NLU-Co at SemEval-2020 Task 5: NLU/SVM Based Model Apply Tocharacterise and Extract Counterfactual Items on Raw Data

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 670–676 Barcelona, Spain (Online), December 12, 2020. 670 NLU-Co at SemEval-2020 Task 5: NLU/SVM based model apply to characterise and extract counterfactual items on raw data MBONING TCHIAZE Elvis ERTIM / 2 rue de Lille, Paris elvis.mboning@inalco.fr Damien Nouvel ERTIM / 2 rue de Lille, Paris damien.nouvel@inalco.fr Abstract In this article, we try to solve the problem of classification of counterfactual statements and extraction of antecedents/consequences in raw data, by mobilizing on one hand Support vector machine (SVMs) and on the other hand Natural Language Understanding (NLU) infrastructures available on the market for conversational agents. Our experiments allowed us to test different pipelines of two known platforms (Snips NLU and Rasa NLU). The results obtained show that a Rasa NLU pipeline, built with a well-preprocessed dataset and tuned algorithms, allows to model accurately the structure of a counterfactual event, in order to facilitate the identification and the extraction of its components. 1 Introduction We can define counterfactuals as thoughts about alternatives to past events (factual), that is, thoughts of what might have been (counter). These counterfactual thoughts are preventive in nature because they help us better anticipate and guide our daily actions. Practically, counterfactual statements require a deep semantic analysis to identify them and extract their characteristic components (antecedent and consequent). Though Natural Language Understanding (NLU) is usually used in chatbots, digital personal assistants, or spoken dialogue systems (SDS), its techniques can also be applied in information retrieval/extraction and more specifically in the semantic classification of events (Catizone et al., 2008), (Wang and Zhang, 2011). This proposal is based on the starting baseline (Yang et al., 2020). We want to exploit the pipelines offered by two mainstream NLU libraries to classify and extract the consequent and antecedent segments occurring in a textual sequence of counterfactual type. For subtask 1, two systems are proposed: the first based on Support Vector Machine (SVM) and the second on NLU. For subtask 2, we only use NLU pipelines. The two NLU solutions are built on the basis of Snips NLU (Coucke et al., 2018) and Rasa NLU (Bocklisch et al., 2017). 2 Background: counterfactual thoughts and statements The counterfactual problem is above all a phenomenon of human thinking and reasoning. It was widely studied quite early in several disciplines, mainly in philosophy with Aristotle, Plato, Leibniz (He describes the existence of several worlds gravitating around an event controlled by laws of logic.)(Epstude and Roese, 2008, p.1). As described by (McMullen et al., 1995), this thought can target reflections on the negativity of an event (Downward counterfactuals = ”what else could I have done to do worse ?”) or on its positivity (Upward counterfactuals = ”what else could I have done to do well ?”). Kahneman and Miller proposed a Norm theory (Kahneman and Miller, 1986) which describes coun- terfactual as ”Unusual experiences tending to result in ‘if only’ thoughts that recapitulate the normal state of affairs.” (Epstude and Roese, 2008). Lewis (Lewis, 1973) on …

Content truncated to 500 words