2.1 Variation in event nominalizations

One of the major discoveries that has been made through the lens of the Functional Nominalization Thesis is that nominalization can apply to different verbal projections, with the upshot that there is substantial variation in the size of the verbal component of complex event nominalizations (Grimshaw 1990; cf. the term argument structure nominals in Borer 1999) across a range of nominalization types both within languages and across languages (Abney 1987, Picallo 1991, Borer 1993, Marantz 1997, Harley & Noyer 1998, Van Hout & Roeper 1998, Borsley & Kornfilt 2000, Alexiadou 2001, Fu et al. 2001, Alexiadou et al. 2010, Baker 2011, Kornfilt & Whitman 2011, Alexiadou 2020).

With respect to language-internal variation, English for instance makes use of two types of -ing nominalizations: nominal gerunds, as in (5), and verbal gerunds, as in (6) (Lees 1960, Abney 1987, Guéron & Hoekstra 1995, Kratzer 1996, Harley & Noyer 1998, Harley 2009b, Panagiotidis & Grohmann 2009, Alexiadou et al. 2010). Despite the apparent similarity of these constructions, it has been argued that the former is built on a VP core while the latter has a more articulated verbal structure containing v/Voice as well, explaining for example why accusative case is licensed only in the latter (Kratzer 1996; see also Harley 2009b and Alexiadou et al. 2010).

As for variation across languages, so-called nominal infinitives in European languages are a well-understood instance, having been argued to show different levels of verbal articulation within the nominal layer—for example, vP in Spanish, as illustrated by (7), versus AspP in German, as illustrated by (8)—with one diagnostic being whether the event denoted by the nominalization may be modified by an adjective (cf. Larson 1998) or an adverb, the latter requiring more verbal structure (Zucchi 1993, Alexiadou et al. 2010, Kornfilt & Whitman 2011, and Alexiadou 2020, among others).

More recently, attention has also been paid to the scope of possible verbal articulation within the domain of event nominalizations overall, with arguments being made that event nominalizations may contain, for example, TP in Turkish (Kornfilt & Whitman 2012; cf. Borsley & Kornfilt 2000), alongside even more striking cases of event nominalizations that contain fully articulated clauses, in languages such as Korean (Kim 2009), as illustrated by (9), and Wá·šiw (Hanink 2016).²

Such examples highlight the wide range of possible variation in the structural articulation of derived event nominals.

However, despite the breadth of literature comparing derived nominals both within and across languages, research on variation of this kind has largely focused on those that characterize events. It is nonetheless well-known that, beyond events, nominalizations may also characterize ordinary individuals—so-called entity nominalizations—which we turn to in the next sub-section.³

2.2 Variation in entity nominalizations

While the transformational era treated event and entity nominalizations alike, as involving transformations from an underlying sentence to a surface noun (Lees 1960, Bach 1968, and Vendler 1968; see Ntelitheos 2006 for discussion), this state of affairs changed starkly following Chomsky 1970. In this work, Chomsky decidedly treats entity nominalizations, of the -er type in particular, as fundamentally different from event nominalizations, uniformly assigning them lexical (as opposed to syntactic) status. (This is termed the “homogeneity hypothesis” by Roy & Soare 2020.)⁴ This perspective was later generalized to the event–entity distinction across languages (Comrie & Thompson 1985, Baker & Vinokurova 2009, Borer 2013), with Comrie & Thompson 1985 notably claiming that, while event nominalizations “retain certain properties of the verbs or adjectives they are related to,” entity nominalizations “… [bear] only morphological and (often unpredictable and idiosyncratic) semantic relations to the associated verb …” (p. 349). One lasting result of this perspective shift has been that entity nominalizations have played less of a role in this line of research (see also discussion in Van Hout & Roeper 1998, Ntelitheos 2006, Alexiadou & Schäfer 2008, Alexiadou & Schäfer 2010, Ntelitheos 2012, Roy & Soare 2014, Harley 2020, and Roy & Soare 2020).

Indeed, even contemporary work echoes the idea that entity nominalizations are fundamentally different from event nominalizations. Most notably, a seminal article on this topic, Baker & Vinokurova 2009, concludes on the basis of a cross-linguistic survey that a universal characteristic of -er type nominalizations in particular distinguishes them from event-characterizing nominalizations: while the latter can be formed out of rather large underlying verbal structures, the former avoid containing a “clause-like syntax” altogether. This fact, in Baker & Vinokurova’s analysis, is crucially tied up with the agentive specification of the -er nominalizer itself, which imposes limits on the structural articulation of the verbal component. A similar claim is reinforced recently in Alexiadou 2022, which proposes some limited variation (building on Alexiadou & Schäfer 2008, Alexiadou & Schäfer 2010, and Roy & Soare 2020) but holds that, while there is some evidence for variation in entity nominals, “we certainly do not have the type of variation that, for example, previous literature has argued for [in] derived nominals, i.e., variation with respect to the presence of Aspect, Tense and CP” (p. 18). Alexiadou makes the observation that, in this domain, we find either reduced nominals of the -er type or the use of a relative clause, with little variation in between. Against this backdrop, we will see in section 4 that Wá·šiw reduced nominalizations offer an interesting test case, as a type of entity nominalization that does not fall neatly into either of these categories.

2.3 Scope and outlook

Whether entity nominalizations as a whole are fundamentally different from event nominalizations remains an open question. One sub-component of this question is what the range of variation possible in the structure of entity nominals is and whether there are in fact cases of entity nominalizations that show more variation in structure than -er type nominals. Another is whether entity nominalizations characterize the kinds of entities they do because they are thematically specified for certain roles. This article argues for two main points that shed light on both of these aspects.

The first point, drawing on empirical evidence from Wá·šiw, is that entity nominalizations are actually more similar to event nominalizations with respect to structural variation than is often thought and can in fact show a “clause-like syntax,” in contrast to the -er nominalizations analyzed in Baker & Vinokurova 2009. The two types of nominalization under investigation are exemplified in (10, 11). The first type involves a fully articulated CP (cf. event nominalizations in Korean, (9)). The second type is structurally reduced but still allows a verbal component that projects up to AspP (cf. nominalized infinitives in German, (8)).

The second, broader point that emerges from this study of structural variation is the following. I will argue that neither type of nominalization illustrated in (10, 11) supports an analysis in which the nominalizer hardwires thematic information; this result supports the long-standing claim that syntax doesn’t make direct reference to such information (Burzio 1981, Fabb 1984, Keyser & Roeper 1984, Burzio 1986, Rappaport & Levin 1988, and Rappaport Hovav & Levin 1992). Instead, the kind of argument characterized in both cases is related solely to the type of dependency involved: the entity is “read off” the structure. While both of the nominalizations in (10, 11) can be understood as characterizing the “subject” of the underlying verb, I show that clausal nominalizations along the lines of (10) differ from reduced nominalizations like (11) in that they are able to characterize other types of arguments as well. As I will argue, this falls out from the type of dependency that derives the nominalization—non-local versus local binding of an open variable—in a way that is directly related to the amount of syntactic structure projected. Along the way, a picture also emerges in which the amount of verbal structure within the nominalization is correlated with the different properties of D-versus n-based nominalizations (in the sense of Alexiadou 2020 and Iordăchioaia 2020).

3.1 Clausal nominalizations/internally headed relatives

We begin with entity-characterizing clausal nominalizations such as the one in (10). We understand these constructions to be entity-characterizing in that the nominalization denotes an individual (not an event); in the case of (10), this individual is the notional subject of the nominalized verb. While the characterized argument in (10) is not overtly expressed with a nominal (translated instead with impersonal ‘one’), a striking characteristic of this nominalization type is that examples are equally well-formed when the characterized nominal is overtly specified, an uncommon state of affairs in entity nominalization. Compare for example (10) to (12), in which the subject of the embedded clause, t’é:liwhu ‘man,’ is overt. The entire nominalization thus constitutes an independent nominal while simultaneously characterizing one of the arguments it contains.

For this reason, this nominalization type is often called an internally headed (or head-internal) relative; Culy 1990, following Langdon 1977, describes this as a type of “nominalized sentence which modifies a nominal, overt or not, internal to the sentence” (p. 27). That is, despite the presence of an articulated clausal structure in these nominalizations (evidence for which is to follow), the resulting meaning characterizes an ordinary individual. Notably, it is likely simply because this construction is called a relative clause that it has been overlooked with respect to possible structural variation within entity nominalizations, despite the fact that event- or proposition-denoting CP nominalizations are now relatively well-known; as noted in Comrie & Thompson 1985 for instance, “in certain languages clausal nominalization is indistinct from relativization” (p. 395).⁵

In the remaining sub-sections of section 3, I review arguments that this type of nominalization contains a fully articulated clausal structure. I also show that this nominalization type can give rise to characterization of a range of entity types, not just subjects. Indeed, the “head-internal” characteristic of this construction is most clearly observed in cases where the characterized entity is the notional object, as in (13), in which the semantic head gó:beʔ ‘coffee’ remains in its typical SOV position within the nominalized clause.

As will be reviewed in more detail in section 3.3, this flexibility in characterization is a result of the binding potential of the nominalizing head common to clausal nominalizations—D—due to the quantificational nature of the binding dependency involved.

3.2 Mixed properties of clausal nominalizations

Entity-characterizing clausal nominalizations in Wá·šiw unsurprisingly display a mix of nominal and verbal properties (Peachey 2006, Hanink 2016, and Hanink 2021a). Beginning with their external properties, the following examples show that they follow a nominal distribution: they can act as the subject of the matrix verb, as in (14), or its object, as in (15), and they can also appear as the argument of a postposition, as in (16).

These nominalizations also exhibit case marking, with the nominalizer alternating for nominative (-gi) and accusative (-ge).⁶ This case distinction tracks whether the entire nominalization is the subject of the matrix verb (14) or not (15, 16). This is consistent with the fact that the same gi/ge is a third person pronoun when used independently and shows the same case alternation between nominative and accusative:⁷

Within the nominalization, however, further properties point to a fully articulated clausal structure. Table 1 offers a simplified verbal template for reference (see Bochnak et al. 2023 for further detail). Taking (19) and (20) as examples, we see that this nominalization type shows both tense and mood morphology in addition to negation and canonical verb agreement.

Clausal nominalizations moreover display switch reference morphology where appropriate, in the form of the different subject–marking suffix -š (Jacobsen 1967). This morpheme is obligatory when subjects across an embedded clause boundary are not co-referential, and it has been argued to be the realization of embedded C in Wá·šiw (Finer 1984, Arregi & Hanink 2018, and Arregi & Hanink 2022). (Same subject marking is null.)

Taken together, these properties point to the structure in (21) (Peachey 2006 and Hanink 2016) for an example such as (19), where word formation is the result of head movement (or similar). The nominalizer -gi/-ge is represented as D due to its otherwise pronominal status.⁸

3.3 Deriving entities with non-local dependencies

One of the analytical challenges posed by entity-characterizing clausal nominalizations is that they look propositional in terms of their fully specified clause structure but are translationally equivalent to externally headed relatives in that they characterize entities (see Hanink 2021a and references therein). For instance, (22a) is a canonical SOV main clause; the addition of the nominalizer (and switch reference, due to the addition of a clause boundary between two subjects with disjoint reference) in (22b) results in an entity interpretation.

Precisely how these nominalizations come to characterize individuals (rather than events or propositions) is therefore not immediately obvious.⁹

While some analyses of internally headed relatives assimilate them to externally headed relatives at LF (Cole 1987, Williamson 1987, Culy 1990, Cole & Hermon 1994, and Grosu 2009), Basilico 1996 argues, building on Jelinek 1987, Culy 1990, Srivastav 1990, Bonneau 1992, and Jelinek 1995, that they are “not cases of relativization semantically, but cases of quantification” (p. 507). This line of analysis, further developed for Wá·šiw in Hanink 2021a, contends that internally headed relatives are derived by binding rather than movement (Salish: Jelinek 1987; Diegueño: Basilico 1996; Lakhota: Grosu 2009; see Grosu 2012 and Jarvis 2023 for discussion).¹⁰ Motivation for an approach that does not adopt covert movement includes the fact that internally headed relative clauses in many languages are not sensitive to islands. The following demonstrates this island insensitivity for Wá·šiw.¹¹

More specifically, analyses along these lines derive internally headed relative clauses through binding of an in situ argument that is introduced as a Heimian indefinite, which consists of a variable and a restriction on that variable (Kamp 1981, Heim 1982).¹² This results in an embedded clause that denotes an open proposition containing a variable that must be bound, similar to binding-based analyses of wh in situ configurations that are likewise not island-sensitive (Aoun & Li 1993, Cole & Hermon 1994; after Huang 1982 and Pesetsky 1987).¹³

While the precise analysis is not crucial for the general point being made here, for the sake of concreteness I will sketch several core (but simplified) aspects of Hanink 2021a’s account in particular. First, (24) shows the representation of the semantic head of the relative clause for the example in (22b), introduced as a restricted variable, while (25) gives the meaning of the binder.¹⁴

The role of the binder is to take the open proposition denoted by the embedded clause, (26a), and replace any instance of g(i) with x, returning the characteristic function in (26b).

In this way, the results of index adjunction as in abstraction-based analyses are achieved without any recourse to movement (cf. Klecha & Martinović 2015).

While the semantic details of the binder vary across analyses (e.g., in Basilico 1996 it is an ι-operator, while in Hanink 2021a it is quantifier-like, as above), what is shared is that (i) the binding dependency is not strictly local, in that it is not necessarily the closest argument that must be bound, and (ii) the nominalizing component is responsible for binding. This is schematized in (27) for the example in (22b).¹⁵

That the binding dependency is not strictly local in internally headed relatives can be observed for example through the following minimal pair, in which the nominalization may come to characterize either ‘girl’ or ‘man’ (with different subject marking only appearing in (28a), as the subjects of the embedded and main clauses differ in this case).

This flexibility in characterization is the result of which noun in the embedded clause enters the derivation as a restricted variable, allowing it to be bound. Because the binding potential of D allows a non-local dependency, the internally headed relative clause is able to give rise to both subject and non-subject dependencies:

Indeed, flexibility in characterization unsurprisingly holds for D-based clausal nominalizations across languages.¹⁶ In (30, 31) are similar examples from Hidatsa (Siouan, United States; see also Williamson 1987 on Lakhota) and Mojave (Yuman, United States), in which both subject and object characterizations are observed.¹⁷

That both subjects and non-subjects may be characterized comes down to the fact that the binding relation may establish a dependency with either a local argument (the subject) or a non-local one (e.g., a non-subject).¹⁸ What the internally headed relative clause construction has in common across many languages, including Hidatsa, Mojave, and Wá·šiw, is (i) an articulated clausal structure and (ii) a determiner-like element at the clausal periphery responsible for nominalization. The importance of D as the nominalizer in these constructions is highlighted for instance in Basilico 1996. Basilico contends that the binding analysis of internally headed relative clauses is made possible by the fact that D is quantificational in the languages he discusses (see also Bonneau 1992 and Watanabe 2004). We return to further discussion of D as a clausal nominalizer in section 4.5.

3.4 Interim summary

We have seen in this section that entity nominalizations with a clause-like syntax can be found in the form of clausal nominalizations in a range of languages. Importantly, because D is able to bind both local and non-local open variables, internally headed relatives are able to characterize a range of arguments, both subjects and non-subjects:

With clear evidence of a clausal core, internally headed relatives offer concrete evidence that entity nominalizations show more variation than has been acknowledged—coming here in the form of full CPs. Moreover, by virtue of the fact that clausal nominalizations may characterize a range of entity types, it is immediately clear that an analysis in which the nominalizer is hardwired with thematic information is untenable. Clausal nominalizations are thus a straightforward test case arguing against a thematic analysis for entity nominalizations as a whole. In sum, not only are “clause-like” entity nominalizations indeed possible in an attested range of languages, but such syntactic nominalizations are clearly incompatible with a thematic analysis.

4.1 Reduced nominalizations

Let us examine structurally reduced nominalizations like those in (33, 34). These nominalizations are formed with the phonologically conditioned prefix t’-/d^e-, which Jacobsen 1964 calls the “third person nominal prefix” (p. 488).¹⁹

Jacobsen describes this prefix as being “added to transitive verb stems to nominalize them while expressing their object,” with Bochnak et al. 2011 later establishing that t’-/d^e- may also nominalize intransitive verbs.

Unlike with the clausal nominalizations discussed in section 3, we will see in the sub-sections that follow that the verbal component of this nominalization does not constitute a fully articulated clause, instead projecting up to AspP only (with the possibility of negation), thus constituting a second instance of entity nominalization in the language with a different verbal profile. Crucially, we will also see that this construction differs from CP nominalizations in that its entity characterization is limited to subjects only (e.g., the above ‘x heals people’ and ‘x burns houses’). A further difference from clausal nominalizations is that the subject in this nominalization type may not be overt:

In what follows, I argue that these facts again do not follow from a thematic analysis; instead they follow from the fact that the verbal component of these nominalizations is reduced in structure. This prohibits case licensing of an articulated subject, necessitating under-specified pro instead. A further, related property of these nominalizations is that they are nominalized by n. Unlike nominalization by D, nominalization by n allows only a local step of binding through λ-abstraction triggered by the nominalizer, and as a minimal pronoun (Kratzer 2009 and Landau 2014), the pro subject will always be the most local open variable available for binding. We begin in the next sub-section by reviewing the mixed nominal and verbal properties of this nominalization type.

4.2 Mixed properties of reduced nominalizations

The t’-/d^e- construction displays the mix of nominal and verbal properties that are expected from deverbal nominalization.

Beginning with nominal properties, we see again (cf. section 3.2) that the t’-/d^e- construction may appear in argument position: examples are given in (36–38) of t’-/d^e- nominals in subject position, in object position, and as the complement of a postposition.

Another diagnostic for the nominal status of this construction is language-internal: t’-/d^e- nominals occur with the set of copular agreement prefixes that appear with nominal predicates. This set of agreement prefixes differs from the canonical pre-vocalic agreement paradigm and is argued in Jacobsen 1977 to occur in equational contexts (p. 63; cf. Bochnak et al. 2011). This categorial contrast is shown in (39, 40), in which the nominal predicate šáwlamhu ‘girl’ triggers the prefix k’- in agreement, while the locational predicate dáŋala ‘in the house’ triggers canonical pre-vocalic third person agreement (∅).²⁰

The example in (41) shows that t’-/d^e- constructions pattern with nominal predicates in this regard.²¹

Turning to the verbal properties of this construction, we can see that the verbal component projects up to AspP, with the possibility of negation.

Morphological evidence that v is projected is, first, the fact that causative morphology (-ha) can appear (see also (33)):²²

In line with Burzio’s Generalization (Burzio 1986), further evidence that vP is present in the structure is the fact that accusative objects are licensed; (43) shows a subject nominalization with an accusative-marked third person pronominal object.

Further supporting the presence of vP, t’-/d^e- nominalizations permit adverbial modification in the form of verbal suffixes. The nominalization in (44) bears the suffix -áŋaw ‘well’; the one in (45) shows -wéwɨš ‘slightly.’

As adverbs have been argued to require (minimally) vP in the structure (Cinque 1999; see also Fu et al. 2001 in the context of deverbal nominalization), their possibility serves as additional evidence for the projection of this phrase within t’-/d^e- nominalizations.²³

Beyond vP, there is also evidence for AspP in the structure (cf. Harley 2020’s proposal for Hiaki; see (74)).

The first piece of evidence is the availability of functional restructuring predicates, based on the idea that these instantiate Asp heads in a monoclausal structure (Cinque 2001, Wurmbrand 2001, Cinque 2004, Wurmbrand 2004, Grano 2015). In Wá·šiw they are bound morphemes that encode aspectual information (see Hanink 2021b) and translate to predicates such as finish, start, and stop:

There is also (more limited) evidence for AspP from grammatical aspect, primarily from the suffix -giš, which is analyzed in Bochnak 2023 as the progressive:

Finally, negation is also allowed, as has already been seen in several examples above, including (48) and (42). This suggests that NegP is also permitted within the verbal component (see also Gotah & Lee 2024 on Ewe).²⁴

Despite this relatively articulated structure, t’-/d^e- nominalizations are structurally reduced in comparison to full clauses. Notably, they do not permit tense morphology or anything higher (i.e., mood marking or switch reference).²⁵ The examples in (50, 51) show that, for example, recent past marking and temporal adverbs are unavailable.

In examples such as (38), moreover, we do not observe switch reference even though the subject referent of the nominalization (the thing that’s big) is different from the subject of the main verb (the person clubbing).

4.3 Against a thematic analysis of t’-/d^e-

Unlike CP nominalizations, which can characterize a range of entity types, AspP nominalizations are restricted in that they characterize subjects only. That is, they may characterize the notional subjects of transitive and intransitive verbs but not direct or indirect objects:

As mentioned in section 2.2, a common proposal for subject nominalizations cross-linguistically is that the nominalizer itself supplies a thematic agent (or external argument; Baker & Vinokurova 2009, Alexiadou & Schäfer 2010, Roy & Soare 2014, and Alexiadou 2022).²⁶ Taking Baker & Vinokurova 2009’s proposal as an example, the subject nominalizer -er has the following meaning.

In this line of analysis, the nominalizer itself takes the place of the external argument, and upon composing with the verbal component of the nominalization it returns the set of agents of generic events of the relevant sort—in the example in (54), company-managing events.²⁷

In conjunction with this meaning for subject nominalizers, Baker & Vinokurova claim that, universally, the size of the verbal component in this type of nominalization is maximally VP (the idea being that, since the nominalizer takes the place of Voice/v, it must select for VP in its stead).

An updated version of Baker & Vinokurova’s analysis is proposed in Bochnak et al. 2011, for Wá·šiw subject nominalizations. Observing however that this nominalization type is also, for example, possible with stative meanings as in (55) (adjectival meanings are always expressed by (nominalized) verbs in Wá·šiw), Bochnak et al. argue that t’-/d^e- encodes an under-specified external argument role, as stated in (56), rather than one that is restricted to agents as in (53).²⁸

However, adopting the Baker & Vinokurova analysis for Wá·šiw faces challenges, beyond the evidence already laid out in the last sub-section that the verbal core is larger than VP (an issue also mentioned in Bochnak et al. 2011). In particular, it is evident that the t’-/d^e- prefix cannot have any external argument thematic role hardwired into its meaning, even one that’s under-specified, because t’-/d^e- nominals may also characterize non-external arguments, as I will now show.

We have already seen that t’-/d^e- nominals are able to characterize a range of subjects, including transitive subjects, as in (57), and unergative subjects, as in (58).

This nominalization type may however also characterize an unaccusative subject (i.e., the sole argument of an unaccusative verb), as in (59a) and (60a). That the verbal source is unaccusative in these examples is demonstrated by its participation in the inchoative–causative alternation (Parsons 1990, Levin & Rappaport Hovav 1995), as (59b, c) and (60b, c) illustrate.²⁹

The fact that the t’-/d^e- construction may characterize unaccusative subjects constitutes strong evidence against any analysis in which an external argument role is built directly into the nominalizer. I take this evidence to show that a thematic analysis of this type of nominalization is untenable: the interpretation of the subject must instead be flexible enough to account for underlying objects. (Given the t’-/d^e- construction’s ability to characterize derived subjects, we would also expect characterization of passive subjects to be possible. However, Wá·šiw lacks a passive; Jacobsen 1979.)

The t’-/d^e- construction in Wá·šiw is thus not a good candidate for an agent nominalization or even an external argument nominalization. Instead, this construction requires an analysis as an instance of true subject nominalizations, which Baker & Vinokurova 2009 suggests are attested but cross-linguistically rare (see also Lee & Ndapo 2025 for recent discussion). In the next sub-section we turn to how this subject characterization is derived.

4.4 Deriving entities with local dependencies

The aim of the previous sub-section was to demonstrate that Wá·šiw subject nominalizations cannot be analyzed as involving a thematic nominalizer. Where, then, does the “subject” characterization come from? In section 3, we saw that a non-local binding dependency in clausal nominalizations gives rise to flexibility in terms of the kind of entity characterized, with both subject and non-subject characterizations being possible. In what follows, I lay out how we can similarly understand the subject restriction in the reduced t’-/d^e- nominalizations as the result of a local dependency instead.

Recalling the discussion in the previous sub-section, a common proposal for subject nominalizations is that the nominalizer itself supplies the thematic subject relation, with the external argument position not syntactically projected at all (see (54a)). However, we saw arguments that a thematic treatment of subject nominalizations is not viable for Wá·šiw t’-/d^e- nominalizations, based on evidence from the range of permissible “subject” types. This state of affairs leaves open the possibility that the subject position is in fact syntactically projected but occupied by some under-specified argument.

Indeed, in addition to the evidence in section 4.2 for an articulated verbal structure, including the possibility of accusative case licensing (43), syntactic projection of the subject position in t’-/d^e- nominalizations is further supported by, for example, the availability of reflexives and reciprocals:³⁰

As pointed out by Baker & Vinokurova 2009, the availability of reflexives indicates that a syntactically projected subject must be present within the verbal component of the nominalization, to serve as a licit local antecedent. There is thus a tension here. On the one hand, we have syntactic evidence for an external argument (and indeed there is enough verbal structure that we would expect one). On the other hand, an overt subject may not appear in this construction (35). The “subject” in subject nominalizations must therefore instead be some under-specified, subject type argument.

Building on proposals by Baker & Vinokurova 2009 and Toosarvandani 2014 for subject nominalizations in Gĩkũyũ and Northern Paiute, respectively, I propose that this under-specified subject in Wá·šiw t’-/d^e- nominalizations is pro and that the structure of these nominalizations is as follows.

From the structure in (62), we can see that subject nominalizations, like the entity-characterizing clausal nominalizations discussed in section 3, might pose a semantic puzzle at first glance: the syntactic projection of a (pro) subject might reasonably be expected to result in a propositional meaning, given that the argument structure of the predicate appears to be complete. Here I follow Toosarvandani 2014, which argues for Northern Paiute subject nominalizations that pro is obligatorily bound, similarly aiming to explain why subject nominalizations in that language do not denote propositions.³¹ Toosarvandani suggests (in fn. 4, pp. 817–818) that this requirement for binding is imposed by D. Here I suggest that it is instead crucially imposed by the nominalizer n. I return to the relevance of the D–n distinction in the next sub-section.³²

Concretely, the composition of a subject nominalization such as (63) then proceeds as in (64).

In (64), the vP is formed with a pro subject saturating the external argument position. As a minimal pronoun (Kratzer 2009 and Landau 2014), however, pro is interpreted as a variable that is then bound following the step of λ-abstraction triggered by n, resulting in a property type meaning for the entire nominalization—in this case, the set of individuals that are holders of people-liking states. The nominalization is thus entity-characterizing, as desired.³³

Let us now return to the nature of the subject restriction. The crucial difference between subject/reduced nominalizations and clausal nominalizations is that the individual ranged over by the characteristic function denoted by the former (64b) may only be the outermost argument: in other words, the bound argument must correspond to the open position most local to n, explaining why non-subject characterizations are never possible. Why should this locality condition hold? We can understand it straightforwardly as a result of deficient verbal structure: in (64), there is no T to case-license a nominative subject, and so pro, interpreted as a variable, will always be the outermost open argument available for binding. Recalling moreover the discussion in section 3 of the role of the quantificational nature of D in the formation of internally headed relatives, which is ultimately what makes a long distance dependency possible in that construction, I now turn to the role of n in regard to binding. Advancements in our understanding of binding (Adger & Ramchand 2005 and Kratzer 2009, among others) have shown that binding may be achieved through functional heads.³⁴ For example, in the verbal domain, reflexivization is achieved through local binding triggered by the functional head v:

Kratzer 2009 points out that this analysis captures a crucial fact about reflexive pronouns: they constitute a class of bound variable pronouns that must be bound locally, by the closest v. This parallels what we find in the case of n-based subject nominalizations in Wá·šiw: if Toosarvandani 2014’s suggestion is on the right track, then n triggers a step of local λ-abstraction, just like v does in the verbal domain.³⁵

While I have thus far only compared D and n in terms of the way they enact binding of an open argument, recent work in the domain of event nominalizations has elucidated further differences between D and n in their role as nominalizers. I now turn to that discussion.

4.5 D- versus n-based nominalizations

We have now seen a contrast in the functional head that serves as the nominalizer: in CP nominalizations the nominalizer is D, while in reduced nominalizations it is n. The aim of this sub-section is to expand on the significance of that distinction in relation to recent advances in the classification of event nominalizations.

Alexiadou 2020 and Iordăchioaia 2020 identify an important distinction in event nominalizations: those that are nominalized by n (cf. De Belder 2011 and Borer 2013) and those that are nominalized by D directly. One of the differences between these two nominalizers is the size of the verbal component they may embed: nominalizations targeting TP or higher may only be nominalized by D directly, while nominalizations applying at the level of AspP or below are nominalized by n, as schematized in the following (adapted from Alexiadou 2020: 100).³⁶

These cut-off points are determined by the kinds of semantic domains (in the sense of Ramchand & Svenonius 2014 and Wiltschko 2014) n or D is able to nominalize (see also discussion in Alexiadou et al. 2007). Offering the schematic in (68), Alexiadou 2020 points out that, depending on the height of the nominalizer (D vs. n), more or fewer nominal projections should be able to co-occur with the nominalization. For instance, if the nominalizer is n, then the nominalization should be able to show gender as well as number distinctions. If it is nominalized by D directly, it should not.

D and n nominalizations are therefore expected to show different properties. One way they differ is with respect to determiner choice. Direct nominalization of CPs by D in, for example, Greek and Polish (see also Pietraszko 2019 on Ndebele) makes use of an invariant definite determiner, as in (69); this can be contrasted with the possibility of a range of determiners in German nominal infinitives like (70).³⁷

In the domain of entity nominalization, such differences between the two nominalization types are as of yet unstudied. However, what we see when we look at the Wá·šiw entity nominalizations addressed in this article is that the same distinction appears to hold in this domain as well.³⁸ In Wá·šiw, for example, I have treated reduced/subject nominalizations as n-based, which has testable predictions according to Alexiadou 2020’s classification. While Wá·šiw lacks gender marking, consistent with the claim that t’-/d^e- nominalizations are n-based is the fact that they (i) may show plural marking (achieved through reduplication; Jacobsen 1964, Yu 2005, Yu 2012) and be counted, as (71) shows, and (ii) are compatible with several types of determiners, as (72) shows.³⁹

What we observe then is that the distinction between D- and n-based nominalizations appears to apply to entity nominalizations as well, in a way that is consistent with the nature of the binding dependency involved. On the one hand, D nominalizes a complete clause and, being quantificational, allows non-local characterizations of any open variable within that clause. On the other hand, n may only nominalize reduced structures and merely triggers λ-abstraction.⁴⁰ For that reason, the binding dependency will always be local: it will always be the open argument closest to n that is bound.

Indeed, the correlation between n-based nominalization and local characterization (in contrast to D-based nominalizations) also appears to hold for the handful of other languages previously shown to have subject nominalizations with articulated verbal components: Gĩkũyũ (Baker & Vinokurova 2009; see also Mugane 1997, Mugane 2003, and Bresnan & Mugane 2006), Malagasy (Ntelitheos 2005, Ntelitheos 2006, and Ntelitheos 2012), Northern Paiute (Toosarvandani 2014), Hiaki (Álvarez González 2012 and Harley 2020), and Ewe (Gotah & Lee 2024). For instance, Baker & Vinokurova 2009 analyzes subject nominalizations in Gĩkũyũ, like the example in (73), as comprising a nominalizing n head atop vP; see also Toosarvandani 2014 on Northern Paiute. Harley 2020 moreover argues that subject nominalizations in Hiaki such as (74)—which, like their counterparts in Wá·šiw, have a slightly more articulated verbal structure consisting of AspP—likewise involve nominalization by n.

While I have focused so far on possible variation in the structure of entity nominalizations within a single language—Wá·šiw—what we find when we consider the subject nominalizations in (73, 74) is that there is a range of possible structural variation across languages as well. Not only do these entity nominalizations show gradience in terms of the amount of verbal structure they allow—no different from the gradience observed in event nominalizations—but they also appear to show the same D–n distinction as their event counterparts in terms of the amount of verbal structure that each of the two nominalizers may target.

Before concluding this sub-section, I note that one question that arises is whether the most local argument will always be the subject. While Wá·šiw lacks dedicated object nominalizations (i.e., nominalizations that may only characterize objects, as opposed to subjects), languages such as Hiaki and Northern Paiute do exhibit nominalizations of this type; (75) exemplifies complex object nominalizations in Northern Paiute.

In both Hiaki and Northern Paiute, the entity nominalizations in question may characterize only non-subjects. On the view that entity characterization is purely structural, this effect should fall out from argument-structural considerations only. Do objects still count as local? Indeed this is precisely the type of analysis that both Toosarvandani 2014 and Harley 2020 independently pursue. In both analyses, the external argument in object nominalizations must vacate to a position outside the nominalization, rendering the remaining object position as the outermost (most local to n) open position.⁴² The relevant factor making n-based object nominalizations possible therefore seems to be whether there is some trigger for the external argument to raise outside the nominalization. (In Wá·šiw, there is evidently not.)

4.6 Interim summary

There are several main takeaways from the discussion in section 4. First, in contrast to full clausal nominalizations, reduced entity nominalizations in Wá·šiw may characterize only the outermost open position of the verbal structure they contain. We have also seen that, despite their apparent “subject” flavor, these types of nominalizations cannot be analyzed thematically. Instead, this subject interpretation is solely “read off” the structure, as summarized in (76). Because the structure is reduced, only pro is licensed. A local step of λ-abstraction triggered by n then binds this pro and is not able to bind any open variable beyond this most local position.

These two factors are independent but confluential in the formation of this nominalization type.

Second, when taken together, the behavior of subject nominalizations across a range of languages constitutes emerging evidence that entity nominalizations show structural variation not only within a language but also from one language to another, with different levels of verbal structure attested. In this way, entity nominalizations are not unlike their event-denoting counterparts.

Finally, distinctions between D- and n-based nominalizations in the event domain were shown to carry over to entity nominalizations as well, further supporting the idea that these two types of nominalizations are not so dissimilar as has previously been claimed.