Constraint Based Phonology



 1. Representative-- Steven Bird’s Homepage

2. Theory--

  Unlike the traditional model of grammatical organization, here the different linguistic modules operate in parallel to provide a pool of constraints which independently ensure the phonological, syntactic and semantic well formedness of an utterance. In the traditional view, there is just the one hierarchy, with phonological units such as distinctive features at the bottom, morphemes and words in the middle reaches, and phrases and sentences towards the top. The associated processing model is serial: recognition involves a phonological component, which passes its output to the morphological or syntactic component, and so on.

We wonder how the conception of constraint is used in phonology? First, a constraint is an empirical generalization about a collection of objects. Second, constraints interact to mutually restrict a solution space. Finally, the order of application of constraints is immaterial to the end result.

    Constraints are familiar to phonologists in the form of morpheme structure constraints and surface structure constraints. The latter are essentially statements about the distribution of phonological primes. A constraint-based phonological analysis is to be evaluated along the same dimensions of empirical and explanatory adequacy as standard generative analyses. In philosophy, artificial intelligence and cognitive science, it has often been found useful to distinguish between intension and extension (or description and object). The distinction between descriptions and objects introduces a degree of flexibility into linguistic notation that is unavailable without this distinction. For example, it is possible for different descriptions to denote the same object (e.g. ‘morning star’/ ‘evening star’), while some descriptions are satisfied by no objects (e.g. ‘unicorn’).  Also, it is possible for the description ‘morning star’ to be interpreted in various ways; it might denote the planet Venus, a brand of dishwashing detergent, or – most relevant here – a particular speech event. If we incorporate the distinction into phonology, we get a fundamentally different view of the relationship between phonology and phonetics from the traditional view. Phonological representations are descriptions of real-world utterances (the objects).

A further consequence of adopting the description/object distinction is that there can be no rule ordering. Rules are just descriptions. Although a rule might be stated in the form of a rewrite rule a → b, this rule is interpreted as a logical implication, equivalent to the expression ﹁ a Ú b. this contrasts with the standard interpretation of these rules, evident from rules like [+voice] →[-voice], where the objects satisfying the description on the left side of the arrow are modified (minimally) so that they now satisfy the description on the right side of the arrow instead.

    A phonological rule such as homorganic nasal assimilation describes the set of all forms which are compatible with the rule.. Thus, rules act as constraints.

    Morpheme structure constraints, which state regularities about certain classes of morpheme, fit naturally into this model. For example, the vowel harmony in Yoruba is restricted to nouns, and so the required constraint on vowel sequences might be formulated as a property of the noun type in the lexicon.

The advent of generative phonology , with its highly procedural model of phonology, drew attention away from constraints. Properties of surface forms were not to be expressed directly but arose out of the combined action of lexical specifications, morpheme structure rules and derivations. No rule could be said to act like a constraint, since the effect of the rule could always be undone by later rules. Although this view has survived to the present intact, it has also been challenged right from the start.

Complaints about the abstractness of generative analyses led to new models which retained the earlier notions about constraints. For example, Shibatani (1973) advocated ‘surface phonetic constraints’, which ‘state possible and impossible combinations of phonetic features at the phonetic level, i.e. represent true generalizations about the phonetic pattern of a language’. To illustrate Shibatani’s approach, we shall consider his analysis of Turkish vowel harmony. Shibatani views vowel harmony as ‘cooccurrence restrictions on vowel features in successive syllables’. Alternating vowels are expressed as archiphonemes (i.e. sets of vowels )in the lexicon, and the selection of the appropriate vowel is done automatically by the surface phonetic constraint for harmony. 

The Turkish word kizlar ‘girls’ is represented morphophonemically as /kiz + lEr/ , where /E/ = {/e/ , /a/ }. The stem vowel is fully specified because it does not alternate, while the suffix vowel is not fully specified. The surface value of the suffix vowel is determined by the above mentioned rule which instantiates /E/ as /a/. If no other constraints reject the phonetic form [kizlar] then this is the allowed form.

3. Related Websites:

 Steven Bird’s publication

HCRC Project: Computational Phonology: A Constraint-Based Approach

 Abstract - Prosodic Morphology and Constraint-based Phonology

 Abstract - Association as Dominance in a Constraint-Based Phonology

 Constraint-based phonology

Computational Phonology:A Constraint-Based Approach

Studies in Constraint-based Phonology

An aspect of dialectal variations in Korean phonology: a constraint-based analysis(PDF)

SFB-Project on Lexical Phonology and Constraint-based Phonology



Last Updated 06/11/08