Constructing A Chinese Semantic Auto Parser Based On An ILP Algorithm Combining Top-down And Bottom-up Methods

There are mainly two approaches to Natural Language Understanding. The first one is using some language rules to analyses the relationship in all components of natural language text. This approach is called "rationalistic approach". The second one is based on data analysis and is called "empirical approach". This approach is based on a huge corpus, using probabilistic methods to get the concomitance probability of every language phenomena. These methods identify the relations by the value of the concomitance probability when they analyses the corpus. The methods based on language rules are essentially deduction reasoning. Its advantage is the qualitative description according context. It can utilize the fruit of modern linguistics. The disadvantage is that they can't deal with the uncertain events, while there are some restriction on the rule's consistency and adaptability. The approach based on statistics is an empirical one. The advantage is that it gets all its knowledge by analyzing the huge corpus. It can achieve better consistency and wide covering. The method based on statistics is an undetermined quantitative analysis. Due to basing on probability, the events with lowest probability are hidden. This paper considers a new empirical approach. Adopting structural data description, ILP methods are used to solve the problems of parser acquiring.The structure analyzing is the base of natural language comprehension. It could be divided into two levels: the first one is to study the semantic representations of all components in a natural sentence; the second one is to establish a map between these representations and natural language sentences. This is one emphases of this paper. In this paper, the system ICASP that builds an automatically semantic parser is presented. The parsing method of a parser constructed by the system ICASP is illustrated by a case-role parsing example in this paper. The basic idea of case-role semantic parsing is: the central verb combined with other components in the sentence forms the "semantic case-role" frame. This frame is used to describe the deep semantic relations within every component of a natural language sentence, expressing the agentives, patients and instruments and other semantic cases in the sentence.The word "parse" is often used to express the action that translating a natural language sentence into a layer structure of the sentence syntactic relations. According to some context-free grammar, a natural language sentence may be parsed into a layer structure with some annotations of components in the sentence. But the parsing on the syntactic lever can be only very small part of comprehension of the sentence. In fact, natural language comprehension should consider the semantic oriented problems. At least, parsing a sentence should point out some important relations in the sentence, such as who have done something to someone, etc. Parsing in semantic level is called semantic parsing. In this paper, a semantic parser constructing system ICASP is designed and implemented. ICASP is based on a new ILP algorithm ICCR. The new...