Comparative Pedagogy of Sentiment Analysis in Digital Linguistics

The article introduces an experiment conducted during classes on text tonality as part of a Master’s Degree course of Computational Linguistics. These classes develop competencies in vocabulary methods, ready-made software libraries, and neural network language models. The experiment involved 16 first-year Digital Linguistics students majoring in Smart Systems in the Humanities at St. Petersburg Polytechnic University in 2024–2025. The didactic design of assignments made it possible to gradually increase the complexity of tasks and combine the acquisition of theoretical knowledge with practical skill development. Module 1 focused on the use of English and Russian tonality dictionaries. The students learned to identify the fundamental limitations of the lexicographic approach, e.g., incomplete dictionaries, as well as difficulties associated with contextual meaning, negations, sarcasm, and cultural specificity. Module 2 involved training a DistilBERT model and working with various datasets. The students reflected on the role of data in prediction quality, as well as on the limitations of modern neural architectures in analyzing complex semantic phenomena. Module 3 involved a group project. The students compared lexicon-based, neural, and library-based (VADER, TextBlob, Flair) approaches for advantages, disadvantages, and applicability. The experiment fostered critical thinking and the ability to make informed decisions when electing tools for specific tasks and available resources. A comprehensive synergy of methods is efficient as it allows students to test both the strengths and the weaknesses of each approach. Integrating sentiment analysis into digital linguistics curricula is a key challenge for modern higher education.