Some results of the digital analysis of the semantic similarity of audiovisual translation texts (based on the films directed by Denis Villeneuve)

The article analyses different translation versions of science fiction films directed by Denis Villeneuve with the aim of determining their semantic similarity through NLP analysis. There are four groups of metrics used in text analysis: semantic (cosine similarity and Word Mover’s Distance), lexical (Levenshtein distance and Jaccard coefficient), vector (Euclidean and Manhattan distance) and correlation (Pearson correlation coefficient and Spearman’s rank correlation coefficient). The material of the study is the subtitled translations into Russian of the ‘Dune: Part One’ and the dubbed and subtitled translations of the ‘Blade Runner 2049’. This study applies linguistic methods such as machine learning method, natural language processing (NLP), semantic similarity analysis and sentiment analysis, descriptive method, comparative analysis method and contrastive analysis method. The results of the complex digital analysis of two translation versions of four different fragments of the films show high scores of semantic metrics, while the overall similarity of the texts is closer to the mid-level, which is influenced by lexical metrics. Lexical metrics are also influenced by the type of translation: dubbing may change the lexical component due to phonetic synchronisation; subtitled translation, on the other hand, is closer to the source, but tends to be compressive. The sentiment analysis indicates that the tone of the source text is consistently conveyed in the translations; thus, the translated texts align with the source text in terms of emotionality as well.