Modeling the N400 ERP component as transient semantic over-activation within a neural network model of word comprehension

Abstract

The study of the N400 event-related brain potential has provided fundamental insights into the nature of real-time comprehension processes, and its amplitude is modulated by a wide variety of stimulus and context factors. It is generally thought to reflect the difficulty of semantic access, but formulating a precise characterization of this process has proved difficult. Laszlo and colleagues (Laszlo & Plaut, 2012, Brain and Language, 120, 271-281; Laszlo & Armstrong, 2014, Brain and Language, 132, 22-27) used physiologically constrained neural networks to model the N400 as transient over-activation within semantic representations, arising as a consequence of the distribution of excitation and inhibition within and between cortical areas. The current work extends this approach to successfully model effects on both N400 amplitudes and behavior of word frequency, semantic richness, repetition, semantic and associative priming, and orthographic neighborhood size. The account is argued to be preferable to one based on “implicit semantic prediction error” (Rabovsky & McRae, 2014, Cognition, 132, 68-98) for a number of reasons, the most fundamental of which is that the current model actually produces N400-like waveforms in its real-time activation dynamics.

Language ERPs reflect learning through prediction error propagation

“…Compared to other models of ERPs, the Error Propagation account has several unique properties. Most ERP models only explain the N400 (Cheyette & Plaut, 2017;Frank et al, 2015;Laszlo & Federmeier, 2011;Laszlo & Plaut, 2012;Laszlo & Armstrong, 2014;Rabovsky & McRae, 2014;Rabovsky et al, 2018). Brouwer et al (2017) model can also explain one P600 study, but the Error Propagation account explicitly captures multiple N400 and P600 effects within the same model.…”

“…Laszlo and Armstrong (2014) refined this approach to explain the effect of repetition on N400 magnitude. Cheyette and Plaut (2017) further extended the approach to cover a wider range of phenomena such as word frequency, semantic richness, priming, and orthographic neighborhood size effects. These models incorporate biologically motivated assumptions about connectivity and when they are combined with an appropriate training regime, they can show peaks around the N400 time window.…”

“…The next set of distinctions relates to their theory for linking the model to ERPs. Laszlo and colleagues (Laszlo & Plaut, 2012;Laszlo & Federmeier, 2011;Laszlo & Armstrong, 2014), as well as Cheyette and Plaut (2017), use mean or total activation at each point in time to represent N400 waveforms. The Rabovsky et al (2018) and Brouwer et al (2017) models use the activation change or update between two adjacent time points to encode ERPs.…”

“…An important feature of the data from Coulson et al (1998) study was that the manipulation of probability only influenced amplitude, but not timing. In models where the timing of ERPs is shaped by learning (Laszlo & Plaut, 2012;Laszlo & Federmeier, 2011;Laszlo & Armstrong, 2014;Cheyette & Plaut, 2017), it is possible that within-experiment learning would change the timing of ERPs. On the other hand, the Error Propagation account proposes that timing is determined by the number of layers that error must be propagated across in the network, so the timing cannot be changed by experience in the same way as amplitude.…”

Quasi-compositional mapping from form to meaning: a neural network-based approach to capturing neural responses during human language comprehension

“…In recent years, there has been growing interest in linking N400s to neural network models [30,[35][36][37][38][39][40][41]. Our own account focuses on modelling N400 amplitude, i.e.…”