Investigation And Applications Of Conversion Between Syntactic Trees In Natural Language Processing

There exist various kinds of tree structures, such as phrase trees, dependency trees, prosodic hierarchies and so on in natural language processing. These tree structures are defined by different grammar frameworks, such as Phrase Structure Grammar (PS-G)[1], Lexicon Function Grammar (LFG)[2,3], Tree Adjoining Grammar (TAG)[4], Combinatory Categorial Grammar (CCG)[5], etc. Different grammar frameworks lead to different tree banks. CCGBank [6] and Penn Treebank [7] are the most wildly used tree banks in natural language processing.Different parsers are developed based on different tree banks. Then how to com-pare the performance of the parsers is a widely concerned research topic. Such com-parison is always based on the conversion between different grammar tree structures. Moreover, the mappings between different grammar tree structures can also be used to solve certain problems creatively, such as the break index prediction task elaborated in this paper.In all, this paper aims at research on the conversion between grammar tree struc-tures defined by different grammar frameworks. On one hand, the conversion driven by linguistics is investigated. The conversion of CCG derivations to PTB trees is ex-plained and the performance of the system is shown to be better compared with those in previous works. On the other hand, the conversion from phrase tree structures to prosodic hierarchies is studied as a example of the conversion driven by applications. Experiments show its potential value in use.