Chinese Machine Translation In The Prepositional Phrase Disambiguation

Machine translation (MT) is an extremely important research field and testing ground of natural language processing (NLP). As an interdisciplinary research field, MT requires the knowledge of Linguistics, Computer Science, Mathmatics etc., which remains a challenging area. Many factors hinder the development of MT, such as lexical ambiguity, syntactic ambiguity, multiple grammatical structures etc., among which various kinds of ambiguity are the most difficult part.This thesis focuses on disambiguation of prepositional phrase (PP) attachment. After a brief introduction to the classification of major MT systems, this thesis reviews on the most important formal grammar in MT: PSG, and demonstrated how prepositional phrase attachment ambiguity is derived when parsing by this grammar. After that, the author mainly analyzes preference-based disambiguation approaches and proposes a tentative model for resolving PP ambiguity. Facing the major problem of lacking of rules, this thesis verifies Brill's transform-based error-driven algorithm and experiment for automatic retrieval of transformation rules. Finally, in connection with the sparse data problem in the processing of PP disambiguation and current statistical language models, the author introduces the backed-off N-Gram based algorithm and similarity-based smoothing.