ZJUKLAB at SemEval-2021 Task 4: Negative Augmentation with Language Model for Reading Comprehension of Abstract Meaning

Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 120–125 Bangkok, Thailand (online), August 5–6, 2021. ©2021 Association for Computational Linguistics 120 LangResearchLab NC at SemEval-2021 Task 1: Linguistic Feature Based Modelling for Lexical Complexity Raksha Agarwal, Niladri Chatterjee Indian Institute of Technology Delhi Hauz Khas, Delhi-110016, India raksha.agarwal@maths.iitd.ac.in niladri@maths.iitd.ac.in Abstract The present work aims at assigning a complex- ity score between 0 and 1 to a target word or phrase in a given sentence. For each Sin- gle Word Target, a Random Forest Regressor is trained on a feature set consisting of lexi- cal, semantic, and syntactic information about the target. For each Multiword Target, a set of individual word features is taken along with single word complexities in the feature space. The system yielded the Pearson correlation of 0.7402 and 0.8244 on the test set for the Single and Multiword Targets, respectively. 1 Introduction Presence of complex words can lead to poor com- prehension of a text. Identification of such complex words in a given text is a core component in the task of Automatic Simplification and Evaluation (Shard- low, 2013). The Lexical Complexity Prediction Task of SemEval 2021 (Shardlow et al., 2021) aims at development of systems for prediction of com- plexity scores for a target word/phrase in a given sentence. In literature, binary classification of tar- get words in a text into complex or non-complex is referred to as Complex Word Identification (CWI) (Paetzold and Specia, 2016; Zampieri et al., 2017; Gooding and Kochmar, 2018; AbuRa’ed and Sag- gion, 2018; Yimam et al., 2018). Unlike previous works, a continuous complexity score is assigned to the target word in the present task which is referred to as Lexical Complexity Prediction (LCP) (Shard- low et al., 2020). For the present work, regression is performed for LCP on a set of linguistic fea- tures covering semantic, syntactic and contextual aspects of the target word as described in Section 3. Additionally, various lexicon based features are used to indicate the rarity of target words. The sys- tem achieves 0.8194 Pearson correlation for Single Word Target and 0.7482 for Multiword Target on the trial set. 2 Task Setup The task is divided into two subtasks, namely Sin- gle Word Target and Multiword Target based on the length of the target. The dataset and evaluation metrics are described below. • Dataset: The dataset consists of an aug- mented version of CompLex (Shardlow et al., 2020). It comprises sentences from three cor- pora, viz. World English Bible Translation, English Portion of the European Parliament proceedings, and articles from CRAFT corpus belonging to biomedical domain. It is split into three subsets Train, Trial, and Test. • Evaluation Metrics: The systems are eval- uated using Pearson correlation coefficient (P), Spearman rank correlation coefficient (S), Mean absolute error (MAE) and Coefficient of Determination (R2). 3 Features In this section we present the details of the feature space used in the present work. 3.1 Corpus Features A feature, named Corpus, is used to indicate to …

Content truncated to 500 words