Conjugation system, statistical morphosyntactic recognition & lemmatization of the modern Greek standard verbal system

Dans cette thèse nous présentons les résultats ainsi que la méthodologie adoptée pour la création d'un système d'analyse morphosyntaxique automatique et de lemmatisation sans dictionnaire des formes verbales monolexicales du grec moderne standard. Avec le modèle rétrograde MOSAIC (Koktova 1985) sur le chech comme point de départ, ainsi que d'autres modèles similaires sur le français (Caradec & Saada 1982) et le grec moderne (Lexifanis, Kotsanis & Maistros 1985), notre recherche a couvert 8.485 lexèmes verbaux grecs, en prenant les données des dictionnaires les plus récents (Kyriacopoulou 1990, Iordanidou 1992, Kriaras 1995, Babiniotis 1998, Institut d'Études Néohelléniques 1998). Il a ainsi été créé: un nouveau système de conjugaison de 385 modèles qui sert à la génération automatique de tous les morphèmes lexicaux/radicaux ainsi que de toutes les formes flexionnelles monolexicales une base de données des séquences graphémiques finales qui permet l'attribution automatique de modèle de conjugaison à n'importe quel lemme verbal une base de données de 151.527 séquences graphémiques finales, statistiquement établie et manuellement perfectionnée, qui peut s'employer pour la reconnaissance automatique de n'importe quelle forme verbale monolexicale et un système des règles morphophonologiques rétrogrades utilisées pour la lemmatisation linéaire des formes flexionnelles, qui fonctionne sur la base du nouveau système de conjugaison de 385 modèles.In this dissertation we present the final results of our 10-year research on the Modern Greek verbal system. The objective of the research has been twofold: i) the development of a statistical database containing word-final grapheme sequences, which, on the basis of Koktova's (1985) retrograde analysis model MOSAIC, allow for the automatic morphosyntactic recognition (tagging) of all monolexical verbal forms of the language without any access to relevant electronic lexicons and ii) the development of a verb lemmatization morphophonological rule system, both providing various applications in all major areas of Text Processing as well as Teaching of Modern Greek Standard. Within this framework, 24Mb of verbal linguistic data have been collected, generated and classified automatically, and manually checked and enriched. These have been submitted to the University in the form of an appendix. Only representative extracts appear in the dissertation. More specifically, they consist of: a) a file of 8,485 Modern Greek verbal lemmas, developed in accordance with the evidence provided by the most recent dictionaries of the language (Dictionary of the Modern Greek Dhemotiki, Kriaras 1995, Greek Dictionary,Tegopoulos-Fytrakis 1993, Abridged Dictionary of Modern Greek, Pagoulatou Publ. 1991, Dictionary of Modern Greek, Babiniotis 1998 and Dictionary of Modern Greek Koine, Triantafyllidis Inst., Aristotle University of Thessaloniki 1998) b) a new conjugation system of 385 paradigmatic models, which allows for the automatic generation of all verbal stems and monolexical forms c) a file of 1st person singular word-final grapheme sequences, which allows for the automatic attribution of paradigmatic model codes to any verbal lemmas of the language d) a file of 27,383 verbal stems characterized solely on the basis of their conjugation model and their permissible suffix set e) 103 files of 519,694 automatically generated and classified verbal forms f) 17 files of 151,527 word-final grapheme sequences, which declare the conjugation model, morphosyntactic content, absolute frequency and lemmatization code of verbal forms g) a linear lemmatization morphophonological rule system, which functions on the basis of the newly developed 385 conjugation model system. More analytically, in the 1st chapter we discuss the role which morphological lexicons and statistical approaches play in the automatic morphosyntactic recognition of word tokens (tagging). In the 2nd chapter we discuss the Modern Greek verb system, including a brief description of the vocabulary of the language (Katharevousa-Dhemotiki): the morphosyntactic categories marked, the conjugation system as presented in various grammatical descriptions over the last 40 years, the verbal stems and inflectional affixes involved, the stress pattern, the external/internal augment and reduplication occurrences, as well as the 2 most recently developed conjugation systems (Kyriacopoulou 1990 and Iordanidou 1992), in an effort to account for the need of developing a new conjugation system. In the 3rd chapter we describe the methodology employed for the collection and processing of data, whereas in the 4th chapter we present extensive extracts from the 10 databases developed all together. A pilot application of the proposed language tool is available on the Internet and can be found on the site of the Language and Education Technology Laboratory of the University of Athens Informatics/Telecommunications Department