Experimentally comparing the learnability of rule interactions

Wang, Yuxuan Melody. (2024). Experimentally comparing the learnability of rule interactions. In Proceedings of the 14th International Conference of Experimental Linguistics, ed. by A. Botinis. Exling Society.

Abstract: This paper aims to test the claim that opaque interactions are harder to learn than transparent ones with artificial grammar learning (AGL) experiments and the ‘povertyof-stimulus’ paradigm. The participants were first taught Vowel Harmony and Palatalisation separately and then tested on whether they preferred to let them feed or counterfeed in ambiguous environments. If transparent interactions are indeed preferred, significantly more subjects should choose the Feeding option. Result indicated that significantly more participants preferred the opaque interaction instead, contrary to predictions. Preference for opacity might be attributed to structural simplicity or a dispreference for the application of the Palatalisation rule.

Keywords: opacity, AGL, learnability, natural bias