Predict and redirect: Prediction errors support children’s word learning.

Abstract

According to prediction-based learning theories, erroneous predictions support learning. However, empirical evidence for a relation between prediction error and children's language learning is currently lacking. Here we investigated whether and how prediction errors influence children's learning of novel words. We hypothesized that word learning would vary as a function of 2 factors: the extent to which children generate predictions, and the extent to which children redirect attention in response to errors. Children were tested in a novel word learning task, which used eye tracking to measure (a) real-time semantic predictions to familiar referents, (b) attention redirection following prediction errors, and (c) learning of novel referents. Results indicated that predictions and prediction errors interdependently supported novel word learning, via children's efficient redirection of attention. This study provides a developmental evaluation of prediction-based theories and suggests that erroneous predictions play a mechanistic role in children's language learning.

The relation between preschoolers' vocabulary development and their ability to predict and recognize words.

“…When primed to interpret an ambiguous frame (e.g., French la petite) as a noun (i.e., "the small one" vs. an adjective: "the small"), children learned action meanings for novel words inserted after the frame (la petite dase), presumably because they predicted that a verb would follow the noun (Havron, de Carvalho, Fiévet, & Christophe, 2019). Further, 3-to-5 year olds' ability to reorient after an incorrect prediction correlates with their skill at learning novel words (Reuter, Borovsky, & Lew-Wlliams, 2019). In an eye-tracking task, children heard sentences like Yummy, let's eat soup!…”

“…First, much of the evidence is consistent with both accounts of how prediction facilitates learning: For example, the fact that structural predictions shape children's word learning (Havron et al, 2019) can be explained both by models in which prediction affects learning directly, via the computation of error signals, and by models in which it affects learning indirectly, because it facilitates fluent language processing and ambiguity resolution. Similarly, the finding that children's ability to reorient after an incorrect prediction is important for word learning (Reuter et al, 2019) could be explained in different ways: It could indicate a direct causal relation between error-revision and learning, or it could be that general cognitive ability means that children who are stronger learners are also better at revising incorrect predictions.…”

“…First, by including neutral, non-predictive sentences (e.g., Now, Craig is looking for the bone) we could measure the efficacy of children's language processing by capturing the speed with which they recognize spoken words without contextual facilitation (Fernald et al, 2006). Second, by varying the final word heard, we could measure children's responses to errors of prediction, capturing the degree to which they can quickly update their comprehension when their predictions are incorrect (Reuter et al, 2019). In particular, we compared word recognition times following neutral sentence contexts, when the final word was no more or less predictable than other options, to word recognition times when the final word was less predictable than a competitor, e.g., comparing recognition of slippers in Now, Craig is looking for the slippers (a neutral context), to Alfie's dog likes to chew on the slippers, where the competitor bone is more predictable than slippers.…”

“…To address this question, we first conducted grouplevel analyses to test whether recognition is indeed slower, in children, following a disconfirmed prediction than when no prediction is disconfirmed. We then assessed how a measure of revision skill ("predict-and-redirect", after Reuter et al, 2019) relates to age and vocabulary.…”

“…The three skills were quantified using the same summary statistics as in the cross-sectional analyses. We captured prediction skill using the combined measure -i.e., through children's preference for predictable pictures minus the dispreference for unpredictable pictures (see Supplement, §7, for evidence that neither the preference nor the dispreference measure alone were strongly predictive of changes in vocabulary size); we captured revision skill thought the "predictand-redirect" measure (Reuter et al, 2019), and finally we captured processing speed using the average timing of the first fixation to the named picture during the recognition window (e.g., Fernald et al, 2006).…”