Poster #52 - Partial Learning of Word Meaning from Referentially Ambiguous Naming Events

• In Event: Poster Session 13

Abstract

Word learners experience naming events that vary widely in their referential quality. How varying events contribute to learning is a matter of great debate. One possibility is that in acquiring words, learners rely primarily on events that are referentially transparent (Medina et al., 2011; Trueswell et al., 2013). Alternatively, learners might rely on a range of events for learning, including those that are referentially ambiguous (Smith & Yu, 2008; Yu & Smith, 2012). Central to this debate then is whether word meaning can be learned from referentially ambiguous naming events (Gleitman & Trueswell, 2020; Smith et al., 2014). The current study presents evidence suggesting that the answer may depend on what it means for a word to be “learned”.
This study revisits Medina and colleagues’ Human Simulation Paradigm (HSP) in which adult learners were presented multiple muted vignettes of parent-toddler interactions during which parents uttered the same target word (e.g., “car”; see Figure 1A). Medina and colleagues found that adults struggled to identify that target word when the vignettes were referentially ambiguous. This study asks whether referentially ambiguous events may provide participants with some knowledge of word meaning even when they fail to provide complete word knowledge.
Adult participants (N = 60) completed a learning phase where they were presented with picture book scenes that shared a common word (e.g., “dog”) in their text. All scenes were presented with text removed and were paired with a novel word (e.g., “MODI”). Importantly, each scene had previously been normed as referentially ambiguous (see Figure 1B). Following the learning phase (see Figure 1C), participants completed two tests: (A) a two-alternative forced choice test in which participants were presented with two new scenes and asked which was more likely to have contained the novel word, and (B) a free-response test in which participants were asked to guess the meaning of the novel word, just as in Medina and colleagues’ study.
As in Medina et al.’s study, many participants failed to identify the exact word meaning in the free response test (M = 56.5%, SD = 24.3%; see Figure 2A). Importantly however, participants’ performance in the forced-choice test was reliably above chance, even for words for which they provided incorrect guesses in the free response test, M = 0.66, SD = 0.18, t(58) = 7.15, p < .001 (see Figure 2B-C). These results suggest that although referentially ambiguous events may be insufficient to yield learning of exact word meaning, they equip learners with reliable partial knowledge about a word’s meaning. Exploratory analyses suggest that the nature of that partial knowledge matters; when participants’ free response errors were semantically related to the target word, they were more likely to succeed on the forced-choice test (see Figure 2D).
Understanding the kinds of input that shape word learning is central to both theories of the mechanisms of learning and interventions that seek to support it. The current work suggests that a better understanding of that input may depend on the output of learning that one seeks to explain.

Authors

• Nina Schoener, University of Connecticut
  Presenting Author

• Sara Johnson, University of Connecticut

• Sumarga H. Suanda, University of Connecticut