It has been argued that English agent nominalizations with the morpheme -er, as in (1), are deprived in syntactic structure (Baker & Vinokurova 2009; Alexiadou & Schäfer 2010). In (1), a tense marker cannot be realized:

The morpheme -er in English is taken to be the head N that selects a bare VP without an external argument (EA) (Baker & Vinokurova 2009). Baker & Vinokurova’s take on the derivation for (2a) is provided in (2b).1

Baker & Vinokurova (2009) argue that there is a type of individual-denoting nominalizations in Gĩkũyũ that is syntactically more articulated in size than agent nominalizations. In addition to transitive and unergative predicates, non-agentive unaccusative predicates are allowed in this type of nominalizations. Examples in (3) are referred to as subject nominalizations, according to Baker & Vinokurova.

For clarity, Figure 1 illustrates the relation between agent nominalizations and subject nominalizations. Agent nominalizations belong to subject nominalizations. Agent nominalizations denote agentive entities (e.g., a dance-er), while subject nominalizations denote either agentive or non-agentive entities (e.g., a dance-er or a die-er (one who dies)). Hence, the former is a proper subset of the latter.

Quite notably, a reflexive anaphor can participate in Gĩkũyũ subject nominalizations, as shown in (4), which is from Baker & Vinokurova (2009: 548) (see also Mugane 1997).

Baker & Vinokurova claim that the EA, PRO, is introduced in the derivation as the antecedent that syntactically binds the reflexive anaphor ĩ. PRO is represented in the tree schematized in (6).

Gĩkũyũ subject nominalizations additionally allow low adverbs, such as ũũru ‘badly,’ as shown in (5) from Baker & Vinokurova (2009: 547) (see also Mugane 1997). This suggests that there are similarities between the clausal domain and the nominal domain in terms of adverbial modification.

Baker & Vinokurova’s analysis of the subject nominalization in (5) is fleshed out in (6). Under their view, the suffix -i is the nominalizer that caps off VoiceP in subject nominalizations. In (6), the syntactic projection c-selected by the ‘aspectual (Asp)’ nominalizer cannot be bigger than VoiceP.3 PRO is the subject, and it is co-referential with the nominalizer -i and the entire NP.

Based on their survey of 78 languages, Baker & Vinokurova suspect that subject nominalizations are rare. They also do not report evidence for clausal properties (e.g., tense) inside subject nominalizations. We show that Oshiwambo showcases subject nominalizations that can be associated with a past tense reading, indicating that the size of the nominal is as big as TP. An implication of this work is that subject nominalizations can be more clause-like than previously assumed in the literature. We also argue that an extended version of the phrasal layering analysis (Alexiadou & Schäfer 2010) captures the facts about Oshiwambo subject nominalizations. Alternative analyses, including the complex head analysis of nominalizations (Wood 2023), do not seem to be applicable here.

The organization of this work is as follows: Section 2 lays out the basic facts about Oshiwambo. Section 3 shows that various types of argument structure and tense can be realized inside subject nominalizations. Section 4 fleshes out the analysis. Section 5 demonstrates that subject nominalizations are different from reduced and headless relative clauses in Oshiwambo. Section 6 concludes.

Oshiwambo is a southern Bantu language of the Niger-Congo family spoken in the northern part of Namibia and the southern part of Angola. The canonical word order is SVO. Similar to other Bantu languages, Oshiwambo displays a number of noun class prefixes. Noun classes 1 and 2 are often used with human-denoting nouns. Noun class 1 is used with singular-denoting entities, and noun class 2 with plural-denoting entities. In addition to noun class prefixes, an augment (aug) prefix is introduced in Oshiwambo noun phrases. We adopt the view that the augment in Bantu is a determiner (D of DP) (see De Dreu 2008; Visser 2008; Carstens & Mletshe 2016; Gambarage 2019, among others). Of the various dialects in Oshiwambo, we focus on Oshingandjera. Oshingandjera is spoken by approximately 42,000 people as their first language. The majority of Oshingandjera speakers live in the Ongandjera district in the northern part of Namibia. The examples presented in this work are confirmed by one of the authors of this paper, who is a native speaker of Oshingandjera. The grammaticality of the examples is additionally confirmed by three other native speakers of the dialect.

(7) shows how simple nouns are realized in Oshiwambo. The noun class 1 prefix, mu-, denotes singular entities, as shown in (7a). The noun class 2 prefix, a-, denotes plural entities, as shown in (7b). The aug prefix undergoes allomorphy depending on the number of the nominal entity. In cases where the entity is singular-denoting as in (7a), aug is spelled out as o-. In cases where the entity is plural-denoting as in (7b), aug is spelled out as a-.

In many cases, agentive nominals host noun classes 1 and 2 since these noun phrases are often human-denoting. Agentive nominals are also realized with the suffix -i. In (8), the suffix -i is realized together with the augment and the noun class prefix. ‘aug-1-V-i’ and ‘aug-2-V-i’ are productively used for Oshiwambo agent nominalizations.

The pattern observed in (8) extends to subject nominalizations. In Section 3, we will see that the suffix -a, in addition to -i, can be realized in this type of nominalizations. An anonymous reviewer asks whether other suffixal forms are attested in Oshiwambo nominalizations. While this paper primarily focuses on how the -i∼-a alternation relates to tense in individual-denoting nominalizations, we note that event-denoting nominalizations can host other suffixal forms. Event-denoting nominalizations can be expressed using the suffix -o, for instance, as shown in (9). Contrast this with the individual-denoting nominalization in (10).

(10) is the kind of nominalization we will attend to in this paper.

Subject nominalizations in Oshiwambo are realized with a noun class prefix and the suffix -i or -a, as will be discussed in Section 3.4. In fact, this is attested in many Bantu languages (see Mugane 1997; Krüger & Pretorius 2006; Mletshe 2010; Lee 2024). Oshiwambo subject nominalizations can host a wide range of predicates, including unaccusative and passive predicates. Reflexive, reciprocal, applicative, and causative markers can also be showcased inside Oshiwambo subject nominalizations. Quite importantly, tense can be realized in subject nominalizations as well. This suggests that Oshiwambo subject nominalizations can host the following syntactic phrases:

In what follows, we provide empirical evidence supporting the presence of these phrases inside subject nominalizations.

We adopt Alexiadou & Schäfer’s (2010) phrasal layering analysis to account for the data in Oshiwambo. The tree structures schematized in (34) provide our analysis of how subject nominalizations in (27) are derived. V-to-T movement derives dheng-w-a in (27a) and dheng-w-i in (27b) based on Baker’s (1985) Mirror Principle.8 Based on the idea that nominalizers trigger nominal concord (see Fuchs & van der Wal 2022 for Bantu nominal concord, among others), we assume that noun class prefixes behave like nominalizers in Oshiwambo. In (33), we see that the noun class prefix triggers nominal concord even in the absence of the suffix -i or -a.

Moreover, noun classes 1 and 2 give rise to singular- and plural-denoting nouns. This implies that noun classes 1 and 2 serve the role of individualizing nouns, making the nouns countable. Note that individuation is a unique property of nouns that can be achieved even in the absence of the suffixes -i and -a, as shown in (7) and (33a). For this reason, we posit that noun classes 1 and 2 are n (nominalizer) of nP, as shown in (34). We have yet to be convinced whether the suffixes -i and -a should also be treated as n. If the noun classes and the suffixes are both treated as a nominalizer, it is possible that Oshiwambo subject nominalizations involve a stacked-nP structure (see Kramer 2015; Lee & Lee 2019; Fuchs & van der Wal 2022; Wood 2023). For now, we posit that -i and -a are spelled out in T since the alternation between the two suffixes is determined by T’s past vs. non-past status in Oshiwambo.

In (34), we assume that PRO is the subject argument, which moves to Spec,TP due to passivization.9 The idea that PRO is present in the derivation accords with Baker & Vinokurova’s (2009) take on Gĩkũyũ subject nominalizations. An anonymous reviewer asks whether an overt nominal argument can replace PRO in (34). This is not possible. We assume that an overt nominal argument needs to be licensed in syntax. Chomsky (2008) proposes that phase heads such as C are the locus of all probing features. Under this view, the probing features on T, if there are any, must be supplied from C. Adapting Chomsky’s (2008) feature inheritance, we assume that Ts initially lack features necessary for licensing nominal arguments. Case could be one of them. Since C is absent in subject nominalizations, the probing features cannot be supplied to T. We argue that this is why the subject inside subject nominalizations cannot be realized as an overt nominal argument.10 This also explains why subject-verb agreement is absent in subject nominalizations. Contrast (12) and (13), for instance. Based on this analysis, we predict that CP-less structures, such as gerunds and infinitival clauses, do not exhibit subject-verb agreement in Oshiwambo. We confirm that this is indeed the case.

(35) schematizes the derivation for (8a), which hosts an active (agentive) verbal structure:

As mentioned in Section 3.4, we posit that past tense T selects a passive VoiceP under subject nominalizations. Non-past tense T, on the other hand, is more flexible in terms of what it can select. In any case, we assume that T inside Oshiwambo subject nominalizations is defective in that it only makes a past vs. non-past distinction. The notion of defective T has been posited for Turkish nominalizations (Kornfilt & Whitman 2011). Kornfilt & Whitman (2011) make a future vs. non-future distinction for Turkish defective Ts. These Ts are defective in that the tense distinction is binary. There is no specified form for past tense, for instance. Kornfilt & Whitman (2011) also point out that the lack of agreement signals the defectiveness of T. As briefly mentioned above, subject-verb agreement is never observed in Oshiwambo subject nominalizations. Once again, feature inheritance becomes relevant in this context. In the absence of C, T cannot be supplied with the necessary features from C to induce agreement. Unlike full-fledged clauses, subject nominalizations do not host C and therefore do not exhibit subject-verb agreement.

Based on our discussion so far, the analysis put forward by Baker & Vinokurova (2009) does not straightforwardly carry over to Oshiwambo. Evidenced by the fact that Oshiwambo subject nominalizations embed tense, subject nominalizations can be more extensive in size than (6). Wood (2023) convincingly argues that a deprived verbal structure that lacks phrasal projections such as VoiceP is associated with Icelandic agent nominalizations. In Oshiwambo subject nominalizations, the presence of the reflexive marker, applied arguments, by-phrases, low adverbs, and temporal adverbs calls for a more articulated structure that includes VoiceP and TP. Overall, the facts in Oshiwambo can be captured under an extended version of Alexiadou & Schäfer’s (2010) phrasal layering analysis where verbal functional projections, including TP, can be established inside the nominal domain.11

Subject nominalizations are different from relative clauses (RCs). RCs in Oshiwambo are postnominal, and the relativizer is overtly expressed, as shown in (36). Note that none of the previous examples showcasing Oshiwambo subject nominalizations contain a relativizer.

Subject nominalizations are not reduced RCs. Reduced RCs do not require a relativizer. Oshiwambo reduced RCs (if there is one) require an overt DP as their head noun, as shown in (37). This is not the case for subject nominalizations, as we have seen in the previous sections.

Also, note that an object inside reduced RCs can bear an augment, whereas an object inside subject nominalizations cannot. If subject nominalizations are taken to be reduced RCs, we would not expect this sort of discrepancy. The contrast between (31) and (37) demonstrates this point. It is not entirely clear why an object inside subject nominalizations does not bear an augment. While future work remains to be done on this issue, we emphasize that this is enough to disambiguate subject nominalizations from reduced RCs.

Subject nominalizations cannot be assimilated to headless RCs. Although the two nominals do not allow an overt DP as their head, they exhibit syntactic differences. (38) presents an example of a headless RC in Oshiwambo. Headless RCs require an overt relativizer, whereas subject nominalizations do not. Moreover, an object in a headless RC can bear an augment whereas an object in subject nominalizations cannot.

Table 1 summarizes the syntactic properties of subject nominalizations, reduced RCs, and headless RCs in Oshiwambo. Based on the empirical data provided thus far, we conclude that subject nominalizations cannot be treated on a par with reduced or headless RCs.

An implication of this work is that the syntactic size of subject nominalizations can be more articulated than previously assumed in the literature. Specifically, we have argued that TP can be showcased inside Oshiwambo subject nominalizations. This calls for a closer look at other (Bantu) languages to see if the findings from Oshiwambo can be replicated. Hopefully, this will shed further light on the typology of subject nominalizations. One aspect of Oshiwambo subject nominalizations that is worth investigating is the past tense reading that is only available with passivization. Future research remains to be done on why this should be the case.