THUTR: A Translation Retrieval System
Chunyang Liu, Qi Liu, Yang Liu, Maosong Sun · International Conference on Computational Linguistics · 2012
We introduce a translation retrieval system THUTR, which casts translation as a retrieval problem. Translation retrieval aims at retrieving a list of target-language translation candidates that may be helpful to human translators in translating a given source-language input. While conventional translation retrieval methods mainly rely on parallel corpus that is difficult and expensive to collect, we propose to retrieve translation candidates directly from target-language documents. Given a source-language query, we first translate it into target-language queries and then retrieve translation candidates from target language documents. Experiments on Chinese-English data show that the proposed translation retrieval system achieves 95.32% and 92.00% in terms of P@10 at sentence level and phrase level tasks, respectively. Our system also outperforms a retrieval system that uses parallel corpus significantly.