Characterizing Verb Argument Structure in Aphasia Using Dependency Parsers

Purpose. Verb argument structure (VAS) is often impaired in post-stroke aphasia. Previous studies investigated VAS impairments using constrained experiments or less often narrative speech, which relies on time-consuming manual annotations of VAS in speech transcriptions. Here, we aimed to develop and validate an automated approach to quantify VAS impairments in narrative speech using natural language processing (a dependency parser).Method. First, we validated this approach by replicating a previous study on VAS in spontaneous speech in aphasia using automatically annotated measures of VAS, in contrast to previous manually annotated VAS (Malyutina et al., 2016). The automatically annotated variables were able to capture similar patterns in the data as the previously manually annotated ones, and in some cases were more sensitive to differences between controls and participants with aphasia. We then applied this approach to a new dataset (n=106) to (i) replicate previous findings and (ii) assess verb argument structure impairments in agrammatism specifically. Results. The findings from the new dataset revealed that nonfluent participants, or, according to grammatical distinctions, agrammatic participants, are affected by VAS complexity in their verb selection and produce fewer arguments than controls. Conclusions. This study shows that dependency parsers are suitable to characterize VAS in spontaneous speech with virtually absent manual labor, showing the feasibility of developing clinically applicable tools for complex language analysis. The results of the study additionally confirm that participants with aphasia vary in their VAS sensitivity and abilities.