ALPHA: Adaptive Mechanisms for Human Language Processing

SFB 378: Resource Adaptive Cognitive Processess

Project Leader: Prof. Dr. Matthew W. Crocker

Project Summary

Research within ALPHA is investigating the extent to which human language comprehension can be explained in terms of adaption, and to develop computational models of the adaptive mechanisms found. In this respect, we are exploring the idea that people resolve structural ambiguities using (near) optimal decision strategies, rather than principles motivated by cognitive limitations. That is, we assume that people exploit their past experience in estimating the best way to resolve ambiguities as they encounter them in immediate context.

One of the central questions we are pursuing is the priority of long-term experience (i.e. frequencies based on our accrued exposure to language) versus short-term based on immediate linguistic and visual context. The research uses large on-line corpora to estimate long-term linguistic biases, and on-line eye-tracking and "visual world" experiments to examine human behavour.

Our computational models need to be designed with human behavior in mind. They need to process language incrementally and robustly, be able to learn from experience, and readily integrate information from diverse information sources. For these reasons, our group is pursuing a class of computational model called experience-based models, such as probabilistic and connectionist systems, in an effort to gain insight into the mechanisms underlying language comprehension. Such insights can in turn inform future experiments and provide useful measures that can be validated against human data.