LECTURETTE #5 DS AND LF In the Aspects model it was assumed that every aspect of a sentence's meaning -- its entire semantic content -- was directly and exclusively related to its 'deep structure', i.e., the initial level in the senten- ce's derivation. This assumption implied (perhaps in fact presupposed) a cognitive hypothesis that a speaker/writer starts out with a meaning hann wishes to express, generates a deep structure that represents that mea- ning, then performs certain (optional or obligatory) transformations upon that deep structure resulting in the surface sentence that hann pronoun- ces or writes; meanwhile, the listener/reader receives the surface form of the sentence and performs the inverse of the relevant transformations on it in order to recover (or, we might say, 'decode') the abstract deep structure on the basis of which the sentence can be interpreted. Work done during the '70's, however, indicated that it was not feasible to associate the meaning of a sentence exclusively with a single deriva- tional level. The relevant evidence concerns ambiguities -- sentences whose unitary surface manifestations mask two or more plausible, but incompatible, interpretations. Some ambiguities can be handled adequately within the Aspects model. For instance, a sentence like (1) is susceptible to two possible interpreta- tions. But each interpretation can be represented by a distinct deep structure as in (2). The sentence in (1) could be derived either from (2a) fairly directly or from (2b) through extraposition of the non-finite VP. (1) The fox saw a goose swimming in the river. (2)a. [The fox [saw [a goose swimming in the river]]] b. [[The fox [swimming in the river]] saw a goose] (The fact -- at least, i believe it's a fact -- that cats are constitu- tionally averse to swimming would explain the relative lack of ambiguity of (1') on the basis of the relative implausibility of the equivalent of (2b), (2'b).) (1') The cat saw a goose swimming in the river. (2')a. [The cat [saw [a goose swimming in the river]]] b. ??[[The cat [swimming in the river]] saw a goose] Thus the ambiguity of (1) would be explained in the Aspects model as due to its being a plausible surface structure corresponding to two distinct deep structures. (Incidentally, this hypothesis would also account for the common-sense experiential evidence that the speaker of a sentence like (1) typically knows exactly what hann means, and is often unaware of the ambiguity, while the hearer is confused. This follows directly if you assume that the cognitive process inside the speaker's mind follows the logical sequence implied above and outlined below: intended meaning --> deep structure --> surface structure Already knowing what hann means, the speaker generates a unique deep structure to represent it, and the alternative deep structure has no place in the process. The hearer, on the other hand, receives the sur- face structure and recognizes that it may plausibly be derived from ei- ther of two plausible but not synonymous deep structures, and is there- fore much more aware of the ambiguity than the speaker.) The ability of this sort of ambiguity to be explained within the Aspects model (and, as far as i know, such ambiguities are still accounted for in REST in similar fashion) is due to its having to do with a basic ambigui- ty in the relative 'deep' grammatical relations between the three consti- tuents 'fox', 'goose', and 'swimming in the river' (specifically, which nominal is subject of the non-finite VP). Since deep structure is sup- posed in Standard Theory to be precisely where such relations are enco- ded, obviously the semantic ambiguity of the surface structure must be related to a deep-structure conflict. But there are other sorts of ambiguities, that do not involve alternative assignments of grammatical relations amongst the major constituents of a clause but rather alternative 'rankings' of quantifiers and quantifier- like expressions. At this point i need to introduce an axiom of REST that is convenient in dialoguing with philosophers and cognitive scientists, but as far as i know is totally unmotivated in terms of linguistics per se. The meaning of a sentence such as (3) is conventionally represented in predicate logic as a string that translates roughly as (4). (As some of you know, in real predicate logic notation 'For all' is represented by an upside- down 'A' which i can't represent here, and 'such that' is unnecessary.) (3) Terry loves everybody. (4) For all x such that x is human, (loves (Terry, x)). For convenience' sake, it is assumed that at some level the semantics of sentences including quantifiers like 'everybody' are actually represented *linguistically* in a form that approximates this predicate-logic repre- sentation. For obvious reasons, this level is called 'Logical Form' (LF). So at LF presumably (3) should look something like (5). This de- finitely implies a difference between LF and DS, since presumably the DS- representation of (3) is something like (6). (5) [everybody(i) [loves(j) [Terry e(j) e(i)]]] (6) [Terry loves everybody] But things get hairier. What about a sentence like (7)? (7) is inhe- rently ambiguous. It is susceptible to either of the interpretations in (8). (7) Everybody loves somebody. (8)a. There exists some person y such that for every person x, x loves y. b. For every person x, there exists some person y such that x loves y. Perhaps the difference is clearer if we put it this way. Interpretation (8b) is true if for every person you can identify some person whom that person loves; doesn't have to be the same person. Greg loves Madeleine, Melissa loves Frederick, Ralph loves Larry, Jennifer loves Morgan, Bob and Carol both love Ted and Alice, and Zachary loves himself. As long as you can set up pairs like this for every single person the statement is true. But interpretation (8a) asserts that there is some special person -- call hann Capra -- whom everybody loves: Greg loves Capra, Madeleine loves Capra, Melissa loves Capra, etc. According to the theory of LF, these two interpretations crucially differ in the hierarchical relationship of the relevant constituents in the LF representation of (7). Interpretations (8a-b) are represented at LF by the structures outlined in (9a-b) respectively (note that these are transparent translations of the predicate-logic representations outlined in (8) into tree structures). That is, for interpretation (8a) 'somebody' has to be higher in the tree than (and in technical jargon we shall get to a little later must *c-command*) 'everybody'; we say '"somebody" has scope over "everybody"'. For interpretation (8b), the opposite must be the case. (9)a. [somebody(i) [everybody(j) [loves(k) [ e(j) e(k) e(i) ]]]] b. [everybody(i) [somebody(j) [loves(k) [ e(i) e(k) e(j) ]]]] Bear in mind, both interpretations of (7) are possible. This means there is nothing in the surface form of (7) that prevents the derivation of either. But the interpretations outlined in (8) and diagrammed in (9) don't have anything to do with grammatical relations in the traditional sense or the verb's valency or anything of that sort, and therefore can't be represented in any way at DS either. As should be clear from the representations in (5) and (9), LF is derived ultimately from DS by Move Alpha; DS is supposed to be the pristine original form of the sen- tence without any movement-transformations. So clearly there must be a distinction between DS and LF. The upshot of all this is that in REST there are two levels that together provide the information necessary for the semantic interpretation of a sentence. DS is a representation of *thematic* relations (indeed, as will become clear later on, in 'orthodox' REST that is *all* DS is). If the main verb of the sentence has certain requirements regarding what sort of object it takes, or whether it takes an object at all, it's at DS that those requirements are crucially met. The DS representation basi- cally says, 'Here's a verb, here are its subcategorization requirements, they are hereby filled by these constituents.' LF, as discussed above, is a representation of *scope* relations; pragmatics enter in here as well (at least to the extent that pragmatics, e.g. discourse structure, can disambiguate sentences like (7) that are superficially ambiguous); that's why in lecturette #4 i said 'LF is the interface between syntax and certain semantic and pragmatic systems'. DS, on the other hand, has nothing to do with pragmatics; it is concerned solely with lexical seman- tics. Nota bene: The whole notion of a distinctive 'LF-representation' in REST is based crucially on the underlying assumption, mentioned back in lectu- rette #2, that ALL relevant relations, including those of semantic scope, are properly represented in terms of constituent structure. Another im- portant, but often tacit, assumption behind REST's LF-theory is that the structure of LF is universal. That is, the translation equivalents of (3) into every single human language will all have the same LF-represen- tation, in spite of the fact that different languages may in terms of grammar organize the relevant expressions differently. So REST requires every single language, every single grammar to be able ultimately to derive LF-representations of the form in (5) or (9) from any sentence in that language. Being sceptical about both of these assumptions, i am sceptical about a lot of REST's LF-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! ***