Combining Linguistic Knowledge and Statistical Learning in French Part-of-Speech Tagging

This paper presents a new part-ofspeech tagger that takes into account both linguistic knowledge and statistical learning. Its novelty relies in several aspects: (a) a fully modular architecture that allows the user flexible use of each independent module, (b) an expanded tagset that gives the user the flexibility to use it directly or use any individually defined tag subset, (c) the exportability of the system to other languages, and (d) the use of a mixed linguistic and statistical approach. Preliminary results are provided, as well as directions for future use of the model. 1 Introduction Tagging part-of-speech on unrestricted texts for a language other than English raises a certain number of questions: if the language has a rich morphology, one needs to analyze the words in the sentence since the inflected forms are not in the dictionary; an electronic dictionary is necessary to write a morphological analyzer. At the morphology level: what should be the level of morphological refi..