Modality Matters: Generalization in Second Language Learning after Production versus Comprehension Practice.

다국어 초록 (Multilingual Abstract)

Generalization is the ability to apply regularities to novel instances, for example, correctly guessing that the plural for the novel English word ‘wug’ should be ‘wugs’. Early language learners make overgeneralization errors like ‘mouses’, applying regularities beyond their attested uses. Theories concerned with the question of how learners learn to correctly generalize regularities, without overgeneralizing, have recently been criticized for being insufficiently mechanistic. Rule learning and statistical learning theories typically do not take into account whether that generalization is happening during production (e.g. coming up with the plural for ‘wug’), or comprehension (e.g. judging whether ‘wugs’ or ‘wugga’ sounds better as a plural for ‘wug’). However, my own prior research showed that training modality affects regularity learning, with production training leading to a more accurate ability to apply regularities to learned words. Thus, modality may provide a potential path to making theories more specific and mechanistic.In this thesis, I first reviewed relevant literature on generalization through a task modality lens. There is very little literature directly contrasting, in a balanced manner, the effects of production versus comprehension training on learning and generalization. However, generalization studies have used both production and comprehension testing to assess generalization performance. Drawing on these results, I identified several different patterns of generalization results by testing modality. I concluded that, if there are any modality differences, production training should lead to better generalization, and production tests should be more likely to elicit overgeneralization errors. I then designed and conducted an artificial language learning experiment that contrasts production versus comprehension training modality between different groups of participants, and uses both comprehension and production testing to assess learning and (over)generalization. People learning the artificial language with production training were better at generalizing and made fewer overgeneralization errors than people learning the artificial language with comprehension training. Surprisingly, comprehension-trained people did do better than production-trained people on vocabulary learning. Finally, people made overgeneralization errors in both comprehension and production tests. I discussed consequences for existing theories as well as practical applications of my findings for second language learners.