re the reproduceability of typological claims

Tue Aug 24 20:28:24 UTC 2004

These are some thoughts on the reproduceability and further testing of language-typological claims.

1. LANGUAGE-TYPOLOGICAL CLAIMS

A language-typological claim will conform to the following schema:

  "In all (or in most; or in a certain percentage of) 

   languages of a given universe, there is X."

X is a structural property; the "given universe" is either the set of all languages (in the case of unrestricted universals) or the set of all languages that have property Y (in the case of restricted universals).

Typological implications will have been arrived at on the basis of a language survey. In most cases, the claim cannot have been arrived at by inspecting all languages in the universe assumed but only a subset - a sample - taken from that universe. 

2. DEGREES OF REPRODUCEABILITY

A typological claim is reproduceable if it fulfills the following two requirements: 

(a) CORRECTNESS OF DATA RELATIVE TO THE SOURCES GIVEN

    The data utilized for every language in the sample are correct given the sources used by the author and given the definitions of the grammatical terms in question adopted by the author. 

(b) CORRECTNESS OF LANGUAGE SAMPLE RELATIVE TO THE SAMPLING 

    METHOD ADOPTED

    The sample chosen conforms to the criteria of the sampling method chosen by the author.

What this means is that a typological claim is open to reproduceability tests only if

   (A) the full list of languages of the sample used is 

       given

   (B) references to the data sources for each language is 

       given (for printed sources, page numbers must be 

       included!)

   (C) definitions of the grammatical concepts utilized are

       given

   (D) the sampling method adopted is identified.

If the actual data are also included in the study, so much the better since it makes the job of the reproducer easier; but as long as the data sources are precisely identified, this does not seem necessary.

3. THE IMPORTANCE OF LISTING THE LANGUAGES IN THE SAMPLE

   It is particularly important to list the languages of the sample for implicational (as opposed to unrestricted) universals. This is for the following reason.

   Suppose a typologist has 100 languages in his database; he wants to test an implicational universal of the form "If a language has Y, it also has X"; and Y, the implicans, is present in only 5 of the 100 languages. Can he then say that he is testing the universal on a sample of 100? The answer is no: the relevant test sample consists only of 5 languages since the remaining 95 are irrelevant: they do not provide test conditions for the implicational statement.

(Analogy: Suppose I have 100 cats, I want to test the claim that cats that have black fur have black whiskers, and only 5 of my cats have black fur. If so, my test sample consists of only 5 cats, not 100.)

The point is this. An implicational universal is a claim about a universe of languages delimited by the implicans and thus, as all universals, it has to be tested with respect to the universe that it is claimed to hold for. 

What this means is that implicational universals with differing implicantia require different language samples each consisting only of the languages that have the implicans. Furthermore, the issue of the sample being or not being genetically and areally balanced needs to be examined with respect to that specific test sample and not for the entire set of languages that the researcher may have inspected or may have available in his database.

This is a simple, common-sense point but I have the impression that it is lost sight of at times.

4. FURTHER TESTING

Suppose a typological claim is REPRODUCEABLE in the sense described above and by reproducing the survey, it has in fact been found to be correct. This means that the author is "off the hook": he did everything right that he set out to do. But it does not yet mean that the typological claim is actually valid. FURTHER TESTING would involve additional checks:

   (a) Using data for the languages from sources other than those cited by the author.

   (b) Using different definitions of the grammatical terms involved.

   (c) Sticking with the author's sampling method but using it to construct a different language sample from the one used in the study. (Cf. Larry Hyman's point in his message of August 24.)

   (d) Using a different sampling method for constructing a new sample for testing the claim. 

Edith A. Moravcsik
Professor of Linguistics
Department of Foreign Languages and Linguistics
University of Wisconsin-Milwaukee
Milwaukee, WI 53201-0413

E-mail (office): edith at uwm.edu
Telephone (office): (414) 229-6794
Fax (office): (414) 229-2741
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/lingtyp/attachments/20040824/84358c4a/attachment.htm>