1. Introduction

In this paper I explore the syntax of noun phrases in Miskitu, a Misumalpan language spoken along the Caribbean Coast of Nicaragua and Honduras. The central concern of the paper relates to the multiple determiners and modifiers that accumulate after the Miskitu noun. Rather than propose a specific theory to account directly for the complexity of this postnominal system, I show that by taking into consideration a few independent properties of Miskitu syntax, we are led to the conclusion that even ordinary-looking noun phrases are structurally ambiguous, and may contain covert relative clause structure. For example, we observe that when two determiners are string-adjacent, there is a possible structure in which the first is buried inside an invisible relative clause. Not only do these observations explain the availability of multiple determiners in the Miskitu noun phrase; they predict this property as the null hypothesis for the language. The observations extend naturally to account for the positioning of adjectival and nominal restrictive modifiers, and would be consistent with the strong claim that such modifiers never need special consideration as part of the theory of noun phrase structure; rather they are always simply the predicate of a clause. This would constitute a substantial simplification of the phrase structure component.

2. Overview of Miskitu grammar

This section provides a sparse introduction to the areas of Miskitu syntax that will be relevant to the rest of the paper. For a much more extensive discussion of these and many other aspects of Miskitu grammar, see CIDCA (1985) and, especially, Salamanca (1988).

2.1. Word order and clause structure

The basic word order in a Miskitu clause is SOV. The only deviations from this are (1) stylistic interchanging of the arguments, often apparently related to the relative sizes of the constituents; (2) postposing of sentential complements; (3) pre/postposing of extra-heavy noun phrases. The categories of INFL(ection) and COMP(lementizer) are clause-final, and all adpositions and case markers¹ must follow the noun phrase. In short, Miskitu behaves like a strictly head-final language. We can illustrate the basic clause structure with the following simple example:²

(1)	yang   prn1   aras   horse   ba   def   kaik-ri   see-pst1   ‘I saw the horse’

We might diagram this as in (2):³

(2)	[IP IP [SUBJ SUBJ [yang] ]SUBJ [I′ I′ [VP VP [OBJ OBJ [aras ba] ]OBJ [V V [t] ]V ]VP [I I [[[kaik] ri]] ]I ]I′ ]IP

2.2. Agreement morphology

I focus here on the two most prominent systems of agreement inflection in Miskitu. The first, as illustrated in section , is realized on verbs, and marks ordinary subject-verb agreement. I will call this type nominative agreement, although, like in English, the subject in a Miskitu clause bears no overt nominative case marker. In contrast, the second type of agreement appears in the nominal system, and has no correlate in English. I use the term genitive agreement to refer to this phenomenon, which is described in section .

2.2.1. Nominative agreement

In a finite clause, as we saw in (2), the verb is inflected to agree with its subject. The examples in (3) exhibit the full inflection paradigm for the verb kauhw-aia (‘to fall’) in the simple past tense:

(3)

a. yang   prn1   (nani)   (plur)   kauhw-ri   fall-pst1   ‘I (we exclusive) fell’ b. man   prn2   (nani)   (plur)   kauhw-ram   fall-pst2   ‘you sing. (plur.) fell’ c. witin   prn3   (nani)   (plur)   kauhw-an   fall-pst3   ‘(s)he (they) fell’ d. Carlos   Carlos   kauhw-an   fall-pst3   ‘Carlos fell’ e. yawan   prn12   kauhw-an   fall-pst3   ‘we (inclusive) fell’

2.2.2. Genitive agreement

Genitive agreement is inflectional morphology that appears on a noun when it stands in a particular type of syntactic relation with a preceding noun phrase. Semantically, this relation commonly encompasses possession, either material (‘X's book’) or inalienable (‘X's arm’), or kinship (‘X's mother’), but since it is in fact much more general, as we will see later on, I will simply refer to this as a genitive construction and to the position occupied by X as the genitive position. In many languages, such as English, the form of the head noun remains constant in such constructions, while the genitive argument is inflected with visible genitive case. The situation is reversed in Miskitu, and parallels a verb's agreement with its subject. In Miskitu, the head noun itself is marked with the agreement features of the genitive argument, which in turn bears no case inflection at all. The inflected form of a Miskitu noun is commonly referred to as its ‘construct state,’ as opposed to the uninflected ‘absolute’ state.⁴ As an illustration, consider the possession paradigm for the noun aras (‘horse’) below:

(4)

a. aras   horse   ‘horse’ b. yang   prn1   (nani)   (plur)   aras-ki   horse-cns1   ‘my (our excl.) horse’ c. man   prn2   (nani)   (plur)   aras-kam   horse-cns2   ‘your (singular (plural)) horse’ d. witin   prn3   (nani)   (plur)   ai   his   aras-ka   horse-cns3   ‘his/her (their) horse’ e. Carlos   Carlos   (ai)   (his)   aras-ka   horse-cns3   ‘Carlos's horse’ f. yawan   prn12   wan   our   aras-ka   horse-cns3   ‘our incl. horse’

One feature common to both nominative and genitive agreement is that neither conveys any number information whatsoever; only person features are distinguished. Thus, in yang aras-ki (‘my horse’) and yang nani aras-ki (‘our (excl.) horse’), the 1st person agreement marker -ki is constant. Moreover, Miskitu systematically identifies 12 (1st person plural inclusive) inflection with that of 3rd person,⁵ resulting in a final three-way distinction between 1, 2 and 3/12.

3. Basic noun phrase structure

3.1. The canonical DP

Much recent work (going back to Brame 1982, and more fully developed in Abney 1986, 1987) has argued that the true structural head of a noun phrase is not the lexical N itself, but rather a functional head that selects the projection of N as its complement. In this way, the noun phrase more directly parallels the X-bar structure now commonly assumed for the clause (exemplified in (2)), in which Infl selects VP. In particular, Abney's ‘DP-analysis’ suggests that the noun phrase is headed by a determiner (D), which takes NP as its complement. Let us take a representative Miskitu noun phrase⁶ such as aras ba (‘the horse’) from example (1). It is plausible to assume at this point that ba (a Miskitu approximation of the definite article, according to CIDCA 1985) corresponds to the determiner, yielding the following structure for the phrase as a whole:⁷

(5)	[DP DP [NP NP [N N [aras] ]N ]NP [D D [ba] ]D ]DP

This analysis concurs with the general head-final property of Miskitu phrase structure noted in section .

The definite marker ba is one member of an assortment of post-nominal ‘articles’ in Miskitu. I concentrate here on a subset of these, which seem to form a natural core of the determiner system. In (6) is a list of these determiners and a rough characterization of their meanings:⁸

(6)	ba ‘definite, (generally) removed from speaker’ na ‘definite, near speaker’ kum ‘singular’ kumkum ‘several’ nani ‘plural’

I must stress that this paper does not purport to give the correct semantics of the determiners, since we simply do not know enough at this stage in the study of Miskitu. For instance, the ‘definite’ determiners ba and na are not quite definite in the same sense as the English ‘the’, neither do they appear to behave as specific in all cases. Often they have no identifiable semantic content. They have identical syntactic behavior, and differ semantically only in that they serve to locate the NP in different proximity to the speaker (whether it be spatial or discourse-related proximity). I will gloss them both simply as ‘def’ and treat them as interchangeable for purposes of the examples. Of the two, ba is far more common, and often places no restriction on the NP's proximity to the speaker, while na definitely indicates that the NP is near the speaker and/or current in the discourse:

(7)	a. aras   horse   ba   def   ‘the/that horse’ b. aras   horse   na   def   ‘this horse’

The other articles, kum, kumkum, and nani differ from the definites in that they encode only number.

(8)	a. aras   horse   kum   sing   ‘a horse’ b. aras   horse   kumkum   several   ‘several/some horses’ c. aras   horse   nani   plur   ‘horses’

Although DP's of the form in (5) are prevalent, the simplest non-null expansion of the noun phrase in Miskitu consists solely of the head noun itself with no visible determiners. Assuming that all noun phrases project to DP, we must now posit a phonologically null determiner.

(9)	[DP DP [NP NP [N N [aras] ]N ]NP [D D [e] ]D ]DP

3.2. Agreement in D

As we saw in section , the genitive construction in Miskitu is characterized most significantly by the fact that the head noun agrees in person with its genitive argument. The symmetry here between the noun phrase and the clause in Miskitu fits well the parallelism between IP and DP structure. In this section I outline an analysis of Miskitu genitive agreement, based closely on similar proposals (for other languages) by Abney (1987), Ritter (1988, forthcoming), Fassi Fehri (1988), Kornfilt (1984), and Siloni (1989, 1990). As in these accounts, I propose that the agreement morphology is actually a special AGR-bearing determiner.

To illustrate the correspondence between IP and DP structure, let us take the following pair:

(10)	a. yang   prn1   kauhw-ri   fall-pst1   ‘I fell’ b. yang   prn1   aras-ki   horse-cns1   ‘my horse’

I propose that at S-structure the following structural relations hold:

(11)	a. [IP IP [DP DP [yang] ]DP [I′ I′ [VP VP [V V [t] ]V ]VP [I I [[[kauhw] ri]] ]I ]I′ ]IP b. [DP DP [DP DP [yang] ]DP [D D [NP NP [N N [t] ]N ]NP [D D [[[aras] ki]] ]D ]D ]DP

The AGR morpheme overtly marks the special SPEC-head relationship which realizes abstract Case on the ‘subject’ in each structure. I and D are bound morphemes, and accordingly, must affix to an appropriate head. This morphological requirement is satisfied by V-to-I raising in IP, and analogously by N-to-D raising in DP, as shown in (11). See Emonds (1978), Pollock (1989) and Chomsky (1989) for discussion of this type of movement. By contrast, the determiners discussed in section are free morphemes.⁹

The genitive agreement morphology shows up in other environments which do not so obviously involve agreement with a referring expression. For example, in constructions involving demonstrative ‘pronouns’ (those shown in (12)), the demonstrative patterns essentially as if it were a 3rd-person genitive argument. It precedes the head noun, which must be inflected for 3 agreement:¹⁰

(12)	naha ‘this’ baha ‘that’ naura ‘that, distant’ bukra ‘that, even more distant’

(13)	a. baha   that   aras-ka   horse-cns3   ‘that horse’ b. baha   that   aras   horse   ‘that horse’

To account for this I propose that the Miskitu demonstrative pronouns are full DP's, occupying SPEC of DP just as any other genitive argument. The structure of (13) is then the same as that of (11b):

(14)	[DP DP [DP DP [baha] ]DP [D D [NP NP [N N [t] ]N ]NP [D D [[[aras] ka]] ]D ]D ]DP

The semantics of this construction is not particularly troublesome. For instance, the demonstrative baha itself seems to mean something like ‘that place’; hence when combined with the locative case ending -ra, the result means roughly ‘there’:

(15)	yang   prn1   baha-ra   that-LOC   iwi-sna   sit-prp:1pr   ‘I live there’

By simple extension of the semantic relation associated with other genitive configurations, baha aras-ka in (13), which I have translated as ‘that horse,’ will evaluate to something like ‘horse belonging/pertaining to that place.’

4. Multiple determiners in the DP

We have given evidence for the plausibility of a DP analysis of Miskitu noun phrases, showing how this hypothesis offers a unified account of both the free determiners in (6) and the bound D morphemes which mark agreement with a genitive argument. However, any serious attempt to uphold this hypothesis will run into the following fact: Miskitu allows more than one determiner in the same noun phrase. Not only is it common for D to combine with determiners from (6), as shown in (16), but we may also have two free determiners, as in (17).

(16)	a. baha   that   aras-ka   horse-cns3   ba   def   ‘that horse (lit. the that horse)’ b. yang   prn1   aras-ki   horse-cns1   kum   sing   ‘a horse of mine (lit. a my horse)’

In fact, as long there is no clash in number features (e.g. *kum nani), any combination of determiners is permitted. This results in noun phrases with up to three free determiners in addition to the morphologically dependent D, as shown in (18).

The central problem of this paper is how to maintain a constrained DP analysis for Miskitu noun phrases while still accounting for the multiple determiner phenomenon (‘D*’ for short, adopting the Kleene star notation). Clearly, if D selects only NP, then there can never be room for more than one determiner in a given DP. To attempt to solve this problem, we might augment our conception of the DP analysis in one of two ways, which can be summarized as follows:

(20)

a. ‘DP-recursion’ D selects either NP or another DP, allowing in principle an infinite number of consecutive D projections. b. ‘Functional proliferation’ The projection of a noun phrase includes a number of different functional categories (e.g. NUMP as proposed by Ritter (1992, forthcoming)) intervening between D and NP.

4.1. Relative Clauses

The first clue to unraveling the phrase structure of the D* noun phrase (or `D*P') in Miskitu, has to do with some interesting facts about relative clauses (RC's) in the language. First, they are often internally headed. In other words, a relativized noun phrase on the surface can look essentially like an ordinary clause which has been combined with the determiner system of a noun phrase. For example, the clause in (23) could, in an argument position, be interpreted as a relative clause with the semantic head upla (‘person’), giving the meaning ‘person who fell.’ This generally involves placing one or more determiners after the clause, whose status as a noun phrase is thereby made explicit. This is shown in (24).

(23)	upla   person   kauhw-an   fall-pst3   ‘someone fell’

(24)	yang   prn1   [      upla   person   kauhw-an   fall-pst3   ba   def   ]      nuu   know   sna   be:pres1   ‘I know the person who fell12’

In the spirit of Williamson's (1987) analysis of Lakhota, I will take a ‘minimalist’ position on the constituent structure of internally-headed relative clause (IHRC) constructions. To begin with, I propose that, in addition to selecting NP, D may also select IP.¹³ Thus, the base component of Miskitu grammar must allow (at least) the following two expansions of D:

(25)	a. DNP D b. DIP D

Relative clause constructions involve the expansion in (25b), while (25a) represents the default DP without relativization. Under this proposal, then, the structure for the DP upla kauhw-an ba in (24) is as follows:

(26)	[DP DP [IP IP [DP DP [NP NP [upla] ]NP [D D [e] ]D ]DP [I′ I′ [VP VP [V V [t] ]V ]VP [I I [[[kauhw] an]] ]I ]I′ ]IP [D D [ba] ]D ]DP

Adopting some terminology from Williamson (1987), I will refer to DP in (26) as the mother and DP as the internal head. I will discuss the compositional semantics of such a structure section .

4.2. Null pronominals in Miskitu

A second property of Miskitu syntax which will be necessary to our understanding of the multiple determiner phenomenon, is the allowance of null arguments in Case-marked positions. In traditional terminology, Miskitu is a ‘pro-drop’ language. The ‘licensing’ of the null pronominal pro has often been tied to the availability of agreement morphology sufficiently rich to recover its grammatical features. Under the current proposal, the syntactic configurations of nominative and genitive agreement are essentially identical, and involve a SPEC-head relationship between an argument and an AGR-bearing functional head. Null pronominals are indeed common in both these environments, as shown in (27) (compare with (10)):¹⁴

(27)	a. pro     kauhw-ri   fall-pst1   ‘I fell’ b. pro     aras-ki   horse-cns1   ‘my horse’

The use of null pronominals is widespread in Miskitu speech, and in particular, pro can often be found as the semantic head of a relative clause. This yields ‘reduced’ relative expressions, such as the following:

(28)	pro     aiwanan   sing:pst3   ba   def   ‘the one who sang, (s)he who sang’

I call these reduced relatives for ease of exposition, although of course there is nothing reduced about their syntactic structure.

An important variety of reduced relative is the ‘bare predicate nominal’ RC, illustrated in (29):

(29)	pro     aras   horse   kata   be:pst3A   ba   def   ‘the horse-PAST (roughly, the x that was of type HORSE)’

Although this type of RC construction is resistant to English translation, it is exceedingly common in discourse, and occurs very naturally in narratives set in the past tense. The following is a typical example of its usage:

(30)	pro     yakan   alone   [      pro     plun   food   kata   be:pst3A   ba   def   ]      pin   eat:pst3   ‘he ate the food (lit. the x that was food) alone’

4.3. SA-deletion

Since bare predicate nominal RC's (as in (29)—(30)) are so common in the past tense, one would expect to find their counterparts in other tenses. However, this expectation is only partially borne out. (31) is a common example of such a construction in the future, or ‘irrealis’ tense:¹⁵

(31)	[      pro     ani   which   aras-ka   horse-cns3   kabia   be:irr3   ]      pain   good   kaia   be:inf   kan   be:pst3   ‘any/whichever horse would be fine’ (lit. x that might be which horse...)

Turning to the present tense, however, it looks surprisingly as if there are no cases of bare predicate nominal RC's. We know what we would expect such a construction to look like. Taking the example from (30), plun kata ba (‘the x that was food’), we should be able to replace kata (the psa3 form¹⁶ of the copular verb k-aia) with the corresponding present tense form sa to yield an expression meaning ‘the x that IS food:’

(32)	pro     plun   food   sa   be:pres3   ba   def   ‘the x that is food’

It would perhaps be fair to expect (32) to occur less frequently than its past tense counterpart, since it is hard to imagine contexts in which the contribution of the present tense in this type of structure would be noticeable, but its shear ungrammaticality is startling. It is difficult to imagine on what basis (32) might be ruled out. However, its ungrammaticality in fact has a simple (although mysterious) explanation. To see this clearly, we must take a closer look at the use of the verb k-aia in Miskitu.

In (3) we illustrated the simple past tense paradigm for the verb kauhw-aia. While there also exists a morphologically simple present tense verb paradigm, its aspectual properties (still poorly understood) make it unappropriate in most present tense situations. Instead, what is generally used is the ‘compound present’ tense. The compound tenses in Miskitu are not really separate tenses at all; rather, they are a kind of serial construction (see Salamanca 1988, Hale 1989) in which the main verb is placed in an uninflected participial form and combined with a finite form of k-aia. In a manner parallel to the English progressive, the tense of these Miskitu compound verb constructions is given by the tense of the k-aia form, which seems to contribute nothing else semantically. A partial table of the inflectional paradigm of k-aia is given below:

(33)	K-AIA (‘to be’) pres pst psa irr 1st person (1) sna kapri katna kamna 2nd person (2) sma kapram katma kama 3rd person (3/12) sa kan kata kabia

The proximate participial (prox) form of kauhw-aia is kauhw-i, and therefore the compound present tense forms are as shown in (34):

(34)	a. kauhwi   fall:prox   sna   be:pres1   ‘I fall/am falling’ b. kauhwi   fall:prox   sma   be:pres2   ‘you fall/are falling’ c. kauhwi   fall:prox   sa   be:pres3   ‘(s)he falls/is falling’

The participial main verb is a dependent form, and cannot normally occur in absence of a finite verb:

(35)	a. witin   prn3   kauhwi   fall:prox   sa   be:pres3   b. witin   prn3   kauhwi   fall:prox   ∅     ‘he is falling’

However, inside a relative clause, we find the opposite pattern. The pres3 form sa in this environment renders (36a) ungrammatical, while suddenly the case with the ‘unsupported’ participial (36b) is well-formed:

(36)	a. witin   prn3   kauhwi   fall:prox   sa   be:pres3   ba   def   b. witin   prn3   kauhwi   fall:prox   ∅     ba   def   ‘(s)he who is falling’

This ‘deletion’ does not occur in the non-3rd-person forms of k-aia, as in (37), nor in non-present forms, as in (38):

(37)	a. man   prn2   kauhwi   fall:prox   sma   be:pres2   ba   def   b. man   prn2   kauhwi   fall:prox   ∅   be:pres2   ba   def   ‘you who are falling’ (i.e. cannot force pres2 reading)

(38)	a. witin   prn3   kauhwi   fall:prox   kata   be:psa3   ba   def   b. witin   prn3   kauhwi   fall:prox   ∅   be:psa3   ba   def   ‘he who was falling’ (i.e. cannot force psa3 reading)

In fact, it appears that the null spellout of pres3 is the only change in the inflectional paradigm of k-aia in RC's:

(39)	K-AIA (‘to be’) in relative clauses pres pst psa irr 1st person (1) sna kapri katna kamna 2nd person (2) sma kapram katma kama 3rd person (3/12) ∅ kan kata kabia

I have no explanation for this gap (which I will refer to as ‘SA-deletion’ for ease of exposition, although I make no claims about its status as an actual deletion rule) in the inflectional behavior of k-aia in the different clause types; for the matter at hand, it is sufficient simply to note that SA-deletion occurs.

As promised above, we now have a simple account of the ungrammaticality of (32). There is nothing wrong with (32) except that SA-deletion has failed to ‘apply.’ The correct form is thus:

(40)	pro     plun   food   ∅   be:pres3   ba   def   ‘the x that is food’

SA-deletion has given us a very interesting result. The RC construction in (40) is string-identical to a DP of much simpler structure. The two contrasting structures are shown in (41)—(42):¹⁷

(41)	[DP DP [NP NP [N N [plun] ]N ]NP [D D [ba] ]D ]DP

(42)	[DP DP [IP IP [pro] [I′ I′ [DP DP [NP NP [plun] ]NP [D D [e] ]D ]DP [[SA]] ]I′ ]IP [D D [ba] ]D ]DP

4.4. Explaining D*

We now in fact predict the existence of D*P's in Miskitu. Let us suppose that neither alternative in (20) is true. Rather, the only selectional possibilities for D are those given in (25). The difference between (41) and (42) now becomes crucial. Semantically, the two are only minimally distinguishable.¹⁸ However, the former contains two D positions, and the latter only one. In fact, we can extract out of (41) the following basic iterative unit, which we can think of as functioning simply to provide room for a D position:

(43)	[DP DP [IP IP [pro] [I′ I′ [XP … [SA]] ]I′ ]IP [D D [∅] ]D ]DP

Now, to arrive at a sequence of n determiners, n-1 of the units in (43) may be stacked on top of a ‘terminal’ DP, which contains the initial (i.e. lowest) D in the sequence. This innermost DP is the predicate of a bare predicate nominal RC, and the pro subject is the internal head.¹⁹ Thus the DP in (44), which contains four determiners (one bound, three free), is realized by the structure in (45), which contains three of the units in (43), one for each of the three top-most D positions.

(44)	baha   that   aras-ka   horse-cns3   kumkum   SEVERAL   nani   plur   ba   def   ‘those several horses’

(45)	[DP DP [IP IP [pro] [I′ I′ [DP DP [IP IP [pro] [I′ I′ [DP DP [IP IP [pro] [I′ I′ [DP DP [DP DP [baha] ]DP [D D [NP NP [N N [t] ]N ]NP [D D [[[aras] ka]] ]D ]D ]DP [[SA]] ]I′ ]IP [D D [kumkum] ]D ]DP [[SA]] ]I′ ]IP [D D [nani] ]D ]DP [[SA]] ]I′ ]IP [D D [ba] ]D ]DP

Summarizing briefly then, for any sequence of two determiners (the first of which may be D), we may assume that the first one is the D of a predicate nominal within an IHRC. The second is the D of the mother. The reason the two determiners appear to be adjacent is that SA-deletion ensures that the intervening copula is phonetically null.

4.5. Why def must come last

We showed in the last sections how the D* phenomenon can be seen as following from the interaction of different areas of Miskitu grammar, without need for any extension of the basic DP model except for the ability of D to select IP (which we proposed to handle IHRC's). However, we still have not provided any motivation for the constraint that def must be the last determiner in any sequence. But we can now see a straightforward explanation for this fact, given our analysis. In a sequence of determiners, the analysis provides a key structural difference between the all the non-final D's and the final one: the non-final ones are the heads of predicate nominal DP's, whereas the final D is the head of the mother DP. It remains only to observe that in Miskitu a predicate nominal may not have def as its determiner:

(46)	baha   that   waitnika   male:cns3   yang   prn1   aisiki   father:cns1   (*ba)   (def)   sa   be:pres3   ‘that man is my father’

I assume, along the lines of Higginbotham (1987), that def obligatorily binds the open place in its NP, and therefore the DP which it heads is necessarily saturated. Since predicates by definition are unsaturated, we have an explanation for the ungrammaticality of (46), and, more importantly, an account for why def must be highest determiner in any Miskitu D*P. Using the iterative unit in (43), there is no way to generate any D positions higher than a def determiner, since there is no way to put that def determiner in predicate position.

5. Postnominal modifiers

In sections and we addressed the issue of the syntactic position of the determiners in Miskitu. We showed that the existence of multiple determiner sequences requires no extra assumptions about the phrase structure of noun phrases, given that we must independently allow for the way Miskitu forms relative clauses, and the fact that SA-deletion occurs. In this section, we turn to the other half of the Miskitu postnominal system, the nominal and adjectival restrictive modifiers.

Adjectives and nominal modifiers come after the noun in Miskitu:

(47)	a. yul   dog   mairin   female/woman   ba   def   ‘the female dog’ b. yul   dog   pihni   white   ba   def   ‘the white dog’

The postnominal modifiers exist along side all combinations of determiners discussed above, with the following generalizations about relative ordering: first, D is of course the innermost, due to its morphological requirement; second, def is always final; otherwise the order is not fixed, although in the more complex cases there appears to be a preference for modifiers to appear inside the determiners. A few examples are shown below:

(49)	a. yul   dog   mairin   female   kum   sing   (ba)   def   ‘a (specific) female dog’ b. yul kum mairin (ba) (same trans.)

(50)	a. yul   dog   mairin   female   pihni   white   kum   sing   (ba)   def   ‘a (specific) white female dog’ b. yul mairin kum pihni (ba) c. yul pihni mairin kum (ba) d. yul pihni kum mairin (ba) (all same trans.)

5.1. The syntactic position of postnominal modifiers

One might assume, even for simple cases like those in (47), that modifiers are adjoined directly somewhere into the structure of an NP or DP, subject to some principle ensuring that they follow the noun they modify. Our analysis of D*P's suggests another option, without further complicating the phrase structure of the DP itself. Specifically, I propose that these restrictive modifiers are actually in predicate position of a RC:

(51)	[DP DP [IP IP [DP DP [NP NP [yul] ]NP [D D [e] ]D ]DP [I′ I′ [AP AP [pihni] ]AP [[SA]] ]I′ ]IP [D D [ba] ]D ]DP

(52)	[DP DP [IP IP [DP DP [NP NP [yul] ]NP [D D [e] ]D ]DP [I′ I′ [DP DP [NP NP [mairin] ]NP [D D [e] ]D ]DP [[SA]] ]I′ ]IP [D D [ba] ]D ]DP

These are ordinary IHRC's, in which SA-deletion occurs. The difference between these cases and those seen above in the discussion of D*P's is that here the head noun is a DP in the subject position of the bottommost RC, while the post-modifier is in predicate position. That is, the head noun in these cases is the internal head of an IHRC. Thus, the way to say ‘white dog’ in Miskitu is literally to say ‘dog that is white.’ This analysis also correctly predicts that adjectives and nominal modifiers must follow the noun they modify, since the predicate follows the subject of a clause in Miskitu. Taking a more complex example such as (53), then, we arrive at the structure in (54).

(53)	baha   that   yul-ka   dog-cns3   mairin   female   pihni   white   kumkum   SEVERAL   nani   plur   ba   def   ‘those several white female dogs’

(54)

[DP1 DP1 [IP1 IP1 [pro] [I′ I′ [DP2 DP2 [IP2 IP2 [pro] [I′ I′ [DP3 DP3 [IP3 IP3 [DP4 DP4 [IP4 IP4 [DP5 DP5 [DP DP [baha] ]DP [D D [NP NP [N N [t] ]N ]NP [D D [[[yul] ka]] ]D ]D ]DP5 [I′ I′ [DP DP [NP NP [mairin] ]NP [D D [e] ]D ]DP [[SA]] ]I′ ]IP4 [D D [e] ]D ]DP4 [I′ I′ [AP AP [pihni] ]AP [[SA]] ]I′ ]IP3 [D D [kumkum] ]D ]DP3 [[SA]] ]I′ ]IP2 [D D [nani] ]D ]DP2 [[SA]] ]I′ ]IP1 [D D [ba] ]D ]DP1

This analysis runs into a potential problem in relation to the constraint mentioned above, that def must appear outside all other determiners and modifiers. The problem is this: in section we accounted for the restriction def by noting that all inner determiners in a sequence are in predicate position, where def is independently not allowed. However, in DP's with postnominal modifiers, some of the inner determiners are not in predicate position, but rather they are in DP's that are the internal head of an IHRC. That is, we have no explanation yet for the ungrammaticality of DP's like (55) (whose structure is shown in (56).

(55)	yul   dog   ba   def   pihni   white   kum   sing   ‘the white female dog’

(56)	[DP DP [IP IP [DP DP [NP NP [yul] ]NP [D D [ba] ]D ]DP [I′ I′ [AP AP [pihni] ]AP [[SA]] ]I′ ]IP [D D [kum] ]D ]DP

We will see in the next section that the answer to this problem lies in a constraint against def on the head of an IHRC.

5.2. The def-constraint

As noted in section , the basic structure I have proposed here for relative clause constructions is taken directly (with a few category changes) from Williamson's (1987) analysis of Lakhota internally-headed relatives. Williamson provides arguments that in Lakhota the semantic head of an IHRC adjoins to its clause at LF. She also notes that there is an indefiniteness restriction on this head noun phrase. Taking Heim's (1982) view of indefinites as essentially open sentences containing an ‘unused’ variable, then LF composition (intersection) of the adjoined indefinite phrase and the clause from which it extracted produces exactly the correct semantics for restrictive relativization. If the adjoined head is not an indefinite expression, and has no free variable position, it will not compose correctly with the open sentence it is adjoined to.

Like the indefiniteness requirement on the Lakhota internal head, Miskitu also places a restriction on the head DP in an IHRC construction. In Miskitu the internal head may not bear a def article. I will assume that this follows from some principle which I will label the def-constraint, whatever its real nature might turn out to be. While this is quite similar in spirit to the indefiniteness restriction of Lakhota, it has a much narrower, more purely formal character. The Lakhota restriction excludes a whole semantically motivated class of ‘quantified expressions’ (Milsark 1974). This set embraces the definite article, demonstratives, proper names, and quantifiers of a specific type—all, every, most, etc. By contrast, in Miskitu, so far as I can determine, the restriction excludes only the def article.²¹

Thus, while (58a-b) are more or less equivalent to (57), (58c) does not have this reading:

(57)	[      pâniki   friend:cns1   sula   deer   kaikan   see:pst3   ba   def   ]      lap tiwan   disappear:pst3   ‘the deer that my friend saw disappeared’

(58)

a. [      pâniki   friend:cns1   baha   that   sualia   deer:cns3   kaikan   see:pst3   ba   def   ]      lap tiwan   disappear:pst3   ‘that deer that my friend saw disappeared’ b. [      pâniki   friend:cns1   sula   deer   kum   sing   kaikan   see:pst3   ba   def   ]      lap tiwan   disappear:pst3   ‘the deer that my friend saw disappeared’ c. [      pâniki   friend:cns1   sula   deer   (*ba)   def   kaikan   see:pst3   (ba)   def   ]      lap tiwan   disappear:pst3   ‘*the deer that my friend saw disappeared’ ‘my friend who saw the deer disappeared’

As the translation of (58c) suggests, there is another reading available for all these cases, where it is the noun phrase pâniki (‘my friend’ or ‘friend of mine’) that is the internal head. This interpretation is not favored, but is forced when the def-constraint prevents the more salient reading. Since Miskitu does not have an overt marker indicating that a noun is the head of a RC, this kind of ambiguity is a general problem for IHRC's with more than one head candidate. Consider for instance the sentence in (59) (from Salamanca 1988:244). The two translations reflect the fact that either of the underlined DP's may be construed as the internal head.

(59)	[      tuktika   child:cns3   mairka-ra   female:3c-acc   kaikan   see:pst3   ba   def   ]      balan   come:pst3   ‘[the child who saw the woman] came’ ‘[the woman who the child saw] came’

The prohibition against def on the head is demonstrated by examples (60—62).²² In (60) and (61) IHRC's are disambiguated by adding def to one of the noun phrases, thus removing the possibility of interpreting it as the head. In (62), both noun phrases are closed off by def, leaving no possible head for the clause and thus rendering it uninterpretable as an IHRC.²³

(60)	[      tuktika   child:cns3   ba   def   mairka-ra   female:3c-acc   kaikan   see:pst3   ba   def   ]      balan   come:pst3   ‘[the woman who the child saw] came’

(61)	[      tuktika   child:cns3   mairka   female:cns3   ba-ra   def-acc   kaikan   see:pst3   ba   def   ]      balan   come:pst3   ‘[the child who saw the woman] came’

(62)	[      tuktika   child:cns3   ba   def   mairka   female:3c-acc   ba-ra   def-acc   kaikan   see:pst3   ba   def   ]      balan   come:pst3

Despite the difference between Miskitu and Lakhota regarding the exact type of restriction on the internal head, let us attempt to apply the spirit of Williamson's analysis to our Miskitu data.

(63)	pâniki   friend:cns1   sula   deer   kaikan   see:pst3   ba   def   ‘the deer that my friend saw’

Taking the IHRC in (57) (repeated in isolation as (63)) as a concrete example, suppose that since the DP sula (‘deer’) meets the def-constraint, it may be interpreted as containing an open variable position. Thus, semantically, sula in this case means ‘deer(x).’ At LF, after sula adjoins to IP, we have the structure in (64):

(64)	[DP DP [IP IP [IP IP [pâniki t kaikan] [DP DP [sula] ]DP ]IP [D D [ba] ]D ]IP ]DP

The original IP now contains a gap, and is itself an open sentence, corresponding to the set picked out by ‘my friend saw x.’ By semantic composition, the entire IP complement evaluates to the conjunction of the two open sentences (or equivalently, the intersection of their corresponding sets): (my friend saw x) AND deer(x). Now at a purely formal level, the function of def might be to bind the open position in its complement, creating a strictly referential expression. We then end up with ‘the x such that (my friend saw x) AND deer(x).’ This appears to be the correct characterization of the translation ‘the deer that my friend saw.’

6. Summary

In this paper I have presented a wide array of facts attesting to the surface complexity of the noun phrase in Miskitu. I made only two specific theoretical proposals, namely that the construct state morpheme is a determiner, and that D selects IP in IHRC's. Neither of these is controversial or necessary to the conclusion of the paper. The problem of the paper, in brief terms, is how to fit all the postnominal material in a the Miskitu noun phrase into any constrained theory of phrase structure. I have not argued against any particular theory of how to do this. Instead, what I have tried to show is that no extra theory is required in the first place. I have pointed to several empirical phenomena in Miskitu grammar: the internally-headed relative clause, pro-drop, SA-deletion, and the constraint against def in predicate nominals and in the internal heads of IHRC's. Whatever the eventual theoretical explanations for any of these individual phenomena, I have shown that they conspire to offer a free solution to the problem of the overpopulated postnominal system of Miskitu. This is a theory-neutral outcome. A theory that relates the facts here to DP-recursion, functional proliferation, or adjunction of modifiers into DP structure, must necessarily include a specific means of ruling out structures like (45) and (54).

Notes

^*This research would not have been possible without the enormous amount of help and support I have received from Ken Hale, of which the hours of profitable discussion of Miskitu grammar are only a part. I am equally indebted to the people of Karawala, Nicaragua, for making my exposure to Miskitu so easy and enjoyable; particularly the following people, whose judgments are directly represented in this paper: Leonzo Knight Julián, Abanel Lacayo Blanco, Lorinda Martínez Lacayo, Francisco Santiago William, Kandler Santiago, and Clementina Simón. Alejandro Avilés, Jorge Matamoros and Melba McClean also provided some crucial judgments. In addition, I want to thank the following people for helpful discussion and comments: Chris Collins, Hamida Demirdache, Irene Heim, Jim Higginbotham, Utpal Lahiri, Alec Marantz, Kumiko Murasugi, Rolf Noyer, Wayne O'Neil, David Pesetsky, Danilo Salamanca, and the participants in the Spring 1991 Syntax-Semantics Workshop: Doug Jones, Toshifusa Oka, Sigal Uziel, and Akira Watanabe. Finally, I thank Danilo Salamanca and CIDCA for making my four trips to Karawala in 1991 and 1992 feasible in the first place. Any errors in this study are of course my own. This research was funded by NSF grant #RCD-9054772 and the MIT Dean's Faculty Development Fund.

¹Not surprisingly from a cross-linguistic standpoint, the two categories are not clearly distinct in Miskitu. For instance, -ra is a morpheme whose function ranges from locative to simple accusative case.

²Throughout the paper, we use the following gloss codes:

³I will not deal with the issue of whether the subject argument is actually generated in SPEC of IP, or whether it raises (to receive Case, for example) from a VP-internal position. Likewise, I assume V-to-I raising without argument (see section ). These issues are of no relevance to the conclusions of this paper.

⁴This terminology, reflected here in the glosses cns1, cns2, and cns3, owes its origin to classical studies of a similar phenomenon in Hebrew.

⁵With nominative agreement, the verb morphology is the same for both 12 and 3 (e.g. kauhw-an in each case in (3)), and for genitive agreement the head noun also bares identical inflection in each case, although in fact the two are differentiated by the addition of the proclitic elements ai and wan. I do not count these clitics as part of the agreement morphology proper, and will ignore them in this paper.

⁶I continue to use the term noun phrase here in the theory-neutral sense to refer to the entire constituent, whatever its exact categorial specification may be.

⁷I will omit wherever possible intermediate projections such as N′ and D′.

⁸Cardinal numbers, along with the corresponding WH-element an (‘how many’) also seem to pattern with these determiners. In addition, CIDCA (1985) reports the use of âp (from English ‘half’) with identical meaning and distribution as kumkum.

⁹Actually, at least na and ba, and possibly kum, are phonologically dependent, in that they are unstressed and must adjoin to some phonologically overt material. The difference is that they do not morphologically subcategorize for an N the way D does.

¹⁰The difference is that demonstratives are incompatible with the proclitic ai, unlike other 3rd-person genitive arguments.

¹¹There is no indication that the presence of nani in such phrases contributes anything to their semantic interpretation, since the meaning of kumkum (‘several’) by itself ensures a plural reading.

¹²Actually, clauses of this sort when followed by def have a non-relative interpretation available also, which is factive in character. So (24) also has the reading ‘I know that someone fell.’

¹³Or possibly CP. The choice has no bearing here.

¹⁴A pro which denotes the current ‘state of affairs’ in the discourse is required to account for the so-called ‘autonomous’ construct (Alpher and Hale, n.d.), in which a noun appears in the (3rd person) construct state with neither an overt genitive argument nor the proclitic ai. The autonomous construct has a kind of definite flavor, which would follow from its compositional interpretation as belonging or pertaining to the current state of affairs in the discourse.

¹⁵In this example, D is null. While a number determiner, kum, for instance, would be possible, def is to my knowledge incompatible with the irrealis verb in the relative clause.

¹⁶The ‘indefinite’ past form kan (be:pst3) of k-aia is also possible, though slightly less common, in bare predicate nominal RC's.

¹⁷In the diagrams I use [SA] to notate the ‘deleted’ pres3 form of k-aia.

¹⁸The difference is on a par with the difference between ‘the red book’ and ‘the book that is red.’

¹⁹I am assuming that a predicate nominal is not an argument and therefore could not itself head a relative construction.

²⁰The ordering among the postnominal modifiers is not entirely free. I will not have anything to say about this, but presumably it has the same status as the restrictions in English, where ‘small white dog’ is strongly preferable to ‘white small dog’.

²¹Proper names and pronouns are acceptable as internal heads. Not surprisingly, they also accept the full range of determiners, including the construct state (D). Thus it appears that Miskitu pronouns and names are at some basic level descriptions rather than intrinsically referential expressions. The same holds for possessed nouns and demonstratives as well. Finally, quantifiers corresponding to the English ‘all,’ ‘every,’ ‘each,’ and ‘most,’ insofar as they can be identified in Miskitu, seem to be possible as modifiers of internal heads, but this has not been investigated systematically.

²²The identification of the semantic head of an IHRC may also be strongly influenced by other factors such as intonation and pausing (while still retaining the restrictive meaning), as well as the animacy of the noun phrases, and there may also be a tendency to favor the direct object over other candidates in the most salient reading.

²³In (62), the phrase tuktika ba mairka ba-ra kaikan ba is presumably not uninterpretable, but rather, as mentioned in footnote above, it could in a different context be interpreted as ‘the fact that the child saw the woman’ or ‘the event in which the child saw the woman’.

References

1. Abney, S. (1986) ‘Functional Elements and Licensing,’ paper presented to GLOW 1986, Gerona, Spain.
2. Abney, S. (1987) The English Noun Phrase in its Sentential Aspect, PhD dissertation, MIT, Cambridge, MA.
3. Alpher, Barry, and Ken Hale (n.d.) ‘On Relativization in Miskito,’ ms., Sydney and MIT, Cambridge, MA.
4. Brame, Michael (1982) ‘The Head-Selector Theory of Lexical Specifications and the Nonexistence of Coarse Categories,’ Linguistic Analysis 10, 321—325.
5. Chomsky, Noam (1989) ‘Some Notes on the Economy of Derivation and Representation,’ MITWPL 10, 43—74.
6. CIDCA (Centro de Investigaciones y Documentación de la Costa Atlántica). (1985) Miskitu Bila Aisanka: Gramática Miskita, CIDCA, Managua.
7. Emonds, Joseph (1978) ‘The Verbal Complex V′−-V in French,’ Linguistic Inquiry 9, 151—175.
8. Fassi Fehri, Abdelkader (1989) ‘Generalized IP Structure, Case and VS Word Order,’ MITWPL 10, 75—113.
9. Hale, Ken (1989) ‘Misumalpan Verb Sequencing Constructions,’ in Claire Lefebvre, ed., Proceedings of the Second Niger-Congo Syntax and Semantics Workshop, section 1, part 2: Serial Verbs, 1—21.
10. Heath, George Reinke (1927) Grammar of the Miskito Language, F. Lindenbein, Herrnhut.
11. Heim, Irene (1982) The Semantics of Definite and Indefinite Noun Phrases, PhD dissertation, University of Massachusetts, Amherst. Higginbotham, James (1987) ‘Indefiniteness and Predication,’ in Reuland and ter Meulen, eds., 43—97.
12. Horrocks, G, and M. Stavrou (1987) ‘Bounding Theory and Greek Syntax: Evidence for Wh-Movement in NP,’ Journal of Linguistics 23, 79—108.
13. Koopman, Hilda (1984) The Syntax of Verbs, Foris, Dordrecht.
14. Kornfilt, J. (1984) Case Marking, Agreement, and Empty Categories in Turkish, PhD dissertation, Harvard University, Cambridge, MA.
15. Milsark, Gary (1974) Existential Sentences in English, PhD dissertation, MIT, Cambridge, MA.
16. Pollock, Jean-Yves (1989) ‘Verb Movement, Universal Grammar, and the Structure of IP,’ Linguistic Inquiry 20, 365—424.
17. Reuland, Eric (1983) ‘Governing -ing,’ Linguistic Inquiry 14, 101—136.
18. Reuland, Eric and Alice ter Meulen, eds. (1987) The Representation of (In)definiteness, MIT Press, Cambridge, MA.
19. Ritter, Elizabeth (1988) ‘A Head-Movement Approach to Construct-State Noun Phrases,’ Linguistics 26.6, 909—929.
20. Ritter, Elizabeth (1992) ‘Cross-Linguistic Evidence for Number Phrase,’ Canadian Journal of Linguistics 37(2), 197—218.
21. Ritter, Elizabeth (forthcoming) ‘Two Functional Categories in Noun Phrases: Evidence from Modern Hebrew,’ to appear in Susan Rothstein, ed., Syntax and Semantics 26, Perspectives on Phrase Structure: Heads and Licensing, Academic Press, New York.
22. Salamanca, Danilo (1988) Elementos de Gramática del Miskito, PhD dissertation, MIT, Cambridge, MA.
23. Siloni, Tali (1989) ‘On the Parallelism between CP and DP: The Case of Hebrew Semi-Relatives,’ in Proceedings of the Leiden Conference for Junior Linguists (LCJL) 1, University of Leiden, 135—153.
24. Siloni, Tali (1990) ‘Hebrew Noun Phrases: Generalized Noun Raising,’ ms., Université de Genève.
25. Szabolcsi, Anna (1987) ‘Functional Categories in the Noun Phrase,’ in Kenesei, ed., Approaches to Hungarian, volume 2, University of Budapest, 167—189.
26. Williamson, Janis (1987) ‘An Indefiniteness Restriction for Relative Clauses in Lakhota,’ in Reuland and ter Meulen, eds., 168—190.