Grammaticality judgment under non-optimal processing conditions: Deficits induced in normal particpants resemble those observed in aphasic patients

Abstract

Department of Linguistics, University of Vienna, USAIntroductionNot all sentence types are affected equally in agrammatic aphasia.For instance, Schwartz, Saffran and Marin (1980) found that Broca’saphasic patients perform much better on semantically reversible activesentences than they do on passive structures. Some researchers haveinterpreted such patterns as indicative of impairments to particulargrammatical representations or processes. One proposal is thatagrammatic aphasic patients are selectively impaired in their ability toprocess structures involving traces of maximal projections (Grodzinsky& Finkel, 1998). However, such claims are challenged by the findingthat, under conditions of acoustic degradation and/or increased pro-cessing load, patterns of deficits can be induced in normal participantswhich closely mirror those seen in agrammatic aphasia (Dick et al.,2001). This suggests that the data from agrammatic aphasia do notnecessarily justify the postulation of localized neural substrates forhighly specific syntactic mechanisms.Most previous work comparing patterns of breakdown in aphasiawith patterns of breakdown in normals under non-optimal processingconditions has been in the domain of sentence comprehension (e.g.,Dick et al., 2001). Another domain in which similar questions can beaddressed is grammaticality judgment. We have recently shown thataphasic patients tend to be impaired in grammaticality judgmentacross-the-board on a variety of sentence types, unlike in sentencecomprehension where more complex structures are differentially af-fected (Wilson & Saygin, submitted). The hypothesis that domain-general factors may account for patterns of agrammatic performancethus predicts that acoustic degradation and increased processing loadshould impact grammaticality judgment roughly equally across allsentencetypes,sincethisisthepatternobservedinagrammaticaphasia.Here we tested this hypothesis by asking normal controls to performgrammaticality judgments under optimal conditions versus three dif-ferent non-optimal processing conditions. Four types of sentences werepresented,crossingtwofactors:whetherornotgrammaticalitydependsupon a trace of a maximal projection, and whether the sentences weremore or less complex.MethodsForty-sixUCSDcollege students participated, andwere assignedtooneofthefollowingfourconditions:Incondition1(controlcondition),participantswere presentedwith sentences bothvisuallyandauditorily,and were asked to make grammaticality judgments. In all other con-ditions, sentences were presented only auditorily. In condition 2, sen-tences were acoustically degraded by compressing them to 80% of theiroriginal length and applying a low-pass filter at 1200 Hz. Condition 3was similar but more challenging: sentences were compressed to 70%and low-pass filtered at 800 Hz. In condition 4, sentences were com-pressedandfilteredasincondition2,plusparticipantshadtoperformadistracting linguistic task in addition to grammaticality judgment:words and nonwords were displayed visually at a rate of 2 per secondfor 3.5 s duringauditorysentencepresentation,andparticipantshadtoread the real words aloud.There were 24 sentences in each condition, half grammatical andhalf ungrammatical. Sentences were recorded in a sound-proof boothby an experienced phonologist who was instructed to avoid prosodiccues to grammaticality status. The stimuli were edited with SoundEdit16 and presented with PsyScope.ResultsA repeated measures ANOVA comparing the performance of thefour groups on the four sentence types revealed significant main effectsof group [F(3, 42) = 101.82, p < .0001], and sentence type [F(3,126) = 10.64, p < .0001]. Unsurprisingly, performance was better un-der optimal conditions, and better on the ‘‘easy’’ sentence types (Fig.1). Crucially, the group by sentence type interaction was not significant[F(9, 126) = 0.97, p = 0.46], implying that all manipulations affectedthe four sentence types across-the-board, without differential effects onany particular sentence type. This is similar to across-the-board deficitsseen in aphasic patients (Wilson & Saygin, submitted), whose pooledresults are also depicted in Fig. 1. In fact, the condition where sen-tences were compressed to 70% and filtered at 800 Hz was statisticallyindistinguishable from the pooled results of all aphasic patients fromthat study. Thus this manipulation succeeded in roughly ‘‘simulating’’the average performance profile of aphasic patients.Brain and Language 87 (2003) 67–68www.elsevier.com/locate/b&l