Abstract

This article addresses a proposal by Katzir (2011) that licensors are crucially involved in definiteness and gender marking in Danish, Icelandic, and Greek. We identify several morphological patterns in Icelandic and Danish that are predicted to be impossible under Katzir’s theory and argue that these languages do not support the existence of licensors. Instead, we propose an account within the framework of Distributed Morphology, arguing that the patterns of gender and definiteness marking seen in these languages are the result of competition for insertion between well-motivated Vocabulary items. In a sense, then, we argue for a more traditional and theoretically simpler analysis of agreement in the languages.

1 Introduction

The morphosyntax of Scandinavian DPs has been used to support a number of theoretical mechanisms, such as movement operations in morphology (Embick and Noyer 2001), syntactic alignment constraints (Börjars and Donohue 2000), and blocking of syntactic processes by morphological ones (Hankamer and Mikkelsen 2002, 2005). In this vein, Katzir (2011) proposes a reanalysis of definiteness marking and concord in Danish and Icelandic DPs in terms of licensors, a new category that is argued to be implicated in agreement processes alongside controllers, targets, and agreement morphology itself. Katzir’s proposal is attractive in that it unifies definiteness marking and gender, number, and case concord analytically, whereas much existing literature focuses exclusively on definiteness marking. Additionally, under Katzir’s analysis quite substantial morphosyntactic differences between Danish and Icelandic DPs are reduced to a single syntactic difference, namely, the structural position of attributive adjectives.

We argue here that Katzir’s analyses of Danish and Icelandic are empirically untenable, and we suggest that the morphosyntax of these languages does not, in the end, support the existence of licensors. First, the trouble with Katzir’s licensors is that they are at once too abstract—to account for the surface form of certain DPs, one needs to appeal to ad hoc phonological processes and irregularities—and too concrete—both Danish and Icelandic exhibit DP types that under Katzir’s analysis require a licensor for definiteness agreement, but there is no licensing morpheme in sight. Second, the assumption that a licensor must c-command the agreement morpheme it licenses requires unlikely and unmotivated structures for certain kinds of DPs. While these structures are required for Katzir’s analysis of Danish and Icelandic nominal morphology, they run into serious trouble with respect to the behavior of certain quantifiers in Danish and the outcomes of possible NP-ellipsis constructions in Icelandic.

Concretely, we document two kinds of problems for the licensor analysis. The first problem, which we dub too few licensors, arises in certain Danish and Icelandic DPs where the realizer of definiteness agreement is present, but unlicensed. This problem is resolved, we propose, by dispensing with licensors and accounting for definiteness agreement solely in terms of a definite controller. Both languages also display the inverse challenge, too many licensors, and in both languages this arises because the morphological exponent of concord is treated as a licensor, as opposed to a realizer of the agreement in question. Consequently, we advocate a leaner theoretical understanding of agreement: there are agreement controllers, agreement targets, and agreement features, but no fourth category of agreement licensors.

We begin, in section 2, by laying out the architecture and motivation for Katzir’s licensor analysis of agreement. In sections 3 and 4, we examine the licensor analysis of definiteness and gender marking in Danish and Icelandic DPs, respectively, and we argue that it comes up short once a broader set of data is considered. We close each section with a sketch of an alternative licensor-free analysis of the data under consideration. We conclude, in section 5, with a more general assessment of the licensor-based analysis.

2 Licensing Theory

The theory of agreement advanced by Katzir (2011) recognizes four kinds of elements involved in agreement relationships. In addition to ‘‘spreaders’’ (underlying loci of agreement features), the features themselves, and ‘‘realizers’’ (the agreement morphemes that express the features), agreement (more properly, concord) systems involve ‘‘licensors’’: elements (i.e., morphemes) that are neither spreaders nor realizers, but intermediaries necessary for the licensing of features and the expression of features by realizers. In this view, every agreement feature (whether expressed or not) is subject to three conditions:

  1. It must be spread by a spreader (the mechanism for this is unclear).1

  2. It must be c-commanded by a licensor for that particular feature.2

  3. It must not be c-commanded by more than one licensor of the relevant kind (a consequence of the ECONOMY condition3).

Realizers and licensors are morphemes, introduced by GRAMMAR via statements such as Katzir’s (8a) ‘‘-en is a suffix’’ and (8b) ‘‘-en can attach to N and d, but not to A,’’ which state conditions on the occurrence of a licensor, and via (10) ‘‘FDEF is realized as -e on all modifying adjectives,’’ which specifies how agreement features are realized. In general, it is very unclear what the syntax of licensors is.4 They seem to have no semantic function, and they do not appear to be syntactic heads—all of the licensors that Katzir proposes are affixes for which he does not propose a category label. In standard Distributed Morphology terms, they would have to be dissociated morphemes (Embick and Noyer 1999). But Katzir does not explicitly adopt either a Minimalist or a Distributed Morphology framework, so it remains unclear what to expect about the syntax of licensors.

In any case, there are mechanisms whereby elements of specified categories in a specified domain (the ‘‘spreading domain’’) receive a particular feature. A distinct mechanism allows licensors for that feature to attach to specified categories of elements, apparently without any direct relation to the distribution of features they are to license. Realizers are introduced by a mechanism that states the shape of the morpheme that realizes a particular feature on a particular kind of element, subject to the condition that that element is c-commanded by the host of the right kind of licensor.

The most striking feature of this theory of agreement is that the licensor for a given feature is not taken to express that feature, nor even (it seems) to necessarily be attached to an element bearing the feature. Licensors are introduced by rules of grammar that do not mention the feature that they license at all. This raises the question of how licensors themselves are licensed. They must be licensed in some way, since we do not find, for example, the FDEF licensor -en anywhere but in definite nominal phrases, where (absent ellipsis or other superficial deformation) there will be at least an N bearing FDEF that needs licensing. The answer must be that a licensor that doesn’t license anything would be superfluous, and the expression containing it ruled out by ECONOMY.

This means, in effect, that the same spreader that spreads a feature defines the domain in which the corresponding licensor can exist; and the spreading domain also determines where realizers can appear. So the spreader and the licensor together license the appearance of realizers, but the spreader also licenses the licensor.

The essence of a licensor account is that ‘‘spreaders’’ (whatever they are) license two kinds of elements: licensors and realizers. Both can be subject to GRAMMAR, which specifies where they can attach. Their interaction is then subject to two constraints:

  1. Every realizer of a feature F must be c-commanded by (the host of) at least one licensor for F.

  2. Structures with superfluous licensors are thrown out.

The claim Katzir makes is that this view of things leads to analyses superior to the best analyses available in a theory without the distinction between licensors and realizers.

3 Danish

Inflection in Danish DPs is sensitive to definiteness (indefinite vs. definite), number (singular vs. plural), and gender (common vs. neuter). Definiteness is inherent to D and is marked on either D or N, and also marked on adjectives. Gender is inherent on N and marked on D and A, and number is marked on D, A, and N. There is no morphological case in Modern Danish, except on pronouns. The table in (1) gives the forms for the articles, and the table in (2) the inflectional paradigm for attributive adjectives (examples of inflected DPs are provided below).

(1)

graphic

(2)

graphic

The traditional understanding is that adjectives show gender and number concord, whereas definiteness marking on A is characterized in terms of strong (indefinite) declension versus weak (definite) declension. In contrast, Katzir analyzes the entire paradigm in terms of agreement with privative features—specifically, a definite feature FDEF and a neuter feature FNt. Katzir does not discuss number, but since plural is the overtly marked value, it would presumably involve a privative plural feature FPl.

3.1 Definiteness Marking: Too Few Licensors

The licensing analysis of Danish definiteness marking inside DPs assumes a definiteness feature, FDEF, which is associated syntactically and semantically with a definite head, which Katzir tentatively identifies as D (see his discussion of this and other analytic possibilities on page 48 and in footnotes 15 and 20). From there, FDEF spreads to all modifying adjectives, where it is realized as the suffix -e, and to the head noun, where it has no morphological realization. The licensor for FDEF is the morpheme -en, which may attach to the definite head, yielding a surface definite article den, as well as to the head noun, yielding a suffixed definite noun (N-en). Only in the former position does -en c-command attributive adjectives, which explains the obligatory use of the prenominal article in the presence of attributive adjectives: these adjectives bear FDEF, which must be licensed under c-command, and only by attaching the licensor -en to D is that structural requirement met. Double definiteness marking is ruled out by Katzir’s ECONOMY condition on licensors: if two structures are identical except for licensors, the one with more licensors is ungrammatical (Katzir’s (12); see footnote 3 above).5 The structure Katzir assumes for simple definite DPs with adjectives is given in (3) (see Katzir’s (16), p. 54).

(3)

graphic

This analysis accounts well for the data in (4)–(7).

(4)

  • den gaml-e hest

  • the old-DEF horse

  • ‘the old horse’

(5)

  • *gaml-e hest-en

  • old-DEF horse-DEF

  • Intended: ‘the old horse’

(6)

  • hest-en

  • horse-DEF

  • ‘the horse’

(7)

  • *den (gaml-e) hest-en

  • the old-DEF horse-DEF

  • Intended: ‘the old horse’

In (4), the definite head has spread agreeing definiteness features onto A (gaml) and N (hest). Both instances of FDEF are licensed by -en (attached to a dummy d to form den) under c-command, accounting for the well-formedness of (4) and the presence of the -e suffix on the adjective. (5) is ill-formed because -en is too low in the structure (attached to N) to c-command the adjective, and thus the definiteness feature on A is unlicensed. When no adjective is present, as in (6), only the definiteness feature on N needs licensing and that is accomplished by suffixing -en to N itself. Finally, (7) is ungrammatical because it contains two licensors where only one is needed: the -en that attached high (to dummy d-) is in a position to license FDEF on both A and N, so the lower instance of -en on N is not needed. Redundant licensing is ruled out by ECONOMY.

In the remainder of this section, we show that this licensing analysis of definiteness agreement cannot be maintained once we consider a larger set of DP types.

First, note that adjectives in possessive DPs show the same inflection as adjectives in DPs headed by the definite determiner; compare (8) with (4).

(8)

  • min gaml-e hest

  • my old-DEF horse

  • ‘my old horse’

Under the licensing analysis, the presence of the definiteness-realizing morpheme -e on the adjective in (8) indicates that (8) contains the definiteness-licensing morpheme -en as well. Following Katzir’s decomposition of the definite article den into a dummy d and -en, we can decompose min in (8) into a possessive root mi- and the definite licensor -en.6 The possessive pronoun is high enough in the structure (Spec,DP) to c-command attributive adjectives, and, by virtue of attaching to the possessive pronoun, so is the licensor -en, and the definite agreement on the adjective is properly licensed. So far so good. But then consider the licensing of definite -e in attributive adjectives in possessive DPs with nonpronominal possessors, as in (9).

(9)

  • Peter-s gaml-e hest

  • Peter-POSS old-DEF horse

  • ‘Peter’s old horse’

By the reasoning above, we need to posit the presence of the definite licensor -en to account for the definite agreement -e on the adjective. Moreover, the licensor -en must c-command the adjective, since licensing takes place under c-command. The possessive clitic -s c-commands the adjective, so that would be an appropriate host for the licensing morpheme, but there is no indication that the surface form -s contains the definite licensor -en. An alternative analysis is that possessive -s is itself a licensor for the definiteness feature on A. That analysis respects the surface form of the possessor ( Peters, not *Peter-en-s or *Peter-s-en) and accounts for the grammatical presence of definite agreement on the adjective in (9). Positing a second licensor for FDEF seems within the spirit of Katzir’s analysis; the main difference between the two licensors of FDEF is that -s is also implicated in the syntax of possession, whereas the sole function of -en is licensing of definite agreement.

If -en and -s are both licensors of definite inflection on A, as suggested by the identical inflection of A in (4), (8), and (9), we expect them to not cooccur, given Katzir’s ECONOMY condition. But they do, as shown in (10), which contains two definite licensors: -s and -en.7

(10)

  • Peter-s den gaml-e hest

  • Peter-POSS the old-DEF horse

  • ‘Peter’s old horse’

Possessive pronouns similarly cooccur with the definite article, as shown in (11). Under the decomposition of min into mi- plus -en, required for (8), this DP thus contains two separate instances of the definite licensor -en.

(11)

  • min den gaml-e hest

  • my the old-DEF horse

  • ‘my old horse’

The possessive examples in (8)–(11) thus create a dilemma for the licensing analysis of Danish definiteness marking: either there is no definite licensor in (8) and (9), and we have no account of these examples, or there are too many licensors in (10) and (11), and we have no account of their well-formedness. (10) and (11) violate ECONOMY just as (7) does, and yet the former are grammatical and the latter is not. Note that this dilemma is entirely absent from licensor-free accounts of definiteness inflection: all we need to say is that definite D (den) and possessive D (-s and min) are all able to control definiteness agreement on A.

Before we leave possessive DPs, we want to respond to a reviewer’s query about the soundness of this argument. The reviewer’s point is that if the qualified definiteness possessive construction in (10) and (11) means something different from the regular possessive construction, perhaps the two structures are not subject to ECONOMY and consequently there would not be too many licensors in the qualified definiteness possessives in (10) and (11). This is an important point. In his discussion of qualified definiteness possessives like (10) and (11), Hansen (1994:113) notes that such possessives are only felicitous when the referent is assumed to be known to both speaker and hearer. The regular possessive construction is not subject to any pragmatic restrictions; it can be used whether the intended referent is known or new. This latter fact shows that Danish does have the morphosyntactic means to express the relevant pragmatic meaning (known referent) without den. Thus, the possessive structure without den and the possessive structure with den can express the same meaning, and when they do, they are consequently subject to ECONOMY. We conclude that the existence of the two possessive constructions does present the licensing analysis with a dilemma: if definite -e is licensed in the regular possessive construction, as it must be in order to account for (8) and (9), there are one too many -e licensors in the qualified definiteness possessives in (10) and (11)—and yet these are grammatical.

The second type of DP that we want to draw attention to is vocative DPs, which to our knowledge have not played much of a role in the literature to date, but are directly relevant for Katzir’s analysis. In vocative DPs, attributive adjectives obligatorily bear the definite -e suffix and do so in the absence of a definite article, as shown in (12).

(12)

  • a.

    kær-e ven

    dear-DEF friend

    ‘dear friend’

  • b.

    gaml-e dreng

    old-DEF boy

    ‘old boy’

  • c.

    sød-e Lise

    sweet-DEF Lise

    ‘sweet Lise’

This configuration poses a special challenge for the licensing account of definite agreement on adjectives. First, a null definite D head has to be posited to host and spread FDEF onto the adjective. That much is shared with licensor-free accounts, unless they take the -e form of the adjective as the unmarked, default form. In addition, the licensing analysis must posit a null licensor for FDEF, since there is no overt element present in these DPs that could plausibly serve as the licensor. If we admit a null licensor for definite agreement on attributive adjectives, then we must ask why that licensor cannot be used in definite DPs in argument position. That is, why is (13) ill-formed?

(13)

  • *Jeg besøgte {kære ven / gamle dreng / søde Lise}.

  • I visited dear friend / old boy / sweet Lise

  • Intended: ‘I visited my dear friend/the old boy/the sweet Lise.’

This points to a general problem: the licensing approach ties definite inflection on A directly to the presence of a particular morpheme, namely, -en; however, -e in fact has a wider distribution than -en, suggesting that the two must be disentangled, and the distribution of -e tied solely to the presence of the definite feature FDEF. The data above show that FDEF is inherent on a range of functional nominal heads, including the definite article, and possessive and vocative D heads. Consequently, these heads all control definite agreement on attributive adjectives.

3.2 Neuter Gender: Too Many Licensors

Katzir identifies the suffix -t as a licensor of neuter gender agreement in Danish, whereas previous analyses treat it as a realizer of such agreement.8 Neuter gender agreement arises from a neuter gender feature FNt being spread, presumably from N, onto A and D. In definite neuter DPs, -t occurs exactly once, on the definite article, and attributive adjectives bear the -e suffix familiar from the discussion of definite agreement above (see (14a)). In indefinite neuter DPs, -t occurs on the indefinite article and on each attributive adjective (see (14b)).

(14)

  • a.

    de-t stor-e gul-e håndklæde

    the-NT big-DEF yellow-DEF towel

    ‘the big yellow towel’

  • b.

    e-t stor-t gul-t håndklæde

    a-NT big-NT yellow-NT towel

    ‘a big yellow towel’

Katzir suggests that this morphological difference stems from a syntactic difference: the definite article, but not the indefinite one, c-commands attributive adjectives. This is so because only the definite article resides in D; the indefinite article is not a D, but ‘‘similar to an adjective’’ (Katzir, p. 61).

Thus, a definite DP like det gamle hus ‘the old house’ has something like the structure in (15), whereas the corresponding indefinite DP, et gammelt hus ‘an old house’, has something like the structure in (16) (we say ‘‘something like’’ because Katzir does not commit to the identity of the functional head that hosts the definite licensor -en; here we treat den as D, which is one of the options suggested by Katzir in his footnote 20).

(15)

graphic

(16)

graphic

In (15), the licensor -t in D c-commands the attributive adjective by virtue of D’s c-commanding the adjective.9 The single licensor -t in D thereby licenses the neuter agreement feature that has been spread onto the adjective. Consequently, no additional licensors of neuter on the adjective are required, accounting for the well-formedness of (14). By ECONOMY, no additional neuter licensor is possible, accounting for the impossibility of *de-t gamle-t hus, where the attributive adjective contains a redundant licensor for the neuter gender feature. In (16), however, the licensor -t on et does not c-command the adjective gammel, because the adjective en itself does not ccommand the lower adjective. Consequently, each adjective in an indefinite DP needs its own -t licensor. This is what accounts for the obligatory -t suffix on each adjective in (14b).

At this point, we would like to draw attention to the quantifier hver ‘each’, which Katzir does not discuss, but which is relevant in that it behaves exactly like the indefinite article. Any adjective following hver shows neuter inflection if the head noun is neuter (17), just as any adjective following the indefinite article does (18).

(17)

  • a.

    hver-t ny-t stor-t hus

    each-NT new-NT large-NT house.NT

    ‘each new large house’

  • b.

    *hver-t ny stor hus

    each-NT new large house.NT

    Intended: ‘each new large house’

(18)

  • a.

    e-t ny-t stor-t hus

    a-NT new-NT large-NT house.NT

    ‘a new large house’

  • b.

    *e-t ny stor hus

    a-NT new large house.NT

    Intended: ‘a new large house’

As these examples further show, hver and en themselves also inflect for gender, bearing the neuter suffix -t in the presence of the neuter head noun hus ‘house’. To account for the obligatory presence of -t on the adjectives following hver in (17a), the licensing analysis must consequently posit a structure in which hver does not c-command these adjectives. This is so because if hver did c-command these adjectives, the -t attached to hver would c-command the neuter gender feature FNt on ny and stor and thereby license FNt on ny and stor, obviating the need for separate licensing -ts on ny and stor. Thus, if hver c-commands ny and stor, ECONOMY makes the wrong predictions about (17): (17b) is predicted to be grammatical and (17a) to be ungrammatical, because (17b) has fewer FNt licensors—namely, one, whereas (17a) has three. (Recall that under Katzir’s analysis -t is the licensor for FNt and the realizer of FNt is systematically null.) Note that similar examples where ny and stor bear the -e inflection are also ungrammatical, since -e is definite and the DPs in (17) and (18) are indefinite. So to account for the attested pattern of inflection, the licensing analysis must posit a structure for (17) in which hver does not c-command the following adjectives. Extending Katzir’s analysis of the indefinite article in (16) above, we arrive at the structure in (19) for (17a).10

(19)

graphic

In (19), hver does not c-command ny or stor, nor do the adjectives c-command each other. Consequently, a licensor -t is required on each of these three elements. The problem with (19) is that it treats hver as a fully formed adjunct phrase, which seems implausible to us, for the same reasons that it would be implausible to treat the English definite article the as a fully formed adjunct phrase. Neither can occur on its own, and both occur in a fixed position within DP—namely, preceding all adjectives. Neither can be coordinated with another prenominal phrase (such as an adjective), and neither can occur solo in predicative position. These are all properties that set hver and the apart from adjectives, and they suggest that hver and the are heads that subcategorize for an NP. Their behavior is accounted for immediately if Danish hver and English the are functional heads in the extended nominal projection in the sense of Grimshaw 2005, but problematic if hver and the are phrasal adjuncts to NP, as shown for hver in (19). We thus conclude that (19) is not a reasonable syntactic structure for (17) and, as a consequence, the pattern of inflection in (17) stands as a challenge to the licensing analysis.11

This brings us to a more general problem with Katzir’s analysis of Danish neuter -t. To ensure that a preceding attributive adjective does not c-command a following attributive adjective, Katzir must assume a unary projection between A and AP, as seen in the structures in (16) and (19). If that unary projection were not there, A would c-command NP and so would the licensor -t on that A, licensing FNt on any following adjectives. In the case of (16), this would yield the string in (20) with no -t on gammel. And in the case of (19), it would yield the string in (21) with no neuter -t on stor, because the -t on ny licenses FNt on both adjectives.12 Both strings are ungrammatical.

(20)

  • *e-t gammel hus

  • a-NT old house.NT

  • Intended: ‘an old house’

(21)

  • *hver-t ny-t stor hus

  • each-NT new-NT large house.NT

  • Intended: ‘each new large house’

Unary projections are incompatible with some Minimalist conceptions of bare phrase structure, in which projection is the result of Merge (Chomsky 1995).13 A reliance on unary projections of this kind thus could be a theoretical liability. One of Katzir’s reviewers notes this issue with respect to the fact that the licensing analysis needs to distinguish N from NP by unary projection in order to avoid generating the ungrammatical *stor-e hest-en ‘big-DEF horse-DEF’ ( p. 55). In response, Katzir says (p. 55):

I should also mention that, while the labels N and NP may suggest that the two projections are related, all that matters for the current proposal is that they are distinct. One way to ensure that they are distinct is to let categories like N project phrasal categories like NP, but the discussion of licensing in this section would remain unaffected if NP were a different category, as long as this category is distinct from N. Such a category can be projected, perhaps, from a null head that takes the highest segment of N as its sister.

One could apply the same fix to APs and avoid a unary projection of AP from A by positing a null head that takes an A complement and then adjoins to NP. Without independent support for the existence of such a null head, this reanalysis is unconvincing, and the unary projection problem stands.

To summarize, Katzir’s licensor-based analysis of Danish definiteness and gender marking fails in four respects. First, it does not account for the behavior of possessive DPs, which have either too few or too many licensors for the agreeing definiteness feature on A. Second, it does not account for the use of definite agreement on adjectives in vocative DPs, which do not contain the definite licensor -en. Third, it does not account for the distribution of neuter -t in DPs headed by the quantifier hver. Fourth, the analysis of neuter -t in DPs with multiple adjectives relies on unary projection from A to AP, which is at odds with current theories of phrase structure.

3.3 Licensor-Free Concord

Katzir identifies definite -en and neuter -t as licensors, of FDEF and FNt respectively. Above, we argued that neither identification is viable, and we proposed that -en (or, rather, the definite article den) is a controller of definite agreement, along with possessive and vocative D, and that -t is a realizer of neuter agreement. We conclude our discussion of Danish by defending the status of -t as a realizer, since Katzir explicitly challenges that assumption.

Katzir’s point is that, if -t is a realizer of neuter agreement, we would expect -t to surface on adjectives also in definite DPs headed by neuter nouns, but such adjectives only bear the definite -e suffix. Compare the inflections on the adjective in the indefinite (22) and the definite (23).

(22)

  • et gul-t håndklæde

  • a yellow-NT towel

  • ‘a yellow towel’

(23)

  • det gul-e håndklæde

  • the yellow-DEF towel

  • ‘the yellow towel’

Under Katzir’s analysis, licensing -t is required on A in (22) because the licensing -t on e- is too low in the structure to c-command FNt on A, and ruled out in (23) because the licensing -t on the definite article is high enough (in D) to c-command FNt on A. Above, we criticized this analysis on syntactic grounds: it relies on analyzing the indefinite article as an adjunct. Such an analysis does not easily extend to hver ‘each’, which apparently fails to license FNt on any As even though its syntax makes it seem very similar to den. Here, we want to draw attention to bare plurals, which behave like singular definite DPs, but do not contain a c-commanding licensor for FNt on A. The relevant pattern is illustrated in (24).

(24)

  • a.

    *{gul-t/-t-e/-e-t} håndklæd-er

    yellow-NT/-NT-PL/-PL-NT towel-PL

    Intended: ‘yellow towels’

  • b.

    gul-e håndklæd-er

    yellow-PL towel-PL

    ‘yellow towels’

The grammatical realization of a plural neuter DP has just the plural suffix -e on the attributive adjective, as in (24b). Any attempt to add the neuter -t is ungrammatical, as seen in (24a). This is strikingly like the situation in definite singular neuter DPs; see (23), where A bears just definite -e and no neuter -t. However, Katzir’s account of the definite singulars cannot be extended to the plural indefinites, since there is no t-bearing c-commanding D to license FNt on A in (24b).

This observation casts further doubt on the licensor analysis of -t, but it also presents a problem for the realizer analysis of -t that we advocate: why is adjectival neuter gender agreement only realized in the form of -t in singular indefinite DPs? The traditional answer invokes syncretism: gender agreement is neutralized in both definite and plural DPs (Hansen and Heltoft 2011: 227). The syncretic form -e can thus realize a multitude of feature combinations, as seen in the paradigm in (25) (repeated from (2)).

(25)

graphic

Distributed Morphology offers a way to improve upon the traditional analysis, while retaining a realizer analysis of neuter -t, alongside plural/definite -e. The analysis assumes full syntactic agreement on A for definiteness, number, and gender, and relies on Pāṇinian competition between variously specified Vocabulary items (VIs) to derive the impoverished morphological realization of this agreement.14 We assume that -t and -e are the morphological realization of postsyntactic Agr nodes (Noyer 1997), which are adjoined to their hosts. Concretely, we posit the VIs in (26) to account for Danish adjectival inflection.

(26)

  • a.

    Agr, [−DEF, SG, NT] ↔ -t

  • b.

    Agr, [−DEF, SG] ↔ -Ø

  • c.

    Agr ↔ -e

The [−DEF] feature is copied from D, which is inherently [±DEF]. Note that we assume that [DEF] is a binary rather than a privative feature. The reason for this is that the simplest way to describe the adjectival inflection pattern is to treat the ‘‘definite’’ -e form as the elsewhere case, which is implemented in our Distributed Morphology account via underspecification.

Neuter -t is the most highly specified VI and thus will only surface in indefinite singular neuter DPs like (22). Neuter -t cannot surface in plural DPs, because the insertion rule for -t specifies singular. That is what accounts for the ill-formedness of (24a). Similarly, -t cannot surface on A in definite DPs, since -t is specified as indefinite. This is what accounts for Katzir’s observation that -t is absent in definite singular neuter DPs like (23).

The feature specifications of the other two VIs, -Ø and -e, also match the adjectival feature specification in indefinite singular neuter DPs like (22); however, by the Subset Principle (Halle 1997), each of them loses out to -t, because -t is more highly specified than either of them. This accounts for the ungrammaticality of the examples in (27) and (28).

(27)

  • *et gul-e håndklæde

  • a yellow towel

  • Intended: ‘a yellow towel’

(28)

  • *et gul-Ø håndklæde

  • a yellow towel

  • Intended: ‘a yellow towel’

The -e inflection on A in a definite singular DP like (23) is due to insertion of the VI in (26c), which is the only one that matches the definite feature on A. Similarly, -e is the only option in plural DPs, since the featural specifications of -t and -Ø both require singular specification of the adjective. This accounts for the pattern in (24), which Katzir’s proposal does not account for.

Indefinite singular common gender DPs show zero inflection on A.

(29)

  • en gul-{Ø/*t/*e} vaskeklud

  • a.CG yellow wash.cloth.CG

  • ‘a yellow washcloth’

Here, -t is impossible, because it has a conflicting gender specification, and -Ø wins out over -e because -Ø is more highly specified than -e. In all other cells of the paradigm (25), including indefinite plural DPs, the adjectival inflection is -e—that is, the elsewhere form in our analysis.

We put forth this alternative analysis of adjectival agreement to counter Katzir’s suggestion (fn. 29) that existing analyses of the alternation between pre- and postnominal definiteness marking, like that of Hankamer and Mikkelsen (2005), are difficult to integrate with an analysis of adjectival agreement. The licensor-free analysis of adjectival agreement sketched above is fully compatible with the analysis of definiteness marking developed in Hankamer and Mikkelsen 2005. A complete analysis should specify the syntactic mechanism for adjectival agreement and also account for the inflection on articles, possessive pronouns, and quantifiers like hver, but we leave that for a future occasion and turn now to Icelandic.

4 Icelandic

As Katzir notes, the structural position of adjectives (adjuncts to NP) is crucial for his account. In that position, they are not c-commanded by the suffixed article, which means the suffixed article cannot license the FNt feature on adjectives. Under Katzir’s approach, this explains the fact that adjectives are only compatible with the prenominal definite article. To provide further support for licensors, Katzir notes that changing the position of adjectives slightly would lead to surprisingly different predictions about patterns of definiteness marking. Katzir thus brings Icelandic into the picture, proposing an analysis of Icelandic whereby adjectives are sisters to nouns instead of being adjoined to NP as in Danish. This is schematized in (30).

(30)

graphic

Notice that adjectives in this position are in the c-command domain of the suffixed definite article (-in in Icelandic) under Katzir’s definition: the definite article -in c-commands everything that is c-commanded by its attachment site, which we take to be the N sister of AP in (30). For Katzir’s analysis of Danish, it was crucial that adjectives adjoined to NP be outside of the suffixed article’s c-command domain.

The prediction that follows from this change is that Icelandic would have the same patterns of gender and definiteness marking as Danish, with the exception that all instances of FDEF and FC2 (a combination of gender, number, and case, which is the counterpart to FNt in Danish) on the adjectives could be licensed by the suffixed article. Thus, as Katzir puts it, ‘‘there will be no need for a prenominal definiteness marker’’ (p. 64). This is indeed what we find: the suffixed article is used even with adjectival modifiers, as (31) shows.15

(31)

  • a.

    hinn gul-i hest-ur

    the yellow-NOM.M.SG.DEF horse-NOM.M.SG

    ‘the yellow horse’

  • b.

    gul-i hest-ur-in-n

    yellow-NOM.M.SG.DEF horse-NOM.M.SG-DEF-NOM.M.SG

    ‘the yellow horse’

  • c.

    hest-ur-in-n

    horse-NOM.M.SG-DEF-NOM.M.SG

    ‘the horse’

  • d.

    *hinn (gul-i) hest-ur-in-n

    the yellow-NOM.M.SG.DEF horse-NOM.M.SG-DEF-NOM.M.SG

    Intended: ‘the yellow horse’16

It is important to note here that, although (31a) is grammatical, it is not in free variation with (31b); the prenominal article and the suffixed article are not completely equivalent from a semantic point of view (Thráinsson 2007). The two are therefore not in competition in Katzir’s sense. We will essentially ignore the prenominal article here, but interested readers should consult Sigurðsson 2006, Thráinsson 2007, and Pfaff 2009, to appear, for thorough discussion. The Icelandic patterns are schematized in (32).

(32)

  • a.

    Indefinite: [Adj – C2] [N – C1]

  • b.

    Definite: [Adj – WK] [N – C1 – in – C2]

In the schematic representation in (32), C2 corresponds to Danish -t, -in corresponds to Danish -en, and WK corresponds to the Danish FDEF realizer -e.17 C1 is a suffix that has no correlate in Danish. C1 attaches only to nouns (including proper nouns) and indicates gender, number, and case.

Katzir asserts that Icelandic is essentially Danish′: a language with the same spreaders, realizers, and licensors as Danish, but in which adjecti Danish′ves occupy a different structural position. If this were so, then it would provide support for Katzir’s licensors as a part of grammar rather than a quirk of Danish. In this section, we will first show that licensors in Icelandic suffer from the same general problems as licensors in Danish. Icelandic is an excellent test case for the claim that what are traditionally described as agreement suffixes/realizers (Danish -t and Icelandic C2) are actually licensors, since the nominal morphology of Icelandic is much less impoverished than that of Danish. When we consider C2, we again find instances of too few licensors and instances of too many licensors.18 The fact that the licensing approach to agreement runs into the same problems in Icelandic despite the change in adjective position suggests that the licensing approach is in systematic trouble: there are no morphemes in either language that behave as licensors are supposed to behave. Thus, we conclude, there is no justification for the additional functional element that Katzir proposes. Furthermore, we show that empirical evidence from other domains does not support Katzir’s structural assumptions about adjectives in Icelandic, thus providing further evidence against an account of Icelandic as Danish′.

4.1 The C2 Suffix Is Not a Licensor

The suffix that Katzir identifies as C2 is traditionally described as an agreement suffix, indicating the gender, number, and case of the DP in which the C2 suffix appears. Katzir proposes that the C2 suffix is just like Danish’s -t suffix in being a licensor. However, it is worth pointing out that the C2 suffix is much more prevalent than the -t suffix. Icelandic distinguishes four cases (nominative, accusative, dative, and genitive), three genders (masculine, feminine, and neuter), and two numbers (singular and plural). While Danish -t only appears in DPs that are neuter and singular, C2 suffixes appear for every combination of gender, number, and case. Thus, in contrast to what we find with -t, neither gender, number, nor case values will affect whether we see a C2 suffix on adjectives. The full paradigm for C2 suffixes is given in (33).

(33)

graphic

It will be clear in section 4.1.2 why we refer to these as the ‘‘default’’ concord markers.

There are two reasons to suspect that C2 is not a licensor responsible for licensing FC2 on modifiers in Icelandic. First, there are several different examples where there appears to be an instance of FC2 but either (a) no C2 suffix at all, or (b) no C2 suffix that is clearly in a position to license features. As with vocatives in Danish, these are instances of too few licensors, and we would expect these structures to be ruled out by GRAMMAR, which requires all instances of FC2 to be licensed. Second, concord in Icelandic presents several examples where C2 surfaces even though its licensing capabilities are not needed owing to the presence of another C2 suffix that is capable of licensing every instance of FC2 in the structure. Reminiscent of Danish gender agreement on A in the presence of et or hvert, these are instances of too many licensors, and we expect these to be ruled out by ECONOMY, which prefers structures with fewer licensors. We will consider these in turn, concluding this section by arguing that it is perfectly reasonable to treat C2 as a realizer, and in fact, this gives us a simple way to explain the fact that C2 and WK are never realized on the same node at the same time.

4.1.1 Weak Adjectives: Too Few Licensors

Following Katzir’s analysis of Danish, let us assume that there is a spreader (or spreaders) responsible for spreading gender, number, and case features to the various elements in the Icelandic DP.

(34) -in, all modifying As, Qs, demonstratives, numerals, and possessive pronouns have FC2- ξ, where ξ corresponds to the gender, number, and case values of the noun.

This rule is on a par with the rule of FNt spreading for Danish (Katzir’s (29), p. 60). Just as in Danish, all instances of FC2 must be licensed by a licensor, which Katzir suggests is the suffix he identifies as C2. In this section, we will present some examples of phrases where there appear to be too few licensors—specifically, instances where an adjective appears with the WK suffix, traditionally called the ‘‘weak’’ form, with no instance of a c-commanding C2 licensor. This is problematic for the licensing analysis, because according to (34), every adjective should have an instance of FC2, and every instance of FC2 must be licensed by a C2 suffix. The examples in this section apparently have unlicensed instances of FC2 and should thus be ruled out by GRAMMAR, contrary to fact.

The first instance comes from possessive constructions. With pronominal possessors, the definite article is generally required on possessed nouns (see Sigurðsson 2006, Thráinsson 2007 for details).

(35)

  • a.

    bók-in mín

    book-the my

    ‘my book’

  • b.

    hús-ið mitt

    house-the my

    ‘my house’

However, there are (at least) two classes of nouns that are exceptions to this generalization. Certain kinship terms (36) and certain abstract nouns (37) either disallow or do not require the definite article in possessive constructions.

(36)

  • a.

    vinur(*-inn) minn

    friend-the my

    ‘my friend’

  • b.

    systir(*-in) mín

    sister-the my

    ‘my sister’

(37)

  • a.

    skoðun(*-in) mín

    opinion-the my

    ‘my opinion’

  • b.

    hugmynd(*-in) mín

    idea-the my

    ‘my idea’

Furthermore, as a reviewer notes, even concrete nouns can appear without the definite article, especially when they refer to something slightly more abstract.19 These examples from Thráinsson 2007:93 are particularly illuminating.

(38)

  • a.

    Bók-in / Bók mín um íslenska setningafræði fékk góða dóma.

    book-the / book my about Icelandic syntax got good reviews

    ‘My book about Icelandic syntax got good reviews.’

  • b.

    Hvar er bók -in / ?*bók mín um íslenska setningafræði?

    where is book-the / ?*book my about Icelandic syntax

    ‘Where is my book about Icelandic syntax?’

In (38a), where bók ‘book’ refers in some sense to the work or content itself, the definite article can be left out. However, if bók refers to an actual copy of a book as in (38b), then the indefinite form (i.e., with no definite article) ‘‘seems quite odd’’ (Thráinsson 2007:93). Interpretational differences aside, the bottom line is that the possession system of Icelandic countenances both nouns with the definite suffix and nouns without it.20

When these possessed nouns are modified by adjectives, those adjectives must (or can) show the weak form, whether or not they have the definite article.

(39)

  • a.

    best-i vinur minn

    best-WK friend my

    ‘my best friend’

  • b.

    stór-a systir mín

    big-WK sister my

    ‘my big sister’

(40)

  • a.

    góL-a bók-in mín

    good-WK book-the my

    ‘my good book’

  • b.

    rauð-a hús-ið mitt

    red-WK house-the my

    ‘my red house’

According to Katzir’s assumptions, the adjectives in (39) have instances of FC2 that must be licensed by a c-commanding licensor: the C2 suffix. Since the adjectives themselves do not have C2 suffixes in (39), there are two remaining possibilities: either the possessive pronoun (in this case, minn) has a C2 suffix, or the noun itself has one.

In fact, as one of Katzir’s reviewers notes (fn. 44), the possessive pronoun does have what looks like a C2 suffix. Possessive pronouns decline almost exactly like the definite article. Katzir argues convincingly that the definite article in Icelandic also bears a C2 suffix, and under Katzir’s account, this explains the loss of the C2 suffix on adjectives when the definite article is present. Following Sigurðsson (1993), Vangsnes (1999), Julien (2005), and Norris (2011), we assume that a reasonable location for Icelandic possessors is Spec,NP. One might propose a base-generated structure along the lines of (41), which places the C2 suffix on the possessive pronoun in a position to c-command the adjective, thus licensing its instance of FC2.

(41)

graphic

This works well for the DPs in (39): the C2 suffix on minn c-commands the instance of FC2 on the adjective, eliminating the need for an additional C2 suffix on the adjective. However, it runs afoul of the more standard possessive constructions in (35). Recall that most nouns require the presence of the suffixed article in addition to the possessive pronoun. The C2 suffix in (41) ccommands not only the adjective, but the noun as well. This in turn would mean that it would c-command the definite suffix attached to the noun. We therefore expect the C2 suffix on minn to render the C2 suffix on the definite article superfluous. This is clearly not the case: whenever the definite article is present, it must have the agreement suffix identified by Katzir as C2. This is similar to the problem we saw for the Danish qualified definiteness construction in (10) and (11).

However, as Katzir points out in footnote 44, the general consensus is that the order of adjective, noun, and possessor in Icelandic possessives is not base-generated. There are two types of proposals: either (a) the noun raises above the possessor and, in so doing, becomes left-adjacent to the adjective (Vangsnes 1999:117–133, Julien 2005:3–11, Norris 2011:110–111), or (b) the adjective and the noun form a unit and raise above the possessor together (Magnússon 1984:100, Sigurðsson 1993:193–195, 2006:10–11, Pfaff 2009:59–80). We believe an analysis of type (a) is incompatible with Katzir’s assumptions, insofar as the definite suffix attached to N must ccommand adjectives. The authors who propose an analysis of type (b) differ in how they formalize the adjective-noun unit, as a consequence of the other assumptions and proposals they make. For concreteness, we assume (like Pfaff (2009), unlike Sigurðsson (1993, 2006)) that the adjective and the noun form a phrasal unit that moves to the specifier of some functional projection, G.21 For rauða húsið mitt ‘my red house’, we have the derivation in (42). Note that, following Katzir’s decomposition, we have decomposed - into -in and -t, and we have decomposed mitt into min and -t.

(42)

graphic

As a result of this movement, the C2 suffix attached to POSS no longer c-commands the instance of FC2 on the adjective rauða. In his discussion of Greek, Katzir proposes that licensing must hold for a feature’s surface position ( pp. 71–72). Thus, in this instance, an extra C2 licensor is needed to license the instance of FC2 on the adjective: the -t on DEF. This is exactly as Katzir’s analysis would predict. Furthermore, as a reviewer of the present article notes, in situations of contrastive focus, the possessive pronoun actually precedes the adjective and noun, and in such cases, the definite article is not possible (Thráinsson 2007), as (43a–b) show.

(43)

  • a.

    mín góð-a bók(*-in)

    my good-WK book-the

    ‘MY good book, (not yours)’

  • b.

    þitt rauð-a hús(*-ið)

    your red-WK house-the

    ‘YOUR red house, (not mine)’

Assuming the possessive pronoun c-commands the noun in this case, the C2 suffix on the definite article is superfluous, and thus the structure without it is predicted to be more economical. This prediction is borne out.

We must now return to the possessives without the definite article in (36)–(39). If we adopt the movement analysis, then the C2 suffixes in these constructions will go unlicensed, as the NP moves out of the c-command domain of the C2 suffix on POSS, and there is no definite article to do the necessary licensing. Such structures should be ruled out by GRAMMAR, because they contain unlicensed instances of FC2. On the other hand, if we adopt the base-generated analysis, then at least one C2 suffix in possessives with the definite article should be superfluous and be ruled out by ECONOMY. Under Katzir’s system of licensing, neither approach to possessives straightforwardly explains the range of possibilities in Icelandic possessives. The existence of both kinds (i.e., with and without the definite article) presents a paradox.

Two immediate options present themselves. We could say that there is a C2 suffix on the noun: the suffix that Katzir glosses as C1. However, the C1 suffix is not identical to the C2 suffix—though there is a high degree of overlap in form between the C1 and C2 suffixes, they are not identical. Furthermore, if we said that C1 could be a licensor of FC2, ECONOMY would predict there would never be a need for a C2 licensor on the definite article, as C1 could do all the licensing work. Therefore, the C1 suffix cannot be a licensor of FC2. Alternatively, we could say that there is a null definite article on the nouns in possessives with no definite article, and that the null definite article has a null C2 suffix. This would require stipulating a null definite article with an extremely constrained distribution (as definite articles must normally be overt), and it would require an additional C2 licensor that is phonologically null and can only attach to phonologically null elements. At best, this is an unmotivated patch, and at worst, it is an unfalsifiable proposal. Since both possibilities for licensors of the instance of FC2 on the adjective run into problems, we consider possessive constructions like those in (36)–(39) as instances of too few licensors.22

As in Danish, some vocatives involving adjectives can have the adjective in the weak form (see (44)), though the strong form is also possible (see (45)).

(44)

(45)

As with possessives, we assume that the spreading rule has applied in the normal fashion and that the adjectives in these examples have FC2. Katzir’s approach predicts the marking in (45): the adjectives have instances of FC2, and all instances of FC2 must be licensed by a C2 suffix. This is exactly what we find. However, in (44), the suffix attached to the adjective is not C2, but the realizer WK, the so-called weak inflection. Vocative phrase like those in (44) thus provide a second instance of an adjective appearing in weak form with no C2 licensor in sight. With no C2 suffix to license the FC2 on the adjectives, we would be forced to posit a null licensor for C2. Without independent evidence for such null elements or explanations of their distribution, vocatives constitute a second example of too few licensors.23

We have thus far shown that, like their Danish counterparts, possessives and vocatives in Icelandic have too few licensors to account for their morphological realization. Next, we turn to two examples that are unique to Icelandic. The first involves superlative adjectives. Superlative adjectives in Icelandic decline just like regular adjectives: they have both strong declensions (C2) and weak declensions (WK). Because of the constructions in which superlatives most often appear (i.e., with definite nouns), superlative adjectives showing the weak form are usually in DPs with the definite article. However, Svavarsdóttir and Jónsdóttir (1998:139) give some examples of superlatives occurring with nouns that have no article but must nevertheless be in the weak form.

(46)

  • a.

    Þjórsá er leng-st-a á á Ísland-i.

    Þjórsá is long-SUPER-WK river in Iceland-DAT

    ‘Þjórsá is the longest river in Iceland.’

  • b.

    Kínverj-ar eru fjölmenn-ast-a þjóð í heim-i.

    Chinese.person-NOM.M.PL are populous-SUPER-WK nation in world-DAT

    ‘The Chinese are the most populous people in the world.’

  • c.

    Jón Páll var sterk-ast-i maður í heim-i.

    Jón Páll was strong-SUPER-WK man in world-DAT

    ‘Jón Páll was the strongest man in the world.’

In the examples in (46), the superlative adjectives are all in the so-called weak form: they all bear the suffix that Katzir calls WK, and none of them bears C2. While the DPs in question might be interpreted as definite, no definite article is present, and thus, there is no C2 suffix. Presumably, these adjectives bear FC2 just like any normal adjective, but there does not appear to be anything to license that instance of FC2.

The second example that is unique to Icelandic involves proper nouns. When nonrestrictive adjectives are used with proper nouns in Icelandic, the adjective must (or may) show weak inflection, as in (47a–b).

(47)

First, we should note that the adjective here appears after the name—not the standard position for adjectives. Julien (2005) attributes this to movement of the name to Spec,DP, under the assumption that names are nouns. The morphology present on names in Icelandic also suggests that names are nouns, as they decline just like nouns, bearing C1 suffixes instead of C2 suffixes. The examples in (47) are slightly irregular in that they are names showing the so-called weak declension for nouns, which is identical to the weak declension of adjectives, but only in the singular. In any case, it is clear that names do not bear C2 suffixes, and thus, the instance of FC2 on the adjectives in (47) is apparently unlicensed.

In this section, we pointed out examples of unlicensed instances of FC2 from four domains: possessives, vocatives, superlative adjectives, and proper nouns. Under Katzir’s approach, we expect these to be ruled out by GRAMMAR, but they are in fact perfectly grammatical. Let us now turn to concord in Icelandic, which provides several examples of too many licensors.

4.1.2 Concord: Too Many Licensors

Icelandic has a rich system of concord in gender, number, and case. Adjectives and the definite article (among other things, as we will soon show) have different forms depending on the gender, number, and case of the DP. Just as in Danish, the endings that adjectives take are different in definite and indefinite contexts, as shown in (48) (Katzir’s (40)).

(48)

graphic

Katzir’s analysis aims to assimilate the C2 ending to the Danish -t and the WK ending to the Danish -e (i.e., treat C2 as a licensor of FC2 and WK as a realizer of FDEF). The idea, then, is that ECONOMY will rule out structures with superfluous instances of C2, just as neuter -t on attributive adjectives in Danish is ruled out by the presence of a licensing neuter -t on the definite article. Katzir has little to say about WK; we assume for the moment that it is exactly parallel to Danish -e (i.e., a realizer of at least FDEF), but we will return to it in section 4.2.24

Katzir restricts his discussion to Icelandic DPs containing at most an adjective and a definite article (as in (31a–d) above).25 However, it is not only adjectives and definite articles that bear C2 suffixes. As Norris (2012) notes, there are at least six different word classes bearing C2 suffixes in Icelandic. The various elements can be seen in (49), and representative examples (one for each gender) are given in (50).

(49) Word classes bearing C2 suffixes in Icelandic

  • a.

    Quantifiers: all-ur ‘all’, sum-ur ‘some’, engin-n ‘none’

  • b.

    Demonstratives: þess-i ‘this’, sá ‘that/the one’, hin-n ‘the other’

  • c.

    Numerals: tve-ir ‘two’, þrí-r ‘three’, fjór-ir ‘four’

  • d.

    Adjectives: lítil-l ‘little’, gul-ur ‘yellow’, falleg-ur ‘pretty’

  • e.

    Definite article: -in-n

  • f.

    Possessive pronouns: min-n ‘1SG.POSS’, þin-n ‘2SG.POSS’, sin-n ‘3.REFL.POSS

(50)

  • a.

    graphic

  • b.

    graphic

  • c.

    graphic

Though there are some instances of suppletion and a bit of variation (e.g., in the demonstratives þess-i and ), the default concord markers can be represented as in (33), repeated here.26 We assume these are the markers that Katzir refers to as C2, as these are unquestionably the endings that surface on strong adjectives in the language.27

(33)

graphic

Katzir proposes that these suffixes are akin to the Danish -t—that is, they are licensors. Under Katzir’s proposal, these C2 suffixes serve the purpose of licensing a feature (FC2). Recall that licensing is done by c-command: every instance of FC2 must be c-commanded by a licensing C2 suffix. Under Katzir’s analysis, multiple instances of the same licensor indicate a lack of a ccommand relationship between the elements hosting the licensors. As in Danish (see section 3.2), the required structures are implausible on syntactic grounds.

Let us consider again the examples in (50). Aside from the adjective and noun, every element in (50) that bears a concord marker is bearing what looks like C2, for a total of five instances of C2 (quantifier allur ‘all’, demonstrative hinn ‘other’, definite article -inn, possessive pronoun minn ‘my’, and numeral fjórir ‘four’). For Katzir’s proposal to be correct, there can be no c-command relationships between any two of those elements. If there were a c-command relationship between any two of them, then at least one of them would be superfluous (i.e., not necessary to license any instances of FC2). ECONOMY would then rule out the structure with a superfluous instance of C2. For example, the quantifier allur ‘all’ has been argued to head its own projection at the top of DP (Sigurðsson 1993), as in (51), which would put it in a position to c-command the definite article, among other things.

(51)

graphic

Some possible support for the claim that quantifiers c-command the rest of the DP comes from (52), which involves a negative quantifier licensing a negative polarity item (NPI) (for evidence that neinn must be licensed by a c-commanding negative element, see Jónsson 1996, Thráinsson 2007).

(52)

  • enginn maður í neinum menningarheimi

  • no man in any(NPI) culture

  • ‘no man/person in any culture’28

We thank an anonymous LI reviewer for suggesting examples of this type.

While the structure in (51) is slightly abbreviated (indicated by the dashed line), the C2 licensor on allur ‘all’ clearly c-commands the instance of FC2 on all as well as the instances of FC2 on everything else in the DP. This is enough to license every instance of FC2. Since Katzir’s ECONOMY prefers structures with fewer licensors, this predicts that a structure with only one instance of C2 (on allur) would be preferred over a structure with five instances of C2. As the ungrammaticality of (53a–c) shows, this prediction is not borne out.

(53)

graphic

These examples involve only one instance of C2 (on allur ‘all’), one instance of C1 (on the noun), and five instances of WK,29 which is what Katzir’s proposal predicts should be grammatical under the accepted structure of nominals in Icelandic (see, e.g., Delsing 1993, Sigurðsson 1993, 2006, Vangsnes 1999, Julien 2005, Pfaff 2009, Norris 2011). The only grammatical way to use those words together in a nominal phrase is as in (50), where every element bears a C2 suffix except the adjective and the noun, which bear WK and C1, respectively. The examples in (53) are ungrammatical, and so are any similar examples with a distribution of C2 and WK suffixes that differs from the grammatical one in (50).

This is inconsistent with the part of Katzir’s analysis that treats C2 as a licensor, and thus, it is a clear reason to reject such an analysis of C2. Arguments analogous to the one made for allur could be made for the other C2 hosts in (49). The only way to make Katzir’s morphological account consistent with the data in (50) and (53) would be to say that there are no c-command relations between any of the elements in the nominal phrases in (50). Such an assertion would be untenable in the context of the other work that has been done on nominal phrase syntax in Icelandic. (As a full discussion of nominal phrase internal syntax in Icelandic is beyond the scope of this article, we refer the reader to the works cited.)

4.1.3 Icelandic: C2 Summary

We have just looked at two broad kinds of evidence against Katzir’s claim that C2 is a licensor: (a) cases where there appear to be too many licensors—that is, more than are needed to license all of the instances of FC2; and (b) cases where there appear to be too few licensors to license all of the instances of FC2. The distribution of C2 is inconsistent with the distribution of a licensor as predicted by both GRAMMAR (some instances of FC2 are left unlicensed) and ECONOMY (some instances of C2 are superfluous).

4.2 Icelandic without Licensors

If C2 is not a licensor, then what is it? We would like to claim that C2 is a realizer, just like -t in Danish. Katzir does not explicitly argue against a realizer analysis of C2, so we simply consider the argument he makes against Danish -t being a realizer. Recall that -t only surfaces on adjectives in neuter singular indefinite DPs; in definite DPs, all adjectives bear the realizer suffix -e. Katzir argues that, if -t is a realizer (say, of FNt), then we would expect it to surface on adjectives with the feature FNt. The fact that there is no -t on adjectives in definite DPs in Danish is thus puzzling for a realizer account. As Katzir notes, ‘‘whatever gender/number features spread onto the adjectives are presumably the same in the definite and the indefinite form’’ (pp. 59–60). In section 3.3, we suggested that the reason -t apparently disappears is morphological: the system of adjectival inflection in Danish is severely impoverished, and, in fact, indefinite plural adjectives bear the same realizer -e as definite adjectives. Given that Icelandic’s morphology is much richer, it is worth revisiting the argument against the realizer analysis of -t/C2.

Katzir’s argument against a realizer treatment of -t does not extend to Icelandic C2, as the gender/number/case distinctions are not fully neutralized in the definite form. This is clear from the definite forms in (54)–(55).

(54) ‘‘Definite’’ adjectives in Icelandic (nominative)

  • a.

    graphic

  • b.

    graphic

  • c.

    graphic

(55) ‘‘Definite’’ adjectives in Icelandic (accusative)

  • a.

    graphic

  • b.

    graphic

  • c.

    graphic

In Icelandic, this WK suffix is traditionally called the ‘‘weak declension’’ of adjectives (as in the other Germanic languages). The weak declension is often assimilated to something like definiteness (Sigurðsson 2006), though as discussed by Thráinsson (2007) and Pfaff (2009, to appear), the distinction between weak and strong adjectives in Icelandic is about more than definiteness.30 This declension paradigm is given in (56)).

(56)

graphic

These WK endings are not random, and while some featural information is lost in the paradigm, some of it is also preserved. These adjectives are still marked for gender, number, and case. Given that inflection is in general more impoverished in Danish than in Icelandic, it is not surprising that gender and number information is totally neutralized in the definite and plural forms in Danish. As for why C2 is apparently absent from the weak form of adjectives in Icelandic, we would like to say that it is because both C2 and WK realize the same terminal node. In Distributed Morphology terms, the WK suffix and the C2 suffix compete for insertion.

C2 and WK are actually only a part of the system of nominal agreement suffixes in Icelandic. Comparative adjectives in Icelandic are often described as having a weak inflection (like WK), but in fact, the paradigm for comparative adjectives is even more impoverished than the WK paradigm. There is also the suffix that Katzir identifies as C1. It attaches only to nouns (including proper nouns), and though there are some similarities between C1 and C2, they are distinct suffixes. The paradigms for comparative adjectives and C1 are given in (57) and (58).31

(57)

graphic
CAE) in Icelandic

(58)

graphic

Thus, there are essentially four different realizers in Icelandic: C2, WK, CAE, and C1.32 We assume the VI schematics in (59), ignoring precise featural specifications of the nodes, as we are concerned here with what drives the choice between C2 and WK, and not with how the various forms of those suffixes are calculated.33

(59) Vocabulary items for Icelandic concord markers

  • a.

    Agr, [ξ] ↔ C1 / N⏜_____

  • b.

    Agr, [ξ] ↔ CAE / COMPARATIVE⏜_____

  • c.

    Agr, [ξ, WK] ↔ WK / A⏜_____

  • d.

    Agr, [ξ] ↔ C2

We use ξ to represent combinations of gender, case, and number features, and we follow Embick (2010) in assuming that allomorphy can be determined under linear adjacency (i.e., concatenation, indicated here by ⏜). The analysis is straightforward. C1, CAE, and C2 all realize identical feature sets. Because C1 and CAE have a more specified distribution, they are in a Pāṇinian relationship with C2. According to the Subset Principle, C1 and CAE should always win over C2, and they do: nouns and comparative adjectives never bear the C2 suffix. In contrast, WK and C2 realize different feature sets: WK also realizes [WK], whereas C2 does not. Thus, again by the Subset Principle, if the node undergoing insertion consists of [ξ, WK], then we expect WK to be inserted instead of C2. C2 is only inserted if none of these conditions are met (e.g., if the adjective does not have the feature [WK]). This analysis also explains why the majority of elements bear the C2 suffix (it is the elsewhere case) and why it is only adjectives that change inflection in definite DPs (WK only attaches to adjectives). While we acknowledge that this is simply a sketch of an analysis, we believe it is a very promising direction for an analysis treating C2 as a realizer instead of a licensor.

4.3 Icelandic Is Not Danish'

We close the section on Icelandic by returning to Katzir’s initial suggestion regarding Icelandic: that Icelandic is Danish', a language identical to Danish with respect to the topics under discussion, except for the structural position of adjectives. Specifically, Katzir suggests that Icelandic APs are sisters to N. The proposed structure is repeated here.

(30)

graphic

As far as we can tell, the main motivation for treating Icelandic as Danish' is to support the licensor-based analysis of definiteness and gender marking. As Katzir notes, changing this aspect of the structure leads to surprisingly different predictions about the marking we expect to see. If those predictions were borne out, the proposal would win convincing support.

As we have shown in the preceding sections, many of these predictions are not borne out, which casts doubt on the utility of the licensor category. Although the main motivation for Katzir’s analysis of Icelandic APs was the distribution of licensors, this does not necessarily mean that the structure Katzir proposes for Icelandic APs must be abandoned as well. However, we believe that there are other reasons to reject Katzir’s proposal for Icelandic APs, and we would like to briefly discuss the matter before concluding our article.

If the relationship between N and AP is intended to be a standard head-complement relationship (as Katzir’s structures suggest), there are a number of fairly standard expectations we would have that are not borne out. For example, we would expect that Ns in Icelandic (but not Danish) could subcategorize for certain adjectives. We are unaware of any such examples. Furthermore, we might expect that a noun could not simultaneously have a PP/CP complement of its own and be modified by an adjective, on the standard assumption that a head can have only one complement. This prediction is not borne out.34 On the subject of complementation, note that the AP-N relationship would have to be head-final, which would make it the only example of head-finality that we are aware of in Icelandic.

Finally, we would like to consider the predictions of Katzir’s analysis in an area that one of Katzir’s reviewers suggested: ellipsis (see Katzir, fn. 24). Katzir’s structural position for adjectives would seem at first blush to prohibit structures with multiple adjectives, as the noun has only one complement position. However, Katzir works around this by suggesting the possibility of ‘‘adjoin[ing] the APs to one another first, attaching the result to NP as a single constituent’’ (p. 61). If we allow this possibility, DPs with multiple adjectives will have the structure in (60).

(60)

graphic

Such a structure makes a very clear prediction about nominal ellipsis: it should be impossible to elide one adjective together with the noun, leaving behind another adjective.35 Katzir himself does not investigate this prediction. While we do not investigate this fully here for reasons of space, we offer some preliminary comments in hopes of inspiring future work on ellipsis(-like) processes in Icelandic nominals. Several possible candidates for nominal phrase ellipsis in Icelandic are given in (61)–(63).

(61)

  • Haraldur vildi kaupa rauðan bíl en ekki þann stóra.

  • Haraldur wanted buy red car but not that big

  • ‘Haraldur wanted to buy a red car, but not the big one (= red car).’

(62)

  • Haraldur vildi kaupa stórt, brúnt hús en hann keypti lítið í staðinn.

  • Haraldur wanted buy big brown house but he bought little in the.place

  • ‘Haraldur wanted to buy a big brown house but he bought a little one (= brown house) instead.’

(63)

  • Bandaríkjamenn drekka léttan bjór og þykir kaldur bestur.

  • Americans drink light beer and consider cold best

  • ‘Americans drink light beer and consider cold (light beer) best.’

In each of these examples, an adjective (italicized) is stranded before the position that the nominal would occupy if overt. In addition, the antecedent (in boldface) includes an adjective that does not surface in the ‘‘elided’’ version. If adjectives formed one large constituent, we would expect either all of them to remain or all of them to be elided. If one (or all) of these examples is a case of genuine ellipsis, then the fact that some adjectives can be elided and some can be left behind suggests that the adjectives do not adjoin to each other first and then merge with N(P). Instead, the adjectives must have some degree of autonomy, either as adjuncts to NP, as specifiers, or as heads taking AP/NP complements—any analysis where they do not form a constituent to the exclusion of the NP.

However, it may also be that these constructions are not cases of genuine ellipsis. If that is so, they may be best analyzed as akin to English one-anaphora but with a silent ‘one’. The ‘‘elided’’ NPs would then be simple NPs modified by one adjective—easily incorporated into any phrase structure for adjectives and nouns.36 Katzir notes that ellipsis would be a good test to determine whether his proposal for the constituency for adjectives is on the right track. If the examples in (61)–(63) are not cases of genuine ellipsis, then we cannot conclude anything from them, but if they can be properly analyzed as ellipsis, then Katzir’s phrase structure for adjectives is not on the right track.37

5 Conclusion

We have focused on cases where, under a licensor analysis, there are either too few or too many licensors. From these cases, we conclude that both of the central assumptions of licensing theory are unsustainable: neither can it be required that every feature have a licensor, nor can the ECONOMY condition hold.

In contrast, we have sketched licensor-free alternative analyses that avoid the problems faced by licensors. Our licensor-free alternatives do not unify definiteness marking and gender marking, which is one of the claimed strengths of Katzir’s analysis. Definiteness and gender do intersect in that, in Danish and Icelandic, adjectives inflect for both, but we believe the two are not intimately linked in the way that Katzir presumes. Indeed, as we have shown for Icelandic, gender is not fully neutralized in any context (except for words that do not decline; see footnote 32); in contrast to Katzir, we have suggested that the apparent neutralization in Danish is the result of morphological syncretism rather than syntactic principles. In short, there is no very good reason to believe that definiteness and gender have anything to do with each other.

We have said nothing about the part of Katzir’s article that deals with Greek, but will simply observe that if the plausibility of that analysis rests on support provided by the analyses of Danish and Icelandic, it is poorly supported.

We conclude with some observations about the relative transparency of licensor-based and licensor-free analyses. Katzir starts from a world where two elements are involved in agreement processes: spreaders (or controllers) and realizers. We would like to point out that once Katzir adds a third—namely, licensors—one of the other two disappears. In the case of definiteness, we lose a spreader. Rather than controlling definiteness agreement, the definite article (or the suffix -en) licenses FDEF, and some other element (a null functional head associated with definiteness) controls the agreement. In the case of gender, we lose the realizer. Traditionally, and in our analysis, -t is treated as a realizer of neuter agreement. Under Katzir’s analysis, neuter agreement happens by a rule of spreading, but it is never actually realized—in contrast to FDEF, there is no morpheme that realizes FNt. We might ask how such an agreement process could come to be. If agreement involves licensors in addition to spreaders (or controllers) and realizers, then we would expect at least some cases of agreement where all three elements are overt. Strikingly, there are no such cases in Danish or Icelandic. Thus, the licensor-free analyses are not only more successful in accounting for the full range of data in Danish and Icelandic, but also formally simpler and more transparent.

Notes

We would like to thank two anonymous reviewers for very helpful feedback on an earlier version and the Morphology Reading Group at the University of California, Santa Cruz, for discussion that stimulated us to write the article in the first place. We would also like to thank Peter Jenks for helpful discussion, as well as consultants for the Icelandic data presented here, especially Edda Arnaldsdo´ttir and Arnaldur Axfjörð. We alone are responsible for any errors.

1 For presentational convenience, we will ignore the possibility of ‘‘syncategorematic’’ introduction of the definiteness feature, hinted at in Katzir’s footnote 6 and assumed in Katzir’s (9), p. 52 (cf. Katzir’s footnote 15), since in any case the features to be spread must be introduced by some mechanism.

2 Making use of a definition of c-command adapted from proposals by May (1985) and Kayne (1994): ‘‘An affix c-commands everything its attachment site c-commands or dominates’’ (Katzir’s (17), p. 54).

3 ‘‘If S1 and S2 are identical except for licensors, and if S1 has strictly fewer licensors than S2, then S2 is ungrammatical’’ (Katzir, p. 52).

4 The distribution of licensors is presumably regulated by GRAMMAR, which ‘‘generates structures and enforces the usual well-formedness conditions on phrase structure, selection, phonology, semantics, and so on’’ (pp. 51–52), but Katzir does not explore this matter in detail.

5 Assuming that the ECONOMY condition is universal, it is hard to see how Swedish, Norwegian, and Faroese, which have double definiteness of exactly the kind excluded here, could exist. Perhaps they could be given some kind of analysis involving movement, along the lines proposed by Katzir for Greek (which must be the intention of the suggestion in his footnote 19: ‘‘perhaps . . . the preadjectival definiteness marker in these languages does not have the noun within its licensing domain’’). However, Katzir does not offer such an analysis and we will not attempt to construct one.

6 The source of the definite feature controlling the agreement on A is either the possessive or some null definitenessrelated head. For our argument, it does not matter which.

7 This construction is discussed in Hansen 1994:112–113 under the label kvalificeret bestemthed (‘qualified definiteness’).

8 Under Katzir’s analysis, the realizer of neuter agreement is consistently null. We return to this issue in section 5.

9 In the sense of c-command introduced in Katzir’s (17), page 54, repeated above in footnote 2.

10 Our argument does not depend on the category label of hver, only on its structural position in the tree. For concreteness, we use QP for Quantifier Phrase.

11 A reviewer asks whether there is any reason to think that (19) is less plausible than (16). We believe that (19) is indeed less plausible than (16), as there is no evidence that hver has a separate life as an adjective, whereas there is evidence for an adjectival version of et, as Katzir shows.

12 If there is also no unary projection from Q to QP, we predict the string in (17b) with no -t on either adjective, since the -t on hver c-commands both adjectives.

13 A reviewer notes that there are proposals within the Minimalist framework that make use of unary projections; see, for example, Kayne 2010, Adger 2012. However, as far as we can tell, the use of unary projections from such proposals in the places where Katzir’s account needs them would be no more principled than allowing unary projections in a theory like the one outlined in Chomsky 1995. Furthermore, there appears to be no principle in Adger’s or Kayne’s work that would force Self-Merge of A or Q in these derivations, and Katzir’s analysis requires obligatory Self-Merge of A and Q to account for the obligatoriness of neuter -t.

14 An alternative approach to weak adjective inflection employing impoverishment has been suggested by Noyer (1998), and such an analysis has been worked out in detail for German by Roehrs (2009). Thanks to an anonymous reviewer for drawing our attention to this kind of analysis.

15 While (31a) is grammatical, it is highly dispreferred in most contexts. The status of the prenominal definite article in Icelandic is quite complicated. As Katzir notes (p. 70), some speakers state that it does not have the same meaning as the form with the suffixed article (as in (31b)). Furthermore, as Thráinsson (2007) notes, there are some instances where the suffixed article is ungrammatical and the prenominal article is required. We follow Katzir in treating the prenominal article as a different syntactic element from the suffixed article, but see Magnússon 1984, Sigurðsson 1993, and Pfaff 2009 for a different view.

16 This is grammatical under the meaning ‘the other ( yellow) horse’. The demonstrative hinn ‘the other’ is identical to the prenominal article in all cases, numbers, and genders except nominative/accusative neuter singular: the demonstrative is hitt and the definite article is hið.

17 Katzir uses v instead of [cf6]WK[cf1] in his article; we avoid this abbreviation because of v’s ubiquity in a different role.

18 The arguments from Danish against -en being a licensor of FDEF/-e can straightforwardly be extended to Icelandic -in and WK. We refrain from doing so for two reasons: (a) Katzir does not explicitly analyze -in as a licensor of FDEF and WK as a realizer of FDEF, and (b) the arguments against treating C2 as a licensor are more revealing.

19 The semantic contrast suggested here is reminiscent of the kind of distinction that Davies and Dubinsky (2003) argue to be operative in governing extraction from NPs: ‘book’ in the abstract reading encodes (nonargumental) participants whereas concrete ‘book’ does not. Davies and Dubinsky argue that it is only participants that can be extracted. Exactly how these distinctions play into possessum marking is an issue that is beyond the scope of this article, but it seems a fruitful line of inquiry for research in this domain.

20 This is not an exhaustive list of constructions that are used for possession in Icelandic. For a particularly thorough survey, see Thráinsson 2007:88–96.

21 We cannot simply use Pfaff ’s (2009) proposed structure, as his structural assumptions about adjectives are incompatible with Katzir’s assumptions about licensing in Icelandic.

22 We are grateful to an anonymous reviewer for very helpful discussion of the issues concerning possessives.

23 We should note that some nouns in Icelandic have slightly irregular declension paradigms. For example, some nouns form the nominative plural with -ir and the accusative plural with -i instead of the normal -ar and -a. The nouns gestur ‘guest’ and vinur ‘friend’ are two such nouns. This is just to say that the suffix on vinir ‘friends’ in (44a) is not C2—evidence against such a proposal can be found in (45a).

24 The distribution of the weak endings in Icelandic is more complicated than presented here; treating them as an exponent of definiteness may be an oversimplification. For discussion of adjective marking in Icelandic, see Pfaff 2009, to appear.

25 The only exception is in Katzir’s footnote 44, where he notes that a reviewer pointed out that possessive pronouns in the language appear with what looks like C2 morphology, even in definite DPs. As Katzir notes—and as we discussed in section 4.1.1—this is puzzling for his approach, since it suggests that the C2 on -in fails to license FC2 on the possessive pronoun.

26 We say ‘‘default’’ here, because there are (at least) four different concord marker paradigms in the language. Elements with concord markers distinct from the default set are nouns/nominals (what Katzir calls C1), comparative adjectives, and the so-called weak declension of adjectives (what Katzir calls v).

27 The three cells in (33) listed as -Ø are not simply null; stems that these endings attach to are subject to u-shift, a pervasive morphophonological rule in Icelandic that fronts, rounds, and raises some or all instances of a in the stem.

28 From the Morgunblaðið newspaper: http://www.mbl.is/greinasafn/grein/1254040/, accessed 14 March 2013. Speakers consulted have accepted this example as well.

29 In (53), we include the WK suffixes as optional on everything but the adjective. With adjectives, there is an alternation between WK and the C2 suffix depending on whether or not the adjective is c-commanded by another C2 suffix, so in these hypothetical examples, we might expect to see other elements bearing WK since their C2 suffix is already licensed. On the other hand, if WK suffixes only attach to adjectives, then perhaps we would expect to see nothing at all.

30 For example, there are instances of adjectives appearing with strong inflection in apparently definite DPs.

(i)

  • Ég horfði upp í blá-an himin-inn.

  • I looked up into blue-C1 sky-the

  • ‘I looked up into the blue sky.’

  • (Thráinsson 2007:3)

The most thorough investigation of these issues that we are aware of is the one provided by Pfaff (to appear). Pfaff argues that strong adjectives in definite DPs as in (i) are merged outside DP, leading to their obligatorily nonrestrictive interpretation. In contrast, weak adjectives are merged within DP, though this says nothing about their interpretation. As Pfaff shows, both nonrestrictive and restrictive interpretations are possible for weak adjectives (given the proper context). We cannot discuss these issues in detail here, but we do note that although examples such as (i) are very interesting, their grammaticality is orthogonal to the point we are making about Katzir’s approach. Even if WK were a marker of mere definiteness, Katzir’s analysis could not account for its distribution.

31 The C1 declension given in (58) is traditionally called the regular and strong declension for nouns in Icelandic. Of course, there are special declension classes to which many words belong (e.g., veggur ‘wall’, bær ‘farm/town’, bekkur ‘bench/grade’), as well as words that are totally irregular in the sense that they constitute their own declension class (e.g., hönd ‘hand’, fótur ‘foot’). Furthermore, there is a ‘‘weak’’ declension of nouns; in the singular, their endings are identical to the WK endings, but in the plural, they have their own forms. We believe it is reasonable to treat these ‘‘weak’’ C1 forms as simple declension classes.

32 Of course, there are also some elements that do not decline. There are some indeclinable adjectives (e.g., hissa ‘surprised’), and present participles (e.g., hlaupandi ‘running’) do not decline either. Einarsson (1949) notes that, historically, they declined like comparative adjectives, and such forms can still be found in the written language. Perhaps we could say that these words make up a fifth class: those words that do not have realizers at all.

33 The choice between strong and weak adjectival inflection in the Germanic languages is often attributed to definiteness, but as Pfaff (to appear) shows, the choice between strong and weak adjectives is not really determined on the basis of definiteness alone. There are strong adjectives in DPs that are apparently definite (see footnote 30), and there are weak adjectives in DPs with no overt marker of definiteness (see (46)). While we do propose a featural difference between strong and weak adjectives, we have chosen to remain agnostic about the identity of the feature. Thus, our VIs in (59) are simplified in this respect. Whether this difference can be captured with features is a question that we leave open here.

34 We are grateful to an anonymous LI reviewer for helpful discussion of this point.

35 Such ellipsis processes are commonly assumed to elide constituents (see, e.g., Lobeck 1995, Merchant 2001).

36 An observation made by a reviewer suggests that, for at least one of the constructions, genuine ellipsis appears to be the wrong approach. The ellipsis analysis makes an interesting prediction about the rigid internal ordering of multiple adjectives, as it is commonly assumed that ellipsis processes can only elide syntactic constituents. For example, an ellipsis account predicts (i) to be ungrammatical.

(i)

  • Haraldur vildi kaupa stóran bíl en ekki þann rauða.

  • Haraldur wanted buy big car but not that red

  • ‘Haraldur wanted to buy a big car, but not the red one (= big car).’

Given that the normal word order is stór rauður bíll ‘big red car’ and not *rauður stór bíll ‘red big car’, it should not be possible to elide stór bíll to the exclusion of rauður ‘red’. However, preliminary fieldwork suggests that (i) is indeed grammatical under the intended reading.

37 It seems that Katzir himself was aware of the difficulties his proposal might face, as he mentions in footnote 43 and at the bottom of page 67. He suggests a very tentative way forward, in which the category to which the definite article and adjectives attach is the same (i.e., by saying there is no difference between N and NP in Icelandic). Katzir would still require there to be a difference between A and AP, though—recall that that is how he prevents licensors on higher adjectives from licensing features on lower adjectives. This would require simultaneously embracing and rejecting bare phrase structure.

References

References
Adger
David
.
2012
.
A syntax of substance
.
Cambridge, MA
:
MIT Press
.
Börjars
Kersti
, and
Mark
Donohue
.
2000
.
Much ado about nothing: Features and zeroes in Germanic noun phrases
.
Studia Linguistica
54
:
309
353
.
Chomsky
Noam
.
1995
.
Bare phrase structure
. In
Evolution and revolution in linguistic theory
, ed. by
Héctor
Campos
and
Paula
Kempchinsky
,
51
109
.
Washington, DC
:
Georgetown University Press
.
Davies
William D
., and
Stanley
Dubinsky
.
2003
.
On extraction from NPs
.
Natural Language and Linguistic Theory
21
:
1
37
.
Delsing
Lars-Olof
.
1993
.
The internal structure of noun phrases in the Scandinavian languages
.
Doctoral dissertation, Lund University
.
Einarsson
Stefán
.
1949
.
Icelandic: Grammar, texts, glossary
.
Baltimore, MD
:
The Johns Hopkins Press
.
Embick
David
.
2010
.
Localism versus globalism in morphology and phonology
.
Cambridge, MA
:
MIT Press
.
Embick
David
, and
Rolf
Noyer
.
1999
.
Locality in post-syntactic operations
. In
Papers on morphology and syntax, cycle two
, ed. by
Vivian
Lin
,
Cornelia
Crause
,
Benjamin
Bruening
, and
Karlos
Arregi
,
265
317
.
MIT Working Papers in Linguistics 34
.
Cambridge, MA
:
MIT, MIT Working Papers in Linguistics
.
Embick
David
, and
Rolf
Noyer
.
2001
.
Movement operations after syntax
.
Linguistic Inquiry
32
:
555
595
.
Grimshaw
Jane
.
2005
.
Words and structure
.
Stanford, CA
:
CSLI Publications
.
Halle
Morris
.
1997
.
Distributed Morphology: Impoverishment and fission
. In
PF: Papers at the interface
, ed. by
Benjamin
Bruening
,
Yoonjung
Kang
, and
Martha
McGinnis
,
425
449
.
MIT Working Papers in Linguistics 30
.
Cambridge, MA
:
MIT, MIT Working Papers in Linguistics
.
Hankamer
Jorge
, and
Line
Mikkelsen
.
2002
.
A morphological analysis of definite nouns in Danish
.
Journal of Germanic Linguistics
14
:
137
175
.
Hankamer
Jorge
, and
Line
Mikkelsen
.
2005
.
When movement must be blocked: A reply to Embick and Noyer
.
Linguistic Inquiry
36
:
85
125
.
Hansen
Erik
.
1994
.
Kvalificeret bestemthed
. In
Møde om udforskningen af Dansk sprog
, ed. by
Mette
Kunøe
and
Erik Vive
Larsen
,
106
114
.
Aarhus
:
Aarhus Universitet
.
Hansen
Erik
, and
Lars
Heltoft
.
2011
.
Grammatik over det Danske sprog, bind I
.
Copenhagen
:
Det Danske Sprog- og Litteraturselskab
.
Jónsson
Jóhannes Gísli
.
1996
.
Clausal architecture and case in Icelandic
.
Doctoral dissertation, University of Massachusetts, Amherst
.
Julien
Marit
.
2005
.
Nominal phrases from a Scandinavian perspective
.
Amsterdam
:
John Benjamins
.
Katzir
Roni
.
2011
.
Morphosemantic mismatches, structural economy, and licensing
.
Linguistic Inquiry
42
:
45
82
.
Kayne
Richard S
.
1994
.
The antisymmetry of syntax
.
Cambridge, MA
:
MIT Press
.
Kayne
Richard S
.
2010
.
Comparisons and contrasts
.
Oxford
:
Oxford University Press
.
Lobeck
Anne
.
1995
.
Ellipsis: Functional heads, licensing, and identification
.
New York
:
Oxford University Press
.
Magnússon
Friðrik
.
1984
.
Um innri gerð nafnliða í íslensku [On the structure of noun phrases in Icelandic]
.
Íslenskt mál og almenn málfræði
6
:
81
111
.
May
Robert
.
1985
.
Logical Form: Its structure and derivation
.
Cambridge, MA
:
MIT Press
.
Merchant
Jason
.
2001
.
The syntax of silence: Sluicing, islands, and the theory of ellipsis
.
New York
:
Oxford University Press
.
Norris
Mark
.
2011
.
Extraposition and definiteness effects in Icelandic NPs
. In
Morphology at Santa Cruz
, ed. by
Nicholas
LaCara
,
Anie
Thompson
, and
Matthew
Tucker
,
97
121
.
Santa Cruz
:
University of California, Linguistics Research Center
.
Norris
Mark
.
2012
.
Towards an analysis of concord (in Icelandic)
. In
WCCFL 29: Proceedings of the 29th West Coast Conference on Formal Linguistics
, ed. by
Jaehoon
Choi
,
E. Alan
Hogue
,
Jeffrey
Punske
,
Deniz
Tat
,
Jessamyn
Schertz
, and
Alex
Trueman
,
205
213
.
Somerville, MA
:
Cascadilla Proceedings Project
.
Noyer
Rolf
.
1997
.
Features, positions and affixes in autonomous morphological structure
.
New York
:
Garland
.
Noyer
Rolf
.
1998
.
Impoverishment theory and morphosyntactic markedness
. In
Morphology and its relation to phonology and syntax
, ed. by
Steven G.
Lapointe
,
Diane K.
Brentari
, and
Patrick
Farrell
,
265
285
.
Stanford, CA
:
CSLI Publications
.
Pfaff
Alexander
.
2009
.
Structural relations between free and suffixed articles in Icelandic
.
Master’s thesis, Eberhard Karls Universität, Tübingen
.
Pfaff
Alexander
.
To appear
.
Inside and outside—before and after: Weak and strong adjectives in Icelandic
. In
Adjectives in Germanic and Romance: Variation and change
, ed. by
Frank
Van de Velde
,
Petra
Sleeman
, and
Harry
Perridon
.
Amsterdam
:
John Benjamins
.
Roehrs
Dorian
.
2009
.
The structural nature of adjective endings in German and some consequences
.
Ms., University of North Texas, Denton. Available at http://ling.auf.net/lingbuzz/000869
.
Sigurðsson
Halldór Ármann
.
1993
.
The structure of the Icelandic NP
.
Studia Linguistica
47
:
177
197
.
Sigurðsson
Halldór Ármann
.
2006
.
The Icelandic noun phrase: Central traits
.
Arkiv for nordisk filologi
122
:
193
236
.
Svavarsdóttir
Ásta
, and
Margrét
Jónsdóttir
.
1998
.
Íslenska fyrir útlendinga: Kennslubók í málfræði [Icelandic for foreigners: Grammar teaching manual]
.
Málvísindastofnun Háskóla Íslands
.
Thráinsson
Höskuldur
.
2007
.
The syntax of Icelandic
.
Cambridge
:
Cambridge University Press
.
Vangsnes
Øystein Alexander
.
1999
.
The identification of functional architecture
.
Doctoral dissertation, University of Bergen
.