Linguistics Department

Colloquium - Alexis Palmer

What does this clause do? Automatic classification of situations in text

Mon, March 26, 2018 | CLA 1.302E

3:00 PM - 5:00 PM

Computational approaches to semantic analysis of sentences/clauses often focus on understanding the propositional content of a clause, identifying the meanings of predicates and the relations between predicates and participants. For any given propositional content, though, speakers also make choices regarding the manner of presentation. For example, a particular situation may be described in different ways, focusing on either the actions of the participants or the attributes of those same participants. Smith (2003) offers a typology of clause types (called "situation entity" (SE) types) identifying the type of SE a clause introduces to the discourse in which it appears. Smith's types are State, Event, Generic Sentence, Generalizing Sentence (~habitual), Fact, and Proposition.


In this talk I will present computational approaches to classifying clauses according to their SE type. The first part of the talk describes a large, multi-genre corpus of English texts in which each clause is manually labeled for SE type and three related semantic features: lexical aspect, genericity, and habituality. Using this corpus, we then build models to predict such labels for new clauses of text. Our best model achieves classification accuracy of roughly 75%, approaching the human upper bound. Finally, I will discuss possible applications of these classifications for textual analysis.


This talk presents joint work with Annemarie Friedrich, Maria Becker, Manfred Pinkal, and Anette Frank.


Carlota S. Smith. 2003. Modes of Discourse. Cambridge University Press.

Bookmark and Share