- Email: [email protected]
The sensitivity of the right hemisphere to contextual information in sentences
Brain & Language 110 (2009) 95–100
Contents lists available at ScienceDirect
Brain & Language journal homepage: www.elsevier.com/locate/b&l
Short Communication
The sensitivity of the right hemisphere to contextual information in sentences Bethanie Gouldthorp *, Jeffrey Coney School of Psychology, Murdoch University, South Street, Murdoch, Western Australia 6150, Australia
a r t i c l e
i n f o
Article history: Accepted 16 May 2009 Available online 13 June 2009 Keywords: Sentence comprehension Hemisphere Visual field Priming Context Sentence length
a b s t r a c t One explanation for the inconsistencies in research examining the sentence comprehension abilities of the right hemisphere (RH) is the presence of confounding variables that have generally served to disadvantage the processing capacities of the RH. As such, the present study aimed to investigate hemispheric differences in the use of message-level sentential information by removing some of the factors known to be inherently disadvantageous for RH comprehension. Thirty-two right-handed undergraduate university students participated in a computer-based lexical decision task where RT and error rates were recorded. The sensitivity of each hemisphere to the message-level contextual information contained in short versus long sentences was compared, as well as the effect of stimulus modality (visual compared to auditory). The results showed that the RH benefited from increased levels of context to at least the same extent as the LH and that, more importantly, this could not be explained by word-level processes alone. This finding, unusual in behavioral research on normal individuals but consistent with neuropsychological, electrophysiological, and neuroimaging approaches, suggests that the RH plays an important role in sentence comprehension, at least in relation to sentences that conform to a relatively simple structure. Ó 2009 Elsevier Inc. All rights reserved.
1. Introduction Although the dominance of the left cerebral hemisphere (LH) in language processing and production has been repeatedly demonstrated over the past few decades, studies have also produced evidence for right hemisphere (RH) involvement in certain areas of language comprehension (e.g., Coulson, Federmeier, Van Petten, & Kutas, 2005; St. George, Kutas, Martinez, & Sereno, 1999). Many differences in the language processing of the left and right hemispheres have been revealed through studies utilizing the differing levels of constraint produced by sentence primes (Faust & Kravetz, 1998). For example, the RH appears to be relatively less affected by linguistic constraint, whereas the LH appears able to take advantage of it (Faust, Babkoff, & Kravetz, 1993). These differences are thought to be the result of the hemispheres using different linguistic mechanisms for sentence comprehension (Faust & Babkoff, 1997). Morris (1994) suggested that the LH interprets contextual information by using ‘‘message-level” mechanisms (i.e., by combining syntactic, semantic, and pragmatic information in order to build a conceptual representation of the meaning of the sentence), whereas the RH uses ‘‘word-level” mechanisms (i.e., by using lexical information and basic relations between individual words in sentences). This proposal is supported by studies that have found that it is only in the LH that normal sentences are much more effective primes than the same words presented in a scrambled list (Faust, Babkoff, & Kra* Corresponding author. Fax: +61 08 9360 6492. E-mail address: [email protected] (B. Gouldthorp). 0093-934X/$ - see front matter Ó 2009 Elsevier Inc. All rights reserved. doi:10.1016/j.bandl.2009.05.003
vetz, 1995; Faust & Chiarello, 1998). Furthermore, increasing the amount of context from one to three to six words produced larger priming effects when presentation was to the RVF/LH but not the LVF/RH (Faust et al., 1993). Additionally, lexical decision latencies were increased for implausible compared to plausible sentence-final words when presented to the RVF/LH but not the LVF/RH. Faust and Kravetz (1998) take these findings as evidence that the LH is more sensitive to linguistic information conveyed at the sentence message-level than the RH. Furthermore, they argue that while the LH is able to suppress contextually inappropriate words or meanings, the RH’s comprehension ability is compromised as a result of the ‘flooding’ of irrelevant information due to its sustained activation of multiple meanings and associations. Nonetheless, the results of these studies supporting a word-level only processing ability of the RH are inconsistent with the findings in relation to patients with RH damage; for example, difficulties with comprehension of metaphors, inferences and coherence maintenance (see Faust, Barak, & Chiarello, 2006 for a review). In addition, ERPs observed in neurologically normal individuals suggest that both hemispheres are primarily driven by message-level congruity during sentence reading, although they utilize this information in different ways (Coulson et al., 2005; Federmeier, Mai, & Kutas, 2005). Similarly, St. George et al. (1999) provided evidence that the way in which the RH is able to integrate information into a coherent context means it may be particularly important in processing at the discourse level. The findings of these sorts of studies suggest that the RH plays an important role in comprehending and integrating language and
96
B. Gouldthorp, J. Coney / Brain & Language 110 (2009) 95–100
in the application of contextual information. Furthermore, several behavioral studies of normal individuals have recently demonstrated a RH sensitivity to context/constraint under certain circumstances (see Gouldthorp & Coney, 2009, for a review). It is possible that the inconsistent nature of behavioral studies of the capabilities and mechanisms of the RH in language comprehension when compared to observations made by the other methodologies (i.e., neuropsychological, neuroimaging and electrophysiological studies) may be the result of several methodological problems. For example, the variation of sentence structure as a means of manipulating different levels of constraint for the same target word (e.g., Faust & Kravetz, 1998) often produces quite dramatic variations in grammatical construction. It is not common in the literature, however, for researchers to carefully match sentence structures across each level of constraint, presumably because of the very considerable difficulties that such matching entails. As a result, there remains the possibility that although these studies purport to be investigating the effect of varying levels of contextual information on hemispheric comprehension, the variation in sentence structure may confound any attempt to properly compare the performance of the two hemispheres. Nonetheless, studies that have attempted to manipulate context by increasing the number of words (e.g., Faust et al., 1993) have still failed to produce results that are consistent with the findings of neuropsychological, electrophysiological, and neuroimaging studies. Other uncontrolled variables may also contribute to the inconsistencies. In several studies, determining the message-level content of the sentence was contingent on appreciating the linguistic function of the verb or verb phrase (e.g., Chiarello, Faust, & Liu, 1999; Faust et al., 1995). There is a substantial body of research that suggests the RH has difficulty in processing verbs (e.g., Gazzaniga & Sperry, 1967; Sereno, 1999). As such, the dependence of previous studies on the critical role of the verb in simple sentences could have partially masked any evidence of RH sentence comprehension. Furthermore, despite some evidence that the noun/verb distinction is better accounted for by differences in imageability than grammatical class itself (e.g., Chiarello, Shears, Liu, & Kacinik, 2005), there continues to be a general failure in the literature to control for imageability in lieu of (or in addition to) grammatical class. One further factor may be of importance in relation to studies of sentence comprehension in the RH; the modality of stimulus presentation for behavioral investigations into hemispheric sentence comprehension has been traditionally visual. There is, however, substantial evidence to suggest that the RH processes language better when received in auditory form than in visual (e.g., Crinion & Price, 2005; Zaidel, 1976). Crossmodal lexical priming paradigms (CMLP) also potentially circumvent some of the difficulties produced by entirely visual lexical decision tasks, such as ensuring central fixation. While relatively few studies have addressed auditory sentence comprehension in particular, none appear to have compared the specific effect that stimulus modality has on hemispheric differences in language capacities and, as such, these potential benefits remain speculative. Based on these observations, the assertion that the RH relies purely on word-level mechanisms needs to be tested using stimuli that are less likely to mask the possibility that the RH is sensitive to message-level mechanisms, particularly given the contradictory evidence from alternative methodologies (e.g., Federmeier, 2007). For example, target words should comprise only short, concrete, highly imageable nouns in order maximize the likelihood that they can be processed appropriately by the RH. Additionally, sentence stimuli should adhere to a consistent grammatical structure across conditions. As observed earlier, this creates particular difficulties for studies attempting to compare the effects of varying levels of contextual information (e.g., Faust & Kravetz, 1998). If sentence
length remains the same, it becomes difficult to retain the same structure while including additional context. On the other hand, sentence length may be increased to accommodate the additional contextual information, provided the longer sentences conform to a broadly similar structure. For example, context could be manipulated by the consistent addition of a single extra predicate to the canonical subject–verb–object structure. While not all factors that might disadvantage the RH can be easily addressed (e.g., the lexical decision task is in itself a concern), the inconsistency of results suggests that alternative approaches within this paradigm must be explored in order to bridge this gap. All of these proposed changes to the usual normal-behavioral methodology were implemented in the present study, of which the main aim was to compare the sensitivity of each hemisphere to the extra contextual information provided in long sentences relative to short sentences that differed structurally in a minor and consistent fashion. Scrambled variants of these sentences were also included as a control measure of word-level processing (i.e., removing the syntactic information). Based on the evidence produced by neuropsychological, electrophysiological, and neuroimaging methodologies indicating that the RH is not only sensitive to message-level information (e.g., Federmeier et al., 2005) but potentially makes better use of contextual information than the LH (e.g., Joanette, Goulet, & Hannequin, 1990), it was hypothesized that the difference in response times for short and long sentences (i.e., low and high context, respectively) would be greater for the right than for the left hemisphere. In addition, it was hypothesized that the RH would be more sensitive to sentential information when presented in an auditory, rather than visual, format. 2. Results RT data for correct responses were initially screened utilizing a deletion criterion of ±3.0 standard deviations from each individual’s mean response times for each condition. This resulted in only 0.44% of the total observations being excluded. An additional screening procedure was applied to the data in order to compare each participant’s mean RTs for each condition with the overall sample mean for the corresponding condition. Data were screened using the Winsorization method (Barnett & Lewis, 1994) at two standard deviations from the mean. This resulted in an additional 4.49% of mean responses being adjusted. An overall 2 2 2 2 repeated-measures analysis of variance (i.e., stimulus modality, visual field, sentence length and sentence type) was performed on the screened RT data (summarized in Table 1). Initial analysis of the results did not reveal a significant main effect of stimulus modality (F(1, 31) = .290, MSe = 7681.2, p = .594). Furthermore, as there was no significant interaction between stim-
Table 1 Mean RT (ms) as a function of target visual field, stimulus modality, sentence length, and sentence type. Standard deviations (ms) appear in parentheses. RVF/LH Normal sentences Visual sentence primes Short sentences 464 (75) Long sentences 438 (76) Auditory sentence primes Short sentences 464 (70) Long sentences 432 (81)
LVF/RH Scrambled sentences
Normal sentences
Scrambled sentences
422 (78)
444 (61)
447 (80)
385 (70)
420 (78)
396 (70)
410 (69)
436 (67)
435 (73)
373 (72)
411 (78)
381 (70)
B. Gouldthorp, J. Coney / Brain & Language 110 (2009) 95–100
ulus modality and visual field (F(1, 31) = .369, MSe = 992.0, p = .548), the results did not indicate that the RH responded faster following auditory presentation than following visual presentation. There were no significant interactions between modality and any of the experimental factors (highest F = 1.720, for the interaction between modality and sentence type) therefore data was collapsed across modality prior to performing subsequent analyses. Analysis of the combined data revealed a significant main effect of visual field. The 19 ms difference between response times following LVF/RH presentation (M = 432 ms) and RVF/LH presentation (M = 413 ms) implied that the LH responded, overall, significantly faster than the RH (F(1, 31) = 41.5, MSe = 1152.8, p < .001). This finding is standard in the literature utilizing lateralized lexical decision tasks (e.g., Chiarello, 1988), and permits the assumption that participants were able to fixate properly during trials and that targets were registered in the designated visual fields. The results also produced a significant main effect of sentence length (F(1, 31) = 117.4, MSe = 1418.8, p < .001). Response times to targets primed by short sentences (M = 441 ms) were, on average, 36 ms slower than for long sentences (M = 405 ms). This result clearly establishes the effectiveness of the sentences in producing differentiated levels of context. Sentence length also interacted significantly with visual field (F(1, 31) = 5.1, MSe = 625.5, p = .031), whereby the difference in RT between short and long sentences was significantly greater for the LVF/RH (41 ms) than for the RVF/LH (31 ms). This suggests that RH responses were enhanced to a greater degree than LH responses by the presence of extra context. In addition, the main effect of sentence type was found to be significant (F(1, 31) = 222.6, MSe = 604.9, p < .001). Participants responded 33 ms faster to targets following normal sentence primes (M = 406 ms) than to targets following scrambled sentence primes (M = 439 ms). Furthermore, an interaction effect of sentence length and sentence type was significant (F(1, 31) = 23.3, MSe = 450.6, p < .001); additional analyses (paired-samples t-tests) indicated that this effect occurred because the difference in response times between normal-long and normal-short sentences (45 ms) was significantly greater (t(31) = 4.8, p < .001) than the difference in response times between scrambled-long and scrambled-short sentences (27 ms) (see Fig. 1). The interaction between visual field, sentence length, and sentence type was non-significant (F(1, 31) = 3.1, MSe = 359.4 p = .088). Nonetheless, in view of the fact that the interaction ap-
Fig. 1. Mean RT (ms) for sentence length, as a function of visual field and sentence type. Error bars represent ±1 standard error from the mean.
97
proached significance, it was thought appropriate to explore it further with individual comparisons. While the difference in RTs for long compared to short scrambled sentences was not significantly different for the LVF/RH compared to RVF/LH (t(31) = 0.696, p = .492), the difference in RT between long and short normal sentences was significantly greater for the LVF/RH (53 ms) than for the RVF/LH (37 ms) (t(31) = 3.1, p < .05). Similarly, an interaction between sentence length and sentence type was significant for the LVF/RH (F(1, 31) = 40.546, MSe = 113.6, p < .001) and approached significance for the RVF/LH (F(1, 31) = 40.546, MSe = 113.6, p = .051), further suggesting that increased sentence length and thus contextual information had a greater impact on both hemispheres when the sentences were normal than when they were scrambled. Analyses of error rates are not reported because these were low (M = 1.7%), closely approaching zero in several conditions, and did not reveal any effects that bear on the main findings of this study.
3. Discussion The hypothesis that the difference in response times for short and long sentences (i.e., low and high context, respectively) would be greater for the right than for the left hemisphere was supported. We obtained an unusual result for a behavioral experiment by showing that the LVF/RH benefited at least as much as the RVF/ LH from increased context. This suggests that the approach utilized in this study enhanced RH processing of both target and sentential information and, as such, produced results that were consistent with observations about context made in neuropsychological, electrophysiological, and neuroimaging approaches. Our results therefore support the view that the RH is sensitive to message-level information in sentences. Faust (1998) points out that an increase in priming due to the additional context provided by longer sentences is not necessarily indicative of processing at the message-level. Longer sentences might simply contain additional associated words and produce an increase in priming, not from the integrated message-level aspect of the sentence, but from the cumulative effect of several associative primes. The stimuli in the present study, however, were designed to minimize associations between words in the sentence prime and the target. It is thus unlikely that the increase in priming effects from short to long sentences resulted from intra-lexical association effects. More compellingly, the comparison between normal and scrambled sentences provided a means by which to determine the effects of context beyond those produced by word-level information. That is, normal sentence primes differed from the scrambled sentence primes in that they provided message-level information (i.e., through syntactic, semantic, and pragmatic analysis) in addition to word-level information (i.e., through lexical processes) (Morris, 1994). Therefore, the difference between lexical decision latencies for normal and scrambled sentences can be taken as a measure of the priming attributable to message-level processes. In the present study, increasing the length of the sentences resulted in a markedly greater increase in priming in the LVF/RH than the RVF/LH for the normal sentences, but had no effect upon visual field differences in priming for the scrambled sentences. This might suggest that the RH was not only utilizing the message-level information contained in normal sentences, but was doing so to a greater extent than the LH. There is, however, the possibility that the relatively fast reaction times observed in this study resulted in the LH approaching a performance ceiling in the long, normal sentence condition. Alternatively, the larger difference between short and long sentences for the RH might be the result of a LH advantage for processing weakly constraining sentences (Wlotko
98
B. Gouldthorp, J. Coney / Brain & Language 110 (2009) 95–100
& Federmeier, 2007), resulting in comparatively less ‘room for improvement’ by the LH. Despite the need to be cautious in assuming from the results that the RH made better use of contextual information than the LH, the results nonetheless demonstrated that the RH’s use of context changed as a result of message-level processing rather than just at the word-level. These results contrast with those of Faust et al. (1995, 1998), who found that while RTs in the RVF/LH were significantly less for normal compared to scrambled sentences, there was no difference in priming between normal and scrambled sentences in the LVF/RH. Faust et al. (1995) used their findings to support the argument that the RH is restricted to word-level processes, whereas the LH is more sensitive to information conveyed at the sentence message-level. The present study, however, provides evidence that the RH is sensitive to message-level information. Furthermore, the results of the present study are consistent with those produced by both Faust, Barlev, and Chiarello (2003) and Chiarello, Liu, and Faust (2001). In both of these studies, inhibition for incongruous targets was observed in both hemispheres, even in the absence of an associated word. This finding suggests, in agreement with the theme of the present results, that both the right and left hemispheres are sensitive to message-level information. It appears that the present study, through the simplicity and consistency of the priming sentences used, the concrete and easily imagined scenarios that they depicted, and the use of highly imageable nouns as probe words, provided a conceptual environment that the RH could readily comprehend. That is, an environment that allowed the RH to reveal an early indication of the global, holistic, synthesizing language skills that have been repeatedly demonstrated in studies of more extended discourse in brain damaged populations (e.g., Delis, Wapner, Gardner, & Moses, 1983; St. George et al., 1999). When the results of the present study are considered in conjunction with the findings of these sorts of studies, one might argue that the RH cannot easily represent precise propositions derived from sentences; instead, it is likely to be specialized for the generation of contextual representations embodied in higher-level discourse models (Beeman, 1998). The sentence primes employed in the present study were of a kind that, due to their simplicity and contextual clarity, might be expected to be relatively easily translated from a proposition to a discourse model, and therefore to be suited to RH processing. In this respect, it is noteworthy that the long form of the sentences used in the present study, although lengthened by nothing more than an extra predicate, embodied an initial clause that permitted the creation of a mental model of the situation that ‘‘set the scene” for the terminating clause. For example, one sentence used in the present study was ‘‘The man read the MENU” in its short form. Its corresponding longer form, however, was ‘‘The man sat at the restaurant and read the MENU”. Unsurprisingly, the extra predicate contained in the clause ‘‘[The man] sat at the restaurant” resulted in significantly faster responses to the target MENU, since a clear context is immediately created by the opening clause that might be expected to activate such related real-world entities as menus. What is interesting in the present results is that this contextual activation apparently occurs to a significantly greater extent in the RH than the LH and it is possible that this might occur due to the creation or utilization in the RH of a superior mental model of the situation depicted in the sentence. In summary, this study demonstrates that behavioral research can reveal evidence of a level of RH sentence comprehension that is more consistent with the results of other methodologies provided careful attention is paid to the simplicity and consistency of sentence structure, to the depiction of concrete and easily imagined scenarios, and to ensuring that sentence comprehension is probed with words that are known to be readily processed by the RH.
4. Method 4.1. Participants Thirty-two undergraduate university students participated in this study were female and 11 were male, with a mean age of 26.9 years, (SD = 9.9 years). All of the participants possessed English as a first language, had normal hearing, normal or correctedto-normal vision and were right-handed (Bryden, 1982). 4.2. Design and stimulus materials A 2 2 2 2 repeated-measures design was used, manipulating the independent variables of visual field (left or right), sentence type (normal or scrambled), sentence length (short or long), and stimulus modality (visual or auditory). The dependent variables were RT and error rate, with RT being the primary experimental measure. 4.3. Visual and auditory sentence fragments The primes consisted of 80 ‘‘short” sentence fragments (5–7 words) and 80 ‘‘long” sentence fragments (9–13 words), where the sentence-final word (the target) was a four-letter noun derived from the MRC Database (1987; Coltheart, 1981) that also had reasonably high imageability (M = 584) and concreteness ratings (M = 577). The short sentences were structured in accordance with the very simple and canonical grammatical structure of subject– verb–object (e.g., ‘‘The boy used the SOAP”), while the long sentences were constructed by inserting an extra predicate into the corresponding short sentence (e.g., ‘‘The boy washed his hands and used the SOAP”). The structural complexity of these sentences was thus greater than the short sentences, but the possible confound of length was outweighed by the benefit of keeping the structure constant within each condition. The extra predicate served to increase the amount of contextual information in the sentence. Prior to determination of the final stimulus set, the Cloze procedure (Taylor, 1953) was utilized in order to quantify the level of constraint of each sentence by determining the probability of the target word being used as a completion. The short sentence fragments were selected according to a probability criterion of between 0.1 and 0.5 (M = 0.3) and above 0.5 for ‘‘long” sentences (M = 0.8). The nouns in each sentence fragment were assessed for association with the sentence-final word (e.g., ‘‘hands” and ‘‘soap”) using the University of South Florida association norms database (Nelson, McEvoy, & Schreiber, 2004) and sentence fragments containing a word with a forward associative strength rating of 0.5 or above were excluded. The short sentence fragments contained individual words with a mean forward associative strength rating of 0.02 (SD = 0.07) and the long sentence fragments had a mean rating of 0.03 (SD = 0.09). Forward associative strength did not differ significantly for short compared to long sentences (t(79) = 1.727, p = .088). See Appendix A for examples of sentence fragment pairs and targets. The stimulus set also included an additional 80 sentences (40 short and 40 long) with the sentence-final word being a non-word. These sentences were not constructed in pairs (i.e., the ‘‘long” sentence fragments had no relation to the ‘‘short” sentence fragments). The sentence fragments were scrambled (i.e., their word order was randomly changed) to produce a control set of 240 scrambled sentence fragments to complement the set of 240 normal sentence fragments. The purpose of including scrambled sentence fragments in the design was to provide a means of parsing out simple word-level priming effects from the overall message-level priming effects of normal sentence fragments.
99
B. Gouldthorp, J. Coney / Brain & Language 110 (2009) 95–100
Four separate stimulus sets comprising 320 sentences each were constructed so that, firstly, each participant would only be presented with either the short or the long version of each sentence, and, secondly, would have equal numbers of target words and non-words for each condition presented to each visual field. This resulted in 20 observations per condition per participant. Participants saw a target word or non-word twice in a session (i.e., preceded by the normal and the scrambled sentence fragment) and to the same visual field. In order to avoid confounding conditions with repetition priming effects, a computer program was written to counterbalance the order of the second presentation of each target word or non-word across participants and to maximize the number of intervening words between the first and second presentation of the target. Each stimulus set contained equal numbers of normal and scrambled sentences, and first and second presentations of normal and scrambled sentences was balanced across participants. The 480 sentence fragments were also used to produce an additional four balanced sets of 320 sentences each for use in the auditory condition. Each sentence fragment (i.e., without the sentence-final word) was recorded by a female native English speaker. 4.4. Apparatus All programs associated with running both the visual and auditory sessions were run on an Intel Pentium 4 processor with a Windows 98 Se operating system and 256 MB RAM. The monitor on which stimuli were displayed was a Mitsubishi Diamond Plus 71 with a screen area of 800 600 pixels, 32-bit color and a refresh rate of 75 Hz. Audio sentences were presented to both ears simultaneously through Sony Stereo MDR-CD 780 headphones. A 2-button micro-switch response box was used to record the participants’ responses. Head position was controlled using an adjustable chinrest positioned 60 cm from the monitor. Ear defenders were used in the visual session to ensure exclusion of any extraneous noise that might occur during testing. 4.5. Procedure The order of presentation of visual and auditory sessions was counterbalanced across participants. Participants attended two 1 h sessions approximately one week apart (aimed at minimizing practice effects across sessions) and were tested individually. Participants were presented with one of the four randomized, balanced stimulus sets of 320 sentences divided into seven blocks. Prior to each presentation, a blue fixation cross would appear on which participants were instructed to focus their gaze. A sentence fragment was then presented either: centrally on the computer screen in black letters in Courier font approximately 1 cm high; or centrally via headphones to both ears (see Love & Swinney, 1996 for cross-modal lexical priming paradigm methodology). The sentences were presented at an individually-thresholded speed (calculated by a separate set of materials prior to commencement of the first session) in order for the presentation rate to replicate that of natural language processing. A black fixation cross would then appear centrally, on which participants were again instructed to focus their gaze. Shortly following the presentation of this fixation cross, and with an inter-stimulus interval of 300 ms, the target word or non-word was flashed to either the left or right side of the screen, at an eccentricity of 2.1° of visual angle. The target word or non-word was presented in black uppercase letters, using the Verdana font, for 150 ms. Participants were instructed to depress both micro-switches simultaneously with the index finger of each hand if they judged the target to be a word, and to refrain from responding if it was a non-word. If participants responded to a non-word or failed to respond within 1500 ms to a
word, an error message (consisting of the word ‘‘ERROR”) was presented briefly on the screen. The inter-trial interval was 4 s. Ethical declaration This experiment has been approved by the Murdoch University Human Research Ethics Committee and has therefore been performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki. All participants gave their informed consent prior to their inclusion in the study, and any details that might disclose the identity of the individuals under study have been omitted. Appendix A. Examples of stimuli. Short sentence fragment
Long sentence fragment
Target
The girl got in the The cat caught the They drove out of the The cook battered the The thief was sent to The man read the
The girl ran the water then got in the The cat watched it land then caught the They drove into the country and out of the The cook cleaned off the scales then battered the The thief was found guilty and was sent to The man sat at the restaurant and read the The boy played and built castles in the
BATH BIRD
The girl lay in the sun and burnt her The boy washed his hands and used the The cat hid in the bushes then climbed up the The waiter uncorked the bottle and poured the The tiger found a hole in the bars and escaped from its
SKIN SOAP
The boy played in the The girl burnt her The boy used the The cat climbed up the The waiter poured the The tiger escaped from its
CITY FISH JAIL MENU SAND
TREE WINE CAGE
References Barnett, V., & Lewis, T. (1994). Outliers in statistical data (3rd ed.). New York: Wiley & Sons. Beeman, M. (1998). Course semantic coding and discourse comprehension. In M. Beeman & C. Chiarello (Eds.), Right hemisphere language comprehension: Perspectives from cognitive neuroscience (pp. 255–284). New Jersey: Lawrence Erlbaum Associates. Bryden, M. P. (1982). Laterality: Functional asymmetry in the intact brain. New York: Academic Press. Chiarello, C. (1988). Lateralization of lexical processes in the normal brain: A review of visual half-field research. In H. A. Whitaker (Ed.), Contemporary reviews in neuropsychology (pp. 36–76). New York: Springer-Verlag. Chiarello, C., Faust, M., & Liu, S. (1999). Cerebral asymmetries in sentence priming and the influence of semantic anomaly. Brain and Cognition, 40(1), 75–79. Chiarello, C., Liu, S., & Faust, M. (2001). Bihemispheric sensitivity to sentence anomaly. Neuropsychologia, 39, 1451–1463. Chiarello, C., Shears, C., Liu, S., & Kacinik, N. (2005). Influence of word class proportion on cerebral asymmetries for high-and low-imagery words. Brain and Cognition, 57, 35–38. Coltheart, M. (1981). The MRC psycholinguistic database. Quarterly Journal of Experimental Psychology, 33A, 497–505. Coulson, S., Federmeier, K. D., Van Petten, C., & Kutas, M. (2005). Right hemisphere sensitivity to word-and sentence-level context: Evidence from event-related brain potentials. Journal of Experimental Psychology: Learning Memory an Cognition, 31(1), 129–147. Crinion, J., & Price, C. (2005). Right anterior superior temporal activation predicts auditory sentence comprehension following aphasic stroke. Brain, 128, 2858–2871.
100
B. Gouldthorp, J. Coney / Brain & Language 110 (2009) 95–100
Delis, D. C., Wapner, W., Gardner, H., & Moses, J. A. (1983). The contribution of the right hemisphere to the organization of paragraphs. Cortex, 19(1), 43–50. Faust, M. (1998). Obtaining evidence of language comprehension from sentence priming. In M. Beeman & C. Chiarello (Eds.), Right hemisphere language comprehension: Perspectives from cognitive neuroscience (pp. 161–185). New Jersey: Lawrence Erlbaum Associates. Faust, M., & Babkoff, H. (1997). Script as a priming stimulus for lexical decision with visual hemifield stimulation. Brain and Language, 57, 423–437. Faust, M., Babkoff, H., & Kravetz, S. (1993). Hemisphericity and top-down processing of language. Brain and Language, 44, 1–18. Faust, M., Babkoff, H., & Kravetz, S. (1995). Linguistic processes in the two cerebral hemispheres: Implications for modularity vs. interactionism. Journal of Clinical and Experimental Neuropsychology, 17, 171–192. Faust, M., Barak, O., & Chiarello, C. (2006). The effects of multiple script priming on word recognition by the two cerebral hemispheres: Implications for discourse processing. Brain and Language, 99, 247–257. Faust, M., Bar-lev, A., & Chiarello, C. (2003). Sentence priming effects in the two cerebral hemispheres: Influences of lexical relatedness, word order, and sentence anomaly. Neuropsychologia, 41(4), 480–492. Faust, M., & Chiarello, C. (1998). Constraints on sentence priming in the cerebral hemispheres: Effects of intervening words in sentences and lists. Brain and Language, 63, 219–236. Faust, M., & Kravetz, S. (1998). Levels of sentence constraint and lexical decision in the two hemispheres. Brain and Language, 62, 149–162. Federmeier, K. D. (2007). Thinking ahead: The role and roots of prediction in language comprehension. Psychophysiology, 44, 491–505. Federmeier, K. D., Mai, H., & Kutas, M. (2005). Both sides get the point: Hemisphere sensitivities to sentential constraint. Memory & Cognition, 33(5), 871–886. Gazzaniga, M. S., & Sperry, R. W. (1967). Language after section of the cerebral commissures. Brain, 90, 131–148.
Gouldthorp, B., & Coney, J. (2009). Message-level processing of contextual information in the right cerebral hemisphere. Neuropsychologia, 47(2), 473–480. Joanette, Y., Goulet, P., & Hannequin, D. (1990). Right hemisphere and verbal communication. New York: Springer-Verlag. Love, T., & Swinney, D. (1996). Coreference processing and levels of analysis in object-relative constructions: Demonstration of antecedent reactivation with the cross-modal priming paradigm. Journal of Psycholinguistic Research, 25(1), 5–24. Morris, R. K. (1994). Lexical and message-level context effects on fixation times in reading. Journal of Experimental Psychology: Learning, Memory and Cognition, 20, 92–103. MRC Psycholinguistic Database (1987). Machine usable dictionary (version 2.00) [Computer Software].. Retrieved 15.01.2008. Nelson, D. L., McEvoy, C. L., & Schreiber, T. A. (2004). The University of South Florida free association, rhyme, and word fragment norms behavior research methods, instruments and computers. Special Issue: Web-based Archive of Norms, Stimuli and Data: Part 1, 36(1), 402–407. Sereno, J. A. (1999). Hemispheric differences in grammatical class. Brain and Language, 70, 13–28. St. George, M., Kutas, M., Martinez, A., & Sereno, M. (1999). Semantic integration in reading: Engagement of the right hemisphere during discourse processing. Brain, 122, 1317–1325. Taylor, W. L. (1953). ‘‘Cloze” procedure: A new tool for measuring readability. Journalism Quarterly, 30, 415. Wlotko, E., & Federmeier, K. (2007). Finding the right word: Hemispheric asymmetries in the use of sentence context information. Neuropsychologia, 45, 3001–3014. Zaidel, E. (1976). Auditory vocabulary of the right hemisphere following brain bisection or hemidecortication. Cortex, 12(3), 191–211.
Contents lists available at ScienceDirect
Brain & Language journal homepage: www.elsevier.com/locate/b&l
Short Communication
The sensitivity of the right hemisphere to contextual information in sentences Bethanie Gouldthorp *, Jeffrey Coney School of Psychology, Murdoch University, South Street, Murdoch, Western Australia 6150, Australia
a r t i c l e
i n f o
Article history: Accepted 16 May 2009 Available online 13 June 2009 Keywords: Sentence comprehension Hemisphere Visual field Priming Context Sentence length
a b s t r a c t One explanation for the inconsistencies in research examining the sentence comprehension abilities of the right hemisphere (RH) is the presence of confounding variables that have generally served to disadvantage the processing capacities of the RH. As such, the present study aimed to investigate hemispheric differences in the use of message-level sentential information by removing some of the factors known to be inherently disadvantageous for RH comprehension. Thirty-two right-handed undergraduate university students participated in a computer-based lexical decision task where RT and error rates were recorded. The sensitivity of each hemisphere to the message-level contextual information contained in short versus long sentences was compared, as well as the effect of stimulus modality (visual compared to auditory). The results showed that the RH benefited from increased levels of context to at least the same extent as the LH and that, more importantly, this could not be explained by word-level processes alone. This finding, unusual in behavioral research on normal individuals but consistent with neuropsychological, electrophysiological, and neuroimaging approaches, suggests that the RH plays an important role in sentence comprehension, at least in relation to sentences that conform to a relatively simple structure. Ó 2009 Elsevier Inc. All rights reserved.
1. Introduction Although the dominance of the left cerebral hemisphere (LH) in language processing and production has been repeatedly demonstrated over the past few decades, studies have also produced evidence for right hemisphere (RH) involvement in certain areas of language comprehension (e.g., Coulson, Federmeier, Van Petten, & Kutas, 2005; St. George, Kutas, Martinez, & Sereno, 1999). Many differences in the language processing of the left and right hemispheres have been revealed through studies utilizing the differing levels of constraint produced by sentence primes (Faust & Kravetz, 1998). For example, the RH appears to be relatively less affected by linguistic constraint, whereas the LH appears able to take advantage of it (Faust, Babkoff, & Kravetz, 1993). These differences are thought to be the result of the hemispheres using different linguistic mechanisms for sentence comprehension (Faust & Babkoff, 1997). Morris (1994) suggested that the LH interprets contextual information by using ‘‘message-level” mechanisms (i.e., by combining syntactic, semantic, and pragmatic information in order to build a conceptual representation of the meaning of the sentence), whereas the RH uses ‘‘word-level” mechanisms (i.e., by using lexical information and basic relations between individual words in sentences). This proposal is supported by studies that have found that it is only in the LH that normal sentences are much more effective primes than the same words presented in a scrambled list (Faust, Babkoff, & Kra* Corresponding author. Fax: +61 08 9360 6492. E-mail address: [email protected] (B. Gouldthorp). 0093-934X/$ - see front matter Ó 2009 Elsevier Inc. All rights reserved. doi:10.1016/j.bandl.2009.05.003
vetz, 1995; Faust & Chiarello, 1998). Furthermore, increasing the amount of context from one to three to six words produced larger priming effects when presentation was to the RVF/LH but not the LVF/RH (Faust et al., 1993). Additionally, lexical decision latencies were increased for implausible compared to plausible sentence-final words when presented to the RVF/LH but not the LVF/RH. Faust and Kravetz (1998) take these findings as evidence that the LH is more sensitive to linguistic information conveyed at the sentence message-level than the RH. Furthermore, they argue that while the LH is able to suppress contextually inappropriate words or meanings, the RH’s comprehension ability is compromised as a result of the ‘flooding’ of irrelevant information due to its sustained activation of multiple meanings and associations. Nonetheless, the results of these studies supporting a word-level only processing ability of the RH are inconsistent with the findings in relation to patients with RH damage; for example, difficulties with comprehension of metaphors, inferences and coherence maintenance (see Faust, Barak, & Chiarello, 2006 for a review). In addition, ERPs observed in neurologically normal individuals suggest that both hemispheres are primarily driven by message-level congruity during sentence reading, although they utilize this information in different ways (Coulson et al., 2005; Federmeier, Mai, & Kutas, 2005). Similarly, St. George et al. (1999) provided evidence that the way in which the RH is able to integrate information into a coherent context means it may be particularly important in processing at the discourse level. The findings of these sorts of studies suggest that the RH plays an important role in comprehending and integrating language and
96
B. Gouldthorp, J. Coney / Brain & Language 110 (2009) 95–100
in the application of contextual information. Furthermore, several behavioral studies of normal individuals have recently demonstrated a RH sensitivity to context/constraint under certain circumstances (see Gouldthorp & Coney, 2009, for a review). It is possible that the inconsistent nature of behavioral studies of the capabilities and mechanisms of the RH in language comprehension when compared to observations made by the other methodologies (i.e., neuropsychological, neuroimaging and electrophysiological studies) may be the result of several methodological problems. For example, the variation of sentence structure as a means of manipulating different levels of constraint for the same target word (e.g., Faust & Kravetz, 1998) often produces quite dramatic variations in grammatical construction. It is not common in the literature, however, for researchers to carefully match sentence structures across each level of constraint, presumably because of the very considerable difficulties that such matching entails. As a result, there remains the possibility that although these studies purport to be investigating the effect of varying levels of contextual information on hemispheric comprehension, the variation in sentence structure may confound any attempt to properly compare the performance of the two hemispheres. Nonetheless, studies that have attempted to manipulate context by increasing the number of words (e.g., Faust et al., 1993) have still failed to produce results that are consistent with the findings of neuropsychological, electrophysiological, and neuroimaging studies. Other uncontrolled variables may also contribute to the inconsistencies. In several studies, determining the message-level content of the sentence was contingent on appreciating the linguistic function of the verb or verb phrase (e.g., Chiarello, Faust, & Liu, 1999; Faust et al., 1995). There is a substantial body of research that suggests the RH has difficulty in processing verbs (e.g., Gazzaniga & Sperry, 1967; Sereno, 1999). As such, the dependence of previous studies on the critical role of the verb in simple sentences could have partially masked any evidence of RH sentence comprehension. Furthermore, despite some evidence that the noun/verb distinction is better accounted for by differences in imageability than grammatical class itself (e.g., Chiarello, Shears, Liu, & Kacinik, 2005), there continues to be a general failure in the literature to control for imageability in lieu of (or in addition to) grammatical class. One further factor may be of importance in relation to studies of sentence comprehension in the RH; the modality of stimulus presentation for behavioral investigations into hemispheric sentence comprehension has been traditionally visual. There is, however, substantial evidence to suggest that the RH processes language better when received in auditory form than in visual (e.g., Crinion & Price, 2005; Zaidel, 1976). Crossmodal lexical priming paradigms (CMLP) also potentially circumvent some of the difficulties produced by entirely visual lexical decision tasks, such as ensuring central fixation. While relatively few studies have addressed auditory sentence comprehension in particular, none appear to have compared the specific effect that stimulus modality has on hemispheric differences in language capacities and, as such, these potential benefits remain speculative. Based on these observations, the assertion that the RH relies purely on word-level mechanisms needs to be tested using stimuli that are less likely to mask the possibility that the RH is sensitive to message-level mechanisms, particularly given the contradictory evidence from alternative methodologies (e.g., Federmeier, 2007). For example, target words should comprise only short, concrete, highly imageable nouns in order maximize the likelihood that they can be processed appropriately by the RH. Additionally, sentence stimuli should adhere to a consistent grammatical structure across conditions. As observed earlier, this creates particular difficulties for studies attempting to compare the effects of varying levels of contextual information (e.g., Faust & Kravetz, 1998). If sentence
length remains the same, it becomes difficult to retain the same structure while including additional context. On the other hand, sentence length may be increased to accommodate the additional contextual information, provided the longer sentences conform to a broadly similar structure. For example, context could be manipulated by the consistent addition of a single extra predicate to the canonical subject–verb–object structure. While not all factors that might disadvantage the RH can be easily addressed (e.g., the lexical decision task is in itself a concern), the inconsistency of results suggests that alternative approaches within this paradigm must be explored in order to bridge this gap. All of these proposed changes to the usual normal-behavioral methodology were implemented in the present study, of which the main aim was to compare the sensitivity of each hemisphere to the extra contextual information provided in long sentences relative to short sentences that differed structurally in a minor and consistent fashion. Scrambled variants of these sentences were also included as a control measure of word-level processing (i.e., removing the syntactic information). Based on the evidence produced by neuropsychological, electrophysiological, and neuroimaging methodologies indicating that the RH is not only sensitive to message-level information (e.g., Federmeier et al., 2005) but potentially makes better use of contextual information than the LH (e.g., Joanette, Goulet, & Hannequin, 1990), it was hypothesized that the difference in response times for short and long sentences (i.e., low and high context, respectively) would be greater for the right than for the left hemisphere. In addition, it was hypothesized that the RH would be more sensitive to sentential information when presented in an auditory, rather than visual, format. 2. Results RT data for correct responses were initially screened utilizing a deletion criterion of ±3.0 standard deviations from each individual’s mean response times for each condition. This resulted in only 0.44% of the total observations being excluded. An additional screening procedure was applied to the data in order to compare each participant’s mean RTs for each condition with the overall sample mean for the corresponding condition. Data were screened using the Winsorization method (Barnett & Lewis, 1994) at two standard deviations from the mean. This resulted in an additional 4.49% of mean responses being adjusted. An overall 2 2 2 2 repeated-measures analysis of variance (i.e., stimulus modality, visual field, sentence length and sentence type) was performed on the screened RT data (summarized in Table 1). Initial analysis of the results did not reveal a significant main effect of stimulus modality (F(1, 31) = .290, MSe = 7681.2, p = .594). Furthermore, as there was no significant interaction between stim-
Table 1 Mean RT (ms) as a function of target visual field, stimulus modality, sentence length, and sentence type. Standard deviations (ms) appear in parentheses. RVF/LH Normal sentences Visual sentence primes Short sentences 464 (75) Long sentences 438 (76) Auditory sentence primes Short sentences 464 (70) Long sentences 432 (81)
LVF/RH Scrambled sentences
Normal sentences
Scrambled sentences
422 (78)
444 (61)
447 (80)
385 (70)
420 (78)
396 (70)
410 (69)
436 (67)
435 (73)
373 (72)
411 (78)
381 (70)
B. Gouldthorp, J. Coney / Brain & Language 110 (2009) 95–100
ulus modality and visual field (F(1, 31) = .369, MSe = 992.0, p = .548), the results did not indicate that the RH responded faster following auditory presentation than following visual presentation. There were no significant interactions between modality and any of the experimental factors (highest F = 1.720, for the interaction between modality and sentence type) therefore data was collapsed across modality prior to performing subsequent analyses. Analysis of the combined data revealed a significant main effect of visual field. The 19 ms difference between response times following LVF/RH presentation (M = 432 ms) and RVF/LH presentation (M = 413 ms) implied that the LH responded, overall, significantly faster than the RH (F(1, 31) = 41.5, MSe = 1152.8, p < .001). This finding is standard in the literature utilizing lateralized lexical decision tasks (e.g., Chiarello, 1988), and permits the assumption that participants were able to fixate properly during trials and that targets were registered in the designated visual fields. The results also produced a significant main effect of sentence length (F(1, 31) = 117.4, MSe = 1418.8, p < .001). Response times to targets primed by short sentences (M = 441 ms) were, on average, 36 ms slower than for long sentences (M = 405 ms). This result clearly establishes the effectiveness of the sentences in producing differentiated levels of context. Sentence length also interacted significantly with visual field (F(1, 31) = 5.1, MSe = 625.5, p = .031), whereby the difference in RT between short and long sentences was significantly greater for the LVF/RH (41 ms) than for the RVF/LH (31 ms). This suggests that RH responses were enhanced to a greater degree than LH responses by the presence of extra context. In addition, the main effect of sentence type was found to be significant (F(1, 31) = 222.6, MSe = 604.9, p < .001). Participants responded 33 ms faster to targets following normal sentence primes (M = 406 ms) than to targets following scrambled sentence primes (M = 439 ms). Furthermore, an interaction effect of sentence length and sentence type was significant (F(1, 31) = 23.3, MSe = 450.6, p < .001); additional analyses (paired-samples t-tests) indicated that this effect occurred because the difference in response times between normal-long and normal-short sentences (45 ms) was significantly greater (t(31) = 4.8, p < .001) than the difference in response times between scrambled-long and scrambled-short sentences (27 ms) (see Fig. 1). The interaction between visual field, sentence length, and sentence type was non-significant (F(1, 31) = 3.1, MSe = 359.4 p = .088). Nonetheless, in view of the fact that the interaction ap-
Fig. 1. Mean RT (ms) for sentence length, as a function of visual field and sentence type. Error bars represent ±1 standard error from the mean.
97
proached significance, it was thought appropriate to explore it further with individual comparisons. While the difference in RTs for long compared to short scrambled sentences was not significantly different for the LVF/RH compared to RVF/LH (t(31) = 0.696, p = .492), the difference in RT between long and short normal sentences was significantly greater for the LVF/RH (53 ms) than for the RVF/LH (37 ms) (t(31) = 3.1, p < .05). Similarly, an interaction between sentence length and sentence type was significant for the LVF/RH (F(1, 31) = 40.546, MSe = 113.6, p < .001) and approached significance for the RVF/LH (F(1, 31) = 40.546, MSe = 113.6, p = .051), further suggesting that increased sentence length and thus contextual information had a greater impact on both hemispheres when the sentences were normal than when they were scrambled. Analyses of error rates are not reported because these were low (M = 1.7%), closely approaching zero in several conditions, and did not reveal any effects that bear on the main findings of this study.
3. Discussion The hypothesis that the difference in response times for short and long sentences (i.e., low and high context, respectively) would be greater for the right than for the left hemisphere was supported. We obtained an unusual result for a behavioral experiment by showing that the LVF/RH benefited at least as much as the RVF/ LH from increased context. This suggests that the approach utilized in this study enhanced RH processing of both target and sentential information and, as such, produced results that were consistent with observations about context made in neuropsychological, electrophysiological, and neuroimaging approaches. Our results therefore support the view that the RH is sensitive to message-level information in sentences. Faust (1998) points out that an increase in priming due to the additional context provided by longer sentences is not necessarily indicative of processing at the message-level. Longer sentences might simply contain additional associated words and produce an increase in priming, not from the integrated message-level aspect of the sentence, but from the cumulative effect of several associative primes. The stimuli in the present study, however, were designed to minimize associations between words in the sentence prime and the target. It is thus unlikely that the increase in priming effects from short to long sentences resulted from intra-lexical association effects. More compellingly, the comparison between normal and scrambled sentences provided a means by which to determine the effects of context beyond those produced by word-level information. That is, normal sentence primes differed from the scrambled sentence primes in that they provided message-level information (i.e., through syntactic, semantic, and pragmatic analysis) in addition to word-level information (i.e., through lexical processes) (Morris, 1994). Therefore, the difference between lexical decision latencies for normal and scrambled sentences can be taken as a measure of the priming attributable to message-level processes. In the present study, increasing the length of the sentences resulted in a markedly greater increase in priming in the LVF/RH than the RVF/LH for the normal sentences, but had no effect upon visual field differences in priming for the scrambled sentences. This might suggest that the RH was not only utilizing the message-level information contained in normal sentences, but was doing so to a greater extent than the LH. There is, however, the possibility that the relatively fast reaction times observed in this study resulted in the LH approaching a performance ceiling in the long, normal sentence condition. Alternatively, the larger difference between short and long sentences for the RH might be the result of a LH advantage for processing weakly constraining sentences (Wlotko
98
B. Gouldthorp, J. Coney / Brain & Language 110 (2009) 95–100
& Federmeier, 2007), resulting in comparatively less ‘room for improvement’ by the LH. Despite the need to be cautious in assuming from the results that the RH made better use of contextual information than the LH, the results nonetheless demonstrated that the RH’s use of context changed as a result of message-level processing rather than just at the word-level. These results contrast with those of Faust et al. (1995, 1998), who found that while RTs in the RVF/LH were significantly less for normal compared to scrambled sentences, there was no difference in priming between normal and scrambled sentences in the LVF/RH. Faust et al. (1995) used their findings to support the argument that the RH is restricted to word-level processes, whereas the LH is more sensitive to information conveyed at the sentence message-level. The present study, however, provides evidence that the RH is sensitive to message-level information. Furthermore, the results of the present study are consistent with those produced by both Faust, Barlev, and Chiarello (2003) and Chiarello, Liu, and Faust (2001). In both of these studies, inhibition for incongruous targets was observed in both hemispheres, even in the absence of an associated word. This finding suggests, in agreement with the theme of the present results, that both the right and left hemispheres are sensitive to message-level information. It appears that the present study, through the simplicity and consistency of the priming sentences used, the concrete and easily imagined scenarios that they depicted, and the use of highly imageable nouns as probe words, provided a conceptual environment that the RH could readily comprehend. That is, an environment that allowed the RH to reveal an early indication of the global, holistic, synthesizing language skills that have been repeatedly demonstrated in studies of more extended discourse in brain damaged populations (e.g., Delis, Wapner, Gardner, & Moses, 1983; St. George et al., 1999). When the results of the present study are considered in conjunction with the findings of these sorts of studies, one might argue that the RH cannot easily represent precise propositions derived from sentences; instead, it is likely to be specialized for the generation of contextual representations embodied in higher-level discourse models (Beeman, 1998). The sentence primes employed in the present study were of a kind that, due to their simplicity and contextual clarity, might be expected to be relatively easily translated from a proposition to a discourse model, and therefore to be suited to RH processing. In this respect, it is noteworthy that the long form of the sentences used in the present study, although lengthened by nothing more than an extra predicate, embodied an initial clause that permitted the creation of a mental model of the situation that ‘‘set the scene” for the terminating clause. For example, one sentence used in the present study was ‘‘The man read the MENU” in its short form. Its corresponding longer form, however, was ‘‘The man sat at the restaurant and read the MENU”. Unsurprisingly, the extra predicate contained in the clause ‘‘[The man] sat at the restaurant” resulted in significantly faster responses to the target MENU, since a clear context is immediately created by the opening clause that might be expected to activate such related real-world entities as menus. What is interesting in the present results is that this contextual activation apparently occurs to a significantly greater extent in the RH than the LH and it is possible that this might occur due to the creation or utilization in the RH of a superior mental model of the situation depicted in the sentence. In summary, this study demonstrates that behavioral research can reveal evidence of a level of RH sentence comprehension that is more consistent with the results of other methodologies provided careful attention is paid to the simplicity and consistency of sentence structure, to the depiction of concrete and easily imagined scenarios, and to ensuring that sentence comprehension is probed with words that are known to be readily processed by the RH.
4. Method 4.1. Participants Thirty-two undergraduate university students participated in this study were female and 11 were male, with a mean age of 26.9 years, (SD = 9.9 years). All of the participants possessed English as a first language, had normal hearing, normal or correctedto-normal vision and were right-handed (Bryden, 1982). 4.2. Design and stimulus materials A 2 2 2 2 repeated-measures design was used, manipulating the independent variables of visual field (left or right), sentence type (normal or scrambled), sentence length (short or long), and stimulus modality (visual or auditory). The dependent variables were RT and error rate, with RT being the primary experimental measure. 4.3. Visual and auditory sentence fragments The primes consisted of 80 ‘‘short” sentence fragments (5–7 words) and 80 ‘‘long” sentence fragments (9–13 words), where the sentence-final word (the target) was a four-letter noun derived from the MRC Database (1987; Coltheart, 1981) that also had reasonably high imageability (M = 584) and concreteness ratings (M = 577). The short sentences were structured in accordance with the very simple and canonical grammatical structure of subject– verb–object (e.g., ‘‘The boy used the SOAP”), while the long sentences were constructed by inserting an extra predicate into the corresponding short sentence (e.g., ‘‘The boy washed his hands and used the SOAP”). The structural complexity of these sentences was thus greater than the short sentences, but the possible confound of length was outweighed by the benefit of keeping the structure constant within each condition. The extra predicate served to increase the amount of contextual information in the sentence. Prior to determination of the final stimulus set, the Cloze procedure (Taylor, 1953) was utilized in order to quantify the level of constraint of each sentence by determining the probability of the target word being used as a completion. The short sentence fragments were selected according to a probability criterion of between 0.1 and 0.5 (M = 0.3) and above 0.5 for ‘‘long” sentences (M = 0.8). The nouns in each sentence fragment were assessed for association with the sentence-final word (e.g., ‘‘hands” and ‘‘soap”) using the University of South Florida association norms database (Nelson, McEvoy, & Schreiber, 2004) and sentence fragments containing a word with a forward associative strength rating of 0.5 or above were excluded. The short sentence fragments contained individual words with a mean forward associative strength rating of 0.02 (SD = 0.07) and the long sentence fragments had a mean rating of 0.03 (SD = 0.09). Forward associative strength did not differ significantly for short compared to long sentences (t(79) = 1.727, p = .088). See Appendix A for examples of sentence fragment pairs and targets. The stimulus set also included an additional 80 sentences (40 short and 40 long) with the sentence-final word being a non-word. These sentences were not constructed in pairs (i.e., the ‘‘long” sentence fragments had no relation to the ‘‘short” sentence fragments). The sentence fragments were scrambled (i.e., their word order was randomly changed) to produce a control set of 240 scrambled sentence fragments to complement the set of 240 normal sentence fragments. The purpose of including scrambled sentence fragments in the design was to provide a means of parsing out simple word-level priming effects from the overall message-level priming effects of normal sentence fragments.
99
B. Gouldthorp, J. Coney / Brain & Language 110 (2009) 95–100
Four separate stimulus sets comprising 320 sentences each were constructed so that, firstly, each participant would only be presented with either the short or the long version of each sentence, and, secondly, would have equal numbers of target words and non-words for each condition presented to each visual field. This resulted in 20 observations per condition per participant. Participants saw a target word or non-word twice in a session (i.e., preceded by the normal and the scrambled sentence fragment) and to the same visual field. In order to avoid confounding conditions with repetition priming effects, a computer program was written to counterbalance the order of the second presentation of each target word or non-word across participants and to maximize the number of intervening words between the first and second presentation of the target. Each stimulus set contained equal numbers of normal and scrambled sentences, and first and second presentations of normal and scrambled sentences was balanced across participants. The 480 sentence fragments were also used to produce an additional four balanced sets of 320 sentences each for use in the auditory condition. Each sentence fragment (i.e., without the sentence-final word) was recorded by a female native English speaker. 4.4. Apparatus All programs associated with running both the visual and auditory sessions were run on an Intel Pentium 4 processor with a Windows 98 Se operating system and 256 MB RAM. The monitor on which stimuli were displayed was a Mitsubishi Diamond Plus 71 with a screen area of 800 600 pixels, 32-bit color and a refresh rate of 75 Hz. Audio sentences were presented to both ears simultaneously through Sony Stereo MDR-CD 780 headphones. A 2-button micro-switch response box was used to record the participants’ responses. Head position was controlled using an adjustable chinrest positioned 60 cm from the monitor. Ear defenders were used in the visual session to ensure exclusion of any extraneous noise that might occur during testing. 4.5. Procedure The order of presentation of visual and auditory sessions was counterbalanced across participants. Participants attended two 1 h sessions approximately one week apart (aimed at minimizing practice effects across sessions) and were tested individually. Participants were presented with one of the four randomized, balanced stimulus sets of 320 sentences divided into seven blocks. Prior to each presentation, a blue fixation cross would appear on which participants were instructed to focus their gaze. A sentence fragment was then presented either: centrally on the computer screen in black letters in Courier font approximately 1 cm high; or centrally via headphones to both ears (see Love & Swinney, 1996 for cross-modal lexical priming paradigm methodology). The sentences were presented at an individually-thresholded speed (calculated by a separate set of materials prior to commencement of the first session) in order for the presentation rate to replicate that of natural language processing. A black fixation cross would then appear centrally, on which participants were again instructed to focus their gaze. Shortly following the presentation of this fixation cross, and with an inter-stimulus interval of 300 ms, the target word or non-word was flashed to either the left or right side of the screen, at an eccentricity of 2.1° of visual angle. The target word or non-word was presented in black uppercase letters, using the Verdana font, for 150 ms. Participants were instructed to depress both micro-switches simultaneously with the index finger of each hand if they judged the target to be a word, and to refrain from responding if it was a non-word. If participants responded to a non-word or failed to respond within 1500 ms to a
word, an error message (consisting of the word ‘‘ERROR”) was presented briefly on the screen. The inter-trial interval was 4 s. Ethical declaration This experiment has been approved by the Murdoch University Human Research Ethics Committee and has therefore been performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki. All participants gave their informed consent prior to their inclusion in the study, and any details that might disclose the identity of the individuals under study have been omitted. Appendix A. Examples of stimuli. Short sentence fragment
Long sentence fragment
Target
The girl got in the The cat caught the They drove out of the The cook battered the The thief was sent to The man read the
The girl ran the water then got in the The cat watched it land then caught the They drove into the country and out of the The cook cleaned off the scales then battered the The thief was found guilty and was sent to The man sat at the restaurant and read the The boy played and built castles in the
BATH BIRD
The girl lay in the sun and burnt her The boy washed his hands and used the The cat hid in the bushes then climbed up the The waiter uncorked the bottle and poured the The tiger found a hole in the bars and escaped from its
SKIN SOAP
The boy played in the The girl burnt her The boy used the The cat climbed up the The waiter poured the The tiger escaped from its
CITY FISH JAIL MENU SAND
TREE WINE CAGE
References Barnett, V., & Lewis, T. (1994). Outliers in statistical data (3rd ed.). New York: Wiley & Sons. Beeman, M. (1998). Course semantic coding and discourse comprehension. In M. Beeman & C. Chiarello (Eds.), Right hemisphere language comprehension: Perspectives from cognitive neuroscience (pp. 255–284). New Jersey: Lawrence Erlbaum Associates. Bryden, M. P. (1982). Laterality: Functional asymmetry in the intact brain. New York: Academic Press. Chiarello, C. (1988). Lateralization of lexical processes in the normal brain: A review of visual half-field research. In H. A. Whitaker (Ed.), Contemporary reviews in neuropsychology (pp. 36–76). New York: Springer-Verlag. Chiarello, C., Faust, M., & Liu, S. (1999). Cerebral asymmetries in sentence priming and the influence of semantic anomaly. Brain and Cognition, 40(1), 75–79. Chiarello, C., Liu, S., & Faust, M. (2001). Bihemispheric sensitivity to sentence anomaly. Neuropsychologia, 39, 1451–1463. Chiarello, C., Shears, C., Liu, S., & Kacinik, N. (2005). Influence of word class proportion on cerebral asymmetries for high-and low-imagery words. Brain and Cognition, 57, 35–38. Coltheart, M. (1981). The MRC psycholinguistic database. Quarterly Journal of Experimental Psychology, 33A, 497–505. Coulson, S., Federmeier, K. D., Van Petten, C., & Kutas, M. (2005). Right hemisphere sensitivity to word-and sentence-level context: Evidence from event-related brain potentials. Journal of Experimental Psychology: Learning Memory an Cognition, 31(1), 129–147. Crinion, J., & Price, C. (2005). Right anterior superior temporal activation predicts auditory sentence comprehension following aphasic stroke. Brain, 128, 2858–2871.
100
B. Gouldthorp, J. Coney / Brain & Language 110 (2009) 95–100
Delis, D. C., Wapner, W., Gardner, H., & Moses, J. A. (1983). The contribution of the right hemisphere to the organization of paragraphs. Cortex, 19(1), 43–50. Faust, M. (1998). Obtaining evidence of language comprehension from sentence priming. In M. Beeman & C. Chiarello (Eds.), Right hemisphere language comprehension: Perspectives from cognitive neuroscience (pp. 161–185). New Jersey: Lawrence Erlbaum Associates. Faust, M., & Babkoff, H. (1997). Script as a priming stimulus for lexical decision with visual hemifield stimulation. Brain and Language, 57, 423–437. Faust, M., Babkoff, H., & Kravetz, S. (1993). Hemisphericity and top-down processing of language. Brain and Language, 44, 1–18. Faust, M., Babkoff, H., & Kravetz, S. (1995). Linguistic processes in the two cerebral hemispheres: Implications for modularity vs. interactionism. Journal of Clinical and Experimental Neuropsychology, 17, 171–192. Faust, M., Barak, O., & Chiarello, C. (2006). The effects of multiple script priming on word recognition by the two cerebral hemispheres: Implications for discourse processing. Brain and Language, 99, 247–257. Faust, M., Bar-lev, A., & Chiarello, C. (2003). Sentence priming effects in the two cerebral hemispheres: Influences of lexical relatedness, word order, and sentence anomaly. Neuropsychologia, 41(4), 480–492. Faust, M., & Chiarello, C. (1998). Constraints on sentence priming in the cerebral hemispheres: Effects of intervening words in sentences and lists. Brain and Language, 63, 219–236. Faust, M., & Kravetz, S. (1998). Levels of sentence constraint and lexical decision in the two hemispheres. Brain and Language, 62, 149–162. Federmeier, K. D. (2007). Thinking ahead: The role and roots of prediction in language comprehension. Psychophysiology, 44, 491–505. Federmeier, K. D., Mai, H., & Kutas, M. (2005). Both sides get the point: Hemisphere sensitivities to sentential constraint. Memory & Cognition, 33(5), 871–886. Gazzaniga, M. S., & Sperry, R. W. (1967). Language after section of the cerebral commissures. Brain, 90, 131–148.
Gouldthorp, B., & Coney, J. (2009). Message-level processing of contextual information in the right cerebral hemisphere. Neuropsychologia, 47(2), 473–480. Joanette, Y., Goulet, P., & Hannequin, D. (1990). Right hemisphere and verbal communication. New York: Springer-Verlag. Love, T., & Swinney, D. (1996). Coreference processing and levels of analysis in object-relative constructions: Demonstration of antecedent reactivation with the cross-modal priming paradigm. Journal of Psycholinguistic Research, 25(1), 5–24. Morris, R. K. (1994). Lexical and message-level context effects on fixation times in reading. Journal of Experimental Psychology: Learning, Memory and Cognition, 20, 92–103. MRC Psycholinguistic Database (1987). Machine usable dictionary (version 2.00) [Computer Software].