TY  - RPRT
TI  - You should evaluate your language model on marginal likelihood over tokenisations
AU  - Kris Cao
AU  - Laura Rimell
PY  - 2021
UR  - https://arxiv.org/abs/2109.02550
ID  - 2109.02550
ER  -