The effects of perceptual training on speech production of Mandarin sandhi tones by tonal and non-tonal speakers

This study examined the effects of perceptual training on the acquisition of sandhi tones by both tonal (Cantonese) and non-tonal (English) speakers. In the pre- and post-training tasks, participants were presented with two monosyllables in each trial and were asked to produce them as disyllables. During perceptual training, they heard trials of disyllabic words with and without sandhi rule applications and were trained to identify the two types of disyllabic words. Based on functional data analyses, Cantonese speakers improved in building up more detailed and native-like allotone representations and in turn exhibited more native-like speech production in many cases. However, American learners showed improvement mainly in the offset f0 values or the turning point, indicating that they might have failed to pay attention to the shape of the entire tonal contour during perceptual training. For tonal speakers, the improvement of allotone production occurred on both real and wug words, indicating that the perceptual training may lead to more accurate production of tone sandhi rules regardless of syllable types. American learners, however, might need more training to further modify the allotone representation to perform more accurate sandhi tone application, especially in wug words. Finally, phonetic bias did not correctly predict the success of allotone acquisition since other factors such as the acoustic cues used in processing allotones during perceptual training might have affected the learning results.