LECTURETTE #22 SUBJACENCY AND BOUNDING THEORY This Lecturette is mostly historical in nature; as shall become clear, more recent versions of REST strive to include much of what used to be discussed under `Bounding Theory' under Government. Such subsumption would be, of course, a Good Thing. First of all, it is heuristically desirable to minimize the primitives and the `operational apparatus', so to call it, of the theory. Secondly, the similarity of the words `Binding' and `Bounding', and more importantly the homonomy between the past (participle) form of `bind' and the root of `bound' has occasionally led to confusion. In case you're wondering why i'm bothering to devote a Lecturette to what is essentially an outmoded part of the theory, it's mainly because you're liable to come across allusions to it here and there in the litera- ture. Subjacency was a live concern from the mid 70's at least up to the mid 80's, and so there's quite a bit of valuable syntactic argumentation written during that period that takes it very seriously. In my opinion, it's important for students of linguistic theory to know something about the history of the theory, not just how it works nowadays. THE `BOUNDING PROBLEM' The history of Bounding Theory in REST begins with the apparent *un*- boundedness *in some cases* of e.g. wh-fronting, as in (1a). It is characteristic of such `garden-variety' examples that wh-fronting can effectively operate across any amount of embedding. However, it was noted in the mid 60's that once one wh-element has been moved out of a clause it became extremely difficult if not impossible to move another out of the same clausse, which was said to have become a `wh-island'; cf. (1b). (My equivocation is due to the fact that (1c), which as far as i can tell ought on structural grounds ot be just as bad as (1b), in fact seems much better, at least to me. I don't know at the moment what to make of this.) (1) a. [Which book](i) did you say Hilary thinks Terry should read e(i)? b. *[Which book](i) did you ask Hilary where(j) Terry bought e(i) e(j)? c. ?[Which book](i) did you say that Hilary asked where(j) Terry bought e(i) e(j)? THE TRANSFORMATIONAL `CYCLE' This `bounding problem' apparently had to do with degrees and types of embedding, which suggested that at least certain movement transformations were sensitive to such things. The conclusion was that each S node con- stituted a sort of boundary in derivational history: movement transforma- tions take place first in the lowest (most deeply embedded) S, then in the one immediately above it, etc. Thus, it was suggested that the deri- vation of a complex clause (one with one or more clauses embedded in it) proceeded by a series of `cycles', each one corresponding to one layer of embedding. In order to constrain the theory as much as possible, it was desirable to define *all* movement transformations as operating in this `cyclic' fashion. Apparent `long-distance' movement such as (1a) was redefined as *successive-cyclic* movement. Instead of the wh-phrase moving diretly from its DS-position in the lowest clause to the front of the highest, as (1a) implies, it is supposed to move at each cycle to the COMP position associated with the S node defining that cycle, as in (2). (This analy- sis assumes the `COMP-to-COMP Condition', basically an extension in spi- rit of the Structure-Preservation Constraint: Once a constituent is in COMP it can only move to a higher COMP.) (2) [(CP) [Which book](i) did you say [(CP) e(i) Hilary thinks [(CP) e(i) Terry should read e(i)]]]? (The fact that it's difficult to get an acceptable string with the wh-phrase in any of the intermediate positions in (2) is perhaps a problem for this analysis.) Supplemented by the Strict Cycle Condition, this analysis accounted successfully (given the theoretical assumptions then current) for the `bounding problem', the discrepancy between (1a) and (1b). The Strict Cycle Condition asserts that once a derivation has finished with a cer- tain cycle, i.e., a given S, and moved on to the next (higher) one it cannot thereafter return to further manipulate the daughters of that S; they are not longer `fair game'. In the DS underlying (1b), the lowest clause contains two wh-expressions, `where' and `which book'. In the first cycle of the (hypothetical) deri- vation of (1b), `where' moves to the COMP position associated with that clause. Prevented from accompanying it by the `Doubly-Filled COMP Fil- ter' (presumed valid for English, though research on several languages by several people, including myself, has shown that it is not universal), `which book' stays put. Once that cycle is over, the Strict Cycle Condi- tion prevents any subsequent cycle from `reaching down into' the lowest S in order to move that phrase. As a result, `which book' becomes effect- ively immovable. QED. Reversing the order in which the two wh-expressions are moved doesn't provide a way out. Obviously `which book' can move into the lowest COMP node and then out of it (by successive-cyclic movement) before `where' can move into it. But once `which book' has moved on to higher COMP nodes the Strict Cycle Condition prevents `where' from moving at all. BOUNDING NODES You will have noted that this approach to the `bounding problem' presuppo- ses the kind of `derivational history' that can be described in (pseudo-) chronological detail: *First* this movement takes place, *then* this one, *followed* by this one, an assumption that in more recent versions of the theory is, to put it mildly, highly undesirable. But early on it was no- ticed that an alternative formulation was possible which didn't depend on such an assumption. This is the `Subjacency Condition' properly so called, which seeks to constrain the power of Move Alpha by reference to `bounding nodes' (you are about to find out why earlier, in discussing the notion of the transformational cycle, i spoke of each S node as con- stituting `a *sort of* boundary'). The Subjacency Condition states that no syntactic `rule' (e.g., movement transformation) can directly relate two positions that are separated by *two* bounding nodes. Why two bounding nodes? This is an empirical question. What it boils down to is that a lot of well-founded transformations necessarily `trans- gress', so to speak, single bounding nodes, but if we assume that that's the limit allowable we can account for the non-generation of strings like (1b). Movement of, e.g., a wh-expression like `where' or `which book' out of its DS-position to the COMP node closest to that position involves the `transgression' of a single S. From there, successive-cyclic move- ment from COMP to COMP transgresses, again, a single S node at each `cyc- lic' stage. However, as can be seen from the figure in (3), any attempt to move something from the lowest S directly to one of the higher COMP nodes, as would be required to generate (1b), would involve the transgres- sion of at least 2 S nodes, and would therefore be prohibited by the Sub- jacency Condition is we recognize S as a bounding node. (3) CP / \ COMP S /|\ /_\ | CP | / \ |__COMP S | /|\ /_\ | | CP | | / \ | |__COMP S | /|\ /_\ |__ X ___|___| PARAMETERIZATION OF BOUNDING NODES This condition constituted the kernel of Bounding Theory; all that was left was the definition of what constituted a `bounding node'. Work done in the mid 70's encouraged the assumption that NP is a bounding node; this would permit the Sentential Subject Constraint and the Complex NP Constraint to be accounted for by Bounding Theory, witness (4). (4) a. *Whom(i) did [(S) [(NP) my letter to e(i)] get lost]? b. *[Which book](i) did [(S) Evan visit [(NP) the store e(i) [(S) that had e(i) in stock]]]? Luigi Rizzi argued in the late 70's that, while NP is probably a bounding node in all languages, S is not; specifically, in Italian CP is a bound- ing node but S isn't. This would account for the fact that (5a) (in which 2 S nodes but only one CP node is transgressed) is acceptable in Italian but its equivalent isn't in English, but (6) (involving as it does the transgression of multiple S- and CP-nodes) is unacceptable in both languages. The general conclusion from this was that languages could choose between S and CP as a bounding node. (5) Tuo fratello, [(CP) [a cui](i) [(S) mi domando [(CP) che storie [(S) abbiano raccontato e(i)]]]], era molto preoccupato. *Your brother, to whom i wonder what stories they have told, was very worried. (6) *Francesco, [(CP) che(i) [(S) non immagino [(CP) [quanta gente](j) [(S) e(j) sappia [(CP) dove(k) [(S) hanno mandato e(i) e(k)]]]]]], ... *Francis, whom(i) i can't imagine how many people know where(k) they have sent e(i) e(k), ... SO WHAT OF IT? You may have noticed that the list of possible `bounding nodes' (NP, either S or CP) shows a strong parallel to the list of `governing cate- gories' in Binding Theory. This and other parallels between Bounding Theory and other modules nagged at the fringes of REST until the mid 80's, when the notion of `barriers' was developed within Government Theory (cf. Lecturette #15). (In fact, in his introduction to his mono- graph on barriers Chomsky explicitly mentions a desire to incorporate subjacency and Bounding Theory into Government Theory, or at least to get the relevant constraints, etc., in both defined in common terms, as an important motivation for the work he was reporting therein.) And, in fact, it can be seen that a lot of what had previously been regarded as subjacency violations involve failure of antecedent government across a barrier. The theoretical problem with this, of course, is that most of the work on which Bounding Theory is based has to do with wh-movement. Remember that a `wh-trace' is not the same as an `NP-trace', in particular that it is not an anaphor and therefore according to the most traditional version of Binding Theory does not need to be licensed by antecedent government. This being the case, in theory the inability of `which book' to govern its trace in (1b) does not, in and of itself, interfere with the genera- tion of such a clause. As i have noted previously, there have been attempts to claim that, regardless of Binding Theory constraints, Govern- ment Theory requires all traces, anaphoric or not, to be antecendent- governed. However, some research, including some of my own, seems to indicate that this can't be the case. So, although Bounding Theory is not a particularly live issue these days (and, as i noted at the begin- ning of this lecturette, this is on the whole desirable), it may end up having to be resurrected in some form. 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! ***