Reproducibility

[G.Lazard]

Dear Frans,

I have been thinking for a long time that linguistics in general and
particularly linguistic typology are in great need of theoretical
clarification. For this reason I am glad that you have initiated a
discussion on the reproducibility of experiments, an issue which necessarily
entails other more basic questions.
The so-called human sciences are not real sciences and linguistics is one of
them. This does not mean that it is an art. Its aim is not to provide
esthetic pleasure (although the linguist may feel pleasure when he has
felicitously achieved a research) nor to fulfil a practical function, but to
produce new knowledge on some aspect of its object, i.e., a language or
languages. However its results are not scientific like those of, say,
physics or biology, which are such that all competent scholars agree on what
is correct and what is incorrect: this is what is called objectivity, the
ultimate proof of which resides in the technical applications of science.
Linguistics has nothing similar. Its results are shaky because they
practically always involve a part of subjectivity.
However the kind of knowledge produced by linguistic research is not pure
fantasy, far from it. It certainly reflects something of the reality of
languages, but it is vague, because the notions on which linguistic research
is founded are ill-defined. The task that faces us is therefore to make it
more precise, by making its foundations more secure. I think that this is
possible, and even that linguistics is probably, among human sciences, a
discipline which is more able than others to be brought closer to the
statute of a science (for more details see what I wrote at some length in
French in "Bulletin de la Societe de Linguistique" 94, 1999, and more
briefly in English in "Folia Linguistica" 36/3-4, 2002, and in the next
issue of "The Linguistic Review" to be published shortly).
The main problem with typology is the necessity of an independent standard
for the comparison of languages. Linguistic categories, being
language-specific, cannot be used as an independent standard, and other
human sciences do not provide objective secure categories. Linguists
therefore either make use of the traditional grammatical categories, which
are neither  adequate for any language nor completely inadequate (this is
the reason why the results are neither mere speculation nor scientific
knowledge), or make a personal choice of a conceptual framework, which may
prove fruitful, but the results achieved by different scholars cannot easily
add up.
Then what about reproducibility? There is no experiment in linguistic
typology, but only observation and comparison. To reproduce a research would
be to choose exactly the same initial framework as in the former research,
use the same sample of languages and of course follow the same procedure. It
would be conclusive only if the initial framework were perfectly clear and
well defined, which is not so easy. So, in the present state of the art, it
may be thought that the most urgent task is not to ensure reproducibility,
but rather to refine the theoretical basis of  linguistic typology and to
make clear the methodological path (or paths).
Best wishes.

Gilbert Lazard
49 av. de l'Observatoire
F-75014 Paris