| The purpose of this thesis is to discuss some methodological issues of machine translation (MT) and to develop a practical model for interlingua machine translation systems.;The intermediate representations for interlingua MT systems vary from ones that are very much syntactically oriented to those that focus heavily on deep semantic analysis. In this thesis, it is argued that semantics should be the basis for intermediate representation for MT, but the design of such a model should be practical and task-oriented.;A practical interlingua MT system needs an intermediate semantic model which has not only a linguistic theory as its foundation but also the capabilities to deal with the urgent needs of MT tasks, i.e. the need to represent the 'meaning' of the language and the need to sufficiently handle ambiguities.;A prototype interlingua English-Chinese MT system, KrosLan, is proposed, which exploits the power of case grammar theory. In KrosLan, case grammar is not only used as the meaning representation framework, it also serves as the basis for verb sense classification, which facilitates property inheritance for verb predicates. This, together with the surface distribution information carried by the Case Matrix Model (Cook 1989), is explicitly used for semantic mapping and sense discrimination.;Generation in KrosLan is based on topic-comment model of Chinese. From the discussion we see that the intermediate representation defined for KrosLan is sufficient and proper for the generation of Chinese, a language which is generically unrelated to English, the source language, and is highly semantically focused. |
