Domain D of principle B
As formulated in (7) above, Principle B requires that a pronoun be antecedent-free in a domain D’. The interpretation of “free” here is not, strictly speaking, the converse of the interpretation of “bound” as required for anaphors. Binding for anaphors entails coreference; whereas freedom for pronouns entails the stronger requirement of disjoint reference—see Lasnik 1976 and Higginbotham 1980a:fn.1). In the following discussion the examples cited do not distinguish between disjoint reference and noncoreference.
The S-paradigm for bound pronouns shows a systematic correspondence with the analogous paradigm for anaphors.
(1)
(2)
The correspondence between (1–2) and the analogous paradigm for anaphors is that bound pronouns cannot occur in positions where anaphors are properly bound, and conversely, anaphors cannot be properly bound in positions where bound pronouns may occur. This correspondence motivated earlier analyses (e.g., Chomsky 1981) in which the domain statements of Principles A and B were taken to be identical.
Although this analysis holds as a first approximation, it fails for the larger class of cases. Thus with the NP-paradigm for bound pronouns (3–4) the generalization fails almost in its entirety.
(3)
With the exception of (3.b.iii) where a pronoun in NP must be free with respect to the subject of that NP. NP does not have the same opacity effects for pronouns as it does for anaphors. NP is an opaque domain with respect to disjoint reference whether or not the NP contains a syntactic subject. This suggests that the notion ‘accessible SUBJECT’ is therefore not the correct notion for formulating the domain statement of Principle B. Since a pronoun—in contrast to anaphors—does not require that an antecedent be present in S, the notion ‘accessible subject’ is unmotivated with respect to disjoint reference (see Huang 1983 for further discussion).
Nonetheless, formulating Principle B in terms of accessible SUBJECT does account for the S-paradigm. An alternative formulation must therefore account for the opacity of finite S and the c-command domain of infinitival subject as well as the opacity of NP. The opacity effects for NP with respect to disjoint reference follow without further stipulation if domain D’ for Principle B is formulated in terms of ‘governing category’, defined in the literature as follows.
(5)
Reference to a governor is superfluous since reference to a maximal phrasal projection entails the existence of a lexical head which governs the phrasal constituents of its projection. Unfortunately there are two defects with this simple definition: 1) it fails to account for the S-paradigm; and 2) it incorrectly predicts that AP, PP, and VP will also be opaque with respect to disjoint reference, as indicated in (1) for VP and (6) for AP.
(6)
Huang 1983 proposes that the domain statement of Principle B be formulated in terms of government and the presence of a SUBJECT. Thus Principle B states that “a pronoun must be free in its governing category,” where governing category is redefined as in (7) (cf. Huang 1983: (14)).
(7)
Huang assumes that the lexical head of NP is a SUBJECT of NP (analogous to AGR which is considered the nominal head of the INFL projection (see Chomsky 1981)). Therefore NP is always a governing category for pronouns since the head N is both governor and SUBJECT with respect to pronominal NP constituents of the NP.
This proposal also encounters problems with the S-paradigm. In (8) VP is the minimal category (=maximal phrasal projection of a lexical head) containing a governor of the pronoun her (=V) and a SUBJECT (=her).
(8)
Since, as Huang notes, the accessibility of SUBJECT to the pronoun is not conceptually relevant to pronoun binding, VP will fit the definition of governing category for the pronoun. To define the matrix S as the governing category for (8), we could require that the SUBJECT be distinct from the pronoun. This comes dangerously close to reintroducing the notion of accessibility into the definition of governing category for pronouns. For (9) however we must do just that if (7) is to be maintained in some form.
(9)
VP is again the minimal maximal projection containing the pronoun her, a governor of the pronoun (=the preposition to), and a SUBJECT (=ei). Requiring that the SUBJECT c command the pronoun reinstates the notion of accessibility.
The problem with formulating domain D’ for Principle B is to characterize what is common to NP and S—excluding VP and AP—and in addition allow for the transparency of subjects of infinitivals without resorting to the notion of accessible SUBJECT. Only in this way can we avoid having to stipulate NP and S as binding categories for pronouns.
One common feature of S and NP—in addition to allowing syntactic subjects—is that both are domains in which θ-roles are assigned by predicates (i.e., verbs, predicate adjectives, and certain nominals), as illustrated by the sentence/nominal pair in (10).
(10)
Since V and A may assign a θ-role to an NP outside of their maximal projections (VP and AP)—e.g., to a subject of S, they do not count as “θ-domains.”
(11)
A θ-domain of, a predicate (=V, A, or N), is the minimal domain in which α assigns it θ-roles to arguments.
A θ-domain constitutes a complete functional construct. Note that this is a type (as opposed to a token) distinction since S functions as a θ-domain even when the sentential subject is a non-θ-position, as in (9).
In order to account for the transparency of infinitival subjects in the S-paradigm for pronouns, it is necessary to incorporate the notions of government and θ-domain in the formulation of the domain statement for Principle B.
(12)
Principle B: a pronoun must be free in the θ-domain of its governor.
In (8), the binding domain for the pronoun her will be the θ-domain of believes—i.e., the matrix S. In (9), the preposition to which governs the pronoun her does not assign the pronoun a θ-role independently of the matrix verb seems. Therefore the θ-domain of the preposition will be identical to that of the verb—i.e., the matrix S. In the case of AGR, however, it is unnatural to talk about its θ-domain since AGR does not assign a θ role to the subject it is coindexed with. This problem can be avoided if we mentioned instead the θ-domain of the predicate which is constructed with AGR. Thus the domain statement of Principle B might be more appropriately formulated as “the minimal θ domain containing a governor of the pronoun.”
It is worth noting in conclusion that the notions “θ-domain” and “domain of subject” are conceptually connected in that a subject marks the periphery of a θ-domain.
