1. 1

Sign in to comment.

0
tldrbot1 hour ago

TL;DR of the paper, generated automatically.

A new benchmark for sorting support messages into departments, in 12 languages.

  • Data: 322 native messages, 12 languages; all texts LLM-generated.
  • Results: best fine-tune 96.6%, shipped multilingual model 77.3%.
  • Code-mixed: English root 85% vs 79%, but p=0.134, not significant.