Individual variability in the distributional learning of L2 lexical tone

This study tested whether successful learners of an artificial tone language exhibit sensitivity to varying degrees of tonal informativeness, which has previously been shown to effect spoken word recognition in native Mandarin speakers. Twenty naïve listeners, whose L1 is American English, learned an artificial language in which each visual nonce symbol was arbitrarily associated with a Mandarin-like monosyllable and tone. The stimuli were designed to mimic Mandarin’s uneven distribution of syllable-tone combinations; syllable frequency and the likelihood of a syllable co-occurring with a particular tone were manipulated across 4 days of training. The results showed that successful learners (those whose perception and production accuracy were consistently above the daily median) most accurately perceived and produced frequent syllables with probable tones and infrequent syllables with probable tones. Successful learners were least accurate in perceiving and producing infrequent syllables with least probable tones. Learners whose daily accuracy was below the median showed no such sensitivity to syllable-conditioned tonal probability. This finding supports the claim that L2 learners can be sensitive to statistical information available from novel input, and further demonstrates that statistical learning takes place even from an early stage of acquisition in successful L2 learners.