Hierarchical timescales of statistical learning revealed by mismatch negativity to auditory pattern deviations

The amplitude of mismatch negativity (MMN) elicited following an unexpected sound reflects a pattern-violation signal that will increase with estimated precision. Precision is inversely related to environmental variance, and should be higher the longer that current regularities have been stable. However, MMN amplitude can be impacted by initial learning such that the relative probability of sounds when first encountered distorts the precision estimates later associated with those sounds. The present study tested the hypothesis that MMN to a pattern violation would be differentially sensitive to both local and global patterning within a sequence, depending on whether the sound was common or rare at sequence onset. Sound sequences consisted of two levels of nested regularity: (1) two tones alternated probabilities as the local standard (p = .875) and deviant (p = .125), and (2) these alternations occurred regularly across four blocks of 2.4 min (stable components) or twelve blocks of 0.8 min (unstable components). Sequences were delivered first in an unstable-stable (“increasing-stability”) and next a stable-unstable (“decreasing-stability”) structure, both inducing a violation to the regular block length at the transition between components. MMN to the tone initially heard as a common repeating standard when later heard as a deviant was not affected by stability of either local (tone probabilities) or global (block length) patterns, reaching equivalent amplitude in all components. In contrast, MMN amplitude to the tone initially heard as deviant was significantly impacted by both local and global pattern stability. MMN amplitude was larger in stable than unstable blocks only if they were heard first (decreasing-stability sequence), and was significantly smaller in both stable and unstable block types after a violation of regular block length. Results are interpreted as local MMN amplitude being “weighted down” by decreased precision in the global structure, but only for the first deviant encountered.