Language as a cue for talker identification in 3- to 5-year-old children

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

Integrative Statement

Being able to identify who is speaking provides useful information for comprehending spoken language, as it tells us what a speaker may be more likely to talk about. Yet, only a small number of studies have investigated what vocal features are distinguishable by children for talker identification (Creel & Jimenez, 2012). One relatively-untested idea is that children associate individuals not just with vocal qualities but also with particular languages. Based on previous findings suggesting that young children use extralinguistic information (i.e., talker age/gender) in speaker recognition and comprehension (Borovsky & Creel, 2014), we ask here whether the language someone speaks can cue talker recognition, and more specifically which linguistic features allow recognition.

In Experiment 1 we tested whether English monolingual 3- to 5-year-olds (n=32) associate voice or language characteristics with individual characters in an eye-tracked talker-voice association paradigm (Allopenna, Magnuson, & Tanenhaus, 1998). In the learning phase, children saw two characters (8 trials, 4 per character). Each spoke sentences in a different language (Spanish or English). The following test phase contained 16 two-alternative trials: both characters appeared, one spoke a sentence, and children were asked to point to the character who had spoken. To assess whether children used voice characteristics vs. language characteristics, the speakers switched languages after the first 8 test trials: if speaker A originally spoke English, they switched to Spanish. If children are associating voice characteristics, not languages, with each character, they should select the character associated with a particular voice. Eye-movements and pointing accuracy were recorded throughout testing.

For these analyses, selection of the same speaker (even if they changed language) was counted as correct. Mean accuracy was high before the talker switch (88%), but low post-switch (13%) (Figure 1). Eye-tracking data patterned similarly, showing a significant main effect of Pre-Switch/Post-Switch in both time windows, F(1,27)=7.11, p=0.013, F(1,27)=64.67, p<0.001. Thus, children pointed and looked towards the character associated with a given language, not voice, suggesting that language is more salient than voice characteristics in children’s talker identification.

It is possible that in order for children to use language as a cue for talker identification, listeners must be familiar with at least one talker’s language. In Experiment 2, we tested whether English monolingual 3- to 5-year-olds (n=32) can recognize talkers of two unknown, dissimilar languages (Spanish and Mandarin). The same procedure described above was used, however, there was no voice-language switch. Pointing accuracy and proportion of looks to the target character were analyzed. Unlike Experiment 1, mean accuracy and looks to the target were not significantly above chance, p>.05 (Figure 2). These data suggest that children cannot differentiate between talkers of unknown languages.

These experiments begin to show that language is a salient cue to talker identity from as early as three years of age, as long as the child knows one of the two languages being spoken. Ongoing studies explore speech-sound features underlying talker recognition, as well as how children of different language backgrounds differ in their use of such information.

Authors

• Reina Mizrahi, University of California San Diego
  Presenting Author

• Sarah Creel, University of California San Diego
  Non-Presenting Author