Inference revision processing in adults with and without aphasia

Abstract

Processing abilities in aphasia, and the nature of processing breakdowns, were the focuses of this investigation. Individuals with either fluent or nonfluent aphasia, plus a control group, participated in a cross-modal lexical priming task designed to elicit priming effects when activation of inference interpretations occurred. Comprehension of inferences was measured by responses to four types of questions that related to the inferences. Results indicated that both the control group, as well as the nonfluent aphasia group, activated the intended meaning of the stimuli whereas the fluent aphasics did not. Comprehension of the inferences was best demonstrated by control participants, nonfluent aphasic participants, and fluent aphasic participants, in that order.

Introduction

Comprehending sentences and discourse often requires generating inferences. For example, in the following sentence pair:

Bill bumped the car in front of him while going around the curve. At the end of the ride, Bill got out of the bumper car.

the inference initially generated by the listener is that Bill was in a car accident; however, after listening to the second sentence, and to successfully comprehend the sentence pair, the listener must revise his or her initial interpretation to conclude that Bill was not in an accident, he was on a county fair ride. The ability to generate inferences between two propositions when one contradicts the other has been investigated in adults with right hemisphere brain damage (RHD) (Brownell, Potter, Bihrie, & Gardner, 1986; Kaplan, Brownell, Jacobs, & Gardner, 1990; Tompkins, Bloise, Timko, & Baumgaertner, 1994). In general, these adults are able to generate an initial inference of a proposition successfully; however, they have difficulty when the initial interpretation must be revised in order to comprehend the intended meaning of the propositions. Though this has been studied in adults with RHD, the ability to generate inferences and revise interpretations in order to comprehend propositions has not been explored in adults with aphasia.

It is well documented that adults with aphasia present with comprehension deficits at various levels and to varying degrees of severity (e.g., Davis, 2000). Aphasic adults who are mild to moderately impaired may not present with comprehension breakdowns at the word level, but will have breakdowns when sentence and discourse are involved; thus, the ability to interpret the meaning of a sentence and to draw an inference is likely impaired in this population. Discourse comprehension requires more than simply accumulating the meanings of the propositions that make up the discourse (Brookshire, 1987). For example, discourse comprehension requires understanding the relationships between the propositions and the general theme or main idea of the discourse.

Nicholas and Brookshire (1986) found that adults with aphasia show a similar comprehension pattern to their nonbrain-damaged peers, in that they comprehend explicitly stated information better than implied information (i.e., constructing inferences). However, these adults with aphasia have significantly more difficulty comprehending the implied information as compared to nonbrain-damaged individuals. Swaab, Brown, and Hagoort (1998) investigated spoken sentence comprehension ability in adults with Broca’s aphasia. From their results, they suggested that their aphasic participants’ sentence comprehension deficit was due to a delay in lexical integration. That is, comprehension requires continuous, real-time integration of word meaning from the preceding context and adults with Broca’s aphasia are slower at integrating word meaning. This, in turn, negatively affects comprehension of the sentence’s meaning.

Levey and Goldfarb (2003) investigated fluent aphasic adults’ accuracy and response time for comprehending indirect requests. They found that their aphasic participants were significantly less accurate and slower when performing the task as compared to nonbrain-damaged participants. Comprehension of indirect requests requires multiple steps. The listener must be able to evaluate the intentional meaning of an utterance rather than its syntactic form. Thus, comprehension of indirect requests involves inferencing. Relatedly, Wapner, Hamby, and Gardner (1981) included adults with RHD and adults with aphasia in their study. They compared these brain-damaged groups’ performances on a variety of complex linguistic tasks including a task involving participants’ ability to comprehend fables as measured by performance on factual and inferencing questions. Results showed that the brain damaged groups performed significantly worse on the questions as compared to the nonbrain damaged group. Further, the adults with aphasia missed significantly more questions than the RHD group. Aphasic participants in the study included adults with fluent and nonfluent aphasia; however, it was not reported if they performed differently on the task. Based on the fact that aphasic participants performed poorly relative to their nonbrain-damaged counterparts on these tasks requiring inferencing ability, the suggestion was made that (a) there is a general sentence comprehension deficit in aphasia and (b) it can specifically include an inferencing deficit.

Tompkins et al. (1994) included adults with left hemisphere brain damage in their study in which comprehension of propositions that required revising an initial interpretation was investigated. They found that these participants, some of whom presented with aphasia, had difficulties with revising initial interpretations to comprehend the propositions. Tompkins et al. then divided the left hemisphere brain damaged group into two subgroups: high and low comprehension. Participants in the lower comprehending subgroup had more difficulty with the task than those in the higher comprehending subgroup, suggesting that the significant deficit in comprehension negatively affected their ability to generate the inferences for the propositions. This further supported the notion that the ability to comprehend sentence meaning requires the ability to generate inferences.

Several studies of the ability to generate inferences during sentence processing have been conducted (Cutler & Swinney, 1978; Long, Oppy, & Seely, 1994; Swinney & Osterhout, 1990; Till, Mross, & Kintsch, 1988; Wright & Newhoff, 2002). The measurement used in a number of these studies has been the point at which activation of lexical information (i.e., the generated inference) occurs. Briefly, activation is determined by the presence of priming. Priming is determined by reaction times; that is, a significantly faster reaction time to a related word than to a nonrelated word.

Using this paradigm, for example, Swinney and his colleagues conducted several investigations with various types of inferences to determine if they could be generated automatically (Cutler & Swinney, 1978; Swinney & Osterhout, 1990). All studies were conducted with neurologically intact participants and results were the same; that is, participants were not able to activate many types of inferences automatically. After sufficient time had elapsed, however, inference activation occurred. This suggested to Swinney and colleagues that generating most inferences requires cognitive control and strategic processing. In reviewing these works, Swinney and Osterhout (1990) proposed a model of inference generation, identifying two categories of inferences: perceptual and cognitive.

Perceptual inferences occur on-line and are automatic, mandatory, and immediate. As an example, Swinney and Osterhout suggested:

The boxer visited the doctor that the swimmer at the competition had advised him to see about his injury (p. 22).

In this sentence, the listener automatically activates the referent (boxer) to the pronoun (him). This occurs on-line, is mandatory, and immediate.

Cognitive inferences occur during on-line processing; however, they are not automatically or immediately derived. Rather, they require controlled processing and are tied to the person’s knowledge base, such as his or her world knowledge. The majority of inference types are cognitive inferences (e.g., topic, predictive, and elaborative inferences). An example of a sentence requiring the generation of a cognitive inference, also from Swinney and Osterhout (1990), is the following:

He cut the juicy meat and began eating his dinner (p. 30).

In this sentence, it is not stated that the man cut his meat with a knife; however, the listener infers from the context that a knife is used to cut the meat. Drawing this inference is not automatic and it requires the listener’s world knowledge to generate it. It is this latter type of inference, cognitive, that is most appropriate for comprehension studies.

More recent investigations of activation of inferences generated during sentence processing have also been conducted with neurologically intact adults and results have been similar to those of Swinney and his colleagues (Long et al., 1994; Till et al., 1988; Wright & Newhoff, 2002). Long et al. (1994) and Till et al. (1988) had participants read sentences and then complete a lexical decision task (LDT) after the sentence had been presented. Participants’ time-course for activating the intended interpretation of the sentence occurred between 500 and 1000 ms after the sentence was presented, but not before 500 ms. Wright and Newhoff (2002), using a cross-modal lexical priming task (CMLP), where sentences are presented aurally followed by visual lexical decision probes, investigated neurologically intact participants’ abilities to activate the initial interpretation of the first sentence of a sentence pair, then the revised, intended interpretation of the sentence pair. In the following example, participants listened to the sentences, then responded to the visually presented lexical probes as they occurred (i.e., following the initial sentence, then the sentence pair).

Bill bumped the car in front of him while going around the curve.^accident At the end of the ride, Bill got out of the bumper car.^fair

Participants were able to activate the initial interpretation at 750 ms after the first sentence was presented; then they activated the revised interpretation 750 ms after both sentences were presented. It is important to note that the results of these studies support the use of priming tasks as appropriate for measuring the interpretation of a sentence’s meaning. More importantly, they also demonstrate that the intended interpretation of a sentence’s meaning is a type of cognitive inference and cannot be activated automatically, but can be measured during on-line processing.

These investigations of inference generation have resulted in consistent findings using various study designs with neurologically intact adults; however, the ability to generate inferences and measure their activation during sentence processing has not been as readily investigated in adults with aphasia. The expectation would be that adults with aphasia would have difficulties with these tasks because of their documented comprehension impairment if, in fact, there is a relation between the two. Inspection of activation patterns for generating intended meanings of sentences would provide additional insight into the sentence comprehension abilities of adults with aphasia and could suggest why comprehension breakdowns, at this level, may occur. Further, an advantage to on-line tasks such as this type is that inference generation is measured as it takes place, not afterwards when additional linguistic and cognitive functions may be required (Kempler, Almor, & MacDonald, 1998). Off-line tasks may consist of follow-up questions, matching, or pointing activities; and, off-line comprehension performance may be affected by numerous impairments, such as motor, language, and cognitive. As a result, individuals may perform differently on on-line measures, such as priming tasks, and off-line measures, such as comprehension questions. Kempler et al. (1998) suggested that on-line comprehension tasks may be an “ideal way of evaluating patients” (p. 63) because they allow for evaluation of language performance without all of the additional resources that are involved in off-line measures (e.g., metalinguistics, memory, and conscious awareness). Of course, during on-line processing tasks it is impossible to reduce additional resources to zero. Although on-line processing measures certainly cannot be considered as over-loading as off-line measures can be, it remains true that some additional resources are required to perform an on-line task. For example, to measure an on-line event requires an off-line task (i.e., pushing a button). However, unlike most off-line measures, this activity occurs quickly and while sentence processing is occurring, thereby markedly limiting the amount of resources recruited.

Though using priming paradigms to measure activation of cognitive inferences generated during sentence processing has not been investigated with aphasic adults, automatic activation of antecedents, as shown in the above example as a type of perceptual inferencing, has been (Blumstein et al., 1998; Swinney & Zurif, 1995; Swinney, Zurif, Prather, & Love, 1996; Swinney, Zurif, & Nicol, 1989). From these investigations, researchers have begun to suggest that the levels at which breakdowns may occur are based on aphasia type/lesion location. For example, in the Swinney et al., 1989, Swinney et al., 1996, activation of antecedents was investigated in adults presenting with Wernicke’s aphasia and Broca’s aphasia. In general, the participants with Wernicke’s aphasia demonstrated automatic activation for the antecedent; however, the Broca’s aphasic adults did not activate the antecedent immediately, but eventually evinced activation at a later time. Thus, Swinney and Zurif (1995) concluded that the comprehension deficit characteristic of Wernicke’s aphasia occurs further down stream; that is, it has a later time course, in comparison to Broca’s aphasia, for occurrence during sentence processing. Broca’s aphasic individuals appear to no longer have this automatic function. This difference highlights the need to further explore inference generation in adults with different types of aphasia; in particular, to determine the time-course for generating cognitive inferences.

Blumstein et al. (1998) used a different design in their investigation of antecedent activation performance by adults with Wernicke’s and Broca’s aphasia. Sentences and lexical decision probes were both presented visually, so participants performed a reading task rather than an auditory task, and the lexical decision frequently occurred at the end of the sentence, rather than within the sentence. Perhaps, as a result, they found the opposite effect; that is, the participants with Broca’s aphasia activated the intended information and the Wernicke’s aphasic participants did not. With the lexical decision task occurring at the end of the sentence, the type of processing occurring may no longer be automatic, suggesting that sentence processing by adults with Wernicke’s aphasia breaks down when strategic processes are required to interpret meaning (Baylogh, Zurif, Prather, Swinney, & Finkel, 1998). However, the additional process of reading the stimuli also may have influenced the results of the study and reduced the possibility for automatic processing to occur.

To summarize, adults with aphasia present with impaired comprehension. Swinney and his colleagues (Swinney & Zurif, 1995; Swinney et al., 1989, Swinney et al., 1996) and Blumstein et al. (1998) demonstrated that the breakdown in aphasic adults’ ability to comprehend the intended meaning of a sentence might occur at various stages during processing depending on the participants’ behavioral presentation. The purpose of the present study, then, was to further explore aphasic adults’ abilities to generate inferences during on-line, as well as off-line, sentence comprehension. We investigated the ability to activate the initial interpretation of the first sentence in a sentence pair, then the ability to activate the revised, intended interpretation of the sentence pair. We also sought to determine if these individuals correctly comprehended the sentence pairs off-line by having them answer four yes/no questions following each sentence pair. Based on the literature, we initially predicted that (a) similar to the neurologically intact participants, the nonfluent aphasic participants would activate the initial interpretation following the first sentence, and then the revised and intended interpretation following the sentence pair; (b) the fluent aphasic participants would not activate the initial or revised interpretations in the allowed time course as the nonfluent aphasic participants and neurologically intact participants would; and (c) the aphasic participants would not comprehend the sentence pairs as well as the neurologically intact participants.