Current Events

Abstract concept formation was once thought to be a uniquely human ability. An increasing variety of species have shown this ability, however. What determines when an individual will focus on general patterns rather than the particulars of specific events? Does the formation of abstract concepts really require greater cognitive sophistication than other types of associative learning? Recent studies of dogs’ abilities to acquire a repeating concept and a same-different concept suggest that flexible use of abstract concepts may depend on repeated experiences and attentional flexibility.

Computational social scientists have expended a considerable amount of effort in recent years to use models based on distributional semantics (most popularly, word embedding models like BERT and word2vec) to capture the meanings of the words and phrases used to describe people. Existing work tends to focus on evaluating measurements for particular phrases on particular dimensions of meaning (e.g. "how can we measure gender stereotypes of occupational identities?"). But phrases about people exist within a broader, more complex, and less well understood space of meanings.  These analyses thus risk biases in what they are measuring, how they are measuring it, and what claims they make based on those findings. In this talk, I will try to convince you that to address these issues, a new measure of semantic similarity, which I call person-centric similarity, is necessary. In the talk, I will introduce person-centric similarity, why it is necessary, and our current empirical and theoretical work underlying it.