LINGUIST List 5.1237

Sat 05 Nov 1994

Sum: Basic word order (and remarks on typology)

Editor for this issue: <>

Directory

1. Frederick Newmeyer, Summary: Basic order (and remarks on typology)

Last week I asked for references to discussions of a problem that comes
up in linguistic typology: When there are conflicting or ambiguous
criteria for deciding whether a particular language is a particular
'type' with respect to some feature (word order, clause alignment, or
whatever), how does one decide how to assign that language? I would like
to thank the following for their helpful replies: George Huttar, Yehuda
Falk, Dan Everett, Larry Trask, Jon Aske, Mike Maxwell, Mark Newson, Bill
Croft, Georgia Green, Ingo Plag, Randy Harris, and Andrew
Carstairs-McCarthy.

I was quite surprised at the small amount of published attention that
there is to this problem. I was pointed to short discussions (no more
than a couple pages) in some of the major works devoted to typology: the
seminal Greenberg paper, Comrie's 'Language Universals and Linguistic
Typology', Croft's 'Typology and Universals', and Hawkins' 'Word Order
Universals'. It was also suggested that I look at Doris Payne's 'The
Pragmatics of Word Order' and to papers on Yagua by Payne and Dan Everett
and on Tzotzil by Judith Aissen.

What prompted my query was a reading of Johanna Nichols' Linguistic
Diversity in Space and Time, which I found extremely impressive. But all
through it I had an uneasy feeling caused by her pigeon-holeing languages
as 'SVO', 'head-marking', 'active-stative', or whatever. Since so many
languages are *not* transparently one particular 'type' on the surface, I
wondered what the basis for these type-characterizations was. There is no
general answer given to this question for an obvious reason: neither
Nichols or anyone else could have profound first-hand knowledge of more
than a small handful of the 174 languages in the data base. I suspect
that in most cases Nichols could not know what criteria were applied to
type a language in the sources she consulted, because many sources are
insufficiently explicit on that point or take as self-evident some
categorization that another would take as controversial or simply wrong.
(Consider, for example, her typing French as VSO.)

There were, to be sure, cases where Nichols threw out some language from
the sample of some particular feature because of its obvious ambiguous
status with respect to that feature. But doing so could have created more
problems than it solved. As both Aske and Croft pointed out in their
postings to me, if a language is 'inconsistent' with respect to a
particular feature, that too is typological data; data moreover that
could be highly relevant to conclusions about stability and diversity
over time. In a sample of 174 languages, misassignment of several
languages within a category with a 3-way division could lead to rather
different conclusions. Likewise, so would postulating a different set of
categories or having categories specifically for 'mixed' types.

This is beginning to sound like a critique of Nichols, but I don't mean
it to be. Rather, it is more a commentary on the shaky art of typological
pigeon-holeing that underlies not just conclusions about language
prehistory, but also much functionalist theorizing and -- increasingly --
generative theorizing as well.

There is also the question of sample *size*. Typologists strive, quite
reasonably, to correct for genetic and areal biases in their samples (the
most heroic effort along these lines that I know of is Dryer's work). But
how confident can we be of any attempt to eliminate bias from the sample,
given Nichols' conclusions that influences can extend half-way around the
globe? And doesn't that present a challenge to purported explanations of
the relative frequency of some typological feature, which are common in
the functionalist literature and increasingly so in the generative? So
much could be the result of historical accident on the one hand and
contact and descent on the other, rather than the product of 'external'
functional forces or the design of UG. The smaller the sample of
languages where mutual influence or common descent is not a possibility,
the more likely that some implicational typological relation is
artifactual. And the more reason we have to think that there are a lot of
typologically possible but -- purely by chance -- nonexisting languages.

Fritz Newmeyer
fjnu.washington.edu

PS: With respect to the last point, Alan Bell has shown that if some
feature appears in 1% of the world's languages (say, 40-50 languages), it
will show up only about 50% of the time in a random sample of 75
languages. You'd need a sample of over 200 languages before it could be
counted on to show up 90% of the time. And we are assuming here, utterly
counterfactually, that there are no genetic relations or areal influences
between languages.