2.1 Determiner-like elements

Malagasy has two determiner-like elements. Conforming to the head-initial nature of the language, these articles appear in a DP-initial position. The default article ny is often translated as ‘the’, but the consensus is that it is not equivalent (Fugier 1999; Law 2006; Keenan 2008; Paul 2009; others). In addition to more usual definite uses, ny allows indefinite or novel readings, as illustrated in (4). There is no overt indefinite article.

In addition to ny, there is an anaphoric article ilay, which Rajaona (1972) calls deictic. It is used when referring to objects that have been previously mentioned or are otherwise salient, as in (5) and (4a).

While ny is unmarked for number, ilay is typically understood as singular. There are, however, cases where ilay is compatible with plural reference (B. Ralalaoherivony, p.c.), such as in (6).

Singular and plural distinctions are overtly marked only on demonstratives.³

We consider ny and ilay as determiner-like in part because they are in complementary distribution, (7a, b), and they do not have intransitive, pronominal uses, (7c). As we will see below, this distribution contrasts with the demonstratives.

2.2 Demonstratives

The demonstrative system in Malagasy is very rich, encoding not only distance (proximal, distal, and neutral) and number (singular and plural), but also visibility (visible and non-visible) and “boundedness”. Bounded and unbounded indicate whether the space that the object occupies is seen as enclosed or open. There are also demonstratives for generics. As shown in Tables 1 and 2, adapted from Rajaona (1972: 633) and Imai (2003), some distinctions are leveled for certain forms (e.g. ireto is the plural, proximal demonstrative, but it is unmarked for bounded versus unbounded).

The reader will note clear morphological regularities (see Rajaona 1972: 623–632 for some discussion), but we set those aside here, focusing instead on the external syntax of demonstratives.

As noted above, the framing pattern for demonstratives is required in Malagasy, with some exceptions to be discussed in 2.3. A second crucial fact about the framing demonstratives is that they must be identical. The initial and final demonstratives must be exactly the same, (8); we have not encountered any text or elicited examples where the demonstratives do not match.⁴

Framing demonstratives are strictly initial and final in the nominal. They frame any dependents that occur in the nominal. We illustrate non-exhaustive possibilities in (9). (9a) shows framing of a noun, adjective, and possessor; (9b) shows framing of a noun, numeral, and PP modifier; and (9c) shows framing of a noun, quantifier, and relative clause. In no cases can either demonstrative appear internal to the nominal instead of in the peripheral positions.

We have encountered two exceptions to this pattern, two elements that can exceptionally appear after the final demonstrative. The first is exceptive phrases that modify a DP and form a constituent with it, which must occur after the final demonstrative, (10). The second case is non-restrictive relative clauses, which may optionally appear after the final demonstrative, (11).⁵ We assume that, in both instances, the modifier must or can adjoin to the right of the nominal, which will place it strictly final under most any structural analysis.

We have nothing more to say about these cases as they do not appear to have any implications for the correct analysis of the framing pattern.

2.3 Non-framing uses of demonstratives

The demonstratives have three non-framing uses which we discuss here. First, the most prominent non-framing use is what we will call the pronominal function, or pronominal demonstratives. The demonstrative appears alone, unmodified, and cannot be doubled, (12).⁶ All demonstratives can appear in this use as far as we are aware, and they can appear in any position where DPs can appear (e.g. subject, object, object of a preposition).

Cross-linguistically, this is a common use of demonstratives (Alexiadou et al. 2007: 95); nevertheless, it is perhaps surprising that pronominal demonstratives are never doubled, where they might be framing a null noun. Accounting for this pattern will be important in the analyses to follow.

Second, a lone final DEM2 demonstrative is sometimes possible if the initial determiner is ilay, (13). The interpretation of such nominals combines the anaphoric meaning of ilay with the parameters of number, visibility, distance from the speaker, and boundedness of the demonstrative that is used. Rajaona (1972: 686) indicates that ilay is only compatible with [+visible] demonstratives but not those that are [–visible], except for izay. We have not systematically investigated which demonstratives are possible in this context.

The presence of ilay precludes the framing pattern. Only the final instance of the demonstrative is possible:

Ilay can cooccur with DEM2 without any intervening nominal material:

Turning to ny ‘det’, a demonstrative can never cooccur with ny, with or without nominal material, (16).

Malagasy has a null determiner, which is possible in object position, (17a); however, DEM2 is not possible here either, (17b).⁷

Finally, we have determined that some initial DEM1 demonstratives can appear alone, without DEM2. This option is both lexically and idiolectally restricted in that most demonstratives are not possible in this pattern and the ones that are possible vary among speakers. All our speakers accept the proximal and neutral plural demonstratives ireto and ireo in this pattern, (18a). Some speakers accept the corresponding singular demonstratives, ity and io, as well, (18b). No speakers that we have consulted accept any of the other demonstrative as a lone DEM1, (18c).⁸

We analyze these cases as a reanalysis of certain demonstratives as determiners (Rajaona 1972). This reanalysis is subject to inter-speaker variation, and we set it aside for the purposes of this paper.

2.4 Against DEM2 as a reinforcer

Before continuing to our analysis, we would like to dismiss treating DEM2 as a demonstrative reinforcer. There is a rich literature on demonstrative-reinforcer constructions in Germanic and Romance languages, exemplified in (19) (Brugè 1996; Bernstein 1997; Leu 2007; 2008; Roehrs 2010).

Despite superficial similarity between Malagasy framing demonstratives and demonstrative reinforcers, we believe that the two are distinct phenomena. There are notable differences between framing demonstratives and reinforcers. First, demonstrative doubling in Malagasy always involves identical demonstratives, while in the reinforcer construction, the demonstrative and reinforcer are always morphologically distinct vocabulary items. Second, reinforcers are optional, while in Standard Malagasy doubling is typically required.⁹ Third, reinforcers are dependent upon the demonstrative and never appear without it, but we have seen that DEM2 can be licensed by the non-demonstrative determiner ilay. Fourth, demonstrative pronouns themselves can be reinforced, as illustrated in the French example in (20); however, we have seen that pronominal demonstrative doubling is impossible in Malagasy.

Fifth, reinforcers are typically locative elements (e.g. here, there) but DEM2 in the doubling construction must be a demonstrative, not a locative.¹⁰ Finally, the position of reinforcers with respect to the head noun varies across languages, as can be seen in the examples in (19), but even in languages where the reinforcer appears to be final (French and Spanish), it precedes possessors and PP modifiers, as in the French examples in (21). The equivalent of (21a) is ungrammatical in Malagasy, as DEM2 must be strictly final (see Section 2.2).

Thus, notwithstanding some points of resemblance, framing demonstratives and reinforcers are syntactically distinct and we do not unify the two.

2.5 Summary

In Section 4, we will be concerned with accounting for the empirical generalizations introduced above and summarized in (22), which we take to be the core facts regarding Malagasy framing demonstratives.

Before turning to possible analyses of these patterns, we first lay out a proposal for the structure of the Malagasy DP.

3.1 The nominal spine

We propose the basic structure for Malagasy nominals in (26).

The nominal is built on an NP, which will contain the lexical head. Above this NP is a series of functional projections which will introduce the various modifiers: relative clauses, quantifiers, numerals, a possessor, and adjectives. Because the focus of the paper is on the analysis of demonstratives, we will not develop a specific proposal for most of these dependents. We will take adjectives (AP) and relative clauses (RC) as representative in order to illustrate our proposal. We assume that they are introduced in specifiers of functional projections above NP (Cinque 2010; 2020), labeled XP, YP, and ZP, above. The topmost projection, XP, is roughly equivalent to PredP/vP in the clausal domain, containing the head noun, its arguments, if any, and diverse modifiers.

Above XP we posit a Dem(onstrative)P(hrase) where demonstratives will originate (Bernstein 1997; Giusti 1997; 2002; Pangiotidis 2000; Brugè 2002; Shlonsky 2004; Roehrs 2010; Cinque 2010; 2020; and others). We will have more to say about the syntax of demonstratives, DemP, and its position after spelling out the remainder of the nominal spine. DemP is dominated by a nominal inflectional projection, labeled NIP, whose more precise identity, if it has one, we will not attempt to determine. While Zribi-Hertz & Mbolatianavalona (1999) identify this projection as Num(ber)P, we will not give it that specific label as we believe it may contain more general inflectional information, beyond number.¹¹ We consider NIP to be equivalent to IP in the clausal domain. In some languages, for example those without tense morphology, IP is a more suitable label for the main inflectional head in a clause compared to T(ense)P that has become standard. Finally, NIP is dominated by D(eterminer)P, following much work within the DP Hypothesis (Abney 1987, inter alia), which we adopt. We view the projections above XP (DemP through DP) as an inflectional domain, roughly equivalent to the projections between vP and CP in the clausal domain.

Returning to the syntax of demonstratives, there are at least two dominant families of analysis in the literature. In the determiner analysis, demonstratives are a kind of determiner. Within the DP Hypothesis, they are base-generated in D, at the top of the nominal spine. The determiner analysis is driven by languages such as English in which determiners and demonstratives are in complementary distribution. Placing demonstratives in the same structural position as determiners straightforwardly accounts for this fact (Jackendoff 1977). On the other hand, the analysis faces several empirical and conceptual difficulties. First, there are languages, such as Greek, in which the two are not in complementary distribution, (27).

Second, the demonstrative is a universal category while determiner is not. There are many languages that lack determiners, most famously the Slavic languages, but none that lack demonstratives (Diessel 1999; Alexiadou et al. 2007: 95). Assigning them to the same position does not yield an immediate explanation for why such an asymmetry might exist. Third, demonstratives can be used intransitively (i.e., pronominally) while determiners generally cannot be (Alexiadou et al. 2007: 95). Finally, the interpretation of demonstratives is different from determiners (Diessel 1999: 35–55; Alexiadou et al. 2007: 98–104).

Responding to these differences, researchers have defended some version of the DemP analysis we adopted above. Demonstratives are base-generated in a lower position in the nominal spine. Cartographic principles or language-specific facts determine the location of DemP. Demonstratives may be phrasal and occupy spec,DemP and/or be heads in Dem. Both options have been proposed in the literature. Either of these elements may ultimately move to a higher position, such as spec,DP or D. Thus, there is a rather large family of DemP-based analyses, schematized in (28), depending upon the base and surface position(s) of the demonstrative and the movements posited within the nominal. We will develop two such analyses for Malagasy below. DP

With this much in place, we develop a derivation for Malagasy nominals.

3.2 Roll Up

We assume that the inverse ordering of adjectives and relative clauses is achieved through Roll Up in the XP domain in (26). Assuming a universal base-generated order of adjectives as in English (quality ≻ size ≻ shape ≻ color ≻ nationality), Roll Up will yield the inverse ordering of adjectives, (24). Following Cinque (2005; 2010), relative clauses may be generated in leftward specifiers above the adjectives. Continued Roll Up will place the relative clause on the right.

A Roll Up, or Snowball, derivation consists in moving a constituent C₁ to a position above a modifier M₁ and then moving a constituent C₂ that properly contains C₁ and M₁ to a position above another modifier M₂. The movement carries along C₁ and M₁ in an inverse order. As each subsequent, larger constituent moves, it picks up a modifier on its right. We illustrate for an example with two adjectives, AP₁ and AP₂, and a RC. The base-generated order seen in (26) is RC ≻ AP₁ ≻ AP₂ ≻ N. The word order resulting from Roll Up is the inverse N ≻ AP₂ ≻ AP₁ ≻ RC, (29).

The derivation has an underlying structure as in (26) above. The first movement, labeled ①, moves NP to the (outer) specifier of ZP, above AP₂. This achieves the ordering N ≻ AP₂. This constituent ZP then undergoes movement, labeled ②, to the (outer) specifier of YP, placing ZP to the left of AP₁ and deriving the order N ≻ AP₂ ≻ AP₁. Finally, YP moves to the (outer) specifier of XP, above RC. This movement is labeled ③. This yields the desired inverse order N ≻ AP₂ ≻ AP₁ ≻ RC, seen in the bold-faced elements.

3.3 Nominal Fronting

We now turn to the framing word order. We assume that the determiner-like elements ny and ilay reside in D and that demonstratives are in Dem. This requires that all other nominal material move to a position between D and Dem. Consequently, we posit Nominal Fronting, which moves the Rolled Up XP constituent to a position between D and DemP, which we identified above as spec,NIP (see Zribi-Hertz & Mbolatianavalona 1999; Ntelitheos 2012). Continuing the schematic derivation from (29), XP moves to spec,NIP, yielding the structure in (30).

The derivation above could correspond to the following nominal:

Moving forward, the details of the Roll Up structure will not be relevant to the analysis of framing demonstratives, and we will replace XP with NP, regardless of its structural make up, even when it contains modifiers. We therefore remain agnostic about the category label, but assume it is the highest projection in the nominal lexical domain. Thus, for the demonstrative-modified nominal in (32), we use the representation in (33).

We will return to the motivation for Nominal Fronting in Section 4, where we discuss the derivation of framing demonstrative examples. Before turning to such examples however, we situate our proposal in the larger context of Malagasy phrase structure.

3.4 Comparison to Malagasy clause structure

The derivation of Malagasy nominals involves the following pieces:

This section briefly shows that our derivation for Malagasy nominals has parallels to the structure proposed for clauses, which typically posit both a Roll Up process low in the structure and fronting of the Rolled Up constituent to a higher, left peripheral specifier.

A Roll Up derivation in the clausal domain has been proposed for two independent purposes in Malagasy. First, Pearson (2000) uses Roll Up to achieve an inverse ordering of double objects compared to English, (35).

Second, Rackowski (1998), Rackowski & Travis (2000), and Pearson (2000) show that the order of preverbal adverbial elements in Malagasy patterns with English and Cinque’s (1999) adverbial hierarchy; however, postverbal adverbials appear in the mirror order. The latter is again achieved via Roll Up. See those references for data. Thus, there is a parallel in the clausal domain for this movement. It takes place in the lower part of the clause, typically identified as vP/PredP. What we have called XP is the nominal equivalent of this phrase.

Nominal Fronting is roughly parallel to Predicate Fronting in the clausal domain. Predicate Fronting derives predicate-initial word order (VOS) from an underlying SVO ordering (see Massam & Smallwood 1997; Rackowski & Travis 2000; Pearson 2001; 2018; Aldridge 2004; Cole & Hermon 2008; Travis & Massam 2021; among others for discussion and motivation). There is a significant amount of literature that derives Malagasy’s predicate-initial word order from an operation that moves the predicate (the verb plus all complements) to a high position in the clause. In other words, a sentence such as (36), has the structure schematized in (37), although the precise details of the category of the moved element and the position it moves to vary among authors.

We believe that the parallels between our nominal structure and that typically posited for clauses provide indirect evidence for our proposal.

Having discussed Roll Up and Nominal Fronting within DP, we turn to the framing demonstratives. Section 4 presents two different approaches to syntactic doubling in the literature—single versus dual representation—and applies them to the Malagasy case. Section 5 then argues for the single representation analysis.

4.1 Single representation analysis

This section presents a single representation analysis of Malagasy framing demonstratives. We first lay out our proposed derivation and then present the machinery, from Saab (2017), that results in demonstrative doubling.

As a reminder, (44) contains our assumptions about Malagasy nominal structure.

To the framing demonstrative example in (45), we assign the structure in (46). In addition to Nominal Fronting of NP to spec,NIP, there is head movement of Dem to D.

We follow Harizanov & Gribanova (2019), which argues that there are two types of head movement: syntactic head movement and post-syntactic morphological amalgamation. The former is a syntactic operation with syntactic properties and subject to syntactic restrictions. Our Dem-to-D is clearly of this type and has a number of its characteristics (Harizanov & Gribanova 2019): i) it affects word order, ii) it is not morphologically driven, iii) it does not result in word formation, iv) it is not governed by the Head Movement Constraint (HMC, Travis 1984) and can skip other heads, and v) it is driven by whatever is assumed to drive syntactic movement (i.e. Agree, feature checking, or a generalized EPP). Of particular note is the claim in iv) that syntactic head movement does not obey the HMC, potentially resulting in what has been called long head movement. Our Dem-to-D is an instance of long head movement. Much recent work (Lema & Rivero 1990; Rivero 1994; Matushansky 2006; Roberts, 2010; Harizanov & Gribanova 2019; Arregi & Pietraszko 2021; others) has argued for the empirical necessity and theoretical viability of long head movement and we follow these works and others in assuming that it is allowed.¹³

We drive Dem-to-D movement with a strong Dem feature, [uDem*], on D. If D bears this feature, Dem will move to D. If there is no Dem in the derivation with this D head, it will crash. There are other D heads in Malagasy which do not bear this feature and result in D being spelled out as ilay or ny according to the vocabulary insertion rules in (47). This correctly derives the low position of the demonstrative and the lack of framing in (48), which has the structure we previously assigned in (49).

As previously discussed, Nominal Fronting is akin to Predicate Fronting in the clausal domain. Massam & Smallwood (1997), Pearson (2001), Massam & Travis (2021), and others invoke an EPP feature on a functional head (e.g. T) that drives Predicate Fronting, and we will simply stipulate that Nominal Fronting is driven by an EPP feature on NI, requiring that its specifier be filled.

We now turn to the theoretical machinery that results in double pronunciation of the demonstrative in (46). Saab’s (2017) I-assignment mechanism was developed to account for verb doubling in the Predicate Cleft (PC) construction. The PC is a widely discussed doubling construction in the clausal domain. PCs consist of a fronted verb or verb phrase resumed by a lower copy or morphologically similar copy of the verb. PCs have received a great deal of attention in the literature since Koopman’s (1984) work on Vata. Examples from typologically-diverse languages are given in (50).

Saab (2017), building on Saab (2008), analyzes the PC in Romance and argues that multiple copies of the verb arise due to the conditions that determine copy pronunciation. He proposes the notion of I-Assignment to achieve this. I-Assignment assigns a feature [I] to certain constituents that then blocks Vocabulary Insertion to that constituent at PF. There are two kinds of I-Assignment, phrasal and head. Phrasal I-Assignment occurs in the syntax under familiar conditions, namely, c-command. Head, or Morphological, I-Assignment occurs at PF.

Phrasal I-Assignment is defined as follows, where for Saab’s purposes, chain links count as identical.

I-Assignment to heads occurs at PF under morphological considerations:

Following Embick & Noyer (2001: 574), Y⁰ is a Morphological Word (MWd) if and only if Y⁰ is the highest segment of an X⁰ not contained in another X⁰. Morphological I-Assignment has the effect of marking for deletion any MWd that is either i) linearly adjacent to an identical head or ii) structurally immediately local to an identical head, where immediate locality is the relation between a head and the head of its complement.

Consider how I-Assignment, in particular Morphological I-Assignment, leads to demonstrative doubling in the structure repeated below. The relevant MWd Y⁰ which could be targeted for I-assignment is Dem in its base position. Given (52), [I] would be assigned to this lower Dem if and only if it is adjacent or immediately local to an MWd containing another instance of Dem, the X⁰. Neither of these conditions is met. The two Dems are not adjacent because the NP boky intervenes. The lower Dem is also not local to D because D does not select DemP as its complement: NIP intervenes. As a result, the base position of Dem is not assigned an [I] feature and both instances of Dem are pronounced at PF.

Constituent labels aside, this is the analysis that Saab (2017: (114)) assigns to the Vata PC in (50e). V-to-Foc is fully parallel to our D-to-Dem and results in verb doubling.

The theory makes the prediction that long head movement will normally lead to head doubling. Whether this is a correct prediction will need to be tested in other domains where long head movement is justified. Space prevents us from exploring it here.¹⁵¹⁶

4.2 Dual representation analysis

This section develops an alternative, dual representation analysis. Section 5 will show that it is not as successful as our single representation analysis in Section 4.1. A dual representation analysis of framing demonstratives necessarily base-generates two instances of the demonstrative. For the sake of parity, we will formulate a dual representation analysis that is similar to our single representation analysis above. In particular, i) we continue to adopt the DemP analysis of demonstratives, with demonstratives being base-generated within DemP, and ii) we posit movement of Dem into DP.

The dual representation structure for framing demonstratives is given in (55). DEM1 is phrasal and occupies spec,DemP, while DEM2 is the head of DemP. DEM1 moves to spec,DP to check the [uDem*] feature of D. DEM1 moves rather than the Dem head DEM2 because it is closer to D. The lower copy of DEM1 is I-assigned according to the Phrasal I-Assignment rule in (51) and is not pronounced. There is Roll Up within NP and Nominal Fronting as above.

This structure gives rise to the correct word order, where DEM1 and DEM2 are initial and final, respectively.¹⁷