I talked about my summer project. We also watched the great “Tono Tono”, and worked on NACLO problems.
About 300,000 people each year suffer some degree of aphasia after strokes. A lot of post-stroke aphasics end up scoring really well on the standard tests after a while, so they’re considered recovered and insurance stops paying for treatment. The problem is that the standard tests only cover phonological, lexical, and syntactic stuff; many of these people (we estimate ~20,000 a year) have lingering discourse-level difficulties that prevent them from having normal conversations or doing their jobs.
We had transcripts of “well-recovered patients” and controls describing Norman Rockwell paintings. I cleaned up the transcripts and scored them for content units. Then I analyzed the data on a word-by-word basis by statistically comparing the usage percent of each word between the two groups (I developed the method). Some 35 words were significant, most of which were detailed descriptors which were used more by controls. We also used a neat statistical method called VLSM to identify brain areas associated with each deficit, which lets us speculate about what might be responsible. The results indicated that patients conserve unnecessary words, as the physical act of speech is difficult. Applied to a much larger corpus, this technique could allow the development of a diagnostic tool to quickly flag patients with discourse difficulties based on a speech sample.