Under greedy decoding, four large tool-using models retain 71-73% of their action policy across languages after normalizing by self-reproducibility, a regularity that breaks below roughly 10B parameters.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
Actions Speak Louder than Words: Measuring Cross-Lingual Policy Retention in Tool-Using Agents
Under greedy decoding, four large tool-using models retain 71-73% of their action policy across languages after normalizing by self-reproducibility, a regularity that breaks below roughly 10B parameters.