School of Psychology

Right hemisphere language processing: Integration of contextual information in sentences

Supervisor:               Dr Jeff Coney

Rationale/hypotheses:
     A major point of contention in the literature is the type of processing mechanisms utilized by each hemisphere. It has been theorized that sentential information can be derived using “word-level” mechanisms (i.e., by using associations between individual words in the sentence) as well as from “message-level” mechanisms (i.e., by building a conceptual representation of the meaning of the sentence through the combination of syntactic, semantic and pragmatic information) (Morris, 1994).  Much of the normal-behavioural research suggests that the Right Hemisphere (RH) is only capable of utilizing word-level information; that is, that the RH is only capable of utilizing intralexical processing and therefore, in effect, treats sentences as strings of single words (Faust, Babkoff & Kravetz, 1995; Faust & Chiarello, 1998).  This proposition is, however, seemingly irreconcilable with the literature derived from the alternative approaches.  For example, this does not account for the findings of neuropsychological research, which indicates the RH is necessary for integrating the elements of an account into a coherent narrative (Delis, Wapner, Gardner & Moses, 1983; Wapner, Hamby & Gardner, 1981), comprehending metaphors (Anaki, Faust & Kravetz, 1998; Brownell, Simpson, Birhle, Potter & Gardner, 1990),  comprehending the relationships between an utterance and its context (Joanette, Goulet & Hannequin, 1990), and correctly interpreting jokes and sarcastic utterances (Brownell, Michel, Powelson & Gardner, 1983; McDonald, 1996).  Many of these tasks require the integration of world-knowledge with the contextual information provided in the sentence.  If the RH was relying purely on word-level processing, it would be unable to integrate the sentential information with the world-knowledge necessary to successfully complete the tasks.  Furthermore, the deficit in these tasks observed in patients with right hemisphere damage (RHD) suggest that the Left Hemisphere (LH) is unable to integrate information gleaned from finer level processing with the coarser, world-knowledge context (Beeman, 1998), resulting in them often ‘missing the point’ (e.g. Gardner, Brownell, Wapner & Michelow, 1983).  That is, the LH appears to use a fine-tuned, ‘propositional’ processing, resulting in a high ability to comprehend syntax and grammar but seemingly missing the overall ‘gist’.  The RH, however, appears to use a coarser, ‘discourse-level’ processing, resulting in it having difficulty with detailed processing but being better able to determine the overall meaning via integration of main ideas with ‘real-world’ knowledge.  RHD patients typically exhibit difficulties relating to the conceptualization of a unit as whole, appreciating its purpose and form and integrating specific elements within these forms (Gardner et al., 1983).  Additionally, systematic changes made to the behavioural methodology in previous experiments that were conducted as an earlier component of my PhD appear to have in many ways resolved the apparent inconsistencies observed between behavioural and alternative approaches.   Based on this evidence, it would appear that the question that should now be being asked is not whether each hemisphere uses message and/or word-level processing, but rather how they each use these mechanisms.  As such, the experiment in which you participated aimed to investigate the relative use of contextual information made by each hemisphere when integration across several sentences was required.  It was hypothesized that there would be a relatively greater increase in facilitation for the RH than for the LH as the number of sentences increased.

Method/Design:
     In order to investigate this question sets of three sentences were constructed, such that: when presented alone were of low constraint (i.e., had a low probability of priming the target, as determined by ‘Cloze’ procedures prior to testing; see Taylor, 1953); when presented in a set of two were of medium constraint; when presented as the full set of three were of high constraint.  That is, for facilitation to have occurred, contextual information from each of the three sentences needed to be integrated.  For example, for the target word “Honey”:
Sentence 1: I taste sweet. = low constraint; there are many things this could be
Sentence 2: I can be put on toast. = med. constraint; there are several things this could be e.g. jam, honey, nutella, vegemite, etc.
Sentence 3: I am made by an insect. = high constraint; jam, nutella etc are not made from an insect
In addition, targets were preceded by a neutral sentence; this was designed to provide a baseline measure of reaction time (RT) from which facilitation effects could be determined (i.e, if RT is significantly faster in any of the conditions compared to the neutral condition then this would reflect facilitation).  You were also given scrambled variants of the sentences; this was designed to remove the structure of the sentences, essentially reducing them to lists of single words and thus providing a means of parsing out simple word-level priming effects from the overall message-level priming effects of the normal sentences.  You may also recall that, while the sentences were always presented centrally on the screen, the target words were presented to either the left or to the right of the screen.  The use of lateralized stimuli in the lexical decision task (i.e., your task was to decide whether the target was a word or nonword) is based on the fact that stimuli presented to either the left or right visual hemifield are initially processed only by the contralateral hemisphere (Zaidel, 1983; Hellige, 1993).  As such, this technique is able to reveal the initial hemisphere-specific computations prior to information being transmitted between the hemispheres (via the corpus callosum) and is therefore a standard paradigm used in hemispheric priming studies (Chiarello, 1991).  You might also be interested to know that all conditions were quite extensively balanced across participants so that you only ever saw the same word or nonword target twice; once in the normal condition and once in the scrambled condition.  In order to minimize any repetition priming effects, each complete set of stimuli (covering 20 trials for each condition type per participant) was divided across two sessions, whereby the normal version and the scrambled version of each sentence was presented in a different session, generally one week apart.  This was designed to allow sufficient time between the first and the second presentation of the same word or non-word.  Each session contained equal numbers of normal and scrambled sentences and whether the normal or the scrambled version was presented in the first or the second session was balanced across participants.

Main Results:
     Statistical analyses performed included repeated-measures ANOVAs and t-tests.

Figure 1. The relative mean facilitation (ms) of each visual field/hemisphere presentation as a function of number of sentences for normal compared to scrambled sentences.

    In these sorts of experiments, facilitation is determined by deducting the RT of the condition of interest from the mean RT for the neutral condition.  In other words, when you were presented with a neutral sentence “I am a neutral sentence”, this should not have affected the time it took you to correctly identify the target word. It therefore provides a RT ‘baseline’.  One the other hand, if when you were presented with, say,  three sentences, you were able to correctly identify the target word much faster than in the neutral condition, then the only thing that could have facilitated your response was the information in the sentences.  So, the more facilitation observed, the greater the use of contextual information is presumed.  As shown in Figure 1, the amount of facilitation increased from one, two and three sentences for all conditions but, interestingly, the relative increase in facilitation did not differ significantly for the LH and RH for normal sentences (LHN and RHN).  The hypothesis was therefore not supported.

     The other major comparison of interest is between the normal and scrambled conditions.  These results were quite interesting because the difference in facilitation between the normal and scrambled sentences was not significant for the LH.  Generally, if no significant difference is observed between the normal and scrambled condition, this would be considered to indicate that no message-level processing has occurred.  Given the large body of evidence showing that the LH does utilize message-level processing, it is extremely unlikely this is an appropriate interpretation.  Rather, an examination of the stimulus sets showed that the ‘scrambled’ sentences were not, in fact, without syntactical structure.  Given the sentences were so short, there were many instances where the scrambled variant still resulted in a syntactically valid, albeit unusual, structure.  From a qualitative perspective, many of you commented to me that you thought the scrambled sentences were a bit like ‘yoda-language’ and therefore understandable!  For example, the sentence “I am fast” was scrambled to “Fast I am”.  Rather than being a problem with the methodology, this actually inadvertently produced an extremely interesting observation.  While the LH seemed to have no problem understanding the unusual sentence structure, the RH was still facilitated to a much greater extent by normal than scrambled sentences, indicating that it was indeed treating the scrambled sentences as strings of single words (as anticipated).  This is most likely a greater problem at low constraint (as in the single-sentence condition) than higher constraint (as in the two and three sentence conditions) as less summation priming would occur.  Indeed, this was reflected in the results; the RH had much less facilitation in the single-sentence scrambled condition compared to the equivalent normal condition, but this difference became less when two or three sentences were presented (although still significantly different). 

Conclusions:
     At very least, both hemispheres appear to have benefited from the increased contextual information derived from the additional sentences.  I suspect that the reason why a RH superiority was not observed was that, given the simplicity of the sentences, the LH may still have been able to use a ‘top-down’ propositional model in order to ‘predict’ the target (as opposed to the RH’s use of a ‘bottom-up’, integrative model).  In other words, both hemispheres were most likely using different processes, but the end result was the same (i.e., facilitation).  My next experiment will be investigating this more specifically, in relation to predictive vs. integrative processing.