Example Based Machine Translation (EBMT)

Similar to statistical method, example-based machine translation method is a corpus-based approach, the basic idea was proposed by the famous Japanese MT expert Makoto Nagao. He studied basic model of foreign languages beginners, and found the foreign language beginners always remember the most basic English sentences and the corresponding Japanese sentences, and then do substitution drill. Referring to this learning process, he proposed example-based of MT theory, which is without a deep analysis, just using the existing empirical knowledge to translate by analogy. The translation process is firstly decomposing source language into sentences then breaking down into phrases fragment, and by analogy, translating the phrase fragments into the target language phrases, and at last combine these phrases to long sentences. To instance method systems, its main source of knowledge is bilingual case library, dictionary, grammar rule base and the like are not necessary. The core problem is to obtain bilingual example database by minimizing statistics.

Example-based MT has significant effect on the translation of the same or similar text, and as the database size increases, its role becomes increasingly significant. For instance the library has the text, direct access to high-quality translation results. And the existing texts in the example database can obtain the high quality translation results. By reasoning and comparing the texts similar to the examples existing in the example database, revise the translation results to construct a similar translation.

In the initial implementation of this approach, it was celebrated by many. However, after a period of time, a problem appeared. Since the method required a large corpus as support, the actual demand in language was huge. Limited by the corpus size, it is difficult for example-based MT to achieve a higher matching rate. Often only limited to relatively narrow or professional fields, the translation effect can reach to using requirements. Thus far, very few MT systems adopt the pure case-based approach, usually regarding the instance-based MT method as one of multi-translation engines, to improve translation accuracy.