Automatic identification of general and specific sentences by leveraging discourse annotations

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 605–613, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Automatic identification of general and specific sentences by leveraging discourse annotations Annie Louis University of Pennsylvania Philadelphia, PA 19104 lannie@seas.upenn.edu Ani Nenkova University of Pennsylvania Philadelphia, PA 19104 nenkova@seas.upenn.edu Abstract In this paper, we introduce the task of identifying general and specific sentences in news articles. Given the novelty of the task, we explore the feasibility of using ex- isting annotations of discourse relations as training data for a general/specific classi- fier. The classifier relies on several classes of features that capture lexical and syntac- tic information, as well as word specificity and polarity. We also validate our results on sentences that were directly judged by multiple annotators to be general or spe- cific. We analyze the annotator agree- ment on specificity judgements and study the strengths and robustness of features. We also provide a task-based evaluation of our classifier on general and specific summaries written by people. Here we show that the specificity levels predicted by our classifier correlates with the intu- itive judgement of specificity employed by people for creating these summaries. 1 Introduction Sentences in written text differ in how much spe- cific content they have. Consider the sentences in Table 1 from a news article about the Booker prize. The first one is specific and details the is- sues surrounding the books chosen for the award. The second sentence is general, it states that the prize is controversial but provides no details. In this work, we present the first analysis of proper- ties associated with general and specific sentences and introduce an approach to automatically iden- tify the two types. The distinction between general and specific sentences would be beneficial for several appli- cations. Prescriptive books on writing advise that sentences that make use of vague and ab- The novel, a story of Scottish low-life narrated largely in Glaswegian dialect, is unlikely to prove a popular choice with booksellers who have damned all six books shortlisted for the prize as boring, elitist and - worst of ...all - unsaleable. The Booker prize has, in its 26-year history, always provoked controversy. Table 1: General (in italics) and specific sentences stract words should be avoided or else immedi- ately followed by specific clarifications (Alred et al., 2003). So our classifier could be useful for the prediction of writing quality. Other applications include text generation systems which should con- trol the type of content produced and information extraction systems can use the distinction to ex- tract different types of information. Our definition of general/specific is based on the level of detail present in a sentence. This def- inition contrasts our work from some other re- cent studies around the idea of generic/specific distinctions in text. Reiter and Frank (2010) present an automatic approach to distinguish be- tween noun phrases which describe a class of in- dividuals (generic) versus those which refer to a specific individual(s). In …

Content truncated to 500 words