On children's variable success with scalar inferences: Insights from disjunction in the scope of a universal quantifier.

Previous developmental studies have revealed variation in children's ability to compute scalar inferences. While children have been shown to struggle with standard scalar inferences (e.g., with scalar quantifiers like "some") (Chierchia, Crain, Guasti, Gualmini, & Meroni, 2001; Guasti et al., 2005; Noveck, 2001; Papafragou & Musolino, 2003), there is also a growing handful of inferences that children have been reported to derive quite readily (Barner & Bachrach, 2010; Hochstein, Bale, Fox, & Barner, 2016; Papafragou & Musolino, 2003; Singh, Wexler, Astle-Rahim, Kamawar, & Fox, 2016; Stiller, Goodman, & Frank, 2015; Tieu, Romoli, Zhou, & Crain, 2016; Tieu et al., 2017). One recent approach, which we refer to as the Alternatives-based approach, attributes the variability in children's performance to limitations in how children engage with the alternative sentences that are required to compute the relevant inferences. Specifically, if the alternative sentences can be generated by simplifying the assertion, rather than by lexically replacing one scalar term with another, children should be better able to compute the inference. In this paper, we investigated this prediction by assessing how children and adults interpret sentences that embed disjunction under a universal quantifier, such as "Every elephant caught a big butterfly or a small butterfly". For adults, such sentences typically give rise to the distributive inference that some elephant caught a big butterfly and some elephant caught a small butterfly (Crnič, Chemla, & Fox, 2015; Fox, 2007; Gazdar, 1979). Another possible interpretation, though not one typically accessed by adults, is the conjunctive inference that every elephant caught a big butterfly and a small butterfly (Singh, Wexler, Astle-Rahim, Kamawar, & Fox, 2016). Crucially, for our purposes, it has been argued that both of these inferences can be derived using alternatives that are generated by deleting parts of the asserted sentence, rather than through lexical replacement, making these sentences an ideal test case for evaluating the predictions of the Alternatives-based approach. The findings of our experimental study reveal that children are indeed able to successfully compute this class of inferences, providing support for the Alternatives-based approach as a viable explanation of children's variable success in computing scalar inferences.