We are very happy to announce the next talk in the Semantic Colloquium, which will take place on Thursday, May 2, 4 – 6 pm in IG 4.301.
Markus Werning (Ruhr University Bochum) will present „Bayesian Pragmatics and the Contextual Modulation of Word Meanings in Online Comprehension: A Quantitative Model of EEG and Cloze data“.
Abstract:
We contrast three quantitative models to explain the contextual modulations of word meanings and how they affect the probabilistic predictions on the completion of a discourse. How words are semantically understood by a listener can not only be studied from the point of view of compositionality, i.e., by asking what they contribute to the truth-conditions the listener recognizes the speaker to express when the latter has uttered a sentence. Word meanings can also be studied by looking at their dynamic effects, i.e., asking in which way they influence the listener’s predictions about the truth-conditions the speaker is going to express when completing the sentence. The Semantic Similarity Model proposes that the main predictor of this dynamic effect are learnt statistical regularities in the framework of distributional semantics. The Relevance Model, in contrast, maintains that the main predictor is the speaker’s intention to maximize relevance. The Bayesian Pragmatic Model, finally, accounts for the rational cooperation between speaker and listener by Bayes’s Theorem. It estimates the influence a word’s meaning has on the expected completion by looking at posterior probabilities that take into account the speaker’s relevance considerations as well as statistically based priors. In a multiple linear regression analysis, the proportions of variance explained by each model are compared with regard to both, Cloze values and N400 amplitudes. The Bayesian Pragmatic model turns out to best explain the data.
You are cordially invited!