A japaneseenglish patent parallel corpus

We describe a Japanese-English patent parallel corpus created from the Japanese and US patent data provided for the NTCIR-6 patent retrieval task. The corpus contains about 2 million sentence pairs that were aligned automatically. This is the largest Japanese-English parallel corpus, which will be available to the public after the 7th NTCIR workshop meeting. We estimated that about 97 % of the sentence pairs were correct alignments and about 90 % of the alignments were adequate translations whose English sentences reflected almost perfectly the contents of the corresponding Japanese sentences. 1