Wonnacott, E. 1 , Gunning, L. . 2 & Brown, H. 2
1 University College London
2 University of Warwick
Artificial language experiments have explored the learning of gender-like noun categories, however these have generally used adult participants. We explore whether 7-year-olds can acquire gender classes via distributional learning using input from a real language (Italian) with two gender classes. Monolingual English speaking children heard singular and plural definite NPs of four types:
masculine singular: il noun[masculine] +o e.g. (IL lettO)
masculine plural: i noun[masculine] +i (I lettI)
feminine singular: la noun[feminine] +a (LA balenA)
feminine plural: le noun[feminine] +e (LE balenE)
(All nouns were inanimate with no semantic cues to gender).
Two (between-subject) conditions manipulated type-frequency: high-type-frequency (8m/8f nouns), low-type-frequency (4m/4f; total exposure matched). We predicted greater generalization under high-type frequency.
Children received input over 5 days via a game (hear NP, identify correct picture), with comprehension and production tests on days 1&5. Data collection is ongoing. Preliminary data (n=20) provide evidence of generalization: when children hear a singular NP with a new noun and produce its plural (or vice versa), they are more likely to produce the correct, rather than incorrect, gendered determiner and vowel, despite neither having occurred with this noun in the input. There was no type-frequency effect, suggesting four instances may be sufficient for generalization.