Presentation

Danielle Fahey¹, Jeremy Yeaton², Brielle C. Stark³, Julius Fridriksson¹, Gregory Hickok², William Matchin¹; ¹University of South Carolina, ²University of California Irvine, ³Indiana University Bloomington

Considerable research has focused on agrammatism, but scant research has focused on paragrammatism (Goodglass, 1997; Matchin et al., 2020; Thompson, 2019; Wilson et al., 2010). Recent work by Matchin et al. (2020) has focused on dissociating behavioral or lesion correlates of paragrammatism from agrammatism, analyzing 53 patients’ discourse samples holistically to determine whether perceptual ratings were reliable/meaningful. Each sample was labeled as AGRAMMATIC, PARAGRAMMATIC, BOTH, or NEITHER. The present study advances Matchin et al., analyzing samples at the utterance level using transcriptions. An initial sampling of 15 patients was randomly selected, 4 previously analyzed as AGRAMMATIC, 6 as PARAGRAMMATIC, 1 as BOTH, and 4 as NEITHER. Each utterance received an overall code of (u)ngrammatical, (g)rammatical, (f)ragment or (h)esitation. Strictly defining agrammatism and paragrammatism, utterances received a secondary code of error type: Ungrammatical utterances were coded as: (o)mission if the sentence was illicit from omission of a single grammatical morpheme, (a)grammatic if there are multiple grammatical morphemes omitted, (p)aragrammatic if the sentence is illicit from disagreeing, inserted grammatical morphemes, (v)erb if the sentence is only missing the lexical verb, or (a+p) if there are violations arising from disagreeing, inserted and from omitted grammatical morphemes. Grammatical utterances were coded as: (g)rammatical if there were no errors, (s)emantic if there were semantic violations, or (n)eologism for utterances characterized primarily by paraphasias. Fragments were analyzed at the constituency level and could be coded with the above codes. (H)esitations were coded twice. Samples original labels were compared to the proportion of codes from transcriptions. On average, 82% of AGRAMMATIC samples’ utterances were coded (a). On average, 61% of PARAGRAMMATIC samples’ utterances were coded (p). 0% of the BOTH sample’s utterances were coded (a), while 67% were coded (p). On average, 6% NEITHER samples’ utterances were coded (a) while 63% were coded (p). Our preliminary results reinforce original perceptual ratings, with caveats. There was no correlation to fluency or aphasia type from either perceptual ratings or transcription codes. Further, AGRAMMATIC samples were symmetrically classified with transcription codes. PARAGRAMMATIC and NEITHER samples had less correspondence, which might be explained by lower inter-rater reliability in Matchin et al. for these samples. Further analyses will target the source of these discrepancies. Another avenue for investigation is the relationship between speech rate (e.g., words per minute (wpm)) and type of grammatical error. Of the initial 15 samples, AGRAMMATIC discourses averaged 27wpm, PARAGRAMMATIC discourses 78wpm, the BOTH discourse 36wpm, and NEITHER discourses 49wpm. AGRAMMATIC labels correlated to wpm (R2=0.523) more than PARAGRAMMATIC labels (R2=0.030). Similarly, (a) codes correlated (R2=0.382), more than (p) codes (R2=0.184). Since potential bias from perception of speech rate was not possible with transcription analyses, these parallel correlations suggest that there may be relationship between speech rate and grammatical processing that could aid in identifying how these syndromes dissociate. Lesion symptom mapping will be conducted with the 53 samples upon completion of utterance coding toward this aim. Thus, this novel coding scheme from transcribed utterances has the potential to clarify double dissociations in grammatical deficits of these syndromes.

Topic Areas: Disorders: Acquired, Syntax