The Domain Dependence of Parsing

The Domain Dependence of Parsing Satoshi Sekine New York University Computer Science Department 715 Broadway, Room 709 New York, NY 10003, USA sekine@cs, nyu. edu http ://cs .nyu. edu/cs/proj ect s/prot eus/sekine Abstract A major concern in corpus based ap- proaches is that the applicability of the ac- quired knowledge may be limited by some feature of the corpus, in particular, the no- tion of text 'domain'. In order to examine the domain dependence of parsing, in this paper, we report 1) Comparison of struc- ture distributions across domains; 2) Ex- amples of domain specific structures; and 3) Parsing experiment using some domain dependent grammars. The observations using the Brown corpus demonstrate do- main dependence and idiosyncrasy of syn- tactic structure. The parsing results show that the best accuracy is obtained using the grammar acquired from the same do- main or the same class (fiction or non- fiction). We will also discuss the relation- ship between parsing accuracy and the size of training corpus. 1 Introduction A major concern in corpus based approaches is that the applicability of the acquired knowledge may be limited by some feature of the corpus. In particular, the notion of text 'domain' has been seen as a ma- jor constraint on the applicability of the knowledge. This is a crucial issue for most application systems, since most systems operate within a specific domain and we are generally limited in the corpora available in that domain. There has been considerable research in this area (Kittredge and Hirschman, 1983) (Grishman and Kittredge, 1986). For example, the domain depen- dence of lexical semantics is widely known. It is easy to observe that usage of the word 'bank' is dif- ferent between the 'economic document' domain and the 'geographic' domain. Also, there are surveys of domain dependencies concerning syntax or syntax- related features (Slocum, 1986)(niber, 1993)(Karl- gren, 1994). It is intuitively conceivable that there are syntactic differences between 'telegraphic mes- sages' and 'press report', or between 'weather fore- cast sentences' and 'romance and love story'. But, how about the difference between 'press report' and 'romance and love story'? Is there a general and sim- ple method to compare domains? More importantly, shall we prepare different knowledge for these two domain sets? In this paper, we describe two observations and an experiment which suggest an answer to the ques- tions. Among the several types of linguistic knowl- edge, we are interested in parsing, the essential com- ponent of many NLP systems, and hence domain de- pendencies of syntactic knowledge. The observations and an experiment are the following: • Comparison of structure distributions across domains • Examples of domain specific structures • Parsing experiment using some domain depen- dent grammars 2 Data and Tools The definition of domain will dominate the perfor- mance of our experiments, so it is very important to choose a proper corpus. However, for practical rea- sons (availability and time constraint), we decided to use an existing multi-domain corpus which has naturally acceptable domain definition. In order …

Content truncated to 500 words