How who is talking matters as much as what they say to infant language learners

Human vocalizations contain both voice characteristics that convey who is talking and sophisticated linguistic structure. Inter-talker variation in voice characteristics is traditionally seen as posing a challenge for infant language learners, who must disregard this variation when the task is to detect talkers' shared linguistic conventions. However, talkers often differ markedly in their pronunciation, vocabulary, and grammar. This is true even in monolingual environments, given factors like gender, dialect, and proficiency. We therefore asked whether infants treat the voice characteristics distinguishing talkers as a cue for learning linguistic conventions that one talker may follow more closely than another. Supporting this previously untested hypothesis, 12-month-olds did not freely combine two talkers' sentences distinguished by voice to more robustly learn the talkers' shared grammar rules. Rather, they used this voice information to learn rules to which only one talker adhered, a finding replicated in same-aged infants with greater second language exposure. Both language groups generalized the rules to novel sentences produced by a novel talker. Voice characteristics can thus help infants learn and generalize talker-dependent linguistic structure, which pervades natural language. Results are interpreted in light of theories linking language learning with voice perception.