Semantic processing of spoken words in Alzheimer's disease: an electrophysiological study.

INTRODUCTION

One of the principal cognitive deficits associated with AD is the impaired performance of AD patients in a variety of semantic tasks. The patients have difficulties in word finding (poverty of content words in spontaneous speech, decreased ability to generate instances of a given category, and difficulties in object naming and naming to definition), deficient knowledge of concept meaning (loss or disrupted organization of attribute knowledge), and anomalies in the effects of semantic context (e.g., abnormal semantic priming in lexical decision)(Chertkow & Bub, 1990; Chertkow, Bub, & Seidenberg, 1989; Chertkow et al., 1994; for a review, see Nebes, 1989). It is unclear whether these impairments are due to a genuine disruption of semantic memory or whether they merely indicate a difficulty in effortful, voluntary access to semantic representations (Bayles, Tomoeda, Kaszniak, & Trosset, 1991; Chan et al., 1993; Hodges, Salmon, & Butters, 1992; Nebes, 1994). Recent studies show that AD patients probably have a combined access and storage deficit. Daum, Riesch, Sartori, and Birbaumer (1996) used a variety of behavioral tasks such as naming, generating definitions, real/unreal object decision, and forced-choice questions about the appearance and functional attributes of objects and animals. AD patients were severely impaired on both verbal and visual tasks and disproportionately impaired on items concerned with animate objects. The authors interpret their findings as indicative of disruption at the structural level of visual representation where the visual features and spatial relations of objects are represented and at the phonological output level where object names are stored for speech production. However, because the patients performed relatively well in some tasks not requiring effortful processing, the authors conclude that semantic networks are partly preserved in AD and that the patients can access semantic knowledge to some degree in an indirect way. Laatu, Portin, Revonsuo, Tuisku, and Rinne (1997) investigated whether AD patients are able to get voluntary access to semantic representations if the retrieval demands of a task are gradually eased without essentially changing the nature of the information to be retrieved. In some tasks the deficits persisted even in passive recognition and forced-choice tasks. The authors' conclusion was that AD patients have a generalized access deficit but some indications of a storage deficit as well.

These observations concerning the existence of semantic access and storage deficits in AD are exclusively based on behavioral measures of semantic function, which do not allow observation of cognitive processes while they occur, but only as they are inferred from overt behavior. However, ERPs provide us with a fresh methodological approach for the measurement of the on-line semantic processing of linguistic stimuli in brain-injured patients. Especially the N400 component of the ERPs is modulated by semantic factors (Kutas & Hillyard, 1980; for a review, see Hagoort & Kutas, 1994). The N400 is a negative-going wave between 300 and 600 msec poststimulus, and its scalp distribution is broad. N400 amplitude to a stimulus word is modulated by the semantic expectancy ("cloze probability") and congruity of the word with the preceding semantic context. The words most expected elicit a minimal N400, whereas the ones semantically incongruous and least expected elicit the largest N400 amplitudes. A standard interpretation of the N400 effect has been that it is sensitive to attention-demanding semantic integration processes, not automatic aspects of lexical access (Brown & Hagoort, 1993). Recent studies show that N400-like waveforms are elicited from pictorial as well as linguistic stimuli (Ganis, Kutas, & Sereno, 1996; Holcomb & McPherson, 1994; Nigam, Hoffman, & Simons, 1992; Pratarelli, 1994). These findings suggest that the N400 might largely reflect some kind of semantic search in a common modality-independent conceptual-semantic network. The issue between these two interpretations of N400 remains unresolved (Osterhout & Holcomb, 1995); assigning cognitive functions to ERPs is, in general, theoretically problematic (Rugg & Coles, 1995).

ERPs can be used to investigate the semantic processing of words in patients with linguistic or semantic deficits (Hagoort, Brown, & Swaab, 1996; Revonsuo & Laine, 1996). ERPs can provide us with information about language processing in real time and with great temporal resolution, and this information can then be used together with results from behavioral studies to construct, test, and constrain models of normal and deficient language processing (Hagoort & Kutas, 1994). Furthermore, ERPs can provide us with indications about the neural mechanisms of language processing, and they can be successfully measured even in the absence of any additional task beyond natural listening or reading. Hagoort and Kutas were the first to present results concerning N400 in AD patients. They refer to an unpublished preliminary study in which AD patients, age-matched controls, and young controls were tested in an N400 paradigm. The subjects were given a short spoken phrase (e.g., "A type of animal") after which the target word was presented visually. The target was either semantically congruous or incongruous with the phrase. Visual inspection of the ERPs to target words showed that the N400 effect (amplitude difference between ERPs to congruous and incongruous words) was smaller for the AD patients than for age-matched controls. The N400 was delayed and smaller for the elderly controls compared with the young. However, no statistical analyses of the effects were presented. In the study of Schwartz, Kutas, Butters, Paulsen, and Salmon (1996), the experimenter uttered the name of a category, which was followed after about 1 sec by a visually presented target word that was the written name of an object. Reaction times and ERPs were recorded. AD patients showed large behavioral priming effects (faster responses to related than to unrelated target words). The amplitude of the N400 effect (difference between ERPs to related and unrelated target words) was, however, much smaller in the AD group and peaked later than in the young and elderly control groups. All groups showed similar patterns of ERP responses to prime words at different levels of category (superordinate, basic, or subordinate). According to the authors, this result indicates that the structure of semantic representations in AD patients is preserved at least to some extent.

The N400 is elicited by semantically incongruous and unexpected final words in written (Kutas & Hillyard, 1980) and spoken (McCallum, Farmer, & Pocock, 1984) sentences. In a recent study involving AD patients (Hamberger, Friedman, Ritter, & Rosen, 1995), the final words of the sentence belonged to one of four stimulus types, which varied as a function of semantic relatedness to a highly expected word. The sentences and final words were presented visually to six AD patients and to old and young control subjects. The N400 amplitude varied as a function of semantic relatedness in young controls and AD patients. The elderly control subjects showed no such orderly variation but instead showed an equally large N400 to all kinds of unexpected final words. In a simultaneous behavioral task the AD patients had difficulties in distinguishing between two of the four stimulus types: the best completion and its semantic associations. Thus the patients' ERPs actually showed better discrimination of different stimulus types than their behavioral responses did. The authors interpret this as suggestive of a semantic access deficit that occurs in a time window after the N400 but before the behavioral response.

Ford et al. (1996) recorded ERPs to final words in spoken sentences. Surprisingly, the N400 amplitude was larger in AD patients than in elderly control subjects, both for unprimed and primed words. The N400 effect (difference between unprimed and primed conditions) was significant for both groups, but it appears to be smaller for the AD group (Ford et al., [ILLUSTRATION FOR FIGURE 1 OMITTED]). The authors suggest that the sentence stem did not adequately prime the final word of the sentence in AD patients and thus reflects a probable impairment in semantic knowledge. Their interpretation, however, is in contradiction with behavioral results (Nebes, Boiler, & Holland, 1986; Nebes & Brady, 1991) that show that a highly constrained sentence stem produces normal priming effects in AD patients when the task is to decide whether the target word is an …