Cross-language Talker Identification

Two groups of monolingual, native English-speaking listeners were trained to identify the voices of ten German-English bilingual talkers. One group of listeners learned to identify the voices from English stimuli only, while the other group learned to identify the talkers from German stimuli only. After four days of training, both groups of listeners were asked to identify the same talkers from novel stimuli in both the language they had been trained on and the language they had not heard during training. No differences were observed in the amount of improvement in talker identification accuracy made by the two groups of listeners during training. In testing generalization across languages, however, the English-trained listeners performed significantly worse on German stimuli than they did on English stimuli, while the German-trained listeners identified talkers just as well from English stimuli as they did from German stimuli. This pattern of generalization across languages suggests that some of the indexical properties of speech are language-specific, while others are language-independent. The Englishtrained listeners apparently learned to identify talkers by relying on language-specific indexical information, while the German-trained listeners learned to identify talkers through language-independent indexical information. This pattern of results suggests that listeners may follow a mandatory perceptual strategy whereby they make use of language-specific indexical information when they can understand the language that is being spoken; otherwise, they learn to identify voices on the basis of languageindependent information alone. This perceptual tendency may result from the influence of automatic linguistic processing on listener performance in a talker identification task that requires conscious control.