The Higgins project is aimed at researching and developing a prototype dialogue system focussed at testing and evaluating different error handling strategies. The goal is split into a number of short and long term partial goals. This page gives a brief overview of some of these.
Long term research issues
The core research issue for the Higgins project is how the various errors that may enter dialogue can be handled or prevented in different ways. The domain was chosen since it can be assumed to give rise to a variety of different error types. The long term research goals of the project include the following items, which also permeate shorter term research issues:
- How do different levels of feedback (information about how the system interpreted the user's utterance) and detail in system responses affect robustness and efficiency of the dialogue?
- Which factors should govern the dialogue strategy regarding feedback and information detail?
- How can the different system modules calculate confidence measures?
- How can we categorise errors regarding the types of error handling they demand?
- How should the system act when it doesn't understand, or only partially understands, what the user says?
- How does the user's beliefs about the system’s understanding correlate to its actual understanding, depending on error handling strategy?
Current research issues
As a first step, a Wizard-of-Oz-related experiment where a speech recogniser was used to introduce errors typical for dialogue systems into human-to-human dialogues was performed. The error situations identified in these tests are the bases for the development of the system.
We aim to build a dialogue system which has features that facilitates error handling studies. The following is a sample of issues we are currently working with, and their present status:
- Which robust interpretation techniques are suitable for more complex utterances? Pickering, the Higgins semantic interpreter, presently achieves robust interpreteation by using grammars allowing insertions, fragments and non-congruence.
- How does a dialogue manager that is able to vary its feedback level make its choices? Pickering, the Higgins interpreter, will produce confidence measures which may be combined with ASR confidence to provide a basis for selecting grounding and feedback levels.
- How can incremental speech recognition be utilised to improve dialogue? Pickering, the Higgins interpreter, supports incremental parsing that facilitates fast feedback, which could help the user detect and correct errors.
- What demands on interpretation and error handling are specific for the chosen domain? Pedestrian navigation dialogues and human error handling strategies have been collected in a modified Wizard-of-Oz setting, as has spatial descriptions in a simulated 3D environment. The data is presently being analysed.
Copyright © 2002-2004 Jens Edlund, Gabriel Skantze and the members of the Center for Speech Technology
|
