Resolving Structural Ambiguity in Generated Speech

Ambiguity in the output is a concern for NLG in general. This paper considers the case of structural ambiguity in spoken language generation. We present an algorithm which inserts pauses in spoken text in order to attempt to resolve potential structural ambiguities. This is based on a simple model of the human parser and a characterisation of a subset of places where local ambiguity can arise. A preliminary evaluation contrasts the success of this method with that of some already proposed algorithms for inserting pauses for this purpose. 1 Avoiding Structural Ambiguity in NLG When a Natural Language Generation system has precise communicative goals, it is essential that there is as little ambiguity as possible in its output. However, it is rare for an NLG system actually to check that its output is unambiguous, except in restricted places such as referring expression generation. This paper considers structural ambiguity, where a generated sentence can have other syntactic analyses than...

• Type: Conference
• Year: 2004
• Where: INLG
• Authors: Chris Mellish
• Keywords: INLG 2004, Language Generation, Local Ambiguity, Natural Language Processing, Structural Ambiguity