Latent error regimes in the learning of plural agreement in SFL

Abstract

The dynamics of three hidden learning regimes in the production of nominal number agreement are analyzed in a corpus of four italian learners of Spanish FL, in a longitudinal case study. Within the framework of complex systems theory, a Bayesian topic discovery model and a hidden Markov model were applied. The dynamics tended to remain in the regime associated with the «correct» attractor (or eventually return to it) with agreement instances: (1) with concrete, familiar and animated controller; (2) when facilitating learning strategies are applied; (3) with more frequent TYPES; (4) with epenthetic «e»; (5) with more than two members. Instead, it tended to move away from the «correct» attractor with: (1) long distance agreement; (2) instances with modifiers that were not definite articles; in particular at low levels of proficiency (A, B1). The simulation was close to the empirical data, but the emission probabilities were underestimated.

Keywords

• second language
• dynamical systems
• nominal agreement
• hidden markov model
• plural