Higgins

Work in progress




The Work in Progress section lists stuff we're working on or testing presently. As the listed items are concluded, they are moved out to other sections of the documentation.

When an item has been moved out, it will no longer be updated here apart from being moved into the Moved Out listing at the bottom of the page, with a note of the date of the move. Both the New and the Moved Out listings are in chronological order, newest first.

The items listed in this section may be work of any kind: implementations, data analysis, user experiments, articles, etc. By the nature of things, some of these are not going to succeed. However, as far as we can, we'll try to keep reports of failure as well as success.

Please take a moment to read the disclaimer and the copyright notice

New

Started: 2005-01-10: Dialogue system prototype tests
A first prototype of the Higgins spoken dialogue system is under testing. A video uptake of a live session can be found on the Scenarios page. A system description will be presented in time. [Expected first half 2005]
Started: 2004-08-18: Web based user test environment
In collaboration with other projects, we're planning to develop a simple user test environment for web use. [Expected end of 2004]
Started: 2004-08-18: Higgins web improvements
More of a vague TODO than a proper in progress, but in any case we're considering adding a RSS 1.0 feed to the web site. In addition to that, we're thinking about opening a Wiki section. The Wiki would mainly be used for internal discussions, however, to keep track of ideas and suchlike. [Expected at some point]
Started: 2004-03-01: ASR language modelling and error detection
The group has previous experience of using the web as a supporting corpus for building statistical language models. Tests where the web is used, for example to hone LM probabilities or to estimate the likelihood of words that Pickering deems to be insertions being misrecognised are underways.[Expected ???]
Started: 2004-02-22: 3D world simulation interface
We're working on methods to control the 3D world from Higgins. This will be useful for user tests, evaluations, and for development in general.[Expected 2005]
Started: 2003-11-11: Colour scheme for logs and displays
The Pickering semantic interpreter is developing nicely and we're working with documenting grammars and grammar functionality. Interpretation results as well as grammars can be presented in a web reader using XSLT, and for the sake of coherence, we're developing a colour scheme for encoding the presentations. You can check out an importable CSS file that holds the colour scheme as well as a HTML test file that illustrates the colour scheme. [Expected end of 2003]
Started: 2003-03-17: Example driven domain/topic predictor/preparser
We're testing ideas for making example based or data driven guesses about the domain or topic of an utterance. This module works seamlessly with the robust interpreter, and may possibly also be able to provide a confidence-like measure for the interpretation. [Expected first half of 2004]
Started: 2003-03-17: Recognition module
Rudimentary recognition module with an API that is fully prepared for more advanced system control. [Expected 1st half of 2003]

Moved out

Started: 2005-03-01: Speech fragment experiments, moved 2005-03-08 to Experiments
Feedback, backchanneling, and grounding are key concepts in Higgins. Currently, we are running a new series of perception, production, and behaviour tests to learn more how short utterances and grunts are interpreted and used to signal these topics.
Started: 2004-05-18: Turn-taking experiments, moved 2004-11-10 to Experiments
Feedback, backchanneling, and grounding are key concepts in Higgins. We're presently running a series of perception and production tests to learn more about the timing issues related to these concepts.
Started: 2004-01-20, moved 2004-08-28 to Pickering: Pickering documentation and manual
We're preparing the Pickering semantic interpreter for beta release by performing final bug testing, overseeing the specifications and by writing a manual. [Expected beginning of 2004]
Started: 2003-11-20, moved 2004-08-20 to Publications: Pickering article and evaluation
We're evaluating the robustness and functionality of the Pickering semantic interpreter using the speech data collected from 24 subjects in late 2003. [Expected beginning of 2004]
Started: 2003-11-01, moved 2004-02-25 to Experiments: Data collection
We're recording more people describing what the see in the 3D world simulation. 16 subjects are included in the second round. [Expected beginning of 2004]
Started: 2004-02-15, moved 04-02-25 to modules: Galatea
Work on Galatea, the Higgins discourse modeller, which is a very important part of the system's dialogue management, is underways [Expected 2004]
Started: 2004-01-15 moved 04-02-10 to modules: Prosody tools
We're working on a listing of ways we may incorporate prosodic information on-line in the system. [Expected 1st half of 2004]
Started: 2003-10-16, moved 2003-10-19 to Experiments: ASR error detection
We're running experiments on machine learning algorithms for detecting ASR errors. Preliminary results are promising. [Expected 1st half of 2004]
Started: 2003-09-20, moved 2003-10-19 to Experiments: Data collections
We're recording people describing what the see in the 3D world simulation. 8 subjects are included in the first round. [Expected end of 2003]
Started: 2003-06-18, moved 03-10-15 to modules: 3D world simulation
Ted Clemmedsson is working 3-6 weeks on building X3D/VRML objects for a 3D world simulation.
Started: 2003-03-17, moved 2003-06-15 to : the robust semantic interpreter Pickering
We've implemented a prototype for a robust semantic interpreter in Oz. The interpreter separates grammar rules from the actual interpretation engine, allows parsing to start from any grammar or grammar node, can allow insertions (at a cost that may be used as a kind of confidence) relative to the sentences generated by the grammar.
Started: 2003-04-24, moved out 2003-06-10 to specifications: Domain model
The domain model is specified in the shape of a publically available XML Schema.
Started: 2003-02-15, moved out 2003-05-20 to Modules and Architecture: Architecture outline
Specification of the systems architectural design, including high level data flow descriptions, communication protocols and level of modularity.
Started: 2003-02-20, moved 2003-03-28 to Experiments and to Publications:
User experiment: detection & correction of speech recogniser errors
We've let eight people correct a number speech recognition sentences in order to see
  1. How well they do, which may be considered a topline for automatic techniques
  2. To what extent context, confidence measures and the presence of N-best lists improves their results, which will help us see on what to concentrate.
[Paper intended for HLT/NAACL 2004]


Valid XHTML 1.1!|Valid CSS!|Level A conformance icon, W3C-WAI Web Content Accessibility Guidelines 1.0