In intransitives, the peripheral suffix indexes the animacy, number, and obviation of third person subjects. If the subject is not third person, there is no peripheral suffix.

While it may look like the 1exc agreement markers in the (a) examples are in complementary distribution with the peripheral suffix, they actually occupy a different slot in the verbal template—the central suffix, typically analyzed as occupying Infl/T (Halle & Marantz 1993; Coon & Bale 2014; Oxford 2014; a.o.). Below, we’ll see forms that feature both the central as well as the peripheral suffixes.

In configurations involving a third person subject and a part object, the peripheral suffix indexes the subject.10

In the (a) examples we have singular subjects and singular peripheral suffixes; in the (b) examples we have plural subjects and plural peripheral suffixes—thus, C varies according to the features of the third person subject. So far, all the data we’ve seen so far seems reducible to simple locality: C indexes the highest argument, and only happens to have overt exponents for third persons.

However, in configurations involving a part subject and a third person object, the peripheral suffix indexes the object. The probe skips over a part subject in preference for a third person.11 This is the core property we would expect from an omnivorous probe: ignoring more local potential goals in favor of more distant ones.

In these examples, we can see that varying the number of the third person object causes the peripheral suffix to vary accordingly. Thus, the peripheral suffix always indexes third person features, no matter whether they come from the subject or the object—the characteristic property of omnivorous agreement.

That brings us to the end of the basic only-one-third-person forms. To summarize: whenever you have only one third person argument, the peripheral suffix agrees with it, skipping over first and second persons if necessary. This is precisely the behavior one would expect from an omnivorous third person probe.

But wait!—the staunch 3noP-er might say. How do we really know that C is skipping over the subject to agree with the object in part>3 configurations? What if the object actually has moved over the subject in those configurations (and only those) prior to C probing, thus rendering it the most local goal to C? Here’s a sketch of what one could propose:

If that was right syntax of mixed configurations (configurations that mix part and third person arguments), then we wouldn’t need to say anything special about C: it would just agree with the most local goal, and happen to only have overt exponents for third person (similar to English 3sg -s).

I don’t think this is the right syntax of mixed configurations. Firstly, if we try to formally work out a syntax that derives the behavior in (13), it’s strikingly difficult to avoid replicating a need for syntactically-active third person features elsewhere. We would need to have a probe somewhere that would move the object over the subject in part>3 configurations. For sake of argument, let’s place it in Voice, the head that introduces the subject. We need Voice to probe down and move the object in part>3 configurations, but crucially to avoid doing so in part>3 configurations. It’s not immediately clear to me how we could derive this state of affairs without the probe being specified to move third person DPs only.12

But even if there were some non-stipulative way to derive (13) without having to make reference to third person features elsewhere in the syntax, cross-Algonquian data points against this picture. While I am not aware of clear evidence relevant to the question of what’s going on syntactically in part>3 configurations in Blackfoot or Plains Cree (or any of the other languages with the same agreement pattern), there is evidence from Odawa (a variety of Ojibwe, Central Algonquian), Listuguj Mi’gmaq (Eastern Algonquian) and Passamaquoddy-Wolastoqey (Eastern Algonquian) that these contexts lack object inversion (Rhodes 1994; Hamilton 2018; and Grishin 2022a; respectively). Strikingly, mixed configurations in those languages also show the same omnivorous third person pattern, with C always indexing the third person argument. While it is conceivable that Blackfoot and Plains Cree could diverge from these languages in the syntax of mixed configurations, there is as of yet no syntactic evidence from any Algonquian language for a syntax of mixed configurations like that in (13). Therefore, at least as a null hypothesis, it seems reasonable to assume parallel syntactic behavior across the family in this domain, especially given the fact that C agreement in mixed configurations is strikingly uniform across Algonquian (Xu 2020; 2021), and thus cannot be used as justification for a distinct syntax of mixed configurations in some Algonquian languages.

The data involves long distance agreement (LDA) constructions that show locality effects: the matrix attitude predicate can only agree with the highest argument in the embedded clause. In Odawa and Mi’gmaq, the relevant LDA construction involves complements in the conjunct order (a distinct agreement paradigm from the one this paper focuses on) of verbs like ‘know’, and in Passamaquoddy-Wolastoqey the relevant LDA construction involves complements in the subordinative mode (a subtype of the independent order that lacks C agreement) to verbs like ‘want’.13 I exemplify below:

In each example above, the embedded clause involves a part subject and third person object. The grammatical examples in (14) show the matrix verb agreeing with the part subject as if it were the object of the matrix clause. In contrast, the ungrammatrical examples in (15) involve the matrix verb agreeing with the third person object. Thus, we seem to find a locality effect here, with the matrix verb only able to show object agreement with the highest embedded argument—the part subject, not the third person object (see Rhodes 1994; Hamilton 2018; Grishin 2022a for detailed argumentation that the agreement pattern is really sensitive to syntactic height, and not e.g. hierarchical agreement or sensitivity to base-generated positions).

While it is true that Odawa, Mi’gmaq, and Passamaquoddy-Wolastoqey do not show the same peripheral agreement pattern as Blackfoot or Plains Cree across the board, they strikingly behave exactly the same as Blackfoot and Plains Cree (as well as the rest of the Algonquian family) when we restrict our attention to configurations with only one third person argument: C agrees with the third person, no matter if it’s the subject or object (see Xu 2020; 2021 for discussion of the cross-Algonquian variation in peripheral agreement):16

In the (a) examples, we see peripheral agreement indexing the third person object; in the (b) examples, we see peripheral agreement indexing the third person subject. This exactly parallels the behavior we see in Blackfoot and Plains Cree—I conclude that across Algonquian, peripheral agreement shows omnivorous third person behavior. I wait until Section 6.2 to discuss the variation we do see in Algonquian.

Thus, the 3noP-er’s syntax for mixed configurations seems unlikely, not only because it seems difficult to implement without just replicating a third person probe elsewhere in the syntax, but also given the comparative Algonquian picture. Though we don’t (to my knowledge) have relevant evidence from the literature on Blackfoot and Plains Cree, the data from Odawa, Listuguj Mi’gmaq, and Passamaquoddy-Wolastoqey strongly suggests that there is no object inversion in mixed configurations. Further research is required to develop relative c-command diagnostics for subject and object in mixed configurations in Blackfoot and Plains Cree.

Let’s turn back to the agreement data in Blackfoot and Plains Cree. In contexts with two third persons in direct configurations (ones where the subject outranks the object; prox>obv, an>in), the peripheral suffix indexes the subject.

The (a) examples feature singular subjects, and the (b) examples feature plural subjects—peripheral agreement varies accordingly. Additionally, since the peripheral suffix is only able to index the number of the subject, the number of the object is ambiguous. Thus, we learn that the peripheral suffix prefers indexing the more local goal—a familiar locality effect that is appropriately captured by a third person probe.

Alternatively, with two third persons, we can get an inverse configuration (object outranks the subject; obv>prox, in>an). In this case, the peripheral suffix indexes the object.

Here, the (a) examples feature singular objects, the (b) examples feature plural objects, peripheral agreement varies in turn, and the number of the subject is ambiguous. These cases may seem to fly against the picture we’ve been building up so far: why are we skipping over the subject to agree with the object? The subject is third person, and thus should satisfy the probe’s needs.

Locality is restored once we realize that the 3>3 inverse crucially involves A movement of the object over the subject, following Bruening (2001; 2005; 2009), Bliss (2013), Hamilton (2015; 2018), Oxford (to appear), and Grishin (2022a), among others. The only assumption we need to make here is that this step of A movement happens before C probes, something that seems quite reasonable given C’s structural height and given the fact that CP is typically an Ā domain.

To show that the inverse, but not the direct, involves A movement of the object over the subject, we can use the classic variable binding diagnostic: A movement should be able to feed variable binding. Put another way, while Ā movement of a quantifier can trigger Weak Crossover (WCO) effects, A movement of a quantifier does not (Postal 1971; Wasow 1972; Koopman & Sportiche 1983; a.m.o.). In both Blackfoot and Plains Cree, the 3>3 inverse reverses binding relations in exactly the way we’d expect from A movement.

Bliss (2013) demonstrates that in direct configurations in Blackfoot, universal quantifier subjects can bind into objects but not vice versa (22), whereas in the inverse, universal quantifier objects can bind into subjects but not vice versa (23). First let’s take a look at the direct configurations:

In (22a) we have a proximate universal quantifier subject, ‘every girl’ (amoksi aakííkoaiks ‘the girls’ plus the quantificational prefix ohkana- on the verb) binding a variable contained inside a relative clause modifying the obviative object, omi sááhomaapii otáákomimmayi ‘the boy she loved’. This is acceptable under the bound reading, which is perhaps what one would expect given that subjects are first-merged c-commanding objects. In (22b), on the other hand, we are attempting to have an obviative universal quantifier object ‘every girl’17 bind into a proximate subject anna sááhomaapi otáákomimmoka ‘the boy that she loved’. The bound reading here is not available—only an unbound reading is.18 This indicates that in direct configurations, subject and object retain their first-merge c-command relationship.19

In the inverse, exactly the opposite state of affairs obtains:

In (23a) the proximate object is able to bind into the obviative subject, in stark contrast to what we saw above in the direct (22b). Conversely, as we can see in (23b) obviative subjects cannot bind into proximate objects, again contrasting with the direct (22a). This suggests that in the inverse, the object A moves above the subject, allowing the object to bind variables in the subject, and additionally suggests that the object cannot reconstruct for variable binding. Parallel conclusions are drawn from parallel data in other Algonquian languages by Bruening (2001; 2005; 2009) for Passamaquoddy-Wolastoqey, Hamilton (2015) for Mi’gmaq, and Oxford (to appear) for Oji-Cree.

Similar binding data can be adduced from the literature on wh questions and WCO in Plains Cree. Dahlstrom (1986; 1991) argues that Plains Cree does not display WCO effects with examples like the following:

In (24a), we see that wh subjects can bind into the object, just like in English. However, in contrast to English, (24b) shows us that it’s possible to wh-extract the object and have it bind into the subject—this looks like it should result in a WCO violation, but it doesn’t.

But there’s an important confound here, as the above discussion (and my bolding) should have primed you to notice: in (24a) we have the direct marker -â (third person object agreement), but in (24b) we have the inverse marker -iko. In fact, all of the grammatical sentences in the literature on Plains Cree that are meant to show that Plains Cree lacks WCO feature the inverse, as Bruening (2001) notes. However, if the inverse involves the object A-moving over the subject, then that should allow the object to bind into the subject without any issues, just like in Blackfoot.

Interestingly, elsewhere in the literature, we can find examples of attempting to get wh objects to bind into subject in the direct, and these examples are ungrammatical, just like in Blackfoot: Blain (1997) provides the following data, which involve direct morphology and attempted “inverse” variable binding:

The result is ungrammatical. Thus, just like in Blackfoot, when you have direct morphology the WCO violation pops into view. This suggests that the inverse is a necessary ingredient in order to avoid a WCO violation, supporting the idea that the inverse involves object movement in Plains Cree.20

Example (25a) requires a bit of explanation: here, the matrix verb form indicates that we have a direct obv>obv configuration (from the combination of the direct marker -â (third person object agreement) and obviative subject agreement -yi). However, awîna is proximate, and given the verbal agreement, you might have expected obviative awînihi (note that awîna being proximate doesn’t automatically make the sentence unacceptable: it’s acceptable when awîna ‘who’ doesn’t corefer with the possessor of otêma ‘his dog’—i.e. when there’s no WCO violation). Blain says the following: “the default (proximate) form awîna ‘who’ is usually acceptable where there is no chance of ambiguity” (Blain 1997: 62). I think it’s conceivable that (25a) is a cleft—e.g. ‘who is it [that his dog is chasing]?’ (Blain actually argues that all Plains Cree wh questions are clefts)—and you get an “obviation reset” at the clause boundary.

Going back to the main point at hand: the variable binding and WCO data tell us that the inverse involves objects A moving over subjects. This results in the object now being the most local goal to the probe in C, resulting in the peripheral suffix indexing the object in the 3>3 inverse—a standard locality effect, just like in the 3>3 direct.

Now that we’ve gone through all the relevant data, let’s take a step back and evaluate where we’ve gotten to. I’ve shown that the peripheral suffix in Blackfoot and Plains Cree displays an omnivorous third person pattern, agreeing with the closest third person, no matter if it’s the subject or object. A crucial point for this interpretation of the data is the idea that in the 3>3 inverse, the object A-moves over the subject. I provided independent justification for this based on the fact that WCO violations are obviated only in the inverse. Below, I sketch the various configurations. I use privative [3] and [part] features here, but that is not crucial—they could be replaced with [–part] and [+part], respectively, if one would like to adopt Nevins’s (2007) featural representations.

The crucial data point here is thus part>3 configurations (26b), as those are the only configurations in which C isn’t just agreeing with the most local goal—there, C must skip over the part subject in preference for the third person object.21 So: omnivorous third person agreement does in fact exist, and thus we need a third person feature in the syntax, contra Preminger (2019). It seems we must favor 3yesP over 3noP.

The idea goes like this: first and second person are not contrastive for animacy, as they’re always animate, but we do get animacy contrasts for third person. So, we don’t need to specify first and second person for animacy, but we do need to do so for third person. Then, we could say that C probes specifically for animacy features, and this will result in the surface appearance of a third-person-only probe.

Unfortunately, that doesn’t work, as we can show that first and second person must be specified for animacy in the syntax. The relevant data comes from the distribution of finals—derivational morphology found across Algonquian that marks the transitivity of the verb and indexes the animacy of the intransitive subject and transitive object. To get a feel for the basic pattern, consider the Menominee verbs in (27). All of these verbs share the same root, panât- ‘be spoiled, spoil’, and they differ in terms of which finals they involve (there are many more finals than just these four, and finals rather idiosyncratically select for different roots, as well as often bear more concrete, lexical meaning, behaviors characteristic of derivational rather than inflectional morphology).

For this particular root, Bloomfield provides forms with four distinct finals:

As should be evident, finals appear to be sensitive to the animacy features of what would essentially be the “absolutive” argument if Algonquian languages featured ergative case marking, to echo the way Rhodes (1976: 80) puts it.

At this point, we need to ask two questions: (i) is the selection of finals really sensitive to morphosyntactic animacy features, rather than notional animacy? and (ii) if only one of inanimate and animate is specified for animacy, which one is it?

The answer to question (i) is yes: the selection of finals is sensitive to grammatical animacy, not notional animacy. While many nouns across Algonquian are predictably animate or inanimate—for instance, nouns referring to notionally animate entities like people and animals are overwhelmingly animate23—there are several cases where animacy must simply be lexically-specified, most commonly for notionally inanimate nouns that are grammatically animate (see Dahlstrom 1995; Goddard 2002; and Quinn 2019 for nuanced discussion of animacy in Algonquian). Below I provide examples of unpredictably animate/inanimate nouns in Meskwaki (Dahlstrom 2015: 3–6, 3–7, 3–8):

As should be evident from comparison of the inanimates in (28) to the animates in (29), it’s not clear what, if anything, could semantically distinguish the two sets of nouns. This behavior is characteristic of animacy across the family.

And when we put these nouns into a sentence, the verb stem (verb root plus finals) tracks grammatical animacy:

In (30), we have two suppletive verbs, mîči- ‘eat in’and amw- ‘eat an’—we might imagine treating these as a portmanteau of an abstract EAT \documentclass[10pt]{article} \usepackage{wasysym} \usepackage[substack]{amsmath} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage[mathscr]{eucal} \usepackage{mathrsfs} \usepackage{pmc} \usepackage[Euler]{upgreek} \pagestyle{empty} \oddsidemargin -1.0in \begin{document} \[\sqrt {\rm EAT}\] \end{document} root plus a TI or TA final. Here, the choice of which verb stem to use is dependent on the grammatical animacy of the object, rather than its notional animacy.

The answer to question (ii) is that inanimate is more likely to be underspecified for animacy. Dahlstrom (2015) argues that “inanimate is the semantically unmarked, elsewhere category of the gender system” (Dahlstrom 2015: 3–11). She notes that, in contexts where you have a single (null) pronoun referring to a mixed set of inanimate and animate nouns, the result is inanimate: in the Meskwaki story wisakea osani okyeni osimeani okomeseani [Wisahkeha, his father, his mother, his younger brother, his grandmother], told by Alfred Kiyana (1913), we find the form mîčikwêna ‘whoever ate them_in’, using the TI stem mîči- ‘eat’, to refer to a person who ate strawberries (inanimate) and fish (animate) (Dahlstrom 2015: 3–11).24 Goddard (2002) makes a similar point, noting that the inanimate indefinite pronoun kêkôhi ‘something’ can be used in contexts where it must be referring to an animate entity, providing the following example, coming from the story “Wapasaya’s Younger Brother” by Alfred Kiyana, where someone admits that they have not killed a (most likely human) enemy:

The conclusion that Dahlstrom and Goddard draw from this kind of data is that inanimates must be unmarked. They can in principle refer to both animate and inanimate referents (as revealed by these examples), but inanimate markers are (usually) restricted to appearing with non-animates only because of some kind of morphological competition—for instance, the Subset Principle in Distributed Morphology (Halle 1997).

So, back to the original question: can we show that first and second person are specified for animacy? The fact: when you have a first or second person subject of an intransitive, or a first or second person object of a transitive, you use an animate final:

In both these examples we have an animate final: the AI final -æhcenæ̂ in (32a), which Bloomfield glosses as ‘snag, immerse’ (Bloomfield 1962: 315), and the TA final -ahw in (32b), which indicates the use of an instrument (Will Oxford, p.c.)—here, the instrument would presumably be money.

Since finals track grammatical animacy, and animate is the specified value out of the pair animate-inanimate, we must conclude that the finals here are sensitive to a specified [an] feature on first and second person DPs. Thus, first and second person are specified for animacy. If so, we cannot analyze the peripheral suffix as probing for animacy, as then it shouldn’t be able to skip over first and second person DPs, counter to fact.25

Another attempt: maybe C probes for obviation features, and first and second person are not specified for obviation? This kind of analysis is somewhat tricky to respond to, since there is no general consensus on what exactly obviation is, and there are several distinct proposals in the literature for how it’s represented morphosyntactically. Here, I’ll focus on three kinds of views, and show that none of them is able to formulate an unproblematic response to the omnivorous third person pattern. The three views are:26

For reference, let us call the first view Contrast Everywhere and the second 3 Only. There are two kinds of 3 Only we find in the literature: first and second person are always proximate (iia), and first and second person are not specified for obviation (iib). I’ll address these views in turn, considering how they would implement the obviation agreement idea, and arguing that the obviation agreement account of the apparent omnivorous third person pattern fails in each case. Contrast Everywhere and the first subtype of 3 Only (part is proximate) only make the argument for 3yesP stronger, and the second subtype of 3 Only reduces to positing a third person feature.

So, I’ve argued that we can’t reanalyze the agreement pattern discussed here as an animacy or obviation probe. There is one last option to consider, which is suggested by Oxford (2014; 2017b) and Xu (2021): agreement for the categorial feature [D].

Oxford (2014), in solving a problem related to his featural representations for various kinds of DPs in (Proto-)Algonquian, proposes that third person nominals are DPs, whereas first and second person nominals are φPs.30 Xu (2021) then capitalizes on this idea in accounting for the peripheral agreement pattern in Menominee (another language with the same peripheral agreement pattern as Blackfoot and Plains Cree).

Oxford provides the following independent argument for this conclusion: first and second person do not participate in obviation and absentative marking, unlike third person, and he proposes that we can understand these facts if those features must be located in D, and first and second person lack D. However, he doesn’t provide independent evidence that obviation and absentative features must be found in D, so this argument ends up somewhat circular.

While this proposal technically works, it doesn’t seem to hold up to the available evidence. Following diagnostics from Déchaine & Wiltschko (2002), Wiltschko et al. (2015) argue that all Blackfoot independent pronouns are φPs, and Déchaine et al. (2015) argue that all Plains Cree independent pronouns are DPs. Neither conclusion involves a structural asymmetry between part and third person pronouns, counter to what the D agreement analysis would want. Oxford (2017b) acknowledges this challenge, and suggests that “since emphatic pronouns are excluded from clausal argument positions, they do not participate directly in subject/object agreement” (Oxford 2017b: 18).

While it is true that overt pronouns in Algonquian languages very often appear in clause-external topic and/or focus positions—a fact that follows naturally from the fact that Algonquian languages are rampantly pro-drop—proposing that Algonquian overt pronouns cannot occupy clausal argument positions at all is too strong, as there are cases where overt pronouns do appear in clause-medial positions. For instance, in Plains Cree, overt pronouns can occur as a kind of “resumptive” element in a left-dislocation construction:

In (37a), the third person pronoun wîya resumes the contrastively-focused John, and in (37b), the first person pronoun nîya resumes the additive pronoun nîsta ‘me too’, a contrastive topic. In these examples John and nîsta are clearly occupying left-peripheral Ā positions, but the pronouns wîya or nîya look more like they are clausal-internal, resuming the left-dislocated constituents.

We also see evidence of clause-medial positioning of overt pronouns in Meskwaki (which has the same peripheral agreement pattern as Blackfoot and Plains Cree): the emphatic series of independent pronouns can occur quite low, between preverb and verb (Dahlstrom 1988)—a use Dahlstrom calls “enclitic”:

Here, the pronoun occurs below an aspectual modifier kwayâši ‘already’ and before the verb root anehkâti ‘be acquainted with each other’. Again we see that overt pronouns can appear in clause-medial positions, contra Oxford’s suggestion.31

It seems to me that the simplest conclusion is to accept the null hypothesis: overt pronouns can indeed occur in clausal argument positions. While they are usually interpreted as topics/foci and therefore usually appear in sentence-peripheral topic/focus positions (Dahlstrom 2017), they can appear overtly even in situ. I conclude that we do actually get agreement with overt pronouns in Algonquian, and therefore, following Wiltschko et al. (2015) and Déchaine et al. (2015) who find no syntactic difference between part and third person pronouns in Blackfoot and Plains Cree, the D agreement account cannot properly derive C’s preference for third person.

So: the alternative analyses of the omnivorous third person pattern have failed. We must accept 3yesP, even in the syntax.

It has often been noted (rightly, I think) that third person occupies some kind of default status in language:

3noP, the idea that third person is underspecified, offers a natural account of Defaultness: in contexts that lack person features, we can only insert underspecified exponents, and underspecified exponents are (definitionally) third person.

While underspecification is a compelling and intuitive way to analyze Defaultness, it’s not the only formal tool we can use to analyze default effects—a conclusion that has been richly explored in the phonological literature on markedness (e.g. Calabrese 1995; de Lacy 2006; a.m.o.)—see also the discussion by Nevins (2007) in the domain of person features. In that sense, Defaultness is not as compelling an argument for 3noP as it might seem. One very basic option to capture Defaultness in a privative 3yesP system is to say that in contexts that truly lack person features, like expletives and default/failed agreement, you insert negative values of binary features as a default. This kind of rule would have to simply be stipulated in the system, and perhaps one might worry about its explanatory adequacy, but it would work to capture the fact that expletives and default agreement are invariably third person. Note that Preminger (2019) also stipulates this kind of rule in order to convert underspecified, privative syntactic representations of person to binary representations in the morphology. Another alternative, proposed by Ackema & Neeleman (2013; 2018; 2019), is to appeal to semantics, noting that the semantics of first and second person would be incompatible in default contexts which lack referents.38 I don’t take a strong stance here on what the right analysis of Defaultness should be—I just want to note that underspecification doesn’t have a monopoly on Defaultness. The real prediction of 3noP is Invisibility, and I hope to have shown that that Invisibility is wrong even in the syntax, as omnivorous third person probes do exist.

So far, I have only mentioned the cross-Algonquian picture in Section 3.3, when discussing the syntax of mixed configurations across Algonquian. There, I noted that other Algonquian languages share the same agreement pattern as Blackfoot and Plains Cree in configurations with only one third person argument: C agrees omnivorously with the third person. However, striking differences emerge when considering configurations with multiple third person arguments. In this section, I’d like to provide a sketch of how we could capture the variation across the family in a unified way under the assumption that C is an omnivorous third person probe.

The standard result when there are multiple matching accessible goals is a locality effect: all else being equal, probes agree with the closest matching goal. This is exactly what we saw in Blackfoot and Plains Cree, with C agreeing with the highest third person in 3>3 configurations (the subject in 3>3 direct, the object in 3>3 inverse). But, strikingly, other Algonquian languages show the exact opposite pattern, seeming to prefer agreeing with the lower third person. Let’s consider the following examples from Southwestern Ojibwe (Nichols 1980) and Passamaquoddy-Wolastoqey (Sherwood 1983) to see this in action:

In the (a) examples, we have direct configurations, and C agrees with the object; in the (b) examples, we have inverse configurations, and C agrees with the subject. This is exactly the opposite of what we saw with Blackfoot and Plains Cree: here, we seem to violate standard assumptions about locality by indexing the lower goal.

Thus, we can make the following split between Highest and Lowest languages: languages where C indexes the highest third person, like Blackfoot and Plains Cree, and languages where C indexes the lowest third person, like Ojibwe and Passamaquoddy-Wolastoqey.39 So far, we’ve seen how an omnivorous third person probe [u3] can capture the behavior of C in Highest languages—what about Lowest languages? How should we analyze lowest-preference?

Grishin (2022b) argues that various existing analyses of lowest-preference—a low probe with Cyclic Expansion (Béjar 2003; Rezac 2003; Béjar & Rezac 2009), syntactic inversion (Myler 2017; Colley 2018), case-sensitivity with ergative-absolutive case (Bobaljik 2008; Preminger 2014), and the Activity Condition (Oxford 2014; Hammerly 2020; 2021; Xu 2021)—all fail to capture all of the facts of Algonquian Lowest languages. Instead, he proposes to analyze Lowest languages as having an insatiable third person probe in an interaction-satisfaction theory of Agree (Deal 2015; 2021), agreeing with all accessible third persons in the clause, resulting an a multiply-valued probe. Drawing a parallel with multiple case-assignment, where there are languages that only expone the outermost, last-assigned case (like Niuean in hyperraising; Seiter 1980; Massam 1985; Béjar & Massam 1999; see also Pesetsky 2013), he proposes that a similar parametrization can apply in the case of probes as well, with there being a general microparameter Expone Outermost40 that regulates the exponence of a multiply-valued head, whether multiply-valued for case or phi features:41

On this view, the Highest languages have a C probe whose satisfaction condition is identical to its interaction condition, [int:3, sat:3], resulting in copying the features from only the highest third person, but the Lowest languages all involve insatiable probes, [int:3, sat:—], agreeing with all accessible third persons, along with a rule of Expone Outermost. The result: C in Highest languages will show agreement with the highest third person, and C in Lowest languages will show (overt) agreement with the lowest third person—but they’ve actually interacted with all third persons on the way there. Thus, we can derive the contrast between Highest and Lowest languages by simply modulating properties of the probe on C. In all cases, however, the probe is restricted to third persons only: we must accept 3yesP.