A Coherence Model Based on Syntactic Patterns

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 1157–1168, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics A coherence model based on syntactic patterns Annie Louis University of Pennsylvania Philadelphia, PA 19104, USA lannie@seas.upenn.edu Ani Nenkova University of Pennsylvania Philadelphia, PA 19104, USA nenkova@seas.upenn.edu Abstract We introduce a model of coherence which captures the intentional discourse structure in text. Our work is based on the hypothesis that syntax provides a proxy for the communica- tive goal of a sentence and therefore the se- quence of sentences in a coherent discourse should exhibit detectable structural patterns. Results show that our method has high dis- criminating power for separating out coherent and incoherent news articles reaching accura- cies of up to 90%. We also show that our syn- tactic patterns are correlated with manual an- notations of intentional structure for academic conference articles and can successfully pre- dict the coherence of abstract, introduction and related work sections of these articles. 1 Introduction Recent studies have introduced successful automatic methods to predict the structure and coherence of texts. They include entity approaches for local co- herence which track the repetition and syntactic re- alization of entities in adjacent sentences (Barzilay and Lapata, 2008; Elsner and Charniak, 2008) and content approaches for global coherence which view texts as a sequence of topics, each characterized by a particular distribution of lexical items (Barzilay and Lee, 2004; Fung and Ngai, 2006). Other work has shown that co-occurrence of words (Lapata, 2003; Soricut and Marcu, 2006) and discourse relations (Pitler and Nenkova, 2008; Lin et al., 2011) also pre- dict coherence. Early theories (Grosz and Sidner, 1986) posited that there are three factors which collectively con- tribute to coherence: intentional structure (purpose of discourse), attentional structure (what items are discussed) and the organization of discourse seg- ments. The highly successful entity approaches cap- ture attentional structure and content approaches are related to topic segments but intentional structure has largely been neglected. Every discourse has a purpose: explaining a concept, narrating an event, critiquing an idea and so on. As a result each sen- tence in the article has a communicative goal and the sequence of goals helps the author achieve the dis- course purpose. In this work, we introduce a model to capture coherence from the intentional structure dimension. Our key proposal is that syntactic pat- terns are a useful proxy for intentional structure. This idea is motivated from the fact that cer- tain sentence types such as questions and definitions have distinguishable and unique syntactic structure. For example, consider the opening sentences of two descriptive articles1 shown in Table 1. Sentences (1a) and (2a) are typical instances of definition sen- tences. Definitions are written with the concept to be defined expressed as a noun phrase followed by a copular verb (is/are). The predicate contains two parts: the first is a noun phrase reporting the concept as part of a larger class (eg. …

Content truncated to 500 words