Slide Slam A17
A study of syntactic acceptability judgments in chronic post-stroke aphasia using lesion-symptom mapping
Danielle Fahey1, Julia Igoe1, Julius Fridriksson1, Gregory Hickok2, William Matchin1; 1University of South Carolina, 2University of California, Irvine
Prior lesion-symptom mapping studies have associated damage to posterior networks with syntactic comprehension deficits and paragrammatic speech. By contrast, lesion-symptom mapping studies have associated damage to frontal networks with agrammatic production deficits and to some extent, deficits in comprehension of complex non-canonical structures, but with little implication in basic syntactic comprehension deficits. However, most studies assess syntactic comprehension indirectly through the use of noncanonical sentence comprehension. Syntactic acceptability judgments (SAJ), which ask participants to gauge sentences’ well-formedness, provide a more direct assessment of syntactic comprehension. We predicted people with aphasia (PWA) would better detect word-order violation than agreement or subcategorization violations. We expected lesion-symptom mapping to show association between comprehension deficits and damage to posterior temporal regions, but no association with frontal damage, consistent with Wilson & Saygin (2004). We adapted Wulfeck & Bates (1991)’s SAJ task into two experiments. We presented our experiment in 2 tasks, with 64 sentences each (eight of each type, examples below), suggesting English was the speaker’s second language. In both tasks, we manipulated sentences’ grammaticality, ½ ungrammatical, counterbalancing placement of the (single) error, on the object (examples a/c) or verb (examples b/d). Task 1 manipulated agreement (examples a/b) and word-order (examples c/d) grammaticality. (a)She is baking a cake/*a cakes. (b)She is/*are baking a cake. (c)They have listened to some teachers/*teachers some. (d)They have listened/*listened have to some teachers. Task 2 controlled whether verbal complements were obligatory (examples e/f) or optional (examples g/h), and manipulated subcategorization selection (e.g., auxiliary selection for verbs (examples e/g) or subordinate complement/adjunct selection by adding/deleting/substituting prepositions on prepositional or noun phrases (examples f/h)). (e)They are chasing a cat/*in a cat. (f)They have signaled/*have signaling to a boat. (g)He is answering a phone call/*to a phone call. (h)They are hurrying/*have hurrying to a meeting. From responses, we performed a repeated measures (RM) ANOVA to identify differential accuracy by error type and location, and regressed participants’ lesions against error detection in both tasks using NiiStat, correlating lesion site to grammatical processing, following Matchin et al. (2020). Eighteen PWA participated. RM ANOVA results showed a significant main effects of error type (p<.001, F(1.622)=3.195, ηp2=.176). Main effect of error location (p=.054, F(1.000)= 4.358, ηp2=.225) and interaction between error type and location (p=.068, F(1.264)= 16.345, ηp2=.521) were approaching significance. Predictably, participants were more accurate on word-order violations (M=65%), but unexpectedly more accurate on subcategorization violations (M=52%) than agreement violations (M=46%). Lesion-symptom mapping showed stronger associations to the posterior temporal ROI (agreement: z=-.886, p=.188; word-order: z=-1.195, p=.116; subcategorization: z=-1.543, p=.061) than the inferior frontal ROI (agreement: z=-.675, p=.250; word-order: z=1.498, p=.933; subcategorization: z=-.581, p=.281). Generally, word-order violations (66% correct) were easier to detect than agreement (49%) or subcategorization (50%) violations; violations showed association to posterior temporal regions rather than inferior frontal regions, consistent with findings by Wilson & Saygin, but contra suggestions that frontal regions primarily support grammatical comprehension processing. Differential accuracy across violation types could be because word-order involves sentence constraint, while agreement and subcategorization require processing hierarchical dependencies; further, agreement is an unbounded dependency.