| Summary: | Sixty one people with aphasia (pwa) and forty one matched controls were tested for the ability to understand sentences that required the ability to assign particular syntactic structures. Participants paced themselves word-by-word through twenty examples of eleven spoken sentence types and indicated which of two pictures corresponded to the meaning of each sentence. Sentences were developed in pairs such that comprehension of the experimental version of a pair required an aspect of syntactic processing not required in the corresponding baseline sentence. The need for the syntactic operations required only in the experimental version was triggered at a “critical word” in the experimental sentence. Listening times for critical words in experimental sentences were compared to those for corresponding words in the corresponding baseline sentences.
We adjusted self paced listening times for word duration by subtracting word durations from tag-to-tag self paced listening times to correct for word duration, yielding what we have previously called “corrected listening times.” Corrected listening times above ceiling (10,000 msec for sentence-final words and 5,000 msec for all other words) were discarded. For controls, this led to 0.2% of data being discarded and for PWAs 2.2% were discarded. Corrected listening times that were more than 3 standard deviations above or below the mean for that sentence type for each subject were adjusted either down to the upper limit or up to the lower limit of the 3SD range (not discarded). For accurate sentences, 1.7% of the control data were adjusted and 1.8% of the aphasic data were adjusted. For inaccurate sentences, 10% of the corrected listening times were adjusted for controls and 3.3% for aphasics.
Our interest is in incremental parsing and interpretation. The measure we used of this process was the residual of a regression of corrected self paced listening times for critical words in experimental sentences (the DV) against corrected self paced listening times in the corresponding words in the baseline sentences (the IV) in correct responses. We call the residuals of these regressions “relative corrected listening times.” Relative corrected listening times are based on listening times at points at which task-related operations are similar, and therefore factor out these effects. They also factor out differences in general speed of processing and motor functioning, which determine the intercepts of the regressions. The relative corrected listening times for each participant therefore reflect the time taken by each pwa or control participant to perform the parsing and interpretation operations needed in the experimental sentences, compared to the time taken by the other controls or pwa.
We performed these regressions separately in each of the four groups of sentences. We analyzed pwa and controls separately. We analyzed groups and not individuals, because performing these regressions on an individual basis would lead to a mean residual for an experimental sentence type compared to a baseline sentence type that approximates zero in each individual (this is a property of linear regression). We analyzed controls and pwa separately to produce separate estimates of basic speed of processing, decision-making and other processes in controls and pwa. We calculated the normal range of relative corrected listening times for each sentence type group by applying the formula in Crawford and Howell (1998) to the results of the regressions in the controls. We then determined whether the relative corrected listening time for each group of sentences in each pwa was within the normal range of residuals.
We focus on a finding that has not previously been reported, which is that, in some pwa, some relative corrected listening times (i.e., residuals of the regression of critical words in experimental sentences against corresponding words in baseline sentences) were lower than those seen in normal individuals. In normal individuals whose comprehension is within the normal range, fast self-paced reading times and short eye fixations are taken as evidence for efficient parsing and interpretation. It is not plausible that lower-than-normal relative corrected listening times in pwa reflect parsing and interpretation that is more efficient than normal. Some other account of these low relative corrected listening times is required.
A possible reason for abnormally low relative corrected listening times is that some pwa make an unusual choice between spending time in processing the current word and maintaining information in memory. The longer incremental processing continues, the longer what has been processed must remain in memory. Pwa whose relative corrected self paced listening times were lower than normal might have balanced time spent in incremental processing and memory load in ways that neurologically normal individuals do not normally do, preferring to spend less time in incremental processing with the consequence of reducing memory load.
In some pwa whose relative corrected self paced listening times were lower than normal, accuracy was normal. In these pwa, the effect of spending less time at this stage was not detrimental. Four of these pwa had abnormally low relative corrected listening times in only one sentence type; all four had normal accuracy on all sentence types. Given their normal accuracy, it is hard to say that their lower-than-normal residual corrected listening times reflect a structure specific deficit. These pwa appear to have set the duration of incremental parsing low for only one structure.
In most pwa whose relative corrected self paced listening times were lower than normal, accuracy was below normal or at or below chance. In these pwa, the lower-than-normal residual corrected times at critical words in experimental sentences could be due to either an intrinsic pathology of the parser/interpreter that limits the time it operates or to a choice between balancing time spent in incremental processing and memory load that leads to more errors than normal. The first of these possibilities is the counterpart of slowed syntactic processing. The second possible mechanism is related to how the control system regulates parsing and interpretation, a mechanism that has some similarity to what has been suggested in the case of lexical processing disturbances in vascular aphasia (Jefferies et al, 2008). If the locus of the disruption is in the control system, these pwa may also have abnormalities in metacognitive functions that detect errors and adjust processing to reduce them.
The six pwa whose relative corrected listening times were lower than normal in one sentence type and whose accuracy was also below normal on that sentence type appear to have had structure-specific deficits, The structures in which these abnormally low relative corrected listening times and accuracy were found showed double dissociations (two cases showed this pattern for passives; three for reflexives; one for pronouns) and none were found only in the most demanding sentence type (object relatives). This argues that the pathological mechanism that produced these behaviors is not resource reduction. These data therefore are consistent with structure specific deficits. It appears that a pathological truncation of parsing and interpretation, or a poor choice between processing and storage, applied when a particular structure was encountered.
The fact that these six pwa had chance performance on at least one other sentence type where relative corrected listening times for critical words were normal indicates that they have other deficits that affected other structures. To our knowledge, this is the first time the possibility that different pathological mechanisms can lead to abnormally low accuracy on different sentence types within individual pwa has been suggested. We are able to draw this conclusion because on-line behaviors differed in different sentence types on which accuracy is below normal in individual pwa, being lower than normal in one sentence type with below normal accuracy and normal or higher than normal in others. It is possible to find similar effects in other on-line behaviors such as pathologically long listening or reading times or abnormal patterns of eye fixations. For instance, it is possible that self-paced listening times or eye fixations would show early interference effects in one sentence type and late interference effects in another in one pwa (this would require testing pwa on two tasks). The present results point to the need to examine on-line data to know whether a pwa has more than one deficit at the level of mechanisms. Examination of accuracy and RT to end-of-sentence responses alone cannot tell us whether this is the case.
|
|---|
