|Title:||Contrast Analysis in Phonological Learning|
|Abstract:||This paper presents a specific proposal for the use of observations about surface contrasts in the learning of underlying forms for morphemes, and investigates the strengths and weaknesses of that proposal. The main formal result presented here is that, if a set of assumptions are satisfied by the grammatical system, then whenever two distinct morphemes contrast on the surface in a particular environment, at least one of the underlying features on which the two differ must be realized faithfully in each of the morphemes in that environment. To put it another way, at least one of the surface features distinguishing the two surface realizations must faithfully reflect a distinction between the underlying forms of the two morphemes. This property is called the Faithful Contrastive Feature property.
The significance of this property is that at least one contrast-causing feature must be overt, in the sense that its underlying values are faithfully presented in the surface forms, where the learner can observe them. This result forms the basis for an algorithm, Contrast Analysis, which starts with a number of features in underlying forms unset (not yet specified), and specifies them for a morpheme only when that morpheme contrasts with another morpheme, and the contrast can only be attributed to a single unset feature. Because the contrast-causing feature must be faithfully realized, the learner can set that feature in the underlying form of each morpheme to match its surface realization for that morpheme.
The contribution of Contrast Analysis will emerge in its role in a larger theory of language learning. We illustrate Contrast Analysis by using it as part of a procedure for initial lexicon construction, designed to construct a working lexicon for use by a subsequent procedure for jointly learning the lexicon and the phonological mapping (a constraint ranking). Initial lexicon construction depends solely on morphologically analyzed surface forms, and does not make reference to phonological mappings. Initial lexicon construction cannot determine all feature values for underlying forms in all cases. In fact, it is likely that in complex cases it will be unable to set a significant number of values for many underlying forms. If common beliefs about the close interrelation between underlying forms and grammatical mappings are correct, it is not possible in general to determine all underlying forms for the morphemes of a language independent of consideration of the grammatical mapping for the language. However, using Contrast Analysis in initial lexicon construction does offer the possibility of setting the values of some features in some morphemes early on. Those values, in turn, could do much to constrain the search involved in subsequent processes of inferring both the mapping and the rest of the underlying forms. Further, Contrast Analysis is defined such that it could be invoked at several points during learning, opening the possibility that it could be invoked during the joint learning of the lexicon and the phonological mapping.