| Summary: | In the research literature on aphasia, the average number of participants is 7.5. To go beyond this narrow sampling, the AphasiaBank project has constructed a database of one-hour interactions with 285 PWAs and 180 controls. The protocol includes story retellings, picture descriptions, procedural discourse, list recall and repetition, verb naming, picture naming, and personal narrative. Using a similar core protocol, parallel databases are being constructed for Cantonese, Mandarin, French, Italian, and Spanish, as well as for certain bilingual groups.
Using various tools we have constructed for automatic computational analysis, 30 studies have examined eight major issues:
1. Discourse. There have been six studies of discourse patterns in AphasiaBank interactions. These studies have used basic measures such as MLU and propositional analysis through CPIDR or Nicholas and Brookshire scoring, as well as more complex frameworks such as Rhetorical Structure Theory, story grammar matching, semantic feature analysis, and main concept analysis. Significant differences are found between PWAs and controls, along with correlations in PWAs with aphasia severity.
2. Grammar. There have been six studies of grammatical effects in aphasia. These studies have shown that, compared to controls, PWAs rely more on frequent, shorter verbs with more subcategorization options. Verb production in the discourse samples is highly correlated with performance on the Verb Naming Test. PWAs used fewer subordinate clauses and had a high level of “that”-omission in subordinate clauses that could not be linked to prosodic factors.
3. Gesture. Because the protocol is videotaped, it provides good material for the analysis of how PWAs and controls use gesture for parallel tasks. Three studies have shown how PWAs produce more deictic and iconic gestures than controls, with Broca’s producing high amounts of pantomime.
4. Lexical Processing. Four studies have examined lexical processing. In Anomics, nouns are retrieved better than verbs, particularly in picture naming, as opposed to free discourse. Measures of lexical diversity through TTR and VOCD were found to be less reliable than those made through the newer MATTR (moving-average type-token ratio) measure.
5. Fluency. Five studies have examined fluency in terms of pause duration and words produced per minute. PWAs showed high levels of disfluency, which correlated only weakly with WAB scores.
6. Statistical Classification. Using variables from the AphasiaBank protocol we were able to classify participants in ways that largely matched their WAB scores, but which added additional information for subtypes of Broca’s and Wernicke’es. We are currently in the process of constructing detailed analysis of variations in error patterns across different aphasia types.
7. Social Factors. Four studies have examined the impact of social factors such as dialect, gender, educational level, and occupation on performance in the protocol.
8. Treatment and Recovery. Several studies are now examining patterns of recovery based on successive administrations of the protocol, as well as repeated work with the scripting treatment.
Work is in progress to extend AphasiaBank to include data on scripted repetition, participation in group therapy discussions, famous people descriptions, interactions with web-based therapy, and dense longitudinal tracking.
|
|---|