Un modèle syllabique pour la reconnaissance de l'écriture

In this paper, we introduce a new syllabic model for handwriting recognition. We propose a supervised syllabification approach of the French language for building a vocabulary of syllables. A statistical n-gram language model of syllables is trained on a Wikipedia corpus. The handwriting recognition system, based on optical character HMM, performs a two pass decoding, integrating the proposed syllabic model. Evaluation is carried out on the RIMES dataset by analysing the performance for various coverage of the syllable model. We also compare the model with lexicon and character n-gram models. The proposed approach achieves interesting performance thanks to its capacity to cover a large amount of out of vocabulary words working with a limited amount of syllables combined with statistical n-gram of reasonable order. MOTS-CLÉS : Syllabe, Syllabation, Ecriture manuscrite, Modèle de langage, Reconnaissance