In recent decades, language acquisition research has focused on explaining how language is learned through domain-general (i.e., non-language specific) mechanisms. One such mechanism that has shown great explanatory power is statistical learning: our ability to pick up on probabilistic regularities in our perceptual world (Erickson & Thiessen, 2015; Romberg & Saffran, 2010). Research in statistical learning has, for instance, repeatedly documented that children and adults alike can implicitly learn syntactic regularities from artificial grammars through passive exposure to novel language input (e.g., Saffran, 2002). However, studies of artificial-grammar learning have typically relied on experimental procedures that poorly reflect the complexity of language learning in the real world: First, language learning requires successfully integrating syntactic, semantic, and pragmatic information, whereas statistical learning research has traditionally focused mostly on syntax (see e.g., Misyak & Christiansen, 2012; Saffran, 2002). Second, studies have relied primarily on passive observational exposure to the target language, which fails to capture the intrinsically interactional nature of real-life language learning (Dale & Christiansen, 2004).Aim of this project is therefore to further develop and test a new experimental paradigm for the study of statistical learning that better approximates real-world language learning. The IMC seed funding will allow the research group — which builds on a collaboration with international experts in the field — to further develop and test a paradigm that will enrich existing methods in statistical learning research by: (1) allowing for the simultaneous and systematic investigation of syntactic and semantic (and potentially also pragmatic) information in one unified artificial-grammar framework; (2) tapping into active rather than passive statistical learning — by making the learning process more interactive, e.g. by providing positive feedback for correct responses; (3) allowing us to observe learning as it unfolds in real time, instead of relying on post hoc testing; (4) allowing us to investigate how non-linguistic factors (e.g., the presence of noise in the environment) may impede learning. At the same time, the paradigm — which we call The Picture Guessing Game (henceforth PGG) — builds on a simple sentence-picture matching method, and as such it retains the simplicity and immediacy that characterizes existing statistical learning paradigms.