Moving beyond “nouns in the lab”: Using naturalistic data to understand why infants’ first words include uh-oh and hi

• In Event: Reorienting the study of language learning using naturalistic data

Abstract

Why do infants learn some words earlier than others? Primarily rooted in investigations of noun and verb acquisition, current theories posit learning mechanisms dependent on stable visual referents or clearly segmentable event boundaries (Golinkoff & Hirsh-Pasek, 2008; Medina et al., 2011; Smith & Yu, 2008) and point to concreteness and imageability as the best predictors of early learnability (McDonough et al., 2011; Swingley & Humphrey, 2018). However, abstract words (e.g., hi, more, uh-oh; Fig.1) are universally among the first-acquired (Frank et al., 2016). To reach a comprehensive account of early word learning, researchers must resolve this contradiction between proposed learning mechanisms and the composition of infants’ early vocabularies. We combined a behavioral experiment with naturalistic observational research to explore how infants learn and represent abstract, high-frequency, routine-based words—hereafter ‘everyday words’.

Experiment: Using the looking-while-listening procedure (Fernald et al., 2008), we tested whether 10-16-month-old infants (N=33, M=13.2m) associate each of 12 everyday words with a corresponding image (e.g., hi: person waving, uh-oh: overturned cup; Fig.1). Following established traditions, images were chosen based on intuition, and parent ratings indicated that referents matched infants’ experience (M=3.11/4 for infants with reported target comprehension, 2.88/4 overall). However, infants did not show above-chance looking to target images [t(32)=1.41, p=0.17]. Thus, this conventional measure failed to capture infants’ comprehension, despite reports that, on average, 60.5% of 12-month-olds understand these words (Frank et al., 2016).

Corpus-based Observational Analysis: Using eye-tracking to investigate early word knowledge forces researchers to intuit imageable, common visual referents for the tested words. Yet, we do not know if the referents of everyday words are imageable or consistent (within or across children). Thus, we used longitudinal, in-home recordings of mother-infant dyads (N=5, range=11-24m, Demuth et al., 2006) to examine infants’ real-world input surrounding these words. We characterized 11,856 referent scenes according to whether they matched the exact objects/human actions from our experimental stimuli, or the same general situation (e.g., uh-oh: fallen cup=‘visual match’, anything dropped=‘situational match’). Visual matches were infrequent (M=6.9%). However, situational matches were common (M=44.8%; Fig.2) and even mapped onto the top usage for 9/12 items, suggesting that infants’ failure in the experiment was not necessarily due to improbable scenes. Critically, though, referent scenes were highly variable both within and across children, with words occurring in 34-1411 unique contexts (M=342.1). The most common object/action referent co-occurred with only 1.4-19.7% of tokens (M=8.2%), but the top situational context, broadly defined (e.g., uh-oh: any object falling), accounted for 12.4-93.2% (M=45.8%), indicating some stability in how these words are used.

These results demonstrate the incongruity between widely-discussed word learning mechanisms, which rely on referent concreteness and imageability, and infants’ early acquisition of everyday words that do not reliably co-occur with consistent visual referents. Current approaches to testing and understanding word learning assume common visual input surrounding a label, both within and across children, and thus may not capture infants’ emerging representations of these highly familiar abstract words. Future studies must continue to leverage naturalistic input to investigate non-canonical features supportive of early word learning.

Authors

• Kennedy Casey, Princeton University
  Presenting Author

• Christine Potter, University of Texas at El Paso
  Non-Presenting Author

• Erica Wojcik, Skidmore College
  Non-Presenting Author

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