Confidence-Ordering Reversal under Contextual Priors in Neural Decoding.

arXiv:2610.08229v1 Announce Type: new Abstract: Contextual priors improve neural-to-language decoding by reshaping candidate scores. However, confidence is read from the same reshaped scores, so the errors a prior leaves behind can become more confident with n