pro-drop etc. (was: questions)

[Alan R. King]

I am not conversant with the language groups cited by D.N.S. Bhat (Salishan
and Munda), but based on what he says about them and on what I know of
other languages with recognisably similar characteristics, I would venture
to say that, in my opinion, these indeed exemplify a "classical" pro-drop
language type.  I get the impression that there are probably quite a lot of
languages around the world with this combination of features, of which the
most nuclear are (perhaps): (1) comprehensive indexing of arguments in the
predicate; (2) so-called "pro-drop".  Some other features that help to
configure "ideal" specimens of this type are probably: (3) noun classes;
(4) lack of case marking on core arguments at least.

In other words such languages are consistently head-marking at the level of
clause structure.  The head is the predicate (typically the verb or verb
group), which incorporates indices that "point" towards arguments.  This
predicate-with-indices complex constitutes the obligatory nucleus of a
clause.  The clause may optionally also contain noun phrases which are
lexical manifestations of the arguments, but these are typically optional,
their omission being what Euro-based linguists tend to think of as "pro-drop".

I have no empirical statistics on the frequency of co-occurrence of the
features mentioned, but the functional coherency of their combined
occurrence seems evident.  For example, we know that pro-drop can occur
without indexing, the arguments being recoverable through pragmatic
operations or other, more indirect of "devious" (so to speak) mechanisms,
Japanese being a well-known example; yet the presence of a well-developed
indexing system in other languages surely "encourages" the
non-obligatoriness of NP arguments (i.e. "pro-drop").

The reason why I mention noun classes in association with this is that
while arguments can be indexed, and are in many languages of course, with
specification of other features such as number, person and/or a limited
range of genders, an extensive set of noun classes (often in conjunction
with some of the aforementioned features) would seem to provide the best
possible basis for a rich indexing system.  And my reason for also listing
the lack of case marking is that a sufficiently highly developed
head-marking system makes cases (a dependent-marking device) unnecessary or
redundant in practice.

Apparently a nice, "well-behaved" exemplification of such a system is the
Papuan language Yimas (unless it only appears well-behaved because my
information about it is so limited!).  It has been described by Foley in
several publications, including his book _The Papuan languages of New
Guinea_.  In Yimas core argument NPs are not case marked.  Verbs index core
arguments for person, number, and noun class.  The noun class system is
highly developed, combining semantic (e.g. male, female, animal and plant
classes) and phonological criteria; for example, there is an -mp class
consisting of noun lexemes ending in -mp such as _impramp_ 'basket'.  Thus
if 'basket' is the subject or object of a clause, an indexing prefix on the
verb will indicate that that argument is in the -mp class.  The presence in
the clause of an NP consisting of the lexeme _impramp_ is itself optional.

Let me say in passing that European languages do not include attestations
of this language type.

The typological polar opposite of this head-marking type of organisation is
the consistently dependent-marking clause structure, whose most notable
feature is dependency on case marking of one sort or another to indicate
grammatical relations.  I say "of one sort or another" because I believe we
may be permitted to consider word order one means of "case marking".  Now
one by-product of this solution arises from the circumstance that if you
want to indicate a relation and the way to do so is to mark the dependent
(i.e. the argument), then you are forced to have an overt dependent to
mark.  Thus in this language type the presence of argument NPs may be
obligatory; their minimal (non-lexical) representation involves the use of
pronouns, which "cannot be dropped".

Some European languages come fairly close to this picture, while others
occupy part of the extensive middle ground that exists between the two
extremes.  Such languages combine certain features of both approaches to
signalling grammatical relations.  Indeed there is no incompatibility
between the different means available; their combination may merely give
rise to redundant marking on both the verb (the head) and arguments
(dependents).  We can think of the result as a "janus"-type clause
structure.  Perhaps to compensate for this potential redundancy, either or
both devices (indexing and/or case marking) may be applied only partially,
underdeveloped, under-differentiated and so on.  In Basque, for example,
there is consistent indexing on the verb for subjects and objects, and also
consistent case-marking of NP arguments; but there are no noun classes, and
indexing is only for person and number.  In other languages there may be a
systematic division of labour between the two devices, e.g. indexing of the
subject only, case marking of objects only, cf. some Romance languages
(although the details are more complex than this).

Now to address Bhat's point: how do we distinguish between what I call an
*index* and an *(NP) argument*.  I agree that this is the crucial point on
which the above typological distinction rests. The only clear criterion I
can see is that of NP-hood: an index is not a noun phrase.  It doesn't look
like one (morphosyntactically); it doesn't behave like one; it doesn't
function as one.  For this theory to make sense, certain things need to be
assumed.  For one: when we say that indices are un-NP-like, there must be
something else that occurs in clauses that *is* NP-like for them to
contrast with.  For another: we will need to establish what the
characteristics of NPs are for the present purpose.  And here there is
admittedly perhaps a danger of circularity creeping into the theory, if NPs
are defined in just such a way as to prove the theory right.  So it would
be good to have some cross-linguistic guidelines on how to identify NPs
and, if possible, discriminate between these and non-NP indices.

I think it is clearly possible for typologists to lose our footing on
issues like this and end up not being sure what we're talking about.  And I
think this danger requires us to be a little bit wary of ad hoc or "exotic"
definitions within particular languages, such as the one according to which
(1) apparent NP arguments are redefined as "attributes" of the "real"
arguments or as not being in the clause at all; (2) the "real" arguments
are identified with indices, never with lexical arguments or NPs.  And I
must point out that this has been attempted from time to time as a way of
accounting for the structure of a number of languages with head-marking or
janus-type clause organisation.

The problem, it seems to me, is not that such "exotic" analyses are
out-and-out wrong; they actually catch certain insights about the
structures of those languages.  The problem is that by using established
terms in unorthodox ways we can end up in total confusion, and thereby lose
even more basic insights.  Things like "clause" and "argument" are possibly
artefacts of the linguist's analytic capacity rather than something
inherent in the linguistic structure per se; let us not get into a
religious-like war over what constitutes a "true" clause or a "true"
argument as if these were god-given.  Typology is after all also a human
product.  So it is up to us to agree to use such terms in the most
convenient way to alow us to go on talking about linguistic structure and
understanding each other.

And this is precisely what I think we need to do with the term "pro-drop"
too (unless we actually decide it doesn't refer to anything worth referring
to, or agree to keep the concept but change the term for it, which might
not be a bad idea if anyone can find a better suggestion).  It is in such a
sense that I continue to think it is *useful* to distinguish between
"pro-drop" languages like Basque, Yimas and the Salishan and Munda
languages, and "non-pro-drop" ones like English.  Of course, the existence
of languages in an intermediate position, or whose classification is made
awkward by certain complications or ambiguities of structure, must also be
allowed for.