Learning to predict and predicting to learn: Before and beyond the syntactic bootstrapper

Abstract

Young children can exploit the syntactic context of a novel word to narrow down its probable meaning. This is syntactic bootstrapping. A learner that uses syntactic bootstrapping to foster lexical acquisition must first have identified the semantic information that a syntactic context provides. Based on the semantic seed hypothesis, children discover the semantic predictiveness of syntactic contexts by tracking the distribution of familiar words. We propose that these learning mechanisms relate to a larger cognitive model: the predictive processing framework. According to this model, we perceive and make sense of the world by constantly predicting what will happen next in a probabilistic fashion. We outline evidence that prediction operates within language acquisition, and show how this framework helps us understand the way lexical knowledge refines syntactic predictions and how syntactic knowledge refines predictions about novel words' meanings. The predictive processing framework entails that learners can adapt to recent information and update their linguistic model. Here we review some of the recent experimental work showing that the type of prediction preschool children make from a syntactic context can change when they are presented with convincing contrary evidence from recent input. We end by discussing some challenges of applying the predictive processing framework to syntactic bootstrapping and propose new avenues to investigate in future work.

Do Children (and Adults) Benefit From a Prediction Error Boost in One-Shot Word Learning?

“…We acknowledge that current data are still insufficient to precisely pinpoint the developmental onset of the effect, but clearly they do not provide strong support for theories that propose a fundamental role for prediction error in early word learning (Chang et al, 2006;Ramscar, 2021). By the age of 5, not only is word learning well underway, but children have also developed very sophisticated linguistic prediction and revision skills (Gambi, Jindal, et al, 2021;Havron et al, 2019;Babineau et al, 2022). Thus, they should clearly benefit from the prediction error boost if indeed the process of word learning is driven by prediction error.…”

“…In other words, larger prediction errors should have larger effects on learning. While not all theoretical accounts of error-based learning formulate this hypothesis explicitly (for a recent discussion, see Babineau et al, 2022), surprisal effects in structural priming studies are usually interpreted as supporting it. Structural priming refers to the tendency to re-use a sentence structure one has just heard (Messenger et al, 2022).…”

“…Thus, Reuter et al's findings are not entirely consistent with the hypothesis that larger prediction errors lead to better word learning. One possibility is that children sometimes fail to revise their expectations: Children who generate particularly strong predictions may not experience any prediction error, and instead associate the novel word to a familiar object (effectively, treating it as a synonym for the expected familiar word; Babineau et al, 2022).…”

Introducing Meta‐analysis in the Evaluation of Computational Models of Infant Language Development

“…The latent representations APC learns have been shown to be applicable to a range of speech tasks, including phone classification (Chung et al, 2019) and subword modeling (Feng, Żelasko, Moro-Velázquez, & Scharenborg, 2021; see also Yang et al (2021)). Central to modeling purposes, APC learning criterion aligns with the idea of predictive processing in human perception (e.g., Friston, 2010;Rao & Ballard, 1999;Babineau, Havron, Dautriche, de Carvalho, & Christophe, 2022) and a similar learning mechanism can thereby be assumed to be available to any mammalian learner. More specifically, APC learning is based on minimizing prediction error of future speech observations, given access to past and current speech stimulus.…”

There might be more to syntactic bootstrapping than being pragmatic: A look at grammatical person and prosody in naturalistic child-directed speech

“…For us, syntactic bootstrapping, pragmatic bootstrapping, prosodic bootstrapping and bootstrapping word learning from context (such as in the case of second-person sentences), can all be combined within a larger framework. According to this framework, young children learn by attempting to predict which word, semantic field or linguistic structure comes next (Babineau et al, 2022;Rabagliati et al, 2016). This larger framework views the child's language-learning task as figuring out the intended meaning of an interlocutor using prediction.…”