Computationally modeling an incremental learning account of semantic interference through phonological influence

Computer models play a vital role in providing ways to effectively simulate complex systems and to test scientific theories and hypotheses. One major area of success for neural network models in particular has been in cognitive neuroscience for modeling semantic interference effects in memory. When a person sees a picture of an object such as a car multiple times, the memory of that object is primed so that it can be retrieved more effectively. When a picture of a similar object is seen, such as a truck, sharing semantic features with the primed object, then the primed memory of a car would interfere with the retrieval of a truck. This is known as semantic interference. A recent hypothesis by Preusse et al. (2013) puts forward that semantic interference is further increased by the sharing of phonemes among two words. In this thesis a new phonological computer model of lexical retrieval is developed based on this hypothesis using a two layer feedforward Artificial Neural Network (ANN). The new model can represent semantic interference effects through increased lexical activation by phonological features. Simulations were performed in a MATLAB environment each using a different variant of the phonological model. The simulations tested three conditions of activating semantic and phonological features. Results demonstrated that semantic interference is significantly increased when phonological features are activated alongside semantic features versus activating semantic features alone thus supporting the hypothesis by Preusse et al. (2013). The characteristics of the new ANN model could make it useful in studying other phenomena related to memory and learning.