Predicting the Presence of Discourse Connectives

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 914–923, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Predicting the Presence of Discourse Connectives Gary Patterson and Andrew Kehler Department of Linguistics UC San Diego 9500 Gilman Drive #0108 La Jolla, CA 92093 {gpatterson,akehler}@ucsd.edu Abstract We present a classification model that predicts the presence or omission of a lexical connec- tive between two clauses, based upon linguis- tic features of the clauses and the type of dis- course relation holding between them. The model is trained on a set of high frequency relations extracted from the Penn Discourse Treebank and achieves an accuracy of 86.6%. Analysis of the results reveals that the most in- formative features relate to the discourse de- pendencies between sequences of coherence relations in the text. We also present results of an experiment that provides insight into the nature and difficulty of the task. 1 Introduction A central goal of natural language generation and summarization systems is to produce interpretable, coherent text that rivals material a human would pro- duce. Doing so requires that systems not only have the ability to generate clauses that are grammatical and easy for people to process, but also the ability to employ the appropriate discourse structuring de- vices needed to yield fluid transitions between these clauses. This is a tricky issue in that it requires that a balance be achieved between the opposing goals of communicative expressiveness and economy. On the one hand, insufficient cueing of inter-clausal re- lationships can lead to a discourse that is at best dif- ficult to process, and at worst misunderstood. On the other hand, too much explicit marking can result in a clunky and even redundant sounding discourse. Here we consider the question of when to ex- plicitly mark the COHERENCE RELATIONS in dis- course, that is, the inter-clausal relationships that the language producer intends the interpreter to in- fer between the meanings of clauses (Hobbs, 1979; Mann and Thompson, 1988; Kehler, 2002; Asher and Lascarides, 2003). Consider, for example, the EXPLANATION coherence relation that holds in (1), in which the second clause provides a cause or rea- son for the eventuality described in the first: (1) a. Max will visit Australia this summer be- cause his father is turning 65. b. Max will visit Australia this summer. His father is turning 65. As example (1) shows, coherence relations can be marked explicitly—by using lexical connectives such as coordinating or subordinating conjunctions (e.g., because in 1a) or certain types of prepositional or adverbial phrases—or left implicit as in (1b). Ei- ther way, establishing the relation itself requires the reader to go through a complex inferential process necessitating that a variety of assumptions be made, typically supported by context and/or world knowl- edge, that are not explicitly asserted by the actual linguistic material. In (1), for instance, such infer- ences would include that Max intends to see his fa- ther when he travels to Australia, that his father …

Content truncated to 500 words