| The human language comprehension system can be characterized by three key properties. First, incoming utterances are interpreted immediately and word-by-word (incrementality). Secondly, multiple sources of information are combined in the interpretation of an utterance (integration). And third, listeners anticipate what a speaker will say next (prediction). Many computational approaches have been proposed to explain specific aspects of comprehension, but no model exists which captures all of these properties. Moreover, existing models have focused on predicting the behavioral correlates of comprehension, such as reading time or eye-tracking data. We propose to devise a computational model of sentence comprehension which incorporates incrementality, integration, as well as prediction. In addition, we aim at modeling the neural correlates of comprehension as provided by Event-Related brain Potentials (ERPs) because ERPs reveal much more fine-grained information about language processing than their behavioral counterparts. Our model will be applied to an important phenomenon in comprehension known as the Semantic Illusion. Semantic illusions occur when anomalous sentences are processed as if they made perfect sense. Experiments with such sentences have shown ERP profiles which challenge previously held views on how the brain handles language processing. In our project, we seek to obtain a unified, computationally precise explanation of these effects. Predictions from our model will be tested in novel ERP experiments. |
