Recent studies of prosodic disambiguation of structural ambiguity have shown consistent disambiguation by talkers of closure- and PP-attachment ambiguities. Schafer et al. (2000, 2001) found evidence of prosodic disambiguation by naïve talkers in a communicative task, while Snedeker et al. (2000) found such evidence only for talkers who were aware of the ambiguity. These clear results contrast with the conflicting results of earlier studies of direct object/sentential complement ambiguities such as (1).
|(1)||Tom noticed his roommate|
|a.||by the salad bar.|
|b.||was kind of depressed.|
Beach (1991) and Marslen-Wilson et al. (1992) found evidence that listeners were sensitive to prosodic differences between productions of sentences like (1), but replications of their experiments by Stirling and Wales (1996) and Watt and Murray (1996), respectively, found no such evidence. Furthermore, the latter three studies provided no acoustic or phonological description of the prosody used for DO or SC sentences.
The present study investigates whether talkers actually produce disambiguating prosody for DO/SC sentences. In Experiment 1, talkers were asked to read sentences like (1), first in a naïve condition, then after being informed about the ambiguity and instructed to disambiguate. Talkers' use of duration and intonation in the ambiguous region was not significantly different between the naïve and informed conditions, nor between sentences with DO and SC continuations.
In Experiment 2, listeners were asked to identify the intended continuations to the ambiguous fragments of the sentences produced by talkers in Experiment 1. In the cases where talkers had produced a large prosodic boundary before the verb's complement, listeners were significantly more likely to correctly identify that utterance as having a sentential complement (p=0.056). All other productions were identified by listeners at chance.
These results raise the question of why talkers so rarely produce such disambiguating cues for DO/SC sentences, given that the cues are helpful to listeners. One explanation is that talkers' production of prosody is motivated not by a need to disambiguate, but by phonosyntactic constraints that are part of the grammar. Because the recursive syntactic structure of an SC sentence does not map straightforwardly onto a strictly layered prosodic structure, prosody that would signal the syntactic structure is difficult to produce. In contrast, the sentences with closure ambiguities so reliably disambiguated by talkers in Schafer et al.'s (2000) study contain two adjacent but separate S-nodes, each of which can be mapped onto its own prosodic phrase.