Language Learning Following Immigration: Modeling Choices and Challenges

Abstract

No agent-based model exists of language learning following immigration to a new country. Language learning has features which make it a good fit to Agent Based Models (ABMs), such as diffusion/adoption effects: people learn language via social interaction and are influenced by other social actors about how and when to invest in learning. Language learning involves positive and negative feedback loops, such that poor progress early in learning can spiral into negativity and avoidance, while early success can accelerate learning. Most importantly, the question of why language learning is difficult for adults is controversial. Should implementers program into models the equations that match the robust age effects observed in data, or should these patterns emerge from multiple factors and actors? To address this, the large literature on foreign language acquisition was reviewed as part of the background of making modeling decisions. Decisions were sufficiently challenging that it was decided to begin with a narrative description, using the Overview, Design Concepts and Details protocol (ODD). The ODD protocol provided an organizing framework in which many details were worked out. These included identifying outcome variables (frequency of use and fluency in the two languages), basic entities (representing individuals, families, neighborhood, global environment), defining rules for initiating and continuing conversation, and rules for agents to move to new locations. Considerable narrative space was used to discuss the rationale for simplifications, as well as decisions that were left for future extensions. Given the complexity of the domain, the narrative description was a necessary foundation to smooth the way for a working simulation.