Prediction and Integration in Human Parsing

Experimental evidence shows that human beings process language incrementally: they assign meaning to a sentence on a word-by-word basis, without waiting for the sentence to end before they interpret it. Moreover, readers or listeners are good at predicting upcoming linguistic material, e.g., they can anticipate the arguments of a verb once they have heard or read the verb. This raises an interesting question for languages such as German and Japanese, where the verb occurs at the end of the sentence. How does prediction work in this case?This proposal is part of ongoing work that attempts to shed light on the incremental processing of verb-final languages. In particular, the aim is to test two hypotheses that have been advanced in the literature. Standardly, it is assumed that holding the arguments of a verb in memory during incremental processing is costly. This means that the more arguments precede a verb, the more difficult it should be to integrate verb into the sentence, which should lead to increased processing time. The alternative hypothesis is that the processor works probabilistically and tries to predict the verb based on the arguments that precede it. This means that the verb should become less difficult to process (more predictable) the more arguments precede it.This proposal requests funding for two trips to conduct research on prediction and integration in the processing of verb final languages. The first trip will foster an existing collaboration and will be used to analyze and interpret a series of eyetracking experiments on prediction and integration. This trip will also be used to prepare the results for publication. The second trip will be used to initiate new collaborations on the computational modeling of prediction in sentence processing. The aim is to develop a mathematical model of the experimental results, as well as computational simulations.