Finite-State Transducers with Multivalued Mappings for Processing of Rich Inflectional Languages

Abstract

This paper proposes a processing for rich inflectional languages such as Russian and Kazakh, based on finite state transducers with multivalued mappings. We propose to simplify grammar of inflectional languages and use multivalued mappings. An advantage of finite state transducers with multivalued mappings proposed in this paper is that it automatically generates possible alternatives of words’ grammatical characteristics, while in existing rule-based technologies alternatives are written by hand. Ambiguity of grammatical characteristics is solved by comparing alternative grammatical characteristics between adjacent words in the source sentence and matching grammatical characteristics are selected. Here an advantage of proposed method should be noted, it does not require explicit description of matching agreements for grammatical characteristics of adjacent words in a sentence, as it is done in existing rule-based methods.