Relex prepositions

[anitzkin]

@ruiting @bgoertzel @linas
I was looking in to doing something with prepositions in r2l and I have re-discovered the fact that prepositions are very confused in Relex; the first thing I wanted to do was have it stop generating these dependencies like prep(verb, prep) . . . and I think . . . am I confused, or is there not a whole bunch of unfinished semantic algorithms about prepositions, using variables like $modified and $prep_source that don't actually refer to anything, algorithms with no implicands. Or do I just fail to understand what is going on in that section???

[linas]

Ah. Heh. You don't understand that section. Its magick.

Observe that unlike all the other algs, this one uses #PrepositionLinkAlg (instead of #TemplateActionAlg) This alg currently appears only 14 times in the algs files.

Other Algs are:
#TwoWordCombineToLeftAlg used once
#TwoWordCombineToRightAlg used twice
#WordSequenceCombineAlg used once
#MorphyAlg used once
#VerbTenseAlg used once
#TenseConjoinAlg used once

... and ..
#PrepositionLinkAlg used 14 times.

oh, and #TemplateActionAlg used 578 times

The $modified and $prep_source are magic tokens that are filled in by this Alg. Yes, the way that they are used seems backwards; they seem to be on the wrong side of the = sign. However, they do get matched and filled in correctly, and if you add a term _mybogusdebug(blah,$prep_source) you can see them get filled in (correctly).

At any rate, #PrepositionLinkAlg is central to making prepositions work: it does all the heavy lifting.

I have not given any thought to the idea of replacing #PrepositionLinkAlg with equivalent #TemplateActionAlg .. however, there is something appealing to making such an overhaul. As long as we don't break a bunch of stuff. Ergo unit tests.

If you can systematically replace #PrepositionLinkAlg in some controllable, consistent way, that would be great. Just .. be aware it might be tricky(???)

[anitzkin]

Okay. Happy to know more :) I will use mybogusdebug first thing Monday and see if I can figure out
how the Preplinkalgs work before I go any further. I test each revision individually as I make them. So
far so good. I will make a list of updates needed for the unit test before I leave for Thailand on the 15th.

---

On Sat, 12/6/14, Linas Vepštas [email protected] wrote:

Subject: Re: [relex] Relex prepositions (#195)
To: "opencog/relex" [email protected]
Cc: "anitzkin" [email protected]
Date: Saturday, December 6, 2014, 6:42 AM

Ah. Heh. You don't understand
that section. Its magick.

Observe that unlike all the other algs, this one uses
#PrepositionLinkAlg (instead of #TemplateActionAlg) This
alg currently appears only 14 times in the algs files.

Other Algs are:

#TwoWordCombineToLeftAlg used once

#TwoWordCombineToRightAlg used twice

#WordSequenceCombineAlg used once

#MorphyAlg used once

#VerbTenseAlg used once

#TenseConjoinAlg used once

... and ..

#PrepositionLinkAlg used 14 times.

oh, and #TemplateActionAlg used 578 times

The $modified and $prep_source are magic tokens that are
filled in by this Alg. Yes, the way that they are used
seems backwards; they seem to be on the wrong side of the =
sign. However, they do get matched and filled in correctly,
and if you add a term _mybogusdebug(blah,$prep_source) you
can see them get filled in (correctly).

At any rate, #PrepositionLinkAlg is central to making
prepositions work: it does all the heavy lifting.

I have not given any thought to the idea of replacing
#PrepositionLinkAlg with equivalent #TemplateActionAlg ..
however, there is something appealing to making such an
overhaul. As long as we don't break a bunch of stuff.
Ergo unit tests.

If you can systematically replace #PrepositionLinkAlg in
some controllable, consistent way, that would be great. Just
.. be aware it might be tricky(???)

—
Reply to this email directly or view
it on GitHub.

[linas]

OK< can you split your other pul requrest into two? Some changes seem harmless, but some of the prep ones make me nervous. The harmless ones should be commited immediately, the prep ones, I'd like to wait till you say that you really get the big picture.

[anitzkin]

Sounds good.

---

On Sat, 12/6/14, Linas Vepštas [email protected] wrote:

Subject: Re: [relex] Relex prepositions (#195)
To: "opencog/relex" [email protected]
Cc: "anitzkin" [email protected]
Date: Saturday, December 6, 2014, 10:58 AM

OK< can you split your other
pul requrest into two? Some changes seem harmless, but some
of the prep ones make me nervous. The harmless ones should
be commited immediately, the prep ones, I'd like to wait
till you say that you really get the big picture.

—
Reply to this email directly or view
it on GitHub.