1 paper · 1 filter
Poulami Ghosh, Preethi Jyothi
The paper investigates how language models handle non-canonical tokenizations across 27 languages, finding that robustness varies by language and token fragmentation, and shows tha…