[agi] NL interface
I'm planning to write an NL interface that uses templates to eliminate parsing and thus achieve 100% accuracy for a restricted subset of English (for example, asking the user to disambiguate parts of speech, syntax etc). It seems that such a program doesn't exist yet. It looks like AGI-level NLP is going to take some time to achieve, so in the meantime we can use this short-cut. I'm looking for someone to develop this jointly. Any takers? Comments? YKY - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=78286483-81b9f2
Re: [agi] NL interface
Hi YKY, I hope that by this time next year the Texai project will have a robust English parser suitable for your project. I am working in collaboration with the Air Force Research Laboratory's Synthetic Teammate Project http://texai.org/blog/2007/12/13/collaboration-initiated-with-the-synthetic-teammate-project/ Our work will be licensed GPL. Stay tuned. -Steve Stephen L. Reed Artificial Intelligence Researcher http://texai.org/blog http://texai.org 3008 Oak Crest Ave. Austin, Texas, USA 78704 512.791.7860 - Original Message From: YKY (Yan King Yin) <[EMAIL PROTECTED]> To: agi@v2.listbox.com Sent: Thursday, December 20, 2007 8:32:56 PM Subject: [agi] NL interface I'm planning to write an NL interface that uses templates to eliminate parsing and thus achieve 100% accuracy for a restricted subset of English (for example, asking the user to disambiguate parts of speech, syntax etc). It seems that such a program doesn't exist yet. It looks like AGI-level NLP is going to take some time to achieve, so in the meantime we can use this short-cut. I'm looking for someone to develop this jointly. Any takers? Comments? YKY This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?&; Looking for last minute shopping deals? Find them fast with Yahoo! Search. http://tools.search.yahoo.com/newsearch/category.php?category=shopping - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=78299398-0c7251
Re: [agi] NL interface
On 12/21/07, Stephen Reed <[EMAIL PROTECTED]> wrote: > > Hi YKY, > > I hope that by this time next year the Texai project will have a robust English parser suitable for your project. I am working in collaboration with the Air Force Research Laboratory's Synthetic Teammate Project http://texai.org/blog/2007/12/13/collaboration-initiated-with-the-synthetic-teammate-project/ > > Our work will be licensed GPL. Some questions: 1. does it output logical formulae? 2. is it interactive? 3. can it be used as a library (preferably under LGPL)? YKY - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=78303015-14de5a
Re: [agi] NL interface
- Original Message From: YKY (Yan King Yin) <[EMAIL PROTECTED]> To: agi@v2.listbox.com Sent: Thursday, December 20, 2007 9:41:10 PM Subject: Re: [agi] NL interface On 12/21/07, Stephen Reed <[EMAIL PROTECTED]> wrote: > > Hi YKY, > > I hope that by this time next year the Texai project will have a robust > English parser suitable for your project. I am working in collaboration with > the Air Force Research Laboratory's Synthetic Teammate Project > http://texai.org/blog/2007/12/13/collaboration-initiated-with-the-synthetic-teammate-project/ > > Our work will be licensed GPL. Some questions: 1. does it output logical formulae? Yes, here is my initial use case "the block is on the table" ===> extracted meanings: (<http://sw.cyc.com/2006/07/27/cyc/equals> ?obj-20 ?obj-18) (<http://sw.cyc.com/2006/07/27/cyc/equals> ?obj-5 ?obj-3) (<http://sw.cyc.com/2006/07/27/cyc/properSubSituations> ?situation-localized-11 ?on-situation-localized-15) (<http://sw.cyc.com/2006/07/27/cyc/situationConstituents> ?situation-localized-11 ?obj-3) (<http://www.w3.org/1999/02/22-rdf-syntax-ns#type> ?obj-18 PreviouslyIntroducedInThisDiscourse) (<http://www.w3.org/1999/02/22-rdf-syntax-ns#type> ?obj-20 <http://sw.cyc.com/2006/07/27/cyc/Table>) (<http://www.w3.org/1999/02/22-rdf-syntax-ns#type> ?obj-3 PreviouslyIntroducedInThisDiscourse) (<http://www.w3.org/1999/02/22-rdf-syntax-ns#type> ?obj-5 <http://sw.cyc.com/2006/07/27/cyc/Book>) (<http://www.w3.org/1999/02/22-rdf-syntax-ns#type> ?on-situation-localized-15 On_SituationLocalized) (<http://www.w3.org/1999/02/22-rdf-syntax-ns#type> ?situation-localized-11 <http://sw.cyc.com/2006/07/27/cyc/Situation-Localized>) (aboveObject ?on-situation-localized-15 ?obj5) (belowObject ?on-situation-localized-15 ?obj-20) (fcgDiscourse-role ?obj-20 external) (fcgDiscourse-role ?obj-5 external) (fcgStatus ?obj-20 single-object) (fcgStatus ?obj-5 single-object) The above propositions have terms expressed in RDF, but are presented in the lispy fashion desired by the original Fluid Construction Grammar implementers. ( ). Note that I include discourse axioms (e.g. PreviouslyIntroducedInThisDiscourse). This grammar is reversible. My next task is to generate the expected sentence given these meaning axioms as input. 2. is it interactive? Yes, my system will be dialog from the start, and Jabber compatible. 3. can it be used as a library (preferably under LGPL)? For your needs, I will dual license it - LGPL for you, GPL otherwise. Understand that I may include third-party Java GPL components that you will have to deal with. Because one is not allowed simply to load a GPL class file into a VM running an LGPL application that you have published, the current practice is to proxy the GPL Java library with a server running in a separate VM and messaging somehow to it. -Steve Stephen L. Reed Artificial Intelligence Researcher http://texai.org/blog http://texai.org 3008 Oak Crest Ave. Austin, Texas, USA 78704 512.791.7860 Never miss a thing. Make Yahoo your home page. http://www.yahoo.com/r/hs - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=78314479-a22df2
Re: [agi] NL interface
On 12/21/07, Stephen Reed <[EMAIL PROTECTED]> wrote: > The above propositions have terms expressed in RDF, but are presented in the lispy fashion desired by the original Fluid Construction Grammar implementers. ( ). Note that I include discourse axioms (e.g. PreviouslyIntroducedInThisDiscourse). This grammar is reversible. My next task is to generate the expected sentence given these meaning axioms as input. Must it be used with Cyc? I see there're Cyc predicates in the output. Integrating it with an AGI other than Cyc might need some extra work... YKY - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=78354320-59d15c
Re: [agi] NL interface
--- "YKY (Yan King Yin)" <[EMAIL PROTECTED]> wrote: > I'm planning to write an NL interface that uses templates to eliminate > parsing and thus achieve 100% accuracy for a restricted subset of English > (for example, asking the user to disambiguate parts of speech, syntax etc). > It seems that such a program doesn't exist yet. > > It looks like AGI-level NLP is going to take some time to achieve, so in the > meantime we can use this short-cut. > > I'm looking for someone to develop this jointly. Any takers? Comments? > > YKY What is the goal of your system. What application? -- Matt Mahoney, [EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=78397761-41d311
Re: [agi] NL interface
The semantics of Texai dialog system will be based upon the vocabulary of OpenCyc, but not upon the OpenCyc executable (which is free to use, but not open source). I have extracted all the OpenCyc content that is directly compatible with OpenCyc and released it at SourceForge: http://sf.net/projects/texai See the files in the open-cyc-rdf package. The TriG-formatted file contains context, which is a new feature of RDF. I persist the OpenCyc content in a Sesame 2 RDF store using the RDF Entity Manager that I wrote this year. I plan to connect this with OpenCog, when Ben designs a suitable RDF/SPARQL interface to the compatible Atoms in OpenCog. I believe that you could develop your system using the Texai dialog system without using the full OpenCyc derived vocabulary. Just a very small portion might work for the limited scope that you described. I'll have much more to say about the dialog system in the months to come. I have only just completed the first use case of parsing text into logic. Next task is to generate the same text from the logical input. -Steve Stephen L. Reed Artificial Intelligence Researcher http://texai.org/blog http://texai.org 3008 Oak Crest Ave. Austin, Texas, USA 78704 512.791.7860 - Original Message From: YKY (Yan King Yin) <[EMAIL PROTECTED]> To: agi@v2.listbox.com Sent: Friday, December 21, 2007 5:47:18 AM Subject: Re: [agi] NL interface On 12/21/07, Stephen Reed <[EMAIL PROTECTED]> wrote: > The above propositions have terms expressed in RDF, but are presented in the > lispy fashion desired by the original Fluid Construction Grammar > implementers. ( ). Note that I include > discourse axioms ( e.g. PreviouslyIntroducedInThisDiscourse). This grammar > is reversible. My next task is to generate the expected sentence given these > meaning axioms as input. Must it be used with Cyc? I see there're Cyc predicates in the output. Integrating it with an AGI other than Cyc might need some extra work... YKY This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?&; Looking for last minute shopping deals? Find them fast with Yahoo! Search. http://tools.search.yahoo.com/newsearch/category.php?category=shopping - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=78394643-f3c0b1
Re: [agi] NL interface
On Dec 21, 2007 11:08 PM, Matt Mahoney <[EMAIL PROTECTED]> wrote: > What is the goal of your system. What application? Sorry about the delay, and Merry Xmas =) The goal is to provide an easy input for AGI, temporarily until full NL capacity is achievable. I guess most AGIers would have realized by now, that a separate NL "module" (such as a chart parser, even with statistical learning) would not work for AGI. The semantics of words, together with syntactic knowledge, should be integrated in one big KB, ie, generic memory. I'm planning ultimately to do that, but this is not happening immediately. That's why I want to build an interface that lets users provide grammatical information and the likes. The exact form of the GUI is still unknown -- maybe like a panel with a lot of templates to choose from, or like the "autocomplete" feature. It will be useful for logic-based / symbolic / hybrid AGIs. YKY - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=79842424-e4f6fd
Re: [agi] NL interface
On Dec 28, 2007 12:45 AM, YKY (Yan King Yin) <[EMAIL PROTECTED]> wrote: > That's why I want to build an interface that lets users provide grammatical > information and the likes. The exact form of the GUI is still unknown -- > maybe like a panel with a lot of templates to choose from, or like the > "autocomplete" feature. I have previously recommended the interface used in the Alice programming environment. (www.Alice.org) The object browser can be directly acted upon, or the objects can be drag/dropped into the programming pane where each of the object's methods are exposed, then the parameters for each method are supplied. It quickly becomes an intuitive process. The resulting statement makes the syntax obvious and each choice can be updated by reselecting from a picklist. Even if you have no interest in animation, the programming interface does a really good job of providing flexibility without being too complicated. - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=79873507-07eb34
Re: [agi] NL interface
--- "YKY (Yan King Yin)" <[EMAIL PROTECTED]> wrote: > On Dec 21, 2007 11:08 PM, Matt Mahoney <[EMAIL PROTECTED]> wrote: > > What is the goal of your system. What application? > Sorry about the delay, and Merry Xmas =) > > The goal is to provide an easy input for AGI, temporarily until full NL > capacity is achievable. > > I guess most AGIers would have realized by now, that a separate NL "module" > (such as a chart parser, even with statistical learning) would not work for > AGI. The semantics of words, together with syntactic knowledge, should be > integrated in one big KB, ie, generic memory. I'm planning ultimately to do > that, but this is not happening immediately. > > That's why I want to build an interface that lets users provide grammatical > information and the likes. The exact form of the GUI is still unknown -- > maybe like a panel with a lot of templates to choose from, or like > the "autocomplete" feature. > > It will be useful for logic-based / symbolic / hybrid AGIs. > > YKY What would you do with the knowledge base after you build it? I know this sounds like a dumb question, but Cyc has built a huge base of common sense knowledge in a structured format, but it isn't useful for anything. Of course that is not the result they anticipated. How will you avoid the same type of (very expensive) failure? What type of knowledge will it contain? -- Matt Mahoney, [EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=80323735-4bde31
Re: [agi] NL interface
Matt, I agree w/ your question... I actually think KB's can be useful in principle, but I think they need to be developed in a pragmatic way, i.e. where each item of knowledge added can be validated via how useful it is for helping a functional intelligent agent to achieve some interesting goals... ben g > > What would you do with the knowledge base after you build it? I know this > sounds like a dumb question, but Cyc has built a huge base of common sense > knowledge in a structured format, but it isn't useful for anything. Of course > that is not the result they anticipated. How will you avoid the same type of > (very expensive) failure? What type of knowledge will it contain? > > > -- Matt Mahoney, [EMAIL PROTECTED] > > - > This list is sponsored by AGIRI: http://www.agiri.org/email > To unsubscribe or change your options, please go to: > http://v2.listbox.com/member/?&; > - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=80324636-31670d
Re: [agi] NL interface
The problem with all traditional knowledge bases (CYC too) it is still all meaningless symbols for the processing computer program, processed according to relational rules in the KB, which are all entered by _human beings_. An AI would need to develop it's own KB, like a child - we all have our own little KB's in our brains, each developed individually be their education and experiences. The KB which the AI would learn would not make sense to us, of course; as little as the neural connections in a human brain make sense to us (in the sense of "concepts"). Regards, Günther Benjamin Goertzel wrote: Matt, I agree w/ your question... I actually think KB's can be useful in principle, but I think they need to be developed in a pragmatic way, i.e. where each item of knowledge added can be validated via how useful it is for helping a functional intelligent agent to achieve some interesting goals... ben g What would you do with the knowledge base after you build it? I know this sounds like a dumb question, but Cyc has built a huge base of common sense knowledge in a structured format, but it isn't useful for anything. Of course that is not the result they anticipated. How will you avoid the same type of (very expensive) failure? What type of knowledge will it contain? -- Matt Mahoney, [EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?&; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?&; -- Günther Greindl Department of Philosophy of Science University of Vienna [EMAIL PROTECTED] http://www.univie.ac.at/Wissenschaftstheorie/ Blog: http://dao.complexitystudies.org/ Site: http://www.complexitystudies.org - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=80329343-0aa736
Re: [agi] NL interface
No, what's required is that the AI develop mappings into it's model of the universe (derived from its own sensors) from the artificially constructed model. This is the way one learns mathematical physics, or reading music (and, I presume, reading in general, but I don't remember that). It's different from the pure kinesthetic learning involved in "how to hit a ball with a stick", but it's still learning, and still a part of what it means to be a GAI. Günther Greindl wrote: The problem with all traditional knowledge bases (CYC too) it is still all meaningless symbols for the processing computer program, processed according to relational rules in the KB, which are all entered by _human beings_. An AI would need to develop it's own KB, like a child - we all have our own little KB's in our brains, each developed individually be their education and experiences. The KB which the AI would learn would not make sense to us, of course; as little as the neural connections in a human brain make sense to us (in the sense of "concepts"). Regards, Günther Benjamin Goertzel wrote: Matt, I agree w/ your question... I actually think KB's can be useful in principle, but I think they need to be developed in a pragmatic way, i.e. where each item of knowledge added can be validated via how useful it is for helping a functional intelligent agent to achieve some interesting goals... ben g What would you do with the knowledge base after you build it? I know this sounds like a dumb question, but Cyc has built a huge base of common sense knowledge in a structured format, but it isn't useful for anything. Of course that is not the result they anticipated. How will you avoid the same type of (very expensive) failure? What type of knowledge will it contain? -- Matt Mahoney, [EMAIL PROTECTED] - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?&; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?&; - This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=80345761-8b71a0