U.S. and Chinese Preschooler’s Generalization of Statistically Learned Preferences, Social Norms, Functions, and Labels

• In Event: The Statistics of Group Cognition: New Evidence for Children’s Statistical Learning of Social Groups

Abstract

Children’s learning about social groups relies in part on the ability generalize the actions of one individual to make inferences about other members of a social group. To do this, children have to distinguish between actions that are idiosyncratic, such as an individual’s preferences (Buresh & Woodward, 2007) from actions that are shared by group members, such as linguistic conventions, object functions, social norms (Henderson & Graham, 2005; Wohlgelernter et al., 2010; Kalish, 2012; Diesendruck et al., 2015). In two studies, we investigate the role of statistical evidence (Xu & Tenenbaum, 2007; Kushnir et al., 2010) in children’s generalizations of these different domains. We also explore the role of culture by sampling two different cultural communities: Children in a northeastern U.S. town and children in Beijing and Shanghai, China.

In Study 1, 3- to 5-year-old children (U.S.: N=72, Mage = 4.06, SDage = 0.73, and China: N=87, Mage = 4.10, SDage = 0.79) saw 3 trials (within participants, counterbalanced) of a puppet selecting 5 objects out of a box that contained either a minority (20%) or majority (80%) of the target object (between participants). Preference trial selections were framed as “one he likes”, Function trial selections were framed as (and demonstrated to) “make a toy light up,” and Label trial selections were “called a toma.” Children were then asked to pick one of the three objects to give to a new puppet: the selected “target” object, the alterative unselected object, or a novel object. This generalization question was framed to correspond to the original action – give the new puppet one that “he likes”, one that “makes the toy go,” or one that is “called a toma.”

Across cultures, children generalized Labels (90%) and Functions (77%) more often than chance to the new puppet (ps < .0001, see Fig. 1). Children didn’t generalize the first puppet’s preference to the new puppet (24%, p < .0001). The domain contrasts were significant (Function/Preference: t(451) = 7.03, p < .0001; Label/Preference: t(451) = 7.77, p < .0001). There was an interaction of statistical learning and culture: In the Majority Sampling Condition only, Chinese children’s generalization of functions were more restrictive than U.S. children (t(451) = -2.56, p = .011).

Study 2 (ongoing) explores the culture effect further with two between-participant conditions, one contrasts generalization of preferences and functions (a property of people vs a property of objects) and another contrasts generalization of preferences and social norms (both are properties of people). We used identical stimuli (from Study 1), but different framing between conditions. Preliminary results (U.S.: N = 26, Mage = 4.15, SDage = 0.78; China: N = 61, Mage = 4.52, SDage = 0.59) echo the contrast found between preferences and functions in Study 1. When choices were framed as “the game’s rules” rather than the object’s functions, children’s generalizations were more sensitive to the sample’s statistical properties.

These findings show that children distinguish between personal and social actions in their generalization from statistical evidence and incorporate domain-specific knowledge into their statistical inferences.

Authors

• Teresa Flanagan, Cornell University
  Presenting Author

• Xin Alice Zhao, East China Normal University
  Non-Presenting Author

• Tamar Kushnir, Duke Univeristy
  Non-Presenting Author