Dango: A Strictly L1-Only Large Language Model for Studying Second Language Acquisition

Dango is a newly introduced 1.8B-parameter large language model specifically designed for studying L1-to-L2 transfer in second language acquisition, focusing on Japanese-to-English. It utilizes a novel filtering method to mitigate L2 contamination during pretraining, enabling more realistic simulations of L2 acquisition. The model has been fine-tuned on L2-learning lessons and demonstrates superior performance in generating human-like L2 production patterns compared to existing multilingual baselines, making it a valuable tool for researchers and practitioners in computational SLA.