On Wed, Apr 3, 2019 at 2:31 AM Linas Vepstas <[email protected]> wrote:
>
> Anton,
>
> I feel like I am torturing you, and for that, I apologize. But when I read
> your attachments, I see things like this: Page 2:
>
> > Obstacles
> > The following problems have been encountered during the research.
> >
> > 1. The MST-parses quality seems to be the main blocking issue preventing
> > learning the reasonable and useful Link Grammar rules.
>
> To me, this claim sounds absurd and outlandish -- it is a direct violation of
> the central limit theorem. The quality of the MST parses should have no
> effect (minimal effect) on the learned grammar. That is what the central
> limit theorem says. So, as bullet/obstacle #1, you claim that a cornerstone,
> a foundation-stone of probability theory is false? Incorrect? Inapplicable?
> What evidence do you provide that your claim is true? Well -- none, so far.
Linas, is it possible for you to spell out in an email what is your
argument that {the central limit theorem implies the MST parses should
have minimal effect on the learned grammar} ?
I don't really get why you believe this, and I have read your various
documents and also obviously spent a fair bit of time thinking about
the unsupervised language learning methods we are playing with...
It seems to me that the MST parsing step in our pipeline is where some
key grammatical constraints are imposed (e.g. no links cross, complete
connectivity etc.), and thus doing this step should impact the learned
grammar even in the large...
But I would like to understand what you're thinking better...
thanks
Ben
--
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 [email protected].
To post to this group, send email to [email protected].
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/CACYTDBehxqiuhZLimUebLtrCbLD2dz2Q-%3D%2BFJ73tqLWycPByZw%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.
