We Hebben Een Serieus Translatie: Modeling Intercomprehension as Probabilistic Inference

Researchers modeled intercomprehension—partial intelligibility of an unfamiliar related language—using a Bayesian noisy-channel framework with an L1 language model and a noise model mapping L2 words to L1. A human behavioral experiment tested speakers of English, Spanish, and Russian on Dutch, Italian, and Ukrainian. The full model aligned more closely with human performance than ablations and outperformed zero-shot prompting of larger models, offering a cognitively plausible account of cross-language comprehension.