AI Zone

history of Suzette and issues in chatlanguage design

2011-07-02T14:23:05-06:00

here is a blog/paper I wrote for gamasutra.com (posted into my papers on this site).
http://www.gamasutra.com/blogs/BruceWilcox/20110622/7840/Suzette_the_Most_Human_Computer.php

2012 CBC - $1750 in Prize Money and Award Medals

2012-01-11T13:12:40-06:00

Decadent Vapours for the second time will sponsors the 2012 CBC.
Peter Cole, botmaster of Brother Jerome, through his company Decadent Vapours, has
agreed to sponsor the 2012 CBC. They will be awarding $1750 to the winning bots along
with award medals to the 1st, 2nd, and 3rd place winners. Also a medal will be awarded
to the “Best New Bot”.

1st Place - $1000
2nd Place - $500
3rd Place - $250

Enter your bot today at the contest website. Free to Enter!

http://www.chatterboxchallenge.com/

^reuse() between different files

2012-02-10T05:47:57-06:00

I’ve decided to incorporate multiple files for my chatbot (i.e. introductions.top, books.top, movies.top etc.) and I was wandering if ^reuse() can be used across files.

I see that you can use it across topics by using ^reuse(~TOPICNAME.LABEL) within the same file. When I attempted to do this for a topic in another file, I get ‘Missing cross-topic label’ error during build.

Is there a way to reference a rule from another file using ^reuse(), or are they meant to be entirely separate?

Cheers.

questions and answers in the chatbot designer

2011-12-01T13:15:01-06:00

I’ve put up a new tutorial on how to handle questions and answers by using context and memory with the chatbot designer. It also demonstrates a technique that allows you to declare general patterns, like:


~subject (has|have|ve) ~object [.]

and have content specific questions (and replies), by attaching topics to thesaurus items and assets (abstract vs concrete). All pretty cool stuff, very flexible.

Skype versus Avatar platform

2012-02-07T12:37:59-06:00

I’ve been racking my brain for quite some time trying to hit on the ideal avatar platform.

However, I’m concluding that Skype, or any VoIP videoconferencing platform, would be best.

Does anyone have Denise or Ultra Hal configured for incoming Skype calls? If so, I’d love to check it out!

Who needs to deploy a mobile app, or even web app, when you can just Skype the chatbot?
I would love to hear from anyone with experience doing this!

Platform Question Concerning Browser Based Bots.

2012-02-09T15:20:58-06:00

On JavaScript, I plan on porting mine to a Windows based application for the Loebner Prize. The only issue would be not the Internet access but access to a browser API on the hosting machine (Internet Explorer preferred. ) that would hosting the competition.

Does anyone know the platform involved? Windows? Apache? Linux?

Thanks!

GRACE/CLUES

2011-01-27T11:39:07-06:00

Small copy and paste of first I/O from the project:

victor@r2d2:/clues$ ./clues

GRACE version 0.1 (on CLUES version 0.8)

<user says> What did Jack do

<Grace says> Sorry, I wasn't able to come up with an answer to that question.

<user says> Jack went to his closet and took out his new suit

Yes, I believe I understand!

<user says> What did Jack do

<Grace says> jack went to his closet and took out his new suit

<user says> What did Jefferson Davis Hogg do

<Grace says> Sorry, I wasn't able to come up with an answer to that question.

<user says> Jefferson davis hogg went to his closet and took out his new suit

Yes, I believe I understand!

<user says> what did jefferson davis hogg do

<Grace says> jefferson davis hogg went to his closet and took out his new suit

<user says>

*Note 1 - this very first sample I/O does not illustrate the power of the engine, since a very simple script with regular expressions could do this.

*Note 2 - the above is NOT using simple templates/pattern matching. Believe it or not, the bot has a complete understanding of what was said by first generating many parse trees… for *BOTH* the ‘fact’ *and* the question.

Among many parse trees, the chosen ‘fact’ parse tree Grace chose was:

pos = simple-sentence
subject.num-noun = 1
subject.noun.1.val = jack
num-predicate = 2
predicate.1.num-verb = 1
predicate.1.verb.1.val = went
predicate.1.verb.1.num-prep-phrase = 1
predicate.1.verb.1.prep-phrase.1.num-prep = 1
predicate.1.verb.1.prep-phrase.1.prep.1.val = to
predicate.1.verb.1.prep-phrase.1.num-noun = 1
predicate.1.verb.1.prep-phrase.1.noun.1.val = closet
predicate.1.verb.1.prep-phrase.1.noun.1.num-adjective = 1
predicate.1.verb.1.prep-phrase.1.noun.1.adjective.1.val = his
predicate.2.num-verb = 1
predicate.2.verb.1.val = took
predicate.2.verb.1.num-prep-phrase = 1
predicate.2.verb.1.prep-phrase.1.num-prep = 1
predicate.2.verb.1.prep-phrase.1.prep.1.val = out
predicate.2.verb.1.prep-phrase.1.num-noun = 1
predicate.2.verb.1.prep-phrase.1.noun.1.val = suit
predicate.2.verb.1.prep-phrase.1.noun.1.num-adjective = 2
predicate.2.verb.1.prep-phrase.1.noun.1.adjective.2.val = new
predicate.2.verb.1.prep-phrase.1.noun.1.adjective.1.val = his
predicate-list-type = and 

Now, also for the question input, Grace also generates parse trees, and, again, using semantic inference, decides which parse tree the user really meant…

pos = simple-sentence
num-question = 1
question.1.val = what
subject.noun.1.val = jack
subject.num-noun = 1
predicate.1.num-verb = 1
predicate.1.verb.1.val = do
num-predicate = 1
predicate.1.verb.1.num-auxiliary-verb = 1
predicate.1.verb.1.auxiliary-verb.1.val = did

So with the “QPT” (question parse tree), Grace goes in and tries to find a “FTP” (fact parse tree) which can answer the question.

In the above, Grace matches the subject (subject.noun.1.val) in the FPT which is “jack”, with the subject mentioned in the QPT.

Since this is all done with FULL grammatical and semantic inference, and not by simple pattern matching, the system will evolve, and learn, to cope with extremely complex sentences and ambiguity.

For example, in the above, note that Grace knows there is one question, but we may ask : “Who or why did….” (in that case, num-question would be 2).

Next, I will be working on tests like:

“Jack went to his closet and took out his new suit because he was going to a dance given by his company”

and ask:

Where was jack going?
Did jack take out his new suit? why?
Where did jack go?
Did jack go to a dance?
Has jack ever been at a dance?

Stay tuned ! Much work ahead of me! !

Getting a Chatbot on Companies Facebook

2012-02-03T09:15:23-06:00

How do I get a chatbot on my companies facebook page and does it work well for customers to use?

Just For Fun

2011-05-21T15:51:37-06:00

As an A.I. Researcher,

what do you do Just For Fun ?

http://www.79b.org/text/

Sparking Together

2012-02-01T08:49:09-06:00

> http://www.sparkingtogether.com/

Does anyone have a clue what this new “Sparking Together” initiative, apparently from Adele Robots (AdeleRobots.com), is all about ??

They’ve got Twitter, Facebook, and some really wacky videos on YouTube ..

> http://twitter.com/Sparking2gether

> http://www.facebook.com/pages/Sparking-Together/180513272042574

> http://www.youtube.com/user/SparkingTogether

They seem to be trying to pull together some kind of new robot community ..

This reminds me of the recent “Project Aigo” from Peter Voss of AdaptiveAI.com to solicit volunteers to help build up their domain specific knowledgebases and ontologies ..

> http://tech.groups.yahoo.com/group/projectAigo

Are things starting to break out all over ?? ;^)