Studies On Automatic Recognition Of Modern Chinese Conjunction Usages And Application

Conjunction is a kind of functional words. In modern Chinese, it has hard grammar presentation tasks and is very important for Chinese grammar analysis and semantic understanding. The same conjunction may have different meanings and different usages in different context. Therefore, conjunction usages should be investigated carefully. Conjunction usage could be automatically recognized through summarying rules artificially or learning rules by machine and describing the rules formally, which is helpful for machine analysis and automatic understanding of Chinese text.Automatic recognition of modern Chinese conjunction usages is one of the important studies on the modern Chinese functional words knowledge base oriented to Natural Language Processing. According to the thought of building "Trinity" knowledge base of modern Chinese functional words put forward by Yu Shiwen, this paper perfects modern Chinese conjunction knowledge base including modern Chinese conjunction usage dictionary, conjunction usage rule base and conjunction usage corpus. The paper researches automatic recognition of modern Chinese conjunction usages based on rules and statistics respectively according to conjunction knowledge base. The rule method is simple, easy and has explicit presentation. But it can’t automatically gain knowledge through methods of machine learning. The statistical method can get language knowledge from training data automatically or semi-automatically. However, it has bad recognition effects for conjunctions with single usage or sparse usage distribution. According to upsides and downsides of rule method and statistics method, the paper preliminarily tries five kinds of a hybrid way of rules and statistics to recognize conjunction usages by combining usage distribution rate, rule recognition precision and statistical recognition precision. Experimental results show that recognition effects of the combination method are better than any single method.On the basis of conjunction usages recognition, the paper studies conjunction structure phrase recognition which is one applications of conjunction usage, which would provide better preprocessing knowledge so that it could improve the quality of machine translation. Initially, the author artificially marks conjunction structure phrases in the corpus which is tagged conjunction usages, summarizes regularities, constructs identification rules of conjunction structure phrases and realizes automatic recognition of conjunction structure phrase based on rules. Subsequently, the paper adopts conjunction usages as feature of statistical model to research automatic recognition of conjunction structure phrase based on statistics through analyzing the shortage of rule method. Experimental results show that recognition effects of statistical method are better than rule method, and the statistics recognition results joined usage feature are higher than recognition results without usage feature. F measure of recoghniton rises1.26%than before adding usage feature and33.3%than rule method.