Lexical Integrity, Checking, and the Mirror: A Checking Approach to Syntactic Word Formation

Lexical Integrity, Checking, and the Mirror:A Checking Approach to Syntactic Word Formation*

MARCEL DEN DIKKEN

Linguistics Program – CUNY Graduate Center, 365 Fifth Avenue – New York, NY 10016-4309,USA (E-mail: [email protected])

Key words: bracketing paradoxes, feature checking, inflectional morphology, lexical Integrity,syntactic word formation, the Mirror Principle

Abstract. Admitting syntactic formation of morphologically complex words is commonlydeemed to be an infringement on the Lexical Integrity Hypothesis. But syntactic word forma-tion, if understood in terms of the checking of features of subparts of words in designatedsyntactic positions, is readily reconciled with strong lexicalism. This paper will argue that achecking approach to syntactic word formation, in tandem with a novel interpretation of theMirror Principle of Baker (1985), yields a straightforward resolution of the otherwise prob-lematic inflectional morphology of the Athapaskan languages, as well as of ‘bracketingparadoxes’ of the unhappier and ungrammaticality type. The syntactically complex structureof unhappier and ungrammaticality that underlies the checking approach to syntactic wordformation is supported on the basis of evidence from polarity item licensing, adverbial modifi-cation, and so-anaphora.

1. Lexical Integrity and the syntax of word formation

1.1. The nominalisation debate

One of the chestnuts of generative linguistic research is the relationshipbetween nominalizations of the type in (1) and sentences of the type in(2).

(1) a. the enemy’s destruction of the cityb. the city’s destruction (by the enemy)

(2) a. The enemy destroyed the city.b. The city was destroyed (by the enemy).

Journal of Comparative Germanic Linguistics 6: 169–225, 2002. 2003 Kluwer Academic Publishers. Printed in the Netherlands.

* Most of the ideas presented in this paper were developed in the course of a morphologyclass co-taught with Chuck Cairns at the CUNY Graduate Center in the fall of 2000. I thankChuck and the students in this class – especially Martin Port – for their feedback. I am alsovery grateful to Anastasia Giannakidou for her help with the polarity facts in section 3.1, to TomRoeper for discussing some of the empirical material in section 3 with me, to an anonymousreviewer for excellent comments on the original version of this paper, and to the guest editorsof this special JCGL issue for their suggestions and guidance.

In seminal work on the topic, Lees (1960) argued for a syntactic approachto nominalizations that would capture the parallels between (1) and (2)straightforwardly. In his (much belated) reply to Lees, Chomsky (1970)instead advocated a lexical treatment of destruction, taking issue inparticular with the idea that (1b) is the passive counterpart of (1a), as(2b) is the passive counterpart of (2a). Chomsky’s key point is that ‘passive’involves two operations (‘Agent-Postposing’ and ‘NP–Preposing’) whichare independent of one another (cf. NP–Preposition without Agent-Postposing or passivization in John’s picture, and Agent-Postposingwithout NP–Preposing or passivization in the destruction of the city by theenemy), neither by itself being an earmark of ‘passive’. So the nominal-ization the city’s destruction by the enemy ‘is only apparently the nomi-nalization of a passive’ (Chomsky 1970, p. 43): after all, the combinationof NP–Preposing and Agent-Postposing evinced by this nominalizationcan arise independently of passivization.1

Chomsky goes on to say that ‘if it were really the nominalization of apassive, this fact would refute the lexicalist hypothesis, since, as wasemphasized earlier, it follows from this hypothesis that transforms shouldnot undergo the processes that give derived nominals’. And that latterpoint, in turn, he establishes in particular on the basis of the fact that anominalization like the growth of tomatoes (or, for that matter, its NP–Pre-posing counterpart the tomatoes’ growth) is unambiguous – unlike verbalgrow, nominal growth only allows for a non-causative reading, not for acausative one. Chomsky (1970, p. 25) explains this by assuming that ‘thesentence John grows tomatoes is derived from a causative construction’.This, combined with a lexicalist approach to nominalizations, will explainthe unavailability of the causative reading for the growth of tomatoes orthe tomatoes’ growth from the fact that transforms cannot be input tolexical operations.

Notice, then, that grow in its transitive incarnation is represented byChomsky in terms of a biclausal structure with an abstract predicate CAUSE

and an embedded sentence [the tomatoes grow]. In other words, Chomskyhere in fact relies on the existence of syntactic word formation: transitivegrow is syntactically derived; if it were lexically derived, there would be

170 MARCEL DEN DIKKEN

1 The conclusion that constructions like (1b) are not the result of passivization seems to bereinforced by the fact that nominalizations corresponding to passive sentences in which, in theabsence of some other modifier, a by-phrase is not omissible (cf. (ia); Grimshaw and Vikner1993) do not require one (cf. (ib)).

(i) a. The building was constructed *(by NP).b. the building’s construction (by NP)

no way to exclude a causative reading of the growth of tomatoes or thetomatoes’ growth along the lines just sketched.2 It is precisely this whichleads Marantz (1997) to conclude that, contrary to common belief,Chomsky (1970) is not in fact a protracted argument in favor of strictlexicalism but instead an ‘argument against the lexicon’.

But even though Chomsky (1970) certainly did not slam the door onsyntactic word formation per se, the conclusion that his paper leaves uswith when it comes to nominalization is quite clear: one had better notmake those in the syntax, or else a causative construal of the growth oftomatoes is impossible to block. Whether this conclusion is warranted ornot (see Marantz 1997 and several other studies in its wake, including DenDikken and Sybesma 1998, for syntactic treatments of nominalizationswhich successfully block the causative reading) is immaterial here; thepoint of interest here is quite simply that, in the aftermath of generativesemantics and Chomsky’s influential rebuttal of a sentence-based accountof nominalizations, subsequent generations of morphologists and syntac-ticians for a long time kept the door to nominalization in syntax securelylocked.

The discussion in the literature typically leaves it there: there are twoapproaches to nominalization, a lexical approach (championed byChomsky 1970) and a syntactic approach (originally developed by Lees1960); the lexical approach reigned supreme throughout the seventies andmuch of the eighties (cf. e.g., Sproat 1985; Randall 1988; Rozwadowska1988), but after having been resuscitated by Baker (1988), the syntacticapproach rapidly started to regain ground (cf. e.g. Murasugi 1990; Hazout1991; Borer 1991, 1999). The idea throughout seems to be that there areprecisely two approaches and that these are necessarily in conflict witheach other – the two could not possibly be reconciled.

LEXICAL INTEGRITY, CHECKING, AND THE MIRROR 171

2 Later in the paper, Chomsky seems to revoke his earlier account of the ungrammaticality ofthe growth of tomatoes on a transitive reading, suggesting tentatively that a lexicalist analysisof both the growth of tomatoes and the growing of tomatoes (the latter supporting the causativereading which the former does not) is possible if one represents transitive grow not bisenten-tially but monosententially, with [+cause] as a feature ‘assigned to certain verbs as a lexicalproperty’ (p. 58): [+cause, grow]. To still make the desired distinction between growth andgrowing, ‘we must restrict the feature [+cause] with respect to the feature that distinguishesderived nominals such as growth from forms such as growing, limiting it to the latter case’(p. 60) – but he admits that this ‘involves an ad hoc step’, and he leaves the issue open.

1.2. Lexical Integrity, checking and the mirror

It is the aim of the present paper to argue that tertium datur. In particular,I will argue that the syntactic approach can indeed be reconciled withstrong lexicalism, via a checking approach to nominalization. The core ofthis approach is the hypothesis that the nominalizing affix, while ‘put on’the verbal stem in the lexicon (which is assumed here to be prior, in toto,to syntactic computation), is ‘syntactically active’ in the sense that it needsto check off its nominal feature in a syntactically projected functionalcategory (D, to be specific). Nominalizations such as destruction orgrammaticality enter the syntactic derivation as verbs and adjectives,respectively, projecting a VP/AP which is included in a functional envi-ronment in syntax which enables the (uninterpretable) nominal feature of-tion or -ity to be checked off and eliminated. This way, the approach laidout here espouses the key ingredients of both the lexical and the syntacticaccounts of nominalization: it shares with the former the idea that themorphological construct (destruction, grammaticality) is indeed formedwholly in the lexicon while it agrees with the latter in recognizing thepresence, in the syntactic structure, of a projection carrying the categorylabel of the stem of the nominalization (VP in the case of destruction, APin the case of grammaticality).

The checking approach, then, assumes that complex words are built inthe lexicon and that some of the ingredients of complex words may havefeatures which are checked against matching features of functional headsin the syntactic structures which these complex words are part of. In linewith Di Sciullo and Williams’ (1987) perspective on the relationshipbetween the lexicon and syntax, the present analysis assumes that thesyntax does not refer to the internal structure of the complex words formedvia lexical word-formation operations; but it is not wholly impervious tothe properties of subparts of those complex words either: the features ofthe inflectional affixes stacked onto the base are visible on the complexword. I follow here the approach sketched in Chomsky (1993, p. 28) andfurther elucidated in Halle and Marantz (1993, pp. 167–168) to thequestion of how the features of inflectional elements in a hierarchicallyarranged complex lexical structure come to be visible on the complexword: the hierarchical structure of the inflected word is ‘translated’ intoa sequence of feature bundles present at the top node (cf. (3), taken fromHalle and Marantz 1993, p. 168), and the syntax ‘sees’ only that sequence,checking off the features of the various inflectional members of thesequence in a set order, starting with the leftmost feature bundle.


The representation in (3) assumes a particular outlook on the way the‘translation’ of inflectional hierarchical structure into a linear sequenceof feature bundles is made – the one suggested in Chomsky (1993, p. 28)as the way to capture the effects of Baker’s (1985) Mirror Principle in (4)(cf. also Brody’s 2000 ‘Mirror’): the features of the affix closest to thestem are put in first position in the sequence of feature bundles and henceget checked first. There is nothing, however, that would intrinsically leadus to prefer this particular ‘translation’ to one that would arrange thesequence of inflectional feature bundles in precisely the opposite order,with Infln leftmost and Infl1 final. While (3) assumes an ‘inside out’ trans-lation of the morphological structure into a linear sequence of featurebundles, one may readily envisage an ‘outside in’ translation which scansthe morphological object from top to bottom. Naturally, random transla-tions of the morphological hierarchical structure into a linear sequence offeature bundles would be sub-optimal; but alongside Chomsky’ssequencing in (3) there is an equally viable alternative that results inexactly the opposite ordering of the feature sequence. Classic MirrorPrinciple ‘violations’ of the Athapaskan type (see section 2) seem toinvolve just such ‘upside down’ mirrors – and there is nothing wrongwith such mirrors from the perspective just outlined.

(4) Mirror PrincipleMorphological derivations must directly reflect syntactic deriva-tions (and vice versa).


(3)

V Af1

Infl1

V

V

Af2

Infl2

V

α

Afn

Infln

Af1 Infl2 . . . Infln

V

[direction of ‘scan’]

1.3. The goals and organization of this paper

In this paper I will exploit the idea that there are two ways, in principle,to translate inflectional morphological structure into a sequence of featurebundles (either ‘inside out’ or ‘outside in’) in a feature-checking analysisof bracketing paradoxes of the type instantiated by unhappier and ungram-maticality. In a nutshell, the discussion to follow can be summed up asfollows:

• The checking approach generates a new perspective on a subset of‘derivational morphology’: nominalizing -tion and -ity, negative un- andcomparative -er are assimilated to inflectional morphology in being non-category-changing morphemes that get their (uninterpretable) formalfeatures checked in functional projections.

• The checking approach leads to a resolution of bracketing paradoxesof the type instantiated by unhappier and ungrammaticality, dissolvingthe bracketing problem in such a way that the morphology andphonology wholly determine the word-internal bracketing of derivedwords, with the appropriate semantic interpretation being read off thesyntactic hierarchical structure.

• The checking approach to unhappier and ungrammaticality-type brack-eting paradoxes leads to the recognition, in the syntactic structuredominating words of these types, of syntactic projections for negation,degree, and adjectival heads whose syntactic activity can successfullybe diagnosed in the empirical facts.

• The checking approach reduces the paradoxical nature of constructs likeunhappier and ungrammaticality to the way the translation from inflec-tional hierarchical structure into a linear sequence of feature bundles isarrived at: the feature-checking sequence is the exact opposite of theorder of attachment as a consequence of the fact that the mirror scansthe hierarchical structure of the complex word from top to bottom(‘outside in’) instead of from bottom to top (‘inside out’).

The paper is organized as follows. In section 2, I will lay out my per-spective on bracketing paradoxes of the unhappier and ungrammaticalitytypes, resolving them in the way sketched out above and discussing theprime theoretical repercussions of the account: the consequences for thevenerable distinction between inflectional and derivational morphologyand for the Mirror Principle. Section 3 will subsequently address a morespecifically syntactic result of the analysis of unhappier and ungrammat-icality developed in section 2: the claim that both feature syntactic pro-jections of Neg and A. Evidence in favor of the syntactic activity of these


projections will be presented and discussed there. In section 4, I thenproceed to developing the structure of de-adjectival nominalizations inmore detail, arguing for the presence of a TP inside the nominalizationand for a syntactic derivation in terms of inversion of the AP predicatearound its subject. Section 5 concludes the paper and offers a possibleextension of the account into the domain of do-support.

2. Bracketing paradoxes as reverse mirrors: A checking account of 2.

unhappier and ungrammaticality

2.1. Bracketing paradoxes – Some introductory remarks

Bracketing paradoxes (cf. (5a–f) for some classic examples; see Williams1981, Pesetsky 1985 etc. for discussion) present the grammar withproblems which, thanks to several decades of careful discussion, are partand parcel of the standard education of every morphologist, phonologist,semantician, and syntactician.

(5)a. unhappier [[un happy] er] [un [happy er]]b. ungrammaticality [[un grammatical] ity] [un [grammatical ity]]c. cross-sectional [[cross section] al] [cross [section al]]d. atomic scientist [[atomic science] ist] [atomic [science ist]]e. eighteen-wheeler [[eighteen wheel] er] [eighteen [wheel er]]f. blue-eyed [[blue eye] ed] [blue [eye ed]]

In each of these cases, the bracketings on the left-hand side reflect thesemantic interpretation of the complex words while the ones on the right-hand side comport well with the morphophonological restrictions imposedby the morphemes involved. Thus, for instance, in the framework ofLexical Phonology and Morphology, un- in (5a,b) is a so-called Level II(or ‘#’) affix and as such wants to attach outside -er and -ity, which bothare Level I (or ‘+’) affixes.

Bracketing paradoxes (or ‘relatedness paradoxes’, as Williams 1981called them) are to some extent a figment of one’s theoretical assump-tions – for instance, if one denies the existence of bracketing per se (cf.work by Anderson, Aronoff, Beard and Matthews), there obviously cannotbe any content to the term ‘bracketing paradox’ at all; and if one gives upthe idea of level-ordering in lexical morphology and phonology, much ofthe problem evaporates as well. On the face of it, though, it would seemthat at least the unhappier case will always stand out as a paradox – quite


independently of theoretical hobbyhorses, we know, on the one hand, that-er attaches only to words that are at most disyllabic; yet, unhappier wouldseem to involve the attachment of -er to a trisyllable, unhappy: unhap-pier is not the negation of the comparative of happy (‘not more happy’)but instead looks like the comparative of the negation of happy.

Even here, however, the question of whether we are confronted with aparadox or a garden-variety case of derivation turns out to be quitecomplex. For there are theories on the market (such as Stump’s 1991paradigm-based theory) for which unhappier is totally unparadoxical.According to Stump (1991, p. 723), ‘[A un- [A [A happi] -er]](= f[DEG:comp]([A un- [A happy]])) doesn’t come from [A un- [A happy]], noris it isomorphic to its logical representation more′(un′(happy′))’. Andespecially as regards this latter point (the semantic interpretation ofunhappier), there was a flurry of debate in the early nineties (in LinguisticInquiry as well as on the Linguist List3), instigated by a squib by RichardSproat (1992) in which the paradox concerning the particular wordunhappier (though not the existence of bracketing paradoxes per se) wasdeclared a hoax on the basis of a more careful inspection of whatunhappier actually means. I will briefly go over this debate here, notintending exhaustivity in any way.

Sproat’s (1992) point is simple: unhappier can happily be assigned justthe bracketing that corresponds to -er’s phonological requirement (i.e.,[A un- [A happi -er]]) – that structure will do both for phonologicalpurposes (as it obviously does) and for the purposes of semantic inter-pretation. The idea here is that the semantics of un- when combined withscalar adjectives (like happy) is not that of contradiction but instead thatof ‘contrary negation’ (Jespersen 1917, Zimmer 1964, Horn 1989): it tellsyou that the contrary of what is designated by the un-less adjective is true.So unhappy means ‘the opposite of happy’, not merely ‘not happy’. Butwith that established, unhappier is no longer paradoxical – with un-attaching outside of happier (as decreed by the phonology of -er), theresulting semantic interpretation is ‘the opposite of happier’ (i.e., ‘lesshappy’), and this, says Sproat, is in fact what unhappier means. In otherwords, the semantics of un- when combined with a scalar adjectivemakes the interpretive component perfectly ready to deal with[A un- [A happi -er]]; there is no need for the [A [A unhappi] -er] struc-ture at all, and the paradox vanishes.


3 See Sproat (1992) and the reply in Kang (1993), as well as the Linguist List postings 3.446,3.454, 3.465, 3.471, 3.472, 3.478, 3.493, and 3.516, contributed by Robert Beard, AndrewCarstairs-McCarthy, Jack Hoeksema, Larry Horn, Richard Sproat, and Gregory Stump.

But apart from the question of whether happier, the comparative ofhappy, is indeed a scalar adjective (a question which a lot of the discus-sion on Linguist List in the wake of Sproat’s squib was devoted to), Kang(1993, p. 789) contests Sproat’s take on the interpretation of unhappierand the conclusion drawn on that basis. Kang claims that (6a) means thatJohn is unhappy (and that Bill is also unhappy) while (6b) ‘states simplythat John’s degree of happiness is lower than Bill’s’ (in other words, itdoes not state that John is unhappy).

(6) a. John is unhappier than Bill.b. John is less happy than Bill.

Kang adduces the examples in (7) to further substantiate the claim thatunhappier and less happy do not mean the same thing. That may be true,but Kang’s claim that (6a) means that John is unhappy needs to be takenwith a sizeable pinch of salt in the light of the fact (pointed out to me byan anonymous reviewer) that it is by no means contradictory to say some-thing like I wouldn’t say John is unhappy, but he certainly is unhappierthan Bill. Kang’s argument for the reality of the bracketing paradox posedby unhappier and the need for the [A [A un- A] -er] structure thereforeneeds to be handled with care.

(7) a.??John is happy but unhappier than Bill.b. John is happy but less happy than Bill.c. John is not happy and unhappier than Bill.d. John is not happy and less happy than Bill.

But even if the [A [A un- A] -er] were shown to be unavoidable forunhappier, the idea that it leads to a bracketing paradox may be, and infact has been, called into question. Robert Beard (in his posting on theLinguist List (3.454, 3 June 1992)), while stressing that ‘[t]he only gram-matically possible bracketing for unhappier is [[unhappi]er]’, asserts thatthe phonological problem that is apparently incurred by this bracketing isillusory. ‘Phonological principles alone cannot define the distribution of-er’: the natural class of possible hosts which the rule of -er affixationmust make reference to (in Beard’s words, ‘all monosyllabic stems anddisyllabic stems ending on an open syllable with a light vowel’4) is not areasonable candidate for a natural class to begin with, according to Beard;


4 This statement is a gross simplification of the facts: polite readily forms its comparativewith -er, for instance. But this does not affect the point; on the contrary, a wholly accuratedescriptive generalization covering the distribution of -er would only end up being an even lessreasonable candidate for a phonologically natural class.

and moreover, there are precisely two types of disyllabic words whichproductively form comparatives with -er (namely, words suffixed with -yor non-adverbial -ly), so one would presumably want to make referenceto those affixes in the rule for -er affixation, but once one admits this,one might as well add the prefix un- to that list as well, making un-A-erbracketed as [A [A un- A] -er] entirely unproblematic.

Beard’s point, while certainly consistent, does little for the ‘emancipa-tion’ of words like unhappier, though – it only aggravates the mystery ofwhy certain disyllabic adjectives but not others support synthetic com-parative formation with -er and superlative formation with -est. Confrontedwith the fact that -y and -ly are the only suffixes which productively allowadjectives formed with them to form their comparatives and superlativeswith -er/-est, a reasonable reply would be to say that in fact -y and -lythemselves serve as hosts for -er/-est (i.e., -er combines first with -y/-ly,the result subsequently combining with A); but unless one were to admitcircumfixes into the grammar of English, such a reply would be unavail-able for un-.

It seems to me, therefore, that attempts to assign unhappier just onesingle structure have failed and that we have to admit that there are indeedconflicting demands put on the structure of unhappier, demands from thelexical phonology and demands from the syntax and semantics of thecomplex word. Now, of course the former type of demand can only be metin the lexicon; but there is nothing that tells us that the semantic demandsneed to be met by the structure assigned in the lexicon. Semantics inter-prets syntax, so we could get the semantics of unhappier under controlby referring to the syntactic structure assigned to the extended projectionof the adjectival phrase headed by this morphological complex and bydeclaring that only that hierarchical structure and not the lexical mor-phophonological bracketing of unhappier feeds the semantics. In whatfollows, I will develop an account of unhappier-type bracketing paradoxesprecisely along these lines, subsequently extending it to bracketingparadoxes of the ungrammaticality type, which I will show are amenableto an account of this sort as well.5


5 Besides unhappier- and ungrammaticality-type bracketing paradoxes, this paper will alsobriefly address in/un-V-abil-ity-type bracketing paradoxes (cf. unreadability), showing that-able, unlike un- and -ity, is not an inflectional, feature-checking element (see section 2.6, below).I will not be concerned in this paper at all with any of the other types of bracketing paradoxcatalogued in the morphology literature (such as (5c–f)); the analysis of unhappier and ungram-maticality may well carry over to those examples (if -al, -ist, -er, -ed can be argued to be inflec-tional elements), but even if it does not, that does not diminish the significance of the overallaccount to bracketing paradoxes in the domain of inflectional, feature-checking morphology.

2.2. Analytic versus synthetic

I will set the stage for the lexicalist yet syntactic analysis of unhappier-and ungrammaticality-type bracketing paradoxes by a brief investigationof the alternation between synthetic and analytic expressions of thecomparative and superlative in English, in comparison to do-support (seealso Graziano-King 1999 and references cited there). Consider the twopairs in (8)–(9):

(8) a. happy happier # more happyb. cheerful *cheerfuller more cheerful

(9) a. leave John left # John did leaveb. leave+NEG *John leftn’t John didn’t leave

For the sake of initial exposition (see section 5.5 for a reappraisal of do-support), we can think of do-support as coming down to something likethe following. An English inflected verb will raise to I (at LF) whereverit can, the use of the dummy do being called upon only in contexts inwhich a special effect is desired (emphasis); but when LF raising is blocked(by negation), an inflected V leads to a crashing derivation, and do-supportis required.

A similar approach unfolds for (8) – more-support (cf. Corver 1997,Graziano-King 1999). Once again raising (this time to the functional head‘Deg’, for ‘degree’; see Corver 1991 for extensive justification of DegPand references to earlier work in the literature) will be performed wheneverpossible, with the dummy more being used only in contexts in which either(i) raising would be possible in principle, but a special effect (here, thepresence of a modifier of the degree: an adverb like much, quite a bit, ora comparative clause with than) is desired,6 or (ii) raising is blocked (thistime not because of the intervention of some functional head like Neg butbecause of morphophonologically imposed complexity restrictions on whatconstitutes a possible host for -er).

The parallel between (8) and (9) will be complete if it is assumed that,just like -ed [+PAST] is merged with the base verb in the lexicon andintroduces morphosyntactic features which need to be checked in a higherfunctional head, so is -er [+COMP] merged with the base adjective in thelexicon. In other words, happier is formed prior to lexical insertion in


6 Graziano-King (1999) points out that in the presence of things like very, quite, and thanthere is a strong tendency for speakers to prefer the analytic comparative and superlative formsto their synthetic counterparts.

syntax, with -er introducing features on the complex word which need tobe checked in a higher F–head, this time Deg. Raising of A+-er to Degis covert in English, just as raising of V+-ed is covert. And just as -ed[+PAST] is an inflectional affix, so is -er [+COMP].

2.3. The unhappier-type bracketing paradox

With this lexicalist, inflectional approach to happier in place, let usconsider its implications for the analysis of the classic bracketing paradoxpresented by unhappier.

The problem with unhappier is that, as far as the phonology is con-cerned, -er must attach before un-; yet unhappier does not mean ‘nothappier’.7 The root of the problem here is the tacit assumption that it isthe morpheme -er itself that expresses the semantics of the comparative.The checking approach to -er taken in section 2.2, however, opens up adifferent perspective.

What we are led to say, given this checking approach, is that -er itselfis semantically perfectly meaningless; all it is is a bunch of uninterpretablemorphosyntactic inflectional features, to be checked against matchingfeatures of a higher functional head. That higher head may itself be seman-tically meaningful or be associated with a semantically meaningfuloperator in its specifier; but the phonological reflex of the inflectionalfeatures is just that: a phonological reflex of inflectional features, itselfperfectly meaningless.

In other words, just like the semantics of tense is located in theT–projection (either in the form of a semantically meaningful T–head or,perhaps more likely, in the form of a null temporal operator in SpecTP),so is the semantics of comparatives located in the Deg–projection (onceagain, either in Deg or in a null operator in SpecDegP).

The paradox that unhappier represents now evaporates. In particular,we are now perfectly free to affix -er prior to the affixation of un- withoutincurring any semantic anomaly: the comparative will take scope overthe negation as a reflex of the way the functional structure of the complexadjectival phrase is projected:

(10) [DegP Deg [NegP Neg [AP . . . [A un-[+NEG] [A happi -er[+COMP]]] . . .]]]


7 I stated the semantic claim in a negative way in the light of the controversy between Sproat(1992) and Kang (1993) (see also the discussion on Linguist List) about what unhappier doesmean. See the discussion in section 2.1, above.

Put differently, the semantics of unhappier is read off the syntactic hier-archical structure, the fact that -er is attached directly to the baseadjective being innocuous as far as semantic interpretation is concerned.

2.4. The ungrammaticality-type bracketing paradox

The paradox presented by ungrammaticality (3b) is subtler and moretheory-dependent than that found in unhappier. For this time, it is notjust ‘phonology vs the rest’: phonology, morphosyntax, and semantics allimpose their own restrictions on the way in which the complex wordshould be analyzed. Phonological considerations (in particular, the theory-based claim that stress-attracting suffixes should be attached at ‘Level I’)force a bracketing in which -ity combines with grammatical prior to theaffixation of un-; but such a bracketing does not just fail to make senseof the semantics of ungrammaticality (which, after all, is the nominaliza-tion of ungrammatical), it also flies in the face of the simple fact that(unrest and a few other such cases aside8) un- does not combine with nouns(cf. *unbook, *uncharm).

To get out of this paradox, we can employ precisely the same strategyas in the case of unhappier. In particular, we can assume that -ity is indeedaffixed prior to un- but that -ity itself does not carry the meaning that wouldbe associated to it if it was a meaning-bearing morpheme. Instead, thesemantics of nominalization is brought in by a functional head (D) intro-duced high up in the syntactic structure of the nominal phrase headed by(un)grammaticality. Put differently, just as -er [+COMP] can plausibly beanalyzed as an inflectional affix, so, I submit, can -ity [+NOM] be taken tobe a feature-checking inflectional element, merged with its host in thenumeration and checking its morphosyntactic features against a higherfunctional head.

Concretely, for the structure of ungrammaticality this leads to thefollowing representation:

(11) [DP D [NegP Neg [AP . . . [A un-[+NEG] [A grammatical -ity[+NOM]]] . . .]]]

where ungrammaticality is treated, category-wise, as an adjective.9

Surprising as this may sound, it has no negative consequences whatsoever.Ungrammaticality never serves as a ‘plain’ adjective (in This sentence is


8 See Horn (1989, p. 284) for a list of cases of this type and some useful discussion of them.9 Or (cf. Marantz 1997; Chomsky 2001, pp. 10, 23) as a category-neutral root in the comple-ment of an adjectival ‘light’ head ‘a’ whose projection is, in turn, embedded in NegP and DP,as in (11).

ungrammatical(*ity) and the like) because -ity carries with it an (uninter-pretable) nominal feature which must be checked against a local D. Ineffect, therefore, -ity A’s will be embeddable only in DP; and even then,only as the head of the (extended) projection in D’s complement – -ityA’s also fail to serve as attributive modifiers inside DP, something whichfollows on the (entirely standard) assumption that raising of the attribu-tive modifier to D is impossible.10 So nothing goes wrong if we assumethat ungrammaticality is an A; on the contrary, it seems to have onlybeneficial effects: un- can now combine with an adjective even if it ismerged after -ity (as desired), and the bracketing paradox is averted.

2.5. Bracketing paradoxes, lexicalism and the reverse mirror

The discussion of unhappier and ungrammaticality-type bracketing para-doxes presented in the foregoing has subscribed completely to the LexicalIntegrity Hypothesis – all the affixes are attached in the lexicon, with thefull complex word entering the syntactic computation as one block andwith the syntactic derivation just being responsible for the checking offof the morphosyntactic features of the various inflectional building blocksof these complex words as listed in the sequence of feature bundles at thetop X0–node. The result is a straightforward account of the morpho-phonology of unhappier and ungrammaticality in the lexicon and anaccount of their semantics read off of the syntactic hierarchical structure.The one remarkable thing about ‘bracketing paradoxes’ of these typesthat remains is the fact that the features of the inflectional affix closest tothe root are checked after the features of more distant inflectional affixes.

The lexicalist feature-checking approach to inflection adopted in theMinimalist program allows us to execute such an approach to unhappierand ungrammaticality without running into a head-on collision with arestrictive approach to the relationship between morphological derivationand syntactic derivation like that embodied by Baker’s (1985) MirrorPrinciple, repeated here:

(4) Mirror PrincipleMorphological derivations must directly reflect syntactic deriva-tions (and vice versa).


10 If attributive modifiers are adjuncts, this follows straightforwardly. The apparent fact thatin Romanian attributive adjectives do seem to be able to raise to D (and pick up the affixaldeterminer; cf. Cornilescu 1992) would have to be analyzed in a different way, e.g., in termsof XP–movement to SpecDP.

The thing to note is that, on a lexicalist approach to inflection, a syntacticderivation which checks the features of un- before the features of the otherinflectional affix (-er in unhappier, or -ity in ungrammaticality) arechecked does actually ‘reflect’ (or mirror) a morphological derivation inwhich un- is attached after the other affix. Such an ‘upside down’ reflec-tion of the morphological derivation in the syntactic feature-checkingderivation will ensue whenever the morphological hierarchical structureis projected onto a sequence of inflectional feature bundles in an ‘outsidein’ fashion, the morphological object being ‘scanned’ from top to bottomrather than from bottom to top (recall the discussion in section 1.2, above).The reflection of the morphological derivation in syntax is perfect – it isexactly upside down but perfectly systematic. The arboreal representa-tion in (12), for unhappier, should make this clear.

On a 1980s-style ‘syntactic word formation’ approach to inflection, bycontrast, a conflict with the Mirror Principle would be unavoidable. Insuch a theory there would be no lexical element unhappier (on the assump-tion that un- and -er are inflectional morphemes, as in the foregoing);instead, the complex word would be built up in the course of the syn-tactic derivation, via head movement of the adjective happy to inflectionalheads in the functional structure dominating the AP, with those inflectionalheads harboring the physical inflectional affixes (un-, -er) themselves.Given that head movement proceeds successive-cyclically and not in azig-zagging fashion, a functional structure of the type in (10) (cf. also thetop section of (12)), with DegP (headed by -er) dominating NegP (headedby un-), would fail to deliver the desired morphological structure (reflectedby the bottom part of (12)).


un-

FF2 -er

FF1

A

A

α FF2 FF1

AP

. . . A . . .

DegP

Neg

FF2

NegPDeg

FF1

(12)

Thus, while an analysis of inflectional morphemes as heads of syntacticphrase markers is irreconcilable with the Mirror Principle in cases likeunhappier and ungrammaticality, a lexicalist approach to inflectionalmorphology of the type adopted in the Minimalist Program allows us toreconcile such apparently unruly objects with a restrictive outlook on therelationship between morphological derivation and syntactic derivationof the type in (4). This is an interesting result – perhaps particularly sobecause Baker (1985) never intended his Mirror Principle to (necessarily)carry over to inflection.11


11 Baker (1985) was concerned with interactions between pieces of derivational morphology(valency-affecting morphemes like causative, (anti)passive, middle, and reflexive affixes), rep-resented as heads of syntactic projections (cf. esp. Baker 1988). A typical example is the inter-play of causative and reflexive morphology in Imbabura Quechua, exemplified in (i). In (ia),the reflexive marker shows up in the morphological string directly adjacent to the base verbmaqa ‘beat’ while in (ib) it is glommed onto the causativizer chi. The logic of Baker’s MirrorPrinciple would have it, therefore, that in (ia) we are dealing with reflexivization at the levelof the causativized VP while (ib) should instantiate reflexivization at the matrix level. This isindeed what we find: in the former, the reflexive links the lower object and the causee (thesubject of the embedded VP), and in the latter we find a reflexive connection between the lowerobject and the matrix subject.

(i) a. Maqa-ku-ya-chi-n. (Imbabura Quechua: Muysken 1981)beat-REFL-DUR-CAUS-3SU

He is causing him to beat himself.

b. Maqa-chi-ku-n.beat-CAUS-REFL-3SU

He lets someone beat him(self).

Baker (1985) relies on Muysken’s (1981) study of Quechua in motivating (4). But in theChumbivilcas dialect of Quechua discussed in Muysken (1988), morphological ordering restric-tions override the Mirror Principle: while the order of chi ‘CAUS’ and ku ‘REFL’ is fixed in thisdialect, (iib) is still a perfectly possible interpretation for Riku-chi-ku-ni. (Bartos 2000 talksabout a superficially similar case of one morpheme order corresponding to two different scopalrelationships, from Hungarian inflectional morphology. That, however, is by no means a straight-forward case of a Mirror Principle problem; I will set it aside here.)

(ii) Riku-chi-ku-ni. (Chumbivilcas Quechua: Muysken 1988)see-CAUS-REFL-1

a. I make someone see myself.b. I make myself see someone.

The Quechua facts may jeopardize a Mirror Principle for valency-affecting (derivational)morphology. In any case, Mirror Principle effects in this domain may come about quite differ-ently from mirroring effects in the realm of inflection, depending on one’s assumptions aboutthe syntactic representation of morphology of the causative, reflexive, (anti)passive, or middletype. I will confine myself here to mirroring effects in the domain of inflection, where (giventhe lexicalist outlook on inflection adopted in Minimalism) they come down to a translation ofinflectional hierarchical structure into a sequence of inflectional feature bundles, as discussedabove.

Indeed, it would appear, prima facie, that the complicated facts ofinflectional morphology in heavily inflecting languages like those of theAthapaskan family (including Navajo) would be very hard to accommo-date in a theory subscribing to (4). For the functional elements ofAthapaskan, Rice (2000, p. 176) arrives at a family-wide morphologicalsequence as in (13).

(13) direct objects < number subjects < qualifiers < situation aspect(or Aktionsart) < viewpoint aspect (or grammatical aspect:(im)perfectivity) < agreement (1/2 person) subjects

This is complicated enough in and of itself (see, for instance, the fact thatthere are two places in the sequence in which subjects are marked: ‘numbersubjects’ is where the plurality of the subject is cross-referenced on thecomplex verb, and ‘agreement subjects’ is where person agreement ismarked for the subject); but the problem is aggravated substantially by thefact that this entire sequence of inflectional elements appears to the leftof the verb stem. Here is a relatively simple example from Navajo:

(14) At’ééd ashkii yidoots’o̧s. (Navajo: Speas 1991b)yi-do-o-∅-ts’o̧s

girl boy 3OBJ-ASP-NONPAST-3SUBJ-will.kiss

The girl will kiss the boy.

On an approach to inflectional morphology that treats the individual mor-phemes as heads of their own phrase markers and that takes the complexverb in (14) to come about via successive-cyclic head movement, theMirror Principle would lead us to surmise that the object agreement markeris attached after the subject agreement marker is attached: after all, itsurfaces further away from the stem than the subject agreement marker,which in fact is the inflectional element closest to the verb. That, in turn,would only be compatible with a syntactic structure in which the func-tional head for object agreement (let us call it AgrO) is higher in the treethan the one for subject agreement (AgrS). But that would be quite out-landish. By contrast, a garden-variety syntactic structure with AgrSP higherthan AgrOP would, nothing else said, seem to incur serious problems forthe Mirror Principle, on a non-lexicalist approach to inflection.

Rice (2000) defends such a non-lexicalist analysis of Athapaskan inflec-tion according to which the various inflectional elements are base-gener-ated as heads of their own functional projections and at the same timetakes very seriously the existence of something like the Mirror Principle– or, in her terms, the idea that an inflectional element that takes scopeover another inflectional element should occur higher in the tree than that


other element. She finds her way around the problem posed by morphemeorder in relation to the desired syntactic functional hierarchy by postu-lating a head-final functional structure as in (15) (adapted from Rice 2000,p. 212; notice that this tree matches (13) perfectly) and by assuming thatthe verb somehow manages to raise across all the functional elements tosurface in a position to their right. This approach certainly yields thedesired result – and it looks more parsimonious and elegant than Hale’s(2001) analysis in terms of templates, different planes at which morphemesare represented, and residual infixation. But it does of course beg the non-trivial question of how the verb manages to raise all the way past all theinflectional elements, a question which she does not give an answer to.

Apart from the fact that Rice’s approach begs the question of how theverb can make its way into its surface position in (15), this structure ofcourse also contravenes antisymmetry (Kayne 1994). As an alternative onemight envision a head-initial counterpart to Rice’s structure and assumethat the surface order results from (i) leaving the verb in situ (which willmake it surface at the end) and (ii) raising all the inflectional affixes upsuccessive-cyclically, their relative surface order resulting from left-adjunction. Such an approach would be incompatible with the facts ofQ–float in Athapaskan, however: the subject can be linked to a floatingquantifier in a position left-adjacent to the verbal complex, as in (16), butno floating quantifier will ever interrupt the string constituted by the inflec-tional affixes and the verb. That the latter is impossible is unexpected onthe head-initial alternative to Rice (2000) just sketched: after all, on thatapproach the sequence of inflectional affixes would not form a complex


AgrOP

V

VP

NumSP

AgrO

QualP

NumS

AspP

Qual

PrepP

Asp

AgrSP

Perf

AgrS

(15)

head together with the verb (which is never incorporated into the inflec-tional cluster), so there is nothing that could prevent a quantifier fromshowing up in a position in between the inflectional string and the verb.

(16) At’ééke Kin

�ání-di la’í ndaalnish. girls Flagstaff-in many are.working

(Navajo: Speas 1991b)

Many girls are working in Flagstaff.

In other words, it looks clear enough that the whole string yidoots’o̧sin (14) is one complex head; but there does not seem to be any straight-forward way of constructing that complex head in the course of the syn-tactic derivation without violating the Mirror Principle.12 But that does notmean that no account of Athapaskan inflection in keeping with a restrictedmapping of morphology onto syntax is forthcoming: it should be obviousby now that a lexicalist approach to Athapaskan inflection, combined withan ‘upside down’ translation of the hierarchical morpheme structure intoa linear sequence of inflectional feature bundles of the type sketched insection 1.2, readily accommodates the complex facts of Athapaskan. Weneed not go into detail here – a single illustration (as in (17), whichcorresponds to (14)) should make things clear.13 A comparison of (17)and our earlier analysis of unhappier-type bracketing paradoxes as in (12)reveals that, in all relevant respects, unhappier and Navajo (14) are justthe same kind of creature: a case of ‘reverse mirroring’. Such ‘reversemirroring’ (by which the sequencing of the inflectional feature bundles isthe image of an ‘outside in’ scan of the complex morphological object) isentirely consistent with Chomsky’s (1993) interpretation of the MirrorPrinciple – and it readily dissolves mirroring or bracketing paradoxesprocured by a non-lexicalist approach to inflectional morphology.


12 Speas (1991a,b) tries out a variety of different approaches but in the end problems for theMirror Principle remain. See also Hale (2001) for a recent attempt.13 See Rice (2000) for eminently detailed discussion, also of the variation within Athapaskanwhen it comes to inflectional morpheme ordering. Rice argues convincingly that affixes corre-sponding to scope-taking functional heads are in a fixed order vis-à-vis each other (one whichreflects the appropriate scopal relationship) throughout Athapaskan while there is variation inthe relative order of inflectional morphemes whenever scope issues do not dictate the ordering.Rice’s approach to this interesting fact (see esp. her chapter 13) readily carries over to thechecking analysis laid out in the text: whenever two functional heads are not in a scope rela-tionship vis-à-vis one another, their relative placement in the syntactic tree is variable, whichis matched by variation in the ordering of the inflectional affixes corresponding to (i.e., checkingfeatures in) these functional heads. I will not flesh this out here, referring the interested readerto Rice (2000).

2.6. Derivation and inflection

In her discussion of Athapaskan morphology, Rice (2000) makes a rigiddistinction between ‘lexical’ and ‘functional’ elements, discussing eachin a separate portion of the book. This is interesting in that it reveals astrong commitment to the distinction between derivational and inflectionalmorphology. In my approach to bracketing paradoxes the nature of specificmorphological elements has also played a key role: for my account to gothrough, an important assumption is that elements like comparative -erand nominalizing -tion or -ity are inflectional elements, not derivationalones. But so far I have not juxtaposed inflection and derivation. I will dothat now.

Let us approach the question of the difference between inflectional andderivational morphology from the perspective of the rationale for thecreation of a linear sequence of feature bundles at the highest X0–node of


AgrSP

Perf

FF2

AgrS

FF1

(17)

PerP

Asp

FF3

AspP

AgrO

FF4

AgrOP

VP

. . . V . . .

α FF4 FF3 FF2 FF1

do-

FF3

yi-

FF4

V

o-

FF2

V

∅

FF1

V

V

a complex morphological object. The reason for this is the need for thefeature-checking process that drives syntactic derivation to be able to seethe features of those parts of the complex object which are ‘relevant tothe syntax’ (as in Anderson’s 1982 famous characterization of inflection)– where ‘relevant to the syntax’ means ‘involved in syntactic featurechecking’. Agreement and temporal/aspectual morphology are certainly inthat envelope. But not all affixes attached to a lexical head are neces-sarily engaged in feature checking against or in the domain of a func-tional head: valency-affecting morphemes (such as causative, (anti)passive,middle or reflexive affixes; cf. fn. 11) are arguably cases of derivational,non-feature-checking morphology. Accordingly, Rice (2000) treats thevoice and valence markers of Athapaskan in her discussion of lexicalelements. I will not address the morphosyntax of these elements here; butwe can safely assume that valency-affecting morphology is in a differentballpark from agreement and temporal/aspectual morphology.

The very fact that agreement and temporal/aspectual morphemes areinvolved in the business of checking their formal features against amatching feature bundle in the syntactic derivation makes it imperativethat their features are included in the linear sequence of features that isthe image of the scan performed on the morphological object. By the sametoken, the very fact that voice/valence morphemes are not thus involvedwill prevent their features from being included in the feature array at thetop X0–node.

When we now return at last to the Germanic language family, we findthat the reflexes of the inflectional/derivational distinction manifestthemselves in the domain of bracketing paradoxes as well. To see this,consider examples (18a,b) (constructed on the basis of Aronoff 1976, pp.121ff.):

(18) a. undemonstrability, uninvestigability, indefensibility, impercepti-bility, indivisibility

b. undemonstratability, uninvestigatability, undefendability, unper-ceivability, undividability

The examples in (18a) all involve the Level I (+) suffix -able/-ible(associated with truncation effects like the one seen in demonstr(at)able,or with allomorphy, as in perceive~imperceptible); those in (18b) instan-tiate the homophonous Level II (#) suffix -able (which triggers no mor-phophonological effects of any kind). Both suffixes accept furtherderivation with a negative prefix (which with Level I -able/-ible is oftenrealized as the Level I prefix in- rather than un-; but this is by no means


a hard and fast rule14) as well as nominalization with -ity – the result ofperforming all three affixations is illustrated in (18).

Now what is important for the point at hand (the derivation/inflectiondichotomy) is that no matter which incarnation of -able we choose, wecan attach both a negative prefix and a nominalizing suffix to the A-ablecomplex. And systematically, the interpretation of the complexes so formedis one in which the nominalizer has highest scope, followed by the negativeprefix, followed by -able. It would seem at first sight that this presents atricky problem for my analysis: for in the case of Level II -able, thereseems to be no way of ‘translating’ the syntactic scope relationship (andconcomitant feature-checking sequence) into a hierarchical morphologicalobject that has the desired bracketing, with Level II un- and -able attachedafter Level I -ity.15 But notice how this problem immediately vanishes oncewe realize that -able (no matter whether it is Level I or Level II: this isreally immaterial for the point to be made below) is not an inflectionalelement: it does not bear a bundle of features which is checked in thedomain of a functional head. It forms adjectives out of verbs, yes; but itcannot form an AP out of a root VP in syntax, in a way analogous to theway -ity can form a DP out of a root AP in syntax. The reason is thateven if we were to succeed in generating an AP on top of the root VP inthe syntactic structure for -able constructions, we still would not manageto make -able a feature checker: lexical heads (like A) do not qualify aselements in whose domain features can be checked; only functional headsdo.

In other words, as far as syntax is concerned, adjectives in -able are justthat: adjectives in -able. They are lexically derived, by attaching a lexical(though affixal) adjective -able to V, not syntactically created out of a rootVP. The feature set of -able does not include any features which are visibleto the scanner that creates a linear sequence of syntactically checkablefeatures. As a consequence, the process that constructs the linear ‘trans-


14 Combinations of in- and -able# are rare (Aronoff 1976, p. 125, fn. 12 mentions onlyinconceivable and indescribable; note that the expected in-V-able+ counterparts of these wordsdo not exist (cf. *inconceptible, *indescriptible) nor do their non-negative bases); as is to beexpected, un- combines more readily with -able+ (see the first examples in (18a)). I will treatLevel I in- and Level II un- on a par in the syntax: as the discussion in section 3.1 shows, thereis evidence for the syntactic presence of NegP for both in- and un-negated As. An anonymousreviewer has raised the important question of why it seems to be the case that only un- entersinto bracketing paradoxes in comparatives: for impolite there is no -er comparative (*impoliter)while for unhappy there is, despite the fact that both polite and happy readily make their com-paratives with -er (cf. politer, brought up already in fn. 4, above). I have no insights to offerhere and leave this question open.15 Thanks to an anonymous reviewer for bringing up this point.

lation’ of the hierarchical structure of a morphogically complex head isentirely oblivious to the presence of -able. For all the words in (18), thatsequence will just include α (the morphological complex) and two inflec-tional feature bundles: one for the negative prefix and one for the nomi-nalizer, in that order (as a result of the ‘reverse’ mirror). For each of thetwo types of cases in (18), this is illustrated in (19).16


16 A few notes on these structures are worth adding for those readers interested in themorphology of -ability nominalizations. Notice that in (19b), -able# combines with -ity+ priorto attaching to the base verb. This morphological derivation, while unavailable on other models,is perfectly well-formed on the assumptions laid out in this paper (esp. the hypothesis that -ityis not a category-changing derivational morpheme but an inflectional element instead): affixa-tion of -ity to -able does not turn the latter into a noun; categorially, -ability in (19b) is anadjective (as the labelling also shows). The idea that Level II -able# combines with -ity beforeattaching to the verbal stem but continues to be an adjective allows us to accommodate (18b)without incurring a bracketing paradox of any sort. Also, while morphophonological rules (stressshift, truncation, allomorphy) readily obtain between -able+ and V in (19a), arguably thestructural distance between -able# and V in (19b) is too large to allow for morphophonologicalalternations.

un-+/un-#

FF2 -ity+

FF1

A

A

α FF2 FF1

AP

. . . A . . .

DP

Neg

FF2

NegPD

FF1

(19) a.

A-able/-ible+

V

Notice, then, that the syntactic structure of an un/in-V-ability nominal-ization does feature projections of D, Neg, and an adjective (syntacticallynominalized by -ity) but does not include a projection of the base verb.This is what the analysis developed in the foregoing leads us to conclude.But, the reader may now ask, is there any evidence for, on the one hand,the presence of syntactic projections of Neg and AP in un/in-A-itynominalizations and, on the other, the absence of a VP in the structure ofV-able constructions? In the next section I will endeavor to show thatindeed there is. Subsequently, section 4 will develop the syntactic accountof de-adjectival nominalizations in further detail.

3. Syntactic evidence for NegP and AP

The structures we arrived at in the foregoing discussion postulate quiteextensive syntactic structures for negative comparatives like unhappierand negative nominalizations like ungrammaticality. The question to askat this time is what syntactic evidence there is for a representation alongthese lines. This question is what I will be concerned with in this section.


un-#

FF2

-ity+

FF1

A

A

α FF2 FF1

AP

. . . A . . .

DP

Neg

FF2

NegPD

FF1

b.

A-able#

V

3.1. NegP

The basic structures in (10) and (11), repeated below, share the presenceof a projection of negation, NegP.

(10) [DegP Deg [NegP Neg [AP . . . [A un-[+NEG] [A happi -er[+COMP]]] . . .]]]

(11) [DP D [NegP Neg [AP . . . [A un-[+NEG] [A grammatical -ity[+NOM]]]. . .]]]

Implicit in these structures, therefore, is the claim that un- representsphrase-level rather than word-level negation – a specimen of negationwhich c-commands not just the stem to which it is affixed but also itssyntactic dependents. We would hence expect this negation to act like sen-tential negation in licensing negative polarity items and, conversely,blocking things which resist being in the scope of Neg0.

The distribution of the Dutch polarity item ook maar X ‘any X’ andthe free-choice reading of the universal quantifiers elk and ieder give usprecisely the desired kind of evidence. Consider first of all the quadru-plets in (20) and (21), featuring polarity items throughout.

(20) a. Niet belast met ook maar enige verantwoordelijkheid,not burdened with also but any responsibility

verliet hij het kantoor.left he the office

Not burdened with any resposibility, he left the office.

b. Onbelast met ook maar enige verantwoordelijkheid, unburdened with also but any responsibility

verliet hij het kantoor.left he the office

c.*Ontlast/Ontheven van ook maar enige relieved of also but any

verantwoordelijkheid, verliet hij het kantoor.responsibility left he the office

d.*Vrij van ook maar enige verantwoordelijkheid, verliet free of also but any responsibility left

hij het kantoor.he the office


(21) a. Niet gehinderd door ook maar enige vorm van not hindered by also but any form of

zelfkritiek, nam hij het woord.self-criticism took he the word

Not hindered by any form of self-criticism, he took the floor.

b. Ongehinderd door ook maar enige vorm van unhindered by also but any form of


c.*Ontstegen aan ook maar enige vorm van zelfkritiek,un-risen to also but any form of self-criticism

nam hij het woord.took he the word

d.*Gespeend/Wars van ook maar enige vorm van devoid/averse of also but any form of


There is a clear contrast, in both cases, between a– and b–examples onthe one hand, and the c– and d–examples on the other: the latter are con-sistently bad. Particularly striking is the contrast between niet belast/onbelast in (20a,b) on the one hand and ontlast in (20c) on the other: whileall based on the same stem and having a negative sense, only the formermanage to license ook maar enig.

These data show that while the presence of the prefixal particle ont- inthe c–examples (a cognate of the ‘reversative’ un- of English verbs likeundo, not the negative prefix) does nothing to help license the any-typepolarity item in the adjective’s domain, the negative prefix on- in theb–examples does create the requisite circumstances under which polarityitems can be licensed. It should be clear that a semantic account of any-type PI licensing couched in monotonicity (as in Ladusaw 1979) will nothelp us here. Ladusaw’s approach has it that only downward entailing (DE)environments license polarity items. But it is easily ascertained that all ofthe examples in (20) and (21) involve DE environments and that they differen bloc from examples with belast ‘burdened’ or gehinderd ‘hindered’:just as (22a) is a valid entailment, so are (22b–d) while clearly (22e) isnot a valid entailment.


(22) a. niet belast met verantwoordelijkheden →niet belast met zware verantwoordelijkheden

b. onbelast met verantwoordelijkheden →onbelast met zware verantwoordelijkheden

c. ontlast van verantwoordelijkheden →ontlast van zware verantwoordelijkheden

d. vrij van verantwoordelijkheden →vrij van zware verantwoordelijkheden

e. belast met verantwoordelijkheden /→belast met zware verantwoordelijkheden

burdened with responsibilitiesburdened with heavy responsibilities

So a semantic approach in terms of monotonicity will not do asGiannakidou (1998) has also shown, quite independently of the facts athand. Nor will a pragmatic approach à la Linebarger (1980), in terms ofnegative implicature. Giannakidou’s (1998) semantic approach in terms ofveridicality does make the desired cut for (20). Her hypothesis is that itis nonveridical contexts that license any-type PIs; and on the assumptionthat veridical contexts license an inference of existence of the object whilenonveridical contexts do not, we correctly distinguish between (20a,b) onthe one hand and (20c,d) on the other: while the truth of the left-handpart of (22a,b) does not presuppose the existence of responsibilities, theleft-hand part of (22c–e) does, which is why only the former two licenseany–PIs.

Now notice that in the examples in (21a,b) the head (on)gehinderd itselfarguably does not c-command the container of the any-type PI. After all,we are dealing here with a passive by-phrase, and passive by-phrases arenot usually assumed to be in the complement of the passive participle. Sothe fact that an any-type PI is successfully licensed in the by-phrase inboth (21a) and (21b) strongly suggests that it is not the adjectival participleongehinderd itself that is doing the licensing here. We are then led toconsider a licenser outside the projection of the adjectival participle whichdoes c-command the by-phrase and which has the potential of licensingPIs. The obvious candidate, of course, is Neg.

So on the basis of the examples in (21) in particular, we have now foundourselves a first piece of syntactic evidence for the presence of a NegP inthe structure of un- prefixed adjectival phrases: the only nonveridicaloperator that is in a position to successfully license the any-type PI in(21b) is Neg0. Even though they are semantically negative and even


downward entailing (though not nonveridical), the adjectives in (20c,d)and (21c,d) fail to license an any-type PI. So semantic negativity in somebroad sense of the term is not sufficient; what is needed is the presenceof a nonveridical operator that c-commands the PI, and in the specific caseof (21b) the only such operator that will do the trick is a syntacticallypresent Neg.

The peculiar free-choice interpretation of the universal quantifiersieder/elk in the example in (23a) is the exact opposite of ook maar–PIswhen it comes to its distribution: though it depends on the presence ofnegative semantics (failing to be licensed in the absence of any negativity),it does not tolerate a syntactically projected negation.17

(23) a. Hij weigerde ieder/elk geneesmiddel.he refused every medicine

b. Hij gebruikte ieder/elk geneesmiddel.he used every medicine

c. Hij vertrouwde (nooit) op ieder/elk geneesmiddel.he trusted (never on every medicine

While all examples in (23) are grammatical, only (23a) allows for an inten-sional, modalized reading (which is actually very salient in these kinds ofsentences) in which ieder/elk behaves like English free-choice any: what(23a) says, on this reading, is that he refused and would refuse anymedicine at all that one offered or would offer him (cf. Blom and Daalder1977, who refer to the discussion of any in Klima 1964). The other twoexamples only allow the garden-variety universal quantifier reading forieder/elk. And since that garden-variety universal quantification readingis impossible to get for noun phrases headed by words like responsibilityin sentences of the type in (24b–c), it follows that these examples aresimply impossible while (24a) is still fine on the free-choice reading ofieder/elk.

(24) a. Hij ontkende iedere/elke verantwoordelijkheid.he denied every responsibility

b.*Hij nam/accepteerde/aanvaardde iedere/elke he took/accepted/accepted e very

verantwoordelijkheid.responsibility


17 See Giannakidou (2001) for discussion of the fact that free-choice items are incompatiblewith extensional/episodic negation, and exemplification from Greek, Catalan and Spanish.

c.*Hij verlangde (nooit) naar iedere/elke he longed (never for every

verantwoordelijkheid.responsibility

With these observations about ieder/elk in mind, let us return to thequadruplets in (20) and (21), replace all tokens of ook maar enig withieder/elk, and see what happens. The result is given in (25) and (26):ieder/elk is grammatical in precisely the contexts in which ook maar enigis bad and vice versa. Since we know that ieder/elk, on its free-choicereading, resists being in the scope of a syntactically projected negation,the ungrammaticality of the b–examples once again confirm the presenceof NegP in the extended projection of un-negated adjectives.

(25) a.*Niet belast met iedere/elke verantwoordelijkheid, verliethij het kantoor.

b.*Onbelast met iedere/elke verantwoordelijkheid, verliet hij hetkantoor.

c. Ontlast/Ontheven van iedere/elke verantwoordelijkheid, verliethij het kantoor.

d. Vrij van iedere/elke verantwoordelijkheid, verliet hij hetkantoor.

(26) a.*Niet gehinderd door iedere/elke vorm van zelfkritiek, nam hijhet woord.

b.*Ongehinderd door iedere/elke vorm van zelfkritiek, nam hij hetwoord.

c. Ontstegen aan iedere/elke vorm van zelfkritiek, nam hij hetwoord.

d. Gespeend/Wars van iedere/elke vorm van zelfkritiek, nam hijhet woord.

Systematically, we find that, with respect to polarity-sensitive items inthe adjective’s domain, adjectives negated with free-standing and affixalnegation behave on a par and as a pair behave differently from implicitlynegative adjectives.

For English, the facts from the domain of polarity item licensing arenot very easily reproduced – primarily because of the fact that Englishseems to be more liberal when it comes to the contexts in which any–PIsare licensable than Dutch is with respect to its ook maar–PIs: while Englishallows for things like devoid of any content (Postal 2001) and ignorant of


any mistakes,18 in Dutch it is really only the on- prefixed adjectives (andadjectivally used participles) that behave on a par with syntacticallynegated APs. But in the realm of the free-choice reading of universal quan-tifiers, English does mimic the Dutch pattern, with every in post-possessorposition (which Postma and Rooryck 1996 refer to as ‘intensional’ every:every possible NP). While perhaps not the best possible illustration, thetriplet in (27) shows this: whereas (27a,b) are clearly impossible, (27c) ismarkedly better.

(27) a.*Not troubled by his every worry/fear/responsibility, he left theoffice.

b.*Untouched by his every worry/fear/responsibility, he left theoffice.

c. Relieved of his every worry/fear/responsibility, he left theoffice.


18 I should add, though, that there does seem to be a tendency for any embedded under ignorantto get a free-choice reading rather than a PI–reading. A random search on the web using ageneral–purpose search engine turned up the examples in (i) with unaware in combination withany, which all clearly have the PI–reading – any cannot be modified with almost or absolutelyhere, a sure sign that we are not dealing with free-choice any. The examples in (ii), featuringignorant, on the other hand, are all grammatical with almost/absolutely inserted before any,suggesting that these involve the free-choice interpretation of any. Those (few) cases in whichan any in the complement of ignorant does get the PI–reading involve a licenser outside theignorant–AP (cf. (iii)). Notice, then, that ignorant does not block the licensing any across itself;conversely, unaware does not seem to block free-choice items in its complement: (iv) givestwo attested examples.

(i) a. A group of 8,500 citizens directly below the boom’s epicenter were unaware thatanything out of the ordinary had taken place.

b. The company is cooperating with authorities and ‘is unaware of any improprieties.’c. He’s unaware of any articles that have linked health problems to the slippery

biofilm that collects inside dentist water lines.d. Clay and I were up and around 5:00 AM but were unaware of anything amiss

until Clay went to get a cup of coffee in the main pavilion around 5:20–5:30 AM.

(ii) a. I’m fairly ignorant about anything beyond basic coverage.b. Counter jockeys and sales help at all computer stores are ignorant of anything other

than whatever complete systems the store sells.c. It’s a talk show that seems to be ignorant of anything that’s happened in late night

television in the past 20 years.d. I mean it’s natural for you Americans to be ignorant about anything outside of

US soil and believing that North America is made up only by the US!

(iii) a. He cannot ever be ignorant of anything.b. Do not assume the Revenue is ignorant of anything.c. If any is ignorant of anything, the ignorance proceeds from his own defect.

(iv) a. I was unaware of the early web (then again I was unaware of pretty mucheverything for 16 years).

b. I sort of blanked out, pretty much unaware of anything at all.

Once again, we find that adjectives negated with prefixal un- behave thesame way as syntactically negated APs; and once again (since (27a,b), justlike (21a,b), involve by-phrases) we can be sure that the parallel is thereflex of the presence, in the extended projection of both types of APs, ofa projection of Neg0.

What appears to be a prima facie problem for my claim that un-Abehaves syntactically like not A when it comes to polarity-related phe-nomena is the fact that the two clearly go separate ways in a variety ofcontexts, including the ones in (28) and (29). While un-A is modifiableby the positively polar modifier pretty, it cannot be modified by negativelypolar at all in the absence of a higher negation (cf. (28a,b)); not A, onthe other hand, can never be modified by a positively polar modifier atall and does not need the help of a higher negation to license negativelypolar at all as a modifier, as the primed examples in (28) show. Likewise,there is a robust contrast between the left-hand examples in (29), featuringprefixal negation, and their right-hand counterparts with n’t/not (cf. Horn1989, pp. 184–185 and references cited there for more such cases).19


19 Stiebels and Wunderlich (1994, p. 918) present a similar case from German, with sondern‘but’:

(i) a. Er war mit dem Computer nicht zufrieden, sondern wollte einen he was with the computer not satisfied but wanted a

neuen.new

He was not satisfied with the computer but wanted a new one.

b. *Er war met dem Computer unzufrieden, sondern wollte einen neuen.he was with the computer unsatisfied but wanted a new

The scope-based account of the text examples carries over directly to these German examples:(ia) is grammatical only if nicht ‘not’ has wide scope; but prefixal negation (un- in (ib)) alwaystakes narrow scope: it scopes over the AP but not over anything outside it.

Note also Hoeksema’s (2000, p. 133) observation that yet and as yet behave differently inthe domain of un-prefixed adjectives (cf. (ii)). As Hoeksema points out, ‘[t]he association of asyet with negation is of a different, looser, kind . . . from the one we encounter in the case ofyet’: it does not need to be c-commanded by Neg at S–structure, for instance, as the grammat-icality of (iii) shows. It is this ‘loose’ link with negation that apparently allows as yet (but notyet) with un-A.

(ii) a. These problems are unsolved as yet.b. *These problems are unsolved yet.b′. These problems have not been solved yet.

(iii) As yet, we have not heard anything.

(28) a. He is*(n’t) at all unhappy.a′. He is(n’t) not at all happy.

b. He is(*n’t) pretty unhappy.b′.*He is(n’t) not pretty happy.

(29) a.*John is unhappy either.a′. John isn’t/is not happy either.

b. That is impossible, {#is/isn’t} it?b′. That isn’t/is not possible, is it?

The problem is only apparent, however. Establishing that there is asyntactically active projection of Neg present in the structure of con-structions with both un-A and not A is not, of course, tantamount toestablishing that the precise structural location and scope of Neg is thesame in both cases. In section 2, we arrived at the structure in (10) forunhappier constructions:

(10) [DegP Deg [NegP Neg [AP . . . [A un-[+NEG] [A happi -er[+COMP]]]. . .]]]

So there is indeed a NegP in the structure dominating unhappier, but it islocated below DegP, right outside the adjective phrase. Now, obviously,in an example like not happier, while NegP will still be present in the tree,its location will be different – at the very least, the semantics as well asthe periphrastic paraphrase not more happy tell us that NegP is higher thanDegP in this case. Regardless of whether it is above or below DegP, Negwill c-command any and all of A’s complements – which is why unhap-pier and not happier behave the same when it comes to the distributionof polarity items in A’s complement position. The relative distance of NegPfrom AP is inconsequential in this domain. But when it comes to modi-fiers of the adjective or polarity-sensitive elements at the level of theclause, all else is not equal: these are higher up in the tree, being able toescape the c-command domain of a low NegP while still being c-com-manded by a higher NegP. This is precisely what we see in (28) and (29):the polar modifiers in (28) are c-commanded by the sentential negationspresent in these structures, but they are outside the reach of the NegPconstrued directly with the un-negated AP, which is very low in thestructure. The account readily extends to (29). With not A, on the otherhand, the NegP construed with the adjective is higher, scoping over thepolarity-sensitive elements in (28) and (29), and hence yields a totallydifferent picture than un-A in these cases: it all depends on scope.

So far, I have focused on morphologically negated adjectives that‘remain’ adjectives – things like unhappy. But in the light of the repre-


sentation I assigned in section 2 to ungrammaticality-type bracketingparadoxes (cf. (11), repeated here), we should expect un-As suffixed with-ity type nominalizers to behave just like ‘purely’ adjectival un-As.


In particular, we should expect these to be able to license negative polarityitems like any and either in Neg’s c-command domain and, conversely, toblock the free-choice reading of the Dutch universal quantifier ieder/elkin Neg’s scope. The facts in (30) show that this is indeed the case.20

(30) a. [Zijn ongeschiktheid voor ook maar één van deze [his unsuitability for also but one of these

rollen] staat buiten kijf.roles stands outside dispute

b.*[Zijn onbereidwilligheid tot iedere/elke concessie] [his unwillingness to every concession

stelde iedereen teleur.disappointed everybody PRT

This, I believe, satisfactorily settles the first question raised by the accountof un-related bracketing paradoxes developed in section 2: Is there anyevidence for NegP?

3.2. AP inside DP

A second question that comes up is specific to adjective-based nominal-izations, like (un)grammaticality. For these, the structure in (11) postu-lates a full-fledged AP as part of the syntactic structure underlying theirderivation. Can we bring this AP to life, so to speak?

To see this, we should first step back to consider Fu, Roeper and Borer’s(2001) arguments for the presence of a VP in the syntactic structure ofverb-based process nominalizations – arguments of two different species:(i) the distribution of adverbial modifiers, and (ii) the distribution of so-anaphora. Let us start with the latter – perhaps the clearer of the two.


20 The Dutch example in (30b) is perhaps interpretable on a garden-variety universal quanti-fier reading for ieder/elk; but such a reading certainly is not very salient.

3.2.1. So-anaphoraFu, Roeper, and Borer (2001) note that, while do so replacement failscompletely in the a–sentences in (31)–(32), the b–examples are gram-matical (though not equally felicitous to all speakers – but speakersgenerally agree that the relative contrast between the a– and b–examplesis real, and that is what is of primary interest to us here).21

(31) a.*Kim’s version of the event and Bill’s doing so were surprising.b.? Kim’s explanation of the event and Bill’s doing so were

surprising.

(32) a.*Sue’s trip to Easter Island surprised us, and Amy’s doing soannoyed us.

b.? Sue’s exploration of Easter Island was impressive, and Amy’sdoing so was a real surprise.

The grammaticality of examples of the type in (31b) and (32b), accordingto Fu, Roeper, and Borer, argues beyond doubt that there is a VP presentin the derived nominal. The facts in (33) and (34) are instrumental to naildown the case:

(33) a. Hankamer: I’m going to stuff this ball through a 6-inch hoop.Sag: I don’t believe you can do so/it.

b. [Hankamer attempts to stuff a ball through a 6-inch hoop.]Sag: I don’t believe you can *so/

✓it.

(34) a. He removed the evidence in the morning, and she did so (*thecounterevidence) in the afternoon.

b. his removal of evidence in the morning, and her doing so inthe afternoon

As Hankamer and Sag (1976) first showed, the surface anaphor so needsa linguistic antecedent (cf. (33a,b)). Moreover, so’s antecedent must belarger than just V (cf. (34a)). In the light of these facts, (34b), which formsa miminal pair with (34a), tells us clearly that the syntactic structure dom-inating removal includes a full-fledged syntactic projection of a verb.


21 The attested examples in (i) (from Kehler and Ward 1995, quoted in Fu, Roeper, and Borer,2001) show that the phenomenon manifests itself in ‘real-life’ prose as well:

(i) a. The defection of the seven moderates, who knew they were incurring the wrathof many colleagues in doing so, signaled that it may be harder to sell the GOPmessage on the crime bill than it was on the stimulus package.

(Washington Post)

b. Even though an Israeli response is justified, I don’t think it was in their bestinterests to do so right now. (token provided by Dan Hardt)

Fu, Roeper, and Borer (2001) stress that do so cannot take a resultnominalization as its antecedent: (35) is ill-formed.

(35) *John’s collection and my doing so.

The cut, then, is between deverbal process nominals on the one hand andother nominals on the other, regardless of whether they are deverbal (cf.the result nominalization in (35)) or denote a process (cf. the non-deverbalnoun phrase headed by trip in (32a)).

With this in mind, let us consider adjective-based nominalizations ofthe grammaticality type. The examples in (36a–c) illustrate the patternwe find with so replacement. And to broaden the empirical perspectivesomewhat, (36d) is an example featuring how replacement.

(36) a.? The acceptability of drugs in Holland – perhaps even more so than tobacco products and alcohol – is on the agenda of theEuropean Parliament this week.

b.? One of the strange quirks of OT is that optimality is apparentlya gradable property: some candidates are said to be more or lessso than others.

c.? The grammaticality of this sentence – or at least, the fact thatit seems so to many speakers – has not been discussed in theliterature before.

d.? The grammaticality of this sentence – or at least, that’s how itsounds to many speakers – has not been discussed in theliterature before.

All of these sentences give the same result as Fu, Roeper, and Borer’scases of so-replacement of the VP in verb-based nominalizations: theirstatus is delicate but grammatical, and they contrast markedly with casesin which we are not dealing with derived nouns, such as the ones in (37),which are totally impossible.

(37) a.*The {(dis)advantage/boon/charm} of living in New York –(which is) perhaps even more so (i.e., {(dis)advantageous/charming}) than living in Boston – is that . . .

b.*The vice/virtue of being critical of the Bush administration –or at least, the fact that it is so judged (i.e., judged asvicious/virtuous) by many of my colleagues – is hotly disputed.

This of course recalls the contrast between (31/32a) and (31/32b) notedby Fu, Roeper, and Borer. The contrast between derived and underivednominals is confirmed particularly poignantly by the quadruplet in (38):


(38) a.*Its charm – or the fact that it seems so to many people – is itsmajor asset.

b.*Its attraction – or the fact that it seems so to many people – isits major asset.

c.? Its attractivity – or the fact that it seems so to many people –is its major asset.

d.? Its attractiveness – or the fact that it seems so to many people– is its major asset.

While charm and attraction are semantically very close to attractivityand attractiveness, only the latter two support adjectival so anaphora –charm does not because it is underived; and even though attraction isunquestionably a derived nominal, it is not de-adjectival while socorresponds to an AP in these examples.

We can even reproduce the contrast between process and result nominals(cf. (31/32b) vs (35)) in the domain of de-adjectival nominalizations. Theminimal pair in (39) illustrates just that.

(39) a.? It was John’s responsibility – or at any rate, the fact that thatwas how he looked – that got him the job.

b.*It was John’s responsibility – or at any rate, the fact that thatwas how he looked – to take care of that.

In (39a), John’s responsibility can be paraphrased as ‘the fact that John isresponsible’, the counterpart in the de-adjectival domain of a processnominalization, and as a result how anaphora succeeds; in (39b), on theother hand, responsibility is not propositional in content, hence howanaphora fails.

So we see that the same kind of reasoning that informed Fu, Roeper,and Borer’s (2001) first argument for recognizing a VP in the syntacticstructure of deverbal result nominals tells us that there is a so-replaceableAP in the structure of de-adjectival nominals like grammaticality.22


22 The discussion here was based exclusively on English; reproducing it in other languages isusually difficult (for instance because of the lack of surface anaphora of the English so type).Thus, Dutch does not have the requisite anaphor to bring to light a syntactically projected VPin the structure of deverbal process nominals. In the domain of de-adjectival nominalizations,however, it seems we might reproduce the argument just laid out in the main text – I am cautious,though, because it is not completely clear that the kinds of anaphors used in the examples beloware really surface anaphors (in the sense of Hankamer and Sag). Consider the examples in (i)–(ii).

(i) a. *Jans ellende – of het feit dat hij zich zo leek te voelen – . . .Jan’s sorrow or the fact that he himself so seemed to feel . . .

b. *Jans verdriet – of het feit dat hij zich zo leek te voelen – . . .Jan’s sadness or the fact that he himself so seemed to feel . . .

3.2.2. Adverbial modificationFu, Roeper, and Borer’s (2001) second argument for a VP in deverbalprocess nominalizations comes from adverbial modification. We are ofcourse well aware of the fact that ‘purely’ nominal phrases generallycannot be adverbially modified – the a–examples in (40)–(42) are straight-forwardly ruled out: there is nothing present in the structure of theseexamples which the adverb could legitimately adjoin to.23 Now, it isinteresting to note that the b–examples in (40)–(42), while once again notperfect and subject to some speaker variation, contrast markedly with thea–cases.

(40) a.*[Kim’s version of the event thoroughly] was a big help.b.? [Kim’s explanation of the problem thoroughly] was a big help.

(41) a.*[Kim’s accident suddenly] disqualified her.b.? [The occurrence of the accident suddenly] disqualified her.

(42) a.??[His metamorphosis into a werewolf so rapidly] was unnerving.b. [His transformation into a werewolf so rapidly] was unnerving.

What makes the b–examples different, from an analytical perspective, isthat they all involve deverbal nominals. Apparently, the verbal base ofthese nominals is syntactically active: it can be modified by adverbial


c. ?Jans verdrietigheid – of het feit dat hij zich zo leek te Jan’s sadness or the fact that he himself so seemed to

voelen – . . .feel . . .

(ii) a. ?Het was Jans verantwoordelijkheid – of in elk geval het feit dat hij it was Jan’s responsibility or in any case the fact that he

zich zo gedroeg– die hem populair bij het volk maakte.himself so behaved that him popular among the people made

b. *Het was Jans verantwoordelijkheid – of in elk geval het feit dat hij it was Jan’s responsibility or in any case the fact that he

zich zo gedroeg – om dat probleem op te lossen.himself so behaved COMP that problem up to solve

The pair in (ii) reproduces English (38); (i) is a close counterpart to English (37), where (ia,b)are both underived nouns with the same gross meaning as the de-adjectival noun verdrietigheidin (ic). My judgments track the ones reported for English; the facts are certainly suggestive,but more careful study is needed to establish the status of zo as a surface anaphor.23 This is not to say that simple noun phrases cannot take adverbial modifiers: in sentenceslike The newspaper yesterday did not report the incident, the adverb yesterday forms a con-stituent with the noun phrase (cf. Barbiers 1996 for discussion and references). But modifierslike thoroughly, suddenly, rapidly, or initially do not seem to be able to behave the wayyesterday can.

material. This, then, is a second piece of evidence for the presence of aVP in the syntactic structure of deverbal process nominals.24

Building on the outcome of this discussion, Roeper and Van Hout (2000)go on to suggest that there is evidence for a VP in the structure under-lying V-ability nominalizations as well:

(43) ? [The acceptability of drugs initially in Holland] was later amatter of dispute.

The adverb initially in (43) cannot be a modifier of the noun phrase headedby acceptability since simple noun phrases do not accept this modifier(*the cue initially); instead, they want an adjectival modifier (the initialcue). The fact that initially is nonetheless acceptable in (43) indicates,according to Roeper and Van Hout, that there is a projection of the verbaccept present in the syntactic structure underlying the nominalizationacceptability – and it is the VP of accept that is being adverbially modifiedin (43).

Roeper and Van Hout’s is not the only conceivable conclusion to drawfrom (43), however. While acceptability is ultimately based on the verbaccept, it is more directly built on the adjective acceptable; and of courseadjectives accept adverbial modifiers just like verbs do. In other words,in (43) initially could also be a modifier of a syntactically projected AP.And as a matter of fact, this latter approach seems closer to the truth thanthe one suggested by Roeper and Van Hout. For, as Fu, Roeper, and Borer(2001) point out, deverbal adjectives fail to show evidence for a syntac-tically projected VP – so-anaphora is impossible, as their examples in (44)show:25


24 Notice that a suggestion to the effect that the b–examples involve reduced relative clauses(cf. his transformation into a werewolf which happened so rapidly) would entirely fail to makesense of the contrast between these examples and the a–sentences. As Fu, Roeper, and Borer(2001, p. 560, fn. 12) point out in this context, a reduced relative approach would also fail toshed light on the fact that only a subset of adverbial modifiers are allowed in nominalizationsof the b–type: while VP–level adverbs like thoroughly, suddenly, and rapidly are fine, sentence-level adverbs such as presumably, probably do not work here at all. I refer to Fu, Roeper, andBorer (2001) for discussion of the restrictions on adverb types in deverbal process nominals; atthe end of section 3.2 I will bring up similar, though not identical, restrictions on adverb typesin the domain of de-adjectival nominalizations, and in section 4 I will analyze those from asyntactic perspective.25 In more recent work, Roeper (2002) has partially revoked his earlier claim that V-able doesnot support do so replacement or adverbial modification of VP, presenting examples such asthe ones in (i) (where (ia) clearly involves Level II -able#). I reproduced these examples directlyfrom Roeper (2002) without adding any diacritics revealing speakers’ judgments. I have found,however, that speakers are generally very reluctant to accept these examples; they seem dis-tinctly worse than the Fu, Roeper, and Borer (2001) examples and the ones presented below.

(44) a.??This act turned out to be amendable, and the British Parliamentdid so last month.

b.*The fish was edible and Kim did so.c.*Kim tried not to be resentful of her cousins, but her husband

did so.

Theoretically as well, there is reason to believe that there is no VP whichdo so could replace in the syntactic structure dominating V-able. Recallfrom section 2.6 that there is no way of resolving bracketing paradoxesof the in/un-V-abil-ity type if -able is taken to be an inflectional affixwhose feature bundle is included in the sequence under X0; and on othergrounds as well, it looks unlikely that -able is a category-neutral inflec-tional element: to all intents and purposes, -able is a lexical (though affixal)adjective which lexically derives As from Vs. The structures in (19)illustrated how in/un-V-abil-ity constructions are derived in the mor-phology and how they project in syntax – and those structures made itperfectly clear that there cannot be a projection of the base verb presentin the syntactic structure dominating in/un-V-abil-ity. The argument carriesover, of course, to non-nominalized in/un-V-able: once again, -able lexi-cally derives an adjective from a verb, preventing the latter from being thesyntactic head of the construction. And since there cannot be a VP presentin the syntax of in/un-V-able, there is nothing for do so to replace: theexamples in (44) are thus ruled out, which seems to be the correct result.

By the same token, if the unavailability of so-anaphora with deverbaladjectives defeats the postulation of a syntactically active V–projection,the example in (43) cannot involve adverbial modification of a VP either;instead, we must be dealing here with a syntactically active AP modifiedby initially. The example in (43), then, procures our first piece of evidencefrom the domain of adverbial modification for the presence of an AP inthe syntactic structure of nominalized adjectives.

This conclusion can be reinforced by an investigation of cases involvingunequivocally A–based nominalizations such as grammaticality. There isno VP present in the structure of grammaticality on anyone’s assump-tions.26 So if we still find that modification with an adverb like initiallyis possible, this will tell us without a doubt that the constituent modified


(i) a. John is clearly defendable, and someone must do so.b. The terrorists became quickly discoverable, although tracing each one by computer

took a long time.26 Unless one were to represent all non-verbal predication in terms of an underlying copularsentence featuring abstract be. I am not aware of any evidence for such an approach, nor do Ibelieve that the copula is a verb (cf. Den Dikken 1995).

by initially is an AP. And that in turn will then tell us that A–based nom-inalizations like grammaticality feature a syntactic structure in which thereis a full-fledged AP present underlyingly. So what are the facts? It turnsout that adverbs like initially can indeed be used in combination withgrammaticality, as in (45a). (45b–d) give three more examples of the sametype. Once again, it is clear that derived nouns behave markedly differ-ently from underived ones: examples like (45e) are impossible. Thisconfirms what we had already seen reflected in the examples in (37)–(39):that there is evidence for an AP in the structure of A–based nominaliza-tions.

(45) a.? [The grammaticality of this construction initially in early OldEnglish] gave way two centuries later to an entirely differentpattern.

b.? [The (il)legality of soft drugs initially in this country] was latercontested in court.

c.? [The infelicity/unfeasibility of such a move ultimately] was notrealized at first.

d.? [The untenability of this proposal eventually/in the end] neverprevented it from being extremely popular in the field forseveral decades.

e.*[The {(dis)advantage/boon/charm} initially of living in NewYork] changed dramatically after 9/11.

I hasten to add, though, that adverbial modification of A–based nomi-nalizations (whether of the acceptability type, where there is a verbal root,or of the grammaticality type, where there is not) is definitely not unre-stricted. Dutch, for example, does not allow this kind of thing at all:

(46) a.*de ongrammaticaliteit van deze constructie the ungrammaticality of this construction

aanvankelijk in het Oudengelsinitially in the Old English

b.*de (il)legaliteit van softdrugs uiteindelijkthe (il)legality of soft drugs ultimately

And even in English, as Roeper and Van Hout (2000) note, ‘the data isparticularly subtle and often yields obscure results’, citing (47a) as a failedattempt at having an -ability adjective modified by a manner adverb suchas quickly.


(47) a.??the learnability of grammar so quickly by children(cf. grammar is learnable quickly by children)

b.*the acceptability of this suggestion very/more/most(cf. this suggestion is very/more/most acceptable)

c.* the ungrammaticality of this construction severely/strongly(cf. severe/strong ungrammaticality; this is severely/strongly ungrammatical)

Likewise, we find that these adjectives cannot at all be modified by degreewords like very or more/most, as in (47b). And this is not, of course,peculiar to -ability type nominals: while grammaticality-type de-adjectivalnominals can be modified by temporal adverbials like initially, as in (45a),modification by degree adverbs like severely/strongly fails completely, asthe ungrammaticality of the example in (47c) shows. Note that there isno semantic incongruity here: the paraphrases given below (47a–c) arewell-formed.

In the following section, I will interpret the split between temporal andother types of adverbials from the perspective of a syntactic analysis ofthe derivation of de-adjectival nominalizations.

4. The structure and derivation of de-adjectival noun phrases in 4. syntax

In the discussion in this paper so far, we have found evidence to supporta syntactic analysis of de-adjectival noun phrases that includes the ingre-dients schematized in (11), repeated here:


There is a syntactically active NegP (see section 3.1), and there is a syn-tactically active AP (see section 3.2), the structure being topped off by aprojection of the determiner, D. But this cannot be all there is to it – inparticular, we need to be wary of the fact that (i) predicates must belicensed by being linked to Tense (as Guéron and Hoekstra 1988, 1993 putit, they must be included in a T–chain; cf. also Déchaine’s 1993, p. 21Predicate Visibility Principle, which says that a predicate is visible onlyif it is c-commanded by Tense) and (ii) predicates must have a subject,which, in turn, must be licensed in its own right.27


27 I have no space here to defend the claims in (48) – I refer the reader to Guéron and Hoekstra

(48) a. Predicates must be licensed by Tense.b. Predicates must have a subject.c. The subject must be licensed.

How do we put these things together? It is clear that (9) as it stands willnot suffice – the elementary requirement of predicate licensing viaT–linking cannot be met: the DP in (9) does not include a projection ofTense, and the presence or absence of Tense outside DP is clearly imma-terial (DP – a ‘strong phase’ in the sense of Chomsky (1999) – breaks achain; cf. also fn. 27). At minimum, therefore, we need a TP inside theDP, to license the predicate.28 That TP will then be put to good use whenit comes to the licensing of the predicate’s subject as well: SpecTP willbe able to serve as the licensing position for the subject. The interim resultis (49) (where I assume that the NegP that checks the features of thenegative prefix un- is located inside TP, in recognition of our earlier findingthat this NegP appears to be stacked immediately on top of the projectionof the lexical adjective):

(49) [DP D [TP SU [T′ T [NegP Neg [AP . . . A . . .]]]]]

Here, I located the subject in its licensing position though this should notbe taken to imply that I would deny the predicate-internal subject hypoth-esis: as far as this study is concerned, it is immaterial whether the subjectoriginates inside or outside AP;29 what matters is that it is licensed inSpecTP, as in (49).

The presence of TP in the structure of de-adjectival nominalizationsimmediately allows us to accommodate the fact that, in English, temporaladverbial modifiers are grammatical in such constructions (cf. (43) and


(1988, 1993) and Déchaine (1993) for extensive discussion. A note of clarification with respectto (48a) is in order, though. What (48a) does not mean is that there has to be a T in every smallclause. All it says is that every small clause predicate must be in a T–chain with T. So wheneverthere is something on the path between the predicate and T that breaks the formation of aT–chain, a more local T–node is needed to ‘save’ the predicate.28 On tense inside DP, see also Larson’s (1998) discussion of John’s former car.29 See Kratzer (1989) and Diesing (1990) for the idea that subjects of individual-level predi-cates are generated outside the projection of the predicate head. In this context, also note theimpossibility of quantifier float in examples like (i), which (esp. if Sportiche’s 1988 approachto Q–Float is on the right track) suggests that there is indeed no lower position for the subjectin the structure of de-adjectival nominalizations. Fu, Roeper, and Borer (2001, pp. 570–571,fn. 21) note that Q–Float also fails in deverbal process nominals; cf. (ii).

(i) a. * these sentences’ ungrammaticality allb. *the ungrammaticality all of these sentences

(ii) a. * the refugees’ deportation all by the soldiersb. *the soldiers’ deportation all by the refugees

(45), above): there is a TP present to which they can serve as adjuncts.To be more precise, adverbs like initially can be adjoined to T′, as is alsoshown by the grammaticality of sentences of the type in (50), where (ona non-split IP analysis, as is now current; cf. Chomsky 1995, Chapter 4and subsequent work) an adjunction position to T′ seems to be the onlyoption.

(50) He initially didn’t get the problem.

I am not suggesting, of course, that adverbs like initially are necessarilyT′–adjoined (they clearly are not: Initially he didn’t get the problem is finealongside (50)); but the fact that they can be so adjoined is precisely whatwe need to account for the nominalization facts discussed: when initiallyand its ilk show up in final position in nominalizations, they are T′–adjoinedand stranded after inversion of the predicate and its subject.30

Dutch apparently does not allow adverbial adjunction to T′ at all. IfTravis (1984) and Zwart (1997) are right that Dutch subject-initial rootclauses are TPs, the ungrammaticality of (51) with aanvankelijk betweenthe subject and the finite verb tells us that T′ adjunction is not an optionin this language, not even for temporal adverbials. And with T′ adjunc-tion barred, we predict, in the light of the analysis presented here, thatadverbial modifiers stranded at the end of nominalizations will be impos-sible in Dutch as well – which is correct, as we saw already in (46).

(51) Hij ⟨*aanvankelijk⟩ begreep ⟨aanvankelijk⟩ het he ⟨*initially understood ⟨initially the

probleem ⟨aanvankelijk⟩ niet.problem ⟨initially not

That degree modifiers like very/more/most/severely/strongly are impos-sible in de-adjectival nominalizations even in English (cf. (47b,c)) isrelatable to the fact that comparatives cannot serve as input to de-adjectival nominalization: while we do have purity and rarity, we do notfind *purerity or *rarerity though it should be perfectly clear what suchforms, had they been available, would have expressed. Apparently, if webuild a DegP on top of the NegP in (49), we ruin the chances of survivalof the derivation built on this structure. But why should this be so?


30 Adverbs like cleverly, which serve as manner adverbs when lower in the structure, can alsobe adjoined to T′ but then assume a subject-oriented interpretation (cf. (i)). Subject-orientedreadings are unlikely to emerge in nominalizations like ungrammaticality; I will leave themaside.

(i) John ⟨cleverly⟩ didn’t ⟨cleverly⟩ answer the question

Clearly, what we cannot say is that that DegP per se is in the wrongplace: in the account of unhappier-type bracketing paradoxes, we gener-ated DegP in exactly the position which we are now saying causes havocin the derivation of ungrammaticality. The problem must lie in some aspectof the syntactic derivation which is specific to ungrammaticality, notshared in common with that of unhappier. That aspect, I believe, is inver-sion.

What I would like to argue is that in the course of the derivation of de-adjectival nominalizations, the predicative AP (accompanied by NegP inun-negated cases) inverts with its subject, much as in things like The doctoris John and, to mention a closer relative, that idiot of a doctor (see DenDikken 1995, 1998 for discussion):31

(52) [DP D [FP (NegP+)APi [F′ F (= of ) [TP SU [T′ T ti]]]]]

Inversion of AP around its subject manoeuvres the de-adjectival noun intothe correct surface position vis-à-vis its subject, and it also introduces thelinker element of in between the two, as a signal of the application ofinversion (cf. Den Dikken 1995, 1998).32 The inversion step depicted in(52) thus immediately kills two birds with one stone.

And in addition, it helps kill a third bird as well – or, to put it more


31 A note is due here about the motivation for inversion. The head of AP has a nominal featureto check in D; raising AP to SpecFP, which is local to D, will make subsequent raising of A toD possible, and the derivation converges as a result. A simple head-movement scenario appar-ently fails: A cannot raise first to T and then on to D. Though further investigation is neces-sary on this point, it may be that, in general, movement of a lexical head to T blocks furtherupward movement. Verb Second in the Mainland Scandinavian suggests that a verb can getinto the Verb Second position without ever raising to T – MSc does not have V–to–T raisingin embedded clauses; unless one is to assume an otherwise unmotivated touch-down in T enroute to the C–domain in V2 sentences, it looks like V2 in MSc skips T altogether (cf. Holmbergand Platzack 1995). One way of addressing the apparent locality problem incurred by this deriva-tion is the assumption that V2 actually involves raising of a maximal projection which, after allother material has been raised out to external licensing positions, dominates just the lexicalverb (cf. Sportiche 1998). Sportiche’s approach to Verb Second is very similar to the deriva-tion depicted in (52), which likewise involves XP–movement to a TP–external position of aphrase which, as its sole lexical content, contains just the lexical head.32 For details, the reader is referred to Den Dikken (1995, 1998); briefly, the idea is thatmovement of a predicate around its subject threatens to violate locality (minimality, shorteststeps) but that the locality violation is successfully averted by raising the head in whose spec-ifier position the skipped subject is sitting (T in (52)) up to the head whose specifier serves asthe host for the raised predicate (F in (52); the nature of F need not concern us here). Thisapplication of domain-extending head movement (in the sense of Chomsky 1995, Chapter 3; itrenders the two specifier positions ‘equidistant’ from any other position in the tree) is signalledovertly by the emergence of a linker element (‘copula’) under F. See the works cited for otherexamples of predicate inversion in both clausal and nominal contexts, likewise giving rise tothe emergence of a linker.

transparently, it helps us account for the fact that degree modification ofAP is impossible in de-adjectival nominalizations of the type in (47b,c),repeated here:

(47) b.*the acceptability of this suggestion very/more/mostc.* the ungrammaticality of this construction severely/strongly

Here we find the degree modifier in a position to the right of the subject,which, in the light of the derivation sketched in (52), means that it wasleft stranded under AP–predicate inversion. On the assumption (exploitedin section 2 as well) that ‘degree’ is represented in the tree in the form ofits own functional projection (DegP), the structure and derivation of(47b,c) will look as in (53):

(53) a. [DP D [TP this suggestion/construction[T′ T [DegP very/more/most/severely/strongly[Deg′ Deg [NegP Neg [AP acceptability/ungrammaticality]]]]]]]

b.*[DP D [FP [NegP Neg [AP acceptability/ungrammaticality]]i

[F′ of [TP this suggestion/construction[T′ T [DegP very/more/most/severely/strongly [Deg′ Deg ti]]]]]

And now it is easy to see why this derivation fails – in (53b) we haveattempted to raise the predicate out of a DegP with a filled specifier, some-thing which delivers an ungrammatical result for the same reason that wecannot say *Nice, I consider him very: we have violated locality. (In thetheoretical terms of Chomsky’s 1999 ‘derivation by phase’ model, whatwe would say is that DegP is a ‘strong phase’.) Along similar lines, theungrammaticality of (47a) can be derived; once again, there will be noway of stranding the modifier to the right of the subject. Temporal mod-ifiers, by contrast, are perfectly well strandable there: they are modifiersof the T–projection (adjoined to T′), and do not interfere in any way withany of the steps taken in the derivation of the examples in (43) and (45),which are hence grammatical.

While this is sufficient to take care of the contrast between (43)/(45)and (47), we still need to ensure that (i) the adverbial degree modifiercannot be dragged along under inversion either (cf. (47b′,c′)) and (ii) adjec-tival degree modification of ungrammaticality, as in (47c″), is fine.

(47) b′.* the very/more/most acceptability of this suggestionc′.* the severely/strongly ungrammaticality of this constructionc″. the severe/strong ungrammaticality of this construction

Let me start with (47b′,c′) and point out that it is not impossible, perse, for a Degree Phrase to invert with a subject inside the nominal phrase


– noun phrases of the type in (54) (recently discussed in detail in Borroff2000; cf. also Bennis et al. 1998, Kennedy and Merchant 1998) illustratejust this:33, 34

(54) a. We have always been [DP the best of friends].b. John is not [DP very good (of) a student].c. John is [DP too good (of) a student].d. [DP How good (of) a student] do you think John is?

But here the adjective does not have a nominal feature to check againstD. So apparently the structure in (55) yields a well-formed output if allthe A–head needs to do is check a feature in Deg, but it crashes if, inaddition, A has a nominal feature to check up in D.

(55) [DP D [FP [DegP very/ . . . [Deg′ Deg [AP . . .]]]i [F′ F (= of ) [TP SU [T′ T ti]]]]]

The desired result can be obtained by assuming that raising of the A–headfrom Deg further on to D is impossible in the structure in (55). And with


33 The linker of is not obligatory in the examples in (54b–d); in fact, its distribution in thesekinds of noun phrases is dialectally/idiolectally restricted. The of-less variants of the examplesin (54b–d) involve A′–fronting of the Degree Phrase, not predicate inversion, hence are lessinteresting in the context at hand.34 Inversion of a degree-modified AP around its subject inside a nominal phrase is notunrestricted. Thus, (ia), though assigning a high-degree interpretation to the inverted predicate,does not accept any overt expression of superlative degree on the inverted predicate (cf. (ic))though degree modification with the aid of, for instance, extreme is fine (cf. (ib)):

(i) a. an/that idiot of a doctorb. an/that extreme idiot of a doctorc. * that [biggest idiot] of a doctor

(I made sure in (ic) that biggest would form a constituent just with idiot, not with the con-taining noun phrase as a whole: the latter has a demonstrative determiner, and demonstrativesare generally incompatible with superlatives, so a construal of biggest as a modifier of all ofidiot of a doctor is impossible; apparently, however, biggest cannot ‘just’ modify the invertedpredicate idiot either.) Similarly, in Hungarian, while (iia) and (iib) are grammatical (cf. DenDikken and Lipták 1997 for discussion of this kind of inversion construction), (iic) is impos-sible. Even when the A–head does not need raise up to D to check a nominal feature there, thenwe find that degree modification of the inverted adjectival predicate is not by any means freelyavailable. It may well be that the text examples in (54) are the exception rather than the rule.

(ii) a. hülye egy srác (Hungarian)crazy a guy

b. de/olyan hülye egy srácthat/so crazy a guy

c. *hülyébb/leghülyébb egy sráccrazier/craziest of guy

DegP sitting in the specifier position of a functional head in (55), this mayultimately be an ‘attract’ problem. D would need to attract Deg to get theadjective+-ity up to D. But Deg is embedded on a left branch and,moreover, has an overt specifier which, in the configuration in (55), isarguably closer to D than Deg is: while Deg is dominated by the leftbranch, its specifier (on the assumption that specifiers are indistinguish-able from adjuncts; cf. Kayne 1994) is not. SpecNegP in (55) hencearguably prevents attraction of Deg from D. And with such attractionblocked, -ity cannot check its nominal feature against D. The derivationthen crashes, as desired. Questions of detail naturally arise at this juncture;but it seems reasonable that with the structure in (55) we can blame theungrammaticality of (47b′,c′) on failure of the checking of -ity’s nominalfeature.

That leaves us with (47c″), the severe/strong ungrammaticality of thisconstruction, which is perfectly well-formed. The mere fact that we findadjectival modifiers here rather than adverbial ones is already a clearindication that they are not construed with the AP that serves as the basisfor the derivation of the de-adjectival nominalization. They must be intro-duced into the structure after everything that we have laid out so far hasalready taken place – so after the inversion of the predicate around itssubject. In other words, severe/strong in (47c″) serves as a modifier of allof what follows. For the account to be complete, we would still need topinpoint the precise location of these modifiers in the tree. I will notdiscuss this issue in any detail here; instead, I will make the simplifyingassumption that a (nominal) Degree Phrase can be introduced on top ofFP and that, since this DegP finds itself in the complement of D, degreemodifiers harbored by this DegP will take the form of an adjective, notof an adverb.

This takes care of all of the facts encountered. So to summarize briefly,what we have seen is the following:

• The structure of de-adjectival nominalizations includes a full-fledgedAP predicate.

• The AP predicate is licensed by Tense, which projects a TP in A’sextended projection.

• The T–projection provides the licensing position for the subject of AP.• The T–projection provides an adjunction site for temporal modifiers like

initially.• The AP predicate inverts with its subject, raising to SpecFP; the linker

of shows up in F.• Stranding degree adverbials under predicate inversion violates locality.


• Carrying degree adverbials along under predicate inversion blocksA–raising to D.

• Adjectival degree modifiers can be introduced between D and FP.

With these results in place, I now proceed to a brief recapitulation of themain conclusions of this paper and a possible extension of one of theminto the realm of do-support in negative clauses.

5. Conclusions and further perspectives

5.1. Bracketing paradoxes dissolved via feature checking

In this paper I have presented a novel account of bracketing paradoxes ofthe unhappier and ungrammaticality type, predicated on the assumptionthat the bracketings imposed by the morphophonological requirements ofthe individual affixes are the sole structures assigned to these constructsin the lexicon but that nonetheless the semantics manages to assign themtheir proper interpretations thanks to the syntactic structures dominatingthem. It is the syntactic projections for negation (Neg), degree (Deg), thedeterminer (D), and the adjective (A), and their hierarchical relations inthe syntactic tree, which feed the semantics; the affixes on the morpho-logically complex heads do not receive a semantic interpretation them-selves: they possess uninterpretable formal features which are checked inthe course of the syntactic derivation via syntactic movement operationsperformed on parts of the syntactic structure.

5.2. The inflection/derivation distinction

Thus, one of the results of this study is that so-called derivational affixeslike -ity are assimilated here to well-established inflectional affixes like-s in being uninterpretable feature-bearers and may hence be called inflec-tional affixes themselves. This depletes the set of derivational affixes; butit does not empty it completely: there still remains a core set of purelyderivational morphemes, like -able for instance, whose behavior is dif-ferent from that of inflectional morphemes: no feature checking butcategory and/or valency change.


5.3. The Mirror Principle

From the checking perspective pursued here, bracketing paradoxes of theunhappier and ungrammaticality-types cease to exist qua bracketingparadoxes and in fact cease to be paradoxes. What is special about themis that they turn the mirror upside down: in compiling the sequence offeature bundles of the inflectional elements, the complex morphologicalobject is scanned from the outside in (or from top to bottom) rather thanfrom the inside out (or from bottom to top). In this respect, as I showedin detail in section 2, unhappier and ungrammaticality-type bracketingparadoxes exhibit the exact same behavior as complex inflected verbs inthe Athapaskan languages: while apparently in conflict with the MirrorPrinciple, these, too, become entirely well-behaved on the assumption thatmorphological objects can translate their inflectional hierarchical structureinto a linear sequence of feature bundles not just ‘inside out’ (as in (56a))but also ‘outside in’ (as in (56b)).


FP3

F2FF2

F3FF3

(56)

FP2

F1FF1

FP1

LP

. . . L . . .

α FF1 FF2 FF3

Af3

FF3

L

a.

L

Af1

FF1

Af2

FF2

L

Why would there be such a choice of ‘mirroring’ when it comes to thecompilation of the sequence of feature bundles on a complex X0 while –on the assumption that Kayne’s (1994) LCA holds – no comparable choicepresents itself in the translation of hierarchical structure into linear wordorder? The answer to this question lies in the following difference betweenthe two processes, feature sequencing (FS) and word ordering (WO). Inthe current derivationalist perspective on syntax, an immediate transla-tion of an asymmetrical relationship between two nodes into a WO of theterminals of those nodes can be made, at every step along the way. Thenatural way to go about the creation of a WO, therefore, is to do it intandem with the derivation, from the bottom up. Not so in the case of FS:crucially, the compilation of the sequence of inflectional feature bundleshas to wait until the entire morphological complex has been constructed.After all, as Halle and Marantz (1993, p. 168) put it, ‘[t]he inflectionalfeature bundles of the affixes attached to the Verb are arranged in asequence with “α” itself ’ (my italics), where ‘α’ is the complex mor-phological object (in Halle and Marantz’ example, an inflected verb). SoFS does not proceed in tandem with the construction of the complexmorphological object: it involves a scan of that morphological object afterit has been completed. That means that, in the case of FS, there is nocompelling reason to go about the job from the bottom up: since FS takesplace ‘after the fact’ anyway, the morphological object may be scanned


FP1

F2FF2

F1FF1

FP2

F3FF3

FP3

LP

. . . L . . .

α FF3 FF2 FF1

Af3

FF3

L

b.

L

Af1

FF1

Af2

FF2

L

either from the inside out (as Chomsky 1993, p. 28 does it) or from theoutside in (as I have argued is the case in unhappier, ungrammaticality,and Athapaskan inflected verbs). The former scan gives rise to ‘straight’Mirror Principle effects, the latter to ‘reverse mirrors’ of the type discussedin this paper. Neither is intrinsically superior to the other: the two are onequal footing.

While this opens up more possibilities when it comes to the mappingof morphology to syntax than does the original Mirror Principle, it isclearly not the case that under my ‘two-way mirror’ approach ‘anythinggoes’: there are precisely two legitimate ways of scanning a morpholog-ical object under FS; ‘zig-zagging’ operations are sub-optimal, perhapsentirely ruled out. The resulting theory is empirically more adequate thanBaker’s (1985) Mirror Principle and Chomsky’s (1993, p. 28) recastingthereof and still theoretically highly restrictive.

5.4. Lexicalism and Lexical Integrity

The approach to bracketing paradoxes and inflection in general taken inthis paper is strongly lexicalist. It should be clear that only a lexicalistapproach to inflectional morphology could give rise to the ‘two-waymirror’ theory developed here.

The checking perspective on unhappier and ungrammaticality in factallows us to eliminate the tension between Lexical Integrity (or stronglexicalism) and syntactic word formation, which has always been espe-cially poignant in the domain of nominalizations. The approach taken hereis syntactic in assigning extensive syntactic structures to de-adjectivalnominalizations like (un)grammaticality; but at the same time it is stronglylexical in fully obeying Lexical Integrity: (i) all the affixes are attachedin the lexical morphological derivation and (ii) the ‘syntactically relevant’(i.e., checkable) features of the constituents of the complex morpholog-ical object are all represented directly at the highest node of the morpho-logical complex (via FS). Thus, the present approach does for so-calledderivational morphology (esp. nominalizing morphology) what Chomsky’s(1995) checking approach did in the domain of run-of-the-mill inflection,resolving the tension between lexicalism and syntactic activity.

5.5. A potential consequence for sentential negation and do-support

With un- treated essentially as an inflectional affix checking featuresagainst Neg0, the possibility presents itself to take a similar approach to-n’t, the contracted sentential negation marker. Suppose, then, that -n’t is


merged with its host verb in the lexicon, with the complex raising throughNeg on the way to I.35 This has at least two profitable consequences.

One immediate one is that we can straightforwardly rule out contrac-tion of sentential negation and the infinitival marker to, resulting in *ton’t.The question of why, in a negative infinitive like It is desirable to notweaken the theory of grammar (where not expresses sentential rather thanconstituent negation; see Beukema and Den Dikken 1989 for other clearcases of sentential not following to), not can never contract with theinfinitival marker is not easy to answer on an approach which takes -n’tto be a syntactic head with clitic-like properties. If, on the other hand,-n’t is merged with its host in the numeration, the fact that it refuses tomerge with to can quite simply be reduced a categorial selectional restric-tion on the part of -n’t: to is formally prepositional, but -n’t cannot dockonto prepositional hosts due to its selectional properties.

A potentially more interesting extension of the selectional restrictionsimposed by -n’t could get us an essentially non-syntactic explanation forthe fact that -n’t cannot attach to lexical verbs. Recall from the discussionin section 2.2 that a peculiar property of comparative -er is that it doesnot combine with complex adjectives (where ‘complex’ here seems to bedefined basically in phonological terms: a syllable count). For -n’t we couldsay something similar, on a lexicalist approach: -n’t does not combine withcomplex verbs. This time, ‘complex’ will not be defined in terms ofnumber of syllables (cf. the difference between main verb do and dummydo) but in terms of the morphological structure of the verb. In particular,if we follow Lebeaux (1988) and Hale and Keyser (1993) in assuming thatthematic structure comes in the form of lexical treelets associated withθ-assigning heads, then the lexical entry of any θ-assigning verb willfeature a little ‘θ-tree’, representable in terms of an l-syntactic structurebelow the word level. This makes all θ-assigning verbs lexically complex.On the assumption, then, that -n’t is not allowed to dock onto lexicallycomplex verbs (e.g., by Kayne’s 1994 LCA), the ban on attaching -n’t toa θ-assigning verb – known elsewhere as the ban on ‘raising to I acrossNeg’ of θ-assigning verbs – follows.36


35 On -n’t as an inflectional affix, see also Zwicky and Pullum (1983).36 A few notes are in order here. I assume that modals are never θ-role assigners, not even intheir ‘root/deontic’ sense (cf. Wurmbrand 2001, pp. 203–204 for a recent defence of the positionthat modals do not assign θ-roles); similarly, I assume that have and be are never θ-role assigners– not even the ‘equative be’ of Cicero is Tully or the ‘main verb’ use of have in I haven’t anymoney (see Den Dikken 1995 for discussion of be and have). Note also that on this account,do-support in negative sentences has nothing to do with ‘adjacency’; hence, we do not needany additional statements (e.g., of the type introduced in Bobaljik, to appear, section 3.3) toensure that adverbs do not trigger do-support. This may be viewed as an advantage of the textapproach.

To make this account of the ban on finite lexical verbs with sententialnegation complete, we need to say something about not, the free-standingcounterpart of -n’t. Why is it that *John left not is equally bad as *Johnleftn’t? The answer here allows itself a slight abstraction: the assumption,in particular, that in sentential negation constructions with non-contractednot there is a null counterpart of -n’t present as a negative affix on thenegated verb. In other words, *John left not is more properly representedas *John left+∅ not, where ‘∅’ is the phonologically null allomorph ofthe carrier of the negative feature on V. On the assumption that ‘∅’ sharesthe ban on affixation to complex hosts with its overt incarnation -n’t,the account of *John left not will be directly parallel to that of *Johnleftn’t.37, 38

One further advantage of the present suggestion in comparison withthe syntactic approaches in Pollock (1989) and Chomsky (1995) is that itis readily made compatible with the facts of Mainland Scandinavian.Modulo Verb Second (which I abstract away from here by consideringembedded clauses only), Mainland Scandinavian is like English in non-negative sentences in not raising its main verbs up to T. In negativesentences, however, while English must employ do-support, MainlandScandinavian simply leaves the main verb in situ. While syntacticmovement approaches struggle quite a bit to make the requisite differ-ence between English and Mainland Scandinavian, all the present analysisneeds to assume is that the latter simply lacks a negative affix altogether– not only does Mainland Scandinavian not have an overt negative affixcorresponding to English -n’t, it does not have its null allomorph either.


37 What remains to be ensured is that when -n’t is used as the negative affix, not may not beused to occupy SpecNegP and that when ‘∅’ is used as the negative affix, not must be used tooccupy SpecNegP. This can be coded fairly straightforwardly in terms of a ‘doubly-filled Comp’effect: since present-day standard English is not a negative concord language, the head andspecifier of NegP may not both have a lexical manifestation at the same time; but in all con-structions with sentential negation there must be something which overtly signals the presenceof the negative feature, so when -n’t is not there, not is used, and vice versa.38 While the text discussion of do-support confines itself to negative contexts, it may be madeto carry over into the realm of do-support in inversion constructions (questions, negativeinversion) in the following way. Languages as diverse as Japanese and Hungarian are well knownfor having a question particle attached to the verb in (a subset of) questions (ka in Japanese,-e in Hungarian). There is variation among these languages when it comes to the exact distrib-ution of their overt question particles: Hungarian uses the question particle only in yes/no-questions, for instance. Suppose, now, that English has a null inflectional particle that obligatorilyattaches to the finite verb in questions and negative inversion constructions. If that null morphemehas the same attachment restriction as -n’t (i.e., it cannot affix to complex hosts, where ‘complex’stands for ‘θ-role assigning’), the distribution of do-support in questions and negativeinversion constructions can be analyzed along the same lines as negative do-support. I will notwork out the details here.

Absent a negative affix of any kind on the finite verb, we will find noeffects of the sensitivity of the negative affix to the complexity of its host.

I advance the possible extension of the analysis of un- to -n’t very ten-tatively here. Whether it is worth our while and genuinely opens up newperspectives on sentential negation and do-support is a question that futureresearch should look into.

References

Anderson, Stephen: 1982, ‘Where’s Morphology?’, Linguistic Inquiry 13, 571–612.Aronoff, Mark: 1976, Word Formation in Generative Grammar, MIT Press, Cambridge.Baker, Mark: 1985, ‘The Mirror Principle and Morphosyntactic Explanation’, Linguistic Inquiry

16, 373–415.Baker, Mark: 1988, Incorporation, The University of Chicago Press, Chicago.Barbiers, Sjef: 1996, The Syntax of Interpretation, Ph.D. dissertation, University of Leiden/HIL.Bartos, Huba: 2000, ‘Affix Order in Hungarian and the Mirror Principle’, in G. Alberti and I.

Kenesei (eds.), Approaches to Hungarian, Vol. 7, JATE, Szeged, pp. 53–70.Bennis, Hans, Norbert Corver, and Marcel den Dikken: 1998, ‘Predication in Nominal Phrases’,

Journal of Comparative Germanic Linguistics 1, 85–117.Beukema, Frits and Marcel den Dikken: 1989, ‘The Position of the Infinitival Marker in the

Germanic languages’, in D. Jaspers, W. Klooster, Y. Putseys and P. Seuren (eds.), SententialComplementation and the Lexicon, Foris, Dordrecht, pp. 57–75.

Blom, Alied and Saskia Daalder: 1977, Syntaktische theorie en taalbeschrijving, Coutinho,Muiderberg.

Bobaljik, Jonathan: to appear, ‘A–chains at the PF–interface: Copies and “Covert” Movement’,Natural Language and Linguistic Theory.

Borer, Hagit: 1991, ‘The Causative–Inchoative Alternation: A Case Study in ParallelMorphology’, The Linguistic Review 8, 119–158.

Borer, Hagit: 1999, ‘The Form, the Forming, and the Formation of Nominals’, ms., Universityof Southern California.

Borroff, Marianne Law: 2000, ‘Degree Phrase Inversion in the Scope of Negation’, ms., SUNYStony Brook.

Brody, Michael: 2000, ‘Mirror Theory: Syntactic Representation in Perfect Syntax’, LinguisticInquiry 31, 29–56.

Chomsky, Noam: 1970, ‘Remarks on Nominalization’, in R. Jacobs and P. Rosenbaum (eds.),Readings in English Transformational Grammar, Baisdell, Waltham, MA.

Chomsky, Noam: 1993, ‘A Minimalist Program for Linguistic Theory’, in K. Hale and S.J.Keyser (eds.), The View from Building 20, MIT Press, Cambridge, pp. 1–52.

Chomsky, Noam: 1995, The Minimalist Program, MIT Press, Cambridge.Chomsky, Noam: 1999, ‘Derivation by Phase’, MIT Occasional Papers in Linguistics 18;

published in M. Kenstowicz (ed.), Ken Hale: A Life in Linguistics, MIT Press, Cambridge,2001.

Chomsky, Noam: 2001, ‘Beyond Explanatory Adequacy’, MIT Occasional Papers in Linguistics20, 1–28.

Cornilescu, Alexandra: 1992, ‘Remarks on the Determiner System of Rumanian: TheDemonstratives al and cel’, Probus 4, 189–260.

Corver, Norbert: 1991, ‘Evidence for DegP’, NELS 21 Proceedings, GLSA, University ofMassachusetts at Amherst, 33–47.

Corver, Norbert: 1997, ‘Much-Support as a Last Resort’, Linguistic Inquiry 28, 119–164.


Déchaine, Rose-Marie: 1993, Predicates across Categories: Towards a Category-Neutral Syntax,unpublished Ph.D. dissertation, University of Massachusetts at Amherst.

Diesing, Molly: 1990: Indefinites, MIT Press, Cambridge.Dikken, Marcel den: 1995, ‘Copulas’, paper presented at GLOW Tromsø; ms., Vrije Universiteit

Amsterdam/HIL.Dikken, Marcel den: 1998, ‘Predicate Inversion in DP’, in A. Alexiadou and C. Wilder (eds.),

Possessives, Predicates and Movement in the DP, John Benjamins, Amsterdam, pp. 177–214.Dikken, Marcel den and Anikó Lipták: 1997, ‘Csoda Egy Nyelv – NominalBinternal Predication

in Hungarian’, in J. Coerts and H. de Hoop (eds.), Linguistics in the Netherlands 1997, JohnBenjamins, Amsterdam, pp. 61–72.

Dikken, Marcel den and Rint Sybesma: 1998, ‘“Take” Serials Light up the Middle’, paper pre-sented at GLOW Tilburg; ms., CUNY Graduate Center and University of Leiden/HIL.

DiSciullo, Anna-Maria and Edwin Williams: 1987, On the Definition of Word, MIT Press,Cambridge.

Fu, Jingqi, Thomas Roeper, and Hagit Borer: 2001, ‘The VP within Process Nominals: Evidencefrom Adverbs and the VP Anaphor Do-So’, Natural Language and Linguistic Theory 19,549–582.

Giannakidou, Anastasia: 1998, Polarity Sensitivity as (Non)Veridical Dependency, JohnBenjamins, Amsterdam.

Giannakidou, Anastasia: 2001, ‘The Meaning of Free Choice’, to appear in Linguistics andPhilosophy.

Graziano-King, Janine: 1999, Acquisition of Comparative Forms in English, unpublished Ph.D.dissertation, CUNY Graduate Center.

Grimshaw, Jane and Sten Vikner: 1993, ‘Obligatory Adjuncts and the Structure of Events’, inE. Reuland and W. Abraham (eds.), Knowledge and Language: Vol. II – Lexical andConceptual Structure, Kluwer, Dordrecht, pp. 145–159.

Guéron, Jacqueline and Teun Hoekstra: 1988, ‘Les chaînes-T et les verbes auxiliaires’, Lexique7, 61–85.

Guéron, Jacqueline and Teun Hoekstra: 1993, ‘The Temporal Interpretation of Predication’, inA. Cardinaletti and M.-T. Guasti (eds.), Small Clauses, Syntax and Semantics 28, AcademicPress, New York, pp. 77–108.

Hale, Kenneth: 2001, ‘Navajo Verb Stem Position and the Bipartite Structure of the NavajoConjunct Sector’, Linguistic Inquiry 32, 678–693.

Hale, Kenneth and Samuel Jay Keyser: 1993, ‘On Argument Structure and the Lexical Expressionof Syntactic Relations’, in K. Hale and S.J. Keyser (eds.), The View from Building 20, MITPress, Cambridge, pp. 55–109.

Halle, Morris and Alec Marantz: 1993, ‘Distributed Morphology and the Pieces of Inflection’,in K. Hale and S. J. Keyser (eds.), The View from Building 20, MIT Press, Cambridge, pp.111–176.

Hankamer, Jorge and Ivan Sag: 1976, ‘Deep and Surface Anaphora’, Linguistic Inquiry 7,391–427.

Hazout, Ilan: 1991, Verbal Nouns: Theta Theoretic Studies in Hebrew and Arabic, unpublishedPh.D. dissertation, University of Massachusetts at Amherst.

Hoeksema, Jack: 2000, ‘Negative Polarity Items: Triggering, Scope, and C-Command’, inL. R. Horn and Y. Kato (eds.), Negation and Polarity: Syntactic and Semantic Perspectives,Oxford University Press, Oxford, pp. 115–146.

Holmberg, Anders and Christer Platzack: 1995, The Role of Inflection in Scandinavian Syntax,Oxford University Press, Oxford/New York.

Horn, Laurence: 1989, A Natural History of Negation, The University of Chicago Press, Chicago.Jespersen, Otto: 1917, Negation in English and Other Languages, A.F. Høst, Copenhagen.Kang, Beom-mo: 1993, ‘Unhappier Is Different from Less Happy’, Linguistic Inquiry 24,

788–794.Kayne, Richard: 1994, The Antisymmetry of Syntax, MIT Press, Cambridge).


Kehler, Andrew and Gregory Ward: 1995, ‘On the Anaphoric Status of Do So’, paper presentedat the LSA Annual Meeting, New Orleans.

Kennedy, Chris and Jason Merchant: 1998, ‘Attributive Comparative Deletion’, ms.,Northwestern University and UC Santa Cruz.

Klima, Edward: 1964, ‘Negation in English’, in J. A. Fodor and J. J. Katz (eds.), The Structureof Language: Readings in the Philosophy of Language, Englewood Cliffs, pp. 246–323.

Kratzer, Angelika: 1989, ‘Stage and Individual Level Predicates’, in Papers on Quantification,NSF Grant Report, Department of Linguistics, University of Massachusetts at Amherst.

Ladusaw, William: 1979, Polarity Sensitivity as Inherent Scope Relations, unpublished Ph.D.dissertation, University of Texas at Austin.

Larson, Richard: 1998. ‘Temporal Adjectives and the DP Hypothesis’, ms., SUNY Stony Brook.Lebeaux, David: 1988, Language Acquisition and the Form of the Grammar, unpublished Ph.D.

dissertation, University of Massachusetts at Amherst.Lees, Robert: 1960, The Grammar of English Process Nominals, Mouton, The Hague.Linebarger, Marcia: 1980, The Grammar of Negative Polarity, unpublished Ph.D. dissertation,

MIT.Marantz, Alec: 1997, ‘No Escape from Syntax: Don’t Try Morphological Analysis in the Privacy

of Your Own Lexicon’, paper presented at the 21st Penn Linguistics Colloquium, appearingin Dimitriadis et al. (eds.), Penn Working Papers in Linguistics 4.

Murasugi, Kumiko: 1990, ‘The Derivation of Derived Nominals’, ms., MIT.Muysken, Pieter: 1981, ‘Quechua Word Structure’, in F. Heny (ed.), Binding and Filtering,

Croom Helm, London, pp. 279–328.Muysken, Pieter: 1988, ‘Affix Order and Interpretation: Quechua’, in M. Everaert, A. Evers, R.

Huybregts, and M. Trommelen (eds.), Morphology and Modularity: In Honour of HenkSchultink, Foris, Dordrecht, pp. 259–279.

Pesetsky, David: 1985, ‘Morphology and Logical Form’, Linguistic Inquiry 16, 193–246.Pollock, Jean-Yves: 1989, ‘Verb Movement, Universal Grammar, and the Structure of IP’,

Linguistic Inquiry 20, 365–424.Postal, Paul: 2001. ‘Overlooked Negative Polarity Items’, ms., New York University.Postma, Gertjan and Johan Rooryck: 1996, ‘Modality and Possession in NPs’, NELS 26.Randall, Janet: 1988, ‘Inheritance’, in W. Wilkins (ed.), Syntax and Semantics 21, Academic

Press, New York, pp. 129–146.Rice, Keren: 2000, Morpheme Order and Semantic Scope: Word Formation in the Athapaskan

Verb, Cambridge University Press, Cambridge.Roeper, Thomas: 2002, ‘Hidden Agents and Themes: What Controls Leftward Movement in

Nominalizations?’, paper presented at the CUNY Linguistics Colloquium, 21 February.Roeper, Thomas and Angeliek van Hout: 2000. ‘The Impact of Nominalization on Passive, -able

and Middle: Burzio’s Generalization and Feature-Movement in the Lexicon’, ms., Universityof Massachusetts at Amherst and Utrecht University.

Rozwadowska, Bozena: 1988, ‘Thematic Restrictions on Derived Nominals’, in W. Wilkins (ed.),Syntax and Semantics 21, Academic Press, New York, pp. 147–165.

Speas, Margaret: 1991a, ‘Functional Heads and the Mirror Principle’, Lingua 84, 181–214.Speas, Margaret: 1991b, ‘Functional Heads and Inflectional Morphemes’, The Linguistic Review

8, 389–417.Sportiche, Dominique: 1988, ‘A Theory of Floating Quantifiers and Its Corollaries for

Constituent Structure’, Linguistic Inquiry 19, 33–60.Sportiche, Dominique: 1998, Atoms and Partitions of Clause Structure, Routledge, London.Sproat, Richard: 1985, On Deriving the Lexicon, unpublished Ph.D. dissertation, MIT.Sproat, Richard: 1992, ‘Unhappier is not a “Bracketing Paradox” ’, Linguistic Inquiry 23,

347–352.Stiebels, Barbara and Dieter Wunderlich: 1994, ‘Morphology Feeds Syntax: The Case of Particle

Verbs’, Linguistics 32, 913–968.Stump, Gregory: 1991, ‘A Paradigm-Based Theory of Morphosemantic Mismatches’, Language

67, 675–725.


Travis, Lisa: 1984, Parameters and Effects of Word Order Variation, unpublished Ph.D. dis-sertation, MIT.

Williams, Edwin: 1981, ‘On the Notions “Lexically Related” and “Head of a Word”’, LinguisticInquiry 12, 245–274.

Wurmbrand, Susanne: 2001, Infinitives: Restructuring and Clause Structure, Mouton, Berlin.Zimmer, Karl: 1964, Affixal Negation in English and Other Languages, Supplement to Word,

Monograph #5.Zwart, C. Jan-Wouter: 1997, Morphosyntax of Verb Movement: A Minimalist Approach to the

Syntax of Dutch, Kluwer, Dordrecht.Zwicky, Arnold and Geoffrey Pullum: 1983, ‘Cliticization vs. Inflection: English n’t’, Language

59, 502–513.


Documents

Lexical Integrity, Checking, and the Mirror: A Checking Approach to Syntactic Word Formation