友清睦子,小倉健太郎,江原暉将
日英機械翻訳システムにおける生成文の評価
-「ダ」型表現を中心にして-(付録)
Abstract:The chief purpose of this work is to establish the criteria for the evaluation of output
phrases in MT systems.
This evaluation is restricted to grammatical side of output phrases of computer-generated
natural language.
We extracted phrasal sentence fragments from the Japanese sentences, for example "TO-IU-KOTO-DESHO-U-KA" and assigned them multiple translations, for example "①That
Means~?" "②Does that mean~?" "③Do you say~?" etc.
The Japanese phrases were chosen to reflect the principal grammatical problems to be
realized in English output.
In cases where the Japanese expression "TO-IU-KOTO-DESHO-U-KA" is translated into
one of ①②③ ,we judged that this expression was correctly translated.
In our next step, we will be programming this evaluation method in order to
automatically estimate the MT system quality.