Domain D of principle A
The condition that an anaphor must be c-commanded by its antecedent is not sufficient to determine the distribution of anaphors. Therefore the basic c-command requirement must be supplemented with a domain statement which specifies the subdomain(s) of a sentence in which an anaphor cannot be free (even when it is bound in the sentence). We say that this subdomain is opaque to binding with respect to an antecedent outside the subdomain. Ungrammaticality due to binding across an opaque domain will be referred to as an opacity effect.
The following paradigms give the basic cases to be accounted for.
(2)
The paradigm shows that an anaphor must be bound in the minimal finite S containing it. This accounts for the illformed examples in (1) as well as in (2.a.i) and (2.b.i), but does not account for (2.b.ii) where the anaphor is bound in the minimal finite S containing it (=the matrix S). Therefore it is necessary to specify a second opaque domain for anaphors—namely the domain a subject. This is independently motivated for binding into NP, as will be discussed below.
The specification of two domains for Principle A (i.e. the domains of subject and finite S) creates an apparent problem with respect to overlap. Both domains pick out (1.b.ii) and (2.b.i). This overlap is problematic if, as in previous work (cf. Chomsky 1973, 1976), the two domains are taken as defining two distinct conditions on the distribution of anaphors (e.g. the Tensed S Condition (TSC) and the Specified Subject Condition (SSC)) and violations of more than one condition are assumed to increase ungrammatically. This second assumption is not borne out for the anaphora paradigms above since (1.b.ii) and (2.b.i) are not significantly more ungrammatical than other illformed examples in the paradigm. Nonetheless, the overlapping specifications have been considered a defect in the formulation of conditions on the distribution of anaphors (see Chomsky 1977b, 1980, and 1981).
Two solutions for this problem have been proposed. Chomsky 1980 sought to characterize two non-overlapping conditions, the SSC plus the Nominative Island Condition (NIC), where the domain of tense is dropped in favor of a prohibition against anaphors bearing nominative case. The NIC proposal has the empirical advantage of accounting for the grammaticality of (3), in contrast to the TSC and its variants (e.g. the Prepositional Island Condition (PIC) of Chomsky 1977).
(3)
In (3) the anaphors are free in the domain of tense (in the embedded S) but are not in the nominative case. Another advantage of the NIC is that it accounts for opacity effects in infinitival complements in Portuguese (see Rouveret 1980) and Turkish (see George and Kornfilt 1981) where infinitival subjects may be marked for agreement with the complement verb. Such subjects are marked in the nominative case. As predicted by the NIC, anaphor subjects are prohibited in these constructions.
Nonetheless, there are several problems with the NIC proposal. While examples like (3) are acceptable in English, they tend to be illformed in other languages. In Dutch, for example, comparable examples with reflexives are ungrammatical (see Lasnik and Freidin 1981 for further discussion). Secondly, the NIC does not account for opacity effects in Turkish gerunds where the subject is marked genitive when the subject agrees with the verbal form, and genitive anaphor subjects are prohibited as illustrated in (4).
(4)
This example suggests that it is not simply a prohibition against anaphors with certain morphological properties which accounts for opacity in these examples, but rather that opacity is determined in terms of syntactic domains. The NIC does not define a syntactic domain, and therefore is not an island condition in the usual sense. For example, although the bracketed NP in (3) is marked nominative, it is not an opaque domain for binding. Icelandic presents yet another empirical problem for the NIC analysis. In Icelandic, finite indicative clauses are opaque to binding, whereas infinitival clauses are not—even when binding occurs across a subject as in (5).
(5)
Icelandic does not appear to observe the SSC. If it does not, then the opacity effects for non-subjects in finite clauses cannot be due to the SSC (see Harbert 1982b for further discussion). For Icelandic the notion “opaque domain” for binding must be defined in terms of some characterization of “finite clause”.
If we reject the NIC, then the opacity effects for subjects of infinitivals in Portuguese and Turkish should also fall under our characterization of “finite clause”. (Note that this characterization does not use the term “finite” in its time reference sense.) Since the agreement element (henceforth AGR) occurs in all the relevant examples under discussion, “finite clause” can be characterized as the domain of AGR—assuming that AGR is a grammatical formative in INFL (=inflection) which is an immediate constituent of S and hence c-commands both VP and the syntactic subject in S. This analysis accounts for (4), which is not covered by the NIC.
Because for English the SSC is needed independently of a finite clause condition on binding (see (2.b.ii), the problem of overlapping domains arises again since the domains of subject and AGR overlap for non-subject NPs. A second solution to this problem is to give a single specification for domain D of Principle A so that the domains of subject and AGR follow from this specification. Initially this was accomplished in the GB framework by attempting to integrate the binding theory with notions of case as follows. Considering the core cases (1–2), we note that an anaphor must be bound in the minimal S-domain in which it receives case.10 Under this unitary characterization (2.a.i) and (2.b) are illformed because the anaphors in them are assigned case in the embedded S-domain and are free in that domain. In contrast, the anaphor in (2.a.ii) is assigned case by the matrix verb and is bound in the matrix S, which is the minimal S-domain in which the anaphor receives case.
In GB theory this analysis is formalized in terms of the notion of government, since case is assigned to NPs under government by a case-assigning element (e.g. V or P). Thus domain D of Principle A is designated as the governing category of the anaphor, where governing category is defined as the minimal S or NP containing the anaphor and a governor of the anaphor. In S, the governors will be V, A, P, and AGR in INFL—in effect, the heads of maximal phrasal projections in (see Freidin 1983) with the additional stipulation that INFL qualifies only with it contains AGR.
While this characterization accounts for the S-paradigm as given in (1–2), it does not generalize to the paradigm for binding in NP given in (6–7).
(6)
(7)
The paradigm illustrates that the subject position in NP is always transparent to binding by a c-commanding NP ((6.a) and (7.a.i–ii)) unless a c-commanding subject intervenes ((6.b.ii) and (7.a.iii–iv)). Thus the paradigm for NP diverges from that of S in two ways: 1) the subject of NP is always transparent, in contrast to the subject of S, which is transparent only when AGR is absent; and 2) the object position in NP is transparent when there is no subject ([NP, NP]), whereas the object position is almost always opaque to binding from outside S (but cf. (19) below). The theory in which domain D of Principle A is specified as the governing category of the anaphor (as defined above) can only account for opacity effects on binding across subjects of NP. The transparency of NP with respect to binding cannot be accounted for by a mechanism similar to (as in the S-paradigm—see note 11) since there is no exceptional case marking into NP to motivate such a proposal.
For NP, the relevant factor which determines its transparency for binding is the presence or absence of a syntactic subject. The SSC is the only relevant opacity condition for NP. Even a c-commanding AGR does not induce opacity in NP, as illustrated in (7.a.i) and (7.b.i). With respect to binding then, there is a basic asymmetry between S and NP. Therefore the governing category analysis of the S-paradigm, which collapses the SSC and the relevant variant of the TSC under a single specification of opaque domain, does not generalize to NP.
Given that the SSC must be included in the domain statement for Principle A, we once again confront the problem of potential overlap with the other part of the domain statement which accounts for opacity of subjects in the S-paradigm. As discussed above, Turkish and Portuguese suggest that AGR rather than tense is the relevant factor for opacity. Since AGR is always linked to a syntactic subject, there is some motivation for considering the agreement element as a kind of (shadow) subject. Using this observation, we can unite AGR with syntactic subjects (i.e. [NP, S] and [NP, NP]) under the designation ‘SUBJECT’ (read “capital subject”)—see Chomsky 1981:209ff. Under this analysis the domain statement for Principle A reduces to “domain of SUBJECT”.
This formulation of Principle A accounts for both the S- and NP-paradigms, with the exceptions of (7.a.i) and (7.b.i). In these latter cases the anaphor is free in the domain of a SUBJECT (= AGR of the sentential complement). This analysis is corrected by modifying the definition of opaque domain so that an AGR element does not function as a SUBJECT for anaphors contained in an NP which is coindexed with it. Note that in the cases where AGR functions as a SUBJECT with respect to an anaphor (i.e. the NIC cases), the anaphor is coindexed with the AGR element. Coindexing of the anaphor in (7.a.i) with AGR in the embedded S gives the representation (8).
(8)
This yields an indexed representation of the sentential complement subject where the referential index of the NP is also carried by a subpart of the NP itself. The interpretation of the coindexing in (8) would be that each other is coreferential with the NP each other’s books—which is of course incorrect. As proposed in Chomsky 1981, this indexing can be excluded by a general prohibition against NPs of the form (9), where i is a referential index.
(9)
(9) is empirically motivated by the impossibility of coindexing a pronoun contained in a NP with the NP itself, as in (10).
(10)
If indexing applies freely as assumed, then some constraint is needed to rule out (10), since it does not fall under Principle B. (See Chomsky 1981:212f. for further discussion).
Following Chomsky 1981, (9) can be incorporated into the formulation of the domain statement for Principle A by modifying the specification “SUBJECT”. Since AGR in (7.a.i) and (7.b.i) should not count as a SUBJECT for anaphor binding, it can be considered ‘inaccessible’ to the anaphors by virtue of (9). Thus the domain statement of Principle A is further modified to (11).
(11)
Accessibility is defined in terms of (9) and “in the domain of” requires that the SUBJECT c-command the anaphor.
The grammaticality of (7.a.i) and (7.b.i) constitutes evidence for this formulation of Principle A with respect to AGR. A similar argument holds for the lexical subjects of infinitival equatives.
(12) Mary believes those to be pictures of herself.
(12) has the indexed representation (13).
(13) Maryi believes [S thosej to be [NPi pictures of herselfi]]
The infinitival subject must be coindexed with the complement of equative be and hence is not an accessible SUBJECT for the anaphor, given (9).
There is a certain naturalness in formulating the domain statement of Principle A in terms of accessible SUBJECT. Since anaphors must get their reference from an antecedent in the sentence in which they occur, it is appropriate to formulate the domain in which they must be bound in terms of a potential antecedent (i.e., syntactic subject or the agreement element which entails the existence of a subject). The notion ‘subject’ is crucial here because the simpler structural notion of c-commanding NP is not sufficient. Thus in
(14) John c-commands each other but does not block proper binding of the anaphor by the subject we.
(14) Wei told John lies about each otheri
Why “subject” rather than “c-commanding NP” is relevant for binding remains to be explained. As noted by Wayne Harbert, this question is related to the question of why in many languages (e.g., Romance and Germanic (excepting English and Icelandic), and Japanese) the only possible antecedents for anaphors are subjects.
As formulated in (11), Principle A still does not cover all the core cases. Consider (15) for example.
(15)
AGR cannot be considered an accessible SUBJECT for the anaphor himself because coindexing of the anaphor with AGR would yield a violation of (9). Thus the anaphor does not occur in the domain of an accessible subject; and therefore (15) is not excluded by Principle A as formulated above. In Chomsky 1981 this problem is resolved by stipulating that an anaphor must be bound in root S. This stipulation is in addition to the domain statement in (11), and in essence constitutes a default condition with respect to (11).
The default condition is unnecessary if we assume that, by definition, an anaphor must be bound to an antecedent (see Freidin 1978:fn.7). Under this assumption, Principle A is a condition on proper binding, where it is given that the anaphor is in fact antecedent bound. We then reformulate (11) as (16):
(16) Principle A: an anaphor cannot be free in the domain of an accessible SUBJECT.
(16) defines the notion ‘opaque domain’ for anaphor binding. Where an anaphor occurs unbound in a sentence (e.g. (1.a.i), (1.b.ii), and (15)), the sentence is illformed because it does not satisfy the basic requirement for anaphors—that they be bound to some antecedent. The requirement is needed independently of (15) to exclude (1.a.i). In the latter example, the anaphor is bound to an accessible SUBJECT—i.e., AGR. The example is illformed nonetheless because AGR is not a possible antecedent since it has no intrinsic reference. This suggests that ‘bound’ should be interpreted as ‘antecedent bound’ rather than ‘coindexed with a c-commanding element’.
So far we have been assuming that the definition of SUBJECT covers all syntactic subjects. This holds for ‘referential’ lexical subjects and their empty category counterparts (i.e., trace), and in addition PRO.
In (17a–b) the anaphor is free in the domain of a trace subject and in (17c) it is free in the domain of a PRO subject. In comparison with referential NP, however, nonreferential (or pleonastic) it does not appear to have a strong opacity inducing effect, as discussed in Freidin and Harbert 1983. For example, (18b) is relatively wellformed in comparison to (18a), which is ungrammatical.
(18)
As illustrated in (19), wellformedness in these constructions is not significantly affected by the addition of AGR to the complement of expect.
(19)
(19b) is on a par with (18b)—i.e., it is relatively wellformed in comparison with the ungrammatical (19a).
Given this statement of the facts, a distinction between “referential” and “nonreferential” subjects is necessary for the formulation of Principle A. Such a distinction can be given in terms of thematic relations (or θ-roles). Referential subjects bear θ-roles and thus constitute θ-subjects, in contrast to non-θ-subjects like nonreferential it. Only θ-subjects qualify as accessible subjects. This account can be extended to the NIC effect if the θ-status of the syntactic subject is somehow transferred to AGR.19 Thus where AGR is linked to a θ-subject, the INFL projection (=S, etc.) constitutes the domain of a θ-SUBJECT. Where S contains a syntactic θ-subject and no AGR, the domain of the θ-SUBJECT is the c-command domain of the syntactic subject. This analysis has a certain naturalness in that non-θ-subjects are not possible antecedents for anaphors since they are inherently nonreferential. (For further details see Freidin and Harbert 1983).
To summarize, lexical anaphors are subject to two conditions:
(20)
(20a) is part of the definition of “anaphor”; whereas (20b) constitutes a condition on proper binding which we call Principle A of the binding theory.
