Adaptive mixed-initiative dialogue management

Abstract

This paper describe mixed-initiative dialc form-filling paradign hierarchy that contai Along with the slot I that allows for adapi on the success or fai zoom-in to more de level questions. The distribution of i the granularity of thc mine the right granul take into account the and the knowledge , suitable in situations cases, the user may However, giving the behaviour less predii on speech-understanc lower-level question! the user supplies U switch to higher-leve 1 Many practical dia filling paradigm. 7 the user, the syste This information i and questions can t One of the merits allows a clear disti changing informat task is described bl values that they m: the order in which ing of slot values, and uncertainties t ticipants attempt t c strategies can be r( that developing a i limited to specifyir tics.0-7803-5028-6/98/$ ABSTRACT in architecture that supports adaptive le. It is based on a generalisation of the lather than a flat slot structure, we use a slots at various levels of abstraction. rarchy, a question hierarchy is defined : mixed-initiative dialogue. Depending -e of certain questions, the system can led questions, or zoom-out to higheriative in dialogue is closely related to nformation that is asked for. To deterry level for system questions we have to fluences on user freedom, predictability the user. Giving the user initiative is iere the user knows what to do. In such 'e all relevant information in one turn. ser more initiative tends to make his ble' and therefore increases the chance g errors. The system should switch to vhen higher-level ones fail, and when olicited information, the system can iestions. TRODUCTIONcue systems are based on the formperform some action on behalf of needs information from the user. ,epresented as a fixed set of slots, isked for each slot. the slot-filling paradigm is that it tion between the dialogue task (ex-I) and the dialogue strategy. The le slots themselves and the possible get. The strategy is concerned with its are discussed, and with ground-:. with inconsistencies, ambiguities t can arise when the dialogue par3uild a common ground. Dialogue ;ed for many different domains, so dogue system for a new domain is a new set of slots and their seman-00 0 1998 IEEE A drawback of the slot-filling paradigm as it stands, is that it results in rather rigid dialogues. The "model dialogue" within the slot-filling paradigm is one in which a number of system questions are interleaved with answers from the user, each answer answering just the question that was asked by the system. In spoken human-human dialogues, this kind of dialogue is very rare; People ask questions that refer to concepts that involve more than one slot, and they are often over-informative and say things which they believe are relevant, even when they aren't asked for. More so, they use these devices intelligently, using high-level concepts to open the dialogue, and zooming into details when problems arise. Many systems (for instance, the one described by Aust et al. [[I]]) are able to ask questions involving ...

Automatic detection of poor speech recognition at the dialogue level

“…This work differs from previous work in focusing on behavior at the (sub)dialogue level, rather than on identifying single misrecognitions at the utterance level (Smith, 1998;Levow, 1998;van Zanten, 1998). The rationale is that a single misrecognition may not warrant a global change in dialogue strategy, whereas a user's repeated problems communicating with the system might warrant such a change.…”

From data to speech: a general approach

“…It is used for the generation of system output in the OVIS system, a spoken dialogue system that gives information about public transport in the Netherlands. In OVIS, planning is performed by the Dialogue Management module, described in Veldhuijzen van Zanten (1998). This module provides the LGM with conceptual representations of the messages to be generated.…”

“…D2S was initially developed with the construction of the Dial Your Disc (DYD) system, which generates spoken monologues in English, giving information about recordings of compositions by Mozart (van Deemter, Landsbergen, Leermakers and Odijk 1994;Odijk 1995;van Deemter and Odijk 1997). D2S is also used for output generation in OVIS, a Dutch travel information system (Veldhuijzen van Zanten 1998;van Noord, Bouma, Koeling and Nederhof 1999). Earlier D2S-related papers concentrated on specific aspects of D2S, such as topic management (Odijk 1995), the use of context in language generation (van Deemter and Odijk 1997), the computation of prosody (Theune, Klabbers, Odijk and de Pijper 1997), and speech generation (Klabbers 1997).…”

Contrast in concept-to-speech generation

“…It was originally developed as part of the Dial Your Disc (DYD) system, where it was used to generate English spoken monologues about Mozart compositions [van Deemter et al, 1994, Odijk, 1995, van Deemter and Odijk, 1997. Since then, D2S has been used for spoken language generation in various systems including GoalGetter, a system that generates Dutch spoken football reports [Theune et al, 2001], and OVIS, a Dutch spoken dialogue system providing train information [Veldhuijzen van Zanten, 1998, van Noord et al, 1999, Klabbers, 2000. In this section, examples from GoalGetter are used to illustrate D2S.…”