Wonnacott, E. , Vujovic, M. & Miller, C. .
Department of Language and Cognition, University College London
Adult grammatical knowledge includes linguistic structures which operate across linguistic items. In naturalistic child language learning, generalizations may emerge gradually, provided structures are witnessed in sufficiently variable exemplars (Bybee 1995; Ramscar et al. 2010; Wonnacott et al. 2012). However, such implicit generalization proves difficult to simulate in lab-based learning experiments (Ferman, & Karni, (2010); Brown et al (pre-print, https://osf.io/prj2k/)), possibly because in these participants (adults and children) are not sufficiently engaging in predictive processing, which would allow prediction error to dissociate structures from uninformative cues.
The current experiment employs a paradigm encouraging explicit prediction. 7-8 year-olds played a computerized game exposing them to instructions in an unfamiliar language (Japanese) involving two postpositional case markers no ue ni (above) and no shita ni (below). Children heard the sentences whilst viewing a grid (Figure 1), and attempted to follow the instruction by moving the pictures. Importantly, incorrect responses were followed by a demonstration of the correct response (though no explicit teaching was provided). Training was followed by a generalization test involving untrained sentences containing untrained nouns (more English cognates).
Critically, there were two between-participant training conditions: (1) high-variability exposure: 28-above and 28-below sentences, each encountered once; (2) low variability exposure: 2-above and 2-below sentences, each repeated 14 times; (number of nouns (8) and total exposure matched across conditions). We predicted high-variability exposure would lead to greater generalization due to better dissociation of the structures from trained instances.
Data collection is ongoing. Inspection of the preliminary sample (N=36) indicates: In training: both participant groups improved, however the low-variability condition showed strongest overall performance training (Cohen?s d=.84), presumably due to repeated testing on identical items. At test: critically, the high-variability condition showed stronger performance in generalization (Cohen?s d=.64). This supports the hypothesis that exemplar variability plays a key role in driving linguistic generalization.