3.1 N-bonding in the verbal domain

N-bonding occurs in both PV and CV sentences, but not in AV sentences, as shown in (11). (11b) and (11c) show PV sentences and (11d) shows a CV sentence. In both cases, N-bonding is observed between the non-Agent Voice verb and the following agent. In the literature, this post-verbal agent is described as being genitive (see e.g. Paul 1996; Keenan & Manorohanta 2001). In contrast, (11a) illustrates an AV sentence for which N-bonding is not observed.

In PV and CV sentences, the verb often takes a suffix voice morpheme. For example, the verb didy ‘to cut’ takes the PV suffix -an in (11b) and the CV suffix -an in (11d). (Recall that the PV suffix is -Vn, where the vowel is lexically determined. For the verb didy, the PV suffix is realized as -an.) When the N-bonding element comes between the agent and preceding verb, where the verb takes a suffix that ends in n, it is not always clear that the N-bonding element is independent from the voice morpheme.⁹ I include (11c) to show the presence of the N-bonding element in the absence of a voice suffix. Rather than taking the PV suffix -Vn, the verb in (11c) takes the prefix voa-.¹⁰ This example shows that the N-bonding element is distinct from the preceding verb and the following demonstrative.

Note also that the presence or absence of N-bonding in these environments is not optional. That is, the presence of N-bonding in an AV sentence leads to ungrammaticality, as shown in (12a). On the other hand, the absence of N-bonding in PV and CV sentences leads to ungrammaticality, as shown in (12b) and (12c).

In the verbal domain, N-bonding also does not occur between a verb and an adjacent argument if that argument is the trigger. This is shown in (13) where the theme is the sole argument of the clause and is thus the trigger. Similarly, N-bonding does not occur between a verb and an adjacent prepositional phrase, as shown in (14). Therefore, within the verbal domain it is not the case that N-bonding occurs between non-Agent Voice verbs and any adjacent constituent. Rather, N-bonding occurs only between a non-Agent Voice verb and a non-trigger (genitive) agent.

3.2 N-bonding in the nominal domain

In the nominal domain, N-bonding is observed in possessive constructions, as shown in (15a), and between certain prepositions and adjectives and their complements, as shown in (15b) and (15c). These examples additionally show that the shape of the D-material (i.e. definite determiner ny, demonstrative ilay, etc.) does not affect the form of the N-bonding element, which consistently appears as n.¹¹

The observation that N-bonding occurs with certain prepositions and adjectives requires further explanation. In Malagasy, complements of prepositions and adjectives are marked for one of three distinct cases. These are traditionally labeled as ‘nominative’, ‘accusative’, and ‘genitive’. I adopt these labels for descriptive purposes, but will return to the distribution of case in Malagasy in Section 4. Distinct case morphology occurs most consistently in the pronominal system, which is presented in Table 2.

I start with prepositions. A given preposition can assign either nominative, accusative, or genitive case to its complement. The sentences in (16)–(18) show that patterns of N-bonding emerge only when the complement is marked for genitive case. In (16), the complement of the preposition noho ‘because of’ is marked for nominative case and as a result no N-bonding occurs; no additional N-bonding element is found in the construction. The same pattern is found when the complement of a preposition is marked for accusative case. This is seen in (17) for the preposition lavitra ‘far from’. Lastly, the examples in (18) show the presence of N-bonding when the complement of a preposition is marked for genitive case. For example, (18a) shows that the preposition aloha ‘in front of’ takes the genitive pronoun -ny and (18b) shows that when the complement of the preposition is a nominal DP, the N-bonding element n appears between the two. We will return to the connection between N-bonding and the pronominal paradigm in Section 4.3.3.

There is a similar division for adjectival predicates in Malagasy, which occur with a following nominal in one of three forms (see Ralalaoherivony 1995) listed in (19).

Examples of the three different forms are illustrated in (20). Crucially, N-bonding occurs only when the nominal appears without accusative case (20c). This provides preliminary support for a licensing account for N-bonding, which I develop in more detail in Section 4.3.

Taken together, the data provided in this section show that the patterns of N-bonding in the nominal domain mirror those described in the verbal domain from Section 3.1 above. To summarize the discussion so far, N-bonding occurs across the verbal and nominal domains in Malagasy and is found between a non-Agent Voice verb and following non-trigger agent, a possessee and possessor, and between certain prepositions and adjectives and their complements. Importantly, these are all environments that are descriptively genitive. All of these facts, I argue, can be explained under a licensing approach.

4.1 Malagasy clause structure

I start by outlining the basic clausal structure of Malagasy, with the assumption that there are only two structural licensors in the language: CT⁰ and v/Voice⁰. In Section 4.1.1, I propose a joint CT⁰ in Malagasy, a high functional head responsible for hosting an Ā-featured DP in its specifier and assigning it nominative case. This follows from previous proposals such as those by Aldridge (2004; 2021), Legate (2014), and Erlewine (2018). In Section 4.1.2, I show that deriving the right outcomes of movement to trigger position relies on a lower functional head, namely v/Voice⁰, the head responsible for merging the external argument and assigning accusative case. I return to the implications of such licensing patterns in Section 4.3.

4.2 Spelling out voice morphology

With the clausal structure outlined above, we can connect the properties of v/Voice⁰ with the different voice morphemes in Malagasy. Recall the voice morpheme distribution repeated in Table 3. In AV the verb takes the AV prefix m- and a verbal prefix (e.g. aN-). I propose that the AV prefix m- is realized on v/Voice_AV, reflecting the absence of the movement-triggering [•D•] feature. In contrast, when there is such a feature on v/Voice⁰ (i.e. in PV and CV), v/Voice⁰ is null.

Following Pearson (2005), I assume that the verbal prefixes are realized on the same head as the PV affixes: Asp.¹⁷ When the theme moves out of the complement of Asp⁰, Asp⁰ is spelled out as a PV affix. Otherwise, Asp⁰ is spelled out as a verbal prefix.¹⁸ Following Pearson (2005), I assume the CV suffix -an is the spell-out of the Appl⁰. As traditionally assumed, all affixes combine with the verb root as it raises via successive head movement. Following Travis (1994; 2006) I assume that verb movement in Malagasy is contained within the heads that are event related and therefore the verb moves only as far as E⁰, an event related functional head situated between CTP and vP which defines the edge of an event (see Travis 1994 for more discussion on E⁰).¹⁹ AV and PV structures with the corresponding voice morphemes are shown in (31).

The observation that Appl⁰ is spelled out as -an only when the applied argument raises out of the specifier position of ApplP, can be explained under a generalized version of the Doubly-Filled Comp Filter, as in (32) (following Sportiche 1992; Koopman & Szabolcsi 2000; Pearson 2005).

The spell-out of v/Voice⁰, Asp⁰, and Appl⁰ for each of the three voices is summarized in Table 4. In AV, there is no movement-triggering feature on v/Voice⁰, resulting in the spell out of v/Voice⁰ as the AV prefix m-. In contrast, the movement-triggering feature is present on v/Voice⁰ in both PV and CV, resulting in the spell-out of a null v/Voice⁰. Verbal prefixes are spelled out on Asp⁰ in AV and CV when the theme remains within the complement of Asp⁰. In PV, on the other hand, when the theme is raised due to the [•D•] feature on v/Voice_PV, Asp⁰ is spelled out as a PV affix. Lastly, the CV suffix -an is the spell out of Appl⁰ (projected only in CV), when the applied argument raises.

In the next section I explain how these assumptions regarding the structural organization and licensing constraints in the language lead to N-bonding.

4.3 Case and licensing in Malagasy

4.4 Interim summary

In Section 3, I presented the distribution of N-bonding across the verbal and nominal domains in Malagasy. I then outlined the proposed clausal structure for the language in Section 4, tying the N-bonding observations to licensing constraints. Specifically, I proposed that Malagasy has only two structural licensors: CT⁰ and v/Voice⁰, which are responsible for assigning nominative and accusative case, respectively. The constructions in which we find N-bonding all contain a nominal that cannot be structurally licensed due to the absence of abstract genitive case in Malagasy. This analysis was further supported by the distribution of the pronominal forms in the language (Pearson 2005). Having established the syntactic conditions that underlie N-bonding, I turn next to an account of the post-syntactic operations that are involved in its derivation.

5.1 N-bonding as a result of Local Dislocation

Recall from Section 4 that the deficiency of structural licensing in Malagasy leaves nominals in certain constructions unlicensed. If Malagasy complies with the Case Filter, defined as requiring all nominals to receive Case within the syntactic derivation, then the analysis thus far suggests that the Case Filter is violated in Malagasy. Following Levin (2015), I assume that languages can make use of post-syntactic licensing strategies and that we can maintain a version of the Case Filter by appealing to the proposal that the Case Filter is satisfied not at the end of the syntactic derivation, but later along the PF branch.

Before turning to the operations that are required to derive the account, I first spell out my assumptions about the overall structure of the grammar. I assume a conventional Y-model as illustrated in (48):

Once completed, the syntactic derivation is sent to Phonological Form and Logical Form. Following the Distributed Morphology framework (Halle 1990; Halle & Marantz 1993; and much subsequent work) I assume that Morphology represents a set of processes along the PF branch which interpret the output of the syntactic derivation (Embick & Noyer 2001). See Embick and Noyer (2001) for a detailed breakdown of proposed operations that take place along the PF branch. The operation relevant for the discussion of N-bonding in Malagasy is Local Dislocation, which I turn to next.

Embick & Noyer (2001) developed Local Dislocation as a variety of Merger, or Morphological Merger, as first proposed by Marantz (1988) (see Embick & Noyer 2001 for detailed discussion on the development of this operation).²³ Local Dislocation is an operation that combines two linearly adjacent terminal nodes to create a complex atomic head (Embick & Noyer 2001; 2007).²⁴ This adjunction operation is schematized in (49), where X•Y denotes a requirement that X must linearly precede and be adjacent to Y.²⁵

I assume following Levin (2015) that nominal licensing can be achieved via linear (i.e. post-syntactic) adjacency. As such, the licensing needs of Malagasy can motivate the implementation of Local Dislocation in the language. More specifically, Levin proposes that adjunction by Local Dislocation allows the nominal to count as part of the verbal extended projection, which obviates the need for the nominal to be Case-licensed. See Levin (2015) for further discussion on licensing by adjacency and proposals on modifying the Case Filter. For the purposes of this analysis, we can think of Local Dislocation as a last-resort licensing mechanism along the PF branch (see Erlewine 2018 for a similar proposal for Toba Batak). This post-linearization operation is also consistent with the empirical data from Malagasy. In a non-AV sentence for example, Local Dislocation can apply to a non-Agent Voice verb and in-situ agent, which will always be linearly adjacent. An example is schematized in (50), for a non-Agent Voice verb voavoha ‘opened’ followed by a non-trigger agent ny vavy ‘the girl’.

We can apply the same machinery to the nominal domain, as schematized in (51) for the preposition aloha ‘in front of’ and its complement ny fiara ‘the car’. Since the preposition and following nominal are linearly adjacent after spell-out, Local Dislocation can apply, rendering the nominal part of the extended projection of the preposition and thus licensed via adjacency.

According to Levin (2015), Local Dislocation will not only result in the observed adjacency requirements (e.g. between non-Agent Voice verbs and following agents), but should also ensure that the adjacent components form a tight phonological unit at PF. Evidence from the presence of domain-internal phonological processes are compatible with this approach. For example, evidence of adjunction is found when the in-situ agent is a proper name, as in (52). In (52a), the non-Agent Voice verb novidin ‘bought’ and proper name Rabe are written as a single word, where the N-bonding element ‘n’ and following ‘r’ fuse to form the prenasalized affricate [ndr], a phonological process that otherwise only applies word-internally in Malagasy (Pearson 2005).²⁶ Assuming that local dislocation is also at play in the nominal domain (52b), the parallel outcomes and phonological effects of ‘bonding’ (i.e. fusion) are simply a reflection of adjunction and are thus unsurprising.

Together, the empirical data is consistent with the proposal that N-bonding in Malagasy is linked to the licensing constraints of the language; N-bonding is observed in constructions where a nominal, as a result of not being structurally licensed, is post-syntactically licensed via adjunction. The licensing analysis described up to this point effectively accounts for the distribution of N-bonding and the phonological consequences are in line with the syntactic account.

5.2 Evidence from Malagasy compounds

So far, we have seen that the N-bonding element reflects the adjunction configuration created by Local Dislocation and carries no specific meaning or function other than to reflect that the adjunction operation has taken place. This provides a similar effect to that of linking elements which occur between two constituents of a compound. The -s- in Mainland Scandinavian and the -o- in Modern Greek are two examples of morphemes that are used in compounds as linking elements but are semantically empty (Josefsson 1997; Ralli 2013; among others). If the N-bonding element surfaces as a reflection of adjunction configurations more generally, then we predict the element to also occur in other head-head constructions in Malagasy, such as compounds. This prediction is borne out.

In Malagasy, a class of compounds called “linking compounds” is formed with a “linking morpheme” (Ntelitheos 2012). The structure of these compounds is also described as being parallel to the structure of possessive constructions in Malagasy (Ntelitheos 2012). The examples in (53) show that in these compound forms, the N-bonding element appears between the two nouns which are concatenated and is subject to word-level morphophonological processes of the language such as consonant mutation and prenasalization (Keenan & Razafimamonjy 1996; Paul 1996; Keenan & Polinsky 1998). For example, in (53a) the initial consonant of vava ‘mouth’ undergoes mutation from ‘v’ to ‘b’ which forms a prenasalized stop with the preceding N-bonding nasal.

This data from compounding provides further support for the use of Local Dislocation within the language and extends the distribution of N-bonding to head-head configurations more generally.²⁷ As a result, we can reformulate our generalization of N-bonding to a language-wide condition as stated in (54).

5.3 The N-bonding element is a morpho-phonological ornament

Since Local Dislocation is a post-syntactic operation and is sensitive only to linear order rather than hierarchical structure, I assume that the insertion of the N-bonding element must also be post-syntactic (i.e. inserted at PF). Elements of this type are described as ornamental (Embick & Noyer 2007), such that they introduce syntactico-semantically unmotivated structure (i.e. nodes) and features which ornament the syntactic representation. In other words, any such ornamentations introduce material into the PF expression but crucially do not add or eliminate information necessary for semantic interpretation. There are two types of elements that can be added in the PF component: features and terminal nodes, as defined in (55) and (56) (Embick & Noyer 2007). These definitions do not explicitly state whether or not the dissociated feature or node is strictly syntactic. I return to this point below.

These inserted elements are deemed dissociated, emphasizing that this material signals the presence of certain syntactic morphemes, features, or configurations, but does not itself represent the actual spell-out of these. I assume, following Embick and Noyer (2007), that the implementation of dissociated features/nodes is governed by language-specific rules. Post-syntactic insertion operations involving dissociated elements have been discussed in recent literature for compounds (see Tat 2013 on Turkish and Dolatian 2021 on Armenian), in which linking vowels/compound markers are semantically empty and are proposed to be added during phonological spell-out in PF (see also Aronoff 1994; Oltra-Massuet 1999; Ralli 2008), as discussed above for Mainland Scandinavian and Modern Greek. Post-syntactic insertion has also been discussed for Korean subject honorification, for which honorific agreement suffixes are implemented by so-called node-insertion (Choi & Harley 2019). More recently, this form of introducing a dissociated or ornamental element has been re-dubbed node-sprouting by Choi and Harley (2019), a term that more explicitly conveys that the insertion (or ‘sprouting’) of such post-syntactic material can only occur when certain conditions are met.

I propose that a similar approach can be adopted for N-bonding in Malagasy. The N-bonding element is a dissociated element and the condition necessary for its insertion is head-head adjunction, which we’ve seen can come about as a result of Local Dislocation or compounding. Extending previous interpretations of dissociated elements, I further assume that the types of dissociated elements that can be inserted, as well as the conditions in which they can sprout, are not restricted to syntactic features/conditions, but can also include phonological features/conditions. More specifically, I propose that the insertion of the N-bonding element involves the insertion of a bundle of phonological features which can sprout as different surface realizations depending on surrounding phonological context. This will correctly derive the surface representation of N-bonding and its phonological variation, which I turn to next.

5.4 Accounting for phonological variation

In order to provide a complete representation of the N-bonding element, we must first examine the variation that exists in its surface realization. This section aims to provide a brief sketch of the variation of the N-bonding element. Specifically, I will show that the N-bonding element has two possible surface representations: /n/ and /i/, and that the preceding phonological context is what determines which representation surfaces. When the N-bonding element is preceded by a non-nasal consonant, it surfaces as /i/. In all other cases, the N-bonding element surfaces as /n/.

Recall the data from (1) and (2), repeated in (57) and (58) below. The examples are similar in that they include (a) a non-Agent Voice verb and an adjacent external argument or (b) a possessee followed by a possessor—both constructions in which we expect to find N-bonding. However, the N-bonding element n is seemingly not present in (58). This is puzzling given the described distribution of N-bonding in the language. If N-bonding is indeed present in (58), then at least two questions arise: (i) is the N-bonding element realized in the surface representation of constructions like those in (58)? and (ii) how do we account for the variation in the form of the N-bonding element? In this section I will argue that the N-bonding element is indeed realized in the surface representation of constructions like those in (58) and that the N-bonding element has an underlying representation /n/ which surfaces as /i/ when it is preceded by a non-nasal consonant and as /n/ elsewhere.

As a starting point to answering these questions, we can examine the phonological patterns that are involved in these constructions. Stress patterns and the presence of epenthetic vowels will be used as the primary cues to signal the presence or absence of the N-bonding element. The predominant stress pattern in Malagasy is to stress the penult, as seen on the verb dídy ‘to cut’ in (59). In the following examples, primary stress is marked with an acute accent ⟨á⟩ on the words under discussion.

Regarding epenthetic vowels, according to Erwin (1996), when a morpheme is consonant final, one resolution is to insert an epenthetic vowel a in order to conform to Malagasy CV syllable structure. This occurs, for example, with the voice morphemes introduced above in Section 2. When a voice morpheme occurs word-finally, a final vowel a is realized on the verb, as shown in (60) where the verb dídy ‘to cut’ takes the PV suffix -an.

If we compare the PV form in (60) to its AV counterpart in (59), we see that the PV suffix -an causes stress to shift rightward by one syllable. In (59), the stress falls on the first syllable of the root dídy, while in (60) the stress falls on the second syllable of the root (didí). However, the final vowel a does not impact stress location. As a result, primary stress falls on the antepenultimate syllable of didíana. These stress patterns are consistent with the analysis that the final vowel a is epenthetic (Erwin 1996).²⁸

We can apply the same diagnostics to the forms in (57) and (58), repeated in (61) and (62) below. Starting with the examples in (61), the root of the verb ‘to open’ is voha and the root of the noun ‘house’ is trano. Both are vowel-final and we do not find an epenthetic vowel, as expected.

Turning to the examples in (62), the root of the verb ‘to cut’ is tapak and the root of the noun ‘foot’ is tongotr. Both are consonant-final thus we expect to find an epenthetic vowel in word-final position. However, the final vowel that appears is y and not the traditionally described epenthetic vowel a.

This could suggest that Malagasy has more than one option for word-final epenthetic vowels and that the N-bonding element is simply absent in these examples. However, if we compare the PV form of tapak ‘to cut’ when the following nominal is a trigger versus an agent, an interesting pattern emerges. When the non-Agent Voice verb is followed by the trigger, the final vowel on the verb is the standard epenthetic vowel a, as shown in (63a). However, when the non-Agent Voice verb is followed by the agent, as in (63b), the final vowel on the verb is y. Crucially, the final vowel y occurs in a construction where N-bonding is predicted, namely between a non-Agent Voice verb and a following agent. I assume following Paul (1996) and Pearson (2005) that the final y is an alternate form of the N-bonding element.

Following Paul (2000) and Pearson (2005), I assume that the inserted N-bonding element is always present in the expected N-bonding constructions and varies in surface form depending on surrounding phonological context. However, while Pearson (2005) analyzes the N-bonding element as a morpheme -ny, I propose that the underlying representation of the N-bonding element is the nasal /n/, which consists of the phonological features [nas] and [cor] such that the feature(s) that can be realized is restricted by surrounding phonological context. When the preceding segment is a vowel, the N-bonding element can be fully realized with both features [nas] and [cor], as illustrated in (64)–(65).

On the other hand, when the preceding segment is a consonant, the N-bonding element must surface as a vowel. Assuming that front vowels are coronal (i.e. have a [cor] feature; see e.g. Clements & Hume 1995), I propose that the [cor] feature of the N-bonding element is preserved and the N-bonding element is thus realized as the coronal vowel /i/ (corresponding to orthographic y). This is schematized in (66), with an example illustrated in (67) for the verb root tapak ‘to cut’.

The phonological analysis proposed here offers a preliminary account of the general properties of N-bonding. However, deeper investigation into the phonological organization of the language is needed. For example, I remain agnostic regarding the framework governing features within segments, but this will need to be examined more carefully in order to fully understand how the variation observed for N-bonding is derived.²⁹

In summary, N-bonding can be analyzed as a reflection of a particular configuration: head-head adjunction, which, in Malagasy, can be derived through Local Dislocation, an operation that is implemented in the language due to structural licensing constraints. Non-trigger agents, for example, are unable to be licensed in their merged position (Spec,v/VoiceP) and therefore must undergo Local Dislocation resulting in a head-head adjunction configuration. This creates the right conditions for [n]-feature sprouting which results in the insertion of the N-bonding element between the verb and non-trigger agent. The same process occurs between possessees and possessors, and between certain prepositions and adjectives and their complements. N-bonding is also found in other head-head constructions such as compounds, lending further support to an adjunction approach. Finally, the variation that is observed in the N-bonding element’s surface forms can be accounted for under the proposal that the underlying form is the nasal /n/, consisting of the features [cor] and [nas] which sprout as either the nasal /n/ or vowel /i/ depending on the surrounding phonological context.