CTPs: Contextual Temporal Profiles for Time Scoping Facts using State Change Detection

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1930–1936, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics CTPs: Contextual Temporal Profiles for Time Scoping Facts using State Change Detection Derry Tanti Wijaya Carnegie Mellon University 5000 Forbes Avenue Pittsburgh, PA, 15213 dwijaya@cs.cmu.edu Ndapandula Nakashole Carnegie Mellon University 5000 Forbes Avenue Pittsburgh, PA, 15213 ndapa@cs.cmu.edu Tom M. Mitchell Carnegie Mellon University 5000 Forbes Avenue Pittsburgh, PA, 15213 tom.mitchell@cs.cmu.edu Abstract Temporal scope adds a time dimension to facts in Knowledge Bases (KBs). These time scopes specify the time periods when a given fact was valid in real life. With- out temporal scope, many facts are under- specified, reducing the usefulness of the data for upper level applications such as Question Answering. Existing methods for temporal scope inference and extrac- tion still suffer from low accuracy. In this paper, we present a new method that lever- ages temporal profiles augmented with context— Contextual Temporal Profiles (CTPs) of entities. Through change pat- terns in an entity’s CTP, we model the en- tity’s state change brought about by real world events that happen to the entity (e.g, hired, fired, divorced, etc.). This leads to a new formulation of the temporal scoping problem as a state change detection prob- lem. Our experiments show that this for- mulation of the problem, and the resulting solution are highly effective for inferring temporal scope of facts. 1 Introduction Recent years have seen the emergence of large Knowledge Bases (KBs) of facts (Carlson 2010; Auer 2007; Bollacker 2008; Suchanek 2007). While the wealth of accumulated facts is huge, most KBs are still sparsely populated in terms of temporal scope. Time information is an important dimension in KBs because knowledge is not static, it changes over time: people get divorced; coun- tries elect new leaders; and athletes change teams. This means that facts are not always indefinitely true. Therefore, temporal scope has crucial impli- cations for KB accuracy. Figure 1: Behavior patterns of context uni-grams for the US presidency state change as seen in the Google Books N-grams corpus: the rise of ‘elect’, immediately followed by the rise of ‘administra- tion’ and ‘president’. Towards bridging the time gap in KBs, we propose a new method for temporal scope infer- ence. Our method is based on leveraging aggre- gate statistics from a time-stamped corpus. First we generate Contextual Temporal Profiles (CTPs) of entities from contexts surrounding mentions of these entities in the corpus. We then detect change patterns in the CTPs. We then use these changes to determine when a given entity undergoes a spe- cific state change caused by real world events. Our main insight is as follows: events that happen to an entity change the entity’s state and therefore its facts. Thus by learning when a given entity under- goes a specific state change, we can directly infer the time scopes of its facts. For example, in the di- vorce event, the person’s state changes from ‘mar- ried’ to ‘divorced’ hence the …

Content truncated to 500 words