Anna Papafragou1 & Julien Musolino2
1 University of Pennsylvania, 2 Indiana University
This paper investigates the development of the ability to compute scalar implicatures (SIs) during language comprehension. SIs arise in examples like 'Some professors are rich' where the speaker's use of 'some' typically indicates that s/he had reasons not to use a more informative term, e.g., 'all'. 'Some professors are rich' therefore gives rise to the implicature 'Not all professors are rich'.
Although very little is known about the psychological status of SIs, recent studies suggest that (a) for adults, SIs "seem to be virtually non-existent in syllogistic reasoning tasks" (Newstead, 1995), and (b) that preschoolers do not show sensitivity to SIs (Noveck, 2001). These conclusions raise two important questions: (1) Are SIs psychologically real?; (2) If so, what factors affect the way SIs are computed during language comprehension?
We present here the results of two experiments designed to address these questions. In the first experiment, we tested a group of 30 adults and 30 5-year-olds on three different scales, 'some/all', 'two/three' and 'start/finish'. In each case, subjects were presented with short stories which satisfied the truth conditions of the stronger term of each scale (i.e., 'all', 'three', and 'finish') but were described by a puppet using the pragmatically infelicitous weaker term of the scale (i.e., 'some', 'two', and 'start').
We found that while adults overwhelmingly rejected the puppet's infelicitous statements (95% of the time on average), children did so only 29% of the time. Moreover, we found that while the rejection rates of adults did not significantly differ across the three scales, children's rejection rate on the 'two/three' scale was reliably higher than on the 'some/all' and 'start/finish' scales (65% vs. 12.5% and 10% respectively, p<.01). In order to address question (2), we trained a group of 30 5-year-olds to detect pragmatic anomaly (e.g., 'This is a brown thing' used to refer to a dog) and then tested them in the way described in Experiment 1. This manipulation gave rise to significantly higher rejection rates (52.5% vs. 12.5%; 90% vs. 65% and 47.5% vs. 10%, p <.01).
We conclude that: (i) adults easily compute SIs when the task is relevant; (ii) children's ability to compute SIs is more fragile than that of adults; (iii) children, unlike adults, do not treat all scalar terms alike; and (iv) children's ability to compute SIs is affected by their awareness of the goal of the task. Theoretical, methodological and developmental implications are discussed.