Centre d'études et de recherches en informatique linguistique,
Université de Marne-la-Vallée The course aims at: We specify the mathematical effect of algorithms in terms of automata
theory (properties taken into account: deterministic, minimal,
reduced, complete, trim, acyclic...; operations: intersection,
complementation, factor decomposition...). We describe algorithmic
techniques for implementing automata of a very large size (e.g.
transition packing for lexical tagging). As to lexical tagging, we consider the zero-silence approach with
dictionary lookup and subtraction of combinations that can be
ruled out on the basis of local grammars (cf. Silberztein, A new
approach to tagging, in Applied Computer Translation 1(4)). In
this approach, we give prominence to the maintainability and readability
of the disambiguation grammars. The other NLP problem is the segmentation
of phonetic strings into words or syllables on the basis of a
phonetic lexicon or of a specification of syllabification. No particular knowledge is required in automata theory, probability
theory, graph theory, object-oriented programming, acoustics or
syntax. The structure of the course assumes that attendants dispose
of some extra time every day to assimilate notions through personal
practice.
RECOGNITION, SELECTION AND TRANSLATION ALGORITHMS WITH FINITE AUTOMATA FOR LEXICAL TAGGING AND SEGMENTATION OF PHONETIC STRINGS
eric.laporte@univ-reims.fr
No specific recommendation