Chatbot not replying

[Pavel Savyuk]

Hi All,

I'm trying to use the chatbot (in nlp/chatbot), using the sample dialog provided on the GitHub page for it, and I seem to not be receiving replies. Specifically:

• Whenever I make a Declarative SpeechAct, I get a confirmation (I assume this is normal behavior)
• Whenever I make an Interrogative SpeechAct, I get a list of everything I said as an array (sometimes the reply is instant, sometimes it's delayed by several seconds with CPU usage spiking, so I assume something is going on...)
• Whenever I make a Truth Query, I get "I can't process Truth Query for now"

eva-chatbot behavior is the same, but it correctly responds to the robot manipulation commands (e.g., look left)

What am I doing wrong? I want to get the possibility of a dialog such as in the "Simple OpenCog inference" video @ https://www.youtube.com/watch?v=LduD7Et_cOs 

My commands are below. Thanks for any help!

Terminal #1

pavel@opencog2:~/relex$ ./opencog-server.sh
[...]
Info: Waiting for socket connection

Terminal #2

pavel@opencog2:~/opencog/opencog/nlp/chatbot$ guile -l run-chatbot.scm
[...]
scheme@(guile-user)>

Terminal #3

pavel@opencog2:~$ rlwrap telnet localhost 17004
Trying ::1...
Trying 127.0.0.1...
Connected to localhost.
Escape character is '^]'.
(process-query "Pavel" "Tom eats apples.")
Entering scheme shell; use ^D or a single . on a line by itself to exit.
guile> Hello Pavel, you said: "Tom eats apples."
You made a Declarative SpeechAct
guile> (process-query "Pavel" "What does Tom eat?")
Hello Pavel, you said: "What does Tom eat?"
You made an Interrogative SpeechAct
(Tom eats apples . what does Tom eat ?)
guile> (process-query "Pavel" "What does Tom drink?")
Hello Pavel, you said: "What does Tom drink?"
You made an Interrogative SpeechAct
(Tom eats apples . what does Tom eat ? what does Tom drink ?)
guile> (process-query "Pavel" "What does Tom eat?")
Hello Pavel, you said: "What does Tom eat?"
You made an Interrogative SpeechAct
(Tom eats apples . what does Tom eat ? what does Tom drink ?)
guile>

[Linas Vepstas]

That interface is a mostly broken experimental scaffolding where ... somethings work and most things don't.   I don't think its maintained; I was probably the last person to mess with it, maybe 6 months ago.  I still have plans to do stuff with it, but that's just me and my personal plans.

Meanwhile, there are several other chatbot interfaces maintained by various people. Not sure where they are, what the status is, or who is doing what.  Man Han has been working on "ghost" but don't know if that can be demoed or not.

This dialog segment: Tom eats apples. What does Tom eat? is a typical question-answering system. We used to have one of those, about 8 years ago, and it mostly worked... it probably could have answered your questions. It was abandoned because of multiple reasons, some technical, some political.  You can look at that old code here: https://github.com/opencog/opencog/tree/master/opencog/nlp/chatbot-old/chatbot

That old code is not worth reviving, however, the lessons learned there were very very useful (to me) and helped clarify various core issues surrounding knowleege, common sense and learning. I cannot transmit those lessons to you quickly and easily, (but I can transmit them, with effort).

Short answer: its not hard to create question-answering systems. But that's not where the action is.   On the other hand, we (someone) really should get the basic question-answering scaffolding working again, on opencog. Its a shame that it was allowed to bit-rot.

--linas

--
You received this message because you are subscribed to the Google Groups "opencog" group.
To unsubscribe from this group and stop receiving emails from it, send an email to opencog+unsubscribe@googlegroups.com.
To post to this group, send email to ope...@googlegroups.com.
Visit this group at https://groups.google.com/group/opencog.
To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/5c2d5140-ba98-40d8-ae30-5f0d2a63a47c%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Linas Vepstas]

p.s. That is Man Hin in the youtube video you attached.

p.p.s. I was wrong, I can very quickly explain how the chatbot-old question-answering system worked.  It took IRC, parsed it with link-grammar, and performed a pattern match on the grammatical parse of the question. As long as there was some "fact" in the atomspace (i.e. bunch of atoms organized in a grammatical fashion) then the pattern matcher could find the "answer".  Your demo "Tom eats apples. What does Tom eat?" are grammatically trivial, and so are trivial to handle by grammatical  pattern matching.

The new "ghost" system is kind-of like the old system, but on steroids. It allows content author to author fairly complex answer-response systems.  Note, however, it is still an authoring system. 

I'm not sure if ghost has the ability to remember factual statements or not -- i.e. to do question answering. Man Hin, can it do that?

If it cannot do that, it should .. at the very lowest levels, its really not hard. Its a good parlor trick. Its the handling of the sophisticated stuff that becomes hard.

I don't know if thee are plans for a common-sense database to be hooked into ghost. I don't know if it will be possible to hook PLN into it.

--linas

[Leung Man Hin]

The one being used in the video is chatbot-psi, more details can be found here: https://github.com/opencog/opencog/tree/master/opencog/nlp/chatbot-psi

I'm currently testing & fixing & adding new functions to "ghost", maybe not quite demo-able yet but hopefully it will soon be. It probably will replace chatbot-psi too.

​Remembering factual statements for question answering is possible, as well as calling PLN...

To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CAHrUA34dSKD%3DEEevb5qKj-bak0ju1XZfkN5frUfqq9bfG%3DG9EQ%40mail.gmail.com.