clac 16/ 2003


Amaya Mendikoetxea

Universidad Autónoma de Madrid

amaya mendikoetxea at uam es


pdf version




Descriptive and theoretical linguistics have often been presented as fields with conflicting interests and goals.  There is an area, however, in which the interface between these two fields is proving to be fruitful and enriching: the writing of modern descriptive grammars. My purpose in this paper is to explore the relation between theory and description and the choices linguists must make when writing modern descriptive grammars.  I focus on some aspects of the content of a new reference grammar of the English language, The Cambridge Grammar of the English Language written by R. Huddleston and G. K. Pullum (H & P, henceforth) and a number of collaborators,[2] while reflecting on the process of writing grammars at the beginning of the 21st century.  Indeed, The Cambridge Grammar of the English Language (CambGR, henceforth) has been marketed by its publishers as the grammar of the 21st century, intended to challenge the well-established position occupied by Quirk et al.’s  A Comprehensive Grammar of the English Language (CompGR, henceforth).[3] My main concern here is to examine critically the conceptual framework within which the linguistic description is carried out in CambGR, especially those aspects in which it differs from CompGR, rather than looking at particular analyses for a variety of constructions.  To highlight some of these differences and to illustrate how grammatical structures are analysed in CambGR, I am going to focus on the treatment of verb complementation in CambGR, as presented mostly in chapter 4 (for phrasal complements) and chapters 11 and 14 (for clausal complements). This will be done after some discussion on the intricate relation between theory and description in general and the type of descriptive work carried out in CambGR.


I have chosen to focus on verb complementation and not, for instance, nouns and noun phrases or lexical-word formation.  Since it is obvious that verbs are central to syntactic analysis and, thus, it is to the description of verbs and their complementation that CambGR devotes a lot of its effort. Though chapter 3 is devoted to the verb (covering mostly issues to do with tense), there is no chapter titled “Verbs and verb phrases” (as opposed, for instance, to chapter 5 “Nouns and noun phrases” and chapter 7 “Prepositions and preposition phrases”), since decisions concerning the analysis of verbs impregnate so many other aspects of linguistic description (e.g. the analysis of complex clauses).  Thus, issues to do with the syntax and semantics of verbs appear as independent items of content in a number of chapters (for instance, chapters 4 & 11, mentioned above).  Finally, from a pedagogical point of view, the analysis of verbs and their complements must be dealt with in detail in any descriptive course on English grammar, and those aspects of it on which we focus here, such as the distinction between phrasal and prepositional verbs and the types of non-finite clausal complements are, or should be, absolutely central in grammar courses for both native and non-native students of English.


In what follows, I first deal with general issues concerning the relation between theory and description (section 2).  I then examine some aspects of the analysis of phrasal and clausal complements of verbs in CambGR, with special attention to the analysis of infinitival complements in chapter 14 (section 3).  Finally, some pedagogical issues are briefly discussed (section 4).  The concluding remarks are in the final section of the paper.





Despite the increasing difficulty of getting funding for research on grammatical studies and the diminishing role given to the study of grammar in school and university curricula, these appear to be good times for grammars.  The end of the 20th century and the beginning of the 21st century have seen the emergence of a number of comprehensive descriptive grammars of English in the tradition set by Jespersen and Poutsma.  To CambGR and CompGR, we must add Biber et al.’s (1999) Longman Grammar of Spoken and English Language, as well as a number of reference and university grammars of more limited scope (among others, Greenbaum & Quirk 1990 and Downing & Locke 1993).  The same is true for grammar of other European languages: Grande Grammatica Italiana di Consultazione (1988-1995) for Italian,  Algemene Nederlandese Spraakkunst (1997, 2nd ed) for Dutch, Svenska Akademiens Grammatik (1999) for Swedish and the recent Gramática Descriptiva de la Lengua Española (1999) for Spanish.[4]


This could be interpreted as one of the signs of the coming of age of linguistics as a science, almost 90 years since the publication of Saussure’s Cours de Linguistique Générale.  It is surely the consequence of more than half a century of unprecedented advances in our knowledge of the structure of  languages, due in large part to the success of the Chomskyan paradigm after the 1950's.  This proliferation of grammars suggests that linguists finally feel confident to reach larger audiences and to present their ideas and their findings in a more accessible way (see sec. 5 here).  It  comes at a point when the view of linguistics as an esoteric discipline with little or no connection with the real world is widespread, especially in the fields of applied linguistics and language pedagogy.  There are many reasons for this.  As pointed out by Newmeyer (1983), the initial appeal of Chomsky’s early work was due to the fact that it “captured the imagination of scholars and pedagogues in numerous fields because it seemed likely to promote solutions to long-standing problems in every area in which language plays a role” (Newmeyer 1983: 130).  But in the early 1970's enthusiasm turned into disillusionment as it became clear that Chomsky’s Standard Theory (and Extended Standard Theory) did not immediately lend itself to fruitful application.  It was partly  the unfounded optimism and unrealistic expectations about the usefulness of the theory for areas such language teaching methodology, machine translation, and so on which led to the view that linguistic theory was concerned with the construction of artifacts with little or no connection with reality.


After several decades of research in linguistics and the development of a wide variety of approaches to the theory of language, the situation is now not as bleak as it appeared to be in the early 1970's.  The work carried out by theoretical linguists has had implications for, and has been applied to, a variety of language-related problems, such as second language learning and natural language processing, among others.   However, the idea of linguistics as a somewhat irrelevant and obscure discipline persists and many theoretical linguists themselves have contributed to that perception with their scorn for practical implications and applications of their work and their scarce interest in making their findings available to a wider audience, especially to those for whom language is the tool of their professional activity.  It is in this that the role of descriptive work of the type found in the grammars mentioned above is so very important.  The usefulness of the discipline has to be reconsidered in the light of these works, which have benefited enormously from the work carried out within the field of theoretical linguistics.  CambGR is just but the most recent example of this trend.  Let us look first at the relation between theoretical and descriptive work in linguistics and then present the characteristics of linguistic description in CambGR.


2.1. On the intricate relation between description and theory


The relation between the fields of descriptive linguistics and theoretical linguistics is still an uneasy one and has recently been the subject of some debate (see Aarts 1993).  The two fields differ regarding their goals.  The goal of a theoretical or scientific grammar (where grammar should not be understood as a ‘volume’) is to construct a theory (a model) which allows us to interpret the data in order to see how that data fits in (or not) within the general conceptual framework of the theory, whatever its orientation: generativist, structuralist, functional, cognitivist or any other approach (and their subdivisions).  Theoretical linguists provide partial analyses of some areas of the grammar, as contributions towards a particular theory of language. On the other hand, the goal of a descriptive grammar is not to validate or refute a particular theoretical construct, but to focus on the empirical data in order to provide a detailed account of the principles governing grammatical categories: their internal structure and the way they combine into larger units (words, phrases, clauses and sentences), focusing on their morphological, syntactic, semantic and discoursive properties. 


Theoretical linguists have often criticised traditional descriptive grammars for their lack of explanation for linguistic facts and for their taxonomic approaches.  The criticisms of descriptive linguists, on the other hand, have focused mostly on the use of data and the methodology employed by theoretical linguists, mostly by those working within Chomsky’s generative paradigm since the 1960's.   Regarding the data, theoretical analyses are based on what the descriptivists tend to generally consider as idealised, unreal and insufficient data.  That is, linguistic theory is concerned with an ideal speaker-listener in a homogeneous speech community (see Chomsky 1965: 3-4) and theoretical linguists build their analysis on the basis of a relatively small sample of sentences, using personal introspection as the sole source of data.  As for their methods, a hypothetical-deductive methodology of the type employed by Chomsky and his associates since the early 60’s is regarded by descriptive linguists as inadequate for language description, who use inductive approaches.[5]


Such differences have led to the view that linguistic theory and descriptive grammar should proceed independently (see, for instance, Stuurman 1989).  However, as argued out by Aarts (1993: 199),  although there seems to be a large gap between the objectives and methodology of descriptive and theoretical linguists, this gap is often made out to be wider than it really is.  Conflict arises when we try to compare and rank the work of theoreticians and descriptivists in general (and often irrational) terms, but once we abandon the idea of comparison (and conflict), it is possible to find areas in which descriptive linguists can benefit enormously from the work carried out by theoretical linguists and viceversa: though descriptivists and theorists “have diverging long-term goals, the short-term aims are less disparate: both disciplines are concerned with grammatical structure and how to characterise it.” (Aarts 1993: 200).  As Aarts (1993) notes, Chomsky himself has referred to the aims of traditional (descriptive) linguistics and theoretical (generative) linguistics as ‘complementary’, where traditional grammar is defined as that which “provides a list of exceptions (irregular verbs, etc.), paradigms and examples of regular constructions, and observations at various levels of detail and generality about the form and meaning of expressions.” (Chomsky 1986a: 6).


Aarts (1993) observes a discernible trend in works of a descriptive type not just to describe particular constructions but to justify the analyses presented.  He advocates a discipline of applied theoretical linguistics (Modern Descriptive Grammar) which “aims at an implementation of the ideas, the concepts and perhaps also to some extent the terminology  of current theoretical work” (Aarts: 1993: 206).  There is no doubt that theoretical ideas, from generative grammar, but also from functionalist and cognitivist perspectives, are being increasingly incorporated into descriptive work. Modern linguistic theory can be an important source of information for descriptive grammarians provided that theoretical proposals are ‘filtered’ in order to present transparent and elegant accounts of native speaker’s intuitions about the facts of their language, from the formalisms of syntactic analyses.[6] In fact, though it has often been said that descriptive grammars must not be tied to a particular theory if they are to be comprehensive (see 2.2.3, below), one cannot deny the influence of particular theories of grammar in modern linguistic description.  The descriptive grammars mentioned at the beginning of this section originate to bridge the gap between theory and description by incorporating many of the insights of modern theoretical linguistics, thus making the work of these linguists accessible to a wider audience.  As Greenbaum (1988: 41) points out, developments in linguistics have turned the spotlight on data previously neglected.  It should be added that new analyses have led to the viewing of grammatical problems in a new light, to the discovery of grammatical properties that had gone unnoticed, and to establishing distinctions among grammatical categories and structures traditionally grouped together, as well as to making generalizations for constructions often considered to be distinct. 


New descriptive grammars have greatly benefited from this research, and it is in this sense that the grammars mentioned at the beginning of this section can be regarded as ‘post-theoretical’, to quote the directors of the Gramática Descriptiva de la Lengua Española in their introductory chapter (Bosque & Demonte 1999: xxiii).  But the relation between description and theory is an intricate one, and though in the grammars mentioned, theory precedes description, one could also regard descriptive work of this type as ‘pre-theoretical’.  In its most common use, this term is employed to refer to largely intuitive analyses which precede an articulated theory.  But what I mean here by pre-theoretical is a type of descriptive work found in modern descriptive grammars which can be most useful and inspirational for the theoretical linguist who collaborates in the construction of a theory of language (see Bosque & Demonte 1999: xxiii).  There is, therefore, a two-way relationship between descriptive and theoretical grammars.  We will return to this issue in the concluding section.  Now let us look at how descriptive work is carried out in CambGR and how it differs from previous descriptive grammars of English, with special reference to CompGR.


2.2. Description and theory in CambGR


The aim of CambGR is set out at the beginning of chapter 1 APreliminaries@: to provide a detailed, descriptive account of present-day, international Standard English, focusing on the principles governing the construction of words, phrases, clauses and sentences. Description involves an underlying theory, no matter how minimally articulated or how implicit this theory is. The need for theory is explicitly acknowledged by the authors of CambGR in Chapter 1: “The primary goal of this grammar is to describe the grammatical principles of Present-day English rather than to defend or illustrate a theory of grammar.  But the languages human beings use are too complex to be described except by means of a theory.” (CambGR: 18).  Description makes use of generalisations and “without a theory there are no generalisations” (CambGR: 18).  In particular, general statements are needed about the way words combine to make sentences (as an alternative to listing all the sentences in a language - an impossible task), which means developing a theory about the ways sentences can be constructed in English.  The theory presupposed in CambGR is one “that classifies the words of the dictionary and specifies ways in which they are combined to form sentences” (p. 19).  That is, one which distinguishes a lexicon and some sort of computational system or grammar.  In this, and in the central role attributed to constituent structure (the idea that sentences have parts which may themselves have parts), CambGR is firmly grounded on phrase structure approaches within the generative grammar tradition.


Before we look in some detail at some aspects of linguistic description in CambGR, it is necessary to provide a brief overview of its contents, as well as some general comments on layout and design.

We have just said that chapter 1 states the goal of CambGR and provides a discussion of general issues (e.g. on the relation between theory and description), as well as more specific issues to do with basic concepts in syntax - the notion of ‘constituent structure’ being central to this grammar. Chapter 2 “Syntactic overview” is a brief survey of the fifteen chapters that deal with syntax, emphasizing those aspects in which the authors’ approach departs from traditional grammars.  This is followed by fifteen chapters that deal mostly with the syntax of English (chapters 3-17) and two chapters which are devoted to morphological matters (chapters 18 and 19).  Issues to do with orthography are dealt with in chapter 20.  This is followed by three sections: “Further reading”, “Lexical index” and “Conceptual index”.


As a large-scale reference grammar, it is not intended to be read from beginning to end.  Thus, there are plenty of cross-references to previous or following chapters.  More detailed technical explanations, which may be skipped without loss of content according to the authors, are given against a blue-shaded background.  We refer to these as the ‘blue sections’ in the pages that follow.  Explanations are illustrated with plenty of examples (numbered separately for each section in the different chapters) and 40 diagrams for sentence structure are also provided (see the index to tree diagrams on p. xiii).  The prospectus emphasizes the user-friendly design and typography of the grammar (though not all readers appear to agree on this point, see Mukherjee’s (2002a) review). 


2.2.1. The nature of syntactic analysis

A lot of space is devoted in CambGR to justifying the right analysis (as well as the right terminology!) within the descriptive framework adopted; much more than in other grammars of the like.  Many traditional claims and analyses which have been challenged by current theoretical frameworks come under scrutiny here.  As an example, the authors abandon the traditional distinction of subordinate clauses into ‘noun clauses’, ‘adjective clauses’ and ‘adverb clauses’ because it suggests a similarity between clauses and parts of speech, which is not justified (see CambGR: 19).  And though the authors make it clear that it is not their purpose to argue in favour of this or another theory of linguistic description and, when possible, they try to present the facts in a way which is neutral between competing theoretical frameworks, in fact a lot of effort appears to go into trying to persuade the reader that the descriptive analysis is the correct one under the perspective adopted in the grammar.


This does not mean that traditional grammars like Jespersen’s and Poutsma’s and, to a large extent, CompGR lack a theoretical perspective in their linguistic descriptions.  But it is fair to say that theory is mostly implicit in these works, which contrasts with the explicitness of the analyses in CambGR.  The result is a tightly woven system, with a high degree of integration between the parts, in which particular analyses for particular constructions are framed throughout the grammar, and in which there is little room for indeterminacy.[7]  Indeterminacy and ‘gradience’ are, however, present in CompGR:

“We recognize that the grammar of a language is an indeterminate system and that grammatical categories are not discrete.  Within a category (for example, the word-class of adjectives) there will be a central class that conforms to all the criteria for the category and peripheral subclasses that conform in varying degrees.  Between related categories there may be no sharp boundary but a gradient so that some subclasses of items are intermediate in the gradient between categories.” (Greenbaum 1988: 50)


A particular type of indeterminacy which the authors of CambGR  try to avoid is where there is a gradient between two analyses such that “sentences may vary in the degree to which one analysis is more appropriate than the other” (CompGR: 90).   This is the case for V(erb)s such as look at and approve of in CompGR (cf. 16.13ff), where the P(reposition)  may be considered (i) as the head of a P(repositional) P(hrase) which functions as an adverbial in sentences of the type S(ubject)-V(erb) -A(dverbial) or (ii) as part of a multi-word V with a N(oun) P(hrase) object in clauses of the type S-V-O(bject):


(1)               i.          S-V-A:            They don=t     [V approve]                  [PP of noisy parties]


ii.         S-V-O:            They don=t     [V-P approve of]           [NP noisy parties]


While there is plenty of syntactic evidence for the SVA analysis, in which of noisy parties behaves as a unit (a constituent) for a number of syntactic processes (fronting, adverb insertion and so on, cf. 16.13), the SVO analysis is grounded on the existence of passive sentences such as Noisy parties are nor approved of and matters of question-formation, in which noisy parties does not seem to behave as an A but as an O.[8]


Given the overwhelming evidence for the bracketing in (1i) CambGR treats Vs like approve of and such like as in the S-V-A analysis (though, crucially, the PP is not an regarded as an>adverbial=, but as a prepositional complement of the V).[9]  It is in aspects like this, that there is no room for indeterminacy. This, of course, is not likely to satisfy readers who would rather think of grammar as a system with a certain degree of indeterminacy and who may feel uneasy with an approach which imposes such tight restrictions on the data.  In his review, Mukherjee (2002a) favours multiple analyses along the lines of (1) and considers it to be “a general weakness of the Cambridge Grammar not to allow for such multiple analyses nor to sketch out descriptive gradients in the first place”. I would argue, however, this is a major point in favour of CambGR, since indeterminacy (or rather ‘vagueness’) of this type is more often than not a consequence of the inadequacy of the research tools employed for description.


2.2.2. The scope of linguistic description and the nature of the data

In its most common use, the term grammar refers to syntax and morphology. CambGR, like CompGR, focuses mainly on syntax, though it includes an excellent chapter on lexical word-formation (chapter 19), and another on inflectional morphology (chapter 18).  As H & P (CambGR: ch 1,  4.3)  indicate, the syntax-morphology division follows the special status of the ‘word’ as the central grammatical unit; while syntax deals with how words combine to form sentences, morphology deals with the form of words, with word formation as a process resembling the formation of larger syntactic units in some respects, but significantly different in other respects.  However, meaning and meaning relations are often present in grammatical description, as a survey of the titles of chapters in CompGR suggests (chapter 8: “The semantics and grammar of adverbials”; chapter 15 “Syntactic and semantic functions of subordinate clauses”; chapter 4 “The semantics of the verb phrase”). Semantic aspects are touched upon in almost all areas of description: “Prepositional meanings” (9.14-59); “Semantic subclassification of adjectives (7.40-44); “The articles in generic reference” (5.52-59), and so on.


Despite the stronger focus of CambGR on syntactic analysis, aspects of meaning are frequently included in the description.  As the authors claim “few would take it to be controversial that a human language such as English is in some sense a system for framing thoughts and making meaningful messages expressible, and this would make it a natural supposition that meaning and grammar would be to some extent intertwined” (CambGR: 33). A careful distinction is established between the domains of semantics (conventionally, the meaning of words and sentences independent of the context) and pragmatics (the way in which utterances are interpreted in context).  A further division is established between truth-conditional and non truth-conditional semantics.  Notions like propositions, entailments, illocutionary meaning, conventional and conversational implicature and pragmatic presupposition are all carefully defined in chapter 1.  The result overall is a more sophisticated approach to the study of meaning and meaning relations to that offered in CompGR, which, although dealing explicitly with pragmatic aspects of meaning in chapters 18 and 19, offers only rather vague semantic explanations.


CambGR follows the current trend that linguistic description should be extended to cover those aspects of meaning and meaning relations which fall within semantics, pragmatics and discourse.  This tendency is more explicit in some chapters.  Thus, chapter 10 AClause type and illocutionary force@ deals with declarative, interrogative, exclamative and imperative sentences.  Chapter 16 AInformation Packaging@ (as chapter 18 in CompGR) deals with a number of non-canonical constructions which differ from their canonical counterparts not in truth conditions or illocutionary meaning but in the way information is presented in the sentence (preposing, postposing, inversion, cleft and so on).  Finally, chapter 17 “Deixis and anaphora” is devoted to the study of deictic and anaphoric expressions.  Careful distinctions are established throughout between syntactic form and categories of meaning and use.  Few people would deny the relevance of these factors for the description of grammatical constructions.  The centre of the debate, however, is whether the inclusion of aspects to do with meaning and use should affect the overall organization of the grammar or not (see some of the contributions in Graustein & Leitner 1989).   Thus, a pragmatically-founded grammar should be text-based and not sentence-based, or at least it should draw its data from authentic texts.  In this sense, both CompGR and CambGR are firmly sentence-based, and, therefore,  more ‘conventional’ than some of the grammars mentioned above (see, especially, Biber et al. 1999 and, to a lesser extent, Downing & Locke 2002). 


The two grammars are also very similar in the nature of the data they use for linguistic description and the way it is presented.  Four sources are mentioned for data collection in CambGR (11): (i) the authors= own intuitions as native speakers; (ii) other native speakers’ intuitions; (iii) computer corpora[10]; and (iv) data from dictionaries and other scholarly work.  In this CambGR does not differ from CompGR  (see Greenbaum 1988: 47).  In both cases, examples are either invented or modified versions of actual utterances, in order to direct reader´s attention more quickly and more clearly to the point and to avoid Airrelevant distractions in the material@ (see Greenbaum 1988: 46; CambGR.: 12).  Since there are no references to sources in either of the two grammars, what was seen by some readers as a major shortcoming of CompGR, also applies to CambGR: the reader is left in the dark regarding whether the data used to illustrate a particular grammatical point is authentic or invented, taken from a corpus or from other sources (see Mukherjee 2002a and references cited within). 


The lack of systematic and consistent use of data from naturally occurring discourse has led Munkherjee (2002a) in his review to regard CambGR  a ‘quaint anachronism’ in comparison with Biber et al.’s (1999) Longman Grammar of Spoken and Written English (LongGR, henceforth): an entirely corpus-based description, based on a 40-million word corpus of spoken and written English.  To be fair, CambGR uses data from corpora for lists of words which appear frequently in a particular grammatical construction, but frequency rates do not play a determining role in linguistic description. For instance, in the description of related structures, decisions about which is the basic structure do not depend on corpus findings on frequency of use, but on grammatical factors (see Mukherjee (2002a) on the different treatments of extraposition in the LongGR and CambGR and Huddleston & Pullum’s (2002) response to his criticism).


The issue here is what counts as valid data for linguistic description (and theory): a question that has divided linguists for decades and that reflects deeper divisions regarding language as an object of study and linguistics as a science (the nature of linguistic analysis and methodological aspects).  Two questions have to be addressed: (i) whether data should be ‘authentic’ (raw data) or whether it should be invented or edited and (ii) the relevance of frequency rates for linguistic description.  Regarding (i), the authors of CambGR, like those of  CompGR, clearly disfavour raw data.  Huddleston & Pullum (2002), in their response to Mukherjee’s (2002a) review, consider it “one of the errors of strictly corpus-oriented grammars to use only raw attested data for purposes of illustration” and think it “counterproductive to quote a sentence with a subject NP containing a long and distracting relative clause when all we are concerned to illustrate is the order of adjuncts in the verb phrase” (p. 3).  As for (ii), the remarks on frequency in CambGR respond to an attempt to separate rare grammatical constructions from sporadic mistakes (see Pullum 2002: 3), but no statistical figures are given.  Whether a construction is ungrammatical or not is in principle independent from whether it is frequently used or not; and decisions about canonical vs. non-canonical constructions are taken on the basis of structural properties, not frequency of use.  This is not to deny the value of frequency percentages as indicators of matters of performance which should be part of a descriptive grammar, or to deny the usefulness of corpus linguistics for linguistic description - but one should not overestimate their value either.  Some reflection is needed on what is to be gained by adding percentage rates to constructions or by providing just raw data.  H & P, like Quirk et al., are writing a descriptive reference grammar of English; they are not to trying to provide a statistical study of frequency of words or structures across genres, varieties of  English or stages of the language, and, hence, their choices in these matters seem fully justified.


2.2.3. Theory what theory?

Before we look in some detail at the type of linguistic analysis present in CambGR in sec. 3, let us say reflect a little more about the theoretical principles underlying CambGR.  In particular, the questions we are interested in here are: Should descriptive grammars be theoretically eclectic or should they be tied to a particular theory? How can CambGR be considered in this respect?


Theoretical eclecticism has often been pointed out as one of the characteristics of descriptive grammars.  The idea is that descriptive grammars should be regarded as works of synthesis, drawing on recent research as well as the grammatical tradition, but should not subscribe any particular theory.  As Greenbaum (1988: 42) puts it, regarding CompGR , “if a grammar is to be comprehensive it cannot be tied to one theory.”  For Greenbaum, grammars that are tied to one theory will necessarily be ‘partial’ grammars (covering only those grammatical aspects of the language that have been investigated by the linguists within that framework), and may date quickly. The problem is that a grammar which tries to include as many analyses from different frameworks as possible may succeed in being theoretically eclectic, and, possibly, ‘neutral’, but may lack coherence. The most satisfactory analysis should be adopted, as long as it fits in with the descriptive framework of the grammar.  That is, decisions have to be made as to what adds to the descriptive value of the grammar, but an overall coherence and consistency has to be maintained insofar as it is possible. 


A clue to the choices made by the authors of both grammars for descriptive analysis is found in the selected bibliography:  the “Bibliographical Note” at the end of each chapter in CompGR, and the “Further Reading” section at the end of CambGR. CompGR simply provides lists of references for different topics dealt with in the particular chapters: no mention is made of the theoretical (or descriptive) approach adopted in any of those references, nothing is said about to what extent the description relies on any of those sources, and no evaluative (or any other type of) comments are made.  The “Further reading” section in CambGR is meant to include only those particular works that the authors have been significantly influenced by or from which they have drawn important analytical insights, as well as other works which the reader may turn to for further research.  The authors also make it clear that the inclusion of a work in this section does not mean they have adopted its position or think its claims are correct: “in some cases the value of a work lies mainly in its defending an analysis with sufficient clarity to permit the reader to see how to improve on it” (CambGR: 1765).  The theoretical framework of the works cited is often mentioned (though often in relation to generative-transformational analyses vs. others).  Evaluative comments are often made.  Alternative analyses are mentioned and it is often made explicit which works have been most influential for the analysis adopted, while the reader is sometimes directed to works which offer additional support for the analysis adopted or, on the contrary, a competing analysis .  In sum, while readers may be left in the dark about the sources for the data used, as we pointed out in the preceding section, they are not left in the dark as to the most relevant sources of the analysis adopted.  The “Further Reading” section offers a careful choice of varied, highly significant works, and the comments made by the authors about the content and nature of these works are invaluable, especially for linguist readers and linguists-to-be.


There is no doubt that the authors have relied on a variety of sources from different theoretical frameworks for their description (see, for instance, the “Further Readings” section for “Deixis and anaphora”), but they also admit that they have drawn many insights from the generativist work of the last fifty years (see Pullum 2002: 2).  The reader, however, should not expect a transformational type of approach, along the lines of the work carried out by Chomsky and his associates.  Generativist approaches trace their ancestry to the pioneer work of Chomsky in the 1950's and 1960's  (Chomsky 1955, 1957, 1965).  Since the mid 1970's, however,  two parallel trends have developed, broadly speaking, within generative grammar.  The first trend is associated with the work of Chomsky, which since the 1980's has been known as the Principles and Parameters (P & P) approach (embodied by Government & Binding Theory (GB) and, more recently, The Minimalist Program).  The second trend consists of a number of theories, which include, among others, Generalized Phrase-Structure Grammar (GPSG) (Gazdar et al. 1985) and Head-Driven Phrase Structure Grammar (HPSG) (Pollard and Sag 1994).[11]  It is this second trend that has most influenced CambGR.


The differences between the P & P approach and the phrase-structure approach adopted in CambGR are significant.  Fundamentally, the former postulates a multi-level theory, transformational rules relating different levels of the grammar, while the latter generates surface structures directly. Moreover, phrase-structure approaches like GPSG provide a semantic translation for each syntactic rule and, thus, it is committed to a model-theoretic account of natural language semantics along the lines of Montague grammar.[12] Indeed, much of the early work in this framework in the late 70's and early 80's was devoted to showing that the adoption of a purely phrase structure model of syntactic analysis, together with a sufficiently developed semantic theory, would permit the construction of descriptively adequate grammars which did not make use of transformational rules (see Horrocks 1987: ch 3).  Some of the differences between these two approaches will become evident in our discussion in the following section (especially, 3.2), in which GB is taken as representative of the P & P approach.




In this section I examine critically the treatment of verb complementation in CambGR, focusing mainly on two constructions: V-P (Particle/Preposition) combinations (3.1) and, in more detail, non-finite clausal complements (3.2).  The term complement is used in CambGR for those functions in clause structure which are  “more closely related to the verb and more clearly differentiated by their syntactic properties” (215) as opposed to adjuncts, which tend to be differentiated by their “semantic properties” (215).  Complements are ‘dependents’ of the V, while adjuncts may be ‘dependents’ (modifiers) or ‘supplements’ (more loosely attached to the V).  In CambGR, complements are not equivalent to ‘objects’, since the term is used to include the subject and ‘predicative’ complements like those underlined in sentences like Ed seemed quite competent; She considered Ed quite competent (217). Whether subjects and predicatives should also be included under the term complement is not an issue to be discussed here, but the use of the term complement henceforth includes only what traditional grammars refer to as direct and indirect objects (as well as prepositional objects). It is indeed to this restricted use that the notions employed in chapter 4 to define complements (licensing, subcategorization, argumenthood, selection restrictions and so on) apply most clearly.[13]


3.1. Phrasal complements: V-P combinations


Chapter 4 in CambGR deals with many interesting issues regarding verb complementation, some of which have hardly been represented in more traditional grammars, e.g. light Vs (sec. 7) (have a look, take a rest and so on), or have received an inadequate treatment, e.g.Vs with multiple complementation patterns (sec. 8, which draws on recent research on verb alternations, especially Levin 1993). Special mention has to be made of the section devoted to V-P(reposition) combinations (sec. 6), which focuses on three issues:(i) Prep(ositional) Vs, where the P is selected by the V (as in Kim referred to your book and He congratulated her on her promotion); (ii) Particles positioning between V and object (as in She put in her application); and (iii) V-P idioms ( as I gave up the struggle) (272).  The term >phrasal V=, which can be used widely to include all the examples in (i)-(iii), is rejected here because it implies that the elements underlined form constituents of the category V.  In the narrow use of the term, phrasal Vs are V + intransitive P combinations, including the examples in (ii) and (iii) (but not those in (i)).[14]  In fact, most of the traditional ‘phrasal Vs’ are in CambGR considered to be an instance of verbal idioms, along with other verbal idioms like You are pulling my leg and This gave the lie to her critics.


In general, the analysis presented in CambGR compares favourably with that of CompGR, especially regarding Prep Vs, for which the analysis in CambGR is firmly based on the syntactic evidence showing that the P (in refer to) forms a constituent with the following NP: [PP to your book] (against the two alternative analyses in CompGR, see (1) above).  Where this sequence does not seem to behave as a PP (e.g. in I came [across some old letters]), the authors resort to the notion of ‘fossilisation’  (see 6.11).  Six different types of Prep Vs are given, of which only two appear in CompGR (Type I: She looked after her son; Type II He invested his money in property (with three subtypes)).  The other four include examples in which both complements of the V are prepositional: He looked [to her] [for guidance] (mentioned under >Other multi-word verb constructions= in 16.17 in CompGR) and  three types with predicative complements: It counts [as too short],  They regard it [as successful], and I think [of it] [as indispensable].  For each of these types lists are given of Vs belonging to the different types and special cases are discussed.  The emphasis in CompGR is, however, on the distinction between Type I Prep Vs vs. multi-word (phrasal) Vs, as well as on the idiomatic status of the Prep Vs in Type II, but in general the discussion is shorter and more superficial.


The analysis of the structures in (ii) and (iii) in CambGR can, however, be confusing at times and is not radically different from the analysis given in CompGR, which deals with different types of phrasal Vs which vary in their idiomatic status: Vs in free combination being the least idiomatic, as opposed to other transitive phrasal Vs with idiomatic meanings (She took in the box vs. She took in her parents). CambGR also distinguishes free combinations from V-P idioms, with varying degrees of fossilisation and lexicalisation, but in all cases the P is meant to be an independent constituent. Given the emphasis on constituent structure throughout the grammar, it is quite surprising that the more ‘lexicalised’ forms are not treated differently from those in free combination: i.e. with the V and the P forming a constituent independently from the NP, as some sort of compound or >multi-word= V, using the terminology in CompGR.  Unless we do that, it is not clear what the function is of the PP headed by the P in idioms; pay back my father that loan is said to contain PP-Od-Oi so that this is a ditransitive structure with an additional PP whose function in the structure is unclear.    


This is an example in which the theoretical framework which underlies CambGR constrains the descriptive analysis.  In  English Grammar. A Generative Perspective, Guéron & Haegeman (1999: 4.2), for example, offer an analysis in which in a sentence like John tore up the letter, the sequence up the letter is originally a PP, but a subsequent movement operation ‘incorporates’ the P up into the V, so that at a less abstract level of analysis, the V tear and the P up form a constituent (incorporation being a reformulation of the traditional generative ‘reanalysis’ account of structures like this).  Arguments in favour and against treating tear up as a single V are examined in detail. In CambGR the notion of fossilisation is seen as an alternative to reanalyis in terms of incorporation for prepositional V’s like come across (vs. refer to) (see p. 277).  Since fossilisation is also used to account for V-preposition idioms, I assume that a similar analysis may be given for cases like those being considered here.  But in the phrase-structure approach favoured by H & P in CambGR there is no room for a multi-layered type of analysis of the type offered by Guéron & Haegeman (1999: 4.2), which can account for both the processes in which V and P appear to behave as a unit, as well as for the processes in which V and P can be considered as independent constituents.[15]


3.2. Clausal complements: content clauses and non-finite clauses


The term clausal complements (or complement clauses) is commonly used for clauses functioning as complements of Vs, As, Ns and Ps.[16]   I will focus here on clauses functioning as complements of Vs, like those in (2) ((2a): that-clause; (2b, c): interrogative clauses; (2d):  to-clause, and (2e):  -ing-clause):


(2)               a.         She knew that some people would not vote for her.

b.         They were wondering whether some people would vote for her.

c.         They asked who would vote for her.

d.         She persuaded them to vote for her.

e.         They remembered voting for her.  


The approach to clausal complements like those in (2) in CambGR departs significantly from the more traditional approach in CompGR, where these are analysed are instances of ‘nominal clauses’.  Nominal clauses (as opposed to adverbial, relative and comparative clauses) are defined as having functions “that approximate those of noun phrases: subject, object, complement, appositive and prepositional complement” (CompGR: 1047).  Their internal classification is based on their form and their function in the structure: subject, object and so on (see 15.3-15.6).  In chapter 16, different types of Vs are distinguished (copular, monotransitive, complex transitive and ditransitive); their (phrasal and clausal) complement types are discussed in some detail and lists are provided of Vs with the same complementation patterns.


I have already said that the functional classification of clauses along the lines of that in CompGR is rejected by the authors of CambGR (see 8.2-8.3).  In CambGR, subordinate clauses are defined according to their form: namely, (i) finite clauses (content clauses (ch. 11), relative clauses (ch.12), and comparative clauses (ch.13)), and (ii) non-finite clauses (ch.14).  Quirk et al.‘s adverbial clauses are analysed in CambGR as adjuncts of the clause in chapter 8 “The clause: adjuncts” and subordinators in CompGR (after, as, before, once...and so on) are regarded in CambGR as Ps heading a PP with a clausal complement (a content clause). Issues to do with clausal complementation of Vs are dealt with in chapters 11 (finite clauses) and 14 (non-finite clauses).  The syntactic properties of these structures are analysed in much more detail in CambGR than in CompGR (especially non-finite complements), but readers used to how information is presented in traditional grammars may find it slightly awkward that subordinate clauses are dealt with in so many different chapters and that aspects of clausal complementation of Vs, for instance, appear in two different chapters depending on whether the clause is finite or non-finite. In what follows, I briefly outline the properties of content clauses and then move on to examine the properties of non-finite clausal complements of Vs in much more detail.


3.2.1. Content clauses

Among finite clauses, content clauses are defined in CambGR as the default category: “they lack the special properties of relative and comparative clauses, and their structure is less different from that of main clauses”(950).  The term ‘content clauses’, taken from Jespersen,  also reflects this default status - the clause is selected simply for its semantic content, not for a special syntactic property. Content clauses are divided according to their illocutionary type (like main clauses): declarative (like (2a)), open/close interrogative (like (2c) and (2b), respectively) and exclamative, with declaratives as the default category.[17]  A typical content clause is represented as in (3):[18]


(3)                                                      Clause


            Marker:                    Head:

                                    Subordinator            Clause


            that                      Subject            Predicate

     NP                         VP



                             your secretary            might be leaving


Descriptive representations like (3) are the result of theoretical choices regarding the status of that and the constituent structure of content clauses.  These choices are mostly explained in the relevant ‘blue sections’ of this chapter, particularly  those in section 8, which deal with the differences between their approach and that of the more traditional grammars.  For instance, regarding the status of that (and whether and if (‘whether’)) as subordinators, in CambGR  a distinction is established between S(ubordinator)-class ‘subordinating conjunctions’ like that (and also whether, if and for)  and  P(preposition)-class conjunctions (while, until, although, before....) (see 8.1 for details), following current linguistic approaches which have convincingly shown that >subordinators= have distinct properties.  As for the binary division of content clauses into [marker:subordinator] and [head: clause], this  is based on evidence suggesting that these elements behave as independent constituents (see p. 955).[19]


Whether one agrees with the authors or not on matters like this, is not crucial, since the descriptive analysis that follows offers extremely interesting and illuminating discussions of a variety of constructions involving content clauses, with an impressive wealth of examples.   Some of the sections in chapter 11 show CambGR at its best: for instance, sec. 4.5 in which content clauses which are complements of nouns are analysed (see the arguments against the analysis of these clauses as >appositive= as in CompGR (1016-7)); the careful distinction between form and meaning throughout sec. 5, which deals with interrogative content clauses; the analysis of the factors favouring whether or if in sec. 5.2, and so on.  There are isolated cases, however, in which the analysis is insightful and promising, but frustratingly undeveloped, leaving crucial questions unanswered, as for instance in the discussion of so/such + content clause (sec. 4.6), where it is argued that the that-clause that follows these elements is a complement in clause structure, not a complement of so/such.


A lot of the issues mentioned here are set off in smaller print in the ‘blue sections’. For the linguist and the training linguist, the discussions there are often most interesting and stimulating, but even for other readers, this is a chapter in which the ‘blue sections’ are essential to understand the various theoretical choices made by the authors.  The same applies to the ‘blue sections’ in chapter 14, which is reviewed in more detail in the next section.


3.2.2. Non-finite clauses

Non-finite clauses are dealt with in chapter 14, which is mostly devoted to a detailed description of the syntactic properties of a distinct type of complement clause, exclusive to non-finite clauses, which appears in what is called the ‘catenative’ construction.  It is a truly ‘post-theoretical’ (see sec. 2.1 here) descriptive analysis, greatly influenced by generative, non-transformational, phrase structure approaches of the GPSG type and, therefore, at times, radically different from traditional analyses of these structures. 


A lot of the effort in chapter 14 is devoted to establishing distinctions between similar structures not distinguished by traditional grammars and, conversely, to offering a unitary analysis of constructions which have been traditionally dealt with separately.  This is a chapter where careful reading of the ‘blue sections’ is essential for a better understanding of the approach adopted.  In what follows, I focus on those aspects in which the description presented differs from the traditional one (e.g. the status of to), as well as GB-type generative analyses (the different types of >catenative= complements).


Three main kinds of non-finite clauses are distinguished, to which H & P refer to as form-types: infinitival, gerund-participial and past-participial, as illustrated in (4) (1174) (where the clauses are complements of the main Vs) :


(4)               a.         Max wanted [to change his name]                                          infinitival

b.         I remember [locking the door]                                                gerund-participial

c.         His father got [charged with manslaughter]                           past-participial


It is mainly inflectional properties that distinguish non-finite clauses from finite clauses, but there are other differences too: for instance, while in finite clauses the subject is obligatory, non-finite clauses like those in (4) have traditionally been analysed are subjectless.  This is also the approach followed in CambGR and it has its roots in the GPSG analysis of these constructions. In GPSG, the bracketed structure in (4a), for instance, is analysed as a bare VP, rather than as a clause.  Notice that the implication here, as well as in traditional grammars, is that the presence of the VP is enough to establish clausal status.  The interpretation of the semantic relation between predicates and their arguments (the participant in the event denoted by a predicate) in this framework does not require the subject position to be projected in sentences like those in (4).  This contrasts sharply with what is perhaps the most influential generative approach in transfomational theories like GB.   In GB, the bracketed structure in (4a), for instance,  is analysed as a clause (an S= or a CP) with an empty pronominal (PRO) in subject position, as required by the (Extended) Projection Principle, which ensures that predicate-argument structures are syntactically represented.   There is no analogue for the Projection Principle in GPSG.


To-infinitival clauses may contain a subject introduced by the subordinator for, as in (5b):


(5)               a.         He arranged [__ to be interviewed first]

b.         He arranged [for her to be interviewed first]


The element for, traditionally a preposition, is analysed in CambGR as a subordinator (a marker) introducing a clause (Head).  It is therefore analysed as the non-finite equivalent to that, though its prepositional source is reflected in a number of properties (see sec. 1.4.1).  More controversial is the status of the element to as VP subordinator  - a marker which introduces a VP (a head) (1187) (as in (6i) below).  This analysis preferred over an alternative analysis in which to is a(n) (auxiliary) verbal head which takes the VP as its complement (as in (6ii) below), which is discussed in the >blue section= on page 1185.  The two competing analyses are represented below:


(6)               i.          [VP        [Marker: subordinator  to]       [Head: VP  be interviewed first] ]


ii.         [VP        [Head: V    to]             [Comp: VP  be interviewed first] ]     


Both analyses capture the fact that to and the following V are neither morphologically nor syntactically bound (i.e. to interview is not analysed as the infinitival form of the lexeme interview).  In CambGR, (6i) is chosen over (6ii) because the relation between to and the following VP is very similar to the relation between that/whether and the following clause, so that all three elements appear to be subordinators.  However, its status as a ‘special’ subordinator is recognized: it can be ‘stranded’, like auxiliary Vs (I don=t have to__; I won=t_ ), and it does not necessarily occupy initial position in the constituent it marks (She taught her children always to tell the truth).  But the strongest argument against an analysis of to as a verbal head is that, unlike all other verbal heads, to can be omitted in certain contexts and that it can only head a VP in subordinate clauses.


A third possibility, not considered in CambGR, and which may account for most of its properties, is that to is neither a subordinator, nor a verbal head, but an inflectional head (I(NFL)), like modals and affixes like -ed and -ing.  This has been the standard GB analysis of this element as represented the two possibilities in (7), which show the pre-Barriers analysis, with I as a constituent of S in (7i) and the Barriers (Chomsky 1986b) analysis, with functional categories like I projecting their own phrases (7ii):


(7)               i.          [S [NP  PRO]     [I to]     [VP be interviewed]]

ii.         [IP [NP  PRO]    [I= [I to] [VP be interviewed]]]


On theoretical grounds, under an analysis along the lines of (7), we would not need to posit a ‘special’ subordinator which introduces VPs.  Notice also, that in (6i), the analysis favoured in CambGR,  to + VP form a VP constituent at a higher level.  This is not true for either of the two possibilities in (7): in (7i) to and the VP are independent constituents, as they are not exhaustively dominated by a common node (the node S, which dominates I and VP also dominates the subject NP); in (7ii), however, to and the VP form a constituent, as in (6i), but that constituent is a projection of I (an I’), unlike in (6i), where it is a higher VP.  There are unfortunate empirical consequences for the analysis chosen in CambGR since there are syntactic processes involving VPs which do not involve to, as shown, for instance, in (8) for VP-ellipsis.  The contrast between (8a) and (8b) is difficult to explain under an analysis like that in (6i), in which mow the lawn and to mow the lawn are both VPs:


(8)               a.         John has to mow the lawn, but I don’t have to__

b.         *John has to mow the lawn, but I don’t have __


An analysis along the lines of (7), however, cannot be contemplated within the descriptive framework of CambGR, which does not recognize the existence of inflectional heads as independent constituents, like the theoretical framework on which it is most firmly based. Adherence to this framework provides a fairly consistent approach to the facts discussed in chapter 14, but it does raise the question as to how strong should be the ties between theory and description, an issue discussed in relation to the analysis of phrasal Vs in sec. 3.1 here.  


The influence of phrase structure approaches of the GPSG type is also evident in the detailed discussion of the different types of ‘catenative’ complements.  Catenatives are a distinct type of complement realised exclusively by non-finite clauses, like the complements of the Vs in  (9) (1177). Catenative constructions can be simple or complex depending on the absence/ presence of an intervening NP:[20]


(9)               i. simple           a.         Emma hopes to go on holiday.

b.         Daniel seems to be worried

ii. complex       a.         Paul persuaded Anna to phone her sister

b.         Anna believes Sue to be ready for the job


Catenative constructions are not exclusive to infinitives. The construction is also found with gerund-participles (I resented their being given such favourable treatment), and past-participles ( I had my car stolen). Additionally, four types of complex catenatives are distinguished (see ch. 14: 1.2):  (i) with prepositional Vs (I rely on them to look after themselves); (ii) with the intervening NP  introduced by for (I arranged for them to go by bus); (iii) with the intervening NP  in genitive case (as in the previous example with resent); or (iv) in plain (accusative) case, as in the examples in (9ii) (a construction often referred to as the ‘accusative + infinitive construction’).  Our discussion is going to focus on examples like those in (9), which have received considerable attention in generative grammar and are central to chapter 14 in CambGR.


In the standard theory transformational analyses of the 60's, constructions like those in (9) were said to contain ‘equi’ and ‘raising’ predicates.  Equi predicates involve ‘control’ of the reference of the (implicit) subject of the infinitive clause by either the matrix subject (10a) or the matrix object (10b) (where indexes express the control relation and e stands for empty subject (= PRO)):


(10)           a.         Emmai tried [ ei to go on holiday]                                          subject-control

b.         Paulk persuaded Annai  [ei to phone her sister]                      object-control


Constructions with raising predicates, on the other hand, involve a transformational rule which places the deep structure subject of the non-finite V as either the subject or the object of the matrix clause, as in (11a) and (11b), respectively (where lines express movement and t stands for the trace left by the moved element in its original position):


(11)           a.         Daniel seems [t to be worried]                                   subject-to-subject raising


b.         Anna believes Sue [ t to be ready for the job]            subject-to-object raising



The standard theory distinction between equi and raising Vs is central to the approach to catenative complements in CambGR, as analysed in non-transformational generative frameworks.  Concerning equi Vs like those in (10), no subject position is projected in the phrase-structure approaches underlying the descriptive analysis in CambGR, but the interpretation of these sentences requires an ‘understood’ subject with a ‘controlled’ interpretation: the “interpretation of the missing subject is controlled by an antecedent in the matrix clause” (1193).  ‘Control’ as employed in GB and GPSG are rather different notions.  In GB, control theory is a module of the grammar that deals with the interpretation of empty anaphoric pronominal elements (PRO) in the subject position of non-finite clauses. In GPSG, control involves an agreement in features between the controller and the controllee, as required by the Control Agreement Principle (CAP), the semantic interpretation (i.e. the assignment of denotations to linguistic expressions) of the missing subject being carried out by other means (as some sort of modified Montague semantics (see Horrocks 1987: 207)).  While it is clear that the notion of control employed in CambGR is not that of GB, since these complements are VPs without subjects, it is not clear what the authors of CambGR actually mean by a ‘controlled’ interpretation.


‘Raised’ and ‘raising’ do not mean in CambGR what they mean in GB, either. The concept of ‘raising’ as a transformational rule does not make any sense outside multi-level generative theories like GB.  Consequently, a non-transformational GPSG-like approach is favoured: Daniel and Sue are generated in the position which they occupy in (11), as corresponds to phrase structure theories with only one level of representation (see sec 2.2.3  here).[21]  Consequently, statements like “the missing subject [in sentences like those in (11b)] is retrievable from the raised complement in the matrix clause” (1193) remain rather mysterious until the syntactic analysis of simple and complex catenative constructions is carried out in the sections that follow.  Regarding simple catenatives, a distinction is established between ‘ordinary’ and ‘raised’ subjects, which correspond to the subjects of equi and raising predicates, respectively: thus, while Emma is an argument of try (an agent) in (10a), Daniel is not an argument of seem in (11a), but rather it is an argument of the predicate be worried  (the meaning being something like. ‘Seemingly, Daniel was worried’).  A raised subject is therefore the syntactic subject of a V which is higher in the structure than the one the subject is semantically related to. A set of arguments, of the type typically found in the generative literature, are given to distinguish Vs like hope (like try) from Vs like seem in sec, 2.1 and the distinction is extended to gerund-participials (We enjoyed sailing vs. We kept sailing) in sec. 2.2. The explanations are clear and to the point, with plenty of relevant examples, showing the grammar at its best.


A parallel analysis is presented for plain-complex catenatives in sec. 3.1.1, where predicates with ‘ordinary’ objects are distinguished from those with ‘raised’ objects, a distinction equivalent to that between object-control and raising-to-object predicates (10b) vs. (11b) .  Like before, raised objects are to be interpreted as the semantic subject of the non-finite V. The arguments given in sec. 3.1.1.to distinguish between intend (with a raised object, like believe) and persuade (with an ordinary object) are amongst those that distinguish ECM (Exceptional Case-Marking) Vs from object-control Vs in GB, and they point towards the subject-like properties of the NP following Vs like intend (or believe), as opposed to the object-like properties of the NP following persuade:  relation with finite complement constructions, relation with passive infinitivals, selectional restrictions and dummy objects.[22]


It is worth mentioning in relation to this that in GB, contrary to the earlier transformational analyses of the 60´s, there is not a raising-to-object rule for structures like (11b), equivalent to the raising-to-subject rule in (11a).  Rather, the NP Sue in examples like (11b) is analysed as the subject of an (exceptional) infinitive clause which is the complement of believe-type Vs (i.e. ECM Vs).. Under this analysis, the object-like properties of Sue (e.g. it can be the subject of the corresponding passive Sue was believed to be ready for the job) are accounted for by the fact that the V believe assigns accusative Case to Sue, the case assigned by transitive Vs to their syntactic NP objects. The arguments given to distinguish persuade from intend (or believe) can be used both to claim that the element Sue in (11b) is a ‘raised’ object, in the standard transformational analysis and in the analysis adopted in CambGR, and that it is actually the subject of the infinitival complement of an ECM V, as in the GB analysis.  There are arguments in favour of the GB analysis and arguments in favour of raising,[23] but a point in favour of the GB approach is that it provides a unitary account of the argument properties of Vs like intend vs. persuade in sentences with finite and non-finite clausal complements.  Notice that while (12ia) is ungrammatical, because persuade requires two complements, as in (12ib), intend patterns in exactly the opposite way in (12ii):


(12)           i.          a.         *Pat persuaded [that Liz should interview both candidates]

b.         Pat persuaded [Liz] [that she should interview both candidates].

ii.         a.         Pat intended [that Liz should interview both candidates]     

b.         *Pat intended [Liz] [that she should interview both candidates].


The standard theory raising-to-object analysis for Vs like intend is well-established and can be easily adapted to a particular descriptive framework, but for a grammar so concerned with constituent structure, it is surprising that facts like those in (12) (and others like pronominalisation) are not taken into consideration in CambGR.


The section on complex catenatives is completed with an analysis of the for-complex construction, the oblique-complex construction and gerund-participials. The description is followed by a classification of catenative Vs in section 5, according to what catenative construction Vs appear in (simple, complex or both), with multiple subdivisions (depending on whether the Vs take to-infinitival complements, and/or gerund-participial and/or past participial complement, whether they have raised or ordinary subject/object, as well as the different types of complex catenatives).  A useful index of Vs is included in sec. 5.5, with the class they belong to, offering overall a much more complete, accurate and systematic overview of Vs taking non-finite clausal complements than that of CompGR (chapter 16).


A surprising addition to the class of catenatives is auxiliary Vs have and be and modals in their core uses (as markers of mood, tense, aspect and voice).  A sentence like She may phone is analysed with phone as a non-finite complement of may and a similar analysis is proposed for the elements underlined in She isn=t listening and I haven=t found it.  This is an example of how adopting a particular descriptive standpoint (i.e. catenatives as special types of complements)  leads to a unitary description of structures which are have not been traditionally dealt with together.  Under the view that non-finite complement clauses do not differ essentially from other objects, structures like those in (10) and (11) and structures with auxiliaries (whose complement is clearly not an object) could not be dealt with together. Sec. 4.2 in chapter 14 is devoted to the justification of the analysis of auxiliaries as catenatives as opposed to ‘the dependent analysis’, where auxiliaries are treated as dependents of a following main V (as in CompGR: ch. 3).  Due to space limitations, it is impossible to review here the arguments presented in favour of the authors’ analysis, but I would urge linguistically-oriented readers to study the relevant ‘blue-sections’ carefully, which are an excellent example linguistic argumentation, because of the subtleties of the arguments and the authors’ efforts to avoid indeterminacy .


Chapter 14 is representative of the main focus of CambGR: to analyse the formal properties of present-day English structures.  Historical factors cannot be used to justify syntactic distinctions, like the distinction between gerunds and present-participles often found in descriptive grammars, which is based on their different historical sources (see 4.3).  Syntactic facts, like the different types of catenative complements, receive syntactic explanations.  The authors make it clear that it is not possible to assign meanings to the different form-types and that the selection of the catenative form is not semantically determined, though it is not altogether random either, in the sense that Vs with similar meanings tend to select the same form-types.  The association of to-infinitivals with temporal projection into the future is due to the historical origin of to as a preposition (goal), while the gerund-participial is commonly associated with what is actual and current, which may be connected with the nominal source of most of these complements.  But these are only “historically-motivated tendencies and associations, not constant elements of meaning” (1241).  Having said that, semantic notions like factivity and discourse factors to do with information packaging play a crucial role in some sections in relation to particular form-types selected in specific contexts (see e.g. 7.1 ‘Subjects and extraposed subjects’).





Writing a grammar does not differ essentially from designing a course.  Decisions are made to suit the purposes of the descriptive framework one adopts, as well as the readers/students one has in mind: what data to consider and how to present it, the type of argumentation, use of linguistic evidence, and so on.  For courses, these choices have a direct impact on the structure of lectures, task design and elaboration of tests and exams.  In this sense, CambGR is an extremely useful tool for courses which focus on constituent structure and whose aim is to provide students with a conceptual framework for the grammatical description of English, as well as for courses centred on linguistic analysis and linguistic argumentation, in which students are expected to adopt an active role and are encouraged to collaborate in the ‘construction’ of a grammar, rather than being passive readers of grammars.  In this sense, independently from whether one agrees with the analyses presented or not, one must be grateful to the authors of CambGR for making the argumentation process explicit at all points


Chapter 14 (some aspects of the content of which have been reviewed in the preceding section) is a clear example of how CambGR has been conceived  from a methodological point of view.  There are numerous references to the choice of data on which the analysis is based.  Rare and unsystematic examples are dismissed as mistakes, like those in which non-raising Vs like try are treated as though they were voice-neutral in sentences like The exam papers are trying to be marked by next week  (‘We are trying to mark them’) (see FN 15: 1197). There are attested examples whose acceptability ranges from marginal (This constant telling tales has to stop) to fully acceptable (There was no telling what he might do next) but which, according to the authors, resist elegant description, as the kind of ‘hybrid’ (nominal/verbal) construction that may arise when a historical change has not been completed (1189).  Acceptability and grammaticality are carefully distinguished and low acceptability is not considered relevant for the rejection of a particular analysis.  Thus, the passive sentence ?The parcel began to be unwrapped by Jill has low acceptability compared to the active Jill began to unwrap the parcel, but this is not against the analysis of begin as a raising V because the two sentences describe the same situation: there is no difference in truth conditions, as opposed to what happens with non-raising Vs (Liz hoped to convince them vs. They hoped to be convinced by Liz) (1197). 


This does not mean that the authors are happy to include ‘unnatural’ examples: different infinitive Vs are used when comparing ask and seem (Kim asked to interview the PM and Kim seemed to intimidate the PM) because Kim seemed to interview the PM sounds somewhat unnatural (FN 28, 1215).  Similarly, the V hope is used throughout chapter 14 as a model non-raising V vs. seem, which is the model raising V, but when discussing whether Vs like hope, which appear in the simple catenative construction, may also appear in the complex catenative, other Vs are used because though hope may take for as in She was hoping for Kim to return safely, a finite construction is much more likely than the complex infinitival (She was hoping that Kim would return safely) (FN 14: 1197).


There are, of course, examples which appear to be compatible with more than one analysis, following the authors’ reasoning.  When this is the case, the authors often present what appears to be the most obvious or immediate analysis, and then provide what they consider to be the most plausible analysis (at closer inspection).  Two instances of this have already been mentioned: the analysis of begin as possibly both a raising and a non-raising V, which is rejected in favour of its analysis as a raising V (1179-1198); and the two competing analyses of auxiliaries, where, initially, the case is argued for their analysis as dependents, an analysis which is later rejected in favour of their status as catenative Vs (sec. 4.2).  While indeterminacy regarding the right analysis is undesirable in the framework they are adopting, it is interesting that the authors do not avoid discussing borderline cases, like the ‘borderline’ members of the catenative construction, which may or may not lend themselves to the analysis put forward for catenative Vs in sec. 4.4.


Comments about the data appear mostly in footnotes and detailed, technical argumentation in favour or against an analysis is often found in the ‘blue sections’, which provide interesting pedagogical possibilities for the design of activities aimed at training students in linguistic argumentation.  To the two examples mentioned here (the analysis of begin and auxiliary Vs) could be added, for instance, the discussion about a possible analysis of ‘hollow’ infinitival clauses of the type Max is impossible to live with as raising predicates (CambGR: 1247) or issues to do with structurally ambiguous sentences involving hollow clauses such as They are ready to eat (CambGR: 1248), to mention another of the issues dealt with in chapter 14.


In sum, CambGR  is a descriptive grammar of English with clear and systematic underlying theoretical principles, which offers a variety of most valuable analyses for an impressive range of linguistic constructions, based on a wealth of empirical evidence and solid argumentation.  It succeeds in showing the complexity of language and the phenomenal task linguists face in their attempt to systematize it.  It is therefore an invaluable pedagogical tool for those interested in encouraging “an exploratory attitude towards data”.  This expression is used by Kilby (1984) in the introduction to his book Descriptive Syntax and the English Verb, with whom, I think the authors of CambGR would fully agree:

"...it is, I think, important to note that one consequence of using English or any other single language to exemplify a theory is that the language needs to be simplified somewhat in order for the essential concepts of the theory to be clearly illustrated.  It follows that anyone looking at one of these works in a critical frame of mind will be able to find many points of detail which are substantially more complex than it is allowed for in such works.  More worrying perhaps, one possible result of training budding linguists through such works is that they may begin to believe that such 'laundered' data is in fact the real thing, and this unwittingly encourages the sort of cavalier attitude to data and variability which is characteristic of much contemporary linguistics.  I feel that it is important to add some counterweight to such works, and to encourage an exploratory attitude towards linguistic data" (Kilby 1984: 2)





This paper has dealt with the relation between descriptive analysis and theoretical linguistics, as it appears in CambGR.  The issues I would like to address in the concluding remarks of this paper have to do with the responsibility of linguists to present their ideas in an appropriate way to a wider audience.


Theoretical linguistics is an exciting field.  Work in this field, as in all areas requiring specialised academic knowledge, often involves a single-minded devotion.  New analyses are constantly being proposed to challenge existing ones, there is a wide range of formalisms and a high degree of complexity.  But linguists cannot remain in their ivory towers.  Stubbs (1989) has discussed the social responsibility of theoretical linguists, who, he believes, must try and present their ideas in an accessible way:

“Knowledge about language is important to professional groups such as educators and politicians.  It has implications for teacher-training and bilingual education, and in fields such as medicine and law where miscommunications between groups of people may have tragic effects.  Linguists therefore have a social responsibility to present this knowledge in a clear and accessible way.”  (Stubbs 1989: 29)

By this, Stubbs does not mean that every individual linguist should undertake this task; rather it is linguists as a group that are responsible for presenting the results of their research in the simplest and clearest form for the benefit of society in general. 


There are several reasons why linguists’ ideas have barely penetrated the widespread everyday beliefs and myths about language and language teaching and learning that are prevalent in our culture (e.g. ideas about the decline of language, the place of ‘knowledge of language’ in language pedagogy and so on; see Thornton 1986 and Stubbs 1989 for an insightful analysis).  Some of these reasons are not specific to theoretical linguistics.  On of these is the inordinate prestige afforded to the theoretician.  As Stubbs (1989: 29) notes, the implications and applications of theoretical work have been denigrated for hundreds of years by Western academics.  Like in many scientific fields, the high prestige goes to the theoretician, as opposed to the applied linguist, the language practitioner or even the descriptive linguist.  Thus, Greenbaum (1987: 40) argues that the term ‘grammarian’, with reference to the descriptive grammarian, is often used pejoratively as opposed to ‘linguist’ or even ‘syntactician’ with reference to the theoretical grammarian.  In addition, according to Stubbs (1989), there is a powerful and influential anti-intellectual tradition which rejects the strong rationalist orientation of current work on linguistic theory and which is evident in traditions of English teaching at university and schools (in Britain). Perceptions that theoretical linguistics is too abstract and that it is irrelevant for language pedagogy are widely held by a perhaps surprising number of second language practitioners and researchers.


Stubbs (1989: 32) claims that this anti-intellectual streak “has been encouraged by generations of university training in the arts and the humanities”.  It is deeply bound to the simplistic view that there are two ways of approaching the world corresponding to the arts and the sciences.  Thus, while complexity is taken for granted in many areas of scientific endeavour, linguists is often negatively characterised as being ‘too complex’.  Sadly, linguists themselves (with a few exceptions) have done very little to counteract this elitist view of their subject. Stubbs (1989: 35) argues for the ‘popularisation’ of linguistic knowledge and gives linguists a three-fold responsibility,  “as intelligent citizens with an informed and analytic interest in social issues; as linguists with a general knowledge about language as a whole; and as specialists in particular branches of language study”. 


Descriptive grammars, like CambGR, may not have the type of audience that Stubbs has in mind, but they certainly reach a much wider audience than specialised academic journals and books. They fulfill the important role of synthesizing and interpreting what is known about language after more than eighty years of the emergence of linguistics as a science, and they can serve as the basis for the elaboration of pedagogical grammars (see Dirven 1986), as well as other practical applications in language-related fields. Their social value is, therefore, indisputable, thus vindicating the ‘usefulness’ of theoretical linguistics as a discipline.  But, as we said in sec. 2 here, since the relation between theory and description is bi-directional, descriptive grammarians also have a responsibility towards theoretical (and applied) linguists.  In Liceras’s (1989: 202) words, a descriptive grammar must “provide useful information for defining formal and typological universals, accounting for learning units, dialectal and social variation, inviting native speakers’ reflection on their own language”. 


To conclude the position advocated here is that the work of  theoretical and descriptive grammarians has implications for each other’s practices and that grammars like CambGR are a great contribution towards bridging the gap between theory and description.  ‘Post-theoretical’ grammars of this type suggest the coming of age of  theoretical linguistics as a coherent and mature scientific discipline that can at last withstand the current broadening of its audience.



Aarts, Bas (1993). Descriptive linguistics and theoretical linguistics: some new thoughts on a (still) uneasy relationship. Moderne Sprachen 37, 4, 197-208.

Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad and Edward Finegan (1999). Longman Grammar of Spoken and Written English.  Harlow: Longman.

Bosque, Ignacio (1997). La búsqueda infinita. Sobre la visión de la gramática en Salvador Fernández Ramírez.  Paper read at the Real Academia Española on becoming an official member on 1/6/1997.

Bosque, Ignacio & Violeta Demonte (dirs.). (1999). Gramática Descriptiva de la Lengua Española (3 vols.). Madrid: Espasa-Calpe.

Bresnan, Joan W., ed. (1982). The Mental Representation of Grammatical Relations.  Cambridge, MA: MIT Press.

Chomsky, Noam (1955). The Logical Structure of Linguistic Theory. Chicago: University of Chicago Press.

Chomsky, Noam (1957). Syntactic Structures. Janua Linguarum Series Minor 4.  The Hague: Mouton.

Chomsky, Noam (1965). Aspects of the Theory of Syntax.  Cambridge, MA: MIT Press.

Chomsky, Noam (1981). Lectures on Government and Binding.  Dordrecht: Foris.

Chomsky, Noam (1986a). Knowledge of Language. Its Nature, Origin and Use. New York: Praeger

Chomsky, Noam (1986b).  Barriers. Cambridge, MA: MIT Press.

Downing, Angela & Philip Locke  (2002). A University Course in English Grammar. Routledge:

            London (first published in 1993 by Prentice Hall).

Gazdar, Gerald, Ewan Klein, Geoffrey K. Pullum & Ivan A. Sag (1985). Generalized Phrase Structure Grammar. Oxford: Basil Blackwell.

Graustein, Gottfied & Gerhard Leitner, eds. (1989). Reference Grammars and Modern Linguistic Theory.  Tübingen: Max Niemeyer Verlag.

Guéron, Jacqueline & Liliane Haegeman (1999). English Grammar. A Generative Perspective.  Oxford: Blackwell.

Greenbaum, Sidney (1988). Good English & The Grammarian. Harlow: Longman.

Greenbaum, Sidney & Randolph Quirk (1992). A Student=s Grammar of the English Language.  London: Longman.

Haeseryn, W., K. Romjin, G. Geerts, J. de Rooij & M.C. van den Toorn (1997). Algemene Nederlandse Spraakunst (2 vols.). 2nd ed. (revised).  Groningen/Deurne: Martinus Nijhoof/ Wolters Plantyn

Horrocks, Geoffrey (1987). Generative Grammar.  Harlow: Longman.

Hualde, José Ignacio and Jon Ortiz de Urbina (2003). A Grammar of Basque.  Berlin/New York: Mouton de Gruyter.

Huddleston, Rodney & Geoffrey K. Pullum (2002). A response concerning The Cambridge Grammar.  The LinguistList (Disc: Language Description, Huddleston & Pullum) (17 July 2002).

Jespersen, Otto (1909-1949).  A Modern English Grammar on Historical Principles (7 vols.). Copenhagen: Munksgaard. (Republished, Heidelberg: Carl Winter; London: George Allen and Unwin.).

Kilby, David (1984). Descriptive Syntax and the English Verb. Beckenham: Croom Helm Ltd.

Levin, Beth (1993). English Verb Classes and Alternations.  Chicago: University of Chicago Press.

Liceras, Juana M. (1989). On linguistic theory and Spanish grammar.  In G. Graustein & G. Leitner (eds.): 184-204.

Mukherjee, Joybrato (2002a). Review of The Cambridge Grammar of the English Language. The LinguistList (3 July 2002).

Mukherjee, Joybrato (2002b). A response concerning The Cambridge Grammar of the English Language. The LinguistList (Disc: Language Description: Huddleston & Pullum) (21 July 2002).

Newmeyer, Frederick J. (1983). Grammatical Theory.  Its Limits and Its Possibilities.  Chicago: The University of Chicago Press.

Newmeyer, Frederick J. (1998). Language Form and Language Function.  Cambridge, MA: MIT Press.

Palmer, F. R. (1987). The English Verb.  2nd Edition. London: Longman.

Perlmutter, David M. (1983). Studies in Relational Grammar .  Vol 1. Chicago: University of Chicago Press.

Pollard, Carl & Ivan A. Sag (1994). Head-Driven Phrase-Structure Grammar.  Chicago: University of Chicago Press.

Postal, Paul (1974) On Raisinf.  Cambridge, MA: MIT Press.

Poutsma, Hendrick (1926-1929) A Grammar of Late Modern English.  Groningen: Noordhoof.

Pullum; Geoffrey K. (2002). Some points of agreement about The Cambridge Grammar.

             TheLinguistList (Disc: Language Description, Huddleston & Pullum) (29 July 2002).

Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech & Jan Svartvik (1985).  A Comprehensive Grammar of the English Language. London: Longman.

Radford, Andrew (1988). Transformational Grammar.  Cambridge: Cambridge University Press.

Renzi, Lorenzo (ed.) (1988) Grande Grammatica Italian di Consultazioni.  Bologna: Il Mulino.

Saussure, Ferdinand (1916) Cours de Linguistique Générale.  Paris: Payot.

Solà, Joan., María-Rosa Lloret, Joan Mascaró and Manuel Pérez Saldanya (dirs.) (2002). Gramàtica del Català Contemporari (3 vols.)  Barcelona: Empúries.  .

Steedman, Mark (1993). Categorial grammar. Lingua 90: 221-258.

Stubbs, Michael (1989). On ivory towers and the market place: Everyday and specialist knowledge in applied linguistics.  In Butler, C., R.A. Cardwell, J. Channell (eds.)  Language and Literature - Theory and Practice.  A Tribute to Walter Granberg.   Monographs in the Humanities VI.  Nottingham: University of Nottingham.

Stuurman, Frits (1989). Generative grammar and descriptive grammar: Beyond juxtaposition.  In G.Grausterin & G. Leitner (eds.): 163-183.

Teleman, Hellberg and Andersson (dirs.) (1999). Svenska Akademiens Grammatik (4 vols.).  Stockholm: Svenska Akademien/Norstedts

Thornton, G. (1986). Language, Ignorance and Education.  London: Arnold



© Amaya Mendikoetxea. Circle of Linguistics Applied to Communication/ Círculo de Lingüística Aplicada a la Comunicación 13, November 2003. ISSN 1576-4737.



contents of clac 16



[1].The content of this paper partially matches that of my review-article ‘English and the good grammarian. A review of The Cambridge Grammar of the English Language’ to appear in Estudios Ingleses de la Universidad Complutense 11.  I thank the editor of that journal, Angela Downing, for granting permission to partially reproduce the contents of my review-article here.  I am also grateful to Jon Ortiz de Urbina for his insightful comments on the EUIC paper.  Ignacio Bosque, Luis Eguren, Joan Solà and Emile Slager have provided valuable references for this paper.  Finally, I must thank Paul Rollinson for his careful reading of the manuscript.

[2].Rodney Huddleston and Geoffrey K. Pullum: The Cambridge Grammar of the English Language.  Cambridge: Cambridge University Press, 2002 (hbk). ISBN 0-521-43146-8, xvii + 1842 pp. Contributors are listed in alphabetical order: Laurie Bauer, Betty J. Birner, Ted Briscoe, Peter Collins,, Anita Mittwoch, Geoffrey Nunmberg, Frank Palmer, John Payne, Peter Peterson, Lesley Sterling and Gregory Ward.  One or both of the authors have worked closely with contributors in co-authoring the chapters concerned.

[3].See Pullum’s (2002) comments on the unfortunate coincidence of acronyms for the two grammars, which is why I use the abbreviations CambGR and CompGR to refer to them.

[4]. The Spanish Gramática Descriptiva de la Lengua Española has inspired grammars of similar character and goals for other languages spoken in the Iberian peninsula: Catalan (Solá, Lloret, Mascaró & Pérez Saldanya (dirs.) 2002 Gramática del Català Contemporari) and Basque (Hualde & Ortiz de Urbina (eds.) 2003 A Grammar of Basque). 

[5].The field of theoretical linguistics can be divided into two general approaches to linguistics: formal vs. functional, with their subdivision.  It must be emphasized that the criticisms of descriptive linguists are mostly directed towards theoretical linguists working within Chomsky’s generative paradigm since the 1960's (see, for instance, Bolinger 1961).  Chomsky’s work is normally viewed as representative of the formal approach (but see sec. 2.2.3 below).  Theoretical linguists working within functional approaches to linguistics have often criticised generative linguistis for exactly the same reasons as those of descriptive linguists.

[6].See Liceras (1989) on how grammarians should proceed in their evaluation of proposal from theoretical linguistics for inclusion on descriptive grammars.

[7].As we said in note 1, though a number of contributors collaborated in the writing of CambGR, either one or the two main authors co-author all the chapters.  This was not the procedure used in the writing of CompGR: the four authors divided the work and worked separately, though they met to avoid incoherences (see Greenbaum 1988).

[8]. It is not at all clear in CompGR (see 16.14) why the possibility of turning the prepositional complement into the S of a passive sentence in ‘Type I Prepositional Vs’ like approve (of)  is a criterion favouring the SVO analysis in (1i).  As Quirk et al. notice, the passive is also found with Vs which follow PP=s with locative meaning (which they refer to as ‘Adverbial’), as in the examples in (i) (CompGR: 1164):


(i)         a.         This field must have been played on last week

b.         These caves were once lived in by primitive men.


If these are clear examples of SVA sentences, why should the existence of prepositional passives like Noisy parties are not approved of lead to a (parallel) SVO analysis for prepositional Vs?.  The problem here is that for Quirk et al. the notion of affectedness (a semantic notion) is associated with objecthood (a syntactic function), so that the subject of a passive is always an affected ‘object’ and hence the parallel SVO analysis for (1i) (and, presumably, for the sentences in (i) in this note) .  A similar confusion is found for the notion ‘adverbial’, which is associated with locative, temporal and manner interpretations, among others, as well as denoting a syntactic function (cutting across what in the generative tradition, is referred to as complements and adjuncts).

[9].The term ‘adverbial’ is not employed in CambGR.  The same is true for terms like ‘adverbial clause’ and ‘adverbial phrase’ (see 612n).

[10].The corpora used by the authors include: Brown Corpus of American English , the London/Oslo/Bergen (LOB) corpus of British English, the Australian Corpus of English (ACE) and the Wall Street Journal corpus.  The British National Corpus was only released to scholars outside the UK after the book was in its final draft.  See Mukherjee’s (2002a) review and the response by the authors (Huddleston & Pullum, 2002) on whether the Wall Street Journal corpus (44 million words) should be considered a corpus or not.

[11].As well as Lexical-Functional Grammar (Bresnan 1982), Relational grammar (Perlmutter 1983) and Categorial Grammar (Steedman 1993).

[12]. In the Introduction to their book, Gazdar et al (1985: 11) say  “our efforts to marry a linguistically interesting generative syntax with an explicitly defined semantics place our work in an arena that few have entered, since most current syntactic research is associated with no theory of semantics whatsoever.” Horrocks (1987: 215-6) defines GPSG as “a combination of Montague’s approach to the semantics of natural languages and a highly sophisticated version of X’-theory”.  Newmeyer (1998) considers the syntax-semantics relation to be one of the attractions of this theory.  The link between theoretical linguistics and the work of psychologists and computer scientists on parsing and information-processing has also been pointed out as one of its assets.

[13].What in CambGR is called complement for descriptive purposes is similar to what is known as ‘argument’ in the generative tradition.  In fact, as the reader can see, the definitions for complement and adjunct in CambGR.  A precise definition of these terms requires use the tools and concepts of a theoretical framework, e.g. in X=-theory, complements are standardly defined as sisters of the V within a V’, while adjuncts occupy adjoined positions (sister to V’).

[14]. It is worth emphasizing that particles (in their idiomatic and non-idiomatic uses) are regarded in CambGR as intransitive P, and not as adverbs, as is the case in traditional grammars like CompGR.  The notion of intransitive P, which has been around for a long time in theoretical grammars is thus incorporated into descriptive reference grammars, a welcome addition.  I thank Jon Ortiz de Urbina for bringing this to my attention.

[15].A different approach is adopted by  Radford (1988: chap. 2), where the element off in  The drunks put off the customers, is not an independent P, heading a PP, as opposed to the equivalent sentence with particle shift The drunks put the customers off. 

[16].See CambGR (fn 31:1017) for why the authors reject the term ‘complement clauses’ for examples like those in (2a, b. c), which are regarded as content clauses (see 3.2.1 here).  Those in (2d, e) are called ‘catenatives’, as we shall see in sec. 3.2.2 below.

[17].In fact, under this definition of content clauses as clauses with no special syntactic features, the inclusion of exclamatives and open subordinate interrogatives like I asked when they were planning to go is rather striking. These are ‘unbounded dependencies’ and their special syntactic properties are not radically different from those of relatives, which are dealt with in chapter 12 “Relative constructions and unbounded dependencies”.  One could say that the reason why they are regarded as content clauses is because their structure is not very different from that of a main interrogative clause. However, the analysis of subordinate interrogatives (and main interrogatives alike) as unbounded dependencies is not emphasised, maybe in order to avoid stressing what these structures have in common with relative clauses, given H & P’s classification of complex structures.

[18].P-Markers in CambGR contain information about both function and categories (see ch. 1: 4.2.2.-3).  The notion of ‘head’ is very different from the standard use of this term in X’-theory.

[19].The explanations given for why that is not the head of the (expanded) clause are, however, less convincing.  Omissibility may work for that, but not for whether.  As for the fact that Vs select the form of the clause following that (e.g.  insist selects a subjunctive clause: We insist [that the work be finished this week], while hope does not *We hope [that the work be finished this week]), one could certainly come up with mechanisms to capture this fact within the framework used by the authors.

[20].The term catenative is a reflection of the fact that the construction can be repeated recursively, thus yielding a concatenation of Vs as in  She intends to try to persuade him to help her redecorate her flat  (1177) (see, for instance, Palmer 1987: ch. 9).

[21].This also reflects a fundamental difference between the two approaches.  The standard theory analysis in (11) is partly motivated on semantic grounds: the NPs Daniel and Sue belong semantically to (are arguments of) the subordinate predicate, not to the main predicate.  The Projection Principle requires that predicate-argument relations are represented at the level of Deep Structure.  There is no analogue to the Projection Principle in GPSG, where semantic interpretation is carried out in a different way (see sec. 2.2.3 here).

[22]. A special case is that of Vs like want and prefer, which since Postal (1974) have resisted a coherent analysis.  These Vs are analysed as a subclass of Vs appearing in both simple and complex constructions, with an ordinary subject in the simple construction and a raised object in the plain-complex construction.  The fact that these Vs do not allow passivisation of the raised object (unlike believe or intend) is left unaccounted for, as has been pointed out to me by Jon Ortiz de Urbina (personal communication).

[23]. In fact, Postal’s (1974) arguments in favour of raising never received an adequate explanation within the GB approach.  Conversely, extraction facts were not accounted for under the raising approach.  I thank Jon Ortiz de Urbina for this observation.