Adaptation d'un système de reconnaissance d'entités nommées pour le français à l'anglais à moindre coût
Abstract
Cross-language portability of Named Entity Recognition systems requires linguistic expertise and needs human effort. Adapting symbolic systems suffers from the cost of developing new lexicons and updating grammar rules. Porting statistical systems on the other hand faces the problem of the high cost of annotation of new training corpus. This paper examines the cost of adapting a rule-based Named Entity Recognition system designed for well-formed text to another language. We present a low-cost method to adapt a French rule-based Named Entity Recognition system to English. We first solve the problem of lexicon adaptation to English by simply translating the French lexical resources. We then get to the task of grammar adaptation by slightly modifying the grammar rules. Experimental results are compared to a state-of-the-art English system.
Origin : Files produced by the author(s)
Loading...