Skip to Content

Department of Linguistics

Grammatical and Semantic Preprocessors

Robert Mannell

Grammatical and Semantic preprocessors in TTS systems are often very rudimentary. Only a small amount of syntactic and semantic information is required to produce reasonably acceptable prosody for many sentences. However, this minimalist approach tends to favour declarative sentences with a single clause or intonational phrase. Complex sentences, for example sentences with embedded clauses, tend to have unacceptable prosody when processed by a TTS system with a minimalist grammatical and semantic preprocessor.

What sorts of problems do grammatical and semantic preprocessors need to solve?

The most important type of information that a TTS system requires from such preprocessors is information that assists in the determination of the prosodic structure of a sentence. What, for example, are the positions of the boundaries between the intonational phrases and which words in each phrase will take the tones? What types of tones will be allocated? Whilst a grammatical processor might help in the placement of phrase boundaries, some degree of semantic processing may be necessary for tone placement. For example, some minimal understanding of the sentence may be required in order to determine which words are "given" and which are "new". There is generally a preference for the placement of tones on "new" information (ie. on content words which carry new information).

Some grammatical information would be helpful for the selection of the alternatively stressed forms of words like "record" (ie. 'record vs. re'cord). Some semantic information would be useful for the disambiguation homographs such as "bow" and "lead".

  • For information on how MU-Talk processes grammatical and semantic information, click here.
  • For more information on how grammatical and semantic preprocessors can facilitate better synthetic speech prosody, click here.
  • For more information on how grammatical and semantic preprocessors can help to disambiguate homographs and words that exhibit grammatical stress alternation, click here.