CSECU-DSG at SemEval-2022 Task 3: Investigating the Taxonomic Relationship Between Two Arguments using Fusion of Multilingual Transformer Models

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 255 - 259 July 14-15, 2022 ©2022 Association for Computational Linguistics CSECU-DSG at SemEval-2022 Task 3: Investigating the Taxonomic Relationship Between Two Arguments using Fusion of Multilingual Transformer Models Abdul Aziz, Md. Akram Hossain, and Abu Nowshed Chy Department of Computer Science and Engineering University of Chittagong, Chattogram-4331, Bangladesh {aziz.abdul.cu, akram.hossain.cse.cu}@gmail.com, and nowshed@cu.ac.bd Abstract Recognizing lexical relationships between words is one of the formidable tasks in compu- tational linguistics. It plays a vital role in the improvement of various NLP tasks. However, the diversity of word semantics, sentence struc- ture as well as word order information make it challenging to distill the relationship effec- tively. To address these challenges, SemEval- 2022 Task 3 introduced a shared task PreTENS focusing on semantic competence to determine the taxonomic relations between two nominal arguments. This paper presents our participa- tion in this task where we proposed an approach through exploiting an ensemble of multilingual transformer models. We employed two fine- tuned multilingual transformer models includ- ing XLM-RoBERTa and mBERT to train our model. To enhance the performance of individ- ual models, we fuse the predicted probability score of these two models using weighted arith- metic mean to generate a unified probability score. The experimental results showed that our proposed method achieved competitive per- formance among the participants’ methods. 1 Introduction Lexical relations are regarded as the most important semantic relations to infer the meanings of words effectively (Khoo and Na, 2006). Therefore, iden- tifying such relations is beneficial to understand the semantic competence and distill the underlying context of the textual expression. If we consider the NLP applications, it has a significant impact on several tasks including semantic search, automatic question/answering, story generation, relation ex- traction, and machine translation (Barkan et al., 2020). However, most of the prior works focused mainly on syntactic (Luu et al., 2014) and contex- **The first two authors have equal contributions. Sentence Language Label / Score Sub-task 1 Apprezzo il vino , ma non il Chianti. It 1 J’ aime les chats, sauf les beagles. Fr 0 I like movies, but not comedies. En 1 I like earrings, except socks. En 0 Sub-task 2 Amo i merli pi ˜A¹ degli uccelli. It 1.9 J’aime les chats, et aussi les canards. Fr 6.17 I like cats, but not sparrows. En 4.77 Table 1: Examples of sub-task 1 and sub-task 2. tual relation (Maksimov et al., 2018) to infer the lexical relation in between words. Considering this PreTENS shared task at Se- mEval 2022 (Zamparelli et al., 2022) introduce a new task that focuses mainly on semantic com- petence with specific attention on the estimating taxonomic relations between two nominal argu- ments. The taxonomic relation here means one argument is a supercategory of the other, or in ex- tensional terms, one argument is a superset of the other. The task is divided into two subtasks. The first one is a binary classification subtask, where a system needs to predict …

Content truncated to 500 words