Overview of Grodner Et Al. (2010)

We constructed 32 stimulus items. Test stimuli were separated into four lists, where each condition was equally represented, and each item appeared once. Across lists, each item appeared in each condition an equal number of times. Test stimuli were presented in random order and intermixed with 40 fillers, designed so that throughout the experiment, all six individual types in a display were equally likely to be a target, and each quantifier was equally likely to designate a member of the target or opposite gender. The distribution of objects in the fillers was like the early and late-summa target displays, but a member of the opposite gender was identified (the boys in Fig. 1). This was so the perceiver could not predict beforehand which of the six individuals was the correct target. Fourteen fillers contained alla, eight nunna, and eight summa. An additional ten items used the definite determiner ‘‘the” to refer to a target with a proper subset of objects. For the main eye-tracking experiment, commands were recorded with the nuclear accent on the sentence-final noun, and secondary (H*) prominence on the quantifier. Care was taken to insure that the quantifier did not receive contrastive stress. To ensure this, the commands were coded according to the ToBI system (Silverman et al., 1992). Ninety-three of the 96 stimulus quantifiers received H* accenting; three were ambiguous between H* and L + H*. The lengths of critical regions are reported in Table 1. Two-tailed comparisons between the Alla and Summa conditions revealed that the interval between the onset of the quantifier and noun was marginally shorter for the Alla commands (F(1, 31) = 3.1, p = .09) as was the interval between the noun onset and POD (F(1, 31) = 2.9, p = .10). Thus, if anything, the identifying acoustic information was delivered earlier in the Alla condition. To determine whether the quantifier + partitive region of our items was phonetically reduced, we compared our auditory stimuli to those used by Huang and Snedeker. Three research assistants who were naïve to the hypotheses being tested hand coded the length of the interval from quantifier onset to determiner onset. Their judgments were highly correlated (all rs > .89, all ps < .0001). Quantifier lengths are given in Table 2. Both summa and alla were reliably shorter in our stimuli (summa: F(1, 46) = 141.2, MSE = .5, p < .001; alla: F(1, 46) = 173.9, MSE = .4, p < .001). To be sure that this was due to selective reduction of the quantifier, and not merely a faster speaking rate in the present study, the duration of this segment was normalized by dividing it by the interval length from the onset of the verb to the onset of the determiner. This region was chosen because it contained an equivalent number of syllables for all stimuli. Even as a proportion, the quantifier was reliably reduced in our items (summa: F(1, 46) = 122.6, MSE = .006, p < .001; alla: F(1, 46) = 168.1, MSE = .006, p < .001). If the reduction of the first syllable was sufficient to identify the partitive in the present items, but not in Huang and Snedeker’s items, then the phonetic cue to the presence of the partitive construction would have been postponed from the onset of the quantifier until the onset of the vowel in of. This corresponds to a mean delay of 226 ms (SD = 32 ms).