SemEval-2 Task 9: The Interpretation of Noun Compounds Using Paraphrasing Verbs and Prepositions

Proceedings of the 5th International Workshop on Semantic Evaluation, ACL 2010, pages 39–44, Uppsala, Sweden, 15-16 July 2010. c⃝2010 Association for Computational Linguistics SemEval-2010 Task 9: The Interpretation of Noun Compounds Using Paraphrasing Verbs and Prepositions Cristina Butnariu University College Dublin ioana.butnariu@ucd.ie Su Nam Kim University of Melbourne nkim@csse.unimelb.edu.au Preslav Nakov National University of Singapore nakov@comp.nus.edu.sg Diarmuid ´O S´eaghdha University of Cambridge do242@cam.ac.uk Stan Szpakowicz University of Ottawa Polish Academy of Sciences szpak@site.uottawa.ca Tony Veale University College Dublin tony.veale@ucd.ie Abstract Previous research has shown that the mean- ing of many noun-noun compounds N1 N2 can be approximated reasonably well by paraphrasing clauses of the form ‘N2 that . . . N1’, where ‘. . . ’ stands for a verb with or without a preposition. For exam- ple, malaria mosquito is a ‘mosquito that carries malaria’. Evaluating the quality of such paraphrases is the theme of Task 9 at SemEval-2010. This paper describes some background, the task definition, the process of data collection and the task results. We also venture a few general conclusions be- fore the participating teams present their systems at the SemEval-2010 workshop. There were 5 teams who submitted 7 sys- tems. 1 Introduction Noun compounds (NCs) are sequences of two or more nouns that act as a single noun,1 e.g., stem cell, stem cell research, stem cell research organi- zation, etc. Lapata and Lascarides (2003) observe that NCs pose syntactic and semantic challenges for three basic reasons: (1) the compounding process is extremely productive in English; (2) the seman- tic relation between the head and the modifier is implicit; (3) the interpretation can be influenced by contextual and pragmatic factors. Corpus studies have shown that while NCs are very common in English, their frequency distribution follows a Zip- fian or power-law distribution and the majority of NCs encountered will be rare types (Tanaka and Baldwin, 2003; Lapata and Lascarides, 2003; Bald- win and Tanaka, 2004; ´O S´eaghdha, 2008). As a consequence, Natural Language Processing (NLP) 1We follow the definition in (Downing, 1977). applications cannot afford either to ignore NCs or to assume that they can be handled by relying on a dictionary or other static resource. Trouble with lexical resources for NCs notwith- standing, NC semantics plays a central role in com- plex knowledge discovery and applications, includ- ing but not limited to Question Answering (QA), Machine Translation (MT), and Information Re- trieval (IR). For example, knowing the (implicit) semantic relation between the NC components can help rank and refine queries in QA and IR, or select promising translation pairs in MT (Nakov, 2008a). Thus, robust semantic interpretation of NCs should be of much help in broad-coverage semantic pro- cessing. Proposed approaches to modelling NC seman- tics have used semantic similarity (Nastase and Sz- pakowicz, 2003; Moldovan et al., 2004; Kim and Baldwin, 2005; Nastase and Szpakowicz, 2006; Girju, 2007; ´O S´eaghdha and Copestake, 2007) and paraphrasing (Vanderwende, 1994; Kim and Baldwin, 2006; Butnariu and Veale, 2008; Nakov and Hearst, 2008). The former body of work …

Content truncated to 500 words