Many Languages, One Network: Cross-Linguistic Semantic Priming Effects in 45 Languages and Large Language Models

Understanding how words relate to each other in human memory is crucial for many scientific fields ranging from artificial intelligence to language education. For this project, scientists build a large-scale dataset to understand how recognizing one word (e.g., cat) can improve recognition of another conceptually related word (e.g., dog), called semantic priming. Researchers examine how the relationship between words is the same or different across 45 languages and compares human responses to artificial intelligence. The project promotes collaborative participation in research and open science practices. Scientists integrate cognitive science, linguistics, and computational modeling to investigate word-level semantic priming effects across languages. The researchers analyze predictors of why certain word pairs exhibit stronger priming effects than others, incorporating psycholinguistic variables and large language model (LLM) word embeddings to model these effects. A key goal is to compare how closely LLMs mimic human semantic processing. Another goal is to explore cross-linguistic patterns in semantic organization across linguistic systems. The project creates an open dataset including linguistic properties, subjective ratings, similarity metrics, and response times for word pairs. This framework for big-team science advances open, collaborative, and reproducible research in language science, bridging the gap between experimental psychology and computational linguistics. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.