Discourse Development Update - Oct 07

We are currently planning two upgrades to our Discourse chatbot engine. The purpose of this note is to inform you of the expected new releases, and to seek your views on some particular changes.

Discourse Release 3.1 - Due Nov/Dec 07

The primary aim of this release is to integrate the MimicBot functionality with Discourse, so that the two systems can share a common code base and data formats.

For Discourse users the main changes will be the introduction of the Learn and AutoLearn chat commands. These enable the chatbot to be programmed from the chat interface, as well as from the normal admin interface. We may also extend the mimic learning mode to Discourse, which allows the bot to learn completely automatically (although this in dangerous in most commercial bots).

Discourse Release 3.2 - Due 1Q08

It is this release that we are seeking comment on. There are two significant changes which we are planning to introduce. These are aimed at making our bots more flexible, but will move them slightly further away from the formal AIML spec (which is itself not moving forward and our changes reflect some of the discussion about the ways in which the spec should move forward).

The two main changes are:

- Pattern Matching Format

We currently use a specific parameter/drop-down to set whether the is defined as a pattern (ie fixed word order), or keyword (ie word order immaterial). Users often forget to correctly set this drop-down, resulting in unexpected behaviour the bot. We therefore intend to move to a "Google" model, i.e.:

- a list of words on their own are treated as a keyword search
- any words enclosed within quotes are handled as fixed word order - ie pattern match

This will do away with the K/P drop-down, but users will have to remember to include the quotes (although we may be able to add Warnings). For commercial users keywords are typically far more common than patterns, so it should not be a significant change.

We will automatically convert existing data.

Note the the K/P drop-down also flags "questions". In future these will be handled as below.

- Context

AIML handles context in a variety of ways, including , and . There has been much discussion on the AIML list about introducing multiple contexts and handling them in a common way. This is what we propose to do. For us the need has been made even greater as one of our bots may find itself operating in multiple environments (eg web and Second Life), and within a virtual world it may be operating in a changing environment (eg a garden, an office), and at different times of the day or week.

The tag/field will be replaced by a tag/field. Within this the user can place a variety of context tags, for instance:

- (as existing)
-
-
-
-
-
-

etc

This should obviate the need in many situations for complex arrangements of conditional tags.

- RDF and Triples

This release will also see additional RDF and triples support.

- Modules

This release will also see more modules aimed at trapping both common chatbot detection ploys, and at answering many basic and common questions using a variety of web based techniques.

Comments

If you have comments on any of these proposed changes please can you email development@daden.co.uk by 30 Oct 07.