The bracketed clauses in (1) have appeared to many a student of introductory syntax to instantiate the same type of syntactic structure.

Introductory classes typically marshal several types of facts (drawn from a long tradition in descriptive grammar) to dissuade the student from this view. Rather than grouping the clauses in (1) together, we must distinguish a (finite) complement clause, (1a), from a (restrictive) relative clause, (1b). Relevant English facts include:

The student is thus led away from the hypothesis of a single structure for the clauses in (1) and toward a perspective on which complement clauses and relative clauses differ in both their internal and external syntax. Let us call this perspective the standard theory. According to the standard theory, internally, relative clauses but not complement clauses contain an Ā dependency. This accounts both for the presence of a gap (the tail of the Ā chain) and for the relative pronoun (the head of the Ā chain), (2a, b). Externally, relative clauses are adjuncts, whereas complement clauses are (as the name suggests) complements; this accounts for the difference in obligatoriness, (2c). Like adjectives, relative clauses are a type of adjunct restricted to nominal projections (a fact potentially to be explained in semantic terms), accounting for (2d). These points together make island effects (2e) unsurprising, even overdetermined: extraction from a relative clause is extraction from an adjunct and from a nominal and across an intervening Ā dependency—three distinct factors all of which are known to give rise to island effects independently.

The impetus for this article is a strand of literature that has nevertheless sought to vindicate the introductory student’s intuition of syntactic commonality between relative clauses and (at least some) notional complement clauses (Manzini & Savoia 2003, Nichols 2003, Aboh 2005, Kayne 2008, Arsenijević 2009, Aboh 2010, Haegeman & Ürögdi 2010, Krapova 2010, Caponigro & Polinsky 2011, Haegeman 2012, Kayne 2014, Manzini 2014, Hanink & Bochnak 2017, Poletto & Sanfelici 2018, Pietraszko 2019, Bochnak & Hanink 2021, and Bondarenko 2022, among others; see De Cuba 2017 and De Cuba 2023 for an opposing view).1 This work generally takes as its point of departure data from various languages showing morphosyntactic commonalities between the two varieties of clauses, often at the clause edge. Some of these commonalities involve indications of DP structure, as for instance in the Washo examples in (3); Hanink & Bochnak 2017 argues that the morpheme ge, glossed “rel,” is in fact uniformly a D head that composes with CP complements.2

A further case where the similarity between notional complement clauses and relative clauses arguably reflects nominalization is discussed by Pietraszko 2019.

Other commonalities involve evidence of Ā movement. In Italian and other Romance languages, for instance, an element (in Italian, che) appears at the edge of relative clauses and finite complement clauses that is implicated in (other) Ā dependencies, such as wh questions:

Looking beyond the clause edge, commonalities in Adyghe in terms of Ā-specific verb morphology are highlighted in Caponigro & Polinsky 2011: both the relative-like example in (5a) and the complement-like example in (5b) feature a marker of relativization of an oblique, namely the verbal “wh-agreement” prefix zə-.

Such facts suggest that relative clauses and notional complement clauses may in fact be more similar that the standard theory had concluded; in particular, it may be that both types of clauses involve a DP layer and contain an Ā dependency. Researchers drawing this conclusion differ in how thorough a rejection of the standard theory they endorse. At one end of the spectrum are Kayne 2008, Arsenijević 2009, and Kayne 2014, according to which relative- and complement-clause structures cross-linguistically are largely identical both in external and internal syntax. Externally, Kayne and Arsenijević propose, both types of clauses involve complementation, not adjunction (a claim that builds on the analysis of relative-clause syntax in Kayne 1994). Internally, clauses of both types involve an Ā dependency between a clause-internal position and an element at the clause edge. Both types of clauses also involve a noun on the edge of the clause; this is the typical nominal head of a relative clause but often a covert noun in a notional complement clause. From this perspective, the differences listed in (2) between finite complement clauses and restrictive relative clauses in English must arise from properties of the particular position relativized, the inventory of relative pronouns, or the inventory of silent nouns. Concerning the presence of a clear gap in relative clauses but not in notional complement clauses, for instance, Kayne and Arsenijević both appeal to the particular position of the tail of the Ā chain in notional complement clauses: for Kayne, complement to a silent P and, for Arsenijević, the specifier of a Force projection in the high left periphery.

A different picture is suggested by Caponigro & Polinsky 2011 on the basis of a thorough study of Adyghe. On Caponigro & Polinsky’s view, the parallel between relative clauses and notional complement clauses is extensive but also language particular. On the one hand, relative clauses and notional complement clauses in Adyghe all involve Ā movement, and a nominal is always present at the clause edge (though it may be silent). On the other hand, unlike Kayne and Arsenijević, Caponigro & Polinsky do not conclude that relative clauses and notional complement clauses have the same structure across languages. Instead, they suggest that variation emerges as a function of different functional inventories in different languages: Adyghe has only a relative complementizer, meaning that all CP embedding must involve relativization, whereas other languages possess non-relative complementizers and therefore allow for non-relative complement clauses. This means the standard theory could well be right for English, even if it doesn’t extend to Adyghe. The idea that some but not all notional complement clauses are similar to relative clauses is further explored by Aboh 2010, Haegeman & Ürögdi 2010, Krapova 2010, Haegeman 2012, Hanink & Bochnak 2017, and Bochnak & Hanink 2021.

In this article I seek to develop this latter type of perspective by exploring the syntax of a class of notional complement clauses in Nez Perce that show a striking resemblance to relative clauses. The Nez Perce relative-clause structure shown in (6a) and the notional-complement-clause type of interest, (6b), have in common the presence of the functional elements yox̂ and ke at their left edge.5 As the glossing of these examples reflects, yox̂ is a (nominative) relative pronoun, and ke is a complementizer linked to Ā extraction (Deal 2016a).

These data thus speak to the type of pattern seen above in Italian (insofar as they involve common morphosyntax at the edge of the clause) and, more generally, to the type of pattern seen in Adyghe (insofar as they involve morphology specific to Ā dependencies).

There are several facets of the morphosyntax of Nez Perce that make this language an interesting one in which to study relative-like notional complement clauses, such as (6b). First, as both (6a, b) show, relative pronouns and complementizers co-occur in this language (i.e., there is no ban on “doubly filled comp”). Previous work has taken different perspectives on whether elements like Italian che, which occurs both in “complementizer” function and in “relative-pronoun” function, as in (4), should be analyzed strictly as a complementizer across these usages (Cinque 1978, Manzini & Savoia 2003) or strictly as a relative pronoun (Kayne 2014, Poletto & Sanfelici 2018). The fact that Nez Perce possesses separate, overt items in these functions makes it clearer what element plays what role in the syntax of this language.

Second, in interesting contrast to Adyghe and Italian, not all notional finite complement clauses have this relative-like character in Nez Perce. The difference is determined by the embedding predicate: some predicates (a subset of the factives) require their notional complement clauses to be “relative,” whereas others do not. The following pair of examples contrasts lilooy ‘be happy,’ whose notional complement is obligatorily relative-like in morphosyntax, with cuukwe ‘know,’ whose notional complement freely lacks this marking.

Nez Perce thus presents the opportunity, internally to a single language—holding constant the inventory of relative pronouns and other functional material (e.g., null nouns, as Kayne and others posit)—to ask: why do some notional complement clauses but not others appear relative-like? What is the syntax of “relative embedding,” that is, notional complement clauses with relative morphosyntax? In addressing similar questions for Washo, it is concluded in Hanink & Bochnak 2017 that the key factor is nominalization: shared morphosyntax between relative clauses and certain notional complement clauses in that language realizes D. The investigation of Nez Perce reaches a different conclusion: notional complement clauses are never DPs in this language. Rather, the distinctive behavior of relative-like notional complement clauses involves Ā movement. The overall picture is one where notional complement clauses vary, within and across languages, in including or not including the familiar syntactic ingredients of relative clauses. I take this to support the general typological perspective emerging from work such as Caponigro & Polinsky 2011 and Hanink & Bochnak 2017. At the same time, we will see that relative-like notional complement clauses in Nez Perce show systematic morphosyntactic differences from their counterparts in other languages, including Adyghe and Washo. This demonstrates that relative-like notional complement clauses are not a unitary phenomenon across languages but rather represent a range of ways in which notional complement clauses may be built up with relative-clause ingredients.

The article is structured as follows. In section 2 I begin with some background about the Nez Perce language, with particular attention to relative-clause syntax. I then present basic properties of Nez Perce relative embedding in section 3. Section 4 discusses the external syntax of relative embedding, arguing that the structures in question do not involve a DP or PP layer above the relative CP. Section 5 then turns to the internal syntax of these clauses, arguing that they involve Ā movement from a position above the TP level. With this picture of the syntax of relative embedding in hand, section 6 turns to the language-internal contrast between relative and non-relative embeddings in Nez Perce, arguing against the position that all notional complement clauses in this language possess an underlying relative structure. Section 7, which doubles as a conclusion, reviews the resulting picture of cross-linguistic variation as concerns relative-like notional complement clauses.

Nez Perce is a Penutian language indigenous to the interior Columbia Plateau region; the traditional territory of Nez Perce speakers encompasses parts of the US states of Idaho, Washington, and Oregon (see Aoki 1994: viii). The language is currently severely endangered with only a small handful of elder native speakers remaining and active language-revitalization projects underway. The data in this article come from work with two speakers, Florene Davis and the late Bessie Scott, conducted on the reservation of the Nez Perce Tribe of Idaho in Lapwai, Idaho, between 2006 and 2019.6

Nez Perce is a morphologically rich language with both head and dependent marking at the clausal level, quite free clausal word order, and very free pro drop of all arguments.7 The case system is tripartite, distinguishing ergative and accusative in transitive clauses, for example (9a), and nominative (unmarked) in intransitive clauses, for example (9b). There is no split ergativity based on clausal properties such as aspect, tense, or negation.

The verbal agreement system is nominative–accusative, with largely separable exponents of person and number agreement. Agreement is overt for plural and (notably) for third person (like in English). Person agreement for the subject and/or object occurs leftmost in the verb word and indexes one or both core arguments, depending on person and number values. The person marker is followed by plural agreement for the subject and/or the object, for example, the object plural marker naas in (9a). Complexities of the agreement system, including relatively idiosyncratic restrictions on combinations of particular affixes, are described in Deal 2015b. Verbal inflection for tense–aspect–mood, which influences the particular form of subject number agreement, is described in Rude 1985 and Deal 2010b.8

Ergative case, accusative case, and object agreement in transitive clauses are tightly linked, and there are two types of circumstances in which all three must be absent (Rude 1985, Deal 2010a, Deal 2010b). (Such patterns will prove helpful in assessing the external syntax of relative embedding in section 4.) The first is when the subject binds the possessor of the object, as in (10). Note that both the subject and the object are nominative (when they are overt at all) and that there is no object agreement on the verb. Subject agreement proceeds as normal in such clauses.9

The second circumstance is when the object is a weak indefinite, that is, an indefinite description that takes narrow scope with respect to all clausal operators. This requirement of narrow scope is shown in (12). Note again that all arguments are in nominative case and that object agreement is absent.

Both circumstances just described result in unmarked case on both the subject and the object: ergative and accusative are lost together.10 These facts are discussed at length and analyzed in an Agree-based case theory in Deal 2010a and Deal 2010b.

Nez Perce relative clauses are described and analyzed in Deal 2016a. As that work notes, most relative clauses in the language are externally headed; all such relatives are post-nominal. They contain a case-marked relative pronoun ko/yox̂, drawn from the demonstrative inventory, the complementizer ke, and a gap inside CP. The same structure is used to relativize on all core argument positions, as well as on obliques. (There is no syntactic ergativity and no pronominal resumption.) The inflectional syntax of a relative clause is generally in line with the structure of simple finite clauses, showing the typical range of tense–aspect–mood values and case arrays. Example relative clauses are shown in (13).

I will first review the evidence that Nez Perce relative clauses involve Ā movement, following Deal 2016a. I will then discuss the morphosyntax of the relative pronoun ko/yox̂ and the complementizer ke, with particular attention to factors that can help us identify their category as D (relative pronoun) versus C (complementizer).

Relative clauses in Nez Perce show familiar evidence of Ā movement. First, relativization is unbounded; the relative pronoun may be found one or more finite clauses away from the gap:

Second, the relative pronoun may not be separated from the gap by an adjunct island or coordinate island:

Third, relative clauses are themselves islands:

Language-internal evidence of an Ā dependency comes from the fact that the same C element, ke, appears in relative clauses and in wh questions (where it is optional). (This example shows ϕ agreement on the complementizer, as do (15) and (16) above; this is discussed below.).

Notably, ke is not a general-purpose subordinator: it cannot occur in the complements of the verbs hi ‘say/tell’ or neki ‘think,’ as in (20a) and (20b), including in cases where these verbs are along the path of relativization, as in (20b).

I suggest in Deal 2016a: 438 that ke is akin to Ā morphology in Chamorro and Hausa in spelling out the features that drive the final step of Ā movement (Chung 1998, Green & Reintges 2001; see general discussion of the morphology of Ā extraction at clause edges in Georgi 2014).

Thus far I have taken for granted that ko/yox̂ is a relative pronoun and ke a complementizer. A triad of behaviors standardly held to differentiate relative pronouns, which are D elements, from complementizers, that is, C elements, is discussed in Poletto & Sanfelici 2018. First is case: relative pronouns inflect for case, but complementizers do not. Second is sensitivity to features of the head noun, for example, animacy: pronouns are sensitive to these features, whereas complementizers are not. Third is compatibility with adpositions: complementizers cannot be combined with adpositions, whereas pronouns can. These factors converge in Nez Perce on a diagnosis of ko/yox̂ as a relative pronoun and ke as a complementizer.

In terms of case, relative pronouns show the same core case-marking pattern found elsewhere in the language, distinguishing nominative, ergative, and accusative. The case of a relative pronoun generally matches the case expected for the relative-clause-internal gap; thus the relative pronoun is accusative in the object-relative example (21a) but ergative in the transitive subject-relative example (21b). (Exceptions to this pattern involve case attraction and are discussed in section 4.).

In terms of features of the head noun, relative pronouns show concord with the head noun in number. A table of relative pronouns by case and number is shown in (22); examples of plural relative pronouns are shown in (23).11

The complementizer does not show any parallel behavior, either regarding case or regarding number. The core form of the complementizer is always ke, regardless of whether the gap is nominative (6a), accusative (13a), or ergative (13b) and whether the head noun is singular or plural. Rather than sensitivity to these factors, which are associated with the D status of the relative pronoun, the complementizer ke shows ϕ agreement with the subject and/or object of the embedded clause. This pattern is analyzed in the interaction–satisfaction theory of Agree in Deal 2015a: the ϕ probe borne by C agrees with all ϕ features, starting with those of the subject, until the feature [addr] is encountered. Person agreement is morphologically overt with first and second person but not with third. The examples in (24) involve relativization of the theme of a ditransitive, which leaves two other nominals available for agreement in the clause. As these examples show, complementizer agreement may target the subject (24a, d), the primary object (24b), or both (24c), depending on the distribution of person features in the clause.

This behavior would be unusual if ke were a D head; we expect a D head to expone its own ϕ features, rather than strictly those of other arguments. Furthermore, as I discuss in Deal 2015a, the Agree algorithm at play in ke agreement considers nominals in an order that begins with the highest argument in the embedded clause (the subject). This follows straightforwardly if the probe originates on the C head and probes into its c-command domain. It is less clear how to derive this pattern on a relative-pronoun analysis of ke.12

Turning now to adpositional relatives, adpositions are generally null in Nez Perce, with their presence detectable through particular cases they assign to their complement DPs: for example, instrumental in (25). Note that the instrumental case shows case concord across the PP-internal DP.13

Adpositional relativization is possible and shows assignment of case to the relative pronoun (and not the complementizer) that parallels what is seen on the D head of a PP complement generally. Thus the same form konki that appears in (25) as a demonstrative also appears in (26) as a relative pronoun.

This behavior is again as expected on the assumption that the relative pronoun is a D head in (26) just as in (25); the diagnosis of ke as a C head explains why its form does not vary with the introduction of the PP structure.

Having reviewed how these three diagnostics for relative pronouns versus complementizers apply in Nez Perce, it should be noted that Poletto & Sanfelici 2018 casts doubt on whether these properties indeed diagnose categorial (D vs. C) status, reviewing evidence from various old Romance varieties. Although Modern Italian che is classifiable as a C head by these properties, Poletto & Sanfelici argue it should instead be treated as a featurally bleached relative pronoun, in line with its historical antecedents. A parallel move for Nez Perce would be to treat ke as a relative pronoun, rather than a C head. In addition to the challenge for this view in connection with agreement, noted above, a further challenge is that ko/yox̂ and ke systematically co-occur: relative clauses in the language must now be taken to systematically contain not just one but two relative pronouns.14 It is not clear why this would be, in syntactic or semantic terms—how is “double” relativization compositionally interpreted? And why is “single” relativization ruled out? What is the connection between ke and other moving elements, such that neither ke nor ko/yox̂ can serve as a relative pronoun by itself? These questions are avoided on the complementizer analysis, where ke is a C head hosting a feature driving Ā movement and ko/yox̂ is the sole relative pronoun, moving to spec-CP. In view of these challenges, I will assume moving forward that ko/yox̂ is indeed a relative pronoun, that is, a D element, whereas ke is a C head originating in the left periphery.

Turning now to notional complementation structures, Nez Perce has a number of attitude predicates that require their notional complement to be “relative” in the sense discussed in the introduction. These are primarily (though not exclusively) predicates whose English translations would be classified as emotive factives.15 Following is a list of such predicates.16

The distinctive behavior of these predicates is that their clausal complement must contain a relative pronoun (yox̂) and a relative complementizer (ke). Omission of these elements produces ungrammaticality:

As previewed in the introduction, I refer to this type of notional complement clause as a relative embedding, or RE (borrowing this term from Caponigro & Polinsky 2011). In this section I discuss the following basic properties of REs. First, the complement is indeed subordinated. Second, complementation possibilities remain constant across both verbal and adjectival uses of the root. Third, all such predicates are factive (even when their English translations are not).

Let us first consider the connection between the two clauses in examples like those in (27). Evidence that the bracketed clauses are indeed embedded (and not, say, high adjuncts) comes from Condition C. The baseline example (29a) shows that a pronoun in the embedded clause may co-refer with the matrix subject. However, the opposite is not possible, as (29b) shows. Example (29c) confirms that this judgment is not due to a general ban on cataphora in the language; thus the inability of the pronoun and the name to co-refer in (29b) suggests that the matrix subject c-commands the embedded clause.

Note that these data directly contrast with data provided by Clem 2022 to show that notional complement clauses in Amahuaca are in fact adjuncts originating high in the matrix clause. The contrast suggests that notional complement clauses may indeed be high adjuncts in some languages—but not in Nez Perce.17

Further reason to think the embedded clause occurs relatively low in Nez Perce RE constructions comes from non-verbal RE predicates. Note to begin with that while various RE verbs in Nez Perce are translation equivalents of English adjectives, these items are clearly verbal rather than adjectival in the examples we have seen thus far. As noted in section 2, Nez Perce verbs inflect for the person and number of their arguments as well as for tense–aspect–mood. None of this inflection is available to adjectives; rather, for tense–aspect–mood to occur in the clause, predicative adjectives must occur with a copular verb. In (30), I exemplify this behavior for an uncontroversial verb (paay ‘to arrive’) and a simplex adjective (himeeq’is ‘big’).

The examples in (31) show the same contrast for forms built from the root lilooy ‘be happy.’ The verbal version, (31a, b), shows ϕ inflection (the prefix hi- for third-person subject) and tense–aspect–mood inflection (the suffix -ca, for imperfective aspect) flanking the root. The adjectival versions, (31c), require a ϕ- and tense–aspect–mood-inflected copula, along with an adjectivalizing morpheme on the root.18

The relative complementation structure is preserved under adjectivalization, as (32) shows. (See (27c) for cicwaay ‘be surprised’ occurring as a verb.)

This fact suggests that adjectivalization occurs above the level at which the root combines with its notional complement, which in turn suggests that the embedded clause occurs relatively low—in a standard complement position or similar—inside the matrix clause.19

We now add an important basic semantic property to the profile of REs that is taking shape: REs are factive. While this is perhaps unsurprising for predicates such as lilooy ‘be happy,’ cicwaay ‘be surprised,’ and timiipni ‘remember,’ whose English translations are factive, it is true as well for predicates such as tim’neeneki ‘be worried,’ whose English translations are not uniformly factive. (On ‘be worried’ in English, see note 15.) In making this assessment I draw on methodological work in Tonhauser et al. 2013 on projective content cross-linguistically. Following the themes of that work, I asked consultants to imagine they had overheard certain snippets of conversation in Nez Perce; I then asked them whether they would draw particular inferences from the overheard sentences. An example is shown in (33) for lilooy ‘be happy.’ (The glossing and parenthesized English translation are for the ease of readers and were not provided to consultants as part of the task.)

In this example, the consultant was asked both about a matrix-clause inference (concerning the husband) and about an embedded-clause inference (concerning Fido, the dog). Crucially, the matrix clause is negated. As expected for a factive predicate embedded under negation, the attitudinal content of the predicate is negated (the husband is not happy), but the content of the embedded clause escapes negation (Fido did go back); that is, it projects. The latter of these judgments is the crucial data point indicative of factivity. An example showing a similar inference to the complement under matrix negation is given in (34), this time with tim’neeneki ‘be worried.’

In a series of trials, sentences containing the RE verbs lilooy ‘be happy,’ cicwaay ‘be surprised,’ timiipni ‘remember,’ and tim’neeneki ‘be worried’ were presented to consultants in this type of format. In nine cases, the predicate in question was presented in an entailment-canceling environment, whether in the scope of negation (as in the examples above), in a question, or in the antecedent of a conditional. Across all of these trials, consultants endorsed the complement of the RE verb. This indicates reliable projective behavior for the complements of these predicates. As expected, these inferences were also drawn when the predicate occurred in a simple upward-entailing environment (without an entailment-canceling operator). Also as expected, these inferences notably contrasted with control trials using the verb neki ‘think’ (which does not take a relative complement). When the matrix verb is neki, consultants do not make an inference to the embedded-clause content:

Recent work on presupposition is careful to distinguish content that is projective, that is, able to escape entailment-canceling operators, versus content that is familiar/given/discourse old. (See Tonhauser et al. 2013 as well as the specific discussion of factivity in Djärv 2019.) Note that the methodology just discussed assesses only projection—thus in claiming that RE verbs are factive, what I claim is that their complement clause content is projective.

Previous work on REs across languages has reached mixed conclusions concerning the presence of a DP layer outside the CP. On one hand, Aboh 2005 argues that REs in Gungbe are relative CPs lacking an outer DP. Similar conclusions are reached by Haegeman & Ürögdi 2010 for certain English notional complement clauses taken to involve a (covert) relativization structure. On the other hand, Hanink & Bochnak 2017 argues for Washo that a nominal layer is in fact the sole commonality between factive notional complement clauses and relative clauses. In a similar vein, Caponigro & Polinsky 2011 analyzes relative complements in Adyghe as DPs, containing both a noun head and a D layer: the relative CP modifies a noun that is usually silent but that can be made overt. Given the factivity of relative complements in Nez Perce, applying such a perspective to this language would of course dovetail with the idea that factive complements in general involve nominal structure (Kiparsky & Kiparsky 1970, Kastner 2015, Hanink & Bochnak 2017, and Bochnak & Hanink 2021, among many others). At an even broader scale, the idea that all apparent CP-complementation structures involve relativization with a (sometimes null) nominal head is advanced by Kayne 2008, Arsenijević 2009, and Kayne 2014.

There are three primary reasons to favor a CP analysis of Nez Perce REs as opposed to a DP analysis. As we will see, each of these reasons involves a contrast between Nez Perce and certain other languages (notably Adyghe), demonstrating cross-linguistic variation.

Consider first the behavior of REs for case and agreement. For Adyghe, Caponigro & Polinsky 2011 demonstrates that RE structures are treated like DP complements in these respects: as (37) illustrates, the complement itself is marked for case as a DP object would be (here, with the absolutive suffix -r), and its presence triggers ergative agreement for the matrix subject.

The data in Nez Perce are very different: in contrast to true nominal objects, REs never agree, bear case, or trigger ergative on the subject. Recall that transitive clauses in the language feature ergative subjects, accusative objects, and both subject and object agreement on the verb. When both arguments are third person singular, as in (38), agreement takes the portmanteau form pee ‘3/3’ (shown in this example with vowel harmony that yields paa).

REs, by contrast, pattern with formally intransitive clauses: nominative subjects, no accusative objects, and only subject agreement on the verb, never object agreement. The constructed minimal pair in (39) contrasts a typical (formally intransitive) RE sentence, (39a), with an attempt at transitive morphosyntax in the matrix, (39b). In the transitive version, the subject is ergative and the verb bears the portmanteau agreement prefix pee ‘3/3’ (here with pre-glottal shortening). This is never attested, regardless of whether the RE complement contains a nominative relative pronoun (yox̂) or its accusative counterpart (konya).

While Nez Perce does allow notionally transitive clauses with nominative–nominative case frames (and no object agreement), as discussed in section 2, recall that this occurs only when either (a) the object is a weak indefinite or (b) the subject binds the possessor of the object (Rude 1985, Deal 2010a, Deal 2010b). The latter possibility seems unlikely here. Concerning the former possibility, the case–agreement behavior seen with REs would certainly follow if the CPs in these examples occurred inside weak indefinite objects. Weak indefinite objects in Nez Perce (unsurprisingly) cannot be anaphoric (Deal 2010b: 219). This, however, is in marked contrast to the idea in the factivity literature that factive clauses are parallel to definite (or referential, specific, familiar, or given) DPs (Haegeman & Ürögdi 2010, Kastner 2015, Hanink & Bochnak 2017, Bochnak & Hanink 2021). What the pattern in (39) shows is that RE clauses in Nez Perce have a case–agreement profile markedly unlike that of DPs with these semantic properties.

Selection provides a second type of evidence against DP status. A long-standing factor motivating nominal analyses of factive complements in English and other languages is that verbs taking factive CPs (also) allow DP arguments, while non-factive verbs often do not:

Similarly, Caponigro & Polinsky 2011 demonstrates for Adyghe that verbs allowing RE complements also allow ordinary nominal complements. The logic is straightforward: if apparent CP complements/factive complements are always DPs, we expect to see them occurring only in positions where DPs are otherwise licensed. For Nez Perce, however, it is not the case that verbs allowing RE complements consistently allow DP complements. Certain verbs taking REs simply cannot take DPs. The sentences in (41) show that ’eey’s ‘be joyful,’ for instance, is unacceptable with a DP object regardless of whether the object triggers transitive case and agreement.

REs are also possible as apparent complements to the unanalyzable particle qe’ciyeew’yew’ ‘thank you,’ which disallows all nominal complements:

The fact that REs do not have the external distribution of nominals again suggests that they are not DPs.

Finally, a third point indicative of a lack of nominal structure in REs comes from the internal syntax of the relative complement. While some proposals for a nominal superstructure in notional complement clauses have called simply for a DP layer above CP (Krapova 2010, Hanink & Bochnak 2017, and Pietraszko 2019, among others), Caponigro & Polinsky show for Adyghe that an overt noun with a meaning such as ‘news’ or ‘validity/verity/truth’ can occur immediately after the CP, that is, in the position where relative-clause heads typically appear in the language:

Nez Perce is again different. As (44) illustrates, the noun tamtaayn ‘news’ cannot be added to the edge of an RE; indeed, I have not found any noun that may be added in this position.

The overall conclusion is that REs lack both the internal and external syntax of DPs: they do not interact with case and agreement as DPs do, do not pattern with DPs for selection, and are not built on nominal projections as DPs are.

A variant of the DP analysis calling for special attention in light of these arguments is that of Krapova 2010. Krapova analyzes REs in Bulgarian as DPs contained inside a larger, sometimes covert, PP layer. The predicates that allow REs in Bulgarian are emotive factives, like many Nez Perce RE predicates are. In support of the PP analysis, Krapova reports that Bulgarian emotive factive predicates allow REs if and only if they also select the preposition za ‘for.’ She proposes accordingly that Bulgarian REs are really za PPs. Extending this analysis to Nez Perce potentially explains why REs do not have the selectional behavior or case–agreement behavior of DPs: there is a DP layer, but only encased within a larger PP.21 And this approach has some initial plausibility: predicates that allow REs also allow phrases marked with the instrumental suffix -ki (discussed in section 2 as a case marker assigned by a null P head; see (25, 26)). Structures featuring ki PPs receive translations that are reminiscent of the meanings of sentences containing REs. The ki PP, like the RE, specifies what the attitude is about:

There remain two important challenges for a PP approach. First is the problem (again) of selectional mismatch: the distribution of ki PPs with ‘about’ readings is wider than the distribution of REs. REs are impossible with hi ‘say/tell’ and neki ‘think,’ but ki PPs with ‘about’ readings are possible:

Ki PPs with ‘about’ readings have the general behavior of adjuncts: they can appear with any predicate, modulo semantic incompatibility. REs, on the other hand, have the general behavior of arguments: they must be selected by a predicate, and some attitude roots but not others are able to select them.22

The second challenge comes from relative-pronoun marking. In Bulgarian, according to Krapova 2010, the za P can be silent when followed by an RE. However, it may also surface overtly with the relative complementizer deto, forming zadeto, as in (49). In a zadeto RE, we see the PP structure overtly.

Nez Perce grammar offers the opportunity for a similar signal of PP structure, though coming from a different morphosyntactic direction. Nez Perce relative clauses permit case attraction: the relative pronoun may surface in the case assigned to the DP containing the relative clause rather than in the case assigned internally to the relative clause (Deal 2016a). In (50), the relative-clause-internal case is nominative, and the case assigned to the larger DP in the matrix is accusative. The relative pronoun appears in the baseline nominative case in (50a) and in the attracted accusative case in (50b).

Relative pronouns can be case attracted to the instrumental ki case, as (51) and (52) show. Note that (52), just like (51), features a true, noun-modifying relative clause, not an RE.

If REs similarly contain a P head, we expect a similar pattern: the relative pronoun should be able to occur in instrumental case here as well. But case attraction to the supposed hidden P is impossible in REs:

I conclude that, despite its initial attraction, maintaining a PP analysis leads to over-generation both in terms of the distribution of REs (for selection) and in terms of their internal shape (morphological case).

A remaining analytical option is that REs are exactly what they look like: CPs. This provides a straightforward approach to the data that were problematic for the DP and PP approaches. CP objects aren’t expected to render the clause formally transitive (presumably because they lack ϕ features), accounting for the case and agreement facts.24 The absence of nouns is expected; nouns are not freely generated on CP edges. CPs, DPs, and ki PPs simply have different distributions; we might expect CPs to appear in positions in which these other phrases are not licensed and vice versa. And finally, the fact that the relative pronoun must be nominative and cannot be case attracted follows from the fact that CPs do not participate in the case system. Since CPs do not receive case, there is no “external” case assigned that the relative pronoun can be attracted to. These conclusions suggest that the view advanced by Kayne 2008, Arsenijević 2009, and Kayne 2014, according to which all apparent CP-complementation structures involve (potentially hidden) nominal structure, is too strong. The overall picture is instead in line with the CP treatment of REs by Aboh 2005 and Haegeman & Ürögdi 2010.

Let us now consider the internal structure of the relative complement, in particular the question of why and how a relative pronoun and an Ā complementizer appear on the edge of these clauses. The appearance of the complementizer ke, which is found otherwise only in cases of Ā movement to spec-CP, provides strong initial evidence that Ā movement is present in these clauses as well. In this section I argue that REs do indeed involve Ā movement of a relative operator to spec-CP and that the relative operator originates in a high functional projection in the relative clause.

Beyond the morphological similarities between relative clauses and REs, one major reason to posit movement of a relative operator in REs comes from island effects. REs have in common with relative clauses the fact that they are fully opaque for Ā extraction, that is, that they are strong islands. Contrast the baseline declarative in (55a) with attempted extraction of the embedded object, (55b), or of the embedded subject, (55c).

Intransitive subjects also cannot be extracted from REs, as shown in (56b); contrast baseline (56a).

This opacity is of special interest in light of two facts. First, as argued in the previous section, REs are CPs and lack a nominal superstructure. The island effect must therefore result from the structure of the CP itself, rather than from a ban of whatever sort on DP subextraction. Second, Nez Perce generally permits cross-clausal Ā movement. Recall that the language distinguishes RE predicates from non-relative-embedding predicates (a distinction we return to in the next section). Complements of the latter type, which lack the distinctive yox̂ ke of RE, are Ā transparent:25

Such data confirm that it is some particular property of the CP in the RE structure that gives rise to the strong island effect. The obvious candidate for this property is the presence of a relativization dependency, given that relativization generally creates strong islands in Nez Perce (and cross-linguistically).

What is the origin site of the relative pronoun in an RE? This is a question that (like DP vs. CP status) has seen diverging views in previous cross-linguistic work. On one hand, several strands of research have converged on a view of REs as involving movement of a relative operator originating relatively high in the functional structure of the clause. Caponigro & Polinsky 2011 makes this proposal for the visibly relative-like complement clauses in Adyghe discussed in the previous section; Arsenijević 2009 and Haegeman & Ürögdi 2010 reach the same conclusion for cases where the purported relativization structure is more covert. In contrast, Aboh 2005 and Aboh 2010 note for Gungbe that either a direct object or a verb copy may be fronted in relative-like embedding, arguing that these relatively low, lexical elements bear an event-related feature that triggers their relativization:

I will now show that Nez Perce REs behave more like Adyghe than like Gungbe in this respect: what is relativized in this language is a functional element that originates high in the clausal structure, rather than a lexical element originating low.

Note to begin with that whatever is relativized in a Nez Perce RE is not obviously an argument. While the language of course permits arguments to be null inside REs (as it does in general), REs are also perfectly well formed with all of their arguments overt and in situ inside the CP (as well as the verb in a typical position). The examples in (59) show this for intransitive and transitive clauses.

Seeing as all lexical elements remain inside CP, these data suggest that what is relativized is a functional element.

It is also notable that the case of the relative pronoun is always nominative in Nez Perce REs:

This is in contrast to true relative clauses, where the relative pronoun surfaces in a variety of cases, depending (modulo case attraction) on its original position in the embedded clause. The data in (61) show that the relative pronoun is nominative when an intransitive subject is relativized, ergative when a transitive subject is relativized, and accusative when an object is relativized.

As noted in Deal 2016a, nominative is the default case used in Nez Perce for hanging topics, base generated in the left periphery:

A high origin site for relativization in REs predicts the nominative-only pattern of the RE relative pronoun: if the operator that is relativized in REs originates in the left periphery, it will never be within the syntactic domain in which other cases are assigned. Default case (nominative) is correctly predicted to be the only option available for its morphological encoding.

A final piece of evidence for high relativization in Nez Perce REs comes from the position of relative pronouns. As discussed in Deal 2016a, in ordinary relative clauses, Nez Perce relative pronouns freely alternate between two positions (without semantic consequences). One is a position to the left of C, between C and the NP head, as seen in (63a). I will call this the high position. The other is immediately to the right of C, apparently inside the relative clause, as seen in (63b). I will call this the low position.27

I show in Deal 2016a that relative pronouns undergo Ā movement regardless of which position they surface in; in both cases, relative-pronoun movement is unbounded, island sensitive, and subject to Ā intervention. In that work I propose that the two options arise due to cyclic movement of relative operators: the relative operator moves through an Ā outer specifier of TP on its way to spec-CP, and it may be pronounced in either position.28 Thus the difference between the two examples above concerns the pronunciation of copies in a movement chain. As to why movement of the relative pronoun transits through spec-TP, in Deal 2016a I argue based on data from English and other languages that TP is a phase in relative clauses. It is notable, then, that in REs, the relative pronoun does not occur in the low position—only in the high one:

This follows if the origin site of the relative operator is not within the TP domain: the relative operator does not originate low enough to pass through the TP phase and so cannot be pronounced in spec-TP.

In sum, we have seen in this section that REs not only possess the CP-edge morphology of relative clauses but also show the same island behavior. This confirms that the presence of similar morphology at the edge of REs and relative clauses is not due to a relatively trivial similarity such as shared CP status (as posited by De Cuba 2017). We have also seen that what is relativized is not an argument but rather a functional element originating above the TP domain. This conclusion echoes Arsenijević 2009, Haegeman & Ürögdi 2010, and Caponigro & Polinsky 2011. The reader is referred to Caponigro & Polinsky 2011 for an explicit semantic analysis of this high relativization that may also be applicable to Nez Perce.

Having now seen central aspects of the internal and external syntax of REs in Nez Perce, let us return to the question of why some notional complement clauses but not others appear relative-like. Nez Perce presents us with the opportunity to ask the question internally to a single language, as some but not all notional complement clauses use the RE morphosyntactic strategy. The verbs neki ‘think’ and hi ‘say/tell’ strictly reject this strategy. The complements to these verbs have the morphosyntax of ordinary matrix clauses, as seen in (65). I will call this simplex embedding.

With the verb cuukwe ‘know,’ the pattern is similar but with a slight nuance. Unlike RE verbs such as lilooy ‘be happy’ and timiipni ‘remember,’ cuukwe ‘know’ typically occurs without RE morphology, as in (66a); the complement looks like a matrix clause. Consultants did on rare occasions accept an RE with this verb, however, and on one documented occasion produced such a structure, (66b).

It should be emphasized that structures like (66a), with simplex embedding, are extremely well documented, whereas the status of (66b) is more marginal. A conservative generalization would be that neki ‘think,’ hi ‘say/tell,’ and cuukwe ‘know’ are distinct from the RE verbs discussed so far in this article in that they permit their notional complement clause to lack relative morphology.

We can set aside the idea that the simplex-embedding examples above involve juxtaposition of clauses (or high adjunction) rather than subordination, thanks to the same type of Condition C data reviewed in section 3 for REs. In simplex clauses, like in REs, the matrix subject c-commands the embedded subject:

The contrast between simplex embedding and REs also helps to clarify the relationship of RE morphosyntax to factivity. Using the same methodology applied to REs in section 3, we can confirm that cuukwe ‘know’ is factive, even when this verb does not take relative morphology in its complement. In (68), for instance, a ‘know’ report is embedded in a conditional antecedent, and consultants conclude that the complement clause is true.

Factivity cross-cuts the distinction between REs and simplex embedding. This suggests that the factive behavior of certain notional complements in Nez Perce cannot be uniformly attributed to some aspect of RE morphosyntax.

There are two primary perspectives to be contrasted on the difference between simplex embedding and REs. One is that the split between the two classes of clauses is merely at the surface level. This is the perspective of Kayne 2008, Arsenijević 2009, and Kayne 2014: all notional complement clauses are really relative clauses, underlyingly. The other is that the surface split is indicative of a deeper reality: notional complements with visible relative morphology have relative structures, whereas those without this morphology lack relative structures. Some but not all notional complements are relativization based, then, as proposed by Haegeman & Ürögdi 2010, Caponigro & Polinsky 2011, and Haegeman 2012. I will present two types of arguments for this second type of perspective.

The first comes from a body of evidence showing that simplex complements need not contain an Ā dependency. We saw above that these complements do not require (and in many cases, do not allow) relative pronoun yox̂ plus Ā complementizer ke. Furthermore, the complementizer ke does not occur by itself on the edges of these clauses; when this configuration is presented to consultants, it is rejected. Consider first a simple structure such as (69), where no visible Ā dependency is present.

Ke is also rejected on the edges of simplex clauses when Ā movement passes through these clause edges but does not terminate there: