1.1 Language overview and data sources

The language discussed in this paper is Hawaiian, the native language of Hawai‘i, which is currently endangered (NeSmith 2016). Hawaiian is a verb-initial language with nominative/accusative case marking. The standard word order is VSO, with embedded clauses occurring in the same position as objects.

There are two primary dialects of Hawaiian in current use: the ‘University’ dialect associated with the language revitalization movement, and the Ni‘ihau dialect, which is spoken on Ni‘ihau and Kaua‘i. However, most linguistic documentation of Hawaiian represents an older variety of the language, and is based on the work of Elbert & Pukui, the latter of whom was a native speaker. The data in this paper comes primarily from their work, including their grammar (Elbert & Pukui 1979) and dictionary (Elbert & Pukui 1986). Additional data comes from Hawkins (1979), which primarily focuses on sentence structure for the same dialect. Therefore, the data in this paper represent the grammar as understood by speakers somewhat removed from the present day, and should not be taken to represent either modern variety (see also Alderete & MacMillan (2015) for discussion).

Elbert & Pukui (1986), the standard dictionary of Hawaiian, is a particularly rich source of data, with nearly 30,000 headwords and extensive examples which include morphological information. In order to perform corpus searching on this dictionary, a publicly available electronic copy of the 1986 dictionary was downloaded, text-normalized, and converted to a single text file.¹ While many of the specific examples in this paper come from Elbert & Pukui’s documentation, several novel generalizations about co-occurrence (or non co-occurrence) have been enabled via corpus search of their (1986) dictionary.

2.1 Root vs. non-root selection: Theoretical framework

Following Arad (2003), the difference between -na and ‘ana nominalization could be characterized in terms of ‘inner/low’ and ‘outer/high’ attachment in a binary-branching, syntactically oriented framework, a line of inquiry that has been very productive cross-linguistically (see e.g. Travis (2000b); Harley (2008); Embick & Marantz (2008), and Bye & Svenonius (2012), inter alia, as well as Kratzer (1996) and Marantz (1997) for antecedent research). Under such an approach, -na and ‘ana may be characterized as allomorphs of one underlying morpheme. In other words, the realization of a morpheme may be conditioned by syntactic position, namely root-attached versus non-root attached.

Arad (2003)’s work on Hebrew and English illustrate this model. Arad, following Marantz (2000), argues that a fundamental difference obtains between morpheme attachment to roots versus morpheme attachment to ‘words,’ which she takes to consist of at least a root and an additional (possibly null) morpheme. Typically, the root-attaching morpheme (but never the root itself) defines the category/part of speech label of the stem, while the root itself is unspecified for category. According to this model, a single morpheme, e.g. a nominalizer, can attach to either a root or a word, but with quite different results. Therefore, properties that superficially appear to be the result of different processes or rules are actually the result of attachment position: root versus non-root.

A key idea for this model is that roots are underspecified both semantically and phonologically. A root becomes a stem/word upon its first, root-attaching affixation, after which semantic and certain phonological properties become fully specified. Additional affixations must refer to the properties that obtained after the initial attachment; in this way, root attachment is privileged over non-root attachment. Arad illustrates the semantic consequences of this model with Hebrew, in which a template root combines with a pattern (also known as a binyan), yielding words with different parts of speech and different, though related, meanings. In (14), the phonological relationship between the various nouns and verbs is clear (they all contain the root elements s-g-r), but the semantic relationship is less clear. While all of the words share a core meaning, each form’s specific meaning only obtains in the context of the (non-concatenative) affixation.

However, further affixation on any word form/stem does not yield the type of semantic instability illustrated in (14). Instead, affixation on a word (root plus affix, also known as a stem) must refer to the meaning of that word, and not to the underspecified meaning of the root. For example, Arad (2003) argues that the verb (15) is derived from the noun (14f). Here, the meaning of the derived verb transparently refers to the meaning of the noun stem. In other words, ‘inner’ or root-selecting affixation is unpredictable, while ‘outer’ or non-root selecting affixation is predictable.

A key property of this morpho-syntactic model of morphology is that roots are underspecified semantically and unspecified categorically; therefore, a root is never allowed to stand alone without at least one (possibly null) affix. At the same time, morphology is accomplished in the same representational space as syntax, as illustrated in the representations below (repeated from Arad (2003)).

Within this framework, root versus non-root attachment also explains phonological effects on both roots and affixes. Arad (2003) illustrates phonological effects on the root with ‘zero-derived’ (null affixed) forms as in English (18), in which forms with null v versus null n affixation result in different stress patterns on the root. However, when a verb has the ‘nominal’ stress pattern (e.g. pérmit, ‘to issue a permit’), this form’s meaning is tightly linked to the meaning of the nominal form; in this case, the stress pattern resulting from root-attached n (resulting in a noun) determines the stress pattern of the noun-derived verb.

Root-attached affixes can also have their spell-out affected by the root, yielding unpredictable forms. Embick (2010) argues that the inner versus outer attached distinction results in the irregular spell-out of nominalizing n as -age, -tion, -al in (19), whereas non-root-attached n always gets the same spell-out for the associated gerunds, namely -ing. A key point for the data set in (19) is that the same n affix is implicated in all of these forms, with their pronunciation determined by morpho-syntactic context.

Returning to the analysis of -na and ‘ana, these appear to have the properties of R- and AS-nominals, respectively. As discussed by Embick (2010), the difference between R-nominals and AS-nominals can be modeled within this framework in the following way. Taking both -na and ‘ana to be allomorphs of n, -na is the root selecting, or inner, allomorph, while ‘ana selects vP, which nominalizes an event with complete argument structure (see also Marantz (1997); Van Hout & Roeper (1998); Borer (1999); Alexiadou (2009) and, Harley (2009) for related proposals, as well as Chomsky (1970) for antecedent research). According to this approach, the verb and related nominalization with -na in (20) have the representations in (21) and (22), in which the relevant category-defining morpheme, v or n, attaches to the root directly.

The representation of a phrase nominalized by ‘ana, such as (23), repeated from (10), has the structure in (24). Because n selects vP in (24), an eventive reading is expected. The spell-out of n for the relevant allomorphs appears as (25).

Because roots are semantically underspecified in this framework, nominalization with -na, as the inner allomorph of n, can have an idiomatic interpretation, as illustrated in (3). According to (Elbert & Pukui 1979: 81), nominalization with ‘ana yields predictable interpretations, as expected given its status as an outer allomorph. In the same passage, Elbert & Pukui note that ‘ana has a wide distribution, while -na “is suffixed to only a few verbs and not to loanwords (it is not productive).” This difference in distribution is explained by (25), insofar as ‘ana combines with vP, while the distribution of -na is tied to specific roots.

The analysis of ‘ana in (25) also predicts that nominalization with ‘ana always has event structure. This prediction is borne out, even for examples in which ‘ana appears to nominalize a noun directly. According to Elbert & Pukui (1979), when ‘ana nominalizes an underived noun, this “seems to give a verblike meaning to the head noun.” (26) and (27) illustrate the pattern of data. The event interpretation of these examples can be accounted for if the root is first selected by null v, prior to nominalization with ‘ana. In other words, while the analysis of (26) and (27) depends on a null v in their representation, this representation is justified given the event interpretation.

To summarize, while -na and ‘ana are both nominalizers, their behavior is quite different. Nominalization with -na, but not ‘ana, can result in idiomatic interpretations and root allomorphy, whereas nominalization with ‘ana, but not -na results in an eventive reading which is nonetheless predictable (similar to English gerunds with suffix -ing). Despite these differences, I have argued that the nominalizers -na and ‘ana are surface realizations of the same underlying morpheme, n. Their surface pronunciations (-na versus ‘ana), as well as their different properties are a result of different attachment positions.⁴ In the next sections, I extend the root- versus non-root attachment analysis to a complex set of valency increasing morphemes.

3.1 Interpretation and interaction

While ho‘o- and ha‘a- are historically related (both being related to Proto-Polynesian faka- (Elbert & Pukui 1986)), it is not immediately obvious if these should be treated as separate morphemes or as allomorphs in the synchronic grammar. While I will argue for an analysis in which ho‘o- and ha‘a- are syntactically conditioned allomorphs, Elbert & Pukui (1979) treat ho‘o- and ha‘a- as separate morphemes.

On the semantic side, Elbert & Pukui (1979) motivate their analysis by the fact that ho‘o- and ha‘a- can occur with the same root, deriving the same meaning (43, 44); the glosses for (43c) and (44c) are reproduced exactly as they appear in Elbert & Pukui (1986). Within Elbert & Pukui’s structuralist framework, the fact that both ho‘o- and ha‘a- can occur on the same root, with no clear conditioning factor governing their distribution, suggests that these be treated as separate morphemes.

Further, Elbert & Pukui (1979) point out that ho‘o- can attach to a ha‘a- prefixed stem (i.e. ho‘o-ha‘a-root), but never the converse. For an example such as (45), prefixation of ho‘o- and ha‘a- results in valency increase with each prefixation, as in (45c); (46) illustrates the same pattern of data. Given Elbert & Pukui’s structuralist perspective, this is taken as evidence that ho‘o- and ha‘a- have separate, dedicated positions in the verbal structure.

Corpus analysis on Elbert & Pukui (1986)’s dictionary bears out Elbert & Pukui (1979)’s intuition that ho‘o- may prefix to ha‘a-, but not the other way around. While forms with the struture ho‘o-ha‘a-root are well attested (see e.g. (45c) and (46c)), corpus analysis (considering also phonologically conditioned allomorphs) yields only one result for a potential ha‘a-ho‘o-root form. However, this form (47) is a noun, which is listed with the ‘rare’ tag in the dictionary; if it does involve the same ha‘a- and ho‘o- discussed here, either in the grammar described here or in some earlier form of the language recorded in the dictionary via textual source, this form does not appear to involve a valency increasing use of ha‘a-. Note that ho‘oili (‘to bequeath’) is, however, derived from ili (‘inheritance,’ ‘to inherit (intransitive)’).

However, ho‘o- and ha‘a- do not behave identically. In particular, prefixation with ha‘a- (like the nominalizer -na) can yield idiosyncratic interpretations, unlike ho‘o-. While it is true that ho‘o- and ha‘a- can attach to the same word and derive the same or similar meanings, as in (43) and (44), some ha‘a- derived forms have less predictable meanings, whereas ho‘o- is generally transparent semantically.

For example, lele is an intransitive verb, and ho‘o-lele is a transitive verb that extends those meanings (48). However, ha‘a-lele has a slightly different set of meanings that are not obvious from the meaning of lele. When both prefixes occur, as in ho‘o-ha‘a-lele, the meaning is derived from the meaning of the ha‘a- prefixed form, not from the meaning of the root alone.⁸

Further, ho‘o-, if present, is always the leftmost causative/simulative prefix; ho‘o- is also fully productive, as discussed above. Ha‘a-, though, does not occur on all roots. Instead, other morphemes occur with the same meaning as ha‘a-, in particular ‘ā- and kā-. Elbert & Pukui (1979: 65, 68–71) also define these as ‘causative/simulative’ (though unproductive) and treat them on a par with ha‘a-. In addition, in some cases more than one of these three morphemes (ha‘a-, ‘ā-, and kā-) can occur on the same root, but with slightly different senses. (49) and (50) illustrate the valency increasing uses of prefix ‘ā- and kā-, respectively.

While ha‘a-, ‘ā-, and kā- are generally in complementary distribution, a limited number of roots take both ‘ā- and kā- (Elbert & Pukui 1979: 68), usually with a slightly different shade of meaning (51, 52).

As with roots prefixed with ha‘a-, roots prefixed with ‘ā- and kā- can have ho‘o- as an additional prefix. And, parallel to the ho‘o-ha‘a-root forms discussed above, ho‘o- increases the valency of the stem, maintaining the derived meaning, typically deriving a transitive from an intransitive, as in (53) and (54).

A preliminary attempt at characterizing the data in this section leads to the following conclusions, though several additional patterns of data are discussed below, leading to a more complete analysis. First, three prefixes, all less productive than ho‘o-, are mostly in complementary distribution with each other: ha‘a-, ‘ā-, and kā-. These three morphemes generally increase the valency of the root to which they attach, though the behavior of these prefixes is more fully specified in the following section. In the rare case that more than one of ha‘a-, ‘ā-, and kā- can attach to a root, there typically obtains an interpretation that is not fully transparent. At the same time, a different form, ho‘o-, can co-occur with the three morphemes ha‘a-, ‘ā-, and kā-, always attached in the ‘leftmost’ position, and with no competing allomorph, setting aside surface phonological alternations.

Within the framework summarized above, I argue that the Hawaiian causative/simulative affixes should be understood in terms of root versus non-root attachment. Specifically, ho‘o- is the outer-attached realization of a type of verbal affix, while the other forms are inner-attached realizations of the same morpheme; direct root attachment can then induce root-conditioned allomorph selection (e.g. ha‘a-, ‘ā-, or kā-) as well as non-transparent semantics, as discussed above with respect to (48), (53), and (54). The ‘outer’-attaching head is not local to the root, and therefore does not participate in root-conditioned allomorph selection. Finally, outer attachment (the ho‘o- forms), being non-local to the root, always results in transparent semantics.

3.2 Morpho-syntactic status of the Hawaiian causative

In this section, I further consider the status of ho‘o- and ha‘a- with respect to their status as potential allomorphs and some theoretical considerations for causatives more broadly. While the analysis of ha‘a- (and related forms) poses some complications, I begin with a discussion of ho‘o-, which patterns with other cognate forms in Polynesian.

For example, Gould et al. (2009) discuss the status of the Niuean prefix faka-. As illustrated in (55), (56), and (57), the core cases of affixation with Niuean faka- pattern with those of Hawaiian ho‘o-. Like with Hawaiian, affixation with faka- can also indicate intention. According to Gould et al. (2009), faka- selects a vP (which itself may have a vP, aP, or nP complement), along similar lines to the analysis presented here for Hawaiian ho‘o-.

Samoan fa‘a- is extensively discussed by Tollan (2018) and Hopperdietzel (2021), with core cases appearing in (58) and (59) (Tollan 2018: 27).

As Tollan (2018) points out, Samoan fa‘a- derives causatives from unaccusative, unergative, and middle predicates. However, an important difference between Samoan and Hawaiian is that fa‘a- appears to be more restricted than ho‘o-, insofar as Samoan fa‘a- does not apply to transitive verbs with ergative case-assigned subjects; while outside the scope of this paper, these differences may be in part due to the difference in case marking alignment between the two languages. Like Hawaiian ho‘o-, Samoan fa‘a- introduces a volitional agent.

Similar to the analysis of ho‘o- proposed here for Hawaiian, Hopperdietzel (2021: 130) argues that Samoan fa‘a- does not select roots based on the following evidence: i) the absence of root-driven allomorphy, ii) the absence of idiosyncratic readings for fa‘a- -prefixed words (contrasted with Samoan prefix ta-, which Hopperdietzel does analyze as root-selecting), and iii) preservation of meaning for fa‘a- prefixed words under root-selecting nominalization. Hopperdietzel’s analysis of fa‘a- as an outer causative allomorph and less productive ta- prefix as an inner causative allomorph therefore parallels the analysis of causatives presented here with respect to Hawaiian.

While having properties of a non-root-selecting affix, Tollan (2018) argues that Samoan fa‘a- is a Voice head with respect to the ‘bundling’ parameter for verbal heads proposed Pylkkänen (2008), according to which specific languages may realize different verbal properties (Voice, causation, transitivity, etc.) on either individual or conflated heads (see also Nie (2020) for extensions of this model). Hopperdietzel (2021) suggests that Samoan fa‘a- has properties of both Voice bundling and non-Voice bundling languages; the restriction against fa‘a- with transitive verbs that have ergative subjects suggests Voice bundling, because fa‘a- introduces causative semantics while also interacting with Voice via the restriction on certain transitives.

Samoan fa‘a- has non-bundling characteristics at the morphological level as well, insofar as fa‘a- co-occurs with passive morphology, according to Hopperdietzel (2021). Legate (2014: 111–118) shows that Malayo-Polynesian Acehnese also allows causative and passive morphology to co-occur, which she argues is strong evidence against a bundling analysis for Acehnese. Like Samoan and Acehnese, Hawaiian also allows ho‘o- with the affixal passive in Hawaiian. For example, (60) and (61) illustrate forms with the passive suffix -C(i)a. In (60), the base ‘ike is modified by ho‘o- resulting in a causative, with the causative form being passivized in (60c). (61) illustrates the same pattern. While I return to a discussion of the Hawaiian passive below, the data here show that ho‘o- and the passive may co-occur.

Given the passive data discussed here, and also considering the lack of Voice/transitive restriction on ho‘o- in Hawaiian (unlike in Samoan), I conclude that Hawaiian is not a Voice bundling language insofar as the causative is concerned. In other words, non-causative v, causative v, and passive v may all occur independently of each other. For example, (60a) and (61a) are verbal non-causatives, while (61b) is a passive in absence of a ho‘o- causative, while (60c) includes both the passive and a ho‘o- causative. Under Pylkkänen (2008)’s analysis, this is evidence that Voice and v are distinct.

Therefore, I will gloss ho‘o- as v_caus. Following Hawkins (1979)’s characterization of the thematic properties of ho‘o-, I take v_caus to be associated with an agentive argument in its specifier, and therefore distinct from transitive/unergative v, which is associated with (at least) agentive and experiencer arguments; in section 4, below, I further explore the relationship between v_caus and v. Therefore, the analysis developed here is similar to that proposed by Harley (2017) for other non-bundling languages, according to which “verbalizing v and causativizing Caus are categorically and morphosyntactically distinct.”

With an analysis of ho‘o- as a non-root selecting allomorph of v_caus in place, I now consider the status of ha‘a- and its alternants. Elbert & Pukui (1986) describe ha‘a- as “a prefix similar in meaning to the causative/simulative ho‘o-,” while Elbert & Pukui (1979: 68) describe the prefixes ‘ā- and kā- as also having meanings similar to ho‘o-. In some cases, the similarity to ho‘o- is clear in terms of valency increase, as in (63, 64), repeated from above.

A key point in the framework adopted here, however, is that roots must have minimally one (often null) affix, with roots themselves being underspecified semantically. Under the analysis in which ha‘a- is a root selecting prefix, it is not attached e.g. to the word lele (63a), but rather to a root which is only indirectly visible (i.e. roots cannot stand alone without an affix). Therefore, the representations of the examples above are arguably as in (65) and (66), under an analysis in which ha‘a- is a root-selecting affix, independent of its status with respect to ho‘o-. Likewise in (66), ha‘a- does not increase the valency of nui the adjective, but rather $\sqrt{nui}$ the root. For this reason, evaluating the status of a root-attached affix is arguably more difficult as compared to outer affixes, which, by definition, attach to stems that have fixed meanings.

In order to characterize the difference between e.g. lele and ha‘a-lele in (65), I propose that v and v_caus are distinct in this grammar. In other words, both (65a) and (65b) represent idiosyncratic interpretations with affix-root structure. Returning to the discussion of Voice-bundling above, a complicating factor is that v_caus, when in root-selecting position, and root-selecting v both act as verbalizers and also argument introducers. However, if the analysis of forms such as (65) is on track, v and v_caus are morphologically distinct heads and therefore not bundled, although both can select roots directly.

Given the available corpus evidence, it is difficult to exactly specify the semantic contribution of v versus v_caus. However, according to Makanani (1973), also cited by Elbert & Pukui (1979), some of the key effects of prefixation with ha‘a- when compared with a form without the prefix include transitivization (67), and verbalization (including turning a form into an unambiguous verb), as in (68)–(70).

In other cases, meanings of verbs may be different without a change in grammatical status, as in (71), (72), and (73). Other core meanings of these prefixes include no change (72a, 72b), as well as the simulative reading discussed above.⁹

As discussed above in connection with (65) and (66), the (a) and (b) examples in the previous pairs of words cannot be compared directly in the framework adopted here; rather, each word consists of a null categorizing prefix (v, n, etc.) in the (a) example and v_caus in the (b) example. For the examples in which there is no difference or only a semantic difference between the (a) and (b) forms (e.g. (72a, 72c)), I conclude that the (a) form (with verbalizing v) and the (b) form (with v_caus) have their meaning due to idiosyncratic interpretation (given the semantically underspecified nature of roots).

Given the evidence here (including the commentary included in Elbert & Pukui (1979)), I conclude that ha‘a-, ‘ā-, and kā- are root-selecting allomorphs of a causative verbal head; grammatically speaking, this causative head adds a specifier, though an understanding of the semantic difference between root-selecting v and v_caus requires additional research. While ha‘a-, ‘ā-, and kā- are generally in complementary distribution, in some limited cases these can be prefixes to the same root, typically with a different shade of meaning (51, 52), which appear similar to morphological ‘doublets’ (e.g. recitation/recital, cover/coverage) discussed by Halle (1973) and Kiparsky (1982). According to the analysis in Embick (2016), doublets of this kind are restricted to root-selecting allomorphs, and typically involve polysemy, exactly as observed in the Hawaiian cases above.

A comparison of ho‘o-, the outer allomorph, and the inner allomorphs ha‘a-, ‘ā-, and kā-, shows that the differences between the outer and inner allomorphs pattern with other analyses of inner and outer causative attachment in other languages in terms of interpretation and productivity (see e.g. Travis (2000a); Svenonius (2005); Harley (2008)). Differences between the interpretations of outer-attaching ho‘o- and the inner-attaching forms are due to their differing structure, which I examine in more detail in the section 4; additional evidence for ha‘a- as a root-attached form is presented in section 5 in terms of affix co-occurrence restrictions.

Returning now to the analysis of ho‘o-/ha‘a- in terms of inner/outer allomorphs, I argue that there exists one general valency increasing morpheme v_caus, with underlying phonology CV(CV) (ignoring length distinctions on the initial V; note that the glottal stop is phonemic in Hawaiian). This one underlying morpheme would then have the outer allmorph ho‘o- and inner allomorphs ha‘a-, ‘ā-, and kā-, with the morpho-syntactic representation in (74), in which specific allomorphs are selected post-syntactically via late insertion in a realizational framework.

Under such an analysis, v_caus would have the insertion rules listed in (75). These insertion rules characterize the fact that ho‘o- is always an outer allomorph in the presence of any one of ha‘a-, ‘ā-, or kā-. Further, as ho‘o- can attach to v_caus, the fact that ho‘o- yields only transparent interpretation is derived. However, (75) cannot be a complete analysis, because there are many instances in which ho‘o- appears to attach to the root directly, as in (28–34).

While I will ultimately adopt an analysis along the lines of (74,75) for v_caus, several complications remain to be addressed for the analysis of valency increase/causative. These complications involve a more detailed view of the argument structure of words with the relevant prefixes.

4.1 Interim summary

Thus far, I have argued for an analysis of ho‘o-, ha‘a-, and related forms in which these are all allomorphs of one valency increasing morpheme. As the selectional properties and semantics of these forms are exceptionally broad compared to more cross-linguistically familiar cases of causatives, I have treated the underlying morpheme as an agentive, general valency increasing type of verbalizer, labeled v_caus, for the purpose of morpho-syntactic analysis. The key idea advanced here is that ho‘o- is to be treated as an outer allomorph of the valency increasing morpheme, with ha‘a-, ‘ā-, and kā- being inner allomorphs.

The descriptive literature on Hawaiian has implicated these allomorphs in a number of constructions, including causatives, deliberatives, and simulatives. These different constructions are modeled here via the combination of various morpho-syntactic properties, including the presence of functional heads such as v, ASP, n, null arguments and coindexation, and inner and outer domains for allomorph selection. The outer domain spell-out of v_caus is reflected in (93a), which includes the property of v_caus selecting v_caus (reflecting ho‘o-ha‘a-root forms); following Nie (2020), I assume that v_caus cannot occur more than twice due to independent limits on argument licensing. The other realizations of v_caus all reflect root selection.

The model in (93) explains some longstanding generalizations in the descriptive literature (Elbert & Pukui 1979; Hawkins 1979), insofar as ho‘o- attaches to a range of functional heads, with the other valency-increasing allomorphs selecting roots. One such generalization is that ho‘o- is far more productive than any of ha‘a-, ‘ā-, and kā-. This is further reflected in Elbert & Pukui (1979)’s claim (confirmed via corpus search) that ho‘o- can be a prefix for any one of ha‘a-, ‘ā-, and kā-, but never the other way around. Additionally, only ha‘a-, ‘ā-, and kā- can yield idiosyncratic interpretations, whereas affixation with ho‘o- is always transparent. At the same time, the fact that Elbert & Pukui (1979) call both ho‘o- and ha‘a- ‘causative/simulative’ prefixes is captured here by their analysis as allomorphs of one underlying morpheme, in contrast to Elbert & Pukui (1979)’s own analysis in which these are separate morphemes.

The primary way that these prefixes differ is with respect to their ability to trigger causative semantics, in the ‘make’ or ‘let’ causative sense of the word. Ho‘o-, but apparently not ha‘a-, can derive forms that are expressed by lexical causatives in English, such as ‘feed’ (29) and ‘show’ (77); also, many ho‘o- forms are glossed in Elbert & Pukui (1986) in the form “to cause to [root]", as in (45c), (46c), and (48d). This is relevant because translations of the form “to cause to [root]" indicate causer-causee argument structure, in which an inner form (null v or inner v_caus) introduces a specifier (the causee) in addition to the presence of a higher phrase, headed by outer v_caus (the ho‘o- prefix), which introduces the causer. According the model of causatives developed above, only ho‘o- forms (with or without inner ha‘a-) should allow this type of argument structure, while ha‘a- type causatives should not have this structure unless further affixation with ho‘o- applies.

In order to further examine this semantic element of valency increase, a corpus search of Elbert & Pukui (1986) reveals the gloss “to cause to [root]” for forms with ho‘o- (and its phonologically conditioned allomoprhs) 145 times; these occur as ho‘o--root or ho‘o--ha‘a- or [other causative alternant]-root. Only one instance of the gloss “to cause to” remains in the dictionary, for the word ho‘i, where it is glossed as “To leave, go or come back; to cause to come back.” Hawkins (1979: 25) specifically refers to this use of ho‘i in her discussion of ho‘o- as one (possibly the only) intradirective verb without a ho‘o- form. She speculates that the phonological similarity between ho‘o- and ho‘i might be the cause of this gap.

By comparison, there are no instances of this gloss for forms with ha‘a- or any of its phonologically conditioned allomorphs in Elbert & Pukui (1986). A complicating factor in the descriptive literature involves one exception to the generalization described here, which is given in Elbert & Pukui (1979)’s grammar (but not Elbert & Pukui (1986)’s dictionary) for ha-‘āpuka, where this word is glossed as ‘to cause to cheat’ (Elbert & Pukui 1979: 68). However, in the dictionary (Elbert & Pukui 1986: 45), the same form is glossed as ‘same as ho‘āpuka.’ Fortunately, for this entry a full example is given, which I repeat here in (94), where I include glosses and a literal translation in addition to the biblical translation given in the text. In (94), the relevant form appears in a passivized relative clause, with the sole argument (outside the relative clause) being ka waiwai ‘the wealth.’ Given that this clause is passivized, an additional argument is suppressed, meaning that, from this evidence, ha-‘āpuka only has two arguments (i.e. if it were in active voice), and does not have causer-causee argument structure.

From this, in conjunction with corpus search on the dictionary, I conclude that only ho‘o- forms can have causer-causee argument structure. This argument-structural property of ho‘o-, in conjunction with its productivity and transparent semantics, is predicted by a model in which ho‘o- is an outer valency-increasing allomorph, where it functions similarly to other examples of outer-domain causatives (see e.g. Folli & Harley (2005), Pylkkänen (2008) and Harley (2017)).

At this point, the analysis developed thus far provides a formal model for the observed properties of the morphemes under discussion (both the causatives and nominalizers). The following section will present further evidence for this proposal, illustrating the interaction of v_caus with nominalization. The interaction of ho‘o- and ha‘a- with nominalizing morphology suggests the same analysis of inner and outer domain allomorphy discussed above.