Speaker: George Kachergis
Date and time: 6/27（Wed）15:30~
Place: Collabo 2

Title: Domain-General Mechanisms for Learning Word-Referent Mappings: Empirical and Modeling Evidence

Abstract:
Associative learning has been closely studied in many species, and various effects have been explained 
using only a few basic assumptions and mechanisms. Human language acquisition is fast and is of great interest, 
but is arguably more difficult to capture in the laboratory. Nonetheless, empirical investigations have led researchers 
to theorize a variety of language learning principles and constraints. While there may indeed be language-specific 
learning mechanisms that are distinct from more universal associative learning mechanisms, I seek to explain 
some basic principles of language acquisition using domain-general mechanisms. Using a model and 
several empirical findings, I show how the principles of mutual exclusivity—an assumption of 1-to-1 word-
object mappings, contrast, and other constraints related to fast mapping nouns may result from attention mechanisms 
attributed to associative learning effects such as blocking and highlighting. However, unlike other associative models, 
the model I propose has selective attention directed by simple competing biases for familiar pairs and for uncertain stimuli. 
Given enough time and audience interest, I can address how the model accounts for effects resulting 
from varied pair frequency, contextual diversity, and active learning. I will also hint at the work that 
I will be doing this summer at JAIST, although this work has only just begun.