A morphological analyser for machine translation based on finite-state transducers

A MORPHOLOGICAL ANALYSER FOR MACHINE TRANSLATION BASED ON FINITE-STATE TRANSDUCERS Alberto Sanchis �, David Pic´o �, Joan Miquel del Val �, Ferran Fabregat �, Jes´us Tom´as �, Mois´es Pastor �, Francisco Casacuberta �, Enrique Vidal � �Institut Tecnol`ogic d’Inform`atica Departament de Sistemes Inform`atics i Computaci´o Universitat Polit`ecnica de Val`encia 46071 Val`encia, SPAIN �asanchis,dpico,moises,fcn,evidal�@iti.upv.es � Servei de Normalitzaci´o Ling¨u´ıstica Universitat de Val`encia - Estudi General 46010 Val`encia, SPAIN �Joan.M.Val,fabregat�@uv.es � Departament de Comunicacions Escola Polit`ecnica Superior de Gandia 46730 Gandia, SPAIN jtomas@dcom.upv.es Abstract A finite-state, rule-based morphological analyser is presented here, within the framework of machine translation system TAVAL. This morphological analyser introduces specific features which are particularly useful for translation, such as the detection and morphological tagging of word groups that act as a single lexical unit for translation purposes. The case where words in one such group are not strictly contiguous is also covered. A brief description of the Spanish-to-Catalan and Catalan-to-Spanish translation system TAVAL is given in the paper. Keywords: Morphological Analysis, Finite-State Transducers, Word Graphs Introduction Rule-based techniques are the usual approach for building general machine translation systems. However, example- based approaches have experienced an increasing interest in different problems of machine translation during the last decade. These approaches have shown competitive per- formance in dealing with translation tasks in a restricted- domain language and have also been useful in particular subproblems that arise in rule-based translation, such as POS tagging (Abney, 1997; Dagan et al., 1997) or finding non-strictly linguistic relations among words or phrases in specific tasks (Brown et al., 1990; Brown et al., 1993; Al- Onaizan et al., 1999). Finite-state machines have been successfully used for the implementation of both rule-based and example-based machine translation systems and tools for natural language processing (Rochar & Schabes, 1995; Oflazer, 1996; Mohri, 1997; Mohri et al., 2000). Finite-state techniques are very appreciated for their simplicity and high time performance. TAVAL is a Spanish-to-Catalan and Catalan-to-Spanish translation system that aims to combine adequately both rule-based and example-based techniques on a finite-state framework. TAVAL takes advantage of the high degree of sequentiality existing between Spanish and Catalan that makes possible to avoid the analysis of certain complex word alignments and changes in syntactic structures. Up to now, there are a few machine translation systems that deal with these two languages. The newspaper El Peri ´odico produces a Catalan version of news that are originally writ- ten in Spanish by using a memory-based translator. The SALT2 system is a rule-based computer-assisted transla- tion system. INCYTA is a rule-based system that needs post-process corrections. InterNostrum is a rule-based sys- tem developed by a team of the Alacant University that uses finite-state technology (Canals et al., 2000). The TAVAL system will be tested for the translation of certain types of bilingual official publications by the Valencian government. The TAVAL translator performs a partial analysis of the segment of text to be translated using a knowledge- based dictionary. An efficient finite-state representation of the dictionary was chosen following some guidelines from the grammatical …

Content truncated to 500 words