Modelling orthographic similarity effects in second language vocabulary acquisition

In this thesis, we aimed to integrate insights from psycholinguistics into models of learning and memory. These models typically treat words as separate facts to be learned by rote memorization, disregarding psycholinguistic evidence of strong interactions between orthographically similar words in memory and processing during learning. In particular, we reexamined the position of vocabulary learning in the KLI framework, taking into account effects of orthographic similarity, and the contrasting work of our colleagues Baxter et al. (2021, 2022). Additionally, we validated effects of wordlikeness observed in the laboratory on a large-scale dataset of vocabulary acquisition in a digital learning environment (Settles et al., 2018). Furthermore, we modified an established model of memory and learning (Pavlik Jr. & Anderson, 2003) to explicitly model interactions between orthographically similar words in memory, informed by the Deese-Roediger-McDermott paradigm (Deese, 1959; Roediger & McDermott, 1995) and spreading activation models of the mental lexicon (e.g., Dijkstra et al., 2019; McClelland & Rumelhart, 1981). Finally, we adapted another established learner model, the deep knowledge tracing model of Piech et al. (2015), by incorporating orthographic embeddings into the model training process. We hope that the theoretical insights and practical recommendations presented in this thesis will aid in the development of the next generation of computer-assisted language learning applications.