Working together to build a conversational Verbot

[MikeA]

I have been thinking about this for a while.

Would anyone be interested in creating a 'group' conversational bot, perhaps with a view to putting it online, entering in a few of the bot competitions, etc?

With a small group probably the advancements will come much quicker. It's also a chance to see how much we could push the capabilities of the Verbot.

[JonC]

Yup,
you can probably count me in on this (but don't ask me to do anything with the "learn" rules!).
Would certainly be interested in knowning more fully what you had in mind.

[MikeA]

I'm thinking about something that is an extra to our individual projects. I reckon we all go about this a slightly different way, and all probably got something unique to add.

Therefore a combined effort would benefit from both extended ideas plus more input. Perhaps it would need to be online (so we could all see it), which means hosting csv elsewhere (which I could do). Although for simplicity I think being compatible with the normal Verbot stuff (editor and player) makes sense.

I'm quite interested in having a bot to enter for one of these competitions, but I don't think I would have the motivation to do that alone. My own projects are not aiming at that type of bot. Still, I think it would be fun to try.

[Mapestone]

I...push the capabilities of the Verbot.

I'm still learning. Count me in, for what it's worth. I'd like to see a conversational verbot that takes advantage of its' weaknesses ~ "I'm just a chatterbot, but I'm a cute little chatterbot." ~ That style of scripting takes talent - I'm short on that too.

[chatterbox-challenge]

Well, I hope you all can pursue this. It was disappointing to see no Verbots enter the 2012 CBC. I know contest aren't for everybody and many Verbots are made for personal use only but it would be nice to see a few entries give it a try. To me entering a contest like the CBC is pretty fun and the $1000, a nice medal and bragging rights are all a plus.

WC

[MikeA]

Yup,
you can probably count me in on this (but don't ask me to do anything with the "learn" rules!).
Would certainly be interested in knowning more fully what you had in mind.

I...push the capabilities of the Verbot.

I'm still learning. Count me in, for what it's worth. I'd like to see a conversational verbot that takes advantage of its' weaknesses ~ "I'm just a chatterbot, but I'm a cute little chatterbot." ~ That style of scripting takes talent - I'm short on that too.

I'm still up for this. And I agree that keeping things Verbot(ish) makes sense. Having the bot online seems logical, so the main question is if we want to include CSV with this. As the online bots can't upload CSV files, we could either work strictly within Verbot's intrinsic capabilities, or host the CSV elsewhere.

I've also got a SDK version working with an SQL database, and could makes changes to that for CSV, but probably, given that that project depends a lot on my c# skills (which are beginner-ish) then it might not be the best approach.

One advantage with keeping things purely vkb is that we could download and test locally with no changes needed. I'm quite interested in pushing the limits of rule input / outputs, conditions, the VerbotStandard vcm (onbeforerulefired, etc), synonyms and replacement profiles.

I'm happy to get the online verbot going and cover the (very minimal) cost of it.

[Jerbot]

I have some ideas. I'm interested to see what we're working with right now, though. If there's no hard code, maybe there's an ERD or something? I could fill in some things or make another one on a tangent.

[Mapestone]

You are the ringleader MikeA. What next?
We all have a few ideas. Even me.

[MikeA]

Ok... great to see a few people interested. Hi jarbot.

I haven't done anything yet, but I'll register an online Verbot (I had one before but stopped using it).

This will mean an easy to access bot, although there is some limit on kbase size, so we'd have to deal with that when we came to it.

I've got an online SDK verbot working with a standard knowledge base and database,but the only problem with this route is that it'd depend on my c# being up to scratch.... and wouldn't like to rely on that!

It would be interesting to push an online bot to its max and then see what's next.....

[Mapestone]

Pass out some assignments, MikeA. ...What's your plans for the database?

[MikeA]

Sorry haven't been around..... had to focus on work to make sure the bills were paid!

Re. a database, I guess that depends on how we want to go with it. Personally I think it might be sensible to avoid that part of it for the time being, focusing more on what Verbot naturally does. What do you all think?

Options for a working area are either 1) Verbots Online Free, 2) Verbot Online Paid, 3) my SDK version.

My SDK version is actually two versions. 1 is used for the test site I setup (link in other thread) that has a Julia and Sylivia running. However, this version doesn't recognised <send > used in rules.

Another version gets aroun this by manually checking the output for <sends >. However this still means currently issues with other tags.

Therefore, I think a better start is to keep to normal standards, and use Verbot Online. The Paid account has one massive advantage - chat logs, it's only $30 a year (or $5 a month) and chat logs are really useful. I'm happy to pay for this for the year.

Using Verbot Online means no CSV or other database directly (because only kbs can be uploaded). However, it is possible to host the database side elsewhere and link to it. However, my suggestion is not doing this at the beginning, but pushing the standard Verbot to it's limits, then once there, seeing what' else might be needed (if at all).

At least this way, we are using the one common strand of knowledge - Verbot coding. JonC is the csv expert, I've got a good grip on sql so perhaps worth discussing something like this further down the line if needed.

What I would be very interested in, is seeing just what a standard Verbot could do using the rules, synonyms, code modules and the VerbotStandard (OnafterRuleFired, etc) to see how far that can go.

[Mapestone]

... had to focus on work to make sure the bills were paid!... What I would be very interested in, is seeing just what a standard Verbot could do ...see how far that can go.

I hear ya, bills first. Also, it makes sense to start at the beginning. Push it to the limits and then fill in the gaps. We might have enough raw material to establish a baseline and a set of quality tests. For quality tests I've thrown together a file of user inputs called SmallTalk: What did you do yesterday? Have any interesting dreams lately? Are you married? Have any kids? Do you like animals? ... I'm surprised at how poorly many chatbots perform, but I haven't established a benchmark.
Your thoughts MikeA?

[MikeA]

For quality tests I've thrown together a file of user inputs called SmallTalk: What did you do yesterday? Have any interesting dreams lately? Are you married? Have any kids? Do you like animals? ... I'm surprised at how poorly many chatbots perform, but I haven't established a benchmark.
Your thoughts MikeA?

That sounds like an excellent idea. Yes, I've noticed small talk shows up a bots weakness. I think a lot of botmasters focus on being clever, knowledgable or funny, rather than being good with small talk.

I've been compiling a list of common questions, along the same lines, but more topic based. I initially started doing this for the setting up of a database bot, but the same questions could be used for this.

I'm also keen to see how, within the confines of Verbot, the conversation can be taken beyound 1 question / 1 answer. This, too me is probably the biggest let-down of some o the more established bots. You often get conversations similar to this:

User:Do you like fruit?
Bot: Yes, I like apples very much.

User: Why?
Bot: Why is a word I am often asked, perhaps you can tell me why?

whereas what would be much better would be something like:

Bot: Because apples are juicy and sweet.

This is leads a bit to your semantic networks thread. Also, I think this can also be done by either keeping within a parent rule or keeping a 'subject' variable, or a combination of all these ideas.

[Mapestone]

... beyond 1 question / 1 answer. ... this can also be done by either keeping within a parent rule or keeping a 'subject' variable, or a combination of all these ideas.

A.L.I.C.E. uses a <topic> tag, though not often seen in the base AIML set. The Verbot Rule Input Condition can be used in a similar fashion. Is there an exceptional chatbot on the web that performs better than most in regard to keeping on topic?

[MikeA]

... beyond 1 question / 1 answer. ... this can also be done by either keeping within a parent rule or keeping a 'subject' variable, or a combination of all these ideas.

A.L.I.C.E. uses a <topic> tag, though not often seen in the base AIML set. The Verbot Rule Input Condition can be used in a similar fashion. Is there an exceptional chatbot on the web that performs better than most in regard to keeping on topic?

I've come across discussions / claims about this ability, but can't say I've ever really seen it. Actually, this is what I'm working on at the moment, using a database-driven verbot. But it really has taken things in a different direction.

Looking again at your semantic network, I wonder if there would be a way to do this using rules (a kb) rather than synonyms. I wonder if this could somehow allow for the forwards / backwards referencing (which synonyms definitely can't do as is). When I get the chance I'll look more at that. I did, ages ago, set up a Verbot that used rules in the same way as a CSV database (by storing the CSV text string as a rule output).

re. the building of the bot, over the next couple of days I'll get an online verbot sorted out that we can access.