Neurobehavioral Correlates of Surprisal in Language Comprehension: A Neurocomputational Model

Abstract

Expectation-based theories of language comprehension, in particular Surprisal Theory, go a long way in accounting for the behavioral correlates of word-by-word processing difficulty, such as reading times. An open question, however, is in which component(s) of the Event-Related brain Potential (ERP) signal Surprisal is reflected, and how these electrophysiological correlates relate to behavioral processing indices. Here, we address this question by instantiating an explicit neurocomputational model of incremental, word-by-word language comprehension that produces estimates of the N400 and the P600-the two most salient ERP components for language processing-as well as estimates of "comprehension-centric" Surprisal for each word in a sentence. We derive model predictions for a recent experimental design that directly investigates "world-knowledge"-induced Surprisal. By relating these predictions to both empirical electrophysiological and behavioral results, we establish a close link between Surprisal, as indexed by reading times, and the P600 component of the ERP signal. The resultant model thus offers an integrated neurobehavioral account of processing difficulty in language comprehension.

Keywords: N400; P600; event-related potentials (ERPs); language comprehension; surprisal theory.

Figure 1

Offline ratings from Delogu et al. (2019) for plausibility (left) and association (middle), and estimated Cloze probability of the target (right) in all three conditions.

Figure 2

Schematic illustration of the neurocomputational model. Each rectangle represents a layer of artificial (leaky rectified linear) neurons, and each solid arrow represents full connectivity between each neuron in a projecting layer and each neuron in a receiving layer. The dashed rectangle is a context layer, and the dashed arrow represents a copy projection, such that prior to feed-forward propagation the integration_output layer receives a copy of the integration layer. All groups except the input and integration_context layer also receive input from a bias unit (not shown). See text for details.

Figure 3

Model predictions: N400-effects (left, plotted downwards; see text) and P600-effects (right), for the event-related condition relative to baseline, and for the event-unrelated condition relative to baseline. Error bars show standard errors.

Figure 4

Model predictions: Surprisal effects for the event-related condition relative to baseline, and for the event-unrelated condition relative to baseline. Error bars show standard errors.

Figure 5

Topographic maps of the ERP effects in the N400 time window (300–500 ms, left column) and the P600 time window (600–1, 000 ms, right column). The upper panel shows the difference between the event-related condition and the baseline. The lower panel shows the difference between the event-unrelated condition and the baseline. Reproduced with permission (CC BY-NC-ND 4.0) from Delogu et al. (2019).

Figure 6

Effects as estimated using regression-based ERP (rERP) estimation: the isolated effects of association in the N400 time-window (300–500 ms, left), and the isolated effects of plausibility in the P600 time-window (600–1, 000 ms, right) for the event-related condition relative to baseline, and for the event-unrelated condition relative to baseline. Error bars show standard errors.

Figure 7

Self-paced reading times (RTs) effects in the target region (left) and the spillover region (right), for the event-related condition relative to baseline, and for the event-unrelated condition relative to baseline. Error bars show standard errors.