3.1 Verb classes and verb inflection

Each noun in Passamaquoddy belongs to one of two grammatical genders, animate or inanimate, with membership determined only partly on a semantic basis. Verb stems are specialized for use with one argument of a particular gender. For intransitive verbs, the subject argument is restricted: Animate Intransitive verbs require a grammatically animate subject; Inanimate Intransitive verbs require a grammatically inanimate subject. For transitive verbs, the (primary) object is restricted: Transitive Animate (TA) verbs require a grammatically animate (primary) object; Transitive Inanimate (TI) verbs require a grammatically inanimate object. (Only TA verbs occur with more than one nominal object.)

Both TA and TI verbs of cognition may subcategorize for a single NP complement, as is the case for kociciy- TA ‘know’ in (12a) and kociciht- TI ‘know’ in (12b).

The TI members of such verb pairs may also be used with a sentential complement that represents a fact as the object of cognition, as shown in (13).

Here the matrix verb is inflected for an inanimate singular object (suffix zero); this argument is presumably the clausal complement itself.

Both TA and TI verbs of cognition also participate in the raising and long-distance agreement constructions. In (14a), for example, TA ‘know’ agrees with skinúhsis ‘boy (an)’ within the complement clause, as reflected by the direct suffix -a. In (14b), TI ‘know’ agrees with mahsúsiyil ‘fiddlehead ferns (in)’ within the complement, as reflected by the inanimate plural suffix -ol on the matrix verb.

Following the proposal outlined in the preceding section, we can take agreement in both examples to be mediated by a null pronoun in object position in the matrix, as shown.

Each Passamaquoddy verb is inflected in several sets of paradigms, known to Algonquianists as orders. In the examples of the raising and LDA constructions that have been cited up to this point, the complement clause verbs have all been drawn from the conjunct order, which includes several types of forms that are primarily used in subordinate clauses. The matrix verbs are forms drawn from a paradigm of the independent order, the most common type of inflected forms used in independent clauses.

Forms of the independent order may also occur, however, in the complements to verbs of cognition in the raising and LDA constructions (as well as in a number of other embedded contexts). The independent paradigms are distinguished by the use of prefixes as inflectional elements in most forms; conjunct inflection is by suffixes alone. An example with an independent indicative form in the complement to a raising verb is given in (15). (The verb in the subordinate clause here bears the third-person prefix /w-/, but this is phonologically reduced to h-, written as an apostrophe.)

3.2 Interrogative complements

The complement clause in the raising or LDA construction may be interrogative. Thus, complement questions are not islands for (apparent) raising or cross-clausal agreement. The type of inflection that the verb in the complement clause receives depends on the type of question involved in the structure. The verbs in most questions made with forms of kèq ‘what’ or wèn ‘who’ are inflected in the conjunct order. Questions with tamà ‘where’ or taywè ‘when’ take independent indicative forms. Questions with tàn ‘how’ use forms from the independent subordinative paradigm.

The examples in (16) below show interrogative complements with ‘what’, ‘who’, and ‘how’ in sentences with raising. The conjunct verb forms in the complements in (16a, b) are bolded, as is the independent form in the complement in (16c). As before, raised nominals are shown as coindexed with an empty position within the complement clause.

The example in (17) illustrates long-distance object agreement between a verb of cognition and an NP located within an embedded question. Since the question word here is tamà ‘where’, the verb of the interrogative complement is an independent indicative form.

A nominal question word may itself be a trigger of LDA; see Section 7.1 below.

3.3 The role of clause-initial eli ‘thus’

Declarative complement clauses in Passamaquoddy raising and LDA constructions that have verbs with conjunct inflection frequently, though not invariably, begin with the particle eli, which I gloss ‘thus’. A brief consideration of the status of this particle is in order, since Bruening (Reference Bruening2001: 165–167) suggests that it should be analyzed as a complementizer. I argue here that this analysis is inappropriate. In fact, there do not appear to be any complementizers in Passamaquoddy, at least in the sense of particles whose primary function is to introduce subordinate clauses. For this reason, it is not clear that any clauses in Passamaquoddy are appropriately analyzed as CPs, an observation that may cast doubt on Breuning’s approach to raising phenomena in the language, which postulates that raised nominals are stationed in Spec of CP.

The particle eli occurs in a variety of constructions other than raising and LDA, such as the complement to ‘be sorry’ shown in (18).

As Sherwood (Reference Sherwood1986: 138) observes, eli makes little or no semantic contribution in examples of this kind. Bruening’s analysis of eli is a complementizer would, of course, account for this fact, and for the frequent clause-initial position of the particle.

Not all uses of eli conform to this pattern, however. From a formal point of view, in fact, eli is a preverb, the initial member in a compound verb. Because eli is a preverb, it need not be separated from the verb with which it is associated, as we see in (19).

The verb in the matrix clause here is a transitive inanimate form, which cannot take the animate noun nmíhtaqs ‘my father’ as its object. Thus, (19) does not involve raising, and nmíhtaqs can only be a constituent of the complement clause in this sentence; eli is not clause-initial in (19).

The morphological unity of the preverb–verb complex is revealed by the workings of an ablaut process known as initial change. Ablaut is most commonly realized as a shift from $o$ (schwa) to $e$ in the first syllable of the compound: the first syllable of the first preverb if one occurs, otherwise the first syllable of the verb stem. Most vowels other than $o$ are not affected, but there are some irregular formations. Eli is the changed form of the preverb oli ‘thus’; compare (16c) above for the basic form.

To see how ablaut works, compare the sites at which change is realized in the examples in (20).

In (20a), the initial component of the stem tol-iht- ‘be making’ undergoes ablaut (irregularly), becoming etol-. In (20b), ablaut is realized on the preverb koti, as it is the first element in the preverb–verb complex here. In (20c), koti remains unaffected by ablaut, since here it is preceded by oli ‘thus’, which undergoes ablaut instead. The fact that eli realizes ablaut in this fashion demonstrates that it is indeed a preverb. Even when it is separated from the verb and appears in clause-initial position, eli realizes ablaut for the preverb–verb complex as a whole.

Passamaquoddy preverbs vary in terms of how readily they may be separated from the verbs with which they are construed by material from outside the preverb–verb complex. Several other preverbs are like eli in their distributional freedom. The changed forms of these preverbs, like eli, typically occupy clause-initial position. A typical example is weci ‘from, for (that) reason’, the changed form of ’ci ‘from, for’, which is illustrated in (21).

In the end, there appears to be little to recommend an analysis of eli as a complementizer. The fact that eli realizes ablaut for the verbal complex is explained if it is a preverb, and its syntactic distribution is typical for one class of preverbs.Footnote ^[11] For these reasons, I do not adopt Bruening’s analysis of eli as a complementizer. I nonetheless represent this preverb as a complementizer in discussing Bruening’s proposals concerning the structure of raising and LDA sentences in the following section.

4.1 Bruening’s proposal

Raising does not always involve extraction from the complement to a raising verb under Bruening’s (Reference Bruening2001, Reference Bruening2009) proposal. Whether extraction takes place depends on the inflection of the matrix verb. In particular it depends on whether the matrix verb form is direct (with suffix -a) or inverse (with suffix -ku or a variant of this). The choice between direct and inverse inflection depends on the relative position of the subject and object of a verb on the Participant Hierarchy in (22), which involves person, animacy, and obviation: First and second persons outrank third, animates outrank inanimates, and proximates outrank obviatives.

If the subject outranks the object, the form is direct; if the object outranks the subject, the form is inverse. (I set aside here the analysis of forms involving first or second persons as both subject and object, which are neither direct nor inverse.)

For Bruening, the contrast between direct and inverse forms reflects argument-movement (A-movement) in the derivation of clauses with inverse forms: In these clauses, the object moves to a functional head above the subject. This movement of the object in a raising sentence poses a potential problem for Bruening’s assumption that the NP in question has undergone non-argument (A-bar) movement within the complement clause. For an NP to undergo first A-bar movement (raising, in the complement) and then A-movement (forming the inverse, in the matrix) would mean that it was subject to the kind of ‘improper movement’ that Chomsky (Reference Chomsky, Anderson and Kiparsky1973) and May (Reference May1979) have argued to be universally excluded. Bruening’s solution is to suppose that the apparently raised NP in a sentence with an inverse verb in the matrix does not in fact undergo raising: Instead, it is directly generated in a specifier (Spec) position at the left edge of the complement clause. It is assumed that this peripheral position in the complement clause is close enough to matrix elements that an NP in this location can undergo matrix operations, since it is at the edge of a projection that constitutes a phase, which places bounds on accessibility to syntactic operations (Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000).

Consider first, however, cases in which the matrix verb in a raising construction is a direct form. On Bruening’s account, the raised NP in such a sentence is extracted from the interior of the complement clause. But the extracted NP does not move to matrix object position. Instead, it shifts into the position of a specifier of Comp, or sometimes into a second such specifier position, and thus remains within the complement clause. Its position is again assumed to be close enough to the matrix verb to permit the establishment of relations with matrix elements. In particular, it can establish an Agree relation with the matrix verb. Thus, the matrix verb shows object agreement with the raised NP, even though the two are not clausemates.

An example of the structure of a raising clause on Bruening’s analysis is given in (23). In this example, the complement of the raising verb is an interrogative clause.

We can see here why Bruening sometimes postulates a second Spec of Comp as the landing site for raising: Since the fronted wh-word kèq ‘what’ is assumed to occupy the first Spec of Comp in a raising sentence with a wh-complement, the raised NP (if it is in Spec in the complement and not in object position in the matrix) must be located in a second Spec position.

The case in which apparent raising does not involve extraction from within the complement clause on Bruening’s account is illustrated in (24).

Here the matrix subject mihtáqsol ‘his father’ is obviative, but the raised NP skinúhsis ‘boy’ is proximate. Obviatives rank below proximates on the Participant Hierarchy, so the matrix verb is inverse. By hypothesis, inverse formation involves A-movement. But this movement cannot be fed by raising, if raising is A-bar movement: Moving an NP first by A-bar movement and then by A-movement is improper movement (Bruening Reference Bruening2001: 275). So the seemingly raised NP skinúhsis ‘boy’ in (24) must not have been raised after all, but must instead have been directly generated in Spec of Comp. The NP in this Spec position is simply coindexed with a null pronoun within the complement clause (Bruening Reference Bruening2001: 284). Since this NP has not reached Spec position by A-bar movement, it is free to undergo the A-movement process involved in deriving the inverse, which moves it into the matrix clause.Footnote ^[12] (An NP in Spec of Comp is assumed to be accessible to inverse formation, just an NP in this position is accessible to object agreement.)

4.2 Problems for the raising analysis

Bruening offers a number of arguments in favor of his analysis of raising. I focus in this section on two of his central claims; see Section 6 for additional discussion. First, he maintains that a raised NP always occupies a position at the periphery of the embedded clause, apart from cases involving inverse forms in the matrix. Second, he asserts that raising respects island constraints in the class of cases where his analysis postulates raising by extraction, which again excludes examples with inverse forms in the matrix. I argue that neither of these claims finds empirical support.

Consider first the assertion that a raised NP can only occupy a position at the left periphery of the complement, so that it necessarily follows all matrix material: Bruening (Reference Bruening2001: 270–271) claims that examples in which the raised NP comes between the matrix verb and another matrix constituent are ungrammatical. He notes that his analysis would account for this alleged fact. However, my consultants find examples that violate this restriction, like those in (25), to be fully acceptable.

In both of these examples, the matrix verb is a direct form, so the raised NP cannot have been extracted from the complement, on Bruening’s analysis. Nonetheless, the raised NP Piyélol ‘Peter (obv)’ comes between the matrix verb and its subject, Asséloma ‘Samuel’ in (25a). In (25b), the raised NP ákomok ‘snowshoes’ comes between the matrix verb and an adverb toké ‘now’ that is construed with it. It seems clear, then, that the predictions of Bruening’s analysis of the structure of sentences with raising are not borne out.

Bruening (Reference Bruening2001: 265–267) also presents examples that suggest that raising respects island constraints in just the class of cases that he takes to involve extraction, those involving direct forms in the matrix. He takes this situation to confirm his analysis of raising as A-bar movement. For my consultants, however, raising in the relevant class of cases is permitted to violate many of the usual islands, including complex NPs, wh-islands, and adjunct islands. There appears to be no difference in this regard between sentences with direct verbs in the matrix and corresponding sentences with inverse forms.

I focus here on examples involving complex NPs. Violations of the Complex NP Constraint generally result in sentences that speakers find thoroughly unacceptable, even unintelligible. An example is given in (26). This represents an attempt to question ‘(which) children’ in a Passamaquoddy sentence rendering ‘I shot the bear that bit two children at Grand Lake Stream’ (the latter a settlement near one of the Passamaquoddy communities).Footnote ^[13]

In contrast to wh-movement, raising out of a complex NP does not in general produce unacceptable results. It is quite easy to find examples that consultants consider well formed. These may have a direct form in the matrix, as in (27a), or an inverse form, as in (27b). The contrast in acceptability between these examples that Bruening’s analysis would predict is not confirmed.

We see, then, that wh-movement is subject to the Complex NP Constraint, but raising is not. This fact argues strongly that raising structures are not formed by a process of A-bar movement. All of the available evidence is in fact consistent with an analysis of the raised NP as a directly generated constituent of the matrix clause, namely the matrix object. Such an analysis explains without stipulation (i) why the raised NP triggers object agreement, (ii) why the raised NP participates in the direct-inverse alternation like any other object, and (iii) why the raised NP can be permuted with matrix material.

5.1 Evidence for two NP positions: Overt pronouns

In the examples of raising and LDA that we have looked at up to this point, one or the other of the two postulated NP positions in these constructions has been empty. But in fact either position in which a null pronoun is possible may also be occupied by an overt pronominal. This is shown by sentences like those in (31).

In (31a) we have an example of the raising construction. The raised NP Píyel ‘Peter’ is matched in the clausal complement of ‘know’ by an (optional) overt pronominal nekòm ‘he, she’. In (31b), the positions of these two NPs are reversed. The result is an instance of the LDA construction. But now it is clear that no agreement at a distance is involved here: The trigger of object agreement on the matrix verb is a pronominal in the matrix clause that may be overt. Examples like (31a, b) provide clear evidence that raising and LDA involve a relationship between two NPs (or their indices), one in the matrix and a second within the complement clause.Footnote ^[16]

5.2 The raising construction with two full NPs

Further confirmation for the claim that the constructions we have been considering involve a relationship between two NP positions (more precisely, between two referential indices) comes from sentences in which there is not only a full NP in matrix object position, but also a second non-pronominal NP, coreferent with the matrix object, within the complement clause. These sentences make a secondary assertion about the referent of the matrix object in addition to the central claim expressed by the complement clause. Two examples are given in (32).

Examples like these are reminiscent of English sentences in which an epithet like the bastard plays a role like that of a pronoun. Note, however, that the complement clause NP in (32b) includes both the demonstrative nòt ‘that (an)’ and the relative clause mecíkit ‘he who is evil-natured’ in addition to the head noun skitàp ‘man’. An analysis of this NP as a pronoun-like epithet is not attractive. The conclusion thus seems inescapable that two full NPs may be related in the raising construction, not just an NP and a (null or overt) pronoun. This situation is expected under the lexical analysis of raising advanced here. The argument structure of a raising verb simply requires its S complement to include some predication structure in its RESTR that (i) can be predicated of the matrix object and (ii) is a function of a semantic argument with the same index as that object. This predication structure may be contributed by any type of NP within the complement clause.

5.3 The absence of Condition C effects

In Section 2, we observed that Passamaquoddy sentences with LDA systematically violate Condition C. As it turns out, Condition C violations are freely permitted in Passamaquoddy. Thus, no special provision is required in the present analysis to accommodate binding in LDA structures.

Consider (33), a schematic representation of an LDA sentence, in which ‘know’ takes a pronominal object pron $_{\text{i}}$ that is coreferent with NP $_{\text{i}}$ in the complement.

Here pron $_{\text{i}}$ c-commands and binds NP $_{\text{i}}$ . This configuration is ruled out on standard formulations of Condition C (and by the HPSG binding theory of Pollard & Sag Reference Pollard and Sag1994).

Violations of Condition C in LDA structures present a problem for the analysis proposed here, however, only if this constraint is operative in the language. There is ample evidence that it is not, a point made also by Bruening (Reference Bruening2001: 26–29). Consider the examples in (35).

The acceptability of (34) is expected. But (35) is also acceptable, while its English counterpart is not, since he _i in the matrix binds and c-commands Joseph _i in the complement. Quite generally, Condition C violations are grammatical in Passamaquoddy, and present no special problem for the analysis of LDA.Footnote ^[17]

6.1 Does raising affect binding?

A quantified expression like psí $=$ te wèn ‘everyone’ located within the complement clause in a Passamaquoddy raising sentence can bind a variable in the matrix. Thus, ‘everyone’ in the complement of the raising verb nmihqitahámak ‘I remember them’ binds the possessor of the matrix object ’tolonapèm ‘his relatives’ in the raising structure in (36). Note that on Bruening’s account of raising, the latter nominal must have been displaced from a position within the complement clause.

Bruening (Reference Bruening2001: 264) assumes that binding requires c-command: The binding quantifier must c-command the bound variable. Since the required c-command relationship does not obtain in the surface form in (36), he proposes that it is established at the abstract level of Logical Form, where binding relations are checked, through reconstruction: This operation will move the raised NP ’tolonapèm back into its original position within the complement clause, thus placing the possessor of this nominal in a position where it is c-commanded by psí $=$ te wèn.

Bruening argues that the role of reconstruction in such examples is revealed by the blocking of binding in cases in which a negative element intervenes in surface structure between a variable in the matrix and a quantifier within the complement that could potentially bind it. The presence of this negative element, he asserts, prevents reconstruction from taking place and establishing the c-command relationship that must hold for binding to take place.

The blocking effect in question is revealed, Bruening maintains, by the absence of the second reading indicated for example (37b).

In (37a), skàt wèn ‘no one’ binds the possessor of nisuwihtícil ‘his/her spouse’.Footnote ^[18] But to establish the same binding relationship in (37b), nisuwihtícil would have to be lowered by reconstruction into its source position within the complement clause so that skàt wèn can c-command it. Reconstruction across a negative element is blocked, so the second reading suggested above for (37b) is excluded. Since this account relies on the premise that reconstruction sets the stage for binding in raising sentences, its success in predicting the readings found for examples like (37b) provides evidence that raising involves movement.

In fact, however, I have been unable to confirm the judgments that Bruening reports for examples like (37b). My consultants have sometimes found this and comparable sentences to be confusing, but once they understand them, they consistently find readings of both indicated types to be fully acceptable. I have found no evidence that would verify Bruening’s claim that negative quantifiers block potential binding relations in Passamaquoddy.Footnote ^[19]

Indeed, there appears to be no evidence for the initial assumption on which Bruening’s argument rests: that a Passamaquoddy quantifier must c-command a variable that it binds. Quite apart from raising sentences, in examples where NP-movement is not at issue, an expression like psí $=$ te wèn ‘everyone’ or psí $=$ te pilsqéhsis ‘every girl’ within a complement clause can bind a (null or overt) pronominal in the matrix, as we see in the examples in (38).

In (38a), psí $=$ te wèn ‘everyone’ in the complement of ‘say’ binds but does not c-command the possessor of the matrix subject ‘his mother’. Likewise, psí $=$ te pilsqéhsis ‘every girl’ in the complement of ‘tell’ binds but does not c-command nekòm ‘she’, the matrix subject in (38b). Yet my consultants judge both of these examples to be entirely acceptable on the indicated readings. Note further that such examples cannot be accounted for, as a JL referee has suggested, by supposing that such cases of binding without c-command actually involve quantifiers that refer to sets, rather than true instances of variable binding. In (38b), in particular, the singular bound pronoun in the matrix can only represent a variable.

I conclude that Bruening’s argument that reconstruction plays a role in determining variable binding in raising sentences in Passamaquoddy does not go through. There is no evidence for movement as the mechanism of raising from this domain of facts.

6.2 Raising and scope relations

Bruening (Reference Bruening2001: 280; Reference Bruening2009: 443) presents a second argument from binding relationships in favor of his movement theory of raising, citing examples (39a, b). He notes that both of these examples are unacceptable and attributes this fact to conditions on binding. First, he suggests, psí $=$ te wèn ‘everyone’ cannot bind the possessor of wikuwóssol ‘his mother’ in (39a) because an object cannot bind into a subject. (This is again presumably because the quantifier must c-command the variable that it binds.) Moreover, he continues, raising psí $=$ te wèn as in (39b) does not give rise to new binding possibilities: ‘an object that raises cannot thereby bind into a subject that it crosses’ (Bruening Reference Bruening2001: 280). Once again the binding potential of a quantifier is determined by its original location, restored by reconstruction; surface positions must accordingly be determined by movement.

As it happens, however, (39a, b) are excluded not for binding reasons, but because they involve impermissible obviation relations. When obviation is properly adjusted, examples parallel in form to these are found to be fully acceptable. To see this, however, we must briefly consider the conditions that determine the distribution of proximate and obviative nominals within clauses.

Obviation is fundamentally a matter of the relative prominence of third-person referents in discourse, but the expression of these prominence relations is subject to certain restrictions. Prominence relationships may be analyzed in terms of pairs of referents within a context. Call the more prominent member of a pair, usually the more topical referent, the controller of obviation for the less prominent member. Two central restrictions come into play: (i) the possessor of a noun is always a controller of obviation for the possessed noun itself; (ii) if a nominal A is located above a nominal B on the relational hierarchy Subject ${>}$ Primary Object ${>}$ Secondary Object, then A is a controller of obviation for B. (An extra provision is needed for clauses with inverse verb forms, although I will not explore this issue here. These are treated for the purpose of (ii) as if the grammatical relations of the Subject and Primary Object have been reversed.)

The primary object is the sole object of a single-object transitive verb, but the goal or benefactive object of a ditransitive verb. The object of a ditransitive verb that bears the theme relation is the secondary object. (The primary object of a ditransitive verb, like the sole object of a simple transitive verb, is involved in determining whether the verb form is direct or inverse.) Consider, then, the examples in (40). (A variety of other words orders are possible in each case.)

Since the primary object wásis ‘child’ in (40a) outranks the secondary object olomussísol ‘puppy’, ‘child’ must be the controller of obviation for ‘puppy’ here. Accordingly, ‘child’ is proximate, while ‘puppy’ is obviative. In (40b), the grammatical relations of these two nominals are reversed, so ‘puppy’ is now the controller of obviation for ‘child’; the former is now proximate and the latter obviative. In (40c), the primary object skinúhsis ‘boy’ outranks the secondary object piléyal ‘temísol ‘his new dog’, so ‘boy’ controls obviation for ‘dog’. This relationship is also guaranteed by the fact that ‘boy’ is coreferent with the possessor of ‘dog’: The possessor controls obviation for the possessed head. Note that ‘boy’ cannot be marked obviative in this example: ‘Boy’ cannot be secondary to ‘dog’, and there is no other third-person referent in the context that could control obviation for ‘boy’. In (40d), by contrast, there is a (null) third-person subject, which is proximate. This subject controls obviation for the primary object ‘boy’, which is therefore obviative. Marking ‘boy’ as obviative does not conflict with its status as coreferent with the possessor of ‘dog’: Nothing prevents an obviative referent from serving as a controller of obviation for another obviative referent.

With this background concerning obviation established, we can see where Bruening’s examples (39a, b) have gone wrong. In both cases, psí $=$ te wèn ${\sim}$ psí $=$ te wénil ‘everyone’ serves as the secondary object (or the coreferent of the secondary object) of a verb whose subject argument is wikuwóssol ‘his ( $=$ everyone’s) mother’. The fact that ‘his mother’ outranks ‘everyone’ on the relational hierarchy means that ‘his mother’ must be a controller of obviation for ‘everyone’. But the fact that ‘everyone’ is (or is coreferent with) the possessor of ‘mother’ means that ’everyone’ must be a controller of obviation for ‘his mother’. Thus, ‘mother’ is both more and less prominent than ‘everyone’. These contradictory requirements on control of obviation have the result that both examples are excluded.

We can amend Bruening’s examples in such a way that the conditions on obviation are satisfied, while the binding relations we are seeking to test remain unchanged. To do this, we need only make ‘everyone’ the possessor of the secondary object in each sentence, rather than the object itself. Examples reformulated along the required lines are given in (41).

All restrictions on obviation are met in (41a). As a noun with a third-person possessor, ‘her mother’ is obviative. This nominal is the subject of ‘hide’, so it outranks the secondary object ‘(everyone’s) doll’.Footnote ^[20] Thus, ‘her mother’ controls obviation for ‘doll’, and the latter must also be obviative. The possessor ‘everyone’ is a second controller of obviation for ‘doll’. Nothing requires ‘everyone’ to be obviative, however, and since ‘everyone’ is coreferent with the possessor of ‘mother (obv)’, it is free to be proximate. (Typically, there is one proximate referent in any context; here ‘everyone’ is selected.) Contrary to Bruening’s assertions concerning binding and c-command, ‘everyone’ within the secondary object in (41a) is free to bind the possessor of the subject of ‘hide’ (‘mother’).

The same obviation relations obtain in (41b). The possessor ‘everyone’ controls obviation for ‘her mother’ and ‘her doll’; the subject ‘her mother’ controls obviation for the secondary object ‘her doll’. Once again ‘everyone’ is free to be proximate. Since all restrictions on obviation are met, the sentence is perfectly acceptable – with the reading that Bruening would exclude, on which the raised NP ‘everyone’ binds the possessor of the complement subject ‘her mother’.

I conclude that the argument that Bruening (Reference Bruening2001, Reference Bruening2009) seeks to formulate on the basis of conditions on binding in raising structures is based on an erroneous understanding of the restrictions that govern the distribution of proximate and obviative nominals. Once these restrictions are clarified, Bruening’s proposed argument dissolves.

6.3 Interrogative complements to inverse raising verbs

On the account proposed in Bruening Reference Bruening2001, the way raising sentences are derived varies according to the type of inflection that the matrix verb receives; see the discussion in 4.1. In sentences with a direct verb form in the matrix clause (suffix -a), the raised nominal is extracted from a position within the complement and moved to a Spec position at the periphery of this clause. In sentences with an inverse verb form in the matrix (suffix -ku- or a variant), the seemingly raised nominal is instead directly generated in a complement Spec position, where it binds a null pronoun within the complement. Recall that, for Bruening, a verb receives inverse inflection when its object undergoes an A-movement operation that shifts it into a position higher within the clause than the subject (although subsequent scrambling may reverse this order again). A nominal in Spec position in the complement of a raising verb may undergo inverse formation, thereby moving into the matrix, just in case it was directly generated in a peripheral location in the complement. (Since a nominal that is directly generated in a Spec position is interpreted as an argument of the complement predicate, it evidently occupies an A-position. Thus, moving such a nominal into the matrix clause does not involve improper movement, the movement of a nominal from an A-bar to an A-position.)

Bruening (Reference Bruening2009: 442) argues that the postulated distinction between raising sentences with direct and inverse forms in the matrix finds support from a restriction on the distribution of wh-words: He claims that the matrix clause in a raising sentence must be direct and cannot be inverse when the agreeing nominal is a complement wh-phrase. He offers the example in (42) in illustration.

Bruening reasons that (42) is excluded because wèn, as a wh-word, must have been shifted into an A-bar position in the complement. Moving this NP again into the matrix via inverse formation would mean moving it into an A-position, a case of improper movement.

As it turns out, (42) is indeed unacceptable, but not for the reason that Bruening supposes. What is wrong with this sentence (at least in the absence of some appropriate discourse context) is again a matter of the conditions that govern obviation. As the matrix subject, ‘Peter’ is the most natural topic in (42), so this NP would be expected to be proximate. There is no discourse motivation for taking ‘someone’ to be topical here, and thus proximate, while taking ‘Peter’ to be secondary, and thus obviative.

We can adjust the expected obviation status of the subject and object in this sentence by instead choosing a possessed subject whose possessor is coreferent with wèn, such as wikuwóssol ‘his mother’, with ‘his’ referring to ‘someone’. As we have noted, a possessor always controls obviation for the possessed noun, so ‘his mother’ will now be obviative with respect to ‘someone’. With motivation supplied in this way for the appearance of an obviative subject, the conditions for the use of an inverse verb form obtain: An obviative subject wikuwóssol ‘his mother’ acts on a proximate object, the latter a null pronoun coreferent with the possessor of the subject and with wèn ‘someone’ in the complement. The raising sentence (43) that we obtain in this way is judged by my consultants to be entirely natural.

Note that the matrix verb here is inverse and that its inflection in determined by a wh-word in the complement: This is precisely the configuration that Bruening (Reference Bruening2009: 442) claims is excluded. I conclude that the constraint on the distribution of wh-words that Bruening (Reference Bruening2009) seeks to use as an argument for his movement analysis of raising dissolves upon closer examination.

6.4 A constraint on extraction from the complements of raising verbs

Bruening (Reference Bruening2001) observes that only certain positions in raising structures may be targets of extraction in wh-questions. He formulates the constraint involved as follows: ‘it is possible to do wh-movement out of the complement of a raising to object verb, but in such a case the verb must agree with the wh-phrase that is extracted’ (p. 304, italics in original). Essentially the same constraint is noted for Innu-aimûn by Branigan & MacKenzie (Reference Branigan and MacKenzie2002: 402). The examples in (44) illustrate the effect in question:

In (44a), the verb wewitahámot ‘you (sg) remember him’ agrees with wèn ‘who’ as its object (conjunct order suffix -ot ‘second-person singular on third person’). Extraction of wèn is permitted: Per Bruening, wèn moves from object position in the complement clause, though the specifier position in this clause, and then into specifier position in the matrix. In (44b), the matrix verb has been inflected for first-person object (suffix -i) plus second-person singular subject (suffix-hin); compare eli—kisi—tokom-í-hin ‘that you (sg) hit me’. This pattern of agreement would reflect raising of the first-person subject of the complement clause. As Bruening’s generalization would lead us to expect, (44b) is unacceptable: Wèn has been extracted from the complement, but the raising verb in the matrix does not agree with it.

Bruening’s explanation for the impossibility of extraction from the complement in cases like (44b), where the raising verb fails to agree with the moving wh-word, is based on the theory of phases advanced in Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000, together with a principle of economy derived from Pesetsky & Torrego Reference Pesetsky, Torrego and Kenstowicz2001.Footnote ^[21] Bruening assumes Chomsky’s theory of phrase structure, which postulates a projection $v$ P that constitutes a layer of structure between VP and TP ( $=$ S). Features are assumed to ‘drive’ movement in the sense that certain heads bear features that must be ‘checked’ against the features of certain phrases by moving the phrases to designated positions with respect to the heads. Bruening assumes in particular that there is a feature [wh] that occurs on the $v$ head of $v$ P as well as on (matrix and complement) C. This feature ‘draws the wh-phrase to the edge of the $v$ P phase, since it must move on to matrix CP’ (Bruening Reference Bruening2001: 305). This step in the derivation is necessary, since movement out of $v$ P can only take place via the edge of this projection, given Chomsky’s (Reference Chomsky, Martin, Michaels and Uriagereka2000: 108) Phase-impenetrability Condition, which limits extraction out of projections that are designated as phases to peripheral constituents. The head of $v$ P is assumed to have not only the [wh] feature that needs checking, but also a set of argument features (A-features) that must be checked. The latter are involved in specifying relationships between a verb and its arguments. Feature checking takes place through various applications of the matching process known as Agree.

This is where economy comes in. Following Pesetsky & Torrego Reference Pesetsky, Torrego and Kenstowicz2001, Bruening hypothesizes that ‘one instance of Agree, checking more than one feature, is more economical than multiple instances of Agree, each checking only one feature’ (Bruening Reference Bruening2001: 305). Since the head of $v$ P must check both its [wh] feature and its A-features, economy dictates that it should do so through an Agree operation with a nominal which will make it possible to accomplish both ends at the same time, if this is possible. Where a wh-expression is undergoing extraction, it functions as just such a nominal: As both a wh-expression and an argument, it can check both the [wh] feature and an A-feature of $v$ by moving to the edge of $v$ P. Thus, economy dictates that $v$ should Agree just with the moving wh-expression and not also with a distinct raised NP. It follows that only the wh-expression may be raised into a position in which the matrix verb can agree with it, given that extraction takes place from the complement clause. No other argument of the complement verb may be raised if wh-movement into the matrix takes place.Footnote ^[22]

Consider, however, how Bruening’s examples may be analyzed under the alternative proposal that we have been entertaining, according to which a raising structure actually involves a verb that takes two arguments: an NP object and a clausal complement that (typically) includes a null NP coreferent with the matrix object. The examples in (45) may serve to illustrate; compare Bruening (Reference Bruening2001: 177 examples (435a, b)).

On the non-movement account of raising that I have proposed, the verb ‘suspect’ in (45a) takes an NP object, plus a complement clause that includes a null pronoun $e$ that is coindexed with the matrix object. There is no reason to suppose that $e$ is the questioned constituent here. In fact, it seems reasonable to suppose that the questioned constituent is the matrix object, indicated in (45a) by $t$ . In (45b), on the other hand, the matrix object must be a null first-person pronoun in the position marked $e$ , since the matrix verb is inflected for first-person object. Thus, the wh-word wèn ‘who’ can only have been extracted from the complement position marked $t$ . Note that extraction is blocked here.

A simple account of the blocking of extraction in (45b) now becomes available: Extraction is blocked from the sentential object of a verb of cognition. In particular, the clausal object of a raising verb is an island.Footnote ^[23] To see that this line of reasoning is on the right track, consider the examples in (46).

The matrix verb in (46a) is wewitaham- ‘remember’, an animate-object (TA) form. Bruening’s economy theory of extraction is compatible with this example: On his account, wèn is extracted from the position indicated by $e$ , checking both the [wh] feature and an A-feature of the complement verb at the edge of the $v$ P phase in the complement before undergoing further attraction to the matrix C. This example is also compatible with the alternative analysis that takes the clausal object of a raising verb to be an island: On the non-movement account of raising, wèn has been extracted here from matrix object position, not from a site within the complement clause.

Consider next the fact that (46b) is not acceptable, even though no raising has taken place here and the matrix verb is an inanimate object (TI) form, which agrees with the clausal complement as its object. The crucial point is that wèn ‘who’ has been extracted from the complement. The unacceptable situation that results is one in which extraction takes place from within the complement without agreement between the matrix verb and the moving nominal. This situation can be ruled out on economy grounds if we assume (i) that matrix $v$ must check its [wh] feature against the moving wh-word, (ii) that matrix $v$ must enter into an Agree relation with a nominal in Spec of Comp regardless of how that nominal reaches this Spec position, and (iii) that it is preferable (because more economical) to achieve these two relations through an Agree operation that targets a single NP. These assumptions have the consequence that matrix ‘remember’ in (46b) must agree with ‘who’ (as it moves through Spec of Comp in the complement) and not with its clausal complement as its object, which in turn has the effect of excluding the TI verb form that appears in this example.

The problem with this account is that it will rule out all wh-movement from within the complement of a TI verb. A wh-word passing through Spec of Comp as it is extracted from such a complement will have exactly the same status as the moving wh-word in (46b). (Recall that no raising takes place in the latter example.) As we see in (46c), extraction from the complement of a TI verb is not blocked in the general case: In this example, wénil ‘who (obv)’ has been extracted from the sentential object of TI nut- ‘hear’, rather than from the complement of a verb of cognition. I conclude that Bruening’s economy theory does not correctly delimit the class of cases in which extraction can take place from a complement clause. What we need instead is a principle that blocks extraction from the clausal objects of verbs of cognition, which include but are not limited to raising verbs: The clausal objects of verbs of cognition are islands.Footnote ^[24]

Further evidence that economy considerations are not responsible for the blocking effects that Bruening points to comes from examples in which the extracted wh-word corresponds not to an argument of the complement verb, but rather to the possessor of such an argument. Consider the examples in (47) in this connection.

Here as before, extraction of wèn is only possible when the matrix verb agrees with it: in (47a), but not in (47b). But wèn is not an argument of the complement verb ‘give away’ in (47b); it is the possessor of the object of this verb. Thus, the $v$ head of $v$ P does not bear any A-feature that indexes this possessor. Considerations of economy in the complement will therefore not block simultaneous extraction of wèn and raising of the first-person subject of ‘give way’ out of the complement clause in (47b). Nor can we say that economy considerations in the matrix will block extraction here by requiring that the matrix verb should agree with an NP that reaches Spec of Comp, regardless of how it gets there. We have already seen that such a principle would incorrectly exclude all wh-movement out of the complement of a TI verb.

On the other hand, the data in (47) are expected if (i) the extracted wh-word in (47a) is a matrix object and (ii) the clausal object of a verb of cognition is an island to extraction. Since nothing blocks extraction from matrix object position, (47a) incurs no violation of the proposed island condition. In (47b), by contrast, wèn can only have been extracted from within the complement clause. This is disallowed.

I conclude that Bruening’s economy theory of the restrictions on extraction out of raising structures in Passamaquoddy fails to generalize to the full range of cases. An account formulated within the proposed non-movement theory of raising appears more promising. Of course, it remains to be explained just why the clausal object of a verb of cognition should be an island to extraction. The answer to this question must be left for further research.Footnote ^[25]

7.1 Question words in long-distance agreement structures

Branigan & MacKenzie (Reference Branigan and MacKenzie2002: 394) discuss a phenomenon they call cross-clausal agreement (CCA) in Innu-aimûn, a Central Algonquian language of Quebec and Labrador (Canada), which parallels Passamaquoddy LDA in certain respects. They argue against an analysis of CCA that would postulate a matrix object (which they term a prothetic object) distinct from the complement NP that appears to govern matrix object agreement in the construction. One argument they present is based on the observation that Innu-aimûn CCA, like Passamaquoddy LDA, may involve agreement with a question word that takes scope in the complement, as in the Innu-aimûn example in (51).

Branigan & MacKenzie (Reference Branigan and MacKenzie2002: 394) suggest that there can be ‘no coherent interpretation for a prothetic pronoun’ in the matrix object position of such a sentence, since this would require coindexation between a (null) pronoun and a question word. They compare the unacceptable English example in (52).

Bruening (Reference Bruening2001: 270) makes the same argument in discussing the following Passamaquoddy example:

He comments that ‘there is no coherent two-argument interpretation in such cases’; compare English *‘Piyel doesn’t remember about him $_{\text{1}}$ who $_{\text{1}}$ I gave a car to’ (p. 270).

The problem with this argument is that there indeed are two NP positions that may be occupied by a pronoun and a coindexed question word in a sentence like (53). To see this, consider (54). Here both positions are filled: The matrix verb of cognition takes an overt object (either an indefinite pronoun or a demonstrative) that is coindexed with the question word that introduces the complement question.

Here wénil in the matrix is presumably an indefinite pronoun rather than an interrogative form, since it may be replaced by the non-interrogative (demonstrative) pronoun níhtol ‘that one (obv.sg)’. On the other hand, wèn can only be an interrogative pronoun in the complement, since it is involved in a long-distance dependency: It is interpreted in the position indicated by $t$ , as the object of ‘dance with’. In any case, it seems clear that we are forced to a two-argument analysis of LDA with question words by examples such as this, despite Bruening’s (and Branigan & MacKenzie’s) misgivings.

7.2 Semantic interpretation

So how can we account for the interpretation of examples in which a question word appears to trigger LDA? To answer this question, we need to place it in the larger context of the interpretation of raising and LDA structures.

On the analysis of these constructions that I have been pursuing, a verb of cognition takes two complements, an NP and a clause. The verb assigns a thematic role to its NP object that we may characterize as ‘object of the subject’s cognition’ or ‘entity about which the subject possesses information’. The complement clause specifies the information in question. Consider the examples in (55).

The declarative complement in (55a) makes a statement that specifies the information that the subject knows about Joseph: (55a) means, ‘I possess information about Joseph, namely that he is going to school in Eastport.’ In (55b), the information that the subject is described as knowing is instead the answer to an indirect question: (55b) means, ‘I possess information about the bears, namely the answer to the question of what they ate.’ So here it is the answer to a question that is about the referent of the matrix object.

The matrix verb may agree with a question word in the complement, as in (56), although agreement in a case of this type is indirect.

On the account of LDA proposed in Section 5 above, matrix object agreement in (56) is actually with the null complement NP indicated by $e$ . The indirect question ‘who first cooked muskrats for her ( $=$ Tihtiyas)’ is about the referent of this null object in the sense that answering this question serves to identify the person $x$ that the speaker is saying that Tihtiyas does not remember – Tihtiyas’s object of cognition. Thus, the example as a whole means, ‘Tihtiyas does not remember information about some person $x$ , namely the answer to the question, which person $x$ first cooked muskrats for her’.

In the case of either a declarative or an interrogative complement in the raising or LDA construction, what it means for the complement clause to specify information about the matrix object is simply that the RESTRICTION associated with the complement S includes a PRD that in turn includes a function of the index of the matrix object.

8.1 Apparent raising of a subset of the target NP: An analysis

We can understand examples like (57), I suggest, in the following terms. In a raising or LDA construction, the complement clause must be about the referent of the matrix object NP. Formally this means that the RESTRICTION set of the complement S must include a PRD that is predicated of the index of the object NP. If the S includes an NP that denotes a set of entities, one of which is the referent of the object NP, then the S is about this referent. The object NP’s index has been restricted by a predication in the RESTR set of S, as required. But if the object NP has a plural referent and nothing in the S’s RESTR set is predicated of that plural referent, then the requirements on this RESTR set are not met. S is not about the object NP in the sense in question.

In developing this analysis, it is useful to consider examples that involve what Schwartz (Reference Schwartz, Barlow and Ferguson1988) calls the plural pronoun construction. In formations of this type in various languages, a plural pronoun is used together with a prepositional phrase or another NP that partly specifies its reference. In Passamaquoddy the construction takes the form of an NP juxtaposed to a plural pronoun: kiluwàw kitàp ${\sim}$ kitàp kiluwàw ‘you (sg or pl) and your (sg) friend’, with kiluwàw ‘you (pl)’ and kitàp ‘your (sg) friend’. The modifying NP is presumably an adjunct to the pronoun, since it can either precede or follow it.

In (58), we have a raising sentence with kìl ‘you (sg)’ as the raised NP and the plural pronoun construction kiluwàw kmóssis, literally ‘you (pl) your older sister’, but understood here as ‘you (sg) and your older sister’, as the related NP in the complement clause.

Here kiluwàw k-móssis ‘you and your older sister’ in the complement S refers to a set that includes the referent of the matrix object kìl ‘you (sg)’. The index of kìl is restricted by the PRD introduced in S by the plural pronoun construction, so the raising structure in (58) in licensed.

A second example may serve to indicate that this account is on the right track. In (59), a third-person raised nominal is interpreted as representing a subset of a second-person plural pronoun in the complement clause. I represent this interpretation here by writing the index of ‘this man’ as $i$ and that of ‘you (pl)’ as i+2, with ‘2’ indicating reference to the addressee.

The second-person plural pronoun kiluwàw ‘you (pl)’, under the indicated interpretation, introduces a PRD in S that restricts the index of the matrix object wòt skitàp ‘this man’. Thus, S counts as being about ‘this man’, and raising is licensed.

If the relationship between the matrix object and the corresponding argument in the complement is reversed, so that the complement NP represents a subset of the matrix NP, raising is not licensed. This is the situation in (60).

Since ‘you (sg)’ in the complement S does not introduce a PRD on the index of ‘you (pl)’ in the matrix, S is not about the matrix object as required.

8.2 Can the movement theory handle the facts?

Up to this point I have considered the facts introduced in this section only from the point of view of the analysis of raising and LDA developed in Section 5 above. What about the alternative analysis that postulates movement as the mechanism of raising?

In fact, it is quite difficult to imagine how a movement analysis can accommodate examples like (57), repeated here as (61). The essential assumption of Bruening’s proposal is that an NP that is initially located within the complement clause in such an example is shifted into a peripheral position, where it can trigger matrix object agreement. But in a case of the type under consideration, the agreeing nominal in the matrix co-occurs with a distinct expression (overt or null) in the complement, and the two are distinct in reference. Not only does the occurrence of one not preclude the occurrence of the other, but the first cannot be analyzed as a simple copy of the second, as some kind of copy theory of movement might postulate.

Noting this problem for his account, Bruening (Reference Bruening2001: 269) offers two observations. First, he takes note of the Plural Pronoun Construction, in which, as we have seen, a type of conjunction may be expressed by stationing a nominal as an adjunct to a pronoun. While this construction does indeed involve a relation between a pronoun and an adjunct that expresses a subset of its referents, it is hard to see how this fact contributes to a solution of the problem for the movement theory that is posed by examples like (61). There is simply no position for kìl to have raised from in this example, since the expression we would need to postulate to provide such a source is *kiluwàw kìl ‘you (pl) including you (sg)’, which is not a felicitous plural pronoun construction. Similarly, the ‘raised’ nominal kìl ‘you (sg)’ in (58) above corresponds to the plural pronoun construction kiluwàw k-móssis ‘you and your older sister’ in the complement, which does not include kìl as a constituent. Again, there is no source in the complement from which the ‘raised’ nominal could have been copied.

Bruening (Reference Bruening2001: 269) takes a different tack. He notes a typologically unusual type of agreement in certain relative clauses in Passamaquoddy which he takes to suggest that ‘movement’ in this language may target just a subset of the referents of the ‘moved’ NP, rather than a pronoun as such. Naturally, if such a subset-movement operation can be shown to be available, we will have license to suppose that the matrix object kìl ‘you (sg)’ in (61) is derived by copying just a subset of the referents of kiluwàw ‘you (pl)’.

The type of relative clause agreement that Bruening (Reference Bruening2001) points to in this connection is illustrated in (62); compare Bruening’s example (702) (p. 269).

Here the verb nacitaham-ti-htí-c-il ‘the one that he and someone hate each other’ is a reciprocal form (suffix -ti-) which is inflected for third-person proximate plural subject (suffix -hti-), but which is also inflected in agreement with the relativized constituent (suffix -il ‘obviative singular’), the latter representing one member of the set of participants reflected by subject agreement. The obviative participant (‘that man’) is one of a pair of referents (‘that man’ and his enemy) who are said to have always hated each other. Thus, Bruening reasons, relativization functions in this case to extract just a subset of the referents of a plural subject.

In fact, however, the obviative singular referent in (62) is a syntactic argument distinct from the syntactic subject of the verb. Both are singular; only subject agreement is plural, reflecting the whole set of participants in the action that the verb describes. There is a class of verbs in Passamaquoddy that have argument structures and agreement with just these properties. They select a subject that represents a subset of a group of individuals who are involved in a joint activity, but also select an object that represents a distinct subset of this group. Subject inflection is then for the combined person and number of the subject and object (the whole set of participants in the activity named by the verb), while object inflection indexes only the properties of the object. This construction is known as split coordination (Bruening Reference Bruening2004, Reference Bruening2006; LeSourd Reference LeSourd, Hele and Valentine2013). Its function is to foreground certain participants in a joint activity, while backgrounding others.

In the independent indicative mode (used in main clauses), verb inflection in the split coordination construction reflects both the subject and the object arguments of a participating verb. The referent picked out by object inflection represents a subset of the set of referents picked out by subject inflection. Example (63) illustrates.

Here nacitaham-tí-ni-ya-l ‘they hate each other, he is enemies with him’ is a reciprocal form, as before. Inflection is for third person proximate plural subject (suffix -ya-), reflecting the combined features of the subject and the object (‘that old man (prox)’ and ‘his wife’s mother (obv)’), and for third-person obviative singular object (suffix -l), reflecting just the grammatical categories to which the object (‘his wife’s mother’) belongs.

Since there are two arguments in the split coordination construction, it is entirely possible for one to be relativized independently of the other. This is exactly what we find in examples like (62). The word-final suffix in a relative-clause form like nacitaham-ti-htí-c-il ‘the one that he and someone hate each other’ indexes the relativized constituent. But the target of relativization in this case is not a subset of the referents of the syntactic subject. Rather, the target is the object argument of the verb, syntactically independent from the subject argument. Only the agreement markers, not the syntactic arguments, overlap in reference. Nothing in such cases of relativization is comparable to what we find in raising structures in which the raised NP represents a subset of the corresponding complement NP.

I conclude that the argument against the movement analysis of raising and LDA that was advanced in 8.1 stands. Bruening’s proposed operation of subset movement, which would circumvent the problems identified here, finds no support. The lexical analysis of raising and LDA presented in Section 5 can readily be extended to handle the cases we have encountered in this section, and must therefore be preferred to the movement alternative.