"forms" and "formalists"

Spike Gildea spikeg at OWLNET.RICE.EDU
Tue Mar 7 17:54:07 UTC 2000


Sorry for the length of this posting, but ideas just kept coming until now,
and I just want to send it before I spend any more time I can't afford on
it.

It seems to me that the recent debate/discussion on FUNKNET has sort of
conflated the analytical category "form" with the social/professional
category "formalist" (one heuristic definition of "formalist" might be 'one
who would introduce him/herself with a phrase like "I am a formal
syntactician"'). As I have reviewed the postings, three issues have stood
out as dividing formalists and functionalists, (and creating internal
divisions in both camps as well).  I wanted to tease them apart and see if
anyone else finds this a useful contribution.  First, the analytical
category of "form" is not homogenous, but consists of a scale from more
concrete, lower-level forms (which I suspect most functionalists accept) to
more abstract, sometimes theory-dependent categories of forms (which appear
to be more to the taste of formalists).  Second, formalists prefer to
"explain" more concrete kinds of form with reference to these more abstract
formal categories, whereas functionalists look for explanation outside the
domain of form.  Third, actual databases of concrete forms collected by
different methods are not equally reliable, and the methods that appear to
be favored by formalists are among the least reliable.  None of these
observations is original to me, but they seem to have gotten obscured in
recent postings.  I can accept that the first two differences might be
matters of preference, but the third is really disturbing if true.

While I have not really encountered much useful typological data in
writings from the professional category of formalists (due in part to the
small number of grammars), I do believe that it is impossible to do
interesting linguistics without some concept of more concrete linguistic
"form".  I think TG's and Edith's postings told a story about such forms
that makes a lot of sense, but that perhaps skirts the more divisive issue,
which doesn't arise in earnest until you get beyond the simplest forms
(like phoneme, or maybe morpheme).  The problem comes when we begin to
categorize "types" or "categories" of forms in individual languages, and
then typologically.  Even if we can agree on some important types (maybe
agreement, parts of speech, phrase structure...), the more abstract and
theory-dependent the formal categories get, the less likely you are to find
widespread agreement that the category is central, or even relevant, to
understanding languages/Language (grammatical relations, passive, or for
more abstract and theory-dependent categories, maybe C-command, government,
etc.).

Some functionalists seem happy to just get a decent phonemic analysis so
they can get reasonably accurate written representations of utterances;
their interest in form stops once they have sufficiently discrete clusters
of phonemes to ascribe meanings/functions to.  Others want to go a little
further up the abstractness scale in search of additional formal patterns
that might be of interest (i.e. that might correlate with some function).
For instance, I am very interested in formal properties usually ascribed to
formal categories like parts of speech and grammatical relations.  Examples
of such patterns in the domain of "subject properties" would be control of
coreference with reflexive possessor pronouns/prefixes, control of
coreference with participants in adjacent/embedded clauses, or control of
agreement (with auxiliaries, verbs, etc.).  My main reason for being so
interested is not anything inherent to the formal patterns, but it is the
fact that these patterns evolve diachronically in ways that suggest their
cognitive reality to speakers, and in ways that allow me to make much
richer functional analyses of the processes of grammaticalization (which,
among other things, could be characterized as change in formal category
membership).

As to the second issue, whereas formalists utilize membership in an
abstract formal category as an "explanation" for formal behavior,
functionalists will look for a function-based alternative or derive a
function-based historical story that creates (and thus "accounts for") the
modern patterns of formal behavior.  For instance, everyone agrees that the
formal category of gender/number suffixes exists in Romance languages, but
not everyone agrees that their distribution is best explained as a
consequence of nominal membership in a formal gender/number category
(although the huge majority can be explained in this way, a significant
number -- as well as certain types of counter-examples -- indicate a
semantic basis to gender/number as well).  As your categories get more
abstract (e.g. C-command as an "explanation" for the domain of control of a
reflexive pronoun), you won't even get a hearing from most functionalists.
Returning to "subject properties", I don't need to refer to (or even to
have) a formal category of "subject" in order to describe these patterns,
and I would reject out-of-hand the suggestion that the formal category of
subject in any way motivates or explains these patterns.  But when the
subject properties all line up to point to the same participant, it sure
makes the existence of a formal category of subject look plausible.  Of
course, when they don't all line up, the formal category breaks down and
people that insist on finding a "subject" in every language/construction
type either have to switch to nonformal criteria to define subject (Givón,
Dixon), select some (nonuniversal) subset of the criteria to define subject
(RG), or decide that "subject" is not a universal category (Dryer).

But leaving aside the issue of the abstract formal category "subject", and
whether such a category might be useful in describing languages, I don't
see that it is possible to question the *existence* of coreference and
agreement patterns in speech -- these are a more abstract type of "form",
but still not particularly theory-dependent.  The problem is collecting
sufficiently reliable samples of speech to verify the nature of such
patterns in individual languages.  This is the third issue that is
intertwined in our debates, and I think it is the most serious issue, as it
speaks to the legacy we leave in our descriptive work.

In my experience to date, elicitation is a reasonably reliable way to
generate a corpus of utterances regarding simple morphology and phrase
structure in relatively simple clauses, but as the sentences get longer and
the morphology gets more complex (i.e. when one needs to code concepts that
do not obviously belong to the simple clause), the elicited corpus gets
less and less reliable -- the results are often not replicable across what
appears to be a relatively homogenous population of speakers, and at least
in the work I'm familiar with first-hand from South America, different
linguists record different sorts of utterances/grammaticality judgements,
even from the same speakers.  I have had the least success replicating
simple grammaticality judgements about utterances cooked up for the purpose
of testing my ongoing hypotheses about abstract patterns like complex
coreference.  Sociolinguistics research has found a parallel decline in
reliability as people switch from producing language to reflecting on
language (cf. especially Labov's publications in the mid-70s).

I believe a part of FUNKNET's antipathy for "formalist" linguistics follows
from our collective impression that it is precisely these sorts of
unreliable data that are at the core of the empirical database for most
formalist theories.  It is not *just* the limitation of the database to
sentence-level data, but to a type of sentence-level data (grammaticality
judgements of unusual sentence types) that does not appear to represent the
way speakers really *use* language.  I have had conversations with some
friendly formalists who do not accept this characterization, either of
their databases or their methods, but my impression remains strong that the
majority of data actually published in formalist syntax articles are of
this type.

Perhaps this impression reveals my theoretical bias, as a simple review of
the examples in any article does not reveal the methods by which each
example was collected.  Would anyone else be interesting in seeing a
session on fieldwork methodology someday at a conference that is attended
by both formalists and functionalists, in which fieldworkers of both
persuasions might really describe their data collection procedures, the
makeup of their database, and the concerns they feel (if any) about the
reliability of their data?

best,

Spike

P.S. Speaking of Labov and methodology, I would be interested to know if
anyone in formalist linguistics has ever refuted (or even addressed) the
main points of Labov's (1975) crushing indictment of grammaticality
judgements produced by linguistically sophisticated native speakers
(especially by native speaker linguists!).



More information about the Funknet mailing list