Dan Schwartz

Inferring Neural Representations of Language

Much of our current understanding of how language is processed in the brain is based on hypotheses generated from observations of language usage and language impairments in normal and abnormal populations, and how particular impairments relate to particular brain area lesions. Observations of lesioned patients have often revealed surprising aspects of the process involved in language production and comprehension, and thus have played a key role in the scientific process. More recently, non-invasive brain imaging has been used to investigate which brain areas are involved in language processing and what types of information correlate with patterns of neural activity. These studies offer many additional insights into the mechanisms underlying language comprehension. However, in most cases brain imaging studies are designed to answer questions about a specific hypothesis of language processing put forward by an experimenter or to compare different hypotheses. In this talk, we consider methods of data analysis to enable us to make naturalistic observations of brain activity that are analogous to naturalistic observations of behavior (language usage). We examine the case where hypotheses about language processing take the form of predefined stimulus representations - for example, representing nouns by the extent to which they exhibit certain characteristics like 'redness', and 'animate'. We argue that the quantitative comparison of these types of hypotheses (stimulus representations) might be misleading in practical cases, and that unsupervised machine learning methods can overcome some of the issues with this comparison, thereby facilitating stimulus representation discovery. This talk will primarily be focused on motivating the use of unsupervised methods and the particular methods we choose to apply, as well as discussion of how the structure discovered by such methods might be interpreted. Feedback is encouraged. The analysis of the data related to the talk is still underway and will not be the focus of the talk.

Emily Braham

Intergenerational Associations in Mathematical Thinking

Educated children and adults have access to at least two different systems for representing and processing numerical information. The first, an approximate number system, can be used to make intuitive estimates and comparisons about the number of items in collections without counting. The second, an exact number system, can be used to label exact quantities (e.g., Arabic numerals) and to perform   exact mathematical operations. In this talk, I will discuss our work on the   intergenerational transmission of these numerical abilities from parents to children.   Furthermore, I will propose follow-up studies using EEG to examine the   electrophysiological underpinnings of numerical processing in both parents and   their children.