SSN-NLP at SemEval-2020 Task 4: Text Classification and Generation on Common Sense Context Using Neural Networks

Proceedings of the 14th International Workshop on Semantic Evaluation, pages 580–584 Barcelona, Spain (Online), December 12, 2020. 580 SSN-NLP at SemEval-2020 Task 4: Text Classification and Generation on Common Sense Context using Neural Networks Rishivardhan K Kayalvizhi S Thenmozhi D Raghav R Kshitij Sharma SSN College of Engineering rishivardhan@ieee.org,{kayalvizhis,theni d}@ssn.edu.in, {raghav18119,kshitij18080}@cse.ssn.edu.in Abstract Common sense validation deals with testing whether a system can differentiate natural language statements that make sense from those that do not make sense. This paper describes the our approach to solve this challenge. For common sense validation with multi choice, we propose a stacking based approach to classify sentences that are more favourable in terms of common sense to the particular statement. We have used majority voting classifier methodology amongst three models such as Bidirectional Encoder Representations from Transformers (BERT), Micro Text Classification (Micro TC) and XLNet. For sentence generation, we used Neural Machine Translation (NMT) model to generate explanatory sentences. 1 Introduction Without common sense it would be difficult to build adaptable and unsupervised NLP systems in an increasingly digital and mobile world. The power of common sense systems is their sense of adaptability to varied topics. This is because common sense systems provide their response based on contextual based understanding of the problem. Its therefore essential to bridge the gap between human language processing and machine natural language processing by improving common sense validation and explanation in the latter. By the Commonsense Validation and Explanation or ComVE challenge (Wang et al., 2019), the task is to directly test whether a system can differentiate natural language statements that make sense from those that do not make sense. We primarily had focused on sub task B and sub task C of the challenge where Task B is to find the key reason from three options why a given statement does not make sense and Task C asks machine to generate the reasons and we use BLEU metric proposed by Papineni et al. (2002) to evaluate them. For Task B we had used a mix of NLP models among BERT by Devlin et al. (2018), Micro TC by Tellez et al. (2018) and XLNet from Yang et al. (2019) to run a multi classification problem and finally post process the results based on majority voting approach. For Task C we had used NMT (Luong et al., 2017) model to generate sentences that provide explanations as to why the particular sentence is against common sense. 2 Dataset Each sub task is associated with a separate data set for its processing. In sub task B the input data is separated as a particular statement with three other sentence options amongst which one of them is the most favourable reason to why the statement is not common sense valid. In sub task C the input data is separated as a particular statement with three supporting statements to why the particular statement is against common sense. The train, trial, dev and test set instances for both tasks are of sizes 10000, 2021, …

Content truncated to 500 words