UC San Diego
Electrophysiological explorations of linguistic pre- activation and its consequences during online sentence processing
- Author(s): DeLong, Katherine Ann
- et al.
Historically, anticipation has played only a minor role in language comprehension theories, with the potential for error precluding the utility of such a strategy, except in conditions of high constraint or degraded input. Such views, however, may not factor in how meaning is stored, retrieved or constructed in the brain. Recently, evidence for the benefits (and to a lesser extent, the costs) of linguistic prediction has begun to accrue, with methodologically diverse findings revealing that various aspects of comprehension (e.g., semantic, syntactic, and phonological) are shaped by top-down processing. Two general difficulties of investigating predictive sentence comprehension are that (a) effects of anticipation and integration are difficult to disentangle, and (b) finding evidence for upcoming events requires clever manipulations and experimental techniques that allow for continuous tracking of responses as contexts unfold. The research presented in this dissertation addresses these concerns by using paradigms (and methods) that allow for pre-target detection of prediction as well as monitoring throughout subsequently predicted targets. We utilize event-related brain potentials (ERPs) and a variety of analytical approaches to investigate probabilistic preactivation across a range of constraining sentence contexts. These studies contrast the contributions of facilitative context (a narrowing of constraint) with semantic fit of particular words. They also examine the consequences of unrealized linguistic predictions - i.e., contextually constrained information presumably preactivated, but ultimately disconfirmed by the input. In the first of four studies we demonstrate anticipation of lexical items via contextual constraint, as well as noting an effect suggestive of an expectancy violation cost. In the second experiment, we demonstrate that predictive processing and its potential costs generalize, to a large extent, across input rate, offering support for fast, unconscious preactivation of linguistic content. In the third study, we test the nature of our proposed prediction cost, ultimately linking it to graded, general constraint violation. In the final experiment, we show that though both cerebral hemispheres are sensitive to message-level constraint, the left hemisphere appears biased toward more predictive sentence processing. Throughout, we take multiple approaches to defining constraint and expectancy, and argue that traditional quantifications of these concepts may conflate rich sources of variability.