Probablistic
Approaches to Modelling Eye Movements in Reading
Scott McDonald
In this talk, I will discuss the
application of probabilistic methods to modelling two
aspects of reading behaviour. The first is the influence of
transitional probability statistics on the duration of eye
fixations, which suggests that the processor is able to rapidly draw
upon statistical information about word contingencies in order to predict
the identity of upcoming words. As the effects of both word frequency
and transitional probability have a similar timecourse, I have
developed a Bayesian model that integrates both sources of information,
where frequency is viewed as the a priori probability of
the fixated word occurring.
In the second part of the talk I'll describe an essential property
of eye movement control during reading - the choice of a
particular word as the target for the next fixation - and
report my attempts to model this as a classification task.
Saccade target selection is goal-directed and
non-deterministic, and so a probabilistic approach that
integrates the predictions of various visual, orthographic and linguistic
features should provide the most appropriate solution.