Statistical word learning at scale: the baby's view is better

Abstract

A key question in early word learning is how children cope with the uncertainty in natural naming events. One potential mechanism for uncertainty reduction is cross-situational word learning – tracking word/object co-occurrence statistics across naming events. But empirical and computational analyses of cross-situational learning have made strong assumptions about the nature of naming event ambiguity, assumptions that have been challenged by recent analyses of natural naming events. This paper shows that learning from ambiguous natural naming events depends on perspective. Natural naming events from parent–child interactions were recorded from both a third-person tripod-mounted camera and from a head-mounted camera that produced a ‘child’s-eye’ view. Following the human simulation paradigm, adults were asked to learn artificial language labels by integrating across the most ambiguous of these naming events. Significant learning was found only from the child’s perspective, pointing to the importance of considering statistical learning from an embodied perspective.

Using Spoken Language Benchmarks to Characterize the Expressive Language Skills of Young Children With Autism Spectrum Disorders

“…However, this difference only emerged when children were grouped according to one aspect of their expressive language, phonology or vocabulary/grammar. Some researchers suggest that early language learning takes place during unstructured social interactions (e.g., Hoff, 2003) during which children must learn to selectively allocate their attention (Posner & Petersen, 1990; for a contrasting view, see Yurovsky, Smith, & Yu, 2013). Thus, children who respond to their communication partner's bids for joint attention are more likely to allocate their attention to relevant information that supports language learning.…”

Nature and origins of the lexicon in 6-mo-olds

“…Finally, we examined referential transparency (i.e., whether parents talk about observable referents), which is suggested to facilitate language learning (33)(34)(35). Notably, previous research examining why infants learn common nonnouns (e.g., "hi," "eat") later than common nouns found that visual copresence varied across word class (36).…”

Gaze and Cross-situational Word Learning

“…Using the HSP, Medina, Snedeker, Trueswell, and Gleitman (2011) found that approximately 90% of learning episodes were ambiguous (< 33% accuracy) and only 7% were relatively unambiguous (> 50% accuracy). In contrast, Yurovsky, Smith, and Yu (2013) found a higher proportion of clear naming events, with approximately 30% being unambiguous (> 90% accuracy). Consistent with this finding, Cartmill, Armstrong, Gleitman, Goldin-Meadow, Medina, and Trueswell (2013) showed that the proportion of unambiguous naming episodes varies across parent-child dyads, with some parents rarely providing highly informative contexts and others' doing so relatively more often.…”

“…Moreover, studies outside the domain of language acquisition have shown that the presence of social cues: (a) produce better spatial learning of 2 The differences in the estimates of referential uncertainty in these studies could be driven by the different sampling procedures used to select naming events for the HSP. Yurovsky, Smith, and Yu (2013) sampled utterances for which the parent labeled a co-present object, whereas Medina, Snedeker, Trueswell, et al (2011) randomly sampled any utterances containing concrete nouns. Regardless of these differences, the key point here is that variability in referential uncertainty across naming events exists and thus could alter the representations underlying cross-situational learning.…”