Statistical Interpretation of Compound Nominalisations

Proceedings of the Australasian Language Technology Workshop 2005, pages 152–159, Sydney, Australia, December 2005. Statistical Interpretation of Compound Nominalisations Jeremy Nicholson† and Timothy Baldwin‡ †‡ Dept. of Computer Science and Software Engineering University of Melbourne Victoria 3010 Australia ‡ NICTA Victoria Lab University of Melbourne Victoria 3010 Australia {jeremymn,tim}@cs.mu.oz.au Abstract This paper presents a method for detecting compound nominalisations from open data, and providing a semantic intepretation. It uses a statistical model based on confidence intervals over frequencies extracted from a large, bal- anced corpus. Using three paraphrases of the given compound nominalisation, and interpre- tation preferences of its components, the algo- rithm achieves about 70% accuracy in classify- ing the semantic relationship as one of subject, and object, and 57% between subject, di- rect object, and prepositional object. 1 Introduction Compound nouns have been a thorn in the side of computational linguistics since its inception, as it is almost impossible to avoid the issue of compound noun interpretation in any lan- guage task with a semantic or lexical seman- tic dimension. For example, an information ex- traction task may need to predict the semantic divergences between the compound nouns news print (“cheap paper on which newspapers are printed”), thumb print (“impression of the pat- tern on a thumb”), and colour print (“printed matter in colour”). Interpreting these divergences has become yet another instance of disaccord between the em- piricists and rationalists of theoretical and com- putational linguistics. While the rationalists contend that compound nouns can be semanti- cally described by some small, hand-crafted set of relations, the empiricists point to discordant examples which defy such natural sets, and rely on data sets for the necessary description. The rationalists then call this approach biased and brittle. Needless to say, the argument is not one which will be resolved here. What we do hope to shed some light on is the applicability of an empiricist approach to the interpretation of an important subclass of compound nouns: com- pound nominalisations, or those compound nouns whose head derives from a verb. One ex- ample is pattern generation, where generation is derived from generate. These compounds tend to have a clearer semantic definition and are well-suited to techniques based on corpus statis- tics. We propose a method for taking English1 raw text input, detecting the compound nouns within, and applying a semantic interpretation for the compound nominalisations. Section 2 details some of the previous work performed in the task. Section 3 outlines the resources used in our study. Section 4 describes the method we applied, with an analysis of the results in Section 5, and a discussion in Section 6. 2 Background The first notable work on interpreting com- pound nouns focussed on the development of discrete semantic classes with which to classify all compound nouns: Levi (1978) proposed a nine-way classification, Warren (1978) identi- fied a basic set of twelve paraphrases, Leonard (1984) developed an eight-way typology, and Finin (1980) put forth a much larger number of possible roles. This early research established …

Content truncated to 500 words