Situation entity types: automatic classification of clause-level aspect

Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 1757–1768, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Situation entity types: automatic classification of clause-level aspect Annemarie Friedrich1 Alexis Palmer2 Manfred Pinkal1 1Department of Computational Linguistics, Saarland University, Germany 2Leibniz ScienceCampus, Dept. of Computational Linguistics, Heidelberg University, Germany {afried,pinkal}@coli.uni-saarland.de palmer@cl.uni-heidelberg.de Abstract This paper describes the first robust ap- proach to automatically labeling clauses with their situation entity type (Smith, 2003), capturing aspectual phenomena at the clause level which are relevant for interpreting both semantics at the clause level and discourse structure. Previous work on this task used a small data set from a limited domain, and relied mainly on words as features, an approach which is impractical in larger settings. We pro- vide a new corpus of texts from 13 genres (40,000 clauses) annotated with situation entity types. We show that our sequence labeling approach using distributional in- formation in the form of Brown clusters, as well as syntactic-semantic features tar- geted to the task, is robust across genres, reaching accuracies of up to 76%. 1 Introduction Clauses in text have different aspectual properties, and thus contribute to the discourse in different ways. Distinctions that have been made in the linguistic and semantic theory literature include the classification of states, events and processes (Vendler, 1957; Bach, 1986), and whether clauses introduce particular eventualities or report regular- ities generalizing either over events or members of a kind (Krifka et al., 1995). Such aspectual distinctions are relevant to natural language pro- cessing tasks requiring text understanding such as information extraction (Van Durme, 2010) or tem- poral processing (Costa and Branco, 2012). In this paper, we are concerned with automat- ically identifying the type of a situation entity (SE), which we assume to be expressed by a clause. Specifically, we present a system for au- tomatically labeling clauses using the inventory of STATE: The colonel owns the farm. EVENT: John won the race. REPORT: “...”, said Obama. GENERIC SENTENCE: Generalizations over kinds. The lion has a bushy tail. GENERALIZING SENTENCE: Generalizations over events (habituals). Mary often fed the cat last year. QUESTION: Who wants to come? IMPERATIVE: Hand me the pen! Figure 1: SE types, adapted from Smith (2003). SE types shown in Figure 1 (Smith, 2003, 2005; Palmer et al., 2007). The original motivation for the above inventory of SE types is the observation that different modes of discourse, a classification of linguistic properties of text at the passage level, have different distributions of SE types. For ex- ample, EVENTs and STATEs are predominant in narrative passages, while GENERIC SENTENCEs occur frequently in information passages. A previous approach to automatically labeling SE types (Palmer et al., 2007) – referred to here as UT07 – captures interesting insights, but is trained and evaluated on a relatively small amount of text (about 4300 clauses), mainly from one rather spe- cialized subsection of the Brown corpus. The data shows a highly skewed distribution of SE types and was annotated …

Content truncated to 500 words