LECTURETTE #15: BARRIERS In Lecturette #14 i talked about c-command as the most basic structural relation involved in Government Theory. But the full theory is, unfortu- nately, more complicated than that. Several attempts were made during the 70's and early 80's to define government in terms of c-command (for reviews, cf. Chomsky's Lectures on Government and Binding, pp. 163-165 and Barriers (1986), pp. 8-22). Eventually, Chomsky concluded that an accurate and useful formal definition of government required a series of logical layers, so to speak: a series of definitions each of which de- pends crucially on a term defined in the previous one, and 'government' being defined only at the very end of the sequence. Unfortunately, due to Chomsky's expository style, this sequence is actually presented (in his 1986 book Barriers) in reverse order, with the result that each defi- nition confronts you with a term whose definition you have to seek in subsequent paragraphs. Speaking for myself, i found this presentation rather irritating when i first struggled through it in 1986, and i'm going to give you the relevant definitions in what i regard as a more logical order. But i'm not going to pretend this is easy stuff. I don't expect anyone to *memorise* the relevant definitions, though after seve- ral years of working with them i will admit to having acquired a pretty good seat-of-the-pants sense of what is and is not a 'proper' government relation. I'm going to give you the straight dope as well as i can, on the understanding that this is to be treated as reference material; you should be able to refer back to it at any time. I. L-Sisters The first stage in this sequence of definitions isn't explicitly highligh- ted in Chomsky's Barriers book at all, though it is stated on p. 13. The label 'L-sisters' is my coinage. As you know, according to the standard definition two constituents are sisters if they are dominated by all the same constituents. They are L-sisters if they are dominated by the same *projections of lexical categories*. I.e., intervening functional heads and their projections don't count for this relation. II. Theta-Marking The second stage in this sequence of definitions is really an expression of Theta-Theory. If a constituent A has a theta-role to assign, then it can *theta-mark* another constituent B if A and B are L-sisters. If B is a phrasal projection, then A equally theta-marks its head. This second condition (the possibility of a constituent theta-marking not only its complement but also the complement's lexical head) is explicitly stated in Chomsky 1986:19; i have argued in a recent paper that it shouldn't be necessary to stipulate this, that theta-marking ought to be able to perco- late down to a lexical head (or, alternatively, that a complement phrase should be transparent to theta-marking) automatically, but we don't need to go into that now. III. Theta-Government Assuming that A theta-marks B as in the previous definition, A also theta- governs B if A happens to be a Bar-0 constituent. Note that this defini- tion is satisfied if A is a functional head such as COMP or AGR, as well as if it is a lexical category like N or V. This distinction is critical at the next stage: IV. L-Marking If A is a *lexical* (as opposed to a functional) category and theta-go- verns B, than A also L-marks B. If B is a maximal projection, than A also L-marks the specifier of B. Note that under these circumstances (i.e., when B is a maximal projection), A can theta-mark (and theta-go- vern) the *head* of B and L-mark its *specifier*. This combination presu- mably ensures the agreement of heads and specifiers (since they are go- verned in common). Note that this doesn't extend to complements: within a phrasal category B the lexical head and the specifier are theta-governed and L-marked, respectively, by the governor of B but the complement is naturally governed by the lexical head of B itself. The L-marking relation can be viewed as assigning an abstract, arbitrary feature '+L' to the L-marked constituent; all constituents are assumed to be marked -L unless explicitly L-marked. V. Blocking Category A maximal projection C is a blocking category (BC) for any of its daugh- ters if it is -L, i.e., not L-marked by anything. Note that a blocking category (likewise a barrier, cf. below) is defined relative to a speci- fic set of constituents of which it is not itself a member; the fact that a VP is (typically) a blocking category for its daughters has nothing to say about the subject (once the subject has moved to the Spec of Infl or Agr position, if you buy the Internal-Subject Hypothesis). VI. Barrier Here's where it gets complicated. If a maximal projection C *immediate- ly* dominates another maximal projection D, D being -L (and therefore a blocking category for all its own daughters), than C is a *barrier* for the daughters of D, *though not necessarily for D itself*. D itself, the blocking category, may also be a barrier for its daughters *if it is not S = IP/AgrP*. As Chomsky puts it, S (or whatever revised label one uses for it) cannot be an 'inherent' barrier, only a barrier 'by inheritance', i.e., only if another blocking category intervenes between it and the re- levant constituent. The 'defectiveness' of S, its inability to be a barrier except by inheri- tance, is one of the complications of the hypothesis that many find irri- tating. It is empirically based: there is plenty of evidence to indicate that, in particular, the Spec position of S (i.e., the SS position of the subject) is vulnerable to government from outside S. For instance, in the following sentences 'him' is in the subject position of non-finite clauses. As shall become clear when we discuss Case-Theory, such a sub- ject position has no proper governor within the clause and therefore must be governed from outside, from the word that governs S itself. In (1a- b), these are respectively the verb 'believe' and the complementizer 'for'. (1) a. I believe [him to be happy](S). b. I prefer very much [for [him to leave first](S)](CP). As we shall see further in discussing Binding Theory, (2) also provides evidence that the subject position of a non-finite clause must be gover- nable from outside. The trace in subject position of the non-finite clause must be governed by its antecedent in the subject position of the matrix. (2) John(i) is believed [t(i) to be happy](S). Finite clauses also appear to be defective; the Case Theory evidence doesn't apply to them, since a finite Infl is quite capable of governing its specifier for Case-Theoretical purposes, but the binding facts are the same as for (2). In (3a) we have a trace in subject position in a finite subordinate clause, necessarily bindable from the matrix. Simi- larly, in (3b) a trace in adjunct position in a finite subordinate clause bound from the matrix. (3) a. Who(i) do you think [t(i) left](IP)? b. When(i) did [he leave t(i)](IP)? That's as much as i'm going to say here about the definition of a barrier and the defectiveness of S implicit in that definition, though the sub- ject is worthy of quite a bit of further discussion; there have been se- veral attempts to simplify the definition of a barrier by either demon- strating that S isn't really a defective category or to make its defec- tiveness dependent on some more basic consideration. Perhaps we can dis- cuss these issues further. You will notice that these arguments depend heavily on concerns of Case and Binding Theory. This is characteristic of the fundamental nature of Government Theory as it developed during the 80's; many of the defini- tions and constraints of other modules of the framework depend crucially on it and are therefore related to it and to each other. Well, now that i've reached the definition of 'barrier' it's time for the full, formal definition PPA definition of a proper government relation. This is still a two-step operation, i'm afraid. First, we have to esta- blish what is a possible 'potential governor'. Any one of the major 'lex- ical' categories (i.e., A, N, P, V) qualifies as a potential governor. So does Infl, or Agr, depending on which version of the framework one is wor- king in. Any Bar-0 constituent belonging to one of these categories is a potential governor. For a given constituent B, another potential governor is another constituent co-indexed, i.e., syntactically defined as co-refe- rential, with it. The distinction between a Bar-0 head (typically dis- tinct from the 'governee' in reference) and a co-indexed constituent, of whatever Bar-level, as a governor is relevant to various areas of the framework, especially Binding Theory, but not to general Government Theo- ry. For the discussion here, all of these are potential governors and the constraints and definitions of Government Theory apply to all of them equally. In order for a potential governor A to be the actual governor of a con- stituent B the following conditions must be met: (a) A must c-command B. In practice, this usually works out actually to an m-command condition, and is occasionally even stated as such. (b) There must be no barrier intervening between A and B. That is, if B is contained within a constituent C that qualifies as a barrier for its daughters but A is outside C, then A cannot govern B. Note one little complication here: If A is adjoined to C, then there is one instance of C that dominates A and another that does not, as in the digram below, then C may be a barrier but for the purposes of Govern- ment Theory A is inside it. In technical jargon, C does not *exclude* A. C / \ A C / \ / \ /_____\ B The last condition in Government Theory is the so-called 'Minimality Con- dition', one of the grounds -- but only one -- for the recent label for the whole framework as the 'Minimalist Program'. In order to qualify as the governor of B, A must be its *minimal* governor, that is, there must be no other constituent Z that itself meets all of the above conditions for a proper governor for B but which is closer to B than A is, i.e., is c-commanded by A. (That's the way the condition is usually put. As such, it ignores the possibility that A and Z are sisters and c-command each other, in which case they are in structural terms equally 'close' to B. Kayne's Binary Branching Hypothesis would prevent such a situation from ever arising, since under this hypothesis it would be impossible for two potential go- vernors to be in a mutual c-command relation while also c-commanding the governee as required by condition (a).) Obviously, the main point of the Minimality Condition is to ensure that for any given constituent there is at most one and only one governor. Re- member what was said above about two distinct kinds of potential gover- nors; it is theoretically possible for a constituent to have both kinds, a governing head and a c-commanding antecedent, and thus to be simultane- ously head- and antecedent-governed. The strict wording of the Minimali- ty Condition rules this out, but as i have argued in a recent paper of mine it may actually be attested in some languages. Best, Steven --------------------- Dr. Steven Schaufele 712 West Washington Urbana, IL 61801 217-344-8240 fcosws@prairienet.org **** O syntagmata linguarum liberemini humanarum! *** *** Nihil vestris privari nisi obicibus potestis! ***