From a sociolinguistic perspective, sociolinguistic research needs to be conducted at two levels: natural language and social language. Among them, natural language research requires emotional analysis of texts. Traditional text sentiment analysis is based on rules or statistical methods, which are subjective and one-sided, making it difficult to meet the data processing needs of the big data era. This article uses deep learning methods to construct an English corpus sentiment analysis model based on multiple tasks, combining deep learning algorithms with natural language processing techniques. After addressing issues such as corpus annotation and data preprocessing, this paper employs tasks such as classification and clustering for training and testing, and implements sentiment analysis tasks based on English corpora. The experimental results show that deep learning methods can effectively solve the problems of subjectivity and one sidedness in text sentiment analysis, with an accuracy of up to 97.3%, which is helpful for language research from a sociolinguistic perspective.