The Typology of Parts of Speech Systems

The Markedness of Adjectives

David Beck

• Publisher: Routledge
• Year: 2013
• Print ISBN: 9780415941556
• eBook ISBN: 9781136069147
• Edition: 1
• Topic: Languages & Linguistics
• Subtopic: Grammar & Punctuation
• Index: Languages & Linguistics

This book presents rigorous and criterial definitions of the major parts of speech - noun, verb, and adjective - that account both for their syntactic behaviour and for their observed typological variation. Based on an examination of languages from five different groups - Salishan, Cora, Quechua, Totonac, and Hausa - this book argues that parts of speech must be defined by combining the criteria of syntactic markedness, which characterizes lexical classes in terms of unmarked syntactic roles, and semantic prototypicality, which delimits their prototypical meanings. Adjectives are shown to be the marked (and, hence, most variable) class because of their inherent non-iconicity at the semantics/syntax interface. The four-member typology of parts of speech systems (languages with three open classes, those that group adjectives with verbs, those that group adjectives with nouns, and those that conflate all three) current in the literature is easily generated by free recombination of these two criterial features. Closer examination of the data, however, casts doubt on the existence of one of the four possible language-types, the noun-adjective conflating inventory, which is accounted here for by replacing free recombination of semantic and syntactic features with an algorithm for the subdivision of the lexicon that gives primacy to semantics over syntax.

Chapter 1
Introduction

Ask a layperson what they know about grammar and you are likely to get an answer that has something to do with parts of speech; ask a linguist what they know about parts of speech and the answer is quite likely to be much less enlightening. Parts of speech systems or, as I will refer to them here, lexical classes are among the most frequently overlooked aspects of linguistic analysis, yet they are at the same time among the most fundamental elements of language. Lexical classes play a key role in most—if not all—syntactic theories, they are the cornerstones of lexicography and lexical semantics, and they are crucial elements in morphological analysis, yet precise and rigorous definitions of these classes have never been successfully formulated. More often than not, lexical classes are treated merely as primitives, either in terms of input to rules, determinants of underlying phrase structure, governors of inflectional patterns, or as sources of valency and subcategorization frames. Thus, class-designations such as "verb," "adjective," and "noun" are the linchpins of semantic, syntactic, and morphological structure, but the terms themselves are rarely defined and their properties, both formal and functional, are often taken for granted. While it is certainly possible to carry out linguistic analysis without a clear definition of the basic units involved (as long as the identity of these units can be agreed upon), any theory which proceeds without a full understanding of its own primitives rests on uncertain foundations.

Another serious drawback to building theoretical models based on tacit assumptions about the properties and definitions of parts of speech is that, as modern linguistics expands its horizons and turns more and more to data from "exotic" and previously undescribed languages, many of these assumptions are being challenged. Most current linguistic theories—whose main proponents are speakers of and researchers in European languages—are built on the model of what Sapir referred to as the "Standard Average European" language type with its familiar three-way division of the lexicon into major open classes of verb, noun, and adjective. Cross-linguistic investigation has shown, however, that not all languages fit this pattern and at one time or another claims have been made that certain languages lack distinctions between adjectives and verbs, adjectives and nouns, or even between nouns and verbs (e.g. Kinkade 1983; Schachter 1985; Sasse 1993; Bhat 1994; Broschart 1997). When confronted with the neutralization of parts-of-speech contrasts, any theory which relies on the three major classes as primitives—or which, at least, has no clear idea of their origin—will have little success in dealing with this variation or of providing an adequate explanation of its provenance.

In terms of cross-linguistic variability in parts of speech systems, perhaps the most salient and widely-remarked upon point of divergence is the frequent absence of the class of adjective. While nouns and verbs appear to be more or less universal, languages that have no or only a few adjectives are a typological commonplace. This seems to imply that there is something marked about the adjectival category, and an investigation of its properties should shed some light on the issues of typological variation in lexical classification and define lexical classes in such a way as to motivate and constrain this variation. It is the aim of the present work to deal with precisely these topics. The discussion will proceed as follows: in Chapter 2, I will outline some previous, unary approaches to defining lexical classes. Traditionally, these have taken three basic tacks—the semantic characterization of parts of speech (Chapter 3), the development of morphological diagnostics (mistakenly taken to be criterial definitions of the lexical classes themselves) (2.2), and definitions based on simple syntactic distribution (2.3). All of these approaches give more or less congruous results when applied to the most typical members of each class and to languages that have Indo-European style three-class parts of speech systems. When confronted with marginal cases and data from other types of language, however, none of them proves adequate in terms of providing rigorous, criterial definitions.

More recent attempts to define parts of speech have tried to deal with variation in terms of syntactic markedness (2.4), a term that is discussed in some detail in Section 2.4.1. One particularly good approach based on markedness is that put forward by Hengeveld (1992a, 1992b) which formulates definitions of parts of speech based on those syntactic roles that different lexical classes can fill without further morphological or syntactic measures being taken. As discussed in Section 2.4.2, when reformulated in terms of contrastive markedness, Hengeveld's "without further measures" turns out to be a useful tool for identifying the marked and unmarked roles of different parts of speech. Languages differ from one another with respect to which lexical classes require further morphosyntactic measures in order to occupy a given syntactic role, and languages that lack a specialized class of adjectives are divided by Hengeveld into "rigid" and "flexible" types depending on the presence or absence of such measures (Section 2.4.3). Further measures include, among other things, recategorization (the acquisition by a word of the typical properties of another part of speech) and decategorization (the loss of properties typical of a word's own lexical class). These processes are described in Section 2.4.4. The major difficulty with syntactic approaches based solely on distributional markedness, however, is that they in no way account for the common semantic core of meanings that are consistently expressed by the same part of speech across so many languages. Section 2.4.5 looks at this issue in the light of work by Croft (1991), which proposes that there are cross-linguistically unmarked mappings between the semantic class and "pragmatic" roles typical of particular parts of speech.

Although Croft's proposal falls short of providing criterial definitions of lexical classes and fails to properly constrain typological variation in parts of speech systems, it does put forward the idea of leaving behind unary definitions based on only one of semantics, syntax, or morphology. Instead, Croft's work suggests a binary approach to the problem and looks at the mapping between the typical properties of lexical classes at two levels of representation. The obvious levels to choose for this are the syntactic and the semantic. The typical syntactic properties of the three major parts of speech having already been discussed (Sections 2.3 and 2.4), Section 2.5 goes on to examine their prototypical semantic properties and shows how cross-linguistic variation in this domain can be dealt with by a theory of semantic prototypes (2.5.1). This theory not only allows the formulation of class prototypes for nouns (2.5.2), verbs (2.5.3), and adjectives (2.5.4), it also predicts the most likely areas for cross-linguistic variation in class membership, meanings lying on the peripheries of the relevant semantic categories being the most variable. Words referring to HUMAN CHARACTERISTICS represent a particularly variable class of meanings and are discussed in detail in Section 2.5.5. Following this some technical issues are discussed (Sections 2.5.6 and 2.5.7) before moving on to the next chapter.

The most important points contained in Section 2.5 are proposals for two criterial semantic properties of nouns and verbs: nouns are said to be prototypically the expressions of semantic NAMEs (2.5.2) and verbs the expressions of semantic predicates (2.5.3). Chapter 3 takes these two semantic criteria, combines them with the unmarked syntactic roles of nouns and verbs identified by Hengeveld (1992a, 1992b), and uses them to develop definitions of the two most basic lexical classes. These definitions are spelled out in Section 3.1, which also provides some terminology and outlines some elementary formalisms borrowed from Meaning-Text Theory (Žolkovskij & Mel'čuk 1967; Mel'čuk 1988). Section 3.2 sketches the role played by the lexicon, or lexical inventory, in the building of syntactic structures and illustrates where it is that lexical classes originate (the lexicon) and how they interact with the rules mapping between the semantic, syntactic, and morphological representations of sentences. Section 3.3 then demonstrates how the new definitions of verb and noun point to a clear and criterial definition of the class of adjective. This definition accounts for the cross-linguistic variability of the adjectival class (that is, for the fact that if a language has only two major lexical classes, it has nouns and verbs and not adjectives) in terms of the non-iconicity of the modificative relation, the unmarked syntactic role of the adjective. The remainder of Chapter 3 deals with a number of subsidiary issues, including the analysis of possessive and attributive constructions (Section 3.4)—both of which are shown to be distinct from modification—and the potential application of the approach being developed here to the definitions of two minor lexical classes (adverbs and adpositions—Section 3.5).

The next chapter, Chapter 4, examines the implications that these definitions of lexical classes have for the typology of parts of speech systems, taking as a starting point a common four-member typology current in the literature (e.g. Schachter 1985; Bhat 1994). This typology proposes that, in terms of the three major lexical classes, there are four possible types of lexical inventory: full NAV inventories that distinguish three lexical classes, N[AV] languages where words that are adjectives in three-class languages are verbs, [NA]V languages where such words are nouns, and [NAV] languages which make no major-class distinctions whatsoever. As it turns out, this typology is easily generated by a feature system using the two criterial features, one syntactic and the other semantic, that make up the definitions of parts of speech put forward in Chapter 3. Thus, full inventory languages are said to be sensitive to both the syntactic and the semantic parameters, the [NAV] inventories are sensitive to neither, the N[AV] inventory is organized along purely semantic lines, and [NA]V inventory would then be subdivided on syntactic grounds alone. The last type of inventory, however, appears to present a problem in that, because the class of nouns inevitably includes the expressions of all prototypical semantic NAMES, it is impossible to completely avoid semantic characteristics when considering the way in which meanings are organized in the lexicon.

N[AV] inventories, however, are less problematic and two concrete examples of these are illustrated in Section 4.1, beginning with a discussion of lexical classes in the Salishan family of languages (4.1.1). Salishan languages have actually been cited in the literature as making no major lexical class distinctions (e.g. Kuipers 1968; Kinkade 1983), although based on the definitions for parts of speech proposed here they can be shown to make the basic distinction between nouns and verbs. What most Salishan languages do not do, however, is make a distinction between adjectives and verbs, all semantic predicates in these languages being both unmarked syntactic predicates and unmarked modifiers. Thus, the Salishan family (with the possible exception of Bella Coola, discussed in Section 4.1.1.4) organizes its lexica on purely semantic grounds and does not accord any special treatment to words expressing semantic predicates when they are used as modifiers. This is a very different situation from that found in Cora (Section 4.1.2), which—like Salish—conflates adjectives and verbs but requires that all modifiers of nouns appear inside relative clauses. Nonetheless, although they differ syntactically, Cora and Salish are essentially identical in terms of their parts of speech systems. As discussed in Section 4.1.2.3, the distinction between the two arises from differences in the syntactic treatment of parts of speech, not from any fundamental distinction in the number or type of lexical classes distinguished in the lexicon.

Section 4.2 returns to the problem of [NA]V inventories. Such inventories are quite common in the literature, the most frequent pattern being that represented by Quechua, where nouns and adjectives seem not to differ morphosyn tactically in either the role of modifier or the role of actant. Closer examination of Quechua (4.2.1), however, reveals that both the use of nouns as "modifiers" and the use of adjectives as actants are, in fact, marked uses when examined at the semantics <=> syntax interface. In the former case, noun—noun "modifier" structures can be shown to be attributive constructions as defined in Section 3.4. Attributives are marked and non-iconic in that they involve the elision of an underlying semantic predicate that is not realized in the syntax. Similarly, adjectives used as actants appear to be elliptical constructions making anaphoric reference to a nominal head whose identity is recoverable from discourse. This implies that adjectival actants used out of context are ungrammatical, a hypothesis that was put to the test in my fieldwork on Upper Necaxa Totonac, another reputed [NA]V language of the Quechua type (Section 4.2.2). In addition, the discussion of Upper Necaxa examines a number of other diagnostics for the noun—adjective distinction. The analysis here demonstrates both the use of primary diagnostics for markedness in criterial syntactic roles and the application of two secondary diagnostics, quantification and pluralization (Section 4.2.2.5). These exemplify the ways in which such tests can be used to sort out lexical class distinctions, as well as ways in which they can lead the investigator astray.

As a result of the discussions in Sections 4.2.1 and 4.2.2, languages of the Quechua—Totonac type are shown not to be eligible for classification as [NA]V languages because of the types of elision that occur both in noun—noun attributive constructions and when adjectives are used as actants. This leaves as the only possible type of [NA]V language a language like Hausa (4.2.3) which does not allow unmarked attributive constructions and avoids ellipsis by giving abstract nominal readings to "adjectives" when these appear in actantial position. Unfortunately, because of this last characteristic and one or two other features of the syntax, it turns out that "adjectives" in Hausa are considered by speakers to be the expressions of semantic NAMES and are, therefore, abstract nouns. This forces the reconsideration of the existence of the [NA]V inventory, which—as shown in Section 4.2.4—turns out to be a logically impossible type of language. This issue is discussed at some length in Chapter 5, which argues that the constraints on typological variation uncovered in the preceding chapter can be easily accounted for by replacing the free recombination of semantic and syntactic features proposed at the beginning of Chapter 4 with an algorithm for the subdivision of the lexical inventory that gives primacy to semantics over syntax. The result is a sufficiently constrained theory of typological variation in parts of speech systems based on rigorous and criterial definitions of each of the three major lexical classes.

Before launching into this discussion, it is probably a good idea to say a few words about the methodology applied here and some of the self-imposed limitations of this study. While this investigation does aspire to being an essay in typology, it is of a substantially different nature than the broad-based typologies inspired by Greenberg (1963) and others which attempt to take data from dozens or even hundreds of languages and distill from them universal patterns and statistical tendencies (a particularly outstanding study of this type is that of Nichols 1992). While this is a feasible task when dealing with highly salient or superficial features of a language such as unmarked word-order or inflectional patterns (although even features like these present a good number of problems), coming to terms with the parts of speech system of an unfamiliar language is a far more complicated task. As will be seen in some of the case-studies below, defining the lexical classes of an individual language involves an understanding of a wide variety of its semantic patterns, syntactic structures, and morphological features and, in the most difficult cases, requires the investigator to go well beyond the type of information included in most ordinary descriptive grammars. Given the degree of familiarity required to sort out the parts of speech systems (at least in problematic languages, which—after all—are the most interesting ones for this type of study) and the type of information available, examining the lexical inventories of hundreds of languages is the work of a lifetime. Instead, I have chosen to let others do much of the inductive work and, based on their findings, have identified four types of lexical inventory that seem to enjoy a great deal of currency in the literature. From these, I examine a small number of languages belonging to two of them.¹ The down side of this is, of course, that there may be among the hundred of languages that I did not examine numerous exceptions or even types that I have not anticipated. However, by focusing on a small number of languages (five, if the Salishan family counts as one), it is possible to speak with a little more assurance and be confident that some measure of justice is done both to the data and to its interpretation.

Also in aid of constraining the task at hand, I will limit the present discussion to the three major classes of noun, verb, and adjective, and I will only be concerned with the classification of "lexical meanings"—that is, with meanings that refer to items or describe real-world qualities and events rather than expressing grammat...

Table of contents

• Cover
• Title
• Copyright
• Contents
• Acknowledgements
• Abbreviations
• Note on phonological transcriptions
• Chapter 1 Introduction
• Chapter 2 Definitions of lexical classes
• Chapter 3 Semantics, syntax, and the lexicon
• Chapter 4 Types of lexical inventory
• Chapter 5 Conclusions
• References
• Index