The Reliance Hypothesis: Chain-of-Thought Faithfulness Rises as Language Resource Level Falls
In brief
- CoT reliability improves in low-resource languages for knowledge-heavy tasks.
- Models depend more on reasoning when they cannot recall the answer.
- The effect is observed across four models and multiple reliability metrics.
A study examines chain-of-thought (CoT) reliability across eight languages with varying resource levels. Models rely more on reasoning in low-resource languages, particularly for knowledge-intensive tasks, but not for pure reasoning tasks.
This summary gives you the essentials. The full article, its figures and quotes are on cs.AI updates on arXiv.org.
Read the original articleAI-generated summary of the source, checked by rules. Report an error