Producing contextually appropriate intonation in an information-state based dialogue system

Abstract

Our goal is to improve the contextual appropriateness of spoken output in a dialogue system. We explore the use of the information state to determine the information structure of system utterances. We concentrate on the realization of information structure by intonation. We present the results of evaluating the contextual appropriateness of varied system output produced with a text-to-speech synthesis system that supports intonation annotation.

Perception of visual scene and intonation patterns of robot utterances

“…Assigning intonation to dialogue system output in a way that reflects relationships between entities in the discourse context can enhance acceptability [3,8]. Previous research has concentrated on the role of linguistic context in processing; dialogue situatedness and the role of visual context in determining accent placement has not been studied.…”

Information structure in topological dependency grammar

“…Particularly in applications that involve human-computer interaction, information structure has thus been found to have a great impact on the understandability of computergenerated language, e.g. question/answering dia-logues (Prevost and Steedman, 1994;Hoffman, 1995;Kruijff-Korbayova et al, 2003) or generation (Kruijff-Korbayova et al, 2002).…”

Thematicity-based Prosody Enrichment for Text-to-Speech Applications

“…Consider, for instance, Steedman's work [5] on the correlation of theme and rheme with rising and falling intonation patterns based on a question-answer setting. An implementation on the grounds of Steedman's characterization was tested in concept-to-speech (CTS) applications [6,7,17]. However, a derivation of thematicity that is fully dependent on a question-answer setting is limited to a specific conversational environment and, thus, fails to deal with the problem of monotony of TTS applications.…”

“…The default synthesized speech output has been enriched using MaryXML prosody specifications 6 , which follow the SSML recommendation 7 .…”

“…Several works in the context of speech synthesis (see, e.g., [5,6,7]) draw upon a theme (i.e., "what is being talked about")-rheme (i.e., "what is being said") division. Such a binary flat division is used to establish a one-to-one correspondence between theme-rheme spans and rising-falling intonation patterns in terms of ToBI labels.…”