Shravan Vasishth and Felix Engelmann

  • Area: LoCo
  • Level: F
  • Week: 1
  • Room:

Abstract

Sentence comprehension, a field of research within psycholinguistics, is concerned with the study of the cognitive processes that unfold when we hear or read a sentence. The focus is on developing theories and models about how sentence comprehension (parsing) works.

The last sixty years have seen significant advances in our understanding of sentence comprehension processes. However, the vast majority of this work is experiment-based, with theory-development being largely limited to paper-pencil models. Although we have learnt a great deal from such informal reasoning about cognitive processes, ultimately the only way to test a theory is to implement it as a computational model. This is a standard approach in research on cognition in artificial intelligence, computer science, and mathematical psychology. Indeed, history has shown that the development of different computational cognitive architectures and mathematical models of cognition has had a huge impact in advancing our understanding of cognitive processes.
This is because computational and mathematical models force the scientist to make detailed commitments, which can then be tested empirically.

The present course brings together these two cultures: informally developed theories of sentence comprehension, and computational/mathematical models of cognition. We develop a series of accounts of sentence comprehension within a specific cognitive architecture that has been developed for modeling general cognitive processes, the ACT-R architecture (version 6.0).

ACT-R is a good choice because it is a mature architecture that has been widely used in artificial intelligence, human-computer interaction, psychology, and psycholinguistics to study cognitive processes.

Some of the issues that have been addressed empirically in sentence comprehension research are: (a) the influence of individual differences in capacity on parsing, (b) the role of parsing strategy, including task-dependent underspecification, (c) the role of probabilistic expectations, (d) the interaction between grammatical knowledge and parsing, (e) the interaction between the eye-movement control system and sentence comprehension, and (f) how impaired processing might arise (e.g., in aphasia). We address all of these topics by presenting computational models of representative phenomena using the ACT-R framework. The source code for the model is already freely available on github:

  1. The ACT-R Parser extended with eye-movement control
  2. The ACT-R Parsing Module
  3. ACT-R simulation of retrieval process
  4. Simple Memory
    Retrieval Model for sentence processing based on ACT-R, originally
    developed by R. Lewis and W. Badecker. With some extensions.

Further information:

course website