2.1 Labeling and wh-questions

In Chomsky (2015), a wh-question such as the following is analyzed on a par with the subject at TP.

By merging which dog in the CP region, we get the structure {wh-DP, CP}, a POP by itself. But C has the Q-feature, parallel to the φ-feature on T. The moved wh-phrase also has this feature, parallel to the DP with the φ-feature moving to TP. With both wh-DP and CP having the same Q-feature, the newly created structure {wh-DP, CP} may be appropriately labeled.² The CP/TP parallels are illustrated below.

Japanese is a typical wh-in-situ language. However, in Old Japanese the wh-phrase overtly moved to the head of the sentence (Nomura 1993; Watanabe 2002). The following is taken from Man’yoshu, an anthology of poems compiled in eighth century CE; the number “1169” is the number of the poem in the anthology (Takagi 1957; 1962).

Without exception, a moved wh-phrase is accompanied by a focus particle such as ka in this example. This overt movement began to disappear from the language beginning in the ninth century, apparently triggered by the separation of the ka particle from the wh-phrase (Miyagawa 2010). In Isobe’s (1990) study of wh-questions in the Tale of Genji, a psychological novel by Lady Murasaki Shikibu in the tenth century, roughly one-third of the examples have ka on a wh-phrase at the head of the sentence, whereas in another third, the wh-phrase occurs in-situ by itself and ka appears at the end of the sentence, just as in modern Japanese. The remaining third are similar to those with clause-final ka, except that the final particle is a different focus particle, zo. The last type is exemplified here.

In other words, once a particle such as ka or zo does not accompany the wh-phrase but instead occurs on the interrogative C, the wh-phrase no longer moves, and thus Japanese became a wh-in-situ language.

Let us assume that the analysis of the movement of the wh-phrase in Old Japanese is essentially the same as that of wh-movement in modern English, although it need not be. For example, given that focus particles are involved, it may be that in Old Japanese, it is focus agreement that is operative as suggested for wh-constructions in general by Rizzi (1997). What we wish to look closely at is the historical change that transformed the OJ wh-movement construction into the modern wh-in-situ construction. We believe that this change sheds light on an important factor affecting labeling that has implications beyond wh-questions.

The key change in the wh-construction from OJ to modern Japanese is the migration of the focus particle from occurring directly on the wh-phrase in OJ to occurring independently as a particle on C in modern Japanese. Once the particle occurs on C, the wh-phrase does not move. From the perspective of labeling, this means that once the particle attaches to it, C is able to project independently without needing a wh-phrase to merge with it, as was the case in Old Japanese. What is the function of the particle on C, such that it makes it possible to forgo the need for the wh-phrase to move to C?

In Miyagawa (2001), it is suggested that the occurrence of the Q/Focus particle on the C_Q/Foc fulfills the EPP requirement of the C head. It does so because the particle has the same feature, Q (or FOC), as the head, C. We will extend this analysis by presenting an analysis based on Contiguity Theory (Richards 2016), but taking the liberty to reinterpret Richards’s ideas as issues of labeling.

A way to describe the idea that the Q-particle meets the EPP requirement of C is to say that the Q-particle makes it possible for C to project without the need for a specifier, which Richards (2016) calls “Support” (he was referring to T, but it is just as applicable to C). We believe that this idea of inducing a head to project by attaching a particle such as a Q-particle is the correct characterization. What is it that the Q-particle does to make it possible to induce projection by C? In Chomsky (2015), it is suggested that certain heads—the head he is concerned with is T, although it could just as well extend to C—are “roots” in the sense of Marantz (1997), and something has to attach to it to turn such a root into a syntactic object that can project, in the same way that v categorizes a verbal element. Although it is uncertain whether the bare form of T or C is a root, we share the intuition that attaching the Q-particle to C renders the C into something that can function in syntax; however, instead of straightforwardly adopting this notion of “root,” we suggest that the Q-particle, by attaching to C, gives C an independent morphosyntactic category status.

We can tell that the C-Q combination in Japanese is an independent category: The C-Q combination occurs as a morphosyntactically independent item that does not affect the accentuation of the preceding element. Most verb stems ending in a vowel such as ta’be ‘eat’ carry accent on the penultimate mora. When the present tense morpheme -ru is affixed to them, the tense morpheme occurs within the same metrical unit, so the rule of assigning accent to the penultimate mora shifts the accent (McCawley 1968). This is the case even when other suffixes such as the potential -(rar)e ‘can’, the causative -(s)ase, and the passive -(r)are intervene between the verb stem and the present tense morpheme, as shown in (12).

These indicate that the passive, the causative, and the tense affixes are all part of the same metrical unit as the stem.³

In contrast, the Q-particle ka, which is an enclitic, does not affect accentuation.

This indicates that while affixes such as the present tense morpheme combine with the preceding element to form a single metrical unit, there is a metrical boundary preceding the Q-particle that is due to the morphosyntactically independent status of the Q-particle relative to the preceding item.

What this indicates is that a head—in this case C—can be induced to project if it has independent categorical status, as indicated by the occurrence of a metrical boundary that precedes it, what Richards (2016) calls “Support” within his Contiguity Theory. This allows the C to fulfill the EPP requirement without the need for a specifier. If this is true, we should be able to find the same kind of effect in the classic case of EPP fulfillment without a specifier, pro-drop in Romance. The head that is involved is T, not C, and indeed, we find precisely the expected metrical boundary that points to T in Romance as having an independent categorical status, as Richards (2016) observes.

As first noted by Oltra-Massuet (1999; 2000) for Catalan verbs, and later extended by Guerzoni (2000) to Italian and by Oltra-Massuet and Arregi (2005) to Spanish, in Romance, stress typically occurs on the vowel that precedes the tense morpheme. This is shown for Spanish below (the data and the basic approach are drawn from Richards 2016).

These linguists suggest that there is a word-internal boundary for assigning metrical stress, and this boundary immediately precedes the tense, as illustrated for cantá -ba-is ‘sing-IMP.IND-2PL’.⁴

This is similar to what we observed for the Q-particle in Japanese, in the sense that both the TENSE.AGR in Romance and the Q-particle in Japanese have independent status, meaning that they are not incorporated into the metrical structure of the preceding metrical unit. This independent status of the head (C, T) induces the head to project without having to merge a specifier.

Pro-drop has been identified with rich agreement (Taraldsen 1978; Rizzi 1982). It is possible that beyond having the full repertoire of agreement features, the agreement morpheme itself may have a morphologically independent status, similar to the Q-particle ka in Japanese. For example, Alexiadou and Anagnostopoulou (1998) suggest that the agreement morpheme is pronominal-like, which would make it an independent lexical item. However, it is difficult to show this, so we will simply focus on the occurrence of the metrical boundary that precedes the T.AGR as an indication that the T.AGR functions as an independent category that can project.

Not all Romance languages allow pro-drop, and we would expect that in these languages the T does not have an independent categorical status. Richards (2016) observes that in French, which does not allow pro-drop, there is no metrical stress preceding tense, but instead the stress is on the final syllable of the word (unless the final syllable has a schwa, in which case the stress shifts to the penultimate syllable).

Thus the tense-agreement segment does not form an independent entity in French. See Richards (2016) for a similar analysis of English, another language that does not allow pro-drop.

We saw that the Q-particle functions to give C an independent categorical status, allowing the head to project without the need for a specifier. We saw precisely the same effect with the rich agreement in Romance, which makes T an independent category, in turn making pro-drop possible. Where T fails to have this independent status, as in French and English, pro-drop is not allowed.

If a particle fails to occur on C, C does not have independent status. It carries a grammatical feature (Q), and projection is made possible by feature sharing, as suggested by Chomsky (2013), by moving the wh-phrase to SPEC,CP. The Old Japanese example is repeated below.

The movement of the wh-phrase is likely triggered by focus, and once it moves to SPEC,CP, a POP arises. We follow Chomsky (2015) in assuming that the POP is resolved by the agreement feature, Q (or FOC), on the wh-phrase and C.

2.2 Expletive construction

Let us return briefly to the Q-particle/pro-drop account. As we argued, the Q-particle attaches to C to give C an independent categorical status, which induces C to project. The same holds for T in Romance, where rich agreement gives T an independent categorical status such that a specifier is not needed. Note that this approach diverges from Chomsky’s (2013) original idea of φ-feature agreement and labeling. In his analysis, agreement must occur on both the head and the specifier.

The sharing of the φ-feature by the DP and TP allows the top node to be labeled as <φ, φ> by drawing from both the DP and TP, thus resolving the POP. In our approach, there is no need for the specifier to carry a φ-feature, but rather, what is important is that T does. And it is this Tφ that requires some enhancement for it to be able to project, either by having some independent X⁰ entity attach to it, as in many Romance languages, or having an XP merge with it, as in English or French. In either case, the enhancement gives T sufficiently independent status to project.

Evidence to support our approach comes from so-called long-distance agreement. The English expletive there constructions are such an example; there appears to be semantically vacuous and inserted only to fill SPEC,TP:

The expletive there sentence differs from nonexpletive ones in that T’s φ-features are valued via a long-distance agreement relation with the associate located lower in the structure rather than in the specifier of TP. This is supported by the fact that the φ-features of T covary with those of the associate DP:⁵

The word order indicates that the associate is in the complement position of the copula. The definiteness restriction that the associate must be indefinite and nonspecific also suggests that it stays low. Hence, what occurs in SPEC,TP is a semantically vacuous element that does not carry φ-features. In these examples, then, T alone projects, which is what we predict.

Long-distance agreement exists not only in English. Icelandic, for instance, is famous for having quirky-case subjects, which occur in idiosyncratic lexical nonnominative cases depending on the main verb (Thráinsson 2007). When a subject bears quirky case, the finite verb, which would otherwise agree with the nominative subject, agrees with the nominative object instead:

This agreement pattern is understood if we think of Icelandic finite T as only targeting nominative nominals for φ-agreement. (21) is clearly a case of long-distance agreement, where T agrees with the object lower down rather than its specifier.⁶

2.3 Root clauses and labeling

We saw that in modern Japanese wh-questions, wh-in-situ is made possible by the Q-particle occurring on C; the particle gives C independent categorical status and induces it to project.⁷ However, there is one exception to the necessity of the Q-particle (Yoshida and Yoshida 1997; we use no as the Q-particle below; Q-particle drop is in the informal register and no is more appropriate in this register).

The Q-particle can never be omitted in embedded contexts.⁸

Why is the omission of the Q-particle possible in a root but not in an embedded clause? We suggest the following.

This makes sense since the root by its nature will not undergo further merger, so it need not be identified as input to Merge.

We can find further evidence that the root need not be labeled from declarative clauses. The root declarative clause normally does not contain an overt complementizer, but an overt complementizer is required in embedded contexts.

This array of data from the declarative construction indicates, again, that the root clause need not be labeled. The root C may have overt manifestation in the way of sentence-final particles such as ne and yo.

There is one interesting exception to the idea that the complementizer in subordinate environments must have overt existence. As originally observed by Saito (1986), in Kansai (Western) Japanese, the complementizer may optionally be left out.

As Kishimoto (2006) points out, this optional omission of the complementizer is only possible with the two verbs in this example, say and think. These verbs typically allow root-clause phenomena in their subordinate clause (Hooper and Thompson 1973; Miyagawa 2017). Thus, for example, Negative Constituent Preposing, a typical root phenomenon (Emonds 1969; 1976), is allowed with say or think (see Hooper and Thompson 1973 for extensive discussion).

From this, we could plausibly analyze complementizer drop as an instance of a root phenomenon occurring in the complements of say and think. On this analysis the subordinate clause without the complementizer does not have a label. A piece of evidence for this is that the subordinate clause without the complementizer cannot be the target of scrambling (Kishimoto 2006).

The ungrammaticality may be accounted for by the fact that the subordinate clause is not labeled, and hence cannot be the target of a syntactic operation such as scrambling.

Finally, we find further evidence for the root not requiring labeling from the topic construction. In Japanese the topic occurs at the left edge, being marked by -wa.

This topic occurs in the SPEC,TOPP position (see Miyagawa 2017); the head TOP itself does not have overt manifestation; thus, the particle -wa is required to avoid a POP, just as in the case of overt wh-movement in Old Japanese. This -wa can optionally drop in root clauses but not in embedded contexts (Kuno 1973; Saito 1985).

3.1 Case-dropping

We begin by looking at a phenomenon commonly called “case-drop,” in which a case marker is optionally omitted. The purpose for starting with this phenomenon is to show that this phenomenon of case-drop adheres to the notion of labeling, thereby showing the relevance of the labeling idea to Japanese inside the TP. This will set the stage for looking at case marking later in the section. It will also lay the groundwork for looking at differential object marking in Turkish.

While Japanese is generally a morphological-case language, there are instances in informal speech and in acquisition where a DP may occur without case marking (e.g., Saito 1985; Fujii and Ono 2000). This most typically happens with the object of a transitive verb. In the example below, the object is a wh-phrase to ensure that what is omitted is the accusative marker and not the topic marker -wa, which can also be optionally omitted (Kuno 1973), but not in the case of a wh-phrase (Saito 1985).

(32a) shows that the object of a transitive verb may occur without case marking; (32b) shows that this is only possible under adjacency. From the perspective of labeling, (32a) is of the form {αP, H}, where αP is the object ‘what’ and H is the head, V. This is an appropriate structure for labeling. In (32b), the locative PP ‘at the department store’ merges with V, which creates a βP. Then the object merges with this structure, most probably by internal merge, creating the new structure {αP, βP}. This triggers a POP that requires some additional mechanism to appropriately label the structure. Very clearly, morphological case marking, in this case the accusative case marker o, has this function. We will turn to the role of case marking below.⁹

In contrast to the accusative case marker, it has been observed that the nominative case marker cannot be omitted (Kuno 1973; Saito 1985).

This makes sense on the idea that the nominative DP is typically not adjacent to a head, but rather is adjacent to the projection of T, forming the structure {αP, βP}, which triggers a POP. However, there is one position in which the nominative -ga may be omitted: If the predicate is unaccusative, the nominative case marker may optionally be omitted (Fujii and Ono 2000).

The sole DP of the unaccusative verb is presumably adjacent to the verb, forming the structure {αP, H}, which is appropriate for labeling. Just as we saw with the object of the transitive verb, if the DP occurs away from the verb, a POP arises without case marking.¹⁰ This includes both a DP that occurs away from an unaccusative verb and the DP of an unergative verb.

In (35a), the DP occurs adjacent to the unaccusative verb ‘come’; because the structure here is {αP, H}, no POP is triggered. In (35b), a locative PP first merges with the unaccusative verb ‘come’, which forms a βP with the verb. To this the DP is merged, creating the structure {αP, βP}; a POP is triggered unless there is morphological case marking on the DP.¹¹

The possibility of leaving out both case markers, -o and -ga, the latter with unaccusatives, also holds in acquisition. Miyamoto et al. (1999) found that Aki (2;3–3;00) tended to drop -o 94% of the time. For -ga, it tended to be omitted for unaccusative/existential verbs anywhere between 63% and 95% depending on the verb, but only 23% of the time with unergative/transitive verbs. Thus, the pattern of case marking omission in informal adult speech for objects of transitive verbs and the subject of unaccusative verbs is also found in early stages of acquisition.

Just as φ-feature agreement functions to resolve a POP that is triggered by the structure {αP, βP}, we believe that morphological case marking functions in a parallel way to allow labeling of this structure. Thus, for example, once the subject adjoins to TP, creating the structure {DP, TP}, in English the POP is resolved by φ-feature agreement (Chomsky 2013), while in Japanese it is morphological nominative case marking that allows this structure to be labeled. We turn below to how case marking has this function.

Before doing so, we wish to note a prediction that our approach makes about a difference between English and Japanese with regard to the external argument. Merging the external argument to its VP-internal subject position creates the structure {DP, VP}. To avoid a POP in English, the DP is moved out of the VP, leaving only the VP as the sole target of search for labeling, and labeling ensues accordingly. In Japanese, the subject could likewise move out of VP to avoid a POP, but there is another possibility: If the subject can be case-marked, we predict that there is no need for it to move.

A particularly interesting demonstration of subject in-situ is found in Kumamoto Japanese, which is spoken on the southern island of Kyushu. Kato (2007) notes that Kumamoto Japanese (KJ) has two nominative markers, -ga for VP-external DPs and -no for VP-internal DPs. The latter is the same form as the genitive case marker, and she glosses it as genitive to distinguish it from -ga. She then notes the following SOV and scrambled OSV examples.

In (36a), the subject has moved to SPEC,TP presumably to satisfy the EPP requirement of T (Miyagawa 2001; see Nishioka 2019 for some additional Kumamoto data beyond the simple EPP cases). Thus it can only be marked with the nominative -ga. In (36b), the object has scrambled to the head of the sentence, presumably to SPEC,TP. This meets the EPP requirement of T (Miyagawa 2001; see Saito 2006 for an alternative to the EPP approach). This makes it possible for the subject to stay in its original SPEC,VP position, which makes it possible to mark it with -no. The possibility of -ga on the subject even in this OSV order indicates a second derivation in which the subject first moves to SPEC,TP, as in the (36a) example, then the object scrambles across it. The important point here is that the subject may stay in its original VP-internal position despite the {DP, VP} structure, because the subject is morphologically case-marked, which makes it possible to avoid a POP without moving the subject out of the verbal projection.

3.2 Case marking as projection blocker

How does the case particle block the projection of the XP that it attaches to? We will adopt Saito’s (2018) idea that case marking turns the node it attaches to “weak” so that it is no longer a candidate for labeling. We will state this in a way that takes advantage of some facts about Q-float.

We focus on the well-known observation that the DP with case marking behaves as a DP, and not as some projection of the case marker. Thus, for example, the DP with case marking c-commands a floated numeral quantifier (Shibatani 1977; Miyagawa 1989):

If the DP is inside a PP, it cannot support a floated numeral quantifier because it does not c-command the quantifier.

In Miyagawa (1989), it is argued that case markers have this property of not inhibiting the DP from c-commanding an element below the DP because the case marker adjoins to the DP.

The higher and the lower DPs together constitute the actual projection of the DP, and this DP may c-command elements outside the DP. In this adjoined structure, the case marker is not dominated by the DP, because it occurs “between” the two segments of the DP (May 1985).

Saito (2018), extending his earlier (2016) work, suggests that case marking renders the node “weak,” which has the effect of blocking projection of the case-marked XP. The search procedure for labeling would then naturally lead to the sister node projecting, which avoids a POP. This idea is formally stated as the following:

This is our rendition of Saito’s (2018) idea of making a node “weak.” In our view, the case marking does so because it is inert for labeling, as we can see by the adjuction structure it takes when adjoining to a DP. We can see this inert nature of case marking in the Q-float data. Because the case marker cannot furnish its own label, the only possible structure is by adjunction, in which the DP itself furnishes the label for the entire structure that contains the case marker. As a result, the DP is able to c-command a floated quantifier. In contrast, if a postposition occurs with the DP, the postposition does project, which prohibits the DP from c-commanding a floated quantifier.

The inert nature of case marking allows us to directly capture its function as a projection blocker, again following the general suggestion of Saito. The case-marked DP has the structure {DP, CASE}; given the adjunction structure, neither member of the pair dominates the other. This is equivalent to {XP, H}, and by the Labeling Algorithm, search would find H as the closest element to project. However, there is a problem. Case marking is inert for labeling, by the fact that it does not project, so the search does not obtain a viable candidate for labeling. The search thus is blocked, and it must turn to the other possible candidate, which is the sister to the DP, such as the TP.¹²

As a final note on case marking and projection blocking, recall in Chomsky (2013) that movement was one way to avoid a POP, by having an offending XP move out of the sister position of a YP, allowing the YP to project. If what we observed for case marking, and what we speculated in footnote 1 about movement are on the right track, they have a surprising commonality. In both cases, an entity that should project is somehow deprived of this capability, allowing the sister node to project. For case marking, it is due to the inert nature of the case marker head. In the case of movement, the lower copy, which still exists in the offending sister position to YP, has been bleached of categorical features that would otherwise inappropriately project, leaving only semantic features necessary to interpret the copy. In this way, the offending copy has been rendered inert for projection purposes to avoid a POP.

3.3 Differential object marking

What we observed for case drop in Japanese resembles differential object marking observed in other languages. Here we will look at differential object marking in Turkish to show that the presence and absence of morphological case marking we observed in Japanese is also found with DOM in Turkish, and that it confirms our view of case marking relative to labeling. In Turkish, the phenomenon directly shows the connection between case marking and labeling: Case marking shows up only when a POP occurs.

In Turkish the object may occur bare, or it may be accompanied by morphological accusative case marking (Kelepir 2001; 2004; von Heusinger and Kornfilt 2005; Kornfilt 2008; Özge 2011; von Heusinger and Kornfilt 2017; Kornfilt forthcoming).

The bare object differs in interpretation from the case-marked one; when bare, the object is interpreted as indefinite (or nonspecific), while the case-marked object receives a definite (or specific) interpretation (e.g. von Heusinger and Kornfilt 2005). We saw this for Japanese case-drop/case-marked environments as well. The difference between these two objects is not only in interpretation, but also in their position in the structure. The bare object cannot occur away from the verb, while the case-marked one may do so, as we saw in Japanese.

In fact, there is evidence that even in the basic SOV word order, the two objects occupy different positions: The bare object occurs in the original complement position of the verb, while the case-marked object occurs higher, outside the VP (e.g. Kennelly 1997; Zidani-Eroğlu 1997; Kelepir 2001). We can see this by the fact that a case-marked object must occur before a VP adverb such as ‘completely’, while the bare object occurs to the right of such an adverb.

Kelepir (2001) argues that case-marked objects “move above the VP to check their case features” and this gives rise to the presuppositional reading of specific/definite (Diesing 1992; Kennelly 1997; Zidani-Eroğlu 1997). We find similar differences in interpretation and position in other DOM languages (Aissen 2003).

The analysis that the cased DP moves outside of the VP is generally assumed in the field; in fact the DP is moving outside of VP if we assume that the adverb ‘completely’ occurs as a VP adverb. The object is thus in the structure {αP, βP}, and case marking is required on the object to resolve the POP.¹³

We assume that, just as in Japanese, the case marker in Turkish adjoins to the DP, creating the set {DP, CASE}. And just as in Japanese, case marking in Turkish is inert for the purpose of labeling, which blocks the projection of DP.

Thus, DOM is a phenomenon that directly reflects the need to label structures. In a structure that does not trigger POP, as in the case of the object occurring adjacent to V, no case marking occurs on the object DP because the structure {DP, H} is sufficient for labeling the projection. On the other hand, if the object moves out of the VP, the object is in the structure {DP_OBJ, XP} and case marking must occur on the object DP to avoid a POP.

3.4 Japanese and Turkish subjects

We have seen that there are two functions that a particle has relative to labeling.

To label a node, one of these is required when a potential POP would occur. In this regard, the subject in Japanese and Turkish illustrates these functions of a particle in a minimal way.

The subject in Japanese is marked with the case marker, which avoids the POP problem for {DP, TP}. In Turkish, the subject does not have any morphological case marking, but instead the T has agreement, which avoids the same POP.

While Japanese consistently works as described in both matrix and subordinate environments, the situation differs for Turkish. The complement subject may not have any case marking, as in the matrix clause, or it may be accompanied by the genitive case marker.

In (48a) the subordinate subject ‘you’ does not have any case marking, and the agreement is in the verbal paradigm, while in (48b) the same subject has the genitive case marker and the agreement is in the nominal paradigm. While (48a) is what we expect given that agreement resolves the POP, (48b) appears to be redundant, having both a case marker and agreement.

There is a recent analysis of the agreement pattern we see in the Turkish complement clauses that does not deem it redundant for the subject with the genitive morphological case marker to coexist with agreement.¹⁴ Contrary to the traditional generative approach that both the verbal and nominal agreements are in the same position (e.g. Kornfilt 2006; Ulutaş 2009), Colley and Davis (2017) argue that the two agreements occupy different positions. While the verbal agreement occurs on T as expected, having inherited it from C (Chomsky 2008), they argue that in the nominalized version, the C with the φ-probe raises to the small n head that takes the CP as its complement, so that the agreement occurs on this C-n complex. Below is the structure they give for the genitive-subject complement clause (slightly simplified to exclude material not relevant to our discussion).

The C⁰ head with the agreement feature raises to n, thereby expanding the domain from CP to nP (cf. Den Dikken 2007; also Gallego 2010, Wurmbrand 2013, Gallego 2018), which forces the agreement to hold in the nominal paradigm instead of the verbal paradigm, similar to simple nominal agreement, as in:

For the verbal agreement (VA), it is possible to argue that it occurs at T instead of C.

Here there is verbal agreement on the embedded verb and there is an overt complementizer. The complementizer does not show agreement but the tensed verb does, which suggests that verbal agreement is on T and not on C.

Nominal agreement (NA) exhibits a different pattern.

This reflects Colley and Davis’s (2017) analysis that agreement is on C that has moved to n and not on T.¹⁵ There is a piece of independent evidence for this idea that the nominal agreement is higher in structure than the verbal agreement. Kural (1993) has argued that the nominalizer is actually a complex that consists of the past tense marker -DI and the complementizer -K (realized as ğ for phonological reasons). Under this analysis, the gloss of (52) would be as in (53) below, and this suggests that we are really dealing with agreement on C.

The account above points to the conclusion that the co-occurrence of the morphological genitive case marking and agreement is not redundant: The agreement that occurs with the nominative DP, which lacks any case marking, occurs on T, while the nominal agreement occurs higher, on C-n. The nominal agreement, failing to occur on T, leaves a POP for TP, {DP, TP}. The morphological case marking addresses this POP, while the nominal agreement allows the C⁰/n⁰ to project. Below, we will see a similar case from Bantu that at first blush appears to be redundant, but on further analysis the DP with morphological case marking and the agreement occur at a distance to address potential POP in two different points in the structure.¹⁶

3.5 Augment vowels in Bantu

Are there dependent morphemes other than case marking that function to block projection? Bantu has what are called augment vowels, which attach as a prefix to DPs. Halpert (2012) argues that the augment vowel functions similarly to case marking; she assumes that it is self-assigned to the nominal as inherent case. This suggests that Bantu in fact has case marking despite its traditionally “caseless” characterization (see Baker 2003 and references therein). We will demonstrate this with the simplest cases.

The example in (54a) shows that ‘woman’ in the object position may occur without the augment vowel, while in (54b), it occurs with the augment vowel in the subject position; the augment vowel matches the noun class of the nominal ‘woman’. The example in (54c) shows that without the augment vowel this DP is ungrammatical in the subject position. As Halpert notes, the distribution of the augment vowel matches the distribution one expects of Case. In the object position it need not occur, but once a DP moves, the augment vowel must occur. And just as with DOM, as we saw in languages such as Turkish, a DP with the augment vowel has topic properties, a point we will return to below.

It is also possible for an object that has not moved to the front of the sentence to take the augment vowel.

This, again, parallels languages such as Japanese and Turkish: Morphological case marking on the object indicates that it is definite/specific, presumably showing that the object has moved from its original complement position to a position higher in the structure.¹⁷ This example also demonstrates that the augment vowel, though functioning like a case marker, is not part of a morphological-case system assigned by a head such as T or V, or the Dependent Case scheme. Augment vowels match the vowel of the noun class that the noun prefix belongs to (Progovac 1993; Baker 2003; Schneider-Zioga 2007), for example, o- for class 1 in example (55). Though not strictly a case marker, we agree with Halpert (2012) that it has the same function with regard to case marking; from our point of view, it is there to avoid a POP.

Assuming with Halpert (2012) that the augment vowel is a self-marking on the nominal that functions in a similar way to case marking in other languages, it would be reasonable to view augment vowels as separate morphemes that carry this function. We can in fact enhance Halpert’s insight about augment vowels by showing that it has the same function as case marking in blocking projection. We can adopt what we said for case marking in Japanese and Turkish: The augment vowel, which is the head in {H, DP}, cannot project because it is inert for projection. As a result, further projection is blocked.

Along with the augment vowel, which essentially functions as case marking, Bantu has agreement in which T agrees with the noun class of the element at the head of the sentence. The following Kinande examples show the subject, the object, and the locative moving to the head of the sentence, with the verbal element agreeing with what has moved.

As already noted, the fronted phrase must be accompanied by an augment vowel. Thus, in these examples, there is agreement on T, which induces projection, and the augment vowel, which blocks projection. Just as we saw for Turkish embedded constructions, this is redundant in appearance, but just as in Turkish, the fronted phrase turns out not to be in SPEC,TP.

As noted by Baker (2003) and Progovac (1993), the agreeing phrase must be interpreted as definite (or specific). This is illustrated with the object reversal order in (57).

In this reversal construction, the object is in a position to trigger agreement on the verbal inflection, and it must be interpreted as definite (i.e., as a topic). This is claimed to be true for all agreeing phrases. Baker notes that “[t]rue polysynthetic languages…always have agreement and always have dislocation” (2003: 112). What Baker means by “dislocation” is that he views the agreeing phrase such as the object in (57) as being somewhere above SPEC,TP—he assumes that it is in a higher SPEC,TP—in a position comparable to the dislocation position in languages such as Italian (e.g., Alexiadou and Anagnostopoulou 1998). On first blush, it appears that Bantu redundantly solves a POP: It has rich agreement (which allows pro-drop), and it has augment vowels that function similarly to case marking in blocking projection. However, Baker’s (2003) analysis shows that there is no redundancy. The agreement allows TP to project without a specifier, much as in Romance. But by the dislocation property of polysynthetic languages, of which Bantu is a typical case, the fronted DP moves to a “dislocated” topic position above the TP, as Baker argues. In that position, there is a POP, and the augment vowel is attached to resolve this POP. In this way, the augment vowel is similar to the topic marker -wa in Japanese.¹⁸