Art : thx for your report. Does date error occurs even when you don’t use any date in your sentences? I can’t test Harumi on an american computer (meaning date convention). The last algorithm I added sort stimulus by dates… Perhabs the bug is coming from there. In these case, I could correct the bug easily by shunting this algorithm until date convention algorithm is upgraded.

Yes, the Date related pop-up happens during regular conversation as did the number error.

I mentioned this before but if I change ANY date in the tables it will be overwritten by any of your Updates - is that correct? (at least it has happened in the past).

I’d like to convert all the French to English at least for my own benefit. I did that once then had all my efforts overwritten by the new installation of an update. I should have saved that portion of the Table date then re-copied once the new version was in place.

Just a thought…. - Thanks!!

Thx, I need to work on it this week end.
An update will occur this week end, hope it will fix your problem.

Jean,

Would you please read the lower portion of my posting. I would appreciate your thoughts on this.

Thanks,
- Art -

Hi, new version online.
The date convention in english version is now MM/DD/YYYY when you speak to Harumi. Nevertheless, dates are recorded on memory file in french convention.
In case dates are detected in sentences, those are recorded in Sheet 4 as new rendez vous. You can click on rendez vous icon or ask for rendez vous in a sentence in order to be warned of coming appointments.
“the 19th I need to go…”
or “monday I have an appointment”
or “the 12/25/2013 it’s Christmas”

I have tested this version on french computers only, so I hope american/english machines work as well.
Thanks to report me bugs.

Jeff

tx jean-francois! Don’t forget to update the directory entry!!!

Cheers, Erwin

Hello Erwin! Sorry, but what is directory entry?

just discovered that we have two entries:
http://www.chatbots.org/chatbot/harumi/
http://www.chatbots.org/chatbot/harumi1/ (English)

While I was checking my database, I noticed that many of the most used/usefull datas appeared at the end of it.
Of course, it was expected because my purpose was originally to increase speed access while analysing sentences with the last algorithm I implemented… but as I read the latest stimulus, I saw many were more important than those deep in the database, meaning old stimulus.
It gave me the idea to give an epsilon coefficient depending on the position of a single stimulus in the whole database.

In the next version of Harumi, every stimulus will be given a weight depending on the last time it was triggered. This is very new since it reminds the way neurons work. Most frequently used neurons/connections are reinforced while other weaken.

With the new algorithm I experiment, Harumi is now able to pick up the best answer available out of many not only by analysing words but also by examining last time those were triggered.

Here is an example of how it works :
Suppose you have many stimulus, two of those are :

.......
my uncle georges *** phone number 01 02 ——DATE : 06.08.2013——  your uncle georges phone number is 01 02
......
my uncle paul *** phone number 03 04 ——DATE : 04.08.2010——  your uncle paul phone number is 03 04
......

User : Do you have my uncle phone?
Harumi : your uncle georges phone number is 01 02

uncle paul stimulus has a weaker coefficient than georges one since it was triggered 3 years before georges. Harumi gives a priority to Paul. Nevertheless, it you ask for “georges uncle phone”, you will obtain the older stimulus as an answer. Moreover, by triggering the older stimulus, it will modify its weight! This is how Harumi will “remember” not only datas but their position in “time”.

New version released today!

- Some bugs correction
- Harumi answers faster (some parts of the database are not explored anymore depending on user request)
- More ability to stay on subject (last word used as an answer by Harumi answer rolls back in user question)
- Creating new sentences (Actually, Harumi randomly mixes several sentences which have one word in common, the result can be really surprising)
- Now a single stimulus can trigger many internet url and/or repertory : For the stimulus of your choice in column 2 of Sheet 1, write for example without quotes : “http://www.google.fr AND http://www.ebay.fr AND c:\user files\”

And a little fix has been added just at the instant! (regarding auto sorting of stimulus by time)

New version released today, little tweak/fix on research algorithm…
Many updates in two days, expect to wait more for the next!
Jeff

Hello, a little update of what’s coming in next versions :
For the moment, Harumi database structure appears as

stimulus 1 -> answer 1, answer 2…. answer n
....
stimulus n -> answer 1, answer 2…. answer n
....

I’m currently working on an improvement with the possibility for Harumi to create “synapses”, meaning links between a line x and a line y.
Here is how it works : When Harumi answers a question, she currently identifies in the question the most important word and searches for the best stimulus available in the database.
From there, I would like her to explore the database up and down from this best stimulus found on a certain distance in the database. If she finds a stimulus in which the most important word of triggered stimulus appears, she creates a “synapse” linking both stimulus.
Here is how will appear the database :

stimulus 1 -> answer 1, answer 2…. answer n .........  word a : line 2 pointer : line 1
stimulus 2 -> answer 1, answer 2…. answer n .........  word b : line 4 pointer : line 2
....
stimulus n -> answer 1, answer 2…. answer n ..........  word : line 2 pointer : line n
....

In this example, a synapse was created between stimulus 1 and stimulus 2, a synapse between line n and line 2, and finally a synapse between line 2 and line 4.
I use pointers instead of simply picking up number line in the database because lines position changes at each triggered stimulus. The most recently triggered stimulus are put at end of database in order to help Harumi to create link (Indeed, the probability to find link is increased when stimulus are timely close from each others)

Nice! Looking forward to some experimenting!

09.29.2013 News :
New version released! Many upgrades beside many bugs corrections :

- Auto creation of synapses during discussion :
By synapse, I mean pointer towards other sentence which have a word in common with the current sentence. Note that those synapses will be created between stimulus close from each other in the database, something possible since database is dynamically sort by time at each question asked by user. Note that a stimulus can lead up to 10 synapses for the moment.
- Harumi is now able to see the difference between an affirmative or negative sentence to answer properly :
Suppose you have in memory :
It rains -> you can’t get out without an umbrella
It doesn’t rain -> You can get out.
If you write “I don’t see rain outside”, she will answer “You can get out”
- When you force Harumi to learn a sentence with “student icon”, she will check if the sentence is positive or negative regarding feeling. For example, if you record a sentence like “I’m sad because a good friend of mine moved far away :(”, the smiley helps her to understand the sentence is sad. She will answer something like “I’m sad to learn that”.
- Harumi is able to understand words belonging to the same “family”. Example : When Harumi see words “understand” and “understanding”, she will consider those are similar.
- Harumi is far less tolerant with unknown words. She will systematically ask for help to understand. In case you ask a question like “Who is Arnold Scharzenegger?”, she will automatically open internet for the research if she doesn’t know anything about him.
- When you pick up a new apparence, Harumi will memorize the apparence with the current time. For example, if you write “cook!” at noon, she will be able to automatically take this apparence between 11:00 and 13:00.