Hi Muidem, Welcome to the list. There's no breaking the rules if you're politely asking a question. As a good friend said to me once: "the only stupid question is the one you never asked."
Francisco at http://CEPT.at has developed a really interesting system which represents words as very much CLA-friendly SDRs. You should take a look at his presentation from the hackathon as well as exploring the tools on his website. I've recently started working with Francisco to explore adding the sequence learning and disambiguation opportunities of a CEPT+NuPIC system. We should have some results to share early in the New Year. At the moment Grok/Numenta seem to be very much focussed on dealing with machine generated data, usually scalar, and especially relatively low-dimensional data which is amenable to using a model-per-stream approach. There's a huge opportunity to prove NuPIC and Grok in this domain and this is what they're concentrating on. This means that it's up to the rest of the community to take a look at questions like yours. Luckily, there are many people on the list who are exploring language questions. It appears to me that a hierarchy is necessary in order to deal with human languages, even if you start at the bottom with CEPT SDRs for words. We're still not sure how to emulate the nested tree structure of human language in CLA - this is for a future project. Regards, Fergal Byrne On Wed, Dec 4, 2013 at 4:26 PM, Brev Patterson <[email protected] > wrote: > Hi Medium, > > We focused on using NuPIC for simple NLP at our recent hackathon: > http://numenta.org/blog/2013/11/06/2013-fall-hackathon-outcome.html > > My hack is actually a learning twitterbot, but I haven't finished it yet. > > May also be helpful: > https://github.com/numenta/nupic/wiki/Natural-Language-Processing > https://github.com/chetan51/linguist > > I think it's all currently based on using a single NuPIC Region layer > ("instance" as you say). > > But, the future goal is like you describe: A hierarchy of regions/layers > tied together, with low-level items (letters) at the "bottom", building up > to higher level items (like sentences) at the "top". > > Thanks, > Brev > > > From: muidem yzarc <[email protected]> > Reply-To: "NuPIC general mailing list." <[email protected]> > Date: Tuesday, December 3, 2013 5:48 PM > To: "[email protected]" <[email protected]> > Subject: [nupic-discuss] new person question about chatbots and nupic > > hello nupic mailing list, > > i searched the archive for chatbot and saw not discussion on chatbots > using nupic. > > this is my first using a mailing list, sorry if i am breaking the rules, i > have a high level conceptual question. > > if i was using nupic in a chatbot would i want one big instance of nupic > (for letter, word, and sentance prediction) or three instances of nupic > (one instance for prediciting the next letter, word, and sentance)? > > is there a better way for new person to ask these beginner types of > questions? (is there a natural language processing sub-group (or list) for > nupic?) > > thank you, > medium > > > _______________________________________________ > nupic mailing list > [email protected] > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org > > -- Fergal Byrne, Brenter IT <http://www.examsupport.ie>http://inbits.com - Better Living through Thoughtful Technology e:[email protected] t:+353 83 4214179 Formerly of Adnet [email protected] http://www.adnet.ie
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
