To sum up, I have suggested that Kayne's ECP can be extended to deal with the lack of object-subject asymmetries in the extraction pattern of Chamorro, a VSO language. A crucial feature of the account is the idea that the ECP notion of government is defined in terms of constituency at S-structure. In Chamorro, however, the government relevant for Case and Binding cannot be so defined. This suggests that the language employs two distinct notions of government that happen to fall together in configurational languages. Traces in Chamorro must not only satisfy the ECP but must also be governed in the Case-Binding sense; in this respect they resemble their analogues in configurational languages, which must be both governed and properly governed. Finally, I have asserted that these results can be extended to VSO languages in general. Despite its speculative character, the overall hypothesis bears on several larger issues that I would like to mention in conclusion. First, the types of syntactic structure relevant to grammatical theory. In positing two types of government for Chamorro, we have essentially recognized a distinction between surface constituency on the one hand and a more abstract relational structure on the other. While the former is language-particular, the latter may be universal; both are required to characterize the class of allowable extraction gaps in the hypothesis developed here. A distinction between constituent structure and relational structure figures prominently, of course, in Relational Grammar and Lexical-Functional Grammar, where it is put to rather different uses. It strikes me as interesting and comforting that the same kind of distinction should prove useful in GB. A second issue concerns Logical Form. While Chomsky assumes that the ECP holds at LF, his discussion (1981, pp. 233–39, 243–44) is careful to leave open the option of stating that principle at S-structure or Phonetic Form instead. This paper supports such an option, in the sense that it accounts for the extraction patterns of English vs. Chamorro by appealing to a difference between configurational and nonconfigurational tree structure that is unlikely to be reflected at LF. But the reality of a level of representation is established partly by the grammatical principles that refer to it. To the extent that the discussion above locates the ECP at S-structure, therefore, it removes one argument for the existence of Logical Form as a ‘syntactic’ level separate from semantic representation. Finally, the results reached here will be meaningful only if they can be extended beyond Chamorro to other VSO languages and, beyond that, to languages of other word order types. In this sense, the hypothesis above implicitly raises the issue of configurationality in other languages. While it is sometimes assumed that all languages in which the verb and its complements are adjacent in linear order have a surface VP constituent, there is no particular reason for this to be so. On the contrary, given standard assumptions about tree structure, the only languages for which the question can be answered immediately are those that cannot have a surface VP: VSO languages, OSV languages, and perhaps W* languages (if these exist). This suggests that before our approach to extraction can be tested on other languages, the surface configurationality or not of those languages must be established independently. Although this complicates the enterprise, it does not — I think — mean that the hypothesis is in principle untestable.