Heads without mentalism

The main point of this chapter is that heads must be among the categories used in the mental representations that ordinary people construct when they hear and utter sentences. In other words, the category ‘head’ and the associated concept ‘dependent’ are psychologically real; and they are real not only in grammars but also in sentence structure. However, we start with some non-psychological considerations about heads.

Heads in grammars and in sentence structure: a survey of theories

The distinction between grammars and sentence structure is an important one, because it is possible to accept the need for heads in the grammar without recognizing them in sentence structure. As Arnold Zwicky pointed out (1988), this possibility is exploited in Generalized Phrase Structure Grammar (GPSG), where the head daughter is represented in the relevant phrase-structure rule as simply ‘H’ (Gazdar et al., 1985: 51), but this is just a conventional abbreviation which is replaced, in the corresponding phrase marker, by all the ‘head features’ of the mother node. But these ‘head features’ are just ordinary features which do not in themselves show explicitly that the head is the head; the head is different from the other daughters only in that its features are the same as those of the mother.

Introduction

This chapter argues that when syntacticians refer to heads, they are referring to one of at least three distinct notions, all of which have a place in the theory of syntax. It can thus be seen as a working out and refinement of the syntactic portions of Zwicky (1985), especially in response to the discussion by Hudson (1987).

My 1985 paper examined several situations in which the assignment of head or dependent status to some participant in a syntactic construction is unclear; for them, tests that pick out the head in straightforward cases like verb or preposition plus object {see penguins, about penguins) do not always make a unique assignment. Hudson (1987) attempted to show that a unique head could be picked out anyway, but at the cost of abandoning some tests and re-interpreting others. Still other syntacticians (for instance, Fenchel, 1989; Warner, 1989; Radford, this volume) have proposed that various problematic cases involve the assignment of head status to more than one participant in a construction, but again at a cost (complications in other parts of the description or other parts of the theory). With these latter authors, I propose to ‘have it both ways’, but not by assigning multiple head status, a step I reserve (see section 13.8) for quite a different set of phenomena.

Instead, what I want to say about heads is rather like what most syntacticians now say about subjects.

Heads in grammatical theory

The majority of current grammatical theories refer explicitly to the head of a phrasal constituent. Yet while the term ‘head’ has entered the common currency of theoretical linguistics, this does not provide evidence of agreement on what it means. Nor does the term's long and varied career in linguistics guarantee that it identifies a notion which is not already identified by some other, more basic notion or interacting set of notions. The purpose of this volumev is twofold: first, it aims to uncover and make explicit the notion (or notions) behind the term ‘head’; second, it aims to investigate the status of the notion (or notions) in linguistic theory.

Most linguists would agree with the informal characterization that the head of a phrase is one of its constituents which in some sense dominates and represents the whole phrase. In an important paper published in the Journal of Linguistics in 1985, Zwicky drew attention to the fact that use of the term ‘head’ had been extended from syntax to morphology (for example, by Lieber, 1981; Williams, 1981; and Kiparsky, 1982) in spite of the fact that there was no generally agreed formal definition of the notion in syntax (though an important contribution had already been made by Gazdar and Pullum, 1981 and Gazdar, Pullum and Sag, 1982). Zwicky therefore set out to find a rigorous, generally acceptable definition for ‘head’. He proceeded by examining the following eight candidate criteria for the identification of a constituent as a syntactic head.

Introduction

Ellipsis, deletability and the question of what fragment of a constituent can stand alone are problems which figure in many discussions of what constitutes the head of a constituent; see Zwicky's definition of base (this volume). The very persistence of this issue in the literature would lead one to expect that there would be distinct cross-linguistic consistency in the deletability and retainability of parts of constituents. In addition, and more generally, it would lead one to expect that the notion of head would lead to useful crosslinguistic generalizations about deletability and obligatoriness of parts of constituents. This chapter offers a first step towards a cross-linguistic investigation of whether and how heads figure in constraining constituent-reducing operations of the type that arise in connected discourse. A comparison of just two of the languages discussed – Russian and Chechen-Ingush – suffices to show that languages can differ substantially in whether heads can be deleted, or dependents left to stand alone, by these discourse operations. In part for this reason, and in part because the relevant tendencies are only statistical, constituent-reducing operations will probably not prove to be a useful cross-linguistic indicator of head and non-head status. On the other hand, they do appear to shed light on some of the larger questions of how constituents are defined, grammaticalized and used in individual languages.

In what follows I will simply assume that the verb is the head of the clause. Since the argument is that there is neither cross-linguistic consistency in what can be deleted in ellipsis nor correlation between deletability and morphological type, the conclusions would be equally valid for any decision as to what is head.

Introduction

This chapter discusses the syntactic and semantic relations between the ‘head of a phrase’ and the phrase itself. In particular, the phrase is a ‘kind of the head since the latter provides both the semantic and syntactic type of the phrase (Hudson, 1987: 115–16). For example, the noun can be treated as head in noun phrases and its systematic priority over other categories in the phrase, such as adjectives and determiners, is manifest in the syntactic and semantic type of the phrase.

There are three reasons for discussing these syntactic and semantic relations. The first is to introduce a unification framework which characterizes them in terms of a dependency approach to combination: categories in a binary phrase are combined as ‘head’ and ‘modifier’ (or ‘dependent’) (Hays, 1964; Anderson, 1977: 92–100; Hudson, 1984: 75–9; Miller, 1985: 25–31). The second reason is to show that this framework provides a better account of these relations than frameworks where categories are combined as functor and argument (Ajdukiewicz, 1935; Vennemann and Harlow, 1977; Flynn, 1983; Calder et aL, 1987; Zeevat, 1988). The third-and most important – reason is to introduce evidence which demonstrates that the semantic relations between heads and phrases are not necessarily transparent: the phrase can exhibit semantic properties which conflict with those of the head. This evidence leads to a revision of the framework where semantic conflicts are resolved on the basis of the head-modifier distinction.

The Head-Modifier Principle

The framework adopted is a unification grammar in which categories combine as head and modifier to yield a result category for the phrase.

Introduction

Head-driven Phrase Structure Grammar (HPSG) is the only current approach to syntax with ‘head’ in its title. Obviously, then, heads play an important role in the framework. As its founders, Pollard and Sag, stress, HPSG is also a unification-based approach (see Pollard and Sag, 1987). This means that whether or not an expression is well formed depends on a variety of factors, none of which takes precedence over any other. I will argue that there is a conflict between these two features of the framework. More precisely, I claim that a fully general characterization of the head-mother relation requires a default principle which is incompatible with a purely unification based approach (as in Pollard and Sag, 1987: 8). It seems to me that this is true within the standard version of HPSG, but it is particularly true if HPSG is revised in ways that I have argued for elsewhere (Borsley, 1987, forthcoming). A unification-based approach is not something that should be given up lightly, since it offers a simple picture of how grammars function and is attractive from a computational point of view. However, if it precludes the capturing of a linguistically significant generalization, it should be abandoned, and it seems to me that this is the case.

In a sense, this chapter is rather narrowly focused, since it is concerned with the correct formulation of one principle in one not very widely assumed framework. However, the argument that I develop is potentially relevant to any framework in which heads have mothers, any constituency-based framework, in other words.

Introduction

In examining the interrelation between grammatical terms as used in the description of individual languages and in general linguistic theory, it is useful to bear in mind that individual languages often present language-specific criteria for a particular grammatical notion which, while not forming part of the general linguistic definition of that notion, none the less provide a language-internal means of identifying instances of the notion in question and thus solidifying our data-base for study of the grammatical notion in general linguistic theory. For instance, the possibility of forming a comparative hungrier shows that English hungry is an adjective in John is hungry, thus providing an instance of a concept that can be adjectivalized even though in many languages non-adjectival constructions are preferred (such as a noun in French Jean a faim, literally ‘John has hunger’, or a verb in Russian Vanja xočet esʹ, literally ‘John wants to eat’); note that cross-linguistically, the existence of comparative forms of adjectives is rather rare, so this is very much a language particular test. In this chapter, I argue that Haruai has a language internal phonological correlate of the head-dependent relation, namely that, in the absence of other factors (contrastive stress, greater stress associated with focus, that is, essential new information), dependents receive greater stress than their heads. The basic wordstress rule in Haruai is for stress to fall on the first syllabic segment of the word; the only consistent exceptions are negative verb forms, which are stressed on the negative morpheme, for example n dw-lm-a ‘I did not go’, literally ‘I go-NEG-PAST(-1.SG)-DEC'.

Introduction

How many heads does a noun phrase have? In the traditional generative conception, the answer is simple: the noun phrase has a single head which is a noun. We shall call this the single-head hypothesis. In the single-head hypothesis, other typical noun-phrase constituents (articles, demonstratives, quantifiers, numerals, adjective phrases, adpositional phrases, relative clauses, etc.) are all treated as modifiers of the head noun.

An alternative to the single-head hypothesis which has recently gained in popularity is the idea that the noun phrase contains at least one head, and possibly a multiplicity of heads, in addition to the noun. Each such head, typically but not necessarily a functional category such as a determiner, numeral or quantifier, has its own phrasal projection and takes another noun-phrase constituent as its complement. We shall call this the multi-head hypothesis.

Section 6.2 of this chapter provides a preliminary discussion of the single-head hypothesis (6.2.1) and the multi-head hypothesis (6.2.2). In section 6.3, which is the heart of the chapter, five specific arguments are then presented which lend support to the single-head hypothesis: incorporation (6.3.1), subcategorization (6.3.2), the position of possessor phrases (6.3.3), apposition (6.3.4) and agreement and government (6.3.5). In conclusion, section 6.4 lists the consequential syntactic and morphological properties which focus on the noun as the central constituent of the noun phrase.

Two hypotheses of noun-phrase structure

In this section, we present an initial comparison of the single-head and multi-head hypotheses.

While our understanding of English grammar at the single sentence level is substantial, we are only just beginning to seriously explore patterns of interclausal relations as they are used in naturally occurring language. Furthermore, even though interactional language use outweighs all other types of language use, the analysis of English discourse within linguistics has tended to concentrate on monologue data and to neglect conversation. One reason for this neglect has to do with the relatively recent development of practical and unobtrusive audio and video recording technology. If we hope to gain an understanding of how grammar emerges and changes with use (i.e., an understanding of grammar as a system adapted to its use [Du Bois 1984]), we must make use of the available technology and look more seriously at language in interaction. The present book is a contribution to our understanding of the use of a clause type that is very common in spoken English interaction: the adverbial clause. I examine the use of adverbial clauses in a corpus of naturally occurring American English conversation.

At a general level, this research is part of a larger program of interest in observing grammar in its “natural habitat”: connected, contextualized discourse. The focus here is on adverbial clause usage, in part because numerous studies have detailed their functions in discourse and in part because of their relative frequency in spoken English. In addition to contributing to our understanding of adverbial clauses in interactional language use, this study has been guided by a methodological goal: The research presented in this book demonstrates the usefulness of conversation analysis as a tool for understanding the emergence of grammar in interaction.

Overview

Unlike initial adverbial clauses in these data, which always end in continuing intonation, when adverbial clauses appear after their associated modified material, they may be connected to that material across continuing or ending intonation. There are 135 final adverbial clauses, making up 66% of the adverbial clauses in the data. Of these final adverbial clauses, 40 (30%) are temporal (when, while, before, after etc.), 18 (13%) are conditional (if), 75 (56%) are causal (because, 'cause), and 2 (1%) are concessive (although, even though).

Eighty (59%) of the final adverbial clauses occur after continuing intonation, while 55 (41%) link back to utterances ending in final falling intonation.

The distinction between continuing and final intonation reflects speakers' decisions to signal that an utterance is not yet completed (continuing intonation), or that an utterance is possibly complete (final intonation). Schiffrin (1987) used this intonation distinction to separate intra-utterance conjunction from inter-utterance conjunction in her analysis of interview data. By distinguishing intra- from inter-utterance intonation patterns, one can describe grammatical connections that occur as parts of intonationally coherent units, and compare these connections to grammatical connections that occur across final intonation boundaries.

In the present chapter I describe the use of adverbial clauses following continuing intonation. In the next chapter, I look at adverbial clauses added to utterances that have been presented as intonationally complete. Final adverbial clauses after continuing intonation are of three types in my corpus: temporal, conditional, and causal. Of these three, only temporal and conditional clauses appear both before and after their main clauses, while causal clauses appear only after the material they modify.

In the preceding chapters, I have explored the work that adverbial clauses do in conversation, and in what positions they do that work. In addition to the usage patterns focused upon in chapters 3–5, there is interesting variation in the distribution and functions of the three major types of adverbial clauses found in the corpus. In the present chapter, I compare the placement and functions of the different clause types, and I also highlight some apparent violations of the general principles discussed in previous chapters. The exceptions serve to further elucidate the principles themselves.

Different clause types in initial versus final placement

Examining variation in the use of different clause types, we find that causal clauses, while not appearing initially, do frequently appear in final position after continuing intonation. In final position, causal clauses do work that is distinct from the work of other final adverbial clauses. These differences are consistent with the semantics of the different clause types, causals presenting explanatory or motivating material, rather than temporal or situational grounding.

As can be seen from Figure 1, of all temporal clauses, most (52.5%) appear finally after continuing intonation. That is, a majority of temporal clauses in this corpus are being used to complete main clause meaning, rather than to structure the discourse (or add more material after an utterance is finished – to be discussed in section 6.2). Utterances in these conversations, then, seem to be regularly provided with some temporal or situational grounding through temporal clauses.