LECTURETTE #24 INTRODUCTION TO RELATIONAL GRAMMAR With this lecturette, i begin a survey of a generative framework of syntactic theory that is in many ways quite different from the one we have been surveying in the previous lecturettes. Relational Grammar (RG) is a framework associated primarily with David Perlmutter, who i thought was affiliated with the Univ. of Rochester but the most recent LSA directory claims is in San Diego, and secondarily with Paul M. Postal (CUNY), Carol Rosen of Cornell, William Davies of the University of Iowa, Barry Blake of La Trobe, David Johnson, for whom i can't seem to find an address anywhere, etc. {I recently heard that Postal is working more in a PPA-type orientation these days; and i have no idea what David Johnson is doing. Of Blake's and Rosen's continued interest in RG i have no doubt, to say nothing of Perlmutter himself.) A highly formalized version of RG, called `Arc-Pair Grammar' (APG), is described in the book of that name by Johnson & Postal, and we shall be touching on some of this as well in the coming run of lecturettes. I wouldn't recommend APG (the book) to anybody who isn't prepared for some pretty heavy technical prose. Other good general book-type resources on RG are Barry Blake's 1990 book Relational Grammar in the Croom Helm Linguistic Theory Guide series and the Studies in Relational Grammar series (3 volumes so far, published sporadically from 1983 to 1990 by University of Chicago Press). Back in Lecturette #2 i mentioned that REST assumes that all syntacti- cally relevant information is representable in constituent structure, in terms of dominance and precendence relations. This assumption, in the strict, absolute sense that it is held in REST, is not shared by any other framework of syntactic theory known to me. I think the one that comes closest to doing so is Generalized Phrase-Structure Grammar (GPSG). But even there, the `rules' that define P-markers have to be annotated for relevant semantic information that is, ipso facto, *not* represented in the P-markers. At the opposite extreme is Relational Grammar. I can't honestly say that RG denies the existence, or relevance, of dominance and precedence relations, or the suitability of tree-diagrams for describing them. But by and large RG and its practitioners regard such matters as of minimal interest. This attitude, i have to confess, has always created problems for me. On the one hand, i have always found RG a very attractive framework. On the other, a lot of my professional concerns have to do with issues of linear precedence, constituent order, that sort of thing, about which RG has little if anything to say. Very early in my graduate student days i came to the conclusion that RG was an `incomplete' theory, in that there are aspects of syntax that it does not and, at least as currently formulated, cannot address. This is not necessarily a serious flaw. Every major theoretical frame- work, as it has become elaborated, seems to have had to recognize cer- tain aspects of human linguistic behaviour that it cannot address. Chomsky has been saying for years that the formal grammar of any given language or language-user, left to itself, generates gobs of garbage, which has to be winnowed by independent modules of the human language faculty. Now when Chomsky says this what he seems to have in mind are semantic and, probably more importantly, pragmatic considerations, about which PPA has little to say. Other theoretical frameworks, encompassing different ranges of human linguistic skill, would naturally need to be `supplemented' by different `winnowing' elements. Just one of several reasons why i insist on balancing work in any one syntactic-theoretical framework with others, and with morphological, pragmatic, and semantic theory as well. RG was born in the early 1970s of the realization that Standard Theory, committed as it is to defining all relevant data in terms of P-markers, was missing certain cross-linguistic generalizations. In the discussion that follows, remember that we're talking about the version of ST that was current in the early `70s, which was basically the Aspects model. In this approach, it was assumed that an English passive clause like (1b) is derived from its active equivalent (1a) (or the DS rather transparently underlying it) by means of a transformational rule along the lines of (1c), which basically says, `Take the NP that follows the verb and put it before the verb, while the NP before the verb can be deleted or put after the verb, preceded by "by".' (1) a. Vivian broke the window. b. The window was broken (by Vivian). c. NP V NP 1 2 3 ---> 3 2 (by 1) For general purposes, this is a reasonably adequate description of the passive in *English*, and to some extent of other languages in the West European Sprachbund. Beginning ca. 1970, Perlmutter, Postal, and some others became increasingly bothered because it (1) didn't help describe languages typologically different from such Western-European languages, (2) seemed as a result to be missing a cross-linguistic generalization, and (3) thus seemed to be completely missing the point of what's really at issue in the active/passive dichotomy. After all, the `Passive Trans- formation', so to call it, isn't about NPs trading places or doing a do-si-do around the verb. What's really going on is, the *Object* becomes the *Subject*. The reordering of constituents in the English clause, or the alternative marking in the Latin example in (2), is, from RG's point of view, an epiphenomenon -- it's merely the outward manifes- tation of the change from object to subject. (2) a. Magister pueros laudat. `The teacher[NOM] praises the boys[ACC].' b. A magistro pueri laudantur. `The boys[NOM] are praised by the teacher[ABL].' {I'm putting this in the baldly simplest language possible. Further re- finements would specify the *Direct Object*, excluding any other kind of object. We shall discuss later the theoretical repercussions of the assumption that passivization involves something happening to the *ob- ject* rather than to the *subject*. It should also be emphasized -- this is potentially very important, at least from a theoretical point of view -- that i am here reporting the standard RG take on the issue. Some other frameworks, most notably LFG, would prefer to regard passivi- zation as primarily something that happens to *verbs, resulting* in differences in the ways in which NPs are marked, arranged, and/or inter- preted.} Note that the argument i have so far outlined takes as its springboard the notions, assumptions, and definitions found in the Aspects Model of ST, in which, at least to a first approximation the Direct Object in English is *defined* as the NP following the verb; RG was to a great extent motivated in the beginning by the difficulty (read *impossibili- ty*) of generalizing such a definition. Partly under the pressure of this critique, ST has evolved a more sophisticated perspective, whose continuing incompatibility with the RG approach is more subtle and shall be gone into shortly. In REST, subject and object are no longer in any way defined in terms of relative order but in terms of hierarchical rela- tions: the Direct Object is the NP that is sister to the verb within the VP, is directly theta-marked and Case-marked by the lexical verb, etc., while the subject is the `external argument', sister not of the lexical verb but of some projection thereof or of Infl. (Strictly speaking, at least in accordance with the Universal-Base Hypothesis the subject is the NP that is Spec of VP at DS and Spec of IP at SS, and is Case-marked by Infl while effectively receiving its theta-role from the VP, derived from the compositional semantics of verb and object, if any.) Within REST as currently practiced, the Passive Transformation involves the Object being *forced* into Subject position (Spec of IP) by the unavailability of Case in its DS position, and thus from this point of view as well could be described in terms of the Object becoming the Subject. In Standard, Chomskyan Theory we have transformations that rearrange constituents. In RG, the constituents themselves are for the most part left alone; instead the *relations between them* are *re-evaluated*. `Revaluation' is a technical term in RG, and represents the core of the framework, corresponding to ST's `Move Alpha'. Revaluations, or `relation-changing rules' as they are sometimes called, are the meat and drink of RG, and most of the research that goes on in this framework is concerned with identifying the constraints on the power of revaluations, constraints which are rather grandly referred to as `laws'. The difference in the ways REST and RG regard grammatical relations has sometimes subtle but ultimately serious repercussions for their respec- tive analytical approaches. To summarize a bit at the risk of oversim- plification, going back to the generation of the passive clause in (1b): Both frameworks agree that the (underlying) direct object becomes the subject. In RG, this is a matter of definition. In REST, saying `the direct object becomes the subject' is merely shorthand for `the NP that is base-generated as the completment of VP moves to the position of the Spec of IP (typically for Case-Theoretic reasons)'. In REST what is fundamental is the generation of a constituent structure that involves an antecendent-trace relationship; in RG, it's the revaluation of the under- lying direct object, however that revaluation may manifest in surface structure; there is no necessary connection, in RG, between the GRs `Subject' and `Direct Object', on the one hand, and the constituent- structural positions `Spec of IP' and `Complement of VP', respectively (in fact, in most if not all RG work such expressions as `Spec of IP' and `Complement of VP' are meaningless). Perhaps more perniciously, in REST (unlike the Aspects approach) the agent-phrase in (1b) is base- generated as a PP adjunct; RG's approach is closer to Aspects in this respect, in that it is essential to the theory that the agent-phrase be `base-generated' as a plain, garden-variety Subject. We shall discuss further later the repercussions of this. What are the relations between constituents that are affected by revalu- ations? At a first approximation, they are such notions of traditional grammar as `subject' and `object'. RG refers to these as `Grammatical Relations' (abbreviated `GRs', which conveniently is just the reverse of the abbreviation `RG'). They definitely do not include dominance and precedence relations; although Johnson & Postal felt compelled at one point in Arc-Pair Grammar to introduce a `Linear Precedence' Relation, RGans have by and large avoided this like the plague. When i was a grad student just beginning to get interested in comparative syntactic theo- ry, i decided to define a set of `syntactic relations' which would in- clude the dominance and precedence relations as well as the more tradi- tional `grammatical relations' that RG is concerned with; so all GRs are syntactic relations but not all syntactic relations are GRs. Within the RG framework, these GRs are posited as *primitives*. That means they are not defined within the theory. Different theoretical frameworks have different primitives; in Standard Theory and in GPSG, the primitives are dominance and precedence relations encoded in consti- tuent structure. In ST, although as noted above the definition is much more sophisticated then it was in the Aspects Model GRs continue to be defined in terms of constituent structure; in GPSG, they're defined in terms of (Montagovian) formal semantics. But in RG they're not defined at all; rather, other things are defined in terms of them. Being theoretical *primitives*, however, doesn't mean they're necessari- ly theoretically *simple*; it's assumed that each grammatical relation, although a formal primitive of the framework, has many `characteristics' or `qualities' that together reveal its nature. (One of my professors, in explaining this, would claim, `I have many qualities, and i'm very primitive.) The `laws' of RG spell out many diagnostics for GRs such as Subject or Direct Object. The practice is kind of like the old saying, `If it looks like a duck, walks like a duck, swims like a duck, quacks like a duck, it's a duck'; basically, in an RG analysis one can say `if it displays characteristics A, B, and C, it's a Subject'. But A, B, and C are not in any way a *definition* of `Subjecthood'. For one thing, Subjects are assumed to be the same in all languages. Their syntactic diagnostics A, B, and C -- the `symptoms', if you will, of Subjecthood, may be language-particular. For instance, in some languages the subject of a relative clause must be coreferential with the subject of the matrix, and in such languages such `relativization control', as it is called, would be a diagnostic for Subjecthood. But that wouldn't carry over to languages (e.g. English) which don't share that restriction. In the next lecturette i shall introduce the basic technical vocabulary of RG, and in #26 will address the question of whether and to what extent it can be described as a `transformational' framework, before getting down to the nitty-gritty of how RG analyses syntactic phenomena. 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! ***