage to learn the relative ordering of heads and complements in such a rapid and error-free fashion. The answer provided by the model is that learning this aspect of word order involves the comparatively simple task of setting a binary parameter at its appropriate value. This task will be a relatively straightforward one if the language faculty tells the child that the only possible choice is for a given type of structure in a given language to be uniformly head-first or uni- formly head-last. Given such an assumption, the child could set the parameter correctly on the basis of minimal linguistic experience. For example, once the child is able to analyse the structure of an adult utterance such as Help Daddy and knows that it contains a Verb Phrase comprising the head verb help and its complement Daddy, then (on the assumption that the language faculty specifies that all heads of a given type behave uniformly with regard to whether they are positioned before or after their complements), the child will automatically know that all verbs in English are canonically (i.e. normally) positioned before their complements. One of the questions posed by the parameter-setting model of acquisition outlined here is just how children come to arrive at the appropriate setting for a given parameter, and what kind(s) of evidence they make use of in set- ting parameters. There are two types of evidence which we might expect to be available to the language learner in principle, namely positive evidence and negative evidence. Positive evidence comprises a set of observed expressions 1.7 Parameter-setting 29 illustrating a particular phenomenon: for example, if children’s speech input is made up of structures in which heads precede their complements, this provides them with positive evidence which enables them to set the Head Position Param- eter at the head-first setting appropriate to English. Negative evidence might be of two kinds – direct or indirect. Direct negative evidence could come from the correction of children’s errors by other speakers of the language. However, (contrary to what is often imagined) correction plays a fairly insignificant role in language acquisition, for two reasons. Firstly, correction is relatively infrequent: adults simply don’t correct all the errors children make (if they did, children would soon become inhibited and discouraged from speaking). Secondly, chil- dren are notoriously unresponsive to correction, as the following dialogue (from McNeill 1966, p. 69) illustrates: (27) CHILD: Nobody don’t like me ADULT: No, say: ‘Nobody likes me’ CHILD: Nobody don’t like me (8 repetitions of this dialogue) ADULT: No, now listen carefully. Say ‘Nobody likes me’ CHILD: Oh, nobody don’t likes me As Hyams (1986, p. 91) notes: ‘Negative evidence in the form of parental dis- approval or overt corrections has no discernible effect on the child’s developing syntactic ability.’ Direct negative evidence might also take the form of self-correction by other speakers. Such self-corrections tend to have a characteristic intonation and rhythm of their own, and may be signalled by a variety of fillers (such as those italicised in (28) below): (28) (a) The picture was hanged . . . or rather hung . . . in the Tate Gallery (b) The picture was hanged . . . sorry hung . . . in the Tate Gallery (c) The picture was hanged . . . I mean hung . . . in the Tate Gallery However, self-correction is arguably too infrequent a phenomenon to play a major role in the acquisition process. Rather than say that children rely on direct negative evidence, we might instead imagine that they learn from indirect negative evidence (i.e. evidence relating to the non-occurrence of certain types of structure). Suppose that a child’s experience includes no examples of structures in which heads follow their complements (e.g. no Prepositional Phrases like ∗dinner after in which the head preposition after follows its complement dinner, and no Verb Phrases such as ∗cake eat in which the head verb eat follows its complement cake). On the basis of such indirect negative evidence (i.e. observing that such structures never occur in English), the child might infer that English is not a head-last language. Although it might seem natural to suppose that indirect negative evidence plays some role in the acquisition process, there are potential learnability problems posed by any such claim. After all, the fact that a given construction does not occur 30 i GRAMMAR in a given chunk of the child’s experience does not provide conclusive evidence that the structure is ungrammatical, since it may well be that the non-occurrence of the relevant structure in the relevant chunk of experience is an accidental (rather than a systematic) gap. Thus, the child would need to process a very large (in principle, infinite) chunk of experience in order to be sure that non- occurrence reflects ungrammaticality. It is implausible that young children pro- cess massive chunks of experience in this way and search through it for negative evidence about the non-occurrence of certain types of structure, since this would impose an unrealistic memory load on them. In any case, given the assumption that parameters are binary and single-valued, negative evidence becomes entirely unnecessary: after all, once the child hears a Prepositional Phrase like with Daddy in which the head preposition with precedes its complement Daddy, the child will have positive evidence that English allows head-first order in prepositional phrases; and given the assumption that the Head Position Parameter is a binary one and the further assumption that each parameter allows only a single setting, then it follows (as a matter of logical necessity) that if English allows head-first Prepositional Phrases, it will not allow head-last Prepositional Phrases. Thus, in order for the child to know that English doesn’t allow head-last Prepositional Phrases, the child does not need negative evidence from the non-occurrence of such structures, but rather can rely on positive evidence from the occurrence of the converse order in head-first structures (on the assumption that if a given structure is head-first, UG specifies that it cannot be head-last). And, as we have already noted, a minimal amount of positive evidence is required in order to identify English as a uniformly head-first language (i.e. a language in which all heads precede their complements). Learnability considerations such as these have led Chomsky (1986a, p. 55) to conclude that ‘There is good reason to believe that children learn language from positive evidence only.’ The claim that children do not make use of negative evidence in setting parameters is known as the No- Negative-Evidence Hypothesis; it is a hypothesis which is widely assumed in current acquisition research. ● ● 1.8 Summary We began this chapter in §1.2 with a brief look at traditional grammar, noting that this is a taxonomic (i.e. classificatory) system in which the syntax of a given sentence is described by assigning each of the constituents in the sentence to a grammatical category, and saying what grammatical function it has. In §1.3, we noted that Chomsky takes a very different cognitive approach to the study of language in which a grammar of a language is a model of the grammatical knowledge (or competence) internalised in the mind/brain of a native speaker (hence a model of the speaker’s I-language). We saw that Chomsky’s ultimate goal is to develop a theory of Universal Grammar/UG which characterises the defining properties of the grammars of natural languages – a theory which is universal, explanatory and constrained, and which provides descriptively 1.8 Summary 31 adequate grammars that are minimally complex and hence learnable. In §1.4, we went on to look at the nature of language acquisition, and argued that the most fundamental question for a theory of language acquisition to answer is why it should be that after a period of a year and a half during which there is little evidence of grammatical development visible in the child’s speech output, most of the grammar of the language is acquired by children during the course of the following year. We outlined the Innateness Hypothesis put forward by Chomsky, under which the course of language acquisition is genetically prede- termined by an innate Language Faculty. In §1.5, we noted Chomsky’s claim that the Language Faculty incorporates a theory of Universal Grammar/UG which embodies a set of universal grammatical principles that determine the ways in which grammatical operations work; and we saw that the syntax of ques- tions in English provides evidence for postulating that syntactic operations are constrained by the following principle: Locality Principle: Every grammatical operation is local in the sense that it affects the closest constituent of the relevant type In §1.6, we went on to argue that the grammars of natural languages vary along a number of parameters. We looked at three such parameters, namely: Wh-Parameter: Some languages (like English) require movement of an inter- rogative wh-expression to the front of an interrogative clause, whereas others (like Chinese) leave interrogative wh-expressions in situ Null Subject Parameter: Some languages (like Italian) allow a null pronoun (= pro) to be used as the subject of any finite (auxiliary or main) verb, whereas other languages (like English) do not Head Position Parameter: Some languages (like English) position head words immediately before their complements, whereas others (like Korean) position them immediately after their complements. We hypothesised that each such parameter has a binary choice of settings. In §1.7, we argued that the syntactic learning task which children face involves parameter-setting – i.e. determining which of two possible settings is the appro- priate one for each parameter in the language being acquired. We further argued that if parameters have binary settings (e.g. so that a given kind of structure in a given language is either head-first or head-last), we should expect to find evidence that children correctly set parameters from the very onset of multiword speech: and we presented evidence to suggest that from their very earliest multiword utterances, children acquiring English as their mother tongue correctly set the Head Position Parameter at the head-first value appropriate for English. We con- cluded that the acquisition of grammar involves the twin tasks of lexical learning (i.e. acquiring a lexicon/vocabulary) and parameter-setting. We went on to ask what kind of evidence children use in setting parameters, and concluded that they use positive evidence from their experience of the occurrence of specific types of structure (e.g. head-first structures, or null subject structures or wh-movement structures). 32 i GRAMMAR ● ● 1.9 Bibliographical background For a fuller account of the grammatical categories discussed in §1.2, see chapter 2 of Radford (2004a) or (2004b). On the nature of determiners, see Giusti (1997), Spinillo (2004) and Isac (2006). On different types of pro- noun, see Cardinaletti and Starke (1999), Wiltschko (2001) and D ́echaine and Wiltschko (2002). On the claim that personal pronouns are D constituents, see Postal (1966) and Abney (1987). On the claim that infinitival to is a tense particle, see Freidin (2004, p. 117, fn. 32). For a technical discussion of tense, see Julien (2001) and Ishii (2006a). The term complementiser dates back to Rosenbaum (1965, 1967) and Bresnan (1970). For more extensive discussion of the notion of I-language introduced in §1.3, see Smith (2004). Chomsky’s Minimalist Pro- gram is developed in Chomsky (1993, 1995, 1998, 1999, 2001, 2002, 2005a, 2005b, 2006). For discussion of Chomsky’s idea that language is a perfect sys- tem of optimal design, see Lappin, Levine and Johnson (2000a, 2000b, 2001), Holmberg (2000a), Piattelli-Palmarini (2000), Reuland (2000, 2001b), Roberts (2000, 2001a), Uriagereka (2000, 2001) and Freidin and Vergnaud (2001). For further discussion of the innateness hypothesis outlined in §1.4, see Lightfoot (1999), Anderson and Lightfoot (2002), Antony and Hornstein (2003), Giv ́on (2002), Hauser, Chomsky and Fitch (2002) and Fitch, Hauser and Chomsky (2005); for a more critical view, see Everett (2005, 2006) and Sampson (2005), and for a reply to such criticism, see Chomsky’s contributions to Antony and Hornstein (2003). For a textbook summary of perceptual evidence that very young infants may be sensitive to syntactic structure, see Lust (2006, §9.2.1). For evaluation of the idea that children learn languages in spite of receiving degener- ate input, see Pullum and Scholz (2002), Thomas (2002), Sampson (2002), Fodor and Crowther (2002), Lasnik and Uriagereka (2002), Legate and Yang (2002), Crain and Pietroski (2002), Scholz and Pullum (2002), Lewis and Elman (2002) and Gualmini and Crain (2005). For discussion of the critical period in language acquisition, see Lenneberg (1967), Hurford (1991) and Smith (1998, 2004); on Genie, see Curtiss (1977) and Rymer (1993). On evidence of a double disso- ciation between linguistic and cognitive abilities, see Clahsen (2008). The idea outlined in §1.5 that grammars incorporate a set of UG principles is developed in Chomsky (1981). The Locality Principle sketched in the same section has its historical roots in a number of related principles, including the Relativised Mini- mality Principle of Rizzi (1990), the Shortest Move principle of Chomsky (1995) and the Attract Closest Principle of Richards (1997). The idea that grammatical differences between languages can be reduced to a small number of parame- ters is developed in Chomsky (1981). A complication glossed over in the text discussion of the Null Subject Parameter is posed by languages in which only some finite verb forms can have null subjects: see Vainikka and Levy (1999) and the collection of papers in Jaeggli and Safir (1989) for illustration and discussion. The discussion of the Wh-Parameter in the main text is simplified by ignoring the complication that some languages allow more than one wh-expression to Workbook section 33 be fronted in wh-questions (see Boškovi ́c 2002a, Grohmann 2006 and Sur ́anyi 2006), and the additional complication that wh-movement appears to be optional in some languages, either in main clauses, or in main and complement clauses alike (see Denham 2000, and Cheng and Rooryck 2000); on wh-in-situ struc- tures, see Pesetsky (1987), Cheng (1997), Cole and Hermon (1998), Reinhart (1998) and Bruening (2007). The claim made in the outline of the Head Position Parameter that all heads of a given type occupy a uniform position with respect to their complements is called into question by the behaviour of prepositions in German, most of which precede their complements, but a few of which (e.g. entlang ‘along’) follow them. Although we assumed in the text that parameters have binary settings, it should be noted that some researchers have assumed that parameters can have more than two alternative settings (e.g. Manzini and Wexler 1987). For discussion of a wide range of parametric variation between languages, see Cinque and Kayne (2005). For a critique of the idea that cross-linguistic variation is reducible to a small number of structural parameters, see Culicover and Nowak (2003), Newmeyer (2004, 2006) and Abeill ́e and Borsley (2006): for a defence of parameters, see Roberts and Holmberg (2006). For a defence of the claim made in §1.7 that parameters are correctly set by children at a very early stage in their development, see Wexler (1998). The claim that no negative evidence is used in setting parameters is made in Chomsky (1981, pp. 8–9); sup- porting evidence can be found in McNeill (1966), Brown, Cazden and Bellugi (1968), Brown and Hanlon (1970), Braine (1971), Bowerman (1988), Morgan and Travis (1989) and Marcus (1993) – but for potential counterevidence, see Lappin and Shieber (2007). On how children set parameters, see Fodor (2001) and Fodor and Sakas (2005). For a technical account of language acquisition within the framework used here, see Guasti (2002) and Lust (2006). ● ● Workbook section ● Exercise 1.1 Word-order parameters like the Head Position Parameter determine the canonical (i.e. ‘basic’, ‘normal’ or ‘underlying’) word order found in particular types of structure in a given language. However (as we will see in subsequent chapters), languages may have a variety of movement operations which allow particular types of expression to be fronted (i.e. preposed) and thereby be moved out of their canonical position into some new position at the front of a particular phrase, clause or sentence. For example, in a head-first language like English, both main and auxiliary verbs immediately precede their complements. Accordingly, in a sentence like John has gone home, the verb gone immediately precedes its complement home, and the auxiliary has immediately precedes its complement gone home. But in a question like Where has John gone? wh-movement means that the complement where of the verb gone is moved to the front of the overall sentence, and so the verb gone no longer immediately precedes its complement where. Likewise, the auxiliary has undergoes auxiliary inversion (thereby moving in front of its subject 34 i GRAMMAR John) and consequently no longer immediately precedes its complement gone where. Below are a number of sentences taken from various plays written by Shakespeare, representing a variety of English sometimes referred to as Elizabethan English (because it was spoken during the reign of Queen Elizabeth the First). Elizabethan English (like present-day English) was a head-first language in which heads were canonically positioned in front of their complements. In relation to the sentences below, show how movement operations which fronted various types of expression could mask the underlying head-first setting of the Head Position Parameter in Elizabethan English. 1 Seawater shalt thou drink (Prospero, The Tempest, I.ii) 2 That letter hath she delivered (Speed, Two Gentlemen of Verona, II.i) 3 Friend hast thou none (Duke, Measure for Measure, III.i) 4 True is it that we have seen better days (Duke Senior, As You Like It, II.vii) (w) 5 She may more suitors have (Tranio, The Taming of the Shrew, I.ii) 6 Run you to the citadel! (Iago, Othello, V.i) 7 Came you from the church? (Tranio, Taming of the Shrew, III.ii) 8 What think you he hath confessed? (First Lord, All’s Well That Ends Well, IV.iii) (w) 9 What will this come to? (Flavius, Timon of Athens, I.ii) 10 What visions have I seen! (Titania, Midsummer Night’s Dream, V.i) Helpful hints Take none in 3, more in 5 and what in 10 to be quantifiers with a noun as their complement (and assume that the negative quantifier is spelled out as no if immediately followed by its complement, but as none otherwise). Note that 1–5 are declarative sentences (used to make a statement), 6 is an imperative sentence (used to issue an order), 7–9 are interrogative sentences (used to ask a question) and 10 is an exclamative sentence (used to exclaim amazement). Model answer for 1 The auxiliary verb shalt ‘shall’ has the subject thou ‘yousingular’ and the complement drink seawater. The main verb drink has the complement seawater. If no movement operations took place in the relevant sentence, we should expect to find the word order Thou shalt drink seawater, with the auxiliary shalt immediately preceding its complement drink seawater, and the verb drink immediately preceding its complement seawater, in keeping with the assumption that Elizabethan English has a head-first setting for the Head Position Parameter. However, the noun seawater undergoes a fronting/preposing operation in order to highlight it, and this means that instead of occupying its canonical position immediately after the verb drink, it is instead moved to a new position at the front of the overall sentence. Likewise, the auxiliary shalt undergoes a separate (subject–auxiliary) inversion operation which means that instead of occupying its canonical position immediately preceding its complement drink seawater, it is instead moved to a new position immediately preceding its subject thou. The effect of these two movement operations is shown schematically below: UNDERLYING ORDER: Thou shalt drink seawater SUPERfiCIAL ORDER: Seawater shalt thou drink In the underlying order, the auxiliary shalt immediately precedes its complement drink seawater, and the verb drink immediately precedes its complement seawater. But preposing seawater and Workbook section 35 inverting shalt means that the verb drink no longer immediately precedes its complement seawater, and likewise that the auxiliary shalt no longer immediately precedes its complement drink seawater. The main theoretical point which our discussion here illustrates is that word-order parameters determine the underlying order of constituents rather than their superficial order (which may be disrupted by movement operations). A point of incidental interest to note in relation to sentence 1 is that inversion was not just restricted to interrogative sentences in Elizabethan English, but could also take place in declaratives and other types of sentence. Moreover (as you will see from other examples in this exercise), it could affect main verbs as well as auxiliary verbs. ● Exercise 1.2 Below are examples of utterances produced by a girl called Lucy at age twenty-four months. Comment on whether Lucy has correctly set the three parameters discussed in the text (the Head Position Parameter, the Wh-Parameter and the Null Subject Parameter). Discuss the significance of the relevant examples for the parameter-setting model of acquisition. CHILD SENTENCE ADULT COUNTERPART 1 What doing? ‘What are you doing?’ 2 Want bye-byes ‘I want to go to sleep’ 3 Daddy play with me (w) ‘Daddy played with me’; this was in reply to ‘What did Daddy do in the park yesterday?’ 4 Mummy go shops ‘Mummy went to the shops’; this was in reply to ‘Where did Mummy go?’ 5 Where Daddy gone? (w) ‘Where’s Daddy gone?’ 6 Gone office ‘He’s gone to the office’ 7 Me have yoghurt? ‘Can I have a yoghurt?’ 8 Daddy doing? (w) ‘What’s Daddy doing?’ 9 Cry (w) ‘(I) cry’; this was in reply to ‘What do you do when Daddy gets cross with you?’ 10 I play (w) ‘I play’; this was in reply to ‘What do you do in the park?’ 11 Whatmehaving? ‘What am I having?’; this followed her mother saying ‘Mummy’s having fish for dinner’ 12 No me have fish ‘I’m not going to have fish’ 13 Want bickies ‘She wants some biscuits’; this was her reply to ‘What does Dolly want?’ 14 What Teddy have? ‘What can Teddy have?’ 15 Where going? ‘Where are you going?’ 16 What Nana eating? ‘What’s Grandma eating?’ 17 Dolly gone? ‘Where’s Dolly gone?’ 18 Watch te’vision ‘I’m going to watch television’ 19 Me have more ‘I want to have some more’ 20 Open door ‘Open the door!’ Helpful hints If Lucy has correctly set the Wh-Parameter, we should expect to find that she systematically preposes wh-expressions and positions them sentence-initially. If she has correctly set the Head 36 i GRAMMAR Position Parameter, we should expect to find (e.g.) that she correctly positions the complement of a verb after the verb, and the complement of a preposition after the preposition; however, where the complement is a wh-expression, we expect to find that the complement is moved into sentence-initial position in order to satisfy the requirements of the Wh-Parameter (if the Wh-Parameter in some sense overrides the Head Position Parameter). If Lucy has correctly set the Null Subject Parameter, we should expect to find that she does not use null subjects in finite clauses: however, it seems clear that many of the sentences produced by two-year old English children like Lucy do indeed have null subjects – and this led Nina Hyams in influential research (1986, 1992) to conclude that English children go through a null subject stage in which they use Italian-style null (pro) subjects in finite clauses. If Hyams is right, this implies that children may sometimes start out with incorrect settings for a given parameter, and then later have to reset the parameter – a conclusion which (if true) would provide an obvious challenge to the simple parameter-setting model of acquisition outlined in the main text. However, the picture relating to the use of null subjects is complicated by the fact that although English does not have finite null subjects (i.e. the kind of null pro subject found in finite clauses in languages like Italian), it has three other types of null subject. One is the kind of imperative null subject found in imperatives such as Shut up! and Don’t say anything! (Imperatives are sentences used to issue orders; they are the kind of sentences you can put please in front of – as in Please don’t say anything!) Another is the kind of nonfinite null subject found in a range of nonfinite clauses in English (i.e. clauses containing a verb which is not marked for tense and agreement), including main clauses like Why worry? and complement clauses like those bracketed in I want [to go home] and I like [playing tennis]: the kind of null subject found in nonfinite clauses in English is usually designated as PRO and called ‘big PRO’ (whereas the kind of null subject found in a finite clause in a null-subject language like Italian is designated as pro and called ‘little pro’. The terms big and little here simply reflect the fact that PRO is written in ‘big’ capital letters, and pro in ‘small’ lower-case letters). A third type of null subject found in English can be called a truncated null subject, because English has a process of truncation which allows one or more words at the beginning of a sentence to be truncated (i.e. omitted) in certain types of style (e.g. diary styles of written English and informal styles of spoken English). Hence in colloquial English, a question like Are you doing anything tonight? can be reduced (by truncation) to You doing anything tonight? and further reduced (again by truncation) to Doing anything tonight? Truncation is also found in abbreviated written styles of English: for example, a diary entry might read Went to a party. Had a great time. Got totally smashed (with the subject I being truncated in each of the three sentences). An important constraint on truncation is that it can only affect words at the beginning of a sentence, not e.g. words in the middle of a sentence: hence, although we can truncate are and you in Are you doing anything tonight?, we can’t truncate them in What are you doing tonight? (as we see from the ungrammaticality of ∗What doing tonight?) since here are and you are preceded by what and hence occur in the middle of the sentence. What all of this means is that in determining whether Lucy has mis-set the Null Subject Parameter and has misanalysed English as a null subject language (i.e. a language which allows the kind of finite null ‘little pro’ subjects found in Italian), you have to bear in mind the alternative possibility that the null subjects used by Lucy may represent one or more of the three kinds of null subject permitted in adult English (viz. imperative null subjects, truncated null subjects and nonfinite null subjects). Since truncation occurs only sentence-initially (at the beginning of a sentence), but finite null (little pro) subjects in a genuine null subject language like Italian can occur in any subject position Workbook section 37 in a sentence, one way of telling the difference between a finite null subject and a truncated null subject is to see whether children omit subjects only when they are the first word in a sentence (which could be the result of truncation), or whether they also omit subjects in the middle of sentences (as is the case in a genuine null-subject language like Italian). Another way of differentiating the two is that in null subject languages like Italian with null finite pro subjects, we find that overt pronoun subjects are only used for emphasis, so that in an Italian sentence like L’ho fatto io (literally ‘It have done I’) the subject pronoun io ‘I’ has a contrastive interpretation, and the relevant sentence is paraphraseable in English as ‘I was the one who did it’ (where italics indicate contrastive stress): by contrast, in a non-null-subject language like English, subject pronouns are not intrinsically emphatic – e.g. he doesn’t necessarily have a contrastive interpretation in an English diary-style sentence such as Went to see Jim. Thought he might help. A third way of telling whether truncation is operative in Lucy’s grammar or not is to see whether expressions other than subjects can be truncated, as can happen in adult English (e.g. What time is it? can be reduced to Time is it? via truncation in rapid spoken English). At first sight, it might seem unlikely that (some of) Lucy’s null subjects could be nonfinite (‘big PRO’) null subjects, since all the clauses she produces in the data given above occur in finite contexts (i.e. in contexts where adults would use a finite clause). Note, however, that two-year-old children typically go through a stage which Wexler (1994) calls the Optional Infinitives/OI stage, during which (in finite contexts) they sometimes produce finite clauses, and sometimes nonfinite clauses (the relevant nonfinite clauses typically containing an infinitive form like go or a participle like going/gone). Hence, an additional possibility to bear in mind is that some of Lucy’s clauses may be nonfinite and have nonfinite (‘big PRO’) null subjects. In relation to the sentences in 1–20, make the following assumptions. In 1 doing is a verb which has a null subject and the complement what. In 2 want is a verb which has a null subject and the complement bye-byes. In 3 play is a verb which has the subject Daddy and the complement with me (and in turn me is the complement of the preposition with). In 4 go is a verb which has the subject Mummy and the complement shops. In 5 gone is a verb which has the subject Daddy and the complement where. In 6 gone is a verb which has a null subject and the complement office. In 7 have is a verb which has the subject me and the complement yoghurt. In 8 doing is a verb which has the subject Daddy, and its complement is a null counterpart of what. In 9 cry is a verb with a null subject. In 10, play is a verb and I is its subject. In 11, having is a verb which has the subject me and the complement what. In 12 no is a negative particle which has the complement me have fish (assume that no is the kind of word which doesn’t have a subject), and have is a verb which has the subject me and the complement fish. In 13 want is a verb which has a null subject and the complement bickies. In 14 have is a verb which has the subject Teddy and the complement what. In 15 going is a verb which has a null subject and the complement where. In 16 eating is a verb which has the subject Nana and the complement what. In 17 gone is a verb which has the subject Dolly and its complement is a null counterpart of where. In 18 watch is a verb which has a null subject and the complement te’vision. In 19 have is a verb which has the subject me and the complement more. In 20 open is a verb whose subject is null and whose complement is door. Model answer for 1 In What doing? the verb doing has an overt object what and a null subject of some kind. Since the object what does not occupy the normal postverbal position associated with objects in English (cf. the position of the object something in Do something!), what has clearly undergone wh-movement: this suggests that Lucy has correctly set the Wh-Parameter at the ‘requires 38 i GRAMMAR wh-movement’ value appropriate for English. Because the object complement what has undergone wh-movement, we cannot tell (from this sentence) whether Lucy generally positions (unmoved) complements after their heads: in other words, this particular sentence provides us with no evidence of whether Lucy has correctly set the Head Position Parameter or not (though other examples in the exercise do). Much more difficult to answer is the question of whether Lucy has correctly set the Null Subject Parameter at the value appropriate to English, and hence (tacitly) ‘knows’ that finite clauses do not allow a null finite pro subject in English. At first sight, it might seem as if Lucy has wrongly analysed English as a null subject language (and hence mis-set the Null Subject Parameter), since What doing? has a null subject of some kind. But the crucial question here is: What kind of null subject does the verb doing have? It clearly cannot be an imperative null subject, since the sentence is interrogative in force, not imperative. Nor can it be a truncated null subject, since truncated subjects only occur in sentence-initial position (i.e. as the first word in a sentence), and what is the first word in the sentence in What doing? (since preposed wh-words occupy sentence-initial position in questions). This leaves two other possibilities. One is that the null subject in What doing? is the ‘little pro’ subject found in finite clauses in genuine null subject languages like Italian: since the verb doing is nonfinite, this would entail positing that the sentence What doing? contains a null counterpart of the finite auxiliary are (raising questions about why the auxiliary is null rather than overt); this in turn would mean that Lucy has indeed mis-set the Null Subject Parameter (raising questions about how she comes to do so, and why she doesn’t mis-set the other two parameters we are concerned with here). However, an alternative possibility is that the structure What doing? is a nonfinite clause (like adult questions such as Why worry?) and has the kind of nonfinite (‘big PRO’) null subject found in nonfinite clauses in many languages (English included). If so (i.e. if What doing is a nonfinite clause which has the structure What PRO doing?), there would be no evidence that Lucy has mis-set the Null Subject Parameter – i.e. no evidence that she ever produces finite clauses with a ‘little pro’ subject. This in turn would mean that we can maintain the hypothesis put forward in the main text that children correctly set parameters at their appropriate value from the very earliest stages of the acquisition of syntax. The error Lucy makes in producing sentences like What doing? would be in not knowing that main clauses generally have to be finite in English, and that main clause questions generally have to contain a finite auxiliary. 2 Structure ● ● 2.1 Overview In this chapter, we introduce the notion of syntactic structure, look- ing at how words are combined together to form phrases and sentences. We shall see that phrases and sentences are built up by a series of merger operations, each of which combines a pair of constituents together to form a larger constituent. We show how the resulting structure can be represented in terms of a tree diagram. We look at some of the principles which underlie sentence formation, and we explore ways of testing the structure of phrases and sentences. ● ● 2.2 Phrases To put our discussion on a concrete footing, let’s consider how an elementary two-word phrase such as the italicised response produced by speaker B in the following mini-dialogue is formed: (1) SPEAkER A: What are you trying to do? SPEAkER B: Help you As speaker B’s utterance illustrates, the simplest way of forming a phrase is by merging (a technical term meaning ‘combining’) two words together: for example, by merging the word help with the word you in (1), we form the phrase help you. The resulting phrase help you seems to have verb-like rather than pronoun-like properties, as we see from the fact that it can occupy the same range of positions as the simple verb help, and hence e.g. occur after the infinitive particle to: cf. (2) (a) We are trying to help (b) We are trying to help you By contrast, the phrase help you cannot occupy the same kind of position as a pronoun such as you, as we see from (3) below: (3) (a) You are very difficult (b) ∗Help you are very difficult 39 40 2 STRUCTURE So, it seems clear that the grammatical properties of a phrase like help you are determined by the verb help, and not by the pronoun you. Much the same can be said about the semantic properties of the expression, since the phrase help you describes an act of help, not a kind of person. Using the appropriate technical terminology, we can say that the verb help is the head of the phrase help you, and hence that help you is a Verb Phrase: and in the same way as we abbreviate category labels like verb to V, so too we can abbreviate the category label Verb Phrase to VP. If we use the traditional labelled bracketing technique to represent the category of the overall verb phrase help you and of its constituent words (the verb help and the pronoun you), we can represent the structure of the resulting phrase as in (4) below: (4) [VP [V help] [PRN you]] An alternative (equivalent) way of representing the structure of phrases like help you is via a labelled tree diagram such as (5) below (which is a bit like a family tree diagram – albeit for a small family): (5) VP V PRN help you What the tree diagram in (5) tells us is that the overall phrase help you is a Verb Phrase (VP), and that its two constituents are the verb (V) help and the pronoun (PRN) you. The verb help is the head of the overall phrase (and so is the key word which determines the grammatical and semantic properties of the phrase help you). Introducing another technical term at this point, we can say that, con- versely, the VP help you is a projection of the verb help, in the sense that the verb helpis projected into a larger structure by merging it with another constituent of an appropriate kind. In this case, the constituent which is merged with the verb help is the pronoun you, which has the grammatical function of being the (direct object) complement of the verb help. The head of a projection/phrase deter- mines grammatical properties of its complement: in this instance, since help is a transitive verb, it requires a complement with accusative case (e.g. a pro- noun like me/us/him/them), and this requirement is satisfied here since you can function as an accusative form (as you can see from the table of pronouns listed under the entry for Case in the Glossary at the end of the book). The tree diagram in (5) is entirely equivalent to the labelled bracketing in (4), in the sense that the two provide us with precisely the same information about the structure of the phrase help you. The differences between a labelled bracketing like (4) and a tree diagram like (5) are purely notational: each category is represented by a single labelled node in a tree diagram (i.e. by a point in the tree which carries a category label like VP, V or PRN), but by a pair of labelled brackets in a labelled bracketing. 2.2 Phrases 41 Since our goal in developing a theory of Universal Grammar is to uncover general structural principles governing the formation of phrases and sentences, let’s generalise our discussion of (5) at this point and hypothesise that all phrases are formed in essentially the same way as the phrase in (5), namely by a binary (i.e. pairwise) merger operation which combines two constituents together to form a larger constituent. In the case of (5), the resulting phrase help you is formed by merging two words. However, not all phrases contain only two words – as we see if we look at the structure of the italicised phrase produced by speaker B in (6) below: (6) SPEAkER A: What was your intention? SPEAkER B: To help you The phrase in (6B) is formed by merging the infinitive particle to with the Verb Phrase help you. What’s the head of the resulting phrase to help you? A rea- sonable guess would be that the head is the infinitival tense particle/T to, so that the resulting expression to help you is an infinitival TP (= infinitival tense projection = infinitival tense phrase). This being so, we’d expect to find that TPs containing infinitival to occur in a different range of positions from VPs/Verb Phrases – and this is indeed the case, as we see from the contrast below: (7) (a) They ought to help you (= ought + TP to help you) (b) ∗They ought help you (= ought + VP help you) (8) (a) They should help you (= should + VP help you) (b) ∗They should to help you (= should + TP to help you) If we assume that help you is a VP whereas to help you is a TP, we can account for the contrasts in (7) and (8) by saying that ought is the kind of word which selects (i.e. ‘takes’) an infinitival TP as its complement, whereas should is the kind of word which selects an infinitival VP as its complement. Implicit in this claim is the assumption that different words like ought and should have different selectional properties which determine the range of complements which they can take. The infinitive phrase to help you is formed by merging the infinitive particle to with the Verb Phrase help you. If (as we argued in the previous chapter) infinitival to is an infinitival tense particle (belonging to the category T) and if to is the head of the phrase to help you, the structure formed by merging to with the Verb Phrase/VP help you in (5) will be the TP in (9) below: (9) TP T VP to V PRN help you 42 2 STRUCTURE The head of the resulting infinitival tense projection to help you is the infinitive particle to, and the Verb Phrase help you is the complement of to; conversely, to help you is a projection of to. In keeping with our earlier observation that ‘The head of a projection/phrase determines grammatical properties of its com- plement,’ the infinitival tense particle to requires an infinitival Verb Phrase as its complement: more specifically, to requires the head V of its VP complement to be a verb in its infinitive form, so that we require the (bare/uninflected) infinitive form help after infinitival to (and not an inflected form like helping/helped/helps). Refining our earlier observation somewhat, we can therefore say that ‘The head of a projection/phrase determines grammatical properties of the head word of its complement.’ In (9), to is the head of the TP to help you, and the complement of to is the VP help you; the head of this VP is the V help, so that to determines the form of the V help (requiring it to be in the infinitive form help). More generally, our discussion here suggests that we can build up phrases by a series of binary merger operations which combine successive pairs of constituents to form ever larger structures. For example, by merging the infinitive phrase to help you with the verb trying, we can form the even larger italicised phrase trying to help you produced by speaker B in (10) below: (10) SPEAkER A: What are you doing? SPEAkER B: Trying to help you The resulting phrase trying to help you is headed by the verb trying, as we see from the fact that it can be used after words like be, start or keep, which select a complement headed by a verb in the -ing form (cf. They were/started/kept trying to help you). This being so, the italicised phrase produced by speaker B in (10) is a VP (= Verb Phrase) which has the structure (11) below: (11) VP V TP trying T VP to V PRN help you (11) tells us (amongst other things) that the overall expression trying to help you is a Verb Phrase/VP; its head is the verb/V trying, and the complement of trying is the TP/infinitival tense phrase to help you: conversely, the VP trying to help you is a projection of the V trying. An interesting property of syntactic structures which is illustrated in (11) is that of recursion – that is, the property of allowing a given structure to contain more than one instance of a given category (in this case, more than one Verb Phrase/VP – one VP headed by the verb help and another VP headed by the verb trying). Since our goal in developing a theory of Universal Grammar/UG is to attempt to establish universal principles governing the nature of linguistic structure, an 2.2 Phrases 43 important question to ask is whether there are any general principles of constituent structure which we can abstract from structures like (5, 9, 11). If we look closely at the relevant structures, we can see that they obey the following two (putatively universal) constituent structure principles: (12) Headedness Principle Every nonterminal node in a syntactic structure is a projection of a head word (13) Binarity Principle Every nonterminal node in a syntactic structure is binary-branching (A terminal node is one at the foot/bottom of a tree, whereas a nonterminal node is one which branches down into other nodes: consequently, the V-trying, T-to, V-help and PRN-you in (11) are terminal nodes because they do not branch down into any other node; by contrast, the VP and TP constituents are nonterminal because they branch down into other nodes.) For example, the structure (11) obeys the Headedness Principle (12) in that the VP help you is headed by the V help, the TP to help you is headed by the T to, and the VP trying to help you is headed by the V trying. Likewise, (11) obeys the Binarity Principle (13) in that the VP help you branches into two immediate constituents (in the sense that it has two constituents immediately beneath it, namely the V help and the PRN you), the TP to help you branches into two immediate constituents (the nonfinite tense particle T to and the VP help you) and the VP trying to help you likewise branches into two immediate constituents (the V trying and the TP to help you). Our discussion thus leads us towards a principled account of constituent structure – i.e. one based on a set of principles of Universal Grammar. There are several reasons for trying to uncover constituent structure principles like (12) and (13). From a learnability perspective, such principles reduce the range of alternatives which children have to choose between when trying to determine the structure of a given kind of expression: they therefore help us develop a more constrained theory of syntax. Moreover, additional support for the Binarity Principle comes from evidence that phonological structure is also binary, in that (e.g.) a syllable like bat has a binary structure, consisting of the onset |b| and the rhyme |at|, and the rhyme in turn has a binary structure, consisting of the nucleus |a| and the coda |t| – vertical bars being used to enclose sounds or sequences of sounds. Likewise, there is evidence that morphological structure is also binary, and hence (e.g.) that the noun indecipherability is formed by adding the prefix de- to the noun cipher to form the verb decipher; then adding the suffix -able to this verb to form the adjective decipherable; then adding the prefix in- to this adjective to form the adjective indecipherable; and then adding the suffix -ity to the resulting adjective to form the noun indecipherability. It would thus seem that binarity is an inherent characteristic of the phonological, morphological and syntactic structure of natural languages: as our discussion develops below, we shall uncover empirical evidence in support of the claim that syntactic structure is indeed binary. 44 2 STRUCTURE ● ● 2.3 Clauses Having considered how phrases are formed, let’s now turn to look at how clauses and sentences are formed. By way of illustration, suppose that speaker B had used the simple (single-clause) sentence italicised in (14) below to reply to speaker A, rather than the phrase used by speaker B in (10): (14) SPEAkER A: What are you doing? SPEAkER B: We are trying to help you What’s the structure of the italicised clause produced by speaker B in (14)? In work in the 1960s, clauses were generally taken to belong to the category S (Sentence/Clause), and the sentence produced by B in (14) would have been taken to have a structure along the following lines: (15) S PRN T VP We are V TP trying T VP to V PRN help you However, a structure such as (15) violates the two constituent structure principles which we posited in (12) and (13) above. More particularly, the S analysis of clauses in (15) violates the Headedness Principle (12) in that the S We are trying to help you is a structure which has no head of any kind. Likewise, the S analysis in (15) also violates the Binarity Principle (13) in that the S constituent We are trying to help you is not binary-branching but rather ternary-branching, because it branches into three immediate constituents, namely the PRN we, the T are and the VP trying to help you. If our theory of Universal Grammar requires every syntactic structure to be a binary-branching projection of a head word, it is clear that we have to reject the S analysis of clause structure in (15) as one which is not in keeping with UG principles. Let’s therefore explore an alternative analysis of the structure of clauses which is consistent with the headedness and binarity requirements in (12) and (13). More specifically, let’s make the unifying assumption that clauses are formed by the same binary merger operation as phrases. This in turn will mean that the italicised clause in (14B) is formed by merging the (present) tense auxiliary are with the Verb Phrase trying to help you, and then subsequently merging the resulting expression are trying to help you with the pronoun we. Since are belongs to the category T of tense auxiliary, it might at first sight seem as if merging are with the Verb Phrase trying to help you will derive (i.e. form) the 2.3 Clauses 45 tense projection/tense phrase/TP are trying to help you. But this can’t be right, since it would provide us with no obvious account of why speaker B’s reply in (16) below is ungrammatical: (16) SPEAkER A: What are you doing? SPEAkER B: ∗Are trying to help you If Are trying to help you is a complete TP, how come it can’t be used to answer A’s question in (16), since we see from sentences like (6B) that TP constituents like to help you can be used to answer questions. An informal answer we can give is to say that the expression Are trying to help youis somehow ‘incomplete’, and that only ‘complete’ expressions can be used to answer questions. In what sense is Are trying to help you incomplete? The answer is that finite (e.g. present/past tense) T constituents require a subject, and the finite auxiliary are doesn’t have a subject in (16). More specifically, let’s assume that when we merge a tense auxiliary (= T) with a Verb Phrase (= VP), we form an intermediate projection which we shall here denote as T (pronounced ‘tee- bar’); and that only when we merge the relevant T-bar constituent with a subject like we do we form a maximal projection – or, more informally a ‘complete TP’. Given these assumptions, the italicised clause in (14B) will have the structure (17) below: (17) TP PRN T' We T VP are V TP trying T VP to V PRN help you What this means is that a tense auxiliary like are has two projections: a smaller intermediate projection (T ) formed by merging are with its complement trying to help you to form the T-bar (intermediate tense projection) are trying to help you; and a larger maximal projection (TP) formed by merging the resulting T are trying to help you with its subject we to form the TP We are trying to help you. Saying that TP is the maximal projection of are in (17) means it is the largest constituent headed by the auxiliary are. Why should tense auxiliaries require two different projections, one in which they merge with a following complement to form a T-bar, and another in which the resulting T-bar merges with a preceding subject to form a TP? The requirement for tense auxiliaries to have two projections (as in (17) above) was taken by Chomsky in earlier work to be a consequence of a principle of Universal Grammar known as 46 2 STRUCTURE the Extended Projection Principle (conventionally abbreviated to EPP), which specified that a finite tense constituent T must be extended into a TP projection containing a subject. However, comparative evidence suggests that there are other languages in which a tense auxiliary does not require a preceding subject of its own – as illustrated by an Italian sentence such as: (18) `E stata arrestata una vecchia signora Is been arrested an old lady ‘An old lady has been arrested’ The absence of any subject preceding the present tense auxiliary `e ‘is’ in (18) suggests that it cannot be a principle of Universal Grammar that every tense auxiliary in every finite clause in every language has a subject. Rather, it seems more likely that this is a property of tense auxiliaries in particular languages like English, but not of their counterparts in some other languages. Recall that we noted in §1.2 that the grammatical properties of categories of word are traditionally described in terms of sets of features, and by convention these are enclosed in square brackets. So, for example, in order to describe the grammatical properties of the auxiliary are (in a sentence like They are lying) as a third person plural present progressive auxiliary, we could say that it carries the features [third- person, plural-number, present-tense, progressive-aspect]. Using this convention, Chomsky suggested in later work that English tense auxiliaries like are carry an epp feature which requires them to have an extended projection into a TP containing a subject. If all finite auxiliaries in English carry an EPP feature, it follows that any English clause structure (like that produced by speaker B in (16) above) containing a tense auxiliary which does not have a subject will be ungrammatical. The EPP requirement (for a finite auxiliary in English to have a subject) would seem to be essentially syntactic (rather than semantic) in nature, as we can see from sentences such as (19) below: (19) (a) It was alleged that he lied under oath (b) There has been no trouble In structures like (19), the italicised subject pronouns it/there seem to have no semantic content (in particular, no referential properties) of their own, as we see from the fact that neither can be questioned by the corresponding interrogative words what?/where? (cf. the ungrammaticality of ∗What was alleged that he lied under oath?and ∗Where has been no trouble?), and neither can receive contrastive focus (hence it/there cannot be contrastively stressed in sentences like 19 above). Rather, they function as expletive pronouns – i.e. pronouns with no intrinsic meaning which are used to satisfy the syntactic requirement for a finite auxiliary like was/has to have a subject. It is interesting to note that theoretical considerations also favour a binary- branching TP analysis of clause structure like (17) over a ternary-branching S 2.3 Clauses 47 analysis like (15). The essential spirit of Minimalism is to reduce the theoretical apparatus which we use to describe syntactic structure to a minimum. Within this spirit, it has generally been assumed that tree diagrams should only contain information about hierarchical structure (i.e. containment/constituent structure relations), not about linear structure (i.e. left-to-right word order), because linear information is redundant (in the sense that it can be predicted from hierarchical structure by simple word-order rules) if we use binary-branching trees. Suppose, for example, that we have a word-order rule for English to the effect that ‘Any constituent of a phrase which is directly merged with the head word of the phrase is positioned to the right of the head, but any other constituent of the phrase is positioned to the left of the head.’ This word-order rule will correctly predict (inter alia) that the VP trying to help you in (17) must be positioned to the right of the tense auxiliary/T are (because the relevant VP is directly merged with the T-head are), and that the pronoun we must be positioned to the left ofare (because we is not merged with T-are but rather with the T-bar are trying to help you). As you can see for yourself, it’s not clear how we can achieve the same result (of eliminating redundant word-order information from trees) under a ternary- branching analysis like (15), since both the pronounwe and the Verb Phrase trying to help you are merged with T-are in (15). It should be noted in passing that an important consequence of assuming that linear order is not a syntactic relation is that it entails that syntactic operations cannot be sensitive to word order (e.g. we can’t handle subject–auxiliary agreement by saying that a finite auxiliary agrees with a preceding noun or pronoun expression): rather, all syntactic operations must be sensitive to hierarchical rather than linear structure. On this view (and assuming the overall structure of a grammar in diagram 16 of §1.3), word order is a PF property (i.e. a property assigned to constituents in the PF component on the basis of linearisation rules like that sketched informally above). An interesting implication of the analysis of clause structure we have presented here is that heads can have more than one kind of projection: e.g. the tense auxiliary are in (17) above has an intermediate (T-bar) projection into are trying to help youand a maximal (TP) projection intoWe are trying to help you. The same is true of other types of head, as can be illustrated by the italicised expressions below: (20) (a) American intervention in Iraq has caused considerable controversy (b) She arrived at the solution quite independently of me (c) He must go straight to bed (d) Nobody expected the film to have so dramatic an ending In (20a) the noun intervention merges with its complement in Iraq to form the intermediate projection (N-bar) intervention in Iraq, and the resulting N-bar in turn merges with the adjective American to form the maximal projection (NP) American intervention in Iraq. In (20b) the adverb independently merges with its complement of me to form the intermediate projection (ADV-bar) independently of me, and this in turn merges with the adverbquiteto form the maximal projection 48 2 STRUCTURE (ADVP) quite independently of me. In (20c) the preposition to merges with its complement bed to form the intermediate (P-bar) projection to bed, and this in turn merges with the adverb straight to form the maximal (PP) projection straight to bed. In (20d), the quantifier (indefinite article) an merges with its complement ending to form the intermediate (Q-bar) projection an ending which in turn merges with the expression so dramatic to form the maximal projection (QP) so dramatic an ending. In clause structures like (17) above, the pronoun we which merges with the intermediate T-bar projection are trying to help you to form the maximal TP projection We are trying to help you has the function of being the subject of the TP. However, the expressions which merge with the relevant intermediate projections to form maximal projections in (20) don’t always have the function of being subjects. If we take a fairly flexible view of what a subject is, we could perhaps say that the adjective American is the ‘subject’ of the expression intervention in Iraq in (20a) by virtue of denoting the entity perpetrating the act of intervention. But we certainly wouldn’t want to say that quite is the subject of independently of me in (20b), or that straight is the subject of to bed in (20c), or that so dramatic is the subject of an ending in (20d). Rather, the expressions which precede the head word in the examples in (20b-d) seem to have the function of being modifiers of the expression that follows them – so that quite modifies independently of me, straight modifies to bed and so dramatic modifies an ending (and perhaps American modifies intervention in Iraq in 20a). What our discussion here illustrates is that it is important to draw a distinction between the position occupied by an expression in a given structure, and its function. In order to get a clearer view of the distinction, let’s take a closer look at the derivation of (20c) He must go straight to bed. As we noted earlier, the preposition to merges with its noun complement bed to form the P-bar to bed which in turn is merged with the adverb straight to form the PP straight to bed. The resulting PP is then merged with the verb go to form the VP go straight to bed. This in turn is merged with the present tense auxiliary must to form the T-bar must go straight to bed. This T-bar merges with the pronoun he to form the TP below: (21) TP PRN T' He T VP must V PP go ADV P' straight P N to bed 2.4 Clauses containing complementisers 49 In a fairly obvious sense, the pronoun he occupies the same kind of position within TP as the adverb straight does within PP: more specifically, he merges with an intermediate T-bar projection to form a maximal TP projection in the same way as straight merges with an intermediate P-bar projection to form a maximal PP projection. Since it’s useful to have a term to designate the position they both occupy, let’s say that they both occupy the specifier position within the expression containing them. More specifically, let’s say that he occupies the specifier position within the T-projection (conventionally abbreviated to spec-T or spec-TP) and that straight occupies the specifier position within PP (= spec-P or spec-PP). However, although he and straight occupy the same specifier posi- tion within the expressions containing them, they have different functions: he is the subjectof the T-bar expressionhas gone to bed, whereasstraight is a modifier of the P-bar expression to bed. In much the same way, we can say that American occupies the specifier position within the Noun Phrase American intervention in Iraq in (20a), quite occupies the specifier position within the Adverbial Phrase quite independently of me in (20b) and so dramatic occupies the specifier position within the Quantifier Phrase so dramatic an ending in (20d). ● ● 2.4 Clauses containing complementisers A question which we have not so far asked about the structure of clauses concerns what role is played by complementisers like that, for and if, e.g. in speaker B’s reply in (22) below: (22) SPEAkER A: What are you saying? SPEAkER B: That we are trying to help you Where does the C/complementiser that fit into the structure of the sentence? The answer suggested in work in the 1970s was that a complementiser merges with an S constituent like that in (15) above to form an S /S-bar (pronounced ‘ess- bar’) constituent like that shown below (simplified by not showing the internal structure of the VP trying to help you, which is as in (11) above): (23) S' C S that PRN T VP we are trying to help you However, the claim that a clause introduced by a complementiser has the status of an S-bar constituent falls foul of the Headedness Principle (12), which requires every nonterminal node in a tree to be a projection of a head word. The principle is violated because S-bar in (23) is analysed as a projection of the S constituent we are trying to help you, and S is clearly not a word (but rather a string of words). 50 2 STRUCTURE An interesting way round the headedness problem is to suppose that the head of a clausal structure introduced by a complementiser is the complementiser itself: since this is a single word, there would then be no violation of the Headedness Principle (12) requiring every syntactic structure to be a projection of a head word. Let’s therefore assume that the complementiser that merges with the TP we are trying to help you (whose structure is shown in (17) above) to form the CP/complementiser projection/complementiser phrase in (24) below: (24) CP C TP That PRN T' we T VP are V TP trying T VP to V PRN help you (24) tells us that the complementiser that is the head of the overall clause that we are trying to help you (and conversely, the overall clause is a projection of the complementiser that) – and indeed this is implicit in the traditional description of such structures as that-clauses. (24) also tells us that the complement of that is the TP/tense phrasewe are trying to help you. Clauses introduced by complementisers have been taken to have the status of CP/complementiser phrase constituents since the early 1980s. An interesting aspect of the analyses in (17) and (24) above is that clauses and sentences are analysed as headed structures – i.e. as projections of head words (in conformity with the Headedness Principle). In other words, just as phrases are projections of a head word (e.g. a verb phrase like help you is a projection of the verb help), so too a sentence like We will help you is a projection of the auxiliary will, and a complement clause like the bracketed that-clause in I can’t promise [that we will help you] is a projection of the complementiser that. This enables us to arrive at a unitary analysis of the structure of phrases, clauses and sentences, in that clauses and sentences (like phrases) are projections of head words. More generally, it leads us to the conclusion that clauses/sentences are simply particular kinds of phrase (e.g. a that-clause is a complementiser phrase). An assumption which is implicit in the analyses which we have presented here is that phrases and sentences are derived (i.e. formed) in a bottom-up fashion (i.e. they are built up from bottom to top). For example, the clause in (24) involves the following sequence of merger operations: (i) the verb help is merged with the pronoun you to form the VP help you; (ii) the resulting VP is merged with the nonfinite T/tense particle to to form the TP to help you; (iii) this TP is in turn 2.5 Testing structure 51 merged with the verb trying to form the VP trying to help you; (iv) the resulting VP is merged with the T/tense auxiliary are to form the T-bar are trying to help you; (v) this T-bar is merged with its subject we to form the TP we are trying to help you; and (vi) the resulting TP is in turn merged with the C/complementiser that to form the CP structure (24) that we are trying to help you. By saying that the structure (24) is derived in a bottom-up fashion, we mean that lower parts of the structure nearer the bottom of the tree are formed before higher parts of the structure nearer the top of the tree. As those of you familiar with earlier work will have noticed, the kind of struc- tures we are proposing here are very different from those assumed in traditional grammar and in work in linguistics in the 1960s and 1970s. Earlier work implic- itly assumed that only items belonging to substantive/lexical categories could project into phrases, not words belonging to functional categories. More specif- ically, earlier work assumed that there were Noun Phrases headed by nouns, Verb Phrases headed by verbs, Adjectival Phrases headed by adjectives, Adverbial Phrases headed by adverbs and Prepositional Phrases headed by prepositions. However, more recent work has argued that not only content words but also func- tion words can project into phrases, so that we have Tense Phrases headed by a tense-marker, Complementiser Phrases headed by a complementiser, Determiner Phrases headed by a determiner – and so on. More generally, the assumption made in work over the last three decades is that in principle all word-level categories can project into phrases. This means that some of the structures we make use of here may seem (at best) rather strange to those of you with a more traditional background, or (at worst) just plain wrong. However, the structure of a given phrase or sentence cannot be determined on the basis of personal prejudice or pedagogical precepts inculcated into you at secondary school, but rather has to be determined on the basis of syntactic evidence of the kind discussed in the next section below. I would therefore ask traditionalists to be open to new ideas and new analyses (a prerequisite for understanding in any discipline). ● ● 2.5 Testing structure Thus far, we have argued that phrases and sentences are built up by merging successive pairs of constituents into larger and larger structures, and that the resulting structure can be represented in terms of a labelled tree diagram. The tree diagrams which we use to represent syntactic structure make specific claims about how sentences are built up out of various different kinds of constituent (i.e. syntactic unit): hence, trees can be said to represent the constituent structure of sentences. But this raises the question of how we know (and how we can test) whether the claims made about syntactic structure in tree diagrams are true. So far, we have relied mainly on intuition in analysing the structure of sentences – we have in effect guessed at the structure. However, it is unwise to rely on intuition in attempting to determine the structure of a given expression in 52 2 STRUCTURE a given language. For, while experienced linguists over a period of years tend to acquire fairly strong intuitions about structure, novices by contrast tend to have relatively weak, uncertain and unreliable intuitions; moreover, even the intuitions of supposed experts may ultimately turn out to be based on little more than personal preference. For this reason, it is more satisfactory (and more accurate) to regard constituent structure as having the status of a theoretical construct. That is to say, it is part of the theoretical apparatus which linguists find they need to make use of in order to explain certain observations about language (just as molecules, atoms and subatomic particles are constructs which physicists find they need to make use of in order to explain the nature of matter in the universe). It is no more reasonable to rely wholly on intuition to determine syntactic structure than it would be to rely on intuition to determine molecular structure. Inevitably, then, much of the evidence for syntactic structure is of an essentially empirical character, based on the observed grammatical properties of particular types of expression. The evidence typically takes the form ‘If we posit that such-and- such an expression has such-and-such a constituent structure, we can provide a principled account of the observed grammatical properties of the expression.’ Thus, structural representations ultimately have to be justified in empirical terms, i.e. in terms of whether or not they provide a principled account of the grammatical properties of phrases and sentences. In order to make our discussion more concrete, we’ll look at how we can test the structure of the following sentence: (25) The chairman has resigned from the board Let’s suppose that (25) is derived as follows. The determiner the is merged with the noun board to form the DP the board. This DP is merged with the preposition from to form the PP from the board. The resulting PP is merged with the verb resigned to form the VP resigned from the board. This VP is then merged with the auxiliary has to form the T-bar has resigned from the board. This T-bar is in turn merged with its subject/specifier the chairman (which is a DP formed by merging the determiner the with the noun chairman), thereby forming the TP shown in (26) below: (26) TP DP T' D N T VP The chairman has V PP resigned P DP from D N the board 2.5 Testing structure 53 The tree diagram in (26) is a representation of (what we take to be) the structure of (25) The chairman has resigned from the board. However, a tree diagram like (26) has the status of a hypothesis (i.e. untested and unproven assumption) about the structure of the relevant sentence. How can we test our hypothesis and determine whether (26) is or isn’t an appropriate representation of the structure of the sentence? The answer is that there are a number of standard heuristics (i.e. ‘tests’) which we can use to determine struc- ture: we shall discuss just three of these here. The first relates to the phenomenon of co-ordination. English and other languages have a variety of co-ordinating conjunctions like and/but/or which can be used to co-ordinate (= conjoin = join together) expressions such as those bracketed below: (27) (a) [fond of cats] and [afraid of dogs] (b) [slowly] but [surely] (c) [to go] or [to stay] In each of the expressions in (27), an italicised co-ordinating conjunction has been used to conjoin the bracketed pairs of expressions. Clearly, any adequate grammar of English will have to provide a principled answer to the question: ‘What kinds of strings (i.e. sequences of words) can and cannot be co-ordinated?’ Now, it turns out that we can’t just co-ordinate any random set of strings, as we see by comparing the grammatical reply produced by speaker B in (28) below: (28) SPEAkER A: What does he do to keep fit? SPEAkER B: Run up the hill and up the mountain with the ungrammatical reply produced by speaker B in (29) below: (29) SPEAkER A: What did he do about his bills? SPEAkER B: ∗Ring up the phone company and up the electricity company Why should it be possible to co-ordinate the string up the hill with the string up the mountain in (28), but not possible to co-ordinate the string up the phone company with the string up the electricity company in (29)? We can provide a principled answer to this question in terms of constituent structure: the italicised string up the hill in (28) is a constituent of the phrase run up the hill (up the hill is a Prepositional Phrase, in fact), and so can be co-ordinated with another similar type of Prepositional Phrase (e.g. a PP such as up the mountain, or down the hill or along the path etc.). Conversely, however, the string up the phone company in (29) is not a constituent of the phrase ring up the phone company, and so cannot be co-ordinated with another similar string like up the electricity company. (Traditional grammarians say that up is associated with ring in expressions like ring up someone, and that the expression ring up forms a kind of complex verb which carries the sense of ‘telephone’.) On the basis of contrasts such as these, we can formulate the following generalisation: (30) Co-ordination Condition Only constituents of the same type can be co-ordinated 54 2 STRUCTURE A constraint (i.e. principle imposing restrictions on certain types of grammatical operation) along the lines of (30) is assumed in much work in traditional grammar. Having established the condition (30), we can now make use of it as a way of testing the tree diagram in (26) above. In this connection, consider the data in (31) below (in which the bracketed strings have been co-ordinated by and): (31) (a) The chairman has resigned from [the board] and [the company] (b) The chairman has resigned [from the board] and [from the company] (c) The chairman has [resigned from the board] and [gone abroad] (d) The chairman [has resigned from the board] and [is living in Utopia] (e) ∗The [chairman has resigned from the board] and [company has replaced him] (f) [The chairman has resigned from the board] and [the company has replaced him] (31a) provides us with evidence in support of the claim in (26) that the board is a Determiner Phrase constituent, since it can be co-ordinated with another DP like the company; similarly, (31b) provides us with evidence that from the board is a Prepositional Phrase constituent, since it can be co-ordinated with another PP like from the company; likewise, (31c) provides evidence that resigned from the board is a Verb Phrase constituent, since it can be co-ordinated with another VP like gone abroad; in much the same way, (31d) provides evidence that has resigned from the board is a T-bar constituent, since it can be co-ordinated with another T like is living in Utopia (thereby providing interesting evidence in support of the binary-branching structure assumed in the TP analysis of clauses, and against the ternary-branching analysis assumed in the S analysis of clauses); and in addition, (31f) provides evidence that the chairman has resigned from the board is a TP constituent, since it can be co-ordinated with another TP like the company has replaced him. Conversely, however, the fact that (31e) is ungrammatical suggests that (precisely as (26) claims) the string chairman has resigned from the board is not a constituent, since it cannot be co-ordinated with a parallel string like company has replaced him (and the constraint in (30) tells us that two strings of words can only be co-ordinated if both are constituents – and, more precisely, if both are constituents of the same type). Overall, then, the co-ordination data in (31) provide empirical evidence in support of the analysis in (26). A second way of testing structure is to use a substitution test. The assumption underlying this test is that a string of words is a constituent if it can be substituted by a single word. In this connection, consider: (32) (a) The chairman has resigned from the board, and he is now living in Utopia (b) The press say that the chairman has resigned from the board, and so he has (c) %If the Managing Director says the chairman has resigned from the board, he must have done (d) If the chairman has resigned from the board (which you say he has), how come his car is still in the company car park? 2.5 Testing structure 55 (The percentage sign in front of 32c indicates that this type of structure is only found in certain varieties of English – notably, British English.) In each of the above sentences, the italicised string can be replaced (or referred back to) by a particular kind ofproform. (Recall that a proform is a function word that can stand ‘in place of’ some other expression.) The fact that the expression the chairman in (32a) can be substituted by a single word (in this case, the proform/pronoun he) provides evidence in support of the claim in (26) that the chairman is a single constituent (a DP/determiner phrase, to be precise). Likewise, the fact that the expression resigned from the board in (32b,c,d) can serve as the antecedent of the proforms so/done/which provides evidence in support of the claim in (26) that resigned from the board is a constituent (more precisely, a VP/verb phrase). Unfortunately, since English has a very limited inventory of proforms, this test is of limited usefulness. A third way of testing structure is to use a preposing test. The core assump- tion underlying this test is that only a string of words forming a constituent can be preposed (and thereby moved to the front of the structure containing it) in order to highlight it in some way (e.g. in order to mark it out as a topic containing familiar/old information, or as a focused constituent containing unfa- miliar/new information). However, there are restrictions on the kind of constituent which can be highlighted by preposing – as can be illustrated by the following contrast: (33) (a) Straight to bed he must go (b) ∗To bed he must go straight As we see from the structure in (21) above, both straight to bed and to bed are constituents of sentence (20c) He must go straight to bed. So how come only the larger of these two constituents can be preposed? The answer would appear to be that straight to bed is a PP and hence a maximal projection, whereas to bed (when modified by straight) is a P-bar constituent, and hence an intermediate projection. And it would seem (from contasts like that in 33) that only a maximal projection can undergo the relevant kind of preposing operation. This being so, one way we can test whether a given expression is a maximal projection or not is by seeing whether it can be preposed. In this connection, consider the following sentence: (34) The press said that the chairman would resign from the board, and resigned from the board he has The fact that the italicised expression resigned from the board can be preposed in (34) indicates that it must be a maximal projection: this is consistent with the analysis in (26) which tells us that resigned from the board is a Verb Phrase which is the maximal projection of the verb resigned. However, an important caveat which should be noted in relation to the prepos- ing test is that particular expressions can sometimes be difficult (or even impos- sible) to prepose even though they are maximal projections. This is because 56 2 STRUCTURE there are constraints (i.e. restrictions) on such movement operations. One such constraint can be illustrated by the following contrast: (35) (a) He resolutely refused to surrender to the enemy (b) Surrender to the enemy, he resolutely refused to (c) ∗To surrender to the enemy, he resolutely refused Here, the VP/Verb Phrase surrender to the enemy can be highlighted by being pre- posed, but the TP/infinitival tense phrase to surrender to the enemy cannot – even though it is a maximal projection (by virtue of being the largest expression headed by infinitival to). What is the nature of the restriction on preposing to+infinitive expressions illustrated by the ungrammaticality of (35c)? The answer is not clear, but may be semantic in nature. When an expression is preposed, this is in order to highlight its semantic content in some way (e.g. for purposes of contrast – as in e.g. ‘Syntax, I don’t like but phonology I do’). It may be that infinitival to has no intrinsic lexical semantic content, and that this makes an infinitival to-phrase an unsuitable candidate for highlighting. If so, this suggests that when preposing material for highlighting purposes, we should prepose as few words as possible. This requirement would seem to be related to Grice’s (1975) ‘Be concise’ maxim (which amounts to ‘Use as few words as possible’), and we can conflate the two together in terms of the following more general condition: (36) Economy Condition Syntactic structures should contain as few words as possible, and syntactic operations should affect as few words as possible Given that only a maximal projection can be preposed for highlighting purposes, it follows from the Economy Condition that the following (more specific) condition will hold on preposing: (37) Preposing Condition When material is preposed in order to highlight it, what is preposed is the smallest possible maximal projection containing the highlighted material So, if we want to highlight the semantic content of the VP surrender to the enemy, we prepose the VP surrender to the enemy rather than the TP to surrender to the enemy because the VP is smaller than the TP containing it. However, this is by no means the only constraint on preposing, as we see from (38) below (where FBA is an abbreviation for the Federal Bureau of Assassina- tions – a purely fictitious body, of course): (38) (a) Nobody had expected that the FBA would assassinate the king of Ruritania (b) ∗King of Ruritania, nobody had expected that the FBA would assassinate the (c) The king of Ruritania, nobody had expected that the FBA would assassinate (d) ∗The FBA would assassinate the king of Ruritania, nobody had expected that (NB that = |ðət|) (e) That the FBA would assassinate the king of Ruritania, nobody had expected 2.5 Testing structure 57 The ungrammaticality of (38b,d) tells us that we can’t prepose the NP King of Ruritaniaor the TPthe FBA would assassinate the King of Ruritania. Why should this be? One possibility is that there is a constraint on movement operations to the effect that a DP can be preposed but not an NP contained within a DP, and likewise that a CP can be preposed but not a TP contained within a CP. One implementation of this idea would be to posit a constraint like (39) below: (39) Functional Head Constraint/FHC The complement of a certain type of functional head F (such as a determiner or complementiser) cannot be moved on its own (without also moving F) Suppose, then, that we want to highlight the NP king of Ruritania in (38a) by preposing. (37) tells us to move the smallest possible maximal projection containing the highlighted material, and hence we first try to move this NP on its own: but the Functional Head Constraint tells us that it is not possible to prepose this NP on its own, because it is the complement of the determiner the. We therefore prepose the next smallest maximal projection containing the hightlighted NP king of Ruritania – namely the DP the king of Ruritania; and as the grammaticality of (38c) shows, the resulting sentence is grammatical. Now suppose that we want to highlight the TP the FBA would assassinate the king of Ruritania. (37) tells us to move the smallest maximal projection containing the highlighted material – but the FHC (39) tells us that we cannot prepose a constituent which is the complement of a complementiser. Hence, we prepose the next smallest maximal projection containing the TP which we want to highlight, namely the CP that the FBA would assassinate the King of Ruritania – as in (38e). However, an apparent problem for the Functional Head Constraint (39) is posed by examples like: (40) (a) Surrender to the enemy, I never will (b) Surrender to the enemy, he resolutely refused to The preposed Verb Phrase surrender to the enemy is the complement of will in (40a), and the complement of to in (40b). Given the analysis in §1.2, will is a finite T/tense auxiliary and to is a nonfinite T/tense particle. If (as we have assumed so far) T is a functional category, we would expect the Functional Head Constraint (39) to block preposing of the VP surrender to the enemy because this VP is the complement of the functional T constituent will/to. The fact that the resulting sentences (40a,b) are grammatical might lead us to follow Chomsky (1999) in concluding that T is a substantive category rather than a functional category, and hence does not block preposing of its complement. Alternatively, it may be that the constraint only applies to certain types of func- tional category (as hinted at in 39) – e.g. D and C but not T (perhaps because D and C are the ‘highest’ heads within nominal and clausal structures respec- tively – and indeed in chapter 9 we shall reformulate this constraint along such lines). 58 2 STRUCTURE It is interesting to note that alongside sentences like (40) above in which a phrase has been highlighted by being preposed, we also find sentences like (41) below in which a single word has been preposed: (41) (a) Surrender, I never will (b) Surrender, he resolutely refused to In (41) the verb surrender has been preposed on its own. At first sight, this might seem to contradict our earlier statement that only maximal projections can undergo preposing. However, more careful reflection shows that there is no contradiction here: after all, the maximal projection of a head H is the largest expression headed by H; and in a sentence like I never will surrender, the largest expression headed by the verb surrender is the verb surrender itself – hence, sur- render in (41) is indeed a maximal projection. This provides another illustration of a point noted earlier – namely that an individual word can itself be a maximal projection, if it has no complement or specifier of its own. The overall conclusion to be drawn from our discussion here is that the prepos- ing test has to be used with care. If an expression can be preposed in order to highlight it, it is a maximal projection; if it cannot, this may either be because it is not a maximal projection, or because (even though it is a maximal projection) a syntactic constraint of some kind prevents it from being preposed, or because its head word has insufficient semantic content to make it a suitable candidate for highlighting. ● ● 2.6 Structural relations and the syntax of polarity items Throughout this chapter, we have argued that phrases and sentences are formed by a series of binary merger operations, and that the resulting structures can be represented in the form of tree diagrams. Because they mark the way that words are combined together to form phrases of various types, tree diagrams are referred to in earlier technical work as phrase-markers (abbreviated to P- markers). They show us how a phrase or sentence is built up out of constituents of various types: hence, a tree diagram provides a visual representation of the constituent structure of the corresponding expression. Each node in the tree (i.e. each point in the tree which carries a category label like N, V, A , T , PP, CP etc.) represents a different constituent of the sentence; hence, there are as many different constituents in any given phrase-marker as there are nodes carrying category labels. As we saw earlier, nodes at the very bottom of the tree are called terminal nodes, and other nodes are nonterminal nodes: so, for example, all the D, N, T, V and P nodes in (26) are terminal nodes, and all the DP, PP, VP, T and TP nodes are nonterminal nodes. The topmost node in any tree structure (i.e. TP in the case of (26) above) is said to be its root. Each terminal node in the tree carries a single lexical item (i.e. an item from the lexicon/dictionary, like dog or go etc.): lexical items are sets of phonological, semantic and grammatical properties/features (with category labels like N, V, T, 2.6 Structural relations and the syntax of polarity items 59 C etc. being used as shorthand abbreviations for the set of grammatical features carried by the relevant items). It is useful to develop some terminology to describe the syntactic relations between constituents, since these relations turn out to be central to syntactic description. Essentially, a P-marker is a graph comprising a set of points (= labelled nodes), connected by branches (= solid lines) representing containment relations (i.e. telling us which constituents contain or are contained within which other constituents). We can illustrate what this means in terms of the following abstract tree structure (where A, B, C, D, E, F, G, H and J are different nodes in the tree, representing different constituents): (42) A B E C D F G H J In (42), G immediately contains H and J (and conversely H and J are the two constituents immediately contained within G, and hence are the two immediate constituents of G): this is shown by the fact that H and J are the two nodes immediately beneath G which are connected to G by a branch (solid line). Likewise, E immediately contains F and G; B immediately contains C and D; and A immediately contains B and E. We can also say that E contains F, G, H and J; and that A contains B, C, D, E, F, G, H and J (and likewise that G contains H and J; and B contains C and D). Using equivalent kinship terminology, we can say that A is the mother of B and E (and conversely B and E are the two daughters of A); B is the mother of C and D; E is the mother of F and G; and G is the mother of H and J. Likewise, B and E are sisters (by virtue of both being daughters of A) – as are C and D; F and G; and H and J. A particularly important syntactic relation is c-command (a conventional abbreviation of constituent-command), which provides us with a useful way of determining the relative position of two different constituents within the same tree (in particular, whether one is lower in the tree than the other or not). We can define this relation informally as follows (where X, Y and Z are three different nodes): (43) c-command A constituent X c-commands its sister constituent Y and any constituent Z which is contained within Y For those of you who find it difficult to conceptualise such abstractions, a more concrete way of visualising this is to think of a tree diagram as representing a network of train stations, with each of the labelled nodes representing the name of a different station in the network, and the branches representing the rail tracks linking the stations. We can then say that one node X c-commands another node Y if you can get from X to Y on the network by taking a northbound train, getting 60 2 STRUCTURE off at the first station, changing trains there and then travelling one or more stops south on a different line. In the light of the definition of c-command given above, let’s consider which constituents each of the nodes in (42) c-commands. A doesn’t c-command any of the other nodes, since A has no sister. B c-commands E, F, G, H and J because B’s sister is E, and E contains F, G, H and J. C c-commands only D, because C’s sister is D, and D does not contain any other constituent; likewise, D c-commands only C. E c-commands B, C and D because B is the sister of E and B contains C and D. F c-commands G, H and J, because G is the sister of F and G contains H and J. G c-commands only F, because G’s sister is F, and F does not contain any other constituents. H and J likewise c-command only each other because they are sisters which have no daughters of their own. We can illustrate the importance of the c-command relation in syntactic descrip- tion by looking at the distribution of a class of expressions which are known as polarity items. These are items which have an inherent ‘polarity’ in the sense that they are restricted to occurring in certain types of sentence. In this connection, consider the quantifier any in English (and its compounds like anyone, anything, anywhere etc.). It has two different uses. One is as a universal (or free choice) quantifier with a meaning similar to every (as in You can have any cake you like): in this use, the initial a of any is stressed, and the relevant word is not a polarity item. The second use of any is as a partitive (or existential) quantifier: in this use, it has a meaning similar to some and can be unstressed (with its initial vowel reduced to schwa or even being truncated in rapid colloquial speech styles – e.g. He wouldn’t do ’nything to help me), and in this second use it is indeed a polarity item. As the sentences in (44) below illustrate, partitive any (and its compounds) can occur in a clause containing a (bold-printed) interrogative expression like how often in (44a), a negative expression like no student in (44b) or a conditional expression like if in (44c) – but not in a positive declarative clause like (44d): (44) (a) I wonder how often we find any morality in business (b) No student will complain about anything (c) If anyone should ask for me, say I’ve gone to lunch (d) ∗I’d like any coffee, please Klima (1964, p. 313) conjectured that negative, interrogative and conditional expressions share ‘a common grammatico-semantic feature to be referred to as affective’. In his terms, expressions like how often, no student and if are all ‘affective’ constituents (by which he seems to mean that they are non-assertive – or nonveridical, to employ the alternative term used by Giannikidou (1997, 1998, 1999)). Using Klima’s terminology, we can suppose that a polarity item such as (partitive) any is restricted to occurring in a structure containing an affective constituent. It turns out that numerous other expressions (italicised below) are similarly restricted to occurring in a structure containing a (bold-printed) affective constituent: 2.6 Structural relations and the syntax of polarity items 61 (45) (a) I didn’t think I would ever pass the exam (b) ∗I thought I would ever pass the exam (46) (a) Nobody dare contradict him (b) ∗Everybody dare contradict him (47) (a) I don’t think he need apologise (b) ∗He need apologise Curiously, the items need and dare are polarity items when they function as auxiliaries (and so do not take the third person singular present tense s-affix, and are not followed by infinitival to), but not when they function as main verbs (e.g. in sentences like Professor Knutter needs to see a psychiatrist). It might at first sight seem as if we can characterise the relevant restriction on the use of polarity items by saying that they can only be used after (i.e. when they follow) an affective constituent. However, any such claim is falsified by contrasts such as the following: (48) (a) The fact that he has resigned won’t change anything (b) ∗The fact that he won’t resign will change anything In both (48a) and (48b), the polarity itemanythingfollows a bold-printed affective item (the negative auxiliarywon’t), and yet only (48a) is grammatical. Why should this be? The answer is that (as originally noted by Klima), polarity items like partitive any are subject to a structural condition on their use which can be characterised in the following terms: (49) Polarity Condition A polarity item must be c-commanded by an affective (e.g. negative, interrogative or conditional) constituent To see how this works, consider whether the polarity item anything satisfies the Polarity Condition (49) in a structure such as the following: (50) TP DP T D NP T VP The won’t/*will N CP V PRN fact change anything C TP that PRN T' he T V hasn’t resigned ' Consider first of all whether the pronoun anything is c-commanded by the bold- printed T-auxiliary will/won’t. Since the sister of [T will/won’t] is the VP change anything, and since anything is one of the constituents of this VP, it follows that 62 2 STRUCTURE [T will/won’t] does indeed c-command [PRN anything] under the definition of c- command given in (43) above. What this means is that using the negative auxiliary won’t in (50) will satisfy the Polarity Condition (49), so correctly predicting that the corresponding sentence (51) below is grammatical: (51) The fact that he hasn’t resigned won’t change anything But now consider whether [PRN anything] is c-commanded by the underlined T-auxiliary hasn’t in (50). The sister of [T hasn’t] is the V-constituent resigned, and it is clear that anything is not a constituent of the verb resigned, and hence that [T hasn’t] does not c-command anything in (50). The Polarity Condition (49) therefore (correctly) predicts the ungrammaticality of a sentence such as the following: (52) ∗The fact that he hasn’t resigned will change anything The overall conclusion which our discussion here leads us to is thus that restric- tions on the use of polarity items can be given a structural characterisation in terms of the relation c-command. ● ● 2.7 The c-command condition on binding A second class of expressions whose distribution can be given a princi- pled characterisation in tems of the relation c-command are so-called anaphors. These include reflexives (i.e. self/selves forms like myself/yourself/themselves etc.), and reciprocals like each other and one another. Such anaphors have the property that they cannot be used to refer directly to an entity in the outside world, but rather must by bound by (i.e. take their reference from) an antecedent elsewhere in the same phrase or sentence. Where an anaphor has no (suitable) antecedent to bind it, the resulting structure is ungrammatical – as we see from contrasts such as that in (53) below: (53) (a) He must feel proud of himself (b) ∗She must feel proud of himself (c) ∗Himself must feel proud of you In (53a), the third person masculine singular anaphor himself is bound by a suitable third person masculine singular antecedent (he), with the result that (53a) is grammatical. But in (53b), himself has no suitable antecedent (the feminine pronoun she is not a suitable antecedent for the masculine anaphor himself), and so is unbound (with the result that (53b) is ill-formed). In (53c), there is no antecedent of any kind for the anaphor himself, with the result that the anaphor is again unbound and the sentence ill-formed. There are structural restrictions on the binding of anaphors by antecedents, as we see from: (54) (a) The president may blame himself (b) ∗Supporters of the president may blame himself 2.7 The c-command condition on binding 63 (55) (a) They may implicate each other (b) ∗The evidence against them may implicate each other As a third person masculine singular anaphor, himself must be bound by a third person masculine singular antecedent like the president; similarly, as a plural anaphor, each other must be bound by a plural antecedent like they/them. However, it would seem from the contrasts above that the antecedent must occupy the right kind of position within the structure in order to bind the anaphor or else the resulting sentence will be ungrammatical. The question of what is the right position for the antecedent can be answered in terms of the following structural condition: (56) Binding Condition A bound constituent must be c-commanded by an appropriate antecedent The relevant bound constituent is the reflexive anaphor himself in (54), and its antecedent is the president; the bound constituent in (55) is the reciprocal anaphor each other, and its antecedent is they/them. Sentence (54a) has the structure (57) below: (57) TP DP T' D N T VP The president may V PRN blame himself The reflexive pronoun himself can be bound by the DP the president in (57) because the sister of the DP node is the T-bar node, and the pronoun himself is one of the constituents of the relevant T-bar node: consequently, the DP the president c-commands the anaphor himself and the Binding Condition (56) is satisfied. We therefore correctly specify that (54a) The president may blame himself is grammatical, with the president interpreted as the antecedent of himself. But now consider why a structure like (58) below is ungrammatical (cf. (54b) above): (58) TP NP T' N PP T VP Supporters may P DP V PRN of blame himself D N the president The answer is that the DP node containing the president doesn’t c-command the PRN node containing himself, because the sister of the DP node is the P node of, and himself is clearly not a constituent of the preposition of. Since there is no 64 2 STRUCTURE other appropriate antecedent for himself within the sentence (e.g. although the NP supporters of the president c-commands himself, it is not a suitable antecedent because it is a plural expression, and himself requires a singular antecedent), the anaphor himself remains unbound – in violation of the Binding Condition on anaphors. This is the reason why (54b) ∗Supporters of the president may blame himself is ungrammatical. Our brief discussion of polarity items and anaphors in this section and the last underlines the importance of the relation c-command in syntax. It also provides further evidence for positing that sentences have a hierarchical constituent struc- ture, in that the Polarity Condition (49) and the Binding Condition (56) are both characterised in structural terms. ● ● 2.8 Bare phrase structure In this chapter, we have used a system of category labels based on the bar notation which has been widely adopted since the 1970s. Within this framework, a sentence like (the title of Gloria Gaynor’s immortal song) I will survive has the structure shown below: (59) TP PRN T' I T V will survive The bar notation used in (59) posits that there are three different levels of projec- tion (i.e. types of expression): (i) heads (also called minimal projections) like the T/tense auxiliary will; (ii) intermediate projections like the T-bar will sur- vive; and (iii)maximal projections like the TP I will survive. However, Chomsky (1999, p. 2) argues that a system which posits three different types of category label for projections of a given head H (viz. H, H-bar and HP) violates a UG prin- ciple called the Inclusiveness Condition which specifies that no new information can be introduced in the course of the syntactic computation – a principle which Chomsky (2006, p. 4) claims to be ‘a natural principle of efficient computation’. The reason why the bar notation used in trees like (59) violates Inclusiveness is as follows. When the word will is taken out of the lexicon, its lexical entry specifies that it has a set of properties which include the grammatical properties represented by the category label T in (59). But the tree in (59) tells us that when will is merged with its complement survive, the resulting string will survive belongs to the category T-bar – in other words, it is an intermediate projec- tion of will. Likewise, the tree in (59) also tells us that the larger string I will survive is a TP – in other words, it is the maximal projection of will. But this information about intermediate and maximal projections is not part of the lexical entry for will, and hence has seemingly been added in the course of the syntactic 2.8 Bare phrase structure 65 computation. However, adding such information about projection levels violates the Inclusiveness Condition. One way of avoiding violation of Inclusiveness is to remove all information about projection levels from trees, and hence replace a tree like (59) above by one like (60) below: (60) T PRN T I T V will survive What our revised tree (60) says is that will, will survive and I will survive are all projections of the tense auxiliary will and hence are all tense expressions. Information about projection levels is omitted in (60) because it is redundant, since it is predictable from looking at the relative positions of constituents within a given structure. Simply by looking at the positions they occupy in the tree (60), we can tell that will is the minimal projection of will (i.e. it is the smallest expression headed by will), that will survive is an intermediate projection of will (by virtue of being neither the smallest nor the largest expression headed by will) and that I will survive is the maximal projection of will (by virtue of being the largest expression headed by will). Similarly, we can tell that the V survive is both a minimal and a maximal projection, in that it is both the smallest and the largest expression headed by survive: hence (e.g.) it can behave like a maximal projection and undergo preposing (as in Survive, I will). In much the same way, we know from looking at the structure in (60) that the pronoun I is likewise both a minimal and a maximal projection: given their status as maximal projections, it follows that pronouns can undergo preposing (as with the pronoun him in Him, I would never trust). Since the information about projection levels in the bar notation is redundant, Chomsky reasons, such information should not be represented in the system of category labels used in tree diagrams: after all, the goal of Minimalism is to reduce theoretical apparatus to the minimum that is conceptually necessary. Chomsky goes even further and argues in favour of a theory of bare (i.e. category-free) phrase structure in which the nodes in trees do not carry category labels. We can illustrate the kind of reasoning behind his thinking in the following terms. In a category-based theory of syntax, the grammatical properties of a pronoun like he are described by assigning it to the category PRN of pronoun. But simply telling us that he belongs to the category PRN does not characterise its other grammatical properties – e.g. the fact that it is a third person expression, it is singular in number, masculine in gender and nominative in case. The traditional way of describing grammatical properties like these is in terms of a set of features like [third-person], [singular-number], [masculine-gender], [nominative-case], with grammatical features conventionally enclosed within square brackets. But Chomsky (1965, 1970) argued that the categorial properties of words can also be 66 2 STRUCTURE described in terms of sets of grammatical features: one such feature might indicate that he is nominal (rather than verbal) in nature, and another might indicate that it is a function word (rather than a content word). If all the grammatical properties of words (including their categorial properties) can be described in terms of sets of grammatical features, the possibility arises that category labels can be entirely replaced by sets of features, so opening up the possibility of developing a theory of bare phrase structure – i.e. a theory in which there are no category labels in syntactic trees. A radical possibility along these lines would be for the structure of I will survive to be represented in terms of an unlabelled tree diagram like (61) below (i.e. a tree containing no category labels): (61) I will survive An unlabelled tree diagram like (61) tells us that the constituents of (61) are I, will, survive, will survive and I will survive. The lexical entries for the items I, will and survive comprise sets of features which include information about their grammatical and selectional properties: e.g. the entry for will tells us that it is a finite auxiliary which selects an infinitival complement. The fact that will selects an infinitive complement (and that survive is an infinitive form and is the sister of will) means that survive must be the complement of will and hence that will survive is a projection of will. Likewise, the fact that will has an EPP feature requiring it to project a subject means that the nominative pronoun I must be the subject of will, and hence that I will survive is an extended projection of will. As before, the relative position of the relevant constituents within the overall structure tells us that will is a minimal projection (of itself), will survive is an intermediate projection of will, and I will survive is the maximal projection of will. The overall conclusion we arrive at is that the information about category labels and projection levels in a conventional labelled tree diagram like (59) above may well be redundant. Given that bare phrase structure is more of a leading idea than a fully developed theory and that it has not been widely adopted in descriptive work, we shall continue to use traditional labelled trees and the bar notation to represent structure, category membership and projection levels throughout the rest of this book, because this remains the notation most widely used in contemporary work in syntax. ● ● 2.9 Summary In this chapter, we have looked at how words are combined together to form phrases and sentences. In §2.2 we showed how more and more complex 2.9 Summary 67 phrases can be built up by successive binary merger operations, each of which combines a pair of constituents to form a larger constituent. In §2.3 we argued that clauses containing a finite tense auxiliary are formed by merging the tense auxiliary with a verbal complement to form an intermediate T-bar projection which is then merged with a subject to form an extended TP/tense phrase projec- tion. On this view, a sentence like It may rain would be formed by merging the present tense auxiliary may with the verb rain to form the T-bar constituent may rain, and then merging the resulting T-bar with the pronoun it to derive the TP It may rain. We also noted that the requirement for English tense auxiliaries to have a subject can be described by saying that a T-auxiliary in English has an epp feature requiring it to have an extended phrasal projection containing a subject. Introducing a new term, we said that the subject occupies the specifier position within TP, and that specifiers are constituents which merge with an intermediate projection to form a maximal projection. We noted that other kinds of constituent can also have specifiers, so that (e.g.) straight occupies the specifier position within a Prepositional Phrase like straight to bed. In §2.4 we argued that clauses introduced by a complementiser/C are formed by merging C with a TP comple- ment to form a CP/complementiser phrase. In §2.5, we looked at ways of testing constituent structure, outlining tests relating to co-ordination, substitution and preposing. We noted that a variety of factors can sometimes prevent constituents from being preposed in order to highlight them; for example, only maximal pro- jections can be highlighted via preposing, and phrases headed by items with little or no substantive lexical semantic content generally cannot be preposed: more- over, there are also syntactic restrictions on preposing, in that the Functional Head Constraint bars the complement of a determiner or complementiser from being moved on its own. In §2.6, we looked at the structural relations between constituents within tree diagrams, noting that the relation c-command plays a central role in accounting for the syntax of polarity items. In §2.7, we went on to show that the relation c-command is also central to any account of the binding properties of anaphors. In §2.8 we discussed the potential redundancy in the system of labels used to represent categories and projection levels in traditional phrase structure trees, and noted that Chomsky has been seeking to develop a theory of bare phrase structure in recent work. For those of you familiar with work in traditional grammar, it will be clear that the assumptions made about syntactic structure within the Minimalist frame- work are somewhat different from those made in traditional grammar. Of course, there are some similarities: within both types of framework, it is assumed that lexical categories project into phrases, so that by combining a noun with one or more other constituents we can form a Noun Phrase, and likewise by combin- ing a verb/preposition/adjective/adverb with one or more other constituents we can form a Verb Phrase/Prepositional Phrase/Adjectival Phrase/Adverbial Phrase. But there are two major differences between the two types of framework. One is that Minimalism (unlike traditional grammar) assumes that function words also project into phrases (so that by combining a determiner/D with a noun 68 2 STRUCTURE expression we form a Determiner Phrase/DP, by combining a (present or past tense) auxiliary/T with a complement and a subject we form a Tense Projec- tion/TP, and by combining a complementiser with a TP we form a Complemen- tiser Projection/CP). This in some cases results in an analysis which is rather different from that found in traditional grammar, so that (for example) the nose would be considered a Noun Phrase in traditional grammar, but is taken to be a Determiner Phrase within the framework adopted here. A further difference between the two frameworks is that Minimalism assumes that all syntactic struc- ture is binary-branching, whereas traditional grammar (implicitly) does not. Key principles/conditions introduced in this chapter include the following: (12) Headedness Principle Every nonterminal constituent in a syntactic structure is a projection of a head word (13) Binarity Principle Every nonterminal constituent in a syntactic structure is binary-branching (30) Co-ordination Condition Only constituents of the same type can be co-ordinated (36) Economy Condition Syntactic structures should contain as few words as possible, and syntactic operations should affect as few words as possible (37) Preposing Condition When material is preposed in order to highlight it, what is preposed is the smallest possible maximal projection containing the highlighted material (39) Functional Head Constraint/FHC The complement of a certain type of functional head F (such as a determiner or complementiser) cannot be moved on its own (without also moving F) (49) Polarity Condition A polarity item must be c-commanded by an affective (e.g. negative, interrogative or conditional) constituent (56) Binding Condition A bound constituent must be c-commanded by an appropriate antecedent An important relation introduced in the chapter is the following: (43) c-command A constituent X c-commands its sister constituent Y and any constituent Z which is contained within Y Recall that – using a train metaphor which treats the nodes in a tree as stations on a train network – we characterised c-command rather more informally by saying that a node X c-commands another node Y if you can get from X to Y by taking a northbound train, getting off at the first station/node, changing trains there and then travelling one or more stops/nodes south on a different line.