Discovering Implicit Discourse Relations Through Brown Cluster Pair Representation and Coreference Patterns

Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, pages 645–654, Gothenburg, Sweden, April 26-30 2014. c⃝2014 Association for Computational Linguistics Discovering Implicit Discourse Relations Through Brown Cluster Pair Representation and Coreference Patterns Attapol T. Rutherford Department of Computer Science Brandeis University Waltham, MA 02453, USA tet@brandeis.edu Nianwen Xue Department of Computer Science Brandeis University Waltham, MA 02453, USA xuen@brandeis.edu Abstract Sentences form coherent relations in a discourse without discourse connectives more frequently than with connectives. Senses of these implicit discourse rela- tions that hold between a sentence pair, however, are challenging to infer. Here, we employ Brown cluster pairs to rep- resent discourse relation and incorporate coreference patterns to identify senses of implicit discourse relations in naturally occurring text. Our system improves the baseline performance by as much as 25%. Feature analyses suggest that Brown clus- ter pairs and coreference patterns can re- veal many key linguistic characteristics of each type of discourse relation. 1 Introduction Sentences must be pieced together logically in a discourse to form coherent text. Many discourse relations in the text are signaled explicitly through a closed set of discourse connectives. Simply disambiguating the meaning of discourse connec- tives can determine whether adjacent clauses are temporally or causally related (Pitler et al., 2008; Wellner et al., 2009). Discourse relations and their senses, however, can also be inferred by the reader even without discourse connectives. These im- plicit discourse relations in fact outnumber explicit discourse relations in naturally occurring text. In- ferring types or senses of implicit discourse re- lations remains a key challenge in automatic dis- course analysis. A discourse parser requires many subcompo- nents which form a long pipeline. The implicit discourse relation discovery has been shown to be the main performance bottleneck of an end-to-end parser (Lin et al., 2010). It is also central to many applications such as automatic summarization and question-answering systems. Existing systems, which make heavy use of word pairs, suffer from data sparsity problem as a word pair in the training data may not appear in the test data. A better representation of two adjacent sentences beyond word pairs could have a significant impact on predicting the sense of the discourse relation that holds between them. Data-driven theory-independent word classifica- tion such as Brown clustering should be able to provide a more compact word representation (Brown et al., 1992). Brown clustering algorithm induces a hierarchy of words in a large unanno- tated corpus based on word co-occurrences within the window. The induced hierarchy might give rise to features that we would otherwise miss. In this paper, we propose to use the cartesian product of Brown cluster assignment of the sentence pair as an alternative abstract word representation for building an implicit discourse relation classifier. Through word-level semantic commonalities revealed by Brown clusters and entity-level rela- tions revealed by coreference resolution, we might be able to paint a more complete picture of the discourse relation in question. Coreference resolu- tion unveils the patterns of …

Content truncated to 500 words