Re: Universal grammar

Tak To <takto@xxxxxxxxxxxxxx> wrote:

In any case, I think "a universal representation of meaning"
exist, which is the collectively state of neurons in one's brain.
It is just that it is not that convenient to "use".

I am playing around with the same ideas, but strictly in the domain of
mathematics. Let me drop off some inputs:

Traditional metamathematics typically (i.e., most, but not all) treat
object math theories as sequences of strings. But mathematicians do not
agree on notation, even though they generally agree on the notions, i.e.,
the semantics.

One might attempt to pin down this semantics using tree structures similar
to ASTs (abstract syntax tree - readers not knowing this stuff might use
the Wikipedia). Take the expression A => B, which may be given prefix
(Lukasiewicz) notation "A B =>" or RPN "=> A B"; but the AST is the same:
      =>
    /   \
   A     B

What do you want to do with the input information?  This ultimately
affects how you want to do the parsing.

For example, it is generally recognized that if one must have
semnatic knowledge if one is to parse correctly.  This is one
of the classical examples:

    The man gives the house plants to charity.
    The man gives the gardener plants to water.

Have your heard of Early's Algorithm,

Those tend to be slow. Do you know how GLR works?

My memory is vague.  Isn't the worst case of GLR a lot worse
than that oF Earley's?   In any case, I was just throwing
alternative ideas.

Take the classical example:
  Time flies like an arrow, fruit flies like a banana.
Both parts are ambiguous, here. It is not difficult to write parser that
keeps track of all possibilities. But this way, all the possible parses of
the first part is s_1, ..., s_k, and the second part is t_1, ..., t_l, get
combined the expansion of the possibilities (s_1, ..., s_k) x (t_1, ...,
t_l), or k*l items.

By contrast, suppose the GLR grammar merges at the "," in the example
above. Then one is forced to write an action, or semantics description, of
the first s_1, ..., s_k possibilities. The parser then starts over with
one single parsing branch to get the t_1, ..., t_l possibilities. When
these merge, one gets a semantic description of (s_1, ..., s_k) x (t_1,
...., t_l), and the parsing is much quicker, as it does not have to carry
all those branching possibilities along after the merges. Natural
languages seem to have this kind of local ambiguities.

So I think this answers your questions. Yes, one is forced to develop
logical models for the natural language semantics.

...you are more likely to find people knowledgeable about
parsing natural languages in comp.ai than in sci.lang.

Thank you. I am aware of that (I'm in fact on comp.ai). I just wanted to
get some inputs here. :-)

--
Hans Aberg
.

Relevant Pages

  • Re: Existence of proof verifiers: A comedy
    ...    "a sequence of wff's such that each wff is either an instance ... There has to be a natural language proof corresponding to ... any formalism in order to have a proof. ... proof is a sequence of wffs, ...
    (sci.logic)
  • Re: Existence of proof verifiers: A comedy
    ... wffs with no natural language satisfy that definition? ...    "a sequence of wff's such that each wff is either an instance ... There has to be a natural language proof corresponding to ... any formalism in order to have a proof. ...
    (sci.logic)
  • Re: XML parsing with python
    ... would like to do this validation through the parsing tool  itself. ... Is lxml part of standard python package? ...
    (comp.lang.python)
  • Re: How to implement this in unit testing
    ... If I want to implement something like this in Ruby Unit Testing: ...     end_choice ... My question is what is the best way to parse previous code i.e. is ... runtime or at least do the parsing and change on the fly instead of ...
    (comp.lang.ruby)
  • Re: Schildt
    ...   syntax descriptions is a matter of current debate: ... is that up to now they haven't inspired any new kinds of parsing ...
    (comp.lang.c)