Natural language processing is one the most interesting study leading to huge research solutions in the modern era. Multilingual toxic comment classification can be served as a huge benefit to the existing social media life where comments, tweets, etc. can be analyzed when a topic is known to the system. This would help in the prevention of false commenting and better interpretation and analysis of the miscommunication that occurred on different social media platforms over a certain issue. Multilingual toxic comment classification refers to the analysis of a hypothetical sentence proposed given a premise. This classification is divided into three categories are the hypothetical sentence proposed can be either an entailment to the premise, neutral, or contradictory to the known premise statement. Natural language inference is considered as one of the most trending problems under the field of natural language processing which helps to determine how two statements given the premise and hypothesis are related to each other. Thus, the paper proposes different models such as CBOW, ESIM, BiLSTM and fine-tuned XML-RoBERTa model to predict the relationship between two statements. The prediction helps in the determination of whether the given hypothesis is in an entailment, neutral, contradictory relation with the given premise. The paper shows a study over various algorithms that can be used to solve the natural language inference problem. After analysis, the paper also proposes a model that obtained an accuracy of 95.35% with a ROC score of 0.9629 for the entailment relationship, 0.97076 for the neutral relationships, and 0.9797 for the contradictory relationships between the sentence pairs.