Thinking ahead: the role and roots of prediction in language comprehension

Abstract

Reviewed are studies using event-related potentials to examine when and how sentence context information is used during language comprehension. Results suggest that, when it can, the brain uses context to predict features of likely upcoming items. However, although prediction seems important for comprehension, it also appears susceptible to age-related deterioration and can be associated with processing costs. The brain may address this trade-off by employing multiple processing strategies, distributed across the two cerebral hemispheres. In particular, left hemisphere language processing seems to be oriented toward prediction and the use of top-down cues, whereas right hemisphere comprehension is more bottom-up, biased toward the veridical maintenance of information. Such asymmetries may arise, in turn, because language comprehension mechanisms are integrated with language production mechanisms only in the left hemisphere (the PARLO framework).

Figure 1

Grand average ERPs (N = 21 young adults) shown at the middle central electrode site (Cz). Negative is plotted up in this and all subsequent plots. At the top is plotted the overall response to auditory sentence-final words that were either (1) most expected in the context (expected exemplars; solid line), (2) unexpected in the context but from the same semantic category as the expected completion (within-category violations; dashed line), or (3) unexpected and from a different semantic category (between-category violations; dotted line). At the bottom is plotted the response to the same three ending types in strongly (left) and weakly (right) constrained sentence contexts. Within-category violations were facilitated relative to between-category violations, especially when presented in strongly constrained contexts, despite their lower plausibility in those sentences. Data are from Federmeier et al. (2002).

Figure 2

Grand average ERPs (N = 24 older adults) shown at the middle central electrode site (Cz). Analogous to Figure 1, at the top is plotted the overall response to expected exemplars (solid line), with in category violations (dashed line), and between-category violations (dotted line). At the bottom is plotted the response to the same three ending types in strongly (left) and weakly (right) constraining sentence contexts. Older adults exhibited a different pattern from that seen for young adults, showing little facilitation for within-category violations, most of which derived from the weakly constrained sentences, consistent with plausibility. Data are from Federmeier et al. (2002).

Figure 3

Grand average ERPs (N = 18 young adults) shown at the middle central electrode site (Cz). Responses to expected exemplars (solid line), within-category violations (dashed line), and between-category violations (dotted line) are plotted at left for presentation in the right visual field (initially apprehended by the left hemisphere) and at right for presentation in the left visual field (initially apprehended by the right hemisphere). Within-category violations were facilitated only for presentation to the RVF/LH. Data are from Federmeier and Kutas (1999a).

Figure 4

Grand average ERPs (N = 32 young adults) shown at a representative frontal electrode site. Responses to sentence-final words in strongly (solid line) and weakly (dotted line) constrained contexts are plotted for presentation to the RVF/LH (at the left) and the LVF/RH (at the right). P2 responses were larger for words in strongly than in weakly constrained contexts only with RVF/LH presentation. Data are from Federmeier et al. (2005).

Figure 5

Grand average ERPs (N = 24 young adults) shown at the middle central electrode site (Cz). Overlapped are responses to words presented to the LVF/RH (solid line) and the RVF/LH (dotted line) when these were high typicality exemplars of the cued category (left), low typicality exemplars of the cued category (middle), or not from the cued category (incongruent, right). Whereas responses to high typicality and incongruent exemplars were similar in the two hemispheres, N400 responses to low typicality exemplars were more facilitated with LVF/RH than with RVF/LH presentation. Unpublished data.

Figure 6

Bar graphs showing memory for words as reflected in response times (RTs) and ERP old/new effects. The top graph plots the RTs for correct endorsements (hits) of centrally presented words that were originally studied in the RVF/LH (black bars) and LVF/RH(white bars) when these words were repeated at short lags (2 or fewer intervening words), medium lags (4–9 intervening words), or long lags (19–49 intervening words). Whereas at short and medium lags RTs were faster for words originally studied in the RVF/LH, at long lags RTs were faster for words originally studied in the LVF/RH. The bottom graph plots the corresponding ERP old/new effects (mean amplitude response to hits minus response to correct rejections, measured at nine central-posterior electrode sites between 250 and 800 ms) for centrally presented words originally studied in the RVF/LH (black bars) and LVF/RH(white bars) at the same three lags. Old/new effects did not differ by study VF for words tested at short and medium lags, but were larger (reflecting more memory signal) for LVF/RH-studied words when these were tested at long lags, analogous to the RT pattern. Behavioral data are from Federmeier and Benjamin (2005) and ERP data are from Evans and Federmeier (2007).

Figure 7

Grand average ERPs (N = 24 young adults) shown at a representative frontal electrode site. Responses to (centrally presented) correctly classified old words (hits; solid line) and correctly classified new words (correct rejections; dotted line) are shown for items originally studied in the RVF/LH (at the left) and the LVF/RH (at the right). A P2 repetition effect (larger P2 responses for old than for new words) was observed only with LVF/RH presentation. Data are from Evans and Federmeier (2007).