theory and formalism in LFG: too much inertia? (long msg)

[Farrell Ackerman]

well, here's my contribution to "naive scientism" and metatheoretic
speculation:  being unaccustomed to thinking so big, i apologize in
advance, but i hope that this continues a (useful) dialogue.

>
>The preceding rationale for splitting theory and formalism assumes a
>the inventory of relevant concepts crystallized in the formalism is
>static.  The linguistic metatheory is essentially viewed as a
>restriction of the formalism.  But we are living in dynamic times,
>cognitively speaking: the old MIT-style generative epistemology which
>held sway when LFG was invented is being overtaken by alien concepts
>such as optimization, probability, comparative grammaticality,
>markedness, and the like.  These are turning out to have interesting
>consequences for explaining typologies.  This suggests that we
>need to put more effort into enriching and expanding our formal
>architectures as a top priority, no?
>

yes, that sounds right.  ernst mayr in his book "this is
biology" suggests that "In biology, concepts play a far greater
role in theory formation than do laws.  The two major contributions
to a new theory in the life sciences are the discovery of new facts
(observations) and the development of new concepts."  He contrasts
this emphasis on concepts within biology to reliance on laws in the
physical sciences (and as the default criterion for evaluating theories
among philosopers): "Regularities are abundant in the living world
too, but most of these regularities are not universal and without
exception; they are probabilistic and very much restricted in
space and time...Of course, at the molecular level many of the
laws of chemistry and physics are equally valid for biological
systems, and these are widespread in biology.  But few, if any,
regularities that have been observed in complex systems satisfy
the rigorous definition of laws adopted by physicists." 1997:62

so, i guess one way of interpreting your remarks is that lfg is
contributing to new theory formation in the two crucial ways
described by Mayr for the life sciences (not really that surprising,
if one considers language as a complex system analogous to those
in biology).  first, there is new data, (or at least, a recognition
that older discoveries of typologists and functional linguists are
data that deserve to be acknowledged and modelled more rigorously than
they have been
in the past):  so lfg researchers
continue to examine cross-linguistic distributions of various
phenomena within morphosyntax (and elsewhere) and attempt to
identify common and less common distributions, as well as
common and less common alignments between different interacting
information types.  from the perspective of concepts, it turns out,
that, as suggested by Mayr for complex biological systems, probabilistic
models seem quite promising, while such notions as harmonic alignment
and markedness are concepts one would like to see organically incorporated
into formalisms which would do a good job constraining the space
of options allowed to human grammars.  (elman et. al. in their book
"rethinking innateness"
invited linguists, among other researchers in cognitive science, to
rethink the conceptualization of language analysis along the model of
biological research.  they, of course, endorse a particular implementation
of probabilistic modelling, i.e.,  pdp models, but this preference
shouldn't obscure
the more important reconceptualization of research that they advocate:
a recononceptualization that can be implemented/formalized in numerous ways.)

i take it as an exciting direction of research within "linking" that
some work within lfg is formally modelling the general conceptual approach to
argument selection proposed by Dowty in a manner that appears to be
compatible with Dowty's original intentions:  i think this is evident in
Asudeh's elegant and appealing
treatment of Marathi subject/object reversal.  i take it that this work
goes in the direction of "enriching and expanding our formal
architectures".  gert and i have tried to address certain markedness
phenomena
in grammar by exploiting the formal device of multiple inheritance hierarchies
and by positing universal grammatical archetypes that provide a template for
the inheritance of chunks of information into language particular systems
(this in some
ways appeals to the markedness evaluation metric of Jackendoff 1975 and
its development in Bochner 1993)).  as we point out, such a
formalism can help us calibrate
degrees of deviation (or markedness) from the archetype as found in specific
languages, but this clearly needs to be supplmented by a story about why we
find
the deviations that exist versus those that don't.

in sum, it seems to me that you're right:  both the data
being looked at and the new concepts being appealing to to talk about them
run far ahead of the formal apparatus used to encode these concepts.
i don't know whether
the enrichment and expansion of formal architectures should be a "top
priority",
(it's not clear we've got the right complement of concepts yet)
but it probably wouldn't be a bad idea to explore alternative
formalizations of
useful concepts.  irina (nikolaeva) and i will be trying to do this as we
develop alternative accounts of highly marked eurasian
prenominal relatives (analogically based on possessives) from both an
ot-lfg perspective and a grammatical archetype perspective.

windily, farrell


Farrell Ackerman
Dept. of Linguistics
University of California, San Diego

http://ling.ucsd.edu/~ackerman/