Phonology Research Paper

1. Goals Of Phonology

The study of a language’s particular sound properties has a number of aspects. First, it must be able to characterize the language’s inventory: which phonetically possible sound types occur in utterances in the language? Second, it must characterize matters of contrast: which of the phonetic differences that occur in the language can serve to distinguish utterances (words, sentences, etc.) from one another? Third is the matter of contextual limitation: even though some property P occurs in language L, are there some environments from which P is excluded? And when P is apparently a property of (some part of ) some element but that element occurs in a position from which P is excluded, what other property—if any— appears in its place? And finally, there is the description of alternation: when the ‘same’ linguistic element appears in different overt forms in different environments, what systematic differences occur? What conditions govern the range of phonetically distinct forms that can count as the ‘same’ word, morpheme, etc.? Different answers to these questions yield different phonological theories.

It should be noted that the present research paper is limited to the sound structure of spoken languages, and ignores the expression systems of manual or signed languages. This is misleading in important respects; research has shown that most of the basic principles of spoken language phonology are also characteristic of the organization of the expression systems of signed languages as well (Coulter 1993). Just as words are composed of sounds, and sounds of component properties, signs are also composed from structured, language-particular systems of more basic constituent elements. Units such as the syllable have close parallels in signed languages. While there are clear differences that depend on modality, these appear on further examination to be relatively superficial. A comprehensive theory of phonology as a part of the structure of natural language ought to take these broader issues into account. Until quite recently, however, the possibility of deep structural parallels between speaking and signing has not been raised, and the discussion below reflects this (undoubtedly unfortunate) limitation as well.

1.1 Some History

Prior to the early twentieth century, studies of sound in language concentrated on the ways in which sounds are made (articulatory phonetics), often confusing the letters of a language’s writing system with its sounds. Toward the end of the nineteenth century, however, increasing sophistication of measurement techniques made it possible to explore a much wider range of differences among sounds, and to lay out the structure of speech in vastly greater detail. Somewhat ironically, perhaps, the explosion of data which resulted from the ability of phoneticians to measure more and more things in greater and greater detail began to convince them that they were on the wrong track, at least as far as increasing their understanding of particular languages was concerned.

Much of what was found, for example, involved the observation that speech is continuous, such that whatever is going on at any particular moment is at least a little different from what has gone on just before and what will go on immediately afterwards. A full characterization of an utterance as a physical event requires the recognition of a potentially unlimited number of distinct points in time, but it is clear that our understanding of an utterance as a linguistic event is hindered, rather than helped, by the recognition of this continuous character of speech. Speech normally is represented as a sequence of a small number of discrete segments, strung out in consecutive fashion like beads on a string; such a segmental representation vastly facilitates the discovery of regularity and coherence, but it must be emphasized that there is no direct warrant for it in either the acoustic or the articulatory observable data of speech, and it constitutes a fairly abstract (though undoubtedly appropriate) theory of how sound is organized for linguistic purposes.

It is clear that the role of particular sound differences varies considerably from one language to another. Thus, in English, the vowel sound in the word bad is much longer than that in bat (more than half again as long), but such a difference in length is always predictable as a function of the following sound, and never serves by itself to distinguish one word from another. In Tahitian, in contrast, essentially the same difference in length is the only property distinguishing, for example, paato ‘to pick, pluck’ from pato ‘to break out.’ A theory of sound that attends only to physical properties has no way of clarifying the quite different functions these properties may have across various languages. This is not to suggest that phonetics is wrong, but rather that there is more to be said.

1.1.1 De Saussure And The ‘Phonemic Principle.’ The great Swiss linguist Saussure (1916) was the first to stress that in order to understand the role of sound in language it is necessary to focus not ( just) on the positive properties of sounds, but on their differences. He suggested that in the study of individual languages, as opposed to general phonetics, utterances should be characterized in such a way that two such representations might differ only in ways that could potentially correspond to a difference between two distinct messages in the language in question. Thus, since long and short vowels never (by themselves) distinguish two distinct utterances in English, the difference should not be indicated in that language; while for Tahitian, it must be. A representation with this property will be called phonological; it will obviously be specific to a particular language, and the distinctive elements that appear in it can be called the phonemes of that language.

While de Saussure enunciated this principle quite forcefully and persuasively, he provided few specific details of just what a phonological representation should look like. There are in fact a variety of ways in which his insight could potentially be realized, and much subsequent discussion in phonology hinges on these differences of interpretation.

1.1.2 The Development Of Phonology As A Theory. Various individual investigators arrived at conclusions similar to de Saussure’s about the importance of attention to language-particular sound contrasts. One of these was the Polish linguist Baudouin de Courtenay (1972), whose work actually antedated de Saussure’s, but attracted little attention due to his isolation in Kazan. He developed a sophisticated view of the relation between phonetics and phonology both in individual grammars and in linguistic change. As transmitted by his later students, Baudouin’s views on the nature of the phoneme constituted an important strand in thinking about language as this developed in Russian linguistics in the early years of the twentieth century. This, in turn, provided the background from which the work associated with the Linguistic Circle of Prague grew in the 1920s and 1930s.

Two of the most prominent members of the Prague Circle were Trubetzkoy (1939) and Jakobson (1941). In their studies of Slavic languages and their histories, they stressed the notion that the collection of potentially contrastive sound types in a language was not simply an inventory, but a highly structured system. This system is organized in terms of a small number of mutually orthogonal dimensions (such as voicing, stop vs. continuant, nasality, etc.), each of which serves in parallel fashion as the basis of multiple contrasts. The notion that the fundamental terms of sound structure in language are these properties themselves and not (or at least not only) the complete sounds they characterize has remained an important component of most subsequent theorizing. The analysis of phonological structure in terms of its constituent basic contrasts, in turn, has served as a model for a variety of other disciplines in the Humanities and the Social Sciences apart from linguistics.

1.1.3 American Structuralist Phonology. Early thinking about sound structure in America was dominated by the anthropological interests of Franz Boas, and focused on an accurate rendering of the sound contrasts in the comparatively ‘exotic’ indigenous languages of the new world. Boas’s student Edward Sapir, however, was concerned to place the study of language in the broader context of an understanding of the human mind and society. As such, he stressed (Sapir 1925) the notion that the elements of sound contrast in a language should be regarded as having a primarily mental reality, part of the speaker hearer’s cognitive organization rather than as external, physical events.

The rise of positivist views of science in the 1930s, and especially of behaviorist psychology, made Sapir’s sort of mentalism quite marginal, and replaced it with more rigorous operational procedures for investigating notions of contrast. Especially associated with the ideas of Bloomfield (1933) and later structuralists such as Bloch and Harris (1951), the result was a theory of the phoneme based exclusively (at least in principle) on a set of mechanical manipulations of corpora of observed linguistic data, from which a set of contrasting minimal elements was to be derived. The central notion of this theory was a phonemic representation related to surface phonetic form in a way that would later be formulated explicitly as a condition of Biuniqueness: the requirement that given either a phonetic or a phonemic representation of an utterance in a given language, that could be converted uniquely into the other (disregarding free variation) without additional information. The perceived rigor of this notion led to its being widely taken as a model not only for the study of other areas of linguistic structure, but for other sciences as well.

1.1.4 Generative Phonology. The phonemic theories of American Structuralists provided a way to characterize linguistic contrasts, the inventories of sound types used in a given language, and the ways sounds can be combined into larger structures, but other aspects of sound structure were less satisfactorily accommodated within those views. In particular, questions of the ways in which unitary meaningful elements change in shape according to their sound context (or ‘allomorphy’: see Morphology in Linguistics) failed to receive systematic treatment. Since any difference among sounds that could serve to contrast linguistic elements was ipso facto a difference between irreducibly basic terms, there was really no way to express the notion that a single item could take a variety of forms (as in the prefixes in inefficient, imprecise, irregular, illegal, etc.) except by simply listing the variants. Such a list is undoubtedly appropriate for cases such as the forms of English to be (am, are, is, was, were, etc.) which are unrelated to one another in form; but in many other cases, the variation is transparently systematic and a function of the sounds in the element’s environment. This sort of variation was recognized by structuralist phonologists, but relegated to marginal status.

Beginning with work of Morris Halle, a student of Jakobson, linguists began to question the centrality of surface contrasts in sound structure. The result was a new view that allowed morphophonemic regularities as well as more superficial ones to be accommodated within a phonological description. The success of this more abstract notion of sound structure in dealing with hitherto irresolvable problems in the description of stress contributed greatly to its success, and the resulting theory of Generative Phonology as developed in the work of Halle together with Noam Chomsky rapidly became the dominant view in the field by the middle of the 1960s.

1.2 Phonology As A System Of Rules And Representations

A basic insight in the development of Generative Phonology was the proposal that it is not only the representation of linguistic elements in terms of basic contrasts that matters: an adequate theory must characterize what a speaker knows about the sound system of the language, and that includes regularities of variation and alternation as well as inventories of basic elements. Combining these two aspects of phonological knowledge required the explicit recognition of a system of rules (expressions of regular patterns of contrast and variation in sound shape) in addition to the theory of representations. Developing an adequate theory of phonological rules, in turn, necessitated a notion of phonological representation that was related to surface phonetic reality in a much more complex and indirect way than the phonemic representations of structuralist linguistics. The central problems of phonological theory came to be articulated in terms of the theory of rules, their nature, and their interaction, and the underlying phonological representations that need to be posited in order to allow them to be expressed in their full generality.

2. Issues In Phonological Theory

As reviewed in Sect. 1.1 above, the problem of phonological description was originally conceived as a matter of discerning the range of contrastive sound units in a language, and arranging them in a system that brings out the dimensions of their contrast. When phonology is seen as a form of knowledge, however, as generative phonologists have stressed, the sound structure of natural language takes on quite a different form and presents quite different problems. Among these are (a) the nature of underlying (‘phonological’ or ‘phonemic’) representations as well as surface (‘phonetic’) representations; (b) the ways in which phonological regularities serve to relate phonological to phonetic representation, including the interactions that may obtain among regularities; and (c) the relation between phonological form and other aspects of linguistic knowledge, such as word structure and sentence structure. Various aspects of these questions are addressed below.

2.1 The Abstractness Of Phonological Representation

Initial reaction to the proposals associated with generative phonology centered on its abandonment of a phonemic representation based on the condition of bi-uniqueness. Relaxing this defining characteristic of Structuralist phonemics led to phonological representations that were considerably more abstract in their relation to phonetically observable properties. The proposals of Chomsky and Halle (1968) concerning the analysis of English, for example, involved positing final ‘silent e’ in words like ellipse, burlesque (phonologically /Ellipse/, /bVrleske/), geminate consonants in words like confetti (/kVnfetti/), a distinction among /k, s, c/ such that acquiesce is phonologically /æckwiesce/, etc. Indeed, there were no constraints whatsoever on the relation between phonological and phonetic form, apart from the desire to set up underlying forms from which as much variation as possible in the shapes of particular morphological units could be predicted. Since much of this variation is the residue of earlier sound changes that have affected the same element in different ways in different environment, many early generative analyses resembled historical accounts of the language more than they did the knowledge of current speakers.

The perception of such apparently excessive abstractness led to proposals for constraining the operation of phonological rules in grammars, and the kinds of representation that should be posited. Kiparsky (1973), in particular, suggested a variety of conditions that would have the effect of prohibiting rules of ‘absolute neutralization’ (by which some posited phonological distinction is eliminated uniformly in surface forms, such that it never corresponds to a phonetic distinction) and other abuses of the theory’s representational freedom. This led phonologists to seek external evidence for the reality of the representations they assumed, in the form of data from poetic systems, historical change, language games, behavioral experiments, and other sources outside of the phonology itself. Such a notion would have been incoherent on the assumptions of Structuralist phonology, for which the properties of the system are a matter of its internal logic, but became important with the shift to a view of phonological form as a matter of speakers’ knowledge.

It is now apparent that phonological representations, while more abstract than assumed by structuralists, are still rather constrained in the extent to which they can deviate from phonetic form. There is no real consensus in the field, however, as to what constraints are appropriate on the operation or form of grammars so as to ensure an appropriate result. In the absence of some motivated grounds for decision in general, the status of specific conditions remains somewhat arbitrary. Replacing the conception of a grammar as a set of rules by that of a set of simultaneously evaluated partially violable constraints, as discussed in a later section, severely limits the potential abstractness of underlying representations in a way that may prove to resolve these issues in a principled way.

2.2 The Interaction Of Phonological Regularities

In the nature of a Structuralist phonemic representation, all of the regularities expressed in the grammar are mutually independent. That is, it is only necessary to know the phonemic environment of a phoneme to predict its phonetic realization; and it is only necessary to know the phonetic environment of a phonetic segment to establish its phonemic correlate. When the bi-unique relation between these levels of representation is relaxed, however, more complex possibilities arise for regularities to interact with one another.

Consider the formation of English regular plurals, for example. For the three words cut, dog, horse (roughly [kæt], [dog], [hors] respectively, phonetically) and others like them, the corresponding plurals are cats, dogs, horses ([kæts], [dog_Z], [hors_IZ). Assume that the phonological representations of the nouns are essentially the same as their phonetic forms, and that the regular plural ending has a single constant phonological form: /z/ for concreteness’ sake, though the point to be made is independent of this choice. Now there are two distinct regularities that are involved in determining the pronunciation of the plural ending:

(a) If the noun ends in a sibilant ([s, z, s, z, c, j]), a vowel [_I] appears between the stem and the ending; and

(b) If the stem-final sound is voiceless, the ending is voiceless ([s]) as well.

Now consider how these principles interact in determining that the pronunciation of horses (phonologically /hors + z/) should be [hors_IZ]. In this case, the conditions for both of the rules above, b as well as a, are met, and we might expect both to apply, yielding (incorrect) [hors_IS]. The relevant observation is the following: the vowel inserted as a result of rule a has the effect that the stem and the ending are no longer adjacent, and so rule b is inapplicable. That is, the regularity represented by rule b in this case presupposes that the set of clusters of consonants which will remain adjacent in surface forms (a matter which is potentially affected by applications of rule a) has been determined. This logical relation of presupposition between rules is generally expressed by saying that rule a applies ‘before’ rule b, though it is important to note that no claim of temporal priority is thereby asserted, but rather a purely logical notion.

Much of the abstractness of early generative phonological analyses was made possible precisely by the availability of rule ordering as a descriptive device: the possibility of specifying rule interactions such as the above as an independent parameter within grammars. Although arguments for the necessity of such ordering formed an important part of the literature, and particular accounts sometimes involved apparently crucial depths of ordering as great as twelve to fifteen, many linguists felt that (as part of the general campaign to reduce abstractness in phonology) stipulated ordering relations ought not to be permitted as a descriptive device. Despite the existence of cases apparently requiring irreducible specification of relative order (e.g., dialects with the same sets of rules and underlying forms, differing in their surface forms only as a consequence of different interactions among the rules), and the fact that all proposed formulations of principles from which observed rule interactions could supposedly be predicted had obvious and well established counter-examples, a consensus developed in the field that such ‘extrinsic’ rule ordering statements ought to be prohibited. This has left a number of kinds of empirically observed interactions unaccounted for, a problem which has persisted as phonologists have turned from rules to constraints as descriptive devices.

2.3 The Structure Of Representations In Phonology

As already noted, the description of speech at either the phonetic or the phonological level as composed of a sequence of discrete segment-sized units is a significant abstraction from physical reality. Its justification comes not from the observable facts, but from the extent to which it allows the analyst to uncover what is orderly and coherent in linguistic structure. By the mid 1970s, however, it had become apparent that a purely segmental organization of representations impeded the description of linguistic regularity in several respects.

One of the first of these problems to be discussed arose from the analysis of systems of tonal contrasts, common in the languages of China, Africa, and in fact much of the world. Careful analysis revealed two ways in which tonal properties were problematic for strictly segmental models: on the one hand, what appeared to be a single tonal specification might take as its scope more than a single segment (perhaps the vowels of several consecutive syllables, or an entire word); and on the other hand, what appeared to be a single segment (a unitary short vowel, for example) might have a tonal specification involving two or even more consecutive tonal levels. If it is assumed that each of the consecutive segmental units of which a representation is composed bears exactly one specification for each potentially distinctive property, and the specifications of distinct segments are independent of one another, both of these situations are anomalous.

As a result of these observations, the segmental view came to be replaced with an autosegmental notion of representation. On this picture, specifications for each potentially distinctive property (or feature) succeed one another discretely, but an additional dimension of representation is the specification of the way these are synchronized or aligned with one another. In the limiting case, where each specification for a given feature is temporally aligned with one and only one specification for each of the other features, the standard segmental picture arises, but more generally, one or more instances of many-to-one alignment may be found. Once this view was articulated, it became apparent that many properties other than those of tone were also most appropriately described in this way. Nasality, in particular, behaves similarly in many languages and the very common phenomenon of assimilation is often best treated as an alteration in the scope of some features rather than a change in their values.

On the other hand, the full flexibility of autosegmental representations predicts much more independence of specification than is actually found. Features of place of articulation typically behave as a unit, despite their logical independence from one another, as do the features characterizing laryngeal activity. This observation gave rise to the notion that phonological features are not just an unordered set, but instead are organized into a sort of hierarchical arrangement, such that for example features of consonantal place can be referred to as a unit. There is considerable consensus on the overall form such a feature geometry should take, but many details and relations among properties remain unsettled.

Another representational issue concerns the existence of structural units larger than the single segment, such as the syllable. Classical generative phonology, as represented by, for instance, Chomsky and Halle (1968), makes no appeal whatever to syllables (or any other unit above the segment). This was not simply an omission on their part, but rather a systematic claim to the effect that segmental specification could always suffice to express any phonological generalization: that is, that any observation involving syllables (of which there are many in Chomsky and Halle 1968) could be satisfactorily reformulated in a way that dispensed with any units other than segments. The boldness of this claim invited immediate attack, and indeed it soon became clear that there were indeed aspects of phonological structure that required reference to syllables and other supra-segmental units. The resulting theory of the prosodic hierarchy, by which segments are grouped into syllables, themselves grouped together into metrical feet, which in turn form constituents of phonological words (perhaps with some additional categories, according to some writers) has become another standard assumption about phonological structure.

2.4 Interactions Between Phonology And Other Areas Of Grammar

The original conception of a generative phonology was as a component of the grammar that served to map lexically specified syntactic structures onto phonetic form, a representation appropriate to serve as the instructions to the language-independent mechanisms of speech production. This picture logically entails the notion that words, phrases, and sentences are already fully assembled in the input to the phonology. This assumption, however, has proven to be problematic.

The notion that phonological properties interact with the syntactic environment in which forms occur is quite a familiar one, especially in the description of accent and intonation. Word formation, too, may depend in some instances on the assignment by phonological rule of phonological properties in a way that contradicts the assumption that phonology applies strictly to the output of this part of the grammar.

The most extensively elaborated picture of how phonology interacts with the rest of grammar is presented by the theory of Lexical Phonology (see papers in Hargus and Kaisse 1993 for introduction and explication). Briefly, this picture distinguishes between lexical and post-lexical regularities, with the former applying in a way that interacts with word-formation, and the latter applying to a representation in which fully formed words appear in their syntactic context. The lexical rules, in turn, may (depending on the language) be divided into two or more strata, for example, rules applying to roots, stems, or words. A process of word formation takes some lexical unit as its input and yields an output (affixed or otherwise altered in form), with this output then subject to appropriate phonological modification. The resultant representation may serve as the input to further cycles of word formation plus phonology, until a complete surface word is arrived at.

These units, assembled into larger syntactic constructions, are then subject to the post-lexical rules. While various forms of lexical phonology generally constitute the background assumption of much phonological discussion, the correctness of such a view of the overall architecture of grammar has come into question. In particular, the rise of constraint based theories implying a single stage in the conversion of underlying to surface form has appeared incompatible with the sequential, derivational character of lexical phonological description. Some researchers have sought to reconcile these two trends in phonological theory, while others have attempted to augment one or the other sub-theory in such a way as to include the results of the other.

2.5 Rules vs. Constraints In Phonological Description

Formulations of phonological regularities since the 1950s have drawn on the mechanisms of automata theory: standardly, these have been expressed as rewriting rules that map an input representation onto an output, one step at a time. Developments in the theory of computation, however, have given rise to an alternative possibility. Instead of converting inputs to outputs in several stages, with each step involving the imposition of a single regularity in the computation of a new representation, one might formulate all of the regularities as elements of a single system of constraints. If these constraints are ranked with respect to one another, it is possible to say that any particular constraint may be violated in the output, provided that such a violation is motivated by the need to avoid violating a more important (higher ranking) constraint. The resulting theory provides an architecture of grammar that is (at least apparently) very different from that which has dominated most of the literature in phonology. For example, the kinds of interaction among regularities treated above as matters of rule ordering can no longer be described in this way, since all of the constraints in the grammar apply (in principle) in a single, simultaneous block. Constraints can have a language-particular ranking among them-selves, but this is a matter of priority, not lgical presupposition. Other differences have been alluded to in the discussion above.

Constraint based theories along the lines of Optimality Theory (Prince and Smolensky 1993) have been shown to have a number of significant advantages over rule-based accounts, and currently constitute the most active area of investigation in phonological theory. It remains to be seen, however, what character a comprehensive theory will have once the substantial body of results achieved within earlier, apparently incompatible frameworks has been incorporated comprehensively.

Bibliography:

1. Anderson S R 1985 Phonology in the Twentieth Century: Theories of Rules and Theories of Representations. University of Chicago Press, Chicago
2. Baudouin de Courtenay J 1972 Selected Writings of Baudouin de Courtenay. Indiana University Press, Bloomington, IN
3. Bloomfield L 1933 Language. Holt, New York
4. Chomsky N, Halle M 1968 The Sound Pattern of English. Harper and Row, New York
5. Coulter G R 1993 Current Issues in ASL Phonology (Phonetics and Phonology 3). Academic Press, San Diego
6. Fischer-Jørgensen E 1975 Trends in Phonological Theory. Akade-misk Forlag, Copenhagen, Denmark
7. Hargus S, Kaisse E 1993 Studies in Lexical Phonology (Phonetics and Phonology 4). Academic Press, San Diego
8. Harris Z 1951 Methods in Structural Linguistics. University of Chicago Press, Chicago
9. Jakobson R 1941 Kindersprache, aphasie und allgemeine lautgesetze. In: Jakobson R (ed.) Selected Writings, Vol. I (English trans. 1968 Child Language, Aphasia and Phonological Universals). Mouton, The Hague, pp. 328–40
10. Joos M 1957 Readings in Linguistics, Vol. 1. American Council of Learned Societies, Washington, DC
11. Kenstowicz M 1994 Phonology in Generative Grammar. Blackwell, Oxford, UK
12. Kiparsky P 1973 Phonological representations. In: Fulimura O (ed.) Three Dimensions of Linguistic Theory. The TEC Corporation, Tokyo
13. Prince A, Smolensky P 1993 Optimality Theory: Constraint Interaction in Generative Grammar. Rutgers University, New Brunswick, NJ
14. Sapir E 1925 Sound patterns in language. Language 1: 37–51
15. de Saussure F 1916 Cours de linguistique generale. Payot, Paris
16. Trubetzkoy N S 1939 Grundzuge der Phonologie. Travaux du Cercle linguistique de Prague 7