IRLab_DAIICT at SemEval-2020 Task 9: Machine Learning and Deep Learning Methods for Sentiment Analysis of Code-Mixed Tweets

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 142–149 Barcelona, Spain (Online), December 12, 2020. 142 BRUMS at SemEval-2020 Task 3: Contextualised Embeddings for Predicting the (Graded) Effect of Context in Word Similarity Hansi Hettiarachchi1, Tharindu Ranasinghe2 1School of Computing and Digital Technology, Birmingham City University, UK 2Research Group in Computational Linguistics, University of Wolverhampton, UK hansi.hettiarachchi@mail.bcu.ac.uk t.d.ranasinghehettiarachchige@wlv.ac.uk Abstract This paper presents the team BRUMS submission to SemEval-2020 Task 3: Graded Word Sim- ilarity in Context. The system utilises state-of-the-art contextualised word embeddings, which have some task-specific adaptations, including stacked embeddings and average embeddings. Overall, the approach achieves good evaluation scores across all the languages, while maintaining simplicity. Following the final rankings, our approach is ranked within the top 5 solutions of each language while preserving the 1st position of Finnish subtask 2. 1 Introduction In natural language, meaning of a word has an influence by its surrounding or context words. This impact is mainly driven by the associated linguistic and cognitive phenomena (Armendariz et al., 2020b). Following these two phenomena, it was found that word meanings have a continuous nature in addition to the commonly known discrete nature. From the cognitive perspective, word meanings can be assigned based on the conceptual structures in human mind (Evans, 2006; G¨ardenfors, 2014). Therefore, meaning of a word can be varied according to the mental state of the reader which is triggered with the contact of context words. Thus, word meanings are not just limited to a discrete nature. From the linguistic perspective, context words can modify the meaning of a word by contextual selection and contextual modulation (Cruse et al., 1986). Contextual selection identifies the meanings of polysemous words (i.e. words with multiple senses) such as ‘cell’, ‘bank’, etc. using the context words. In this case, the most appropriate meaning will be picked from a set of discrete senses. For an example, phrase ‘prison cell’ implies that the word ‘cell’ refers a room. Contextual modulation modifies the meanings of single sense words by highlighting their characteristics. Thus, unlike the contextual selection, modulation makes a continuous effect on the meaning and it is widely available, because majority of the words are general to a certain extent. For an example, ‘butter’ is a single sense word. But, if it appears in the phrase ‘poured the butter’, context words reveal that the mentioned butter is in liquid state. Focusing on the above-mentioned facts, it is important to consider both discrete and continuous effects while predicting the meaning of words in natural language text. Even though there is a continuous effect on meaning, majority of the previous research were only focused on the discrete effect. Considering this limitation, SemEval-2020 Task 3 was designed by targeting the continuous (graded) effects of context. With the involvement of continuous effects, the goal of this task is predicting the effect of context in human perception of similarity. For an example, given the phrases: ‘Her prison cell was almost an improvement over her room at the last hostel.’ ‘His …

Content truncated to 500 words