Is there a little pro?
Anders Holmberg
University of Durham / CASTL, University of Tromsø
1. Introduction
Sentence (1) is a standard example of a so called null subject.
(1) Están cansadas.
be-PRES.3PL tired-F.PL
‘They are tired.’
(Spanish)
The sentence has no overt subject, yet, according to standard Principles-andParameters theory it has a phonetically empty subject pronoun, so called (little)
pro, formally licensed and interpreted by virtue of the agreement on the finite
verb or auxiliary.
This paper is a scrutiny of this hypothesis, which goes back to works such
as Chomsky 1982 and Rizzi 1986, in the light of more recent developments in
syntactic theory, particularly the Minimalist program. I will argue that there is a
null (unpronounced) pronoun in the subject position in (1), but with properties
which are quite different from those of pro in standard Principles-andParameters theory. I thereby disagree with claims made in Anagnostopoulou &
Alexiadou 1998 and Manzini & Savoia 2002, according to which constructions
such as (1) do not involve any kind of subject pronoun. Crucial evidence comes
from null subject constructions in Finnish.
2. The GB theory of empty categories
Pro started its life in Chomsky 1982 as part of a theory of empty NPs,
which in turn formed part of the theory of NP-types and Binding, one of the
cornerstones of the Government-Binding (GB) program According to this
theory the different types of empty nominal categories that had been identified
and whose properties had been intensely investigated, namely NP-trace, whtrace, PRO, and now also pro, were really special cases of the same category, an
empty nominal category with no inherent properties apart from (presumably)
nominal categorial features and maximal X-bar level: [NP e]. The grammatical
properties of [NP e] are ‘functionally determined’, that is, they are determined by
the syntactic relations that it enters into, particularly the binding relation, as
follows (in the Chomsky 1982 version):
(2) [NP e] is
a. wh-trace if it is locally A-bar bound,
b.
c.
d.
NP-trace if it is locally A-bound from a non--position,
PRO if it is locally A-bound from a -position,
pro if it is governed by strong enough INFL or by a clitic.
The four types of empty categories correspond to three types of overt NP,
defined by the two binary features [+/-pronominal , +/-anaphor], plus one type
which can only be empty.
(3) a. [-anaphor, -pronominal]: R-expressions (referential NPs) and whtraces;
b. [+anaphor, -pronominal]: anaphors (reflexives, reciprocals) and NPtrace;
c. [-anaphor, +pronominal]: pronouns and pro;
d. [+anaphor, +pronominal]: PRO, which in principle cannot have an
overt counterpart: the so called PRO Theorem (see Chomsky 1981,
1982).
This is a theory of almost unparallelled elegance within formal linguistic
theorizing. It has basically all the properties we want from a scientific theory: A
range of phenomena (the different empty categories), each with their distinctive
properties, are seen to be special cases of a single phenomenon (the featureless
empty NP), their distinctive properties derived from other independent
properties, primarily the binding relation they enter into. This phenomenon is
then further unified with another set of phenomena, namely different forms of
overt NPs, their distinctive properties being ultimately derived from variation
with regard to the value of two primitive binary features. Together with certain
other axioms of the theory, especially the Theta Criterion, the Case Filter, and
the Empty Category Principle (ECP), this theory could explain a truly
impressive range of phenomena to do with binding, movement, and generally
the distribution of NPs, across a wide range of languages, potentially including
all natural languages.
Impressive though it was, it didn’t take long before the theory began to
crumble. There were empirical shortcomings. For example, as formulated in (2)
the theory could not accommodate the East Asian type of null pronouns, found
in Chinese, Japanese, Korean, etc., which seem not to be licensed by ‘strong
INFL’ or clitics, as the languages in question don’t have any subject-verb
agreement (the overt marker of strong INFL) or pronominal clitics. The theory
could be modified, though, to accommodate these null pronouns as instances of
pro but licensed in some different fashion (see Huang 1989). Another, possibly
worse problem was the discovery of various forms of NP types that didn’t fit
into the restrictive framework of the classical Binding Theory, including long
distance reflexives (see Huang 2000 for an overview), the SE vs. SELF type
anaphors (see Reinhart & Reuland 1993), and dependent pronouns (see Fiengo
& May 1994).
An even greater challenge for the theory of empty categories in (2) was the
emergence of the Minimalist Program (initiated, of course, by the same person
who initiated the GB program, namely Noam Chomsky). The theory in (2) is
radically incompatible with the Minimalist Program. Perhaps most strikingly
this is the case for traces. Consider (4), where (4b) is the rough analysis of the
passive (4a) in terms of the GB program.
(4) a. John was arrested.
b. Johni was arrested [NP e]i
In (4b) John and the trace are two distinct categories. They form a Chain,
formally indicated by the subscript, which means that they share a Theta-role
and a Case, but they are nevertheless two distinct categories representing two
types of NP, an R-expression and an anaphor, respectively.
In Minimalist derivational theory the passive (4a) is derived by first
merging the lexical item [D, John] (assuming that the name John belongs to the
category D) with the verb [V, arrest] forming a VP, as in (5a). The structure is
subsequently expanded by merging more categories one by one with the already
derived construct (indicated by   ). At a certain point [D, John] is
remerged with TP, satisfying the EPP-feature of T. The result is (5b).
(5) a. [arrest John]   
b. [John T [be [arrested John]]] 
c. John was arrested
When the structure is eventually spelled out, the lower copy of [D, John] is not
spelled out (see Chomsky 1993, Nunes 2000). In (5), and corresponding to the
Chain in (4), there are multiple occurrences of the same category, the name
John, only one of which is ultimately pronounced/spelled out. That is to say,
there is no trace in the sense of an empty category with its own distinct
properties and coindexed with an overt category. The derivation of wh-questions
and other structures involving wh-movement (or, more generally, A-bar
movement) follows the same principles: A whP is first merged as an object,
subject, or adverbial, etc., and subsequently remerged with CP, potentially
several times (in the case of long-distance extraction), but generally only one of
the copies is spelled out. That is to say, there is no empty category with the
properties of wh-trace in (2).
The two types of traces gone, all that remains of the theory of empty
categories in (2) is PRO and pro. Let us put PRO aside for the time being, and
consider the position of pro in the Minimalist Program.
3. Little pro and uninterpretable features
The most authoritative theory of pro within the GB program is the one
articulated by Rizzi 1986, building on earlier work by Chomsky 1981, 1982,
Rizzi 1982, Bouchard 1984, among others: pro is inherently unspecified for
feature values, as stated in (2). Its distribution is determined by the following
two conditions:
(6) a. pro must be licensed.
b. pro must be identified.
In (8), representing the (relevant part of) the structure of (7), pro is licensed
because it is governed by I (INFL), which in Spanish is sufficiently strong (or
rich) to be a licenser of pro.
(7) Están cansadas.
(Spanish)
be-3PL tired-F.PL
‘They are tired.’
(8)
IP
proi
I’
[3PL]
Ii
[3PL]
The content of pro is identified by Agr, the person-number-gender features of I;
pro inherits the feature values of I because pro and I are coindexed.
The theory can explain the well known correlation between rich agreement
morphology and null subjects: In languages with weak agreement morphology I
fails to license pro. It may also fail to identify pro. However, the theory allows
for the possibility that a language can identify pro in some persons (for instance
1st and 2nd person plural in French or in construction with 1st singular am in
English), yet never allows null subjects.
The theory of pro outlined above cannot be maintained in a theory making
the distinction between interpretable and uninterpretable features which plays a
crucial role in Chomsky 1995: ch. 4 and subsequent works by Chomsky and
others. Chomsky argues that there are two varieties of features: an
interpretable and an uninterpretable variety. The person, number and gender
features of an NP (or DP), are interpretable, restricting the denotation of the NP.
The person, number, or gender features which appear on a verb, auxiliary or
adjective, as in (9), are uninterpretable as they do not restrict the denotation of
these categories.
(9) Las chicas están
cansadas.
the girls be-F.3PL tired-F.PL
(9) asserts that a group of female individuals excluding the speaker and the
addressee (the denotation of the NP las chicas) each have (some degree of) the
same, indivisible and genderless property of being tired (the denotation of the
predicate estan cansadas). The sentence does not, for example ascribe to the
girls a particular female way of being tired, or, at least not necessarily, repeated
occurrences of being tired. 1 By definition uninterpretable features cannot
survive until LF, so they must be eliminated in the course of the derivation of
LF. However, they may be, and typically are, visible in PF. According to
Chomsky 2000, 2001a, 2001b their role in the grammar is to drive syntactic
operations, particularly movement.
Chomsky 2001a furthermore proposes that the formal difference between the
interpretable features and their uninterpretable counterparts is that the latter
enter the derivation unspecified, being assigned values as part of the process of
derivation by virtue of entering into the relation Agree with an interpretable
counterpart (see Zwart 1997: 189 for a version of the same idea). This theory
gives formal expression to the intuition that agreement is directional. In for
example (9), the auxiliary verb and the adjective agree with the subject NP, not
vice versa. Once the uninterpretable features are assigned values, they are
removed from the syntactic derivation being handed over to phonology, the
derivation of PF.
4. Two hypotheses
Within this theory of agreement it is obviously not possible for an
inherently unspecified pronoun to be specified by the features of I, as those
features are themselves inherently unspecified. The following are two
alternative hypotheses consistent with the feature theory sketched above:
* Thanks to the organizers of GLOW in Asia 4 for inviting me and giving me the
opportunity to present this paper, and thanks to the audience at the Colloquium for their
comments. Thanks also to Annabel Cormack, Theresa Biberauer, Ian Roberts, and Dave
Willis
1 It can denote a plurality of distinct occurrences of being tired, but crucially it need not
do so. There are languages that have an interpretable plural form of the verb; see
Newman 1990, Schmidt 2002. This verb form is “used primarily to indicate plural
action, either on the part of several agents /…/ or applied to several objects on one or
several occasions, or an event that is taking place on several occasions. (Schmidt 2002:
10). In these languages the plural (or ‘pluractional’) form of the predicate can only
denote a plurality of events or states.
Hypothesis A:
There is no pro at all in null-subject constructions. Instead Agr, the set of
features of I, is itself interpretable; Agr is a referential, definite pronoun,
albeit a pronoun phonologically expressed as an affix. This presupposes
that Agr is also assigned a subject theta-role, possibly by virtue of heading
a chain where the foot of the chain is in vP, receiving the relevant thetarole.
A version of this hypothesis is articulated in Alexiadou &
Anagnostopoulou 1998, another one in Manzini & Savoia 2002, yet another one
in Platzack (to appear). If Agr is interpretable, it could, on that account, specify
the features of pro. But if Agr is interpretable, there is no need for pro. The role
of (subject) pro in Chomsky 1982, Rizzi 1986 and related work is to carry the
subject theta-role, possibly nominative Case, and satisfy the EPP. But if Agr is
interpretable, hence referential, then Agr must itself carry the subject theta-role.
This means that there could at most be an expletive pro in specIP. If Agr
absorbs nominative Case as well, as seems most plausible if it is referential and
heads a chain, then it would be a Caseless expletive pro. Expletive pro is a
dubious category, particularly in a minimalist framework, as it has no interface
properties at all, neither at LF not PF. But even granting the theoretical
possibility, the only condition that could conceivably require an expletive pro in
specIP would be the EPP. Alexiadou & Anagnostopoulou 1998 and Manzini &
Savoia 2002 exclude this possibility by stipulating that the EPP is (effectively)
satisfied by Agr in null-subject languages. We may, however, for the sake of
argument, retain as a theoretical possibility that specIP can be filled by a covert
expletive pro in null-subject languages.
A consequence of Hypothesis A is that a preverbal agreeing subject,
lexical or pronominal, in a null-subject language is in an A-bar position. It is a
position neither associated with Case nor with a theta-role, and furthermore, if
Agr bears the subject theta-role, the lexical subject’s relation to the sentences
must be that of an adjunct or apposition.
Hypothesis B:
The null subject is specified for interpretable features, and values the
uninterpretable features of Agr just like any other subject. This implies
that the nullness is a phonological matter: The null subject is a pronoun
which is not pronounced.
The following is a clear empirical difference between hypotheses A and B:
According to A, in finite null-subject constructions the subject position specIP is
either vacant or filled with expletive pro. According to B, the position is
occupied by a pronoun, and is therefore not available for filling by some other
category. Hypothesis A therefore predicts that a language that has null subjects
but also has overt expletives should allow, or even require, specIP in finite null-
subject constructions to be filled with an overt expletive. Hypothesis B predicts
this to be impossible.
Null-subject languages are generally assumed not to have (overt) expletive
pronouns, especially not ‘pure’, there-type expletives (see Chomsky 1995:
288).2 There is at least one null-subject language which has overt pure expletive
pronouns, though, namely Finnish, see Holmberg & Nikanne 2002. As will be
demonstrated below, Hypothesis B makes the right prediction for Finnish.
5. Null subjects and agreement in Finnish
Finnish is a partial null-subject language in that 1st and 2nd person
pronouns are optionally null, in any environment.
(10) a. (Minä) puhun englantia.
I
speak-1SG English
b. (Sinä) puhut englantia.
you speak-2SG English
c. *(Hän) puhuu englantia.
he/she speak-3SG English
(Me) puhumme englantia.
we speak-1PL English
(Te) puhutte englantia.
you-PL speak English
*(He) puhuvat englantia
they speak-3PL English
A 3rd person definite subject pronoun can be null only in certain environments,
essentially when it is bound by a higher subject (see Vainikka & Levy 1999).
(11) Pekka väittää [että (hän) puhuu englantia].
Pekka claims that he speaks English
The null variety is necessarily bound by the matrix subject. Generic pronouns
can, and must, be null; this will be discussed briefly in section 12. Quasireferential pronouns in construction with weather predicates or extraposed
clauses can also be null. In standard (as opposed to colloquial) Finnish the
subject pronoun must be null in weather expressions.
(12) a. Nyt (se) taas sataa.
now it again rains
‘Now it’s raining again.’
There-type, ‘pure’ expletives do not trigger agreement and appear not to be assigned
Case. Their only function is to satisfy the EPP. For it-type expletives one may argue that
they are excluded from specIP in a finite null-subject construction since they are not
assigned Case, if nominative Case is absorbed by (interpretable) Agr. Hypothesis A
therefore makes a clear prediction only for pure expletives.
2
b. (Se) oli hauskaa että tulit käymään.
it was nice that came-2PL visiting
‘It was nice that you came to visit.’
As shown by (10), Finnish has rich agreement morphology. 3rd person
agreement is a bit less rich than 1st and 2nd person agreement, in that 3SG is null
in the past tense and in the conditional mood where tense is neutralized; see
Holmberg & Nikanne (1993), Holmberg & al. (1993). In many varieties of
colloquial Finnish there is no distinction between 3SG and 3PL. In these
varieties, 3PL is also null in the past tense and the conditional mood. In the
present tense indicative 3SG is, however, phonologically visible in the form of
vowel lengthening, and 3PL has a suffix –vat (or –vät, subject to vowel
harmony) in all tenses and moods, except in those colloquial varieties where it is
= 3SG. I will focus on 1st and 2nd person in this paper, as these are the forms
which uncontroversially allow a null subject.
Most varieties of Finnish do not use definite null pronouns much. The use
is largely restricted to formal varieties, including standard written Finnish. It is
nonetheless clear that the null subjects are part and parcel of Finnish ‘core
grammar’, since Finnish speakers have largely uniform intuitions about nullsubject constructions. See Holmberg & Nikanne 2002 for discussion.
6. The Finnish expletive sitä
Finnish has an expletive pronoun which is obligatory in certain contexts. In
general, Finnish does not tolerate verb-initial declarative sentences, hence (13a)
is ungrammatical. You either have to move a referential category (an argument
or a referential adverbial) to specIP, or merge the expletive sitä.
(13) a. *Sattui minulle onnettomuus.
happened to-me accident
b. Minulle sattui onnettomuus.
to-me happened accident
c. Sitä sattui minulle onnettomuus.
EXPL happened to-me accident
’I had an accident.’
The same pattern is illustrated in (14) and (15).
(14) a. *Meni nyt hullusti.
went now wrong
b. Nyt meni hullusti.
now went wrong
c. Sitä meni nyt hullusti.
EXPL went now wrong
’Now things went wrong.’
(15) a. *Viihtyy
saunassa.
feels-good in-sauna
b. Saunassa viihtyy.
in-sauna feels-good
c. Sitä viihtyy saunassa.
EXPL feels-good in-sauna
’One feels good in the sauna.’
The expletive sitä is the partitive form of the pronoun se ‘it’. As shown in
Holmberg & Nikanne 2002, it is essentially a there-type, pure expletive, merged
in the position where the subject is found in unmarked sentences. As also shown
by Holmberg & Nikanne, this is not a subject position per se, as it can be filled
by other referential categories. If anything, it could be characterized as a topic
position. I will continue to refer to it as specIP. It is not a specCP position, as
shown, for instance, by the fact that it can invert with the finite verb in
questions (where the finite auxiliary or verb moves to C, and is affixed with a
question particle):
(16) Menikö sitä taas hullusti?
went-Q EXPL again awry
’Did things go wrong again?’
There is not necessarily any interpretive difference between the (b) and the (c)sentences in (13, 14, 15). Fronting the argument or temporal or locative adjunct
is the solution to a formal condition, the Finnish version of the EPP, as is merge
of the expletive. The fronted argument or adjunct can have more specific
information-structural implications, but need not have any. In fact, in written
Finnish the use of the expletive is proscribed, leaving fronting as the only
acceptable means to satisfy the EPP.
A couple of caveats are in order: First, a verb-initial sentence is acceptable
with ‘verb-focus’, or rather polarity focus, as in (17), where the verb tends to
have focus stress and possibly a focus particle.
(17) SATTUI(-pas) minulle onnettomuus.
happened (FOC) to-me accident
‘I did have an accident.’
In this case the finite auxiliary or verb, or more precisely, the functional head
which incorporates the finite auxiliary or verb and encodes polarity is moved to
C: see Holmberg 2001b.
Second, verb-initial impersonal sentences are allowed if the sentence
contains no category that can move to specIP to satisfy the EPP. A category can
move to specIP if it is referential, in the sense that DPs and certain adverbials
(locative, temporal, instrumental, but not for instance manner or reason) are
referential. According to Holmberg & Nikanne (2002) they must be ‘potential
topics’. For instance (18) is therefore acceptable. Compare (14a) and (18):
(18) Meni hullusti.
went awry
‘Things went wrong’
The manner adverb hullusti is not a potential topic, while the time adverb nyt
‘now’ in (14) is, hence the difference: The time adverbial must move to specIP
in the absence of an expletive. In the absence of a referential category, the
expletive is optional; compare (18) and (19).
(19) Sitä meni hullusti.
EXPL went awry
’Things went wrong.’
See Holmberg & Nikanne 2002 for discussion of verb-initial clauses. The
following is thus a viable formulation of the EPP in Finnish:
(20) EPP in Finnish: If the sentence contains a referential category, SpecIP
must be overtly filled either by a referential category or by an expletive
pronoun.
It is important to note that the expletive is not restricted to impersonal and
generic sentences, but also occurs in construction with, for instance, 1 st and 2nd
person finite verbs and ditto subjects, if the latter are not moved to specIP, as in
(21) (INE = Inessive case):
(21) a. Sitä olen
minä-kin käynyt Pariisissa.
EXPL have-1SG I-too
visited Paris-INE
‘I have been to Paris, too (actually).’
b. Minä sitä olen
käynyt Pariisissa.
I
EXPL have-1PL visited Paris-INE
‘I’ve been to Paris (would you believe it).’/’I’m the one who has been
to Paris.’
In (21a) the subject pronoun is in a sentence-medial focus position with the
expletive filling specIP. In (21b) the subject pronoun has moved to the sentenceinitial focus position (a specCP position; see Vilkuna 1995, Holmberg &
Nikanne 2002), in which case, again, the expletive can fill specIP.
7. Testing hypotheses A and B
The stage is now set for testing the two hypotheses presented in section 2.
According to Hypothesis A, Finnish 1st and 2nd person Agr is made up of
interpretable features, and so is essentially an affixed definite pronoun. An overt
1st or 2nd person pronoun is therefore not required, and if included, it is
(presumably) in a higher, A-bar-type position. The prediction is, then, that
specIP, the position immediately preceding the finite verb or auxiliary in a
declarative sentence, or immediately following it in a yes/no question, could, or
even should be filled with an expletive pronoun. The prediction is false, as first
observed in Hakulinen 1975; see also Holmberg & Nikanne 2002.
(22) a. *Sitä puhun englantia.
EXPL speak-1SG English
b. Oletteko
(*sitä) käyneet Pariisissa?.
have-2PL-Q EXPL visited Paris
’Have you been to Paris?’
This is predicted by Hypothesis B, according to which a null pronoun occupies
specIP. (22a,b), with the expletive, are ill formed for the same reason that
(23a,b) are: The subject pronoun occupies specIP satisfying the EPP, thus
leaving neither space nor function for the expletive to fill.
(23) a. *Sitä minä puhun englantia.
EXPL I speak-1SG English
b. Oletteko te (*sitä) käyneet Pariisissa?
have-2PL you EXPL visited Paris
(23a,b) compared with (21a,b) show that a subject pronoun can’t be combined
with an expletive in pre-verbal position in either order (pronoun-expletive or
expletive pronoun) except if the pronoun is focused, in which case it may
occupy a position other than specIP, leaving it up to the expletive to satisfy thye
EPP.
8. A counterproposal
There is another interpretation of the facts, compatible with Hypothesis A:
(a) The EPP in Finnish is a requirement that every finite sentence should have a
Topic, and (b) Agr, particularly 1st and 2nd person Agr, being a referential,
definite pronoun, is a Topic satisfying the EPP, and therefore precluding merge
of an expletive. In the following I will show that this hypothesis is untenable.
As discussed by Vilkuna (1989, 1995), there are strictly two positions in
the left periphery preceding the finite verb or auxiliary, in Finnish: A contrastive
focus position which is also the landing site for a fronted wh-phrase, and a
Topic position, which in the unmarked case is occupied by the subject. For
example, (24a) (adapted from Vilkuna 1995) can only be interpreted with
Annalle ’to Anna’ as Contrastive Focus and the subject Mikko as Topic, (24b)
can only be interpreted with the fronted object kukkia as Contrastive Focus and
Annalle as Topic (the subject in that case being Information Focus or ’Main
News’ in Vilkuna’s 1995 terms), while (24c) is ill formed, having one argument
too many preceding the finite verb; see Vilkuna 1995 (ALL = Allative case).
(24) a. Annalle Mikko antoi kukkia.
Anna-ALL Mikko gave flowers
’It was to Anna that Mikko gave flowers.’
b. Kukkia Annalle antoi Mikko
flowers Anna-ALL gave Mikko
’Flowers, Anna received from Mikko.’
c. *Kukkia Annalle Mikko antoi.
flowers Anna-ALL Mikko gave
It does not matter whether the arguments are pronouns or lexical NPs; the
information-structural interpretations remain the same. That is to say, a sentence
cannot be introduced by two Topics. 3 Now if 1st and 2nd person Agr is itself a
Matters are complicated by the fact that an object or adverbial may ‘scramble’ to
preverbal position when (a) the sentence is introduced by a focused category, and (b) the
object or adverbial is not Main News. If the subject is then Topic, the result will be a
sentence with two Topics preceding the finite verb, in the loose sense of Topic employed
here. Thus for instance (i) is well formed:
(i)
Kukkia Mikko Annalle antoi.
flowers Mikko Anna-ALL gave
3
Topic, the prediction is that an overt pronoun preceding Agr should have
Contrastive Focus interpretation.
(25) Minä olen
käynyt Pariisissa.
I
have-1SG visited Paris-INE
The prediction is false. The initial pronoun can, but certainly need not have
Contrastive Focus interpretation. For example, it is a perfectly licit answer to a
question ”Where have you been?”, even in the formal register where null
subjects are most common. Consequently it can be preceded by a contrastively
focused category:
(26) Pariisissa minä olen käynyt.
Paris-INE I
have visited
‘It’s Paris I’ve been to.’
What if 1st and 2nd person Agr is a Topic only optionally, being optionally
unmarked for information structural features? That would allow (25) and (26),
but then (22a,b) are again wrongly predicted to be grammatical, that is when the
unmarked option is taken.
I conclude that UG allows for the possibility of null-subject constructions
with a null subject pronoun. Following the Chomskyan approach to agreement,
the null pronoun has interpretable features, and assigns values to the
inherently unspecified features of Agr. In other words, the null subject pronoun
identifies Agr (= the finite verb or auxiliary agrees with the null pronoun), not
vice versa.
9. Is the null pronoun derived by absence of spell-out or deletion?
Consider the following three alternative versions of Hypothesis B:
(i) null pronouns are derived as ordinary pronouns except that they are not
spelled out;
(ii) null pronouns are derived exactly as ordinary pronouns, and spelled out, but
are deleted in PF;
(iii) null pronouns is a subcategory of pronouns which a language may or may
not include among its lexical items, in the same way as a language may or
may not have weak pronouns or clitic pronouns.
See Vilkuna 1995, Holmberg 2001a,b. This does not, however, impact on the argument
in the text: The generalization still holds that the sentence cannot be introduced by two
Topics.
Can we decide which of the three alternatives is right? As for (i) I will make the
following conjecture:
There is a form of null pronoun which plausibly is null because it is not
spelled out, namely PRO. Given the theory of syntactic features in Zwart 1997
and Chomsky 2001a, a category can’t be spelled out before its
uninterpretable/unspecified features have been valued. PRO is a nominal
category in a non-Case position, typically the subject position of a non-finite
clause. Its Case feature therefore does not get assigned a value, and
consequently the pronoun does not get spelled out. This is not a problem in
itself, as long as the pronoun can either be interpreted as a bound anaphor
(controlled PRO) or, as a default option, be assigned a generic interpretation
(arbitrary PRO). In this view, an unspecified Case feature does not in and of
itself cause a derivation to crash, as long as the interpretation of the NP in
question at LF can be ensured. This is essentially a version of the classical
analysis of PRO in Chomsky 1981, 1982. The reason why PRO is illicit in for
example (27), whether it is controlled by the subject (’John hit himself’) or has
an arbitrary interpretation (’John hit anyone’) is, then, that it cannot avoid being
assigned Case in this position by the transitive verb, and thus be spelled out (as
him, himself, or anyone, etc., depending on which features it is composed of) –
on the assumption that Case must be assigned where it can be.
(27) *John hit PRO.
More precisely, the operation Agree (see Chomsky 2000, 2001a, 2001b), by
which an NP is assigned a Case value by a functional head (by finite I in the
case of the subject, by v, the abstract transitivizer head, in the case of the
object), and the functional head is assigned -feature values by the NP in return,
must apply where it can apply.
What about pro in specIP? Arguably specIP in a finite clause is not a Caseposition in Finnish, or indeed any language. Although nominative Case
assignment is concomitant with movement to specIP in many languages, the
movement is not crucial to Case-assignment; see Chomsky 2000, Holmberg
2001a. However, whether the pronoun moves from its first-merged position in
spec,vP or not, it will unavoidably get assigned nominative Case by finite I
through the operation Agree. We are now led to conclude that the subject
pronoun in finite clauses is assigned Case and thereby gets spelled out, just as
the object pronoun in (27) does (ruling out an empty pronoun in object
position). In the same process, the features of I get assigned values by the
pronoun.
If so, the nullness of the null definite subject pronoun in finite sentences
must be the effect of deletion of phonological features in PF. That is to say, we
are back to a version of the original pro-drop analysis in Perlmutter 1971: null
subjects are derived by deletion of a pronoun. This hypothesis will be elaborated
and qualified in the next section.
10. Discourse pro-drop and argument pro-drop
A number of languages have null subjects and null objects even though
they have no agreement at all. Chinese, Japanese, Korean, and Thai are among
the languages which exhibit so called discourse pro-drop, that is pro-drop which
appears to be licensed purely by discourse-antecedents. (28a,b) are from
Mandarin Chinese, (28c) from Korean. 4
(28) a. Nŭháimen líkāi-le,
yīnwéi [e] lèi-le.
girls
leave-ASP because
tire-ASP
‘The girls left, because (they) were tired.’
b. [e] méi chī zǎofàn.
no eat breakfast
‘(I/you) have not had breakfast.’
c. John-i
Mary-ka
[e] miwha-n-ta-k
malha-ss-ta.
John-NOM Mary-NOM
hate-ASP-DECL-COMP say-PAST-DECL
‘John said that Mary hated (him).’
By the assumptions made in the previous section, the subject and object
pronouns in these examples are assigned Case in Narrow Syntax, and thereby
get spelled out, but they can then be deleted in Phonology subject to
recoverability under ordinary rules and conventions governing ellipsis, broadly
speaking. In (28a, c) the ellipsis is recoverable from the subject of the matrix
clause, in (28b) it is recoverable from the discourse.
The question is, why can’t all languages do that? Languages like English,
German, French, etc. have plenty of other forms of ellipsis, so why not ellipsis
of pronouns, in positions corresponding to the positions of the (putative) deleted
pronouns in (28a,b,c)? It seems that non-pro-drop languages have something
which prevents ellipsis of pronouns, regardless of identification or
recoverability.
Some non-pro drop languages arguably have an EPP condition which
requires specIP to contain an overt category in finite clauses, a subject or an
expletive, or even other overt material, as in Icelandic according to Holmberg
(2000). Insofar as this condition truly requires presence of a phonologically
overt category in specIP, it will effectively prevent deletion of a subject
4
Thanks to Nianling Yang for the Chinese examples. The Korean example is from Yang
(1985).
pronoun. It is, however, not obvious that the EPP always requires the presence
of an overt category. For example, English clauses with an extracted wh-subject
have no overt material in specIP.
A more interesting hypothesis is the following, based on Tomioka 2003:
First, a characteristic of the East Asian languages which allow discourse prodrop is that a noun phrase may lack a determiner and number marking but still
may function as an argument, denoting an individual or a group. In other words,
a bare noun or NP can denote an individual or a group. Tomioka (2003) gives
the following example.
(29) Ken-wa ronbun-o yon-da.
(Japanese)
Ken-TOP paper-ACC read-PAST
‘Ken read a paper/papers/the paper/the papers.’
The noun ronbun is multiply ambiguous, its interpretation being determined
pragmatically.
In other languages a bare noun or NPs can only be a predicate, and needs a
determiner and/or number marking to function as an argument (except, in some
languages, for mass nouns); see also Chierchia (1998).
Second, as shown by Tomioka (2003), elided pronouns in Japanese have
the same range of interpretations as overt N/NP. A natural conclusion is, then,
that elided pronouns in Japanese are bare nouns, not Ds or DPs (as also argued
by Hoji 1998).5 Now, there is one nominal ellipsis which is found even in
English and perhaps generally in non-pro-drop languages, namely the one
exemplified in (30), often referred to as N’-ellipsis.
(30) a. John ate four bananas, but I ate only three [e].
b. Om du tar en röd slips, så tar jag en blå [e].
if you take a red tie, then take I a blue
‘If you take a red tie, I’ll take a blue one.’
(Swedish)
Tomioka’s claim is that Japanese pro drop (and likewise in other languages of
the Japanese type) is always ‘N-bar-ellipsis’ (that is ellipsis of N or NP, as
opposed to DP). The wide range of contexts where pro-drop is found in these
languages, and the range of interpretations assigned to the elided structures, is a
consequence of the property that bare nouns can be arguments in these
languages.
This implies that nouns (N/NP) can be deleted across languages, while
deletion of D/DP is more restricted. I will now make the following conjecture:
5
Assuming standard X-bar theory, they could be bare NPs. This does not make sense,
however, under the Bare Phrase Structure hypothesis (Chomsky 1994 and subsequent
works) widely adopted in the Minimalist Program.
(31) D/DP cannot be deleted, in any language, except as part of a larger deleted
constituent.
I don’t profess to understand why there would be such a principle, but as will be
shown, once it is assumed, a number of facts fall into place.6 If arguments are
necessarily Ds or DPs in English, Swedish, French, and other non-pro-drop
languages, then it follows that they can never be deleted. What about null
subjects in ‘agreement pro-drop languages’ like Spanish, Finnish, etc.?
Particularly in Spanish (along with other Romance languages) bare N-arguments
are quite impossible (see Chierchia 1998), yet Spanish has null subjects. 7
Assume that what is characteristic of agreement pro-drop languages is that
they have a D in I. Not only do they have, say, an uninterpretable D-feature in I
matching a D-feature on the subject, but they actually have an interpretable
determiner as a sentential head, either forming a feature complex with T, or
acting as a separate head. Assume, furthermore, that this means that the subject
itself can be a bare noun, which can be construed as being in the scope of the D
in I, perhaps crucially if it moves to specIP.
Since overt lexical subjects clearly cannot be bare nouns in Spanish,
Italian, etc. we have to assume some sort of doubling of D in this case, or even,
as suggested in unpublished work by Dominique Sportiche, that the D-in-I ends
up on top of the subject N/NP as a result of a movement operation. However,
there is one case where this need not be assumed, namely null subjects. Instead,
the analysis of null pronouns from discourse pro-drop languages can be
extended to agreement pro-drop languages: The null pronoun is a bare N. In
agreement pro-drop languages the bare N must be construed as being in the
scope of D. In discourse pro-drop languages this is not necessary.
The claim is, now, that Spanish and Finnish have a null pronoun in null
subject constructions, but this null pronoun is a bare N, while overt pronouns are
Ds. The facts from Finnish discussed in section 4 show that this null N occupies
SpecIP.
6
An alternative, slightly more general, formulation of (31) is (i):
(i) Arguments cannot be deleted, except as part of a deleted predicate.
7 In Finnish nominal arguments do not need a determiner. For instance, the object in (1)
can be interpreted as either definite or indefinite.
(i)
Juha luki artikkelin.
Juha read article-ACC
However, unlike Chinese, Japanese and Korean, Finnish will always have number
marked on argument nominals. The object in (i) cannot be interpreted as anything but
singular. That is to say, nominal arguments are never as bare as they can be in the ‘true
discourse pro-drop languages’. Note also that Finnish doesn’t even allow pro-drop in all
persons. I therefore count Finnish as belonging squarely to the argument pro drop
languages.
It is widely assumed in current literature that pronouns are (bare) Ds.
However, one of the most detailed studies of pronouns from recent years,
namely Cardinaletti & Starke (1999) argue that null pronouns are not Ds but Ns.
More precisely, they argue that the subcategory of pronouns called weak
pronouns are Ns (or a projection of N, but crucially lacking D), and that null
pronouns in pro-drop languages are a subtype of weak pronouns.
Is the null subject the result of deletion of a weak N-pronoun, or do some
languages have null N-pronouns in their lexical inventory, and use them in
appropriate contexts? Compare versions (ii) and (iii) of Hypothesis B, discussed
at the beginning of this section. Cardinaletti and Starke (1999) appear to take the
latter view, as does Cardinaletti (in press). I will continue to assume that the null
pronouns are the result of deletion of spelled-out weak pronouns. This seems a
more parsimonious approach, given that there is ellipsis of various other
categories, anyway, including ‘N-bar deletion’. That is to say, I tentatively
maintain version (ii) of Hypothesis B.
11. On the role of rich agreement
We conclude that what we used to call little pro is a deleted bare N, a weak
pronoun in Cardinaletti and Starke’s (1999) terms, in agreement pro-drop
languages as well as in discourse pro-drop languages. At least in agreement prodrop languages this bare N is specified for person and number, and values the features of I, which are eventually spelled out as agreement on the finite verb or
auxiliary. Also in these languages the N occupies specIP in Narrow Syntax; this
was shown in section 4 to be the case in Finnish. I assume it is the case
generaly, at least in agreement pro-drop languages.
How does this theory capture the correlation between rich agreement and
null subjects? In GB theory it was explained by the licensing and identification
conditions on little pro. In the present theory there is no licensing of little pro.
There is still, in a sense, ’identification of little pro’, but not as part of Narrow
Syntax. The content of the deleted pronoun obviously has to be
recovered/identified, but this is now part of sentence processing: The hearer
needs to undo the deletion of the subject pronoun, and identify the referent of
the null subject. In this process the status of the features as interpretable or
uninterpretable does not matter. The form of the inflection on the auxiliary,
verb, or adjective encodes feature values, and these feature values entail the
existence of a pronoun with the corresponding interpretable feature values.
If a language has rich agreement, this will be made use of in recovering the
content of a deleted pronoun (as detailed in Cole 2000). On the other hand, we
know that rich agreement is not a precondition for null subjects, as recovery
may rely, in part or entirely, on contextual cues (as in Chinese, etc). So the
question remains how to explain the correlation between rich agreement and
pro-drop, assuming that the correlation is generally true.
It is tempting to assume that D-in-I, the property characterizing agreement
pro-drop languages, is crucially connected with rich subject-verb agreement.
What we cannot do, however, is to somehow equate D and rich agreement, say,
taking rich agreement to be the overt realization of D, that is if we accept the
arguments given in this paper that subject-verb agreement is not itself a
referential category, but the realization of an inherently unspecified set of features which receive their feature values from the subject by Agree. 8 The
question is if the correlation is as crucial as often assumed. There are at least
two languages that have rich agreement, yet do not allow pro drop, namely
Icelandic and German. In our terms, these languages do not have D-in-I in spite
of having rich agreement morphology. Can a language have D-in-I, yet have
deficient subject-verb agreement or even no subject-verb agreement, relying
perhaps entirely on discourse for the interpretation of null subjects? I leave this
question for future research.
12. Some residual issues
We can now address a residual problem concerning Finnish pro-drop: It
only applies to 1st and 2nd person pronouns, except in certain well-defined cases.
Note that this cannot be explained as an effect of deficient agreement: As shown
in (10), 3rd person agreement is morphologically visible. I can now suggest a
formal account of this: The 3rd person pronouns are Ds, hence are not deletable.
A piece of evidence in support of this conjecture is the fact that the 3 rd person
pronouns se ‘he/ she/ it’ and ne ‘they’ double as demonstratives, and in
colloquial Finnish, as definite articles.
(32) a. Se
vanha talo on nyt myytävänä.
that/the old house is now for-sale
b. Se on nyt myytävänä.
it is now for-sale
As noted, generic pronouns are null in Finnish.
(33) Nykyään väsyy helposti.
nowadays tire easily
’One gets tired easily these days.’
8
Besides, when the subject is an overt DP, the agreement morphology is realized on the
finite verb while the subject has its own determiner, so at least in that case the agreement
is not literally the morphological realization of D.
My impression is that generic pronouns are commonly null, cross-linguistically.
A much discussed form of generic null pronoun is arbitrary PRO. Another case
discussed in detail in Rizzi (1986) is the case of generic null objects in Italian.
(34) a. It’s nice [PRO to be a millionaire].
b. Questa decisione non rende [e] felici
this
decision not make happy
‘This decision doesn’t make people happy.’
(Italian)
This fits potentially very well in the typology of nominal argument types
sketched above, based on Tomioka (2003). Generic arguments do not denote
individuals but kinds, losely speaking. As such they can be bare (determinerless)
nouns or NPs. In some languages, including the Germanic ones, the prototypical
generic argument is a bare plural (Tigers are dangerous). As bare nouns or NPs
they can be deleted, or perhaps more generally (if we do not want to analyze
(33) or (34) as the result of deletion), they can be null.
13. Conclusions
Consider again Hypotheses A and B:
Hypothesis A:
There is no pro in null subject constructions. Instead Agr , the
-features of I, is itself an interpretable category, a referential, definite pronoun
phonologically expressed as an affix.
Hypothesis B: There is a null subject in specIP in null subject constructions.
The null subject is a specified for interpretable -features, and values the
uninterpretable features of Agr just like any other subject does.
I conclude that Hypotheses A and B, formulated in section 3 are both right.
Although the idea was rejected that Agr is interpretable, Hypothesis A is partly
right in the sense that I contains an interpretable nominal feature in agreement
pro-drop languages, namely interpretable D. Hypothesis B is right in that there
is a subject pronoun in specIP in null subject constructions, namely a bare Npronoun.
These are the substantial claims made in this paper. Other claims have a
more tentative character. I have opted for a PF-deletion analysis of null subjects
in finite clauses. This may prove to be untenable. In that case, the fall-back
position is either that the null subject is a pronoun (crucially a bare N, equipped
with -features) which is not spelled out, or that the null subject is a special type
of pronoun included in the dictionary of some languages but not others. The
analysis of null subjects as spelled-out but deleted pronouns was connected with
the assumption that PRO is a nominal argument which does not get case, and
therefore cannot be spelled out. This traditional view of PRO has been
challenged on many fronts in recent years. For example, Sigurdsson (1991)
argues that Icelandic PRO has case. Recently the same claim is made for Basque
by San Martin (2003). According to another recent theory control is actually
movement, which means that PRO is a ’trace’, or rather a copy of a moved DP,
which is to say that it is not an empty category at all (following the discussion in
section 1, above); see Hornstein (1998, 2001). If we give up the claim that PRO
is a caseless pronoun, then we may reconsider analyzing null subjects in finite
clauses, and other forms of null pronouns, as (bare N-) pronouns which are not
spelled out.
References
Alexiadou, Artemis & Elena Anagnostopoulou. 1998. Parametrizing AGR:
Word order, V-movement, and EPP-checking. Natural language and
linguistic theory 16. 491-540.
Bouchard, Denis. 1984. On the content of empty categories. Dordrecht: Foris.
Cardinaletti, Anna (in press) Towards a cartography of subject positions. In L.
Rizzi (ed.) The Structure of CP and IP. The Cartography of Syntactic
Structures, vol. 2, Oxford University Press, Oxford - New York.
Chierchia, Giennaro. 1998. Reference to kinds across languages. Natural
Language Semantics 6: 339-405.
Cardinaletti, Anna & Michal Starke. 1999. The typology of structural deficiency.
Clitics and other functional categories in European languages, ed. Riemsdijk,
H. van, Berlin: Mouton de Gruyter.
Chomsky, Noam. 1981 Lectures on government and binding. Dordrecht: Foris.
Chomsky, Noam. 1982. Some concepts and consequences of the theory of
government and binding. Cambridge MA: MIT Press.
Chomsky, Nom. 1995. The minimalist program. Cambridge MA: MIT Press.
Chomsky, Noam. 2000. Minimalist inquiries: the framework. In R. Martin, D.
Michaels, and J. Uriagereka (eds.), Step by Step. Essays on Minimalist Syntax
in Honor of Howard Lasnik. Cambridge, MA: MIT Press. 83-155..
Chomsky, Noam. 2001a. Derivation by phase. In M. Kenstowicz (ed.) Ken
Hale: A Life in Language, Cambrdige MA: MIT Press. 1-59.
Chomsky, Noam. 2001b. Beyond explanatory adequacy. Ms. Department of
Linguistics and Philosophy, MIT.
Cole, Melvyn. 2000. The syntax, morphology, and semantics of null subjects.
Doctoral dissertation, University of Manchester.
Fiengo, Robert & Robert May. 1994. Indices and identity. Cambridge MA:
MIT Press.
Hakulinen, Auli. 1975. Suomen sitä: Pragmatiikan heijastuma syntaksissa.
Sananjalka 17: 25-41. [Finnish sitä: A reflection of pragmatics in syntax.]
Hoji, Hajime. 1998. Null object and sloppy identity in Japanese. Linguistic
Inquiry 29. 127-152.
Holmberg, Anders .2000. Scandinavian stylistic fronting: How any category can
become an expletive. Linguistic Inquiry 31: 445-484.
Holmberg, Anders. 2001. The syntax of yes and no in Finnish. Studia
Linguistica 55: 141-175.
Holmberg, Anders. 2001. Expletives and agreement in Scandinavian passives.
Journal of Comparative Germanic Linguistics 4: 85-128.
Holmberg, Anders & Urpo Nikanne. 1993. Introduction. In A. Holmberg & U.
Nikanne (ed.) Case and other functional categories in Finnish syntax.
Berlin: Mouton de Gruyter. 1-20.
Holmberg, Anders, Urpo Nikanne, Irmeli Oraviita, Hannu Reime & Trond
Trosterud. 1993. The structure of INFL and the finite clause in Finnish. In
A. Holmberg & U. Nikanne (ed.) Case and other functional categories in
Finnish syntax. Berlin: Mouton de Gruyter. 176-206.
Holmberg, Anders & Urpo Nikanne. 2002. Expletives, subjects, and topics in
Finnish. In P. Svenonius (ed.) Subjects, expletives, and the EPP. New
York, Oxford: Oxford University Press. 71-106.
Hornstein, Norbert. 1998. Movement and Control. Linguistic Inquiry 30: 69-96.
Hornstein, Norbert. 2001. Move! A minimalist theory of construal. Malden and
Oxford: Blackwell.
Huang, James. 1989. Pro-drop in Chinese: A generalized control theory. In O.
Jaeggli and K. safir (eds.) The null subject parameter. Dordrecht: Kluwer.
Huang, Yan. 2000. Anaphora. A cross-linguistic study. Oxford, New York:
Oxford University Press
Jaeggli, Osvaldo. 1982. Topics in Romance syntax. Dordrecht: Foris.
Manzini, Rita & Leonardo Savoia. 2002. Parameters of subject inflection in
Italian dialects. In P. Svenonius (ed.) Subjects, expletives, and the EPP.
Oxford: Oxford University Press. 157-200.
Newman, Paul. 1990. Nominal and verbal plurality in Chadic. Dordrecht: Foris.
Nunes, Jairo. 1999. Linearization of chains and phonological realization of
chain links. In S.D. Epstein & N. Hornstein (eds.) Working minimalism.
Cambridge MA: MIT Press.
Perlmutter, David. 1971. Deep structure and surface structure constraints in
syntax. New York: Holt Rinehart & Winston.
Platzack Christer. (to appear) Agreement and null subjects. To appear in Papers
from the 19th Scandinavian Conference of Linguistics.
Reinhart, Tanya & Eric Reuland. 1993. Reflexivity. Linguistic Inquiry 24: 657720.
Rizzi, Luigi. 1982. Issues in Italian syntax. Dordrecht: Foris.
Rizzi, Luigi. 1986. Null objects in Italian and the theory of pro. Linguistic
Inquiry 17: 501-557.
Samek-Lodovici, .V. 1996. Constraints on Subjects, An Optimality Theoretic
Analysis, Unpublished Ph.D dissertation, Rutgers University.
San Martin, Itziar. 2003. PRO receives regular structural case in Basque. Talk at
GLOW in Asia 2003.
Itziar San Martin (U of Basque Country)
Schmidt, Bodil K. 2002. Introduction. In B.K. Schmidt, D. Odden, & A.
Holmberg (eds.) Some aspects of the grammar of Zina Kotoko. Berlin:
Lincom Europa. 1-14.
Sells, Peter. 1987. Aspects of logophoricity. Linguistic Inquiry 18: 445-480.
Tomioka. Satoshi. 2003. The semantics of Japanese null pronouns and its crosslinguistic implications. In K. Schwabe & S. Winkler (eds.) The interfaces.
Deriving and interpreting omitted structures. Amsterdam: John
Benjamins.
Vainikka, Anne & Y. Levy. 1999. Empty subjects in Finnish and Hebrew.
Natural Language and Linguistic Theory 17: 613-671.
Vilkuna, Maria. 1989. Free word order in Finnish. Suomalaisen Kirjallisuuden
Seura. Helsinki
Vilkuna, Maria. 1995. Discourse configurationality in Finnish. In K.É. Kiss
(ed.) Discourse- configurational languages. Oxford: Oxford University
Press. 244-268.
Zwart, C. Jan-Wouter. 1997. A minimalist approach to the syntax of Dutch.
Dordrecht: Kluwer.