Approaches to contrast

The notion of contrast has been central to linguistics since Saussure's famous dictum that ‘dans la langue il n'y a que des différences … sans terms positifs’ [‘In a language there are only differences, and no positive terms’] (Saussure 1972 [1916]: 166). ‘The sound of a word’, according to Saussure, ‘is not in itself important, but the phonetic contrasts which allow us to distinguish that word from any other’. That is, a phoneme is identified not only by its positive characteristics – for example, the fact that it sounds like [i] – but also by what it is not – that is, by the sounds it contrasts with.

The notion of contrast can be understood at several different levels. At the most basic level, it can refer simply to whether two sounds contrast in a language or not. In English, for example, [i] is different from [i], and these vowel sounds alone are able to differentiate words in the language: sheep [∫Ip], for instance, is different from ship [∫Ip]. This contrast recurs in many other word pairs, such as cheap ∼ chip, seat ∼ sit, seen ∼ sin, meal ∼ mill, reed ∼ rid, and so on. Compare this situation with that obtaining in Israeli Hebrew, which has a single phoneme in the part of the vowel space where English has two. This phoneme, which can be represented as /i/, is pronounced somewhere between English /i/ and /i/ (Chen 1972).

Contrastive specification: an elusive problem

It is far from obvious how to decide, for a given phoneme in a given language, which of its features are contrastive and which are not. The problem is made even more elusive by the fact that it does not appear to be difficult. In particular situations we may have intuitions about what the answer must be. But our common-sense intuitions may lead us astray, in this area as in others. Or we may find that we can follow more than one logical chain of reasoning, each of which may appear to be sound, but which lead to different and incompatible conclusions.

To give something of the flavour of this problem, both its seeming obviousness and real difficulty, I would like to begin with a quote from Stephen Anderson (1985:96–7). Anderson is illustrating Trubetzkoy's (1939) notion of phonemic content, intended to be the sum of the contrastive properties of a phoneme: ‘If we consider [English] /t/, for example, we can see that this segment is phonologically voiceless (because it is opposed to /d/), non-nasal (because opposed to /n/), dental (because opposed to /p/ and /k/), and a stop (because opposed to /s/ and to /θ/).’

Anderson is not proposing a detailed analysis of English; he is simply illustrating what some of the contrastive features of English /t/ would be in a Trubetzkoyan analysis, and presumably in any analysis of contrast that used these features.

Introduction: Jakobson and his collaborators

The work of Roman Jakobson and his colleagues merits a separate discussion. Many of the main ideas in Trubetzkoy's Gründzuge were worked out in collaboration with Jakobson. To Jakobson is due the notion of the distinctive feature. Whereas Trubetzkoy allowed features to be gradual (multivalued), Jakobson eventually proposed that all features are binary. The notion that language and cognition crucially involve binary dichotomies became central to Jakobson's thinking, and this idea was later taken over into generative phonology.

In the 1950s and early 1960s, Roman Jakobson, Morris Halle and their colleagues wrote a series of publications that laid the groundwork for the next phase of distinctive feature theory and what eventually became the theory of generative phonology. Much critical attention, then and afterwards, was directed at various aspects of their theory that struck observers as being the most controversial and innovative, such as the nature of distinctive features, the relation of phonemes to allophones, and the organization of the grammar.

What is of greatest interest for our present purposes is an aspect of their work that attracted relatively less attention. This is their development and utilization of a contrastive hierarchy of distinctive features. We have seen that such a hierarchy is already implicit in much of Trubetzkoy (1939), and it could be that this was one source of the idea. It also fits well with Jakobson's general emphasis on dichotomies as a fundamental aspect of cognition: if the distinctive feature is grounded in a binary discrimination, then it makes sense that an inventory is built up via a succession of such binary splits.

Introduction

In section 3.7 I sketched the outlines of a theory of phonology that was distilled from the leading ideas discussed in chapter 3. This theory adopts the Contrastivist Hypothesis, which holds that phonology computes only contrastive features. It determines what the contrastive features in a language are by applying the SDA to a contrastive feature hierarchy for that language. In keeping with the Contrastivist Hypothesis, phonological activity serves as the chief heuristic for determining what the feature hierarchy is for a given language.

Though the ingredients for such a theory were in place by the 1930s, phonological theory did not develop in this direction; why it did not was the subject of chapters 4–6. These chapters show that the theory of section 3.7 has never properly been put to the test. In this chapter I argue that these ideas remain viable and indispensable to an explanatory theory of phonology.

Of course, any contemporary effort to implement such a theory must take account of advances in phonology since the 1930s. For example, the diagnostic given in (38d) of chapter 3, that a contrastive feature must be present in all the allophones of a phoneme, is not consistent with the generative phonological conception that phonology is relatively abstract with respect to phonetics. In keeping with Chomsky and Halle's arguments against taxonomic phonemics, it is unlikely that we can put limits on the degree to which a segment may be modified in the course of a derivation.

This book has been about the contrastive hierarchy in phonology. In one sense, ‘phonology’ refers to phonological theory as it has been understood and practised from the 1920s until today. I have tried to trace the history of the notion of a contrastive feature hierarchy, and its various guises as it has appeared and disappeared in different versions of phonological theory. In another sense, ‘phonology’ can be understood as a component of the language faculty whose properties phonologists are trying to discover. In this sense, I have looked for evidence of the role that the contrastive hierarchy plays in contributing to an understanding of this cognitive domain.

In both aspects of this study the fate of the contrastive hierarchy has been closely connected to another central concept, the Contrastivist Hypothesis. The Contrastivist Hypothesis states that only contrastive features are active in the phonology. To properly test whether this hypothesis is correct, we need to have reliable ways of identifying the ‘contrastive features’. We also need to know what it means for a feature to be ‘active’, and what place ‘in the phonology’ refers to; but my main concern here has been contrast. I have argued that the contrastive hierarchy is crucial to the Contrastivist Hypothesis, because it provides the proper way to identify contrastive features. Conversely, the Contrastivist Hypothesis is crucial to the contrastive hierarchy, because the former supplies the main empirical reasons for being interested in the latter, at least as far as phonology is concerned.

Introduction

In this chapter I will consider some other approaches to phonological contrast that have been advanced in the recent phonological literature. I will start with theories that are conceptually quite different from the approach I have been taking and then consider those that have more in common with it.

In section 8.2 I consider an alternative explanation of the typology of labial harmony triggers reviewed in section 7.4.3. There, I argued that the observed relation between harmony triggers and inventories supports the Contrastivist Hypothesis. Kaun (1995) advances what appears to be a very different explanation, grounded in perceptual-functional phonetic constraints. I will argue that her account is not, in fact, a real alternative to the explanation I proposed in chapter 7.

Dispersion Theory is concerned with contrast at a perceptual phonetic level, and in section 8.3 I review a dispersion-theoretic account of a phonological change in the history of Russian, proposed by Padgett (2003a). I will argue that this version of Dispersion Theory is enmeshed in descriptive and explanatory complications that may be difficult to sort out. I will present an MCS-style alternative solution that is conceptually much simpler.

Section 8.4 looks at Structured Specification, or natural classes, theory (Broe 1993; Frisch 1996; Frisch, Pierrehumbert and Broe 2004). Rather than designate features as being ‘contrastive’ or ‘redundant’, as the SDA does, this theory is indirectly sensitive to the contrasts in an inventory.

Introduction

Optimality Theory (OT; Prince and Smolensky 2004) is a radical departure from the derivational model of previous versions of generative phonology. Any new theory puts old questions into a new light, and it is illuminating to explore the relationship between OT and the Contrastivist Hypothesis. I will show that the contrastive hierarchy, being a static set of conditions, lends itself very easily to formulation in terms of a set of constraints, and hence to OT. At the same time, I will argue that OT is not itself a theory of contrast, but is capable of instantiating a wide range of such theories.

The relationship between OT and the Contrastivist Hypothesis is more complex. It appears that the insights of a contrastivist approach can best be captured in a serial (derivational) model of OT in which some of the restrictions of the standard parallel version are relaxed.

I will begin with a brief review of the essentials of OT (section 6.2), and then I will present some early treatments of contrast within OT(section 6.3). I will consider how the contrastive hierarchy might be incorporated into OT in section 6.4. Related issues are discussed in sections 6.5–6.7. Dispersion-theoretic approaches to contrast developed within OT are considered in section 8.4.

Optimality Theory

OT (Prince and Smolensky 2004, first published 1993) is a theory of constraint interaction that posits violable ordered constraints. For example, to account for the universal preference for syllables to have onsets and to avoid codas, Prince and Smolensky posit the constraints in (1).

Chomsky and Halle's revolution in phonology

Chomsky and Halle's approach to phonological theory, as represented in their major work The sound pattern of English (Chomsky and Halle 1968, henceforth SPE), represented a sharp break with the main currents of American linguistics that immediately preceded them. Nevertheless, some elements of the Jakobson-Halle approach to phonology were continued in generative phonology. In general terms, Chomsky and Halle's theory of generative phonology was a synthesis of Jakobson and Halle's theory of distinctive features and phonemic analysis, revised in the light of Chomsky's emphasis on formal explicitness, simplicity, and abstractness and autonomy of mental representations (Dresher 2005).

One aspect of Jakobson and Halle's theory that did not make it into SPE was the contrastive hierarchy. As discussed in the previous chapter, the atmosphere was not conducive to assigning a special role to contrastive specifications. One of the major points of difference between the older structuralist approach and generative phonology was the status of the taxonomic phoneme (Chomsky 1964), alluded to above in section 4.6. Classical generative phonology posits only two significant levels in the phonology: the underlying lexical, or ‘systematic phonemic’ level (known as the ‘morphophonemic’ level in structuralist theory), and the surface phonetic level (a level that was not recognized as a systematic level in structuralist phonology, following the arguments of Bloomfield 1933). Instead of a phonetic level, the neo-Bloomfieldians posited what Chomsky called the ‘taxonomic phonemic’ level, a level defined largely in terms of surface-oriented criteria designed to simplify the acquisition problem (see Dresher 2005 for discussion).