Human aging brings with it declines in sensory function, both in vision and in hearing, as well as a general slowing in a variety of perceptual and cognitive operations. Yet in spite of these declines, language comprehension typically remains well preserved in normal aging. We review data from functional magnetic resonance imaging (fMRI) to describe a two-component model of sentence comprehension: a core sentence-processing area located in the perisylvian region of the left cerebral hemisphere and an associated network of brain regions that support the working memory and other resources needed for comprehension of long or syntactically complex sentences. We use this two-component model to describe the nature of compensatory recruitment of novel brain regions observed when healthy older adults show the same success at comprehending sentences as their younger adult counterparts. We suggest that this plasticity in neural recruitment contributes to the stability of language comprehension in the aging brain.
In this review we have two goals. One is to present current views on the neural structures that underlie sentence comprehension when listening to everyday speech. Our second goal is to suggest why, in spite of age-associated declines in sensory and cognitive function and microscopic changes in the CNS, language comprehension typically remains well preserved in normal aging.
Although varying in detail from one language to another, all languages of the earth have certain fundamental features in common. First, all languages have a lexicon. This is a vocabulary keyed to specific concepts such as objects (nouns), actions (verbs), and modifiers of nouns (adjectives) and verbs (adverbs). Although many sub-human species can use lexical-like elements to communicate with others of their species, humans are unique in the power to develop sophisticated language structures. These structures free us from the here and now, allowing us to communicate with others of our species in a highly flexible manner about events, objects, states of mind, the past and the future. A second universal component of language accounts for this flexibility. This is a syntax: a system of rules that specify allowable combinatorial relationships between lexical elements and the additional cognitive resources we need to implement these rules. This system allows us to communicate using meaningful sentences to express complex thoughts and intentions.
Lesion studies to functional brain imaging
Much of the early research into the brain structures that carry language was motivated by the desire to understand the loss of language comprehension or production after stroke or other sources of brain injury. Such studies attempted to correlate the site of the lesion with the consequences to language function and for many years served as the sole means of understanding brain-language relations. These studies of aphasia (the disruption of language after brain damage) had already by the late 19th century isolated the “language area” in the left cerebral hemisphere in right-handed (and most left-handed) individuals (Goodglass and Wingfield 1998).
Within a broad perisylvian region, deficits in language production were shown to be associated with damage to the ventral inferior frontal cortex (vIFC; BA 44/45/47), which includes Broca's area, and deficits in language comprehension associated with damage to the posterior lateral temporal cortex (PLTC; BA 21/22/42), a part of which is commonly referred to as Wernicke's area. This early picture was completed by discovery of a white-matter tract, the arcuate fasciculus, connecting these two regions. The aphasic syndromes of Broca's aphasia, Wernicke's aphasia, and conduction aphasia were thus neatly accounted for by damage to, or disconnections between, these centers and pathways (Geschwind 1965, 1969). We show these three components of the core sentence processing region colored in blue in Fig. 1, with the arcuate fasciculus represented by the connecting blue double-headed arrow.
Such lesion-based research was significantly enhanced by technological advances in the development of computerized axial tomography (CT) and structural magnetic resonance imaging (MRI) that allowed contemporary visualization of lesion location, freeing investigators from reliance on postmortem data to study cognitive-cortical correlations. The reliance on correlations between damaged brain areas and the loss of particular functions to draw inferences about normal function in nondamaged individuals, however, remained the same.
The past 10 years have witnessed a dramatic development in our understanding of brain-language relations. Evidence-based models of large-scale neural networks can now be constructed by examining cerebral activation during the performance of language functions like speaking, reading, and understanding syntactic relations between the words of a sentence. This was made possible by the development of positron emission tomography (PET) and functional magnetic resonance imaging (fMRI), which allow investigators to visualize patterns of brain activity while a subject is performing a cognitive or linguistic task.
As we shall see, armed with such techniques it is now also possible to examine the special challenges of language processing in older adulthood. To be sure, some aspects of language ability decline as we age, but an equally intriguing question is how language comprehension remains so relatively stable even though the older brain undergoes a variety of neurobiological changes. Notable among these changes are data from in vivo structural MRI studies that show age-related reductions in brain volume, especially in frontal and medial temporal areas (Raz et al. 1997, 2004).
In Fig. 2 we reproduce scatter plots of adjusted brain volume from four cortical areas (selected from Raz et al. 2004) that make two important points. The first of these is to demonstrate that the degree of volumetric decline is not uniform across all cortical areas. Two areas that show the strongest correlation between adjusted volume and age are the lateral prefrontal cortex (top left) and hippocampus (top right). These areas are important, respectively, for working memory and maintenance of information for memory consolidation. One can illustrate the point that volumetric decline is not uniform across the cortex with two selected areas from Raz et al.'s data that show quite limited age correlations: inferior parietal cortex (bottom left) and primary visual cortex (bottom right). These cross-sectional data, which are scatter plots of individual subjects, make the second point, which is the wide variability among individuals across the age range.
Contrary to earlier notions that this reduced brain volume reflects significant cell death with normal aging, much of this volumetric loss may result from reduced synaptic density in these areas. Of presumed great functional significance are accompanying regional changes in dendritic morphology, cellular connectivity, and changes in neurotransmitter integrity (Bannon and Whitty 1997; Wong et al. 1997). Albeit more subtle in normal aging than in age-related neuropathology such as Alzheimer's or Parkinson's disease, these changes nevertheless have measurable impact on network dynamics that carry cognitive function (see Burke and Barnes 2006; for a discussion and review from the human and animal literature). In addition to these cortical changes, diffusion tensor imaging (DTI), a new magnetic resonance technique that allows mapping of white matter tracts by imaging the direction of water diffusion through tissue, shows age-related deterioration of callosal white matter fibers, especially in anterior brain regions (Head et al. 2004).
In the review that follows, we first outline the cognitive changes associated with normal aging and how they constrain language comprehension and production. We will also recognize the spared abilities that keep language among the best preserved of cognitive functions in normal aging. Second, we will draw from neuroimaging studies to describe our current understanding of the critical regions and large-scale neural circuits involved in sentence comprehension. Finally, we will examine current views on how the aging brain recruits neural resources to support successful comprehension in the face of sensory-cognitive decline. By examining language comprehension in the aging brain, we address one of the most fundamental questions in current neurobiology: how stable behavior can be produced in spite of changes in underlying neural structures and circuit parameters (Prince et al. 2004).
Sensory and cognitive constraints on older adults' language processing
A central feature of normal aging is a general slowing in a variety of perceptual and cognitive operations (Salthouse 1996). Although there is little loss of word knowledge in all but the oldest-old (Verhaeghan 2003), word retrieval while speaking becomes slower and more difficult. This can lead to more frequent occurrences of “tip-of-the-tongue” states, where a desired word or person's name is known, but there is difficulty in its retrieval (Burke et al. 1991).
The issue of age-related slowing takes on special significance in the realm of spoken language comprehension. Speech rates in everyday conversation range from 140 to 180 words per minute (wpm). While speech is arriving at this rapid rate, individual words must be identified and structural relations between the words must be established (which are the objects, which are the actions, and who is doing what to whom), and this material must be integrated with what has already been heard. Processing operations that cannot be accomplished as the speech is being heard must be conducted on a transient trace of that speech in memory. It would thus not be surprising if older adults had more difficulty comprehending and storing in memory rapid speech than their young adult counterparts. This turns out to be the case (Gordon-Salant and Fitzgibbons 1997; Wingfield 1996; Wingfield et al. 1999).
The aging process is also typically accompanied by a reduction in the capacity of working memory. This is the ability to temporarily hold and simultaneously manipulate information in immediate memory (Baddeley 1996). Understanding who gave whom the present in the sentence, “Mary who liked Catherine gave her the present” requires the listener to detect the overall sentence frame, “Mary gave her the present,” and the embedded clause, “who liked Catherine,” and to integrate the two correctly in working memory. As such, working memory is known to constrain the comprehension of sentences with complex syntactic structures (see the review in Wingfield and Stine-Morrow 2000) and also to result in older adults' avoidance in producing syntactically complex utterances the organization of which puts a special burden on working memory (Kemper 1992).
Although there is a question of whether all aspects of language processing are constrained by a single working-memory resource or by a complex of specialized resources (e.g., Caplan and Waters 1999), there is no doubt that working-memory limitations constrain cognition in aging adulthood. As might be predicted from these two observations, complex syntax and rapid speech rates operate in a multiplicative fashion in affecting sentence comprehension with their combined effects being more challenging for older than for younger adults (Wingfield et al. 2003).
A third feature that affects comprehension of spoken language is the increased incidence of hearing decline among the aging population. Although many adults retain good hearing well into old age, some degree of age-related hearing loss (presbycusis) is not uncommon among older adults. This loss tends to be differentially greater for the higher sound frequencies that are critical for speech perception (Morrell et al. 1996). In addition to simple acuity losses due to cochlear hair cell loss, there can also be higher-level auditory-processing deficits that leave speech sounding unclear or garbled (phonemic regression), even when it is amplified (Jerger et al. 1989; Schneider and Pichora-Fuller 2000).
Hearing loss may not merely cause the older adult to miss critical words in a conversation. An additional concern is that successful perception in the context of poor auditory acuity may come at the cost of perceptual effort drawing on resources that might otherwise be available for encoding what has been heard in memory or for higher-level comprehension operations. That is, even a modest decline in hearing acuity can have a negative effect on memory and comprehension even when it can be demonstrated that the words themselves have been correctly identified (see Wingfield et al. 2005 for data and a discussion).
In spite of impediments such as these, healthy older adults communicate with good effectiveness. This is so because linguistic knowledge, and the procedural rules for implementing this knowledge, remain well preserved in normal aging (Wingfield and Stine-Morrow 2000). Indeed, in all but late-stage Alzheimer's disease, the formal qualities of speech production (syntactic form, melodic line) and the ability to comprehend at least the surface meaning of speech are maintained (Kempler 2005).
Until recently the study of compensation in adult aging has been confined to behavioral studies that have elucidated the factors that come into play at a descriptive level, such as the way older adults can effectively use linguistic context to aid word recognition (Pichora-Fuller et al. 1995; Wingfield et al. 1991) and to use the intonation pattern of speech to help identify the major clauses, or “thought units,” in a syntactically complex sentence (Kjelgaard et al. 1999). The development of neuroimaging techniques has, for the first time, allowed us to gain insight into how the aging brain compensates for the abovementioned sensory and cognitive declines in terms of patterns of upregulation (an increase in brain activity evidenced by increased blood flow) that correspond to the compensatory allocation of neural resources.
Neural recruitment as a mechanism for maintained language performance
One central question must be how anatomic and physiological changes in the aging brain yield declines in certain aspects of language function. At the same time, we must also address the question of how performance is nevertheless typically maintained at a relatively high level in spite of these demonstrable changes. Functional imaging research has begun to answer these questions.
The primary technique used in most of the studies we will describe makes use of blood-oxygen-level-dependent (BOLD) functional magnetic resonance imaging (fMRI). This technique is based on the finding that changes in neural activity are accompanied by alterations in the deoxyhemoglobin concentration of the blood in the region of this activity. The resultant alteration in the magnetic property of the blood is detected in the MRI signal and serves as an indicator of increased mental activity. The brain images shown in such studies represent subtractions of activity levels during performance of an experimental task against a resting baseline or the activity level consequent to some other control condition.
A number of imaging studies contrasting performance by healthy young and older adults have shown a major difference in observed patterns of neural activation when young and older adults are asked to perform the same cognitive task. In general, there is a shift from more focal activation in young adults to more widespread patterns of activation in older adults (e.g., Cabeza 2002; Logan et al. 2002). Such pattern changes in older adults can include a decrease in neural activation in some brain regions and increases in others relative to young adults. This pattern has been observed in a number of cognitive domains, from encoding pictures (Gutchess et al. 2005) to studies of episodic memory (Cabeza 2001). At least two hypotheses can be entertained concerning these age-related differences.
One hypothesis attributes such pattern differences to reduced cerebral specialization in the course of the aging process. This can be referred to as the dedifferentiation hypothesis. This account holds that more widespread activation seen in older adults is a consequence of cortical regions losing their dedicated contribution to a specialized function. Such a decline in specialization of regional brain function with age would result in additional brain areas being activated in the older adult in regions that are often engaged by a similar kind of task. As a consequence one might see two cognitive tasks activating distinct but anatomically related brain regions in young adults, whereas older adults would show activation of both sets of regions for both tasks.
The appearance of widespread activation when older adults are confronted with a cognitive task could thus reflect a general decline in neural efficiency, resulting in the loss of differentiated function that is ordinarily observed when young adults engage the same activity (Park et al. 2003). A key question here is whether an observed increase in areas of activity in older adults helps or hinders task performance. There are certainly cases where more widespread activation is thought to interfere with cognitive success (Cabeza 2001; Logan et al. 2002).
A diminution of task differentiation in the older brain might have several consequences. The more widespread activation associated with age-related dedifferentiation could lead to activation of, for example, areas of frontal cortex the engagement of which compensated for the decreased specificity in neural activation (Park et al. 2003, 2004). Alternatively, there may be cases where one might interpret an increase in areas of cortical activation as reflecting the older brain strategically recruiting additional neural resources to maintain performance at a high level. We will refer to this alternative as the compensation hypothesis. The compensation hypothesis argues that the additional brain regions activated during task performance in older adults reflect focused recruitment in a specific anatomic distribution in response to the diminishing cortical resources that are available as we age. Unilateral activation during cognitive-task performance by young adults, for example, may be supplemented by recruitment of homologous regions in the contralateral hemisphere in older adults (Cabeza et al. 2002). When performing tasks that are primarily sensory in nature, older adults show recruitment of frontal cortex not activated by younger adults (Cabeza et al. 2004). Because frontal brain regions are associated with executive functions like working memory but not sensory functions, this would be consistent with older adults employing higher-level activation to compensate for sensory decline.
In weighing the evidence for the compensation hypothesis, the key question is whether increased activation in brain regions during task performance by older adults that does not appear when young adults perform the same task necessarily implies that these additional areas of activation are contributing meaningfully to successful task performance. This interpretive question can be met in two ways. One is to use an event-related imaging design in which one focuses on patterns of cerebral blood flow only for test trials in which performance has been successful. This strategy allows imaging in younger and older adults to be equated for performance accuracy. Another strategy is to contrast patterns of neural activation for older adults who tend to do well in task performance versus those who do not. This strategy takes advantage of the wide variation in task performance in older subjects while equating for age. Both approaches are represented in the studies to be described.
A second concern is that age-related physiological functioning may confound the interpretation of observed activations. These might include changes in the cerebrovascular tree that can slow transit time or in vascular-neuronal coupling (D'Esposito et al. 1999). Concerns such as these are addressed by performing comparable contrasts within groups of healthy older adults that provide results compatible with the contrasts of older adults with younger adults.
Anatomic model of sentence processing
Sentence processing offers an important test condition for examining these hypotheses. This is so because people engage in sentence processing naturally and with remarkable efficiency, even in healthy aging. This minimizes interpretive confounds that would be related to performance of unfamiliar or artificial tasks that may represent an additional source of difficulty for older relative to younger adults. Sentence processing also allows one to isolate factors of interest in the stimulus materials, such as demands on working memory, that can be manipulated experimentally. Finally, because language processes are understood in such detail, it is possible to examine the specific sources of compensation that contribute to age-related cognitive performance.
We will examine these questions from the perspective of a neural model of sentence processing that hypothesizes two major dissociable components: the previously noted core sentence processing component that subserves many of the central or “universal” facets of language that are implicated in sentence processing and an associated large-scale network of brain regions that supports executive resources such as working memory deployed as needed to help support sentence comprehension.
Figure 1 illustrates the previously described core sentence processing network (in blue) built around the ventral inferior frontal cortex (vIFC, which includes Broca's area) and the posterolateral temporal cortex (PLTC, which includes Wernicke's area) in left peri-Sylvian cortex. As first implied by the effects of focal brain damage, this core network for sentence processing has been confirmed and more exactly specified by neuroimaging studies in healthy adults (e.g., Caplan et al. 2000; Friederici 2002; Luke et al. 2002).
The resource network involves at least several frontal cortical regions and extends to include right hemisphere structures as well as subcortical structures. The left hemisphere structures associated with the recruitment of working-memory and executive resources include dorsolateral prefrontal cortex (dlPFC) and dorsal portion of left inferior frontal cortex (dIFC). Regions in the right hemisphere include PLTC and dIFC. These regions are shown in red in Fig. 1.
From the perspective of this model, these resources are recruited to supplement the core sentence-processing network when working-memory demands are increased in the face of long and/or syntactically complex sentences. These areas associated with working-memory support are distinct from activation in other areas such as anterior cingulate that respond to an attentional challenge as imposed by, for example, significantly accelerated speech rate (Peelle et al. 2004). Subcortical regions may also contribute to comprehension by upregulating structures such as the striatum during error monitoring when performing a difficult task.
This two-component model predicts that in healthy aging the core sentence-processing network is activated in a relatively consistent manner to support the generally good quality of older adults' sentence comprehension. Evidence for the stability of this core language area comes from sources such as activation studies of single-word comprehension that show little age-related change (Madden et al. 1996, 2002; Lustig and Buckner 2004). The model also predicts that cortical regions supporting executive resources are upregulated by the older brain to compensate for declining working-memory resources during language processing. Such activation would be especially noted in tasks that involve comprehension of complex sentences. Evidence compatible with this hypothesis comes from functional imaging studies showing increased activation in these areas in healthy aging during nonlinguistic working-memory challenges (Rypma et al. 2001). Compensation may involve not only a change in the anatomic distribution of regional brain activity but also a change in the temporal course during which the activation occurs. Finally, upregulation of working-memory resources and prolonged activation may be among the mechanisms that help explain the basis for stable language functioning in healthy aging despite the microscopic changes that occur in the aging brain.
Modified compensation hypothesis: the interaction of biological and cognitive factors during sentence comprehension in adult aging
The compensation model of directed cognitive and neuroanatomic activity during healthy aging was examined in a series of BOLD fMRI experiments that manipulated a grammatical component and a working-memory component of sentences (Cooke et al. 2002; Grossman et al. 2002a,b). The premise underlying this work was that although sentence comprehension is fundamentally preserved in the face of an aging CNS, comprehension can nevertheless suffer, especially when the sentences contain complex syntax and wording that may tax working-memory resources.
Consider, for example, a relatively simple active-conjoined sentence such as “The man insulted the woman and hired a lawyer.” In this sentence, there are several important thematic roles: the agent (the man) who performs both actions (insulting and hiring) and the recipient of the actions (the woman and the lawyer). A more complex sentence type is one in which the meaning is expressed with a subject-relative embedded clause, such as “The man, who insulted the woman, hired a lawyer.” In this case, the agent and the recipient retain their thematic roles, but the main clause (The man hired a lawyer) is interrupted by a relative clause (who insulted the woman). It is possible to determine who hired the lawyer, but it requires a bit more cognitive effort.
An even more complex syntactic form is one in which the meaning is expressed using an object-relative embedded clause, such as “The woman, who the man insulted, hired a lawyer.” In object-relative clause sentences, the embedded clause not only interrupts the main clause, but the head noun phrase (The woman) functions as both the agent of the main clause (who hired a lawyer) and the recipient of the relative clause (who the man insulted). Because the thematic roles in object-relative clause sentences are not canonical and require more extensive thematic integration, they are more difficult to comprehend than subject-relative clause or active-conjoined sentences.
In addition to this purely syntactic manipulation, one can also independently and orthogonally manipulate demands on working memory by separating the major elements of the sentence by either a short three-word adjectival phrase (e.g., The woman with blond hair, who insulted the man, hired a lawyer) or a longer six-word adjectival phrase (e.g., The woman with blond hair and dark glasses, who insulted the man, hired a lawyer). Numerous behavioral studies have shown these factors to have a significant and graded effect on comprehension errors and processing times in young (e.g., Just et al. 1996) and older (Fallon et al. 2006; Wingfield et al. 2003) adults.
Grossman et al. (2002a) presented sentences that embodied these principles to young and older adults while the subjects were in the bore of an MRI scanner. The subjects' task was simply to answer a probe question about who performed the action described in the sentence with each of the sentences randomly including a male or female as the agent (see Cooke et al. 2002 for the specific stimulus sentences used and a more detailed account of their features). The behavioral data showed that although on average sentence-comprehension accuracy was poorer for older adults relative to young adults, there was much broader variability in the comprehension performance of the older adults compared with their younger counterparts. This suggested two groups of comprehenders among these otherwise healthy elderly adults.
One group of elderly subjects was as accurate as their younger peers in their sentence comprehension. We refer to these as the good comprehenders. A second group showed significantly impaired sentence comprehension relative to the young adults. These we refer to as the poor comprehenders. The poor comprehenders were quite similar to the young adults in comprehending simpler sentences. It was when sentences were more syntactically complex and especially challenging to working-memory resources that the poor comprehenders separated themselves from the young adult subjects and the good comprehender elderly group. As we will describe, Cooke, Grossman, and their co-workers observed distinct brain activation profiles for the young subjects and for the good and poor comprehender elderly groups while comprehending easier and more difficult sentences.
PATTERN FOR GOOD ELDERLY COMPREHENDERS.
Consider first the contrasts of brain activation in young adults versus the older adults whose comprehension accuracy was relatively close to that of the young adults. The top two brain renderings in Fig. 3A were created by subtracting the activation levels measured in the various brain regions for the good comprehender older adults from the young adults' activations. Subtracting in this direction shows that the young adults were producing a significantly greater degree of activation than the older adults in the posterolateral temporal-parietal cortex in the left hemisphere. This region is thought to support a short-term auditory-phonological buffer that retains information transiently during the course of processing (Chein and Fiez 2001; Jonides et al. 1998). This region also includes the core left posterior sentence processing component. (These renderings follow the usual convention of representing larger activation differences with brighter colors.)
In contrast to this area in which the elderly good comprehenders showed less activation than the young adults during sentence comprehension, the compensation hypothesis would lead one to expect to see the successful older adults recruit other brain regions to maintain their successful performance. This contrast can be revealed by subtracting the activation levels in the young adults from the older adults' activations. The consequences of this subtraction are shown in the two brain renderings in Fig. 3B, where one sees these successful older adults showing significant upregulation in two areas. One of these is increased activity in the dorsal portion of left inferior frontal cortex. This area is thought to be important for maintaining and rehearsing stored verbal information in working memory (Chein and Fiez 2001; Smith et al. 1998) and is anatomically and functionally distinct from the ventral portion of left inferior frontal cortex (Amunts et al. 1999) that appears to be more strongly associated with grammatical aspects of sentence processing (Cooke et al. 2002, 2006).
In addition to this increased activation in the left hemisphere, one also sees in Fig. 3B that the successful older adults showed additional activation in the right posterolateral temporal-parietal region. The young adults also showed increased activation in this area but only for sentences that especially stressed working-memory demands (Cooke et al. 2002). For the older adults, these areas of activation were evident for all types of sentences.
These findings for the elderly good comprehenders are thus largely consistent with the compensatory hypothesis in so far as left dorsal inferior frontal and right temporal-parietal regions are upregulated to compensate for the reduced left temporal-parietal activation they show relative to the young adults. They also illustrate the involvement of a network of cortical areas upregulated in support of sentence processing beyond the left perisylvian core sentence-processing area. In particular, the network upregulates by augmenting working-memory resources needed to support sentence processing.
PATTERN FOR POOR ELDERLY COMPREHENDERS.
It is important to affirm that the activation pattern described in the preceding text was operating in a compensatory fashion in support the above older adults' good comprehension. That this is likely the case can be seen in Fig. 3, C and D, in which the patterns of activation of the elderly good and poor comprehenders are contrasted (Grossman et al. 2002b).
C represents the result of subtracting the activation pattern observed while the elderly poor comprehenders were engaging in the sentence-comprehension task from the activation pattern of the elderly good comprehenders. This subtraction shows that the poor comprehenders engaged significantly less activation of the core sentence-processing areas in inferior frontal cortex and posterior-superior temporal cortex of the left hemisphere relative to the good comprehenders. It should be noted that the reduced activation in this area in the poor comprehenders did not interfere with their comprehension of grammatically simpler sentence forms; as previously noted, their performance deficit relative to the young adults and the elderly good comprehenders appeared only when the sentences were more syntactically complex.
In contrast with this area of reduced activation, subtracting the activation pattern of the poor comprehenders from that of the good comprehenders, shown in Fig. 3D, revealed the poor comprehenders to be upregulating dorsolateral prefrontal cortex to a degree not shown by the good comprehenders. Dorsolateral prefrontal cortex is often activated during general problem-solving activities, regardless of the verbal or nonverbal nature of the material (Paulus et al. 2001; Prabhakaran et al. 2001; Ramnani and Owen 2004). This prefrontal region is not activated in young adults during sentence-comprehension tasks. This would suggest that the older adults who are less successful in their comprehension of grammatically complex sentences were attempting to understand these sentences by adopting a general-purpose problem-solving approach to the task.
Adopting this general problem-solving approach may not hinder comprehension of grammatically simpler sentences because their interpretation is quite straightforward. However, treating a grammatically complex sentence like any kind of cognitive challenge that requires planning and mental manipulation, rather than engaging specialized linguistic processes to understand the specific grammatically determined features of a complex sentence, appears to undermine comprehension that depends on complex grammatical processing. These findings resonate in an interesting way with suggestions that dealing with complex syntax engages prefrontal cortex, including vIFC, in controlling selection among competing parsing representations as would be the case when noncanonical sentences are encountered (Novick et al. 2005).
The findings for the poor comprehending elderly are also generally consistent with the compensatory hypothesis because, like those for the elderly good comprehenders, they too illustrate a change in the characteristic brain activation pattern in response to a cognitive challenge. However, we see here the interaction of biological and cognitive factors in a modified compensatory approach to their sentence processing. This elderly group appeared to have developed a specific cognitive strategy during sentence processing in their attempt to compensate for their age-related difficulty. As we saw, this strategy activated dorsolateral prefrontal cortex, a cortical distribution normally activated for general-purpose problem-solving but not typically activated during sentence processing. As we also saw, however, this cognitive strategy and the associated brain activation pattern were unfortunately not sufficient to yield fully successful sentence comprehension.
Extending the modified compensatory hypothesis: the temporal course of activation during comprehension of grammatical features that vary in resource demands
The interaction of biological and cognitive compensation can be further evidenced by examining sentence processing with a task that manipulates resource demands in a more subtle manner. In this study, we assessed executive resources inherent to more difficult grammatical features of a sentence. This is based on the seminal electrocortical event-related potential (ERP) work of Angela Friederici and her colleagues (Friederici 1995; Friederici et al. 2003). Friederici and her co-workers showed a change in the anatomic distribution of the ERP depending on the resource demands of the grammatical elements being judged in a sentence. These investigators also demonstrated changes in the temporal course of processing, with the ERPs for resource-demanding materials extending into a later time window than with less demanding sentences.
In a study of young adults, Cooke et al. (2006) pursued the neuroanatomic substrate and time course associated with processing three different kinds of grammatical features in a sentence and the presumed working-memory demands needed to sustain success. The task was a grammatical acceptability test in which the subject was required to judge the coherence of sentences that did or did not contain a grammatical violation. One feature examined was the inflectional form of the past participle, where in some sentences the morpheme –ed was omitted (e.g., “The test is being explained/explain to excited new college students.” (The incorrect form in this and subsequent examples are given in bold type.) With such sentences one may also add to the working-memory demands by inserting additional wording, such as “The test is being vaguely yet slowly explained/explain to excited new college students.” (The words used to increase the working-memory resources needed for the sentences are shown in italics.)
More demanding in terms of working-memory demands would be acceptability judgments for sentences containing a noun-verb substitution (e.g., “The dance is being not too seriously rehearsed/rehearsal prior to the debut performance.”). The final type of sentence contained a transitivity violation; a sentence containing a verb that cannot be expressed in a passive form because the verb is intransitive (e.g., “The woman is being brazenly and rudely harassed/winked by the building workers outside.”). Identifying transitivity violations are thought to involve relatively prominent working-memory demands relative to the required linguistic operations.
Examination of activation patterns for young adults during correct grammaticality judgments in an early time window showed left inferior frontal recruitment. Whereas less-demanding judgments activated vIFC, more-demanding judgments recruited both ventral and dorsal portions of IFC. More-demanding grammaticality judgments were also associated with right inferior frontal activation during this early time window (Cooke et al. 2006). Our question was whether fMRI could be used to look at activation patterns appearing during a later time window, paralleling the ERP studies of Friederici and her co-workers. This would allow us to establish whether the age-related compensatory mechanisms discussed in the preceding text play out both in terms of their anatomic distribution and in terms of their temporal course.
Although temporal resolution of fMRI is poor, we were nevertheless able to monitor BOLD activity levels 2 s later than the point at which the BOLD hemodynamic response is typically monitored. The imaging findings seen during this later time window are illustrated in Fig. 4. For the young adults, the relatively simple grammaticality judgment with few resource demands (recognizing -ed violations) resulted in no late activation. By contrast, these young adults showed suprathreshold activation in the late time window when judging the two more resource-demanding grammatical features. This is illustrated in Fig. 4. The activation in the late time window for young subjects can be seen in A for recognition of noun-verb substitutions and in B for recognizing transitivity violations. In both cases, the young adults sustained into the later time window left inferior frontal cortex recruitment. As can be seen, there was no increased right hemisphere activation relative to the simpler grammatical judgment. This result for the young adults is consistent with the ERP findings of Friederici and her colleagues.
Although the young adults showed activation in the late time window only for resource-demanding grammatical judgments, older adults tested with the same sets of materials show a different pattern. One aspect of this difference appears in Fig. 4C which shows older adults upregulating the ventral portion of left inferior frontal cortex even for the simpler grammatical feature in the late time window, activation that appeared for young adults only with more challenging grammatical judgments. To the extent that this age-specific left hemisphere activation appearing in the late time window reflects sustained grammatical processing over time for a grammatical agreement with fewer resource demands, it is consistent with the compensatory hypothesis rather than the dedifferentiation hypothesis.
When older adults were faced with more resource-demanding grammatical challenges that produced increased activation in only left hemisphere regions in young adults (Fig. 4, A and B), the older adults showed in addition right frontal activation in the later time window for noun-verb substitutions (Fig. 4D) and transitivity violations (E). Activation of right inferior frontal cortex in the late time window thus may be related to the cognitive strategy of sustaining working-memory resources over time to support decisions about the grammaticality of resource-demanding sentence features, consistent with the modified compensatory hypothesis. These data thus extend the compensatory hypothesis in the temporal dimension.
Language comprehension is not unique in showing changing patterns of neural activation in older adults that compensate for what we have argued would otherwise be far more dramatic performance declines. Other cognitive domains, such as working memory, recall from episodic memory, and memory-dependent response inhibition, also appear to show compensatory changes in older adults. These other functions, however, still show markedly poorer levels of behavioral performance in older adults compared with young adults (e.g., Cabeza et al. 2004; Gutchess et al. 2005; Neilson et al. 2002). Two points need to be made in this regard. The first is that age differences in language comprehension do appear when older adults are faced with rapid speech rates or when challenged by grammatical forms that put considerable pressure on working memory. Nevertheless, language retains a special place among preserved functions in normal aging. The question thus remains as to why compensatory activations are enough to preserve many aspects of language performance but not enough to preserve equal levels of performance in these other domains. We have attempted to address this question by suggesting that sufficiently flexible neural and cognitive resources can conspire to circumvent much age-related change to maintain access to the goal of communicative efficacy.
This compensation is driven by the fact that language holds a special place in human interaction. Language emerges early in childhood, requires no formal training despite its complexity, and is present in every society on earth. Where spoken language is not possible, sign languages mediated by manual movements have developed to express objects, actions, and their syntactic relations. In sum, the human brain is “built” to do language, it is an activity continuously practiced across the life span, and no less important, it is carried by a rich network of brain regions that allow sufficient compensation to operate at a remarkably good level.
Functional neuroimaging has over recent years brought converging evidence to human brain lesion studies by expanding the picture beyond just the role of the left perisylvian region for language processing. This is crucial because it appears that nontraditional language-related brain areas are upregulated during healthy aging to maintain performance in the face of the neuronal decline that accompanies normal aging.
Beyond confirming the importance of the traditional left hemisphere language regions centering on Broca's and Wernicke's areas and their interconnection, functional imaging studies conducted in our laboratories and elsewhere have identified a large-scale neural network that is activated during the course of sentence processing. We characterize this network in terms of a core perisylvian sentence-processing component together with both left and right hemisphere extrasylvian cortical regions that support the executive and working-memory resources necessary to process complex sentences.
These studies offer a window on the strategic recruitment of critical brain regions by older adults in response to otherwise limited working-memory resources not seen during sentence comprehension in young adults. We have shown that older adults augment regional brain activation within the language-processing system both spatially and temporally to help compensate for age-related neuronal changes. Moreover, healthy older adults appear to develop cognitive strategies to help compensate for the age-related decline in the neuronal substrate for sentence processing. This compensation for age-related changes in neuronal mass and efficacy of neuronal transmission addresses in human cognitive function the fundamental question (Prince et al. 2004) of how stable behavior can be produced in spite of changes in underlying neural structures.
We have attempted to show how the aging brain can maintain a relatively high level of performance despite the biological changes associated with aging. The stability in performance is maintained by engaging novel brain regions in space and over time and implementing novel cognitive strategies in response to a constant cognitive goal, in this case, effective sentence comprehension. This highlights the adaptive and plastic nature of the neurobiology of brain function that supports optimizing goal attainment with available resources even when in older adulthood these resources may become diminished.
The authors' work is supported by National Institutes of Health Grants AG-04517, AG-19714, and DC-05432 to A. Wingfield and AG-17586, AG-15116, NS-35867, and NS-44266 to M. Grossman.
- Copyright © 2006 by the American Physiological Society