Each participant saw 4 experimental conditions (4 trials each), 2 control ... WSI or SSI; (iv) the false condition was not consistent with either WSI or SSI, nor with ...
Children’s interpretation of sentences with multiple scalar terms Summary: Previous research has found that children typically do not compute the scalar implicatures (SIs) associated with sentences containing a single scalar term, such as (1a). This study investigated children’s comprehension of sentences containing multiple scalar terms, such as (2a). The main finding was that with such sentences, children computed SIs on a par with adults. We suggest that children succeeded in computing SIs for two reasons: (i) the scalar alternative every appeared in the sentence; (ii) the visual context explicitly depicted the truth conditions associated with the alternative scalar terms some and every. Current study: Much previous literature has shown that children, unlike adults, typically accept sentences such as (1a), in contexts in which their SIs (1b), are false, e.g., (3A) (Noveck 2001; Gualmini et al. 2001; Papafragou & Musolino 2003; Katsos & Bishop 2011; among many others). (1) a. The lion ate some of his cookies. b. The lion didn’t eat every one of his cookies (2) a. Every lion ate some of his cookies. b. Not every lion ate every one of his cookies (weak) c. No lion ate every one of his cookies (strong) However, children’s performance varies depending on task and methodology. For instance, children’s performance improves if the linguistic alternatives required to compute the relevant SI are made explicit to the child during the trial. This fact has led to the proposal that the source of children’s difficulties with SIs lies in the retrieval of the required alternatives from the lexicon (see Chierchia et al. 2001; Reinhart 2006; Barner et al. 2011; Tieu et al. In press). We designed an experiment to investigate this factor further by looking at sentences containing multiple scalar terms (2a). To our knowledge, this has not been previously investigated with children. In an experiment with adults, Chemla & Spector (2011), found that participants computed two kinds of SIs in response to sentences like (2a): a weak inference (WSI) (2b), and a strong inference (SSI) (2c). The SSI can be generated either globally or locally (Chemla & Spector 2011). For present purposes, the pertinent observation is that, either way, (2a) explicitly contains the scalar term needed to compute either kind of SI, namely every. Given previous findings that children’s performance improves when alternatives are explicitly mentioned, we were led to make the otherwise counter-intuitive prediction that children would be more successful in computing SIs in sentences with multiple scalar terms (2a), than they have typically been with sentences containing a single scalar term (1a).
(3A) (3B) (3C) Note: Blank circles indicate eaten cookies. These pictures indicate the critical elements of the relevant conditions; actual test items involved acting out stories with toy props. Experiment: We used a Truth Value Judgment Task (Crain & Thornton 1998) to investigate the interpretations assigned by children and adults to sentences like (2a). Eighteen adults and 13 children (4;01-5;07, M=4;06) were included in the final data set (5 children were excluded due to failing controls). On each trial, one experimenter acted out stories in front of the participant and a puppet (manipulated by a second experimenter). Every story contained three characters (e.g., lions), and each of the characters performed some action (e.g., eating) on none/some/all of a set of four objects (e.g., cookies). At the end of each story, the puppet described what it thought had happened in the story. The participant’s task was to judge whether the puppet’s description was right or wrong. Each participant saw 4 experimental conditions (4 trials each), 2 control conditions (2 trials each), and 5 filler items. The experimental conditions varied in which inferences were made true: (i) the strong condition was consistent with both WSI and SSI; (ii) the weak condition (3B) was only consistent with WSI; (iii) the literal condition (3C) was not consistent with either
%Acceptance
WSI or SSI; (iv) the false condition was not consistent with either WSI or SSI, nor with the literal meaning. The two control conditions ensured children interpreted the quantifier every in a target manner; they also ensured that children were not rejecting in the weak condition simply because not all of the characters performed an action on the same number of objects. Finally, the filler trials allowed us to balance the number of yes-/no-responses, and to ensure the participants were paying attention. The experiment was carried out across two sessions, presented roughly 7-9 days apart. Results: Fig. 1 displays the proportion of yesresponses across conditions. We ran a mixed model analysis with the literal condition in adults as the 1.00 reference level. The model revealed: (i) no effect of group in the literal condition; (ii) significant dif- 0.75 ferences between literal and the other three congroup ditions (all p-values
