Colin Dawson

From domain-generality to domain-sensitivity: 4-Month-olds learn an abstract repetition rule in music that 7-month-olds do not

Learning must be constrained for it to lead to productive generalizations. Although biology is undoubtedly an important source of constraints, prior experience may be another, leading learners to represent input in ways that are more conducive to some generalizations than others, and/or to up- and down-weight features when entertaining generalizations. In two experiments, 4-month-old and 7-month-old infants were familiarized with sequences of musical chords or tones adhering either to an AAB pattern or an ABA pattern. In both cases, the 4-month-olds learned the generalization, but the 7-month-olds did not. The success of the 4-month-olds appears to contradict an account that this type of pattern learning is the provenance of a language-specific rule-learning module. It is not yet clear what drives the age-related change, but plausible candidates include differential experience with language and music, as well as interactions between general cognitive development and stimulus complexity.

Surprise! Infants consider possible bases of generalization for a single input example

Infants have been shown to generalize from a small number of input examples. However, existing studies allow two possible means of generalization. One is via a process of noting similarities shared by several examples. Alternatively, generalization may reflect an implicit desire to explain the input. The latter view suggests that generalization might occur when even a single input example is surprising, given the learners current model of the domain. To test the possibility that infants are able to generalize based on a single example, we familiarized 9-month-olds with a single three-syllable input example that contained either one surprising feature (syllable repetition, Experiment 1) or two features (repetition and a rare syllable, Experiment 2). In both experiments, infants generalized only to new strings that maintained all of the surprising features from familiarization. This research suggests that surprise can promote very rapid generalization.

Bayesian 3D Tracking from Monocular Video

We develop a Bayesian modeling approach for tracking people in 3D from monocular video with unknown cameras. Modeling in 3D provides natural explanations for occlusions and smoothness discontinuities that result from projection, and allows priors on velocity and smoothness to be grounded in physical quantities: meters and seconds vs. pixels and frames. We pose the problem in the context of data association, in which observations are assigned to tracks. A correct application of Bayesian inference to multi-target tracking must address the fact that the models dimension changes as tracks are added or removed, and thus, posterior densities of different hypotheses are not comparable. We address this by marginalizing out the trajectory parameters so the resulting posterior over data associations has constant dimension. This is made tractable by using (a) Gaussian process priors for smooth trajectories and (b) approximately Gaussian likelihood functions. Our approach provides a principled method for incorporating multiple sources of evidence, we present results using both optical flow and object detector outputs. Results are comparable to recent work on 3D tracking and, unlike others, our method requires no pre-calibrated cameras.

Auditory masked priming in Maltese spoken word recognition

This study investigated lexical access in Maltese, an understudied Semitic language. We report here on a series of four lexical decision experiments designed to test the hypothesis that the consonantal root and the word pattern may each prime lexical access in Maltese. Priming of morphologically related forms is generally taken as evidence consistent with morphological decomposition in processing. Here, we used two speech priming techniques: auditory priming in which primes and targets were equally audible, and auditory masked priming in which primes are masked from conscious perception by volume-attenuation and compression. Our results show priming of targets by forms sharing a consonantal root, but not by forms sharing a word pattern.

Can rational models be good accounts of developmental change? The case of language development at two time scales.

Rational models of human perception and cognition have allowed researchers new ways to look at learning and the ability to make inferences from data. But how good are such models at accounting for developmental change? In this chapter, we address this question in the domain of language development, focusing on the speed with which developmental change takes place, and classifying different types of language development as either fast or slow. From the pattern of fast and slow development observed, we hypothesize that rational learning processes are generally well suited for handling fast processes over small amounts of input data. In contrast, we suggest that associative learning processes are generally better suited to slow development, in which learners accumulate information about what is typical of their language over time. Finally, although one system may be dominant for a particular component of language learning, we speculate that both systems frequently interact, with the associative system providing a source of emergent hypotheses to be evaluated by the rational system and the rational system serving to highlight which aspects of the learners input need to be processed in greater depth by the associative system.