Automatic Information Extraction in the Medical Domain by Cross-Lingual Projection

This research tackles the automatic annotation of texts written in a language L1 by exploiting resources and tools available for another language L2. Our approach involves the use of a parallel corpus (L1-L2) aligned at the level of sentences and words. To address the lack of annotated French corpus in the medical field, we focus on the French-English language pair to annotate French medical texts automatically. We focus in this article on Medical Entity Recognition (MER). We evaluate our MER method on the English corpus and the projection of the annotations on the French corpus. We also discuss the problem of scalability since we use a parallel corpus extracted from the Web and propose a statistical method to handle heterogeneous corpora.