TRINDI Annual Report 1999

Information exchange in dialogue

TRINDI (Task Oriented Instructional Dialogue) is an EU-funded research project concerned with
  • human-machine interaction using natural dialogue
  • task-oriented instructional dialogues - dialogues that enable the user to make choices in the performance of a certain task
  • route planning and instructional texts for xerox machine maintenance (as particularexamples)
  • finding general and portable technology that will enable the construction of dialogue systems
  • increasing the flexibility and knowledge richness of dialogue systems by representing the information which is exchanged in a dialogue
  • developing a dialogue toolbox which allows developers to define dialogue rules for information update and to experiment with different kinds of information representing the current state of the dialogue

Summary of activities in 1999

The second year of the project began 1st February, 1999. During this year we
  • developed a theoretical model of dialogue moves and information state revision in dialogue, in the form of a general framework with three different instantiations
  • implemented and documented a first version of the TRINDIKIT dialogue toolbox
  • implemented preliminary versions of three restricted dialogue systems using TRINDIKIT: Midas, SRI-Autoroute and GoDiS
  • conducted preliminary work on evaluating these three systems against a grid of levels on interaction of the user and the system
  • conducted preliminary research on adapting our model of dialogue moves and information state revision to the analysis and generation of instructional texts
  • conducted preliminary research on the possibilities of accommodating focus and prosodic information into the TRINDI technology
  • conducted preliminary work on extending some of the previous work in the FraCaS project on underspecification to dialogue

Market prospects -- menus vs dialogues

Much of our interaction with automated task oriented systems is currently menu driven. For example, an automated system to book a cinema ticket can involve first choosing the city, then the film, the cinema and the time. Current technology using a telephone interface involves listening to a number options and responding by pressing a number.

An example of a more complex menu structure is present in the programming facilities of mobile phones where the menu can be viewed as a tree structure which the user has to descend starting at the root node.

While menu interfaces are ubiquitous in modern technology they are often tedious and frustrating.

Current dialogue technology can straightforwardly implement menu structures but the effect can be even more tedious since the user is forced to use the natural and efficient modality of speech in a very unnatural way descending the menu structure one node at a time.

TRINDI technology offers the possibility of allowing the user to present several pieces of relevant information at one time or to present information in the order in which the user finds most natural (either in speech or written interfaces). The system will request information that is missing. If information presented by the user is relevant to a specific branch of the menu system then the system will be able to jump directly to that branch without requiring the user to step through all the intervening nodes. If the user does not know what to do the system can present the options. This means that users can use their own conception of the knowledge space and not be locked to that of the designer of the system. This will allow for more natural dialogic interaction with computers in situations where the conventional interface of keyboard and screen is impossible or undesirable.

While TRINDI technology is not limited to the "dialogization" of menus, this application alone is of huge market potential given the omnipresence of menus in current interfaces.

Innovative technology outlook

Why do people engaged in a dialogue say what they say when they say it? What effect do their contributions have on the information available to the participants in the dialogue? These are central questions for the automation of dialogue.

Machines need to have simple strategies that simulate aspects of human dialogue behaviour. The strategies need to be simple enough to be computationally robust and yet complex enough to make the dialogues coherent and not too irritating for the user. The simplest strategies involve just producing particular utterances in sequence or directly computing a response on the basis of the preceding utterance from the user. A common more sophisticated approach is to treat dialogues as games and utterances as instantiations of a limited set of moves (such as question, answer, clarification, request, confirmation, etc.). Allowable moves can be specified in terms of transition networks.

Our approach is to enhance this conception by associating with dialogue moves representations of the information that the dialogue participants have. The main effect of an utterance is to change this information in some way, and the information is used by the participants to decide what to do next.

The TRINDIKIT dialogue toolbox

TRINDIKIT is a toolbox for the development of dialogue systems. It focusses on the development of dialogue move engines. A dialogue move engine updatesthe information state of the dialogues system on the basis of observed dialogue moves and selects appropriate moves to be performed.

Apart from proposing a general system architecture, the TRINDIKIT also specifies formsats for defining information states, update rules, dialogue moves and associated algorithms.

TRINDIKIT provides a general modular architecture for the construction of and experimentation with dialogue systems and in addition provides:

  • definitions of datatypes
  • a language and format for specifying upodate rules
  • methods for accessing information states
  • an algorithm definition language for dialogue move engines and Control modules
  • default modules for input, interpretation, generation and output
  • methods for visually inspecting information states
  • debugging facilities

The TRINDIKIT comes along with three example systems:

  • MIDAS, Multiple Inference-based Dialogue Analysis System
  • GoDiS, Gothenburg Dialogue System
  • Autoroute
TRINDIKIT's homepage for further information.

Promotion and awareness

The project has created an International Consultation and User Group (ICUG).

Members of the ICUG were invited to a the TRINDI workshop held on 6th May, 1999 which was held in conjunction with Amstelogue'99 (the third in a series of international workshops on the Semantics and Pragmatics of Dialogue), Amsterdam, 7th-9th May. The workshop was open to other interested participants and invited talks were given by two ICUG members: David Sadek, CNET France Telecom and Bob Carpenter, Bell Labs, Lucent Technologies.

The TRINDI project is mainly responsible for organizing the next conference in the series: Götalog 2000 - Fourth Workshop on the Semantics and Pragmatics of Dialogue to be held in Göteborg, June 15-17 2000. During this conference there will be a TRINDI event to which the ICUG and others will be invited.

Future work

The project will end in July 2000. During the final seven months we will
  • complete our work on dialogue dynamics and levels of interaction
  • complete our work on restrictive dialogue systems
  • complete our work on the application of the TRINDI dialogue model to instructional texts
  • complete our work on information structure and prosodic cues
  • examine issues in underspecified and robust interpretation
  • produce an extended version of the TRINDIKIT toolbox

Further information

Further information about the project can be obtained on the TRINDI web page or from the project coordinator:

Robin Cooper
TRINDI Administrator, Department of Linguistics, Göteborg University, Box 200, S-405 30 Göteborg, Sweden

Printable version of the TRINDI document: