Wendy C. West, Timothy O’Rourke, Phillip J. Holcomb
Tufts University

     It has become evident, particularly over the past decade, that a full understanding of cognitive processing in humans will not be forthcoming unless a better understanding of the neural "hardware" in which cognition takes place is available. Put another way, brain structure/function places very real restrictions on the types of perceptual and cognitive processes that are biologically plausible (Churchland & Sejnowski, 1992). Therefore, according to this view, the study of human cognition is critically dependent upon a Cognitive Neuroscience approach. Unfortunately, this task has been made difficult in the case of understanding the neurological underpinnings of language because of an inability to model this skill in animals, where brain structure/function relationships can be most easily studied through invasive procedures.

     The goal of this chapter is the familiarize the reader with the use of one type of Cognitive Neuroscience methodology, event-related brain potentials (ERPs), that is suitable for studying the relationship between cognitive and neural processes in intact human subjects. Although we briefly review studies of one ERP that has been associated with general intellectual or cognitive function (the P300), the major emphasis is on ERP measures of language comprehension. An overview of the ERP language processing literature is presented and is followed by a more detailed presentation of three representative experiments from our laboratory where we have focused on some of the processes involved in the comprehension of written words. Note in particular that we did not exhaustively review the literature on intelligence and ERPs. This is because of space limitations and because much of this literature is of questionable validity (e.g., see the discussion of ERP studies of individual differences below). Those interested can find information on this topic in a review by Kounios (1994).

Event-Related Brain Potentials

     Studies of brain-damaged patients and studies with normal subjects using modern imaging techniques (e.g., PET, SPECT, fMRI) have contributed important information regarding the anatomical organization of the language system and other cognitive processing systems. These methodologies do not, however, possess the high temporal resolution provided by electrophysiological recordings. In particular, ERPs complement well those techniques with high spatial but not temporal resolution.

     ERPs are voltage fluctuations in the ongoing electroencephalogram (EEG) that are time locked to sensory, motor, or cognitive events (Coles & Rugg, 1995). These voltage changes are manifested as positive and negative peaks (or components – see below) on the ERP waveform that are distributed across time. Since the fluctuations in voltage associated with an event are too small to be recognized in the ongoing EEG, the resulting ERP is derived by averaging the waveforms over many repetitions of the event. This serves to reduce the amplitude of the background EEG which is unrelated to the event (since it is different for each repetition). Meanwhile, the time-locked electrical activity associated with the event remains constant and is thus easily identified in the resulting average waveform.

     While early ERP studies tended to concentrate on quantifying entire waveforms, the last twenty years of research has tended to focus on particular regions or temporal windows of the ERP referred to as components. At the scalp a component is usually associated with a specific peak, which is typically bounded on either side by other peaks (see Figure 1). Components are traditionally labeled with a letter denoting their relative polarity (P or N) and either a peak latency value, denoting the time post-stimulus-onset at which the component peaks (e.g., N400 would be a negative component peaking at 400 msec), or a number denoting the ordinal position of the component (e.g., P3 would be the third positive-going component). A substantial body of research has demonstrated that specific components reflect activity in more-or-less distinct perceptual or cognitive systems (Rugg & Coles, 1995). One caveat that is important to keep in mind is that components, as seen in the scalp recorded ERPs, are somewhat distorted because they reflect the summation of activity from a number of distinct, temporally overlapping neural sources or generators. This so-called "component overlap" problem is assumed to be due to the brain being a highly parallel information processor. One partial solution to this problem is to, whenever possible, compare ERPs from conditions that isolate a single perceptual/cognitive process.

Figure 1. Plotted in this figure are two visual ERPs averaged over 16 subjects from the Cz (central-midline) electrode site. As in all of the figures in this chapter stimulus onset (time zero) is marked by the vertical calibration bar which also provides the scale for the y-axis (5 microvolts). Each x-axis tic mark represents 100 msec and the 100 msec prior to word onset was used as a baseline reference (i.e., the average in this period was subtracted from each post-stimulus point). Notice that it is customary in ERP research to plot negative voltages in the upward direction. The stimuli for these ERPs were the final words in sentences presented in a word-by-word fashion (so that ERPs could be isolated for each word in the sentence). The solid line represents ERPs evoked by final words that fit with the context of the sentence and the dashed line ERPs evoked by final words that were anomalous. The N400 is the negative-going deflection which peaks at 400 msec and is largest in the anomalous (dashed line) condition. The subsequent large positivity that peaks between 600 and 800 msec and can be seen in both conditions is probably related to the P3b. Both the N400 and P3b are thought to be endogenous components. The deflections prior to 300 ms, which include a small positivity at 50 msec (P1), a negative-going peak at 100 msec (N1 or N100) and a second positivity at 225 msec (P2 or P200), are examples of exogenous components. Notice that the manipulation of a linguistic variable (semantic congruency) did not differentially effect these early components.

     The components of the ERP waveform can be classified as either exogenous or endogenous. Generally, speaking exogenous, or stimulus bound, components tend to occur in the first 100 to 200 msec after the onset of a stimulus. With the possible exception of selective attention effects (see Mangun & Hillyard, 1995), characteristics of these early components are primarily determined by the physical nature of the eliciting stimulus and are relatively insensitive to changes in the psychological state of the subject (Rugg & Coles, 1995). In contrast, many of the later "endogenous" components are highly sensitive to changes in the state of the subject, the meaning of the stimulus, and/or the processing demands of the task (Rugg & Coles, 1995). It is these endogenous components, in particular the P300, that have been of primary interest to those studying intellectual functions.

     Finally, the spatial distribution of the ERPs across the scalp provides some information as to the location of the underlying neural generators. However, only limited inferences can be made about the locations of these sources due to several factors (Nunez, 1981, 1990). First, any spatio-temporal voltage pattern at the scalp could be the result of more than one configuration of sources. Second, smearing, or low-pass filtering, of the final voltage pattern results since the electrical currents are volume conducted through the brain and skull. Finally, the orientations of the source dipoles determine the pattern of activity detected on the scalp. So, for example, a potential recorded at a right hemisphere scalp site may actually have been generated in the left hemisphere.

     Despite these limitations in spatial resolution, ERP recordings have several distinct benefits. First, this technique is noninvasive; ERPs can be recorded from the intact scalp in normal subjects. It also yields multidimensional data points which provide measurements of latency, amplitude, polarity, and scalp distribution of the components of interest. Most importantly, the high temporal resolution (on the order of milliseconds) of the technique allows the study of the time course of neural activity underlying higher brain functions. This is particularly crucial for the study of language. Since language comprehension occurs in "real time", a rapid and continuous on-line measure is needed to delineate language processes as they unfold. Other dependent measures used in the study of language, such as reading time, decision-making time, or eye movement measurements, are "after-the-fact" or are restricted to a single point during comprehension. For this reason, electrophysiological recordings can provide extremely useful information for the study of sequential processing functions, such as language.

The P300 Component

     In 1965 Sutton, Braren, Zubin and John were the first to report on a positive-going ERP component with a peak latency of 300 msec. A substantial amount of subsequent research elaborated on the conditions that modulate both the amplitude and latency of this "P300" or "P3" (see Picton, 1992, for a recent review). Briefly, the amplitude of the P300 tends to increase as a function of at least two variables: declining stimulus probability or expectancy (e.g., Duncan-Johnson & Donchin, 1977) and increasing task relevance or value (e.g., Johnson, 1986). On the other hand, with the possible exception of response related processes, the peak latency of the P300 varies as a function of many of same factors that effect behavioral reaction time, (e.g., McCarthy & Donchin, 1981). This has lead some investigators to conclude that P300 latency is a measure of stimulus evaluation time.

     P300s have been reported for stimuli in all of the primary modalities. Most studies have reported that the P300 has a posterior (parietal maximum), bilaterally symmetric scalp distribution. However, there have been a number of reports suggesting that under certain circumstances a more anterior distribution can be found (e.g., Courchesne, Hillyard & Galambos, 1975). The more anterior distribution has typically been attributed to a separate, but related component, usually referred to as the P3a (in contrast to the more traditional and posterior P3b). P3a is usually largest to highly novel events suggesting its relationship with the orienting response. P3b on the other hand tends to be large to any low probability event (even the same event), as long as it is task relevant (Courchesne et al., 1975). (From here on we will use the label P300 in a more generic sense when referring to the "family" of positivities in this latency range, and will use the labels P3a and P3b when referring to the more specific sub-components.)

     While there is no consensus as to the nature of the psychological process(es) reflected by the P3b, most accounts suggest that it plays some role in memory. Perhaps the most widely quoted theory is that of Donchin (e.g., Donchin & Coles, 1988). He has proposed that the P3b reflects a ubiquitous process whereby working memory is updated -- events that require the most significant updating (low probability, meaningful events) are those that generate the largest P3bs. Moreover, as some events might be expected to take longer to integrate into working memory (e.g., more complex stimuli), the time course of this process is reflected in the latency of the P3b.

     Individual differences in the amplitude and latency of P300s have been reported in a large number of studies (see Picton, 1992). Factors influencing this component include type of psychopathology, age, personality, diet and, in a few studies, intelligence. For example, McGarry-Roberts, Stelmack and Campbell (1992) reported that the latency of the P3b were negatively correlated with IQ in a group of 30 women, while P3b amplitude had a more complicated relationship with IQ (it was larger in some cases and smaller in others for subjects with higher IQs). One problem with this correlational approach to studying intelligence and to much of the clinical research on individual differences using ERPs, has been a general lack of specificity of findings. In other words, virtually every disorder and condition has shown either a smaller or later P300. Such lack of specificity has greatly hindered the potential usefulness of this measure as either a research or clinical tool.²

The N400 Component

     The N400 component was first described by Kutas and Hillyard (1980). They reported a negative ERP component with a peak latency of 400 msec which was elicited by semantically inappropriate (or anomalous) terminal sentence words, but not by semantically appropriate terminal words (see Figure 1). The amplitude of the N400 component was larger for "strong" incongruities (e.g., "He took a sip from the transmitter") than for "moderate" incongruities (e.g., "He took a sip from the waterfall") and was absent for semantically appropriate but physically deviant (larger letter size) words. The latter elicited a late positive complex of waves (P560). Kutas and Hillyard reasoned that the N400 wave may be "an electrophysiological sign of the `reprocessing' of semantically anomalous information" (p. 203).

     Word expectancy and sentence context. The change in N400 amplitude with degree of semantic incongruity in the previous study suggests that the N400 may be sensitive to word expectancy. Kutas and her colleagues (Kutas & Hillyard, 1984; Kutas, Lindamood, & Hillyard, 1984) further examined the effect of word expectancy by varying within-subjects the cloze probability of sentence final-words (cloze probability of a word is determined by the proportion of subjects who fill in that word as the most appropriate ending for a given sentence). They found that N400 amplitude progressively increased as a function of decreasing cloze probability. Furthermore, N400 amplitude was larger for low probability (but appropriate) words semantically unrelated to the best completion word (e.g., "Don't touch the wet dog", where the best completion is "paint") than for low probability words semantically related to the best completion word (e.g., "He liked lemon and sugar in his coffee", where the best completion is "tea"). N400 amplitude was also larger for semantically anomalous words unrelated to the best completion word than for semantically anomalous words related to the best completion word. These results indicate that semantic anomaly is not necessary to elicit the N400 component. Rather, N400 amplitude is sensitive to word expectancies produced by sentence context and to associations among words. Based on these findings, Kutas has suggested that the N400 may be an index of the amount of semantic priming or activation that a word receives from prior context.

     The N400 is not only elicited by terminal words. Rather, it appears that all open class or content words (e.g., nouns, verbs, and adjectives that make reference to specific objects and events) in a sentence elicit an N400 negativity (Kutas, Van Petten, & Besson, 1988). However, as a sentence progresses the amplitude of the N400 decreases (Kutas et al., 1988; Van Petten & Kutas, 1990). This finding suggests that the N400 reflects modifications in linguistic processing due to the build-up of context. In other words, recognition of a word benefits more and more from associative priming as the meaning of the sentence becomes more constrained.

     Semantic priming using word lists. The N400 component has been found to be elicited not only to words in sentence contexts but also to isolated words in word lists. Although experiments involving such word lists may not represent situations of natural language comprehension, they do provide a paradigm for controlled manipulation of semantic processing. In a semantic priming lexical decision experiment, Bentin, McCarthy and Wood (1985) reported a larger N400-like negativity in response to words preceded by a semantically unrelated word than to words preceded by a semantically related word. Similar semantic priming effects have been observed in a number of studies to words presented visually (e.g., Boddy, 1986; Holcomb, 1986, 1988; Rugg, 1985), auditorily (Holcomb & Neville, 1990) and cross-modally (Holcomb & Anderson, 1993).

     Semantic priming within sentences. Additional evidence supporting the interpretation that the N400 is an indicant of semantic processing has been provided by studies using a sentence verification task. Fischler, Bloom, Childers, Roucos, and Perry (1983) showed subjects true and false sentences whose form was either affirmative or negative (e.g., true affirmative: A robin is a bird, false affirmative: A robin is a tree, true negative: A robin is not a tree, false negative: A robin is not a bird). The subjects' task was to judge whether each sentence was true or false. They found that the ERPs for false affirmative sentences were more negative than those for true affirmative sentences in the region from 250 to 450 msec after the onset of the final word. Conversely, the ERPs for true negative sentences were more negative than those for false negative sentences in this time window. Apparently, the negative component was sensitive to whether or not the subject and object of the sentence were semantically related. It was not dependent upon the truthfulness of the sentence or the decision and response required by the sentence. Using a similar sentence verification task, Kounios and Holcomb (1992) found that N400 amplitude was affected by the relatedness of subject and predicate terms, but not by the quantifiers all, some, or no. The N400 component was also more negative for exemplars (e.g., dogs) than for category names (e.g., animals). This effect was observed in a concrete/abstract judgment task using word lists as well as in the sentence verification task. The results of Fischler et al. and Kounios and Holcomb suggest that the N400 is sensitive to the structural (non-propositional) aspects of semantic memory for words both in isolation and in relation to other words.

     Syntactic versus semantic processing. It also appears that the N400 component is not sensitive to manipulations of the grammatical structure of a sentence. Osterhout and Holcomb (1992) showed subjects grammatically correct sentences (e.g., "The broker hoped to sell the stock") and sentences containing syntactic anomalies, i.e., grammatical violations (e.g., "The broker persuaded to sell the stock"). For the anomalous sentences, a slow positive-going wave with an onset around 500 msec was observed at the point where the syntactic anomaly should be detected, i.e., the word "to". This "P600" component was elicited consistently for words which violated the "preferred" structure of a sentence and was quite distinct from the N400. On the other hand, final words in the syntactically anomalous sentences elicited an N400-like effect, indicating that grammatical structure can influence semantic processing. Furthermore, the P600 had a widespread scalp distribution which was largest at fronto-central and right hemisphere sites, while the N400 had the posterior distribution characteristic of this component. These findings suggest that syntactic and semantic processing involve separate psychological and neurological systems.

     Language-specificity of the N400. It can be concluded from the aforementioned study that the N400 is not sensitive to grammatical structure, except when it interferes with sentence meaning. Consistent with this hypothesis is the finding that words which serve grammatical functions, i.e., closed class or function words (e.g., articles, conjunctions, prepositions, and auxiliaries), do not typically elicit much if any N400-like activity (Kutas, et al., 1988; Neville, Mills, & Lawson, 1992). Illegal nonwords (random letter strings or nonwords which do not comply with the phonological or orthographic rules of a language) also do not elicit an N400 (Holcomb & Neville, 1990). On the other hand, pseudowords (nonwords which follow the phonological and orthographic rules of a language) produce N400s that are comparable to, or larger than, those for open class or content words (Holcomb, 1988; Holcomb & Neville, 1990). In short, word-like nonwords elicit a large N400, while un-word-like nonwords do not. Holcomb and Neville interpreted this result as evidence that the N400 is specific to language and language-like stimuli.

     Further support for the proposal that the N400 is "language specific" is provided by studies using non-linguistic stimuli. Besson and Macar (1987) replicated the Kutas and Hillyard anomalous sentence paradigm. In addition to using sentences with congruous or incongruous final words, non-linguistic stimuli with congruous and incongruous endings were tested. As expected, an N400 was observed in response to semantic incongruities within sentences. Deviations in the expected endings of well-known melodies, musical scales, or series of geometric patterns did not elicit an N400. Rather, the non-linguistic deviations produced a late positivity. These results suggest that the N400 is not simply a response to a mismatch or violation of expectation, but is specific to semantic or linguistic processing.

     The cognitive process underlying the N400. There is no consensus in the literature as to the identity of the cognitive process or processes which are reflected by the N400 component. Evidence exists for both pre-lexical and post-lexical accounts. For example, Osterhout and Holcomb (1995) note that if the peak latency of the component is taken as the important temporal marker, then the N400, which peaks about 400 msec after the onset of a stimulus, occurs too late for it to be an index of lexical access (the process whereby a words representation in lexical memory is first activated), which is believed to occur at about 200 msec (Sabol & DeRosa, 1976). However, if the latency at which the waveforms diverge is taken as the crucial marker, then it is possible that the N400 reflects some aspect of lexical access, since differences in the ERPs for contextually appropriate and inappropriate words have been found as early as 50 msec after stimulus onset (Holcomb & Neville, 1991).

     Studies utilizing a semantic priming/lexical decision task have traditionally focused on an automatic spreading activation account of priming (Collins & Loftus, 1975). Some of the ERP findings seem to support this automatic processing account. For example, Boddy (1986) found corresponding increases in N400 amplitude to unrelated words in a LDT at 1000 and 200 msec stimulus-onset-asynchronies (SOAs). Similarly, Besson, Fischler, Boaz, and Raney (1992) found N400 effects to unrelated target words at a short SOA (300 msec) in a task involving graphemic judgment of the word pairs. Further support for an automatic lexical processing account of the N400 was obtained by Kutas and Hillyard (1989) in a delayed letter search task. Since letter search does not require access of word meanings, they concluded that the N400 amplitude differences found for the second words of related and unrelated word pairs were due to automatic spreading activation.

     Conversely, masked priming LDT, a task which involves purely automatic priming, evidenced by the fact that the prime is unidentifiable, has produced behavioral but not N400 effects (Brown & Hagoort, 1993). As a result, it is suggested that the N400 represents post-lexical controlled processing. Two types of post-recognition processes have been proposed to explain the significance the N400 component: a conceptual memory process and a higher level "integrative" process. Evidence that the N400 may reflect activity in an a-modal conceptual/semantic system has been obtained in tasks involving cross-modal priming (Holcomb & Anderson, 1993) and priming between words and objects (Nigam, Hoffman, & Simons, 1992). Integrative processing interpretations propose that the N400 "reflects a post-lexical process that is invoked in proportion to the degree that the evoking stimulus and its context form an unfamiliar or unexpected conjunction" (Rugg, 1990 p. 375). Stated another way, the more effort involved in integrating a word into the ongoing context, the larger the N400.

     The N400 and the organization of the lexicon. Studies using ERPs have begun to reaffirm evidence from neuropsychology and brain imaging that verbal representations are organized into distinct subdivisions subserved by separate brain regions. Neville, Mills, and Lawson (1992) found that words with semantic and grammatical functions are processed by distinct language subsystems. Subjects read sentences which ended with either a semantically anomalous or highly expected word and were required to decide if each sentence made sense. Sentence internal words were coded as either content words (open class: nouns, verbs, and adjectives) or function words (closed class: articles, prepositions, conjunctions, etc.). Content and function words elicited similar early sensory ERP responses. However, after 150 msec the pattern of results was quite different for the two word types. Function words elicited a negative component peaking at 280 msec (N280) that was largest over frontal and anterior temporal sites and was only evident over the left hemisphere. However, content words elicited a negative peak at 350 msec (N350) that was largest over posterior sites in both hemispheres.

     Neville et al. propose that "these differences in morphology, timing, and distribution of ERPs elicited by open and closed class words are generated by the activation of different neural systems that are organized to process the different kinds of linguistic information that these word classes provide" (p. 251). Only function words elicited an N280 suggesting that they are accessed by a system specific to grammatical processing or parsing of sentence structure. Furthermore, the distribution of the N280 over left anterior sites is consistent with clinical reports of deficits in comprehension and production of grammar after damage to these regions. The N350 reported here is likely an N400, involving the semantic processing of words. Function words elicited a small N350. This may reflect dual processing of function words by both the semantic and grammatical subsystems.

     Similar evidence that content and function words are processed by distinct neural systems was obtained by Nobre and McCarthy (1994) using a word list category detection task. Much larger N400 amplitudes were elicited by content words (concrete nouns) than function words. This result suggests that the attenuation of the N400 for function words in sentence paradigms may not be due to contextual or semantic priming since a similar effect was obtained with single words. Rather, the reduced amplitude may reflect some inherent processing difference based on word type. Furthermore, a left frontal negativity at 290 msec elicited by function words that was not evident for content words was observed which was similar to the N280 reported by Neville et al. The results of Neville et al. and Nobre and McCarthy support the idea that function and content words are processed by distinct neural systems (or at least to different degrees by these systems). These studies also demonstrate that electrophysiological recordings are useful for identifying differential processing of various word types.

Concreteness

     Representations in the language system may also be organized along the dimension of concreteness-abstractness. Highly concrete concepts have direct sensory referents; they are modality-specific representations of objects and events. Verbal materials rated high for concreteness are generally also rated high for imageability (i.e., they readily evoke mental images; Paivio, Yuille, & Madigan, 1968). Abstract concepts, on the other hand, lack specific objective correlates and are not necessarily tied to any modality.

     Empirical evidence generally suggests that concrete verbal materials are comprehended faster and more easily than abstract verbal materials, although there are a few exceptions (e.g., Paivio & Begg, 1971). The time to comprehend a sentence is generally shorter when the sentence is concrete rather than abstract (Haberlandt & Graesser, 1985; Marschark, 1979; Schwanenflugel & Shoben, 1983). Subjects respond faster to concrete than to abstract sentences in meaning classification tasks, in which meaningful and anomalous sentences must be distinguished, (Holmes & Langford, 1976; Klee & Eysenck, 1973) and sentence verification tasks, which require a judgment of the meaningfulness or truth value of a sentence (Belmore, Yates, Bellack, Jones, & Rosenquist, 1982). It has also generally been found that subjects both encode and retrieve concrete words and sentences faster and more completely than abstract words and sentences. This has been demonstrated in recognition, free and cued recall, and paired associate learning (Holmes & Langford, 1976; Marschark & Paivio, 1977; Nelson & Schreiber, 1992; Paivio, 1967; but see Marschark, Richman, Yuille, & Hunt, 1987).

     The effects of concreteness in the above experiments are usually attributed to relatively late comprehension or decision processes. Concreteness effects have also been examined in tasks designed to tap earlier stages of processing (i.e., tasks in which retrieval of word meaning is not required). The most frequently used tasks have been lexical decision and word naming. Significantly faster reaction times (RTs) for concrete words have been found in some but not all of the many experiments employing these methods (see Schwanenflugel, 1991). Schwanenflugel (1991) identified several factors associated with the occurrence of concreteness effects in these tasks. First, effects are more likely to occur when low frequency words are used (DeGroot, 1989; James, 1975; Kroll & Merves, 1986). This may be due to an enhanced need to consult semantic information when making lexical decisions to words that appear infrequently in the language. Second, concreteness effects occur in neutral lexical decision or naming trials when those trials are intermixed with trials requiring the processing of word or sentence meaning (Schwanenflugel, Harnishfeger, & Stowe, 1988). Third, concreteness effects are larger and more prevalent in the lexical decision task, which requires that a decision be made in addition to lexical access, than in the naming task, which only requires lexical access. These findings suggest that concreteness effects may only occur when deep semantic processing (i.e., the retrieval of meaning) is required by the task.

     The evidence discussed above demonstrates that, in general, concrete words and sentences are comprehended, encoded, and remembered faster and more easily than abstract words and sentences. Theories espousing a common semantic system maintain that factors such as context availability (Bransford & McCarrell, 1974; Kieras, 1978; Schwanenflugel, 1991) or age of acquisition (Brown, 1957; Gilhooly & Gilhooly, 1979; Schwanenflugel, 1991) are responsible for differences in the ability to process concrete and abstract materials. This argument is closely related to the notion that information from different modalities (e.g., words vs. pictures) is processed by separate sensory and perceptual systems but meaning is shared in a common system or store (REFS). On the other hand, multiple system hypotheses advocate separate and oftentimes redundant semantic systems for information from different modalities (Kosslyn, 19??; OTHER REFS). A leading theory with this viewpoint is Paivio's dual-coding theory (1971, 1986, 1991) which postulates functionally (and perhaps anatomically) distinct, yet interconnected, representational systems for processing verbal and nonverbal information. According to dual coding theory, all verbal stimuli, whether concrete or abstract, are initially processed by a single verbal (representational) system. Concrete words have access to both referential (activation of the nonverbal system) as well as verbal-associative processes, and therefore can easily evoke mental images. Abstract words have primarily associative interconnections, and therefore do not easily evoke mental images. In other words, concrete (high-imagery) words should be processed by both the verbal and the image-based systems while abstract (low-imagery) words should be processed primarily by the verbal system. The involvement of two systems and thus two forms of representation should lead to superior performance with concrete words on tasks involving the retrieval of semantic information

     If, in fact, the verbal and nonverbal systems are interconnected yet distinct, they should be functionally independent and, therefore, should produce additive effects. Examples of such effects include the findings that (a) pictures (which have access to the verbal system through naming in addition to the imagen system) are remembered in free-recall and recognition tasks more readily than words (Paivio, Rogers, & Smythe, 1968), (b) concrete words (which also have access to the imagen system) are retrieved more easily than abstract words (Paivio, 1967), (c) interactive imagery instructions improve performance for concrete but not abstract words (see Richardson, 1980), and (d) recall for words that are both imaged and pronounced is better than for words that are only imaged or only pronounced (Paivio & Csapo, 1973).

     Evidence from visual hemifield studies and from patients with localized brain lesions suggests that, indeed, more than one neuronal region or system is involved in the processing of concrete and abstract words. For example, Ellis and Shepherd (1974) found that when concrete/abstract word pairs were briefly presented (150 msec) tachistoscopically to the left visual field (LVF) or the right visual field (RVF), concrete words were more accurately recognized than abstract words when presented to the LVF (right hemisphere). However, the two word types were recognized equally well when presented to the RVF (left hemisphere). For criticisms see Boles (1983). In addition, Day (1977) found that reaction times for lexical decisions and semantic category judgments for concrete nouns did not differ with field of presentation, but reaction times for abstract nouns were faster for stimuli in the RVF than in the LVF. Similarly, Chiarello, Senehi, and Nuding (1987) found that concrete words were equally effective in priming lexical decisions to words (concrete) presented in the LVF as to words presented in the RVF. Abstract words, on the other hand, were less effective in priming lexical decisions to words (concrete) in the LVF than in the RVF. Furthermore, patients with deep dyslexia, a disorder which results from wide-spread lesions localized to the left hemisphere, have greater difficulty reading abstract low-imagery words than concrete high-imagery words (Coltheart, 1980). This suggests that the intact right hemisphere permits more efficient processing of concrete than abstract words and that representations for high-imagery words and the concrete objects to which they refer are available in the right hemisphere (but see Warrington, 1981).

     Paivio's dual coding model of word meaning asserts that concreteness effects are due to the additional form of representation and processing afforded to concrete words by their connection to the nonverbal system. Accordingly, it is not a word's concreteness value per se that predicts its superiority, but rather its imageability (i.e., its ability to evoke mental images). Imageability and concreteness are highly correlated (e.g., 0.83 in the Paivio et al., 1968, norms). However, in some instances, words can be rated high for concreteness but low for imageability (e.g., antitoxin or magnesium) or low for concreteness but high for imageability (e.g., concert or gaity). It has been found that a word's imagery value, not its level of concreteness, is the most highly predictive factor for its memorability in tasks of recall and recognition (e.g., Paivio, 1968). Furthermore, the evidence from neuropsychological observations and visual hemifield studies suggest that concrete words may have access to the right hemisphere and, therefore, to regions of the brain that are not specialized for language. Consequently, imagery, as a nonverbal form of mental representation, may play an important role in the encoding and retrieval of words and their meanings.

     Semantic memory processes involved in imagery, as in tasks requiring identification of faces, familiar objects, geometric forms and verbal stimuli likely occur in both hemispheres, in particular the posterior temporo-occipital regions (Paivio & te Linde, 1982). On the other hand, evidence suggests that episodic memory for nonverbal material and image-mediated episodic memory for words occurs primarily in the right temporal lobe, particularly in the hippocampus (Paivio & te Linde, 1982). For example, Jones (1974) found that patients with left temporal lobe lesions performed more poorly than controls in immediate and delayed recall of verbal paired associates in the absence of imagery instructions, but were able to use visual imagery to improve their performance with concrete pairs, although not to reach a normal level. In contrast, patients with right temporal lobe lesions, who have memory deficits for nonverbal materials, showed impairment in recall of concrete pairs of words when instructed to use imagery to learn that material, but performed as well as controls in learning pairs of abstract words either with or without the use of a sentence mediation verbal mnemonic (Jones-Gotman & Milner, 1978). Similarly, relative to controls, patients with large right temporal lobe lesions were impaired in incidental free-recall of concrete words after instructions to generate images to those words (Jones-Gotman, 1979). These studies imply that the right hemisphere of these patients was important for mediating verbal tasks involving imagery. Further, it is important to note that all of these patients were able to generate images to concrete words during the study phases. Their impairments lay in their memory for those generated images (Paivio & te Linde, 1982). This distinction supports the idea that imagery is not a unitary phenomenon but reflects a collection of processing mechanisms.

     The N400 and concreteness. The N400 potential is a likely candidate for determining the relationship between imagery and language. Not only does the N400 appear to reflect the semantic processing of a word but it may also be sensitive to the concreteness value of that word. Concrete words have been found to produce a more negative-going ERP in the region of the N400 than abstract words in a concreteness judgment task (Paller, Kutas, Shimamura, & Squire, 1988) but not in a repetition priming lexical decision task (Smith & Halgren, 1988). Manipulation of concreteness value was not the primary focus of either of these studies, however. In the second study, only 16 words of each type were used and these varied in normative frequency.

     Kounios & Holcomb (1994) more closely examined the effect of concreteness on ERPs using both a lexical decision task (Experiment 1) and a concreteness judgment task (Experiment 2) in a repetition priming paradigm. Overall, concrete words elicited a more negative ERP component between 300 and 500 msec (N400) than abstract words. This effect was more pronounced in the semantic concreteness judgment task than in the lexical decision task. However, the pattern of results was similar for both tasks. For both tasks, the negativity for concrete words was larger over the right hemisphere than the left and over anterior scalp sites than posterior sites. The greater N400 for concrete words is consistent with the idea that concrete words have more semantic associations than abstract words. The different scalp topographies suggest that more than one neural generator is involved in processing these two word types. Topographic differences were also revealed in effects of word repetition. Large repetition effects were seen for concrete words over both hemispheres. In other words, across the scalp there were smaller N400 negativities for concrete words at the second presentation than at the first presentation. In contrast, repetition effects for abstract words were only evident over the left hemisphere in Experiment 2 and were nonexistent in Experiment 1. If the effects were due simply to semantic context one should observe larger effects for abstract words. This result, thus, suggests that context and concreteness are distinguishable factors, both of which may be acting on the same comprehension process and which interact to modulate the N400. Furthermore, the different scalp distributions of the N400 negativity for concrete and abstract words support the notion that two neuronal populations exist which are unequally responsible for the processing of these two word types.

ERPs and Imagery

     To determine if the N400 component was reflecting activation of an image-based representational system in these tasks, we performed another experiment designed to manipulate the types of processing strategies used by subjects. This was accomplished by controlling the type of information that the subject was required to obtain in order to complete the task

     Subjects in this experiment were assigned to one of three groups: imagery, semantic, and surface. Each group performed a variation of a sentence verification task. In these tasks subjects were required to judge the truthfulness of sentences with either concrete or abstract final words. The sentences presented to each group varied according to the type of processing required to judge the sentences as true or false. Specifically, for the imagery group the sentences required subjects to try to form an image of the final word (e.g., "It is easy to form a mental image of a canoe."). For the semantic group the sentences required subjects to decide if they make sense semantically (e.g., "It is common for people to have a canoe."); these sentences were constructed so as not to encourage subjects to use imagery. While it is possible that subjects may have formed images to words in this condition, what we might call implicit imagery, imagery presumably would not provide any information useful for making the required judgment. Finally, for the surface group the sentences required subjects to decide if a probe letter was present in the final word (e.g., "There is a ‘t’ in the word canoe."); these sentences do not encourage the use of imagery or the extraction of any semantic information.

     Concreteness effects were first assessed by the latency from the onset of the final word to judge a sentence as true or false. The time to respond to all sentence final words was shorter for subjects in the surface group than for those in the imagery or semantic and response time was generally shorter for concrete than abstract words. This difference was significant for subjects in the imagery and semantic groups but not the surface. Furthermore, response times to concrete words were shorter for subjects in the imagery group than for those in the semantic group, while response times to abstract words were equivalent for the two groups

     The N400 component was evaluated by measuring the mean amplitudes of the ERPs elicited to the sentence final words between 300 and 550 msec after stimulus onset. No significant differences between the ERPs for concrete and abstract words were observed at any scalp locations in the surface group (see Figure 2). For the semantic group, concrete words were associated with significantly more negative waveforms than abstract words (see Figure 3). The magnitude of this effect was consistent across the scalp. For the imagery group during this epoch, concrete words were also associated with more negative-going waves than were abstract words (see Figure 4). However, the difference in amplitude between concrete and abstract words at lateral sites increased toward more anterior locations. At the most posterior sites (Pz, O1,O2) there were no differences between the word types.

     Data for the imagery group were also analyzed in terms of which words the subjects judged to be imageable or not imageable (Figure 5). Since very few concrete words were judged by any subjects to be non-imageable, the analysis compared only concrete imageable, abstract imageable, and abstract non-imageable words. At midline sites, abstract imageable words showed no voltage differences from concrete imageable words at any location, while abstract non-imageable words were more positive than concrete imageable words at all locations. At lateral sites a more complex pattern was observed. Again, abstract non-imageable words were more positive than concrete imageable words at all locations. Also, there were no voltage differences between concrete imageable words and abstract imageable words at occipital and Wernicke's sites. However, from temporal to frontal sites there was an increasing difference in the waveforms for these two word types. Furthermore, at the left occipital site, concrete imageable words were associated with a more positive wave than were abstract imageable words, while the waveforms for the two word types were equivalent at the right occipital site.

     The epoch between 550 and 800 ms was also analyzed. Again the surface group showed no concreteness effects. The effects for the imagery and semantic group were also similar to the previous epoch. However, for the imagery group, concrete words were actually more positive than abstract words at occipital sites and the difference between them was larger over the left than the right hemisphere.

     Data for the imagery group were also analyzed according to the subjects' ratings of each word's imageability as was done for the previous epoch (Figure 5). Concrete imageable words were more negative than abstract imageable words at frontal sites but became more positive at posterior sites. This positivity was more pronounced over the left hemisphere than the right hemisphere at occipital locations. Concrete imageable words were more negative than abstract non-imageable words at all scalp locations although this difference increased toward the front of the head. This comparison also showed a trend similar to the concrete imageable versus abstract imageable comparison in that concrete words were more negative at anterior sites. When abstract imageable words were compared with abstract nonimageable words, another effect was revealed. Waveforms at anterior locations were similar for imageable and nonimageable abstract words. At posterior locations, the nonimageable abstract words became more positive than the imageable abstract words. The overall picture that emerged appeared to be that at frontal sites there was a difference between concrete words and abstract words (concrete more negative) and at posterior sites there was a difference between imageable words and nonimageable words (imageable more negative). This trend was complicated, however, at the most posterior sites where abstract imageable words were actually more negative than concrete imageable and abstract nonimageable words.

     Subjects who were engaged in surface level processing during the first task were much faster to verify sentences as true or false than subjects engaged in semantic processing or imagery. These subjects also responded to sentences with concrete and abstract final words equally quickly. This result is evidence that the subjects in this group were, in fact, processing only the surface characteristics of the final words and were not accessing the meaning of the words. These findings are in agreement with Craik and Lockhart's (1972) levels of processing model which states that shallow sensory analysis of a word precedes deeper levels of processing. Subjects who were engaged in deeper modes of processing (the imagery and semantic groups) exhibited the classic concreteness effect; they responded faster to concrete final words than abstract final words. In addition, subjects in the imagery group responded to concrete words faster than subjects in the semantic group, while response times to abstract words were equivalent for the two groups. This may be evidence for a processing advantage for concrete words when imagery is used. Furthermore, this effect was not dependent on whether the subjects described themselves as "good" imagers or "poor" imagers.

     Electrophysiological recordings for the surface group also confirmed that only shallow processing was being used by these subjects. There were no differences in the waveforms elicited by concrete and abstract words at any scalp locations. This result was similar to the finding by Chwilla et al. (1995) in which N400 priming effects were absent when subjects performed a shallow task. The task used by Chwilla (upper or lower-case decision) did not involve lexical access, since subjects responded equally quickly to both word and nonword stimuli. The task performed by our surface group probably did involve lexical access. In similar letter search tasks subjects generally have responded faster to words than nonwords (Word Superiority Effect; Reicher, 1969). These results are supportive of the hypothesis that the N400 and subsequent components reflect post-lexical processing.

     Large concreteness effects were evident in the ERPs for subjects in the imagery and semantic groups. Waveforms to concrete words were generally more negative-going than waveforms to abstract words. This effect was more pronounced for the semantic group than for the imagery group. A slight increase in the magnitude of the effect toward anterior sites was evident for the semantic group, although the effect was still quite prominent at even the most posterior locations. On the other hand, the imagery group showed a dramatic increase in effect size toward frontal sites. At Wernicke's and parietal sites only a very small difference between concrete and abstract words was apparent. At occipital sites there was actually a reversal in the polarity of the effect, such that waveforms to concrete words became more positive than those to abstract words during the 550 to 800 msec epoch. The different scalp distributions of concreteness effects for the imagery and semantic groups are suggestive of different patterns of activity in the underlying neural generators.

     Examination of the difference waves produced when the ERPs to abstract words are subtracted from the ERPs to concrete words revealed that the time course of these effects may also differ (see Figure 6). The peak effect for the semantic group occurred 400 to 500 msec after stimulus onset. The effect for the imagery group peaked later, at about 650 to 750 msec. For the imagery group in particular, there appeared to be two distinct negative components: the first in the region of the N400, the second between 550 and 800 msec. Both of these components became larger at anterior scalp sites.

     Additional evidence that different mechanisms may be involved in the semantic and imaginal processing of these words was visible in the ERPs to words which the subjects in the imagery group judged as imageable and nonimageable. There was a large difference between concrete and abstract words (concrete more negative), regardless of imageability, at anterior regions. Conversely, there was a large difference between imageable words and nonimageable words (imageable more negative) at posterior regions. Furthermore, during the 550 to 800 msec epoch, abstract imageable words were more negative than concrete imageable words at these posterior locations. This effect may reflect greater effort in forming images to abstract words, perhaps because these images are more complex.

     Taken together, these results suggest that more than one source is contributing to the observed effects. An anterior generator peaking at about 400 msec seems to be involved with the retrieval of semantic information, while a posterior generator peaking at about 700 msec is involved with the retrieval of imaginal information. This hypothesis coincides with existing knowledge of the brain regions involved in language and imagery. The N400 component has been implicated in semantic information retrieval, as well as in integration of words into the current context. A field potential peaking at 400 msec, with response characteristics similar to the scalp-recorded N400, has been demonstrated to arise from the anterior-medial temporal lobe (McCarthy, Nobre, Bentin, & Spencer, 1995; Nobre & McCarthy, 1995; Smith, Stapleton, & Halgren, 1986). Functional imaging studies and neuropsychological observations have implicated visual cortical areas (including occipital, posterior temporal, and parietal regions) in visual mental imagery (see Farah, 1995).

     The results of the current study strongly support the notion that concreteness and imageability, while highly correlated, are independent dimensions by which words and their meanings are organized in the brain. Based on the spatial distributions of the potentials evoked in the current study, it appears that these dimensions are controlled by different brain regions. A more anterior region seems to be involved in processing a word for its contextual or inherent verbal associations, while a more posterior region seems to be involved in the retrieval of images associated with a word. One brain region which may be involved with the concreteness or verbal factor is the anterior medial temporal lobe, which has been linked to the N400 (McCarthy et al., 1995, Nobre & McCarthy, 1995). Another candidate is the pre-motor left inferior frontal lobe, which has been found to cause deficits in verbal fluency when damaged (Damasio, 1985) and in PET studies has been reported to be active in normal subjects during verbal associative processing (Peterson, Fox, Posner, Mintun, & Raichle, 1988). Conversely, regions of the brain devoted to visual perceptual processing have been demonstrated to subserve visual imagery (e.g., see Farah, 1995, Kosslyn, 1994).

     The existence of an ERP component related to the use of imagery in response to reading verbal material is strong evidence for an image-based representational system. Furthermore, it indicates that the verbal and nonverbal semantic systems are functionally and anatomically distinct (they are subserved by different neural structures) yet they are interconnected. This finding strongly supports Paivio's dual coding model of mental representation. Likewise, it is strong evidence against single code models which reject imagery as a significant symbolic system related to language. More generally, the findings reported here are supportive of semantic system which is organized along a number of different dimensions. According to this model, the meanings for words can be redundantly stored in modality specific representations. This redundancy can facilitate processing for some types of verbal material (such as imageable words) and may be the factor underlying concreteness effects. Future experiments will focus on imageability as a dimension by which words are organized in the brain and on imagery as a system for encoding and retrieving semantic information.

ERPs and Lexical Similarity

     Lexical similarity has emerged as an important construct for theories of lexical processing. Generally speaking, it refers to redundancy among lexical elements. For example a word like dove and a pseudoword like gake have a high degree of lexical similarity because they share the same letters with many other lexical elements. Conversely, a word like soak and a pseudoword like arat have low degree of lexical similarity because only a few lexical elements share such letter combinations. Much research has focused on the extent to which lexical similarity facilitates or inhibits lexical processing because the answer to this question has significant implications for models of word recognition. For example, serial models (e.g. Forster, 1976) predict that lexical similarity will inhibit processing while both local (McCelland & Rumelheart, 1981) and distributed (Seidenberg & McCelland, 1989) models allow for lexical similarity to facilitate or inhibit processing.

     Unfortunately, no clear effect of lexical similarity has been established. To date, both facilitory effects (e.g., Andrews, 1989; 1992; Forster & Davis, 1984; Segui & Grainger, 1990; Snodgrass & Mintzer, 1993) and inhibitory effects (Andrews, 1989; 1992; Colombo, 1986; Forster, 1987; Forester et al., 1987; Granger 1990; Grainger, O'Regan, Jacobs, & Segui, 1989; Lupker & Colombo, 1994; Pugh, Rexer, Peter, & Katz, 1994; Segui & Grainger, 1990; Snodgrass & Mintzer, 1993) have been reported.

     Several issues persuaded us to evaluate lexical similarity using ERPs. First, a common thread in all of the existing research on lexical similarity is that it has primarily relied on reaction time (RT) measures. Secondly, several studies have shown (e.g. Holcomb, 1993; Kounios and Holcomb, 1992) that RT, in particular, is sensitive to aspects of decision making, while the N400 component of an ERP is relatively impervious to decision based processing. Finally, for some time we known that tasks such as lexical decision and naming can provide a distorted glimpse of processes involved in lexical processing (see Balota & Chumbly, 1984). The consensus is that these tasks, in varying degrees, introduce additional information processing related to making a decision that would not otherwise be involved in word recognition. Our initial goal was to demonstrate a consistent ERP effect of lexical similarity while replicating existing behavioral findings. Our broader goal is to provide converging behavioral and electrophysiological evidence that can be used to constrain models of lexical processing.

     As a starting point we decided to focus on an effect in the literature that has received little attention, but has important implications. In two reports Andrews (1989; 1992) demonstrated a facilitory effect of lexical similarity; RTs were quicker to low frequency words with more lexical similarity than to those with less similarity. However, Andrews (1989) also showed an inhibitory effect of lexical similarity for pseudowords; RTs to pseudowords with higher lexical similarity were slower than RTs to pseudowords with less lexical similarity. So, there seems to be a qualitatively different effect of lexical similarity for words than there is for pronounceable nonwords. To account for this difference Andrews speculated that a decision criterion was employed for making lexical decision to pseudowords (Balota and Chumbly, 1984). According to this argument when pseudowords are not similar to any real words, "no" responses in a LDT can be made more easily. Conversely, when pseudowords are similar to a number of real words, "no" responses to pseudowords are slower because the subject is initially inclined to answer "yes". This would likely have the effect of increasing errors and latencies for "no" responses. For our purposes, it’s important to note that this account treats words and pseudowords in a fundamentally different way. Either a pseudoword is easy to spot as different and confirm as not a word, or a pseudoword is easily confused with real words and requires some additional processing.

     Another possibility exists. It is strategic in nature, such that information processes all together separate from the lexicon are engaged. The logic is a follows. A reasonable strategy for subjects in a LDT would be to simply monitor the level of activity within the lexicon. Since subjects only have to decide whether a stimulus is or is not a word, there is no reason to completely process each stimulus. If the lexicon indicates that there is little or no activation, then subjects can respond "no". If activation is apparent in the lexicon then subject can save processing time in this task and respond "yes". Low frequency words with many neighbors can be responded to faster because they generate activation throughout the lexicon therefore provoking a quicker "yes" response. The catch with this strategy is that pseudowords with lots of neighbors will also produce more activation in the lexicon. This will incline subjects to say "yes" to these pseudowords, which in turn, will make latencies and errors for "no" responses increase. This explanation still allows for nonwords, overall, to have longer latencies than words. Moreover, the critical aspect of this argument is that the effect of neighborhood size within the lexicon could be the same for words and pseudowords. That is, differences in RT between words and pseudowords due to manipulation of lexical similarity may simply be attributable to subjects' response strategies. What is needed is evidence for the effect of neighborhood size that is not influenced by strategic/decision making processes.

Lexical Decision Task

     In a lexical decision task we manipulated neighborhood size for an equal number of low frequency words and pseudowords while controlling for bigram frequency (see Andrews, 1992). Stimuli came from either large neighborhoods or small neighborhoods. Neighborhood size was operationally defined with the N metric (Coltheart, Davelaar, Jonasson, & Besner, 1977), which includes in a stimulus' neighborhood any element in the lexicon which shares N-1 letters (where N is the total number of letters in the stimulus). Subjects saw each stimulus in isolation for 300 msec and were instructed to respond as quickly as possible while minimizing errors. In addition to collecting data from our standard ERP sites, overt responses were collected so that ERPs could be compared directly to RT data and discussed in relation to previous findings. We made two predictions. First, given how closely our stimuli were patterned after Andrews (1989; 1992) we expected to replicate her behavioral findings (i.e., faster word and slower pseudoword RTs for larger neighborhoods). Second, it was predicted that stimuli with more neighbors would generate larger N400s. The rationale for this prediction is that the more neighbors activated the more feed-forward activity there is to be expected to semantic level representations. Since the N400 is thought to reflect activity at the semantic level, the more semantic representations receiving partial activation the larger the expected N400.

     The RT data replicated Andrews’ findings (1989; 1992) very closely. Low frequency words from large neighborhoods were responded to more quickly than low frequency words from small neighborhoods. Moreover, pseudowords from large neighborhoods were responded to more slowly than pseudowords from small neighborhoods.

     Figure 7 displays the grand mean ERPs for 13 scalp sites for pseudowords from large neighborhoods (solid) and pseudowords from small neighborhoods (dotted). A small, but reliable, effect of neighborhood size was obtained (e.g. see Pz). Pseudowords from large neighborhoods produced larger negativity in the region of the N400 than pseudowords from small neighborhoods. Figure 8 displays the grand mean ERPs for 13 scalp sites for words from large neighborhoods (solid) and words from small neighborhoods (dotted). While the overall effect in the region of the N400 was in the same direction as the effect for pseudowords, the difference was not reliable (e.g. see Pz). Despite the fact that this ERP effect is modest, the direction of the effect appears to be the same for both words and pseudowords. Our tentative conclusion, needing further confirmation, is that the ERP effect of lexical similarity is qualitatively different than the RT effect for lexical similarity.

Semantic Categorization Task

     There are two possibilities for why a reliable effect of lexical similarity did not emerge for words in the lexical decision task. First, there is considerable evidence that overt behavioral responses, like those solicited in a lexical decision task, generate late positive components in ERP waveforms (see Donchin and Coles, 1988). It may be that the positivity adjacent to the N400 (see Figures 7 & 8) acted to attenuated or reduced the lexical similarity effect. Second, the effects of lexical similarity my also be attenuated by subjects in a lexical decision task who adopt a processing strategy where they do not completely engage the lexicon (see above) and subsequent semantic processes. Moreover, it seems that this type of processing strategy would affect lexical decision to words more so than lexical decision to pseudowords.

     The goal of our next experiment was to generate a more reliable effect of lexical similarity for words that would be visible in the N400. We used the same stimuli from our lexical decision task, but added foils that belonged to a specific semantic category. The task for subjects was to respond to stimuli that belong to the category of body parts (go / no go). It was assumed that to perform this task subjects must access an item’s meaning in semantic memory and that this would enhance the N400 since it appears to reflect processing at the semantic level. In addition, subjects only made overt responses to the foils. This change should eliminate the late positivity that could have been attenuating the effect of lexical similarity in ERPs to words with large and small neighborhoods.

     Figure 9 displays the grand mean ERPs from the semantic categorization task for 13 scalp sites for pseudowords from large neighborhoods (solid) and small neighborhoods (dotted). Pseudowords from large neighborhoods produced larger negativity in the region of the N400 than pseudowords with small neighborhoods, however the effect was not reliable. Figure 10 displays the grand mean ERPs from the semantic categorization task for 13 scalp sites for words from large (solid) and small neighborhoods (dotted). Here a reliable effect of neighborhood size did emerge. Words with large neighborhoods produced larger negativities in the region of the N400 than words with small neighborhoods. Overall, then, the change in task had the desired effect. Having subjects evaluate the meaning of each word and removing the late positivity due to overt responses allowed a reliable effect of lexical similarity to emerge for words. The fact that pseudowords did not produce a reliable effect of lexical similarity is not surprising given the task. It is not unreasonable to suppose that subjects did not pay as much attention to pseudowords in this task as they would have in a lexical decision task. To perform this task subjects may have first determined if a stimulus was a word or a pseudoword. Then they could terminate processing if the stimulus was a pseudoword. Nevertheless, as in the lexical decision task, the direction of the effect of lexical similarity was in the same direction for both words an pseudoword.

     Taken together, these two studies establish that ERPs have a consistent response to manipulations of lexical similarity, regardless of whether a stimulus is a word or a pseudoword. This is an important finding because it demonstrates that the mental lexicon responds in a similar way to all stimuli that conform to the rules of English orthography. In addition, given that the materials used to generate this ERP result also replicated existing behavioral results, it seems reasonable to conclude that differences in behavioral responses between words and pseudowords are due to post-lexical decision based processing. Finally, with these results in hand it will now be possible to use ERPs to explore further issues concerning the organization of word knowledge that have proven intractable for traditional behavioral measures.

Concluding Remarks

     The findings from these experiments illustrate how electrophysiological data can be used to augment more traditional measures such as RT in studies of language comprehension. Moreover, they demonstrate how in some cases, such as the differences between concrete and abstract words or words high and low in imagability, that ERPs can be used to more forcefully argue for separate processing or representational systems. These and other findings suggest that the application of similar procedures to certain populations of cognitively impaired subjects (e.g., those with language comprehension problems), might yield interesting and important insights into the information processing and neurological basis of these impairments. Moving towards this goal our laboratory has recently embarked on a series of language studies in normal and reading disabled children. Much of what we intend to do over the next several years in these children will be motivated by the results from the research reported here in normal college aged adults. Finally, as previously mentioned, while ERPs have excellent temporal resolution, in most cases it is difficult to isolate the exact brain region(s) producing an effect. For this reason it will be important to begin efforts to combine ERP findings with those from other procedures which have better spatial resolution (e.g., PET or fMRI).

Andrews, S. (1989). Frequency and neighborhood effects on lexical access: Activation or search? Journal of Experimental Psychology: Learning, Memory, and Cognition, 15(5), 802-814.

Andrews, S. (1992). Frequency and neighborhood effects on lexical access: Lexical similarity or orthographic redundancy? Journal of Experimental Psychology: Learning, Memory, and Cognition, 18(2), 234-254.

Balota, D. A., & Chumbley, J. I. (1984). Are lexical decisions a good measure of lexical access? The role of word frequency in the neglected decision stage. Journal of Experimental Psychology: Human Perception and Performance, 10, 340-357.

Becker, C. A. (1976). Allocation of attention during visual word recognition. Journal of Experimental Psychology: Human Perception and Performance, 2, 556-566.

Belmore, S. M., Yates, J. M., Bellack, D. R., Jones, S. N., & Rosenquist, S. E. . (1982). Drawing inferences from concrete and abstract sentences. Journal of Verbal Learning and Verbal Behavior, 21, 338-351.

Bentin, S., McCarthy, G., Wood, C. C. (1985). Event-related potentials associated with semantic priming. Electroencephalography and Clinical Neurophysiology, 60, 343-355.

Besson, M., McCar, F. (1987). An event-related potential analysis of incongruity in music and other non-linguistic contexts. Psychophysiology, 24, 14-25.

Besson, M., Fischler, I., Boaz, T., Raney, G. (1992). Effects of automatic associative activation on explicit and implicit memory tests. Journal of Experimental Psychology: Learning, Memory and Cognition.

Boddy, J. (1986). Event-related potentials in chronometric analysis of primed word recognition with different stimulus onset asynchronies. Psychophysiology, 23, 232-245.

Boles, D. B. (1983). Dissociated imageability, concreteness, familiarity in lateralized word recognition. Memory and Cognition, 11, 511-519.

Bransford, J. D., & McCarrell, N. S. (1974). A sketch of a cognitive approach to comprehension: Some thoughts on what it means to comprehend. In W. W. D. Palermo (Ed.), Cognition and the symbolic processes (pp 189-230) . Hillsdale NJ: Lawrence Erlbaum Associates.

Brown, C. M., & Hagoort, P. (1993). The processing nature of the N400: Evidence from masked priming. Journal of Cognitive Neuroscience, 5, 34-44.

Chiarello, C., Senehi, J., & Nuding, S. . (1987). Semantic priming with abstract and concrete words: Differential asymmetry may be postlexical. Brain and Language, 31, 43-60.

Churchland, P. S., Sejnowski, T.J. (1992). The computational brain. Cambridge, Ma: MIT Press.

Chwilla, D. J., Brown, C.M., & Hagoort, P. . ((1995).). The N400 as a function of the level of processing. Psychophysiology,, 32, 274-285.

Coles, M. G. H. & Rugg, M.D. (1995). Event-related brain potentials: An introduction. In M. D. Rugg & M.G.H. Coles (Ed.), Electrophysiology of mind. . Oxford: Oxford University Press.

Collins, A. M., Loftus, E. F. (1975). A spreading-activation theory of semantic processing. Psychological Review, 82, 407-428.

Coltheart, M. (1980). Deep dyslexia: A right hemisphere hypothesis. In K. P. J. M. M. Coltheart (Ed.), Deep dyslexia . London: Routledge and Kegen Paul.

Coltheart, M., Davelaar, E., Jonasson, J.T., & Besner, D. (1977). Access to the internal Lexicon. In S. Dornic (Ed.), Attention and performance IV (pp. 535-555). Hillsdale, NJ:Erlbaum.

Colombo, L. (1986). Activation and inhibition with orthographically similar words. Journal of Experimental Psychology: Human Perception and Performance, 12(2), 226-234.

Courchesne, E., Hillyard, S.A., & Galambos, R. (1975). Stimulus novelty, task relevance and the visual evoked potential in man. Electroencephalography and Clinical Neurophysiology, 39, 131-143.

Craik, F. I. M. L., R. S. . (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior, 11, 671-684.

Damasio, A. R. (1985). The frontal lobes. In E. V. K. M. Heilman (Ed.), Clinical Neuropsychology . New York: Oxford University Press.

Day, J. (1977). Right-hemisphere language processing in normal right-handers. Journal of Experimental Psychology: Human Perception and Performance, 3, 518-528.

DeGroot, A. M. B. (1989). Representational aspects of word imageability and word frequency as assessed through word association. Journal of Experimental Psychology: Learning, Memory & Cognition, 15, 824-845.

Donchin, E., & Coles, M. G. H. (1988). Is the P300 component a manifestation of context updating? Behavioral and Brain Science, 11, 355-372.

Duncan-Johnson, C. C. & Donchin, E. (1979). On qunatifying surprise: The variation of event-related potentials with subjective probability. Psychophysiology, 14, 456-467.

Ellis, H. D., & Shepherd, J. W. . (1974). Recognition of abstract and concrete words presented in left and right visual fields. Journal of Experimental Psychology, 103, 1035-1036.

Farah, M. J. (1995). The neural bases of mental imagery. In M. S. Gazzaniga (Ed.), The cognitive neurosciences . Cambridge, MA: MIT Press.

Fischler, I., Bloom, P. A., Childers, D. G., Roucos, S. E., Perry, N. W. (1983). Brain potentials related to stages of sentence verification. Psychophysiology, 20, 400-409.

Forster, K. I. (1976). Accessing the mental lexicon. In I. R. J. W. E. W. (Eds.) (Ed.), New Approaches to language mechanisms.  Amsterdam: North-Holland.

Forster, K. I. (1987). Form priming with masked primes: The best match hypothesis. In Coltheart, M. (Ed.) Attention and Performance XII: The Psychology of Reading. Hillsdale NJ: LEA.

Forster, K. I., & Davis, C. (1984). Repetition priming and frequency attenuation in lexical access. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10, 680-698.

Forster, K.I., Davis, C., Schoknecht, C., & Carter, R. (1987). Masked priming with graphemically related forms: Repetition or partial activation? Quarterly Journal of Experimental Psychology, 39A, 211-251.

Gilhooly, M. L., & Gilhooly. (1979). Age-of -acquisition effects in lexical and episodic tasks. Memory and Cognition, 7, 214-223.

Grainger, J. (1990). Word Frequency and Neighborhood Frequency Effects in Lexical Decision and Naming. Journal of Memory and Language, 29, 228-244.

Grainger, J., O=Regan, J. K., Jacobs, A. M., & Segui, J. (1989). On the role of competing word units in visual word recognition: The neighborhood frequency effect. Perception & Psychophysics, 45(3), 189-195.

Haberlandt, K. F., & Graesser, A. C. . (1985). Component processes in text comprehension and some of their interactions. Journal of Experimental Psychology: General, 114, 357-375.

Holcomb, P. J. (1986). Electrophysiological correlates of semantic facilitation. Electroencephalography and Clinical Neurophysiology, 38, 320-322.

Holcomb, P. J. (1988). Automatic and attentional processes: An event-related brain potential analysis of semantic priming. Brain & Language, 35, 66-85.

Holcomb, P. J., Neville, H. (1990). Auditory and visual semantic priming in lexical decision: A comparison using event-related brain potentials. Language and Cognitive Processes, 4, 281-312.

Holcomb, P. J., Neville, H.J. (1991). The electrophysiology of spoken sentence processing. Psychobiology, 19, 286-300.

Holcomb, P. J., Anderson, J. (1993). Cross-modal semantic priming: A time-course analysis using event-related brain potentials. Language and Cognitive Processes, 8, 379-412.

Holmes, V. M., & Langford, J. . (1976). Comprehension and recall of abstract and concrete sentences. Journal of Verbal Learning and Verbal Behavior, 15, 559-566.

Hunt, E., & Agnoli, F. (1991). The Whorfian hypothesis: A cognitive psychology perspective. Psychological Review, 98, 377-389.

James, C. ((1975)). The role of semantic information in lexical decisions. Journal of Experimental Psychology: Human Perception and Performance, 104, 130-136.

Johnson, R. (1986). A triarchic model of P300 amplitude. Psychophysiology, 23, 367-384.

Jones-Gotman, M., & Milner, B. . (1978). Right temporal lobe contribution to image-mediated memory. Neuropsychologia, 16, 61-71.

Jones-Gotman, M. (1979). Incidental learning of image-mediated or pronounced words after right temporal lobectomy. Cortex, 15, 187-197.

Kieras, D. (1978). Beyond pictures and words: Alternative information processing models for imagery effects in verbal memory. Psychological Bulletin, 85, 532-554.

Klee, H., & Eysenck, M. W. . (1973). Comprehension of concrete and abstract sentences. Journal of Verbal Learning and Verbal Behavior, 12, 522-529.

Kosslyn, S. M. (1994). Image and brain: The resolution of the imagery debate. Cambridge, MA: MIT Press.

Kounios, J., Holcomb, P. (1992). Structure and process in semantic memory: Evidence from event-related brain potentials and reaction times. Journal of Experimental Psychology: General, 121, 460-480.

Kounios, J. H., P.J. . ((1994)). Concreteness effects in semantic processing: Event-related potential evidence supporting dual-coding theory. Journal of Experimental Psychology: Learning, Memory & Cognition, 20, 804-823.

Kounios, J. (1994). Psychophysiological measures of intelligence. In R. Sternberg (Ed.). Encyclopedia of intelligence. NY: Macmillan.

Kroll, J. F., & Merves, J. S. . (1986). Lexical access for concrete and abstract words. Journal of Experimental Psychology: Learning, Memory and Cognition, 12, 92-107.

Kutas, M. & Hillyard, S. (1980). Reading senseless sentences: Brain potentials reflect semantic incongruity. Science, 207, 203-205.

Kutas, M., Lindamood, T., Hillyard, S. A. (1984). Word expectancy and event-related brain potentials during sentence processing. In S. Kornblum, & Requin, R. (Ed.), Preparatory States and Processes (pp. (pp. 217-238)). Hillsdale, New Jersey: Erlbaum.

Kutas, M., Hillyard, S. (1984). Brain potentials during reading reflect word expectancy and semantic association. Nature, 307, 161-163.

Kutas, M., & Hillyard, S. A. (1989). An electrophysiological probe of incidental semantic association. Journal of Cognitive Neuroscience, 1, 38-49.

Kutas, M., Van Petten, C., Besson, M. (1988). Event-related potential assymetries during the reading of sentences. Electroencephalography and Clinical Neurophysiology, 69, 218-233.

Lupker, S. J. & Colombo, L. (1994). Inhibitory effects in form priming: Evaluating a phonological competition explanation. Journal of Experimental Psychology: Human Perception and Performance, 20(2), 437-451.

Mangun, G. R., & Hillyard, S.A. (1995). Mechanisms and models of selective attention. In M. D. R. M. G. H. Coles (Ed.), Electrophysilogy of mind . Oxford: Oxford University Press.

Marschark, M., & Paivio, A. . (1977). Integrative processing of concrete and abstract sentences. Journal of Verbal Learning and Verbal Behavior, 16, 217-231.

Marschark, M., Richman, C. L., Yuille, J. C., & Hunt, R. R. . (1987). The role of imagery in memory: On shared and distinctive information. Psychological Bulletin, 102, 28-41.

McCarthy, G., & Donchin, E. . ((1981)). A metric for thought: A comparison of P300 latency and reaction time. Science, 211, 77-79.

McCarthy, G., Nobre, A. C., Bentin, S., & Spencer, D. D. . (1995). Language-related field potentials in the anterior-medial temporal lobe: I Intracranial distribution and neural generators. Journal of Neuroscience, 15, 1080-1089.

McClelland, J. R., D. . ((1981)). An interaction activation model of context effects in letter perception: Part 1, An account of basic findings. Psychological Review, 88, 375-407.

McGarry-Roberts P.A., Stelmack, R.M., & Campbell, K.B. (1992). Intelligence, reaction time and event-related potentials. Intelligence, 16, 289-313.

Nelson, D. L., & Schreiber, T. A. . (1992). Word concreteness and word structure as independent determinants of recall. Journal of Memory and Language, 31, 237-260.

Neville, H. J., Mills, D.L., Lawson, D.L. (1992). Fractionating language: Different neural subsystems with different sensitive periods. Cortex, 2, 244-258.

Nigam, A., Hoffman, J.E., & Simons, R.F. (1992). N400 to semantically anomalous pictures and words. Journal of Cognitive Neuroscience, 4, 15-22.

Nobre, A. C., & McCarthy, G. (1994). Language-related ERPs: Scalp distributions and modulation by word type and semantic priming. Journal of Cognitive Neuroscience, 6, 233-255.

Nobre, A. C. M., G. . (1995). Language-related field potentials in the anterior-medial temporal lobe: II Effects of word type and semantic priming. . Journal of Neuroscience, 15, 1090-1098.

Osterhout, L., Holcomb, P. (1992). Event-related brain potentials elicited by syntactic anomaly. Journal of Memory and Language, 31, 785-806.

Osterhout, L., Holcomb, P.J. (1995). The electrophysiology of language comprehension. In M. G. H. C. M.D. Rugg (Ed.), Electrophysiology of mind . Oxford: Oxford University Press.

Paivio, A. (1971). Imagery and verbal processes. New York: Holt, Rinehart & Winston.

Paivio, A. (1986). Mental representations: A dual coding approach. New York: Oxford University Press.

Paivio, A. (1991). Dual coding theory: Retrospect and current status. Canadian Journal of Psychology, 45, 255-287.

Paivio, A. (1967). Paired-associate learning and free-recall of nouns as a function of concreteness, specificity, imagery, and meaningfulness. Psychological Reports, 20, 239-245.

Paivio, A., Rogers, T. B., & Smythe, P. C. . (1968). Why are pictures easier to recall than words? . Psychonomic Science, 11, 137-138.

Paivio, A., Yuille, J.C., & Madigan, S.A. (1968). Concreteness, imagery, and meaningfulness values for 925 nouns. Journal of Experimental Psychology: Monograph Supplement, 76, 1-25.

Paivio, A., & Begg, I. (1971). Imagery and comprehension latencies as a function of sentence concreteness and structure. Perception and Psychophysics, 10, 408-412.

Paivio, A., & Csapo, K. . (1973). Picture superiority in free recall: Imagery or dual-coding? Cognitive Psychology, 5, 176-206.

Paivio, A. t. L., J. . (1982). Imagery, memory, and the brain. Canadian Journal of Psychology, 36, 243-272.

Paller, K. A., Kutas, M., Shimamura, A. P., & Squire, L. R. . (1988). Brain responses to concrete and abstract words reflect processes that correlate with later performance on test of recall and stem completion priming. In J. R. R. Johnson, and R. Parasuraman (Ed.), Current trends in brain potential research . Amsterdam: Elsevier.

Peterson, S. E., Fox, P. T., Posner, M. I., Mintun, M., & Raichle, M. E. . (1988). Positron emission tomographic studies of the cortical anatomy of single-word processing. Nature, 331, 585-589.

Picton, T. W. (1992). The P300 wave of the human event-related portential. Electroenchephalography and Clinical Neurophysiology, 9, 456-479.

Pugh, K. R., Rexer, K., Peter, M., & Katz, L. (1994). Neighborhood effects in visual word recognition: Effects of letter delay and nonword context difficulty. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20(3), 639-648.

Reicher, G. M. (1969). Perceptual recognition as a function of meaningfulness of stimulus material. Journal of Experimental Psychology, 81, 275-280.

Richardson, J. T. E. (1980). Mental imagery and human memory. London: Macmillan.

Rugg, M. D. (1985). The effects of semantic priming and word repetition on event-related potentials. Psychophysiology, 22, 642-647.

Rugg, M. D. (1990). Event-related brain potentials dissociate repetition effects of high- and low-frequency words. Memory & Cognition, 18, 367-379.

Rugg, M. D. & Coles, M.G.H. (1995). The ERP and cognitive psychology: Conceptual issues. In M.D. Rugg & M. G. H. Coles (Ed.), Electrophysiology of mind. Oxford: Oxford University Press.

Sabol, M. A. DeRosa., D.V. (1976). Semantic encoding of isolated words. Journal of Experimental Psychology: Human Learning and Memory, 2, 58-68.

Schwanenflugel, P. J., & Shoben, E. . (1983). Differential context effects in the comprehension of abstract and concrete verbal materials. Journal of Experimental Psychology: Learning, Memory and Cognition, 9 , 82-102.

Schwanenflugel, P. J., Harnishfeger, K. K., & Stowe, R. W. . (1988). Context availability and lexical decisions for abstract and concrete words. Journal of Memory and Language, 27, 499-520.

Schwanenflugel, P. J. (1991). Why are abstract concepts hard to understand? In P. J. Schwanenflugel (Ed.), The psychology of word meanings, (pp. 223-250) . Hillsdale, NJ: Lawrence Erlbaum Associates.

Seidenberg, M. S., & McClelland, J. L. (1989). A distributed developemental model for word recognition and naming. Psychological Review, 96, 523-568.

Segui, J., & Grainger, J. (1990). Priming word recognition with orthographic neighbors: Effects of relative prime-target frequency. Journal of Experimental Psychology: Human Perception and Performance, 16(1), 65-76.

Smith, M. E., & Halgren, E. . (1988). ERPs during lexical decision: Interaction of repetition with concreteness, frequency, and pronounceability. In J. R. a. R. P. R. Johnson (Ed.), Current trends in brain potential research . Amsterdam: Elsevier.

Snodgrass, J. G., & Mintzer, M. (1993). Neighborhood effects in visual word recognition: Facilitory or Inhibitory? Memory & Cognition, 21(2), 247-266.

Sutton, S., Braren, M, Zubin, J. & John, E.R. (1965). Evoked potential correlates of stimulus uncertainty. Science, 150, 1436-1439.

Van Petten, C., Kutas, M. (1990). Interactions between sentence context and word frequency in event-related brain potentials. Memory & Cognition, 18, 380-393.

Van Petten, C., & Kutas, M. (1991). Influences of semantic and syntactic contex on open- and closed-class words. Memory & Cognition, 19, 95-112.

Warrington, E. K. (1981). Concrete word dyslexia. British Journal of Psychology, 72, 175-196.

1. To appear in: S. Soraci & W. McIlvane (Eds.), Perspectives on fundamental processes in intellectual functioning. Send correspondences to P. Holcomb, Department of Psychology, Tufts University, Medford, Ma 02155. The research reported in this paper was supported by NICHD grant number HD25889 to the third author.