Sren Kierkegaard at SemEval-2023 Task 4: Label-aware text classification using Natural Language Inference

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1871–1877 July 13-14, 2023 ©2023 Association for Computational Linguistics Søren Kierkegaard at SemEval-2023 Task 4: Label-aware text classification using Natural Language Inference Ignacio Talavera Cepeda and Amalie Brogaard Pauli and Ira Assent Department of Computer Science, Aarhus University, Denmark ignacio@post.au.dk, ampa@cs.au.dk, ira@cs.au.dk Abstract In this paper, we describe our approach to Task 4 in SemEval 2023. Our pipeline tries to solve the problem of multi-label text classification of human values in English-written arguments. We propose a label-aware system where we reframe the multi-label task into a binary task resembling an NLI task. We propose to include the semantic description of the human values by comparing each description to each argument and ask whether there is entailment or not. 1 Introduction Shared tasks are competitions organized by dif- ferent academic institutions that put the scope on specific research problems that are challenging for the community, typically because of low data avail- ability. Natural Language Processing (NLP) has many shared tasks annually (Palmer et al., 2021; May et al., 2019; Rodríguez-Sánchez et al., 2022). In these tasks, a problem that currently concerns the community is stated, and datasets are provided to the participants, which then have to try to solve that problem to get the highest score possible, and to contribute to establishing and comparing state- of-the-art approaches. SemEval 2023 is the 17th edition of the workshop, and it features 12 different NLP shared tasks, divided in several categories that include semantic structure, discourse and argumen- tation, medical application and social attitudes. This paper addressed Task 4 which is in the field of discourse and argumentation mining. The task is a text classification task where participants have to classify whether or not an argument belongs to certain human value categories (Kiesel et al., 2023). The categories are compiled from a Social Science study by Kiesel et. al. (Kiesel et al., 2022). The par- ticipants are able to submit runs that detect either a subset of the seven most common categories or all of them. Each argument is divided into a premise text, a conclusion text and a binary stance of the premise to the conclusion, which can be either "in favour of" or "against". Kiesel et al. (2023) provide a dataset with the challenges of identifying a subset of fundamental values behind written arguments. The task is a multi-label text classification. We propose to address the task by reframing the task from a multi-label task to a binary task which includes the textual semantic information of the labels. We reformulated the task to resem- ble a Natural Language Inference (NLI) task. The structure of NLI tasks is given a tuple with premise- hypothesis to decide if there is entailment, neu- trality or contradiction between the premise and hypothesis (MacCartney and Manning, 2008). We propose a similar structure to our task at hand; we generate pairs of each argument with a textual de- scription of each human-value. We then set the …

Content truncated to 500 words