Translational_NLI

Natural Language Inference (NLI) is a crucial task in natural language processing, involving the determination of the logical relationship between two given statements - a premise and a hypothesis. In this context, the FEVER dataset, initially designed for Fact Extraction and Verification tasks, serves as a valuable resource for NLI-based research. The dataset consists of pairs of premises and hypotheses in English, which are then translated into French for multilingual evaluation. These pairs are annotated with labels indicating whether the statements exhibit an entailment, contradiction, or neutral relationship. This project explores the application of NLI models in a multilingual setting by evaluating the performance of translated English premise and hypothesis pairs. The project uses state-of-the-art machine translation techniques to convert English pairs into French. The translated French pairs undergo an NLI classification task to predict logical relationships between the statements.The NLI classifications of the translated pairs are compared with actual French premise and hypothesis pairs to assess the accuracy and reliability of the machine translation process. Through this comprehensive evaluation, valuable insights are gained into the cross-lingual transferability of NLI models, as they encounter diverse linguistic contexts. Ensuring the quality of the translation process is crucial in preserving the integrity of the NLI task outcomes. The outcomes of this study significantly contribute to the advancement of multilingual NLI research, guiding the development of robust NLI models capable of effectively handling multiple languages. Moreover, the project’s insights have practical implications in various applications, such as cross-lingual information retrieval and automated fact-checking, enabling more sophisticated understanding and processing of multilingual data in real-world scenarios.