730 likes | 884 Views
Turning an L1 three-way contrast into an L2 two-way contrast. Paola Escudero University of Utrecht and McGill University Paul Boersma University of Amsterdam Second International Conference on Contrast in Phonology Toronto, May 3, 2002. Introduction.
E N D
Turning an L1 three-way contrast into an L2 two-way contrast Paola Escudero University of Utrecht and McGill University Paul Boersma University of Amsterdam Second International Conference on Contrast in Phonology Toronto, May 3, 2002
Introduction • Learning an L2 two-way contrast is problematic if it has an L1 three-way contrast as a starting point. • The initial state of L2 speech comprehension provides evidence of an intermediate perceptual level. • The perception of L2 learners improves during development. • L2 perceptual development need not affect L1 performance.
Case:the perception of front vowels by Dutch learners of Spanish
L1 and L2 production environments Dutch Spanish
Evidence for an intermediate discrete perception level • target-language /i/ associated with L1 /i/ • target-language /e/ identified with L1 // • (/I/ |i|: identification task reflects recognition)
L1 and L2 production environments Dutch Spanish
Perception modes • The model requires that L2 boundaries can shift without affecting L1 perception. • Therefore, we must assume separate perception grammars for L1 and L2 within every single speaker. • Is there independent evidence for such a distinction? Set up the two alleged modes by language-dependent priming, then compare L1 classification in the two modes.
Beginning Dutch learners of Spanish Mode: Dutch Spanish
Intermediate Dutch learners of Spanish Mode: Dutch Spanish
Advanced Dutch learners of Spanish Mode: Dutch Spanish
Bilingual Dutch-Spanish Mode: Dutch Spanish
Formalization: OT constraints • “an F1 of 200 Hz is not /a/” • “an F1 of 200 Hz is not /E/” • “an F1 of 200 Hz is not /I/” • “an F1 of 200 Hz is not /i/” • “an F1 of 450 Hz is not /a/” • “an F1 of 1000 Hz is not /a/” • ...
L1 perception if there’s a lexicon • Recognition phase undoes misperceptions.
How recognition mismatches change the rankings in the perception grammar
L1 computer simulation • Initial state: all constraints ranked equally high. • Learner hears 1000 tokens/month, drawn from the Dutch F1 distribution. Learner is also told (by recognition) which was the correct category. • Stochastic OT, evaluation noise 2.0. • Plasticity (size of the learning steps): starts at 10.0 (much larger than the evaluation noise); decreases by 3% every month; ends at 0.014 after 18 years. First fast, then accurate.
L2 computer simulation • Initial state: final state of L1. • Learner hears 500 tokens/month, drawn from the Spanish F1 distribution. Learner is also told (by recognition) which was the correct category (/A/, /E/, /i/; never /I/). • Stochastic OT, evaluation noise 2.0. • Plasticity (size of the learning steps): stays constant at 0.01 slow but accurate.