What controls gain in gain control? Mismatch negativity (MMN), priors and system biases.

Repetitious patterns enable the auditory system to form prediction models specifying the most likely characteristics of subsequent sounds. Pattern deviations elicit mismatch negativity (MMN), the amplitude of which is modulated by the size of the deviation and confidence in the model. Todd et al. (Neuropsychologia 49:3399-3405, 2011; J Neurophysiol 109:99-105, 2013) demonstrated that a multi-timescale sequence reveals a bias that profoundly distorts the impact of local sound statistics on the MMN amplitude. Two sounds alternate roles as repetitious "standard" and rare "deviant" rapidly (every 0.8 min) or slowly (every 2.4 min). The bias manifests as larger MMN to the sound first encountered as deviant in slow compared to fast changing sequences, but no difference for the sound first encountered as a standard. We propose that the bias is due to how Bayesian priors shape filters of sound relevance. By examining the time-course of change in MMN amplitude we show that the bias manifests immediately after roles change but rapidly disappears thereafter. The bias was reflected in the response to deviant sounds only (not in response to standards), consistent with precision estimates extracted from second order patterns modulating gain differentially for the two sounds. Evoked responses to deviants suggest that pattern extraction and reactivation of priors can operate over tens of minutes or longer. Both MMN and deviant responses establish that: (1) priors are defined by the most proximally encountered probability distribution when one exists but; (2) when no prior exists, one is instantiated by sequence onset characteristics; and (3) priors require context interruption to be updated.

Initial Uncertainty Impacts Statistical Learning in Sound Sequence Processing.

This paper features two studies confirming a lasting impact of first learning on how subsequent experience is weighted in early relevance-filtering processes. In both studies participants were exposed to sequences of sound that contained a regular pattern on two different timescales. Regular patterning in sound is readily detected by the auditory system and used to form "prediction models" that define the most likely properties of sound to be encountered in a given context. The presence and strength of these prediction models is inferred from changes in automatically elicited components of auditory evoked potentials. Both studies employed sound sequences that contained both a local and longer-term pattern. The local pattern was defined by a regular repeating pure tone occasionally interrupted by a rare deviating tone (p=0.125) that was physically different (a 30msvs. 60ms duration difference in one condition and a 1000Hz vs. 1500Hz frequency difference in the other). The longer-term pattern was defined by the rate at which the two tones alternated probabilities (i.e., the tone that was first rare became common and the tone that was first common became rare). There was no task related to the tones and participants were asked to ignore them while focussing attention on a movie with subtitles. Auditory-evoked potentials revealed long lasting modulatory influences based on whether the tone was initially encountered as rare and unpredictable or common and predictable. The results are interpreted as evidence that probability (or indeed predictability) assigns a differential information-value to the two tones that in turn affects the extent to which prediction models are updated and imposed. These effects are exposed for both common and rare occurrences of the tones. The studies contribute to a body of work that reveals that probabilistic information is not faithfully represented in these early evoked potentials and instead exposes that predictability (or conversely uncertainty) may trigger value-based learning modulations even in task-irrelevant incidental learning.

Initial uncertainty impacts statistical learning in sound sequence processing.

This paper features two studies confirming a lasting impact of first learning on how subsequent experience is weighted in early relevance-filtering processes. In both studies participants were exposed to sequences of sound that contained a regular pattern on two different timescales. Regular patterning in sound is readily detected by the auditory system and used to form "prediction models" that define the most likely properties of sound to be encountered in a given context. The presence and strength of these prediction models is inferred from changes in automatically elicited components of auditory evoked potentials. Both studies employed sound sequences that contained both a local and longer-term pattern. The local pattern was defined by a regular repeating pure tone occasionally interrupted by a rare deviating tone (p=0.125) that was physically different (a 30msvs. 60ms duration difference in one condition and a 1000Hz vs. 1500Hz frequency difference in the other). The longer-term pattern was defined by the rate at which the two tones alternated probabilities (i.e., the tone that was first rare became common and the tone that was first common became rare). There was no task related to the tones and participants were asked to ignore them while focussing attention on a movie with subtitles. Auditory-evoked potentials revealed long lasting modulatory influences based on whether the tone was initially encountered as rare and unpredictable or common and predictable. The results are interpreted as evidence that probability (or indeed predictability) assigns a differential information-value to the two tones that in turn affects the extent to which prediction models are updated and imposed. These effects are exposed for both common and rare occurrences of the tones. The studies contribute to a body of work that reveals that probabilistic information is not faithfully represented in these early evoked potentials and instead exposes that predictability (or conversely uncertainty) may trigger value-based learning modulations even in task-irrelevant incidental learning.

Altering the primacy bias--how does a prior task affect mismatch negativity?

The role in which two tones are first encountered in an unattended oddball sequence affects how deviance detection, reflected by mismatch negativity, treats them later when the roles reverse: a "primacy bias." We tested whether this effect is modulated by previous behavioral relevance assigned to the two tones. To this end, sequences in which the roles of the two tones alternated were preceded by a go/no-go task in which tones were presented with equal probability. Half of the participants were asked to respond to the short sounds, the other half to long sounds. Primacy bias was initially abolished but returned dependent upon the go-stimulus that the participant was assigned. Results demonstrate a long-term impact of prior learning on deviance detection, and that even when prior importance/equivalence is learned, the bias ultimately returns. Results are discussed in terms of persistent go-stimulus specific changes in responsiveness to sound.

Mismatch negativity (MMN) to pitch change is susceptible to order-dependent bias.

Pattern learning facilitates prediction about upcoming events. Within the auditory system such predictions can be studied by examining effects on a component of the auditory-evoked potential known as mismatch negativity (MMN). MMN is elicited when sound does not conform to the characteristics inferred from statistical probabilities derived from the recent past. Stable patterning in sequences elevates confidence in automatically generated perceptual inferences about what sound should come next and when. MMN amplitude should be larger when sequence is highly stable compared to when it is more volatile. This expectation has been tested using a multi-timescale paradigm. In this study, two sounds of different duration alternate roles as a predictable repetitive "standard" and rare MMN-eliciting "deviation." The paradigm consists of sound sequences that differ in the rate at which the roles of two tones alternate, varying from slowly changing (high stability) to rapidly alternating (low stability). Previous studies using this paradigm discovered a "primacy bias" affecting how stability in patterning impacts MMN amplitude. The primacy bias refers to the observation that the effect of longer-term stability within sequences only appears to impact MMN to the sound first encountered as deviant (the sound that is rare when the sequence commences). This study determines whether this order-driven bias generalizes to sequences that contain two tones differing in pitch. By manipulating (within-subjects) the order in which sounds are encountered as deviants the data demonstrate the two defining characteristics of primacy bias: (1) sequence stability only ever impacts MMN amplitude to the first-deviant sound; and (2) within higher stability sequences, MMN is significantly larger when a sound is the first compared to when it is the second deviant. The results are consistent with a general order-driven bias exerting modulating effects on MMN amplitude over a longer timescale.

Implementing conditional inference in the auditory system: what matters?

In prior studies, we have used a conditional linkage between rare deviations in a regular sound pattern to determine if the auditory system can use the first deviation to anticipate the probable features of the second deviation (i.e., make a conditional inference). This study was designed to test two hypotheses about why the mismatch negativity (MMN) to a duration deviant sound seems more susceptible to conditional inference effects. The MMNs to duration and frequency glide deviant sounds were significantly smaller when their occurrence was conditionally linked to the identity of a prior deviant as opposed to when they occurred randomly in a sequence. Results provide support for the learned conditional inference interpretation of reduced MMN size to linked deviants. We discuss alternate explanations and conclude that conditional inference studies could provide insight into the dynamics of probability-based prediction in the auditory system.