LECTURETTE #4 NUMBER OF LEVELS In the Aspects model, a derivation could be indefinitely long. It was possible (or at least conceivable) for a derivation to consist of a sin- gle level, a clause being 'base-generated' in its surface form. Or it could consist of any number of levels arranged in a linear sequence, each one differing from its predecessor by exactly one movement-transforma- tion. And there was a clear sense that, just as in the reconstruction of earlier stages of a language's history it is possible to claim that cer- tain sound changes necessarily precede others, so in a given grammar cer- tain transformations might have to be ordered before others. Thus, in the Aspects model a sentence like (1) would be derived from a 'deep structure' roughly along the lines of (2) by means of a series of almost a dozen transformations, each of which would correspond to a new level and some of which (especially those relating to the 'lower' clau- ses) would have to precede others (relating to 'higher' clauses) in or- der to achieve the desired result. For instance, in order for Equi- Deletion to delete the NP 'Sam' in the lowest clause by identity with the object of the middle clause, the lowest clause would have had to be pas- sivized in order to get 'Sam' into the subject position from which it can be deleted. (1) The man who persuaded Sam to be examined by a specialist was fired. (2) {somebody} fired the man [the man persuaded Sam that [a specialist {should} examine Sam {by }]] {by } With the redefinition of all movement-transformations as a instantiations of the single, generic transformation Move Alpha, a lot of the motivation for these assumptions fell by the wayside. Does it make any sense to say 'First *this* alpha moves *here*, then *this* alpha moves *here*, etc.' when the transformational component itself is supposed to be unable to distinguish between one alpha, or one structural position, and another? As a result of various developments during the 70's, at least some of which we shall be discussing soon, the 'deep structure' represented by (2) above was replaced by something more along the lines of (3) as the 'deep structure' underlying (1). (3) {} was fired the man [who persuaded Sam [to be examined Sam by a specialist]] In which case, independent, general constraints would *require* the Passive transformation to take place in both the lowest and highest clauses. It would no longer matter whether Equi-Deletion took place before or after the lowest clause had been passivized; if the lowest clause isn't passivized, the whole derivation bombs, and if it is then Equi-Deletion would presumably also take place automatically. There is no longer any need to extrinsically impose an order on transformations; rather, constituents move around to satisfy general constraints and inso- far as their movement doesn't violate such constraints. And there is no longer any point to talking about the 'level' by which certain syntactic transformations have taken place but before certain others have. ('So what? It's still necessary for Passive to precede Equi, in order to set up the conditions that make Equi possible. How is this different from the Aspects approach?' i can hear some of you ask. At least i hope so. Yes, in actual fact in this situation Passive must precede Equi, just as it did in the earlier approach. The difference is that in REST you don't have to *stipulate* that it does. Whereas in the Aspects model you had to specify somewhere in the explicit grammar of English which you are trying to formulate that such-and-such transformations had to precede such-and-such other transformations, now it happens automatically. The point is not to make the actual generation of strings any simpler but to simplify the explicit grammar of a given language by reducing the number of specifications it has to make. To offer a computer-science analogy, the hardware can be as complicated as anyone likes; what we're after is a maximally simple software.) As a result of these developments, a new principle of derivational orga- nization was defined for the framework. Whereas in the Aspects model a derivation could be viewed as a linear sequence of any number of discrete levels, in REST a derivation is viewed in terms of precisely four dis- crete levels arranged in what has come to be called the 'T-Model' even though it is typically represented in the shape of an inverted 'Y': DS | SS / \ PF LF 'DS' is typically understood to stand for 'deep structure', 'PF' for 'phonological form', and 'LF' for 'logical form'. Contrary to some first impressions, 'SS' definitely does not stand for 'Surface Structure'. I have heard it interpreted as 'Shallow Structure', not a particularly satisfactory label since it invites the question, 'Just how shallow is "Shallow Structure"?' As will become clear shortly, a better interpre- tation of 'SS' is probably 'Syntactic Structure'. Note that in the non- linear organization of the 'T-Model' SS is central and linked directly to all the other levels. PF and LF are referred to as the 'interface' levels. PF represents the interface between syntax and phonology, the point at which phonological rules take over in order to get the string into a final, pronounceable form; it is the real equivalent to what in the Aspects model would be considered 'surface structure', the last level in the linear derivation. LF is the interface between syntax and certain semantic and pragmatic systems. DS can also be regarded as an interface level, but it's the interface between syntax and the lexicon. In the lext lecturette i will be concentrating on the nature and function of DS and LF, and the diffe- rences between them. vThe assumption in REST is that the base-generated form of a string is re- presented by DS. Various transformations operate on the DS-representa- tion to satisfy various constraints, etc. of the grammar, and the result is the representation at SS. PF and LF then read off of SS whatever they require to serve their interfacial functions. Note that Move Alpha is assumed to operate between any two levels. Between DS and SS, it operates pervasively, and nothing meaningful can be said about the order of its operations; as long as the cumulative result at SS of all transformations is grammatical, everything's fine. Move Alpha, possibly supplemented or qualified by minor operations, also ope- rates to derive PF and LF from SS. The difference in the character of movement-transformations between DS and SS on the one hand and that between SS and either of the interface levels on the other is due to the fact that different constraints operate at different levels. I'll going into this in more detail in later lecturettes, but just to give a few brief examples (for those acquainted with the relevant terms), the 'Theta-Criterion' is relevant at DS, the 'Case Filter' at SS and PF (in different ways), and the 'Empty Category Principle (ECP)' primarily at LF. Note also that in the 'T-Model' PF and LF are linked only through SS, not directly. 'Orthodox' REST insists on this, that everything that is rele- vant to both PF and LF must by definition be inherently syntactic and therefore true of the only purely syntactic level, SS, as well. Some arguments have, however, been presented to the effect that PF and LF ought to be directly related, on the grounds that certain phonological, or at least prosodic, phenomena in certain languages seem to have a di- rect relevance to semantic/pragmatic interpretation in spite of being irrelevant to syntax; i would direct you to Anthony Woodbury's 1987 paper 'Meaningful Phonological Processes' (Language 63:685-740). As was pointed out to me at a workshop last year, if PF and LF can interface directly with each other we may not need any distinct syntactic levels (DS and SS) at all. To which i responded that this is precisely what some non-transformational and functionalist frameworks claim. The principal new development in this area in the 'Minimalist Program' is a tendency to break down the distinction between DS and SS. There is an increasing sense that a distinct interface between syntax *sensu stricto* and the lexicon is unnecessary and that syntactic processes and lexical insertion just take place helter-skelter until a point is reached from which both PF and LF can be derived. The joint derivability of PF and LF is what is crucial here; if this double operation is sucessful the deri- vation is said to 'converge', and the string is both grammatical and interpretable. If it fails it is said to 'crash' and we're left with ungeneratable garbage. 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! ***