SUO: Re: Link Grammar and Parser - RRGs
What's CHUG? I guess the UG part is 'universal grammar', but somehow I'm
missing the CH. :-)
Arguments against (or for) grammatical relations aren't necessarily the
same as arguments against (or for) universal grammar. From what I can
tell, it is certainly the case that many linguists don't believe in
universal grammar. Van Valin apparently does, and the RRG book (Van Valin
and LaPolla, Syntax: Structure, Meaning, and Function) devotes a chapter to
Grammatical Relations GR. This includes a section on a theory of
grammatical relations. I gather that some folks think that grammatical
relations such as subject, object, indirect object are somehow 'primitives'
of grammar, while others think that they are somehow derived.
An overview of GR can be found at
http://ai.ato.ms/MITECS/Entry/woolford.html
Haven't finished reading it yet, but it seems to cover the subject
reasonably well.
I raised the RRG subject --- I hasten to add that I said 'its not for
everyone'--- so I feel like I should say something more about it.
1) The fact that RRG doesn't use traditional phrase structure or
traditional grammatical relations doesn't detract from its usefulness as a
theory of English. The universal aspects, if true, apply to English and
from what I've seen so far are every bit as simple as the traditional
analysis of English. Whether better or not, I'm not yet ready to argue.
One would also have to ask, 'better for what purpose.'
However,
2) The fact that the analysis of English that results from RRG is
non-traditional does mitigate against it's being ready for prime time. For
a non-traditional theory to take hold, it is necessary that there be a
pretty clear case for it being better (for some purpose) than the
traditional theory. One also recalls that relative to paradigm shifts in
physics someone (wish I had the reference handy) said words to the effect:
new paradigms aren't accepted by old physicists -- they are accepted only
when the old physicists die off.
3) Another mitigation against RRG being ready for prime time is the lack
of any (publicly available?) significant lexicon based on RRG.
4) I might add to the latter: the 'formal' language VV&LP use to formalize
their recommendations for the lexicon is never (as far as I can tell) given
a formal statement. It is pretty clear that they are using a structured
approach to the Logical Structure (their caps) to be used in the lexicon,
but they never say what it is explicitly. Moreover, it uses different type
styles for different 'sorts', including bold face lower case roman, plain
roman, SMALL CAPS, slanted SMALL CAPS, and very small caps with the same
baseline as the rest of the representation. And, between the lexicon an
the logical structure of an utterance that includes the operators, there
are paired parentheses, square brackets, angle brackets, (not sure about
curly brackets), each used for a different purpose. Its a significant job
just to translate this into a language that uses plain ASCII.
These problems being noted, I've learned a lot from VV&LP so far, much of
it relevant to choosing an ontology adequate for mapping English into CGs
or some symbolic logic. Perhaps this is partly due to my general ignorance
of linguistics to begin with.
I note that my 3) and 4) by themselves would be reason enough for someone
interested in generating a real application in the near future (or in
his/her lifetime?) to stick to some more well developed approach -- even if
the theory isn't as elegant :-).
I think the lattice versus acyclic graph question is interesting.
Certainly, there is a lot of work going on in linguistics relative to
typology (as they say). Placing languages in a lattice structure looks
like a good thing for someone to do, but I haven't seen it anywhere.
Best,
John Velman
On Fri, Dec 05, 2003 at 10:59:24AM -0800, Richard Cooper wrote:
> Regarding RRGs and CHUG, I found the following reference:
>
> Bhat, Darbhe Narayana Shankara. 1991. Gramatical Relations:
> The Evidence Against Their Necessity and Universality.
> London: Routledge.
>
> This paper isn't available to me, but the title indicates
> that at least some linguists aren't convinced that a CHUG
> really exists.
>
> Could languages be organized into a lattice, instead of
> a directed acyclic graph? If we believe ontologies are
> organized in lattice form, perhaps there is something in
> the human mind that makes lattices, rather than DAGs,
> the most convenient way to compartmentalize mental life.
>
> And if there is no single CHUG, then why should RRGs
> be extended beyond the target language (English in my case)?
>
> Does anyone have any other references related to this topic?
>
> Thanks,
> Rich
>