Modeling and dynamical analysis of cortical network activity in semantic priming

Abstract

The semantic priming paradigm, involved in language comprehension, refers to the facilitated processing and retrieval of a word (known as target) following the previous processing of another semantically related word (known as prime). Literature on semantic priming reveals a vivid debate about the nature of priming: it can be associative (e.g. afraid-scared), semantic, that is “true relations of meanings”, (e.g. sheep-goat) or a combination of both (e.g. cat-dog). This debate impacts then how the semantic memory, coding words’ meaning, is modeled and how the priming is thought to occur.

Brunel and Lavigne (2009) designed a network model that studies semantic priming as a function of a set of parameters. In addition, they used an input-output relationship that is mathematically good-looking but rather difficult to manipulate numerically. This master thesis thus focuses on assessing whether using a more standard and a more numerically stable input-output relationship, such as a sigmoid function, would give a qualitatively similar dynamic behavior to the original model. Furthermore, the thesis investigates the parameter sensitivity. To these ends, the network model is simplified into a one-dimensional model and the dynamic behavior is investigated for both input-output relationships. The modified model is then tested with experimental-like stimuli to mimic real psychology experiments and to understand semantic memory functioning.

Dynamical analysis performed on the derived one-dimensional model reveals that the dynamic behavior remains qualitatively the same when using one or the other input-output relationship. Results also suggest parameter sensitivity of the original model. The modified model with experimental-like stimuli suggests that the semantic memory system should be in a bistable regime to observe semantic priming. Activation of a word in semantic memory depends then on the amplitude and the duration of the stimulus. Extension to higher-order dimensions is also discussed.