UIRISC at SemEval-2023 Task 10: Explainable Detection of Online Sexism by Ensembling Fine-tuning Language Models

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 10–15 July 13-14, 2023 ©2023 Association for Computational Linguistics lasigeBioTM at SemEval-2023 Task 7: Improving Natural Language Inference Baseline Systems with Domain Ontologies Sofia I. R. Conceição † , Diana F. Sousa †, Pedro M. Silvestre †, Francisco M. Couto ‡ LASIGE, Faculdade de Ciências, Universidade de Lisboa, 1749-016 Lisbon, Portugal †{sconceicao,dfsousa,psilvestre}@lasige.di.fc.ul.pt ‡fjcouto@edu.ulisboa.pt Abstract Clinical Trials Reports (CTRs) contain highly valuable health information from which Natu- ral Language Inference (NLI) techniques deter- mine if a given hypothesis can be inferred from a given premise. CTRs are abundant with do- main terminology with particular terms that are difficult to understand without prior knowledge. Thus, we proposed to use domain ontologies as a source of external knowledge that could help with the inference process in the SemEval-2023 Task 7: Multi-evidence Natural Language In- ference for Clinical Trial Data (NLI4CT). This document describes our participation in subtask 1: Textual Entailment, where Ontologies, NLP techniques, such as tokenization and named- entity recognition, and rule-based approaches are all combined in our approach. We were able to show that inputting annotations from domain ontologies improved the baseline systems. 1 Introduction Natural Language Inference (NLI) determines whether a given hypothesis can be deduced from a given premise (Romanov and Shivade, 2018). Par- ticularly in the clinical text, where there is much variation in terminologies, NLI is more challenging (Romanov and Shivade, 2018). The systems have to deal with obstacles that are constantly present in the text, such as homonyms, acronyms, or ab- breviations, leading to ambiguity since, without context or background, some can be mapped to di- verse expanded forms that are not entirely correct in the context (Krallinger et al., 2008; Couto and Krallinger, 2020). Take, for example, a part of a sentence with this text "Amyotrophic lateral scle- rosis (ALS) patients..." ALS here is the acronym for the disease Amyotrophic lateral sclerosis, but ALS is also the gene symbol for the human gene SOD1 superoxide dismutase 1. Without context, when only presented with the ALS word, it is not possible to know if it is referring to the disease, the gene, or even another possibility not explored in this example. Clinical trials play a fundamental role in discov- ering new health information (National Library of Medicine, National Center for Biotechnology In- formation, 2023). In these studies, the aim is to per- form pre-defined interventions in volunteers. For instance, the goal can be for clinicians to study the effects of different drug concentrations on patients. The outputs of these studies are registered in Clin- ical Trials Reports (CTR). These reports store all the information about the conditions to be selected to participate in the trial, groups of participants, information about dosage and duration, results, and adverse events. CTRs are rich in domains with specific terms that are not easy to grasp without prior knowledge. Domain ontologies can provide this external knowledge. Ontologies are defined by Gruber (1993) as being the "specification of conceptualization" and …

Content truncated to 500 words