Case Study: True Wh-in-Situ in Palauan
Wh-movement, wh-agreement, and wh-in-situ in Palauan are investigated by Georgopoulos (1985, 1991), whose work we draw on here. In this Austronesian language, overt wh-movement is flagged by a type of wh agreement that registers the Case of the moved wh-phrase (similar to Chamorro ‘‘wh-agreement’’) but is signaled by mood morphology on the verb.7 Compare the simple declarative clause in (1) with the constituent questions in (2a) and (2b). In (2a), the nominative Case of the wh subject is flagged by realis morphology and the absence of subject-verb agreement; in (2b), the nonnominative Case of the wh-object is flagged by irrealis morphology (even though the semantic mood is realis).
(1)
(2)
The long-distance questions in (3a–b) reveal that Palauan wh-agreement is of flavor (i): it surfaces on every verb on the path of wh-movement.
(3)
Alongside questions formed by overt wh-movement, Palauan productively allows wh-in-situ, as (4a–b) show. In (4a), the wh-in-situ is a subject; in (4b), a direct object. Crucially, wh-agreement is not triggered by wh in-situ, as can be seen by comparing these examples with (3a–b). Notice further that wh-in-situ is possible with matrix scope (as in (4a)) or embedded scope(as in(4b));for discussion, see Georgopoulos 1991,152–155.
(4)
We therefore expect there to be other Palauan evidence that argues that wh-in-situ does not involve movement in overt syntax, followed by pronunciation of the lowest copy. There is such evidence, though of a rather surprising sort.
Overt wh-movement in Palauan can evidently cross the boundaries of certain islands, including complex DP islands and embedded question islands (see Georgopoulos 1991, 80–82). But wh in-situ observes embedded question islands in the following sense: it is impossible for a wh-in-situ to take scope outside an embedded question. Consider (5), in which the matrix verb obligatorily selects an embedded question. Even though this embedded question contains two instances of interrogative wh-in-situ, neither interrogative phrase can take matrix scope.
(5)
Georgopoulos (1991, 171–173) also shows that the judgments do not change if one of the interrogative phrases in (5) undergoes overt wh movement within the embedded question. The broader generalization, then, is that wh-phrases that surface within an embedded question, including instances of wh-in-situ, cannot take scope outside it.
Further evidence for the correctness of this generalization is provided by question-answer pairs like (6).
(6)
In the question in (6), the matrix verb can select either an embedded question or a declarative clause as its complement. One argument of the matrix verb is a displaced wh-phrase; both arguments of the embedded verb are wh-phrases that surface within the embedded clause, one displaced and the other in situ. Structures of this type are clearly questions in which the matrix wh-phrase has wide scope. What are the possible interpretations of the embedded wh-phrases? The answers in (6) reveal that the embedded wh-phrases can both take matrix scope (in which case the matrix verb selects a declarative clause; see answer (b)) or they can both take embedded scope (in which case the matrix verb selects an embedded question; see answer (a)). But it is impossible for one embedded wh-phrase to have wide scope while the other has embedded scope (see answer (c) and the discussion in Georgopoulos 1991, 168 171). In other words, it is (once again) highly deviant for wh-phrases that surface within an embedded question, including instances of wh-in-situ, to take scope outside that question.
To recapitulate, overt wh-movement in Palauan can evidently cross the boundaries of embedded question islands, whereas wh-in-situ appears to be sensitive to embedded question islands in the sense just outlined. This surprising difference confirms that wh-in-situ in Palauan is not, after all, derived via syntactic wh-movement. Were it (counterfactually) the case that all instances of wh-in-situ involved movement in overt syntax, the asymmetry between wh-in-situ and overt wh-movement with respect to island violations would be hard to explain. But the Palauan pattern receives a straightforward account if we adopt an analysis in the spirit of Pesetsky 1987 and much subsequent work (see, e.g., Cheng 1991; Cole and Hermon 1994, 1998; Reinhart 1997; Cheng and Rooryck 2002), according to which some types of wh-in-situ are indefinites, interpreted as free variables eligible for unselective binding. Suppose we assume that in Palauan, in-situ wh-phrases are interpreted as free variables that must be bound by the nearest c-commanding Q operator (i.e., the nearest c commanding C that bears the feature [Q]). Then the fact that wh-in-situ cannot take scope outside embedded questions, and in this sense appears to respect embedded question islands, will follow immediately.
The analysis that we are suggesting is not unusual from the standpoint of the typology of indefinites; see Matthewson 1999 on St’at’imcets ku, another morphologically marked indefinite that must be bound by a c commanding operator and might conceivably force narrowest scope. At the same time, the analysis makes further predictions whose status remains unclear. For instance, if wh-in-situ in Palauan never undergoes movement of any sort, it should be able to take scope outside adjunct islands. Evidence bearing on this prediction is, as far as we know, not available. Setting aside these uncertainties, we provisionally conclude that wh-in-situ in Palauan does not undergo movement; rather, it is licensed by unselective binding.
