Infants' selective use of speaker cues across contexts

• In Event: 3-235 - Contending with talker variability across development

Integrative Statement

To navigate complex language environments, infants must find the most informative cues to extract important information. In monolingual environments where all speakers use one language, infants learn to ignore speakers and instead prioritize other cues, such as sound-based regularities (Houston & Jusczyk, 2000; Thiessen & Saffran, 2007). But in multilingual environments, there are systematic, learnable relations between speakers and the input they provide (Grosjean, 2001). It has been proposed that bilingual infants might benefit from hearing languages spoken by separate people (Ronjat, 1913), as speakers could be used to segregate different channels of information. We tested whether infants exploit speaker differences to detect the structure of incoming speech.

Study 1 tested whether infants better learn linguistic structures separated by speaker. Monolingual 9-10-month-olds heard two streams of speech produced by a male and female talker. 50% of the input followed a learnable pattern (Target stream); 50% had no learnable structure (Non-target stream). Infants were assigned to either the Consistent condition, where each stream was produced by one speaker, or Variable condition, where each speaker produced half of each stream (Fig.1). Learning of the Target regularities was tested using the Headturn Procedure.

In Study 1a, the Target and Non-target stream used entirely non-overlapping sounds. Infants (n=40) listened reliably longer to Familiar items that maintained the Target regularity vs. Unfamiliar items that violated that pattern [F(1,38)=13.05, p<.001], but there was no interaction with speaker condition [F(1,38)=.30, p=.59], suggesting when there was a phonological distinction between the streams, infants’ learning was unaffected by correlated speaker cues.

In Study 1b, the two streams used an identical set of syllables. Infants (n=40) failed to differentiate Familiar vs. Unfamiliar items [F(1,38)=.68, p=.42], and there was no interaction with condition [F(1,38)=.32, p=.58]. Even though speakers provided a perfectly correlated cue, infants showed no evidence of using speakers to segregate their learning. While infants in Study 1a overcame speaker differences to detect underlying structure, speaker cues alone did not support learning in this artificial language environment. It could be that with richer, naturalistic stimuli, infants might attend to the relation between speakers and linguistic input.

Study 2 examined infants’ ability to form associations between speakers and real languages by testing whether infants detect changes in the language used by individual speakers. Monolingual 9-10-month-olds (n=20) were familiarized with two bilingual talkers (one male, one female). One speaker (e.g., male) always used English; the other (female) always used Arabic, a novel language. After being habituated to these pairings, infants were tested on Familiar trials that maintained the speaker/language pairings vs. Novel trials with switched pairings (Fig.2). Infants listened marginally longer to Novel trials [t(19)=1.50, p=.07], suggesting they track the language used by each speaker in natural contexts. Ongoing studies are testing whether younger infants, for whom indexical properties are more salient (Houston & Jusczyk, 2000), can attend to and benefit from such associations.

Together, these studies suggest infants dynamically adjust their attention to speaker differences as they attempt to exploit the reliable cues in their auditory and social environments.

Authors

• Christine Potter, Princeton University
  Presenting Author

• Casey Lew-Williams, Princeton University
  Non-Presenting Author