PolyU CBS-Comp at SemEval-2021 Task 1: Lexical Complexity Prediction (LCP)

Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 565–570 Bangkok, Thailand (online), August 5–6, 2021. ©2021 Association for Computational Linguistics 565 PolyU CBS-Comp at SemEval-2021 Task 1: Lexical Complexity Prediction (LCP) Rong Xiang Department of Computing The Hong Kong Polytechnic University csrxiang@comp.polyu.edu.hk Jinghang Gu Chinese and Bilingual Studies The Hong Kong Polytechnic University jinghang.gu@polyu.edu.hk Emmanuele Chersoni Chinese and Bilingual Studies The Hong Kong Polytechnic University emmanuelechersoni@gmail.com Wenjie Li Department of Computing The Hong Kong Polytechnic University cswjli@comp.polyu.edu.hk Qin Lu Department of Computing The Hong Kong Polytechnic University csluqin@comp.polyu.edu.hk Chu-Ren Huang Chinese and Bilingual Studies The Hong Kong Polytechnic University churen.huang@polyu.edu.hk Abstract In this contribution, we describe the system presented by the PolyU CBS-Comp Team at the Task 1 of SemEval 2021, where the goal was the estimation of the complexity of words in a given sentence context. Our top system, based on a combination of lexical, syntactic, word embeddings and Transformers-derived features and on a Gradi- ent Boosting Regressor, achieves a top corre- lation score of 0.754 on the subtask 1 for sin- gle words and 0.659 on the subtask 2 for mul- tiword expressions. 1 Introduction The notion of complexity has often been debated in linguistics and, depending from the disciplines, it might have different meanings. In linguistic typology, for example, complexity is generally studied as a property of the language system as a whole, it is conceived as the number of (morphological, syntactic, semantic etc.) distinc- tions that a speaker has to master, and it is assessed by comparing different languages (McWhorter, 2001; Parkvall, 2008). On the other hand, in the perspective of psycholinguistics and cognitive sci- ence, the notion of complexity can be described as the difficulty encountered by language users while processing concrete linguistic realizations (sentences, utterances etc.) (Blache, 2011; Cher- soni et al., 2016, 2017, 2021; Iavarone et al., 2021; Sarti et al., 2021). Finally, in the Computational Linguistics community, the assessment of complex- ity at the lexical level is often related to readability applications (Shardlow et al., 2020), with the goal of determining if a word in a given text will be difficult to understand for the language users. Such applications are extremely useful for second lan- guage learners, for speakers with relatively low literacy and for people with reading disabilities, helping to tailor the difficult level of the texts to the needs of the target users. Task 1 of SemEval 2021 (Shardlow et al., 2021) aims at the development of systems for the estima- tion of lexical complexity in context, both for single words and for multiword expressions. The organiz- ers provided two datasets with the target words in a sentence context, with annotations consisting of a mean of the complexity ratings assigned by humans. In our paper, we present the system developed by the PolyU CBS-Comp team for the competition. Our top system achieves a Pearson correlation of, respectively, 0.754 on the single words dataset and 0.659 on the multiword expressions one. 2 Related Work In the earliest shared …

Content truncated to 500 words