LECTURETTE #21: TYPOLOGY OF EMPTY CATEGORIES (II) In Lecturette #20 i introduced three of the four types of empty catego- ries defined by REST's Binding Theory. Now for the Biggee, the one i'm sure many of you have been waiting for. According to all i've been say- ing, there ought in principle to be such a thing as a constituent that is both an anaphor and a pronominal. What would such a thing be like? It's called PRO. That's `big PRO', to distinguish it from `little pro', which is merely a garden-variety pronominal. {Notice to anybody out there who may be acquainted with LFG: REST's PRO is not to be confused with LFG's PRO. That both are typically represented all in caps is coinci- dence; at least, the capitalization is independently motivated in each case. LFG's PRO is really more like REST's pro, an ordinary pronominal that happens to have no phonological manifestation.} PRO is by definition both an anaphor and a pronominal. As an ananphor, it is subject to Prin- ciple A of Binding Theory, obligating it to be bound in its governing cat- egory. As a pronominal, it is subject to Principle B, obligating it to be free in its governing category. REST's way out of this dilemma is to assert that PRO is inherently ungoverned and ungovernable, and therefore has no governing category. This is known as the `PRO Theorem.' (The PRO Theorem has occasionally been challenged in the literature. The example best known to me is a 1991 paper by Halld'or 'Armann Sigurdsson, `Icelandic Case-Marked PRO and the Licensing of Lexical Arguments' (Natu- ral Language and Linguistic Theory 9:327-363). By and large, however, the PRO Theorem is taken as unchallengeable orthodoxy within REST, as firmly grounded as anything in Euclid; in practice, PRO is regarded as *by definition* ungoverned.) In effect, what this means with reference to reference is that PRO may or may not have an antecedent; it has no govening category, so any attempt to determine whether its antecedent is inside or outside its governing category collapses, never mind identifying an antecedent at all. Consi- der the sentences in (1). In the first, we have the `generic PRO'; PRO has no syntactic antecedent and is assumed to refer generally and indes- criminately to any member of an appropriate set of entities (in this case, presumably the entire human population). In the second, PRO is understood to be coreferential with the first-person subject of the matrix. (1) a. It is important [PRO to eat a good breakfast]. b. I want [PRO to eat a good breakfast]. According to the modular organization of REST outlined in Lecturette #7, the question of defining the antecedent of PRO is the province of Control Theory, about which i'm afraid i will have virtually nothing to say since i have seen very little literature on it. (If anybody out there has some good references, let me know and i'll spread the word!) But for all in- tents and purposes PRO behaves in fact like a pronominal, in that it may but need not have an overtly expressed syntactic antecedent; the main difference is that PRO is ungoverned. Since Case is assigned/checked under government and PRO is ungoverned, PRO never bears Case. The only way it can evade the Case Filter is by haning no phonological manifestation. (Remember, the Case Filter expli- citly refers only to NPs with phonological content -- and now you know why!) Thus, the fact that PRO is an empty category is a corollary of the PRO Theorem. And this is why, as noted in Lecturette #19, although Bind- ing Theory defines four different types of categories on the basis of the two binary features +/- anaphor and +/- pronominal, only three are ever manifest on the surface: no phonologically overt constituent can ever be both +anaphor and +pronominal. Being inherently ungovernable, PRO is restricted to ungoverned positions. The archetypical such position is the subject of infinitive clauses, since infinitives are held (in REST at least) not to govern their sub- jects. Overt subjects of infinitives have to get Case by rising to the position of the object of the matrix verb; that's how `John' in (2) mana- ges to get past the Case Filter. But PRO can remain in its DS-position, since it doesn't need Case and doesn't need to be governed, in fact can- not be. (2) [I want John(i) [t(i) to drink this wine]]. It is of course a matter of definition that all empty categories are un- pronounced, but it's not strictly correct to conclude therefrom that they are necessarily totally devoid of phonological manifestation. You may have heard of the `_wanna_ contraction'. There are certain verbs in Eng- lish which take infinitive complements and can, at least in a colloquial/ informal register, optionally undergo a phonological contraction with the infinitive auxiliary/marker `to'. The paradigmatic example is the verb `want', which under these circumstances can become `wanna', as in (3). Another example is `supposed', which can result in `sposta' as in (4). (3) a. Do you want to watch TV? b. Do you wanna watch TV? (4) a. Are you supposed to visit your aunt? b. Are you sposta visit your aunt? Certain empty categories in the subject position of the infitive clause have no noticeable affect on this contraction. PRO, the normal subject of infinitive clauses, does not block _wanna_ contraction, and neither does the ordinary anaphoric trace, in the (relatively rare) circumstances in which it appears in the subject position of an infinitive. Cf. (5-6), respectively. (5) a. What do yo want PRO to watch? b. What do you wanna watch? (6) a. [These papers](i) want t(i) to be finished t(i) by tomorrow. b. These papers wanna be finished by tomorrow. {Be it understood that in (6a) the NP `these papers' is base-generated as the object of `finish', then moved (pardon the procedural-metaphorical language) to subject position thereof because of the passive voice of the clause, and from there to the subject position of the matrix, leaving tra- ces in both the subject and object positions of the lower clause. It's the trace in subject position that's at issue here.} But a variable in the subject position of an infinitive complement does block _wanna_ contraction, as in (7). (The _wanna_ contraction is possi- ble in (5) because the variable is in object position, not in subject po- sition.) An overt NP in the same position, of course, also blocks it; cf. (8). In this respect, the `empty R-expression' patterns with overt ones and not with other empty categories. (7) a. Who(i) do you want e(i) to drink this wine? b. *Who do you wanna drink this wine? (8) a. I want John to drink this wine. b. *I wanna John drink this wine. That being said, i have to admit that i'm not entirely convinced by this argument, since in my own usage both clauses in (9) are acceptable; at least, i find myself saying things like this in very informal contexts, although i admit i would very surprised to see them in print (of course, i'm seeing them in print right now, but i mean in a normal context, as opposed to the context of a discussion of grammatical theory). This may be partly due to `contamination' from the pattern in (10), involving ano- ther matrix verb, `go', which when it takes an infinitive clausal comple- ment (thereby functioning as a sort of future-tense auxiliary) can under- go something very like _wanna_ contraction, even when the subject of the lower clause is at least coreferential with a wh-operator as in (10b) -- and i have seen this sort of thing in print. (9) a. ?Who do you wanna drink this wine? b. ?Which wine is sposta be drunk? (10) a. Who(i) you gonna call e(i)? b. Who's gonna clean up this mess? But i suspect this is not strictly parallel to the _wanna_ contractions in (5-8). I think what we have here is a case of what is known as a *par- asitic gap*, in which the subject of `clean up' has not actually moved anywhere but is rather a syntactically independent empty category (presu- mably PRO) that is coreferential with (= `controlled by') the subject of the matrix. Thus, (10b) should probably be analyzed as in (11), in which the index on PRO indicates coreferentiality pure and simple, without any movement being implied. {Of course, that's all such indexing is supposed to indicate, but one gets so used to indices on traces that one gets into the habit of think- ing there's a direct connection between notational indexation and hypothe- tical movement. A facile assumption to watch out for.} (11) [Who(i) [e(i) is going [PRO(i) to clean up this mess](IP)](IP)](CP)? Be that as it may, i suspect that the acceptability of strings like those in (10) makes it a lot easier for me to accept the ones in (9), whatever may be the case for other native speakers of English. I offer this to you for what it's worth, and because you're liable to hear about the _wanna_ contraction from time to time, so you might as well know what it's about. That pretty much wraps up my survey of REST's Binding Theory. As i said in my recent general posting, i intend in Lecturette #22 to say a little something about Bounding Theory & Subjacency and maybe tie up some loose ends. Unless there's something about REST/PPA/Minimalism that anybody is particularly concerned about that hasn't been discussed yet, we'll be moving on soon to some alternative frameworks of syntactic theory. 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! ***