Poster #77 - Learning Words From Referentially Ambiguous Naming Events: Evidence From a New Artificial Word Learning Paradigm

• In Event: Poster Session 14

Abstract

Most scholars would agree that when toddlers hear new words (e.g., “ball”) in contexts where their referents are transparent, word learning is optimized. There’s less agreement, however, on whether toddlers also learn when words are heard in referentially ambiguous contexts (Gleitman & Trueswell, 2020; Zhang & Yu, 2021). In one notable study, Medina and colleagues (2011) found that even adults struggle to learn words when they are presented with vignettes of referentially ambiguous naming events. Using a new artificial word learning paradigm, two experiments present evidence that learners’ ability to acquire word meaning from referentially ambiguous events may depend on task dynamics, the type of ambiguity, and how learning is defined.

Our experimental design (see Fig 1) combines elements of Medina and colleagues’ (2011) Human Simulation Paradigm and Posner and Keele’s (1968) Prototype Distortion Task. Adult participants saw scenes from picture books (all text removed) that either did (target trials) or did not (distractor trials) contain a target word (e.g., “dog”) in its original text. They then guessed whether they thought the scene contained the target word. Because the word’s identity was not known, participants began the study guessing randomly. Participants did however receive feedback after each guess, allowing them to learn across trials. Following 40 randomized categorization trials (20 targets, 20 distractors), participants were asked for the target word’s identity.

In both experiments, all target scenes were previously normed and determined to be referentially ambiguous. In Experiment 1 (n = 16), all target scenes contained the referent somewhere in the scene. In Experiment 2 (n = 16), 10% of target scenes were “absent referent” events. Of interest was whether participants learned the target word, how learning differed across the experiments, and whether participants successfully categorized scenes even when they did not successfully identify the target word.

In contrast to Medina and colleagues’ results, participants readily learned from referentially ambiguous events when referents were reliably present (i.e., Experiment 1). Because several design features distinguished the current experiments from Medina and colleagues’ (e.g., the nature of the stimuli, the quantity of naming events, task demands, etc.), it is unclear what underlies the differing results. These results do however suggest that there are contexts in which referentially ambiguous events lead to learning (Yurovsky et al., 2013; Zhang & Yu, 2021). Separately, these results underscore just how much absent referent events negatively impact learning. The simple addition of two absent referent events was enough to cut word learning in half (see Figure 2A-B). Finally, these results highlight the importance of “partial learning” in conclusions about the input that lead to learning. Across both experiments, even when participants failed to identify the target word’s precise meaning, they nonetheless distinguished events that did and did not contain the target word (see Figure 2C-D).

How different naming events contribute to word learning is a matter of great interest and debate. The current results highlight that partial, if not precise, word learning may be possible despite referential ambiguity of naming events.

Authors

• Julie-Ann Williams, University of of Connecticut
  Presenting Author

• Nina Schoener, University of Connecticut

• Bethany Stoddard

• Sumarga H. Suanda, University of Connecticut