How do learners make sense of their intricately structured visual and auditory environments? One important learning mechanism available to both infants and adults is the ability to detect statistical structure within visual and auditory inputs. Despite the scope of this “statistical learning” (SL) ability and the large literature that now surrounds it, the processes underlying SL, and how those processes may differ across modalities and change across development, remain unclear.

The goal of this dissertation was to clarify these mechanistic questions. Paper 1 reviews previous work on visual and auditory SL across development, highlighting the debate concerning the domain-generality of SL. Paper 2 investigates the domain-generality of infants’ SL by employing a visual SL task comparable to previous auditory SL tasks. Findings from four experiments with 8-month-olds suggest that visual SL is constrained compared to auditory SL in infancy. Individual differences in visual SL performances were unrelated to visual short-term memory performances, but were related to overall cognitive ability. These findings suggest that visual SL ability is important for, and constrained by, early cognitive development.

Papers 3 and 4 examine the representations that result from visual SL. I evaluate the ability of two major classes of SL models – statistical models and chunking models - to correctly predict adults’ and infants’ performances on visual SL tasks. Across five experiments, adults and 8-month-old infants discriminated between high- and low-probability visual sequences, providing strong evidence of SL. Critically, adults represented pairs of items embedded within larger sequences, but did not represent so-called “illusory sequences.” These results support the competitive chunking model of Servan-Schreiber and Anderson (1990), and suggest that adults represented visual sequences in terms of hierarchical levels of chunks. In contrast, 8-month-olds did not show evidence of representing embedded or illusory sequences, suggesting that infants represented visual sequences in terms of only the highest level of chunks. Together, these studies not only suggest that the representations that result from visual SL are best captured by chunking models, but also that the type of chunking learners engage in may change between infancy and adulthood.

Learning to understand others’ emotions from an early age is crucial for long-term social-cognitive development. However, much remains unknown regarding potential mechanisms behind the development of early emotion understanding. Some theoretical work has proposed that language may facilitate the development of emotion understanding (e.g., Barrett, 2017), while other theoretical work argues that language should not be crucial for such development (e.g., Ekman & Cordaro, 2011). Thus, to better understand how children learn to understand emotions, what may explain some of the large individual differences observed in early emotion understanding, and what theoretical perspective may best explain early emotion understanding development, it is important to further examine the role of language in early emotion understanding development. The present studies use a novel approach to examining the relation between language and emotion understanding across early development using 1) Eye tracking, 2) Corpus analysis, and 3) Live action assessment of emotion category learning methodologies. Study 1 examined whether infant performance on a completely non-linguistic emotion categorization task related to infant language abilities. Participants were 50 infants between 15- and 18-months of age who completed the emotion categorization task using an eye tracker and whose parents provided the infants’ vocabularies. Results from this study revealed that non-linguistic emotion categorization related to vocabulary size for girls, but not for boys, providing some evidence that language may begin to relate to early emotion categorization abilities in infancy. Study 2 investigated young children’s natural language environments and how this environment relates to children’s propensity to talk about emotion words. Nearly 2,000 transcripts of natural interactions between mothers and their 15- to 47-month-old child were drawn from the CHILDES (MacWhinney, 2000) database. Findings from this study revealed that child use of emotion language was significantly predicted by their age and their mother’s use of emotion language, but not by the child or mother’s general language complexity. These results suggest that exposure to emotion language, but not necessarily general language, may be important for children’s developing ability to talk about and understand emotions. Study 3 evaluated whether exposure to specific emotion words would causally influence children’s ability to learn about new, complex emotions. Across two experiments, 72 3-year-old children took part in a pre-test post-test emotion category learning assessment where they were tasked with identifying which face best matched how a character would feel in a scenario. Between pre-test and post-test, children saw a face paired with a scenario and either heard an explicit emotion label, a vague emotion label, or irrelevant information. Results revealed that children learned the face-scenario associations better when given an explicit emotion label versus irrelevant information (Study 1), but vague emotion labels did not provide any advantage over irrelevant information (Study 2). Taken together, these results suggest that explicit emotion labels may be particularly important for helping young children learn about emotions. In sum, these three studies provide some evidence that language may be important for the development of early emotion understanding. However, the strength of this evidence varied by study, suggesting that other factors such as the particular methodology being used, child age, and use of emotion language versus general language, may play important roles in understanding this relation. Altogether, these results present evidence that may be useful for more comprehensively understanding the potential mechanistic role of language in the early development of emotion understanding.

The goal of this dissertation was to examine the scope and development of early visual short-term memory (VSTM) abilities in 6–12-month-old infants. Study 1 investigated individual- and age-related differences across three unique VSTM tasks by examining the effect of increased delay on memory performance. Results suggest longer-term memory processes are quantifiable utilizing 500-1250 ms delays by 8 months with the Change Detection paradigm and spatial- attention cueing processes are quantifiable by 10 months with the Delayed Response paradigm. Performance improved from 6-12 months and longer delays impaired performance. Study 2 examined whether fragile VSTM abilities were responsible for infants’ A-Not-B error, a phenomenon often observed when searching for an object hidden at one of two locations (A and B). Six to 12-month-olds were assessed on a looking A-Not-B task as well as on Delayed Response and Change Detection to assess links between performance. Individual differences in VSTM performance were not predictive of A-Not-B error performance, providing evidence that VSTM accounts may not solely provide explanation of A-Not-B error. Study 3 assessed whether individual differences in 8-month-olds’ neural band power was predictive of their VSTM performance. Infants’ continuous EEG was recording during Change Detection as well as during a baseline resting-state. Results suggest the early alpha band modulates VSTM performance, with decreases in alpha band power shown for higher performers and increases in alpha band power shown for lower performers, relative to baseline. Theta band power was not found to be related to VSTM performance. These outcomes help inform the understanding of infant VSTM and its emergence throughout early development.

This work aimed to understand the limits and capabilities of implicit sequence learning, or internalizing statistically-defined regularities in a stream of stimuli without awareness or intention to learn. Both spatial and nonspatial informational components were manipulated in presented streams of stimuli, and undergraduates learned regularities in both categories of information and applied this knowledge to test trials. Evidence for a unique contribution of a spatial component to visual implicit sequence learning was presented, and changes in performance across the quarter were examined. This work supported a model of implicit learning as a general principle of neural processing (Reber, 2013) and a model of unidimensional and multidimensional systems involved in implicit sequence learning (Keele et al., 2003).

Presented here are a trio of computational projects that, on the surface, have seemingly little in common: (1) proposing a new bioactive conformation of epothilone; (2) creating an amyloid binding model that rationalizes the affinities of a series of DDNP analogs that act as neuroimaging probes; and (3) designing a program for the rapid filtering and ranking of enzyme designs. However, the solution for all of these projects involved focusing on structural distortion energies or penalties. In the case of epothilone, a crystallographic structure, a transferred cross- correlated relaxation NMR structure, and a slew of computational structures all claiming to correctly identify the bioactive conformation of epothilone have already been published. We initially set out to provide corroborating computational evidence for one of these previously proposed conformations, but by correlating distortion energies incurred upon binding and experimental IC50 values, we have discovered a new bioactive conformation of epothilone that fits computational models and experimental data. In a similar fashion, we have shown that the degree of steric strain introduced upon the binding of DDNP to the tau amyloid correlates well with the experimentally determined binding affinities. We have created a model that qualitatively explains binding affinity data and structure-activity relationships for a new set of DDNP analogs. Finally, we have shown that analyzing the distortion from the ideal quantum mechanically computed geometry for an enzyme-catalyzed transition state can be an efficient way of filtering and ranking large numbers of computationally designed enzymes. This led to the creation of two novel computer programs: EDGE (Enzyme Design Geometry Evaluation), that performs these computations in an embarrassingly parallel fashion; and an enzyme design scientific workflow in Kepler that harnesses the power of the UC Grid to perform enzyme design in a massively parallel fashion.