SO WHAT IS LF GOOD FOR, ANYWAY? a bit of discussion A couple of our participants have asked me to elaborate a little further on my comments at the end of lecturette #5: > The whole notion of a distinctive LF-representation in REST is based > crucially on the underlying assumption, mentioned back in lecturette > #2, that ALL relevant relations, including those of semantic scope, > are properly represented in terms of constituent structure. Another > important, 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- > representation, 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. Many of you have probably noticed my scepticism about the assumption that all syntactically-relevant information can and should be representable in terms of constituent structure (some of you i have talked to personally about this). My scepticism about the assumption of the universality of LF in general and LF-representations in particular is due in large part to the fact that much of the discussion in the literature of LF, LF-represen- tations, and how they need to be derived from SS, etc. seems at least to me to bear a more than passing resemblance to 17th-century attempts to develop 'philosophical languages', i.e., forms of written or, in some cases, spoken communication that would provide direct, objective represen- tations of reality without any ethnic or cultural overlay and which would therefore be equally valid for all people everywhere. Of course, these attempts merely objectified or hypostasized philosophical and semantic notions current in Western European academic circles of the time. Such attempts are still with us, and after all never failed completely. Their spirit lives and thrives in much of the communcation methods taken for granted in many of the 'hard' sciences. I'm thinking now not only of rich mathematical formalisms such as those commonly in use among physi- cists but also of the whole edifice of Linnaean terminology in biology (i remember being told that the word 'ladybug' has rather restricted cross- cultural value, but that 'Adalia bipunctata' means the same thing all over the world). (And not because it's Latin, either, but because the international commu- nity of biologists has covenanted that it should be so. As far as i know the 17th-century exponents of 'philosophical languages' -- at least the more perspicacious of them -- admitted that their proposals, in order to succeed, would also depend upon convention, just like natural languages do. LF, however, including at least the general outline of its structure -- e.g., the fact that a scope-bearing element like a negative or a quan- tifier at a certain position in the tree takes scope over everything lower in the tree -- is supposed to be part of Universal Grammar, which means that it is in some sense 'hard-wired' into our brains by ultimately genetic means. The theory seeks precisely to avoid any possibility of conventional adaptation of LF, at least partly because LF is a 'covert' level, with no direct surface representation. Since the typical human child has no direct access to the LF-representations of the sentences hann hears yet is able to interpret them consistently in ways that pro- vide a reasonable facsimile to the interpretations intended, the theory deduces the conclusion that there must be some universal internal human 'language' -- sometimes referred to as 'mentalese' -- into which senten- ces of actual human languages can with a reasonable expenditure of effort be easily translated.) But these forms of 'universal communication' remain successful precisely because they are deliberately limited to narrow ranges of intellectual enterprise. Matrix theory may be great for quantum mechanics, but it's not very useful for ordering pizzas or discussing the weather or explai- ning exactly what your spouse or parents or children or home or friends or pets or vegetable garden or the starry vault of heaven mean to you. I can translate the Gospels and other major Christian documents into Sanskrit, and have done some of that sort of thing, but i certainly wouldn't want to try to translate them into 'scientific language', nor would i try to do the same to Eliot (George or TS), either. The 17th- century attempts to develop a 'philosophical language' that could convey all that natural human languages convey but in a universally agreed-upon form foundered on the inability of such a formal, analytical mode to get a handle on all the rich variety and nuances of meaning that we are ac- customed to in our natural languages. (Similar problems nowadays seem to plague current approaches to lexical semantics. At the Mid-America Linguistics Conference recently someone was speaking about the attempts by such theoreticians as Ray Jackendoff to develop a strict, compositional semantics by decomposing the meaning of any given verb into a set of semantic atoms belonging to a restricted set of logical functions. These formal definitions may be able to cover gross meaning differences between verbs, but have so far, according to the speaker, proven incapable of describing the difference between the verbs 'break, fracture, sever, shatter' or 'rip, tear'. And, he went on to report, cross-linguistic organizations of semantic space seem not to fit very well into this kind of universal, 'philosophical' approach.) Just to give one example of the kinds of problems i see in an excessively strong hypothesis of the universality of semantic representations, Eng- lish uses ordinary, straightforward transitive sentences to express mea- nings which in other languages are expressed by means of what seem at least to native English-speakers like more complex constructions. For instance, the verb 'like' in English is a simple transitive verb, taking a normal, nominative-case experiencer-subject as in (1a). But other lan- guages typically express the same meaning by means of a verb meaning li- terally 'please' and assigning dative case to the experiencer, as in (1b-c). (Furthermore, in some languages, such as Hindi, this dative experiencer nevertheless functions in certain respects as the subject, that is, some syntactic tests that are language-internally sensitive to subjecthood pick out this NP as opposed to the nominative-case one.) The fact that all the sentences in (1) mean the same thing suggests that at LF they all ought to have the same representation, apart from the lan- guage-particular vocabulary distinctions. But i really don't know how equivalent LF-representations can be derived from the obviously struc- turally distinct SS-representations. (1) a. I like mangoes. (English) b. Les mangues me plaisent. (French) 'mangoes'-nom. 'me'-dat.cl. 'please'-3pl.pres. c. Mujh-ko aam pasand h~e. (Hindi) 'me'-dat. 'mango(es)' 'pleasing' 'are'-3pl.pres. Some versions of LF-theory, perhaps recognizing this difficulty, likewise restrict themselves to a very limited range of issues. Such as scope, the relative ranking of quantifiers, negatives, etc. vis-a-vis each other and other parts of a sentence. Or control, the ability of an element in one part of a sentence to be co-indexed with a syntactically independent element in a different part. According to some versions of LF-theory, it is relations like these, and these alone, whose representations should be formally universal and derivable somehow from SS, that need to be encoded at LF. Such a restricted version of LF-theory i can deal with and at least provisionally accept. But there are linguists who insist that LF has to be able to cover all semantics and pragmatics, and i for one feel that we know much too little about how the human mind actually represents reality (both objective, natural reality, and the social realities we deal with from day to day), and to what extent that representation is culturally conditioned versus to what extent it is the same for all hu- mans, to be able to say with any confidence what the sentences of natural human languages get turned into in order to be interpreted. The best we seem to be able to do, it seems to me, is to take a sentence of a given natural language and convert it into a different sort of thing which is still in a sense a sentence in the same or another natural language but which has been organized in such a way that language scientists, i.e., we, can discuss it scientifically. I will admit that LF-theory is not one of my strong points. I've slogged through one of the standard book-length treatments of it, Robert May's Logical Form: its Structure and Derivation (Linguistic Inquiry monograph #12, MIT Press), without getting much out of it. Does anybody out there have a better handle on this stuff? Could we get some discussion going here on the value of LF as a distinct level of analysis? 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! ***