A comparative syntax of internally-headed relative clauses in Gur

Gur (or Mabia) languages which are spoken in West Africa have so-called internally-headed relative clauses (IHRCs), but they have not received serious attention in syntactic and typological research on IHRCs. In this article, building on detailed first-hand data, we describe the syntax and semantics of IHRCs in five Gur languages: Buli, Dagaare, Dagbani, Gurene, and Kabiyé. It is demonstrated that their IHRCs refute the syntactic and semantic generalizations proposed in the literature (Gorbet 1976; Cole 1987; Grosu 2002; Watanabe 1991; 2004). We also compare IHRCs in Gur and Japanese and argue that the existing semantic typology of IHRCs must be reconsidered, showing that properties of two types of IHRCs—restrictive and maximalizing IHRCS—do not necessarily show predicated correlations.


Introduction
Relativization in natural languages comes in limited varieties.Depending on a structural position in which a relativized head noun (H) appears, relative clauses are classified into two types: an externally-headed relative clause (EHRC) and an internally-headed relative clause (IHRC) (see especially Bodomo & Hiraiwa 2010;also Lehmann 1984;Keenan 1985;Bianchi 2002;de Vries 2002;Andrews 2007;Cinque 2013).Some languages have only one type (e.g.Lakhota; Williamson 1987), but others have both types (e.g.Japanese; Kuroda 1992), as illustrated in (1).In EHRCs, the relativized head noun (H) appears structurally outside the relative clause CP, as shown in example (1a).On the other hand, in IHRCs, it appears structurally inside the relative clause CP, as shown in example (1b) (we will present a more detailed definition shortly in Section 2).

Naomi-top
Ken-nom apple-acc buy-come-ben-past c-acc eat-past 'Naomi ate the apple that Ken bought for her.' (IHRC) Glossa general linguistics a journal of Hiraiwa, Ken, et al. 2017.A comparative syntax of internally-headed relative clauses in Gur.Glossa: a journal of general linguistics 2(1): 27.1-30, DOI: https://doi.org/10.5334/gjgl.40 IHRCs have attracted much attention in light of the search for universals and parameters: namely, an ultimate question that generative linguists want to answer is what makes IHRCs possible in some languages and impossible in others.For example, in the generative literature, it was often observed, in particular in the 1970s and the 1980s, that IHRCs are only found in OV languages (see Kuroda 1974;Gorbet 1974;Langdon 1977;Downing 1978).Cole (1987) asks a deeper question how such a word order generalization can be derived and goes on to link the word order generalization to the availability of a null pronoun, as formulated in (2). (2) Cole's Generalization (Cole 1987) IHRCs are restricted to languages with (i) SOV word order and (ii) a null anaphor.
According to Cole (1987), a null anaphor is subject to the Command Condition that an anaphor cannot both precede and command its antecedent (Langacker 1969, Ross 1969).
He assumes that IHRCs have the same structure as EHRCs, with a difference being which of the head nouns is pronounced, as shown in (3a)-(3b) (see also Ito 1986;Erlewine & Gould 2016 for a similar proposal).In OV languages, IHRCs and EHRCs do conform to the Command Condition.Therefore, IHRCs form a legitimate structure.On the other hand, in VO languages, IHRCs are not licit, because the null pronoun pro precedes and commands its antecedent, namely, the internal relativized head noun, as shown in (3c). (3)

IHRC
Gur (or Mabia) languages which are spoken in West Africa offer an indispensable testing ground for Cole's Generalization in (2) as well as for the simple OV generalization.Tellier (1989) was the first, to the best of our knowledge, to point out a counterexample to Cole's Generalization.He observes that Mooré, a Gur language spoken in Burkina Faso, has SVO word order and disallows null pronouns, but still allows IHRCs, as shown in example (4) (see also Gil 2000;Aldridge to appear;and Wilbur 2016 for similar evidence from Riau Indonesian, Tagalog, and sign languages).1 (4) Mooré (Tellier 1989; see also Peterson 1971;1974) [Fo sẽ yã daw ninga zaamẽ wã] kula me.you c saw man rel yesterday dem went-home prt 'The man who you saw yesterday went home.' (IHRC) Mooré is not a single, accidental exception, however.Hiraiwa (2003;2005b;2009a;2009b;in press) have shown that Gur languages in general present consistent evidence against Cole's Generalization or the word order generalization. 2As we show in the following sections in detail, (at least) five Gur languages allow IHRCs, even though they are VO languages, their noun phrases are modified postnominally, and none of these languages allows null anaphora.An example is given in (5) from Buli.
(5) Buli Amoak nya [Atim ale sua naa buui la].Amoak saw Atim c own cow rel dem 'Amoak saw the cow that Atim owned.' (IHRC) As we will see below, various linguists have made alternative generalizations, but none of them predicts that Gur languages can have IHRCs.Despite such empirical and theoretical importance of Gur IHRCs, however, they have not duly attracted attention in the literature on IHRCs.Thus, the main goal of this collaborative article is to present the result of our careful study on IHRCs in five Gur languages-Buli, Dagaare, Dagbani, Gurene, and Kabiyé-and to advance the current understanding of the syntactic mechanism underlying IHRCs.We show that data from the Gur languages demand reconsideration of the various existing proposals on the syntax (and semantics) of IHRCs.We also compare the syntax of IHRCs in Gur with the syntax of IHRCs in Japanese.This is important in two respects.First, most, if not all, of the languages that attest IHRCs are not easily accessible, but highly delicate grammatical judgement is vastly available in Japanese.Second, for that reason, Japanese is one of the very few languages whose IHRCs have been studied in great detail since the 1970s.Thus, we believe that it is illuminating to compare IHRCs in Gur and Japanese.The organization of this paper is as follows.Section 2 gives an overview of general syntactic characteristics of IHRCs in the Gur languages.Section 3 examines various syntactic and semantic generalizations proposed in the literature and shows that Gur IHRCs do not fall under any of them.Section 4 compares IHRCs in Gur and Japanese and argues that the Gur languages resist the existing semantic typology.Section 5 concludes the article and discusses future issues.

General syntactic features of IHRCs in Gur
Following Lehmann (1984), Bodomo & Hiraiwa (2010) A mere linear typology is not sufficient because a relative clause can be internally-headed or externally-headed when H appears at either edge.Namely, the fact that H appears at the edge does not automatically mean that it is located outside the relative clause CP.Among the six possibilities, in-situ EHRCs are logically impossible.On the other hand, right-headed IHRCs have not been attested yet to the best of our knowledge, even though they are logically possible. 4,5 In the following subsections, we will show that the four Gur languages (except Dagaare) have in-situ IHRCs (Section 2.1) and that all the five Gur languages also have left-headed IHRCs (Section 2.2).It should be emphasized, however, that what is important for our main goal of this article is the fact that the Gur languages have in-situ IHRCs (except Dagaare) and hence discussions of left-headed IHRCs will be kept to minimum.

In-situ IHRCs
4 An anonymous reviewer points out that what we call left-headed IHRCs have a structure syntactically equivalent to Kayne's (1994) raising analysis of EHRCs (or to a similar proposal by Bianchi 1999).Even though that is the case, Kayne's original analysis of externally-headed relative clauses (i.e. the structure in which a D takes a CP as its complement and H remains in the specifier of the CP) has met various objections (see Bhatt 2002, andin particular Donati &Cecchetto 2015).Most notably, Donati & Cecchetto (2011;2015), rightfully point out that Kayne's analysis suffers from a problem that a D in those languages with EHRCs (such as English and Italian) cannot take a CP as its complement.To the extent that their objection is correct (and we think it is), EHRCs cannot have the same structure as that of left-headed IHRCs in (7).Rather, they have the structure in which an external D takes an NP outside the relative clause (see Iatridou et al. 2001;Bhatt 2002;Donati & Cecchetto 2011;2015).This in turn suggests an interesting possibility that syntax of (at least a certain type of) IHRCs is dependent on a clausal determiner that takes a CP.See Hiraiwa (2005b;in press) for relevant discussions. 5An anonymous reviewer asks whether the apparent absence of right-headed IHRCs might be due to the prohibition against rightward movement (see Kayne 1994).It is a little more complicated, however, because it is possible for H to be moved leftward (e.g. to the specifier of CP), followed by a leftward movement of a remnant constituent (e.g.TP).This strands H at the right periphery, but it may still count as internallyheaded in that H is not completely outside the entire relative clause.Therefore, right-headed IHRCs should be logically/syntactically possible, but still they are unattested so far.
IHRCs in the Gur languages focused on in this article are illustrated below (we will return to Dagaare in Section 2.2).The following examples are in-situ IHRCs, in which the relativized head noun appears in its original (i.e. in-situ)  We assume that the post-nominal placement of the nominal determiner is due to movement of NP to the specifier of DP (we assume that D has an EPP/edge feature). 9There is a semantic difference between maa and la.Both are definiteness markers, but la is used when both interlocutors know what is being referred to and its existence is part of the world knowledge of the speaker and the hearer.In that sense, it is more like a distal demonstrative and similar to yɔ in the Kabiyé example in (17).In this article, we will use maa in all the Dagbani relative clause examples for consistency. 9"NP" here is simplified and there should be more functional projections, such as noun class/gender.Some Gur languages (e.g.Buli and Kabiyé) still retain a productive noun class system with different class pronouns and agreement, and others only have a relatively simplified system.
Similarly, the right-edge placement of the clausal determiner in the Gur languages results from essentially the same derivation.In this case, CP moves to the specifier of DP ( assuming the same feature on D).  20)-( 23) and hence an indefinite interpretation is not allowed in IHRCs.This indicates that its function is to mark definiteness of the noun phrase modified by the relative clause. 11In some Gur languages such as Dagbani, some verbs may allow an indefinite interpretation under certain metaphorical interpretations, as shown in example (i).We will not go into details here, because relative clauses in Dagbani are generally followed by definite determiners and an investigation of the nature of such a phenomenon as (i) goes beyond the scope of this article.

(i) Dagbani
[dp[cp Ata ni bori pag so] (maa)] ka na.Ata c like.imperf woman rel d come loc 'The woman that Ata likes has come.' 'A woman with (some) qualities that Ata likes has come.' (in-situ IHRC) Turning our attention to the composition of internal head nouns in these Gur languages, note that they are obligatorily marked by determiner-like elements, glossed as rel(lativizer).Williamson (1987) makes an important observation that internal head nouns in IHRCs must be indefinite and this is called the Indefiniteness Restriction.Thus, the definite internal head nouns are excluded as shown below.
( The relativizer in Kabiyé does not function as a specific-indefinite determiner by itself, either.But at least it has an indefinite meaning, in that it is the same form as a wh-pronoun.As shown in (32), the relativizer in Kabiyé is decomposed into a homorganic nasal morpheme N and a noun class pronoun.The homorganic nasal morpheme also appears in demonstratives, wh-expressions, numerals, etc.Notice that the wh-expression in (32b) is exactly the same as the relativizer in (32a).Furthermore, the distal demonstrative in (32c) is also morphologically identical with the relativizer except the tone on the homorganic nasal morpheme (HL vs. HH).
( Given the semantic difference between (32a)/(32b) and (32c), we assume that the high tone has definite semantics, while the low tone has (specific) indefinite semantics in Kabiyé. 14t the current stage of understanding of the syntax of noun class pronouns, we are not yet in a position to propose a precise position of those relativizers within the noun phrase.This has to be left for future research.

Left-headed IHRCs
In addition to in-situ IHRCs, all the five Gur languages have what we call left-headed IHRCs (see also Basilico 1996 for The Mesa Grande dialect of Diegueño and Mojave). 15 In the following examples, the relativized head noun appears at the left periphery-leftheaded IHRCs.Left-headed relative clauses can be internally-headed or externally-headed, but there is reason to believe that those left-headed relative clauses in the Gur languages are internally-headed, considering some evidence presented in Hiraiwa (2005b;2009a;2009b) and Bodomo & Hiraiwa (2010) Basilico (1996) refers to this kind of movement of H as head fronting.In (ib), the relativized head noun is fronted to the left edge position of the relative clause (cf.(ia)).The fact that the fronted H in example (ic) cannot receive subject case-marking from the main clause shows that the head noun still remains internal the the relative clause.
(i) The Mesa Grande dialect of Diegueño (Basilico 1996: 501, 505) a. [xaṭcok(-∅) wi:m tuc-pu]-c n y iL y .dog-obj rock-comit I.hit-dem-subj black 'The rock that I hit the dog with was black.' (in-situ HIRC) b. ['wil y 'xaṭ(-∅) n y i-m 'tu:-pu]-c n y iL y cis.rock dog-obj that-comit I.hit-dem-subj black.indeed'The rock that I hit the dog with was black.' (left-headed HIRC) c. *['wil y -pu-c 'xat(-∅) n y i-m 'tu:-pu]-c n y iL y cis.rock-dem-subj dog-obj that-comit I.hit-dem-subj black.indeed'The rock that I hit the dog with was black.' Tellier (1989) observes that a head noun in IHRCs in Mooré (a Gur language) can undergo VP-internal short movement from the base order in (iia) to the derived order in (iib).
(ii) Mooré (Tellier 1989: 310-311) a One piece of evidence that the relativized head noun of left-headed relative clauses in the Gur languages is still located internally to the relative clause CP comes from pied-piping in PP relativization and possessor relativization (see Hiraiwa 2005b;2009a;2009b;Bodomo & Hiraiwa 2010).
In the left-headed IHRC example in (39c)-(40c), the postposition and the possessed noun phrase are pied-piped with the head noun, respectively.If [gbong kuui zuk] 'roof rel on' in (39c) were outside the relative clause CP, it would be incompatible with the matrix predicate zyuagi 'big' as its subject, just as example (39d) shows.Thus, the pied-piped relativized head noun [gbong kuui zuk] in (39c) must be still internal to the relative clause CP.Similarly, in example (40c), [gban kaai naang-ka] 'book rel cover-d' appears at the left edge of the relative clause CP, but example (40d) shows that when the verb da 'buy' takes the same pied-piped phrase as its object, it necessarily means 'Atim bought the cover of the book'.Again, this shows that [gban kaai naang-ka] in (40c) is still inside the relative clause CP.These facts indicate that left-headed relative clauses are internally-headed in the Gur languages.In the remainder of this article, we will mainly use in-situ IHRCs (Buli, Dagbani, Gurene, and Kabiyé) and left-headed IHRCs (Dagaare), in examining the existing typological generalizations.
Before closing this subsection, let us comment on some differences exhibited by Dagaare.The definite determiner a in Dagaare also functions as a clausal determiner in relative clauses, just as in the other Gur languages.This determiner is, again, obligatory in relative clauses, as shown in (41).Within a DP, the same element functions as a nominal determiner.Dagaare is distinct, however, in that the determiner precedes an NP and a CP, unlike the other four Gur languages.We can make sense of this fact if we assume that D in Dagaare uniformly lacks an EPP/edge feature and does not attract NP/CP to its specifier.The relativizer na in Dagaare is also in a sharp contrast with the other Gur languages in that it is identical with the distal demonstrative na and optional even in definite relative clauses (and its omission does not affect semantics), as shown in (44a)-(44b).In other words, Dagaare lacks a dedicated relativizer element.Neither does it have any indefiniteness function, unlike the other Gur languages.17Thus, this morpheme is completely distinct from the specific-indefinite determiner kanga in (44c).Each of these asymmetries between Dagaare and the other Gur languages might be contributing factors for the absence of in-situ IHRCs in Dagaare.The internal head noun, being definite and unmarked in Dagaare, must always be positionally marked at the left periphery of the relative clause (see Basilico 1996).But we do not have any conclusive evidence at this moment (see also Hiraiwa 2009a; b for relevant discussions).

Structure of IHRCs in Gur and summary
Given the right-most placement of the determiner in these Gur IHRCs (except Dagaare), we have proposed that the word order is derived from a remnant CP movement, as shown in (36a)-(36b We have shown that the four Gur languages (except Dagaare) have both in-situ IHRCs and left-headed IHRCs.In the remainder of this article, we will focus on examples of in-situ IHRCs (except Dagaare) as they are uncontroversially internally-headed.

Syntactic generalizations and counter-examples
In this section, we examine syntactic and semantic generalizations proposed in the previous literature and argue that Gur languages provide important counter-evidence against them.
As mentioned in Section 1, the Gur languages refute the word order generalization because they have consistent VO order and disallow null pronouns.In addition to the word order generalization, there have been other typological generalizations proposed in the literature on IHRCs (see also Basilico 1996;Watanabe 2004 for summaries).In this section, we will test each of the generalizations against IHRCs in Gur and show that none of them predicts that IHRCs are allowed in Gur.Watanabe (1991) proposed that a mechanism of wh-in-situ is crucially at work in making IHRCs available. (47) Wh-in-situ Generalization (Watanabe 1991) IHRCs are limited to languages with wh-in-situ.
In wh-in-situ languages, only a null operator moves to the specifier of CP in wh-questions, leaving the wh-phrase itself in-situ.Watanabe argues that the same happens in IHRCs: only a null operator moves, making it possible to leave the head noun in-situ, as shown in ( 48).

C
As his generalization predicts, wh-movement is optional in all the Gur languages above (except Dagaare), as shown in ( 49)-( 52).Overt wh-movement in Gur, unlike relativization, is either accompanied by focus marking or a different complementizer, but an overt complementizer does not appear in wh-in-situ. 18 18 Bodomo (1997) notes an example of wh-in-situ.
The following examples demonstrate that wh-movement out of a relative clause is simply ungrammatical, whether it is overt or covert.
This seems to be limited to a copular construction, however, and in fact, wh-movement results in ungrammaticality in this case.
(ii) Dagaare (Bodomo 1997 Although the Gur languages apparently conform to the wh-in-situ generalization, the generalization itself is not without an exception: as Watanabe ( 2004) admits, Imbabura Quechua has obligatory wh-movement, while it allows IHRCs (Cole & Hermon 1994).In light of this, Watanabe (2004) extends the notion of wh-in-situ and its analysis in (41a) to focus-in-situ as well.This is because Imbabura Quechua still allows focus-in-situ, while it has obligatory wh-movement.He argues that wh-in-situ and focus-in-situ share the same mechanism in which some element agrees with/ checks a feature without movement as shown in (48).Furthermore, being a wh-insitu or focus-in-situ language is not sufficient for a language to have IHRCs because there are languages like Mandarin Chinese, which is a wh-in-situ language, but lacks IHRCs.
Watanabe, then, goes on to argue that languages also make use of the same mechanism in determiner systems, proposing the HIRC-Indeterminate Generalization.
(59) HIRC-Indeterminate Generalization (Watanabe 2004: 88) Languages with an indeterminate system make available for ordinary nominal expressions the long-distance dependency (checking or binding) used by the indeterminate.This recruitment makes HIRC possible.
The correlations are illustrated in the following table in (60).Note that in these languages with IHRCs, wh-pronouns and indefinite pronouns show morphological similarity.Such a system is called an indeterminate system in Watanabe (2004).Setting theoretical details aside, the generalization in (59) states that IHRCs are only found in languages in which indefinite pronouns are built based on wh-pronouns.Under this generalization, English lacks IHRCs because it does not have an indeterminate system.
Mandarin Chinese, although it has an overt indeterminate system, disallows IHRCs because it lacks a determiner system that instantiates the long-distance dependency required by ( 59).
( In conclusion, we have shown that IHRCs in Gur disconfirm two of the existing syntactic generalizations-the word order generalization and the indeterminate generalization, while the wh-in-situ generalization is refuted by Imbabura Quechua. 19 19 An anonymous reviewer raises a question whether the word order generalization could be reinterpreted as a generalization about the directionality of D: IHRCs may be found only in D-final languages.While this can straightforwardly explain the absence of in-situ IHRCs in Dagaare (because it is the only D-initial language

A comparative syntax-semantics of IHRCs in Gur and Japanese
As we have seen above, the Gur languages pose an important challenge for existing syntactic typological generalizations.Grosu (2002), in contrast, investigates semantic aspects of IHRCs and their correlations with syntactic properties.He proposes to classify IHRCs into two different semantic types: restrictive IHRCs and maximalizing IHRCs.
Restrictive IHRCs allow existential interpretation, while maximalizing IHRCs do not.Consider the following paradigm from Grosu (2002).While Lakhota IHRCs allow existential interpretation as shown in (62), Quechua IHRCs do not.In other words, the Quechua example in (63) cannot be continued by '... and two were bad.' (Srivastav 1991:683) Quechua (Maximalizing IHRC) (Srivastav 1991) [Nuna ishkay bestya-ta ranti-shqa-n alli] bestya-m ka-rqo-n.man two horse-acc buy-perf-3 good horse-valid be-past-3 'The two horses that the man bought were good horses.'According to Grosu (2002), there are further interesting correlations between the semantic types and syntactic properties.As shown in table (64), restrictive IHRCs allow existential interpretation and stacking, and are insensitive to island.Maximalizing IHRCs do not allow existential interpretation or stacking, and exhibit island-sensitivity.In the rest of this article, we will examine whether the predicted correlations exist in Gur IHRCs, through a comparison with Japanese.Given the island-sensitivity of IHRCs in Gur (Section 3.1) and Japanese (Watanabe 1991, Kuroda 1999), it is predicted that they belong to the maximalizing type. 20In other words, Grosu's (2002) typology expects Gur IHRCs (i) to resist existential interpretation and (ii) to disallow stacking.Moreover, they should also pattern with IHRCs in Japanese (as well as Quechua and Navajo) in other important respects.As we will see, however, the expected correlations are not observed.
among the five Gur languages), we are aware of a number of D-initial languages that have IHRCs.Passamaquoddy is one such language (Bruening 2001).The Word Atlas of Linguistic Structures (WALS) also lists 10 languages that are D-initial and have IHRCs (accessed on July 29, 2016). 20Island-(in)sensitivity of Japanese IHRCs has been a point of controversy.Uchibori (1991), Mihara (1994), andHoshi (1995) argue that IHRCs in Japanese are not sensitive to islands.But Kuroda (1999) has detailed counterarguments that they are indeed subject to the Complex NP Constraint.We also add here that constructing an island-insensitive example of IHRC requires a lot of control over contexts and other factors, while it is quite simple to show that EHRCs in Japanese are island-insensitive.

Existential interpretation
First of all, we show that semantic interpretation of Gur IHRCs is the same as Japanese IHRCs: IHRCs in these languages do not allow existential interpretation.In example (65a), the second sentence leads to a contradiction because the IHRC indicates that the three mangoes are the only mangoes that were there.On the other hand, the EHRC is felicitous in the same discourse as shown in example (65b).
(66) Buli a. #[Atim ale da mango-ta nga-ta tii la] masa.Alege nga-ta an Atim c bought mango-pl nc-3 rel dem good but nc-3 neg masa.good 'Three mangoes that Atim bought were good.But three were not good.rel Eso bought d 3sg-be good but 2 be.neg good 'Two mangoes that Eso bought were good, but two were not good.'(left-headed IHRC) (70) Dagaare #[A mongo-ri ata na Dakoraa nang da] veɛlɛ la kyɛ ka ata na meng D mongo-pl 3 rel Dakoraa c buy be.goodf but and 3 dem also ba veɛlɛ.neg be.good 'Three mangoes that Dakoraa bought were good, but three were not good.'(left-headed IHRC) The Gur languages exactly pattern with Japanese IHRCs in the absence of existential interpretation.This means that the IHRCs in these languages belong to the maximalizing type, in Grosu's typology.

Stacking of IHRCs
With this in mind, let us look at stacking of IHRCs.According to the semantic typology in (64), IHRCs of the maximalizing type are expected not to allow stacking.This prediction, however, is not completely borne out by the Gur languages.
Buli While the split among the Gur languages remains to be explained ultimately, it is at least worth pointing out that there is no principled correlation between stackability of IHRCs and the other properties of the IHRCs.Thus, the data from the Gur languages call the semantic typology into question.
In fact, Japanese also refutes the alleged correlation between maximalizing IHRCs and stackability.In the following example, an IHRC is stacked by another IHRC. 23i) Japanese (Grosu 2002 on-dat apple-nom be-past c-acc kitte-kure-ta no]]-o tabe-ta.cut-ben-past c-acc eat-past 'Ken ate the apple that was on the table that Naomi cut for him.'

Question-answer
Another hallmark property of maximalizing IHRCs is that they cannot be used as an answer to a wh-question, as they are not semantically restrictive.Matsuda (2002) makes an observation that an IHRC in Japanese cannot be used as an answer to a wh-question.Consider the discourse in (77).Japanese IHRCs are allegedly not restrictive (Kuroda 1992;Shimoyama 1999;Grosu 2002) and they are not felicitous when used to answer a whquestion (see Hiraiwa in press).
(77) Japanese a. Dono ringo-o tabe-ta no? which apple-acc eat-past c? 'Which apple did you eat?
b. [[Teeburu-no-ue-ni at-ta] ringo]-o tabe-ta yo.table-gen-on-loc be-past apple-acc eat-past prt 'I ate the apple that was on the table.' (EHRC) c. #[Ringo-ga teeburu-no-ue-ni at-ta no]-o tabe-ta yo.apple-nom table-gen-on-loc be-past c-acc eat-past prt 'I ate the apple that was on the table.' (IHRC) In all of the five Gur languages, however, it is perfectly grammatical to answer a whquestion with an IHRC.(left-headed IHRC) However, this example is ungrammatical not because of the stacking, but because of two other conditions being not satisfied.First, there is a conflict in the temporal order of the events expressed by the relative clauses.It would mean that Mary read a long paper before John wrote it.Second, the two IHRCs do not satisfy Kuroda's Relevancy Condition (Kuroda 1974;1975-76;1976-77;1992).
(ii) The Relevancy Condition (Kuroda 1975-76: 86) For a pivot-independent [KH: internally-headed] relative clause to be acceptable, it is necessary that it be interpreted pragmatically in such a way as to be directly relevant to the pragmatic context of its matrix clause.
( These asymmetries between Japanese IHRCs and Gur IHRCs suggest, again, that the latter is not of the maximalizing type, if we take Grosu's (2002) typology as its face value.A weaker interpretation of the results is that the cluster of properties bifurcating maximalizing and restrictive IHRCs may not correlate as previously thought.It should be noted that stacking may not offer a reliable test for semantic classification.sg = singular, sid = specific indefinite determiner, subj = subject marker, top = topic, valid = validator nominal determiner makes a noun definite, a clausal determiner makes the entire relative clause necessarily definite.The four Gur languages with in-situ IHRCs require the clausal determiner at the edge of the relative clause, as shown in ( a mongo (na) Dakoraa nang da].1sg eat f d mango rel Dakoraa c buy 'I ate the mango that Dakoraa bought.' (left-headed IHRC) Buli a. Ka mango kuna ate fi de? f mango which c 2sg eat 'Which mango did you eat?' b.N de-ka [Atim ale da mango kuui la].1sg eat-f Atim c buy mango rel d 'I ate the mango that Atim bought.' (in-situ IHRC) c. N de-ka [mango kuui ate Atim da la].1sg eat-f mango rel c Atim buy d 'I ate the mango that Atim bought.' , and Hiraiwa (in press), we assume the following (representational) definitions of structural classification.First, relative clauses are divided into two types, depending on structural positions of H.A relative clauses is externally-headed when H appears structurally outside the relative clause (CP by our definition) and projects an NP that dominates it (or has it as an adjunct).In contrast, a relative clause is internally-headed if H appears structurally inside the relative clause and the NP is dominated by the relative clause.In addition to this structural classification, relative clauses are also divided into three types, depending on linear positions of a relativized head noun H.A relative clause is left-headed, if H appears to the left of the relative clause, while it is right-headed, if H appears to the right.It is called in-situ, when H appears within the relative clause.This is summarized in (7) below.
CP ... [ NP H ] ... ]] Left-headed [ DP [ NP H [ CP ... e H ... ]]] [ DP [ CP [ NP H ] ... t H ... ]] Right-headed [ DP [ NP [ CP ... e H ... ] H ]] The relativizer in Buli is not a specific indefinite determiner per se, but still it yields indefiniteness.As shown in (30), the first part of the relativizer changes according to the noun class of the head noun it attaches to.Thus, it is possible to analyze the relativizer baai as consisting of a noun class pronoun ba (accompanied by vowel lengthening) and an element expressing indefiniteness -i, which is in parallel with the specific indefinite determiner in Dagbani and Gurene.In fact, Buli expresses an NPI/an indefinite pronoun by combining the relativizer and another noun class pronoun, as shown in (31) (wa: Class 1, ba: Class 2).
The relativizers in Gur languages share a function of indefiniteness marking.Indeed, just as Williamson (1987) observes for Lakhota, they are morphologically identical with specific-indefinite determiners in Dagbani and Gurene, as shown in examples (28)-(29).
Atia buy.perf market yesterday d lose.perfprt 'The goat that Atia bought at the market yesterday got lost.' (left-headed IHRC) Bodomo & Hiraiwa 2010 different from the other languages in that its in-situ IHRCs are very marginal.Therefore, in this article, only examples of left-headed IHRCs are provided for Dagaare.An example of a left-headed IHRC and a highly marginal in-situ IHRC example in Dagaare are illustrated below (seeBodomo & Hiraiwa 2010: 959 for the left-headed structure of IHRCs and its supporting evidence in Dagaare).Thus, left-headed IHRCs in the Gur languages have the following schematic structures.
Watanabe (2004)t (e.g.Cocopa and the Imperial Valley dialect of Diegueño(Gorbet 1976)), we would like to note here that the five Gur languages discussed in this article fail to replicate such short VP-internal head fronting and parasitic gap licensing as stated inTellier (1989)(and one of the co-authors' brief consultant work with a Mooré speaker(October, 2003)could not confirm Tellier's data).It remains to be seen if short VP-internal head fronting is a peculiar feature of Mooré IHRCs among Gur languages, calling for more extensive work.It is also worth noting here that none of the relativizers in the five Gur languages has any clear focusing function, contrary to the focusing function ascribed to the relativizer ninga in Mooré inWatanabe (2004).(38).a. [ DP [ CP H i -rel ... (C) ... t i ... ] D ] (Buli, Dagbani, Gurene, Kabiyé: left-headed IHRC) b. [ DP D [ CP H i -rel ... (C) ... t i ... ] ] (Dagaare: left-headed IHRC) An anonymous reviewer asks whether the interpretive differences between (39c)/(40c) and (39d)/(40d) is due to the difference between the relativizing suffix and the definite determiner.However, Dagaare shows that it is not the case, because Dagaare lacks a dedicated relativizing suffix.
movement to the specifier of DP, while H is obligatorily dislocated to the edge of CP, as shown in (46).The movement of the head noun to the specifier of CP is probably best analyzed as A-bar movement because Gur languages lack scrambling and the movement of H requires the same complementizers as the ones required by wh-movement and focus movement (see Hiraiwa 2005b).
). In-situ IHRCs have H in-situ within CP, while left-headed IHRCs have H dislocated to the left-edge of CP (and H remains there).Dagaare, which only has leftheaded IHRCs, is different in that CP does not undergo None of the five Gur languages, however, possesses an indeterminate system.The data in the following table indicate that their wh-pronouns and indefinite pronouns show no morphological similarity.Thus, the Gur data demonstrate that the HIRC-Indeterminate generalization must be reconsidered. .
, Dagbani, and Gurene permit stacking of relative clauses as shown in examples (71)-(73), where two relative clauses are stacked without any overt coordinator. 21en that coordination of clauses in these languages require overt coordinators, it is reasonable to think that the relative clauses above are stacked.In contrast, Kabiyé and Dagaare do not allow stacking of relative clauses.Instead, both languages require the coordinator ni/ka, as shown below.22 21Notice the absence of a clausal determiner in the inner IHRCs.Recall, as we observed in Section 2, that a clausal determiner makes the entire relative clause definite.If there were a clausal determiner in an inner IHRC in the stacking examples, that would violate the so-called Indefiniteness Restriction (seeWilliamson  1987), because the inner IHRC would become definite.In other words, stacking data show that the Indefiniteness Restriction is also at work in the Gur languages.
.perf Atia c buy.perf goat rel market yesterday d 'I saw the goat that Atia bought at the market.'(in-situIHRC)Atiabuy.perf market yesterday d 'I saw the goat that Atia bought at the market.'(left-headedIHRC)Thiswouldbe unexpected if IHRCs in Gur were really of the maximalizing type as those in Japanese are.The fact that they can answer a wh-question indicates that they are restrictively interpreted, just as English relative clauses can answer a wh-question, as shown in example (83).I ate the mango that Dakoraa did not buy.' (left-headed IHRC)The results of our study are summarized in the table below.