Keywords...

[bobrush]

I have just started using Hal and I am very happy with it so far. My question is:

In the brain editor I added the keywords BECKEY and BECKY and told it to say :Hello Beckey, nice to meet you" Priority level is 1.

When I ask Hal "I would like you to meet Beckey" Hal does not use the comment I requested.

Why isn't he reading the keywords???

[Don Ferguson]

Hello Bobrush,

One thing I thought I'd better check...

You mentioned that you entered the keywords BECKEY and BECKY.  Did you enter those words as single-word keys on two completely separate entries?  That would be the way to do it.

If you entered the two words on the SAME entry, you would have created an "and-logic" situation, in which Hal would have to see BOTH "Beckey" and "Becky" in the same sentence in order to trigger.

It's just a possibility, but I thought that I would mention it.

I tested the action you're looking for by inserting the following two lines of code into my Ultra Hal in the "Enhanced_Main.brn" database:

" BECKY ","",""," Hello Becky! ","","",1,#FALSE#,#FALSE#,""
" BECKEY ","",""," Hello Beckey! ","","",1,#FALSE#,#FALSE#,""

When I say anything to UltraHal with Becky's name, Hal says "Hello Becky," so the routine does work.

Please let me know if this fixes the problem!

Sincerely,

Don

[bobrush]

You are good........thank you. I now understand how Hal learns so I now know how to teach him! Another learning question....

When Hal reads text to learn how does Hal work that info into conversation or does he just wait to be asked?

[Don Ferguson]

Hi,

The "learn from text" feature takes each individual sentence from a text file and formats the sentences into individual entries in user database files.  By looking in Hal's default directory, and sorting the list for "newest modified," you should be able to discover the specific files to which Hal is appending.

The strength of this routine is that large numbers of sentences can be processed quickly into Hal's memory.

The weakness of this method is that the current routine doesn't do all the processing that Hal does during a conversation.  For instance, Hal doesn't relate sentence 9 back to sentence 8, even if sentence 9 contains a lot of pronouns, and making that link would make sense (Hal can do this during conversation).  Therefore, you are more likely to get "out of context" responses if Hal is full of "learn from text" materials.

There has been some talk of making the "learn from text" function operate with all the conversational learning logic "on," but the new versions and new features from Zabaware have been coming so fast and furious over the years, I think that issue has been on the back burner.  (I started working with Hal at 2.0 and thought it was terrific, and 3.0 was a drastic improvement, 4.0 was amazing, 4.5 was a major upgrade, and now 5.0 is out today with features that I wouldn't have ever imagined!)  If the "learn from text" got changed in the future so that it took each conversation through the entire list of conversational processes, it would run much SLOWER, but would give you more sophisticated linkages and analysis of the sentences.

Now, how does Hal use the information?  It is stored among all the other databases at more-or-less "medium" priority, which means that Hal will consider the material along with everything else he "knows," and it will come out in conversation sometimes, but sometimes not.  It will be available at all times in conversations with that user.  (Remember that most of Hal's learning is put into databases that are specific to a user, so that Hal has a different viewpoint when speaking to different users with different names.)

I hope this is helpful, and I hope that you continue to teach Hal and enjoy Hal!

Sincerely,

Don

[bobrush]

Don.....U DA MAN....Thanks again.