Priming of strong semantic relations in the left and right visual fields: a time-course investigation

Abstract

Prior time-course investigations of cerebral asymmetries in word processing have sometimes reported hemisphere differences in the onset and duration of semantic priming. In the current study, very strongly related word pairs (categorical associates such as arm–leg) were employed in a low relatedness proportion lexical decision priming paradigm. A range of prime–target stimulus onset asynchronies (SOAs: 150–800 ms) was included. Only very weak evidence was obtained for a LVF priming lag at the briefest SOA, while priming was bilateral at moderately long SOAs. We consider these data in the context of previous time-course studies and suggest that, when highly semantically similar word pairs are used, a right hemisphere priming lag is, at best, a very small effect.

Introduction

It can no longer be claimed that language is the sole province of the left cerebral hemisphere (LH). Evidence from right hemisphere-injured patients [32], functional neuroimaging [36], event-related potentials [1], [28], and lateralized stimulus presentation in split-brain and neurologically normal individuals [8], [14], [27], [53] indicates that the right hemisphere (RH) also contributes to linguistic processing. In particular, comprehending the full meaning of words, phrases, and texts must include an appreciation of alternative and figurative meanings, distant allusions, inferences, and pragmatics, and it is in this domain that processes subserved by the right hemisphere are most evident [10]. But although a right hemisphere role in such phenomena is well-documented, the underlying processing mechanisms remain to be elucidated.

Techniques such as visual half-field presentation may provide an optimal method to identify hemisphere-specific processing mechanisms because large numbers of participants are readily available, both right and left hemisphere measurements can be taken within the same individual, and the method itself imposes few constraints on the design and length of experiments. Systematic experimental analyses of the conditions which differentiate left versus right hemisphere processing can be accomplished relatively quickly, allowing us to hone in on the differential mechanisms and test competing theories. Indeed, a great deal has been learned in the past several decades about the information processing basis of cerebral asymmetries using this technique [29].

Within the domain of language comprehension, it is critical to determine whether the RH role in discourse and higher level language processing involves asymmetries in word meaning activation and retrieval. Semantic priming techniques have been used to address this issue. When a target word, presented to the right or left visual field (RVF or LVF), is preceded by a semantically related prime word, responses are faster and more accurate than following an unrelated prime. Such semantic priming indicates that prime and target meanings were available to enhance processing in the receiving hemisphere. A large body of research indicates that words presented to either hemisphere are primed by related meanings, although not necessarily under the same conditions [14], [41]. Several factors have been identified which differentially modulate the priming obtained in the LVF and RVF, such as the nature of the semantic relation between prime and target words [16], [38], whether the task encourages attention to the prime–target relation (manipulations of relatedness proportion [16], [19]), and perhaps whether the prime word is made available to both hemispheres simultaneously (central prime) or to a single hemisphere (lateral prime) [16], [38]. Thus, the question is not whether semantic relatedness can enhance processing in the RH as well as the LH, but to determine the mechanisms via which semantic relatedness facilitates processing within each hemisphere.

One important consideration is the time course of meaning availability. The temporal availability of word meanings within each hemisphere will place constraints on the efficiency of sentence-level language processes which depend upon word meaning access (such as meaning integration, ambiguity resolution, and the construction of a “message”). By varying the stimulus onset asynchrony (SOA) between the prime and target both the onset of word meaning availability, and its duration, can be investigated. According to Neely [47] (see also [31]), semantic priming can be attributed to meaning activation if there is a brief SOA, particularly if only a low proportion of the experimental trials are semantically related. Expectancy processes or post-access relatedness checking would be implicated as priming mechanisms at longer SOAs, especially when there is a high proportion of related trials, and/or a high non-word ratio (proportion of unrelated trials with non-word targets). It has also been suggested that binary response tasks might facilitate post-access processing [47]. Although data have appeared which challenge some aspects of Neely’s model (e.g. [33], [50]), it still provides a useful starting point for conceptualizing asymmetries in semantic priming processes. In this study we consider whether left hemisphere superiority for language comprehension may in part be due to its more rapid accessibility to word meanings. If so, then this lexical–semantic “head start” could snowball and provide a benefit for many subsequent linguistic processes. On the other hand, if meaning accessibility has similar onsets across hemispheres, then overall LH language superiority would depend more on processes that extend beyond the word level.

Prior lateralization studies that have manipulated SOA are consistent with the view that, at different moments in processing, some word meanings are available in one hemisphere, but not another. However, asymmetries in the onset and duration of priming appear to vary depending upon the nature of the semantic relation between prime and target words. Below we review, in turn, time-course studies that have investigated four different types of semantic relations: the availability of alternate word meanings [4], [5], [6], [11], and word meanings related via association [2], [23], [46], category membership without association [3], [22], [37], and associated category members [7]. These priming studies are summarized in Table 1. We then outline how the time-course experiment presented here can address issues that prior studies have not investigated.

Alternate word meanings: When the availability of multiple meanings of an ambiguous prime word is probed, bilateral priming is observed for the dominant (most frequent) meaning at very early time courses (SOAs below 100 ms) [5], [6], [11], [17]. However, at SOAs of 35–50 ms less frequent (subordinate meanings) are more likely to be primed in the RVF than in the LVF, and an exhaustive pattern of meaning access (all word meanings primed) is only observed for RVF targets [5], [6], [11]. At slightly longer SOAs (80–200 ms), exhaustive access is observed for both VFs [17]. Much later in processing (750 ms SOA), the LVF/RH maintains exhaustive priming for multiple word meanings and senses, whereas within the RVF/LH word meaning availability is narrowed, with usually only the most dominant meaning maintained [5], [6], [11]. This pruning of word meaning availability could be due to the use of more controlled and/or post-access processes within the left hemisphere. The “early exhaustive–later narrow” pattern of meaning availability is also observed within the RVF for words that have both literal and metaphorical meanings [4]. In contrast, very early word meaning availability within the LVF/RH is reliably obtained only for a word’s most dominant meaning; it appears that metaphorical or subordinate meanings are available to the RH somewhat later in processing [4].

Associative relations: It is interesting that, although early priming of the dominant meanings of ambiguous words is seen in the LVF/RH, priming between strongly associated words (e.g. bee–honey) appears to have a later onset within the RH than the LH. That is, the LVF/RH obtains facilitation for these words at SOAs of 450 ms [2] or 750 ms [46], but not at briefer SOAs (67 ms [46]; 250 ms [2]). Whereas, priming for strongly associated words in the RVF is obtained uniformly across these same intervals [2], [46]. Collins and Coney [23] also investigated associative priming, but included only word pairs with low image primes and high image targets (e.g. series–book). In this case, associative priming was observed only when the prime word was presented to the RVF (for LVF targets at 250 ms SOA, both LVF and RVF targets at 450 ms SOA). This novel result may indicate that the left hemisphere must be directly stimulated to obtain early onset priming for low image associated words (see [23] for further discussion). In general, however, it appears that strongly associated meanings are available early in the left hemisphere, and maintained at later moments of processing, perhaps accompanied by suppression of more remote associations [46]. In contrast, associated meanings appear to have a more delayed onset of availability within the RH.

Category-only relations: The third group of studies investigated priming for non-associated category members (e.g. dog–goat). Here there is early onset priming within the RVF, but not the LVF (SOAs between 165 and 450 ms) [3], [22], [37]. LVF/RH priming is only obtained at longer SOAs [22], [37]. It is unclear whether RVF/LH categorical priming is maintained at these longer SOAs—Koivisto [37] using a low (0.25) relatedness proportion found no RVF priming, while Collins [22] using a higher (0.50) relatedness proportion did obtain RVF priming. This may suggest that this relatively weak semantic relation is not maintained within the LH, unless the experimental conditions encourage more controlled processing of the prime words. However, this suggestion will need to be verified by further experimentation. In any case, it appears that the onset of categorical priming is much delayed in the RH, relative to the LH.

Associative + category relations: Finally, Audet et al. [7] observed bilateral priming for strongly associated category members (cat–dog) at both short (150 ms) and long (800 ms) SOAs. Bilateral priming was also obtained for word pairs which instantiated a locative relation (boat–harbor), but only at the long SOA. It is interesting that the locative pairs had a delayed onset of priming, but it is difficult to interpret this effect as neither strength of association nor degree of semantic similarity for these stimuli was reported. However, the early onset and sustained availability of the category associates suggests that priming for very strongly related word pairs may be symmetrical across hemispheres.

Thus, although prior studies suggest some interesting hemisphere differences in the time course of meaning availability, the data are not entirely consistent. It does appear that the onset of priming for weakly related meanings (subordinate meanings of ambiguous words, non-associated category members) occurs somewhat later in the LVF/RH than in the RVF/LH. Furthermore, the RH appears to maintain weakly related or metaphoric meanings at later intervals (750–800 ms) when such meanings are no longer available in the LH. However, the data for more strongly related word pairs is less consistent. On the one hand, several studies have reported early, and bilateral, priming for the dominant associates of ambiguous words which are by definition strongly related [5], [6], [11], [17]. On the other hand, when the primes are not restricted to ambiguous words, priming for strongly associated word pairs has a later onset in the LVF/RH, than in the RVF/LH [2], [46]. One possible explanation for this discrepancy could relate to the nature of meaning representations for associates of ambiguous versus non-ambiguous words. The ambiguous words used in prior studies have meanings which are inconsistent with each other (e.g. financial bank, bank–money versus earth formation, bank–river) and which may involve some competition for attentional resources [49]. Meanings which are strongly associated to unambiguous words (e.g. bee–honey) would be unlikely to have competitors with inconsistent meanings. However, it is difficult to imagine a theoretical account that would posit earlier RH meaning availability for strongly related words with competing inconsistent meanings, but not for those where meaning competition is absent. Moreover, there are some significant methodological differences between the experiments investigating ambiguity and association. Primes were centrally presented in the ambiguity experiments, but laterally presented in the association studies (see Table 1). Thus, it might be argued that the priming observed in the ambiguity studies involved some degree of LH and RH collaboration, while that obtained in the association experiments might represent a “truer” reflection of RH processing capabilities [38], [42].

In addition, one can question whether strength of association is necessarily the best way to operationalize the degree of semantic relatedness. Recently, some investigators have argued that automatic semantic priming is more a function of the degree of meaning overlap between prime and target which may be independent of associative strength [44], [45], [50]. Furthermore, connectionist models have successfully simulated priming effects using semantic feature overlap as a metric of relatedness [43], [45], [48]. Categorical associates (e.g. dog–cat, arm–leg) have a high degree of semantic feature overlap relative to non-associated category members [14], [16].¹ Hence such pairs may represent the most strongly related word meanings, regardless of whether strength is measured by association or semantic feature overlap. Audet et al.’s [7] finding of early bilateral priming for such pairs may indicate that meaning activation within the RH is not delayed for strongly related word meanings. However, this conclusion may be premature because Audet et al. employed central primes which may engender inter-hemispheric collaboration. Further, the moderately high relatedness proportion used in that study could have allowed attentional and/or relatedness checking strategies to influence priming (a limitation that also characterizes most prior time-course investigations—see Table 1).

Finally, with the exception of Koivisto [37] prior time-course studies have either contrasted only two SOAs (e.g. [3], [11], [46]) or have investigated SOAs within a very narrow time window [17]. This makes it difficult to determine just when delayed priming within the RH may onset, or when LH priming is narrowed. The time-course review presented above required comparisons across studies (e.g. [11], [17]) in order to make such inferences, and methodological differences between studies make such conclusions tentative at best.

It is quite important theoretically to determine whether strongly related word meanings are equally available to each hemisphere at various moments of processing, or whether there is a delayed RH onset of priming for such pairs. If even strongly related meanings have a delayed onset within the RH, this would suggest that meaning activation in general is slowed within the right hemisphere. This would imply either that RH language processing depends on earlier meaning activation within the LH, or that semantic activation within the right hemisphere is more sluggish. On the other hand, if strongly related meanings are available bilaterally at very early moments of processing, then a more robust RH language processor would be implicated. In this study, we investigate the time course of meaning availability across hemispheres using word pairs that are very strongly related via both association and semantic similarity (i.e. category associates), lateralized primes, and a low relatedness proportion across five SOAs (150–800 ms). It is interesting that all prior studies using such pairs have reported bilateral equivalent priming [7], [16], [19]. However, those which used a low relatedness proportion utilized a moderate to long SOA [16], while those including brief SOAs employed higher relatedness proportions and centrally presented primes [7].

If there is a delayed RH onset of meaning availability for strongly related words, then we should observe an earlier onset of priming when both primes and targets are presented to the RVF/LH, than to the LVF/RH. However, if priming is obtained at the same SOAs across visual fields, then this would support the idea that at least some word meanings are bilaterally available early in processing.