|Date||Mar 07, 2014|
|Title||Situation Entities and Discourse Modes|
|Abstract||As readers we are sensitive to various levels of structure within texts. Theline of research described in this talk addresses two levels: the contributionof individual clauses and that of an intermediate level of text structure --passages of several sentences or more. We follow Smith (2003) in discussing theDiscourse Modes of individual text passages. Smith introduces the five modes ofNarrative, Description, Report, Information, and Argument/Commentary andfurther identifies specific linguistic correlates of these different modes. Onegroup of correlates points to the contributions made to the discourse byindividual clauses of text. As a first step toward Discourse Mode classification,we address the problem of identifying and classifying references to situationsexpressed in written English texts. The situation entity (SE) classificationtask as construed here is the classification of clauses according to the typeof entity they introduce to the discourse. The classification scheme we useincludes, among others, events, states, abstract entities, and genericsentences.|
In this talk Iwill discuss both previous and ongoing work on SE classification. In earlierwork (Palmer et al. 2007), we annotate a portion of the Brown corpus with SElabels and then use this for training and testing models for SE classification. The two key strategies employed are (1) useof linguistically-motivated features and (2) implementation of SEclassification as a sequence labelling task. We find that deep linguisticinformation improves classifier performance. Our results also provide empiricalsupport for the notion of modes of discourse, as we see clear genre effects inSE classification. Our ongoing work refines the SE classification scheme fromearlier work and aims to more carefully annotate a much larger corpus of textsfrom a number of different genres, thus laying the foundation for moreextensive computational treatment of SE type and related linguistic features.
** This talkpresents joint work with Annemarie Friedrich.
|Bio||Alexis Palmer is a postdoctoral researcher at the Department of ComputationalLinguistics and Phonetics, Saarland University (Saarbrücken, Germany). Hercurrent research focuses primarily on computational approaches to discourse andsemantics, including work on discourse modes, situation entities, short-answerscoring, and working with learner language. In addition, Alexis has a secondresearch focus on computational linguistics for low-resource languages; her2009 PhD thesis "Semi-Automatic Annotation and Active Learning forLanguage Documentation" was completed at the University of Texas at Austinunder the joint supervision of Jason Baldridge and Katrin Erk.|