Characterizing Stylistic Elements in Syntactic Structure

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 1522–1533, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Characterizing Stylistic Elements in Syntactic Structure Song Feng Ritwik Banerjee Department of Computer ScienceYejin Choi Stony Brook University NY 11794, USA songfeng, rbanerjee, ychoi@cs.stonybrook.edu Abstract Much of the writing styles recognized in rhetorical and composition theories involve deep syntactic elements. However, most previous research for computational sty- lometric analysis has relied on shallow lexico-syntactic patterns. Some very re- cent work has shown that PCFG models can detect distributional difference in syn- tactic styles, but without offering much in- sights into exactly what constitute salient stylistic elements in sentence structure characterizing each authorship. In this paper, we present a comprehensive ex- ploration of syntactic elements in writing styles, with particular emphasis on inter- pretable characterization of stylistic ele- ments. We present analytic insights with respect to the authorship attribution task in two different domains. 1 Introduction Much of the writing styles recognized in rhetor- ical and composition theories involve deep syn- tactic elements in style (e.g., Bain (1887), Kem- per (1987) Strunk and White (2008)). However, previous research for automatic authorship at- tribution and computational stylometric analy- sis have relied mostly on shallow lexico-syntactic patterns (e.g., Mendenhall (1887), Mosteller and Wallace (1984), Stamatatos et al. (2001), Baayen et al. (2002), Koppel and Schler (2003), Zhao and Zobel (2007), Luyckx and Daelemans (2008)). Some very recent works have shown that PCFG models can detect distributional differ- ence in sentence structure in gender attribution (Sarawgi et al., 2011), authorship attribution (Raghavan et al., 2010), and native language identification (Wong and Dras, 2011). However, still very little has been understood exactly what constitutes salient stylistic elements in sentence structures that characterize each author. Al- though the work of Wong and Dras (2011) has extracted production rules with highest informa- tion gain, their analysis stops short of providing insight any deeper than what simple n-gram- level analysis could also provide.1 One might even wonder whether PCFG models are hing- ing mostly on leaf production rules, and whether there are indeed deep syntactic differences at all. This paper attempts to answer these questions. As an example of syntactic stylistic elements that have been much discussed in rhetorical the- ories, but have not been analyzed computation- ally, let us consider two contrasting sentence styles: loose (cumulative) and periodic:2 a loose sentence places the main clause at the begin- ning, and then appends subordinate phrases and clauses to develop the main message. In con- trast, a periodic sentence starts with subordi- nate phrases and clauses, suspending the most 1For instance, missing determiners in English text written by Chinese speakers, or simple n-gram anomaly such as frequent use of “according to” by Chinese speak- ers (Wong and Dras, 2011). 2Periodic sentences were favored in classical times, while loose sentences became more popular in the modern age. 1522 Hobbs Joshi Lin McDon SˆROOT → S , CC S PPˆPRN …

Content truncated to 500 words