Artificial grammar learning and training effects across modalities and domains

Lukács, &. 1, 2 , Dobó, D. 1, 2 & Lukics, K. S. 1, 2

1 Department of Cognitive Science, Budapest University of Technology and Economics
2 MTA-BME Lendület Language Acquisition Research Group, Budapest, Hungary

Modality and domain effects have been demonstrated in statistical learning by several studies (SL; e.g. Conway & Christiansen, 2005, 2006; Frost et al., 2015). Our aim was to directly compare the efficiency of SL in extracting linguistic (L) and non-linguistic (NL), visual (V) and acoustic (A) patterns in an artificial grammar learning (AGL) task in young adults. We also tested how characteristics of input presentation (starting small--incremental presentation of stimuli of different length versus random presentation; and 1x versus 2x training length) influence learning across these conditions. Learning was tested with serial presentation of items with the following stimuli: acoustic nonsense syllables (AL), pure tones (ANL), visual nonsense syllables (VL) and nonlinguistic symbols (VNL). Significant domain and modality effects were observed: learning was more effective in the acoustic than in the visual modality, and in the verbal versus nonverbal domain (ANL~<VL=VNL<AL). The linguistic advantage was only present in the acoustic modality. Starting small only improved learning with AL materials, and with shorter training. Doubling training length enhanced performance only in the visual modality. Our findings support the acoustic advantage of sequential learning for linguistic stimuli, and suggest that the starting small effect might be language-specific.