An associative model of adaptive inference for learning word–referent mappings

Abstract

People can learn word-referent pairs over a short series of individually ambiguous situations containing multiple words and referents (Yu & Smith, 2007, Cognition 106: 1558-1568. Cross-situational statistical learning relies on the repeated co-occurrence of words with their intended referents, but simple co-occurrence counts cannot explain the findings. Mutual exclusivity (ME: an assumption of oneto-one mappings) can reduce ambiguity by leveraging prior experience to restrict the number of word-referent pairings considered but can also block learning of non-one-to-one mappings. The present study first trained learners on one-toone mappings with varying numbers of repetitions. In late training, a new set of word-referent pairs were introduced alongside pretrained pairs; each pretrained pair consistently appeared with a new pair. Results indicate that (1) learners quickly infer new pairs in late training on the basis of their knowledge of pretrained pairs, exhibiting ME; and (2) learners also adaptively relax the ME bias and learn two-to-two mappings involving both pretrained and new words and objects. We present an associative model that accounts for both results using competing familiarity and uncertainty biases.

The company that words keep: comparing the statistical structure of child- versus adult-directed language

“…For example, increased contextual diversity has been shown to facilitate word segmentation (e.g. Hayes & Clark, 1970;Newman, 2008 ;, artificial language learning (Gillette, Gleitman, Gleitman & Lederer, 1999 ;Kachergis, Yu & Shiffrin, 2012;Plaut & Kello, 1999 ;Recchia, Johns & Jones, 2008 ;Yu & Smith, 2007), and rapid and diverse vocabulary acquisition (e.g. Hoff & Naigles, 2002;Hurtado, Marchman & Fernald, 2008 ;Huttenlocher, Haight, Bryk, Seltzer & Lyons, 1991 ;Naigles & Hoff-Ginsberg, 1998 ;Rowe, 2008).…”

Cross-situational word learning is better modeled by associations than hypotheses

“…This model does not use episodic memory of prior contexts nor multiple hypotheses per word to learn meanings. For comparison, we use a recent associative model of word learning that has competing biases for familiar pairings and for uncertain stimuli, but that approximately stores all co-occurrences on each trial with noisy retrieval [6]. In the following, we describe the models, fit them to the experiment, and determine whether they can both reproduce the human learning trajectories.…”

“…We fixed α = 0.97 for one experiment, and λ = 5.0 for another. As a learning rate, χ is a function of both the amount of time and the number of possible pairings on a trial, and this experiment differs on both dimensions from that in [6], χ was allowed to vary freely.…”

“…The single-hypothesis model has two free parameters: retrieval failure probability f and storage probability s. The associative model has three free parameters: learning rate χ, uncertainty vs. familiarity scalar λ, and decay parameter α. To equate the models' number of free parameters, we fixed the value of one parameter to the best-fitting value found in [6]. We fixed α = 0.97 for one experiment, and λ = 5.0 for another.…”

“…At issue in this study is whether learners store in memory multiple candidate meanings of each word (e.g., bosa-{o 1 , o 2 } and manu-{o 1 , o 2 })-albeit noisily, or instead store only a single hypothesis per word (e.g., bosa-o 1 and manu-o 2 ). Although biases such as mutual exclusivity-a preference for having one-to-one word-object mappings-are observed in two-year-olds [4,5] as well as adults [6], these learning biases can be explained using an associative model with competing biases for prior knowledge and uncertain (e.g., novel) stimuli [7,6]. This model comes from a line of research suggesting that evidence for multiple word meanings is accumulated incrementally across several situations [3,8].…”