Pronominal Clitics - Oxford Scholarship Jump to ContentJump to Main Navigation
Aspects of the Theory of Clitics$

Stephen Anderson

Print publication date: 2005

Print ISBN-13: 9780199279906

Published to Oxford Scholarship Online: September 2007

DOI: 10.1093/acprof:oso/9780199279906.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (www.oxfordscholarship.com). (c) Copyright Oxford University Press, 2015. All Rights Reserved. Under the terms of the licence agreement, an individual user may print out a PDF of a single chapter of a monograph in OSO for personal use (for details see http://www.oxfordscholarship.com/page/privacy-policy). Subscriber: null; date: 28 August 2016

Pronominal Clitics

Pronominal Clitics

Chapter:
(p.227) 8 Pronominal Clitics
Source:
Aspects of the Theory of Clitics
Author(s):

Stephen R. Anderson (Contributor Webpage)

Publisher:
Oxford University Press
DOI:10.1093/acprof:oso/9780199279906.003.0008

Abstract and Keywords

This chapter surveys some of the substantive syntactic properties of the most widely studied class of clitics, those traditionally analysed as pronominals. It begins by examining the nature of (predicate-argument) agreement, comparing it with well-known phenomena arising in the analysis of special clitics. This requires an elaboration of the analysis of the Morphosyntactic Representations of categories to account for phenomena such as (the presence versus absence versus optionality of) clitic doubling, clitic climbing, and the like. While the bulk of the literature devoted to pronominal clitics focuses on object clitics, some languages (including several spoken in northern Italy and in nearby areas of Switzerland) also have special clitics referring to subjects. Their properties are explored, including those of Surmiran and a range of northern Italian dialects. The significance of the morphological approach to special clitics for the syntax of functional categories and the proposal that all such categories constitute syntactically autonomous heads (each with its own projection) in syntactic representation are discussed.

Keywords:   special clitics, pronominal clitics, agreement, Morphosyntactic Representations

The canonical examples of special clitics in the minds of many linguists are pronominals such as those found in association with the main verb in Romance languages1 or those found in second position in languages like Warlpiri or Tagalog. Pronominal clitics have a number of interesting properties in their own right. Some of this interest is more or less independent of their special clitic status, and other matters are important to explore in connection with the very notion that these elements can be treated linguistically in the same way as other, non-pronominal special clitics. The purpose of this chapter is to deal with (or at least acknowledge) some of these issues. Given their complexity, and the richness of the existing literature dealing with them, this survey will inevitably have something of the character of a superficial whirlwind tour. Nonetheless, it is important to show that the present framework provides a basis for discussing and analyzing these classic problems.

I have been assuming we can call pronominal clitics the functional morphology of phrases, but what does that mean as far as the underlying grammatical structure is concerned? Pronominal clitics are also frequently assimilated in various ways to agreement markers—what connections are there between the two? A key role in understanding these matters is played by Clitic Doubling, a construction whose properties help to clarify the relation between pronominal clitics and the argument positions they are related to. A further topic which has been prominent is the construction known as “Clitic Climbing.” While a relatively traditional analysis of this construction fits well with the general framework adopted here, its extension to cases of Long Distance Agreement is perhaps less familiar. Finally, while attention to pronominal clitics has generally focused on elements corresponding to objects of various types, some (p.228) languages also display clitics linked to subjects, which pose particular issues of their own.

8.1 Pronouns and Agreement

What is the structural relation between a pronominal clitic and an argument position which it is interpreted as specifying or filling? In “classical” generative grammar, this was considered a straightforward instance of movement. Such clitics were generated as pronouns of a certain sort, occupying an argument position just as any other nominal expression might. By virtue of some identifiable property (e.g., “[+Clitic]”) they then underwent movement to adjoin to the Verb—or wherever else they might appear in a particular language, such as second position.

The movement analysis explained one important property of clitics: in many cases, they are mutually exclusive with overt nominal expressions in a given argument position, as illustrated in sentences (8.1) from (standard) French.2

(8.1)

a.

Les enfants

l'=

ont

déjà

mangé

The kids

it

have

already

eaten

The kids have already eaten it

b.

*Les enfants

l'=

ont

déjà

mangé

le gateau

The kids

it

have

already

eaten

the cake

(The kids have already eaten the cake)

Unfortunately, this explanation goes a bit too far. In some languages, at least some pronominal clitics are not mutually exclusive with overt nominals, as in the Spanish sentence in (8.2), where the presence of the clitic is consistent with the simultaneous presence of the full phrase al professor.

(8.2)

Le=

entregué

el libro

(al professor)

to-him

I gave

the book

(to the professor)

I gave the book to him/the professor

In Spanish, dative (and sometimes accusative) clitics can correspond to an overt argument expression. This is called clitic doubling, and it obviously raises problems for the movement analysis. It is hard to see how the argument (p.229) position could remain in place, filled by a full nominal or prepositional phrase, while also undergoing movement as a clitic.3

On the theory of the present book, of course, pronominal special clitics are not elements that have been moved to where they appear in the surface form from some argument position. Rather, they are the overt reflection of properties of that position, construed as part of the functional content of the clause and realized by a principle of phrasal morphology as a modification of the phonological form of the clause. In fact, the phenomena surrounding pronominal special clitics are rather similar to what we find in verbal agreement (as various authors have noted). I will thus develop a view of these clitics by starting from corresponding facts in that domain.4

Kinds of Agreement

In many (perhaps most) languages, agreement as marked on the verb registers certain properties of an argument, and does not supplant the overt expression of the argument. This is true in French, for example, where agreement and a full nominal subject are both present in all finite clauses. In other cases, however, agreement morphology and overt argument expressions are in complementary distribution. A particularly clear example is provided by the Venezuelan Carib language Pemon, as illustrated by the examples in (8.3).5

(8.3)

a.

kamicha

ke

Antonio-da

mure

ponte-'po

clothes

with

Antonio-ERG

child

dress-PAST

Antonio dressed up the child with clothes

b.

kamicha

ke

mure

ponte-'po-i-ya

clothes

with

child

dress-PAST-3-ERG

He dressed up the child with clothes

c.

kamicha

ke

i-ponte-'po

Antonio-da

clothes

with

3-dress-PAST

Antonio-ERG

Antonio dressed him up with clothes

d.

kamicha

ke

i-ponte-'po-i-ya

clothes

with

3-dress-PAST-3-ERG

He dressed him up with clothes

(p.230) An extreme variant of this situation is proposed by Baker (1995) for Mohawk (and other “Polysynthetic” languages). Baker argues that the Mohawk verb always includes agreement (and/or an incorporated Noun), but that overt nominal argument expressions appear in the position of adjuncts, rather than that of arguments. The presence of agreement is claimed to preclude the presence of a nominal in the argument position, as in Pemon. The difference is that the presence of agreement itself (ignoring for now the case of incorporated nouns, to which I will return in Chapter 9) is obligatory in a language of this type, and thus we do not see the kind of alternation that occurs in (8.3).

A third possibility in agreement systems is what we find in languages with (perhaps generalized) “pro-drop.” Here also agreement with one or more arguments may always be present. While such agreement is compatible with the simultaneous presence of an overt argument, though, that argument may also be absent from the surface form, in which case the argument it refers to is interpreted in the same way a pronoun would be. An example is provided by Georgian, as in (8.4). The verb momc˙era ‘he wrote it to me’ agrees simultaneously with all three of its arguments (although there are only two overt affixal markers of agreement present in the form). Any one of these may be phonologically unrealized in the sentence, and interpreted as pronominal.

(8.4)

(vano-m)

(me)

(c˙erili)

mo-m-c˙er-a

(Vano-ERG)

(me)

(letter.NOM)

PFX-1OBJ-wrote-3SG.AOR

Vano/he wrote a letter/it to me

We need to provide, then, for several distinct possibilities. Sometimes agreement marking simply reflects the properties of the argument (as for instance in French subject–verb agreement). Sometimes it alternates with a full argument (as an option, as in Pemon, or perhaps obligatorily, as in Baker's analysis of Mohawk). Finally, its presence sometimes sanctions (while not requiring) a null argument.

I propose to regard the case of mere “registration”6 of an argument as involving an operation that copies some relevant morphosyntactic features (typically, person, number, and possibly gender; see Corbett 2003 for a survey) from the argument to the Morphosyntactic Representation7 of the verb, where rules of the morphology may spell them out. Alternative views on which these features are generated independently in the Morphosyntactic Representation or its equivalent, and then checked for the required identity at some point (p.231) in the syntax, can be regarded for present purposes as notational variants of this picture.

The second type of agreement is supplied by Baker's (1995) analysis of Mohawk, which I adopt here at least for the purposes of discussion. In this case, we have obligatory agreement coupled with the impossibility of full DPs in argument position. Baker argues at length that the argument positions are in fact present, and occupied by phonetically null (but referential) pro elements whose features are reflected in verbal agreement.

To describe this situation, we need to establish a link between the particular syntactic configuration posited for Mohawk and something in the morphology of the language. For Baker, that link is provided by the Morphological Visibility Condition (“MVC”). This is a parametric choice in the grammars of individual languages: in some languages, such as Mohawk, the condition holds, while in others, like English, it does not. Where the MVC obtains, it requires that all arguments of a head (such as the verb) must be reflected in the morphology of that head. Baker assumes that the markers which appear in this function will then prevent full nominals from receiving Case in argument position. As a result, they must appear (if at all) in some external, adjoined position. This yields the particular clause structure which Baker argues for in Mohawk.

A key role in this explanation is played by the assumption that agreement elements in Mohawk absorb the structural case which the verb would otherwise assign to its arguments. This must obviously be interpreted as a fact about Mohawk, at least in part, since agreement is not universally incompatible with the appearance of full nominals in argument positions. Something extra thus needs to be said in the grammar of such a language if the desired consequences are to follow.

Mohawk verbs do, of course, subcategorize argument positions, but on Baker's analysis these can only be filled by phonetically null pro with some appropriate set of features. In this language, these are (roughly) [± ME], [± YOU],[±SG], [±PL], [±MASC], [±FEM], [±ZOIC], where dual number can be represented as [-SG, -PL] and neuter gender as [—MASC,—FEM,—ZOIC]. Because of the presumed Case absorption effect, overt nominals appear in clause-external adjunct position, where they form a chain with the presumed pro.

An alternative to Baker's assumption that Mohawk agreement absorbs Case is simply to say that Mohawk verbs do not assign structural Case at all—or rather, that the only principle of structural-case assignment in the language is the one licensing nominals in adjoined position. Notice that since agreement is obligatory, any hypothetical structural Case related to the verb would never, in fact, be assigned on Baker's assumptions.

(p.232) The exclusion of full DPs from argument positions need not, however, be related to the assignment of abstract Case at all. I will suggest below that this can be derived instead from Binding relations, which would disallow anything but pro in positions with which the verb agrees referentially (as is the case in Mohawk). On that assumption, the assignment of Case is quite orthogonal to the issue of what fills argument positions.

The presence of agreement establishes reference, so let us assume (with Baker) that argument positions are also indexed referentially in the verb by agreement. This relation involves a coindexing relation in addition to the copying (or identity) of features characteristic of agreement in general. Baker believes that what is at work here is the presence in syntactic representations of functional categories (instances of AGR) which bear indices. These are subsequently incorporated (via adjunction) into a single word with the head verb, thus satisfying the MVC. You might notice that the agreement elements thus look like little clitic pronouns.

There are (at least) two ways in which Baker's assumptions differ from those of the present book. First, like much recent work within the Minimalist Program (but for different reasons), I consider it important to avoid positing independent functional categories like AGR as syntactic heads. In that case, however, there is no obvious constituent to bear the referential burden of Baker's AGR elements. Second, in line with the Lexicalist Hypothesis as articulated in Anderson (1992), the present framework does not allow for the creation of words within the syntax.

What is necessary is to coindex something in the representation of the verb with the argument positions to which that verb assigns θ-roles. In the present terms, that “something” clearly has to be the verb's Morphosyntactic Representation.

I thus want to say that agreement rules come in two basic flavors, corresponding to the difference between registration and agreement mentioned above. Note that Baker needs to say that AGR absorbs Case in some languages and not in others: the difference to be developed is intended to have the same consequences.

The primary function of an agreement rule is to copy some set of features from an argument to the head, as for instance in English or French subject agreement. As a result, the Morphosyntactic Representation of the verb (or other predicate) will contain those features, and they can then trigger the introduction of overt morphological markers in the derivation of the appropriate word form. This kind of agreement—registration—has no further syntactic consequences.

(p.233) As a second possibility, an agreement rule can also establish a relation of coindexation between the argument position in question and the Morphosyntactic Representation itself, as is arguably the case in Mohawk. Such a relation is obviously very similar to the one Baker assumes, except that the coindexation is not with a “morpheme” internal to the Verb, but rather with a subset of the featural content of the verb's Morphosyntactic Representation. In Mohawk, then, we could say that agreement is of this second, coindexing sort. Where some pro i is coindexed with a layer of agreement features, the two must agree in those features.

Now suppose we take seriously the fact that, as a result of agreement, the verb (through its Morphosyntactic Representation) is coindexed with those of its argument positions with which it agrees. Since the verb is the head of the VP, the “Head Feature Convention” (which I borrow from work in GPSG, Generalized Phrase Structure Grammar: see Gazdar et al. 1985) says that features in its Morphosyntactic Representation are transmitted to that of the VP. In case the constituent type IP is distinct in a given language from VP, when the verb moves to I (the head of IP) its features are also transmitted to the Morphosyntactic Representation of the IP node.

Note that in many languages, Agreement establishes a relation between the verb and more than one of its arguments. In such a case, the agreement material corresponding to various arguments must in general be kept distinct. The “Layering Convention” of Anderson (1992) covers just this issue for word-level morphology, and extends directly to the case of Morphosyntactic Representations for phrasal categories.

Just as the content of agreement may be transmitted from the verb to superordinate phrasal categories which it heads, it may itself inherit other properties from such a phrasal category. Suppose we say that Tense, for example, is a property of IP—part of the Morphosyntactic Representation of that phrasal node. The Head Feature Convention, by asserting identity between the properties of a phrase and of its head, has the effect of transmitting Tense properties to a verb when it occurs in I, the head of IP, and the morphology of the verb may thus reflect this phrasal property.

In the case of interest to us here, features and/or referential indices assigned to the verb by Agreement are inherited by the Morphosyntactic Representation of the clause which it heads, where they constitute (perhaps a portion of) the “functional content” of the clause. On this view, “functional categories” such as Tense and Agreement are characteristic of a clause, while not projecting additional layers of syntactic structure. When realized as inflectional morphology on the finite verb, this content characterizes that word as well by virtue of the identity established by the Head Feature Convention.

(p.234) Now let us return to the case of genuine agreement: the sort which involves coindexing as well as feature identity. Here a referential index will be present by the mechanisms just described at the level of the clause, coindexed with the agreeing argument position. Now suppose an overt nominal expression were to appear in such an argument position, one with which the verb (and thus the clause node) is coindexed via Agreement. This ought to produce a violation of the Binding Conditions, since the nominal would thus be bound (coindexed with a C-commanding category) within its clause.

This would exclude both R-expressions and lexical pronominals from appearing in positions with which the verb stands in a relation of agreement (and not merely registration). It is plausible, however, to assume that phonologically null pronominal elements (pro) are not similarly excluded. In this way, we could derive the exclusion of overt nominal expressions in argument positions from the referential nature of agreement in a language (e.g., Mohawk), without further Case-theoretic stipulation.

On this view, phonetically null pronouns are associated with referential (i.e., coindexing) agreement. Assume also that most languages8 do not have null pronouns in their lexicon: the only way a motivated argument position can be phonetically unfilled is when lexicalization is blocked by virtue of referential agreement. Thus pro is not really a pronoun, but rather just a kind of empty category which (like other empty categories) must be sanctioned by principles of grammar. In this case, the relevant principle is grounded in the binding relation between referential agreement material and the position pro occupies.

Based on these considerations, it is possible to propose a typology of the relations between agreement and the arguments it indexes. English or French represent one extreme, where agreement is always non-coindexing registration. As a result, argument positions in such languages are never required (or allowed) to be empty in the absence of other specific displacements or deletions.

In Georgian, or in Italian and the other classic pro-drop languages, the agreed-with position can optionally remain empty. This can be regarded as an optionality in the operation of Agreement: this always copies the features of the argument, and it may optionally also coindex. Where coindexation appears, we get pro; where it does not, we have a normal pronoun or full lexical nominal expression.

In contrast, accepting Baker's extensively argued proposal for clause structure in Mohawk, Agreement in that language is always coindexing, and thus (p.235) argument positions are always empty. Mohawk deals with this situation by licensing the formation of chains which relate an empty (but referential) argument position to the content of a nominal in adjunct position.

In Pemon, a given argument position can be identified either by an overt nominal or by verbal agreement, but not both. We can describe this by saying that Pemon agreement is of the referential kind, but Agreement itself is optional. When it occurs, it triggers the associated morphology and precludes an overt nominal; when it does not, there is no morphology, phonologically null pro is not licensed, and the relevant argument position must be filled with an overt expression.

This difference between coindexing “agreement” and non-coindexing “registration” is a way of reconstructing at least one sense of “strong” versus “weak” agreement. It corresponds to the traditional notion that agreement in some languages “identifies” the corresponding position(s). It also comes as close as we are likely to get to formalizing the notion of “strong” agreement, in treating this as a parameter of grammatical variation rather than deriving it from some aspect of the formal nature of agreement itself, such as the precise combination of features represented, or the extent to which various forms are distinct from one another—two tempting but ultimately unproductive approaches that have been taken in the literature to grounding the distinction. I suggest that the present proposal provides a more nearly satisfactory reconstruction.

A Complex Example: Finnish Agreement

A more intricate agreement system than those considered above is found in Finnish, and it is worth exploring how this is to be characterized in the present framework.

Finnish has a set of markers whose status has been the subject of much discussion, appearing on head nouns to indicate properties of a possessor. On the basis of their failure to trigger the well-known process of consonant gradation in a previous syllable, as opposed to clear inflectional affixes, some authors (e.g., Nevis 1986) have described these elements as (special) clitics. Despite their phonology, Kanerva (1987) shows clearly that they must be treated as word-level affixes, and not as clitics. They thus constitute markers of agreement within the nominal phrase. They also show interesting similarities to the markers of verbal agreement in the language, and both the nominal and the verbal markers should fall within a theory of agreement relations. The phonology of these affixes in the broader context of Finnish noun inflection is analyzed by Kiparsky (2003), and will not concern us here. The account of their syntax below follows in large part that of Toivonen (2000), an analysis in (p.236) the terms of Lexical Functional Grammar and translated here into the framework of the present book.

The facts concerning the nominal agreement system are as follows. When a nominal has a first- or second-person possessor, the head of the phrase bears obligatory possessor marking, and an accompanying genitive pronoun may or may not appear for emphasis. This is illustrated in the sentences of (8.5), where (following the descriptive tradition) I gloss the possessive markers as “Px” with the appropriate person.

(8.5)

a.

Poika

myi

(minun)

marsu-ni

boy

sell.3SG.PST

1SG.GEN

guinea pig-px1SG

The boy sold my guinea pig

b.

(Sinun)

kissa-si

on

sairas

2SG.GEN

cat-PX2SG

is

sick

Your(sg) cat is sick

c.

Pekka

näkee

(meidän)

ystävä-mme

Pekka

see.3SG.PRES

1PL.GEN

friend-PX1PL

Pekka sees our friend

d.

(Teidän)

kissa-nne

on

ruma

2PL.GEN

cat-px2pl

is

ugly

Your(pl) cat is ugly

This is straightforward: when the possessor is first or second person, Agreement copies its features of person and number onto the head noun, and may optionally coindex that noun's Morphosyntactic Representation with the possessor. If the coindexation option is taken, the presence of that referential indication prevents an overt possessor phrase, and the position of the possessor is filled by phonologically null pro. If there is no coindexation, pro is not licensed in that position and a full pronoun must appear. This is the normal state of affairs described above for “pro-drop” languages.

In the third person, the situation is more complicated. Here we must distinguish three separate cases as far as the applicability of Agreement is concerned. First, when a human pronoun (hänen ‘his/her’ or heidän ‘their’) appears as possessor, as in (8.6), the head noun must bear the possessive suffix.

(8.6)

Pojat

näkevät

heidän

ystävä-nsa

boy.PL

see.3PL.PRES

3PL.GEN

friend-PX3

The boysi see theirj,* i; friend

Agreement here is obligatory and consists only in copying the person feature(s) onto the head noun. Since this does not involve coindexation, an empty (p.237) pro possessor is not licensed and the pronoun must appear. From independent considerations of binding, this must always be interpreted as disjoint in reference from the subject of the sentence.

When an overt possessor expression appears that is either (a) non-pronominal, or (b) non-human,9 no possessor suffix is possible, as illustrated in (8.7).

(8.7)

a.

Minä

pesen

Pekan

autoa(*-nsa)

I

wash.1SG.PRES

Pekka.GEN

car(*-PX3)

I am washing Pekka's car

b.

Minä

annan

kissa-lle

sen

ruokaa(*-nsa)

I

give.1SG.PRES

cat-ALLATIVE

it.GEN

food(*-PX3)

I give the cat its food

With overt possessor phrases other than human genitive pronouns, Agreement does not apply at all. No possessive affix appears on the head noun, and the possessor cannot be replaced by null pro.

The third case, that of sentences like (8.8), is particularly interesting. When a third-person possessive marker appears on the head noun, and no overt possessor is present, the nominal is interpreted as possessed by the subject of the clause. Note that there is no requirement that the possessor in this case be human.

(8.8)

a.

Pekka

näkee

ystävä-nsä

Pekka

see.3SG.PRES

friend-PX3

Pekkai sees hisi friend

b.

Se

heiluttaa

häntää-nsä

It

wiggle.3SG.PRES

tail-PX3

It wiggles its tail

The interpretation of (8.8) shows that a possessor is present, which must be a phonologically null anaphoric element controlled by the subject of the sentence. We might plausibly take this to be PRO, the element that also appears as the subject of certain non-finite complement clauses. When the possessor in a nominal expression is PRO, then, agreement copies its person and number features onto the head noun along with its referential index. This referential index, in turn, is controlled by the subject of the clause in which the nominal appears.

In the third-person case, then, the difference between (non-coindexing) registration, (coindexing) agreement, and no agreement at all is not an option, (p.238) as in the first and second person. Rather, each of these possibilities corresponds to a specific set of circumstances.

We can now compare this agreement system, as it operates in nominals, with the way Agreement operates in finite clauses headed by a verb. Here Agreement is always obligatory, and the morphology of the verb reflects the person and the number of its subject. When the subject is first or second person, it can be phonologically null as shown in (8.9).

(8.9)

a.

(Minä)

ammuin

karhun

1SG

shoot.1SG.PST

bear

I shot a/the bear

b.

(Sinä)

näit

karhut

2SG

see.2SG.PST

bear.PL

You saw the bears

Agreement here is exactly parallel to what we saw above in the nominal case. With a first-or second-person (possessor or) subject, Agreement copies the person and number features onto the head, and may or may not also introduce coindexing. When coindexing occurs, the possessor or subject expression is replaced by phonologically null pro.

In the third person, the verbal and nominal cases diverge, but not entirely. First, verbal Agreement is obligatory regardless of the nature of the subject as human or non-human, pronominal or non-pronominal, as illustrated in (8.10).

(8.10)

a.

Poika/Hän

myi

kissa-nsa

boy/he

sell.3SG.PST

cat-PX3

The boyi/Hei sold hisi cat

b.

Kissa-ni

kuoli

cat-PX1SG

die.3SG.PST

My cat died

c.

Se

oli

sairas

it

was.3SG

ill

It was ill

In all of these cases, the subject must be present overtly, even if it is pronominal and recoverable from the context. This suggests that third-person agreement, as opposed to first or second person, consists only in copying the person and number of the subject to the verb, and cannot involve (even optionally) coindexation.

(p.239) There is one circumstance in which a third-person subject can be dropped, however. Sentences with third-person singular agreement but no overt subject are interpreted as having an arbitrary or generic subject, as in example (8.11).

(8.11)

Voi

mennä

ulos

can.3SG

go.INF

out

(One) can go out

This sort of generic interpretation is often found in other languages with non-finite clauses where there is no controller for the subject position, such as English To know her is to love her. We saw this in the discussion of ins in Surmiran in section 7.5 of Chapter 7, where I suggested that we should regard such sentences as having the element PRO as their subject, with an “ARBITRARY” reading assigned when this appears as the subject of a finite clause. Similarly, Anderson (1982) argues that Breton finite clauses with a distinctive “impersonal” verbal inflection have PRO ARB as their subject.

I propose that the same is true in Finnish, and that sentences like (8.11) also have PRO ARB as their subjects. But this, of course, brings the verbal agreement system into a closer relation with the nominal pattern. Recall that in that case, coindexing agreement was limited to the case of controlled anaphoric PRO as possessor. It seems then that the major difference between the verbal and the nominal systems is that in the verbal case, Agreement is always obligatory at least to the extent of copying person and number features. Coindexation is also possible if the argument triggering Agreement is first or second person; and obligatory if the trigger is PRO (interpreted as ARBITRARY in the absence of a controller, a possibility that is only relevant to the verbal subcase). In the nominal case, no agreement takes place with a non-human or nonpronominal overt third-person possessor, but otherwise the two systems are entirely similar.

There are further phenomena to be discussed in connection with Finnish agreement, such as the facts of non-finite clauses of various types, the exact nature of the control relation relevant to PRO possessors, among others. Nevertheless the present typology of agreement relations seems to provide an appropriate framework for making the distinctions necessary to an adequate analysis.

8.2 Clitics, Agreement, and Doubling

Now let us return to the analysis of pronominal clitics. I propose to regard clitic pronominals as in fact a form of agreement, differing from verbal agreement only in whether the functional content is realized as the morphology (p.240) of a phrase or a word. This is not a novel proposal: Miller and Monachesi (2003) note that it arises fairly naturally within theories such as Lexical Functional Grammar and Generalized (or Head-Driven) Phrase Structure Grammar based on feature systems, and similar suggestions have been entertained within other frameworks, such as the proposals of Franks and King (2000).

The overt manifestation of agreement material by pronominal special clitics can appear in various places, as we have already seen. Clitics may appear with reference to the beginning of the clause—in second position, often, as in Warlpiri or Tagalog. The relevant agreement representation, a set of features associated with the S (or IP) node, triggers the introduction of phonological material (the clitic(s)) within this domain. The linear position in which the clitics are found is determined by Alignment constraints in the ways developed in chapter 6.

Alternatively, as in Romance, we may find the clitics associated with the finite verb. The functional content of the clause is shared by its Morphosyntactic Representation and that of the Verb node which is its head by virtue of the Head Feature Convention, assuming the relevant features have the status of “Head Features” in the language at issue. In Romance languages, those parts of this feature structure relating to non-subject positions are realized within the restricted syntactic domain of the Verb constituent, while the subject material is realized by inflection on the verb itself. In other languages, such as Georgian, the features inherited by the verb are all realized as inflectional properties of that word.

If pronominal special clitics are closely comparable to verbal agreement, they ought, like agreement, to come in both “weak” and “strong” flavors, corresponding to the distinction drawn above between registration and agreement per se. And indeed, it is not hard to see how to pursue this analogy.

The “weak” form is what we find in clitic-doubling constructions, where the presence of the clitic merely registers the properties of the corresponding nominal. Consider the Bulgarian examples in (8.12).

(8.12)

a.

Decata

=ja

običat

neja

children.DEF

her.ACC

love.3PL

her.ACC

The children love her

b.

Na

Svetozar

=mu

xrumna

edna

misâl

to

Svetozar

him.DAT

dawned.3SG

one

thought

A thought occurred to Svetozar

c.

’te

=ti

kaža

az

na

tebe

koj

e=

predatel

will

you.DAT

tell.1SG

to

you

who

is

traitor

I'll tell you who's a traitor

d.

Uvažavat

=go

edin

učitel

zaradi

vseotdajnostta

=mu

respect.3PL

him.ACC

one

teacher

for

devotion.DEF

his

A teacher is respected for his devotion

e.

Kogo

kazvaš

=sa

=go

uvolnili?

who.ACC

say.2SG

AUX.3PL

him.ACC

fired

Who do you say they fired?

f.

Na

nego

=mu

=e

mâčno

to

him

him.DAT

is

homesick

He is homesick

(p.241) As summarized from a variety of authors by Franks and King (2000), the source from which the sentences in (8.12) are taken, doubling occurs in Bulgarian when an object nominal is topical and specific, generic, or a wh-expression. It is optional in all of these cases, except for impersonal sentences such as (8.12f) where it is obligatory. On the other hand, doubling is not possible in other non-specific contexts, such as sentence (8.13).

(8.13)

Târsjat

(*=go)

nov

učitel

seek.3PL

him.ACC

new

teacher

They are looking for a new teacher

The description of this state of affairs is not trivial, but neither are the facts themselves. We say that Agreement with object nominals (which will of course be realized by the introduction of a clitic rather than directly by verbal morphology) is optional, unless the argument is non-referential, as in the intensional reading of (8.13) or an idiom chunk. When it occurs, it is optionally coindexing, except in the case of impersonal sentences with no nominative subject such as (8.12f), where it is obligatory. As in the case of agreement realized as verbal morphology, where this agreement is referential (or coindexing), it precludes the presence of an overt nominal in the corresponding argument position, and thus the clitic gives the impression of substituting for such an expression. Where the option of coindexation is not taken, simple registration of an overt argument phrase results.

A well-known case of a similar sort is supplied by standard Castilian Spanish, in examples such as those of (8.14).

(8.14)

a.

*(Lo=)

veo

a

él

him.ACC

see.1SG

PREP

him

(I see him)

b.

(* lo=)

veo

a

Juan

him.ACC

see.1SG

PREP

John

I see John

c.

(* la=)

veo

la

mesa

her.ACC

see.1SG

the

table

I see the table

d.

*(Le=)

hablo

a

él

him.DAT

speak.1SG

to

him

I speak to him

e.

(Le=)

hablo

a

Juan

him.DAT

speak.1SG

to

John

I speak to John

(p.242) In Spanish, doubling is possible (but not obligatory) with full indirect object nominals, but not with full direct objects. In contrast, when the object is a pronoun, doubling is obligatory in both cases. As Miller and Monachesi (2003: 87) describe matters, some dialects, particularly Latin-American, double full indirect object expressions whether pronominal or not, and have a strong preference for doubling when the direct object is animate and specific. As in Bulgarian, the facts are in some ways rather intricate, but the descriptive parameters are clear. We say that where clitics can occur, Agreement (at least in the sense of feature copying) can apply. Where doubling is impossible, we say that this agreement must also be coindexing; and where doubling occurs, we say that coindexation does not.

The kind of complications which we see in these cases are not at all peculiar to clitic constructions. Bresnan (2001: 146-7) notes that verbal agreement and clitics may both be sensitive to the same factors, such as pronouns versus full nominals, the character of a nominal as definite, specific, animate, etc. We saw above that the nature of possessor agreement in Finnish depends on whether the possessor expression is a human pronoun or not. In general, it is clear that agreement systems (in the broad sense of this notion) can involve the same parameters, regardless of whether the results are realized by verbal morphology or by pronominal special clitics.

In a language like (standard) French, the situation is somewhat simpler. The agreement represented (optionally) by an object clitic is of the “strong” (coindexing) sort, and no overt nominal of any sort can appear to double the clitic. In this language, subject agreement is obligatory, weak, and word-based, while object agreement is optional, strong, and phrasally realized.

A clitic analog to the agreement pattern of Pemon is furnished by Surmiran, the form of Rumantsch discussed above in section 7.5. In this language, argument positions can be occupied freely by full nominals or by pronouns. Alternatively, non-subject arguments can be referred to by clitics associated with the verb, but there is no doubling: that is, any given argument is (p.243) represented either by a nominal expression (including pronouns) or by a clitic, but not both. These points are illustrated by the examples in (8.15).

(8.15)

a.

Ursus

ò

purto

las bulias

a nous

Ursus

has.3SG

bring.PP

the mushrooms

to us

Ursus brought the mushrooms to us

b.

Ursus

ans=

ò

purto

las bulias

(*a nous)

Ursus

us

has.3SG

bring.PP

the mushrooms

to us

Ursus brought us the mushrooms

c.

Ursus

las=

ò

purto

(*las bulias)

a nous

Ursus

them.FEM

has.3SG

bring.PP

the mushrooms

to us

Ursus brought them to us

d.

Ursus

las=

ans=

ò

purto

(*las bulias)

Ursus

them.FEM

us

has.3SG

bring.PP

the mushrooms

(*a nous)

to us

Ursus brought us them

Non-subject agreement in Surmiran is thus like French, with one difference. In French, non-subject (“disjunct”) pronouns are not generally available except in special contrastive or other strongly stressed contexts, while in Surmiran, pronouns are freely usable in non-subject positions. Another difference comes from the fact that French freely allows combinations of multiple clitics, while combinations such as that found in (8.15d) are tightly restricted in Surmiran. In particular, the only acceptable combinations consist of a third-person direct object clitic followed by a non-third-person indirect object. Substituting clitics for both nominals in (8.16) thus leads to an ill-formed sentence.

(8.16)

a.

Tgi

dat

igl matg

a Gelgia?

who

gives.3SGPRES

the bouquet

to Gelgia

Who is giving the bouquet to Gelgia?

b.

?*Tgi

igl=

la

dat

Who

it

her

gives

c.

?*Tgi

l'=

igl

dat

Who

her

it

gives

Pronominal special clitics are generally used to index arguments of a predicate, but in some languages, clitics appear that do not correspond to any argument. An example is provided by verbs in French (and many other languages) that require the presence of a reflexive clitic without projecting any corresponding argument, as in (8.17). These “lexical reflexives” are sometimes known as “pronominal verbs” in the literature. (p.244)

  1. (8.17) a. Marie s'est évanouie ‘Marie fainted’

  2. b. Je me suis comporté comme un fou ‘I behaved like a madman’

Again, there is a parallel with word-level agreement. As discussed in Anderson (1991), verbs in some languages may show agreement with more arguments than they have. For instance, in most Algonquian languages, the conjugation of a verb indicates clearly whether it is treated as transitive (i.e., taking an object) or intransitive; and whether the Theme argument (intransitive subject or transitive object) is grammatically Animate or Inanimate. In some instances, however, verbs are conjugated in a way that indicates the presence of an argument that they do not project. The Menomini examples in (8.18) all involve syntactically intransitive verbs whose inflection includes “agreement” with a non-existent Inanimate object.

  1. (8.18) a. noqnonam ‘he swims’ (Transitive Inanimate inflection)

  2. b. nema▾mi▾qtεhko▾skanan ‘I go bare-legged’ (Transitive Inanimate inflection; plural object)

  3. c. mo▾hkanam ‘he uncovers it’ or ‘it (a heavenly body) rises’ (Transitive Inanimate inflection)

Conversely, other Algonquian languages such as Maliseet (Sherwood 1986) may fail to show agreement with a core argument that they do in fact project. For instance, the verb in sentence (8.19) is conjugated as Animate Intransitive, despite the fact that it takes a direct object.

(8.19)

can

pitkәme

cikәni

John

pack.AI.INDIC.3

apple.ANIM.OBV.PL

John packed the apples

The complex inflectional patterns of Georgian include a number of deviations between syntactic and morphological argument structure. Some of these are illustrated by the sentences in (8.20), all of which include inflection (underlined) for an extra argument which is not present in the syntactic representation.

  1. (8.20) a. kari uberavs ‘the wind blows’ (inflected for 3SG 10)

  2. b. sonat˙a dauḳari (čem-tvis) ‘you played a sonata (for me)’ (verb inflected for 3SG 10)

  3. c. mjinavs ‘I'm sleeping’ (3SG DO)

As discussed in more detail in Anderson (1991), these facts follow if the agreement representation which characterizes the verb (and thus, in the clitic (p.245) case, is inherited by the including phrase) is not necessarily isomorphic to its subcategorization or argument structure. In such cases, the exact collection of features that appears in the Morphosyntactic Representations triggering morphological Agreement and/or the introduction of special clitics is motivated only in part by the syntactic representation. Such a situation has sometimes been asserted to be impossible by those working in heavily syntactic theories of morphology, such as Distributed Morphology, but in fact it is not uncommon in the languages of the world.

Sometimes a clitic results from something other than the argument structure of the clause or the lexically determined agreement structure of the verb. Jaeggli (1986) discusses “ethical datives” such as me ‘(on) me’ in example (8.21).

(8.21)

Juan

me

le

arruinó

la vida

a

esa chica

Juan

me

her

ruined

the life

to

that girl

Juan ruined that girl's life on me

These datives do not appear to correspond to any argument position present in the syntactic representation, and are apparently limited to first and second person. One possible account would make appeal to a “derivational” rule (along lines hinted at in the discussion of Tagalog in section 6.4) which maps the Morphosyntactic Representation of the clause onto a new one with an extra agreement layer corresponding to the ethical dative while also adding the semantic content corresponding to “on me/us/you” to the meaning of the sentence. The extra Morphosyntactic content then triggers the introduction of a clitic, without alteration in the basic syntactic structure of the clause. Much remains to be done to fill out and substantiate this analysis, but it seems promising.

8.3 Clitic Climbing

Another construction which has figured prominently in the literature on pronominal clitics is known as “Clitic Climbing.” In Italian examples such as (8.22), clitics find themselves associated with a verb other than the one that subcategorizes for the argument they specify.

(8.22)

Mario

lo=

vuole

leggere

Mario

it

wants

to.read

Mario wants to read it

This arises with a limited set of matrix verbs that take non-finite complements. Rizzi (1978, 1982) and Burzio (1986), among others, argue that it is (p.246) associated with a syntactic reanalysis of the embedded structure that is triggered by the matrix verbs in question. While their complements are introduced with the structure of a clause (including some sort of CP layer), the result of this Restructuring (as the reanalysis has come to be called, following Rizzi) is that the matrix and embedded verbs are no longer separated by a clause boundary after it applies. Restructuring is an optional process; when it does not apply, the clitic(s) must appear with the embedded infinitive as in sentence (8.23).

(8.23)

Mario

vuole

legger

=lo

Mario

wants

to.read

it

Mario wants to read it

Let us assume that the result of Restructuring is a structure something like that in (8.24).

(8.24) [VP [V vuole] [VP [V leggere] pro 3SG.M ]]

What we want in this case is for the clitic lo that specifies the phonologically null pro i that is the object of leggere to be introduced within the domain of the finite verb vuole rather than in the domain of the embedded verb. Let us assume that Agreement operates as expected, adding content to the Morphosyntactic Representation of [Vleggere] in agreement with the object argument pro i. This content, by the Head Feature Convention, is inherited by the VP that this verb heads.

Of course, since Italian has no rule introducing pronominal clitics at the VP level (but only within V), the presence of this material at this level of structure has no direct effect. However, we can then say that features characteristic of an embedded VP are also inherited by an including VP. This requires a notion akin to that of Grimshaw's (2000) “Extended Projection” to characterize the scope of the relation involved here, but the intuition seems clear. Such inheritance will be blocked by the intervening CP structure when Restructuring has not taken place. Given that the agreement material thus characterizes the matrix VP, it is naturally inherited (again, via the identity required by the Head Feature Convention) by the verb that projects that phrase—in this instance vuole, where it can trigger the introduction of a pronominal special clitic.

Where more than one clitic is present, either all of them “climb” or none does, as illustrated in (8.25).

(8.25)

a.

Mario

vuole

dar

=glie

=lo

Mario

wants

to.give

him

it

Mario wants to give it to him

b.

Mario

glie=

lo=

vuole

dare

Mario

him

it

wants

to.give

Mario wants to give it to him

c.

*Mario gli= vuole dar=lo

d.

*Mario lo= vuole dar=gli

(p.247) This is true even in languages where it is not possible to realize both clitics in association with a single verb. Recall that in Surmiran, sequences of two third-person clitics are not allowed: one or both of the arguments of a verb like dar ‘give’ must be represented by a full pronoun in argument position, since only one can be represented only by a clitic. Clitic Climbing in Surmiran is obligatory with causatives (formed with lascher ‘let’ and far ‘make, do’). According to the normative grammar (Signorell, Wuethrich-Grisch, and Simeon 1987), it is not possible with “modal” matrix verbs like leir ‘want’, but many speakers accept sentences with a clitic associated either with the matrix or with the embedded infinitive with such verbs, as well as with the past-tense auxiliary, as illustrated in (8.26).

(8.26)

a.

Nous

lagn

la=

tarmetter

dumang

we

want.1PL

it.FEM

to.send

tomorrow

We want to send it tomorrow

b.

(?) Nous la= lagn tarmetter dumang

c.

Te

n'=ast

betg

igl=

cumpro

you.SG

NEG-have.2SG

not

it

bought

You haven't bought it

d.

Te n'=igl= ast betg cumpro

When the embedded infinitive has two associated third-person arguments, it is not possible to have one clitic with each verb, even though it is also excluded to have both together. Some of the possibilities are laid out in (8.27).

(8.27)

a.

Ia

vi

dar

el

ad ella

I

want.1SG

to.give

it.MASC

to her

I want to give it to her

b.

Ia igl= vi dar ad ella

c.

Ia la= vi dar el

d.

Ia vi igl= dar ad ella

e.

Ia vi la= dar el

f.

*Ia vi la= igl= dar

g.

*Ia la= igl= vi dar

h.

*Ia la= vi igl= dar

i.

*Ia igl= vi la= dar

(p.248) In Surmiran, it appears that the Morphosyntactic Representations that would result from applying Agreement to two separate third-person arguments are ill-formed, regardless of where one might try to realize the agreement material as clitics.

On the analysis proposed here, pronominal special clitics are a kind of phrase-level agreement phenomenon. If the functional content that triggers the introduction of such clitics can “climb” in the way we have seen, we would expect that the same might be true of functional content triggering word-level agreement in the form of verbal inflection. And in fact, when we look for such a parallel, we find it. Bobaljik and Wurmbrand (2002) show that the Siberian language Itelmen has a restructuring construction comparable to that of Italian, and that when Restructuring applies, agreement material associated with an embedded verb can show up on the matrix verb. This construction, which Bobaljik and Wurmbrand refer to as “Long Distance Agreement (LDA),” is illustrated in (8.28).

(8.28)

a.

t' -utu-z-in

әlčqu-aɬ-iɬ

1SG.SBJ-unable-PRES-2SG.OBJ

see-FUT-INF

I am unable to see you

b.

na

әntxa-βum=n I n

kma

jeβna-s

he

forget-1SG.OBJ-3

me

meet-INF

He forgot to meet me

While the Itelmen facts appear to provide a rather close analogy to the better-known construction in Italian, it must be noted that other instances of apparent “LDA” do not yield to the same analysis. Polinsky and Potsdam (2001) for example discuss a similar construction in the North-East Caucasian language Tsez, and show in detail that it is not based on Restructuring. This construction has no clear analog in the grammar of clitics, and I will not discuss it here.

Another construction which should probably be mentioned here is the well-known phenomenon of Complementizer Agreement found (at least) in some dialects of Dutch and Flemish and in some German dialects as well (Haegeman 1992; Hoekstra and Smits 1998). The examples in (8.29) from the Limburg dialect of Dutch are illustrative.

(8.29)

a.

veurtot-s

tiech

te

bruk

zuu-s

before-2SG

you.SG

the

bridge

see-2SG

before you see the bridge

b.

(iech waet neet)

boe-t

ger

zuu-s

(I know not)

where-2PL

you.PL

be.2PL

(I don't know) where you (pl) are

(p.249) The analysis that suggests itself here assumes that agreement material is inherited by all of the heads in the verb's “extended projection,” up to the level of a complementizer head of CP. Morphological agreement can then take place not only on the verb itself, but also on the complementizer. The properties of this construction have been the basis of an extensive literature; much more discussion is required to incorporate the results of that literature and resolve its outstanding issues, but the line of attack suggested here does appear to provide a basis for that work.

Obviously, there is much more to be said about Clitic Climbing, and the present remarks do not purport to constitute a full theory of the phenomenon. An extensive survey of a broader class of these “Complex Predicate” constructions in the principal Romance languages is provided by Abeillé and Godard (2003), and I have only touched the surface of the intricate array of properties they discuss. Nonetheless, it seems clear that these facts do not in themselves pose special problems with regard to the overall framework proposed here for the grammar of pronominal special clitics. Indeed, that framework appears to offer new possibilities for unifying the analysis of phenomena that have previously been treated in isolation from one another.

8.4 Subject Clitics

Most of the attention in the literature dealing with pronominal special clitics is devoted to clitics that correspond to (direct and/or indirect) objects, but these are not the only pronominals that a theory must account for. In at least some languages we also find clitics that correspond to subjects, and these present issues of their own. I refer here not simply to the phonologically weak pronouns that may occupy subject position, as in English, but rather to special clitics introduced in association with the verb (or perhaps elsewhere), which may (and in some cases, must) be “doubled” by a full nominal in subject position.

Surmiran is one such language, as we have already seen. In section 7.5, we saw that when the verb is “inverted” (that is, displaced to a position outside the VP from which it C-commands the VP-internal subject position), special clitics may be introduced post-verbally in agreement with the subject. (p.250) These do not necessarily replace, but may rather double the subject nominal. Similar subject clitics are found (under slightly different conditions) in Engadine Rumantsch (Puter and Vallader); their occurrence in Surselvan, the best-known form of Rumantsch, is much more limited and associated with “non-standard” speech. These pronouns are one of the topics of Linder (1987), though the Surmiran data there are quite limited and not drawn from current usage. In some languages of northern Italy (Poletto 1993, 2000) and Franco-Provençal (Roberts 1993a), as well, special clitics associated with subject position have been discussed, and we will turn briefly to those facts after a summary of the Surmiran data and some reasons for calling the elements in question clitics rather than verbal agreement markers.

The full set of Surmiran subject clitics was given in Figure 7.4 in section 7.5 of the preceding chapter. Recall that these include overt markers in all cases except the second-person plural. But why should we call these elements (special) clitics, rather than treating them as aspects of the morphology of the verb? They do meet standard criteria for analysis as clitics, as discussed in Chapter 2: for instance, they are not sensitive to lexical (as opposed to purely phonological) properties of their hosts, and they are attached “outside of” clearly affixal material. But is this sufficient? And if they are clitics, what is their nature? Do they have a referential function, like the object clitics of various Romance languages (including Surmiran), or do they simply register the person, number, and gender of the subject, like the pre-auxiliary subject clitics of Franco-Provençal Valdôtain (Roberts 1993a)?

The post-verbal elements in Figure 7.4 cannot simply be morphology registering the properties of the subject, like agreement markers. One piece of evidence in support of this is the fact that they play an essential role in sanctioning the omission of a pronominal subject. As we have seen, (cf. examples (7.47) and (7.48) in Chapter 7), subjects cannot in general be omitted in Surmiran, despite the presence of relatively “rich” agreement on the verb. This shows that normal verbal agreement is not of the strong or coindexing type. It is only the presence of a post-verbal clitic that allows (though it does not require) the subject to be omitted. The agreement that gives rise to the subject clitics, that is, involves optional coindexing with the subject position; and when this occurs, phonologically null pro occupies that position and excludes an overt nominal. At a minimum, then, the operation of Agreement which gives rise to verbal morphology is distinct from the Agreement that gives rise to the subject clitics.

The referential nature of the clitics is also supported by the fact that they cannot appear in association with non-referential subjects, such as navot (p.251) ‘nothing’, nign ‘none, no one’, etc., as in (8.30), although such sentences do show the expected subject agreement on the finite verb.

(8.30)

a.

Nign

n'am

ò(*'l/*'la)

anvido

no one

NEG=me

has-3SG(=3SG-M/F)

invited

No one invited me

b.

Ossa

n'am

ò(*'l/*'la)

nign

anvido

today

NEG=me

has-3SG(=3SG-M/F)

no one

invited

Today no one invited me

Another difference between the clitic elements in Figure 7.4 and verbal agreement is a phonological one, which shows up in the second-person singular form. The regular second-person singular desinences in all tense forms in Surmiran end in -s (e.g., cantas ‘you (sg) sing’). A very general rule of Surmiran phonology replaces /s/ by [∫] before all consonants except [l]. We would therefore expect the form cantast ‘sing-2SG=2SG’ to be pronounced *[kántә∫t], parallel to scolast [∫kólә∫t] ‘teacher’, but this does not happen: cantast is pronounced [kántәst], with [s] and not [∫]. In this respect, the sequence of 2SG ending followed by =t is comparable to what we find in compound forms like (las) clavs-tgesa [kláfsʨèzә] ‘(the) house-keys’, not *[kláf∫ʨèzә]. Evidently, the pre-consonantal s→∫ process (whether this is expressed as a rule or as some set of constraint(s) that require this replacement) is an aspect of the lexical phonology which is no longer active at the point post-verbal clitics are introduced (or when compounds are formed). These facts suggest that the elements in Figure 7.4 are added post-lexically, as would be expected of clitics (though not of inflectional agreement markers).

I conclude, then, that the elements in Figure 7.4 are clitics and not verbal morphology. They are introduced optionally (and post-lexically) in constructions that call for subject–verb inversion, and when present, sanction the (otherwise ungrammatical) omission of pronominal subjects. Such omission is not obligatory, however, and the clitics can be doubled by an overt nominal or pronominal subject phrase, corresponding to the optionality of the coindexation aspect of this distinct Agreement process.

The great bulk of work on clitics linked specifically to subjects (as opposed to subject clitics forming part of a much more general system, as in Tagalog or Warlpiri) has focused on the languages of Northern Italy, including both those considered “dialects” of Italian and others, such as Friulian and Franco-Provençal. Poletto (2000) provides many useful references and discusses a wide range of data from these systems, which differ in many intricate details from (p.252) one another. It is quite impossible to cover anything like the whole of that ground here, and in any event the differences between the theoretical assumptions of most of this literature and those of the present book would require even more extensive discussion. All I can realistically hope to accomplish is a rapid sketch of the ways in which some of these facts might be expected to fit into the present framework.

Poletto (2000) distinguishes four types of subject clitic. She argues that each of these is associated with a different functional category in clause structure, an account which is at odds with the assumptions of this book and thus not directly available here. Fortunately, that is not a necessary interpretation of the facts, since the required distinctions can be represented in other ways.

The first of the four types, called “invariable clitics” by Poletto, is a marker that is the same regardless of the person and number of the subject. For instance, the Swiss Lombard dialect of Lugano has an initial particle a which occurs with all subjects as in (8.31).

(8.31)

a.

A=

vegni

mi

CL

come.1SG

I

I come

b.

A=

ta=

vegnat

ti

CL

2SG

come.2SG

you.SG

You (sg) come

c.

A=

vegn

luu

CL

come.3

he

He comes

d.

A=

vegnum

CL

come.1PL

We come

e.

A=

vegnuf

CL

come.2PL

You (pl) come

f.

A=

vegn

lur

CL

come.3

they

They come

It is not at all clear why such an element should be called a “subject clitic,” since it does not depend in any way on the properties of the subject. As noted first by Benincá (1983), invariable “subject clitics” are found in sentences that convey new information or in exclamative contexts. Since there is no apparent (p.253) morphosyntactic property of a clause corresponding to this characterization, it appears that they should be treated along the lines of the Tagalog “particle clitics” discussed in section 6.4. I will therefore assume that they are introduced derivationally at the level of CP, contributing their phonological content together with the semantic/pragmatic interpretive content identified by Benincá.

A second type of “subject clitic” is illustrated by the Friulian dialect of San Michele al Tagliamento. As illustrated in (8.32), an initial element in this language appears as i when the subject is first or second person, but as a with third-person subjects.

(8.32)

a.

I=

mangi

CL

eat.1SG

I eat

b.

I=

ti=

mangis

CL

2SG

eat.2SG

You (sg) eat

c.

A=

1=

mangia

CL

3SG.M

eat.3SG

He eats

d.

I=

mangin

CL

eat.1PL

We eat

e.

I=

mangè

CL

eat.2PL

You (pl) eat

f.

A=

mangin

CL

eat.3PL

They eat

Poletto calls these “deictic clitics.” For reasons to be discussed below, I will treat them as special clitics introduced in the domain of the head of CP, and thus as a type of clitic analogous to the phenomenon of Complementizer Agreement mentioned above.

The third and fourth types of subject clitic treated by Poletto involve clitics attached to the verb that reflect specific person and number combinations in the subject. Some of the languages she considers have such clitics specifically for second-person singular subjects, and some for second- and third-person singular; these she refers to as “person clitics.” Other languages have such clitics for all third-person subjects, distinguishing gender and number, and these (p.254) are labeled “number clitics.” For reasons that seem primarily theory-internal, these two types are distinguished and assigned to distinct functional categories in Poletto's (2000) analysis. I will treat them together as “personal subject clitics” and consider their central property to be their orientation with respect to the main verb, as opposed to the head of CP. I assume they are introduced as special clitics within the domain of the head of IP.

The northern Italian “subject clitics” thus fall into two major groups, which we can think of as “CP-clitics” and “IP-clitics.” In treating the “person” and “number” clitics as associated with the IP system, while the “invariable” and “deictic” clitics are associated with the CP system, I follow the basic architecture of Poletto's analysis, though not in detail.

Support for this broad division comes from several sources, as Poletto shows. First, we can note that CP clitics and IP clitics can co-occur. In example (8.31b) we see an invariable clitic combined with a 2SG personal clitic, while examples (8.32b,c) show a deictic clitic combined with personal clitics. On the other hand, we do not find combinations of two distinct CP clitics or IP clitics.

The possibility of combining CP and IP clitics follows from the proposal that agreement material is (potentially) inherited by all of the heads within the extended projection of the agreeing verb, and can potentially be realized in any of several ways: as agreement on the main verb; as a clitic associated with this verb or (in the case of clitic-climbing constructions) as agreement or a clitic on a higher main verb within a complex predicate (or restructuring) structure; or as either agreement or a clitic associated with the head of the dominating CP. Each of these realizations is potentially a rule distinct from the others, and as we have seen, they can co-occur in various combinations. The one kind of combination that apparently is not possible, however, is to have multiple instances of the same realization driven by the same agreement material. Thus we find (as in Surmiran) that in clitic-climbing structures the clitic may be realized either on the lower verb or on the higher one, but never on both simultaneously. The possibility of having “subject clitics” of the CP and IP classes co-occur, then, argues that these are two quite distinct types of realization of agreement material.10

Another property distinguishing the two types of clitics, noted by Poletto, is the fact that CP clitics may obligatorily coalesce with the complementizer (p.255) phonologically, In the Loreo dialect of the Veneto region, an invariable clitic a occurs, which obligatorily coalesces with a vowel-final complementizer as shown in examples (8.33).

(8.33)

a.

Ara

ch'=a

vegno

look

that-CL

come.1SG

Look, I am coming

b.

*Ara che a vegno

c.

No

so

s=a

vegno

not

know.1SG

if-CL

come.1SG

I do not know whether I will come

d.

*No so se a vegno.

On the other hand, the personal subject clitics in this language coalesce only optionally with a complementizer, as shown in example (8.34).

(8.34)

a.

Ara

che

el=

vien

look

that

3SG.M

come.3SG

Look, he is coming

b.

Ara ch'el vien

Since CP clitics like Loreo invariable a are introduced in association with the complementizer, it is reasonable that they could trigger allomorphy in that element. Personal subject clitics like el, on the other hand, are introduced in association with the verb, and their combination with a preceding complementizer element is governed by the phonology. In this case, the reduction of such vowel sequences in the post-lexical phonology is evidently an optional process.

A final difference between subject clitics of the two types is the fact that “invariable” and “deictic” clitics consistently precede “strong” negation markers (in the sense of Zanuttini 1997), while pronominal subject clitics generally follow them. This follows if such negation appears within the IP, typically at its left edge. A CP clitic will be outside this structure, and thus precede the negation, while a clitic introduced within IP in association with the verb will follow it.

Poletto's (2000) study and the literature on which it is based explore a number of additional properties of subject clitics in the languages of northern Italy and adjacent regions, and I cannot address them all. Nevertheless, it seems clear that the basic distinctions to be made in that work are natural ones within the present framework, just as I have argued above for other aspects of the behavior of pronominal special clitics. (p.256)

Notes:

(1) Vast amounts have been written about the Romance pronominal clitics, and it is quite impossible for me to cover the complete research history relevant to these elements, although much of the discussion below is devoted to them. A very useful survey of relevant phenomena and analyses, from which I have drawn at several points, is provided by Miller and Monachesi (2003). Many properties of Romance pronominal clitics noted in that paper are not discussed here because these would take us too far afield from the main point, but I do not think any of these matters pose serious problems for the analysis to be presented.

(2) Sentence (8.1b) is acceptable if the final phrase le gateau ‘the cake’ is preceded by a significant pause, indicating a right-dislocated structure. This is irrelevant to the point here, which concerns the complementarity of clitics and full nominals specifying the same argument within a clause. It should also be pointed out that some forms of spoken French differ from the “standard” language in permitting much freer use of clitics that “double” a full nominal expression as in (8.1b).

(3) This is true even if one adopts a “copy and delete” view of displacement, since the clitic is not literally a copy of the nominal it doubles.

(4) Much of what I propose in this section is quite close to the range of possibilities and theories entertained for agreement within Lexical Functional Grammar. See Bresnan (2001: ch. 8) for an extended discussion in that framework.

(5) The data below are from José Alvarez, unpublished field notes. These facts were reported by Alvarez on the Linguist List, vol. 6, no. 574, in 1995.

(6) The useful terminological distinction between “registration” and “agreement” proper was introduced in early work by Perlmutter and Postal on a theory of Relational Grammar. Unfortunately, it does not seem to have survived into current discourse, but it is well worth reviving.

(7) See Anderson (1992) for this notion.

(8) Exception may need to be made for Chinese, Japanese and similar systems where the syntax of phonologically null DPs is somewhat different and more like that of genuine lexical alternants of full DP expressions.

(9) The usual sort of qualification about what counts as “human” must be made here, but does not affect the basic point.

(10) If “personal subject clitics” are a single category, as assumed here, that accounts for the impossibility of having both “person” and “number” clitics in Poletto's sense in the same language. Some other factor must be responsible for the fact that “invariable” and “deictic” clitics do not co-occur, though this may in fact be an accidental gap, since the number of languages attesting either is relatively limited.