In 2 experiments, eye movements were monitored as participants followed instructions containing temporary syntactic ambiguities (e.g., “Pour the egg in the bowl over the flour”). The authors varied the affordances of task-relevant objects with respect to the action required by the instruction (e.g., whether 1 or both eggs in the visual workspace were in liquid form, allowing them to be poured). The number of candidate objects that could afford the action was found to determine whether listeners initially misinterpreted the ambiguous phrase (“in the bowl”) as specifying a location. The findings indicate that syntactic decisions are guided by the listener’s situation-specific evaluation of how to achieve the behavioral goal of an utterance. As a sentence unfolds in time, the grammatical relationships among its constituents are often temporarily ambiguous. For example, the phrase italicized in (1) may indicate the location where an egg is being poured, or may specify which of several eggs is intended. (1) The baker poured the egg in the bowl... (2) a....while stirring continuously.

We present a computational model that uses intention recognition as a basis for situated word learning. In an initial experiment, the model acquired a lexicon from situated natural language collected from human participants interacting in a virtual game environment. Similar to child language learning, the model learns nouns faster than verbs.

Previous work on early language acquisition has shown that word meanings can be acquired by an associative procedure that maps perceptual experience onto linguistic labels based on cross-situational observation. A new trend termed social-pragmatic theory [27] focuses on the effect of the child’s social-cognitive capacities, such as joint attention and intention reading. In this paper, we argue that statistical and social cues can be seamlessly integrated to facilitate early word learning. To support this idea, we first introduce a statistical learning mechanism that provides a formal account of cross-situational observation. The main part of this paper then presents a unified model that is able to make use of different kinds of social cues, such as joint attention and prosody in maternal speech, in the statistical learning framework. In a computational analysis of infant data, we report the quantitative results of our unified model in computing word-meaning associations, which outperforms the purely statistical learning method. 1

thank members of the UBC Baby Cognition Lab for their help with data collection, and Paul Bloom, Geoff Hall, and Terry Regier for helpful discussion. We owe a particular debt to Liz Bonawitz, for discussions and pilot work on an earlier version of this work.

In two studies, the authors investigated 2- and 3-year-old children’s awareness of the normative structure of conventional games. In the target conditions, an experimenter showed a child how to play a simple rule game. After the child and the experimenter had played for a while, a puppet came (controlled by a 2nd experimenter), asked to join in, and then performed an action that constituted a mistake in the game. In control conditions, the puppet performed the exact same action as in the experimental conditions, but the context was different such that this act did not constitute a mistake. Children’s normative responses to the puppet’s acts (e.g., protest, critique, or teaching) were scored. Both age groups performed more normative responses in the target than in the control conditions, but the 3-year-olds did so on a more explicit level. These studies demonstrate in a particularly strong way that even very young children have some grasp of the normative structure of conventional activities.

Most theories of word learning fall into one of two classes: hypothesis elimination or associationist. We propose a new approach to word learning within a Bayesian framework. Tenenbaum and Xu (2000) presented a Bayesian model of adults learning words for hierarchically structured categories. We report two experiments with 3- and 4-year-old children, providing evidence that the basic principles of Bayesian inference are employed when children acquire new words at different hierarchical levels. Implications for theories of word learning are discussed.

The current article proposes a new theory of infant pointing involving multiple layers of intentionality and shared intentionality. In the context of this theory, evidence is presented for a rich interpretation of prelinguistic communication, that is, one that posits that when 12-month-old infants point for an adult they are in some sense trying to influence her mental states. Moreover, evidence is also presented for a deeply social view in which infant pointing is best understoodFon many levels and in many waysFas depending on uniquely human skills and motivations for cooperation and shared intentionality (e.g., joint intentions and attention with others). Children’s early linguistic skills are built on this already existing platform of prelinguistic communication. Human beings communicate with one another in unique ways. Most obviously, humans communicate with one another linguistically, that is, with socially learned, intersubjectively shared symbols of a type not used by other animal species in their natural forms of communication. But humans also communicate with one another in unique ways gesturally. Many of the most important gestures humans useFfor example,

Performance metrics for machine intelligence (e.g., the Turing test) have traditionally consisted of pass/fail tests. Because the tests devised by psychologists have been aimed at revealing unobservable processes of human cognition, they are similarly capable of revealing how a computer accomplishes a task, not simply its success or failure. Here we propose the adaptation of a set of tests of abilities previously measured in humans to be used as a benchmark for simulation of human cognition. Our premise is that if a machine cannot pass these tests, it is unlikely to be able to engage in the more complex cognition routinely exhibited by animals and humans. If it cannot pass these sorts of tests, it will lack fundamental capabilities underlying such performance. 1.

This thesis describes a self organizing learning algorithm that learns phonemes from acoustic input. For generating acoustic input a highly simplified articulatory model is used and the phonemes are expressed in these articulatory terms. The simplification allows us to concentrate on specific issues like the effect of articulatory effort on learning and sound change without having to deal with rather complex realistic models that only hinder these phenomena and make extensive computational simulations impossible. The thesis is also concerned with methodological issues in general linguistics. The approach taken here is motivated by arguing against structurally inclined ways of looking at linguistic problems that still dominate the field.

There are an infinite number of possible word-to-world pairings in naturalistic learning environments. Previous proposals to solve this mapping problem focus on linguistic, social, and representational constraints at a single moment. This paper investigates a cross-situational learning strategy based on computing distributional statistics across words, across referents, and most importantly across the co-occurrences of these two at multiple moments. We briefly exposed adults to a set of trials containing multiple spoken words and multiple pictures of individual objects with no information about word-picture correspondences within a trial. Nonetheless, subjects learned over trials the word-picture mappings through cross-trial statistical relations. Different learning conditions compared the degree of within-trial reference uncertainty, the number of trials and the length of trials. Overall, the remarkable performances of learners in various learning conditions suggest that they calculate cross-trial statistics with sufficient fidelity and by doing so rapidly learn word-referent pairs even in highly ambiguous learning contexts. Quine (1960) famously presented the core problem for learning word meanings from their co-occurrence with perceived events in the world. He imagined an anthropologist who observes a speaker saying “gavagai ” while pointing in the general direction of a field. The intended referent (rabbit, grass, the field, or rabbit ears, etc.) is indeterminate from this experience. The solution to this indeterminacy problem requires that the learning system be somehow constrained. Research on children’s word learning has concentrated on how this learning might be constrained in a single trial such that the word is correctly mapped to the referent on that trial. This literature suggests attentional (Smith, 2000), social (Baldwin 1993,