Marked unergatives: Syntactic ergativity and nominalizations

Abstract

Samoan deverbal nominalizations show a crosslinguistically rare tripartite-inactive alignment where unaccusative, unergative, and transitive subjects are distinguished by inalienable genitive, alienable genitive, and ergative case, respectively, with objects being marked like unaccusative subjects (Mosel 1992). In addition, subject clitics exhibit a marked unergative alignment, where only unergative subject clitics are distinctly marked by alienable genitive case, whereas all other arguments receive inalienable genitive case. In this study, we demonstrate that these alignments follow naturally from a language-specific combination of independently established phenomena, including (i) prepositional ergativity (Polinsky 2016), (ii) split (in)alienability (Myler 2016; Alexiadou 2003), (iii) split-intransitivity, (iv) the unaccusative restriction on nominalizations (Imanishi 2014; Alexiadou 2001), and (v) a nonuniform nature of clitic pronouns (Bleam 2000), and therefore provides novel evidence for each of these phenomena. Comparing the distribution of ergative case in nominalizations crosslinguistically, we argue that the source of ergativity varies across languages and suggest that the split between syntactic and morphological ergativity cannot be reduced to a category-split of ergative subjects.

Introduction

Deverbal nominalizations as mixed extended projections show properties traditionally related to both the nominal and the verbal domain, such as argument and event structure, plural marking, adjectival and adverbial modification, etc. (Alexiadou and Borer 2020; Borer 2014; Harley 2009; Alexiadou et al. 2007; Alexiadou 2001; Grimshaw 1990 and many others); it has also been acknowledged that different types of nominalizations vary in the exact distribution of such properties: so-called complex event nominalizations entail the event and argument structure of the nominalized verbal predicate whereas result nouns do not (Grimshaw 1990 and much subsequent work). Complex event nominalizations have thereby been commonly described to exhibit specific restrictions on the syntactic realization of their arguments: On the one hand, nominal arguments are typically marked by nominal genitive/possessive case instead of verbal cases like nominative (1)a or accusative (1)b, as illustrated by Greek below.1

1. 

On the other hand, certain nominalizations typically require the external argument of transitive verbs, i.e., the agent, to be introduced by an (optional) prepositional by-phrase, similar to passive constructions (2), instead of a DP (Bruening 2013; Grimshaw 1990). As a result, even accusative languages like Greek show an ergative alignment in the nominal domain (Alexiadou 2017, 2001).

• (2)

Based on this restriction, nominalizations have been argued to be subject to an unaccusativity requirement (3), according to which only unaccusative structures can be nominalized (Imanishi 2020; Alexiadou 2001).

• (3)

While most studies focus on the subject of transitive verbs, the unaccusative requirement also affects the realization of the external argument of nominalized unergative verbs, although their status as complex event nominals has been contended (Grimshaw 1990, cf. Borer 2013). In Greek, such arguments lose their agentive properties in nominalizations and are instead introduced as a possessor, as indicated by the ungrammaticality of agent-oriented modifiers in (4) (Alexiadou 2001).

• (4)

Although the initial motivation for the unaccusative requirement on nominalizations came from Indo-European languages like English, Catalan, and Greek (see Alexiadou 2001 and references therein), parallel restrictions have been described for typologically unrelated languages like Kaqchikel (Mayan; Imanishi 2020, 2014) and Mẽbêngôkre (Jê; Salanova 2007), suggesting a more universal constraint on argument structure in complex event nominals. Yet, crosslinguistic data on the status of external arguments in nominalizations is still rare, especially from nonaccusative languages. Moreover, recent studies challenge the universal character of the unaccusative restriction based on potential counterexamples from a diverse set of languages (Šereikaitė 2023 on Lithuanian, Imanishi 2020 on Chuj and Ch’ol (Mayan), Smirnova and Jackendoff 2017 on Russian, Massam 2000 on Niuean (Polynesian)), suggesting a parametrized view on nominalizations (Imanishi 2020, cf. Alexiadou to appear).

In this study, we present novel data from Samoan (Polynesian) which provides independent evidence for the unaccusative requirement from a combination of verbal and nominal case alignments in complex event nominalizations. While Samoan exhibits an ergative alignment in the verbal/clausal domain and a split-genitive alignment in the nominal domain, nominalizations show a tripartite-inactive alignment on regular DP arguments that distinguishes unaccusative, unergative, and transitive subjects (Mosel 1992; Mosel and Hovdhaugen 1992, cf. Chung 1973): Transitive subjects maintain their ergative marking (5)c, whereas intransitive unergative (5)b and unaccusative subjects (5)a are marked by alienable and inalienable genitive case, respectively.

• (5)

Based on their syntactic and semantic properties, we demonstrate that this typologically rare alignment follows naturally from two independent properties of the Samoan grammar that combine with the more general unaccusative requirement in complex event nominalizations: Firstly, the prepositional nature of transitive subjects in a syntactically ergative language like Samoan (Hopperdietzel 2020; Polinsky 2016; Mosel 1985), which makes them similar to passive by-phrases in accusative languages and allows them to escape the unaccusative requirement on nominalizations. Secondly, inherent alienable genitive case on unergative subjects indicates that only unergative subjects must be based-generated outside of the verbal domain, namely as possessors in the nominal domain, as only they fail to satisfy the unaccusative requirement (cf. Tyler 2021; Myler 2016; Alexiadou 2003).

• (6)

Additional support for the special status of unergative subjects comes from subject clitics which in verbal clauses exhibit a neutralized alignment with the three types of subject being marked by absolutive case in the absence of ergative marking (Mosel and Hovdhaugen 1992). In complex event nominalizations, however, unergative subjects are distinctly marked by alienable genitive case (7)b, whereas transitive subjects (7)c pattern with unaccusative subjects being marked by inalienable genitive case (7)a (Mosel 1992).

• (7)

Based on the presence of the prepositional resumptive pronoun ina only in transitive contexts, we develop a nonuniform analysis of subject clitics in Samoan (cf. Bleam 2000): While intransitive subject clitics receive both their case and thematic role in their original argument position before moving to a preverbal position (cf. Uriagereka 1995), transitive subjects are base-generated in a nonthematic position (cf. Sportiche 1996), receiving the unmarked case of the respective domain (Baker 2015), with their thematic role assigned via resumption. As a result, in nominalizations, unergative subject clitics are both case and thematically licensed in the nominal domain, indicated by alienable genitive case and interpretative shifts.

The marked unergative alignment in Samoan nominalizations therefore supports the absence of external arguments in complex event nominalizations and provides novel crosslinguistic evidence for the unaccusative restriction on nominalizations from an ergative language. Crucially, marked unergatives make further predictions about the source of syntactic ergativity in the language, which we argue, is tied to the prepositional nature of the ergative subject in Samoan. Contrasting the Samoan data with nominalizations from the morphologically ergative language Niuean, which does not display syntactic ergativity and rejects ergative marking in complex event nominalizations (Massam 2020, 2000), our findings support the view that ergative subjects in morphologically and syntactically ergative languages differ regarding their lexical category, i.e., DP vs. PP (Polinsky 2016). However, given that ergative subjects have been argued to violate the unaccusative requirement on nominalizations, e.g., in the Mayan language Kaqchikel (Imanishi 2020, 2014, cf. Burukina 2023), we suggest that syntactic ergativity is not a crosslinguistically uniform phenomenon related to prepositional ergativity (pace Polinsky 2016, cf. Coon et al. 2014). Consequently, the distribution of ergative case in nominalization may be used as a diagnostic for the categorial status of subjects of transitive verbs in future research.

The paper is structured as follows: In Sect. 2, we start our investigation with an overview of the case alignments in Samoan, showing that it exhibits an ergative alignment in the clausal domain and a split-genitive alignment in the nominal domain that is sensitive to (in)alienability. We then turn to the tripartite-inactive alignment found in Samoan complex event nominalizations, in which unaccusative, unergative, and transitive subjects are marked by distinct morphological cases. After demonstrating that Samoan complex event nominalizations embed a VoiceP under the nominalizer n, we explain this alignment by the unaccusative restriction on nominalization, according to which n selects for unaccusative complements. While unergative subjects are forced to be introduced in the nominal domain, we argue that ergative subjects obey this requirement due to their prepositional nature in a syntactically ergative language like Samoan. In Sect. 3, we apply our analysis to the marked unergative alignment of subject clitics. Developing a nonuniform analysis of subject clitics in Samoan, we propose that only transitive subject clitics are base-generated in T, receiving default case, whereas intransitive subject clitics are merged in the original argument position before moving to T. Due to the unaccusative restriction and the absence of prepositional resumption, unergative clitics must be merged in the nominal domain where they receive both inherent alienable genitive case and the possessor role in Spec, PossP. In Sect. 4, we discuss how our findings support a prepositional account of syntactic ergativity in Samoan and how nominalizations can inform the analysis of source of (syntactic) ergativity crosslinguistically. Section 5 concludes.

Tripartite-inactive case

Although argument structure properties of deverbal nominalizations have received much attention in the typological and formal literature (e.g., Alexiadou and Borer 2020; Koptjevskaja-Tamm 1993), the morphosyntactic relations between the nominalized verb and its arguments have been investigated primarily from the perspective of better-studied accusative languages. In such languages, arguments in deverbal nominalizations typically exhibit an ergative alignment where subjects of transitive verbs are distinctly marked from intransitive subjects and objects (1)/(2) (Alexiadou 2017, 2001). Based on this observation, certain nominalizations in these languages have been argued to be subject to an unaccusative requirement that prevents the syntactic projection of an external argument in Spec, VoiceP, making nominalizations similar to passives (Imanishi 2020; Bruening 2013; Alexiadou 2001). Despite its strong crosslinguistic prediction, data from a typologically more diverse set of languages, especially languages that do not exhibit an accusative alignment, is rarely discussed (but see Burukina 2023, 2021; Imanishi 2020, 2014 on various Mayan languages, Polinsky 2016 on Tongan, Salanova 2007 on Mẽbêngôkre, Massam 2000 on Niuean).

In this section, we will contribute to this discussion data from the Polynesian language Samoan, which combines an ergative alignment in the clausal domain with a split genitive alignment sensitive to (in)alienability in the nominal domain to a crosslinguistically rare tripartite-inactive case alignment in deverbal nominalizations (e.g., not discussed by Comrie 2013), where unaccusative, unergative, and transitive subjects are distinguished by their case marking. Crucially, Samoan differs from Mayan languages in that its ergative is not syncretic with the genitive (Coon 2013), providing more direct morphological access to the syntactic status of the respective arguments. Adopting a prepositional analysis of syntactic ergativity in Samoan (Hopperdietzel 2020; Polinsky 2016; Mosel 1985, pace Collins 2021; Tollan 2018), we then argue that the tripartite-inactive alignment naturally follows from the unaccusative restriction on Voice-under-n in complex event nominalizations in the presence of an inherent alienable genitive case assigned by Poss (cf. Tyler 2021 on Mississippi Choctaw, Alexiadou 2001). The tripartite-inactive case alignment in Samoan nominalizations therefore provides independent support for the unaccusative restriction on nominalizations, a syntactic approach on (in)alienability, and a prepositional analysis of syntactic ergativity in the language (see Sect. 4.1 for a more detailed discussion).

Typological and methodological background

Samoan is a Polynesian language of the Oceanic subbranch of the Austronesian language family and spoken by roughly 510 000 speakers of which half of the population lives on the islands of the Samoan archipelago, i.e., the Independent State of Samoa and American Samoa (Lewis et al. 2015). In addition, larger communities reside in Australia, New Zealand, and the USA, especially in Hawai`i and the West Coast. The language is primarily VSO, with other word orders likely sensitive to information structure, and exhibits preverbal TMA marking and postverbal adverbial modifiers, as illustrated below (see Mosel and Hovdhaugen 1992 for a reference grammar).

• (8)

The structure of the DP is parallel to the clausal structure with prenominal case-marking and articles and postnominal modifiers and possessor arguments.

• (9)

To account for the verb-initial syntax, we adopt an analysis based on phrasal VP movement to the Spec, TP and additional T to C movement (Collins 2017, cf. Medeiros 2013; Massam 2001 on remnant movement in Hawai’ian and Niuean, also Middleton 2021; Otsuka 2005 in T-to-C movement in Polynesian languages), combined with distributed deletion of nonverbal (phasal) constituents, including DPs, PPs, and CPs, in the higher copy (van Urk 2022, cf. Clemens 2014).2 This is illustrated by the structure in (10) for the sentence in (8), without the postverbal adverbial vave ‘quickly’.

• (10)

Following Kahnemuyipour and Massam (2006 on Niuean) and Pearce (2005 on Maori), we assume a parallel account for noun-initial syntax with the NP movement to Spec, DP, stranding of phasal arguments in postnominal position, and subsequent D-to-K movement. As linearization is not crucial to our study here, we will present hierarchical structures prior to movement for presentational convenience.

Most data presented in this paper comes from elicitation sessions with five native speakers of Samoan, of which two speakers (one male, one female; born and raised in Western Samoa) live in the United States and three speakers on the island of Upolu in the Independent State of Samoa (two female, one male). Dialectal and interspeaker variation that occurred during the elicitation sessions are mentioned, where relevant. Samoan exhibits two registers, a more formal and a more colloquial one, called t- and k-style. All examples are presented in the t-style, which was voluntarily offered by our consultants. Apart from phonological differences, e.g., /t/ to /k/, while registers only have been described to exhibit a minor impact on the morphosyntax, case marking may be affected (Collins 2014, cf. Mayer 2001; Ochs 1982; Duranti 1981). The data was elicited in on- and offline meetings between 2021 and 2023 via judgment tasks based on manipulated examples, partly adapted from existing sources on Samoan nominalizations (Mosel 1992; Mosel and Hovdhaugen 1992; Chung 1973), as well as translation tasks based on English stimuli. While all examples and stimuli were tested in the context of full clauses, we decided to present only the relevant part, i.e., the nominalization itself for presentational clarity.

Samoan case

In the following, we focus on Samoan case alignment in the verbal and the nominal domain before we turn to the combination of both alignments in deverbal nominalizations. In particular, we adopt a prepositional analysis of syntactic ergativity and a structural analysis of the split-genitive alignment sensitive to (in)alienability, which will be crucial for the analysis of the observed tripartite-inactive in deverbal nominalizations.

The clausal domain: Ergative/absolutive alignment

In the clausal domain, Samoan exhibits an ergative/absolutive case alignment (Mosel and Hovdhaugen 1992).3 The transitive subject is marked by the ergative marker e (11),4 whereas intransitive subjects and objects are marked by tonal absolutive case, which is realized by a pitch rise on the preceding syllable (Yu 2021).5

• (11)

Notably, Samoan shows properties of syntactic ergativity as ergative arguments reject A’-movement to a clause-initial position (Muāgututi’a 2018; Polinsky 2016; Mosel and Hovdhaugen 1992). In topicalizations, for example, absolutive arguments move to a clause-initial position without any additional material present (12)a (see Hohaus and Howell 2015; Potsdam and Polinsky 2011 for a discussion of `o in Samon and its cognates in other Polynesian languages). Topicalized ergative subjects however drop their ergative marking and the resumptive pronoun ina shows up in the original subject position (12)b (Hopperdietzel 2020; Cook 1988, also Ershova 2024).6

• (12)

Ergative subjects thus resemble other oblique PPs in the language which drop their oblique marker i in topicalized position and require the presence of a resumptive pronoun, like i ai, in postverbal position.

• (13)

Adopting Polinsky’s (2016) prepositional analysis of syntactic ergativity, we take ergative and oblique marking to be inherently prepositional. In particular, we assume that ergative subjects (left) adjoin as (passive-like) prepositional adjuncts to VoiceP, saturating the agent role introduced by Voice (10) (cf. Mosel 1985 on the adjunct-like optionality of ergative arguments in isolated sentences). Since Samoan lacks both preposition stranding and pied-piping (13)b, prepositional constituents, including the ergative subject, are trapped in their base-generated position. The topicalization of prepositional arguments therefore requires resumption, in which the argument is base-generated as a DP argument in the left periphery and resumed by a prepositional anaphoric pro-form in its original VoiceP-internal position.7 This is illustrated for topicalized ergative subjects in (14).

• (14)

This analysis is supported by the morphological markup of the ergative resumptive pronoun, which can be decomposed into the general oblique marker i and the deictic demonstrative proform nā ‘that there’ (Mosel and Hovdhaugen 1992: 131f).8 As discussed in detail by Polinsky (2016) for Tongan, prepositional ergativity is expected to correlate with additional morphosyntactic phenomena, particularly anaphoric binding, raising, and control. Polinsky (2016) also offers tentative evidence that these patterns hold in Samoan (pace Chung 1978), though a systematic investigation lies beyond the scope of this paper: Firstly, Samoan lacks reflexive anaphoric pronouns (Mosel 1991). Secondly, potential (hyper-)raising structures seem to be only apparent, as ‘raising’ can target nonsubjects, including obliques (e.g., masani-raising), can (optionally) leave a pronominal copy in the embedded clause, and lacks case connectivity of the ‘raised’ argument (15) (cf. Chung 1978 for a discussion).9

• (15)

Finally, potential control structures show properties of nonobligatory control, as they allow binding by noncommanding antecedents (16) (Polinsky 2016), and optional overt pronouns in the subordinated clause.

• (16)

Note that the label transitive/ergative “subject” as used in this study is therefore somewhat misleading, since ergative PPs are more similar to oblique PPs, such as passive by-phrases in English, with the absolutive argument being the “true” subject of the clause (cf. Chung 1978; Hohepa 1969 on syncretism of ergative subjects and passive by-phrases in Western and Eastern Polynesian languages). Despite this complication, we continue to refer to ergative PPs as transitive subjects, following the tradition of Polynesian grammar description (but see Cook 1991; Mosel 1985 for a critical discussion of the notion of “subject” in Samoan). We further emphasize that our analysis of the ergative case marker as a preposition is specific to Samoan (and potentially to other Polynesian languages) and is not intended as a universal claim about ergativity across languages. In particular, we make no claims about the source of syntactic ergativity in unrelated language families, where similar surface patterns may arise from distinct underlying structures; for example, high absolutive syntax in Mayan languages (cf. Coon et al. 2014).

In contrast to inherent/prepositional ergative marking, we treat absolutive case as the unmarked case in the clausal domain (Baker 2015; McFadden 2004; Marantz 1991), though our analysis is fully compatible with traditional agree-based approaches. In line with a high-absolutive analysis of Samoan (Hopperdietzel 2020; Tollan 2018; Koopman 2012), we assume that absolutive case is assigned to the highest DP in the TP. Evidence for this assumption comes from the unavailability of absolutive case in nonfinite contexts in which T is absent. This is illustrated by bare nominalizations below, which lack both T and absolutive case (see Sect. 2.3.2 for a detailed discussion; pace Collins 2014).

• (17)

Absolutive case in Samoan is therefore equivalent to nominative case in accusative languages like English or Greek (ABS=NOM; Legate 2008), where nominative is assigned to internal arguments in passive constructions. If Samoan absolutive case were accusative case in transitive clauses (ABS=DEF), we would expect absolutive case to be available in nominalizations, especially in the presence of ergative marking on the transitive subject (but see FN15 for a more complex empirical picture). As regular DPs, absolutive arguments can freely undergo A’-movement to a clause-initial position in topicalization, as illustrated for (12)a below.

• (18)

To summarize, Samoan exhibits an ergative alignment with ergative marking being an inherent prepositional case. Ergative subjects thus resemble passive by-phrases, suggesting that semantically transitive clauses are syntactically intransitive in the narrow sense (cf. Mosel 1985, see Otsuka 2011 for a similar intuition on Polynesian).10

The nominal domain: Split-genitive alignment

In the nominal domain, Samoan like other Polynesian languages distinguishes between alienable and inalienable possessors morphologically by distinct genitive cases on the possessor (Mayer 2001; Mosel and Hovdhaugen 1992: 282ff, cf. Clark 2000; Wilson 1982; Biggs 1969). On the one hand, inalienable possessors, including body parts (19)a, part–whole relationships (19)b, and characteristic properties (19)c, are marked by the case marker o.11

• (19)

On the other hand, alienable possessors are marked by the genitive case marker a, including nonpermanent and initiated possession (20)a/b and transitory relationships (20)c.

• (20)

Some nouns can combine with both alienable and inalienable possessors, depending on the possessive relationship that holds between the two nouns.

• (21)

• (22)

Notably, alienable and inalienable genitive marking can cooccur within a single noun phrase, suggesting independent morphosyntactic sources for inalienable and alienable genitive case (see also Ball 2009 on Tongan).12

• (23)

Adopting a structural analysis of (in)alienability (Armstrong 2024; Tyler 2021; Myler 2016; Alexiadou 2003, 2001; Szabolsci 1994 and references therein), we assume that inalienable possessors merge nP-internally as direct arguments of relational nouns. As such, they receive inalienable genitive case, which we take to be the unmarked case in the DP domain, equivalent to absolutive case in the verbal domain (cf. Baker 2015; see Sect. 4 for further parallels between absolutive and (inalienable) genitive case), as illustrated for example (19)a in (24) below.

• (24)

In contrast, alienable possessors are merged in the specifier of an additional functional head Poss, parallel to Voice in the verbal domain, which on the semantic level, introduces an abstract possessive relation between the two nouns (Barker 1995). While in languages like English or Greek, where inalienable and alienable possessors are marked alike, the structural distinction between alienable and inalienable possessors is not reflected by morphology, alienable genitive case is inherently assigned to nP-external argument DPs in Spec, PossP in languages like Samoan (cf. Tyler 2021 on inherent dative case on alienable possessors in Mississippi Choctaw).13

• (25)

Alienable genitive case in Samoan is therefore an inherent case in the nominal domain, similar to morphological ergative case in languages like Basque, where vP-external arguments, i.e., subjects of unergatives and transitive verbs, are uniformly marked by ergative case (e.g., Holmer 1999).

Deverbal nominalizations: Tripartite-inactive alignment

In mixed domains like deverbal nominalizations, we expect the two alignments, i.e., syntactic ergativity from the clausal domain and split-genitive from the nominal domain, to interact in the way predicted by their internal structure (Alexiadou 2020; Iordăchioaia 2020a; Alexiadou et al. 2007 and references therein). In particular, the availability of verbal and nominal cases reflects the presence of verbal and nominal layers. For example, the availability of accusative case in English -ing-nominals correlates with the presence of an (accusative-assigning) transitive Voice head (26)a, whereas the unavailability correlates with the absence thereof (26)b (Alexiadou 2013; Harley 2009).

• (26)

While distinct types of nominalization exist in Samoan, e.g., nominalizations derived by nominalizers like -ga or -Caga, our focus lies on bare nominalizations, i.e., nominalizations in which verbal predicates combine with an article in the absence of overt morphology (see Mosel and Hovdhaugen 1992; Mosel 1992, also Ball 2009 on Tongan, Cablitz 2000 on Marquesean, Massam 2000 on Niuean, Hooper 1996 on Tokelauan, cf. Moyse-Faurie 2016; van Lier and van Rijn 2013; Clark 1981; Chung 1973 for further discussion). Bare nominalizations thereby denote the eventuality of the nominalized verb. They have the same distribution as regular DPs but are often used as arguments of psych (27)a, perception (27)b, and communication verbs, or as headlines or (book/story) titles (27)c.

• (27)

In the absence of absolutive case, bare nominalizations exhibit a tripartite-inactive alignment, in which all three types of subjects are marked by distinct case markers. Firstly, subjects of unaccusative verbs, including dynamic verbs like pa‘ū ‘fall’ and tanu‘u ‘arrive’ as well as stative verbs like laulelei ‘be beautiful’ are marked by inalienable genitive case.

• (28)

Secondly, subjects of unergative verbs like pese ‘sing’, ata ‘laugh’, and galue ‘work’ are marked by alienable genitive case. As a result, the choice of inalienable and alienable genitive case reflects the distinction between two types of intransitive subjects and can be used as a diagnostic for split-intransitivity (Collins 2014, cf. Burzio 1986; Perlmutter 1978; see Sect. 3.3 for further discussion).14

• (29)

Finally, as expected from unaccusative nominalizations, objects of transitive verbs receive inalienable genitive case.15 Subjects of transitive verbs however maintain their ergative marking from the clausal domain and cannot appear with either alienable like unergative subjects or inalienable genitive case.

• (30)

As mentioned by Muāgututi’a (2018), speakers often tend to insert the resumptive pro-form ina into bare nominalizations of transitive verbs; though its presence is not obligatory as such nominalizations without ina are frequently found in the literature (see Mosel and Hovdhaugen 1992) and judged grammatical by our consultants.16

• (31)

It is important to note that optional ina insertion is also observed outside of nominalizations in verbal clauses. Here, its distribution is sensitive to information structure properties, indicating the salience of the agent in the given discourse, if present (Mosel 1985, also Cook 1978).

• (32)

As in-situ ina-doubling is generally optional and not restricted to nominalizations, we take ina-doubling as an independent phenomenon without direct impact on the analysis of argument structure in nominalizations, especially in comparison to respective verbal clauses. While a detailed analysis of the morphosyntactic and pragmatic properties of ina doubling, especially in nominalizations, is far beyond the scope of this paper, we tentatively adopt a big-PP approach, in which ina and the ergative subject cooccur within a single PP, for explanatory purposes (Grewendorf 2008, cf. Cardinaletti 2019; Uriagereka 1995 for big XP approaches).17

• (33)

To summarize, the combination of the ergative/absolutive alignment in the clausal domain and the split-genitive alignment sensitive to alienability in the nominal domain results in a tripartite-inactive alignment, in which all three types of subjects are marked by distinctly: unaccusative subjects and objects, are marked by inalienable genitive case, unergative subjects are marked by alienable genitive case, and transitive subjects maintain their ergative marking from the clausal domain (Table 1).

Table 1.

Case marking in verbal clauses and nominalizations

	clausal domain	nominalization
Sunacc	Habs	oinal
Sunerg	Habs	aal
A	eerg	eerg
O	Habs	oinal

Voice-under-n

To investigate the source of the tripartite-inactive alignment, we first establish the internal structure of Samoan bare nominalizations. Crosslinguistically, it has been argued that complex event nominalizations as mixed projections vary regarding the presence of nominal and verbal structure on the one hand, and the nominalizer, namely n or D, on the other hand (Alexiadou 2020; Iordăchioaia 2020a; Alexiadou et al. 2010). Crucially, the type of nominalizer determines the presence of nominal structure, as only n-based complex event nominalizations exhibit all properties of regular nouns (34), i.e., possessives, number, etc., whereas D-based complex event nominalizations lack lower nominal projections (35). Moreover, only D-based nominalizations can embed larger verbal structures, such as TPs or CPs.

• (34)

• (35)

In the following, we demonstrate that Samoan bare nominalizations are n-based nominalizations that maximally embed AspP complements. As such, they are expected to obey the unaccusativity restriction on nominalizations, according to which DP arguments are banned from Spec, VoiceP under n (Alexiadou 2017, 2001; Imanishi 2020, 2014; Bruening 2013; Salanova 2007).

Verbal properties

Initial evidence for the presence of a vP in bare nominalizations comes from the availability of verbalizing morphology, such as the causative prefix ta- (36)a and the anticausative prefix ma- (36)b that form anticausative and causative variants from a-categorial roots (cf. Hopperdietzel 2020; Mosel and Hovdhaugen 1992: 184ff, 188f).

• (36)

Likewise, the verbalizing prefix liu- which derives inchoative verbs from nouns, such as liu-suavai ‘become liquid, melt’ from suavai ‘(a) liquid’, can show up in bare nominalizations (Hopperdietzel 2020).18

• (37)

Further evidence comes from verbal agreement, marked by partial reduplication (38)a, designated plural morphology (38)b, or suppletion (38)c (Zuraw et al. 2014; Mosel and Hovdhaugen 1992: 220ff, 442ff). While the locus of plural agreement has been identified as v (Thornton 2019, cf. Bobaljik and Harley 2017; Toosarvandani 2016; Haji-Abdolhosseini et al. 2002), the verb agrees with the absolutive argument in verbal clauses but with the inalienable genitive argument in nominalizations (see also Sect. 2.4.2).19

• (38)

Moreover, Samoan verbs optionally indicate event number independently of argument number by pluractional morphology such as full reduplication (Thornton 2019; Mosel and Hovdhaugen 1992: 224ff, cf. Haugen 2011; Haji-Abdolhosseini et al. 2002), which is also grammatical in bare nominalizations.

• (39)

In addition to prefixed verbs, more complex directional vPs can be nominalized where a motion verb combines with a directional particle like ifo ‘down’ which introduces a prepositional goal argument (see also example (38)c above; Mosel and Hovdhaugen 1992: 558).

• (40)

Finally, vP-internal modifiers like oblique causers are grammatical in nominalizations.20 The last two diagnostics indicate that bare nominalizations embed a phrasal vP (Hopperdietzel 2021, cf. Alexiadou et al. 2015).

• (41)

Evidence for the presence of a VoiceP in bare nominalizations comes from Voice-related morphology that cooccurs with ergative-marking on the subject (Hopperdietzel 2021, 2020; Tollan 2018): Firstly, the ergativizing suffix -Cia which derives ergative clauses from verbs (42)a that cannot take an ergative subject otherwise (42)b, adding a higher degree of agency (cf. Cook 1996; Mosel and Hovdhaugen 1992: 198ff).

• (42)

Secondly, the causative prefix fa‘a- obligatorily introduces (ergative) agentive causers (43)a but is ungrammatical with (oblique) nonagentive causers (43)b (Hopperdietzel 2021; Koopman 2012, cf. Mosel and Hovdhaugen 1992: 175f).

• (43)

That both types of Voice-related morphology are grammatical in bare nominalizations indicates that bare nominalizations are derived from a verbal core which minimally includes a VoiceP.

• (44)

Additional support for this analysis comes from the availability of instrumental modification in bare nominalizations, as illustrated by the modifier i lona lima tusi ‘with its forefinger’ below.

• (45)

Further, the presence of post- and preverbal aspectual modifiers suggests that the nominalized verbs can also come with aspectual layers (Mosel and Hovdhaugen 1992: 381ff, 557ff). This holds for postverbal aspectual modifiers like so‘o ‘frequently’ (46), as well as for preverbal aspectual modifiers like toe ‘again’ (47). In nominalizations, the latter occur in between the article and the verb.

• (46)

• (47)

While the presence of (low) aspectual adverbs signals the availability of aspectual projections in bare nominalizations (cf. Cinque 1999; Alexiadou 1997), bare nominalizations do not embed constituents of the size of a TP. Unlike certain types of nominalizations in Turkish and Japanese (Kornfilt and Whitman 2012), tense morphology such as the past tense marker sā is ungrammatical in Samoan (Collins 2014).

• (48)

As absolutive case is associated with a TP-layer in Samoan (Tollan 2018; Koopman 2012), its absence supports a TP-less analysis (pace Collins 2014, cf. de Miguel 1996 on nominative case in Spanish verbal infinitives).

• (49)

As expected by the absence of a TP-layer, also higher sentential adverbs that occur to the left of tense/aspect particles (50) as well as complementizers are ungrammatical in bare nominalizations (51) (Collins 2014).

• (50)

• (51)

Summarizing our investigation (Table 2), bare nominalizations embed a reduced verbal structure of maximally the size of an Asp(ectual)P, with higher projections such as TP and CP generally absent. This indicates that Samoan bare nominalizations are complex event nominals that come with a verbal argument structure.

Table 2.

Verbal layers in Samoan bare nominalizations

Layer	Verbal clauses	Nominalizations	Diagnostics
vP	Yes	Yes	– verbalizing morphology
			– number agreement
			– pluractionality
			– directionals
			– (oblique) causers
VoiceP	Yes	Yes	– ergativizing morphology (-Cia)
			– causative morphology (fa‘a-)
			– instrumental modification
			– ergative marking
AspP	Yes	Yes	– postverbal aspectual modifiers
			– preverbal aspectual modifiers
TP	Yes	No	– T/A marker
			– absolutive case
CP	Yes	No	– sentential adverbs
			– complementizer

Nominal properties

Based on the morphosyntactic type of the nominalizer, two types of nominalizations have been proposed crosslinguistically (Alexiadou 2020; Iordăchioaia 2020a; Alexiadou et al. 2010). On the one hand, n-based nominalizations exhibit all nominal layers and thus the properties of regular nouns, including possessives, diminutives, number, and different types of determiners (52)a. On the other hand, D-based nominalizations are somewhat defective as they only include a D-layer with projections for number, diminutives, and possessives missing (52)b.

• (52)

We now demonstrate that Samoan bare nominalizations pattern with n-based nominalizations, even though the nominalizer itself is not overtly realized. Initial evidence comes from alienable genitive case on subjects of unergative nominalizations, which is associated with the presence of a PossP in the nominal domain (see Sect. 2.2.2; Tyler 2021; Alexiadou 2001).21

• (53)

In Samoan, nominal properties are usually realized on the prenominal determiner (Mosel and Hovdhaugen 1992: 149): Firstly, the article si is used for diminutives which are “implicitly specific and connotes sympathy and belittlement” (Mosel and Hovdhaugen 1992: 264). Notably, the diminutive article can combine with both regular nouns (54)a and bare nominalizations (54)b (Mosel and Hovdhaugen 1992: 538).

• (54)

Secondly, number marking of specific plural noun phrases is indicated by the absence of an overt article. Again, both regular nouns (55)a and nominalizations (55)b can appear in their bare forms; though plural marking on the latter is rather rare with speakers preferring verbal plural marking instead (Mosel and Hovdhaugen 1992: 539).22

• (55)

Finally, nominalizations cannot only be combined with specific articles but also with nonspecific articles like the singular se, primarily in negative and irrealis contexts (Mosel and Hovdhaugen 1992: 537f).

• (56)

The observation that different types of determiners can occur in Samoan bare nominalizations therefore provides support for n-based analysis, since D-based nominalizations have been argued to be restricted to default determiners, due to their defective nature (i.e., the lack of D-n agreement; Iordăchioaia 2014).

Depending on their syntactic position, bare nominalizations can take absolutive case (27)b or combine with oblique (27)b, and more rarely ergative (57), prepositions, further supporting the assumption of the presence of a DP layer in bare nominalizations.

• (57)

To summarize, bare nominalizations show the same morphosyntactic properties as regular DPs and can thus be classified as n-based nominalizations that entail all nominal layers, including PossP, DimP, NumP, and DP (Table 3).

Table 3.

Nominal layers in Samoan bare nominalizations

Layer	Regular nouns	Nominalizations	Diagnostics
nP	Yes	Yes	– variable determiners (via D-n agreement)
PossP	Yes	Yes	– alienable genitive case on unergative subjects
DimP	Yes	Yes	– diminutive article si
NumP	Yes	Yes	– plural article ø
DP	Yes	Yes	– determiners
			– case marking
			– prepositions

The unaccusative requirement on n

Based on their morphosyntactic properties, we can classify Samoan bare nominalizations as n-based complex event nominals that embed verbal constituents maximally of the size of AspP.

• (58)

In the following, we first focus on nominalizations embedding a VoiceP complement, before returning to nominalizations embedding an AspP in Sect. 3.3. Crosslinguistically, Voice-under-n in many languages have been argued to be subject to an unaccusative requirement (59) (Alexiadou 2017, 2001; Imanishi 2014; Bruening 2013; Salanova 2007, but see Šereikaitė 2023; Imanishi 2020; Smirnova and Jackendoff 2017 for a more nuanced view), i.e., n selects for VoiceP complements that lack a specifier. Specifically, Bruening (2013) argued in detail that n in nominalization is similar to a passive head: next to changing lexical category it also demotes the external argument, which may then be introduced as an adjunct in the verbal domain, i.e., a by-phrase,23 or as a possessor argument in Spec, PossP in the nominal domain.24

• (59)

This restriction has consequences for both transitive and unergative subjects which are merged vP-externally in Spec, VoiceP, as illustrated by Greek: Transitive subjects, on the one hand, can only be realized by an optional by-phrase. Consequently, even languages that display an accusative alignment in the clausal domain show an ergative alignment in the nominal domain (see Alexiadou 2017 for a detailed discussion and crosslinguistic data).

• (60)

Unergative subjects, on the other hand, are merged as possessors in the nominal domain, as they violate the unaccusative requirement in the absence of a prepositional (passive) alternative. They therefore differ from transitive and unaccusative subjects in not being thematically licensed within the verbal domain, as indicated by the infelicity of agent-oriented modifiers like siniditi ‘conscious’ in (61).25

• (61)

As a result, the unaccusative requirement on n determines how external arguments can be introduced in nominalizations, namely either as oblique by-phrases in the verbal domain in the context of nominalizations of transitive verbs or as a possessor in the nominal domain in the context of unergative verbs.26

Three types of subjects

In the following, we demonstrate how the tripartite-inactive alignment in Samoan nominalizations follows naturally from the interaction of the unaccusative requirement on n-based nominalizations with the split-genitive marking based on (in)alienability and prepositional ergativity.

Unaccusatives

As vP-internal arguments, unaccusative subjects obey the unaccusative requirement on nominalizations and are merged in their canonical position as complements of v below an unaccusative expletive Voice head in both the clausal domain and bare nominalizations. In the absence of T in nominalizations, unaccusative subjects do not receive absolutive case but inalienable genitive case, the unmarked case in the nominal domain (cf. Alexiadou 2017; Baker 2015).

• (62)

Unergatives

As vP-external arguments, unergative subjects violate the unaccusative requirement on nominalizations, as they are merged in Spec, VoiceP in the clausal domain (Alexiadou et al. 2015; Kratzer 1996). In nominalizations, unergative subjects are therefore merged as possessors in Spec, PossP in nominalizations to satisfy the unaccusative requirement. Consequently, unergative subjects receive inherent alienable genitive case from Poss instead of unmarked inalienable case like unaccusative subjects. Unlike in languages like Greek or English, the two types of intransitive subjects are therefore distinguished due to the split-genitive alignment in the nominal domain.

• (63)

Crucial evidence for the special status of unergative subjects in nominalizations comes from the observation that unlike unaccusative subjects, number agreement (via partial reduplication; see Sect. 2.3.1) on nominalized unergative verbs is strongly disfavored (64)b; although it is grammatical in verbal clauses (64)a. As number agreement has been argued to be subject to locality constraints (Thornton 2019; Bobaljik and Harley 2017), this contrast suggests that unergative subjects in nominalizations appear in a nonlocal position to v, which we identify as Spec, PossP.

• (64)

However, nominalized unergative verbs still exhibit a verbal core, as they allow full reduplication to mark pluractionality (65) and take hyponymous (66)a and cognate objects (66)b, which in Samoan are introduced as oblique arguments (cf. Tollan and Massam 2022; Tollan 2018 for a discussion).27

• (65)

• (66)

Although the unergative subject is introduced in the nominal domain, nominalized unergative verbs still contain an (unaccusative) agentive Voice head, as they cooccur with Voice-related morphology, such as the causative prefix fa‘a- (67)a and the ergativizing suffix -mia (67)b, e.g., in the context of object omission or pseudo noun incorporation (cf. Mosel 1992).28

• (67)

In nominalizations, unergative subjects therefore differ from unaccusative subjects in that they are merged as possessors outside of the verbal domain, in Spec, PossP, where they receive inherent alienable genitive case, resulting in morphological distinct marking.

Transitives

As discussed in Sect. 2.2.1, transitive subjects in a syntactically ergative language like Samoan are merged as PPs, resembling passive by-phrases in accusative languages like English and Greek. As such, transitive subjects in Samoan satisfy the unaccusative requirement on nominalization since Spec, VoiceP is not occupied. Therefore, ergative marking survives nominalization in Samoan. In the absence of absolutive case, objects as internal arguments receive unmarked inalienable genitive case, parallel to unaccusative subjects, giving rise to the observed tripartite-inactive case alignment.29

• (68)

However, if transitive subjects bear absolutive case, e.g., due to pseudoincorporation (69) or antipassivization of the object (70) (Collins 2017; Shibatani 2006, cf. Mosel and Hovdhaugen 1992: 108ff, 738ff), transitive and unergative subjects behave alike, i.e., being introduced as possessors marked by alienable genitive a case (Mosel 1992).

• (69)

• (70)

The examples above therefore suggest that the grammaticality of ergative marking in bare nominalizations follows from its prepositional nature, as external DP-arguments become ungrammatical in such contexts, providing novel crosslinguistic evidence for the unaccusative requirement on nominalizations.

To summarize, we derive the tripartite-inactive alignment from a language-specific combination of independent phenomena: (i) split-intransitivity, (ii) split (in)alienability, (iii) prepositional ergativity, and (iv) the unaccusative requirement on nominalizations. Unergative DP subjects are thus merged in the nominal domain, receiving alienable genitive case, whereas transitive subjects can maintain their prepositional ergative marking. In the absence of absolutive case, internal arguments, i.e., unaccusative subjects and objects, get unmarked inalienable genitive case.

Marked unergatives

Further evidence for the special status of unergative subjects in bare nominalizations comes from subject clitics, which in the clausal domain show a neutralized alignment. In nominalizations, both unaccusative and transitive subject clitics receive inalienable genitive case, whereas unergative subject clitics receive alienable genitive case (Mosel and Hovdhaugen 1992; Mosel 1992). Motivating a nonuniform analysis of Samoan clitic pronouns (cf. Bleam 2000; Uriagereka 1995; Sportiche 1996, see Anagnostopoulou 2017 for an overview), we take intransitive subject clitics to be derived by movement out of a big-DP, where they are assigned their respective case. In contrast, transitive subject clitics are merged directly to a higher syntactic head, being resumed by prepositional ina in their original theta position in Spec, VoiceP, parallel to left dislocation of regular transitive subjects. As a result, transitive subject clitics are exceptionally marked by the unmarked case of the respective domain, i.e., absolutive in the clausal and inalienable genitive in the nominal domain (cf. Baker 2015; Cardinaletti and Starke 1999). We therefore derive the distinct marking of unergative subjects again from the unaccusative requirement on nominalizations, as only unergative subject clitics require both case and thematic licensing in the nominal domain.

Towards the end of the section, we also address apparent variation in genitive marking of intransitive subjects, i.e., alienable genitive case on unaccusative subjects and inalienable genitive on unergative subjects. We demonstrate that this alternation has syntactic and semantic effects, which further supports our analysis of Samoan bare nominalizations and provides novel evidence for a variable intransitivity (cf. Neu 2023; Sorace 2000).

Subject clitics

We begin our investigation of marked unergatives with an investigation of preverbal subject clitics, which exhibit a neutralized alignment with all three types of subjects being marked by absolutive case. Based on the distribution of the resumptive pronoun ina (Mosel 1985; Cook 1978), we argue that transitive subject clitics do not result from movement out of a big-DP but are directly merged in the inflectional domain (cf. Bleam 2000).

Subject clitics vs. independent pronouns

In verbal clauses, clitic pronouns in Samoan appear in between T/A markers and the verbal predicate, typically in the second position of the clause (with the exception of the generic T/A-marker which precedes the clitic), realizing the person and number feature of (in)transitive subjects (Cook 1994; Mosel and Hovdhaugen 1992: 455ff, cf. Moyse-Faurie 1997 for a cross-Polynesian overview).30 While unaccusative and unergative subject clitics maintain their absolutive case in preverbal position (71)a/b, transitive subject clitics are not marked by ergative but by absolutive case instead (71)c. Therefore, subject clitics exhibit a neutralized alignment, including double absolutive case in the context of transitive verbs (see Yu 2021 for a discussion of tonal absolutive on preverbal subject clitics). Similar to left dislocation of transitive subjects, ergative drop on preverbal transitive subject clitics shows a strong tendency to cooccur with the prepositional resumptive pronoun ina (also Cook 1996; Chung 1978).31

• (71)

Phonologically, subject clitics are reduced independent personal pronouns, which appear in postverbal position and pattern like regular DPs showing an ergative alignment; though third-person singular pronouns tend to be marked by the presentative marker `o, when they are appear as absolutive arguments, as illustrated by the object in (72)c (Mosel and Hovdhaugen 1992: 121f).

• (72)

Table 4 provides a (simplified) overview of the morphophonological relations between different pro-forms, indicating that the plural forms differ from the singular and dual forms in having the same phonological form for both clitic and independent pronouns (Mosel and Hovdhaugen 1992: 121f).32 Our examples will primarily come from singular contexts, leaving potential variation between singular and plural clitics to future research.

Table 4.

Independent and clitic pronouns in Samoan (adapted from Mosel and Hovdhaugen 1992: 121f)

	sg		dual		pl
	Independent	Clitic	Independent	Clitic	Independent	Clitic
1.incl	a‘u	ou / =u	ta‘ua	tā	tatou	tātou
1.excl	—	—	ma‘ua	mā	matou	mātou
2	‘oe	e	‘oulua	lua	‘outuo	‘outuo
3	ia	ia, na	la‘ua	lā	latuo	lātuo

Subject clitics can be identified as true clitics (in the sense of Cardinaletti and Starke 1999), as they can be doubled by independent pronouns in postverbal position, which gives to an emphatic reading related to (contrastive) focus (Mosel and Hovdhaugen 1992: 457, cf. Moyse-Faurie 1997).33

• (73)

In transitive contexts (73)c, this can result in clitic tripling with the subject expressed by the preverbal absolutive subject clitic as well as the postverbal prepositional pro-form ina and the ergative independent pronoun.

A nonuniform analysis of subject clitics

We tentatively adopt a nonuniform analysis of subject clitics in Samoan, according to which transitive subject clitics are similar to agreement markers being merged directly to T, whereas intransitive subject clitics are derived by movement out of a big-DP (see Bleam 2000 on the movement vs. base-generation distinction in Spanish accusative and dative clitics, cf. Anagnostopoulou 2017; Cuervo 2003). This analysis is motivated by the parallelism of dislocated transitive subjects: In both left dislocation (74)a and cliticization (74)b, the transitive subject appears without ergative marking but the postverbal prepositional resumptive pronoun shows up in the postverbal subject position.34

• (74)

In Sect. 2.2.1, we have developed an analysis of dislocated transitive subjects as being base-generated in clause-initial position, being resumed by ina in the original subject position, due to their prepositional nature prohibiting their movement. Adopting a base-generation account (Baker and Kramer 2018, cf. Sportiche 1996inter alia, also Polinsky 2016 on Tongan), we propose a related account for transitive subject clitics. In particular, we assume that transitive subject clitics are merged directly to T, where they, as D-elements, receive unmarked absolutive case and undergo T-to-C movement. In this position, they are resumed overtly by the prepositional ina, and sometimes covertly by pro (see FN31), in the thematic subject position in Spec, VoiceP (see Polinsky 2016 on pro resuming preverbal clitics in Tongan).35

• (75)

For intransitive subject clitics, we again notice the parallelism to left dislocation, suggesting that preverbal clitic intransitive subjects clitics do not require resumption in such contexts.

• (76)

We therefore adopt a movement analysis of intransitive subject clitics (Cardinaletti 2019; Grewendorf 2008; Uriagereka 1995 inter alia, cf. Anagnostopoulou 2017 for an overview, also Collins 2017 on Samoan). Accordingly, intransitive subject clitics are base-generated together with an independent pronoun within a big-DP, where they both are assigned absolutive case before moving to T, as illustrated for unergative (77)a and unaccusative subject clitics (77)b below.

• (77)

To summarize, we proposed a nonuniform analysis of Samoan subject clitics which follows from the prepositional nature of ergative subjects: Transitive subject clitics are base-generated in T and resumed by the prepositional pronoun ina, whereas intransitive subject clitics are born within a big-DP in the original argument position before moving to T.

Possessive clitics

The distribution of clitics is not limited to the verbal domain as they also occur as prenominal possessive clitics in the nominal domain, where they are transparently marked by either alienable or inalienable genitive case (Mosel 1992; Mosel and Hovdhaugen 1992: 549ff). Subject and possessive clitics exhibit the same morphophonological form, except for the second-person singular, which has the suppletive form =u, and the third-person singular, which is restricted to na in prenominal position (Table 5). If a clitic starts with a vowel, the initial vowel is fused with the genitive case marker a or o. Given those minor morphophonological differences, we take subject and possessive clitics to instantiate the same underlying phenomenon which primarily differ in their case marking.

Table 5.

Subject and possessive clitics in Samoan (adapted from Mosel and Hovdhaugen 1992: 124)

	sg		dual		pl
	Subject	Possessive	Subject	Possessive	Subject	Possessive
1.incl	o‘u / =u	=‘u	tā	=tā	tātou	=tātou
1.excl	—	—	mā	=mā	mātou	=mātou
2	e	=u	lua	=lua	‘outuo	=utou
3	ia, na	=na	lā	=lā	lātuo	=lātuo

In the following, we show that possessive clitics exhibit the same (in)alienable split as possessors realized by regular DPs and independent pronouns. In contrast to the neutralized alignment of preverbal subject clitics in the clausal domain and tripartite-inactive alignment of regular DPs and independent pronouns in bare nominalizations, we observe a marked unergative alignment, in which only unergative clitics receive alienable genitive case. We then demonstrate how this unexpected pattern supports our analysis of bare nominalizations in Samoan.

(In)alienability split

Like regular DPs, independent possessive pronouns that appear in postnominal position are marked for either alienable or inalienable genitive case, depending on the possessive relation between the possessor and the possessum (Mosel and Hovdhaugen 1992: 280).

• (78)

Although independent possessive pronouns are available, speakers usually prefer the use of prenominal possessive clitics, which exhibit the same (in)alienability split as other possessive noun phrases (Mosel and Hovdhaugen 1992: 278). Phonologically, possessive clitics clitisize with their genitive case marker and the article.36

• (79)

Building on our analysis of subject clitics, we propose that possessive clitics are base-generated in-situ within a (silent) big-DP where they are assigned the respective genitive case, before moving to a higher position. Inalienable possessive clitics are therefore born in the complement position of n, where they are assigned unmarked inalienable genitive case (compare Sect. 2.2.2; cf. Baker 2015; Myler 2016; Alexiadou 2001). Parallel to the verbal domain, possessive clitics as defective elements move to D (Alexiadou et al. 2007; Cardinaletti 1998).37

• (80)

Alienable possessive clitics are introduced in Spec, PossP, where they are assigned inherent alienable genitive case by Poss before moving to D.

• (81)

The (in)alienability split of possessive clitics therefore reflects the distinct syntactic positions in which possessive clitics enter the derivation: While clitics bearing inalienable genitive case are base-generated in the complement position of n, clitics bearing alienable genitive case are base-generated in Spec, PossP.

Possessive clitics in nominalizations

In bare nominalizations, postverbal independent pronouns exhibit the same tripartite-inactive alignment as regular DPs with unaccusative subjects (82)a and objects being marked by inalienable genitive, unergative subjects by alienable genitive case (82)b, and transitive subjects by the ergative (82)c.

• (82)

As expected, speakers generally prefer the use of possessive clitics in bare nominalizations, obligatorily referring to the subject of the nominalized predicate (cf. Mosel and Hovdhaugen 1992: 549ff, Mosel 1992). Instead of tripartite-inactive alignment of regular DPs and independent pronouns in bare nominalizations and the neutralized alignment outside of nominalizations, possessive clitics exhibit a crosslinguistically rare marked unergative alignment (again not mentioned by Comrie 2013): While transitive subject clitics (83)c receive inalienable genitive case like unaccusative subject clitics (83)a, unergative subject clitics are distinctly marked by alienable genitive case (83)b. Nominalized transitive verbs therefore exhibit a double inalienable genitive alignment (83)c.

• (83)

Building on our analysis of regular DP arguments, we take unaccusative subject clitics (83)a to be base-generated within a big-DP in vP-internal position, where they get unmarked inalienable genitive case before moving to D.

• (84)

While unaccusative subject clitics obey the unaccusative requirement on nominalizations, unergative subject clitics (83)b must be introduced outside of the verbal domain, namely as possessors in Spec, PossP in the nominal domain. In this position, they receive inherent alienable genitive case as part of a big-DP and move to D.

• (85)

For transitive subject clitics (83)c, we note that parallel to the clausal domain, inalienable marking cooccurs with ina in postverbal subject position. Applying our nonuniform analysis of subject clitics to nominalizations, transitive subject clitics are merged outside of the verbal domain directly to D, where they receive unmarked inalienable genitive case, in local configuration with D (cf. Baker 2015). Being base-generated in an athematic position, prepositional ina resumes the transitive subject clitic in Spec, VoiceP, in line with the unaccusative requirement. Objects likewise receive unmarked inalienable genitive case, resulting in a double inalienable genitive alignment.38

• (86)

The distinct marking of unergative subjects in Samoan bare nominalizations therefore follows naturally from a combination of the unaccusative requirement on nominalizations, split-(in)alienability marking of possessors, and a nonuniform syntax of subject clitics in a prepositional ergative language like Samoan.

Agents vs. possessors

As unergative and transitive subjects are assigned distinct thematic roles, i.e., agent in Spec, VoiceP and possessor in Spec, PossP, we predict interpretative differences between the two types of subject, since the possessor relation has been argued to be generally underspecified, allowing for various pragmatically salient interpretations (cf. Harley 2009: 324, Alexiadou 2001; Barker 1995). This prediction is borne out by the data: Alienable genitive marking on unergative subjects gives rise to additional readings, including an ownership reading, either associated with a specific way in which the event is performed or a recording of the event (which for some speakers is the reading most readily available; cf. Hooper 2000, 1996 on nominalizations in closely related Tokelauan).

• (87)

Inalienable transitive subjects crucially lack such readings and must instead be interpreted as event participants, i.e., agents, as a consequence of thematic linking via ina resumption (88)a. The same holds for unaccusative subjects which obligatorily receive a patient interpretation. To express an ownership reading with transitive nominalizations, the object must undergo pseudo noun incorporation (88)b, which then requires the subject to be merged in Spec, PossP to satisfy the unaccusative requirement, as illustrated by the contrast below.

• (88)

As transitive subject clitics receive their agent role already in the verbal domain via ina resumption, they are blocked from being merged in another theta position such as PossP, violating the theta criterion (Chomsky 1981). D as an athematic clitic-hosting position is therefore the only option for transitive subject clitics to be introduced, ruling out alienable genitive case on transitive subject clitics. The observed interpretative differences between inalienable genitive and alienable genitive possessors in nominalizations therefore provide independent support for our assumption that inalienable genitive case is the unmarked case in the nominal domain that is not associated with a specific theta role, whereas alienable genitive case is inherently assigned to alienable possessors.

Variable intransitivity

Our discussion of the distribution of genitive cases in Samoan nominalizations so far suggests that genitive case assignment on intransitive subjects is categorial, with unergative subjects being marked by alienable and unaccusative subjects by inalienable genitive case. However, as already noted by Chung (1978) and Mosel (1992), genitive case can alternate on intransitive subjects. In the following, we demonstrate that these alternations are semantically meaningful and reflect distinct syntactic positions in which intransitive subjects can be merged (cf. Neu 2023; Tollan 2018; Massam 2009; Sorace 2000), which further supports our analysis.

Alienable genitive case on unaccusative subjects

Although unaccusative subjects are usually marked by inalienable genitive case, they can also receive alienable genitive case. Crucially, the case alternation has a semantic effect on the interpretation of the genitive argument: While inalienable-genitively marked arguments are obligatorily interpreted as event participants, i.e., patients (89)a, alienable-genitively marked arguments are interpreted as possessors (89)b.

• (89)

Parallel to unergative subjects (see Sect. 3.2.3), alienable-genitive marking therefore gives rise to a more flexible possessor relation, in which the genitive argument can be also interpreted as the owner of the event, e.g., a record of a falling event on social media, or as the creator of a specific manner/method of falling, e.g., in high diving. In addition to the verb-internal position, nominalizations of typically unaccusative verbs thus allow their internal argument to be merged in Spec, PossP in the nominal domain, parallel to unergative verbs.39

• (90)

The variation of genitive marking in Samoan nominalizations thereby resembles variation in unaccusativity diagnostics, observed crosslinguistically, e.g., the choice of perfect auxiliaries in Italian (Sorace 2000). In general, alienable genitive case on the sole argument of otherwise unaccusative verbs is rather rare and restricted to dynamic unaccusative verbs that potentially allow for agentive interpretations (in certain contexts). Stative unaccusative verbs like matua ‘be.tall’, which otherwise receive inalienable-genitive case, cannot occur with alienable genitive case at all (cf. Sorace and Shomura 2001 for similar restrictions on variable intransitivity in Japanese).

Inalienable genitive case on unergative subjects

Inalienable genitive case on unergative subjects is more common and primarily observed in two contexts (Mosel 1992): On the one hand, unergative subjects receive inalienable genitive case, when additional material intervenes in between the nominalizer n and VoiceP. This holds for unaccusative restructuring verbs like fia ‘be wanted’ (91) (cf. Collins 2017),40 the negative particle lē ‘be not’ (91) (cf. Mosel and Hovdhaugen 1992: 478, Mayer 1976: 344f), as well as aspectual particles like preverbal toe ‘again’ (92) and postverbal so‘o ‘frequently’ (93).

• (91)

• (92)

• (93)

As the presence of intervening material, such as an AspP and a NegP breaks the selectional relationship between the nominalizer and the embedded verb, we expect that the unaccusativity constraint is lifted in such contexts, and VoiceP can introduce an external argument, as observed in the data. This also includes unaccusative restructuring verbs like fia ‘want’ (94), which independently satisfy the unaccusativity constraint.

• (94)

On the other hand, if unergative subjects receive inalienable genitive case in the absence of unaccusative restructuring verbs, Mosel and Hovdhaugen (1992) note that they are perceived as less agentive and less deliberate (95)b than their alienable-genitively marked counterparts (95)a (also Mosel 1992, cf. Moyse-Faurie 2000; Chung 1973 on the closely related Polynesian language Faka’uvea and Pukapukan, respectively). Notably, the genitive alternation also has an effect on the morphosyntactic properties of unergative nominalizations, as inalienable genitive case on subjects makes plural agreement available (95)b, which is otherwise ungrammatical on unergative subjects marked by alienable genitive case (see Sect. 2.4.2).41

• (95)

We take this variation as further indication that intransitive verbs do not necessarily split neatly into an unergative and an unaccusative class in Samoan, but allow their subjects to be merged either vP-internally or vP-externally: While high agentive intransitive subjects are located in Spec, VoiceP like transitive subjects, low agentive intransitive subjects are introduced within the vP (cf. Neu 2023, also Kouneli 2021; Tollan 2018; Massam 2009 for related intuitions). Consequently, low agentive subjects of otherwise unergative verbs do not violate the unaccusative requirement on nominalizations (96). In this context, unergative subjects therefore receive unmarked inalienable genitive case instead of alienable genitive case.42

• (96)

Samoan genitive case alternations therefore provide further crosslinguistic support against a categorial split of intransitive verbs into unaccusative and unergative types, as intransitive subjects can be variable, being introduced within or outside the vP, as previously observed for languages like Turkish or Italian (Neu 2023; Sorace 2000), though with interpretative consequences.

To summarize, we derived both the tripartite-inactive and the marked unergative alignment of Samon bare nominalizations from a language-specific combination of crosslinguistically established phenomena, i.e., split and variable intransitivity, prepositional ergativity, the unaccusative requirement on nominalizations, and the nonuniform nature of (subject) clitics. Our investigation therefore does not only provide additional support for each of those phenomena but also highlights the special status of unergative subjects in Samoan.

On the source of ergativity

In closing, we briefly address further implications of our findings for the source of ergativity in Samoan and other Polynesian language based on the distribution of ergative morphology in nominalizations across languages. In particular, we demonstrate that inherent and dependent case approaches cannot readily explain the alignment types found in Samoan nominalizations. The prepositional analysis is also further supported by the distribution of the passive by-phrase in nominalizations of Polynesian languages with an accusative alignment, like Māori (Pearce 1997; Waite 1994). Contrasting the distribution of ergative marking in Samoan and Niuean nominalizations (Massam 2001), we then provide further evidence for a split between syntactically and morphologically ergative languages in Polynesian sensitive to the syntactic category of the ergative subject (Polinsky 2016).

Against alternative analyses of ergativity in Samoan

While we have adopted a prepositional analysis of (syntactic) ergativity in Samoan (Polinsky 2016), there are previous accounts of the Samoan case system that treat ergative case as either an inherent case assigned by Voice (Tollan 2018), or as a dependent case, assigned to the highest DP in the clausal domain (Collins 2021). We now demonstrate that both approaches fail to account for the distribution of ergative and alienable genitive case in bare nominalizations, at least without further assumptions.

In dependent case approaches, ergative case is assigned to the structurally higher DP in the clausal domain, mirroring the assignment of accusative case in languages like English (Baker 2015; McFadden 2004; Marantz 1991inter alia). For Samoan, a version of dependent ergative case is adopted by Collins (2021) (see also Collins 2017, 2014). To account for syntactic ergativity, one may adopt a case discrimination account as proposed by Deal (2017) and Otsuka (2010), according to which ergative extraction is blocked, due to the sensitivity of A’-movement to morphological case (cf. Drummond 2023 on the Polynesian language Nukuoro).

• (97)

In contrast, inherent case approaches treat ergative case as being inherently assigned to the DP argument in the specifier of a transitive Voice head (Legate 2008; Aldridge 2004; Woolford 1997inter alia). Adopting such an inherent case approach for Samoan, Tollan and Massam (2022) argue that syntactic ergativity follows from high absolutive case in the language, which requires the object to move across the ergative subject to the edge of the VP to get absolutive case from T (also Tollan 2018, cf. Clemens and Tollan 2021; Coon et al. 2014). As a result, the object argument freezes the ergative subject in its base-generated position for further (A’-)movement (see also Hopperdietzel 2022; Collins 2017 for independent motivation for object movement in Samoan).43

• (98)

A crucial difference between the prepositional and both inherent and dependent case approaches to ergativity is the syntactic category of the ergative subject. While it is a PP-adjunct to VoiceP in the former, it is a DP argument situated in Spec, VoiceP in the latter. Inherent and dependent case approaches therefore predict the ergative subject to violate the unaccusative requirement on nominalizations and to pattern with unergative subjects, i.e., marked by alienable genitive case in Samoan nominalizations, contrary to the facts.44

• (99)

To save a DP-ergative analysis, one may assume that the unaccusative requirement does not hold for Samoan bare nominalizations (see Imanishi 2020 for a parametrized view on Mayan languages). However, without the unaccusative requirement, inherent alienable genitive case on unergative subjects is left unexplained, as unergative subjects would be expected to be introduced in Spec, VoiceP in the complement domain of n, receiving unmarked inalienable genitive case instead, again contrary to the facts.

• (100)

Therefore, inherent and dependent case approaches to Samoan ergativity face the same challenges as they are unable to account for (alienable genitively) marked unergative subjects in Samoan nominalizations without further assumptions.45 The prepositional account instead naturally derives the distribution of ergative and alienable genitive case nominalizations and links it to more general properties of the Samoan grammar, e.g., ina resumption for extracted subjects in the context of topicalized and clitisized subjects.

Prepositional ergativity and passives: Māori

Crosslinguistic support for a prepositional analysis of Samoan ergativity comes from the related Eastern Polynesian language Māori (Harlow 2007). Unlike Samoan, Māori is described to exhibit an accusative-like alignment with intransitive and transitive subjects appearing in nominative case (marked by a for personal names) and objects in oblique case.

• (101)

In addition, Māori has a passive construction marked by the passivizing suffix -Cia where the former subject is demoted to an oblique PP marked by e and the former object is promoted to the nominative subject. Note that the preposition e in the Māori passive is homophonous with the ergative marker e in Samoan.

• (102)

Similar to Samoan, Māori exhibits a type of bare nominalizations in which nominative case becomes unavailable (Pearce 1999; Waite 1994). In such nominalizations, unergative subjects are marked by inherent alienable (103)b and unaccuative subjects (103)a and objects by unmarked inalienable genitive case (see also Harlow 2007).

• (103)

As expected from our investigation of Samoan nominalizations, transitive subjects do not receive inalienable genitive case. Instead, they can be realized as oblique PPs of nominalized passivized verbs, resembling passive by-phrases in Greek and ergative PPs in Samoan.46

• (104)

The passivization of transitive verbs together with the unavailability of inalienable genitive on transitive subjects in Māori nominalizations can again be explained by the unaccusative requirement, which prohibits the transitive subject to merge as a DP-argument in Spec, VoiceP (105). Unlike prepositional ergative languages like Samoan, accusative languages like Māori reflects this restriction morphologically with passive morphology on the verb.

• (105)

Crucially, in the Polynesian literature, passive by-phrases in Eastern Polynesian languages and ergative PPs in Western Polynesian languages have long been assumed to be diachronically related, providing a potential pathway for prepositional ergativity in Polynesian languages (cf. Chung 1978; Hohepa 1969).

Morphological vs. syntactic ergativity: Niuean

In Sect. 4.1, we have argued that due to the unaccusativity restriction, ergative case is only expected to survive nominalization, if it is a prepositional case. Crucially, the difference between syntactic and mere morphological ergativity has been linked to the syntactic category of the ergative subject, i.e., PP- vs DP-ergatives (Polinsky 2016). For morphological ergative languages, we therefore predict the absence of ergative marking in nominalizations, as ergative DPs violate the unaccusative restriction. This prediction is borne out by Niuean.

In contrast to Samoan, Niuean (Polynesian) does not exhibit an extraction restriction on ergative subjects (Tollan and Massam 2022; Clemens and Tollan 2021; Polinsky 2016), as both absolutive and ergative arguments are topicalized without resumption (Massam 2020; Seiter 1980).47

• (106)

Consequently, Niuean ergativity can be classified as purely morphological. We therefore assume that ergative subjects are DP arguments merged in the specifier of transitive Voice with ergative case inherently or dependently assigned (Tollan and Massam 2022; Clemens and Tollan 2021; Massam 2020).48

• (107)

Unlike Māori however, Niuean lacks passive voice, i.e., it is unable to introduce the external argument as a PP. As is typical for many ergative languages (Dixon 1994), Niuean instead exhibits an antipassive construction to detransitivize transitive verbs, especially in imperfective contexts (Massam 2020, cf. Polinsky 2017 on the limited productivity of antipassives). In this construction, the former absolutive DP object is realized by an oblique locative PP with the subject in absolutive case.

• (108)

In contrast to passives in accusative languages like Greek or Māori, detransitivized verbs in Niuean are thus more similar to unergative than to unaccusative predicates, i.e., their Spec, VoiceP position is filled by a DP argument.

• (109)

Like other Polynesian languages, Niuean exhibits a type of bare nominalizations, which has been proposed to exhibit selectional properties similar to their Samoan (and Greek) counterparts (Massam 2000, cf. Massam 2020 for different types of Niuean bare nominalizations).49 In contrast to Samoan however, ergative marking becomes unavailable in such nominalizations.50

• (110)

Instead, intransitive subjects (111)a and transitive subjects (111)b receive genitive case, whereas objects are introduced as PPs. Unlike Samoan and Māori, Niuean lacks split-alienability and assigns just a single genitive case, resulting in an accusative-type alignment in bare nominalizations (Massam 2000, but see Massam and Sperlich 2000 for a more complex picture).

• (111)

We can therefore attribute the variation between Samoan and Niuean to the interaction of the unaccusative requirement on nominalizations and the categorial status of transitive subjects. As ergative-marked transitive subjects are merged as DPs, they violate the unaccusative requirement. In the absence of a passive construction, transitive subjects must be merged as possessors in Spec, PossP in the nominal domain, where they receive unmarked genitive case (cf. Massam 2020). Due to the limited case licensing properties in Niuean nominals, we suggest that the internal argument must be introduced within a (locative) PP, akin to antipassive constructions above.

• (112)

The absence of ergative case in Niuean nominalization therefore provides independent support for our hypothesis that only prepositional ergative case survives nominalization, whereas inherent/dependent ergative case does not, as it violates the unaccusative restriction on nominalizations. Consequently, nominalizations may be used as a diagnostic to identify the categorial status of ergative subjects, i.e., PP vs. DP, crosslinguistically.

Conclusion

In this paper, we investigated the argument structure of Samoan bare nominalizations which show a crosslinguistically rare tripartite-inactive alignment of regular nouns that distinguishes unaccusative, unergative, and transitive subjects with an additional marked unergative alignment of subject clitics that singles out unergative subjects (Mosel 1992). By examining the morphosyntactic and semantic properties of bare nominalizations, we showed that these alignments follow naturally from a language-specific combination of independently established phenomena, including split intransitivity, split (in)alienability, prepositional syntactic ergativity, and a nonuniform status of subject clitics, for each of which our study thereby provides novel crosslinguistic support.

On the one hand, adopting a structural account of (in)alienablity, we identified alienable genitive case as an inherent case assigned to the DP in Spec, PossP, whereas inalienable genitive case represents the unmarked case in the nominal domain. The restriction of alienable genitive case to unergative subjects thus indicates that only unergative subjects are merged as possessors in the nominal domain, whereas all other arguments are introduced as event participants in the verbal domain. Classifying Samoan bare nominalizations as VoiceP-embedding n-based nominalizations, we explained the special status of unergative subjects by the unaccusative requirement on nominalization (Alexiadou 2001), according to which DPs are banned from Spec, VoiceP under n. Analyzing ergative marking as an inherent prepositional case in Samoan (cf. Polinsky 2016), transitive subjects satisfy this requirement, and ergative marking survives nominalization. In the absence of absolutive case, unaccusative subjects and objects receive unmarked inalienable genitive case, resulting in the tripartite-inactive case.

On the other hand, based on the prepositional nature of ergative subjects, we developed a nonuniform analysis of Samoan subject clitics (cf. Bleam 2000): Intransitive subject clitics are base-generated within a big-DP in their original vP-internal argument position, where they are assigned absolutive case before moving to T; transitive subject clitics are merged directly to T instead and require ina-resumption. In this position, transitive subject clitics receive unmarked absolutive case, resulting in a neutralized alignment. To satisfy the unaccusative requirement, only unergative subject clitics must be introduced in Spec, PossP, as transitive subject clitics can be merged directly to D. As a result, only unergative subjects are introduced as possessors of the nominalized event.

Our findings have broader theoretical implications: Firstly, the observation that the unaccusativity restriction on nominalizations holds in languages of typologically unrelated families, including Indo-European (Alexiadou 2001), Mayan (Imanishi 2020), and Polynesian, highlights the tight relation between nominalizations and unaccusativity. This restriction therefore calls for further examination across language types to gain a better understanding of its nature. Secondly, it supports a prepositional analysis of syntactic ergativity in Samoan, as analyses based on inherent or dependent ergative case cannot readily explain the case alignment of bare nominalizations. Thirdly, comparing the distribution of ergative case in nominalizations across ergative Polynesian languages, only PP-ergatives survive nominalizations (also Polinsky 2016 on Tongan), whereas DP-ergatives, as in the morphologically ergative language Niuean, violate the unaccusative requirement and are therefore ungrammatical in nominalizations (Massam 2020). However, the unaccusative requirement does not necessarily align with syntactic ergativity, since some syntactically ergative languages like Kaqchikel (Mayan) reject ergative case in nominalizations (Imanishi 2020, cf. Burukina 2023, 2021 for further discussion). This suggests that syntactic ergativity cannot be reduced to the categorial status of the ergative subject, PP vs. DP (pace Polinsky 2016) and therefore challenges a uniform crosslinguistic analysis (cf. Coon et al. 2014 for a related intuition).

Funding Information

Jens Hopperdietzel acknowledges that this paper was funded by ERC-2017-COG769192 to Andrew Koontz-Garboden. Artemis Alexiadou acknowledges that this paper is part of a project that has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (Grant agreement No. 856421).

Declarations

Competing Interests

Not applicable.