A conditional unigram tokenizer trained on parallel data does not improve machine translation but consistently lowers language-model perplexity per byte.
F., Della-Pietra, S
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Conditional Unigram Tokenization with Parallel Data
A conditional unigram tokenizer trained on parallel data does not improve machine translation but consistently lowers language-model perplexity per byte.