The neural architecture of discourse compression

Abstract

Re-telling a story is thought to produce a progressive refinement in the mental representation of the discourse. A neuroanatomical substrate for this compression effect, however, has yet to be identified. We used a discourse re-listening task and functional magnetic resonance imaging (fMRI) to identify brain regions responsive to repeated discourse in twenty healthy volunteers. We found a striking difference in the pattern of activation associated with the first and subsequent presentations of the same story relative to rest. The first presentation was associated with a highly significant increase in blood oxygen level dependent (BOLD) signal in a bilateral perisylvian distribution, including auditory cortex. Listening to the same story on subsequent occasions revealed a wider network with activation extending into frontal, parietal, and subcortical structures. When the first and final presentations of the same story were directly compared, significant increments in activation were found in the middle frontal gyrus bilaterally, and the right inferior parietal lobule, suggesting that the spread of activation with re-listening reflected an active neural process over and above that required for comprehension of the text. Within the right inferior parietal region the change in BOLD signal was highly correlated with a behavioural index of discourse compression based in re-telling, providing converging evidence for the role of the right inferior parietal region in the representation of discourse. Our findings demonstrate, for the first time, the existence of a neural network underlying discourse compression, showing that parts of this network are common to re-telling and re-listening effects.

Introduction

According to current linguistic views discourse processing is hierarchical (Graesser et al., 1994, Greene et al., 1992, Kintsch and van Dijk, 1978). Microstructure is formed through the initial extraction of propositions which are organized into a multileveled network, operating at the level of word meanings, syntactic rules, and local cohesion within sentences through mechanisms such as substitution and ellipsis. Memory constraints dictate that only a subset of propositions is retained, forming a macrostructural representation of the message. Refinement of the macrostructural representation occurs through processes such as deletion, generalization and constitution, and most important in the present context, can be facilitated by re-telling (Bloom, 1994).

The mental representation of discourse is not simply a function of the text itself, but is shaped and determined by a variety of cognitive, sociolinguistic, and situational factors including repetition (Johnson-Laird, 1983, Van Dijk and Kintsch, 1983, Xu et al., 2005). Re-telling progressively produces a more compact and concise narrative. This is seen as a decrease in speaking time and word count, as well as the duration and distribution of pauses across repeated cycles (Goldman-Eisler, 1968). These changes will be referred to here as the compression effect. A similar effect is seen on tasks of discourse comprehension where silent reading of the same passage of text is associated with a decrease in reading time (Levy & Burns, 1990). In theoretical terms, refinement through repetition reflects the progressive loss of direct links between the microlinguistic aspects of the text or language output and the overarching mental representation. Failure to compress a message with repetition might be general marker of inefficient discourse representation, as suggested by the absence of a compression effect on re-telling in neurologically impaired individuals, such as those with temporal lobe epilepsy (Field, Saling, & Berkovic, 2000).

The cerebral organization of discourse compression is of considerable interest. To the extent that compression is a marker of effective representation, the existence of a neural network that is progressively recruited on repeated engagement with narrative, either in re-listening or re-telling, could be hypothesized. We are aware of only one neuroimaging study that throws some light on this issue. While not designed to isolate a cerebral network underlying the compression effect, a PET activation study on comprehension and memory of discourse conducted by Maguire, Frith, and Morris (1999) showed recruitment of the left middle frontal gyrus and medial parietal cortices when a second hearing of a story was compared with a first hearing.

The aim of the present study was to study the functional neuroanatomy of compression using functional magnetic resonance imaging (fMRI), and an activation task in which subjects listened to a story presented in on three consecutive occasions. Our major contrast of interest was the change in blood oxygen level dependent (BOLD) signal between the first and final presentations of the story. To establish if the regions identified in this way were specifically associated with discourse changes on re-telling, we conducted correlational analyses between fMRI data and performance on an established, out-of-scanner measure of the compression effect (Goldman-Eisler, 1968).