Spanish and pro-drop (LONG)

[Alan R. King]

With regard to the question of object pro-drop (and the more or less
related issue of object indexing), the case of modern Spanish (in its
numerous varieties) strikes me as a rather difficult case; so much so that
even after having spoken Spanish daily for over half of my life (some 25
years), I am very hesitant to make confident generalizations on the
subject.  But then I have never specialized in Spanish grammar; I am just a
non-native speaker who happens also to be a linguist.  On the other hand,
as a typologist, I suppose that precisely the fact that Spanish is
difficult to classify should make it worthy of interest.  With some
trepidation, therefore, I shall attempt to address and expand upon Fernando
Zuniga's (¿o no será "Zúñiga"?) point.

First I shall recall that while Zuniga refers to Spanish, rather similar
phenomena occur in the other Romance languages of the Iberian Peninsula,
obviously with some differences of detail, comparable to those noted by
Zuniga to exist among varieties of Spanish itself.  So these phenomena are
perhaps common to Catalan, Portuguese and Galician.  To a lesser degree,
they can also be seen to be related to patterns of clitic use in the other
Romance languages, as Zuniga noted, though there is rather more variation
at this level so that generalizations across Romance require caution.  I
think Iberian Romance has taken the obligatory use of co-referential
clitics (to use a hopefully neutral designation that does not prejudge the
theoretical interpretation of the fact) referred to in Zuniga's
contribution further than any of the other Romance languages, and within
Iberia perhaps Spanish.  In other words, Spanish (and its peninsular
relatives) have probably come closer to acquiring both object indexing in
the verb complex and a concomitant object pro-drop principle than the rest
of Romance, and for that matter than the rest of Indo-European Europe as
far as I am aware; perhaps, indeed, than any other Indo-European languages?

Assuming what I have just said is right, then it surely deserves to be
mentioned that the only non-Indo-European language of the Iberian
Peninsula, Basque, is decidedly object-indexing and object-pro-drop; the
former fact is built into the language's morphology, and the latter can be
seen in relation to it.  This suggests to me that we may possibly be
looking at an local areal phenomenon.  At the same time we should not
forget, however, that Iberian Romance, in developing these tendencies, has
built on trends seen in varying degrees across Romance which provide the
"groundwork" or "infrastructure" for the subsequent Iberian evolution.

Now to the grammatical discussion.  At the risk of repetition and stating
what is obvious to many readers, I would like to begin by looking at the
nature of *subject* indexing and pro-drop in Spanish first.  I believe this
makes an interesting backdrop against which to evaluate and perhaps
contrast the state of object indexing and pro-drop, since the facts for the
subject are relatively cut-and-dried.  I hope this is not seen as
belabouring the point unnecessarily, but I am persuaded that the issue is
of sufficient interest and importance to warrant dwelling upon, and would
like to see more discussion of the assertions I shall make here.

Like its ancestor Latin and much of Romance, Spanish has subject indexing
and subject pro-drop.  Thus:

(1)
SPANISH
Ve-o                unas casas.
see-PRS.SPI1s some houses
'I see some houses.'

(in idiomatic English we'd normally say 'I can see some houses' with the
same meaning.)  (Abbreviations: PRS = present tense; SPI = subject person
index, 1s = first person singular.)

The -o in veo is a portmanteau morph which combines tense and mood
information (present indicative) with a first person singular subject
index.  The presence of such marking is obligatory; finite verbs thus must
contain a subject index.  The above example thus illustrates the two points
that interest us: (1) the verb is obligatorily subject-indexed; (2) the
sentence contains no subject pronoun (thus illustrating pro-drop).

Now it may seem an obvious point, given the 2000-year tradition of Latin
and Romance grammar, but this analysis depends crucially on the assumption
that -o neither constitutes nor contains a pronoun.  If it were a pronoun,
then we would not need to refer to indexing here, and also Spanish would
then be non-pro-drop for subject, since the -o cannot be dropped.  It may
seem very obvious that -o is not a pronoun, but how could we state our
reasons for that judgment explicitly?

First of all: it has often been said that person-indexing morphemes usually
arise diachronically, if only we could go back far enough, from earlier
pronominal elements, and I don't believe we can discard the possibility
that long, long ago (earlier than proto-Indo-Europan, in fact) this -o
and/or other person indices in Spanish might have started life as pronouns.
 But even if we were sure of that fact and the source pronoun were known to
us, that would be judged irrelevant to the synchronic analysis of Spanish
under discussion.  Let us generalize the point: the possible source of
indices in original pronominal forms in no way affects our synchronic
evaluation of indices as something other than pronouns.

Secondly: there is a temptation to say that -o is not a pronoun because it
is an inflection, but I am not sure the point is as pertinent as it may
first appear to be.  Morphological realization is one matter, syntactic
function is another, and what we are most interested in is the latter.
Indeed pronouns may be realised as clitics, and person indices may appear
as particles, which are themselves perhaps clitic in form.

In my opinion the important thing to demonstrate here is the negative point
that -o IS NOT A PRONOUN, and this can surely be done by observing that
pronouns (in Spanish) have certain charateristics which the indices do not
share.  And I think that in this regard the most basic characteristic of
true pronouns is that they constitute NOUN PHRASES.  True, pronominal NPs
may and do have certain features to distinguish them from NPs having a noun
as head (or other types of NP), but they also have other important features
in common.  Whereas indices have no syntactic features in common with NPs.
I believe we need to insist on this criterion for identifying pronouns; not
to do so leads, potentially and indeed often in practice, to all sorts of
misunderstandings.

If we must insist that the ability to stand as a true NP head is a
necessary condition for something to be a true pronoun, then we must also
remember that, therefore, having the category of person (and often also
number etc.) is NOT a sufficient condition for calling something a personal
PRONOUN.  It is pretty obvious that while verbs (or verbal inflections)
often vary for person, this does not make them pronouns.  But the same is
also true of clitic items in some languages which may function as indexing
particles but lack, synchronically in any case, a pronominal character.  I
cannot insist too strongly that grammars, generally of non-European
languages, have often got this wrong.

Next, and crucially, we need to look at the way the indices are distributed
in clauses, and contrast that with the distribution of NPs (including
pronouns).  In Spanish, as already noted, subject indices are obligatory in
finite clauses; the presence of a subject NP is in every case optional.  In
the following examples, parentheses indicate the optionality of subject NPs:

(2)
SPANISH
(Yo) ve-o          unas casas.
(I)     see-SPI1s some houses
'I see some houses.'

(3)
(Tú)   ve-s          unas casas.
(you) see-SPI2s some houses
'You see some houses.'

(4)
(Él/Juan)   ve-0        unas casas.
(he/Juan) see-SPI3s some houses
'He/Juan see some houses.'

(5)
(Ellos/Juan y María)   ve-n           unas casas.
(they/Juan and Maria) see-SPI3p some houses
'They/Juan and Maria see some houses.'

etc.  There exists the complication that the third-person-singular index is
zero in (4).  Given the grammatical tradition of Spanish, nobody is likely
to argue that this proves that these elements are anything other than
indices, and certainly not that they are really pronouns and there is
furthermore "pro-drop" in the case of ve 'he/she sees'; but where
alternative analyses are possible this is precisely the kind of doubt that
can get us into trouble.

With the exception of cases like ve (whether we choose to analyse this as
the lack of an index or an index realized morphologically as zero), it
seems to be typically the case that indices are obligatory elements.
Whether pronouns are obligatory or optional, on the other hand, depends, of
course, on whether or not the language is pro-drop.  

Furthermore, the possibility of a pronoun occurring depends in most cases
on the absence of another coreferential noun phrase in the clause.  This is
just the opposite of the expected behaviour with indices.  Except where we
have 'zero indices' as just mentioned, indices are able to co-occur with
noun phrases that they index; normally these may have either nominal or
pronominal heads.

Thus, taking sentence (5) above, we find that the subject index may occur
without a subject NP (Ven unas casas), and may also co-occur with one
subject NP (Ellos ven
; Juan y Mónica ven
).  But the pronominal subject
ellos cannot co-occur with another NP such as Juan y Mónica: *Ellos Juan y
Mónica ven
; *Juan y Mónica ellos ven


There may be special cases where it is possible for two coreferential
subjects (or objects) to be present, as in the English sentence:

(6)
ENGLISH
My friend Jack, he lives in New York.

This is not a counter-example as long as it is clear that (a) such
"doubling" is not obligatory, i.e. the normal English sentence is "My
friend Jack lives in New York", and (b) the occurrence of "doubling" is
limited to specific cases (such as the topicalization of the subject, in
this instance).

These are, I believe,  the kinds of things that tell us whether we have
indexing or not, and pro-drop or not.  Now, what about object pro-drop in
Spanish then?

In Latin there was no indexing of objects and no object pro-drop.  A
development common to all Romance languages was the emergence of a series
of clitic object pronouns, deriving diachronically from the ordinary object
pronouns in unstressed position.  It is interesting to note that clitic
object pronouns are also found in non-Romance languages of southern Europe,
namely in the Balkan peninsula, so this could be an areal feature.

(Balkan Slavic languages vary regarding both the placement rules and
syntactic functions of their clitic pronouns in a way roughly comparable to
the variation among Romance languages.  Consider on the one hand the two
Croatian sentences:

(7)
CROATIAN
Vid-im      Rudolf-a.
see-SPI1s Rudolf-ACC
'I see Rudolf.'

(8)
Vid-im      ga.
see-SPI1s him.ACC
'I see him.'

(ACC = accusative); now contrast (7) with the Macedonian example in (9):

(9)
MACEDONIAN
Go         vid-ov              Grozdan-a.
OPI3sm see-PST.SPI1s Grozdan-ACC
'I saw Grozdan.'

(OPI = object person index, 3sm = third person singular masculine; PST =
past tense).  Apart from the fact that Croatian and Macedonian clitic
placement rules are different -- Croatian "vidim ga" versus Macedonian "go
vidov" -- which is not in any way essential to the present issue, note that
the cognate clitics, ga/go, have different syntactic functions in the too
languages.  Macedonian go seems to have become an object index, but not
Croatian ga, at least going by these examples.)

Clitics are characterized by tighter syntactic restrictions than free
forms; however there are differences of detail as to just how the clitic
pronouns are syntactically restricted in each language possessing them.  It
would seem likely that at an early stage, clitic pronouns in each of these
languages were subject to a second-position rule of some kind.  This state
of affairs still obtains in Serbo-Croatian, as (8) above also illustrates.  

But in the Romance languages the nature of the clitic placement rules has
evolved over time, giving various different results.  The language or
languages (depending on whether they are considered diverging forms of a
single language or two languages; this is currently under debate) that
has/have probably stayed closest to the presumed early common Romance norm
is/are Portuguese and Galician.  According to that norm, the clitic
precedes the verb if another element precedes, but follows the verb if the
verb is sentence-initial, cf. Galician:

GALICIAN
(10)
Quéro-te.
love.SPI1s-you
'I love you.'

(11)
Quen te quere?
who you love.SPI3s
'Who loves you?'

Spanish once had a similar rule, but like the remaining Romance languages,
has generalized the pre-verb position for clitics with finite verbs in most
cases, cf.

(12)
SPANISH
Te quiero.
you love.SPI1s
'I love you.'

(13)
Quién te quiere?
who you love.SPI3s
'Who loves you?'

It is my opinion (or at least suspicion) that there is no correlation
between the precise rules of clitic placement in the various Romance
languages and the syntactic functions that they assume in each language; I
would therefore tend not to attach so much importance to the placement
rules that happen to operate in Spanish as Zuniga suggests in his post.

Furthermore, as the above Slavic examples suggest, cliticization, which is
a change in form (in a syntactic as well as a morphological sense) need not
imply a change of function; clitic pronouns may still be pronouns, although
pronouns of a special type.  By saying that they are "of a special type", I
am trying to make allowance for the fact that, as clitics, they cannot
retain all of the "nouny" features of independent pronouns, which is
significant in view of my earlier insistence that pronouns constitute NPs.
At least they remain in complementary distribution with non-clitic NPs, as
is the case with ga in Croatian, which can occupy the same place in
argument structure as the noun Rudolfa (cf. (7-8) above).  Likewise in
standard literary French:

(14)
FRENCH
J'aime Paul.
I love   Paul
'I love Paul.'

(15)
Je l'   aime.
I   him love
'I love him.'

Here there is no reason for calling the clitic l(e) an index; it only
occurs when another object NP that would be coreferent DOESN'T occur in the
clause.  (This is of course different from the case of colloquial
constructions such as:

(16)
COLLOQUIAL FRENCH
Paul, je l'    aime.
Paul  I   him love
'(As for) Paul, I love him.'

but in French this only occurs when the object is topicalized and le can
then be considered a resumptive pronoun, as in the English gloss; see also
below.)

Cliticization, then, doesn't necessarily imply a change of function (from
pronoun to index), but it does help greatly to prepare the way in that
potential direction.  In diachronic perspective, cliticization is of course
a weakening of form, and the passage from pronoun to index may be
considered a weakening of function (from clause constituent to grammatical
marker), so both steps look like natural links in a probable
grammaticalization chain.  Furthermore, cliticization may well be a
*necessary* precondition to what we might call "indexization", since it
allows the originally pronominal element to lose its positional autonomy
(if it had it, as free NP arguments mostly do in Romance and Slavic, for
example) and perhaps be drawn into the verbal cluster.  (I say perhaps,
because cliticization doesn't necessarily involve that, cf. the
Serbo-Croatian clitic placement rule.)  This means, on the one hand, that
the clitic no longer occupies the same position in the clause as the free
NP, which presumably makes it easier for them to co-occur coreferentially,
because they no longer compete for the same positional "slot"; while also,
the clitic already fulfills a presumed condition for indices, that of being
a bound morpheme whose position is fixed.

The question is thus now: HAVE the Spanish clitic object pronouns actually
become indices, i.e. stopped being true pronouns? -- in which case we can
talk about Spanish also having object pro-drop, provided that in the
sentence (Zuniga's (2a)):

(17)
SPANISH
Lo               vi.
OPI?/PRN? see.PST.SPI1s
'I saw him.'

(PRN = pronoun) lo is not a pronoun but an object index; or, on the
contrary, is lo to be considered a pronoun, in which case this sentence
does not illustrate object pro-drop (only subject pro-drop).  Zuniga
suggests that in some varieties of Argentinean Spanish we have an object
index, hence also object-pro-drop, on the evidence of the acceptability, in
these varieties, of (18), with lo co-occurring with Jon (his (2d)):

(18)
SOME ARGENTINEAN SPANISH
Lo     vi                    a      Jon.
OPI? see.PST.SPI1s ACC Jon
'I saw Jon.'

while he recognises that in some other dialects (18) is ungrammatical (lo
and Jon cannot co-occur here), hence lo is presumably a pronoun, and
therefore (17) would contain a pronoun and no pro-drop would be involved.

In my opinion the question is perfectly well formulated but the answer is
not a simple one, because it requires us to raise a number of other related
questions.  Furthermore I admit that there is something worrying about this
whole matter, namely that what surely ought to be an important typological
parameter (pro-drop), and at the same time a rather important categorial
assignment (pronouns versus indices), all seems to hinge on what are in
fact very subtle nuances of usage, register, and regional variation
pervading Spanish in its many varieties (because this very kind of
vacillation is not restricted to Argentinean or even American Spanish).  I
think that the procedure of asking these kind of questions is the right
one, but I am still not certain whether the answers we find are going to be
realistic and meaningful.  I say this in a spirit of self-doubting, not of
facile criticism!

There are some contexts, such as that exemplified by (18), where different
Spanish varieties differ or where even the judgments of native speakers may
be uncertain, but there are other contexts that are relatively
cut-and-dried.  This is the case, for example, with pronominal objects, or
rather, where the English translation, at least, contains a pronominal
object.  For example, 'I love you' is normally expressed in Spanish as in
(12), "Te quiero", but if the "you" is emphasised, it may be rendered by
the free pronoun "ti".  This constitutes an NP, and so must, in this
context, be preceded by the accusative marker "a".  When the object is a
personal pronoun as here, the presence of the clitic "te" is generally
obligatory, in any variety of Spanish:

(19)
SPANISH
Te              quiero        a     ti.
OPI?/PRN? love.SPI1s ACC you
'I love *you*.'

(19')
*Quiero       a     ti.

Similarly when the pronominal object is third-person:

(20)
La              quiero.
OPI?/PRN? love.SPI1s
'I love her.'

(21)
La              quiero        a     ella.
OPI?/PRN? love.SPI1s ACC her
'I love *her*.'

(21')
*Quiero       a     ella.

(However, (19') and (21') may perhaps be acceptable, at least marginally,
in a very marked style, e.g. poetic, archaic...; but not, at any rate, as
normal modern Spanish.)  If te and la are indices, then indexing is
obligatory with personal pronoun objects.  If they are pronouns, then
pronoun "doubling" is obligatory.  This contrasts, as already seen, with
noun objects, when indexing or doubling is optional for Spanish as a whole;
here the difference may again be considered stylistic:

(22)
(La) quiero a María.
       love.SPI1s ACC Maria
'I love Maria.'

Another clearcut rule is with fronted topical objects, where doubling is
again obligatory in modern Spanish:

(23)
A     María la quiero.
ACC Maria     love.SPI1s
'I love Maria(TOPIC).'

(23')
*A María quiero.

I believe this may be generally true of all modern Romance languages (but
stand to be corrected).  

This fact, however, does not argue strongly in favour of the indexing
analysis.  Because (23) contrasts with (22), where "Maria" isn't fronted
and la is optional, it is sufficient to consider la a resumptive pronoun in
such cases.  If it were an index, we would expect it to be equally
obligatory in both sentences, just as the subject-indexing suffixes are
equally obligatory whatever the position of the subject NP.

Another important constraint on Spanish clitic doubling is that it is
obligatorily lacking when the object is inanimate (except under
topicalization, as just seen).  Thus:

(24)
Veo          esa  casa.
see.SPI1s that house
'I see that house.'

(24')
*La veo esa casa. (cf. 22)

(25)
Esa  casa  la veo.
that house     see.SPI1s
'I see that house(TOPIC).'

(25')
*Esa casa veo.

(I think (25') is starred, but it actually sounds okay to me because it is
sometimes used where I live, though I suspect this is a calque from Basque,
the local "indigenous" language!)  This rule is generally fulfilled,
although my formulation in terms of the single feature [animate] is an
oversimplification; other features such as definiteness also intervene in
determining what objects take doubling.  (A similar but not quite identical
combination of features involving the animacy hierarchy also determine
whether direct object NPs do or do not take the accusative marker a.)

This distinction does not actually condemn the indexing hypothesis.  One
way we could get around it would by to posit a (non-traditional in Spanish
grammar) syntactic "gender" distinction for object NPs opposing "animate"
and "inanimate", and say that, just as third-person indices differentiate
the standard masculine and feminine genders, so they also differentiate
between these and "inanimate", and that the inanimate object "index" is
"zero".  But taken in conjunction with the other provisos needed
(optionality of the index for "animate" objects, obligatoriness of the
non-zero index under topicalization even for "inanimate" objects, etc.,
etc.), this looks a bit too ad hoc to be fully convincing.

And then there is the question of dative objects, which I have so far
studiously avoided.  Zuniga says that in American Spanish dative objects
are obligatorily indexed.  At least in European Spanish, it seems not to be
quite fully obligatory: I believe that (26) (= Zuniga's (1c)), which he
gives as starred, is a possibility, although it is nevertheless very common
indeed to prefer (26') (= Zuniga's (1b)):

(26)
Presté el libro al estudiante.
lend.PST.SPI1s the book DAT.the student
'I lent the book to the student.' (starred by Zuniga)

(26')
Le presté el libro al estudiante.
'ditto' (preferred)

Other criteria may again interact.  With an inanimate dative object (which
is relatively unusual), indexing is less common, though still possible:

(27)
Ceuta pertenece      a  España.
Ceuta belong.SPI3s to Spain
'Ceuta belongs to Spain.' (preferred)

(27')
Ceuta le pertenece a España.
'ditto'

On the other hand, with the verb gustar 'be pleasing, i.e. like', the index
seems to be practically obligatory:

(28)
*Tu casa gusta a Juan.

(28')
Tu    casa  le gusta   a      Juan.
your house     please DAT Juan
'Juan likes your house.'

All in all, I can't quite go along with Zuniga's observation that "the _le_
indirect object marker for 3s ... is not less obligatory than the _-é_
portmanteau coding 1s subject, mode and tense" where European Spanish is
concerned at least.

The Spanish clitics could be firmly regarded as indices, as opposed to
pronouns, if all sentences with an NP object but without the corresponding
clitic were ungrammatical.  I don't know of any varieties of Spanish for
which this is the case across the board.  

Actually my reservations arise from looking at the whole issue not
atomistically (as I have just been doing), in terms of a variety of
different instances, but in its entiretly.  An indexing language is one in
which it is fundamental for indexed arguments to be marked (possibly by
zero) in the predicate; if the indexing language is also pro-drop for those
arguments that are indexed (as is often though not invariably the case, cf.
e.g. German which has subject indexing and is non-pro-drop), then we can
talk about the optionality of an NP being present in the clause referring
to the argument in question, as opposed to the obligatorily occurring index.

This principle is *not found consistently* for objects in Spanish.
"Indexing" of objects in Spanish is not obligatory across the board, so
object "indices" are not an essential feature of the Spanish clause.  They
are sometimes required, but only sometimes.  Some such cases, such as
topicalization, can be accommodated naturally into a "pronominal" analysis;
the rest must be covered by particular "doubling" rules.

I believe that the same can be said of the other Romance languages, though
the details obviously vary and the evidence for object indexing is weaker.

And if there are no object indices, only pronouns, then there is
consequently no object pro-drop in Spanish.

That is what I consider to be probably the best analysis of present-day
Spanish.  On the other hand, Spanish does look like it is moving in the
*direction* of developing into an object-indexing language.  And if so,
then its present situation, including the vacillations and
"inconsistencies" (e.g. doubling being obligatory with personal pronouns
but not with other NPs), may teach us about how indexing may arise in
languages.  It may also be worth looking more closely at other southern
European languages (Romance, Slavic...) for similar insights and contrasts
concerning the kinds of changes that are underway in some of these languages.

Most of these points were touched on by Zuniga.  He asked for further
discussion, and I am myself interested, not so directly in Spanish per se,
but certainly in the general issues raised, including implications of the
rather complicated and even ambivalent Spanish data for those issues.  I
also believe that in such cases it is important for typologists to pause
long enough to assimilate explicit, detailed data on individual languages
and their specific range of problems if one's broader conclusions and
methods are to be valid.

Alan