FRDCSA:

UniLang

Brief: Stream of consciousness multiagent system
Jump To: Parent Description
Code: GitHub

UniLang is a system for providing the services you would expect from a Multi-Agent System (MAS), and more. It allows IPC between different Agents connecting through sockets, and also has packages which client Agents can use to do most of the work, based on Event.pm.

From a users perspective, the UniLang-Client Agent allows us to communicate with a wide variety of agents through a single text based client, from inside the provided Emacs major-mode called ushell.

Ushell now incorporates a bidirectional interface between Emacs and UniLang, allowing us to greatly enhance integration between Emacs and all other systems. For instance, the long awaited Emacs Agenda system was completed with just one function - a call to PSE from within Emacs through UniLang.

Automatic classification of commands is working (see corpus). Therefore, one can simply write what they are thinking and UniLang will dispatch this to the proper agent.

Here is an example of automatically classified commands. Note that the classifier, which usually works well, is not giving accurate classifications for some as simple but unknown reason.
```
The classifications are not correct, but, whatever.
	[icodebase-capability-request, capability-radar, capability-unilang, solution-to-extant-problem, goal]

I think we should have some kind of autoformatting for UniLang.
	[capability-manager, unilang-specific-message, observation, shopping-list-item, complex-statement]

Great - corpus is working well, if a bit slow.
	[shopping-list-item, complex-statement, capability-manager, unilang-specific-message, observation]

Now all we have to do is fix that data and worry about complexity/efficiency issues with Corpus doing autoclassification.  We also have to add more elegant classification than bayes, as well as talking amongst agents to support better categorization and command execution from our notes.
	[shopping-list-item, complex-statement]

I wonder how much memory this takes up.
	[icodebase-capability-request, goal, solution-to-extant-problem, capability-radar, capability-unilang]

Maybe there's a way to save classifications models so the whole thing need not be reloaded.
	[complex-statement, icodebase-capability-request, capability-radar, capability-unilang, solution-to-extant-problem]

UniLang agents should, defined through Audience, have a nominal state that allows UniLang while dynamically starting them upon messages received for them (so they don't all have to start at once), that indicates they are ready for processing.
	[goal, icodebase-capability-request, capability-radar, solution-to-extant-problem, capability-unilang]

Should define a general purpose classification scheme - so that the system can get really specific about what it just classified.
	[capability-manager, shopping-list-item, complex-statement, observation, icodebase-capability-request]

Complex-statement for instance can be checked for using get_sentences
	[observation, shopping-list-item, complex-statement, capability-manager, unilang-specific-message]

Debugging UniLang is rather difficult and we don't want to fall prey to the same problem as with GNU Hurd.
	[capability-manager, goal, observation, shopping-list-item, complex-statement]

Now we have enough demos to definitely get people working on our projects.  Mike is right that I should learn persuasion.
	[shopping-list-item, complex-statement, unilang-specific-message, capability-manager, observation]

We have to hurry, though.
	[unilang-specific-message, observation, complex-statement, shopping-list-item, icodebase-capability-request]

You know, should set up regression testing soon.
	[shopping-list-item, complex-statement, capability-manager, unilang-specific-message]
```
Here is a random example of typical UniLang-Client traffic from the log file. Note that this is only traffic from the user, not the other agents. I have annotated them with very rough sample classifications. Often, one entry will initiate or continue a great amount of activity.
- (PSE - add to BOSS todo) Replace all the uses of use Package qw ( F1 F2 ... ) and simply export these functions instead.
- (PSE - add to BOSS todo) Document all programs soon, since I am already forgetting what they do.
- (Bugzilla - file bug on RADAR) Status: Available through apt-get, yet sudo apt-get install libnet-google-perl says its already the latest.
- (PSE - add to RADAR and Predator todos) Compute error probabilities for every phase of the RADAR-Predator system.
- (BOSS - add to Machiavelli's requirements) Machiavelli should also try to estimate what other people are working on, so that we can be sure not to redouble efforts.
- (BOSS - add to RADAR's requirements) Incorporate that special stuff for tracking upstream that I read in Wed Aug 4 21:23:53 EDT 2004 debian weekly news.
- (MyFRDCSA - search capabilities) What are our video editing options?
- (BOSS - add to Predator's requirements) automatic analysis of whether a package belongs to free/non-free, etc.
- (Machiavelli - assert related group) http://acmsoft.org/ looks to be in a similar position to us.
- (Audience - add to Eric's queue) send to Eric http://www.nitle.org/semantic_search.php
- (BOSS - add to FRDCSA-website requirements) Make a website which poses questions, and allows people to answer them. For instance, What is a good system that acts like a personal wayback machine?
Currently, the corpus system still is mainly responsible for the classification of these logs. Additionally, we are working on adding translation capabilities between the various languages encountered, hence the name, UniLang. We will utilize a great many tools for this from the knowledge representation, linguistic ontological engineering, and machine translation fields. Examples forthcoming. Humorous unforeseen interaction.
```
  Lord, don't let me fall, don't let me FALL!
1UniLangUniLang-ClientMon Oct  2 22:41:05 CDT 2006No one here by that name.
```