As was shown, intransitive subjects (5b) and objects (6a) are consistently marked to the exclusion of transitive subjects (6a), which surface unmarked.

Consider first transitive constructions, and how the structure of such clauses bears on the generalization developed in (26). As argued for in §1.3, the Nias transitive clause involves the subject DP raising to a clause-medial position, i.e. Spec,FP, out of the vP and the vP-remnant fronting with the object remaining vP-internal, as in (33). Importantly, this means that while the object remains within the vP phase, the subject is located outside of the vP phase in Spec,FP. Consequently, the subject is not located within the same phase as the lexical V. Thus, given the structure of transitive clauses the arguments adhere to the generalization in (26) where the object surfaces marked but the subject is unmarked.

Consider now intransitive constructions, as in (5b), where the subject always surfaces marked. I argue that intransitive subjects uniformly remain within the vP phase, which I provide independent evidence for below. Because of this, they are always marked, as per the generalization in (26). Thus, I propose that the intransitive clause has the structure in (34):

Evidence that intransitive subjects are uniformly structurally lower than transitive subjects comes from causativization patterns. As shown, when an unaccusative predicate, e.g. a stative as in (35b), is marked with the morphological causative f(a/e)-, the theme argument is retained and an agent/causer argument is introduced.

In contrast, when a transitive predicate is causativized, as in (36b), a causer subject is introduced and the causee (i.e. demoted subject) must surface as an oblique argument.

In the case of causatives of unergatives, the realization of the intransitive subject patterns with the unaccusative in (35b). As shown in (37b), the causee (i.e. the unergative subject) is not realized as an oblique argument (as with transitive subjects) and no argument demotion is visible.

I interpret the above pattern as indicating that subjects of unergatives are base-generated lower than transitive subjects, which are base-generated in Spec,vP (Massam 2009; Tollan & Oxford 2018; Kouneli 2021). Thus, since vP uniformly lacks an argument merged into its Spec in intransitives, when such predicates are causativized the causer can merge to Spec,vP without requiring subject demotion— i.e. the causee being realized as an oblique.

Further data that indicates this structural difference between intransitive and transitive subjects concerns differences regarding these arguments associating with quantificational elements. In Nias, the quantificational adverb ʔoi can induce a quantificational interpretation of an argument that is within its scope, as in (38) (Brown 2001). As (38a,38b) show, ʔoi can associate with both unergative and unaccusative subjects, inducing a quantificational interpretation.11 But in transitive constructions, as in (38c), while ʔoi can associate with an object, it cannot with the subject (Brown 2005).

I interpret this split between transitive and intransitive subjects as indicating that transitive subjects are located in a position that is outside ʔoi’s domain, and, thus, cannot associate. Conversely, given that intransitive subjects can associate with ʔoi, this indicates that intransitive subjects are in a position below ʔoi. Concretely, following the analysis of quantificational particles in Clemens 2019 and Tollan & Massam 2022, I propose that ʔoi heads a phrase, which can associate with an argument in its c-command domain. In Nias, I suggest that this phrase merges low to the vP. Evidence that ʔoi merges low comes from scope data. As (39) shows, when negation and ʔoi co-occur, ʔoi scopes below negation. Assuming that NegP projects low in the structure, within the extended verbal projection, this indicates that ʔoi must merge to a phrase below NegP—i.e. vP.

Thus, given the causativization and interpretative facts regarding ʔoi, I conclude that while transitive subjects are vP-external, intransitive subjects are uniformly vP-internal in Nias.12

As was shown in §1.2, there is an exceptional case regarding the absence of marking in transitive subjects, which concerned experiencers. Unlike agentive transitive subjects, experiencer subjects were shown to consistently surface marked, as in (21) (repeated below). Interestingly, unlike agentive transitive subjects, experiencers can also associate with oi, as shown in (40).

Given the above facts in (21) and (40), following Belletti & Rizzi (1988) among others, I argue that experiencer subjects are base-generated lower in the structure than agentive transitive subjects. Moreover, given that experiencer subjects consistently surface marked and can associate with ʔoi, these subjects remain vP-internal and are spelled out in the same phase domain as V (as with intransitive subjects), as sketched in (41). Thus, given the structure in (41), experiencer constructions adhere to the generalization in (26).13

Zooming out, this section demonstrated that the argument marking pattern in Nias is captured by the generalization in (26). Crucially, this generalization does not make reference to the grammatical role of the argument. Rather, I argued that the relevant factor regulating argument marking is the structural position of the argument at PF. When the DP is spelled-out in the same phase domain as a lexical head, it surfaces marked; when it is spelled-out in phase distinct from a lexical head it is unmarked. Moreover, I showed that this generalization can capture the argument marking pattern observed with core arguments. Specifically, I showed that the apparent ‘marked absolutive’ surface pattern is due to a systematic difference in structural position between intransitive subjects and objects, and agentive transitive subjects.

As shown are Nias’ vowel (42) and consonant (43) inventories. As discussed in Brown (2001; 2005) the Nias syllable adheres to a (C)V template (where V may be bimoraic).

While medial syllables may lack an onset, Nias has a strict ban on initial syllables being onsetless, i.e. no V-initial words (*[_ωV). In such cases, where the underlying initial syllable lacks an onset, a glottal stop [ʔ] is inserted as a default onset (44) (Brown 2001). This process is illustrated in Table 3. As shown in Table 3, when the V-initial stem is prefixed with the intransitive m-, no glottal stop appears on the initial segment of the stem, e.g. [m-uta] (‘to vomit’), cf. *[m-ʔuta]. But in contexts where the stem is the initial element, i.e. with the transitivizing suffix -iö, a glottal stop surfaces as the initial element [ʔuta-iö] (‘to throw up x’) (cf. *[uta-iö]).14

As shown in Table 4, in the case of consonant-initial nominals, argument marking is realized in two grades. In cases where the initial consonant is voiceless, e.g. faxe, argument marking is realized as voicing of the initial segment, e.g. vaxe. In cases where the initial consonant is a voiced obstruent, e.g. doi, argument marking is realized as trilling d^roi. In cases where the initial segment is a sonorant, e.g. the nasal initial nagole, argument marking does not occur even in contexts where arguments typically surface marked. As (45) illustrates, the object is nasal-initial and surfaces unmarked.

In the case of vowel-initial nominals, the pattern is somewhat different. As shown in Table 5, the surface form of marked vowel-initial nominals appears to be divided into two classes. Nominals where argument marking is realized as a prothetic [n] and nominals where it is realized as a prothetic [g]. In cases where the nominal is unmarked, it surfaces with a prothetic [ʔ], per the glottal-insertion rule in (44).

Further, as observed in Brown 2001, there are several cases where the marked forms of vowel-initial nominals are seemingly not predictable on the basis of their unmarked surface forms. For example, as Table 5 shows, [ʔöri] ‘federation’ and [ʔöri] ‘amulet’ are only phonologically distinguished in their marked forms: [nöri] and [göri], respectively. Subsequently, Brown (2001) interprets this observation as indicating that argument marking for vowel-initial nominals is not predictable and that the marked form of such nominals are lexically specified. Although Brown’s (2001) observation on this point may suggest that the marked forms of vowel-initial nominals are not phonologically predictable, I argue that this division regarding the surface form of marked vowel-initial nominals is, in fact, predictable on the basis of the underlying form of the nominal stem, which is obscured by the surface forms due to glottal-insertion. I propose that for vowel-initial stems, the surface form of marked nominals uniformly involves a prothetic [n]. In cases, where the vowel-initial stem is unmarked, glottal-insertion (i.e. (44)) applies and the nominal surfaces with a prothetic [ʔ]. In cases where the surface form of the marked nominal involves an initial [g], the underlying form of the nominal stem involves an initial [ʔ], which is visible in contexts where the nominal surfaces unmarked. In contexts where [ʔ]-initial nominals are marked, the general pattern for voiceless consonant-initial nominals applies and the initial [ʔ] undergoes voicing surfacing as [g]. Thus, Nias’ seeming ‘unpredictable’ vowel-initial pattern is an artifact of the glottal-insertion rule (with applies to vowel-initial forms) obscuring the difference between the underlying forms between the two classes of nominals.15

To summarize, I conclude that the surface forms of marked arguments are phonologically predictable on the basis of the underlying form of the nominal stem. In cases where the nominal is consonant-initial the nominal undergoes one of two grades of lenition and is either voiced or trilled. In the case of vowel-initial nominals, a prothetic [n] surfaces on the stem. Given this empirical situation, I propose that Nias’ argument marking pattern is result of a phonological operation, which occurs in the argument marking contexts previously discussed. Specifically, that in such contexts, a phonological operation is triggered where a prothetic nasal is inserted to the nominal, as in (46) (where M stands in for the marking context).

Given (46), the vowel-initial argument marking pattern (as in Table 5) follows straightforwardly from the application of (46), as shown in (47).

In the case of the argument marking pattern observed with consonant-initial segments, I argue that (46) applies in these contexts as well and a prothetic [n] is inserted to the stem. However, in such contexts, I argue that the insertion of [n] triggers lenition on the stem. Specifically, when the initial segment is a voiced obstruent, the segment becomes sonorant, which is realized as trilling (48a) and in contexts where the initial segment is voiceless, i.e. lacks a voice feature, the segment undergoes voicing (48b). For stems where the initial segment is a sonorant, neither (48a) or (48b) occurs.16

Finally, given that Nias quite generally disallows complex onsets, i.e. CC clusters, I argue that the repair rule in (49) is operative and the nasal deletes.

To illustrate how the above processes conspire to output Nias’ argument marking pattern, consider two stems faxe and bavi. In the first case, as sketched in (50), (46) occurs and creates a context for the voicing process in (48b) to occur. Finally, given the prohibition on CC sequences, the nasal deletes, i.e. (49), generating the surface form vaxe. In the second case, as in (51), (46) occurs and creates a context for trilling (48a). Here as well, (49) then occurs and [n] deletes, thus, generating the surface form bavi.

Note as well, that that ‘g-mutation’ pattern in Table 5 is straightforwardly derived as an instance of voicing mutation. As illustrated in (52), with the stem ʔöri, after nasal insertion, the initial [ʔ] undergoes voicing to [g]. Then the [n] deletes, outputting the surface form: [göri].

To summarize, I showed that Nias’ mutation pattern is a phonologically predictable process. Specifically, in mutation contexts a prothetic nasal is inserted. In cases of a vowel-initial stem, the nasal surfaces. But in contexts where the stem is consonant-initial, the insertion of the nasal triggers lenition of the adjacent segment plus the deletion of the nasal.

Given the present analysis, Nias’ mutation system displays a striking parallel to mutation systems observed cross-linguistically. For example, mutation in Welsh. Welsh exhibits a phonologically predictable mutation pattern, where voiceless segments undergo voicing and voiced stops undergo spirantization (Pyatt 1997; Roberts 1997; Tallerman 2006; Borsley et al. 2007, i.a). This phenomenon is illustrated in Table 6, where the 3sg.m possessive pronoun ei triggers mutation on the nominal complement (examples from Pyatt 1997).

Importantly, as with Nias’ mutation system, Welsh mutation contexts do not strictly correlate with grammatical role(s). Rather, mutation occurs when the target is realized in a particular syntactic configuration with the mutation trigger. As illustrated in (53), in transitive VSO clauses, the direct object is c-commanded by the triggering element (i.e. the subject here) and the object undergoes soft mutation (i.e. beic ≈ feic). However, when the direct object is fronted and is not dominated by a trigger, as in (54), it surfaces unmutated.

Additionally, elements that are non-arguments also may undergo mutation. As shown below, predicates of copular clauses (55) and non-finite verbs (56) may also be mutation targets when they are realized in mutation triggering syntactic contexts.

Thus, the key parallel between the distribution of mutation in Welsh and Nias, is that the relevant factor regulating mutation is not grammatical role, but whether or not a potential mutation target is realized in a specific triggering syntactic configuration. As was shown in §1, the environments where an argument mutates in Nias does not correlate with the grammatical role of the argument but rather the position that the nominal surfaces in. As with Nias, in Welsh when an element is realized in a triggering syntactic configuration it undergoes mutation; when it is realized in a non-triggering configuration it does not.

While argument marking on lexical nouns was shown to be predictable, Brown (2001; 2005) observes that pronouns participate in a distinct, non-phonologically predictable alternation— termed ‘pronoun mutation’. While Brown (2001) identifies this alternation as tracking grammatical role—i.e. case, I argue that this alternation is best explained without reference to the grammatical role of the pronoun.

As shown in Table 7, Nias has two pronouns series which are described in Brown 2001; 2005 as ‘unmutated’ and ‘mutated’.

The motivation for assimilating the two pronoun patterns with the mutation pattern observed with lexical nouns is due to the distribution of pronouns in the two series. Specifically, ‘mutated’ pronouns occur in the same contexts as marked lexical nouns, i.e. intransitive subjects (57a) and objects (57b) but do not surface as transitive subjects. Conversely, ‘unmutated pronouns’ consistently appear as transitive subjects (57b), as with unmarked lexical nouns.

In this respect, the distribution of the two pronoun series appears at first glance to track with the distribution of marked and unmarked lexical nouns that are core arguments. Namely, that mutated pronouns pattern with marked core arguments in being absolutive arguments. While unmutated pronouns pattern with unmarked core arguments in being ergative arguments.

Despite the distributional similarity between (un)mutated pronouns and (un)marked core arguments, there are several important differences that indicate that the alternation observed with pronouns is not due to an underlying phonological process of the type observed with lexical nouns. First, as shown above in Table 7, while the initial segment of some pronouns appear to undergo the same lenition process characteristic of marking on lexical nouns, e.g. the approximate [j] undergoing trilling [d^r] with the 1sg, in other contexts no lenition process occurs. For example, in the 3sg the initial [j] does not undergo lenition in the mutated case. Second, the surface forms of mutated pronouns suggest that pronoun mutation is a subtractive process. While the surface form of marked lexical nouns never involve phonological reduction of the stem, e.g. syllable deletion, pronoun mutation consistently involves this type of reduction. Throughout the entire mutated pronoun series the stems are consistently phonologically reduced compared to their unmutated counterparts, as in Table 7. Given the above differences between ‘pronoun mutation’ and argument marking on lexical nouns, I conclude that the alternation observed with pronouns in Nias cannot be the result of the same phonological process that is operative with argument marking on lexical nouns.

Note as well, that explaining the observation that ‘mutation’ on pronouns involves (morpho)-phonological reduction is especially challenging for an account that uniformly identifies mutated forms with abs-case. If mutated forms are the realization of abs—i.e. the morphologically marked case, then abs-marked pronouns are expected to be more (morpho)-phonologically marked than their erg, i.e. morphologically unmarked, counterparts. While this is the case for lexical nouns, in the case of pronouns the reverse is observed. Namely, that pronouns that are absolutives are (morpho)-phonologically less marked than pronouns that are ergative.

While the alternation observed with Nias’ pronouns does not correspond to the alternation observed with (un)marked lexical nouns, Nias’ pronoun alternation does appear to show behavior of characteristic of clitic and prosodically independent pronouns. As shown below, Nias’ ’mutated’ pronouns display properties that are characteristic of clitics (Cardinaletti & Starke 1999; van Riemsdijk 1999). As illustrated in (58a,58b), when pronouns occur in cleft constructions or marked with the focus particle ha, they surface in their unmutated forms. Conversely, cases of mutated pronouns occurring in contexts where they are foci are unattested in Brown 2001; 2005.

That ’unmutated’ pronouns may be focused, while ‘mutated’ pronouns occurring in focus contexts are unattested in Brown 2001; 2005 suggests that the difference between the two pronoun classes is one of prosodic strength. Namely, that while ’mutated’ pronouns are prosodically deficient clitics, unmutated pronouns are prosodically independent (Cardinaletti & Starke 1999). Further, this difference in strength accounts for the distribution of ‘mutated’ pronouns being restricted to objects (57b) and intransitive subjects (57a). Concretely, I propose that at PF, a clitic (Cl⁰), owing to its prosodic deficient status, attaches to a host within the spell-out domain. Given the syntactic structure of transitive (33) and intransitive (34) clauses, both the object and intransitive subject clitic are within vP domain, and, thus, can cliticize to V (as in (59)).

Consider now transitive subject pronouns, which never surface as reduced ‘mutated’ pronouns, i.e. (57b). Given that transitive subjects, unlike objects and intransitive subjects, evacuate vP, they are no longer within the vP spell-out domain. Thus, cliticization on to V is impossible (as in (60)) and only a full, i.e. prosodically independent, pronoun may surface here.17

Interestingly, this account can explain the exceptional behavior of subject pronouns in experiencer constructions, which surface ‘mutated’, as in (61a,61b).

As was shown in §2.1.1, experiencer subjects are base-generated below the vP and do not raise out. Due to experiencers remaining vP-internal, such arguments will be spelled out in the same domain as the verbal host. Thus, experiencers can be realized as clitic pronouns, which can attach to the verb at PF, as in (62).

To summarize, I conclude that the alternation observed regarding Nias’ pronoun system is not the same as the mutation system observed with lexical nouns. Rather, in the case of pronouns the relevant difference is a difference in pronoun type—i.e. clitic vs. independent pronoun. Moreover, this alternation between clitic and independent pronoun is regulated by the position that the nominal is spelled-out in— i.e. whether it is spelled-out vP-internally or externally. In this way, Nias’ pronoun alternations parallels the mutation on lexical nouns, where the underlying factor that regulates these alternations is not the grammatical role of the argument, but the structural position it is spelled-out in.

On the present analysis, the surface form of marked or ‘mutated’ arguments is due to the insertion of a nasal segment to a nominal stem. Moreover, it was shown that the relevant contexts where this process occurs is when a DP is in the same phase as a lexical head and is spelled-out, as stated in (26). Thus, in light of the present analysis of argument marking as nasal insertion, the generalization can be restated as follows:

Given the generalization in (63), I propose that the aforementioned phonological process that underlies argument marking is due to the application of (64) at PF. As shown, (64) states that nasal insertion occurs on a nominal in contexts where the nominal phrase is within the same phase (i.e. YP), as a lexical head (where X⁰= V, P, N).

Given the rule in (64), nasal insertion is a process that is regulated by the presence (or absence) a lexical terminal syntactic node being located in the same phase as the DP that is spelled-out. In this way, nasal insertion is a morpho-phonological process that, crucially, makes reference to syntactic contexts. Given that (64) is sensitive to the syntactic context in this way— i.e. whether a lexical head is in the same phase as the nominal, I propose that (64) applies early in the PF module when terminals are replaced with Vocabulary Items. Specifically, when the terminals inside the DP are associated with their corresponding Vocabulary Item(s), if they are spelled-out within a phase with a lexical head, e.g. V, (64) applies, and the DP is realized with a nasal as its initial segment.18^,19 Subsequently, after (64) is applied in the cycle, in the case of nominals with a stem-initial consonant, lenition is triggered before the nasal deletes (as discussed in §2.2.2).

As was argued in §2.1.1, that argument marking is sensitive to phase boundaries was supported by the alternation observed with argument marking between transitive and intransitive subjects. The former are systematically outside the vP phase and never surface marked; the latter are systematically within the vP phase and always surface marked. I propose the reason that (64) is sensitive to phase boundaries is due to the cyclic nature of Spell-out and that (64) makes reference to syntactic contexts. Consider first intransitive constructions, as in (34). Upon the merger of the next higher phase head, i.e. v, Spell-out is triggered on the DP and terminals within the DP are associated with Vocabulary Items. Subsequently, when the phase head C merges, the next Spell-out cycle occurs, and the vP phase is sent to the interfaces. Since, there is a lexical head V and a DP in the vP phase, (64) is triggered, as illustrated below in (65); Spell-out cycles are boldfaced).

Consider now the transitive construction, as in (33). When Spell-out is triggered for the vP phase, the lexical V triggers nasal insertion on the object nominal. Moreover, V is also replaced its associated Vocabulary Item. Later, in the next Spell-out cycle, the subject DP is spelled out in the CP phase. However, although the subject DP is spelled-out inside the CP phase, there is no lexical syntactic head that could trigger (64). As (66) shows, when the subject DP is spelled-out, the lexical V has already been replaced with phonological material (which occurred in the earlier vP Spell-out cycle). Thus, due to the cyclic nature of Spell-out and that (64) makes reference to syntactic contexts, transitive subjects are systematically unmarked.

In this respect, Nias’ nasal insertion process and its sensitivity to phase boundaries parallels a number of other morpho-phonological process, which exhibit a similar sensitivity. This type of phase domain sensitivity can be observed with several external sandhi processes. As shown in (67a,67b) in Mende the initial segment of the verb undergoes voicing in the presence of an immediately preceding overt object. However, when the object is null (68a) or moves to a vP-external position (68b), this process is neutralized (Cowper & Rice 1987).

Cowper & Rice (1987) argues that the the key difference between (67a,67b) and (68a,68b) is that while in (67a,67b) the adjacent nominal, i.e. the sandhi trigger, is in the vP/VP with the verb, in (68a,68b) the nominal is vP/VP-external. Assuming that vP delineates a phase, then the Mende data indicate that a requirement on mutation is that the trigger and target be located in the same phase (for similar cases of where sandhi processes are conditioned by phases see: e.g., Pak 2008; d’Alessandro & Scheer 2015, and Bošković 2017).

The same effect occurs in Welsh, where phase boundaries can neutralize mutation. As shown in (69a), the first element after the subject surfaces mutated. However, when the subject and the adjacent element are in distinct CPs, as in (69b) with the wh-item, mutation is neutralized (Pyatt 2003).

Assuming that CP delimits a phase in (69b), what the above data show is that Welsh mutation can be neutralized in cases where the trigger and target occupy distinct phases (see Pyatt 2003 for a proposal of this pattern in terms of intonational phrase boundaries, which have been argued to correspond to phases, cf. Kratzer & Selkirk 2007; Selkirk 2011).

In this respect, I argue that Nias’ argument marking pattern is an instance of this more general pattern, where a (morpho)-phonological process can be neutralized in contexts where the trigger and target are in distinct phases. For Mende, this is observed when the object and verb are both realized in vP. In Welsh, mutation is neutralized when the trigger and target are in distinct phases. For Nias this is observed with transitive subjects, which are outside the vP and, thus, argument marking is neutralized.