Modelling Entity Instantiations

Proceedings of Recent Advances in Natural Language Processing, pages 268–274, Hissar, Bulgaria, 12-14 September 2011. Modelling Entity Instantiations Andrew McKinlay School of Computing University of Leeds, UK scs4ajm@comp.leeds.ac.uk Katja Markert School of Computing University of Leeds, UK markert@comp.leeds.ac.uk Abstract We introduce the problem of detecting En- tity Instantiations, a type of entity relation in which a set of entities is introduced, and either a member or subset of this set is mentioned afterwards. We perform the first, reliable, corpus study of Entity In- stantiations, concentrating on intersenten- tial annotation. We then develop the first automatic instantiation detector, which in- corporates lexical, contextual and world knowledge and shows significant improve- ments over a strong baseline. 1 Introduction In this paper we annotate and classify Entity In- stantiations. An Entity Instantiation is a non- coreferent entity relationship, where a set of en- tities is mentioned, and then a member or subset1 of this set is introduced. Example 1 shows a pair of sentences with the set in bold and set member in italics.2 Examples 2 and 3 show a pair of sen- tences with a set in bold and subset in italics. (1) a. Some European funds recently have skyrocketed. b. Spain Fund has surged to a startling 120% premium. (2) a. Bids totalling $515 million were sub- mitted. b. Accepted offers ranged from 8.38% to 8.395% 1When we refer to a subset, we mean a proper subset. We consider two equal sets to be coreferent, and not participating in an Entity Instantiation. 2Examples 1, 2, 3, 8 and 9 are adapted from the Penn Treebank Wall Street Journal Corpus (Marcus et al., 1993). (3) a. In the aftermath of the downturn many manufacturers have struggled. b. Those relying on foreign imports have had the most difficulty. The detection of Entity Instantiations is not tack- led in ACE (ACE, 2000–2005) or MUC (MUC, 1987–1998), the two most popular schemes of se- mantic relation annotation. It is, however, im- portant as it can supplement knowledge about the member or subset. In Example 4 below, the Entity Instantiation between ‘several EU countries’ and ‘the UK’ gives us the knowledge that not only are interest rates dropping in the UK, but inflation is rising as well. Entity Instantiations can also aid the interpretation of sentiment — in Example 5, the author’s thoughts about the pay of Wayne Rooney can be inferred from the negative sentiment of the first sentence. In some instances, the member or subset is even uninterpretable without the set. In Example 3, ‘Those relying on foreign imports’ re- quires ‘many manufacturers’ to interpret the miss- ing head noun. The problem of detecting these types of Entity Instantiation overlaps with bridg- ing anaphora. (4) a. Inflation has increased sharply in sev- eral EU countries. b. In the UK, this has accompanied a drop in interest rates. (5) a. Footballers are vastly overpaid. b. Manchester United pay Wayne Rooney £200,000 per week. The interpretation of Entity Instantiations can of- ten be difficult. a variety …

Content truncated to 500 words