This study reports on a self-paced reading experiment in which native and non-native speakers of English read sentences designed to evaluate the predictions of usage-based and rule-based approaches to second language acquisition (SLA). Critical stimuli were four-word sequences embedded into sentences in which phrase frequency and grammaticality were crossed in order to examine whether grammatical processing is modulated by phrase frequency. The magnitude of grammaticality effects for native speakers did not differ by phrase frequency, indicating that phrase frequency does not modulate native grammatical processing. Phrase frequency did, however, modulate the magnitude of non-native grammaticality effects. This modulating effect of phrase frequency on grammatical processing for non-native speakers depended on proficiency, showing a u-shaped change in the size of grammaticality effects relative to speaker proficiency. The overall pattern of change in grammaticality effects suggests a gradual developmental shift in grammatical processing from an initial reliance on phrase frequency to an eventual abstraction of generalizable rules from the linguistic input once sufficient experience has accumulated. Results suggest that second language (L2) grammatical development relies on a combination of both usage-based and rule-based knowledge and processing, rather than exclusive reliance on one or the other.

No abstract

Mixed effects regression models are widely used by language researchers. However, these regressions are implemented with an algorithm which may not converge on a solution. While convergence issues in linear mixed effects models can often be addressed with careful experiment design and model building, logistic mixed effects models introduce the possibility of separation or quasi-separation, which can cause problems for model estimation that result in convergence errors or in unreasonable model estimates. These problems cannot be solved by experiment or model design. In this paper, we discuss (quasi-)separation with the language researcher in mind, explaining what it is, how it causes problems for model estimation, and why it can be expected in linguistic datasets. Using real linguistic datasets, we then show how Bayesian models can be used to overcome convergence issues introduced by quasi-separation, whereas frequentist approaches fail. On the basis of these demonstrations, we advocate for the adoption of Bayesian models as a practical solution to dealing with convergence issues when modeling binary linguistic data.

Late second language (L2) learners experience pervasive difficulty mastering grammatical gender, and a comprehensive account of this deficit has yet to emerge. We investigate a previously unexamined aspect of L2 gender use: the time course of lexical feature retrieval. Using event-related potentials (ERPs) with a covert production task, we examined whether L2 gender retrieval is delayed relative to phonology and to the time course of feature retrieval in native speakers for familiar nouns whose gender participants had strong knowledge of. Results find that L2 gender retrieval is not fundamentally delayed, and that L2 lexical feature retrieval may be more susceptible to top-down influences. These findings place important constraints on accounts of L2 acquisition and processing with respect to how lexical features are represented and retrieved. Our results further suggest that deficits in online L2 gender use may stem from post-retrieval processes and/or retrieval errors rather than inherent delays in gender retrieval.

No abstract