| This dissertation is aimed at studying the problems of resolving ambiguities found in Mandarin Chinese (hereafter, Chinese), providing mechanisms for disambiguation, and applying them to a Chinese-English machine translation system called ALICE.;Different languages have different types of ambiguity because they have different structures. In contrast to English, Chinese is a language without inflection. As a result, these two languages have rather different types of ambiguity. We address mainly three interesting disambiguation problems: (1) serial verb constructions (SVCs), (2) serial nouns, (3) linguistic attributes of tense, number, and definiteness.;Different types of ambiguity need different mechanisms for disambiguation. We have classified ambiguity in Chinese into five types: syntactic, semantic, lexical, context, and morphological. We propose a disambiguation model, called Four-Step, for resolving syntactic ambiguity, especially in SVCs. Semantic ambiguity in serial nouns is resolved through a new mechanism called Semantic Functional Grammar, which combines Lexical-Functional Grammar with Semantic Grammar. Resolving context ambiguity makes use of a set of heuristics implemented in transfer rules. Lexical ambiguity is resolved by the traditional mechanisms of semantic restriction and referral to a concept lexicon.;In a tradeoff between the two major approaches for machine translation, interlingua and transfer, we propose a lexical-interlingua, syntactic-transfer approach, called Mini-Transfer (for minimal transfer). In light of the study of sublanguages, we propose an approach of Paraphrasing before Translation (PBT).;The mechanisms introduced above (except PBT) are implemented in the ALICE system, which has been tested with several sets of sentences (news headlines and linguistic data) and has produced acceptable translations. |
