LINGUIST List 7.691

Wed May 15 1996

Disc: Syntactic Typology

Editor for this issue: Anthony M. Aristar <aristartam2000.tamu.edu>

Directory

1. John Verhaar, Re: Frederick Newmeyer on syntactic typology
2. Matthew Dryer, Syntactic typology and syntactic change
3. Simon Kirby, Re: 7.687, Qs: Syntactic typology
4. Martin Haspelmath, typology, change, frequency (Re: Newmeyer)

LINGUIST-NETters,
	Frederick Newmeyer's query and comments of May 10 represent some vital
questions about typology facing us today. Let me first grant that sampling
methods are important, and that both geographical spread and genetic variety 
are important. Then, there is the "contingency" (accidental nature) of 
even the most perfect sample, and of diachronic data.
	I do not find the example of the 30% - 70% distribution of features 
X and Y convincing. For whatever we would discover about language change 
(X arising about twice as often as Y arising, where cross-linguistically 
there is, -ex hypothesi_, no _tertium quid_ apart from X and Y), no matter 
whether new findings would confirm of disconfirm that devlopment, that would 
be as "contingent" as all the rest.
	The problem is, I think, really philosophical. It seems to me 
interesting that certain changes exclude certain other changes (for example, 
Johanna Nichols's law of head-ward migration, never dependent-ward migration, 
of constituents--one of the more interesting recent findings). But 
such findings are, again, contingent on historical accidents.
	If we want to know what language is like apart from the historical
developments, some or all of which is accidental, we would want to know about
some atemporal, suprahistorical, "essence" or "nature" of language (for 
example, that language is "mental"--in some noncontingent, and thus 
noncircular, sense). But there is, I suggest, no such object of knowledge, 
just as there is no metaphysical "nature" of anything else of the kind that 
we could know. We linguists, too, are contingent. . The philosopher Colin 
McGinn recommends "species modesty"--we are, after all, only limited human 
beings. But if we drop the impossible questions, and pursue laws like 
those found by Nichols and by sampling specialists, linguistics is 
still fascinating beyond one's wildest dreams.
John Verhaar

Some comments on Fritz Newmeyer's remarks about the problem of deciding whether
typological generalizations reflect "real" linguistic generalizations rather
than accidents of history. Newmeyer's basic idea is that if some language
type is twice as common as a second language type, and if that difference is
linguistically real, then we would expect to find syntactic changes in the
direction of the more frequent type twice as often.

Newmeyer brings up an apparently puzzling case. On the one hand, I argued in a
1989 paper that there is a linguistic preference for SOV order over SVO order,
based on their crosslinguistic distribution On the other hand, Newmeyer
suggests that there seem to be more known cases of changes towards SVO order
than changes to SOV order, which might suggest that SVO order is linguistically
preferred. While Newmeyer's question relates to syntactic preferences in
general, I illustrate some of the problems associated with the specific
question of SOV vs. SVO.

1. While I share Newmeyer's impression that changes to SVO seem more common
than changes to SOV, this would need to be systematically examined. My own
work on word order shows that impressions that linguists have had about word
order correlations are often not borne out when a sufficient amount of evidence
is examined carefully.

2. Note that among the four most common orders SOV, SVO, VSO and VOS, the
order SVO is the most natural result of change, since all three other orders
can easily change to SVO (changes from VSO or VOS to SOV or vice versa being
more dramatic changes). Thus if one were to examine historical changes to SOV
versus SVO, one should really restrict attention from SOV to SVO and not
include changes from VSO or VOS to SVO.

3. The frequency of different word order changes is immensely confounded by
the fact that many cases of word order change are clearly to be understood in
terms of language contact. To cite just one example, a number of Austronesian
languages in Papua New Guinea have apparently changed from VO to SOV due to
contact with non-Austronesian Papuan languages. Neither of these changes has
any bearing on the question of the relative "naturalness" of SOV vs. SVO. Nor
is it easy to factor out the role of internal factors and contact factors in
word order change; in many cases the two sorts of factors may be working in
tandem.

4. Behind the question of SVO vs. SOV is lurking an issue that is almost a
taboo in the field of linguistics, that surrounding what Bill Croft calls the
assumption of uniformitarianism. Namely, is it at least possible that SVO
order has been increasing in recent millenia, not due to some linguistic
preference for SVO order, but due to some nondeterministic association between
SVO order and certain characteristics of culture that have increased in
frequency in recent millenia? In fact, Givon proposed just this in his 1979
book On Understanding Grammar. The issue is not whether Givon's story is
compelling. The issue is whether we can simply dismiss the possibility of such
on a priori grounds. While the question again needs systematic examination,
SVO order is distinctly less frequent in Australia, New Guinea and the New
World than it is in mainland Old World and Austronesia: in my current database,
SOV outnumbers SVO by 69 to 17 (in terms of genera, groups comparable to
subfamilies of Indo-European) in the former "area", but only by 59 to 52 in the
latter area. More specifically, SOV order seems to be much more common than
SVO order in language groups involving small populations and relative genetic
isolatedness. Does this reflect a linguistic preference for SOV or does it
reflect some statistical association between culture type and word order? I
think that we DO have reason to suspect that there is a relationship between
word order and culture in the context of creoles: certain cultural situations
lead to creoles and creoles are apparently more likely to be SVO.

5. Even if the difference noted in the preceding point is unrelated to culture
features, the existence of areal patterns going across large areas of the world
(see also relevant work on this by Johanna Nichols) makes the entire enterprise
of determining whether differences in frequency are due to historical accident
ultimately unsolvable.

Matthew Dryer
Permanent email address: lindryerubvms.cc.buffalo.edu

Message 3: Re: 7.687, Qs: Syntactic typology

Newmeyer suggests looking at relative frequencies of attested language
changes of types X->~X and ~X->X to test for the significance of a
typological generalisation that X has a different frequency
cross-linguistically than ~X. Surely this is a non-starter. The original
problem is whether or not our sample of synchronic states is
representative, but this sampling problem is going to be at least as bad
for the diachronic cases. In other words, why should we be any less
circumspect about the significance of relative frequencies of changes of
types than relative frequencies of types themselves (especially
considering that the sample will necessarily be smaller). 

Another approach is to look at the significance of the typological
generalisations themselves. If our typological statement is that X is
somewhat more common than ~X, then I think that there are serious grounds
for worrying about sampling errors etc. However, compare this to the
statements embodied in typological hierarchies. For example, the classic
hierarchy in typology, the relative clause accessibility hierarchy of
Keenan and Comrie, sets out a series of chained implicational statetments
of the sort: "if a language has oblique relatives, then it will have
indirect object relatives, if a language has indirect object relatives,
then it will have direct object relatives" etc. For a hierarchy that
contains n such chained implications, there are (n+2) predicted language
types in a space of 2^(n+1) possible language types. The accessibility
hierarchy therefore predicts 7 out of 64 possibilities. For these kinds
of universal, even _if_ there were many many exceptions we should not
reject the hierarchy as a significant statement of tendency. Another way
of looking at it is: what are the chances that an "accident" in the
sampling technique would give us this sort of result? 

What does everyone think about this? Does this mean that we should ignore
any cross-linguistic statements that only refer to one feature of Language
in favour of those that make statements about the interaction of several
features? 

Simon

- 
Simon Kirby -- Department of Linguistics, University of Edinburgh
simonling.ed.ac.uk ------------ http://www.ling.ed.ac.uk/~simon/

A comment on Fritz Newmeeyer's discussion note regarding typology and 
syntactic change:

I think Newmeyer asks an interesting question that needs to be pursued, 
but omits an important consideration. His point is that in order to test 
whether a frequency skewing in the world-wide distribution of two features X 
and Y is significant, we might look at the frequency of change to X and to 
Y. For instance, if X is significantly more frequent than Y, then we 
should also expect changes to X to be significantly more frequent than 
changes to Y.
 What Newmeyer fails to take into account is the frequency of changes 
away from X and away from Y. I think that world-wide distribution is 
basically a sound method for determining whether a particular feature is 
universally preferred or dispreferred (in OT parlance, whether the 
corresponding constraint tends to be ranked high or low). I expect a 
universally preferred feature, e.g. voiced liquids, to be frequent in the 
world's languages, whereas the dispreferred feature, voiceless liquids, 
is rare. Newmeyer would now say that changes to voiced liquids should be 
much more frequent than changes to voiceless liquids, but this need not 
necessarily be so: It could be that voiceless liquids arise equally 
frequently, but are much less stable, i.e. there are many more changes 
away from voiceless liquids than away from voiced liquids.
 Let me give a concrete example from morphosyntax where I think 
something like this happens. Universally, inflection-outside-derivation is 
strongly preferred over derivation-outside-inflection. Now as Bybee (1985) 
has argued, this also has a diachronic correlate, but there are certainly 
changes by which dispreferred derivation-outside-inflection orders arise, 
e.g. when a reinforcing particle is added to a suffix-inflecting 
demonstrative (cf. Old Latin i-pse, eum-pse, eam-pse, etc., analogous to 
English this-here). In this way, a new (derivational) series of 
demonstratives arises, which has the derivational suffix outside the 
inflectional suffix (details and many examples in my 1993 paper in 
Lingustics). However, this situation is very unstable--typically further 
changes soon begin to reverse the order of inflection and derivation, to 
conform to the universally preferred state of affairs (Latin ips-e, 
ips-um, ips-am, etc.). Thus, the frequency of occurrence of a change to 
derivation-outside-inflection is significantly higher than the 
distribution in a world-wide sample, because the frequency of a change 
AWAY from derivation-outside-inflection is also significantly higher than 
the frequency of a change away from the reverse order.
 An analogous case from phonology might be the frequency of /y/, the 
high front rounded vowel, which is rare in the world's languages, but 
apparently arises more often than it is found synchronically--precisely 
because it is diachronically unstable and tends to disappear quickly.
 Thus, Newmeeyer's method, though in principle valid, is more difficult 
to apply than he assumes. Probably world-wide distribution will remain 
the more eeasily testable indicator of universal preferredness.

Martin Haspelmath
(Free University of Berlin)