• Corpus ID: 11640604

The Goldilocks Effect: Infants' preference for stimuli that are neither too predictable nor too surprising

@inproceedings{Kidd2010TheGE,
  title={The Goldilocks Effect: Infants' preference for stimuli that are neither too predictable nor too surprising},
  author={Celeste Kidd and Steven T. Piantadosi and Richard N. Aslin},
  year={2010}
}

• Celeste Kidd, S. Piantadosi, R. Aslin
• Published 2010
• Psychology

The Goldilocks Effect: Infants’ preference for stimuli that are neither too predictable nor too surprising Celeste Kidd ([e-mail]) Steven T. Piantadosi ([e-mail]) Brain & Cognitive Sciences, Meliora Hall Rochester, NY 14627 USA Brain & Cognitive Sciences, 43 Vassar Street Cambridge, MA 02139 USA Richard N. Aslin ([e-mail]) Brain & Cognitive Sciences, Meliora Hall Rochester, NY 14627 USA Abstract Even before birth, infants attend to the statistical properties of their sensory environments to learn about events in world. Tracking these statistics is crucial to mastery of visual, social, linguistic, and cognitive tasks. However, the degree to which their sampling follows prescriptions of rational statistical infer- ence is unclear. Do infants’ attentional preferences reflect ef- ficient information gathering? We investigated using an ideal observer model (a Markov Dirichlet-multinomial). We pre- dicted infants’ attention to sequential events would be moder- ated by information content. We tested infants (7-8 months) with 32 unique event sequences (objects popping out of boxes) on a Tobii eye-tracker. Each sequence continued until look- away. Controlling for other variables, we found infants were significantly more likely to look away at either highly informa- tive or uninformative events according to the model. This sug- gests infants allocate visual attention to maintain intermediate rates of information processing, avoiding committing cogni- tive resources to either overly predictable or surprising events. This “Goldilocks effect” may reflect a general strategy for ef- ficient learning from environmental statistics. Keywords: Statistical learning; statistical inference; ideal- ized learner; infant gaze behavior; infant methods; infant eye- tracking; Bayesian modeling; information theory; infant visual attention. Introduction Infants have a lot to learn in the first few years of life, and a limited set of resources with which to do it. The world is brimming with potential sources of information, but where among this spatiotemporal array of events should infants be- gin their learning? From birth, infants survey their sensory environments, sampling the visual data that surrounds them at the incredibly rapid pace of two or more fixations a second during 90% of their waking hours (Haith, 1980). This pro- cess, of surveying and sampling, provides infants with rich in- formation from which they can start to learn about the world. Previous empirical work has demonstrated that infants are able use the statistical properties of their environment in a di- verse array of learning tasks pertaining to sounds, words, peo- ple, shapes, and objects (Fiser & Aslin, 2002; Maye, Weiss, & Aslin, 2008; Saffran, Aslin, & Newport, 1996; Saffran, Johnson, Aslin, & Newport, 1999; Yu & Ballard, 2007) 1 . In many complex cognitive systems (e.g., object recognition, 1 The literature detailing these statistical learning abilities is so large, in fact, that if it were printed and stacked in a pile, it would be more than 140 infants tall (based on 161,000 unique articles cat- aloged by Google Scholar at time of publication and 26-inch mean height of 8-month-olds). language), infants obtain the representations of higher-level structures by tracking the low-level statistical coocurrences. For example, newborn infants must track the distribution of acoustical properties of speech sounds in their target lan- guage in order to infer its phonological categories (White, Peperkamp, Kirk, & Morgan, 2008). Researchers of visual, conceptual, and social learning find similar patterns. Recent technologies (e.g. eye-tracking, brain imaging techniques) have elucidated many of the mechanisms infants employ in building high-level structures from low-level environmental statistics. Though the topic has been of great interest to re- searchers, the mechanisms and representations infants em- ploy during the process of collecting environmental statistics are still not well understood. Amid the unbinned and unsorted masses of sensory data available in the world, an undirected search would be ineffi- cient. Infants have too many things to do—motor actions to program, words to learn, categories to form—to waste time. How then should the infant allocate her visual attention? Several researchers attempted to unify this work by iden- tifying an overarching stimulus feature that could generally account for all of infants’ preferences for various stimulus properties. Sokolov (1960) postulated that the primary driver of infants’ attention is stimulus novelty. Consistent with this theory, infants commonly prefer the novel stimuli in prefer- ential looking/listening tasks such as those use in the Fantz paradigm (Fantz, 1964), high-amplitude sucking procedure (Siqueland & DeLucia, 1969), and head-turn preference pro- cedure (Kemler Nelson et al., 1995) . The novelty account is also consistent with habituation behavior, during which infants’ attention to recurring stimuli decreases over time. However, the novelty hypothesis does not account for infants’ familiarity preferences in many preferential looking and lis- tening studies. Notable examples include infants’ affinity for their native languages and for faces, especially those of their mothers. Roder and others attempted to reconcile infants’ preference patterns by relating preference and processing load (Hunter & Ames, 1988; Roder, Bushnell, & Sasseville, 2000; Wag- ner & Sakovits, 1986). Roder suggested that the process of memory formation was responsible for preference. Under this theory, infants would be expected to exhibit a familiar- ity preference early in processing as they form a memory of