LECTURETTE #20 THE TYPOLOGY OF EMPTY CATEGORIES In Lecturette #19, i introduced the Principles of Binding Theory and the typology to which they refer, and mentioned that the importance of Bind- ing Theory lies mostly in its defining of the typology of empty catego- ries. You will remember that Binding Theory refers to a pair of binary features, +/- anaphor and +/- pronominal, that together define four pos- sible types of categories relevant to Binding Theory. Each of these four can be `empty', i.e., entirely lacking in phonological manifestation (indeed, as already hinted, one of them cannot be otherwise), although REST assumes they have syntactic reality and, as we shall see, some of them may have phonological repercussions. Now it is time to introduce these four types of empty categories, in the same order in which i in- troduced the more general terms for the four types in Lecturette #19. 1. +Anaphor, -Pronominal. The empty anaphor is commonly called a `trace' and is often represented by the letter `t'; you will sometimes come across the variant label `NP-trace', which is used to distinguish it from `wh-trace', about which more below; these terms were more common in the late 70's then they are now. As explained back in Lecturette #6, the assumption in REST is that when- ever any constituent moves from any position, it leaves behind a trace. UNLESS THERE IS GOOD REASON FOR THINKING OTHERWISE (cf. below), this trace is equivalent to the so-called `NP-trace', even if the constituent in question is actually, say, a verb (one of the reasons we are no longer so prone to calling them `NP-traces'); it's an empty anaphor. Hence, it is subject ot Principle A of Binding Theory, which imposes limits on how far its antecedent can move. Remember, Principle A says that an anaphor must be bound within its governing category; therefore, the trace must have an antecedent within said governing category, i.e., not separated from it by a barrier to government. Since the antecedent of a trace is the constituent whose movement away from the trace's current position is what created the trace in the first place, you can see that this stipu- lation constitutes a constraint on the operation of Move-alpha. A constituent can to some extent effectively get around this constraint by successive movement (not precisely the same as what used to be called `successive-cyclic' movement, about which more in Lecturette #21), in which the constituent moves gradually up the tree, each step of its path -- more correctly, each link in its chain -- within the proper distance of both its immediate predecessor and successor to meet the requirements of Binding Theory. Thus, in the tree diagrammed in (1), the constituent X has gradually climbed the tree; as long as each of the t's is within the governing category of the one below it, and X in its final position is within the governing category of the highest t, then each link in the chain is properly bound by the one above it. Note from this that it is possible for an anaphor to be an antecedent of another. (1) /\ X a / \ t(5) b / \ t(4) c / \ t(3) d / \ t(2) e / \ t(1) f 2. -Anaphor, +Pronominal. The empty pronominal, typically represented in text and diagrams by `pro' (note utter absence of capitals, cf. below) and called `little pro' or `small pro', is perhaps the easiest of all empty categories to get a handle on. It behaves in all respects like a personal pronoun -- it replaces a fully referential NP, it just can't be pronounced itself. Cross-linguistically, pro is assumed to be the subject of imperatives and imperative-like finite constructions such as those in (2). (2) a. Put that down! b. Gotta run! (assumed 1st-sing. subject) c.(Fr.) Tombera? Tombera pas? (context requires a reading with a 3rd-sing. subject -- very colloquial) (Many languages have imperative forms for persons other than the 2nd. French, for instance, has an imperative of the 1st plural -- witness the exhortation `Marchons!' in the Marseillaise. And Classical Sanskrit had a full conjugation of the imperative in all persons and numbers. As far as i know, languages like English which lack these forms typically express the same idea through subordination under a verb of permission: `May i ...', `Let's ...', `Qu'il ...', etc. In such constructions, the subject of the subordinate clause is typically expressed, even if as in the `let's' construction it's as an oblique object of the matrix.) What else can pro be? There is debate and, as far as i know, lack of consensus on whether optional object positions as in (3) can be occupied by pro. More importantly, there are languages (Italian and Spanish are the ones most commonly mentioned in the literature, but there are lots of others) in which overt subject pronouns are optional in all types of clauses; cf. (4). In REST it is typically assumed that in such clauses the subject position is occupied by pro. (3) a. Evan ate. b. You see/understand? (4) Cerco/Cerca/Cerchiamo/Cercate/Cercano un'libro. I am/(s)he is/we are/you are/they are looking for a book. (I have chosen Italian as representative of all such languages here because it is one i know reasonably well. Unfortunately, such a choice could misleadingly encourage one to suppose that it is richness of overt subject-agreement marking that makes it possible for a language to dis- pense with overt subject pronouns. It should be pointed out that Chinese and Japanese, languages with no subject-agreement marking whatsoever, also have this option.) These languages are typically referred to as `pro-drop' languages, which within this theoretical context has always struck me as a misnomer. After all, we're claiming that `pro' -- the empty pronominal -- is *present*, not absent, in the subject position in clauses like those in (4), that the grammars of Italian, Spanish, Japanese, etc. allow pro in the subject position of ordinary declarative and interrogative clauses (which e.g. English does not). It's not pro that's being dropped, rather it's an overt pronominal that is being dropped in favour of pro. A logically separate problem is the subject of weather verbs. These verbs have no theta roles to assign, even to their subjects, and on the basis of Theta Theory one can claim that at DS they therefore have no subjects at all. The fact that in English such verbs nevertheless have overt subjects as in (5a) is due to independent licensing principles of English grammar. In the `pro-drop' languages, to the best of my know- ledge, the subject position is always empty with weather verbs. Can one claim that in clauses such as those in (5b) that position is occupied by pro, as it is in (4)? Or is it really and truly absolutely, utterly emp- ty? As far as i know there is no consensus on this point, but personally i prefer the latter hypothesis. (5) a. It's raining/snowing/etc. b. Piove/neve/etc. 3. -Anaphor, +Pronominal I said that the empty pronominal was probably the easiest of the empty categories to get a handle on. Well, in many ways this is the hardest. It's sometimes called a `variable' because it's roughly equivalent to a variable in predicate logic (such as would be used in predicate-logic translations of sentences using it), and sometimes it's called a `wh- trace'. This last label is justified in part because it captures the fact that, like the empty anaphor mentioned in section 1. and unlike the empty pronominals, it typically arises through movement. The difference between the `variable' (or `wh-trace') and the `(NP-)trace' has to do with the details of movement (and their consequences for Binding Theory), *not* with the nature of the moved constituent (referred to consistently hereafter as the antecedent of the trace). Just as an `NP-trace' can result from the movement of something that isn't an NP, so a `wh-trace' can result from the movement of something that has no `wh-features' (i.e., doesn't involve any sort of relative or interrogative pronominal). What's critical is not the categorial character of the antecedent but its position, and, contrariwise, the position of the `trace'. The antecedent of an `NP-trace' is in an A-position. Head positions are A-positions, but of course even if they are empty they can only accept Bar-0 constituents. This is how a finite verb can move to Agr or some- thing like that and leave behind an anaphoric trace. Otherwise, an A- position is typically one to which either a theta-role or Case might be assigned. Typically, of course, it's the latter; an NP moves to an A- position in order to get Case (or have its Case properly checked, in the Minimalist version) and pass the Case Filter, leaving behind a trace (usually in a theta-position, which, therefore, is also an A-position). This is what is referred to as A-Movement; it involves *anaphoric* traces. The antecedent of a `wh-trace' is in an A-bar-position. This may be the Spec position of a head that lacks the power to assign a theta-role or check Case (e.g., Comp). Or it may be an adjunction site. Bear in mind that when adjunction takes place, the `host' position may actually be an A-position, but if so the theta-role or Case or whatever properties which accrue to that position by virtue of its being an A-position are the exclusive prerogative of the constituent that `belongs' there, not shared with the adjunct. (Please forgive all the legalese metaphors; this is the best language i've been able to find so far for explaining this stuff.) Remember what i said about chains back in Lecturette #16: each chain must include one theta-position and one Case position. Obviously, neither of these is an A-bar-position, and therefore the antecedent of a variable cannot occupy either. So what does? Typically, the variable itself occu- pies at least one of them. In fact, a corollary of Binding Theory says that `a trace (i.e., a -pronominal empty category) in a Case-marked posi- tion is a variable'. The archetypal case of A-bar-movement, of course, is wh-extraction as in (6). That's why these empty categories are traditionally called `wh- traces'. From here on, however, i expect i will refer to them as `varia- bles', because that's what i'm used to calling them, and infact in my own research i've dealt with them primarily in circumstances not involving wh-extraction. Outside the context of this lecturette, when i speak of `traces' you can pretty much assume that i'm referring to empty anaphors. (6) a. Who(i) do you thnk e(i) will win? b. Whom(i) did Sam give the monkey to e(i)? c. Where(i) did Morgan find the keys e(i)? d. Hilary showed it to the farmer whom(i) you had seen e(i) in the drugstore yesterday. >From the point of view of Binding Theory, the most important thing about variables is that, being neither anaphors nor pronominals, they are `sub- ject' to Principle C and are therefore completely free; as stated in the previous lecturette, in theory at least the syntax won't bother looking for antecedents for them. This is not quite true; or at least it mustn't be taken to mean that variables, like overt R-expressions, can exist with- out antecedents. There is a fundamental assumption, or stipulation if you like, underlying Binding Theory that -pronominal empty categories can only arise through movement. Thus if a variable exists in a syntactic construction it must by definition have an antecedent somewhere. That's why it's called a `variable', after all; it's bound by an `operator', to use the terminology that REST has borrowed from predicate logic. Seman- tically bound, that is; being -anaphoric, it doesn't need to be syntacti- cally bound by its antecedent. Remove the word `who' from (6a) above and you will get an unacceptable string; but it's unacceptable not because the variable has lost its antecedent but because, having no antecedent (whose movement could have created it), it isn't a variable at all but pro, and English doesn't allow pro in subject position in ordinary decla- rative clauses. Translate the same clause (without `who') into Italian and as far as i know it comes out fine. I think now is a good time to introduce the Empty Category Principle, which is really a corollary of the more basic Principles of Binding Theory but in analytical usage has tended to be elevated to the status of a principle itself. The ECP, as it is commonly abbreviated, applies only to `traces', i.e. to empty categories created by movement, which are by definition -pronominal. It says: A -pronominal empty category must be properly governed. There has been some debate and controversy over what constitutes `proper government'; the consensus is that a government relation must obtain between the trace and some other constituent, without a barrier interve- ning. The traditional definition (i.e., from the early 80's) is that there are two kinds of proper government or two ways the ECP can be met. One is `antecedent government', in which the other constituent is the trace's antecedent; this is simply a version of Principle A, and applies strictly to `NP-traces'. The other possibility is known as `theta-govern- ment', in which the other constituent is not the trace's antecedent but rather a head with the power of assigning theta-roles. This is what hap- pens to variables; whatever else may be said about them, they may be sepa- rated by quite a distance from their antecedents but usually sit within the government ambit of some lexical head. Some attempts have been made to eliminate the possibility of theta govern- ment from the ECP, reducing proper government to antecedent government. Such attempts have not struck me as convincing, and my own research has convinced me that the ECP needs to maintain the distinction between (ana- phoric) traces and variables, and that theta-government must remain an option for satisfying it. One more thing needs to be said about the ECP: it, and indeed all the Binding Theory principles, applies crucially at LF, though in some cases (languages? grammars?) it may also apply at SS. But given the fact that Binding Theory is to some extent concerned with the interface between syntax and semantics, as alluded to in the previous lecturette, it is reasonable that it should be an aspect of the derivational level that encodes that interface. This lecturette has gotten awfully long, and there's still a fair amount i feel i should say. So i'm going to cut it off here; Lecturette #21, with the rest of what i have to say on the typology of empty categories, should be following in another week or so. 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! ***