Wh-in-Situ in Bahasa Indonesia and Choice Function *

(1) Wh-Questions in BI a. [CP1 Apai yang kamu pikir [ CP2 Esti kira [CP3 Pak Yanto beli ti kemarin]]]? what that you think Esti expect Mr. Yanto bu y yesterday? ‘What do you think Esti expects Mr. Yanto bought yesterday?’ b. [CP1 Kamu pikir [CP2 apai yang Esti kira [ CP3 Pak Yanto beli ti kemarin]]]? you think what that Esti expect Mr. Yanto buy yesterday ‘What do you think Esti expects Mr. Yanto bought yesterday?’ c. [CP1 Kamu pikir [CP2 Esti kira [CP3 Pak Yanto beli apa kemarin]]]? you think Esti think Mr. Yanto buy what yesterday ‘What do you think Esti expects Mr. Yanto bought yesterday?’


Introduction
This paper addresses the question of what is the licensing mechanism of wh-in-situ in Bahasa Indonesia/BI.We argue that the relevant mechanism is choice function à la Reinhart (1997Reinhart ( , 1998)).

Wh-in-situ in Bahasa Indonesia
BI has three ways of forming wh-questions: i) overt syntactic movement to the scopal [Spec, CP], ii) partial syntactic movement to the non-scopal [Spec, CP], and iii) wh-in-situ.These three strategies for wh-questions are illustrated in (1a-c).
( In (1a), the wh-phrase apa 'what' undergoes overt syntactic movement to the scopal, matrix [Spec, CP].This option is always available for nominal wh-phrases such as siapa 'who' and apa 'what' but obligatory for non-nominal wh-phrases such as kenapa 'why ' and bagaimana 'how'. (1b) illustrates the partial syntactic movement option in BI, where the same wh-phrase undergoes movement into the intermediate, non-scopal [Spec, CP] though the example itself has a matrix wh-interpretation as in the fully moved example in (1a).This option is available for nominal wh-phrases but not for non-nominal wh-phrases.Finally, (1c) illustrates the in-situ option.This is possible for nominal wh-phrases but impossible for non-nominal wh-phrases.This section provides an overview of the structural and interpretive properties of wh-in-situ in BI.The discussion in this section draws heavily on the description and analysis of this construction presented by Saddy (1991).Saddy observes that the wh-in-situ construction in BI exhibits a spectacular range of syntactic and semantic characteristics that would not be accounted for under standard analyses of the corresponding constructions in other languages such as English, Chinese, and Japanese.We review his main arguments in the rest of this section to show that the two most widely assumed analyses of wh-construal, overt/covert syntactic movement and unselective binding, are not applicable for wh-in-situ in BI.Thus, many of the examples in this section are drawn from his work, unless otherwise indicated.However, we also note that there is a non-trivial divergence between the judgments reported by Saddy (1991) and those reported by the second author of this paper.

Overt syntactic movement?
The first analysis of wh-in-situ in BI that is easily dismissed is the overt null operator/Q-feature movement analysis as proposed by Watanabe (1992) for wh-in-situ in Japanese.Overt movement in BI shows island effects, as in (2a-c).This analysis predicts that the in-situ counterparts should be as ungrammatical as (2a-c) but (3a-c) are all fully grammatical. 12) a.* Apa i yang kamu katakan [dimana kita beli t i ]?

Covert syntactic movement?
Saddy presents several arguments that the covert movement analysis of wh-in-situ as in Huang (1982) is also incorrect for wh-in-situ in BI.First, the covert/LF movement in this language obeys island constraints as the overt/syntactic movement.Specifically, wh-phrases that remain within syntactic islands in overt syntax still give rise to ungrammaticality, as shown in (4a-c).
(4) a.  (Saddy 1991: 195, 196) In (4a), the wh-phrase siapa 'who' undergoes partial wh-movement into the intermediate, non-scopal specifier of CP.Since this short extraction does not cross any syntactic island, it cannot be the source of the ungrammaticality.The ungrammaticality follows if we assume that the LF/covert movement of the partially moved wh-phrase into the matrix specifier of CP obeys island constraints in BI.According to this analysis, the wh-phrase siapa 'who' undergoes covert movement into the scopal specifier of CP for the purposes of scope taking.This movement thus renders (4a) ungrammatical due to its crossing the syntactic island.The same story holds for (4b) and (4c).Thus, (4a-c) show that covert movement obeys island constraints in BI.Now, if the covert movement analysis of wh-in-situ in languages like Chinese, Japanese, and English (under multiple interrogative questions) is correct for BI, the in-situ counterparts of (4a-c) should be ungrammatical because the LF representation of the insitu variants would be identical to that of (4a-c).This prediction is falsified by (5a-c).
(5) a.  (Saddy 1991: 195, 196) The second argument against the LF movement approach to wh-in-situ in BI is based on the fact that this language does not allow complements that contain a wh-in-situ for verbs such as ingin tahu 'want to know, wonder' that are obligatorily subcategorized for the [+WH] complement, as in English.This is illustrated by the contrast between (6a) and (6b Prof his TR-think I TR-love who (Saddy 1991: 207, 208) The standard assumption on the crossover effect is that it arises when a pronoun fails to be ccommanded both by a binder and by its variable at the surface/derived structure to be construed as a variable.Under the movement analysis, this effect can be formalized as the filter of the form  , b) are deemed ungrammatical.We maintain here, following Cole and Hermon (1998: 234), that the presence of the weak crossover effect does not mean that the in-situ wh-phrase in (7a, b) undergoes syntactic movement because the crossover effect can be formulated in non-movement terms as a constraint on the representation.Specifically, Cole and Hermon (1998) argue that the crossover effect can be analyzed as the byproduct of the Bijection Principle of Koopman and Sportiche (1983) that prohibits a single operator from binding more than one variable.This principle allows us to correctly block (7a, b) without also assuming the syntactic movement because the basegenerated wh-operator in [Spec, CP] binds both the pronoun and the variable.For this reason, we conclude, contrary to Saddy, that the presence of the crossover effect itself does not show that wh-insitu undergoes syntactic or LF movement.
The last argument made by Saddy against the LF movement analysis of wh-in-situ in BI is based on his observation that this construction does not support a pair-list reading.Consider (11).( 11 Saddy reports that this multiple wh-question can only be interpreted as a request for a single pair as in John bought a book; thus, the answer as in John bought a book, Mary bought a magazine, Bob bought a shirt is not a possible reply to this question.Since Higginbotham and May (1981) on English multiple interrogatives, the availability of the pair-list reading for English sentences like who bought what has been taken to be driven by the association of the two wh-phrases in the same Comp at LF (or the multiple specifiers of the same C in the more modern terminology) known as absorption.To the extent that this analysis is correct, the lack of the pair-list reading in (11) shows that apa 'what' does not undergo any movement either in overt syntax or LF.Accordingly, this example provides evidence against the movement approach to the wh-construal in BI.Again, however, the second author has reported that the pair-list reading is available in sentences like ( 11) above.This is also the judgment elicited from speakers of Malay by Cole and Hermon (1998: 225), who report that their Malay informants had no problem with a list interpretation for sentences as in ( 12).
(12) Siapa kamu fakir beli apa? 'Who did you think bought what?' who you think buy what (Cole and Hermon 1998: 225) This judgment, therefore, indicates that the argument against the LF movement based on the pair-list reading is not strong as Saddy wanted it to be.We come back to this point in section 4.
To sum up this section, we have reviewed a total of four arguments presented in Saddy that the covert movement analysis is not an adequate mechanism of licensing wh-in-situ in BI.Though the two arguments based on the crossover effect and the unavailability of the pair-list reading in multiple questions do not necessarily argue for or against the LF movement analysis, the other two other arguments from the lack of island effects and the [+WH] subcategorization requirements provide relatively clear evidence that this analysis is not applicable to BI wh-in-situ.

Unselective binding?
The third potential analysis of wh-in-situ, which is perhaps the most widely held analysis for wh-insitu in languages such as Japanese and Chinese, is that of unselective binding (Pesetsky 1987); we defer the variant of this approach presented recently by Cole andHermon (1998, 2000) until section 4.4).Pesetsky (1987) proposes that wh-interpretation is achieved not only by syntactic movement but also a non-movement mechanism called unselective binding.Pesetsky claims that the choice between these two options is determined by the notion of D(iscourse)-Linking, which roughly corresponds to the morphological distinction of English wh-words between "which-X" (which man, which book, etc) and everything else (who, what, etc).As Pesetsky (1987: 107, 108) remarks, which-phrases are discoursed-linked (D-linked), because "when a speaker asks a question like which book did you read?, the range of felicitous answers is limited by a set of books both speaker and hearer have in mind" whereas "no such requirement is imposed on wh-phrases like who, what, or how many books."Based on this discourse-related observation, Pesetsky argues that if a wh-phrase is D-linked, it contains a variable that is unselectively bound by a Q-morpheme located in the scopal C head position and thereby is licensed without syntactic movement.On the other hand, if a wh-phrase is not D-linked, it must undergo syntactic movement, be it overt or covert, to be properly licensed by the scopal C. Pesetsky draws various types of evidence concerning the presence/absence of superiority effects in English questions as well as the behavior of what he calls aggressively non-D-linked whphrases such as what the hell in English and its equivalent in Japanese to support this hybrid approach to wh-construal.
Saddy, however, point outs a couple of potential problems with Pesetsky's version of unselective binding analysis when applied to wh-in-situ in BI.The first problem concerns the morphological composition of wh-phrases in BI.As we have seen above, Pesetsky's analysis rests upon the correlation between the morphological composition of a wh-phrase and its interpretive mechanism.This correlation, however, does not hold in BI because wh-phrases in this language all have "Dlinked" expressions corresponding to English "which-X" form.For example, orang siapa 'which person", which would be analyzed as a D-linked phrase in Pesetsky's terms, is used interchangeably with the non-D-linked form siapa 'who' but this difference in morphological composition does not change the interpretive and structural constraints observed so far in this section.Though this observation may not be a problem for Pesetsky's theory directly, it indicates that Pesetsky-style Dlinking is not directly applicable to BI wh-questions.The second potential problem with the extension of Pesetsky's analysis to BI is based on the quantificational uninformativeness of wh-insitu in BI as reported by Saddy. Pesetsky (1987) employs D-linking to account for the triplet interpretation available for examples as in ( 13) below so that the D-linked phrase which prize may get matrix scope without movement by being bound by the matrix Q, as shown in ( 14).Importantly, this analysis crucially assumes that D-linked in-situ phrases such as which prize must be able to interact in scope with other scope-bearing elements such as every athlete; for, the triplet interpretation otherwise would be unavailable in examples like (13).As Saddy (1991: 205) puts it, "it is a necessary property of Pesetsky's Q-bound D-linked WH expressions that they interact quantificationally with other elements in the matrix clause."When applied to BI wh-insitu constructions akin to (13), Pesetsky's analysis predicts that this type of construction also should allow the triplet interpretation.Saddy observes that this prediction is false because, according to his informant work, this reading is precisely the kind of interpretation that BI wh-insitu resists, as shown in examples like (15a).
(15) a. Setiap orang men-cintai siapa?'Who did every person love?' every person TR-love who (who>every, *every>who) b.Siapa i yang setiap orang Ø-cintai t i ?'Who did every person love?' who Foc every person love (who>every, every>who) (Saddy 1991: 199) The example in (15a) with the wh-phrase in situ only allows the wide scope reading of the in-situ phrase with respect to the universal quantifier setipa orang 'every person' in subject position; the reading where the value of the person loved co-varies with that of the lover is impossible.This latter reading becomes available only when the wh-phrase must undergo overt syntactic movement, as shown in (15b).Saddy notes that the same contrast holds in the triplet of examples as in (16a-c).
( what Foc every person know Tom buy (what>every, every>what) (Saddy 1991: 200) According to Saddy, the in-situ wh-phrase apa 'what' necessarily takes wide scope over the universal quantifier setiap orang 'every person' in (16a), even though the relative structural height of the latter with respect to the former leads us to expect the opposite reading.Again, the wide scope reading of the universal quantifier over the wh-phrase is only possible when the latter undergoes movement, either partially, as in (16b), or fully, as in (16c).Saddy takes the interpretive outcomes seen in these examples as evidence that wh-in-situ in BI is quantificationally uninformative with respect to other scope-bearing expressions, unlike moved wh-phrases.This result would remain mysterious under Pesetsky's Qbinding analysis of triplet questions.Saddy thus concludes that Pesetsky's analysis is inadequate for BI.
We would like to add, however, is that we could not reproduce the same judgments as elicited by Saddy from his BI consultants.According to the second author, both (15a) and (16a) allow the narrow scope reading of the in-situ wh-phrase with respect to the universal quantifier, the reading where the value of the thing bought and the person loved can vary with the value of the universal quantifier.This result, therefore, shows that the wh-in-situ in BI is scopally informative, contrary to what Saddy reports.At this moment, we have no idea how scope judgments can diverse in such a clear manner, as we do not know the linguistic backgrounds of Saddy's language consultants; it may be a reflex of the ongoing change that BI experiences through its interaction with languages like Dutch, English, and many other local languages spoken in BI.For the purposes of this paper, we assume that the second author reflects the majority judgment, keeping in mind, though, that Saddy's elicited judgments might also hold for certain dialects of BI.We come back to this in section 4.2.If so, we have lost one major argument against Pesetsky's version of the unselective binding approach to wh-in-situ in BI.However, in section 4.1, we review Reinhart's (1997Reinhart's ( , 1998) ) evidence that casts doubts on the general applicability of unselective binding as a possible in-situ strategy based on the scope behavior of wh-in-situ in multiple questions in English and BI, by extension.Anticipating this discussion, we conclude here that Pesetsky's analysis is not suitable for wh-in-situ in BI.Saddy (1991) proposes that wh-in-situ in BI behave as an interrogative definite description, drawing on an impressive range of syntactic and semantic parallelisms that hold between this class of expressions and words of the form "this-X/these-Xs" in game show questions in English.An example of English game show quizzes is given in (17).

Wh-in-Situ in Bahasa Indonesia is not an interrogative definite description
(17) Question: For $100, every armchair general watched this television station.
Answer: What is NBC? (slightly modified from Saddy 1991: 208) The question in (17) does not have interrogative force in the standard sense as a wh-question because it is syntactically a declarative statement; rather, it gains such force from the very context that this sentence is uttered in a game show question; a host utters this sentence to challengers, expecting them to make a question such that it is an appropriate answer to the definite NP this television station.In other words, the interrogative requirement here is that challengers identify the member (s) of the definite description of the form 'this-X/these-Xs.' As mentioned above, the reason Saddy brought up this type of game show in his work is because of his observation that statements as in (17) in the game show context exhibit exactly the same range of structural and interpretive properties that we have seen to characterize wh-in-situ in BI.Space limitations prevent us from reproducing all the relevant examples from Saddy (1991: 210-212).Saddy argues that the definite description analysis of wh-in-situ in BI provides a unified account of all the properties we have seen to hold for this class of wh-questions.The scopally uninteractive behavior of in-situ wh-phrases is a natural consequence of the fact that it is a definite description, namely, that they "pick out a specific individual or a set of individuals."(p.212).Wh-in-situ does not satisfy the WHcomplement requirement of verbs like ingin tahu 'wonder' because it does not move at all into the specifier of the complement CP but instead is licensed in situ by a non-quantificational mechanism by virtue of its definite nature.Similarly, the lack of pair-list reading and weak/strong crossover effects and the insensitivity to syntactic islands for the purposes of scope taking are all derived because wh-insitu in BI is interpreted in situ.Saddy's analysis is extremely ingenious in a number of important ways.For example, it provides a unified, non-stipulatory account of all the otherwise mysterious syntactic and semantic characteristics associated with wh-in-situ in BI from the single fact that this class of wh-words is an interrogative definite description.More importantly for the purpose of this paper, his analysis suggests that natural languages may well develop a non-syntactic mechanism of licensing wh-in-situ in its base position without relying on syntactic movement.However, there are problems that cast doubts on Saddy's treatment of wh-in-situ in BI.The first argument is that there is evidence internal to BI that wh-in-situ in this language contains a variable.Cole and Hermon (1998) observe that nominal wh-words in Malay can be used as a variable bound by non-wh-operators, as shown in (18a, b) and (19a, b).This observation also holds for BI. ( 18 (Cole and Hermon 1998: 239) In ( 18a) and (19a), the wh-word is bound by the existential quantifier that is overly represented by the reduplication of the question word itself.Similarly, in (18b) and (19b), the wh-word is bound by the existential quantifier realized in the form of -pun 'also'.This use of the in-situ wh-words, therefore, shows that this class of words contain a variable.This result is problematic for Saddy's analysis because definite descriptions as a rigid designator do not contain a variable under the most commonly held assumption.The second argument against Saddy's analysis is that it misses the important generalization that wh-in-situ in BI behaves more like existential indefinites rather than definite descriptions.It is widely acknowledged that certain weak/existential indefinites such as singular NPs (e.g., someone, something) and cardinal plurals (e.g., two men, many women) are insensitive to syntactic island for scope-taking, as the contrast between (20a-c) and (21a-c) shows.
(20) a. Someone reported that Max and all the ladies disappeared.some>all, * all>some b.Someone will be offended if we don't invite most philosophers.some>most, *most>all c.Many students believe anything that every teacher says.many>every,* every>many (Reinhart 1997: 338) (21) a. Everyone reported that Max and some lady disappeared.every>some, some>every b.Most guests will be offended if we don't invite some philosopher.most>some, some>most c.All students believe anything that many teachers says.
all>many, many>all (Reinhart 1997: 339) (20a-c) show that strong quantifiers such as all, most, and every cannot violate one or the other island constraints to take wide scope over another scope-bearing element in the matrix clause.This is not surprising if we assume that Quantifier Raising, an instance of LF movement, is constrained by the island constraints, as is overt syntactic movement.What is surprising, then, is the fact, illustrated in (21a-c), that weak existential indefinites such as some and many take wide scope over the quantifier in the matrix subject position in apparent violation of the island constraints that we have just seen to constraint the Quantifier Raising operation.This wide scope reading of certain existential indefinites has been a source of endless controversies in formal semantics (see Kratzer 1998, Reinhart 1998 and references cited therein).Whatever the ultimate analysis might turn out to be, this island-insensitive behavior is similar to that of wh-in-situ in BI.We have seen in section 2 that this class of phrases can freely take widest scope in a massive violation of the standard set of island constraints on movement.Given this parallelism, the null hypothesis is that wh-in-situ in BI should be treated also as existential indefinites.This indeed has been a standard assumption on wh-phrases in the literature (Karttunen 1977).Based on the above considerations, we reject Saddy's approach to wh-in-situ in BI and seek an alternative account that captures the insight behind his work that wh-in-situ in BI is interpreted in situ.3

The choice function analysis of wh-in-situ in BI
Reinhart (1997,1998) claims that there is an interpretive mechanism available for indefinite expressions including wh-phrases that allows existential quantification over choice functions.
Reinhart argues that introducing this way of licensing allows for a unified explanation for the set of problems with traditional analyses in terms of LF movement and unselective binding/absorption.

Choice function
Reinhart (1997,1998) starts by showing that neither the LF movement (Huang 1982) and unselective binding (Pesetsky 1987) /absorption (Higginbotham and May 1981) analyses of wh-insitu in multiple questions in English are tenable on the ground that they cannot derive several properties associated with this type of expressions.Consider like (22a-d).
( 22 (Reinhart 1998: 31, 44) It has been standardly assumed since the seminal work by Huang (1982) that overt syntactic movement obeys both subjacency and the ECP whereas covert LF movement is only constrained by the ECP.Reinhart notes, however, that this line of analysis cannot account for the contrast between ( 22b) and ( 22d); it would incorrectly predict the latter to be ungrammatical because what way is an adjunct just as how.Another problem with this analysis is that this way of assigning matrix scope is untenable in the first place within the Minimalist Program (Chomsky 1995), under which movement is subject to the "shortest steps" requirement.Consider (23a), which is assigned the LF representation in (23b) under the matrix scope reading of the in-situ wh-phrase.
(23) a.Who knows where to find what.b. for which <x, y>, x knows where to find y (Reinhart 1998: 33) Reinhart notes that the very movement here is impossible within the minimalist framework because it is less economical in terms of the shortest movement requirement than its potential movement into the specifier of the embedded CP.Thus, this example shows that the scope assignment of wh-in-situ via LF movement is untenable and that a non-movement licensing is in need.
Reinhart further shows that the non-movement approach to wh-in-situ in terms of unselective binding/absorption also fails in light of the interpretation of examples as in (24).
(24) Who will be offended if we invite which philosopher?(Reinhart 1998: 36) Reinhart assumes the semantics of questions proposed by Karttunen (1977): the denotation of a question is the set of propositions which constitute true answers to it.The unselective binding/absoportion mechanism would assign the interpretation shown in (25a), which is formally represented as in (25b) under Karttunen's model, for the example in ( 24).
(25) a. for which <x, y>, if we invite y and y is a philosopher, then x will be offended.
b. {Pé(∃<x, y>)} & P = ^ ((we invite y and y is a philosopher) → (x will be offended) & true (P)) c.Lucie will be offended if we invite Donald Duck.(Reinhart 1998: 36) In (25a), the restriction is contained in the implication of an if-clause.Given the truth-theoretic conditions on such a clause, (24) would come out true in cases where the value of y is a member of the non-philosopher set; for example, the sentence would be true if Donald Duck is inserted as in (25c), since he is not a philosopher.This is, however, what (24) does not mean.What we need, thus, is to pull out the restriction from the implication as in (26a) or its Karttunen-style equivalent in (26b).
(26) a. for which <x, y>, y is a philosopher, and if we invite y, x will be offended.b. {Pé(∃<x, y>)(y is a philosopher) & P = ^ ((we write y) → (x will be offended)) & true (P))} (Reinhart 1998: 36) Thus, examples as in ( 24) above show that the absorption/unselective binding is not adequate for assigning scope for wh-in-situ phrases.At the same time, they indicate that the ultimate mechanism for this purpose is such that it allows us to ensure that the value of the wh-in-situ will be necessarily chosen from the set of members that satisfy its accompanying restriction.
Reinhart argues that the problems with LF movement or unselective binding/absorption noted above are naturally solved once we allow existential quantification over choice function in the form of existential closure.Choice function is defined as in ( 27). ( 27) A function f is a choice function (CH (f)) if it applies to any non-empty set and yields a member of that set.(Reinhart 1997: 372) For example, the LF representation of (28a) under the wide scope reading of the indefinite is shown in (28b) under the choice function approach.
(28) a.Every lady read some book.
In the LF representation, the indefinite book is replaced by a function variable to be bound by an existential operator that is base-generated in the highest level.The choice function here thus applies to the non-empty set of books and picks up one member out of this set.This representation says that there is a function f such that for every z, if z is a lady, z reads the book selected by this function.More informally, this representation means that there is a book that is read by every lady.
The choice function analysis provides a straightforward solution to the problems noted above with LF movement or unselective binding/absorption approaches.First, the contrast between (22a, d) and (22b) follows under the standard assumption from Szabolcsi and Zwarts (1993) that adverbial whphrases do not have an N-set and that they denote functions ranging over higher-order entities.(22a, d) are grammatical because whom and what way contain an N-set, a necessary condition for the choice function to work.On the other hand, (22b) is ungrammatical because how cannot be evaluated by choice function due to its lack of N-set; the required LF movement would be blocked by the shortest movement requirement.The "Donald Duck" problem, which we have seen to arise with unselective binding/absorption above, is also solved as the direct consequence of the choice function because, as defined in ( 27), a choice function applies to a non-empty set of individuals and yields a member out of this set.The value for y is, then, correctly ensured to be selected from the set of philosophers.

Deriving the properties of wh-in-situ in BI
Let us now see whether all the syntactic and semantic properties discovered by Saddy (1991) with respect to wh-in-situ in BI follow under the choice function approach.The relevant properties are summarized in (29a-e).
Kansas Working Papers in Linguistics, Vol. 30 (2008) (29a) is directly derived from the simple fact that wh-in-situ remains in situ throughout the syntactic derivation (including on the mapping to LF). ( 29b) is derived for the same reason; this class of expression cannot satisfy the [+WH] subcategorization requirement of verbs such as ingin tahu 'wonder' because it does not move under the assumption that this requirement can only be satisfied by syntactic movement of the wh-phrase into the specifier of the CP selected by such verbs.Similarly, to the extent that (29c) holds (recall our earlier conclusion that this property itself is silent about whether the movement has occurred or not), the lack of crossover effects is a natural consequence of the fact that wh-in-situ does not undergo movement.The properties in (29d, e) need more elaborate discussion.
As we have seen in section 2.3, Saddy observes that in-situ wh-phrases do not show scope interaction with quantifiers that c-command them but instead take only widest scope.To the extent that this judgment is real, his observation naturally follows from Reinhart's (1997Reinhart's ( , 1998) assumption that existential closure of a function variable introduced by an NP can be inserted in the highest possible position.Thus, sentences like (15a), repeated as (30a), would receive the LF representation in (30b).In (30b), choice function applies to a set of persons in a given world and picks out one member from this set.This representation corresponds to the wide scope reading of the in-situ wh-phrase over the universal quantifier in subject position.We have seen in section 3, however, that Saddy's characterization of the scope of wh-in-situ cannot be reproduced in our grammatical judgment task.This raises the question of whether the existential operator that binds the function variable does not always have to be introduced at the highest possible scopal position but instead can be introduced in the scope of another quantifier.If the answer is yes, the scope interaction between wh-in-situ and ccommanding universal quantifiers, as reported by the second author, is predictable.Indeed, Reinhart (1998) shows, based on (31a), that the operator can be inserted within the scope of another operator, as in (31b).
(31) a.Most linguists have looked at every analysis that solves some problem.
b.For most linguists x, (∃(f)) (CH (f) & (∀y) (analysis (y) and y solves f(problem)) → (x looked at y)). (Reinhart 1998: 40) According to Reinhart (1997: 40), "the choice of a problem may vary with the choice of a linguist, in which case some problem is not "specific."Nevertheless it can take scope over every analysis."Thus, this intermediate reading of the indefinite some problem is naturally accounted for if we assume that the existential operator is below another quantifier most, as shown in (31b).This analysis predicts that the corresponding wh-in-situ in BI should also be able to take this intermediate scope in sentences like (32a).This prediction is indeed confirmed.The LF representation for (32a) then looks like (32b).
This observation, on the other hand, would remain mysterious under Saddy's account because it crucially depends on his observation that wh-in-situ in BI always takes widest possible scope.

New predictions: The NP vs. non-NP Asymmetry
We have introduced Reinhart's approach to wh-in-situ and shown that all the properties associated with BI wh-in-situ can be straightforwardly derived from the notion of choice function.It is important that, given the definition of choice function in ( 27), we can make the prediction that the availability of choice function crucially depends on whether a given in-situ wh-phrase can denote an N-set.In other words, we predict that the NP vs. non-NP asymmetry should be observed essentially in the same way as in (22a, b, d).Cole and Hermon (1998) show that this prediction is indeed borne out in Malay.Examples in (33a-g) are constructed in BI based on the related but partial paradigm from Malay reported in Cole and Hermon (1998: 226) A slightly different implementation of the non-syntactic, in-situ approach to wh-in-situ has been independently proposed by Cole andHermon (1998, 2000) based on the data from Malay.Cole and Hermon (1998: 240) propose that "in wh-in-situ in Malay the (wh-OP) question operator is merged at the root Spec CP, and, therefore, unselectively binds a wh-variable in its scope."This analysis derives essentially the same set of facts concerning BI wh-in-situ as the choice function analysis.Then, their analysis might amount to the same thing as choice function.Indeed, they note (p.240), for example, that "since it does not affect the issues under consideration in this paper, we will maintain the pretense that the question operator binds the wh-variable directly rather than through the mediation of a choice function, and shall continue to employ the term 'unselective binding'."The same position is maintained in Cole and Hermon (2000: 106), who remark that "Reinhart 1995 argues that the correct mechanism for in situ interpretations of wh is a choice function rather than unselective binding.We leave this issue open since the precise mechanism for in situ interpretation is irrelevant for our analysis."Cole and Hermon (2000), however, is more explicit in their analysis of wh-in-situ, as shown in their proposed schematic representation given below.
(34) Unselective Binding of Wh-in situ: [ CP OP i [... [ CP ...wh i ]]] where wh is a variable in a base-generated position and OP is base-generated in scopal position and binds wh.(Cole and Hermon 2000: 109, their (17)) The explication of Cole and Hermon's analysis above might give us the impression that the choice function approach is a notational variant of their proposed version of unselective binding.However, we show below that the predictions do diverge with respect to the "Donald Duck" Problem and the intermediate scope reading noted by Reinhart and that the choice function analysis is superior to Cole and Hermon's analysis.The first divergence between the choice function approach and Cole and Hermon's version of the unselective binding concerns the interpretation of in-situ wh-phrases in BI contained within an if-clause.Recall that the LF representation of (25a), which would be derived under selective binding, fails to express the fact that the value of the denotation of the in-situ wh-phrase must be selected from the set of philosophers because this approach would leave the restriction (philosopher) in the implicational clause at LF, as shown in (25b), and render (25a) true even though the value of y is Donald Duck, as shown in (25c).This "Donald Duck" problem won't arise under the choice function approach because the value of y must be selected from the non-empty set of philosophers in a given model/world.Therefore, Cole and Hermon's analysis cannot capture the correct interpretation unless it is accompanied with special mechanisms of pulling out the restriction out of the antecedent of an implicational clause.Several technical additions would not be inconceivable to avoid this problem.The point here, however, is that none of such special additions is required under the proposed approach.The second domain in which the predictions of the two competing approaches would diverge concerns the intermediate scope reading illustrated by (32a).The LF representation in (32b) is derived under the choice function, which correctly captures the intermediate reading.It is not clear whether this intermediate scope reading would be derived under Cole and Hermon's version of the unselective binding approach.In their 1998 paper, they assume that the wh-in-situ in Malay is bound unselectively by the operator base-generated in the root [Spec, CP].Accordingly, it would falsely predict that the intermediate scope reading would be impossible.The same problem remains with their analysis updated in their 2000 paper because it base-generates the operator in the scopal [Spec, CP].Crucially, however, the intermediate reading in (32a) requires that the operator must be base-generated in a position in the matrix clause that is lower than the specifier of the matrix TP but higher than the complement of the matrix VP.The relevant reading would be impossible, contrary to facts.Based on these two divergences between Cole and Hermon's analysis and the choice function analysis, we conclude that the two analyses are not entirely the same.The latter analysis makes better predictions concerning the "Donald Duck" Problem and the intermediate scope reading.The two problems could be technically solvable by several special amendments on the mapping from syntax to LF under Cole and Hermon's analysis but the fact that these amendments are not necessary but instead derived from the way choice function independently works provide strong support in favor of the choice function analysis over the unselective binding analysis.

Conclusions
The correct licensing mechanism for wh-in-situ in BI is choice function à la Reinhart (1997Reinhart ( , 1998)).

( 13 )
Who did every athlete expect to win which prize? triplet answer: Gretsky expected Milli Vanilli to win an Oscar, Gefrion expected George Burns to win Grammy, etc. (Saddy 1991: 204) (14) [ S′ [ Comp Q i, j who i [ S e i every athlete expect….win which prize j ]] ) a. Who fainted when you attacked whom? b. * Who fainted when you behaved how? c. * How did Max faint when you behaved?d.Who fainted when you behaved what way?
). Prof dia i meng-ira saya men-cintai siapa i ?'*Who i does his i professor thinks I love?' If the [+WH] subcategorization of the verb ingin tahu 'wonder' [+WH] must be satisfied by the [+WH] feature within its complement CP, then the fact that in-situ wh-elements do not 16) a. Setiap orang tahu Tom mem-beli apa? 'What does every person know Tom bought?' every person know Tom TR-buy what (what>every, *every>what) b.Setiap orang tahu apa yang Tom beli t i ?'What does every person know Tom bought?' every person know what Foc Tom buy (what>every, every>what) c.Apa yang setiap orang tahu Tom beli t i ?'What does every person know Tom bought?' , p. 282 (29) Syntactic and Semantic Properties of Wh-in-Situ in BI a. Wh-in-situ is insensitive to syntactic island/ECP effects.b.Wh-in-situ is not able to satisfy verbs' +wh subcategorization requirement.c.Wh-in-situ is immune to weak/strong crossover effects.d.Wh-in-situ is scopally uninteractive, always taking non-overt wide scope.e. Wh-in-situ is not able to support a pair-list reading for multiple wh-questions.