Taler, V. . 1, 2 , Johns, B. 3 & Jones, M. 4
1 University of Ottawa
2 Bruyere Research Institute
3 University of Buffalo
4 Indiana University Bloomington
While most accounts of statistical learning focus on development, i.e.
the mechanisms by which environmental information is acquired and stored
within the brain, the present research examines the opposite end of that
spectrum: how that information is lost in neural degradation. To do so,
we use a cognitively-plausible learning mechanism to derive
representations of words from the natural language environment. We then
examine how this information is lost in people who are developing
cognitive impairment; that is, we examine how the large-scale
statistical regularities that are contained in semantic memory are lost
as a function of stages of neural loss. These simulations are then used
to understand data from a range of verbal fluency tasks (e.g., name as
many animals as you can beginning with the letter F) in three different
participant groups: young adults, older adults, and people with mild
cognitive impairment. These comparisons show how the model is sensitive
to the structure and decay of the representations in specific semantic
categories. Collectively, this work provides an important perspective
on large-scale statistical learning as a function of typical and
atypical aging.