Poster #14 - What Changes Over Development in Cross-Situational Word Learning Behaviour and How Do We Explain It?

• In Event: 3-206 - Poster Session 13
  In Poster Session: PS 13 Section - Attention, Learning, Memory

Integrative Statement

Studies show that children as young as 1-year-of-age can learn new word-object mappings in cross-situational word learning tasks (Smith & Yu, 2008). One-year olds are also sensitive to novelty and stimulus novelty can decrease learning in cross-situational word learning tasks (Smith & Yu, 2013). These results suggest that novelty attracts bottom-up visual attentional processes in ways that do not support but rather compete against top-down word-object mapping processes. On the other hand, adults can exploit such novelty effects to learn word-object mappings faster. How such developmental changes in the interaction between the top-down and bottom-up processes occur is unclear both empirically and mechanistically.

We employed both modelling and empirical approaches to investigate this question. We use a neutrally-grounded model of word learning that integrates bottom-up visual attentional dynamics with top-down word learning processes across multiple timescales. WOLVES (Word-Object Learning Via Visual Exploration in Space) implements real-time interactions between processes of attention, memory traces, and working memory to examine and measure in-the-moment visual exploratory behaviour and word learning processes as they unfold over time. Further, because we can specify these processes at different strengths in the model, we can simulate the cross-situational experimental paradigms from Smith & Yu (2008; 2013) with and without novelty effects for various age ranges. In parallel, we ran the same cross-situational experimental paradigms on 3-4 year old children to dissociate the mechanisms underlying word learning when compared to results from studies with 1-year-old infants.

The model successfully captures multiple findings in cross-situational tasks in one-year olds. The overall word learning performance and the visual behavioral measures such as looking times to objects, fixation counts and duration of fixations etc., of the model and those reported in empirical studies correspond well. Introducing strong local novelty effects as Smith & Yu’s (2013) task makes the learning of word-referent mappings more difficult, as observed in empirical data. Figure 1 shows less looking to the target in the novelty version of the task while Figure 2 shows fewer overall words learned by both the model and children.

Using the correspondence between our model and prior empirical data as a foundation, we are currently exploring the influence of novelty on older children’s performance in the cross-situational word learning task; including the possibility of better learning when novel stimuli are introduced. These data will then be captured in our model to better understand how novelty driven processes and word-object association processes interact in the task.

Our model is the first neural theory to show that novelty-driven attention competes with, rather than supports, cross-situational learning in one-year olds. Our ongoing experiments and model simulations inform the on-going debate between hypothesis testing and associative learning accounts of cross-situational learning (Smith, Suanda & yu, 2014). Our simulations indicate that factors including visual object recognition, memory trace consolidation, strength of working memories, top-down and bottom-up attentional influences operate in non-linearly complex ways to give rise to a diverse range of developmental trajectories.

Authors

• Larissa K Samuelson, University of East Anglia
  Presenting Author

• John Spencer, University of East Anglia
  Non-Presenting Author

• Laura Colosimo, University of East Anglia
  Non-Presenting Author

• Ajaz A Bhat, University of East Anglia
  Non-Presenting Author