research.att.comEditor for this issue: <>
flint.bucknell.edu>I would like to do two things in response to the Stump- Sproat debate: (1) present a simple explanation of the relation of comparison and negation and (2) take issue with Sproat's point (4), that all morphological theories render bracketing paradoxes for cases like _unhappier_. Obviously theories without bracketing have no bracketing paradoxes; the question is: do they have comparable problems. I don't see any. 1a. There are only two logically possible bracketings for _unhappier_: (a) [unhappi]er and (b) un[happier]. (a) implies the neutral, lexically negated adjective _unhappy_ is compared; (b) implies that the compared adjective _happier_ is lexically negated. (Syntactic negation is marked by the particle_not_ in English. The evidence that it is syntactic derives from the facts that _not_ (a) moves in syntax and (b) simply negates; it doesn't involve such conditioned variables as contrariness, reversivity, etc.) 1b. Comparative and Superlative forms are never lexically negated because Comparison is an inflectional category and is therefore formed after lexical (L-) derivation. Evidence: many languages have an analytical comparative, _more happy_, _less happy_, which require syntactic structure. Since inflectional rules apply in syntax (please don't risk much of your career on the assumption that it doesn't), Comparison must apply after all L-derivation is complete. It follows that *un[happier] is impos- sible for the same reason that *_unbetter_, *_unsicker_, *_unfaster_, *_untaller_, etc. are. 1c. Conclusion: The only grammatically possible bracketing for _unhappier_ is [unhappi]er], precisely what we get. Phono- logy has to work with that and, as long as we are not trying to salvage lexical phonology, it can. Given the Peripherality (Outward Sensitivity) Constraint, there are always only two ways to bracket any word and those two ways are determined by whether the affixation involves a prefix or suffix. Moreover, bracketing makes no sense at all with infixation, revowelling, stem muta- tions. How do you bracket the noun _cook_ so that the corres- pondence rules interprets it identically as _baker_? The question makes no sense. As Steve Anderson, Mark Aronoff, P. H. Matthews, myself, and many others have long argued, there is no bracketing therefore there can be no bracketing paradoxes. That is, all bracketing is predictable from the minimum definitions of the morphemes involved. 2. It is not true therefore wrong that all theories predict that _unhappier_ is a bracketing paradox and therefore produce a story for it. Lexical morphology predicts paradoxes here but Lexeme-Morpheme Base theories (LMBM) do not. The lexicon under LMBM contains all and only open-class items, i.e. N, V, A excluding pro-N, pro-V, and pro-A: the same division Garrett gets in his production model, the same division which haunts aphasio- logical studies. The Separation Hypothesis now derives from the architecture of this model and the strongest form of modularity which keeps all lexical categories and operations in the lexicon, all syntactic categories and operations in the lexicon and all morphological operations, the grammatically empty modification of lexemes conditioned by features added by lexical and syntactic operations, in an autonomous morphological spelling component. So long as derivational operations apply in the appropriate order, spelling operations--affixation, revowelling, mutation, nothing--apply blindly and come out in the correct order, stra- tally or not, this makes no difference. The data. Phonological principles alone cannot define the distribution of _-er_. Here's why: the distribution of synthetic comparision in English cannot be predicted purely on phonological principles. (What sort of natural class is "all monosyllabic stems and disyllabic stems ending on an open syllable with a light vowel" anyway?) In fact, PRODUCTIVELY, only disyllabic lexemes containing the morphemes -_y_ and -_ly_ (when used on adjectives but not adverbs: _friendlier_ but *_quicklier_) form synthetic comparatives. Phonology alone certainly does not determine that synthetic forms are allowed only if two adjectives are not compared: My car is redder than yours *My car is redder than orange My car is more red than orange So when comparative constructions arrive at Phonology, the deci- sion as to which comparative is synthetic and which, analytic has already been made and it is a syntactic and/or lexical, not phonological, decision. The question as to what kind of features condition this distinction, therefore, is wide open and clearly the lexicon and syntax are involved. If we allow morphology to handle all affixation, the prefix _un_- is simply added to the modest list of two affixes, -_y_ and -_ly_ which permit _-er_ within syllabic limits. LMBM inflection has already inserted a grammatical feature, say, [+Comparative] into Spec\A and how that feature is used, i.e. where and with what phonological substance is spelled in depends upon the morphology. The decision between /mor/ in Spec position or reading Spec and suffixing the head is the same decision involved in cliticizing the head under Anderson's (_A-Morphous Morphology_, chapter 8) combined principles of clitic-affix distribution. An autonomous morpholgy can read the phonological matrix or the grammatical representation of the lexeme, certainly its derivational history. Since both inflectional and derivational morphology is handled by the same component, there is no possibility that a marker of some lexical derivation feature will be placed outside an inflectional marker. No, I'm afraid that there are far grander differences between morphological theories than Sproat leads us to believe (see what Carstairs-McCarthy and Spencer, for example, think).Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue
mbeya.research.att.com>For those who care, please emend my nonsensical parenthetical to read: (Note that `very' fails for clear scalars like `all', so perhaps what is really tested by `very' is gradability rather than scalar-hood. Thanks to Gregory Ward for this and other points.) Richard Sproat Linguistics Research Department AT&T Bell Laboratories tel (908) 582-5296 600 Mountain Avenue, Room 2d-451 fax (908) 582-7308 Murray Hill, NJ 07974 rwsMail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue
UKCC.uky.edu>The following is a reply to Richard Sproat's reply to my comment on his _unhappier_ squib (LI 23 (1992), 347-352). Responding to my question of whether the comparative form of a scalar adjective is itself necessarily scalar, Richard says I guess it rather depends upon definitions here, but if by `scalar' one means `scalar predicate' in the sense of Horn (1972) or Hirschberg (1991), then surely _unhappier_ is scalar but offers no justification for this conclusion. I think, though, that my original question is more subtle than this somewhat glib response gives it credit for being. I don't have access to Hirschberg's dissertation, but Horn's (1972) conception of scalar predicates is intuitively this: scalar predicates are predicates that designate contrasting points or intervals on some scale; thus, _warm_ and _hot_ are scalar predicates, because they designate contrasting intervals on a scale of temperature. Clearly _happy_ is scalar in this sense; but whether _happier_ is scalar in this same sense is far from clear, and it is precisely this question that my earlier posting was intended to raise. The simplest examples of scalar adjectives are gradable adjectives (i.e. those denoting properties which one may possess to a greater or lesser degree), and it is pretty clear that _happier_ isn't gradable. As an analogy, consider the following model: (i) A B C Is B any less on A's right than C is? No, because the `right of' relation is one of direction, not of distance. (One could, of course, say that C is FARTHER to the right of A than B is, but this is because _far_ interjects the parameter of distance.) Thus, in the one-dimensional model in (i), _right of_ is a non-gradable predicate; B doesn't possess the property of being on the right of A to any lesser degree than C does. Now, suppose that (i) represents three points on a one-dimensional scale of happiness (where individual A is at the point of abject sadness and individual C at that of delirious elation). On this assumption, the `happier' relation seems simply to be the analogue of the `right of' relation, again pertaining to direction but not to distance: B may be less happy than C, but it's not clear that B possesses the property of being happier than A to any lesser degree than C does; either you're happier than A or you're not. The assumption that _happier_ is non-gradable correctly predicts the ungrammaticality of expressions like *_very happier_, *_less happier_, and *_as happier as Sandy_. Note that while sentences like _C is much happier than B_ might seem to suggest that _happier_ is gradable, they are probably comparable to sentences like _C is farther to the right of A than B is_; _much_, like _far_, seems to add a gradable parameter of distance to a non-gradable parameter of direction. In and of itself, the conclusion that _happier_ isn't gradable doesn't entail that it isn't scalar, since there are scalar adjectives that aren't gradable (e.g. the adjective _universal_). There are, however, independent criteria that can be used to test the claim that _happier_ is scalar. For instance, the semantic properties of scalar predicates associated with the same scale cause them to exhibit an asymmetry with respect to contexts like `_____ if not actually _____' and `not only _____ but _____': (ii) The sandwich was warm if not actually hot. *The sandwich was hot if not actually warm. The sandwich was not only warm but hot. *The sandwich was not only hot but warm. (iii) Their performance was good if not actually excellent. *Their performance was excellent if not actually good. Their performance was not only good but excellent. *Their performance was not only excellent but good. (iv) What ensued was widespread if not actually universal pandemonium. *What ensued was universal if not actually widespread pandemonium. What ensued was not only widespread but universal pandemonium. *What ensued was not only universal but widespread pandemonium. Note, however, that _happy_ and _happier_ do not participate in these asymmetries: (v) Sandy is happier (than she was last year) if not actually happy. Sandy is happy if not actually happier (than she was last year). Sandy is not only happier (than she was last year) but happy. Sandy is not only happy but happier (than she was last year). That is, with respect to these criteria, _happier_ doesn't behave as if it were a scalar adjective associated with the same scale as _happy_. Intuitively, this makes good sense, since unlike _happy_, _happier_ doesn't clearly designate a fixed point or interval on the scale of happiness; if anything, it designates a direction on that scale. But if _happier_ isn't scalar, then Richard's claim that the structure [ un [ happi er ]] is semantically workable becomes rather hard to defend. In response to my assertion that the _unhappier_ paradox becomes completely unparadoxical in the Paradigm Function Theory advocated in my article in _Language_ 67 (1991: 675-725), Richard says The point is that, *all* analyses of bracketing paradoxes have presented what seemed to the authors at the time to be well- motivated theories wherein bracketing paradoxes ceased to be such. Indeed, it would be bizarre if things were otherwise, since presumably nobody who treats an apparently paradoxical construction wants to argue that the construction in question remains a paradox in their theory. But the point in my earlier posting is that the _un-ADJ-er_ construction doesn't actually cease to be paradoxical in the approach envisioned in Richard's squib. On the approach he advocates, _unhappier_ is given a single structural analysis (viz. [ un [ happi er ]]), valid both for phonological and for semantic purposes; by contrast, cases like _uneasier_ and _slaphappier_ each have to have two different structural analyses ([ un [ easi er ]] and [ slap [ happi er ]] for phonological purposes, but [[ un easi ] er ] and [[ slap happi ] er ] for semantic purposes). Thus, the paradox remains, though in a somewhat disguised form: despite the obvious parallelism between the three expressions, _uneasier_ and _slaphappier_ have to be analyzed differently from _unhappier_. This paradox vanishes in the Paradigm Function Theory, in which the three expressions receive a uniform analysis at all levels.Mail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue
HKUCC.bitnet>Concerning Rick Moreau's query, the same claim about internal arguments being incorporated into compounds is made (in a finer form appealing to the thematic hierarchy) in Jane Grimshaw's recent book, "Argument Structure" which may well contain discussion of Rick's counterexamples. The solution surely has to do with the fact that examples like "man-made hill" contain passive participles. It is questionable whether "man" here is an external argument of the verb "make". In one GB account of the passive developed by Ian Roberts, the optional object of the by-phrase is an "implicit argument". Under a lexical approach to the passive, it would clearly not be an external argument. At any rate, the relevant argument structure underlying "man-made hill" is that of "hill (is) made by man" rather than that of "man makes hill." Osvaldo Jaeggli's paper on the Passive in Linguistic Inquiry (1986) deals with the status of the agentive by-phrase. IT may be that the compounding theorists would have to modify the definition of internal argument to deal with these cases. Stephen Matthews, University of Hong KongMail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue
essex.ac.uk>Some (I hope) helpful remarks for Rick Moreau: I think the reason Rick Morneau's example are not (necessarily) counter-examples to claims about how internal and external arguments operate in compounds is that the verbs that are the heads of the compounds are actually passive. These are his examples: man-made hill customer-selected colors snake-infested swamp In `man-made', you may seem to have a verb (make) getting its exeternal argument (man) inside the compound (cf `men make X'), but in fact, what you have is a passive (made), and an internal argument (made by man). This is clearest with the last example: ?snakes infest the swamp vs. the swamp is infested by/with snakes. As regards the (non-) existence of VP, and coordination facts, there has been a good deal of work in Categorial Grammar recently: Steedman's paper in Language is a starting point ( J. Steedman, "Dependency and coordination in the Grammar of Dutch and English," Language, vol. 61, pp. 523-568, 1985). --------------------------------------------------------------------------- Doug Arnold, dougMail to author|Respond to list|Read more issues|LINGUIST home page|Top of issue