Poster #41 - Apples to Oranges: The Effects of Semantic Context on Referential Ambiguity

• In Event: Poster Session 05

Abstract

When an infant hears a new word (e.g., “shoe”), they must figure out its correct referent. The Human Simulation Paradigm (HSP) illustrates the difficulty of this seemingly simple task (Gillette et al., 1999). In the HSP, naïve adults watch muted vignettes of parent-infant interactions and guess a “mystery word” that parents uttered in those vignettes (see Figure 1A). Adults struggle mightily in the HSP, shining a light on the challenge that infants must face at this task (see Cartmill et al., 2013; Medina et al., 2011). Interestingly, in standard HSP studies, the to-be- guessed mystery words are semantically unrelated (e.g., “ball”, “shoe”). Critically however, infant-directed speech has a great deal of semantic structure to it, with semantically-related words (e.g., “socks”, “shoe”) often occurring adjacently (see Tamis-LeMonda et al. 2019). Here, we modified the standard HSP to examine whether participants’ identification of mystery words improved when those words formed semantically-coherent sets, as they do in infants’ experience.

In an initial norming study, we created a picture-book version of the HSP where participants saw textless scenes from picture books and guessed a word present in the original text (see Fig 1B). From this study, we identified six referentially ambiguous scenes, defined as scenes in which fewer than 25% of participants guessed the mystery word (see Medina et al., 2011). In the main study we asked whether participants’ guesses for referentially ambiguous scenes would improve when they were presented adjacent to scenes depicting the relevant semantic category.

36 adult participants were assigned to either the semantically-structured or semantically-random condition. Semantically-structured participants viewed seven scenes (“cue trials”) that clearly depicted words in the same semantic category as the referentially ambiguous scene (“target trial”), which was presented immediately following the cue trials (see Figure 1C). Semantically- random participants were also presented seven cue trials followed by the same target trial. However, cue trial scenes did not depict words belonging to the target semantic category (see Figure 1C).

Participants in the semantically-structured condition were more likely to guess the mystery word than participants in the semantically-random condition, t(34) = 2.34, p < .05 (see Figure 2A). Furthermore, the percentage of participants that outperformed norming levels was greater in the semantically-structured condition (61%) than in the semantically-random condition (28%), X2(36) = 4.05, p < .05 (see Figure 2B). In an exploratory analysis of the mechanism underlying these effects, participants who consistently offered category words in the cue trials were more accurate in the target trial (see Figure 2C-D).

The HSP is a compelling demonstration of the challenge of determining word meaning in the moment. The current data highlight that the in-the-moment challenge is reduced when words are experienced in a semantically-coherent manner. Fortunately for young learners, they experience words in this way (see Roy et al., 2015; Tamis-LeMonda et al., 2019). The implications of this work for the measurement of input quality and for the mechanisms of word learning will be explored in more depth.

Authors

• Marissa Renee Ciccarini, University of Connecticut
  Presenting Author

• Sumarga H. Suanda, University of Connecticut