Dialogue managers (DMs) are of increasing interest to interface developers, but have not been as useful as they could be. Most existing dialogue managers are not built from existing components; instead, they are knocked together by interface developers without input from linguists. The resulting interfaces are not flexible enough to handle conversations of moderate complexity, multiple modalities, or more than two participants, or to be adapted to new conversational tasks or domains without considerable effort.
Existing dialogue toolkits either focus on the needs of linguists and are difficult to field and evaluate, or focus on the needs of individual developers and provide little or no visibility into what linguistic theories they may embody.
Our objective with Midiki is to bridge that gap. We want to develop a toolkit that is friendly to linguists, supporting sophisticated behavior while providing full visibility into its internal operations. We are striving to build a toolkit that is agnostic with regard to data distribution method, operating system, programming language, and theory of dialogue. With enough users, we might have a shot at it.
An initial version of the software has been developed, and is being hosted at SourceForge. We appreciate every bit of feedback we can get from actual users. Please don't hesitate to use the tools available from Sourceforge to report bugs, request features, or vent your frustrations (constructively).
DownloadingIf possible, you should use anonymous CVS to access the latest version of the software. During the alpha stage, bugs have been identified and fixed in the repository but we haven't had the time to update all of the files for a formal release for every change. If you prefer not to use CVS, or if you are unable to access the CVS repository for updates, then we recommend you get the latest nightly tarball. This is a copy of the complete CVS repository, in bzip2 format, generated every night.