OT and diachronic functionalism

[Martin Haspelmath]

I'm somewhat surprised to see that Joan Bresnan dissociates herself from
the competence/performance terminology. OK, it probably has a lot of
baggage attached to it that I may not be aware of, but my impression is
that we all pretty much agree that there is a necessary conceptual
distinction between language use and language structure, or speech and
grammar, or more generally between processing and storage, or cognitive
events and cognitive patterns. I used "competence/performance" as a
convenient shorthand, following a well-known functionalists's (John
Hawkins's) usage.

True, the competence/performance distinction has often been attacked by
functionalists, but I believe that this attack is misguided. The problem
with the Chomskyan approach is not that it draws this distinction, but
that all the ways in which performance influences competence are
systematically ignored. (See also my review article on Fritz Newmeyer's
1998 book, soon to appear in Lingua).

So in my view, three tasks of linguistics are:
(i) describing competence,
(ii) describing performance, and
(iii) explaining competence, which must apparently often make reference
to (ii).

I understand the Chomskyan approach as claiming that (iii) is
irrelevant, (ii) is moderately interesting, and (i) is the core task of
linguists (especially concerning the innate aspects of competence). Joan
Bresnan's OT functionalism draws the lines quite differently, it seems,
and I haven't understood yet, how. So far my impression of OT had been
that the Chomskyan program is virtually unchanged: The main goal is a
maximally elegant description (or "characterization", the euphemism that
is often used instead) of the competence grammar, but the archtecture of
the formal grammar is different: The machinery makes use of abstract
entities called "constraints" which often happen to correspond closely
to the actual constraints on language use that speakers grapple with
while speaking. (I call these "user constraints" in my ZS paper.)

Joan Bresnan asks:

>> Speakers are thus constantly "tempted" to devoice consonants
(partially
>> or completely) under these conditions, and sometimes they give in to
the
>> temptation.

>Is the "constant" presence of this "phonetic temptation" any different
from
>the OT hypothesis that the contextual markedness of voiced obstruents
>is a universal constraint that is present in every individual by virtue
of
>the human articulatory and perceptual systems?

Until recently I thought yes, crucially, in that the OT constraint is
part of the competence grammar (and is perhaps even innate as part of
UG), whereas the "user constraint" is a constraint on the articulatory
system and has nothing to do with the conventional language system (=
competence). (But now Joan Bresnan's contributions to this discussion
have confused me.)

Joan continues:
>Martin suggests that once speakers "give in to the temptation" to
>devoice, devoicing may spread throughout the speech community and
>result in language change.  Yes, but when we try to model precisely
>what "giving in to the temptation" means, we may come up with the idea
>that the "phonetic temptation" becomes more dominant compared to the
>temptation to preserve constrasts.

Sure, intuitively that's what is going on, and linguists have been
talking in such terms for centuries (though not as precisely as in OT,
of course). But how can a "phonetic temptation" become more dominant
than another temptation? After all, these temptations or constraints are
universal ? identical for all speakers and all languages! Clearly, what
happens is that the linguistic *conventions* change: It suddenly becomes
socially acceptable to "give in to the temptation". Or in other words,
the grammar has changed, and the functional motivation has left its mark
on the competence system. Since the grammar is so clearly shaped by the
performance constraint, it is possible (and elegant) to describe the
grammar in terms of a competence constraint that mimics the performance
constraint. My point in my ZS paper is that such descriptions are
clearly better than arbitrary descriptions (so I welcome OT), but that
we still need the link to the performance constraints if we want to
explain grammar.

Fritz Newmeyer asks:
>(A)     Constraints are universal, and may or may not be functionally
>motivated (e.g. Jane Grimshaw's LI paper, where the issue does not come

>up).

>(B)     Constraints are universal and functionally motivated (Martin
>Haspelmath's forthcoming Z fuer S paper).

>(C)     Constraints are universal and are the actual functional
>motivations themselves. That is how I interpret the following remark
from
>Joan's December 10 posting: 'Couldn't conflicting constraints such as
>iconicity and economy be universal, but prioritized differently across
>different domains and different languages?'

>Martin and Joan: does that put the finger on the different ways that
you
>view the constraints?

I thought so initially, but I haven't understood Joan yet.

Fritz mentions some word order facts as difficult for an
OT-functionalist position like Joan Bresnan's, and again Joan might
reply that her general approach is only in its infancy, so it's too
early to judge whether it might not work after all.

But I agree with Fritz's general skepticism: Languages are so full of
patterns that are there because of speakers' conservatism, not because
they allow speakers to give in to some temptation. Fritz is absolutely
right in pointing out that if one wanted to make these structures follow
from an OT-functionalist grammar, one would have to introduce a
functional constraint such as "conventionality".

Another consideration that makes me skeptical about OT is that the
descriptions rely so heavily on non-functional, language specific
constraints such as alignment. HEAD-LEFT might be seen as a constraint
in English grammar, but it surely cannot be functionally motivated,
because HEAD-RIGHT is just as good from the language user's point of
view. As John Hawkins has argued convincingly, the real motivation for
the Greenbergian word order correlations is Early Immediate
Constituents, and it seems difficult to integrate this into an OT-style
grammar. But let's wait and see.

Martin

--
Martin Haspelmath (haspelmath at eva.mpg.de)
Max-Planck-Institut fuer evolutionaere Anthropologie, Inselstr. 22
D-04103 Leipzig (Tel. (MPI) +49-341-9952 307, (priv.) +49-341-980 1616)