Agrammatic comprehension of relative clauses

Four hypotheses that attempt to account for the comprehension deficit in agrammatism are put to an empirical test. The interest in them is in that they all view the deficit as highly selective. The first, proposed by D. Caplan and C. Futter (1986, Brain and Language, 27, 117-134), argues that agrammatic patients cannot carry out normal syntactic analysis beyond the category label of each incoming lexical item and are reduced to the use of a cognitive strategy that commends assignment of thematic roles to noun phrases merely by their linear position in the string. A second, less radical hypothesis (Y. Grodzinsky, 1986a, Brain and Language, 27, 135-159), accounts for the deficit differently, by deleting a particular kind of syntactic object (trace) from the otherwise normal representation, and augmenting the resulting, underspecified representation by a strategy, whose use is quite restricted. A third account that is tested contends that agrammatic aphasics fail to comprehend perceptually complex constructions, where the metric for complexity is determined by results obtained from comprehension tests of normal listeners. The fourth account (M. F. Schwartz, M. C. Linebarger, E. M. Saffran, and D. S. Pate, 1987, Language and Cognitive Processes, 2, 85-113) argues that the thematic transparency of a construction (whether or not thematic roles are assigned directly to positions) is the best predictor of the manner by which agrammatics can handle it. An empirical test is thus constructed, both to extend the evidential basis concerning the comprehension skills of these patients and to distinguish between the accounts. Four types of relative clauses are presented to the patients, where embedding type (center vs. right) is one variable, and location of gap (subject vs. object position) is the other. The patients are tested in a sentence-picture matching paradigm. The finding, that is rather robust, is that gap location is the best predictor of agrammatic performance: the patients perform well above chance on both types of subject gap relatives, and at chance levels on object gaps. It is then shown that the Trace-Deletion Hypothesis (Grodzinsky, 1986a) is the only one among the accounts considered that is compatible with these data.