An Analysis of Causality between Events and its Relation to Temporal Information

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 2097–2106, Dublin, Ireland, August 23-29 2014. An Analysis of Causality between Events and its Relation to Temporal Information Paramita Mirza Fondazione Bruno Kessler, University of Trento Trento, Italy paramita@fbk.eu Sara Tonelli Fondazione Bruno Kessler Trento, Italy satonelli@fbk.eu Abstract In this work we present an annotation framework to capture causality between events, inspired by TimeML, and a language resource covering both temporal and causal relations. This data set is then used to build an automatic extraction system for causal signals and causal links between given event pairs. The evaluation and analysis of the system’s performance provides an insight into explicit causality in text and the connection between temporal and causal relations. 1 Introduction Causality is a concept that has been widely investigated from a philosophical, psychological and logical point of view, but how to model its recognition and representation in NLP-centered applications is still an open issue. However, information on causality could be beneficial to a number of natural language processing tasks such as question answering, text summarization, decision support, etc. The lack of information extraction systems focused on causality may depend also on the lack of unified annotation guidelines and standard benchmarks, which usually foster the comparison of different systems performances. Specific phenomena related to causality, such as causal arguments (Bonial et al., 2010), causal discourse relations (The PDTB Research Group, 2008) or causal relations between nominals (Girju et al., 2007), have been investigated, but no unified framework has been proposed to capture causal relations between events, as opposed to the existing TimeML standard for temporal relations (Pustejovsky et al., 2010). The work presented in this paper copes with this issue by i) proposing an annotation framework to model causal relations between events and ii) detailing the development and the evaluation of a supervised system based on such framework. We take advantage of the formalization work carried out for the TimeML standard, in which events, temporal relations and temporal signals have been carefully defined and annotated. We propose to model causal relations in a similar way to temporal relations, inheriting from TimeML the notion of event, relation and signal, even though our approach to causality is well rooted in the force dynamic model by Talmy (1985). Besides, we focus our preliminary annotation on TimeBank (Pustejovsky et al., 2006), a corpus widely used by the research community working on temporal processing. This should possibly enable the adaptation of existing temporal processing systems to the analysis of causal information, given that we rely on well-known standards and data. On the other hand, this makes it easier for us to straightforwardly investigate the relation between temporal and causal information, given that a causing event should always take place before a resulting event. 2 Related Work Research on the extraction of event relations has concerned both the analysis of the temporal ordering of events and the recognition of causality relations. However, the two research lines have progressed quite independently from …

Content truncated to 500 words