抄録
To facilitate retrieving patent information across languages, we are developing a multi-1ingual patent retrieval system, where user queries are translated into the target language by way of a dictionary. In this paper, aiming to enhance the translation dictionary, we propose a method to automatically extract translations from Japan-US patent families consisting of Japanese/English comparable texts. Our method computes the association score for each combination of Japanese/English words in patent families, and selects those with greater scores as translations. We also show the effectiveness of our method through experiments.
