When Opposite Premises Translate Identically: A Source-Conditioned Diagnostic for Multilingual Commonsense Items

Key points

• Automatic translation scores can rank source-target pairs, but an acceptance decision alone cannot certify a counterfactual benchmark field.1
• We test a source-contrast diagnostic for ACCORD premises: reverse the English polarity, translate both arms independently, count identical target outputs, and score each arm against its own source.1
• Across 71 matched question identities at two containing-item depths, five target languages, and three translation checkpoints, Swahili has 63, 113, and 67 exact collisions among 710 pairs per checkpoint.1
• Each collision proves at least one unfaithful translation, but does not identify its arm.1

Sources (1)

• [1]When Opposite Premises Translate Identically: A Source-Conditioned Diagnostic for Multilingual Commonsense Items
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 17, 03:44 AM

  Multilingual reasoning comparisons depend on translated premises preserving the relations in their English sources.

  Automatic translation scores can rank source-target pairs, but an acceptance decision alone cannot certify a counterfactual benchmark field.

Extractive summary: sentences quoted from the sources.