The effects of perceptual salience on cross-situational word learning behaviour and mechanisms

Abstract

CSWL has been explained by models of associative learning which assume that learners implicitly accumulate statistical evidence about multiple referents for each word until one word-referent pairing becomes dominant, and by models of hypothesis-testing which assume that learners explicitly track a single hypothesised word-referent pairing across learning instances, and only abandon this conjecture if proven incorrect on subsequent instances. What remains unexamined are the effects of bottom-up auditory salience and visual salience on adult CSWL performance, and on the learning mechanism adopted during learning, despite salience being a well-known predictor of enhanced attention and learning. To examine this, we manipulated auditory salience by presenting words in adult-directed speech (ADS-only), infant-directed speech (IDS-only), which contains varied prosodic features relative to ADS, such as higher and more variable pitch, or half in ADS and half in IDS (Mixed condition). Since the ADS-only condition approximates the register adults typically hear in everyday situations, we compared accuracy in this condition to when all words are salient (IDS-only) and the effect of relative salience with a CSWL session (Mixed). Accuracy was above chance in all conditions, and there was no evidence that accuracy significantly differed between the conditions, nor was there evidence that accuracy in the Mixed condition differed depending on the speech register assigned to the target word or the word associated with the distractor object. We then examined the effect of visual salience on CSWL by comparing differences in accuracy across a Static-control condition in which all objects remained stationary, an Object Movement condition in which half the objects were stationary while the remaining half moved, and a Cross-Modal Salience condition in which labels of static objects were produced in ADS while labels of moving objects were produced in IDS. Again, accuracy was above chance in all conditions, and there was no evidence of a difference in accuracy between the conditions. In both the Object Movement and Cross-Modal Salience condition, participants were more accurate when the target object was moving than stationary, and were better at identifying the target word’s object in the presence of a stationary distractor than a moving distractor. Lastly, we investigated which learning mechanism best represents the data collected by examining the effects of preceding trials on performance, and whether exposure to immediate relative salience influences the learning mechanism adopted during CSWL. Adults were more accurate at identifying the target word’s referent after guessing correctly than incorrectly on the previous testing instance in which the same word occurred, and when the target appeared in immediate succession than when interleaved across two or more instances. They were also more accurate on immediate and interleaved instances containing a salient target than a nonsalient target. Overall, the findings suggest that while bottom-up salience may influence attention and single referent tracking during CSWL, it may not necessarily promote better CSWL performance relative to the control. Instead, our findings suggest that visual salience may hinder word learning, with object movement distracting learners from forming correct word-referent associations in ambiguous situations.

Date of Award	2023
Original language	English
Awarding Institution	Western Sydney University
Supervisor	Gabrielle Weidemann (Supervisor)