Information acquisition with message-oriented machine translation.

摘要

Three models of machine translation co-exist today: Direct (word-for-word), Transfer (sentence and grammar) and Interlingua (contextual).;This research proposes a message-oriented (Interlingua) translation. It is based on the idea that grammatical elements are not always necessary for the translation process. It is possible to neutralize text by removing grammatical details, setting them aside for possible future use, and using dictionary look-up to translate concepts, messages and their relations into the target language.;To measure the performance of the suggested message translation, a research design was devised to pioneer an objective and quantitative method for assessing translation qualities. English text was translated using simulations of the Direct approach and the proposed message-oriented approach. Questionnaires were prepared for three sample documents in the original English, and two Chinese translations. Questions were designed to measure the subjects' comprehension of document content. Using the English version as the benchmark, results of responses were compared to evaluate translation qualities.;The difference between the two Chinese groups was not statistically significant although the message translation was clearly favored over the word-for-word translation. The major factor uncovered was that the preexisting knowledge of the source language (e.g. English) by native Chinese subjects remedied the deficiency of less comprehensible word-for-word translation in the target language (e.g. Chinese). Detailed item analyses of individual questions found that message translation was better when inferences of word meanings and/or reorderings of words were required to answer a question correctly. Correspondingly, when exact match of keywords was sufficient without inference to choose the right answer, word-for-word translation would perform nicely without understanding the document content.