Han, Chao2023-05-152023-05-15https://udspace.udel.edu/handle/19716/32735Mismatch Negativity (MMN) studies have utilized the varying-standard oddball paradigm to investigate the effect of linguistic category on speech perception. But the evidence is missing for an MMN driven by a within-category acoustic contrast between standards and deviants. With three experiments, the current dissertation asks whether the memory trace in the varying-standard paradigms retains gradient information that could lead to an MMN response. ☐ Experiment 1 looked for an MMN response using the varying-standard paradigm without a categorical contrast between standards and deviants. The standards were realized by [tæ] with voice onset times (VOTs) of 42, 48, and 55ms VOT [tæ]s, and the deviants were realized by [tæ] with a 119ms VOT. The MMN was computed as the difference in brain response to deviants minus the same stimulus in a roving-standard control condition. A within-category MMN was observed, suggesting that the memory trace in the varying-standard paradigm must contain gradient information. ☐ Experiment 2 asked what drove the within-category MMN in Experiment 1. The observed within-category MMN could be due to the deviant VOT contrasting to the phonetic knowledge retrieved from long-term memory. Alternatively, it could be driven by a contrast between the deviant VOT and a statistical summary based on the presented standard VOTs. In that case, the MMN magnitude should be modulated by the statistical structure of the presented stimuli (Garrido, Sahani, & Dolan, 2013). Experiment 2 presented one group with a 128ms VOT [tæ] deviant embedded in a normal distribution of varying standards with a mean VOT of 64ms and a “wide” standard deviation of 15ms. It presented a second group with the same deviant and mean standard VOTs but with a “narrow” standard deviation of 5ms. The result replicated Experiment 1 with a within-category MMN, but no difference between the two standard deviation groups was observed. The lack of difference in the MMN magnitude suggests that a statistical summary of the presented stimuli could not explain the within-category MMN obtained in Experiment 1. ☐ As the conclusion of Experiment 2 was based on a null result, Experiment 3 was conducted to provide positive evidence for a statistical summary of the presented stimuli. Experiment 3 swapped the standards and deviants used in Experiment 2, such that the standard VOTs formed a distribution with a mean VOT of 128ms while the deviant VOT was 64ms. If the varying standards activate phonetic knowledge as a memory trace, the deviant VOT should not lead to an MMN, as the deviant VOT corresponds to the most typical phonetic realization of /tæ/. On the other hand, if the brain computes a statistical summary of standard VOTs, the acoustic difference between the standard and deviant VOT values should lead to an MMN response. The result was that the deviant VOT elicited a robust MMN, indexing the brain’s sensitivity to the statistical structure of the presented stimuli. ☐ The results of the three experiments confirmed that a speech MMN could be driven by a within-category difference other than a categorical contrast. This finding suggests that the memory representation in the varying-standard paradigm retains gradient information along with a category representation. Furthermore, the gradient information comes from the statistical summary based on the acoustic properties of the presented stimuli. The present work thus extends the prior research emphasizing the role episodic memory played in speech perception.Mismatch negativitySpeech perceptionAcoustic propertiesMemory representationEpisodic memoryThesis1379048721https://doi.org/10.58088/rs7t-my592023-03-22en