back to

Seminar: 12/7 - ICBS mini-conference

11:00 to 1:00 PM      at:  5101 Tolman Hall

Five short talks about research in the cognitive and brain sciences on campus

11:10-11:30     John Case

Task specificity, temporal dynamics, and network properties within the prefrontal cortex during lexical retrieval and response selection

The prefrontal cortex, including the inferior frontal gyrus (Broca’s area) and the middle frontal gyrus (DLPFC), has been shown to play a key role in lexical response selection, including phoneme and word repetition, word generation, and lexical decision. FMRI and EEG research indicates that prefrontal activation increases with task difficulty and the level of semantic processing required during the task. However, the mechanisms and temporal dynamics of prefrontal engagement during lexical retrieval and response selection remain largely unexplored and highly debated. We examined when and how patterns of prefrontal activation vary as a function of difficulty and lexical requirements of the task using intracranial electrocorticography (ECoG) - a technique with unique temporal and spatial resolution. ECoG signals were recorded in 4 subjects with intractable epilepsy (3 with left and 1 with right fronto-temporal grids). Participants completed two tasks: word repetition (where they were asked to repeat aurally presented words of varying length) and word generation (where they had to provide a semantic antonym to each word). Increases in spectral power in the broadband gamma range (70-150 Hz) were used to index functional activation of the underlying cortex. The word generation task was more difficult, as evidenced by fewer completed trials and increased reaction times (p < .01). Word repetition elicited greater gamma increases during the word perception stage (150-750 ms) over the posterior STG and perisylvian cortex (p < .05), suggesting greater focus on the acoustic properties of the stimulus. During the same period, electrodes within Broca's area and the DLPFC exhibited a decrease in gamma power, which was greater during word generation, indicating deactivation of these areas during stimulus perception. At most sites, this initial decrease was followed by an increase in gamma power, which was greater for word generation (p < .001). Changes in gamma power were observed as early as 150 ms from stimulus onset and often lasted until the response, suggesting the specific role of these areas in lexical search and response selection. Trial-by-trial cross-correlation analyses performed on the averaged amplitudes of gamma responses revealed a coherent network in the prefrontal areas. Importantly, the extent and the strength of this network were substantially diminished during word repetition. Taken together, these findings indicate that Broca’s area and the DLPFC engage early, stay on-line until the appropriate response is selected, and act in a more coordinated fashion when the task difficulty is increased and lexical search and retrieval are required.

11:30-11:50     Matthew Crossley

Unlearning in procedural categorization

Most efforts to erase or modify (i.e., unlearn) undesirable behaviors have met with little success. This is clearly evidenced by the huge amounts of money invested every year in prisons, psychotherapy clinics, drug treatment facilities, and schools - all of which suffer from high recidivism rates. Unfortunately, little is known about why many behaviors are so difficult to unlearn. Recently, however, we proposed a neurobiological theory that addressed this question for one important class of behaviors - specifically, those obtained via procedural learning (Crossley, Maddox & Ashby, 2012, J. of Experimental Psychology: General). We then used this theory to derive an effective intervention protocol and reported empirical results that showed promising initial signs of unlearning. Briefly, this theory assumed that dopamine-dependent reinforcement learning at key synapses in the striatal microcircuit is modulated by action-outcome contingency. This talk will elaborate on these empirical and theoretical findings.

11:50-12:10     Erv Hafter

Prosthetics that are redundant with natural perceptual processing may be useful simply because they reduce the cognitive load of perceiving

There is a long history of the use of Noise-Reduction (NR) algorithms engineered to improve speech understand in noise in hearing aids. Reports from listeners with NR claim that they are valuable and make them happy. This is a good thing but, alas, the measured intelligibility of the speech does not improve.  As a solution to this conundrum, we viewed NR more as a cognitive crutch that does for the perceiver what his or her auditory nervous system could do. Evidence in support of this hypothesis comes from a study with various dual tasks.

12:10-12:30     Joseph Williams

Why does generating explanations help learning? Bringing basic research and online education together.

Generating explanations is widely recognized to be beneficial for learning, but little experimental work has probed the underlying mechanisms. In contexts as varied as explaining category membership, people's behavior, and statistical relations, lab experiments provide evidence for a Subsumptive Constraints account: Explaining "why?" a particular fact is true drives learners to understand how it can be subsumed as an instance of a broader principle or generalization. Rather than providing a general boost to engagement, explaining "why?" exerts selective constraints which promote discovery of subtle principles and the use of relevant prior knowledge, but can impair learning when patterns are unreliable. These effects of explanation are further examined in the context of students doing homework exercises, an ecologically valid but precisely controlled environment.

12:30-12:50     Lotfi Zadeh

Toward Human-Computer Dialogue (HCD)

Achievement of human-computer dialogue has long been one of the principal achievements of AI. There is a close connection between HCD and the Turing test. Achievement of HCD is still a distant objective. There is a reason. Traditional approaches are based on the classical, Aristotelian, bivalent logic. Bivalent logic is intolerant of imprecision and partiality of truth. Imprecision and partiality of truth are basic characteristics of human language.

A concept which is needed to move toward realization of HCD is that of a restriction. Informally, a restriction is an answer to a question of the form: What is the value of a variable, X? For example, how long will it take me to drive from Berkeley to San Francisco airport? An answer such as usually about 1 hour is a restriction. A restriction is a carrier of information. A proposition drawn from a natural language may be viewed as a restriction. Fundamentally, a dialogue may be viewed as an exchange of restrictions. What is important about restrictions is that they can be used as objects of reasoning and computation.