1991 Volume 6 Issue 4 Pages 592-600
Traditional rule-based machine translation has a serious problem, called knowledge acquisition bottleneck. To overcome the problem, Nagao proposed Translation by Analogy or Example-Based Translation. The basic idea of it is very simple: translate a source sentence by imitating a translation example of a similar sentence. According to this idea, we propose MBT1, which is a method to select the best target words in verb-frame translation based on examples. MBT1 consists of three components: translation database, definition of metric, and translation process. Translation database is the collection of translation examples. A translation example is a pair of verb-frames. A verb-frame is one verb with several nouns as its arguments. We define a metric between a translation candidate (a pair of verb-frames) and a translation example in the database. In the translation process, MBT1 generates all candidates of translation. For each candidate, MBT1 retrieves the most similar translation example and get the score of the candidate based on the metric. MBT1 uses the score to evaluate the correctness of the candidate. We implemented MBT1 in English-Japanese translation.