Prosodic disambiguation of syntactic ambiguities in sentence production (and comprehension)

Michael Walter, Lars Konieczny , and Barbara Hemforth
miwalter@cognition.iig.uni-freiburg.de
Center for Cognitive Science, University of Freiburg

Schafer & al. (1996, 2000) showed that during comprehension listeners use prosodic cues to disambiguate syntactic structures. They present evidence that the distribution of pitch accents and prosodic boundaries such as intermediate phrase boundaries strongly influences attachment decisions: when presented with acoustic materials spoken by trained speakers, listeners prefer to attach a relative clause that can be attached to either NP1 or NP2 to the NP that has a stronger pitch accent, while a prosodic boundary between the complex NP and the RC results in an NP1 preference. With regard to naïve speakers, Schafer & al. (2000) present evidence that they also use prosody to disambiguate syntactic ambiguities.

We investigated whether naïve speakers of German prosodically disambiguate ambiguous structures of the type NP1-NP2-RC. We had 16 subjects read sentences like (1) and (2) in which we varied the attachment site of the relative clause (NP1 vs. NP2). Subjects were instructed to read each sentence silently first and were then presented with a question referring to the attachment of the RC in order to check whether the subjects had correctly understood to which NP the relative clause refers. Subjects were then told to read the sentence aloud.

(1)
Die Anwältin der Verbrecher, die schlief, war sehr beliebt.
The lawyer (fem./sing.) of the criminals (pl.) who slept (sing.) was very popular.
Question: Who slept?
(2)
Die Anwälte der Verbrecherin, die schliefen, waren sehr beliebt.
The lawyers (pl.) of the criminal (fem./sing) who slept (sing.) was very popular.
Question: Who slept?

The sentences speakers produced were recorded on DAT-tape, digitalized and then analyzed using the program xwaves. Although there was considerable individual variation in the prosodic contours produced, we found two prosodic parameters that were consistently used to disambiguate structures: Firstly, speakers produced a stronger pitch accent on NP2 as compared to NP1 if an NP2 attachment was produced. In the case of NP1 attachments, there was no difference between the pitch accents on NP1 and NP2. Secondly, speakers produced stronger boundary tones (intermediate phrase boundaries) before the RC in the case of NP1 attachments than in the case of NP2 attachments.

Our results show that prosodic parameters are used during production to disambiguate syntactic structures. However, although the prosodic parameters we found are the same that Schafer & al. (1996) claim to be relevant for comprehension in English, we have no evidence yet that the prosodic parameters used for disambiguation by German speakers also guide comprehension. We are currently preparing a comprehension experiment in which subjects will be presented with materials from the production study to check whether the disambiguating prosodic parameters used in production are used by listeners to disambiguate syntactic structures.



AMLaP Conference, Saarbrücken, September 2001