Preschoolers’ flexible use of talker information during word learning

• Preschoolers concurrently learned novel labels and talkers’ referent preferences.
• Children rapidly corrected looks and accuracy after misleading voice cues.
• Children may encode multiple person-referent mappings during word learning.
• Children use cues more flexibly than in previous work (Trueswell, Sekerina, Hill, & Logrip, 1999).

Previous research suggests that preschool-aged children use novel information about talkers’ preferences (e.g. favorite colors) to guide on-line language processing. But can children encode information about talkers while simultaneously learning new words, and if so, how is talker information encoded? In five experiments, children learned pairs of early-overlapping words (geeb, geege); a particular talker spoke each word. Across experiments, children learned labels for novel referents, showing an advantage for original-voice repetitions of words which appeared to stem mainly from semantic person-referent mappings (who liked what referent). Specifically, children looked to voice-matched referents when a talker asked for their own favorite (“I want to see the geege”) or when the liker was unspecified (“Point to the geege”), but they looked to voice-mismatched referents when a talker asked on behalf of the other talker (“Conor wants to see the geege”). Initial looks to voice-matched referents were flexibly corrected when later information became available (Anna saying “Find the geege for Conor”). Voice-matching looks vanished when talkers labeled the other talker’s favorite referent during learning, possibly because children had learned two conflicting person-referent mappings: Anna-likes-geeb vs. Anna-talks-about-geege. Results imply that children’s language input may be conditioned on talker context quite early in language learning.