Emmanuel Prochasson

Language Models for Handwritten Short Message Services

Handwriting is an alternative method for entering texts composing short message services. However, a whole new language features the texts which are produced. They include for instance abbreviations and other consonantal writing which sprung up for time saving and fashion. We have collected and processed a significant number of such handwriting SMS, and used various strategies to tackle this challenging area of handwriting recognition. We proposed to study more specifically three different phenomena: consonant skeleton, rebus, and phonetic writing. For each of them, we compare the rough results produced by a standard recognition system with those obtained when using a specific language model.

Bilingual Terminology Mining from Language for Special Purposes Comparable Corpora

In this paper we study the problem of compiling bilingual lexicon from language for special purposes (LSP) comparable corpora. We first define what would be the comparability for specialized comparable corpus and stress the distinction between expert and non-experts documents. We then turn to the contextual information method that concentrates on bilingual lexicon extraction and show its limits: the context vectors do not discriminate very much due to the small amount of data available and, the translation of the context vectors is more difficult due to the lack of specialized translations in the dictionary. For each problem, we propose a solution that relies on the LSP linguistic properties. For the first limit, we propose to strengthen the representativeness of the lexical contexts based on domain-specific vocabulary, called anchor points, notably the neoclassical compounds and the transliterations. For the second limit, we propose to use the parallel sentences present in the comparable corpus in order to build a specialized bilingual lexicon directly correlated to the specialized vocabulary of the comparable corpus. Our experiments illustrate that these two strategies are well-founded and show that candidate translations are of better quality.