No problem Linas. >From my point of view I'm encouraged that OpenCog is closer to an interesting language model than I thought.
I was surprised to see you discussing category theory in the context of a language model. Category theory is motivated by formal incompleteness. To see this applied to language is something I argued for long and hard. I remember a thread on U. Bergen's "Corpora" list in 2007 with very little traction on exactly this point. People could not see the relevance of formal incompleteness for language. To see you, and others, embracing this is progress. I'm glad you are deconstructing the grammar. You are probably forced to it by the success of distributed representation these last few years. But at least you are doing it. I feared some ghastly fixed Link Grammar with neural nets just disambiguating. Instead I see Ben is right. My basic data formulation of the problem may well be compatible with what OpenCog are doing. That's good. Though I am still confused by Ben's statement that "we can't currently feed as much data into our OpenCog self-adapting graph as we can into a BERT type model". What does an OpenCog network look like that it is hard to feed data into it. Can you give an example? What does an OpenCog network with newly input raw language data look like? -Rob On Wed, Feb 20, 2019 at 4:21 PM Linas Vepstas <[email protected]> wrote: > > > On Tue, Feb 19, 2019 at 5:33 PM Rob Freeman <[email protected]> > wrote: > >> Linas, >> >> OK. I'll take that to be saying, "No, I was not influenced by Coecke et >> al. >> > Note to self: do not write long emails. (I was hoping it would serve some > educational purpose) > > I knew the basics of cat theory before I knew any linguistics. I skimmed > the Coecke papers, I did not see anything surprising/unusual that made me > want to study them closely. Perhaps there are some golden nuggets in those > papers? What might they be? > > So, no, I was not influenced by it. > > For all that, I can't figure out if you are contrasting yourself with >> their treatment or if you like their treatment. >> > > I don't know what thier treatment is. After a skim, It seemed like > word2vec with some minor twist. Maybe I missed something. > >> >> I quite liked their work when I came across it. In fact I had been >> thinking for some time that category theory has something the flavour of a >> gauge theory. >> > > Yellow flag. Caution. I wouldn't go around saying things like that, if I > were you. The problem is that I've got a PhD in theoretical particle > physics and these kinds of remarks don't hold water. > > I have no problem with the substance of it. I just don't think it is >> necessary. At least for the perceptual problem. The network is a perfectly >> good representation for itself. >> > > To paraphrase: "I know that the earth goes around the sun. I don't think > it's necessary to understand Kepler's law". For most people, that's a > perfectly fine statement. Just don't mention black holes in the same > breath. > > > I say you can't resolve above the network. Simple enough for you? > > Too simple. No clue what that sentence means. > > > '"fixed"? What is being "lost"? What are you "learning"? What do you > mean by "training"? What do you mean by "representation"? What do you mean > by "contradiction"?'... > > But if you haven't understood them, it will probably be easier to use > your words than argue about them endlessly. > > ??? > > > Anyway, in substance, you just don't understand what I am proposing. Is > that right? > > I don't recall seeing a proposal. Perhaps I hopped in at the wrong end of > an earlier conversation. > > I'm sorry, this conversation went upside down really fast. I've hit dead > end. > > --linas > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T581199cf280badd7-M931e84b52e289fb3c776903b Delivery options: https://agi.topicbox.com/groups/agi/subscription
