LCP-RIT at SemEval-2021 Task 1: Exploring Linguistic Features for Lexical Complexity Prediction

Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 548–553 Bangkok, Thailand (online), August 5–6, 2021. ©2021 Association for Computational Linguistics 548 LCP-RIT at SemEval-2021 Task 1: Exploring Linguistic Features for Lexical Complexity Prediction Abhinandan Desai, Kai North, Marcos Zampieri, Christopher Homan Rochester Institute of Technology Rochester, NY, USA ad2724@rit.edu, kn1473@rit.edu, mazgla@rit.edu, cmh@cs.rit.edu Abstract This paper describes team LCP-RIT’s sub- mission to the SemEval-2021 Task 1: Lex- ical Complexity Prediction (LCP). The task organizers provided participants with an aug- mented version of CompLex (Shardlow et al., 2020), an English multi-domain dataset in which words in context were annotated with respect to their complexity using a five point Likert scale. Our system uses logistic regres- sion and a wide range of linguistic features (e.g. psycholinguistic features, n-grams, word frequency, POS tags) to predict the complexity of single words in this dataset. We analyze the impact of different linguistic features on the classification performance and we evaluate the results in terms of mean absolute error, mean squared error, Pearson correlation, and Spear- man correlation. 1 Introduction Lexical complexity prediction (LCP) is the task of predicting the complexity value of a target word within a given text (Shardlow et al., 2020). Com- plexity within LCP is used as a “synonym for diffi- culty” (Malmasi and Zampieri, 2016)1. A complex word is therefore a word that a target population may find difficult to understand. Various LCP sys- tems have been designed to identify words that may be found to be complex for children (Kajiwara et al., 2013), language learners (Malmasi et al., 2016), or people suffering from a reading disability, such as dyslexia (Rello et al., 2013). These sys- tems have been utilized within assistive language technologies, lexical simplification systems, and in a variety of other applications. LCP is related to complex word identification (CWI) (Paetzold and Specia, 2016). CWI is mod- eled as a binary classification task by assigning each target word with a complex or non-complex 1The term “complex” within LCP is not necessarily related to the terms simplex and complex used in morphology. label. The shortcomings of modeling lexical com- plexity using binary labels have been discussed in previous work (Zampieri et al., 2017; Maddela and Xu, 2018), motivating the organization of SemEval- 2021 Task 1: Lexical Complexity Prediction.2 LCP models complexity in a continuum and the goal is to predict a target word’s degree of complexity by assigning it a value between 0 and 1. This value may then correspond to one of the following la- bels: very easy (0), easy (0-0.25), neutral (0.25- 0.5), difficult (0.5-0.75), or very difficult (0.75-1) (Shardlow et al., 2020). In this paper, we describe (in detail in Section 4) the LCP-RIT entry to SemEval-2021 Task 1. We approached LCP from a feature engineering per- spective with a particular focus on the adoption of psycholinguistic features, such as average age- of-acquisition (AoA), familiarity, prevalence, con- creteness, and arousal, alongside the use of prior complexity labels. Our submitted system utilized a combination of these …

Content truncated to 500 words