A General Framework for the Annotation of Causality Based on FrameNet

A general framework for the annotation of causality based on FrameNet Laure Vieu†, Philippe Muller†, Marie Candito⋆, Marianne Djemaa⋆ †IRIT CNRS & Toulouse University, ⋆Alpage INRIA & Paris Diderot Abstract We present here a general set of semantic frames to annotate causal expressions, with a rich lexicon in French and an annotated corpus of about 4000 instances of causal lexical items with their corresponding semantic frames. The aim of our project is to have both the largest possible coverage of causal phenomena in French, across all parts of speech, and have it linked to a general semantic framework such as FN, to benefit in particular from the relations between other semantic frames, e.g., temporal ones or intentional ones, and the underlying upper lexical ontology that enables some forms of reasoning. This is part of the larger ASFALDA French FrameNet project, which focuses on a few different notional domains which are interesting in their own right (Djemaa et al., 2016), including cognitive positions and communication frames. In the process of building the French lexicon and preparing the annotation of the corpus, we had to remodel some of the frames proposed in FN based on English data, with hopefully more precise frame definitions to facilitate human annotation. This includes semantic clarifications of frames and frame elements, redundancy elimination, and added coverage. The result is arguably a significant improvement of the treatment of causality in FN itself. Keywords: semantic role labelling, causality, FrameNet 1. Introduction A lot of information in natural language is of a causal na- ture: relations between events, explanations, argumenta- tions are all important to the understanding of texts, and thus useful in question-answering (Oh et al., 2013), infor- mation extraction (Khoo et al., 2000; nan Cao et al., 2014) or textual entailment (Gordon et al., 2012). Davidson, in his work on individuation of events (Davidson, 1969), went as far as saying causality defines events, as “Events are iden- tical if and only if they have exactly the same causes and effects”, and causality is an active area of research in lin- guistics (Neeleman and van de Koot, 2012). Causality is also an important notion in discourse analysis, being one of the most important categories of discourse relations (Hovy and Maier, 1992; Mak and Sanders, 2013). Numerous studies in NLP have focused on extraction of cause-effect relationships between nouns denoting events (Kozareva, 2012), or relations between events denoted by verbs (Riaz and Girju, 2013; Do et al., 2011), or combi- nations of nouns and verbs (Riaz and Girju, 2014). Other researchers try instead to gather general knowledge about typical causal links, mostly expressed by verbs (Hashimoto et al., 2009; Szpektor and Dagan, 2008; Chklovski and Pan- tel, 2004), essentially for textual entailment or knowledge mining. They all target realized causalities: an event occurred and caused another event, or is likely to be the cause of another event. This is also the focus of annotation efforts, integrat- ing time and causal information, as in (Mirza and Tonelli, 2014; Bethard and Martin, …

Content truncated to 500 words