arxiv.org
Tokenization and Representation Biases in Multilingual Models on Dialectal NLP Tasks
Dialectal data are characterized by linguistic variation that appears small to humans but has a significant impact on the performance of models. This dialect gap has been related to various factors (e...
