The N400 in a semantic categorization task across 6 decades

Abstract

Objectives: To characterize the effects of normal aging on the amplitude, latency and scalp distribution of the N400 congruity effect.

Methods: Event-related brain potentials (ERPs) were recorded from 72 adults (half of them men) between the ages of 20 and 80 years (12/decade) as they performed a semantic categorization task. Participants listened to spoken phrases (e.g. `a type of fruit' or `the opposite of black') followed about 1 s later by a visually-presented word that either did or did not fit with the sense of the preceding phrase; they reported the word read and whether or not it was appropriate. ERP measurements (mean amplitudes, peak amplitudes, peak latencies) were subjected to analysis of variance and linear regression analyses.

Results: All participants, regardless of age, produced larger N400s to words that did not fit than to those that did. The N400 congruity effect (no-fit ERPs−fit ERPs) showed a reliable linear decrease in the amplitude (0.05–0.09 μV per year, r=0.40) and a reliable linear increase peak latency (1.5–2.1 ms/year, r=0.60) with age.

Conclusions: In sum, the N400 semantic congruity effect at the scalp gets smaller, slower and more variable with age, consistent with a quantitative rather than qualitative change in semantic processing (integration) with normal aging.

Introduction

Since development and aging span a lifetime, there is good reason to look at every brain and psychological function through the perspective of age. Indeed we believe this is unavoidable if our ultimate goal is to understand the mind–brain link. In this report we examine the effects of normal aging on semantic memory use. By semantic memory we refer to the repository of knowledge that one has about words, objects and events in the world, knowledge that supports categorization, inference and aspects of natural-language processing. It is generally believed that semantic memory is relatively unaffected by normal aging, in contrast to the age-related decline in retrieval of memories for particular episodes in an individual's life. And, in fact, when the integrity of semantic knowledge is inferred from performance on vocabulary tests, there is not only little evidence of age-related decline but even occasional reports of an age-related increase in vocabulary size (e.g. O'Dowd, 1984; Bowles and Poon, 1985). However, this apparent equivalence in vocabulary scores across the ages may at times mask qualitative differences in the nature of the definitions that younger and older subjects provide. For example, younger adults tend to define a word with a single good synonym whereas older adults are more likely to give a less specific and more verbose description (e.g. Botwinick et al., 1975).

Other estimates of whether or not and, if so, how semantic memory organization and access might change with age have been based on variants of so-called semantic-priming paradigms. The most common among these are the lexical decision, word and picture naming, and semantic categorization tasks. Methodological differences notwithstanding, these paradigms are similar in showing that context¹ can have a facilitatory effect on task performance; the accuracy, the speed, or both, of a task decision are generally improved when a (target) word is preceded by a semantically or associatively related (prime) word or a congruent sentence fragment as opposed to an unrelated word or an incongruent sentence fragment (see Neely (1991)for a review). This facilitatory effect (priming) of reaction times (RTs) by an appropriate context is the norm whether a subject is 20 or 70. However, there is some question as to whether the size of the semantic-priming effect remains constant across the adult lifespan; the results are inconsistent. Most investigators have found that semantic priming² remains stable across the adult lifespan, even in the presence of deficits in explicit memory (Byrd, 1984; Cerella and Fozard, 1984; Bowles and Poon, 1985; Rabinowitz, 1986; Burke et al., 1987; Balota et al., 1992; Mueller et al., 1997). Others, however, have observed significant age-related changes in the size of semantic-priming effects, in some cases smaller and in other cases larger than normal (e.g. Petros et al., 1983; Howard et al., 1986; Chapman et al., 1994; for meta-analysis see Laver and Burke, 1993; Nyberg et al., 1996). These discrepancies are difficult to tease apart, as a number of factors, such as the duration of the interval between the context (prime) and target stimuli, the proportion of related to unrelated items, and the strength of the relationship between the prime and target modulate the automatic and attentional contributions to the overall priming effect.

On the one hand, it is generally assumed that semantic information remains largely intact across the lifespan, i.e. that the structure of the semantic network is not qualitatively altered by normal aging. On the other hand, the literature also suggests that the speed with which semantic information is utilized (including access and integration), whether consciously or unconsciously, is significantly slowed with advancing age. Elderly volunteers typically show slower response times in lexical decision, naming, categorization, and various verbal fluency tasks (e.g. Birren and Fisher, 1995). In fact, some researchers have suggested that the disproportionately greater than normal amount of semantic priming sometimes observed in elderly populations is an artifact of the slower response times and that it vanishes when overall slowing is factored out (see Milberg and Blumstein (1981)for a similar argument in aphasic patients). In any case, to date there is no consensus as to whether this effect is merely a consequence of general, domain non-specific cognitive and motoric slowing or specific to semantic or verbal processing.

We decided to examine aspects of this issue using a measure of both brain activity and cognitive processing that is sensitive to semantic relations and does not require a motor response. Specifically, we chose to measure the electrical activity recorded from the human scalp synchronized to target events that need to be categorized relative to a semantic context. These time-locked recordings known as event-related brain potentials (ERPs) provide a multidimensional (amplitude, latency, topography) measure of the difference between two or more experimental conditions – in this case, a word that has been semantically primed and one that has not.

A number of investigators have recorded ERPs from college undergraduates as they rendered decisions about category membership (e.g. Fischler et al., 1983; Fischler et al., 1985; Boddy, 1986; Neville et al., 1986; Kounios and Holcomb, 1992; Heinze et al., 1998). A finding common to all these studies is the pattern of ERPs for words that are actually category members versus those that are not. Regardless of the type of response required (including none), ERPs elicited by a category non-member are more negative between 200 and 600 ms than those elicited by a category member. This negative difference, termed the N400 effect, typically has a posterior (centro-parietal) maximum, and shows a slight amplitude and durational asymmetry favoring the right hemisphere.³

The N400 component of the ERP was first described in response to semantically-anomalous words within isolated sentences or within written text (Kutas and Hillyard, 1980; Kutas and Hillyard, 1983). Subsequent experiments demonstrated an N400 to every word in a sentence, as well as in response to letter strings (words and pseudowords) in word lists or wordpairs (for review see Kutas and Van Petten, 1994). N400 amplitudes have been found to vary with a number of factors: (1) more predictable words elicit smaller N400s than less predictable words, (2) open class words elicit larger N400s than closed class words, (3) low-frequency words elicit larger N400s than high-frequency words, especially for words occurring for the first time, (4) words occurring at the beginnings of sentences elicit larger N400s than those occurring near the ends of sentences, (5) words occurring for the first time in an experiment have larger N400s than upon repetition, although this effect interacts with the similarity of the context within which the various presentations occur and the interval between the repetitions, (6) words preceded by a semantically or associatively related word elicit smaller N400s than words preceded by unrelated words and (7) real words and pseudowords (e.g. orthographically legal and pronounceable non-words) elicit larger N400s than do non-words (i.e. orthographically illegal, unpronounceable letter strings). Thus, it seems that both lexical and contextual (especially semantic) factors influence the amplitude of the N400 component.

A critical finding was the dissociation of the truth or falsity of a statement from the semantic relation between its subject and predicate (e.g. Fischler et al., 1983; Fischler and Raney, 1991; Kounios and Holcomb, 1992). Specifically, Fischler et al. (1983)demonstrated that large N400s were elicited by the final words of propositions whenever the subject and predicate on N400 amplitude were semantically unrelated, regardless of the validity of the sentence. Subsequent work from Fischler's laboratory revealed that if all the stimulus sentences were affirmative, then sentence validity (i.e. sentence-level constraints) did modulate the size of the N400 somewhat. Nonetheless, the lexical relation between the subject and predicate remains a powerful factor in determining the overall amplitude of the N400 in such short sentences.

In summary, across several laboratories N400 amplitude has been found (1) to vary with a number of different semantic variables (Kounios and Holcomb, 1992) and (2) to be relatively insensitive to decision-related and response-selection mechanisms (e.g. Heinze et al., 1998) that do, however, affect the speed of the eventual response (response time). Thus, it has been suggested that the N400 can be used to ask questions about the organization of semantic memory and how it is utilized in such tasks, irrespective of task demands and specific decision-related effects. It is in this vein that we examined the hypothesis that, insofar as age-related changes are observed in certain semantic-categorization or semantic-priming tasks, the young–old differences arise from differences in access to or decisions about the memory rather than its structure per se. To this end, we studied the effects of normal aging on semantic categorization via ERP recordings.

Although most of what we currently know of the N400 and the factors that modulate it is based on work with college undergraduates, a few studies provide some data on the changes in N400 with development and late-life aging. Holcomb et al. (1992)reported on a large study of individuals between 5–26 years of age as they heard or read sentences that either ended with a highly-expected ending or a semantically-anomalous word (in the reading version, the minimum age was 7). While all age groups showed an N400 effect in both modalities, the size of the effect was inversely correlated with age. The results were interpreted as consistent with other reports showing that as children acquire good language skills they rely less on semantic context (but see also Juottonen et al., 1996; Gonzalez-Garrido et al., 1997).

Harbin et al. (1984)were the first to examine the effects of aging on ERPs in a categorization task. They recorded ERPs from 3 midline locations of a younger (mean age 21 years) and an older group (mean age 71) making decisions about the 5th of a series of visually-presented words in an Identity and a Category condition. In the Category condition, the first 4 words all belonged to the same semantic category and subjects were asked to indicate whether or not the 5th word also belonged. Given their interest in the P3 component of the ERP and the sensitivity of the P3 to event probabilities, the 5th word matched the previous 4 on only 15% of the trials. Both reaction times and the latency of the N400 to mismatches were noticeably longer in the older group, although the difference N400 (mismatch−match ERP) reportedly peaked at approximately 540 ms for both groups. Gunter et al. (1992)compared the ERPs elicited by congruous and incongruous endings of sentences of medium-to-high contextual constraint from a group of young students with those from a group of highly-educated middle-aged academics (mean age 55 years). The N400 effect in the older group was both delayed in latency (by 120 ms) and reduced in amplitude. These basic findings have been replicated by others in small groups of middle-aged and elderly individuals (Gunter et al., 1995; Gunter et al., 1996; Ford et al., 1996).

The present experiment was aimed at replicating and extending these findings to men and women from 20 to 80 years of age. Congruent and incongruent words were flashed visually after the context (category name) was spoken by the experimenter. We chose this paradigm with auditory context and visual target word because we have found that it is a task that can be performed by patients of various mental capabilities⁴ and that it produces a robust N400 in response to semantically-incongruent words (see Kutas et al., 1988; Iragui et al., 1996).

Given the known reduction in N400 amplitude with increasing strength of semantic association and its inverse correlation with degree of contextual constraint, we thought it useful to make a similar comparison in this simpler semantic categorization task (Kutas and Hillyard, 1984; Kutas and Hillyard, 1989; Kutas et al., 1984; Van Petten and Kutas, 1991a; Van Petten and Kutas, 1991b; Van Petten et al., 1991). To that end, the stimulus set included approximately equal numbers of highly constraining antonym contexts (e.g. `the opposite of black') for which there is only one reasonable outcome (e.g. `white') and moderately constraining category contexts (e.g. `a piece of furniture') for which there are several reasonable alternatives (e.g. `table', `chair', `couch', `cabinet', etc.). Antonymic and categorical relations also map onto the distinction that has been made between semantic and associative priming (for review see de Groot, 1990) as well as the distinction between a prediction-based versus an expectancy-based strategy for utilizing contextual information, respectively (e.g. Becker, 1980; Becker, 1982). Thus, while we expect to obtain large N400 effects for both stimulus types, the proposed differences in the lexical and/or contextual mechanisms underlying semantic and/or associative priming with antonyms and categories would predict some ERP differences as well.