Semitic Morphological Analysis and Generation Using Finite State Transducers with Feature Structures

Proceedings of the 12th Conference of the European Chapter of the ACL, pages 309–317, Athens, Greece, 30 March – 3 April 2009. c⃝2009 Association for Computational Linguistics Semitic Morphological Analysis and Generation Using Finite State Transducers with Feature Structures Michael Gasser Indiana University, School of Informatics Bloomington, Indiana, USA gasser@indiana.edu Abstract This paper presents an application of finite state transducers weighted with feature structure descriptions, following Amtrup (2003), to the morphology of the Semitic language Tigrinya. It is shown that feature-structure weights provide an effi- cient way of handling the templatic mor- phology that characterizes Semitic verb stems as well as the long-distance de- pendencies characterizing the complex Tigrinya verb morphotactics. A relatively complete computational implementation of Tigrinya verb morphology is described. 1 Introduction 1.1 Finite state morphology Morphological analysis is the segmentation of words into their component morphemes and the assignment of grammatical morphemes to gram- matical categories and lexical morphemes to lex- emes. For example, the English noun parties could be analyzed as party+PLURAL. Morpho- logical generation is the reverse process. Both processes relate a surface level to a lexical level. The relationship between these levels has con- cerned many phonologists and morphologists over the years, and traditional descriptions, since the pioneering work of Chomsky and Halle (1968), have characterized it in terms of a series of ordered content-sensitive rewrite rules, which apply in the generation, but not the analysis, direction. Within computational morphology, a very sig- nificant advance came with the demonstration that phonological rules could be implemented as fi- nite state transducers (Johnson, 1972; Kaplan and Kay, 1994) (FSTs) and that the rule ordering could be dispensed with using FSTs that relate the surface and lexical levels directly (Koskenniemi, 1983). Because of the invertibility of FSTs, “two- level” phonology and morphology permitted the creation of systems of FSTs that implemented both analysis (surface input, lexical output) and gener- ation (lexical input, surface output). In addition to inversion, FSTs are closed un- der composition. A second important advance in computational morphology was the recognition by Karttunen et al. (1992) that a cascade of composed FSTs could implement the two-level model. This made possible quite complex finite state systems, including ordered alternation rules representing context-sensitive variation in the phonological or orthographic shape of morphemes, the morpho- tactics characterizing the possible sequences of morphemes (in canonical form) for a given word class, and one or more sublexicons. For example, to handle written English nouns, we could create a cascade of FSTs covering the rules that insert an e in words like bushes and parties and relate lexical y to surface i in words like buggies and parties and an FST that represents the possible sequences of morphemes in English nouns, including all of the noun stems in the English lexicon. The key fea- ture of such systems is that, even though the FSTs making up the cascade must be composed in a par- ticular order, the result of composition is a single FST relating surface and lexical levels directly, …

Content truncated to 500 words