Aldrin Aldrin - Mar 19, 2012:

Hello I am new member in this forum. My language is Spanish, and I have not much control with the English language, I am using a translator to communicate with you. I apologize if I not understand well.

My first question relates to this thread. I am creating my first chatbot in pandorabots. And I want to change my html page, but I have no knowledge of html.

I tried to copy the source code of the pages that appear in the links, but do not serve me, because my chatbot fails.

Someone can help me create my own html page for my chatbots?.

Aldrin,
At last & hopefully someone is trying to make a bot chatting in Spanish!
First, let me show you some facts that made me run away from AIML

1) Spanish has many inflections (I think too much) and they are used a lot! with AIML you should code too much transformations to catch a verb (70-180 conjugations) there are many complex and irregular verbs, most nouns might beat at least 4 common forms (singular-plural, male-female, all 4 combinations) and some of them have over 30 more uncommon such as diminutive, augumentative, pejorative, and many variants (all combinations) etc.
2) AIML is hardly procedimental, you cannot plan exactly how it should behave because many trasnformations losses information, and without it you cannot ellaborate a good response.
3) AIML lacks NLG (Natural Language Generation) so you must code it apart, and the flow of the replacements is uncertain so you must try to re-code all this jumping out of the model.
4) AIML is incredibliy easy for English, specially because you get easy access to a lot of work done by many botmasters, so you can blend all this into a bot easily and make him chat in a decent way. All the contrary holds for Spanish, there is no good AIML set to start with and if you take an English one and try to translate it, you run hardly out of resources because each “replacement” will multiply into 20-40 at least, so your 40k AIML statement files will convert into >1M statements, barely impossible to hold in memory and give an answer under a second having a very fast processor.
5) Many members of this forum, are very clever and have made excellent jobs with AIML, but I guess, they never tried other languages, and felt into the ‘serious’ AIML limitations I found.
6) ChatScript and RiveScript are evoultions over AIML and they are rather better, you might start from there, but remenber.. they only include improvements useful for English stuff, (WordNEt, and Linguistic functions, etc. )
7) For AIML extensions, many English dictionaries, ontologies etc, are easily and freely available (WN, etc.) as long as Spanish ones are difficult or even impossible (ate least for me) to get something decent and w/o errors.
8) AIML lacks spell correction, it might be introduced externally (as a pre-process) but its painfull not to have it inside the model.
9) AIML has been only slightly improved in 10 years (1.0 and recent 2.0 versions), rather stable.. when everything moves on… is stuck for me (sorry all)
10) My experience (little) with AIML It was rather complicated to hold themes, manage context, set and clear variables and manage many categories without a geometric explosion of the number of AIML statements needed resulting into a very complex way to plan and follow what the f# is going on inside the answer-engine.

All above and more subtle ones, made me run away from it, but not without having learned some good lessons, and they inspired me to create my own engine!

The Upsides are:
1) AIML is free (you get it, run it and wow! its answering..)
2) AIML runs on many platforms
3) AIML is based on XML, easily readable with apropriate tools
4) There are lots of resources like editors, text converters (Q&A into AIML) etc.
5) You can have a bot up and running in no time (but it will be a clone/blend of other bots)

... and may be many more! (I cannot remember they all)

If you can live with it, use it and enjoy!
If not you might need to create create your own engine like I am doing, for the last 5 years. ;(

Despite all,AIML is a good start to see what you might design, then make something better.

good luck!

Andy! Good to see you again. Maybe it would help Aldrin if you also posted your reply (or part of it) in Spanish? Especially since he used translation software to write his original post in English…

(Just a thought. Since moving to Europe I’ve become much more sensitive to the challenge language barriers pose to technical discussions…)

C R Hunt - Mar 26, 2012:

Andy! Good to see you again. Maybe it would help Aldrin if you also posted your reply (or part of it) in Spanish? Especially since he used translation software to write his original post in English…

(Just a thought. Since moving to Europe I’ve become much more sensitive to the challenge language barriers pose to technical discussions…)

Thanks for the warm welcome, hope you enjoy Europa’s cultural offer!

:( I think not a good Idea to post in Spanish because most of the readers here are english-readers (I think most of them) and if Aldrin uses google translate in one way he shold be able to use the other way also!.. I encourage him also to learn English or he will be dropped out while not being able to read important papers and publications, books etc,. which are not translated anywhere.

I think he will understand and I won’t have to make a looooong translation… since in this themes, also I write and explain better in English, just all literature of this NLP stuff IS in English )

In the meantime, I have been improving my AI-NLP dialog platform, allowing it to learn from the experience, while this is complex, my new partner encouraged me to do this tweak, now… lets see how it works… F-Scores seems to be high enough (>90%) by using an agglomerative clustering algorithm for learning!

¿did you see my platform description document?
I can send it to you to show the internal diagram. (it is written in English and only some surrounding explanations are in Spanish.. just google-translate them)

cheers!

Aldrin

BTW if you read Spanish you can go to my homepage and read lot on NLP and chat-agents.
I also wrote a whole NLP platform to build a full Spanish/English chatbot, although its not open-sourced. At the end you can download many documents in PDF to read about NLP and the project itself called: AndyBot. Here is the link:

http://web.fi.uba.ar/~ahohenda

I actually live in Argentina, have a small company called PandoraBox, do AI-NLP Research and teach at the University of Buenos Aires, Facultad de Ingeniería (UBA), where I am currently doing the last part of my Phd-These.

Enjoy!

http://www.pandorabots.com/pandora/talk?botid=fe3102758e348a72

Max - That is an example of the default Pandorabot page and is not what the thread is about.