Case Study: Internally Headed Relative Clauses in Passamaquoddy
Relative clauses in Passamaquoddy, an Eastern Algonquian language, may be either externally headed, as in (1), or internally headed, as in (2). The understood head of the relative clause is italicized in each ex ample. In general, external heads precede the relative clause; internal heads occur nested within the relative clause or at its right edge.
(1)
(2)
It is generally not possible for a relative clause to include both an external and an internal head. Thus, example (3) with a resumptive demonstrative pronoun nı´htol ‘that (obviative)’ is unacceptable. (The personal pro noun ne´kom ‘he, she, it (animate)’ has no obviative forms, so the only possible pronoun here is a demonstrative.)
(3)
Evidence that external heads surface outside the relative clause, whereas internal heads surface inside, is provided by the language’s system of obviation. Roughly speaking, any expression that refers to the most topical or otherwise most prominent third person within a particular span of dis course is proximate. Expressions referring to any other third person in the same context are obviative. (However, proximate or obviative status is overtly marked only for nouns and pronouns that belong to the animate grammatical gender.) Within an obviation span, only one DP can be proximate; all other third person expressions that differ in reference from the proximate DP must be obviative. The minimal such domain in Passa maquoddy is ordinarily the clause.
As Bruening (2001, 206) has observed, however, the external head of a relative clause can be proximate even when some other NP within the relative clause is proximate. Thus, no`t otu´hk ‘that deer’ is proximate in (4), but the subject NP nta´tat ‘my dad’ within the relative clause itself is also proximate. This argues that the external head lies outside the minimal obviation domain—here, the relative clause.
(4)
The obviative status of the relativized correspondent of the NP no`t otu´hk ‘that deer’ is registered by two verbal markers, one of which is the so called direct theme sign-a that is used in forms with a third person subject (as here) only if the object is third person and obviative. The other is the participle agreement suffix-il, which indexes the relativized object NP as obviative and singular. Since the head of the relative clause in this example is proximate, but is nonetheless coreferential with the obviative object of nehpah-a´-c-il ‘he (proximate) kills him (obviative)’, this head can only be external to the relative clause itself.
The distribution of proximate and obviative forms also provides evidence that when the relativized constituent appears on the right periphery of the clause, the clause can only be internally headed. In (5), for ex ample, the obviative status of the relativized constituent otu´hkol ‘deer (obviative)’ is expected if the relative clause here is head internal, since the subject NP within the clause, nta´tat ‘my dad’, is proximate.
(5)
If ‘deer’ could instead be an external head here, however, then we would expect (6) with ‘deer (proximate)’ to be acceptable, since the head of a relative clause need not be construed as occurring within the same obviation domain as nominals contained within the clause. But (6) is, in fact, unacceptable. This argues that a right-peripheral head lies within the relative clause.
(6)
Bruening (2001, 117–124) analyzes the obviation system of Passama quoddy as analogous to abstract Case. Though we do not adopt the details of his analysis, our approach is similar in that it takes proximate and obviative morphology to encode formal features (minimally, the feature [proximate]), which are uninterpretable in the sense that they have no truth-conditional effects on semantic interpretation. This assumption will come into play shortly.
As Bruening (2001, 161–162) observes, relative clauses are derived by wh-movement whether their heads are internal or external. This is shown by the fact that both types of relatives exhibit island effects.
(7)
(8)
In Bruening’s analysis, internally headed relatives undergo covert wh movement in Logical Form (LF). We, too, assume that Passamaquoddy has both externally and internally headed relative clauses: the relativized NP in relative clauses of the first type is a null operator, whereas the relativized NP in relative clauses of the second type is the internal head. But unlike Bruening, we propose that in both types of relative clauses, the relativized NP undergoes wh-movement in overt syntax, landing ultimately in the specifier of the C bearing the active feature [pred] at the top of the relative clause. In the phonology, the lowest copy of the moved wh-phrase is pronounced (if it can be); higher copies are not pronounced. These details of pronunciation emerge clearly in internally headed relatives; in externally headed relatives, they are concealed by the fact that the moved wh-phrase is the null operator, which has no pronunciation.
Evidence for our approach is supplied by participle agreement in relative clauses, which we analyze as wh-agreement. This agreement, which surfaces optionally on V, registers features of the relativized NP, including number and obviation. Consider, for instance, the externally headed relatives in (9), where the participle agreement suffix-il indicates that the relativized NP is obviative and singular.
(9)
Participle agreement also occurs in internally headed relatives like (10).
(10)
In long-distance relative clauses, participle agreement shows itself to be wh-agreement of flavor (i): the agreement surfaces (optionally) on every V along the path of wh-movement. Crucially, this happens whether the relative clause is externally headed, as in (11), or internally headed, as in (12).
(11)
(12)
Bruening (2001, 229) takes this pattern to indicate that in relative clauses of type (25), the internal relative head undergoes covert wh-movement in LF. However, if that were so, the participial agreement on the higher V could not reflect any Agree relation holding in overt syntax between it and the internal head. The problem is that in overt syntax, the two would be lodged in separate phases: because the internal head would not yet have moved, it would remain embedded within the lower CP, which forms its own strong phase, and so the Phase Impenetrability Condition would prevent the higher V from having access to it.
It is worth asking whether the higher V’s participial agreement could reflect any LF relation of Agree holding between it and the internal head, where by ‘‘LF relation’’ we mean one established in the covert part of the derivation, after Spell-Out. (This is evidently Bruening’s proposal; see also Bruening 2001, 209.) The issues here are controversial. So far, we have taken the Agree relation reflected by participial agreement to involve the uninterpretable feature [ proximate]. We have further been assuming, following Chomsky (2001, 5), that ‘‘Spell-Out removes LF uninterpretable material from the syntactic object K...shortly after the uninterpretable features have been assigned values (if they have not been assigned values at this point, the derivation will crash, with uninterpretable features at the interface).’’ This set of assumptions would make it impossible for [ proximate] to persist past Spell-Out and therefore for this uninterpretable feature to be involved in any LF relation of Agree. Alternatively, one could take the position that uninterpretable features can indeed persist past Spell-Out to the covert part of the derivation. If so, the challenge would be to ensure that an Agree relation not present in overt syntax but established only in covert syntax could have phonetic effects. While an analysis along these lines could be constructed, it would be deeply subversive to the spirit of the grammatical architecture assumed by Chomsky in much work (e.g., 1981, 1995, 2001), since that architecture claims that covert syntax (LF) and PF are independent interpretive components that do not directly interact.
No such difficulties arise in our analysis. Under this analysis, participle agreement is analogous to ‘‘wh-agreement’’ in Chamorro. This agreement transparently reflects the relation of Agree in overt syntax that holds between V and the relativized NP, which bears the active feature [wh]. The main difference between Passamaquoddy participle agreement and Chamorro ‘‘wh-agreement’’ lies in which copy of a wh-moved NP is pronounced. In Chamorro, only the highest copy is pronounced, whereas in Passamaquoddy, the lowest copy is pronounced, leading to an internal head (when the relativized NP is not a null operator to begin with).
