Identifying Source-Language Dialects in Translation

In this paper, we aim to explore the degree which translated texts preserve linguistic features of dialectal varieties. We release a dataset augmented annotations Proceedings European Parliament that cover speaker information, and analyze different classes written English covering native varieties from British Isles. Our analyses discuss discriminatory between reveal words whose usage differs same language. perform classification experiments show automatically distinguishing is possible with high accuracy, even after translation, propose new explainability method based on embedding alignments in order specific differences dialects at level vocabulary.