Hi Amirouche,
On Sat, May 25, 2019 at 6:32 PM Amirouche Boubekki <
amirouche.boube...@gmail.com> wrote:
> Hello,
>
>
> Is there existing work or work you want to recommend to me regarding the
> topic of extracting triples from text.
>
> Here are some example:
>
> - Bloomberg is an analytics company. => (Bloomberg, is a, analytics
> company)
> - Microsoft created Windows => (Microsoft, created, windows)
> - Google is a search engine => (Google, is a, search engine)
>
>
> How does opencog proceed to refine text into knowledge?
>
> Good question, no good answer. I mean, back in 2002, you could use the
atomspace to represent
EvaluationLink
PredicateNode "created"
ListLink
ConceptNode "Microsoft"
ConceptNode "Windows"
but so what? What can you do with that? Well, there are 3 or 4 or 5
answers, of varying levels of sophistication.
* Run a question-answering chatbot. So, there's some dead code, from 10
years ago, that was a question-answering chatbot. You could ask it
plain-english questions "Who created windows?" and it would answer it,
based on a search of triples like the above. It worked. It had three
problems: (a) it only supported a small number of question types (b) it
gave one-word answers (c) it could not track conversation state (d) it
could not reason.
* There was/is a non-functional subsystem called relex2logic+sureal (I call
it non-functional, because I never saw it actually work) that solved (b)
and sort-of solved (a) and attempted to solve (d). It would be great if
someone stepped up and kept this in a maintained state, so that we could
provide at least a basic demo. But no one wants to do this.
* There is a subsystem called "ghost" which is a chatbot, i.e. it solves
problem (c) unfortunately, it cannot do (a) (b) or (d) - i.e. it cannot do
basic question-answering. It's internal data format is not documented
anywhere, so no one knows how to integrate it with question answering.
* There is a system called PLN which can do (d) in principle, but it has
trouble scaling up to larger knowledgebases.
* As I mentioned to buj, it would really be cool if someone (e.g. you)
stapled a crsip-logic reasoner/solver to opencog. I would strongly
recommend the potsdam ASP solver. It is really lighting-bolt fast, and
could solve many of the basic reasoning examples that PLN struggles with.
I don't think it would be that hard to glue these together. That would
solve problem (d) but leave the integration of (a), (b), (c) in the open.
So this is what we are up against: multiple systems, mostly incompatible,
partly-unmaintained, all mostly undocumented. So, in that environment,
nothing works. Importing data is not a problem. Keeping an end-to-end
demo working, and not broken, is a problem.
-- Linas
--
cassette tapes - analog TV - film cameras - you
--
You received this message because you are subscribed to the Google Groups
"opencog" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to opencog+unsubscr...@googlegroups.com.
To post to this group, send email to opencog@googlegroups.com.
Visit this group at https://groups.google.com/group/opencog.
To view this discussion on the web visit
https://groups.google.com/d/msgid/opencog/CAHrUA35_Oavvich5anyaM9TsBp1SxsxqtZWQ9w-vxufybCZnxw%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.
