UC Irvine

People are highly skilled at extracting socially-relevant information from the movements of others. The primary human movements analyzed here involve movements of the articulatory organs (which produce speech sounds capable of transmitting a wealth of talker-specific characteristics) and movements of the entire body (from which we regularly and rapidly infer others' goals and intentions). In the first study, I examine how acoustic correlates of articulatory kinematics shape perceived \textit{attractiveness} of the voice. I test the hypotheses that spectral and temporal correlates of precise articulation are relevant predictors of vocal attractiveness through the roles of sexual dimorphism and processing fluency accounts of preferences. In a sample of talkers producing vowels in carrier words, I find that a high proportion of variance in voice preference ratings for female talkers can be explained by measures related to the acoustic-phonetic distinctiveness of speech. The next study shifts focus to brain regions encoding representations of actions performed by human bodies. Contemporary models of action observation now posit a special role for the posterior superior temporal sulcus (pSTS) in integrating low-level perceptual cues with top-down influences of attention. This implies that action representations in the pSTS are not immutable but instead are dynamic and context-dependent. Multivariate pattern analysis (MVPA) evaluated how task demands shape the specific information in the pSTS during action observation. The statistical structure of multivariate patterns in the pSTS was found to be highly susceptible to feature-based attention, revealing that the pSTS plays an important intermediary role at the interstices of bottom-up and top-down cues. The last chapter serves as an important guide for researchers seeking to optimize experimental design, data preprocessing and machine learning parameters for rapid event-related MVPA. Here, I evaluate the independent and joint effects of four methodological data processing choices aimed at reducing the effects of trial-, voxel-, scan-, or motion-related noise sources. Two of these choices in particular interacted to produce large increases in classifier performance in cases where there is true signal present, a finding which is consistent across both real and simulated datasets.