Global Learning of Noun Phrase Anaphoricity in Coreference Resolution via Label Propagation

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 978–986, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP Global Learning of Noun Phrase Anaphoricity in Coreference Resolu- tion via Label Propagation ZHOU GuoDong KONG Fang JiangSu Provincial Key Lab for Computer Information Processing Technology School of Computer Science and Technology Soochow University. Suzhou, China 215006 Email:{gdzhou,kongfang}@suda.edu.cn Abstract Knowledge of noun phrase anaphoricity might be profitably exploited in coreference resolu- tion to bypass the resolution of non-anaphoric noun phrases. However, it is surprising to no- tice that recent attempts to incorporate auto- matically acquired anaphoricity information into coreference resolution have been some- what disappointing. This paper employs a global learning method in determining the anaphoricity of noun phrases via a label propagation algorithm to improve learning- based coreference resolution. In particular, two kinds of kernels, i.e. the feature-based RBF kernel and the convolution tree kernel, are employed to compute the anaphoricity similarity between two noun phrases. Experi- ments on the ACE 2003 corpus demonstrate the effectiveness of our method in anaphoric- ity determination of noun phrases and its ap- plication in learning-based coreference resolu- tion. 1 Introduction Coreference resolution, the task of determining which noun phrases (NPs) in a text refer to the same real-world entity, has long been considered an important and difficult problem in natural language processing. Identifying the linguistic constraints on when two NPs can co-refer re- mains an active area of research in the commu- nity. One significant constraint on coreference, the anaphoricity constraint, specifies that a non- anaphoric NP cannot be coreferent with any of its preceding NPs in a given text. Therefore, it is useful to skip over these non-anaphoric NPs rather than attempt an unnecessary search for an antecedent for them, only to end up with inaccu- rate outcomes. Although many existing machine learning approaches to coreference resolution have performed reasonably well without explicit anaphoricity determination (e.g., Soon et al 2001; Ng and Cardie 2002b; Strube and Muller 2003; Yang et al 2003, 2008), anaphoricity determina- tion has been studied fairly extensively in the literature, given the potential usefulness of NP anaphoricity in coreference resolution. One common approach involves the design of heuris- tic rules to identify specific types of non- anaphoric NPs, such as pleonastic pronouns (e.g. Paice and Husk 1987; Lappin and Leass 1994; Kennedy and Boguraev 1996; Denber 1998) and existential definite descriptions (e.g., Vieira and Poesio 2000). More recently, the problem has been tackled using statistics-based (e.g., Bean and Riloff 1999; Bergsma et al 2008) and learn- ing-based (e.g. Evans 2001; Ng and Cardie 2002a; Ng 2004; Yang et al 2005; Denis and Balbridge 2007) methods. Although there is em- pirical evidence (e.g. Ng and Cardie 2002a, 2004) that coreference resolution might be fur- ther improved with proper anaphoricity informa- tion, its contribution is still somewhat disap- pointing and lacks systematic evaluation. This paper employs a label propagation (LP) algorithm for global learning of NP anaphoricity. Given the labeled data and the unlabeled data, the LP …

Content truncated to 500 words