Cognitive Science Colloquium
All meetings take place on Thursdays, 3.30-5.30 pm in Bioscience Research Building 1103.
February 4 Thomas Carlson (Psychology, Maryland)
Title: "Seeing Myself in the Dark: Interactions between Vision and Proprioception Reveal an Extended Sense of 'me'."
Abstract: Effectively interacting with our environment requires a multimodal sense of the body, the putative body schema. Extending this sense of the body beyond the physical body can lead to improvements in efficiency, as evidenced by salmon and trout exploiting vortices in rivers, and new capabilities, exemplified by tool use in humans. The notion of an extended sense of the body has been the topic of philosophical discussion for more than a century. Only recently has it been subject to empirical testing. In this talk, I will present a series of experiments that examine if, and under what conditions, our body sense can be extended. The first set of experiments investigated the cognitive factors that support the integration of visual and proprioceptive information into the body schema. Our experiments show that that a sense of perceived ownership is critical. A second set of experiments explicitly tested if external objects can be integrated into the body schema. We found that without any experience or training objects can be quickly incorporated into the body schema, but also that these extensions have limitations. In particular, we found that second-order extensions of the body schema, using tools to interact with objects, were not possible. This suggests that physical contact plays a role in assimilating objects. A third series of experiments examined the relationship between the body schema and self-awareness. These experiments examined the possibility that passing the so-called mirror test a test for self-awareness that only a few species with higher-order cognitive abilities have passed (e.g. humans and dolphins) may be mediated by an extended body sense. Our experiments found support of this notion by showing that humans maintain a bodily connection to our reflections in the mirror. This result in conjunction with our findings that perceived ownership mediates the integration of visual and proprioceptive information suggests that success in passing the mirror test may be derived from a sense of ownership over our reflection in the mirror.
February 18 Renee Baillargeon (Psychology, Illinois at Urbana-Champaign)
Title: "Making Sense of Others Actions: Psychological Reasoning in Infancy."
Abstract: Beginning early in the first year of life, infants attempt to make sense of others intentional actions. Although the nature and development of infants psychological reasoning (or theory of mind, as it is sometimes called) remain the subjects of intense controversy, the notion that infants already possess some understanding of others actions is becoming widely accepted. In much of the research on this topic, infants watch simple scenes in which an agent acts on objects (e.g., a person reaches consistently for chocolates as opposed to carrots). Investigators examine what mental states infants attribute to the agent, and how they use these mental states to interpret and predict the agents actions. Results indicate that infants in the first year of life are able to attribute at least two kinds of mental states to an agent: motivational states (e.g., goals, dispositions), which specify the agents motivation in the scene, and reality-congruent informational states (e.g., knowledge, ignorance), which specify what accurate information the agent possesses or lacks about the scene. Over the past few years, experiments on reality-incongruent informational states have focused on the question of whether infants also realize that an agent can hold false or pretend beliefs about a scene. In my talk, I will review evidence that, when attempting to make sense of an agents actions in a simple scene, infants take into account the agent's motivational, reality-congruent informational, and reality-incongruent informational states.
March 4 Victoria Southgate (Psychology, Birkbeck, UK) -- POSTPONED TO APRIL 29
March 25 Janet Metcalfe (Psychology, Columbia)
Title: "Metacognition of Agency: Brain Monitoring of Control"
Abstract: The question addressed by this research is how does a person know that he or she is the agent? According to many, having the metaknowledge that one is, oneself, the actor (or the thinker) is simply a given. It is direct knowledge. This is what Ryle called the Official Doctrine. Descartes believed that this was, in fact, the only sure knowledge and based his entire epistemological system upon it. In contrast to this view, I will take the position that people's knowledge about their own agency is inferential , like other metacognitive judgments. All researchers in metacognition agree that all metacognitive judgments that have been studied to date are based on cues. So, we can look for cues that feed into the metacognitive judgments of agency, investigating both their behavioral profiles and their brain prints. Additionally, given that metacognitions of agency are based on cues, we also investigate individual differences, age differences and mistakes that people make in their knowledge of when they are the agent.
April 8 Lisa Feigenson (Psychological and Brain Sciences, Johns Hopkins)
Title: "Constraints and Flexibility as Hallmarks of Early Object Representations"
Abstract: A central focus of cognitive psychology is: What is the format of the mental representations that store information, and what computations can be performed over those representations? Here I explore answers to these questions for the case of the concept "individual," drawing from work on object cognition, working memory, and cognitive development. I build a case that representations of individuals (e.g., object) are in some ways strikingly constrained by the architecture of working memory. At the same time, the variety of computations that can be performed over these representations allow for great representational flexibility. For example, although the canonical individual may be an object, multiple objects may be bound into sets which then function as an individual for the purposes of attention and memory. This is true starting in infancy. The emerging portrait is one of continuity across the lifespan, with both infants and adults making use of a representational system that is at once both remarkably limited and surprisingly flexible.
April 22 Josh Tenenbaum (Psychology, MIT)
Title: "How to Grow a Mind: Statistics, Structure, and Abstraction"
Abstract: How do humans come to know so much about the world from so little data? From sparse and noisy fragments of experience, we draw generalizations that successfully guide our actions in future situations and tasks we have never faced before. Even young children can infer the meanings of words, the hidden properties of objects, or the existence of causal relations from just one or a few relevant observations -- far outstripping the capabilities of conventional learning machines. How do they do it? And how can we bring machines closer to these human-like learning abilities?
These questions are instances of the classic "problem of induction", and they have a classic answer. Rich sources of prior knowledge must be available to constrain human learners' hypothesis spaces and enable meaningful generalizations. In this talk I will describe some attempts to capture and study this insight more formally in computational theories of cognition. I will argue that to understand the nature, use and origins of knowledge guiding human inferences, we need to bring together several ideas that are familiar to cognitive scientists but are traditionally viewed more as competitors rather than as complementary pieces of the puzzle. These ideas are statistics, structure, and abstraction. By "statistics", I mean specifically Bayesian inference in probabilistic generative models. The hypothesis spaces and priors of Bayesian learning provide a natural language in which to describe the inductive biases that guide human generalization. Formalizing these inductive biases, these knowledge-based Bayesian priors, will require us to define probabilistic models over structured symbolic representations such as graphs, grammars, predicate logic, schemas, theories or programs. To explain the origins of these priors, we will adopt a hierarchical Bayesian framework. Inference occurs in parallel at multiple levels of abstraction, allowing the knowledge that serves as background or inductive constraints for one level of learning to itself be generalized from experience, via simultaneous inferences occurring at higher levels of the hierarchy.
More specifically, this talk will focus on models of learning and "learning to learn" about categories, word meanings and causal relations. I will show in each of these settings how human learners might balance the need for strongly constraining inductive biases -- necessary for rapid generalization -- with the flexibility to adapt to the structure of new environments, learning new inductive biases for which our minds could not have been pre-programmed. I will also discuss briefly how this approach extends to richer forms of knowledge, such as intuitive psychological and social inference, physical reasoning and natural number.
April 29 Victoria Southgate (Psychology, Birkbeck, UK)
Title: "Predicting Actions and Outcomes in Infancy."
Abstract: The ability to form on-line predictions about the likely outcomes of ongoing events is a prerequisite for a number of social cognitive abilities, including coordinating one's actions with others, the basis of our human ability to cooperate with one another. Although cooperation and collaboration are hypothesized to be defining features of human ontogeny, it was unclear whether human infants had the prerequisite ability to generate predictions concerning others' actions. In this talk, I will present both behavioral and neuroimaging data from infants suggesting that indeed this ability is present in the first year of life, and I will discuss potential cognitive and neural mechanisms that may be recruited in support of this ability.
May 6 Jeff Heinz (Linguistics and Cognitive Science, University of Delaware)
Title: "Modular Phonological Learning."
Abstract: Debates between domain-specific (Gallistel 1999) and domain-general (Marescal et. al 2007, Spencer et. al 2009) learning strategies often focus on the possible existence of a module specialized for speech and language (Liberman 1996). This talk argues that within the domain of language itself there are specialized learning mechanisms. In particular, formal language-theoretic and formal learning-theoretic analyses of the sound patterns in the world's languages imply three different learning mechanisms. Additionally, a broad interdisciplinary research program for investigating this hypothesis is outlined, and recent relevant results are discussed.