3.1 The clitic view

Halle & Marantz (1993) propose that the person prefix on Independent Order verbs in Potawatomi is a pronominal clitic. They give exactly one argument to this effect, which is that the prefix can appear separated from the verb stem (Halle & Marantz 1993: 141). We have already seen this in Passamaquoddy-Maliseet, in examples (2/13) and (15). Halle and Marantz say that the prefixes appear “at the front of CP,” and “their location depends on what else occurs within the CP.” They do not spell out their analysis, but they seem to be proposing that the prefix is a pronominal clitic high in the clause, at the left edge of CP, from which position it attaches to a host.

Halle and Marantz’s description is not accurate. The prefix is actually quite selective. It can only go on the main verb itself, or on a preverb. All of Halle and Marantz’s Potawatomi examples involve preverbs. I repeat all three of their examples below:

Hockett (1948: 140) lists all three of these elements as preverbs. Ku is listed as a preverb indicating “assent to a request, and perhaps other things.”⁵ ‘Quickly’ and the inceptive are typical preverbs in other Algonquian languages. The Passamaquoddy-Maliseet correspondents are the preverbs nokosa (‘quickly’) and mace (‘start’). From Hockett’s description of the verbal morphology it is clear that the prefix only goes on the verbal complex in Potawatomi, which is to say either the verb stem itself, or the first preverb. Potawatomi is just like all other Algonquian languages in this respect.

If the prefix were a pronominal clitic that cliticized to whatever was first in CP, we would expect there to be many things it could attach to. For instance, a negative particle and the modal particle op obligatorily precede the verb and any preverbs in Passamaquoddy-Maliseet, in that order:

Wh-words used as indefinites also typically precede the verb (keq here). But the prefix never attaches to any of these preverbal elements, instead it goes on the first preverb (underlined). On standard analyses of clause structure, a Neg projection and a Mod projection would be lower than C. Examples like that in (18) then show very clearly that the prefix is not at the left edge of CP and does not attach to whatever is first in CP (that would be kat in (18)).

The negative and modal particles which obligatorily precede the verb also do not need to be initial in the clause, meaning that they also are not always the first thing in CP:

The prefix also does not go on whatever is to the left of negation (here a freestanding subject pronoun), it goes on the first preverb, far to the right. Note that in this example it also fails to attach to an adverb meaning ‘again’ and a demonstrative pronoun (which, like wh-indefinites, tend to immediately precede the verb).

Other particles obligatorily appear at the left edge of CP but cannot host the prefix. For instance, the particle on, ‘then’, obligatorily appears at the left edge of CP, and in the following example it is followed by the second-position future clitic (whose form is oc after consonants, hc after vowels). As stated above, this clitic follows the first prosodic word in the CP. Yet the prefix never attaches to on, or the future clitic, or the adverb ‘again’ which follows it, or the subject NP. It attaches to the first preverb:

Just to belabor this point fully, Passamaquoddy-Maliseet is a wh-movement language. Wh-phrases obligatorily move to Spec-CP in interrogatives (Bruening 2001; 2004a; 2007). We can therefore use the position of an interrogative wh-phrase to decisively locate the left edge of CP. If the prefix were a clitic that attached to whatever was at the front of CP, we would expect it to attach to a wh-phrase. It does not, however:⁶

We could also interpret Halle & Marantz (1993) such that the prefix attaches not to what is leftmost in CP, but to whatever is leftmost starting from the head C. The two examples above show that this could not be right, either. Both of these examples have a freestanding subject pronoun between the wh-phrase in Spec-CP and the verb. In (21b), the modal particle op also comes between them (and it must, suggesting an intervening ModP projection). The prefix does not attach to any of these preverbal elements, however, it goes on the verb or preverb. (Note that left-dislocated elements precede wh-phrases and the types of particles listed above that occur at the left edge of CP, so the pronouns and NPs in none of the above examples could be viewed as left-dislocated. See Bruening 2001: 34–35.)

The prefix is very selective, then, and does not attach to whatever is at the left edge of CP. In examples like (18–21b), the prefix actually seems to be quite far from the edge of CP.

Halle & Marantz (1993) also propose that the verb raises to C in Potawatomi. The word order facts presented here show that this could not be correct for Passamaquoddy-Maliseet. In fact the verb must remain quite low in Passamaquoddy-Maliseet. It must be lower than the preverbal negative particle and modal particle in the examples above. It has to be lower even than a low ability modal. Such root modals are generally thought to be very low in the clause (e.g., Cinque 1999). In the following example, however, the preverb that encodes the root modal (Able) is separated from the main verb. This means that the main verb has not raised even as high as the low root modal.

Similar arguments against verb raising to a high position have been given for some other Algonquian languages. Johnson (2016) argues that the verb in Potawatomi stays low, giving similar word order arguments to those produced here for Passamaquoddy-Maliseet. The facts of Potawatomi seem to be very similar to those in Passamaquoddy-Maliseet, which I take to justify applying arguments from Passamaquoddy-Maliseet to the clitic analysis of Potawatomi. Lochbihler & Mathieu (2009) give some similar word order arguments for Ojibwe and point out that positing V movement to C incorrectly predicts that the verb should precede all preverbs in Algonquian languages. (The position of the verb and the preverbs will be important in building an analysis in section 4.)

Since the one argument that has been given in the literature does not actually indicate that the prefix is a pronominal clitic, it is necessary to scrutinize the behavior of the prefix using diagnostics that have been proposed in the literature for distinguishing affixes from clitics.

3.2 Diagnostics from Zwicky and Pullum (1983)

To begin, Zwicky & Pullum (1983) propose six diagnostics that are meant to tell clitics from affixes. I will go through them, although most of them are not very reliable. In particular, the first four are only tendencies, if they are even that. For recent critical discussion of these diagnostics, see Thoms et al. (2023). Diagnostics that I consider much more telling will be presented in section 3.3.

Zwicky and Pullum’s first diagnostic is that clitics can exhibit a low degree of selection with respect to their hosts, while affixes exhibit a high degree of selection with respect to their stems. We have already seen that the Algonquian prefix is very selective. It can only attach to either the verb stem itself, or a preverb. This contrasts with clitics like the English possessive’s and cliticized (contracted) auxiliaries, which attach to whatever is adjacent to them, regardless of category. Second-position clitics in Passamaquoddy-Maliseet like oc above are also not selective, and will come after whatever is first in the clause. However, some clitics, like Romance object clitics, are selective, and only attach to verbal elements. This diagnostic is therefore not particularly reliable.

Zwicky and Pullum’s second diagnostic says that arbitrary gaps are more characteristic of affixes than clitics. For instance, they point out that affixal negation in English has an arbitrary gap, as for most dialects there is no *amn’t. There are no arbitrary gaps that I know of in Algonquian, but since this diagnostic seems to only work in one direction, this is also not telling. (Thoms et al. 2023 point to the “person case constraint” as a gap in clitic clusters, but I note that this gap is generally thought not to be arbitrary; see Anagnostopoulou 2017.)

Zwicky and Pullum’s third diagnostic says that morphophonological idiosyncrasies are more characteristic of affixed words than of clitic groups. In all Algonquian languages, the prefixes have one allomorph for consonant-initial stems, and one for vowel-initial stems. In Passamaquoddy-Maliseet, the allomorphs are n-, k-, w/’- before stems that begin with a consonant, and nt-, kt-, ’t- before stems that begin with a vowel (the apostrophe is an /h/, which is typically only audible in its effect on the following consonant). This is entirely regular. However, there are also a few vowel-initial stems that idiosyncratically take the n-, k-, w/’- allomorphs rather than the nt-, kt, ’t- ones. Morphophonological idiosyncrasies mark the prefix as an affix, then, not a clitic. Again, however, this is only a tendency, as clitics, like the object clitics in Romance, can have idiosyncrasies (see Thoms et al. 2023 for one from European Portuguese).

Zwicky and Pullum’s fourth diagnostic says that semantic idiosyncrasies are more characteristic of affixed words than of clitic groups. There are no semantic idiosyncrasies that I know of in Algonquian. The prefixes simply mark the person of one of the arguments. They are quite regular in this use. This diagnostic also only works in one direction, however, and so this is also not particularly telling (and Thoms et al. 2023 dismiss this criterion altogether, arguing that semantic idiosyncrasies are almost non-existent even with affixes).

Zwicky and Pullum’s fifth diagnostic involves syntactic rules. According to them, affixes can undergo syntactic rules along with their hosts, but clitics are claimed not to. It is difficult to find such rules in Alqonquian. However, we could think of the separation of the preverb and prefix together from the main verb stem as in (13) above as a movement rule affecting the preverb and prefix together (as proposed by, e.g., Dahlstrom 1995; Branigan 2012). It would be quite reasonable to propose that in (13), the preverb and prefix have moved together across the future particle. Note that it is not just second-position clitics that can separate preverbs from the main verb, other things like argument pronouns can, too (example (23b) repeats example (15) from above):

In these two examples a subject pronoun separates the preverb (or even two preverbs) from the main verb. Here it really appears that we need a movement rule to dislocate the preverb plus prefix away from the main verb. (Note that in these two examples, the prefix is not audible for phonological reasons.) It might be possible to use non-syntactic mechanisms to get a second-position clitic in between a preverb and the main verb (the prosodic inversion of Halpern 1992, for instance), but that could not work for non-clitic pronouns and other non-clitic material. We need a movement rule that moves the preverb plus prefix away from the main verb. Given this, the prefix is acting like an affix, and not like a clitic. (But note that the analysis I will propose in section 4 will not have the prefix+preverb moving away from the verb; rather, the highest preverb/verb will move to the location of the prefix.)

Zwicky and Pullum’s final diagnostic is the ability to attach to material already containing clitics. According to Zwicky and Pullum, clitics may attach to other clitics, but affixes may not. Affixes always have to attach inside of clitics. In Passamaquoddy-Maliseet, clitics like the future oc (hc after vowels) can come between a preverb and the verb stem, but never between the prefix and the preverb or verb:

It is impossible for the prefix to attach to a clitic. In this respect it behaves like an affix and not a clitic.

To summarize so far, the diagnostics of Zwicky & Pullum (1983) are not particularly useful, as most of them merely describe tendencies. Some are dismissed by Thoms et al. (2023). Suppose we were to take them all equally seriously as diagnostics, and we were to err on the side of the prefix being a clitic whenever it did not clearly pattern as an affix. Then the score would be the following:

Four of Zwicky and Pullum’s diagnostics indicate that the prefix is an affix, while only two indicate that it is a clitic. This scorekeeping is really only an exercise, however; I would not take the first four diagnostics seriously. Only the last two seem like they are more than tendencies. On these two diagnostics, the Algonquian prefix patterns as an affix.

3.3 More telling: Diagnostics from Kramer (2014) and Preminger (2014)

Kramer (2014) and Preminger (2014) have proposed other diagnostics for telling clitics from affixes. These diagnostics are aimed specifically at telling agreement affixes from pronominal clitics. They are therefore more directly useful to the particular case under investigation, and I would argue that their diagnostics are also much more telling than those of Zwicky & Pullum (1983). Looking at Algonquian, all of their diagnostics that are applicable indicate that the prefix is an affix.

First, agreement affixes are obligatory, whereas pronominal clitics can be optional. The prefix in Algonquian is obligatory, even when it indexes a freestanding pronoun:⁷

Compare Romance object clitics, which are generally not obligatory in the presence of an overt argument (although dative ones can be).

A second diagnostic is the ability to index more than one argument. In Romance languages, it is possible to have more than one object clitic. In contrast, in Algonquian, the prefix can index only one argument. It can index both first and second persons, as both subjects and objects (27a–b), but if both arguments are first or second person, it cannot index them both simultaneously, it can only index one (always the second person; 27c–d):

This again makes the prefix an affix, not a clitic.⁸

Romance object clitics also behave independently in ditransitives. It is possible to have two, one for each object, or one for the lower object in the absence of one for the higher object. In contrast, agreement may be limited by locality to only indexing the higher of the two objects. The latter is true of the prefix in Algonquian. In the inverse, it can index an internal argument (as in 27b), but with a ditransitive, it can index only the higher object:

It is not immediately obvious which argument the prefix is indexing in an example like this, but the prefix can only ever index a proximate argument. In this case, this is the applied argument (the benefactive). The direct object of the verb (what is made) has to either be inanimate (and inanimates are never indexed by the prefix) or obviative with respect to any other third person arguments. This means it can never be indexed by the prefix (see Bruening 2001: 134–136 on the lower or “secondary” object). Additionally, the first object (the applied object) can be first or second person and then be indexed by the prefix in the inverse or in first-second person interactions:

In contrast, the lower object is limited to third person. I take all of these facts to indicate that the lower object is never indexed by the prefix. This then identifies the prefix as an affix, not a clitic.

Clitics may also index only certain types of arguments, for instance just specific indefinites or definites. They may also have semantic effects, for instance in inducing a specific or emphatic reading. In contrast, affixes index all arguments, regardless of their semantics and without inducing any particular interpretation. The Algonquian prefix behaves like an affix in this regard: it indexes all arguments of the relevant grammatical role, and it has no semantic effects. For instance, it obligatorily indexes both definite pronouns (26a) and weak indefinites, like wh-words used as indefinites that take narrowest scope:

This diagnostic also identifies the prefix as an affix, and not a clitic.

Another diagnostic concerns binding relationships. According to Kramer (2014) and Preminger (2014), clitics can change binding relationships, but agreement affixes cannot. In particular, clitics can circumvent weak crossover. In Algonquian languages, we have to look across orders to see whether the prefix has any effect on binding, since the prefix is obligatory in the Independent Order. However, it does not appear in the Conjunct Order. If the prefix were a pronominal clitic, we might expect it to make binding relationships different in the Independent Order compared to the Conjunct Order. It does not. As Bruening (2001; 2005; 2008; 2009) shows, the inverse enables the logical object to take scope over and bind into the logical subject. However, this is true regardless of whether the verb is in the Independent and has the prefix, or is in the Conjunct and does not. I illustrate with variable binding below. The example in (31a) is in the Independent Order and has the prefix, while the example in (31b) is a wh-question in the Conjunct and does not. In both cases, the object quantifier is able to bind a pronoun inside the subject as a variable:

In contrast, when the verb is not in the inverse, the object cannot bind a pronoun as a variable in the subject, regardless of the presence of the prefix:⁹

It is clear that the presence or absence of the prefix makes no difference to binding relationships in the language. In this respect the prefix behaves like an agreement affix, and not a clitic.

Additionally, according to Kramer (2014) and Preminger (2014), affixes may agree in just a subset of the features relevant in the language, whereas clitics tend to index all features. In Algonquian, the prefix agrees like an affix in just a subset of features: it indexes person and not number. This is visible in examples (27c) and (28) above, where the number of the agreeing argument is indexed by the suffix -pa or -ya. It should also be pointed out that the prefix behaves unlike freestanding pronouns in this respect, which index all features. For instance, second person pronouns in Passamaquoddy-Maliseet have the forms singular kil, inclusive first person kilun, and second person plural (excluding first person) kiluwaw. The prefix is therefore behaving very differently from pronouns in the language.

According to Kramer (2014) and Preminger (2014), clitics also tend to appear in all clause types, including in non-finite clauses and imperative clauses. In contrast, affixes may disappear, depending on the clause type (cf. Nevins 2011). The Algonquian prefix behaves like an affix in this respect, as we have already seen: It is present only in the independent order (33a) and is absent from the Conjunct Order inflection (33b) and in imperatives (33c):

Compare Romance object clitics again: They appear in finite clauses, non-finite clauses, imperatives, all tenses, etc. (though their position famously varies by clause type).

Kramer (2014) and Preminger (2014) propose two other diagnostics, but these are not applicable to Algonquian. First, according to Preminger (2009; 2014), if an agreement relation is broken, agreement will be realized as a default. In contrast, clitics will just disappear in a similar situation. What breaks agreement, according to Preminger, is the presence of a “defective intervenor”: another NP that is closer to the agreeing head that is not capable of agreeing with that head. I have not been able to find any kind of situation like this in Algonquian languages. The prefix indexes the grammatical subject, which is always the highest NP in the clause, or it indexes the object in the inverse, but in this case the subject is inert and causes no intervention. In ditransitives, as described above, the second object can never be indexed by the prefix, while the first can (in the inverse), so there is never a situation where the second object could have been indexed but was blocked by the higher object. Algonquian languages do have long-distance agreement (e.g., Frantz 1978; Branigan & MacKenzie 2002). However, this is often truly long-distance, being able to skip over any intervening NP (e.g., in Passamaquoddy-Maliseet, Bruening 2009; LeSourd 2019), or it can only index the highest NP in the lower clause (e.g., in Plains Cree, Dahlstrom 1991). In neither case is there ever a situation where something that is incapable of agreeing blocks the agreement relation.

The second and final diagnostic is the ability to appear on a passive or reflexive verb. According to Kramer (2014), clitics can appear in passive/reflexive forms, but (object) agreement typically will not. It is not clear whether Algonquian languages have passives; I will therefore leave them aside. Reflexive verbs are detransitivized with a suffix, as in the following example:

The prefix can appear, as in this example (though it is inaudible for phonological reasons). However, the appearance of the prefix follows the pattern of intransitives generally. The verb ‘kill’ in this example is inflected in the Subordinative (one of the contexts for the N morpheme). All intransitives in the Subordinative have the prefix (if they have an animate subject):

This is just the pattern of intransitives generally. Reflexive verbs, being intransitives, are agreeing exactly as one would expect of them. I therefore do not take this diagnostic to show anything about the Algonquian prefix.

Baker & Kramer (2018) add another diagnostic for clitics versus affixes. According to them, affixes can index singular universal quantifiers and NPs that contain a bound variable, but (doubled) clitics cannot. According to this diagnostic, the Algonquian prefix is an affix, since it can index a singular universal quantifier and an NP containing a bound variable:

We also saw a prefix indexing a negative quantifier or negated existential in example (30). If Baker & Kramer (2018) are correct that this diagnostic distinguishes affixes from clitics, then the Algonquian prefix must be an affix.¹⁰

The following table summarizes the diagnostics from Kramer (2014), Preminger (2014), and Baker & Kramer (2018):

As can be seen, all of these diagnostics that are applicable indicate that the prefix is an affix. On not a single one does it come out as a pronominal clitic.

3.4 Summary

The single piece of evidence given by Halle & Marantz (1993) for the clitic view was not correct. Their proposal that the prefix is a pronominal clitic high in the clause was also shown to be untenable. In addition, all of the reliable diagnostics that have been proposed for telling agreement affixes from pronominal clitics show that the Algonquian person prefix is an agreement affix, and not a clitic. I conclude that the Algonquian person prefix is a canonical affix, which is exactly how most work on Algonquian languages since Bloomfield (1946: 95) has treated it, other than the literature cited above that followed Halle & Marantz (1993).

The issue now is how to analyze the prefix, given its unusual distribution. This is the subject of section 4.

4.1 The suffixes

The verb can have several valence-changing suffixes on it. I will assume that these are syntactic heads that combine with the V through head movement, as in the vast literature following Baker (1988). Consider the following example, which has a base transitive verb that is turned into a ditransitive with a Ditrans(itive) morpheme and then a reciprocal with a Recip(rocal) morpheme. The reciprocal lowers the verb’s valence by one (so, from a ditransitive back to a transitive). The morpheme closest to the root (-om-) indicates the valence of the base verb (transitive with an inanimate object):

We can analyze the Transitive Inanimate (TI) suffix as v (Brittain 2003) and the ditransitive morpheme as an Appl(icative) head (Marantz 1993) above that. Recip(rocal) can be analyzed as a version of the Voice head (Bruening 2004b; 2006b); this head combines next. V moves through v and Appl to Recip:

This gives us a simple analysis of the valence-changing morphology. Turning to the inflectional morphology, a lot of it is agreement. I will analyze agreement morphemes as Agr heads that merge with other heads in the syntax, following roughly Halle & Marantz (1993), except that the Agr heads adjoin in the syntax, not at any post-syntactic level (there is no need for any post-syntactic level of grammar). For instance, the head Voice, including the Recip version of it, will have an Agr head merged with it as its sister.

In a transitive verb with an animate object, there are three Agr suffixes: (1) the Theme Sign (marking Direct/Inverse or 1/2 interactions); (2) the Central ending, which indexes the same argument as the prefix (but in number in addition to person); (3) the Peripheral ending, which indexes the other argument (the terms “central” and “peripheral” come from Goddard 1974; 1979; Nichols 1980):

In negative clauses, there is also a Neg morpheme between the Theme and Central Agr heads. This Neg morpheme is not sufficient by itself to make a clause negative, there also has to be a preverbal negative particle (ma in (42)).

In Bruening (2004b; 2006b), the head Recip is a variety of Voice. I propose that all clauses with a verb have a VoiceP. I further suggest that Voice, including the Recip variety of it, merges with an Agr head that is the Theme Sign. See Oxford (2019) for an analysis of the theme sign that locates it in Voice and treats it as object agreement. Immediately above Voice there is a head Σ that merges with another Agr. Σ is the negative morpheme on the verb if the clause is negative. I follow Zanuttini (1997); Poletto (2008); De Clercq (2013) in hypothesizing the existence of (at least) two positions for negation in the clause. One is this low Σ that is between Voice and any tense, aspect, or mood heads. The other is a higher Pol(arity)P that hosts the preverbal negative particle (ma in 42). I assume that there is an agreement relation between Pol and Σ, with the result that there is only a single semantic negation. Σ is present but null in a positive clause. Whether it is null or not, it merges with an Agr that is the Central Ending. Finally, there is another, unidentified head that I will call “H” that merges with the Agr that is the peripheral ending. The tree below shows the proposed structure. (The external argument is merged in Spec-VoiceP, as in Kratzer 1996.)

The V moves first to v, then to Voice, then to Σ, and finally to H. This results in the form monu-w-a-wi-wa-k. Note that H has to be below the lowest preverb, so the verb does not move very high in the clause. The heads that host the Agr morphemes (or the Agr morphemes themselves) engage in agreement relations with the arguments of the verb. I will not spell these relations out here; for some worked out analyses, see, among others, Bruening (2001) and Oxford (2019) (but note that I do not endorse Oxford’s view that the prefix and the central ending are a discontinuous agreement morpheme; see section 4.4).

While this is by no means a complete analysis of the Algonquian suffixal verbal morphology, it will suffice for our purposes here.¹¹ The suffixes are correctly located on the verb stem through head movement of the lexical V to the head H.

4.2 Preverbs

It is not entirely clear what grammatical category preverbs are. They could be auxiliary verbs; there is a large literature showing that the Algonquian preverbs are like auxiliary verbs in other languages in the functional categories they express and in their order. See, for example, Leavitt (1985); Costa (2002); Cook (2003; 2014); Shields (2005); Branigan (2012); Oxford (2016). However, they can also express various modificational functions, things that are not typically expressed by auxiliary verbs, like ‘good’ in (14). I will remain uncommitted on their grammatical category. I will propose, however, that they share a feature with verbs, and no other grammatical category has this feature. For lack of a better term, I will use the feature [+V]. Verbs and preverbs are both [+V]. Preverbal particles, second-position clitics, NPs, adverbs, adjectives, and any remaining grammatical categories that I have overlooked are not [+V].

I propose that preverbs are heads that merge into the clause immediately above the head H in (43). I will illustrate with example (23b), repeated below:

There are two clear preverbs in the second clause of this example, nokomasi (‘easily’) and kisi (ability modal). It is not entirely clear how to treat kinalo. It forms a tighter bond with the verb stem, which cannot be separated from it. Algonquianists call such morphemes initials, although many of them can also be used as preverbs. Since morphemes like this are not the focus of this paper, I will simply treat kinalo-ke as a single lexical verb, although one could also treat kinalo as a third preverb. I will focus on the two preverbs nokomasi and kisi. I propose that the ability modal merges immediately above H, and the modificational preverb ‘easily’ merges above that. I will label them “Pvb1” and “Pvb2” in the diagrams. The subject pronoun has moved and adjoined to HP in this example (its starting position was Spec-VoiceP, as above):

I assume that HP and any of the PvbPs are possible adjunction sites for moved argument NPs (and other things). Recall that the lexical verb moves to H, so it immediately follows the subject pronoun in this example. (If we instead choose to analyze kinalo as a preverb, then the pronoun would be adjoined to the lowest PvbP instead of to HP.)

The preverbs can either form prosodic words on their own (and they must, if something separates them from the verb stem), or they can optionally form a single prosodic word with the verb in H, if they are adjacent to it. This seems to become obligatory with certain verb stems that are prosodically deficient on their own, for instance -ke- in (44) and -ya- in the following:

The second-position clitic yaq follows the preverb-verb combination here, indicating that they form a single prosodic word. However, the very same preverb can be separated from a different verb by a second-position clitic, as in the following example:

I will not attempt to spell out when preverbs form their own prosodic words and when they form a prosodic word with an adjacent verb, but will leave it as optional in principle. For detailed discussion of some of the factors involved in Passamaquoddy-Maliseet, see Leavitt (1985).¹² On similar issues in other Algonquian languages, see Goddard (1988); Branigan (2012).

4.3 The prefix

Section 3 showed that the prefix is an agreement affix. In the current analysis, that makes it an Agr head that merges with some other head in the syntax. I propose that, immediately above the highest preverb, there is a head which I will identify as “Infl” (only because it is vaguely reminiscent of Infl in analyses of English and other languages).

I will assume that Infl includes a specification of the clause type, as Independent or Conjunct or Imperative. When Infl is Independent, an Agr head merges with it (but not otherwise):

I also assume that Infl attracts a [+V] head to it. Given locality, this will be the highest [+V] head. In the structure in (48), this will be Pvb3. This corresponds to the second clause of the example in (14). That clause is repeated below:

The highest preverb moves to Infl:

If there is no preverb, the highest [+V] head will be the verb itself in H. Consider the following example:

‘Where’ questions (and ‘how’ questions in 49) use the Independent Order rather than the Conjunct, so Infl in this example is specified as Independent and therefore has Agr adjoined to it. There is no preverb, so after the verb has moved to H, it will move on to Infl:

It should be clear that the proposed analysis correctly locates the prefix on the highest [+V] head. The highest [+V] head moves to Infl, which has an Agr adjoined to it in the Independent Order.

In the tree in (52), I have shown C taking IP as complement directly, but this is a simplification. Recall that many things can and do come before the preverb with the prefix on it. The negative particle and the modal particle op obligatorily precede the prefix. I repeat the example in (19) below:

This example has at least a PolP and a ModP above IP, all of which I assume are below C. Moved NP arguments and adjuncts can adjoin to these projections:

Thus, many things can, and some things must, come in between the left edge of the clause and the highest preverb which combines with the prefix.

This analysis successfully accounts for the different distributions of the prefix as opposed to the suffixes. It also accounts for the word order facts that have been presented throughout this paper.

4.4 Discontinuous person-number realization?

The analysis proposed here does not connect the person prefix to the central suffix, which indexes the number of the same argument that the prefix indexes. (I assume that the correct theory of the agreement relations will relate them, but only indirectly, since they both agree with the same argument.) The Algonquian pattern is an instance of a very common cross-linguistic one, where agreement is realized discontinuously with a person prefix and a number suffix (see Trommer 2002; Harbour 2008; 2023 and references there). One might wonder whether the current analysis is missing something by not attempting to capture this pattern. There are several reasons why this is not the case, and the two affixes should not be connected formally. These are also reasons why the analysis of the pattern proposed by Harbour (2008; 2023) will not work for Algonquian, so I will present that first. (Harbour’s analysis is adopted for Algonquian languages by Oxford 2019.)

In a nutshell, Harbour (2008; 2023) proposes that the morphology attempts to insert two morphemes into the position of the prefix. The agreement head is structured, with person over number. So the morpheme realizing person takes precedence, and the morpheme realizing number has to get shunted off to the end to avoid disrupting adjacency between the person prefix and the stem (see Harbour 2008; 2023 for the details).

This will not work for Algonquian languages. As we have seen, it is actually the suffix that is stable across paradigms while the prefix appears or disappears depending on the conjugation. The following table presents the facts in a slightly different format:

From this format it can be seen that the central suffix (boldface) appears in all paradigms, in the same position, while the person prefix only appears in the Independent Order. It would make more sense to say that the morpheme is first inserted into the suffix slot, and only when it cannot be fully realized there does it split off to become a prefix. Of course, one could say that the morphology first tries to insert the agreement into the prefix slot, finds that it cannot in the Conjunct and Imperative, and then shunts the morpheme off to the end. But this presupposes a prior analysis of when and where the prefix appears, which is exactly the coverage of the current analysis. So the current analysis is logically prior to any analysis of the distribution of the features across the agreement morphemes.

Second, the prefix and the suffix do not even occur on the same complex head in Algonquian languages. As we have already seen, the host of the prefix and the host of the suffix can be separated by prosodic word boundaries, and even by whole phrases. This makes it unlikely that the prefix and the suffix are related in any direct way, and it makes Harbour’s analysis unworkable. In his analysis, the suffixal part of the agreement head would only be shunted off to the end of the complex head that the prefix is part of. In Algonquian languages, this would incorrectly put it as a suffix on the same preverb that hosts the prefix. As the following example reiterates, the central ending goes on the verb stem, while the prefix goes on the first preverb, which is a separate prosodic word. In this example it is also separated from the verb stem by another prosodic word (another preverb):

It would be completely ungrammatical to put the central agreement suffix on the same preverb that the prefix appears on (56b).

Consider also the coordination facts discussed above in relation to example (39), repeated in (57). Here the prefix appears on a preverb, while the suffixes appear on two coordinated verbs below it. In (57), the subject happens to be singular, and so there is no central ending, but if the subject were plural, I assume that each of the main verbs would bear a central suffix (I have not been able to verify this, but all the other suffixes go on both verb stems). If this is correct, then in Harbour’s analysis, a single morpheme would have to be duplicated and realized twice, once in each conjunct. This is an extremely dubious analysis. Examples of coordination like this rather show that the prefix and the suffix are independent and are not related formally, except indirectly, insofar as they agree with the same argument.

Additionally, other morphemes can also agree with the same NP that the prefix does. The theme sign, for instance, commonly treated as object agreement (e.g., Oxford 2019), can index the same NP in person. In the following example (repeated from (23a)), the theme sign suffix -ol indicates a first person subject with a second person object, duplicating the second person features of the prefix (inaudible for phonological reasons):

Yet no one that I know of has proposed formally relating the prefix and the theme sign in the way that Harbour (2008) and Oxford (2019) have proposed to relate the prefix and the central ending. Why not is unclear; but turning it around, if we are not to relate the theme sign and the prefix in our formal model, then we should not relate the prefix and the central ending, either.

I take all of these considerations to indicate that the prefix and the suffix are independent and are not a single instance of agreement that is realized discontinuously. I therefore treat them independently in the analysis. I also conclude that any analysis of the cross-linguistic pattern that relates the prefix and the suffix formally, like that in Harbour (2008; 2023), cannot work for Algonquian and is therefore not a viable explanation for the cross-linguistic pattern as a whole.

4.5 Summary, and an issue

The proposed analysis explains the different distributions of the prefix and suffixes. The suffixes are placed on the verb stem by movement of the lexical verb through a sequence of heads to a low head H. The prefix is an Agr head on Infl, which is higher. Infl attracts the highest [+V] head. This correctly puts the prefix on the main verb if there is no preverb, but on the highest preverb if there are any. In the proposed analysis, the main verb moves to H and then to Infl if there is no preverb. If there is a preverb, the verb stays in H while the highest preverb moves to Infl. There are multiple projections to the left of Infl, so the highest verbal element has not moved very high and can be preceded by multiple words and phrases.

While this analysis accounts for all the facts discussed in this paper, there are some additional suffixes that can appear on the verb that raise non-trivial issues of clause structure. In addition to the negative suffix that I have provided an analysis of, there can also be suffixes indicating “preterite” and “dubitative preterite.” The preterite is apparently a past tense, while the dubitative preterite is a modal or evidential of some kind. In many cases the preterite is one morpheme while the dubitative preterite is another, mutually exclusive one, as in the following:

However, in some forms, the dubitative preterite seems to add a morpheme to the preterite:

In these forms we can also see that the preterite and dubitative preterite come in between the central agreement and the peripheral agreement. A natural analysis in the current proposal would be that the preterite/dubitative is the head H, to which the peripheral ending adjoins (the Obv suffix in the above examples). I repeat the proposed analysis of the inflectional suffixes from (43) below, with the verb changed to match the examples here, and the preterite -pon located in H:

However, if the dubitative preterite is actually two morphemes, then it is not clear how to analyze them. It is also not clear what tense and modal categories are doing in such a low position, below the preverbs that are ability modals and aspectual categories. In most languages, tense and modals are higher (and Passamaquoddy-Maliseet has another modal, op, which is higher than even the highest preverb; see (53)). Of course, we could just say that in Passamaquoddy-Maliseet, preterite and dubitative preterite are low, and there is a language-specific clause structure that differs from what we tend to find in other languages. This may or may not be unsatisfying. I will have to leave this issue unresolved here.