fMRI evidence that left posterior temporal cortex contributes to N400 effects of predictability independent of congruity

Highlights

• In fMRI posterior MTG responds to both lexical content and predictability.

• Posterior MTG response aligns with N400 amplitude in identical ERP manipulation.

• Middle frontal gyrus responds to both lexical content and incongruity.

• Anterior temporal ‘structure’ region is not modulated by predictability or incongruity.

• Predictive facilitation likely reflects pre-activation, not facilitated integration.

Abstract

Previous electrophysiological work argues that predictability and semantic incongruity rapidly impact comprehension, as indicated by modulation of the N400 component between ~300 and 500 ms. An ongoing question is whether effects of predictability in fact reflect pre-activation in long-term memory as opposed to modulating the kind of integration processes triggered by incongruity. Using fMRI, we compared the impact of predictability and incongruity in adjective-noun phrases, in regions identified with lexical and phrasal localizer scans. We found that predictability impacted activity in left posterior middle temporal gyrus (pMTG), while incongruity impacted activity in left precentral gyrus. Together with parallel data from ERP, these data are consistent with the hypothesis that left pMTG activity is a key contributor to N400 effects of predictability and that the relevant mechanism is reduced activation of stored lexical representations. We tentatively suggest that the left precentral region may play a role in reanalysis when incongruity is encountered.

Introduction

Much psycholinguistic research suggests that during language comprehension readers and listeners routinely predict the content, and perhaps the form, of the upcoming input, and that predicted input is correspondingly processed more easily (Federmeier and Kutas, 1999, Altmann and Kamide, 1999, Wicha et al., 2004; and many others). However, from the early days of this work, a central challenge for studying prediction in language has been dissociating effects of prediction from other processes. In particular, predictability manipulations were often confounded with congruity, such that in one condition the critical word was both highly predictable and semantically congruous, and in the other condition the critical word was both unpredictable and semantically incongruous (e.g. He spread the warm bread with butter/socks; Kutas & Hillyard, 1980). If neural activity is greater in the latter condition than the former, the difference could be attributed either to facilitated lexical or conceptual access, which would reduce activity in the predictable condition, or to increased effort towards computing the sentence- or discourse-level meaning, which would increase activity in the incongruous condition.

This ambiguity has resulted in ongoing debate about the functional interpretation of the well-known N400 effect in ERP, and in uncertainty about the extent to which the N400 effect can be used to study predictive processes in language comprehension. The N400 response refers to a broad negative deflection in the ERP response to words and other meaningful stimuli which peaks around 400 ms post-stimulus onset. Kutas and Hillyard (1984) and hundreds of studies since have observed that the amplitude of the N400 is smaller for words that are predicted by the context. This could be taken to indicate that the N400 indexes activation of lexical and conceptual networks, which can be activated less broadly when a predictive context allows the comprehender to narrow in on the correct candidate (Federmeier & Kutas, 1999). However, much of this data could also be taken to indicate that the N400 only indexes the integration of incoming input into the sentence- or discourse-level meaning, with unpredictable sentences describing scenarios that are less common/plausible and thus require more effort to construct the meaning (Van Berkum, Brown, Zwitserlood, Kooijman, & Hagoort, 2005).

Recent ERP work from our group and others provides new evidence that N400 amplitude is modulated by predictive facilitation. Deconfounding predictability and semantic congruity in sentence contexts has been challenging because even large corpora are too sparse to provide precise estimates of the likelihood of particular sentence continuations, and human sentence continuation data provide precise estimates for highly predictable endings but less precise estimates for unpredictable endings. Lau, Namyst, Fogel, and Delgado (2016) introduced a two-word adjective-noun paradigm, which allows better-supported estimates of predictability from corpus data. We compared the separate effects of predictability (runny nose vs. dainty nose) and semantic incongruity (yellow bag vs. innocent bag). When predictability was controlled, we found that congruity elicited only small, barely reliable N400 differences, while predictability drove a very large N400 reduction. These results suggest that predictability can modulate the N400 response independent of integration difficulty, by hypothesis through facilitated lexical and/or conceptual activation.¹

Here we used the adjective-noun paradigm from Lau et al. (2016) to provide converging evidence from fMRI for the hypothesis that N400 effects of contextual predictability indeed reflect pre-activation of stored lexical and/or conceptual representations. We first used a localizer run to identify regions that were engaged in lexical and sentential processing across participants. Then, just as in our ERP study, participants were presented with two-word adjective-noun phrases and attended to the materials in anticipation of a post-experiment memory recall test. If N400 effects of predictability reflect mechanisms that support basic access of stored lexical representations, we would expect to see effects of predictability in posterior temporal areas that respond to manipulations of lexical content. Alternatively, if N400 effects of predictability reflect mechanisms that integrate stored lexical or conceptual representations, we would expect to see effects of predictability in frontal or anterior temporal regions that respond to manipulations of sentential structure and meaning.

Which brain regions generate N400 context effects has also been a matter of some debate. As reviewed by Lau, Phillips, and Poeppel (2008; see also Van Petten & Luka, 2006), fMRI effects in mid-posterior temporal cortex appear to best track semantic priming manipulations that generate N400 effects, and MEG work using sentence contexts is also consistent with this (although see e.g. Maess et al., 2006, Hagoort, 2008 for dissenting perspectives). However, the results of fMRI studies using congruity/predictability manipulations in sentences have been much more variable. This likely reflects the fact that the temporally ‘sluggish’ fMRI response sums the response to most or all words in the sentence, resulting in increased variability that can mask the differential response to the single word that is manipulated. An additional virtue of the adjective-noun paradigm used here is that only the single context word and the target word contribute to the response.