Syntactico-Semantic Approach to Binding: Evidence from Serbian*

1. Introduction

One of the most controversial questions in the literature on anaphor binding is whether the binding theory can be defined in purely syntactic (i.e. configurational) terms or in purely semantic (i.e. theta-theoretic) terms. The aim of this paper is to show that Serbian binding facts provide good evidence for defining binding theory in terms of both syntactic and semantic notions.

This paper is organized as follows. In Section 2, I illustrate the relevant binding facts, which show that the notion of SUBJECT, necessary for defining a governing category, must be defined in both syntactic and semantic terms. In Section 3, I discuss the licensing requirements of both anaphors and pronouns in Serbian. In Section 4, I propose a binding theory for Serbian that treats both local and long-distance anaphor binding (i.e., binding across subjects of certain noun phrases) in a uniform manner. Section 4 summarizes my conclusions.

2. The Relevant Binding Facts

In keeping with Pica's (1987) generalization, Serbian reflexives, 'sebe' and 'svoj', being morphologically simple in form, are subject oriented, i.e., they tend to favor subjects as their antecedents (cf. (1)).

(1) Jovan (i) je ispricao Mariji (j) sve o sebi (i, *j).
John AUX told Mary-DAT everything about self
'John told Mary everything about himself/herself.'

In Serbian, 'subject' means grammatical subject, rather than logical subject. This is especially evident in passive constructions, in which an anaphor is bound to the grammatical subject (cf. (2)).

(2) Marija (i) je ubijena u svom (i)/njenom (*i) stanu.
Mary AUX killed in self's / her apartment
'Mary was killed in her apartment.'

As indicated above, the possessive reflexive 'svom' refers to the subject of a passive verb, whereas the regular possessive pronoun (cf. njenom) must be disjoint in reference from the subject.

Based solely on the above facts, a SUBJECT, necessary for defining a governing category, can be simply defined in terms of a syntactic or a configurational notion, namely, as an NP realized in the specifier position. However, when we consider the binding possibilities within the Serbian noun phrase (cf. (3-4)), we find a curious variation in what counts as a SUBJECT.

(3) Jovan (i) je primetio [Marijinu (j) losu brigu o sebi (*i, j)].
John AUX noticed Mary's-ADJ bad caring about self
'John noticed Mary's poor caring of herself.'


(4) Jovan (i) je procitao [Marijin (j) clanak o sebi (i, ?*j)].
John AUX read Mary's-ADJ article about self
'John read Mary's article about herself.'

In (3), in which the reflexive 'sebi' is embedded in a process nominal (cf. brigu 'caring'), the reflexive is bound within the local nominal domain by the possessive referential adjective Marijinu 'Mary's', showing that 'subjects' of process nominals count as SUBJECTs for the purpose of defining a governing category. We find the contrary pattern in (4), in which the reflexive 'sebi' is embedded in a non-process nominal (cf. clanak 'article'). In this example, the reflexive is bound in a 'non-local' domain, i.e. outside the noun phrase and only marginally within a local nominal domain, indicating that specifiers of non-process nominals count as a 'non-subject' for the purpose of the subject orientation of the reflexive. Although not shown, the same binding facts obtain for the possessive reflexive 'svoj'. The main question that arises is what makes process nominals opaque for anaphor binding. In other words, why do only subjects of process nominals act as SUBJECTs whereas SUBJECTs of other nominals do not? In answering these questions, I resort to semantic differences between the two types of nouns, which result in defining the SUBJECT in terms of semantic, i.e. theta-theoretic notion. Before doing this, I briefly discuss the issue of licensing of anaphoric elements.

3. Licensing of Anaphoric Elements

Serbian provides good evidence that licensing of lexical anaphoric elements (pronouns and reflexives) must include both the type of antecedent that can bind the anaphoric elements and the minimal domain for their binding. Section 3.1 illustrates the first requirement and Section 3.2 the second.

3.1 Antecedent Requirements

As seen in (1) and (2) above, the Serbian reflexives 'sebe' and 'svoj' are subject oriented, i.e., they tend to choose subjects as their antecedents. In example (1), the reflexive 'sebe' and its antecedent are arguments of the same predicate, i.e., they belong to the theta-grid of a single verb. The subject orientation of Serbian reflexives is also retained in constructions in which the reflexive in question is an adjunct, i.e., it is not a part of the same theta grid as its antecedent (cf. (2) above and (5) below).

(5) Marija (i) je videla zmiju pored sebe (i) / nje (*i)
Mary AUX saw snake next self / her
'Mary saw the snake next to her.'

As indicated in (5), as well as in (2), Serbian pronouns require non-coreference with the local subject, i.e. they are 'anti-subject' oriented. This binding pattern, i.e. that reflexives be subject-bound while pronouns be subject-free, is found to be typical of languages with morphologically simple reflexives (e.g., other Slavic languages such as Polish (Reinders-Machowska, 1991), Czech (Toman, 1991), and Russian (Rappaport, 1986), and also non-Slavic languages such as Japanese (Katada, 1991), and Norwegian (Hestvik, 1991, 1992)).

3.2 Domain Requirements

According to Browne (1993), the binding domain for reflexives is the tensed (i.e. finite) clause or the noun phrase. The following example illustrates the clausal binding domain.

(6) Jovan (i) zna [da Petar (j) voli sebe (*i, j) / njega (i, *j)].
John knows that Peter loves self / him
'John knows that Peter loves himself/him.'

Here we still retain the complementary distribution of reflexives and pronouns. Specifically, the reflexives must be bound within a tensed clause, while pronouns must be subject-free within a tensed clause. The binding domain in infinitival complement clauses cannot be tested since they are nearly absent in Serbian.

The second binding domain for the reflexives is the noun phrase, "if it contains a possessor interpretable as a subject" (Browne 1993: 367). We saw above that only process nominals containing the subject establish a binding domain. In other words, 'subjects' of process nominals, whether prenominal (as in (3) above) or postnominal (as in (7) below), obligatorily serve as the antecedents for the reflexives.

(7) Jovan (i) je primetio [losu brigu studenata (j) o sebi (*i, j)].
John AUX noticed bad caring students-GEN about self
'John noticed students' poor caring of themselves.'

On the other hand, the binding domain with non-process nominals is generally extended to the entire clause. More precisely, the clausal subject can serve as an antecedent for the reflexives even in the presence of the intervening subject of these NPs (cf. (4) above, repeated below as (8)), indicating that the subjects of non-process nominals are transparent with respect to anaphor binding [1].

(8) Jovan (i) je procitao [Marijin (j) clanak o sebi (i, ?*j)].
John AUX read Mary's-ADJ article about self
'John read Mary's article about him/herself.'

What is especially interesting with these types of noun phrases is that regular pronouns can be bound by a local subject, i.e. the subject of the noun phrase. This is illustrated below.

(9) Jovan (i) je procitao [Marijin (j) clanak o njemu (?*i) / njoj (?j)].
John AUX read Mary's-ADJ article about him / her
'John read Mary's article about him/herself.'

Crosslinguistically, this behavior, i.e., that pronouns are bound in a local domain while reflexives are only marginal in the same domain, is very unusual. In the next section, I offer an account for this unusual behavior of reflexives and pronouns embedded in non-process nominals.

4. A Proposal

In this section, I propose a binding theory for Serbian that is able to account for the above binding facts, which are summarized below.

(10) a. Reflexives ('sebe' & 'svoj') are subject-bound either
within a tensed clause or within a process nominal.
b. Pronouns are subject-free either within a tensed clause or within a process nominal.

From the above descriptive generalization, we observe that process nominals act as an opacity factor for binding, whereas non-process nominals do not. In answering the question as to why there arises a difference between the two types of nouns, I adopt the view held by many linguists (e.g., Grimshaw 1990; Valois 1991; Zucchi 1993) that non-process nominals have pure nominal characteristics while process nominals are a mixture of verbal and nominal features. Some of the verbal characteristics of process nominals are the ability to take aspectual modifiers, or to obligatorily take the same arguments as their corresponding verbs[2]. As to why process nominals act as an opacity factor for binding, following Grimshaw (1990), I claim that process nominals have the same argument structure as their morphologically related verbs. More precisely, subjects of process nominals are arguments while subjects of other nominals are adjuncts, with no unique thematic role specification. Only subjects of process denoting nominals, having a specific thematic role, (i.e., a role "thematically selected" by the head noun as in Higginbotham 1983), count as SUBJECTs. Subjects of other nominals, having no fixed relation to a head noun (cf. "Possessor" or R-relation of Higginbotham 1985) do not count as SUBJECTs, hence being transparent to anaphor binding [3]. A similar conclusion was reached by Giorgi (1987) for the similar binding possibilites within Italian noun phrases.

At this point, the above descriptve generalization can be formalized in terms of the following binding conditions.

(11) Condition A: An anaphor must be SUBJECT-bound
in its binding category.
Condition B: A pronoun must be SUBJECT-free in its
binding category.

where SUBJECT is defined as:

(12) The SUBJECT is a theta-argument of a predicate realized in the Spec position.

where a Binding Category is defined as:

(13) A binding category for an element A is a minimal XP that contains A and a SUBJECT [4].

Finally, the term 'SUBJECT-bound' means both coindexed with and c-commanded by a SUBJECT, and 'SUBJECT-free' means not SUBJECT-bound.

The two binding conditions stated above are inverses of each other, indicating that reflexives and pronouns are in complementary distribution. Thus, as far as Serbian binding facts are concerned, the "disjoint reference" principle (i.e. Condition B) does not need to be stated as a separate principle, but should, as Bouchard (1984) and Burzio (1989) have already proposed, be viewed as an Elsewhere Principle. Based on the facts from Polish, Reinders-Machowska (1991) reaches a similar conclusion.

We further observe that both orientation and locality effects are encoded in the above binding conditions. In this respect, we adopt Vikner's (1985) and Wexler & Manzini's (1987) proposal about the existence of two parameters: antecedent parameter and domain parameter. However, contrary to their view about the independence of these two parameters, we claim that these two parameters are interconnected. This is captured by the notion SUBJECT, which has two functions: (i) to act as a binder and (ii) to define a binding domain (i.e. binding category).

Furthermore, the binding theory proposed above, does not assume reflexivization to occur within a single predicate, i.e., an antecedent and a reflexive do not need to be thematic coarguments (as assumed, for example, by Reinhart & Reuland 1991, 1993 and Pollard and Sag 1992). This is especially evident from the examples such as (2) & (5) above, in which a reflexive is embedded in an adjunct position (i.e. locative PPs). Those examples clearly show that the antecedent requirement, i.e., the notion of SUBJECT as defined above, plays a crucial role in defining a 'governing' category.

We also note that the definition of SUBJECT proposed in (12) refers to the syntactic, i.e. configurational notion, Specifier, where the Specifier means both prenominal and postnominal specifiers (cf. (3) & (7))[5]. The notion of Specifier is necessary because in passive sentences a grammatical subject, but not a semantic subject (i.e. an external argument), binds a reflexive (cf. (2)).

The definition of SUBJECT is not stated in terms of an accessible SUBJECT. This is due to the fact that Serbian, like many other languages, does not obey the i-within-i condition of Chomsky (1981). The binding condition A in (11) is able to account for the lack of the i-within-i effect in Serbian (cf. (14)) because the locality requirement would be violated in such constructions, i.e., an anaphor would not be bound within its binding category (i.e., embedded finite clause).

(14) Jovan (i) zna [da je [prica o sebi (*i) / njemu (i)] netacna.
John knows that AUX story about self / him not+true
'John knows that a story about himself is not true.'

Finally, it is worth pointing out that I use the term 'bound' to mean both coindexed and c-commanded by an antecedent. The ungrammatical example below illustrates that an anaphor must be c-commanded by an antecedent.

(15) *[Jovanova (i) knjiga] je proslavila sebe (i).
John-ADJ book AUX celebrated self
*'John's book made himself famous.'

In sum, simply by parametrization of SUBJECT, the binding theory proposed above is able to account for the complex set of Serbian binding facts . In this regard, it is worth noting that the LF movement theory (e.g. Pica 1987), which assumes that monomorphemic reflexives move to INFL at LF (a position from which reflexives can be c-commanded by a subject) would not be able to account for the Serbian facts. Specifically, this theory would fail to account for the NP-internal binding (involving process nominals), since a nominal domain has no INFL node for the reflexive to move to. In addition, this theory would be unable to account for the subject orientation of Serbian reflexives embedded in adjunct positions (cf. (2) & (5)). This is because the movement of a reflexive out of an adjunct phrase would involve crossing a barrier (adjuncts being non-L(exically) marked), hence, violating the ECP (i.e. antecedent government).

5. Conclusions

In this paper, I proposed a binding theory for Serbian that is based on the parametrization of the notion of SUBJECT, which is defined in syntactico-semantic terms, such as a theta-argument of a predicate occupying the specifier position. SUBJECT, defined in this way, plays two roles: it acts as a binder and it defines a minimal binding domain (i.e. binding category). Through the notion SUBJECT, we thus capture the dependence between two licensing requirements for anaphoric binding: the antecedent requirement and the domain requirement. The question of the extent to which SUBJECT can perform this double role in other languages is put forward as an issue for further research.

Furthermore, by defining SUBJECT in this way, both local and "long-distance" binding (i.e., binding across 'Specified Subjects') are given a unified treatment, treating them as instances of local binding. In this respect, the binding domain for anaphors in Serbian would, according to Reuland & Koster's (1991) classification, fall into the group of "structurally" defined domains, i.e., local domains in which a SUBJECT is an opacity factor for binding.

Finally, it is worth noting that the above definition of binding category is more akin to the standard definition of governing category (cf. Chomsky 1981) than to other current definitions that try to eliminate the SUBJECT condition from the definition of a governing category (e.g., Johnson 1987; Hestvik 1991; Reinhart & Reuland 1993).

Notes

* I would like to thank Lee Baker, Ileana Comorovski and Arnim von Stechow for their helpful discussions and encouragment. I am also grateful to those friends of mine, native speakers of Serbian, who provided me with their judgments of the sentences discussed in this paper.

[1] Rappaport (1986 : 106) also reports that the Russian "NP specifier is transparent to reflexive binding". However, Rappaport provides examples only with NP-specifiers of non-process nominals. Similarly, Reinders-Machowska (1991) and Toman (1991) show that Polish and Czech NP-specifiers can be transparent for anaphor binding. Toman (1991) tries to account for the transparency of subjects of NPs in Czech by saying that this is due to the adjectival nature (i.e. possessive adjectives) of such subjects. Although the same is true in Serbian (the prenominal subjects of NPs are possessive adjectives), subjects of process nominals are still an opacity factor for binding.

[2] Serbian process nominals (generally derived from imperfactive verbs) also exhibit these characteristics. For detailed morpho-syntactic distinctions between process and result nominals in Serbian, see Zlatic (1994).

[3] As suggested by A. Von Stechow (personal communication), a difference between process vs. result nominals can also be captured by the fact that only process nominals have an external, non-thematic EVENT argument.

[4] I borrow the term 'binding category' from Bouchard (1985) (originally used by Chomsky 1981), instead of a standard term 'governing category' that includes the X-zero governor of an anaphor. This is because I agree with Bouchard that "an X-zero governor has very little effect in the binding theory" (except to account for the ECM constructions present in English, but not in Serbian).

[5] In order to account for Russian binding facts, Rappaport (1986) also assumes that both prenominal and postnominal subjects of NPs are specifiers.

REFERENCES

Bouchard, D. (1984) On the Content of Empty Categories, Foris, Dordrecht.

Bouchard, D. (1985) The Binding Theory and the Notion of Accessible SUBJECT, Linguistic Inquiry 16, 117-133.

Browne, W. (1993) Serbo-Croatian, in B. Comrie and G. C. Corbett, eds., The Slavonic Languages, Routledge, London.

Burzio, L. (1989) On the Non-Existence of Disjoint Reference Principles, Rivista di Grammatica Generativa 14, 3-27.

Chomsky, N. (1981) Lectures on Government and BindingForis, Dordrecht.

Giorgi, A. (1987) The Notion of Complete Functional Complex: Some Evidence from Italian, Linguistic Inquiry 18, 511-518.

Grimshaw, J. (1990) Argument Structure, MIT Press, Cambridge, MA.

Hestvik, A. (1991) Subjectless Binding Domains, NLLT 9, 455-496.

Hestvik, A. (1992) LF Movement of Pronouns and Antisubject Orientation, Linguistic Inquiry 23, 557-594.

Higginbotham, J. (1983) Logical Form, Binding, and Nominals, Linguistic Inquiry 14, 395-440.

Higginbotham, J. (1985) On Semantics, Linguistic Inquiry 16, 547-593.

Johnson, K. (1987) Against the notion 'Subject', Linguistic Inquiry 18, 354-361.

Katada, F. (1991) The LF Representation of Anaphors, Linguistic Inquiry 22, 287-313.

Koster, J. & E. Reuland, eds. (1991) "Long Distance Anaphora", Cambridge University Press, Cambridge, MA.

Pica, P. (1987) On the Nature of the Reflexivization Cycle, NELS 17, 483-499.

Pollard & Sag (1992) Anaphors in English and the Scope of Binding Theory, Linguistic Inquiry 23, 261-303.

Rappaport, G. C. (1986) On Anaphor Binding in Russian, NLLT 4, 97-120.

Reinders-Machowska, E. (1991) Binding in Polish, in Koster & Reuland 1991, 137-151.

Reinhart, T. and Reuland, E. (1991) Anaphors and Logophors: an Argument Structure Perspective, in Koster & Reuland 1991, 283-323.

Reinhart, T. and Reuland, E. (1993) Reflexivity, Linguistic Inquiry 24, 657-720.

Reuland, E. & Koster, J. (1991) Long Distance Anaphora: an Overview, in Koster & Reuland 1991, 1-27.

Stevanovic, M. (1991) Savremeni Srpskohrvatski Jezik, Fifth edition, Naucna Knjiga, Beograd.

Toman, J. (1991) Anaphors in Binary Trees: an Analysis of Czech Reflexives, in Koster & Reuland 1991, 151-171.

Valois, D. (1991) The Internal Syntax of DP, Doctoral. Dissertation, University of California, Los Angeles.

Vikner, S. (1985) Parameters of Binder and Binding Category in Danish, Working Papers in Scandinavian Syntax 23, University of Trondheim.

Wexler, K. & Manzini, R. (1987) Parameters and Learnability in Binding Theory, in T. Roeper & E. Williams, eds., Parameter Setting, Dordrecht, Reidel.

Zlatic, L. (1994) An Asymmetry in Extraction from Noun Phrases in Serbian, in G. Fowler & J. Ludwig, eds., Indiana Slavic Studies 7, 207-217.

Zucchi, (1993) The Language of Proposition and Events, Kluwer Academic Publishers, Dordrecht.

If you wish to make a public comment about this paper; please send your message to: linconf@tamvm1.tamu.edu

Discussion of this paper is available at the Discussion Web Site at: