Can a tog be a dog? A meta-analysis of experimental factors influencing infants' mispronunciation sensitivity

• In Event: 1-041 - Big Ideas to Tackle Small Samples

Integrative Statement

During the second year of life, infants experience a vocabulary explosion during which they are not only learning the distinct combinations of sounds that serve as referents for distinct objects (e.g., cat versus hat) but also have to remain flexible enough in these representations in order to account for variability in the signal (e.g. speakers, accents). Two decades of research using the Intermodal Preferential Looking Procedure (IPLP; Golinkoff, Hirsh-Pasek, Cauley, & Gordon, 1987) have examined infants' recognition of correctly pronounced (dog) and mispronounced (tog) words. Experimental evidence suggests three potential possibilities for the developmental trajectory of mispronunciation sensitivity: a) decreasing with development (Trajectory 1; Mani & Plunkett, 2011); b) increasing with development (Trajectory 2; e.g. Altvater-Mackensen & Mani, 2013); and c) unmodulated by development (Trajectory 3; e.g. Swingley & Aslin, 2000). Yet, only Trajectory 1 (Best, 1994) and Trajectory 2 (Curtin & Werker, 2007) are supported by existing theoretical frameworks. We attempt to resolve these diverse possibilities using a meta-analytic approach. We furthermore examine how experimental factors may modulate infant mispronunciation sensitivity, such as whether mispronunciation sensitivity increases as more phonological features are changed (White & Morgan, 2008); and whether using an unfamiliar distractor image, and thus a possible referent, paired with an image of the target word may provide a more reliable estimate of mispronunciation sensitivity (Mani & Plunkett, 2011).

In a meta-analysis, we capture the developmental trajectory of mispronunciation sensitivity, focusing on experiments testing infants using IPLP. The final sample included results from 32 papers reporting on data from infants aged 6 to 31 months (2252 infants). Experimental factors were also coded, including number of phonological features changed (1, 2, 3) and distractor image familiarity (unfamiliar, familiar) .

Infants consistently recognized correctly pronounced (Hedges’ g = 0.91, SE = 0.12, p < .0001) and mispronounced (Hedges’ g = 0.25, SE = 0.06, p < .001) words. Effect size Hedges’ g was significantly greater for correct pronunciations compared to mispronunciations (Hedges’ g = 0.50, SE = 0.03, p < .0001), Neither recognition nor mispronunciation sensitivity were modulated by infant age. As the number of features changed in mispronounced words increased, the effect size of mispronunciation sensitivity increased (β = -0.31, SE = 0.03, p < .0001). Furthermore, mispronunciation sensitivity was greater in experiments using an unfamiliar compared to familiar distractor image (β = 0.18, SE = 0.09, p < .05).

These results suggest that although infants associate a mispronounced word with the target image, this is reduced in comparison to correct pronunciations, providing evidence of mispronunciation sensitivity. Mispronunciation sensitivity is not modulated by infant age, however, awarding support to Trajectory 3. Furthermore, experimental factors modulated the mispronunciation sensitivity effect, i.e. infants were sensitive to the number of features changed and an unfamiliar distractor image may be a more likely referent for a mispronounced label than a familiar distractor image with an unrelated known label. The results of this meta-analysis have important implications for the study and understanding of infants’ phono-lexical development as well as the design and interpretation of future mispronunciation sensitivity studies.

Authors

• Katie Von Holzen, University of Maryland
  Presenting Author

• Christina Bergmann, Max Planck Institute for Psycholinguistics
  Non-Presenting Author