Skip to main content
Przejdź do strony domowej Komisji Europejskiej (odnośnik otworzy się w nowym oknie)
polski polski
CORDIS - Wyniki badań wspieranych przez UE
CORDIS
Zawartość zarchiwizowana w dniu 2024-05-27

Specification, Interaction and Reconfiguration in Dialogue Understanding Systems

CORDIS oferuje możliwość skorzystania z odnośników do publicznie dostępnych publikacji i rezultatów projektów realizowanych w ramach programów ramowych HORYZONT.

Odnośniki do rezultatów i publikacji związanych z poszczególnymi projektami 7PR, a także odnośniki do niektórych konkretnych kategorii wyników, takich jak zbiory danych i oprogramowanie, są dynamicznie pobierane z systemu OpenAIRE .

Rezultaty

Assessment of the usability of TrindiKit and GoDIS for the development of tutorial dialogues: We assessed the feasibility of developing a tutorial dialogue system using the TrindiKit. Tutorial dialogue has been chosen as an example of a difficult type of dialogue to handle, because of the complex interplay between domain knowledge, tutorial strategies and linguistic realization thereby exhibited. We used GoDIS (questions under discussion) for action-oriented dialogue as a starting point, and discussed concrete modifications and extensions needed to model the phenomena typically occurring in tutorial dialogue. More information on the SIRIDUS project is available on the project's web page: http://www.ling.gu.se/projekt/siridus/
Experimental implementation of information structure realisation through intonation with Mary and Festival TTS in GoDIS: We developed an implementation in the GoDIS (questions under discussion) system, which enables to experiment with the realization of information structure by means of intonation, using two off-the-shelf text-to-speech (TTS) synthesis systems, namely the Festival TTS for English and the Mary TTS for German and English. These TTS systems support high-level control of intonation by using either the SABLE standard or a ToBI-based intonation markup. We defined experimental mappings from information structure categories to intonation annotation, and ran a small-scale evaluation of the synthesized output in different versions. The evaluation results indicate that controlling the output intonation by determining intonation according to information structure with respect to the information state leads to improving the appropriateness of the system output. More information on the SIRIDUS project is available on the project's web page: http://www.ling.gu.se/projekt/siridus/
Spoken language dialogue systems, such as automated telephone enquiry systems and hands-free in-car device control, are rapidly becoming a commercial reality. SIRIDUS aims to improve the understanding of what is required to provide reusable, robust and user-friendly spoken dialogue systems. The project demonstrators include an automated telephone operator, and an integrated toolset for dialogue researchers. One of the objectives of the SIRDIUS project was to provide architectures which allow appropriate sharing of information between modules, for example, enabling dialogue systems to generate appropriately stressed output. Conditional responses are responses such as the ones (1Sa,b) below: U: Can I fly on Monday? - S: Not if you want to fly business class. - S: Yes, if you fly economy class. We carried out a small corpus investigation of such responses using the AMEX travel-agency data and the Verbmobil corpus of appointment-scheduling dialogues. We analysed the contexts in which these responses are used and the effect speakers achieve by using them. Subsequently, we provided a theoretical analysis of conditional responses and their function in dialogue. Then we implemented a treatment of conditional responses as an extension to the repertoire of dialogue moves handled in GoDIS. More information on the SIRIDUS project is available on the project's web page: http://www.ling.gu.se/projekt/siridus/
Specification and implementation of information structure determination from information state in GoDIS: Information structure is a level of utterance meaning that specifies a partitioning of the meaning according to how it fits into the context, in particular, what parts of the meaning are context-dependent vs. context-affecting, and what parts of the meaning should be distinguished from alternatives in the context. We have employed one prominent existing approach to information structure, and specified how information structure partitioning can be derived from the information state in the GoDIS (questions under discussion) dialogue-modelling framework. In this way, different information structure partitioning of the same prepositional content can be derived in different contexts. Different languages realize information structure differently, e.g., by means of combining intonation, word order, syntactic structures, morphological marking. Determining information structure provides a uniform level of representation. More information on the SIRIDUS project is available on the project's web page: http://www.ling.gu.se/projekt/siridus/
TRINDIKIT is a toolkit for building and experimenting with dialogue move engines and information states. We use the term information state to mean, roughly, the information stored internally by an agent, in this case a dialogue system. A dialogue move engine, or DME, updates the information state on the basis of observed dialogue moves and selects appropriate moves to be performed. Apart from proposing general system architecture, the TRINDIKIT also specifies formats for defining information states, update rules, dialogue moves, and associated algorithms. It further provides a set of tools for experimenting with different formalizations of implementations of information states, rules, and algorithms. As from version 2.0, TRINDIKIT supports asynchronous (concurrent) control. To build a dialogue move engine, one needs to provide definitions of update rules, moves and algorithms, as well as the internal structure of the information state. One may also add inference engines, planners, plan recognisers, dialogue grammars, dialogue game automata etc. The current version of TrindiKit is 3.0 (released end December, 2002) and is downloadable from the TrindiKit web page (http://www.ling.gu.se/projekt/trindi/trindikit/). More information on information states and related concepts and a link to the TrindiKit web page is available on the SIRIDUS web page http://www.ling.gu.se/projekt/siridus/

Wyszukiwanie danych OpenAIRE...

Podczas wyszukiwania danych OpenAIRE wystąpił błąd

Brak wyników

Moja broszura 0 0