Comparing the output of a prosodic segmenter to syntactic analysis

OData support
Dr. Szaszák György József
Department of Telecommunications and Media Informatics

Prosodic features like intonation, stress or rhythm play a very important role in speech communication. In spoken conversation this information not only helps listeners to locate word and phrase boundaries, but also helps to determine the intention or the emotions of the speaker.

In linguistics, syntax is the study of the principles and rules for constructing sentences and syntagms in natural languages.

Prosody and syntax are highly related, even if the prosodic structure cannot be directly mapped to the syntactic one and vice versa.

The aim of my work was to analyze, in what degree the syntactic structure can be predicted based on the prosodic features extracted from the speech signal.

First, a prosodic segmenter was used on the speech material, and in parallel, the sentences were also fed into an automatic syntactic analyser, Hunpars. After matching the outputs of these two analysers, the correspondence of prosodic and syntactic boundaries was examined.

Results show that based on prosodic structure, three syntactic levels can be recovered. Recall of clause boundaries was about 82%, while recall of lower levels ranged around 45-50%. Precision of clause boundary detection was very high, around 90%, and for lower levels around 60%. The syntactic structure – to some extent – can be recovered based on the aligned phonological phrase.

These findings can be exploited in speech understanding systems.


Please sign in to download the files of this thesis.