LINGUIST List 8.672

Wed May 7 1997

Disc: OT (Optimality Theory)

Editor for this issue: Anthony Rodrigues Aristar <aristarlinguistlist.org>

Directory

1. D.B. den Ouden, disc. on OT
2. Mark Donohue, Disc: OT

Although I have unfortunately lost the original questions posed by
Chris Hogan on the falsifiability of OT, I believe the response of
Joaquim Brandao de Carvalho (LINGUIST 8-670) presents an image
of OT that does not necessarily have to be correct (how's that for
prudence?). Note that these remarks contain a very rough sketch of
some ideas on language and that I am more than open to
positive criticism and adaptation of these ideas.

Carvalho writes:

> Even if OT specified which constraints are permitted in the theory,
>the entire theory would probably not be affected whenever constraints
>might be shown to be 'falsified'. The reason is that constraint
>falsification could always be viewed by OT supporters as constraint
>violation.<

I believe this is precisely what does NOT apply if OT specifies
what type of constraints are permitted in the theory. This is done by
many OT'ers (cf. the discussion on phonetics and OT on the Optimality
List last February). This has to do with the second point Carvalho
makes:

> The important point, from which it can be shown that OT is
>NOT a linguistic theory nor a theory tout court, is that constraints
>are formally arbitrary and thereby circular. Let us take constraints
>like 'syllables must have onsets' or 'syllables must not have codas'.
>These constraints are arbitrary insofar as nothing but empirical
>observation tells us why onsets should be obligatory, while codas
>should not.

This is simply not true. There are a number of, for example,
functional explanations for the fact that VC syllables are more
marked than V syllables, which in turn are more marked than CV
syllables. The sounds are better perceived and/or articulation is
easier (rule number 1: Do as little as possible!). Functional
considerations such as those posed by, for example, Flemming (1995):

I More contrast is good
II Bigger contrast is good (there should be some minimal distance
between two characteristics)
III Easy articulation is good

can well by applied to make this point.

Once it is acknowledged that OT constraints should NOT be allowed to
be randomly stated, but should find their origin in `functional'
or even `cognitive' or `biological' factors, it serves as a highly
useful description of how different and often competing factors
influence the way we use language, the way language develops, and
the possible ways in which natural human languages can be variable.
In this view, language is a result of certain possible cognitive
structures/processes, the wish to communicate as much and as nuanced
as possible, and the natural characteristic to make this cost as
little energy as possible. It is even quite possible to imagine how
the latter two have influenced, or even formed the first, during the
many years of the development of our cognitive and genetic
characteristics (and vice versa?), but it is not essential to the
framework. I wish to emphasize (for obvious reasons) that this is not
a behaviorist point of view. Constraints CAN thus be falsified if
they are not rooted in factors such as those described above. I do
believe this may serve as an argument against Carvalho's statement
that OT is necessarily circular in its approach to, for instance,
syllable structure. On the other hand, it has to be acknowledged that
many OT'ers still do not apply these considerations in the
definitions of constraints and that in these cases Carvalho's point
is totally valid.

Dirk den Ouden


===========================

Dirk-Bart den Ouden
Department of Dutch
University of Groningen
P.O. Box 716
9700 AS GRONINGEN
Tel: +31 (0)50 - 3637412

===========================

Joaquim Brandao de Carvalho (LINGUIST 8-670) makes some interesting
points with regard to the formal status of OT, as regards other
'theories'. As far as I follow his remarks, he argues for two points:

1.
>constraint falsification could always be viewed by OT supporters as
>constraint violation. Now, as is well-known, OT constraints are violable by
>definition.

As Joaquim points out, the fast that this is alowed in OT is its main point
of difference with other linguistic theories; it takes account of the
exceptions better than do the other theories out there, and as Joaquim
points out, "let us not be afraid of exceptions ; exceptions do exist, and
are, thereby, marked". All generalisations in all theories of linguitics
are violable/find exceptions, and this should not be a stumbling block to
using them (provided the exceptions don't outnumber the rules!).

He then makes an unusual critique about OT methodology:

2.
>The important point, from which it can be shown that OT is NOT a linguistic
>theory nor a theory tout court, is that constraints are formally arbitrary
>and thereby circular. ...

[example follows showing that, in essence, OT constraints are based merely
on an examination of a large corpus of lingustic data rather than a
theory-internal reason, concluding with:]

>Why is a given system unmarked rather than marked ?
>Because it fails to violate the constraints posited by the theory. But why
>does the theory posit a given constraint rather than its contrary ?
>Because this is what we find in unmarked systems.

now, it seems to me that two points are raised here:

1: the basis for positing any lingusitic theory

2: details of the architecture of that theory

With respect to 1, surely we are going to posit any model of linguistic
patterning on a large corpus of data (either from one language or from
many), and on the basis of the observations we make with respect to that
data, conclude that certain elements need to be incorporated or discarded
from the assumptions that we come to view the data with. Rather than being
arbitrary, this process of observation, inference, testing, and refinement
is in fact a responsible approach to modelling, and through that deriving a
theory of, lingusitic systems. An approach that simply decides on an
approach before any data is considered would be guilty of not responding to
the experimental stimulus, in this case data.

With regard to 2, the way in which the architecture of any theory works is
going to reflect the need to model the greatest amount of data possible -
the least marked form - as economically as possible, and account for the
left-over s- the marked forms, the exceptions - with as little extra
architecture as possible. The wayn in which these differences in markedness
are handled will of course vary from model to model, and in the case of OT
it depends on a set of ranked constraints.

Finally, Joaquim says that:
>In sum, OT must run after the facts because it does not provide an
>independent theory of markedness (for syllable structure in the present
>case), contrary to what should be expected from a real linguistic theory.

Again, I find two points are salient here:

3: OT has no theory-internal reasons for judging markedness

4: a 'real' lingusitic theory should have its own motivations

With respect to 3:
This is a very good point, and the fact that OT is very data-grounded, and
contains very little intervention between the data and the output
(predictions, selection of candidates) is a reason for perhaps considering
it to be a model of the data, rather that a theory. However, as pointed out
above, all theories judge markedness only on the basis of a corpus of data.
This ties in with a discussion of 4:

However, in essence that is all that any of the linguistic 'theories' in
widespread use do: model data, with various degrees of success,
arbitrariness, and complexity. Some (most) theories include large
mechanisms between the data and the predictions / gramamticality
judgements, but they still derive these mechanisms from a study of real
lingusitic data and an attempt to generalise a principle over these data.
In that sense, all of the 'theories' are really more or less abstract
models of lingusitic competence, rather than independent construals of the
way languages should work. The fact that all theories are based on
linguistic data, rather than idealised conceptions, says a great deal about
the empiricism built into the field.since its beginnings all those
thousands of years ago. The fact that one theory has managed to rach its
predictions and gramamticality judgements without too much (essentially
arbitrary, if we factor out actual linguistic data) architecture, should
not be seen as a drawback, but rather as an enviable asset. It means, in
effect, that the model with the least internal complexity is most free to
adapt and learn from other models of language - witness the use of OT by
phonologists, LFG syntacticians, GB syntacticians, and many more (sorry for
not listing everything - see the recent linguist posting of the Hopkins OT
workshop's set of papers and posters for a very topical reminder of the
range of practitioners of OT).

In sum, the motivations for any model/theory of linguistics should be, and
are, based in langauge data; most models/theories have internal structure
of varying degrees of opaqueness, and it happens that OT is on the low end
of that scale, proceeding from input ot output without too much
architecture. So be it.



_____________________________
Mark Donohue
Department of Linguistics
University of Manchester
Manchester M13 9PL
England

mark.donohueman.ac.uk
+ 44 - (0)161 275 3259
_____________________________