Patterns of Wisdom: Discourse-Level Style in Multi-Sentence Quotations

Proceedings of NAACL-HLT 2016, pages 1139–1144, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics Patterns of Wisdom: Discourse-Level Style in Multi-Sentence Quotations Kyle Booten and Marti A. Hearst UC Berkeley Berkeley, CA 94720 kbooten@berkeley.edu, hearst@berkeley.edu Abstract Quotations are kernels not just of wisdom but also of beautiful and striking language. While recent studies have characterized the stylistic features of quotations, we characterize the or- der of stylistic information within quotations. Analyzing a corpus of two-sentence quota- tions collected from the social network Tum- blr, we explore the ways that both low-level features and high-level features tend to occur in either the first or second sentence. Through analysis of examples, we interpret these ten- dencies as manifestations of rhetorical pat- terns. Results from a prediction task suggest that stylistic patterns are more prominent in quotations than in a comparison corpus. 1 Introduction The ancient arts of rhetoric described ways of wield- ing language to make it particularly persuasive or memorable. Central in this endeavor—a predecessor to modern linguistics (Dolven, 2013)—was the de- scription of rhetorical “tropes” or “figures,” patterns of language from the level of the phoneme (as in the case of rhyme or alliteration) to higher-level syntac- tic and even logical structures (Peacham, 1954). In the figure of epistrophe, successive clauses end with the same words. In pysma, the speaker (or writer) launches a series of sharp and vehement questions; this and other rhetorical figures describe language at the level of discourse—that is, the relationship be- tween linguistic elements across sentences. Recent studies of quotations have described what makes certain fragments of text more memorable than others (Danescu-Niculescu-Mizil et al., 2012; Guerini et al., 2015); this ongoing project can be seen as a contemporary, empirical investigation of the rhetorical arts. Inspired by the way that clas- sical rhetoricians described complicated and high- level linguistic patterns, the present study takes the novel step of analyzing the way linguistic elements are sequenced within quotations. After describing a minimalistic set of stylistic features designed to cap- ture common patterns in quotations, we demonstrate that some features tend to occur in certain positions within a quotation. We investigate whether quotes are more predictable than other genres in a “Quote Ordering Task” in which the goal is to distinguish the correct version of a quotation from one whose sentences have been reversed. 2 Related Work 2.1 Style of Memorable Language Danescu-Niculescu-Mizil et al. (2012) used a vari- ety of features to distinguish popular movie quota- tions from unmemorable lines from the same movie. By modeling word and part-of-speech sequences of quotes and non-quotes, they found that quotations tended to use less common words but that these words were placed in more common syntactic pat- terns. The researchers evocatively hint at some “common syntactic scaffolding” that structures quo- tations, and we build on this finding by characteriz- ing these patterns. They also found that quotations tend to contain linguistic features that make them more “generalizable,” such as tendency toward in- definite over definite …

Content truncated to 500 words