Virtual Humans Forum
Virtual Humans Forum
Home | Profile | Register | Active Topics | Members | Search | FAQ
Username:
Password:
Save Password
Forgot your Password?

Virtual Humans
 All Forums
 Virtual Humans
 Artificial Intelligence
 Link Grammar API
 New Topic  Reply to Topic
 Printer Friendly
Previous Page
Author Previous Topic Topic Next Topic
Page: of 3

abstemious
Curious Member



5 Posts

Posted - May 31 2009 :  19:48:50  Show Profile  Reply with Quote
Has nothing else happened here, or am I in the wrong place?

The Link Grammar Parser is actively maintained and enhanced at http://www.abisource.com/projects/link-grammar/

There is a sibling project built on top of it: ReLex extracts semantic relationships from the parser linkages. http://opencog.org/wiki/RelEx

I think a chatbot that is of the quality of Zabaware or Chatterbot contenders has to be built on techniques at least as advanced as LGP and ReLex. AIML was nice for a start, but does NOT scale. IMHO writing a zillion Visual Basic routines will not be the final solution. Natural languages have so many ways to express things that pattern processors are required for any decent coverage, but patterns on letters (regular expressions) are not enough. Regular expressions on words aren't, either.

LGP essentially matches patterns on word senses (additionally informed by information on part of speech, tense, etc) to expose the structure of an utterance.

ReLex finds patterns of structures.

There are several more layers needed above that, probably including: capturing implications of user utterances in an ontology-based database, logical inference to answer questions from that data, dialog management that simulates how humans take turns, and an AI planner to manage and create goals. The good news is that a lot of that exists, but most researchers do not have the resources to extend their approaches to a practical scale.

This is a long-winded way of saying that I think that a good-quality open source chatbot will require the collaborative efforts of many (hundreds?) of non-programmers who are willing to learn a somewhat arcane syntax and contribute to (and debug!) the resultant massive rulebases.

Representation (the notation) matters: You cannot describe the workings of an automobile engine in terms of quantum mechanics.

Although a proprietary system, Suzette by Bruce Wilcox seems to have the right kind of approach on notation.
http://www.gamasutra.com/view/feature3761/beyond_aiml_chatbots_102.php
Also see related posts on http://aidreams.co.uk/forum
Go to Top of Page

Nomeneiste
Intrigued Member



USA
121 Posts

Posted - Sep 10 2009 :  21:33:09  Show Profile  Click to see Nomeneiste's MSN Messenger address  Reply with Quote
Mik,

Did you continue work on this LinkDLLDemo? Can you share what you learned from it, if anything?

Thanks!
Brandon

- The answer is oorple.
Go to Top of Page

mikmoth
Dedicated Member



1624 Posts

Posted - Sep 12 2009 :  20:49:55  Show Profile  Visit mikmoth's Homepage  Reply with Quote
Yeah, I still have it. The last time I used it was for grammar correction. I use markov chains to generate random sentences. Then I pass them through the DLL and 90% of the time it's able to weed out grammaticaly incorrect sentences. If the sentence is correct it will return an output - otherwise it wont. It's not full proof tho.

http://www.lhandslide.com/
http://www.airpoof.com/


Me: It's so nice to be Home God.
God: Where'd you go? You've been sitting here the whole time.
Home: I have? Oh...
Go to Top of Page

Nomeneiste
Intrigued Member



USA
121 Posts

Posted - Sep 22 2009 :  22:13:48  Show Profile  Click to see Nomeneiste's MSN Messenger address  Reply with Quote
Good morning,

Very few things are fool proof, from my experience but it is exciting (to me) that the ability for an AI to read and grasp what it is reading is not terribly far off. This is the main reason why I am very excited about the prospect of the DLL.

Thanks!
Brandon

- The answer is oorple.
Go to Top of Page

lcutrone
Curious Member



1 Posts

Posted - Nov 12 2009 :  04:14:08  Show Profile
quote:
Originally posted by drgold

tiny_g, two things: First, rather than going old-school with Visual Studio 1.0, I would highly recommend downloading VB.NET 2008 Express at http://www.microsoft.com/express/vb/Default.aspx ... Microsoft has finally made it completely free.

Second, although I once heard tell that some version of the Link Grammar API was supposed to be accessible from VB.NET, I could never get it to work. What you *can* do, however, is have your code secretly launch the compiled version of the Link Grammar parser in the background (sending it whatever sentence you want it to parse), redirect its output to a file, and then have your code read in that file. A little clunky, and it takes a few moments to load the parser anew each time you wanna parse a sentence (meaning short sentences take 1 to 2 seconds to parse on my computer)... but it works, and it all happens invisibly without the user's knowledge.

I'm feeling charitable today, so I wrote a VB.NET project that does just that, which you can d/l at http://floatingcontinent.com/Linky.zip Probably only works in VB.NET 2005 or higher, so probably won't open in your VS1.0... but again, VB.NET 2008 is free




p.s. The 10000 in the code gives it a 10-second time-out, so if the link grammar parser is taking more than 10 seconds on a sentence, it'll cancel the operation... so if it seems to be hanging for an uncomfortably long time on some sentences, you might want to lower that number.




I know this post is a bit old, but I would be interested in your linky.zip file. The link is no longer valid. Are you still feeling charitible???
Go to Top of Page
Page: of 3 Previous Topic Topic Next Topic  
Previous Page
 New Topic  Reply to Topic
 Printer Friendly
Jump To:
Virtual Humans Forum © V.R.Consulting Go To Top Of Page
This page was generated in 0.33 seconds. Snitz Forums 2000