Finite-State Descriptions of Various Levels of Linguistic Phenomena

Abstract

Finite State Automata (FSA) and their variants are natural tools adapted to the description of various linguistic phenomena which must be dealt with at various points in different types of automatic processing of texts written in Natural Language: - at the orthographical(morphological) level, one has to describe together each word and all its orthographical (morphological) variants. FSA represent such irregular families in a natural way; - during the lexical parsing of texts, one tries to construct the complete list of the words of the texts with minimal grammatical information. Compiling such an elementary list is a necessary step in any linguistic analysis. Because of various types of ambiguities which occur between simple words (e.g., voler: to steal, to fly, etc.), between free sequences of simple words and compound words (e.g., cordon bleu: blue rope, or good cook), between compounds (e.g, Pied noir: Frenchman born in Algeria, or American Indian), and because of cross constraints between ambiguities, the result obtained by a lexical parser cannot be represented by a mere list, and requires the use of a particular FSA (a Direct Acyclic Graph, DAG);