Supervisor: Gerold Schneider
When sentences get complex, ambiguous, or employ unusual lexis or argument structure, it takes readers considerably longer to read, or they may even hesitate and re-read parts of the text, for example in garden-path sentences. Eye-tracking corpora have now become available. In order to predict reading ease or warn writers about infelicitous constructions, researchers are interested in building language models that correlate well to the observed data. We use information theoretical tools like n-gram models, surprisal, syntactic parsers, to assess and improve correlations. The ultimate goal of this type of research is to come up with models of native speakers.