2. Basic properties of Isbukun Bunun and its correlative construction

Bunun is one of the Formosan languages spoken in Taiwan. Throughout this paper I confine my attention to the dialect of Bunun known as Isbukun Bunun and I use the term ‘Isbukun’ to refer to this particular Bunun dialect.Footnote ² In particular, the current study is based on the Isbukun dialect spoken in Taitung County.

Isbukun Bunun is predicate-initial, and has two NP case markers (see He et al. Reference He, Zeng, Li and Lin1986, P. J.-K. Li Reference Li and Li1997, Huang Reference Huang and Li1997, Jeng Reference Jeng, Zeitoun and Li1999, Zeitoun Reference Zeitoun2000). The nominative marker is a and the non-nominative, or oblique, one is mas.Footnote ³ As can be seen in (3a), both of them can be omitted in colloquial speech. When not omitted, they are placed immediately before the NP, as in (3b).

1. (3)

   1. a. Ma-ludah Tahai Aping.

      av-beat Tahai Aping

      ‘Tahai beats Aping.’

   2. b. M-in-aun  a  Tahai  mas  utan.

      av<pfv>eat  nom  Tahai  obl  sweet.potato

      ‘Tahai ate sweet potatoes.’

Though the case system for NPs is somewhat impoverished in Isbukun Bunun, the case paradigm for personal pronouns is complete and case specification is obligatory for pronouns. Table 1 provides the complete list; examples are given in (4).

1. (4)

   1. a. H-in-uud saikin davus

      av<pfv>drink 1sg.nom wine

      ‘I drank the wine.’

   2. b. Ma-ludah  kasu  zaku

      av-beat  2sg.nom  1sg.obl

      ‘You beat me.’

Table 1:

Personal pronouns (adapted from Jeng (Reference Jeng, Zeitoun and Li1999))

Like personal pronouns, demonstrative pronouns can also be marked with their associated case,Footnote ⁴ as shown in Table 2 and example (5).

1. (5) Manah  bunun-an  tulkul-tan

   av.shoot  person-this.nom  chicken-this.obl

   ‘This man shot the chicken (here).’

Table 2:

Demonstrative pronouns

Like many other Formosan and Western Austronesian languages, Isbukun Bunun permits a range of arguments to serve as the syntactically most prominent NP, marked by the a nominative case marker of the clause. Such an a-marked phrase has been known by a number of names in the Austronesian literature, including “subject”, “topic”, and “pivot” – here I will refer to it as the subject for ease of presentation, though this term should be regarded with caution. The syntactic subject relation can be borne by a wide range of arguments: not only by the external argument (the actor), but also by noun phrases bearing semantic roles such as patient, theme, goal, benefactive and instrument. Importantly, all verbs in Bunun must be inflected with their associated morphology marker; that is, the semantic role or grammatical function of the surface subject has to be encoded on the verb.Footnote ⁵ Example (6) gives the voice marking system in Isbukun Bunun.Footnote ⁶

1. (6)

   1. a. Actor: ma-, m-, Ø

      m-aun  a  Tahai  mas  acipul-tan  laupaku.

      av-eat  nom Tahai  obl  corn-this.obl  now

      ‘Tahai is eating corn now.’

   2. b. Patient: -un

      kaun-un-in  a  acipul-a  mas  Tahai.

      eat-pv-pfv  nom corn-that. nom  obl  Tahai

      ‘That corn has been eaten by Tahai.’

   3. c. Locational phrase: -an Footnote ⁷

      na-sabah-an     adi  lumah  mas  Tahai  aip.

      fut-sleep-lv     this  room  obl  Tahai  today

      ‘This room will be slept in by Tahai today.’

   4. d. Benefactive phrase: is-

      is-baliv  a  Tahai  mas  ahil-tan.

      bv-buy  nom Tahai  obl  book-this.obl

      ‘This book is bought for Tahai.’

Furthermore, as in many Austronesian languages (Chang Reference Chang1997, Richards Reference Richards, Paul, Phillips and Travis2000, Rackowski Reference Rackowski2002, Aldridge Reference Aldridge2004, Pearson Reference Pearson2005), Ā-extraction in Isbukun, including wh-construction formation and relativization, is restricted to the nominative-marked argument, obeying the so-called subject-only restriction (Keenan and Comrie Reference Keenan and Comrie1977). This extraction restriction is illustrated in the interrogative paradigm of (7).Footnote ⁸

1. (7) The subject-only restriction

   1. a. Ma-i-baliv a Dahu mas tulkuk.

      av<pfv>buy nom Dahu obl chicken

      ‘Dahu bought the chicken.’

   2. b. Sima ma-i-baliv tulkuk?

      who av< pfv>buy chicken

      ‘Who bought the chicken?’

   3. c. *Maaz ma-i-baliv a Dahu?

      what av< pfv>buy nom Dahu

      Intended: ‘What did Dahu buy?’

   4. d. Maaz baliv-un mas Dahu?

      what buy.pv obl Dahu

      ‘What did Dahu buy?’ = ‘What is bought by Dahu?’

Example (7c) is ungrammatical since it involves argumental extraction of the patient/theme object, namely a non-subject; however, note that, as shown in (7d), once this object is promoted to be the surface subject, as reflected by the verbal morphology, Ā-extraction in question becomes legitimate again. In other words, only the subject can undergo Ā-operation but not the object. Similar restrictions are attested in relativization as well. Like numerous languages, Isbukun Bunun has embedded headed relatives, in which the nominal is construed as the head of relativization. As shown in (8), headed relatives in Isbukun Bunun are always prenominal and such a relative clause is linked to its modified noun by the associative linker tu.Footnote ⁹ Crucially, similar to the restrictions observed earlier, only the subject can be relativized: (8b) is ruled out since the head noun ahil-an ‘this book’ is an object that has been relativized and extracted.

1. (8)

   1. a. Sa-i-du-an-ku a b-in-aliv mas cina tu ahil-an.

      see<pfv>lv-1sg.obl nom buy<pfv> obl mom lnk book-this.nom

      ‘I have seen the book Mother bought.’

   2. b. *Sa-i-du-an-ku a ma-i-baliv cina tu ahil-an.

      see< pfv>lv-1sg.obl nom av<pfv>buy mom lnk book-this. nom

I turn now to the focus of this study, the correlative construction in Isbukun.

Generally speaking, a correlative construction is composed of a correlative clause and a matrix clause. The correlative clause, which is an adjunct CP, contains a relative or wh-operator, and the matrix clause contains an anaphoric element associated with the correlative clause. For ease of exposition, in the following discussion we refer to the adjunct CP in a correlative construction as the correlative CP and the anaphoric noun phrase as the anaphoric nominal. Sentences exemplifying the correlative construction in Isbukun are provided in (9).Footnote ¹⁰

1. (9) Isbukun Bunun correlatives

   1. a. Sima  ma-tushung  uvaaz  (hai),  na  ludah-un-ku  saia  tu  bunun.

      who  av-bully  child  top  fut  hit-pv-1sg.obl  3sg.nom  lnk  person

      ‘I will hit the person who bullies a child’ or ‘I will hit whoever bullies a child.’

      Lit. ‘Who bullies a child, that person will be hit by me.’

   2. b. Maaz  suu  kazima-un  (hai),  na  kahaitas-un-ku  saia  tu

      what  2sg.obl  like-pv  top  fut  hate-pv-1sg.obl  3sg.nom  lnk

      luhi.

      (small)-animal

      ‘I will hate the animal which you like.’ or

      ‘I will hate whatever animal you like.’

      Lit. ‘What is liked by you, that animal will be hated by me.’

   3. c. Na  ku-isa  kasu  (hai),  na  ku-saincia  amin  saikin.

      fut  to-where  2sg.nom  top  fut  to-there  also  1sg.nom

      ‘I will go to the place where you go,’ or ‘I will go wherever you go.’

      Lit. ‘Where you go, that place I also go.’

   4. d. Na  lakua  kasu  mudan  (hai),  na  tudiip  mudan  amin  saikin.

      fut  when  2sg.nom  av.go  top  fut  that.time  av.go  also  1sg.nom

      ‘I will go at the time when you go,’ or ‘I will go whenever you go.’

      Lit. ‘When you go, that time I also go.’

As can be observed in (9), several distinct properties, which have been argued to hold of correlatives cross-linguistically, can be found in the preceding Isbukun sentences. For one, an adjoined CP, occurring to the left of the matrix clause, always contains a wh-operator that relates to some corresponding nominal element in the matrix clause. This placement conforms to the general property that correlative CPs predominantly occur in the left periphery, in a position that is not necessarily adjacent to the anaphoric nominal expression (Dayal Reference Dayal1996). The Isbukun structure resembles the Hindi correlative example in (1) and is similar to the schematic representation in (2), repeated here as (10a) and (10b).

1. (10)

   1. a. [_CP  jo CD sale-par hai], Aamir us CD-ko khari:d-egaː.

      rel CD sale-on be Aamir that CD-acc buy-fut.m.sg

      ‘Aamir will buy the CD that is on sale.’

   2. b. [_CorCP…wh-XP_i …]_i [_IP…correlate-XP_i …]

Interestingly, while Isbukun is identical to the Hindi case in that its correlatives obligatorily have an anaphoric nominal in the matrix clause, it differs from Hindi in that the Isbukun correlative allows a wh-indefinite or a pronoun to serve as the anaphoric nominal correlate, instead of the demonstrative phrase.Footnote ¹¹ Examples are given in (11).

1. (11) More instances of Isbukun Bunun correlatives

   1. a. Sima  ma-tushung  uvaaz  (hai),  na  {sima/saia}  ludah-un-ku.

      who  av-bully  child  top  fut  who/3sg.nom  hit-pv-1sg.obl

      ‘I will hit the person who bullies a child,’ or ‘I will hit whoever bullies a child.’

      Lit. ‘Who bullies a child, who/he will be hit by me.’

   2. b. Maaz  suu  kazima-un  (hai),  na  {maaz/sain}  kahaitas-un-ku.

      what  2sg.obl  like-pv  top  fut  what/3sg.nom  hate-pv-1sg.obl

      ‘I will hate the thing which you like,’ or ‘I will hate whatever you like.’

      Lit. ‘What is liked by you, what/it will be hated by me.’

   3. c. Na  ku-isa  kasu  (hai),  na  ku-isa  amin  saikin.

      fut  to-where  2sg.nom  top  fut  to-where  also  1sg.nom

      ‘I will go to the place where you go,’ or ‘I will go wherever you go.’

      Lit. ‘Where you go, where I also go.’

   4. d. Na  lakua  kasu  mudan  (hai),  na  lakua  amin  saikin  mudan.

      fut  when  2sg.nom  av.go  top  fut  when  also  1sg.nom  av.go

      ‘I will go at the time when you go,’ or ‘I will go whenever you go.’

      Lit. ‘When you go, when I also go.’

Thus it seems that although Isbukun and Hindi both require an anaphoric nominal expression in correlativization, the two differ in the types of elements that can serve such a function. In the following discussion, I refer to this phenomenon as the correlate requirement; we will return to this issue again in section 3 and discuss what groups the three types of NPs together in Isbukun correlatives.

The Isbukun correlative can denote a kind of universal reading, but it also admits of a kind of definite reading. This fact is reflected by the English translation in (9) and (11). That is, a sentence such as (11a) has a meaning analogous to ‘I will hit anyone who bullies a child’, which expresses the speaker's general wish to, say, protect children. On the other hand, given a context in which some child looks injured and there is a known, specific person who caused the injury, (11a) can also be uttered for the speaker to communicate that (s)he will hit that particular person as revenge. The availability of both a universal and a definite reading is also widely reported in the correlative literature, as mentioned earlier and observed in the Hindi example. Given these similarities, it seems to be well grounded to treat these Isbukun sentences on a par with the correlatives investigated in other languages.Footnote ¹²

To recapitulate, I have shown that the treatment of sentences in (9) and (11) as correlativization is well grounded as the constructions under investigation conform to the properties generally observed in correlatives. After securing the correlative analysis to be on the right track, its relevant properties and structure will be examined in the following sections.

3. The syntactic representation of the Isbukun correlative

In the literature we find two basic types of approaches centering around the question of how the correlative clause syntactically combines with the main clause (Srivastav Reference Srivastav1991, Dayal Reference Dayal1996, Izvorski Reference Izvorski and Kusumoto1996 and Bhatt Reference Bhatt2003). The two plausible structures, differing along the line of positing base-generation vs. movement of the relative clause, are referred to as the base-generated IP-adjunction approach and IP adjunction via movement approach respectively in the following discussion.

The base-generated IP-adjunction approach, first defended extensively in Srivastav (Reference Srivastav1991) and Dayal (Reference Dayal1996), assumes that the correlative CP is initially merged as an adjunct to the IP projection of the main clause, illustrated in (12). Crucially, the anaphoric nominal expression and the correlative CP do not form a constituent at any point in the derivation. Within this structure, the correlative CP acts as a generalized quantifier and the anaphoric nominal as a variable, which is akin to an Ā-trace.

1. (12)

On the other hand, under the IP adjunction via movement approach (Bhatt Reference Bhatt2003), the correlative CP is first merged as an adjunct to the nominal correlate it is associated with; subsequently, the correlative CP moves out of the projection containing the nominal correlate and adjoins to the matrix IP. The correlative CP thus modifies the correlate phrase locally by forming a complex structure with it in the base. From this low position, the correlative CP (optionally) moves out to adjoin to IP via an Ā-operation, illustrated in (13).Footnote ¹³

1. (13)

In what follows, arguing from non-constituency facts, island effects and condition C effects, I contend that the non-local base-generation approach is the proper analysis of the Isbukun correlatives.

The first piece of evidence that the correlative clause and its associated anaphoric element never form a constituent at any level comes from the fact that, unlike in Hindi, it is never possible to have the [correlative CP–correlate] sequence at any stage in Isbukun, as shown in (14) (cf. (9)).

1. (14) *Na  ludah-un-ku  [sima  ma-tushung  uvaaz  saia  tu  bunun].

   fut  hit-pv-1sg.obl  who  av-bully  child  3sg.nom  lnk  person

   Intended: ‘I will hit the person who bullies a child.’

On the other hand, the [correlative CP–correlate NP] sequence is generally sanctioned in languages that are independently shown to allow IP Adjunction via movement. As Bhatt shows, it is also possible to coordinate two [correlative CP–correlate] sequences. Assuming that coordination can only apply to constituents, coordinate structures such as (15) suggest that the Hindi correlative CP and its associated correlate phrase can combine to form a single constituent. This is expected under the adjunction-via-movement approach, under which the correlative CP is first locally merged as an adjunct to the nominal correlate.

1. (15) Rahul  a:jkal [[[  jo  kitaːb  Saira-ne  likh-i:]_k  vo_k]  aur  [[jo  cartoon

   Rahul  nowadays  rel  book  Saira-erg  write-pfv  that  and   rel  cartoon

   Shyam-ne  bana:-ya:]_j  vo_j ]]  parh  rahaː  hai.

   Shyam-erg  make-pfv  that  read  prog  be

   ‘Nowadays, Rahul is reading the book that Saira wrote and the cartoon that Shyam made.’

   Lit. ‘Nowadays, Rahul is reading what book Saira wrote, that and what cartoon Shyam made, that.’    (Bhatt Reference Bhatt2003: 504)

In sharp contrast, since the [correlative CP–correlate] sequence is not permissible in Isbukun Bunun, coordinating such sequences is expected to be problematic:

1. (16) *Na  ludah-un-ku  [sima  ma-tushung  uvaaz  saia  tu  bunun]

   fut  hit-pv-1sg.obl  who  av-bully  child  3sg.nom  lnk  person

   mas  [sima  tanhaiu  tulkuk  saia  tu  bunun].

   and  who  av.steal  chicken  3sg.nom  lnk  person

   Intended: ‘I will hit the person who bullies a child and the person who steals chickens.’

As a result, the fact that Isbukun correlativization only allows the correlative clause to appear non-locally to the left of the nominal correlate indicates that only the base-generated adjunction structure is available in Isbukun grammar.

Moreover, since the derivation in (12) does not involve overt syntactic movement, we expect it not to be sensitive to islands; in contrast, if the structure in (13) is available for Isbukun correlatives, island-sensitivity between the correlative CP and the correlate phrase would be an automatic consequence.Footnote ¹⁴ The relevant data regarding island effects support the base-generation structure in (12):

1. (17) Sima  su  kazima-un  hai,  kaun-un-ku  saicin  s-in-pitia  tu

   who  2sg.obl  like-pv  top  eat-pv-1sg.obl  3sg.obl  bv<pfv>cook  lnk

   haising.

   Rice

   Lit. ‘Who is liked by you, I eat the rice that (s)he cooked.’

2. (18) Sima  minsuma,  kazima-un-ku  saicin  s-in-baliv  tu  patasan.

   who  av.come  like-pv-1sg.obl  3sg.obl  bv<pfv>buy  lnk  book

   Lit. ‘Who comes, I like the book that (s)he bought.’

As can clearly be seen here, the relation between the correlative CP and the anaphoric nominal phrase is island-insensitive, exemplified by the inapplicability of the Complex Noun Phrase Constraint (Ross Reference Ross1967). In other words, when the correlative CP is adjoined to the matrix clause, the nominal correlate can be contained within a relative clause. Such immunity to island effects is expected under the base-generated IP adjunction view but not under the movement approach.

The last argument against the IP adjunction via movement account comes from Binding Condition C effects. If the syntactic relation between the correlative CP and the anaphoric nominal is derived by initial merger of the correlative CP to the nominal correlate and subsequent movement of the correlative CP, we should expect that a pronoun cannot c-command the anaphoric nominal and co-refer with a R-expression contained within the correlative CP. If it did, the R-expression would be bound, inducing a Condition C violation (Lebeaux Reference Lebeaux1998). However, this expectation is not met, as illustrated in (19).

1. (19) [Sima  mazima  Abus_k]_j,  na  kahaitas  sain_k  saicia_j.

   who  av.like  Abus  fut  av.hate  3sg.nom  3sg.acc

   ‘Abus hates the person that likes her.’

   Lit. ‘[Who likes Abus_k]_j, she_k will reject him/her_j.’

(19) shows that a pronoun like sain in Isbukun may c-command the anaphoric nominal saicia and co-refer with a name (i.e., Abus) contained within the correlative CP. Such a pattern could only arise if speakers have access to the base-generated adjunction structure. Specifically, by initially merging the correlative CP to the matrix IP, sentence (19) could be derived without any Condition C effect resulting. On the other hand, if the Isbukun grammar had recourse only to the ‘IP adjunction via movement’ structure, then sentence (19) would be ruled out as a Condition C violation, just as its Hindi counterpart presumably is. As Bhatt observes in cases like (20), if a pronoun c-commands the anaphoric nominal in the Hindi correlative, then that pronoun cannot co-refer with a name contained within the correlative CP. The ungrammaticality of (20) in Hindi would thus follow from the non-availability of the base-generated IP adjunction structure in this language as well as standard assumptions regarding the evaluation of Condition C.

1. (20) *[jo  laRkii  Sita-ko_k  pyaar  kar-tii  hai]_j  us-ne_k  us-ko_j  ţhukraa di-yaa.

   rel  girl  Sita-acc  love  do-hab  be  that-erg  that-acc  reject give-pfv

   Intended: ‘She_k rejected the girl who loves Sita_k.’

   Lit. ‘[Which girl loves Sita_k]_j, she_k rejected her_j.’ (Bhatt Reference Bhatt2003: 513)

Therefore, the base-generated IP adjunction account correctly predicts that the Isbukun correlative sentence in (16) can be generated without inducing any Condition C problem. In this section I have presented arguments from non-constituency, island-insensitivity, and absence of Condition C effects that lend support to the base-generated status of the correlative clause. This range of facts would be mysterious if the correlative clause occupied its surface position as a result of movement; instead, these phenomena receive a natural explanation if we accept that the syntactic relation between the correlative CP and the nominal correlate is built on the base-generated configuration. Granted this, next we further examine how the non-local link between the correlative clause and its nominal correlate is established.

4. Semantic representation and bound variable interpretation

In Isbukun Bunun the correlative clause must precede the main clause. As demonstrated earlier, repeated here as (21), the link between the two can be mediated by a demonstrative NP, a pronoun or a wh-indefinite, which occurs in the main clause. Following Srivastav (Reference Srivastav1991) and Dayal (Reference Dayal, Bach, Jelinek, Kratzer and Partee1995), I suggest that the correlative clause is adjoined to the matrix IP and acts as a quantificational phrase binding a position inside its sister IP, as schematized in (22).

1. (21) Sima  ma-tushung  uvaaz,  na  {saia  tu  bunun/saia/sima}

   who  av-bully  child  fut  3sg.nom  lnk  person/3sg.nom/who

   ludah-un-ku.

   hit-pv-1sg.obl

   ‘I will hit the person who bullies a child.’

2. (22)

Specifically, I propose that the wh-element in the correlative CP is an operator that combines with the predicate inside the correlative CP to yield a quantifier-level meaning for the whole CP. That is, the correlative CP is a quantifier whose meaning is compositionally built upon the meaning of the wh-operator inside it.Footnote ¹⁵ On the other hand, the anaphoric nominal in the main clause (i.e., the demonstrative NP, pronoun or wh-indefinite) is a bound variable quantified over by the correlative CP, and the full structure is interpreted by standard rules of quantification.Footnote ¹⁶ Under this view, a correlative CP like (21), for example, denotes the set of properties of some individual(s) who bullies children and the main clause denotes the property of being hit by me. Therefore, sentence (21) is true if and only if the property of being hit by me is one of the properties of the people who bully children.

There are several reasons for adopting this analysis. To begin with, it captures an important empirical generalization: there must always be a wh-phrase in the Isbukun correlative CP, while other non-wh words are cannot appear in the same position.

1. (23) *Aiza  bunun  ma-tushung  uvaaz  (hai),  na  ludah-un-ku  saia  tu

   exist  person  av-bully  child  top  fut  hit-pv-1sg.obl  3sg.nom  lnk

   bunun.

   Person

   Intended: ‘If a person bullies a child, I hit him.’

This fact follows from the current analysis: A lexical indefinite cannot function as an operator and subsequently combine with the open sentence denoted by the IP inside the correlative clause. Therefore, a correlative CP without a wh-phrase serving as an operator is thus ill-formed in Isbukun. In other words, non-wh words (e.g., lexical indefinites) cannot fulfill the function because they fail to serve as operators. They cannot enable the correlative CP to generate the quantificational interpretation. This generalization poses a challenge to the donkey anaphora treatment (Cheng and Huang Reference Cheng and Huang1996); under this view, there is no principled reason to prohibit unselective binding of a lexical indefinite since an indefinite NP can usually be unselectively bound by a proper operator (Lewis Reference Lewis and Keenan1975, Saito Reference Saito and Yoon2004). Therefore, treating the Isbukun correlative as involving a quantificational structure provides a straightforward explanation for the obligatory presence of a wh-phrase in the correlative CP.

Along the same lines, the current analysis also captures a significant difference between a donkey pronoun and the anaphoric nominal in a correlative. A donkey pronoun is known to be optional, while relative clause binding is not (cf. Srivastav Reference Srivastav1991). As shown in (24b), the donkey anaphor can be omitted, and the main-clause VP need not contain any reference to the antecedent. However, this is not possible in (25); there, the main clause must contain an argument referring back to the wh-word in the correlative CP. In other words, in a correlative construction the number of wh-operators must match the number of anaphoric nominals in the matrix clause. Under the present view, the correlate requirement can be better understood in the context of (22): the mandatory presence of an anaphoric nominal follows from the fact that the correlative clause, being a quantifier, must bind a variable; otherwise, the ban on vacuous quantification is violated (Kratzer Reference Kratzer, Carlson and Pelletier1995). As such, some anaphoric nominal must occur so as to provide the variable to be bound by the correlative clause, hence the correlate requirement.

1. (24)

   1. a. Every farmer who owns a donkey beats it.

   2. b. Every farmer who has a donkey is happy.

2. (25) *Sima  ma-tushung  uvaaz  (hai)  makuang. (cf. (21))

   who  av-bully  child  (top)  av.bad

   Intended: ‘(People) who bully children are bad.’

More importantly, the proposed binding structure is able to capture the types of anaphoric nominals allowed in the matrix clause. Suppose that the Isbukun Bunun correlative involves a quantificational structure in which the correlative CP binds a variable in the main clause. Other things being equal, a proper characterization of the phenomenon would then amount to the identification of appropriate variables in a given language. I propose that this is where natural languages can vary, as languages are known to differ parametrically in the availability of bound variable interpretations for different types of lexical items.Footnote ¹⁷ For example, languages such as Chinese (Y.-H. A. Li Reference Li1992, Tsai Reference Tsai1994, Reinhart Reference Reinhart1998), Japanese (Kuroda Reference Kuroda1965, Baker Reference Baker1970, Nishigauchi Reference Nishigauchi1990), Korean (Suh Reference Suh, Kuno, Lee, Whitman, Bak, Kang and Kim1989), Malay (Cole and Hermon Reference Cole and Hermon1998, Reference Cole, Hermon, Lutz, Müller and von Stechow2000), Malayalam (Jayaseelan Reference Jayaseelan2001), Indonesian and numerous others (see Cheng Reference Cheng1991 and Haspelmath Reference Haspelmath1997) allow wh-indefinites as bindable variables, though this possibility is not necessarily available in other languages. In the same fashion, there are languages that allow pronouns to be free variables that pick up quantificational force of the operator which scopes over them, while this option is not available for other languages. Returning to our earlier puzzle: what makes demonstrative phrases, pronouns and wh-indefinites form a class in the Isbukun correlative? The current proposal thus provides a principled account: the grammar of Isbukun Bunun allows demonstrative phrases, pronouns and wh-indefinites to serve as anaphoric nominals in correlatives since they can independently be construed as bound variables in this language, and thus can satisfy the quantification requirement imposed by the structure of (22). The evidence for the bound-variable status of each of these elements is provided in (26), where it is clear that demonstrative phrases, pronouns and wh-phrases can serve as bound variables. In each case the referential value varies with the value-assignment of the quantificational antecedent. To put it differently, the indicated indexing possibilities are possible only when the meaning of a demonstrative phrase, a pronoun or a wh-indefinite can include a free variable bound by a quantifier.

1. (26) Bound variable interpretation

   1. a. Dem-phrase

      [tacinicini  tu  bunun]_i  hai  sasadu  amin  kazima-un  saicin  tu

      everyone  lnk  person  top  av.stare  also  like-pv  3sg.obl  lnk

      binanauaz-a_i.

      girl-dem

      ‘Every person_i is staring at that_i girl he_i likes.’

   2. b. Pronoun

      [tacinicini  tu  bananaz]_i  hai  mazima  amin  isaicia_i  pingaz.

      everyone  lnk  husband  top  av.like  also  3sg.gen  wife

      ‘Every husband_i loves his_i wife.’

   3. c. Wh-phrase

      tacinicini_i  hai  na  ma-baliv  tu  cikis  maaz_i  ma-pudaan  sia  suan.

      everyone  top  fut  av-buy  lnk  some  what  av-send  at  2sg.loc

      ‘Everyone_i will buy something_i to send to you.’

The fact that the Isbukun correlative permits the three types of NPs in question to be the anaphoric nominal can therefore be attributed to the fact that these NPs qualify as bindable variables, and can be bound in a proper binding configuration in Isbukun. In other words, the current proposal straightforwardly explains why the three NP types form a natural class as eligible anaphoric nominals in the Isbukun correlative.Footnote ¹⁸