Elsevier

Brain and Language

Volume 100, Issue 1, January 2007, Pages 23-43
Brain and Language

Hemispheric differences in strong versus weak semantic priming: Evidence from event-related brain potentials

https://doi.org/10.1016/j.bandl.2006.06.117Get rights and content

Abstract

Goals

Research with lateralized word presentation has suggested that strong (“close”) and weak (“remote”) semantic associates are processed differently in the left and right cerebral hemispheres [e.g., Beeman, M. j., & Chiarello, C. (1998). Complementary right- and left-hemisphere language comprehension. Current Directions in Psychological Science, 7(1), 2–8]. Recently, this hypothesis has been challenged [Coney, J. (2002). The effect of associative strength on priming in the cerebral hemispheres. Brain and Cognition, 50(2), 234–241]. We predicted that foveal presentation of strong and weak associates would elicit different patterns of hemispheric activity, as indexed by high-density event-related brain potentials (ERPs), and that source localization of the scalp potentials would help clarify the nature of hemispheric contributions to semantic organization.

Methods

128-channel ERPs were recorded in two experiments as subjects performed a lexical decision task. Word trials were equally divided into strongly related, weakly related, and unrelated word pairs. All words were foveally presented. SOA was 800ms in Experiment 1, and 200ms in Experiment 2.

Results

Topographic analyses revealed medial frontal (MFN) and parietal (N400/LPC) effects for both strong and weak associates. Between ∼450 and 550 ms, the magnitude of the N400/LPC effect indicated priming for both strong and weak associates over left parietal sites, while priming over right parietal sites was restricted to strongly related word pairs. During this interval, spatiotemporal source modeling showed that these scalp effects were best accounted for by ipsilateral sources in the medial temporal lobe. The observed pattern of asymmetries for strong versus weak associates is not consistent with certain proposals regarding the complementarity of right- and left-hemisphere contributions to semantics. It is, however, consistent with findings from visual half-field studies (Hasbrooke and Chiarello, 1998). We discuss the relevance of these results for theories of hemispheric asymmetry and meta-control in lexical semantic access.

Introduction

When two word meanings are related in memory, activation of one meaning in response to a visually presented prime word facilitates subsequent access to a related (target) word meaning. This facilitation or priming manifests as improved speed or accuracy in response to related targets, as compared with an unrelated or “neutral” baseline condition (Neely, 1991). The semantic priming effect has been observed across a range of tasks (e.g., lexical decision, pronunciation, semantic categorization) and has proven to be an important tool for probing the nature of language and cognition, resulting in a large corpus of semantic priming data in psychology of language and related fields (Hutchison, 2003; Neely, 1991).

Despite the widespread interest in semantic priming, some basic questions remain unresolved. One key question is how strength of association (or semantic distance) between concepts affects the nature of semantic priming within and across the two hemispheres. Are strongly and weakly related word pairs processed similarly in the left and right cerebral cortex, or are there qualitative differences in how the two hemispheres process strong (“close”) versus weak (“distant”) semantic relationships?

Two prevailing responses to this question have emerged. According to one view—referred to here as the Complementarity Hypothesis—strong and weak associates should evoke different patterns of activity over the left and right hemispheres, reflecting qualitative differences in left and right hemisphere processing of semantic information. For example, Chiarello, Beeman, and associates have argued that weakly (or “remotely”) related word meanings are processed more easily in the right hemisphere, which functions to maintain a wide range of meanings, while the left hemisphere can support priming of both strong and weak associates at short SOAs, but mainly supports strong semantic priming due to meaning selection (focused attention) at later time points (see Beeman & Chiarello, 1998 for a review).

An alternative view is that strong and weak associates engage the very same cognitive and neural processes, and any differences in strong and weak priming are strictly quantitative, rather than qualitative, in nature (Coney, 2002). Thus, according to this position, which is referred to here as the Equivalence Hypothesis, strong and weak associates should be processed similarly by the two hemispheres.

To address this issue, the present study measured high-density event-related potentials (ERPs) to semantic associates that varied in strength of association (i.e., strong, weak, and unrelated word pairs). Two experiments were performed, using a lexical priming paradigm and two prime-target SOAs to examine time course of hemispheric priming effects. The basic logic was that if strong and weak associates elicited different patterns of lateralized brain activity, this would support the Complementarity Hypothesis. On the other hand, if the same spatiotemporal patterns of brain activity were observed, this result would be more consistent with the Equivalence Hypothesis. Because the topography of scalp ERPs does not point directly to underlying patterns of brain activity, a spatiotemporal dipole method was used to generate a neural source model of left and right hemisphere activity in strong versus weak semantic priming.

Studies of semantic priming have revealed multiple mechanisms that may influence activation and retrieval of word meanings in different contexts (Hutchison, 2003; Neely, 1991). According to one influential theory of semantics, semantic priming can arise through automatic spreading activation (ASA), a process by which activation of one concept automatically spreads to related (“linked”) concepts. On this view, word meanings are conceptualized as nodes within a Hebbian network: related concepts are connected via links, and the strength of a particular semantic relationship is represented by the “distance” between nodes (Collins and Loftus, 1975, Meyer and Schvaneveldt, 1971). Activation of one semantic node will therefore spread to, or activate, related nodes in memory. This process is “automatic” in the sense that it can occur in the absence of executive control processes that require effort and awareness (Posner & Snyder, 1975). In addition to ASA, researchers have identified several types of non-automatic or controlled semantic processing (Neely, 1976, Neely, 1977, Neely, 1991, Posner and Snyder, 1975; Shiffrin & Schneider, 1977), including postlexical “matching” and verification strategies (Becker, 1980), and semantic integration (e.g., Brown & Hagoort, 1993).

Given that multiple mechanisms have been implicated in semantic processing, it is not surprising that research is uncovering a widespread network of areas that are active during meaning comprehension and that vary as a function of task parameters, such as degree of associative strength or SOA (Braeutigam et al., 2001, Frishkoff et al., 2004, Rossell et al., 2003). A current challenge is to parcellate these areas according to function, and to clarify when and how these different mechanisms operate to give rise to semantic priming.

Some prior research has reported differential patterns of hemispheric asymmetry in response to strong and weak semantic relationships, in support of the Complementarity Hypothesis (e.g., Beeman & Chiarello, 1998). According to Beeman and Chiarello, weakly (remotely) related word meanings appear to be preferentially processed by the right, as opposed to the left, hemisphere, while the left hemisphere appears to support priming of either strong or weak associates, depending on the time course (but cf. Koivisto, 1997).

A now-classic study by Burgess and Simpson (1988) provided evidence for hemispheric differences in time course, as well as distance or strength, of semantic processing. In this experiment, Burgess and Simpson examined semantic distance effects in the two hemispheres using ambiguous primes (e.g., “bank”) that were followed by words encoding either the dominant or the subordinate meaning of the prime word (e.g., MONEY versus RIVER). To examine hemispheric contributions, target words were presented to either the left or right visual field, and SOA was varied to map the time course of hemispheric differences in priming. At a short (35 ms) SOA, the two meanings showed similar facilitation in the RVF/LH, but only the dominant meaning was activated in the LVF/RH. Conversely, at a long (750 ms) SOA, the subordinate meaning was suppressed in the RVF/LH, but showed equivalent or greater facilitation than the dominant meaning in the LVF/RH. These results are consistent with the idea that the LH may initially activate both meanings, but that the subordinate meaning is later suppressed. By contrast, dominant meanings appear to be primed in the RH at short SOAs, but then decay at longer SOAs, just as the subordinate meaning is becoming more activated over the RH.

On the other hand, data from Coney (2002) has suggested that strong and weak associates evoke similar patterns of priming in the two hemispheres. Coney examined priming of words that were strongly, moderately, or weakly associated; word pairs were presented to the right or left visual field, and SOA was varied (250 versus 1000 ms). As expected, he observed a linear decrease in reaction time with increases in associative strength. Moreover, responses to related words were faster over the RVF/LH, and priming effects were somewhat stronger at the long SOA. Importantly, however, there was no interaction between associative strength and visual hemifield, and no interaction between hemifield and SOA. These results cast some doubt on the existence of qualitative differences in priming of strong and weak associates as a function of time course and hemispheric asymmetry.

The hypothesis that strong and weak associations should evoke qualitative differences in brain activation rests on the idea that different cognitive processes support activation of close (or strong, focal) versus remote (or broad, diffuse) semantic relationships. The very concept of semantic “distance” between concepts is implicit in models of spreading activation. Indeed, in their original formulation of the spreading activation model, Collins and Loftus (1975) suggested that related concepts are “stored close together” in psychological and neuronal space. This metaphorical description in turn raises important questions about how to operationalize the notion of semantic “distance” and about the theoretical assumptions implicit in the use of different measures.

One common measure of distance considers the number of intervening concepts between prime and target (so-called direct versus indirect associative “links”). The use of direct and indirect associates assumes a model of spreading activation in which activation of concept A results in automatic spreading of activation (ASA) to related concepts. Assuming that activation is allowed to spread across multiple related concepts (“nodes”), this implies that activation of A can lead to activation of concepts that are not directly associated with A: i.e., A  B  C, Thus, A primes C, even though A and C are not directly associated. An example of an indirect association is ‘lemon’–‘sweet,’ where the intermediate link is the word ‘sour’ (Kiefer et al., 1998, Weisbrod et al., 1998, Weisbrod et al., 1999). The strength of this indirect priming will depend on a decay function: how quickly activation decreases over time and “distance.”

Work by McNamara (1992) and by Spitzer and associates (Kischka et al., 1996, Moritz et al., 2001, Spitzer et al., 1993, Weisbrod et al., 1998, Weisbrod et al., 1999) has demonstrated qualitative, as well as quantitative, differences in the processing of direct and indirect associates. Most important for the present discussion, Kiefer et al. (1998) observed differences in lateralization of ERP priming effects for direct versus indirect semantic associates: Whereas direct associates elicited a negativity over bilateral inferior prefrontal sites, indirect associates elicited only a right hemisphere negativity over lateral inferior prefrontal sites. These results are consistent with the hypothesis that “close” (direct) and “remote” (indirect) semantic associations engage the two hemispheres in qualitatively different ways. Thus, studies using direct and indirect priming suggest possible differences in the distance or scope of semantic priming between the two hemispheres.

An alternative is to characterize semantic distance in terms of associative strength, where strongly related concepts are considered to be closer in conceptual–semantic space (e.g., de Groot et al., 1982, Rodel et al., 1992). Strength of association can be quantified through empirical measures, such as discrete free association (Nelson, McEvoy, & Schreiber, 1998) and relatedness ratings. In comparison with direct/indirect associates, the use of strong and weak associates has several advantages. First, associative strength has intrinsically more gradations: whereas direct/indirect priming extends to two or three levels at most, strength of association can be defined on a continuous scale. Through the use of parametic (item) analyses, this approach can in turn provide a more powerful test of hypotheses regarding semantic distance (cf. Dien, Frishkoff, Cerbone, & Tucker, 2003). Second, associative strength is a more transparent measure. As underscored by the work of McKoon and Ratcliff (1992), it is not obvious how indirect associates can be distinguished from weak associates: words such as ‘lemon’ and ‘sweet’ may be assumed (on intuition) to have no direct association, but this can be difficult to verify empirically. On the other hand, strength of association can be measured in several ways, but the notion that some words are more strongly associated than others rests on solid footing: studies of learning and memory over the past several decades have shown many effects of varying the strength of association between words and concepts (Coney, 2002, Deese, 1965, de Groot et al., 1982, Rodel et al., 1992).

Finally, the construct of associative strength may be more theoretically neutral. It is not necessary to maintain that associative strength is isomorphic with “distance in semantic space”; effects of associative strength may be equally compatible with other theories of semantic priming, such as those based on distributed semantic architectures (McRae et al., 1999, McRae et al., 1997).

Given the hemispheric asymmetries observed for direct and indirect priming in previous research an important question is whether strong and weak associates show a similar pattern that would point to qualitative, as well as quantitative, differences for strong and weak semantic priming. To examine this question, two experiments were conducted, in which high density (128-channel) ERPs and behavioral measures were recorded as subjects read words (targets) presented on a computer screen. Targets were preceded by single words (primes) that were either semantically related or unrelated to the target words, where degree of relatedness was varied parametrically. Experiment 1 used a relatively long SOA (800; 200 ms interword interval). The long interword interval (IWI) was selected to minimize overlap of the ERPs to the prime and target stimuli, permitting easy identification of the major spatiotemporal components of the response to the target. At the same time, the 800 ms SOA allowed for multiple influences on semantic priming, including forward-acting expectancy for the target (Neely, 1991). Experiment 2 used a comparatively short SOA (200; 50 ms IWI) in order to minimize the role of expectancy (cf. Hill et al., 2002, Rossell et al., 2003). Although the short IWI resulted in overlap of the ERPs to the prime and target, the prediction was that ERP results from Experiment 1 would provide a blueprint for interpreting condition differences in Experiment 2. In addition, the use of long and short SOAs would permit us to examine the time course of priming for strong and weak associates, as measured by different ERP components.

One advantage of ERP measures is that hemispheric differences can be detected even with centrally presented stimuli. An important limitation of visual half-field studies is that they separate hemispheric differences in semantic processing under the unusual conditions of initial input being isolated to one hemisphere’s visual cortex. By contrast, the present study employed used foveal, i.e., bilateral visual field, presentation of words, with the rationale that words are typically foveated during normal reading processes (Reichle & Perfetti, 2003), and that hemispheric asymmetries in processing would be observable in asymmetric patterns of electrophysiological activity. An important caveat is that the lateral distribution of scalp-recorded ERPs does not point unequivocally to neural generators in a particular region: a right-lateralized distribution can be due to either a left or right hemisphere generator. For this reason, in the present study traditional scalp ERP analyses are complemented by the use of spatiotemporal dipole modeling, using the methodology detailed in Frishkoff et al. (2004). This approach allows us to examine intensity of activation in right versus left cortical regions with millisecond time resolution and provides a way to parcellate the multiple stages and processes involved in lexical semantic priming.

Section snippets

Subjects

Fifty participants were recruited from introductory courses in Psychology at the University of Oregon. Subjects were native English speakers who were right-handed, with normal or corrected-to-normal vision. Academic course credit was given in exchange for participation. Five subjects had two few good EEG trials after artifact rejection; three additional subjects had excessive alpha or frontal EMG. This left 42 subjects (20 male; mean age = 19.7, SD = 3.1; 24 with at least one left-handed relative)

Methods

Methods were identical to those in Experiment 1, except that the prime–target SOA was 200 rather than 800 ms (150 ms prime; 50 ms ISI). Sixty-two subjects were run in the experiment. One session was aborted due to equipment failure, six datasets had too few trials after rejection of artifacts (<20 per cell), and eight subjects had data that were otherwise noisy due to excessive alpha, EMG, or blink recovery artifacts. The 47 remaining subjects were all right-handed, average age 19.9 (SD = 2.9).

Meta-analysis: Semantic relatedness × SOA

To examine differences in time course of strong and weak semantic priming, we treated SOA as a between-subjects variable, and combined the data from Experiments 1 and 2.

General discussion

Since the inception of priming research, researchers have been curious about the time course and scope of semantic priming: how far and how fast does activation spread from one concept to another? The objective of the present research was to characterize behavioral and ERP effects of associative strength (or semantic “distance”), and to examine hemispheric differences in strong and weak semantic priming as a function of SOA. In contrast with most research on hemispheric differences in

Acknowledgments

This research was supported by a research fellowship award from the U.S. National Institute of Mental Health, #MH66544. I wish to thank Don M. Tucker for helpful discussions and for his review of previous drafts, Tara Torrassa for assistance with data acquisition, and Karsten Hoechstetter for advice on the presentation of source localization data.

References (63)

  • G.A. Frishkoff et al.

    Frontal and posterior sources of event-related potentials in semantic comprehension

    Cognitive Brain Research

    (2004)
  • H. Hill et al.

    Automatic vs. controlled processes in semantic priming—differentiation by event-related potentials

    International Journal of Psychophysiology

    (2002)
  • B.W. Johnson et al.

    High-density mapping in an N400 paradigm: evidence for bilateral temporal lobe generators

    Journal of Clinical Neurophysiology

    (2000)
  • M. Kiefer et al.

    Right hemisphere activation during indirect semantic priming: evidence from event-related potentials

    Brain Language

    (1998)
  • U. Kischka et al.

    Dopaminergic modulation of semantic network activation

    Neuropsychologia

    (1996)
  • M. Koivisto

    Time course of semantic activation in the cerebral hemispheres

    Neuropsychologia

    (1997)
  • M. Kutas et al.

    Electrophysiology reveals semantic memory use in language comprehension

    Trends in Cognitive Science

    (2000)
  • K. Marinkovic et al.

    Spatiotemporal dynamics of modality-specific and supramodal word processing

    Neuron

    (2003)
  • C.M. Michel et al.

    EEG source imaging

    Journal of Clinical Neurophysiology

    (2004)
  • M. Rodel et al.

    Hemispheric dissociation in judging semantic relations: complementarity for close and distant associates

    Brain Language

    (1992)
  • S.L. Rossell et al.

    The anatomy and time course of semantic priming investigated by fMRI and ERPs

    Neuropsychologia

    (2003)
  • M. Scherg et al.

    Two bilateral sources of the late AEP as identified by a spatio-temporal dipole model

    Electroencephalography and Clinical Neurophysiology

    (1985)
  • M. Spitzer et al.

    Indirect semantic priming in schizophrenic patients

    Schizophrenic Research

    (1993)
  • M. Weisbrod et al.

    Electrophysiological correlates of direct versus indirect semantic priming in normal volunteers

    Cognitive Brain Research

    (1999)
  • Y. Wang et al.

    Common spatial subspace decomposition applied to analysis of brain responses under multiple task conditions: a simulation study

    Journal of Clinical Neurophysiology

    (1999)
  • K.S. Binder et al.

    Extraction of information to the left of the fixated word in reading

    Journal of Experimental Psychology: Human Perception and Performance

    (1999)
  • C.A. Becker

    Semantic context effects in visual word recognition: an analysis of semantic strategies

    Memory & Cognition

    (1980)
  • M.J. Beeman et al.

    Complementary right- and left-hemisphere language comprehension

    Current Directions in Psychological Science

    (1998)
  • S. Bentin et al.

    Erp manifestations of processing printed words at different psycholinguistic levels: Time course and scalp distribution

    Journal of Cognitive Neuroscience

    (1999)
  • J. Boddy

    Event-related potentials in chronometric analysis of primed word recognition with different stimulus onset asynchronies

    Psychophysiology

    (1986)
  • C. Brown et al.

    The processing nature of the n400: Evidence from masked priming

    Journal of Cognitive Neuroscience

    (1993)
  • Cited by (41)

    • Understanding associative vs. abstract pictorial relations: An ERP study

      2019, Neuropsychologia
      Citation Excerpt :

      Interestingly, N400 amplitude is known to be modulated by relations' strength, with strongly-related pairs evoking the weakest amplitude, moderate relations yielding an intermediate amplitude, and unrelated pairs giving rise to the strongest, most negative amplitude. This was found both for object pairs (McPherson and Holcomb, 1999; see also Barrett and Rugg, 1990; Ganis et al., 1996 for use of objects, yet without manipulating relations strength) and for words (Frishkoff, 2007; Kutas and Hillyard, 1984). In the metaphors literature, a similar modulation of N400 amplitude was reported on the literality-metaphoricity or ‘novelty’ axis, such that literal pairs showed the weakest N400 amplitude, followed by conventional metaphors, which in turn elicited weaker waveforms than novel metaphors, and lastly unrelated pairs evoked the most negative, strongest N400 (Arzouan et al., 2007; De Grauwe, Swain, Holcomb, Ditman & Kuperberg, 2010; Coulson and Van Petten, 2002).

    • The effect of associative strength on semantic priming in schizophrenia

      2018, Psychiatry Research
      Citation Excerpt :

      To our knowledge, it is the first time that the strength effect is explored in schizophrenia patients in studies of semantic priming. Consistent with previous studies in healthy groups (e.g. Coney, 2002; Frishkoff, 2007), our results show associative strength effect on semantic priming. Reaction times were faster for strongly than for weakly associated pairs, and faster for weakly associated pairs than for non-associated pairs.

    View all citing articles on Scopus
    View full text