Typed Dependency Relations for Syntactic Analysis of Thai Sentences

Typed Dependency Relations for Syntactic Analysis of Thai Sentences* Siripong Potisuk Department of Electrical and Computer Engineering The Citadel, the Military College of South Carolina 171 Moultrie Street Charleston, South Carolina 29409 USA siripong.potisuk@citadel.edu Abstract. This paper describes a preliminary effort in identifying many different types of relations among words in Thai sentences based on dependency grammar. The relation is represented as a triple containing the pair of words and their relation. So far, the current representation contains 35 grammatical relations. The dependencies are all binary relations. That is, a grammatical relation holds between a governor and a dependent. The analysis makes use of the Thai “Orchid” corpus part-of-speech tags and the Stanford typed dependencies definitions. Keywords: Dependency grammar, Thai syntactic analysis. 1 Introduction Language modeling is one of the many important aspects in natural (both written and spoken) language processing by computer. For example, in a spoken language understanding system, a good language model not only improves the accuracy of low-level acoustic models of speech, but also reduces task perplexity (the average number of choices at any decision point) by making better use of high-level knowledge sources including prosodic, syntactic, semantic, and pragmatic knowledge sources. A language model often consists of a grammar written using some formalism which is applied to a sentence by utilizing some sort of parsing algorithm. One popular example is a set of context-free grammar (CFG) production rules, which is based on a phrase-structure representation of syntax, can be used to parse sentences in the language defined by that grammar. Another example is constraint dependency grammar (CDG) proposed by Maruyama. CDG parsers rule out ungrammatical sentences by propagating constraints. Constraints are developed based on a dependency-based representation of syntax. Some parsing algorithms combine phrase-structure and dependency grammars. In the next subsection, we present a contrastive description of the two major approaches to representing syntax of natural languages in order to motivate our choice of dependency grammar for parsing Thai sentences. 1.1 Dependency vs. Phrase-structure Grammar In the theory of the syntax of natural languages, there are currently two major methods of representing the syntactic structure of natural languages: dependency grammar and phrase- structure (constituency) grammar. There has been no third approach developed although combinations of the two major methods above have been used, e.g., lexical-functional grammar, case grammar, relational grammar, word grammar, etc.. ______________ * The author would like to thank the Citadel Foundation for its support in the form of a presentation grant. Copyright 2010 by Siripong Potisuk PACLIC 24 Proceedings 511 As a formal syntactic representation, dependencies have been studied and explored for centuries by traditional syntacticians of European, Classical, and Slavic languages. Lucien Tesnière was credited as the first syntacticians who formalized and laid the groundwork for subsequent investigations of the theory. Unfortunately, the dependency formalism has not gained great popularity among syntacticians who favor constituency grammar. Constituency or phrase-structure grammar was formulated in North America in the early 1930’s by Leonard Bloomfield, primarily for describing the syntax of English. The theory …

Content truncated to 500 words