Robert Levine (levine@ling.ohio-state.edu)
Sat, 25 Jan 97 01:46:59 EST
I agree with Ivan that the argument from the binding theory is not conclusive, because we don't really know everything that we need to to say when condition C-type violations can and cannot occur, whether point of view or shifts in the discourse perspective or whatever are sufficient to account for such effects by themselves or whether these must interact in a specific way with certain syntactic conditions, etc. What I really wanted to focus on at the outset of this discussion is what seems to me the much more fundamental issue of what is entailed by a lexical rule treatment of adjuncts such as has been proposed by van Noord & Bouma, Rob Malouf, Ivan and various others alluded to in one of Gosse's postings. I'm thinking of the kinds of points Daniele raised in her posting and which the ACKMS group has stated in their recent work, about the fact that complementhood, i. e. selection by a head, should be carry certain empirical consequences with it. The crucial issue seems to me to be that when you have different valence specifications you have different lexical entries, and that where you have different lexical entries you have the possibility that somewhere along the line there will be differences in more than just valence features. That seems to me to be the core of the arguments for complementation as negation, and it's what I'd really like to see at the center of our discussion. The question that needs to be addressed is this: Is there any evidence, comparable to the subcategorization and cliticization evidence for negative adverbs from French, that verbs associated with extractable adverbials demand differences in their lexical entries from verbs unaccompanied by such adverbs? Putting it another way, is extractability sufficient to justify complement status? It's pretty clear, for example, that when we agree that a verb selects an adverb, it's because there are plenty of other indications besides the appearance of the adverb that we have a different lexical item from the one which does not select the adverb, viz. examples like (1)a. Robin reads slowly. b. This book reads slowly. (2)a. Robin reads to children slowly. b. *This book reads to children slowly. There are differences between the valence lists that are attendant on the selection, or not, of an adverbial complement in such cases, apart from the adverbial itself, as well as significant differences in the way the ARG-ST list is linked to the components of the CONT specification. What we are in effect forced to take on with the particular LR approach to adjunct at issue is the claim that the relation among the tokens of _reads_ is the same in all three examples in (3): (3)a. Robin reads. b. Robin reads slowly. c. The book reads slowly. That is, there are three different, independent lexical entries. The difference between (3)a and (3)b---where the only difference in the corresponding entries or in the arbitrarily large number of other separate entries that will be captured by the LR in question is the specification of the number of adjuncts that can in fact cooccur with the head---is just as great as the difference between (3)b and (3)c. It is, on this account, purely an *accident* that the only difference between the lexical entries corresponding to (3)a and b (or those required for a second, third or nth adjunct to show up following _reads_) is the number of COMPS list entries for adverbs, while that in (3)c cannot have the optional PP[to] complement and has its least oblique ARG-ST element structure shared with the `readee' rather than the `reader' roles in CONT. And it is also an accident that this kind of contrast among (3)a-c will be paralleled by every one of the verbs that can show up in these `middle' constructions, so that no matter *what* verb appears place of _reads_, the lexical entries corresponding to (3)a and (3)b will never differ to any greater extent than these do, while that corresponding to (3)c will always differ in the same way with respect to the first two types of entry. Note further that this behavior cannot be explained by the lexical rule itself; lexical rules just capture generalizations about lexical entries that already exist, don't they? The fact that one assigns separate positions in the lexicon to two homophonous items means that they are free to go their separate ways; as Aronoff and other morphologists have noted, once a given form its own niche in the lexicon it can pretty much go its own way and develop the various idiosyncrasies that, e. g., derivationally related forms are noted for. Isn't this exactly the kind of reason that AGKMS adduce differences in cliticization behavior and subcategorization as strong evidence of a lexical difference between adverbially negated and non-negated forms. And while, as AG note in their paper `French word order and lexicality', `we consider negative adverbs... to be included among the complements of finite Vs , by a specific Lexical Rule' (p. 27), it is also true that the subcategorization differences between certain negated vs. nonnegated verbs, mentioned in their earlier paper on the syntax of French negative adverbials, cannot be built into the lexical rule itself, since verbs which do not take take clausal complements won't manifest these differences. It follows then that there can in principle be differences between lexical entries that are related by lexical rules which aren't specified in those rules themselves. So why do we never find differences between lexical entries corresponding to (3)a and (3)b *independent* of the putative difference in the valence list allowing an adverbial to appear in (3)b? Or *do* we find such differences? If we do, that ends the argument, so far as I'm concerned, and Ivan's position will be completely vindicated. But I've never heard of evidence of that sort. And if it turns out that there isn't any, what kind of plausibility does the LR-based treatment of adjuncts have? What I find a bit hard to understand is this: the adjuncts-are-complements position seems to have default status in this discussion, but it only does so if you assume that there really are no traces, so that tracelessness is really being taken to be the null hypothesis here. It seems to me that a priori it is no more natural to assume that a sign can never have a null phonology than to assume that it can. If assuming tracelessness leads you to a position where you have to posit an arbitrary number of homophones corresponding to a single head, all of which have their own lexical entries and yet never differ from each other except in the one respect you posited them for, why is this a more plausible outcome than accepting the existence of a sign with a null PHON specification? Bob
This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:36:03 PST