Effects of short-term memory deficits on comprehension

A surprising finding coming out of studies carried out during the late 1980s and early 1990s was that a phonological STM deficit appeared to have few consequences for sentence comprehension, even for complex sentences with a large intervening distance between the words that needed to be linked together (e.g., Butterworth, Campbell, & Howard, 1986; Caplan & Waters, 1999; Martin et al., 1994; Martin & Romani, 1994). Martin and Romani (1994) and Martin and He (2004), however, reported that a semantic STM deficit (but not a phonological STM deficit) was related to deficits in comprehending certain sentence types. Specifically, in a task in which subjects had to indicate whether sentences were sensible or anomalous, the patients had difficulty making those judgments for sentences in which several adjectives came before a noun (e.g., “rusty old red wagon” versus “rusty old red swimsuit”) or several nouns came before a verb (e.g., “glasses, vases, and mirrors cracked” versus “rugs, vases, and mirrors cracked”). They did much better when the adjectives followed the noun (e.g., “the wagon was old, red, and rusty”) or the nouns followed the verb (e.g., “the movers cracked the mirrors, vases, and glasses”). These authors argued that in the “before” conditions, individuals had to maintain the meanings of the adjectives or nouns in an unintegrated fashion until the respective noun or verb was processed, thus overloading their restricted STM capacity. In contrast, in the “after” condition, each adjective could be integrated as a modifier of the noun as it was heard and each noun could be integrated as an object of the verb as it was heard. The assumption was that once these integrations had been made, the STM representation was no longer needed. Given the poor performance of the patients with semantic STM deficits and the good performance of the patient with a phonological STM deficit, these authors argued that the retention of semantic information, rather than phonological information, was critical for maintaining the words prior to integration.

Effects of short-term memory deficits on production

A role for STM in comprehension may seem intuitively necessary, whereas a role in production may seem less obvious. However, in production, one does not typically retrieve one word and then utter it, retrieve the next word and utter it, and so on. Instead, speech is typically produced in fluent stretches, which implies some degree of advance planning of two or more words. This advance planning would presumably rely on STM resources to support the representations of several words prior to their articulation.

Martin and colleagues have reported evidence from aphasic patients suggesting that semantic STM capacity is critical for this advance planning in production. In one study, Martin and Freedman (2001) showed that patients with a semantic STM deficit had difficulty producing adjective-noun phrases (e.g., “long blonde hair”), tending to produce these utterances in piecemeal fashion (Figure 9-5). For example, for the target utterance of “small leaf”, patient AB produced, “It’s a leaf. It’s small.” These same patients did better when asked to produce the same information in a sentence frame (e.g., “The leaf is small”). A patient with a phonological STM deficit performed at a normal level for both utterance types. In a related study, Martin, Miller, and Vu (2004) showed that patients with a semantic STM deficit had difficulty producing sentences that began with a conjoined noun phrase (e.g., “the ball and the block moved”) compared to sentences that began with a single noun (e.g., “the ball moved”). A patient with a phonological STM deficit showed a latency difference between the two sentence types that was well within the range of controls. Martin and colleagues concluded that speakers plan at a phrasal level, attempting to plan all of the lexical-semantic representations for a phrase prior to beginning phonological retrieval for that phrase. The normal performance for the patient with a phonological STM deficit in the two studies might be attributed to the fact that there is very limited planning at the phonological level (which was within the capacity of the patient) or to a dissociation between input and output phonological retention (Martin et al., 1999). That is, there may be separate capacities for maintaining phonological representations involved in speech perception and those involved in speech production (Shallice & Butterworth, 1977; Romani, 1992). The patient studied by Martin and colleagues may have had a deficit on the input side but not on the output side (Martin et al., 1999).

Effect of short-term memory deficits on learning

Although a phonological STM deficit appears to have few consequences for sentence comprehension and production, it does have consequences for the learning of novel phonological forms (i.e., the learning of new words or names) (Baddeley, Papagno, & Vallar, 1988; Freedman & Martin, 2001). Freedman and Martin (2001) showed contrasting patterns of learning impairments for patients with phonological and semantic STM deficits. Those with phonological STM deficits had greater difficulty learning foreign translations of English words than in learning new meanings for known words. Those with semantic STM deficits showed the reverse pattern, indicating that semantic STM was needed to transfer new semantic information to LTM. Thus, there was a code-specific relation between short-term retention and long-term learning.

Interpretive versus postinterpretive aspects of sentence processing

Caplan and Waters and colleagues (see Caplan & Waters, 1999, for an overview) carried out several studies examining the role of WM in sentence comprehension. They demonstrated that both DAT and Parkinson disease patients had preserved phonological retention and rehearsal but had very reduced WM capacities as measured by tasks like the reading span task (e.g., a WM span of less than one in the case of the DAT patients). For both patient groups, they showed that the effect of syntactic complexity was no greater than for controls. However, both groups showed an exaggerated effect of the number of propositions in the sentence. (See examples of variations in syntactic complexity and number of propositions in Table 9-1.) Caplan and Waters (1999) explained these findings on the grounds that different WM resources support interpretive versus postinterpretive aspects of sentence processing. By interpretive processing, they refer to “the processes of recognizing words and appreciating their meanings and syntactic features; constructing syntactic and prosodic representations; and assigning thematic roles, focus, and other aspects of propositional and discourse-level semantics . . . ” (p. 78). By postinterpretive processing, they refer to “remembering the content of a sentence, using the meaning of a sentence to plan action, reasoning on the basis of sentence meaning” and other related processes (p. 79). For interpretive aspects, they assume a dedicated WM system for language processing that is not tapped by standard span tasks (either simple or complex). For postinterpretive processing, general WM resources are drawn upon. Thus, in order to accommodate their findings, they have to assume that variations in syntactic complexity affected interpretive processes whereas variations in number of propositions affected postinterpretive processes.

Working memory and executive function

A number of studies in the literature on healthy individuals have demonstrated a relation between WM capacity and the ability to allocate attention to external stimuli and internal representations and to inhibit attention to interfering material. For instance, there is a negative correlation between WM capacity and the size of the standard Stroop effect (Kane & Engle, 2003), which reflects the degree of interference in naming the ink color of a written color word (e.g., saying “green” to the word “red” written in green ink). Attention allocation is thought to be one aspect of executive function (EF), which involves coordinating goal-directed behavior. A recent study by McCabe, Roediger, McDaniel, et al., (2010) found a very high correlation between factor scores for WM (derived from several measures of complex span) and EF (derived from several standard EF tasks such as the Wisconsin Card Sorting task).

These findings suggest that deficits in WM might be related to deficits in aspects of EF. Hamilton and Martin (2005, 2007) presented evidence that an aphasic patient (ML) who showed a semantic STM deficit had difficulty inhibiting verbal information, but not nonverbal information. For instance, he showed a verbal Stroop effect that was many standard deviations outside the normal range but showed an effect within normal range on a spatial analogue to the Stroop effect. On a recognition probe task (that is, a task in which subjects decide if a probe word matches an item in a previous list), ML showed highly exaggerated interference effects if the probe matched an item in a previous list or was even phonologically or semantically related to a previous list item. Hamilton and Martin (2005) proposed that the patient had a specific difficulty in inhibiting irrelevant verbal representations. A study by Biegler, Crowther, and Martin (2008) showed that this patient and another with a semantic STM deficit had exaggerated difficulty in repeatedly naming items from the same semantic category. They suggested that this repeated naming from the same category led to an overactivation of all of the terms in the category. Difficulty in inhibiting the incorrect names led to increasingly longer naming latencies for same-category items relative to different category items. They hypothesized that the left inferior frontal gyrus (Figure 9-6), which was affected in these patients, is involved in verbal inhibition. A deficit in inhibition could cause difficulty in spontaneous speech, where one needs to inhibit what has already been talked about to produce the next words.

Recently, Barde et al. (2010) have offered another interpretation of the interference effects demonstrated by ML in the recognition probe task described earlier (Hamilton & Martin, 2007). They suggest that the effects can be accounted for by reactivation of the prior list item by the rhyming or semantically related probe word and a rapid loss of the semantic or phonological features of the probe that makes it difficult for the patient to verify whether the item matches the probe or not. They present data from 20 aphasic patients with varying degrees of semantic and phonological STM deficits. Consistent with their hypothesis, they show that the degree of phonological interference is accounted for by the degree of phonological STM deficit and the degree of semantic interference is accounted for by the degree of semantic STM deficit. It is unclear how their hypothesis can account for ML’s difficulties on the Stroop task (or more recent data indicating an exaggerated picture-word interference effect; Biegler et al., 2008). However, it is possible that these are simply co-occurring deficits for ML and not causally related to the STM deficit. Only the investigation of a large number of patients on STM as well as single word tasks involving interference can address this issue.

Summary of short-term memory and working memory deficits and their role in language processing

Reduced STM span, as often observed in aphasia, has few consequences for language comprehension and production if the span deficit reflects difficulty maintaining phonological information. In contrast, a deficit in maintaining semantic information causes difficulties for both comprehension and production. Phonological STM is important, however, for learning new phonological information; similarly, semantic STM is important for learning new semantic information. There is controversy whether semantic STM deficits reflect a rapid loss of semantic information or a deficit in various EFs (such as an inhibition deficit). Deficits in more complex WM appear to affect the ability to carry out what have been termed postinterpretive aspects of sentence processing rather than more on-line interpretive aspects (Caplan & Waters, 1999).

Semantic dementia

Semantic dementia is a subtype of frontotemporal dementia (and a further subtype of primary progressive aphasia) that results from atrophy of the anterior temporal lobes (see Figure 9-6) bilaterally, though often the atrophy is greater in the left hemisphere (Mummery, Patterson, Price, et al., 2000). Patients suffering from semantic dementia show a progressive loss of both productive and receptive vocabulary, stemming from a loss of semantic knowledge (Hodges & Patterson, 2007; Snowden, Goulding, & Neary, 1989; Warrington, 1975). Semantic dementia patients typically first lose knowledge of fine-grained features of concepts while retaining knowledge of more common or typical features. Thus words tend to be replaced with more general terms (e.g., “animal” instead of “camel”) and less typical features are omitted when drawing (e.g., if asked to draw a camel, a semantic dementia patient would likely include a head, two ears, and four legs, but omit distinctive features such as the hump). This correspondence between word production and drawing suggests that the loss of semantic knowledge in semantic dementia is multimodal, as is further evidenced by a corresponding loss of knowledge of how to use objects in the later stages of the disease (Bier & Macoir, 2010; Hodges, Bozeat, Lambon Ralph, et al., 2000). Semantic dementia patients also typically develop symptoms of surface dyslexia, where irregularly spelled words (e.g., yacht) are read as if they were regular (Woollams, Lambon Ralph, Plaut, & Patterson, 2007) demonstrating the role of semantic memory in processing of exception words (i.e., words with irregular grapheme-to-phoneme correspondences).

Semantic dementia has been particularly influential because the deficit seems to be limited to semantic memory. Despite their severe semantic memory impairment, semantic dementia patients seem to have preserved syntactic processing, phonology, calculation ability, and drawing skills (at least in the earlier stages; Ash, Moore, Antani, et al., 2006; Hodges et al., 1999). Semantic dementia contrasts with progressive nonfluent aphasia, another form of frontotemporal dementia, which shows essentially the opposite pattern: impaired syntactic and phonological production in the context of spared semantic knowledge (Grossman, 2010). Importantly, patients with semantic dementia also appear to have relatively preserved episodic memory (i.e., memory for events, though the semantic memory deficits can make this difficult to demonstrate; Hodges & Patterson, 2007).

The amodal nature of the semantic memory deficit in semantic dementia suggests a relatively general role of the anterior temporal lobes in semantic memory. One recent proposal along these lines is that the anterior temporal lobes serve as a “hub” that integrates distributed modality-specific semantic information (Jefferies, Baker, Doran, & Lambon Ralph, 2007; Jefferies & Lambon Ralph, 2006; Lambon Ralph & Patterson, 2008; Patterson, Nestor, & Rogers, 2007; Rogers, Lambon Ralph, Garrard et al., 2004; see also Damasio, 1989). Some of the major motivations for this type of theory come from differences between patients with semantic dementia and stroke patients with semantic memory deficits.