Fw: Call for Debate: Reproducibility in Typology

[Alice Harris]

Through a mistake on my part, my message did not go out to the entire
listserve on Monday.  Although Paolo left it intact in his response, some
members of the list have contacted me about the problem.  Therefore, I am
sending it again, below.  I apologize.  ACH

My understanding of reproducibility is somewhat different from Paolo's.  I
don't think the intention is to reproduce historical events, such as p >
f, or even the discovery of this sound change.  Biologists, for example,
do not try to reproduce diachronic events, such as the evolution of fish
or zebras.  Economists and historians don't conduct experiments (and
therefore can't reproduce them), but linguists do.  As Paolo pointed out,
scientists do reproduce or try to reproduce experiments.  Their aim,
clearly, is not to publish the results; it is to verify the results. Often
reproducing an experiment is the first step in going on to a further
experiment.  The need for reproducibility is the reason that scientists
devote so much space in an article to describing the exact conditions
under which they conduct each part of an experiment.

In typology, in most cases the equivalent of describing the exact
conditions is listing the languages in the data set, the sources of
information, and the essential criteria for identifying the items studied.
 For example, if one were doing a study of the order of articles and nouns
as related to the order of adjectives and nouns, it would be necessary to
define the concepts of "article", "noun", and "adjective", or to provide
criteria by which the investigators decided to include some items and
exclude others.  Another linguist should, then, be able to reproduce the
results using this same set of sources on the same languages with the same
criteria.  But the point is not only that someone should be able literally
to reproduce the results, but also that defining the conditions exactly
allows others to evaluate the results.  If, in my example, the
investigator defined "adjective" in such a way that it excluded
participles, another investigator might think the results would be
interestingly different (perhaps more significant) if participles were
included.  The second investigator might then repeat the experiment with
that condition changed. The idea of reproducibility as I understand it is
similar to "full disclosure".

Alice

Alice C. Harris
Professor
Department of Linguistics
SUNY Stony Brook
Stony Brook, NY  11794-4376
Phone: 631-632-7758, 631-632-7777
Fax: 631-632-9789
e-mail: alice.harris at stonybrook.edu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/lingtyp/attachments/20040901/33eca3ae/attachment.htm>