Chuck, your analogy of the dynamics of volley ball as applied to NLP is well taken.

*Update*
I’ve got a keyword generator functioning now. I copy an article with several thousand words into a text box. I have 1000 common English words in a worksheet.  The program iterates through each word in the text. It counts each word that is not ‘common’ and produces a total count. I’ve got a ratio setup to define the minimum number of occurrences to flag a word as a keyword.

I’ll try tying this into my reader tomorrow that scrapes text from various file formats.  I was going to work on improving the simple grammar parser but I got distracted by this task.

Regards,
Chuck

*Update*
I’ve got code now that iterates through a folder full of XLS, PDF, PPT, and DOC files….extracting all text…comparing it to a list of 1000 common words…and then ranking the remaining words in terms of occurrence frequency.  The code calculates a minimum number of occurrences before the word becomes a ‘keyword’. The keywords are sorted and listed next to the filename.

The next step is to work keywords into a ‘context’ definition One thing that became apparent after testing 100 large files is that even though one can collect keywords….the ‘context word’ is not always in the list. 

It’s sort of like saying “He got a home run!”  “He stole 3rd.”  “The game went into extra innings.” - but I never mention “baseball”. I see I need a feature that will allow me to predefine and to allow for subsequent definitions of a context word.

That’s about it for now.

Oh! I did learn about ‘tag sets’ and English grammar last night. I’ve printed out a document to read tonight.

Regards,
Chuck

*Update*
I’ve conducted about 25 chatbot related experiments/activities this year in C++, VB, and VBA. I’ve gotta say I can do a lot of things now as a result of this research…that I hadn’t done before.

I can parse an article, extract words, eliminate common words, implement pattern matching, program basic human/bot chatting, download data from the internet to feed a virtual world, evaluate grammar, etc.

HOWEVER, I’m in a quandary!  I still don’t have a solution to human-bot dialog that doesn’t rely simply on ‘pattern matching’.  Sure I have ‘variables’ included in responses but I’m not sure that is a significant achievement.

Dialog between two individuals may be very complex. There are often dual meanings, over tones, emotions, sarcasm, etc.  So, I’m spending a lot of time thinking about this.  More later…

Regards,
Chuck

Hi Chuck

You should post the very latest sample conversations with your bot.  Does Walter still ‘twitter’ ?

Hi Vic,
I posted my last session a couple pages back along with an image http://www.chuckbolin.com/walter/xbot.jpg. It shows the questions and responses.  The log is included in that post. I haven’t added any more because I need a more sophisticated model to supplement the existing model. The existing model does two things.

1) Human questions require the bot to extract data and format a response. This works.
2) Plain old pattern matching of human input and bot responses. This works.

However, I starting thinking about extracting meaning and your past references to objects and properties. So I’m working on a simple concept to start testing the creation of objects behind the dialog. Something like.

Human: The boy’s name is Roy.
[behind the scenes]: Create instance of object of type boy. Set name attribute to ‘Roy’.
Bot: Where is Roy?  The bot’s response is driven by the necessity to ‘fill in’ incomplete information.

I’m thinking in this sense I can extract meaning from the dialog. Of course, I need to design a file format where I can define objects, properties, and questions a bot might ask to obtain the data.
This is all limiting because dialog doesn’t just deal with people or things.  There are a lot of interactions between objects.

I’ve got a few experiments planned to test these ideas.

Regards,
Chuck

Yes, that idea presented itself to me as well a few months ago—the “necessity to ‘fill in’ incomplete information.” as you say. 

In fact, if the bot can figure out what it does not know, it can form a response that is a question.

The big challenge of course is something like:

human: I went to New York last weekend
bot: Excellent, did you enjoy it?  —-good: - it knows that you didn’t tell it how you liked it

human: I went to New York last weekend and had a great time!
bot: Excellent, did you enjoy it? —- bad: it didn’t realize you did tell it that you liked it

perhaps if it could at least know that it doesn’t know what “and had a great time!” meant, it could ask.

I think this is the biggest challenge we face, if we can overcome it, we’ll have a great bot!!

Vic,
I’ve been thinking this over for the past 3 days.  I think adults have a strong need to ‘fill in the blanks’.  However I think young children, before they learn to ask ‘why’, are quite content with the information they have. They seem to process it a bit differently.

I told a story to my 5 year-old grandson over the weekend. It was interesting to see him respond to my words, expressions, vocals, etc. He somehow managed to take the story and created something imaginary, without the need for more input. After a while, he stood up on the couch to look out the window to see if the ‘giant boy’ was really stealing the Petunias.  His imagination obviously created a virtual experience that seemed very real.

*Update*
I’ve got some more meat to the design I’ve expressed above. I need a framework to manage the information conveyed during a chat or in a short story.  I’ve purchased several children books this week for the purpose of analyzing.  I’ve started marking them up. In addition I’ve been sketching a “free flowing database” system that may be used in this experiment. I’ll post more later.

Regards,
Chuck

Chuck,

That’s a very good observation.  I think both natural language generation and   comprehension are both powered by imagination !!!!!!

I’ve once heard about US research on the ability or people to think creative:

Children of 2 years, 100% creative
Children of 5 years, 40% creative
Man of 40, only 2% creative!!!!!

I’ve heard it during a conference in Australia about 2 years ago. Unfortunately, I’ve never been able to find the source for more details.

Yes and compare that with the time it takes those age groups to learn a naturual language. . is there a connection ?  I think so!

Erwin Van Lun - Oct 1, 2010:

I’ve once heard about US research on the ability or people to think creative:

Children of 2 years, 100% creative
Children of 5 years, 40% creative
Man of 40, only 2% creative!!!!!

I’ve heard it during a conference in Australia about 2 years ago. Unfortunately, I’ve never been able to find the source for more details.

It wouldn’t surprise me to find that there’s an almost direct correlation between the amount of creativity exhibited and the amount of functional experience the individual displays, with perhaps an “overlap” of a few percentage points between them. Thus, a child of 2 will display a creativity level of 100%, and a functional experience level of 2-3%, whereas our man of 40 exhibits only 2% creativity, and functional experience of nearly 100%.

whereas our man of 40 exhibits only 2% creativity, and functional experience of nearly 100%.

And showing signs that physical functionality is fastly reaching 0% 

PS: previewing this message took about 3 minutes     then second preview was fast all of a sudden

@Jan, offtopic: We’ve run some test proccess that unfortunately drastically decreased performance. So we cancelled them. We need better tools to index Chatbots.org, identify all external links and find ‘bad’ links. The good news: new version of Chatbots.org (2.6.5), to be launched in several weeks, will be a LOT faster! pffff.. Sorry for the inconvenience!

Hi,
Those are some interesting percentages about ‘imagination’ versus ‘functional ability’.  I like this point about age vs. learning a language. That is akin to ‘teaching old dogs new tricks’.

This suggests that a chat bot design should be aimed at creating ‘child-like’ bot framework…as opposed to a ‘mature’ bot. Not sure what that means but it is certainly something worth thinking about.

*Update*
This update is certainly VERY cool…from my perspective. However, I’ve built a chat program that allows two Excel apps on two different PCs to chat. It took me a while to figure out WINSOCK but it is certainly working. This opens up the exciting possibility that I might be able to connect XBot to the internet. All any of you would need would be a chat app that allows you to specify both a port and an IP address. It doesn’t have to be Excel.

I know I’ll have to sort out all the firewall stuff but I’ll save that for later.

Right now I’m going to work on a multi-chat capability within XBot. If I get this working I’ll need a volunteer or two to help me test. I can provide the Excel program.

Regards,
Chuck