Oblique nominals, a verbal affix and late merge

4.1 Sentience restriction

Sentience refers to real-world or semantic animacy, or the ability to sense or perceive (Speas and Tenny 2003). Note that in Blackfoot sentience is not the same as grammatical animacy: grammatical animacy includes both sentient (e.g., ‘boy’) and non-sentient (e.g., ‘wagon’ or ‘blanket’) entities as discussed in Section 2.

Sentience plays an important role in various parts of Blackfoot grammar. In particular, a sentience restriction in the language has been recognized to be a defining property that distinguishes nominals that carry a grammatical function - subject or primary object - from those which do not, such as adjuncts (e.g., Frantz 2009; Kim 2015a; Ritter and Rosen 2010; Wiltschko and Ritter 2015): subjects and primary objects must be sentient, unlike adjuncts. In recent studies on Blackfoot, the sentience restriction has been proposed to be a part of narrow syntax, represented as syntactic agreement (Kim 2017b; Wiltschko and Ritter 2015).^[15] In light of these studies, the sentience restriction is an important criterion in Blackfoot to determine whether a given oblique nominal is a primary object or an adjunct. The nominals introduced by the benefactive suffix and by linker Ps show a difference in the sentience restriction. Beneficiaries must be sentient (Bliss 2007; Frantz 2009), unlike nominals introduced by a linker P.

First consider the beneficiary nominal illustrated in (10a). As a benefactive suffix, -o(mo) is a type TA final, so the primary object of the derived applicative verb is predicted to be grammatically animate. In accordance with this prediction, in (10a), the beneficiary ‘the baby’, which is animate, is introduced by the suffix -o(mo). If the beneficiary is grammatically inanimate, as is ‘the bed’ in (10a), the sentence becomes ungrammatical.

Without the suffix -omo, a beneficiary nominal cannot be introduced, as shown in (10b). The verb root in (10b) is the same as in (10a); however, it is in a TI form with the TI suffix –ii, and lacking -omo.^[16] The fact that it is in a TI form is supported by the data in (10c): the verb only allows an inanimate object noun such as ‘the leather’, but not an animate object such as ‘the blanket’. As a TI verb without the beneficiary suffix -omo, it cannot have a beneficiary nominal, regardless of its animacy status, as shown in (10b): neither animate ‘my daughter’ nor inanimate ‘the chair’ is allowed.

Importantly, the range of possible nominals allowed by the suffix is more restricted than would be predicted by its status as a TA verb. The beneficiary introduced by the suffix must be sentient (Bliss 2007; Frantz 2009). That is, nouns which are grammatically animate but semantically inanimate are not allowed as beneficiaries. This is illustrated in (11a) with a newly elicited example.

Here, the beneficiary is the grammatically animate noun ‘wagon’ (11a). Despite its animate classification, it is not sentient; thus, the sentence is ungrammatical. That the noun ainaka’si ‘wagon’ is a grammatically animate noun is shown by the fact that the noun is allowed as the object of a TA verb (11b), but not of a TI verb (11c).

By contrast, oblique nominals introduced by linkers do not show this type of restriction; linkers allow either sentient or (in)animate nominals. As illustrated in (12) with the instrument linker, either the sentient ‘John’ or the inanimate ‘soup’ is allowed. Moreover, a grammatically animate but non-sentient instrument ‘the knife’ is also allowed.

The associative linker ohp- shows the same pattern. As shown in (6b) earlier, it can introduce a sentient nominal (‘the boy’). As shown in (13), it can also introduce a grammatically animate noun (‘vase’, 13a) or an inanimate noun (‘water’, 13b).

4.2 Person prefix marking

In Blackfoot, verbs can be marked with affixes that indicate the person value of their primary object (Frantz 2009). In particular, first, second, and third persons are indicated by prefixes (nit-, kit-, ot-, respectively) on verbs. A beneficiary introduced by the suffix –o(mo) can appear as a person prefix as shown in (14a), which suggests that it is a primary object. The person prefix can co-occur with an optional full pronoun such as niisto for added emphasis, as shown in (14a). However, the sentence is ungrammatical if this beneficiary is expressed solely with the pronoun, entirely lacking the person prefix (14b).

In contrast, nominals introduced by linkers cannot be expressed via person prefixes, unlike beneficiaries introduced by the suffix -o(mo). This is illustrated in (15a): when the associative linker introduces the associate ‘me’, the person prefix nit- as the expression of the associate results in ungrammaticality. (15b) shows the grammatical variant of this sentence, where the associate is expressed via an independent pronoun ‘I’, niisto.^[18]

The contrast shown in this section between the beneficiary morpheme and the linkers indicates that the nominals introduced by linkers cannot be treated in the same way as an applied beneficiary, i.e., a primary object.^[19]

4.3 Theme (DIRECT/INVERSE) marking

In Blackfoot, like all other Algonquian languages, the direction of the action described by the verb is indicated by theme markers, as was briefly mentioned in Section 2: by either DIRECT or INVERSE morphology (Frantz 2009) (cf. Table 1 below). DIRECT-INVERSE systems make reference to a person scale such as that shown in (16), which is simplified for the purpose of this paper. If the direction of action is from a 1st/2nd person to a 3rd person (1 > 3 or 2 > 3) or from a 1st person to a 2nd person (1 > 2), the verb is marked as DIRECT. If the direction is the opposite, from a 3rd person to a 1st/2nd person (3 > 1 or 3 > 2) or from a 2nd person to a 1st person (2 > 1), then the verb is marked as INVERSE. As theme markers encode grammatical relations between two entities in the clause, they appear obligatorily only on transitive verbs such as TA or TI verbs, but they cannot appear on intransitives, i.e., AI or II verbs. Unlike TA verbs, however, only a DIRECT theme marker is allowed with TI verbs, while an INVERSE theme marker is not allowed.^[20]

Table 1 below provides a summary of the theme suffixes (in bold) for TA forms for the simplified person scale in (16). The examples in this section are all in TA forms, and thus theme suffixes for the TI paradigm are not presented. In the table, the pronominal prefixes are indicated in brackets: 1st person nit-, 2nd person kit-, and 3rd person ot-. These prefixes are in brackets as they do not appear when the relevant person is not in a pronominal form. Note that in Blackfoot only one person prefix can appear per verb, and the 2nd person prefix always takes precedence over other person prefixes (Frantz 2009), as reflected in Table 1 (see (17a) for a concrete example).^[21] As for the action between 1st person and 3rd person pronominals, it is the 1st person that takes precedence over the 3rd person regardless of the direction of the action; thus, the 3rd person prefix does not appear in any case presented in the table.

Consider (17a) where the TA verb has the DIRECT theme marker -a. This marker appears because the 1st person (‘I’) acts on a 3rd person (‘the boy’). In contrast, in (17b), the second person (‘you’) acts on the 1st person (‘I’); thus, the INVERSE marker -ok appears on the verb. In this particular case, the verb also has the suffix -i which indicates that the object is in 1st person; this is preceded by the INVERSE marker (Frantz 2009).

Like the direct object of the verb in a transitive clause as shown in (17), the nominal introduced by the suffix -o(mo) must be indexed by a theme marker as shown in (18). The third person subject ‘John’ acts on the 1st person beneficiary; consequently, the verb is marked with the INVERSE marker -ok. Without this INVERSE marker, the sentence is ungrammatical.

Recall that a verb suffixed with the beneficiary morpheme is a TA verb (Frantz 2009). TA verbs require that their object be grammatically animate; thus, the beneficiary ‘me’ must be analyzed as the primary object in (18), rather than the inanimate theme ‘bread’. Accordingly, the INVERSE marker indicates the direction of action between the subject ‘John’ and the animate beneficiary, not the inanimate theme.

Unlike the beneficiary in (18), a nominal introduced by a linker cannot be indexed by a theme marker. For example, in (19a), the animate DP anna si’kaan ‘the blanket’ is introduced by the instrument linker oht-, and the direction of the action is from a 2nd person (‘you’) to a 1st person (‘me’). This is indexed with an INVERSE marker –ok, and the sentence is grammatical. In this example, the 2nd person subject appears as a prefix to the verb, and the 1st person object is indicated by the suffix –i, similar to the example in (17b). As the instrument ‘the blanket’ is grammatically animate in the language, it might be considered an object of the TA verb and indexed by the theme marker.^[23] If this were the case, the sentence in (19a) would be marked with a direct theme marker –a, and the 3rd person instrument would have to agree with the verb, indicated by a 3rd person suffix such as –wa, as illustrated in (19b).^[24]

In (19b), the intended meaning is such that the direction of the action is from the 2nd person (‘you’) to the 3rd person animate instrument (‘the blanket’); however, the sentence is ungrammatical with the intended meaning, suggesting that an instrument nominal introduced by a linker cannot be indexed via a theme marker. The grammatical meaning of this sentence where the direct object ‘someone’ is indexed by the theme marker is indicated for comparison.^[25]

4.4 Suffixal agreement

In Blackfoot, verbal suffixes on TA/TI verbs indicate the person and number of objects including beneficiaries, and those on AI verbs mark these features of the subject (see Section 2). For example, for ordinary TA and TI verbs where a direct object appears, the suffix -wa indicates that the direct object is in 3rd person singular and the suffix –yi indicates that the direct object is in 3rd person plural. In a ditransitive clause such as beneficiary clauses, the same forms of the suffixes indicate the number and person of the beneficiary, i.e., the primary object, not the theme nominal, a secondary object. Following Frantz (2009) and Ritter and Wiltschko (2014), I assume that this type of suffix is an agreement morpheme. Consider the beneficiary clause in (10a), repeated as (20) below. In this example, the verb ‘sewed’ shows suffixal agreement with the beneficiary ‘the baby’, as indicated by the 3rd person singular marker -wa. It does not agree with the secondary object, ‘the blankets’, which is plural, as the ungrammaticality of the third person plural agreement marker –yi shows.

Nominals introduced by linkers, however, do not show suffixal agreement with the verb, as illustrated below with the instrument linker (21). In (21), the plural nominal ‘sticks’ is introduced by the instrument linker oht-, while the theme nominal of the verb ‘hit’ is singular ‘John’. The verb shows agreement with the theme, as indicated by the 3rd person singular suffix -wa. If the verb agrees with the instrument, with the 3rd person plural marker –yi, the sentence is ungrammatical, as indicated in (21). The ungrammaticality of the suffixal agreement shown in (21) suggests that an oblique nominal introduced by a linker cannot be a primary object.

The data provided in this section demonstrates that oblique nominals introduced by linker prefixes cannot be primary objects, unlike those introduced by the benefactive suffix –o(mo). The results of this section also suggest that oblique nominals introduced by a linker P cannot be analyzed in terms of Appl, unlike beneficiary nominals in the language or the applied objects in the previous literature (see Section 3). The differences between the nominals introduced by the linker and the benefactive morpheme discussed in this section are summarized in Table 2 with the addition of the differences between them noted in Section 3.

Building on the data presented in this section, I propose that linker PPs are adjuncts, as with the previous studies that identified oblique PPs in Bantu-type languages as adjuncts for similar reasons (see the discussion in Sections 1 and 3).

A remaining question is how to account for the fact that oblique nominals of linker Ps do not show primary object properties even though the linker appears as if it forms a complex verb, much like the complex verb formed with the applicative affix. Addressing this question is the topic of the next section.

5.1 Late merge of linker PP at the phase level

In addressing the question raised at the end of Section 4, I propose a late merge analysis of linker PPs, taking the view that adjuncts are integrated into a given derivation late, unlike non-adjuncts such as objects, as with Lebeaux (1988) and Stepanov (2001). In particular, I adopt Stepanov’s (2001) proposal for the late merge of adjuncts laid out in the framework of Chomsky (1995, 2000, in which the derivation proceeds by Merge via feature checking. Assuming two types of Merge, namely adjunction and non-adjunction Merge, within Chomsky’s system, it is proposed that a choice between these two Merges depends on the featural makeup of the syntactic object that is undergoing Merge.^[27] A syntactic object labeled α merges with another syntactic object β via adjunction only when α has no uninterpretable features to be checked. More specifically, lacking uninterpretable features, it cannot participate in the derivation cyclically but merges late after the cyclic derivation has completed. On the other hand, α merges with ß via non-adjunction Merge only when α has uninterpretable features to be checked. Being a non-adjunct, in this case, α undergoes cyclic derivation checking its uninterpretable features.^[28]

In Section 4, linker PPs were demonstrated to be adjuncts lacking a range of properties of primary objects, unlike applied objects in the language. I propose that the lack of object properties reflects the lack of relevant uninterpretable features on linker PPs, as assumed in minimalist syntax (see also Footnote 28). Lacking the relevant features, linker PPs merge late via adjunction. Casting Stepanov’s (2001) proposal into an Agree-based model (Chomsky 2000, 2001), I further propose that linker PPs, lacking relevant uninterpretable features, cannot participate in the derivation via Agree, but instead merge late.^[29] Assuming strongly cyclic syntax where the derivation proceeds by phases (Chomsky 2008), I propose that the specific point of the derivation where adjuncts such as linker PPs late merge is the end of a given phasal derivation (see Nissenbaum 2000 for a similar proposal for adjunct clauses generally). As for linker PPs, being modifiers of the event denoted by the verb phrase, they merge late at the end of (both agentive and non-agentive) vP derivations (e.g., see (25c)), assuming that (both types of) vP are phasal (Chomsky 2000; Marantz 2001; Legate 2003).^[30]

Now I turn to illustrating the proposed account using data with linker Ps. Consider the data in (22), where the instrument linker oht- introduces the DP ‘those arrows’. Unlike non-adjuncts (e.g., objects) that participate in the syntactic derivation cyclically, the linker PP is built up separately (“on a different plane” (Chomsky 2001)), as in (23a), from the cyclic derivation of the rest of the verbal complex, vP (23b). The head of vP is the locus of the final suffix (e.g., TA) (see Section 3). Thus, the derivation of the linker PP (23a) is separated from the cyclic portion of the derivation of vP (23b). Within vP (23b), the verb stem V remerges to v resulting in the order of V-v, similar to the familiar verb movement to v in English (see Section 5.2 for detailed discussion), which is also consistent with other studies on Blackfoot (e.g., Ritter and Rosen 2010).

(23)

vP being phasal, the adjunct PP (23a) merges with vP via adjunction, as shown in (23c), once vP is built up as in (23b). Assuming that the linearization is established phase-by-phase when spell out applies (Fox and Petsetsky 2005; Matushansky 2006; Nunes and Uriagereka 2000; Uriagereka 1999), in (23c), spell out applies at phasal vP which results in the linearization of the linker P with the relevant elements. As with Julien (2002), I propose that linker Ps linearize in their base position with the verb complex [V-v], resulting in their surface appearance as prefixes, as in a [P-V-v] complex, oht-yissino’to.TA.^[31] Through an investigation into affix orders in genetically unrelated languages, Julien (2002) has shown that prefixes may not be the result of the traditional version of incorporation via head movement (Baker 1985, 1988), but result from linearization in their base position with other relevant syntactic materials.^[32] Taking the view of Julien (2002), I propose that linker Ps, on the surface, look as if they are incorporated into the complex [V-v] to which they are adjacent. Under this view, linkers do not undergo head movement that results in incorporation, unlike applicative affixes which undergo incorporation via head movement (see Sections 1 and 3). Consequently, a linker does not form a complex verb with [V-v] and the lack of primary object properties shown by linker PPs is not an unexpected outcome.

In conclusion, oblique nominals introduced by linker Ps are unlike applied nominals in that the introducer, a linker, does not undergo incorporation via head movement. They also differ from oblique adjunct PPs in other languages in that the linker has the surface appearance of an affix to the verb. The proposed late merge analysis captures these properties of linker PPs, which were left unaccounted for in the previous analysis of incorporation via head movement or the adjunct analysis discussed in Section 3.

5.2 The position of a linker P

In the previous section, a linker P initially merges with its complement, and it is proposed that in the course of the derivation the linker P linearizes with the [V-v] complex. However, it remains unclear how it comes to appear as a prefix on the verb, separated from its complement nominal. The affixal status of the linker P is important with respect to the main question raised by this paper (e.g., Sections 1 and 4): linkers in Blackfoot are verbal affixes similar to applicative affixes in other well-known languages, but the oblique nominals unexpectedly turn out to be adjuncts, which are usually introduced by independent Ps across languages. Thus, the current proposal should account for how linkers appear as verbal affixes, not as independent Ps.

I propose that the observed surface position of linker P on the verb, displaced from its initial merging position, can be accounted for by further remerge of [P-V-v] to higher heads, Infl and O(uter)-Asp, projected in the extended projections of verbs in the language. As will be discussed shortly in this section, I further propose that this remerge is motivated by Agree relationships between V and other higher functional heads in the extended verbal projection.

Before discussing the proposed account, I introduce an assumption about the extended verbal projection for Blackfoot (e.g., Bliss et al. 2014; Ritter 2014), as illustrated in (25), where irrelevant details are omitted. Recall that v is the locus of the final suffix (see Section 2); above vP is O-AspP, which is the locus of theme markers (Bliss et al. 2014) – a DIRECT or INVERSE suffix as discussed in Section 4.2.^[33] Above O-AspP is InflP instantiated by an order suffix, –hp or a zero morpheme (Ritter and Wiltschko 2014).^[34] In (24), a null order morpheme appears between the DIRECT theme marker –a and the 3s suffix –wa, which has not previously been indicated in this paper. I only indicate it in this example and relevant examples in this section for convenience. Also, note that person prefixes such as nit- appear in the specifier of IP following Ritter and Wiltschko (2014), which will play a role in accounting for the morpheme order between linkers and the [V-v] complex discussed later in the section.^[35] This is sufficient detail to understand the remerge of [V-v] to O-AspP and Infl, and I do not discuss in detail how Infl and O-AspP are the loci of the relevant morphemes in the language, as these details are not relevant to the current discussion (see Footnotes 33 and 34 for a brief discussion).

Now I turn to the proposal of remerge of V into the higher functional heads, v, O-Asp, and Infl. Within minimalist syntax (e.g., Chomsky 1995, Chomsky 2001), although details differ, classical incorporation phenomena (in the sense of Baker 1985) have been viewed as involving feature checking (e.g., Julien 2002; Pesetsky and Torrego 2001; Matushansky 2006; Roberts 2010). For example, as schematically presented in (26), the uninterpretable feature of a probe X is checked by a goal Y with the matching interpretable feature in its c-command domain. As a consequence of the feature checking, the remerge of goal Y to X takes place.

Likewise, I propose that the remerge of V in Blackfoot into higher heads is driven by an Agree relation between V and the higher functional heads v, O-Asp, and Infl. In particular, I propose that the remerge of V to the higher functional heads checks an uninterpretable c(ategorial)-feature, as shown in many studies such as Julien (2002) and Matushansky (2006). Like these studies, I assume that c-features can appear on both functional and lexical heads. A head can c-select the syntactic category of its complement, and the relevant c-feature of the selecting head is the uninterpretable counterpart to the c-feature of its complement, e.g., [uV] for v that c-selects VP. Moreover, the c-selected head of the complement has an interpretable matching c-feature, e.g., [iV] for V. Importantly, in a case where more than one functional head appears in the extended projection, the lexical head of the projection determines the c-features of those functional heads (Julien 2002). What this means for Blackfoot V and its extended projection is that the functional heads, v, O-Asp, and Infl, all share a [uV] feature, and the remerge of V takes place all the way upward, as illustrated in (27). Put differently, in (27), the probes v (27a), O-Asp (27b), and Infl (27c) establish Agree relations with V at each derivational stage: the [uV] feature on these probes are checked by an [iV] feature and thus the remerge of V into v (27a), O-AspP (27b), and Infl (27c) sequentially takes place.

In (27a), linearization of the linker P and the [V-v] complex takes place, resulting in [P-V-v], as discussed in the previous section. Subsequently to the linearization, [V-v] remerges to O-AspP along with P, as in (27b). Here, curly brackets around the linker P are employed in order to indicate that its displacement is not a usual remerge resulting from feature checking. In the next step of the derivation in (27c), [P-V-v-O-Asp] remerges to Infl. I propose that the remerge to O-AspP shown in (27b) yields the stranding of the complement of the linker. Thus, the surface position of the linker appears to be a by-product of an Agree relation between a higher functional head (i.e., O-Asp) and the [V-v] complex to which the linker PP merged.

The current paper proposes that linker PPs merge at vP, a projection lower than IP. Upon spell out, the linker P linearizes with a [V-v] complex and at the next step of the derivation, the [P-V-v] sequence remerges to Infl (27c) via O-AspP (27b). The prediction of this proposal is that a person prefix that merges in the specifier of IP (see (25/27c)) (e.g., Déchaine and Wiltschko 2014; Ritter and Wiltschko 2014 among others), if present, will always precede the linker, assuming that the person prefix is also the result of spell out in its base position (Julien 2002) (see Section 5.1). This prediction is borne out by the data, as shown in (24): the 1st person prefix nit- appears before the sequence of [P-V-v]. As linker Ps move to the higher Infl through O-Asp as a part of [P-V-v], linker Ps will always appear after a person prefix, which appears in the specifier of IP (Ritter and Wiltschko 2014), as in (24) (see also (27c)). Moreover, the proposed account rules out the incorrect order in (28) where a linker (iiht-) precedes a person prefix (nit-). In other words, if the linker PP were an adjunct to a higher phrase than vP, e.g., an IP, (28) would be a correct order, contrary to fact.^[36]

In the sections to follow, I rule out the possibility of linker PPs merging lower than vP. For example, the morpheme orders observed in (24) could be claimed to result from the linker PP merging between vP and VP; I demonstrate using evidence from abstract nominalization and the distribution of idioms in Blackfoot that linker PPs cannot be added to a lower phrase than vP, such as I(nner)-Asp between vP and VP. This further supports the account proposed in this section.

6.1 Abstract nominalization

Abstract nominalization in Blackfoot shows that linker PPs must appear above I-AspP. Abstract nominalization in Blackfoot is one of the several nominalization types available in the language, and it targets intransitives only, such as AI verbs (Bliss et al. 2012; Frantz 2009).^[37] Abstract nominalization is morphologically marked with -n or –hsin: the first variant attaches to stems ending in –aa, and the second allomorph appears elsewhere. Abstract nominalization indicates either the state or process described by the verb (Frantz 2009). An example of abstract nominalization is illustrated in (30).^[38]

In a recent study of abstract nominalization in this language (Ritter 2014), it is shown that the source of the nominalization is I-Asp (in the sense of Travis 2010), located between vP and VP, as represented in (31).

(31)

Abstract nominalization will be contrasted with another type of nominalization in the language, namely conjunct nominalization (glossed as CN), which is illustrated in (32) (Frantz 2009). Unlike abstract nominalization, conjunct nominalization is proposed to target IP (Bliss 2014), which contains a vP, which this paper also assumes.

The proposal that abstract nominalization targets I-Asp (31) predicts that elements outside I-Asp cannot undergo abstract nominalization, while elements inside I-Asp can. In particular, regarding the linkers, it would be predicted that they should not undergo abstract nominalization if they appear outside I-Asp. In contrast, under the proposed analysis of this paper, the linkers are predicted to be grammatically able to undergo conjunct nominalization, as they are adjoined to vP contained within IP.

The predictions are borne out by the data. First, with respect to abstract nominalization, linker PPs, and just linker Ps, are ungrammatical with abstract nominalization, as shown in (33–34), which are nominalizations of (12) and (6b), respectively. Note that this is the case regardless of whether an oblique nominal is absent or present.^[39]

This ungrammaticality suggests that linker PPs attach to a phrase higher than I-AspP, supporting the proposed analysis of linker PPs as a vP adjunction. Non-linker PPs contrast with linkers in this respect. They are grammatical with abstract nominalization, as shown in (35) with the non-linker waamis ‘up’. This grammaticality suggests that non-linker PPs attach to a phrase below I-Asp, unlike linker PPs.

With respect to conjunct nominalization, linkers are grammatical, as shown in (36). The grammaticality of linkers in conjunct nominalization follows from the proposed analysis in this paper: as a linker PP adjoins to vP, it is contained within IP; therefore, nominalizing the IP that contains the linker PP yields a grammatical result, as shown in (36).^[40]

To summarize, evidence from abstract nominalization indicates that linker PPs appear above I-AspP (which is lower than vP), in contrast with non-linker PPs, which appear below I-Asp; this is represented in (37). Together with the evidence that linker PPs cannot be IP-adjuncts as discussed in Section 4, the conclusion illustrated in (37) provides further support for the analysis proposed in this paper that a linker PP merges at vP, which appears higher than I-AspP (as an adjunct).

Furthermore, the positions of linker and non-linker PPs as proposed in (37) predict that they can co-occur. This is borne out as illustrated in (38). In (38), the non-linker waamis- ‘up’ appears with the instrument linker oht- and the sentence is grammatical.^[41]

6.2 Idioms

In this section, building on evidence from idioms, I show that non-linker PPs appear VP-internally, while linker PPs appear above I-Asp, consistent with the conclusion drawn from abstract nominalization. This is represented in (39).

A generalization drawn from studies on idioms in the verbal domain is that material outside VP does not generally form part of an idiom with the verb, while material inside VP commonly tends to do so (e.g., Laron 1988; Marantz 1997). For example, Marantz (1997) proposes that agents outside VP (introduced by v/Voice) usually do not take part in idiom formation, while objects inside VP easily form idioms with the verb. Furthermore, recent approaches to idioms (e.g., Harwood 2015; Kim 2015b; Svenonius 2005) suggest that this is not merely a distinction between agents and objects, but that the distinction can map onto a phasal boundary. For example, a phase head such as v can be a boundary for idiom formation; material inside v can participate in idiom formation, while material outside v does not usually participate in idiom formation.^[42]

To illustrate, I discuss Slavic prefixes, which to some extent parallel Blackfoot’s linkers and non-linkers. Slavic is known to have two types of prefixes: lexical and superlexical. These largely belong to the class of prepositional elements (see Svenonius 2004, 2005 and references therein). Examples of lexical and superlexical prefixes are illustrated with za- in Russian in (40). It is used as a lexical prefix meaning ‘into’ in (40a), and as a superlexical prefix in (40b). Under this latter usage, the prefix has the aspectual meaning ‘begin to’; thus, the verbal complex has the meaning ‘starting to throw’.

Putting aside other properties of lexical prefixes (see Svenonius 2004 for details), these prefixes easily form idiomatic expressions, as illustrated with a Russian example in (41).

On the other hand, it is rare that superlexical prefixes form idiomatic expressions (Svenonius 2004). The typical meanings of superlexical prefixes are transparent, as shown in (42).

Building on the idiomatic patterns of these prefixes (and other properties; see Svenonius 2005 for details), Svenonius proposes that lexical prefixes merge below the phase head, v (e.g., in VP), while superlexical prefixes are realized above the phase head v (e.g., in AspP). Thus, the pattern of idioms involving prefixes in Russian supports the view that a phasal head such as v can be a boundary for idiom formation.

If the phasal boundary for idioms in Blackfoot is also a v head, to whose projection linker PPs adjoin (see Section 5), the positions of linker PPs (above v) and non-linker PPs (below v), as proposed in (39), predict that linker PPs may not belong to idioms, being outside the boundary for idiom formation, i.e., v, and, conversely, that non-linker PPs may form idioms with the verb, being inside the boundary. This type of split in idiom formation patterns is exactly what is found with linkers and non-linkers; in Blackfoot it is easy to find idioms formed with non-linkers but this is not the case with the linkers. The Blackfoot dictionary (Frantz and Russell 1995) shows numerous idioms that consist of a non-linker and a verb, some of which are illustrated in (43). Strikingly, in the dictionary, there are no idioms formed with linkers.^[43]

Building on the fact that non-linkers appear below I-AspP and that they can form idioms with the verb, I propose that non-linkers appear below v, inside the VP as in (39). The lack of idioms containing linkers supports the account proposed in this paper in which linker PPs adjoin to vP above the phase head v, the boundary for idiom formation.^[44]