Peter Chew's doctoral dissertation

A Computational Phonology of Russian: detailed plan

1 Introduction:

This chapter describes the goal of the thesis ¾ to develop a broad-coverage phonology of Russian which explains the following in a coherent manner, as far as possible in line with established phonological theory:

The chapter reviews the relevant literature on phonological theory, Russian phonological classification, syllabification, morphological parsing and stress. It also introduces the computational techniques to be used, clarifies the distinction between linguistic frameworks and theories, and describes methods of evaluating linguistic theories.

2 Phonological classification (may be omitted from thesis):

The principles of phonological classification are explored in this chapter. In particular, the following issues are discussed:

The relevance of these concepts to other modules in the grammar is examined. It is shown how the process of constructing classification lattices for phonological systems can be automated and I discuss the results of constructing a lattice in this way for Halle’s Russian-phoneme feature matrix. In the light of this, the framework of Structured Specification will be evaluated for (a) its ability to shed light on the issues listed above and their relevance to other modules of the grammar, and (b) its helpfulness in determining the optimal classification for Russian. Also presented in this chapter is a program which converts from Russian orthography to phonemic and phonetic transcription.

3 Syllabification:

This chapter describes the computational implementation of the Sonority Sequencing Principle (SSP) as a phrase structure parser and evaluates its ability to explain consonant clusters observed in Russian, in particular by reference to word-initial onsets and word-final codas. The exceptions to the SSP are used to highlight possible weaknesses in the theory and I make proposals to account for some consonant clusters which are unexplained by the SSP. I also discuss the difficulty of determining the ‘correctness’ of a given syllable structure.

4 Morphological parsing:

In this chapter I describe how a morphological parser was implemented as a phrase structure grammar, just as the syllable parser was in Chapter 3. I state the necessity for a basic stock of morphemes (morphological objects) in order to do this. Using the list of morphemes from Oliverius (1976), I develop a grammar of Russian word-formation which is able to parse words into morphemes. I show how morpheme frequencies can be used to estimate the probability of a given parse being correct, and that this method improves the performance of the parser in cases where more than one parse is possible. I evaluate the morphological parser using Oliverius’s list of parsed words, and show how the information encoded in the parser can be extrapolated to parse words not listed by Oliverius. I also discuss the relevance to morphological parsing of issues such as the following:

I show how the morphological parser hinges in part on successful classification of morphological objects, and propose how this, like phonological classification, can be done in the framework of Structured Specification.

5 Stress assignment:

In this chapter I describe how stress assignment can be integrated with morphological parsing to produce a computational account of word-stress in both word-derivation and word-inflection in Russian. Since generative studies of Russian stress do not include comprehensive lists of morpheme underlying forms, I describe how such a list can be obtained using Halle’s and Melvold’s theories and ‘backwards phonology’. I then show how the list can be used by a stress assignment algorithm to prove that Halle’s theory of stress is reliable for inflected nouns. I discuss how Melvold’s stress assignment rules which deal with derived forms can then be incorporated into the algorithm to improve its performance on derived words. Finally, I compare Halle's and Melvold's theories with a competing theory of Russian stress, that of Zalizniak (1985).

6 Integration of phonological classification, syllabification, morphological parsing and stress assignment:

The final chapter shows how the modules of the grammar described so far can be integrated within a declarative phrase-structure framework, providing a coherent account of central aspects of Russian phonology. I show that the integration of these modules requires a detailed, composite method of representing Russian words, and give examples. I discuss avenues for further research which would complement the account of Russian phonology which I have given in this thesis.

Return to Peter Chew's Home Page

Contact Peter Chew