Optimized Hindi Negation Detection Using a Hybrid Rule-Based and BERT Model

A significant aspect of language is the ability to negate, which is also an essential part of the process of extracting information from written text. The fields of finance, healthcare, and education all place a significant amount of weight on this particular subtask. Throughout the years, a variety of strategies have been analyzed in order to find a solution to this issue. Negation Detection and Scope Resolution is crucial for Sentiment Analysis and Chatbots, as it affects the meaning and relevance of text. Much of the research on this problem has focused on medical data and financial data where negation cues are prevalent (e.g., medical reports). Negation is easy for people to recognize, but computer systems struggle to identify the phrases that signal it and define its extent. Early systems employed rules and heuristics to handle the problem of discovering negation cues and their contexts. However, human language is complicated, therefore these methods were inaccurate. After exploring several strategies, Deep Learning-based approaches emerged as promising. CNNs, particularly with BERT and, more recently, BiLSTMs, rule-based systems, machine learning classifiers, and conditional random field models are all examples of such techniques. The purpose of this review paper is to discuss the topic of negation detection in Hindi through the use of natural language processing, specifically the BERT approach and the hybrid rule-based approach. Additionally, a literature survey of over 15 papers will be discussed, which will include objectives, the research contribution, as well as its limitations and future work. Additionally, the importance of negation detection in hind language will be discussed, as well as how the BERT model is useful for negation detection and the challenges that should be addressed.