Number marking and non-count nouns

[Alan R. King]

David Gil's rejoinder to my arguments concerning the obligatoriness (or
not) of plural marking in English in the light of non-count nouns raises
important issues, which I feel transcend the limits of the original
discussion and deserve to be addressed in their own right.

Actually at issue in the present phase of discussion, as I see it, are two
quite different kinds of question, both of great typological interest
thought in different ways.  We have a question about methodology: in any
typological study, what phenomena do we include among the variables to be
tabulated, cross-referenced or whatever - and how do we decide this?  But
in any case, David has also thrown out some statements regarding the
empirical facts of +-count distinctions among languages with number marking
that beg to be addressed in further depth and breadth.  I suggest we go on
discussing both of these issues, either separately or jointly.

Let me first recap the methodological issue, as I see it.  I don't know if
my previous analogy using ergativity and universal subject properties was
helpful or even accurate.  Anyway, what I'm trying to say is that, if one
of typology's main aims is to find out how languages differ (the other
presumably being to find out how they don't differ), and if (although I am
no defender of imposing any sort of aprioristic theoretical strait-jacket
on our investigations) we want to move towards whatever degree of
insightful systematization of our findings those findings genuinely admit
and justify, then language data must at some point be passed through
certain types of filter.  This is a principle that can be (and in my
opinion sometimes has been) abused in modern linguistics, but that in no
way invalidates it if applied sensibly.  It is also a principle that comes
dangerously close, perhaps, to merely stating the obvious.  We know that in
perceiving, and processing our peception of, the world, "filters" are
always needed, otherwise we would be eternally stuck at the level of
multitudinous chaotic impressions.  All I'm saying, basically, is that in
typological work too, we must always try, eventually, to move beyond "the
level of multitudinous chaotic impressions" to that of SOME sort of
UNDERSTANDING.  So we need filters; but of what kind?

I am also no reductionist or minimalist, with either a small or large "m".
On the other hand, we cannot allow ourselves to get eternally stuck at the
kind of impasse illustrated by a dialogue in the American comedy show of
years ago, "The Munsters", in which, father Munster having disappeared, his
wife, giving the police his description, in response to the question
"Eyes?", answers "Two."  Of course this is elementary pragmatics:
utterances should be not merely true but relevant.  Is Mrs. Munster's reply
relevant?  Well, that depends on the assumptions made about the number of
eyes people are *expected* to have.  Which in turn hopefully rests on
empirical facts about how many eyes people do have, or can (in practice) have.

Transposed to the domain of linguistic typology, this entails a problem of
potential circularity.  Because the question of "how many eyes a person can
have," i.e. of "what sort of X's a language can have", is *both* an
empirical question to be discovered by the typologist *and* needed as a
presupposition for the typologist to be able to investigate - and "filter"
in an appropriate manner - empirical data about her or his object of
enquiry.  So there are lots of opportunities - far too many, unfortunately
- to stack the cards, or simply, without necessarily meaning to or
realising it, to produce results that are only true or meaningful in
relation to other results: we demonstrate, effectively, that either
propositions P and Q are both true, or both false (well, more or less).
That doesn't necessarily make them true, but we sometimes tend to behave as
if it did.

Despite the inherent danger of circularity, it seems to me that we have no
other way to proceed, at least at some stages in our research, if our aim
is to produce, in some sense, systematic findings rather than a jumble of
raw observations.  We must take the risk.  And certainly sometimes we will
back the wrong horse and have to backtrack.  But that's not so terrible, in
the long run.  Still, I see it as a typologist's dilemma.

My reaction, when my affirmation that Germanic languages have obligatory
"unrestricted" nominal number marking was challenged by David's reference
to non-number-marking non-count nouns, was that which the policeman
presumably had when Mrs. Munster replied "Two."  Of *course* non-count
nouns don't number-mark, that's actually how we know they are non-count
nouns (I think).  Don't ALL languages with number marking have non-count
nouns?  I actually said that I didn't know, but bet they did.  And now
David disagrees, affirming that he knows of languages that disprove my
assumption.

So here are the two very interesting (though completely different)
questions (or clusters of questions).

(1) Am I right, as a typologist, to suppose that I should identify and use
"filters" based on universal principles that I believe are true to decide
how to classify my raw data?

Can we, and do we need to, theorize and perhaps establish some principles
or recommendations of general application on this methodological point?

(2) Do all number-marking languages, in fact, have a grammatically relevant
+-count distinction among nouns?

If so, does this give me as a typologist the right to *discount* the
existence of that distinction in an obligatory / non-obligatory
classification of number marking systems?

And is it important FOR THE PURPOSE OF THIS QUESTION that Language A and
Language B, both of which have both (a) number marking and (b) non-count
nouns that cannot be number-marked assign the non-count feature to
*different* sets of nouns?

I've already said enough for now about question (1).  Let me end with a few
thoughts about question (2) and David Gil's observations on the subject.

>And -- in response to Alan's questions -- [the mass/count distinction]
clearly is a
>parameter along which languages may vary.  Some of this variation is
>probably rather idiosyncratic.  But much of it is in fact systematic,
>following a specific "animacy" hierarchy, as pointed out by Smith-Stark
>and recently much elaborated upon by Grev Corbett.  For example, there
>are languages where number marking is only possible for animates, or
>only for humans, or only for a subset of important humans.  Surely, the
>fact that in such languages a word such as "table" isn't marked for
>number is relevant to a discussion of ways in which number marking can
>fail to be obligatory across languages.

In spite of myself, I find David's reasoning somewhat convincing, or at
least disquieting.  As I previously observed, Hawaiian is an example of a
language which, at least as regards *morphological* number marking,
differentiates "a subset of important humans" from all other nouns in
allowing them to be so marked (the same is true of all the other Polynesian
languages for which I've checked).  I can probably save myself this time
round because the other nouns, which lack morphological marking, are also
marked for number: analytically.  Whew!  But there no doubt will be
languages, and I'm sure they've already been collected together in the
literature (I happen to have rather limited access to the full range of
literature, unfortunately), in which such a subset of nouns are the only
ones that can be number-marked at all.

Now we could deal with this, as far as I can see, in two ways.  We might
say that since *some* nouns in such a language vary for number, the
category exists, even if most nouns cannot manifest it.  Like saying that
number is there all right, but most nouns behave morphologically like
English "sheep".  Of course the rest of the grammar (e.g. verbal agreement)
might be able to resolve that hypothesis one way or another, but then again
perhaps it doesn't (because there is no verbal agreement, etc.).  What
then?  It seems I must decide, rather arbitrarily, whether the largest
subset of the noun class are "potentially" singular or plural or not.  And
like a line of dominoes, this might push my typological results one way or
the other.  So we really do want to have some principles.  (Sorry, I
strayed back to Question (1) there.)

Well, having strayed, let me say something a bit nasty about our discipline
at present.  Faced with such dilemmas, I perceive in a certain amount of
(often otherwise excellent) typological work the insidious presence of an
initial Global Hypothesis (about the issue studied, or about Language
altogether) which is allowed to sway the interpretation one way or the
other, with the cumulative effect that the desired conclusion is reached,
but to my mind the whole outcome is thereby made something less than
kosher.  So, are we to condemn Global Hypotheses in typology?  Help, I'm lost!

In peace,

Alan