The Order of Wh Words in Multiple Wh Extraction
in Serbo-Croatian*
Vedrana Mihaliček
The Ohio State University
1 INTRODUCTION
Bošković (1997, 2002) has claimed that, at least in some Serbo-Croatian constructions
involving multiple wh extraction, there are constraints on the order of fronted wh words.
His contention is that in embedded questions, correlatives and long distance extraction
the subject wh phrase has to linearly precede any fronted object or VP adjunct wh
phrases, in line with the Superiority Condition. In contrast, there are no such ordering
restrictions in main clause questions, according to Bošković. Then, it is argued that when
a multiple wh question in Serbo-Croatian obeys the Superiority condition, the highest wh
phrase was fronted by wh movement, which is subject to Superiority. On the other hand,
when a main clause question involves a Superiority violation 1, Bošković argues that it
was derived by some other, non-syntactic mechanism. This other grammatical mechanism
is claimed to be focus fronting (Stjepanović 2003, Bošković 2002).
In this paper, I will present data from the literature, as well as constructed and naturally
occurring examples which challenge Bošković's empirical generalization about the
possible orders of fronted wh words in Serbo-Croatian. I will argue that there are no
syntactic constraints on the order of fronted wh words in any construction involving wh
extraction. A corollary of this claim is that both the questions that involve so-called
Superiority violations and those that do not have exactly the same syntactic distribution.
That being the case, there is no principled reason to posit distinct grammatical
mechanisms for multiple wh questions with Superiority violations and those without. It
simply appears that whatever the mechanism responsible for wh fronting in Serbo-
Croatian, it is not constrained by the Superiority Condition.
*The author would like to thank Peter Culicover, Mary Beckman, Carl Pollard, Brian Joseph, the
participants of the 5th Graduate Colloquium on Slavic Linguistics and members of Synners, for helpful
discussions, comments, suggestions, guidance and/or encouragement. Many thanks also to Anastasia
Smirnova and Lauren Ressue for their helpful feedback on an earlier version of this paper.
1I will use the expression “contains/involves a Superiority violation” to refer to multiple wh questions in
which a non-subject wh phrase linearly precedes a wh phrase, out of mere terminological convenience. The
use of this term should not be interpreted as my commitment to the existence of the Superiority Condition.
1
2 THE DATA
2.1 Main Clause Questions
It has previously been noted in the literature that there are no ordering constraints on the
fronted wh phrases in Serbo-Croatian. Below is an example adapted from Rudin (1988a):
(1) [56] a. Ko je šta kome dao?
whonom is whatacc whodat given
'Who gave what to who?'
b. Ko je kome šta dao?
c. Šta je ko kome dao?
d. Šta je kome ko dao?
e. Kome je šta ko dao?
f. Kome je ko šta dao?
The above example shows that the three fronted wh words, ko, šta and kome, can appear
in any logically possible order. Note that the questions in (1) are all main clause multiple
wh questions.
2.2 Bošković's (1997, 2002) Data and Analysis
However, Bošković (1997) argues that only in main clause questions such as those in (1),
the fronted wh words can be freely ordered. In correlatives, embedded questions, and
long distance extraction, it is claimed that the subject wh word has to linearly precede any
object or VP adjunct wh words. Examples (2-8) and (12) below have been adapted from
Bošković (1997), and all the judgments indicated are Bošković's.2
(2) [16] a. Ko je šta tražio, taj je to i dobio
whonom is whatacc asked-for that-onenom is thatacc also got
'Everyone got what they asked for'
b.*Šta je ko tražio, taj je to/to je taj i dobio
(3) [17] a. Ko kako radi, on tako i zarađuje
whonom how works henom that-way also earns
'Everyone makes money the way they work'
2 It has to be noted that the present author disagrees with Bošković's judgments, and accepts all the
examples that Bošković judges ungrammatical. The author is a native speaker of a Bosnian dialect (spoken
in Sarajevo), and has also had extensive experience with a Croatian dialect of Serbo-Croatian (spoken in
Zagreb).
2
b.*Kako ko radi, on tako/tako on i zarađuje.
Examples (2) and (3) are supposed to show that in correlatives, the subject wh word has
to precede the object or the adjunct wh word, otherwise the sentence in ungrammatical.
Bošković also claims that sentences (2b) and (3b) are ungrammatical regardless of the
order of pronouns in the second clause - that is, whether the order of the pronouns in the
second clause parallels the order of wh phrases in the first clause or not.
(4) [11] a. Znate ko je koga istukao.
know-2pl whonom is whoacc beaten
'You know who beat who.'
b. Znate koga je ko istukao.
(5) [12] a. Jovan i Marko ne znaju ko je koga istukao.
Jovan and Marko don't-know whonom is whoacc
beaten
'Jovan and Marko don't know who beat who'
b.?*Jovan i Marko ne znaju koga je ko istukao.
(6) [13] a. Jovan i Marko ne znaju ko je kako istukao Ivana.
Jovan and Marko don't-know whonom has how beaten Ivan
'Jovan and Marko don't know who beat Ivan how'
b.?*Jovan i Marko ne znaju kako je ko istukao Ivana.
Bošković accepts both (4a) and (4b), even though in (4b) the subject wh word follows the
object wh word. He argues that this is because the wh questions in examples in (4) are not
truly embedded: What looks like matrix clause material (znate) is really adsentential, and
the wh question is in fact a matrix wh question. He suggests that this analysis is further
supported by the fact that there is a pause between znate and the wh question, while there
is no such pause between the matrix clause and the embedded question in examples in (5)
or (6) (Bošković 1997:7).3 Examples (5) and (6) are claimed to involve truly embedded
questions, because there is enough matrix clause material to preclude an adsentential
analysis, and, according to Bošković, only (5a) and (6a), where the subject wh word
precedes the other wh word, are fully grammatical.
As for long distance extraction, it has been previously noted that not all speakers of
Serbo-Croatian accept it at all (Rudin 1988a). Bošković (1997), however, claims that for
3The author made recordings of herself pronouncing examples (4-6) and found no phonetic evidence for a
pause between the matrix clause material and the embedded question in any of the examples. Since all the
wh words in these examples start with a voiceless stop, there is a "pause" corresponding to the stop closure
at the beginning of all the embedded wh questions, but this cannot be what Bošković (1997) has in mind.
Spectrograms available upon request.
3
those speakers who do accept long distance extraction, the subject wh word has to
precede any other wh words:
(7) [9] a. ?Ko ste gdje tvrdili da je zaspao?
whonom are where claimed C is fallen-asleep
'Who did you claim fell asleep where?'
b.*Gdje ste ko tvrdili da je zaspao?
(8) [10] a. Ko tvrdiš da koga voli?
whonom claim C whoacc loves
'Who do you claim loves who?'
b.*Koga tvrdiš da ko voli?
Stjepanović (2003) discusses another wh construction where, she claims, the order of wh
words is not free, and the subject wh word must precede any others, namely, sluicing with
multiple wh remnants. Below are two examples adapted from Stjepanović (2003), with
her judgments indicated4:
(9) [23] A: Neko je udario nekog.
somebodynom is hit somebodyacc
'Somebody hit somebody'
B: Ko koga?
whonom whoacc
'Who whom?
B': ?*Koga ko?
(10)[25] A: Neko je negdje sakrio blago
somebodynom is somewhere hidden treasure
'Somebody hid the treasure somewhere'
B: Ko gdje?
whonom where
'Who where?
B': ?*Gdje ko?
Reminiscent of Bošković's claim about the irrelevance of the order of pronouns in the
second clause of correlatives, Stjepanović (2003) claims that the degraded status of B'
questions in the above examples does not have anything to do with the order of indefinite
pronouns in the original statement A. That is, even if in (12A) nekog preceded neki,
(12B), it is claimed, would be preferred to (12B').
4 The author disagrees with Stjepanović's (2003) judgments also.
4
Effectively, the data that Bošković (1997) and Stjepanović (2003) present, if truly
accurate and representative, would imply that the multiple wh questions in Serbo-
Croatian where the subject wh word precedes any other wh words, and those in which it
doesn't, have a different syntactic distribution. It is important for Bošković (1997) to
show that this is the case so that the Superiority Condition can still be said to be operative
in Serbo-Croatian. Here is how the Superiority Condition is usually stated:
(11)The Superiority Condition (Chomsky 1973, Bošković 1997):
a. No rule can involve X, Y in the structure
... X ... [... Z ... WYV... ] ...
where the rule applies ambiguously to Z and Y and Z is superior to Y
b. The category A is superior to category B if every major category
dominating A dominates B as well but not conversely
So, since subjects are superior to any object or VP adjuncts, wh movement rule should
apply to subjects5. As a consequence, in multiple wh extraction, the subject wh phrase
should linearly precede any other wh phrases. If it does not, the question is said to
involve a Superiority violation. If questions with Superiority violations are
ungrammatical in a language, then that language is said to exhibit Superiority effects.
If Rudin's (1988) original generalization about the order of fronted wh phrases were
correct, then it would appear that whatever rule is responsible for fronting wh phrases in
Serbo-Croatian, it is insensitive to Superiority. By arguing that questions with Superiority
violations have a different syntactic distribution that those without - namely, that only the
latter can be embedded or appear in correlatives or involve long distance extraction or
sluicing - Bošković (1997) sets the stage for claiming that there are distinct mechanisms
responsible for the formation of each kind of multiple wh question. In those without
Superiority violations, the superior wh phrase was syntactically wh moved. In questions
with Superiority violations, all wh phrases were fronted by some other, non-syntactic
mechanism. It is argued that this mechanism is focus fronting (Stjepanović 2003,
Bošković 2002).
This way, Bošković (1997, 2002) can claim that there is wh movement in Serbo-Croatian,
and that Serbo-Croatian does exhibit Superiority effects. When Serbo-Croatian appears
not to show Superiority effects, i.e. when questions with Superiority violations seem to be
grammatical, that is because no syntactic wh movement occurred in those questions.
5 In Serbo-Croatian all wh phrases have to be fronted, but the clitic cluster comes after the first wh word.
Assuming a syntactic approach to the clitic cluster placement (e.g. Progovac 1996), if there are multiple
fronted wh words, only the linearly first one is analyzed as having landed in SpecCP via wh movement, the
clitic cluster being in C, and any other fronted wh phrases IP adjoined (this is basically the story that
Bošković (1997) accepts).
5
The claim that when the subject wh phrase is linearly first in a multiple wh question then
it was wh moved, and when it's not then it wasn't, would be completely circular without
the purported difference in syntactic distribution of the two kinds of questions. As another
independent piece of evidence for the existence of different wh fronting mechanisms in
Serbo-Croatian, Bošković (1997) claims that the interrogative complementizer li can only
occur in questions with no Superiority violations, i.e. those which involve syntactic wh
movement:
(12)a. [25a] Ko li je koga istukao?
whonom C is whoacc beaten
'Who on earth beat whom?'
b. [26a]*Koga li je ko istukao?
The interrogative complementizer, it is claimed, has to be introduced in overt syntax and
trigger overt wh movement. So it cannot occur in questions such as (10b) in which
arguably no wh movement took place.
2.3 Counterexamples to Bošković's (1997, 2002) empirical
generalization
In this section, data will be presented to challenge the claim that there are ordering
constraints on the fronted wh words in certain wh constructions in Serbo-Croatian. The
data presented here include constructed data, naturally occurring data, and examples from
elsewhere in the literature. It will be argued that in whatever syntactic environment
multiple wh questions with no Superiority violations can appear, so can the questions
with Superiority violations. Further, the data cut across major dialects of Serbo-Croatian,
so that the absence of Superiority effects cannot be explained away as a feature of a
particular dialect.
Consider first some counterexamples involving embedded questions. Godjevac (2000)
notes that she and her consultants, all speakers of a Serbian dialect, accept examples of
embedded wh questions which Bošković (1997) judges to be ungrammatical. She also
offers an original example:
(13)[137c] Pitam se šta ko kome sada govori.
ask-1s REFL whatacc whonom whodat now says
'I wonder who is saying what to who now'
In (13), the accusative wh word precedes the nominative wh word, which precedes the
dative wh word. The order of wh words in this question clearly violates Superiority, so on
Bošković's analysis it doesn't involve wh movement, and it shouldn't be possible to
embed it. Of course, Bošković could argue that the matrix clause material in (13) is
meager and possibly adsentential. However, in at least some of the naturally occurring
6
data presented below there in no doubt that the question with a Superiority violation is
really embedded, accepting Bošković's contention that the sheer amount of matrix clause
material precludes an adsentential analysis.
(14)a. Jel znaš koga ko možda ima nasnimano na kazeti??6
Q know whoacc whonom maybe has recorded on tape
'Do you maybe know who has who recorded on tape??'
b. Slažem se da ne treba ništa komentarisati
agree-1s REFL C doesn't-need nothing comment
kome je ko zašto dao glas pre nego što se
whodat is whonom why given vote before than C REFL
objave rezultati7
announce results
'I agree that it shouldn't be commented on at all who has given a vote to who and
why before the results are published.'
c. Sarajevo tačno zna gdje je ko, šta je ko radio,
Sarajevo exactly knows where is whonom whatacc has whonom done
ko je odakle i ko gdje živi.8
whonom is from-where and whonom where lives
'Sarajevo knows exactly who is where, who has been doing what, who is from
where, and who lives where.'
d. Film ipak nedaje onu neizvjesnot prvog djela već
movie still doesn't-give that suspense first part but
se većinom svodi na to
REFL mostly comes-down to that
kako je ko prošao9
how is whonom fared
6 http://www.index.hr/forum/default.aspx?q=t&idf=15&idt=34341&p=1
7 http://dizajnzona.com/forums/lofiversion/index.php/t6227.html
8 http://strippyvamprica.blogger.ba/arhiva/2005/02/15
9 http://www.blog.hr/print/?id=1622784039
7
'The movie still doesn't have the suspense of the first part, but mostly comes down to
(showing) who fared how.'
All the sentences in (14) contain an embedded multiple wh question with a Superiority
violation. Two things are important to note: First, at least for sentences in (14b-d), there is
no question that there is enough matrix clause material to preclude the adsentential
analysis. According to Bošković's own criterion, these must be instances of true
embedding. Second, these examples cut across different dialects of Serbo-Croatian,
which strongly suggests that the absence of Superiority effects cannot be analyzed as a
parameter of dialectal variation.
In particular, (14a) was clearly produced by a speaker of a Croatian dialect, because the
lexical variant kazeta occurs instead of kaseta, the latter being prevalent in Bosnia and
Serbia. (14b) was definitely produced by a speaker from Serbia for two reasons: Ekavian
variant pre occurs instead of prije which excludes ijekavian Bosnian dialects and most
Croatian dialects. Moreover, the use of the lexical variant komentarisati clearly excludes
ekavian North Western Croatian dialects, where komentirati would be required. Finally,
(14c) was most likely produced by a Bosnian speaker because the dialect is ijekavian
(gdje is used instead of gde), but Croatian dialects are excluded on account of the use of
the word tačno (the Croatian variant being točno).
It is also worth noting that Godjevac (2000) and her informants, and the present author
collectively represent the three major dialects of Serbo-Croatian: Serbian, Bosnian and
Croatian, yet we all accept embedded questions involving Superiority violations.
Now consider a few naturally occurring examples of correlatives containing a wh
question with a Superiority violation:
(15)a. Kao fol, šta je ko tražio to je i dobio??10
like trick whatacc is whonom asked-for that is also got
'Supposedly, everybody got what they asked for??'
b. Istorija je nešto mnogo kompleksnije od stupidnih zaključivanja
history is something much more-complex than stupid conclusions
"šta je ko tražio to je i dobio".11
whatacc is whonom asked-for that is also got
10 http://www.sarajevo-x.com/forum/search.php?
search_id=790485176&start=25&sid=58ce467159f7a337b666bfdc2a17a008
11 http://www.forum.hr/showthread.php?t=60120&page=4
8
'History is something much more complex than stupidly concluding "Everybody
got what they asked for".'
c. Kako tko radi tako mu i je.
how whonom works that-way him also is
'Everybody fares depending on how they work'
d. Zato i pišem, a slažem se,
for-that-reason also write-1s and agree-1s REFL
kako ko radi tako mu i bude.12
how who works that-way him also will-be
'That's why I'm writing, and I agree, everybody will fare depending on how they work.'
All the examples in (15) involve multiple wh questions with a Superiority violation, that
Bošković (1997, 2002) claims cannot occur in correlatives. Although it is less clear which
dialects these examples represent, it is obvious that (15c) was produced by a Croatian
speaker for two reasons: First, tko is used, a spelling (rarely phonological) variant of ko,
which only occurs in Croatia. Second, the sentence ends with je, a third person singular
verbal clitic. Only in (some) Croatian dialects can je be stranded at the end of a sentence
and not occur as a part of the clitic cluster. A Bosnian or Serbian speaker would have to
use the corresponding full form jeste in this context. On the other hand, (15b) was clearly
not produced by a Croatian speaker because the Croatian variant of istorija is povijest.
So, it seems that Superiority effects are absent in correlatives at least in a Croatian and
some non-Croatian dialect.
Recall that Bošković (1997, 2002) claims that in any long distance extraction, the subject
wh word has to precede any other wh words. Moreover, it is also claimed that if one wh
word remains in situ in the downstairs clause, it cannot be the subject wh word, because
that would constitute a Superiority violation (Bošković 1997: 6). First consider some
constructed examples of long distance extraction which this author accepts and which
involve Superiority violations:
(16)a. Gdje ste ko mislili da će nastupiti?
where have whonom thought C will perform
'Who did you think would perform where?'
b. Šta ko želite da vam pokloni za rođendan?
whatacc whonom want C you give for birthday
12 http://www2.serbiancafe.com/lat/diskusije/mesg/78/010537844.html?47
9
'Who do you want to give you what for your birthday?
(17)a. Kako Amerika hoće da se ko otcijepi?
how America wants C REFL whonom secede
'Who does America want to secede how?'
b. Šta misliš da će ti ko dati za rođendan?
whatacc think C will you whonom give for birthday
'Who do you think will give you what for your birthday?'
In examples in (16), both wh words are on the left periphery of the matrix clause, and the
subject wh word does not precede the other wh word. In examples in (17), the subject wh
word remains in situ in the embedded clause, and the non-subject wh word from the
embedded clause is fronted.
In addition, Rudin (1988b) contends that if in long-distance extraction in Serbo-Croatian
one wh word remains in situ, it can be either the subject or the non-subject one. Here is
an example she offers:
(18)[25b] Šta želite da vam ko kupi?
whatacc want-2pl C you whonom buy
'What do you want who to buy you?
Since the object wh word is fronted over the superior subject wh word, (18) clearly
involves a Superiority violation.
Finally, consider a few naturally occurring examples in which an object or adjunct wh
word from the embedded clause is fronted to the left periphery of the matrix clause and
precedes the matrix clause subject wh word:
(19)a. U Smaku je svirao, neću reći šta je ko hteo,
in Smak is played not-will say what is who wanted
ali šta je ko mislio da je najbolje da se uradi. 13
but whatacc is whonom thought C is the-best C REFL do
'He played at Smak, I won't say who wanted what, but who thought what the best
thing to do is.'
13 http://www.znaksagite.com/diskusije/viewtopic.php?p=118994
10
b. ...e sad, druga je tema kako ko misli da ti psi žive...14
but now another is issue how whonom thinks C thosedogs live
'Now, it's a different issue how who thinks that those dogs live'
c. [S]tvar je izbora, kako ko želi da hrani svog psa.15
thing is of-choice how whonom wants C feed POSS dog
'It's a matter of choice, who wants to feed their dog how.
Examples in (19) presumably also involve Superiority violations in long distance
extraction, because the matrix subject wh word is clearly superior to any object or adjunct
wh words from the downstairs clause, yet the subject wh word is not left-most.
As for sluicing with multiple wh remnants, it is first worth noting that Stjepanović (2003)
marks all the questions she wishes to exclude with '?*', and not '*' (see examples (9) and
(10) above). This choice in notation can be inferred to indicate that even in her judgment
these examples are at most severely degraded but not completely unacceptable.16
This author finds examples of sluicing involving Superiority violations perfectly
acceptable. However, contrary to Stjepanović's (2003) claim, it is important that the order
of wh remnants parallel the order of indefinite pronouns in the initial sentence. Consider
the following example:
(20) A: Negdje je neko sakrio blago.
somewhere is somebodynom hidden treasure
'Somebody hid the treasure somewhere'
B: Gdje ko?
where whonom
'Who where?'
B': ?Ko gdje?
If the order of indefinite pronouns in the A sentence is adjunct (negdje) then subject
(neko), it is much better to respond with (22B) than (22B'), even though (22B') doesn't
involve a Superiority violation, while (22B) does.
14 http://svijet-ljubimaca.com/forum/viewtopic.php?p=169426
15 http://www.sdcafe.co.yu/viewtopic.php?p=171634
16In fact, the same is true of examples involving embedded questions with Superiority violations that
Bošković (1997) presents (see examples (5) and (6) above).
11
While there are no naturally occurring data of sluicing with multiple wh remnants to be
presented here, since Stjepanović's (2003) does not categorically exclude examples with
Superiority violations, and this author finds them perfectly acceptable, the correct
empirical generalization cannot be that multiple wh questions with Superiority violations
simply do not occur in that construction. At best, such examples appear to be degraded
for some speakers, such as Stjepanović. But this is a very weak claim that cannot of itself
establish that questions with Superiority violations and those without have a different
syntactic distribution.
Since in this section a fairly convincing and data-driven case was made that Superiority
violations can occur in correlatives, long distance extraction, embedded questions and (at
least for some speakers) sluicing, it appears that there is no empirical basis for the claim
that multiple wh questions in Serbo-Croatian have a different syntactic distribution
depending on whether the subject wh word occurs linearly first or not. The two kinds of
questions seem to have exactly the same syntactic distribution. So, based on distributional
facts alone, there is no need to posit two distinct fronting mechanisms, wh movement and
focus fronting.
Recall that another piece of evidence offered in support of positing two distinct wh
fronting mechanisms was the placement of the interrogative complementizer li -
Bošković (1997) claimed that it can only occur in questions without Superiority
violations, so that a different, syntactic, mechanism must be involved in those questions,
than in questions with Superiority violations in which li supposedly cannot occur.
Consider the following naturally occurring examples:
(20) Mislim da BiH ima prečih problema nego
think C BiH has more-pressing problems than
šta li ko radi između svoja četri zida17
whatacc C whonom does between one's four walls
'I think that BiH has more pressing problems than who does what in their own
home.'
(21) Šta
li je
ko
toliko
zgrešio,
kad
desetak
whatacc
C has
whonom
that-much
done-wrong
when
about-ten
ljudi
(glumaca)
unose
"dokaze"? 18
people actors bring-in evidence
17 http://www.sarajevo-x.com/forum/viewtopic.php?p=606350&highlight=&sid=85fabd8bc169ab3a040b8eb061b4d5bb
18 http://www.danas.co.yu/20021217/dijalog.htm
12
'Who on Earth has done so much wrong, when about ten people (actors) are
bringing in "evidence"?
In (21), li occurs in a matrix question that contains a Superiority violation. In the question
in (20), the object wh word precedes the subject one, yet the question contains the
interrogative complementizer li. Moreover, remember that it should also be impossible to
embed a question with a Superiority violation, but this is exactly what we see in
(20).
It is likely that (20) was produced by a speaker from Bosnia, both because of the content
and the fact that it was found on a Bosnian website. (21) was most likely produced by a
Serbian speaker because of the use of zgrešio, which is an ekavian dialect form (in
Bosnia and most of Croatia it would be zgriješio).
In sum, the naturally occurring data and this author's judgments seem to indicate that,
contrary to Bošković (1997), the use of the interrogative complementizer li has nothing to
do with whether the question involves a Superiority violation or not.
Since multiple wh questions with and without Superiority violations can occur in all the
same constructions, and can co-occur with the interrogative complementizer, there
appears to be no empirical justification for the claim that the two kinds of questions are
syntactically different and formed by distinct grammatical mechanisms.
3 DISCUSSION AND CONCLUSION
In the preceding section an attempt was made to show that the data do not support
Bošković's claim that there are ordering constraints on the fronted wh words in Serbo-
Croatian. Whether the subject wh word precedes the other wh words or not does not
affect the syntactic distribution of the question or the placement of the interrogative
complementizer li. So, there is no empirical motivation for claiming that the two kinds of
questions are syntactically different and involve distinct grammatical mechanisms. It
appears to be the case that whatever single grammatical mechanism is responsible for the
formation of multiple wh questions in Serbo-Croatian, it is not constrained by Superiority
and it allows the wh phrases to be freely ordered.
One question that this paper doesn't attempt to answer is what the exact nature of the
mechanism of extraction is, and which grammatical component it properly belongs to, i.e.
whether it a syntactic or a non-syntactic mechanism. At this point, though, the author
would like to suggest that it is a syntactic mechanism since reconstruction effects are
observed. Consider the following example:
13
(22)[Koje [svoje]i/*k članke]j je Markok rekao da će [[njegov]*i/k profesor]i objaviti tj?
'[Which of [SVOJ]i/*k articles]j did Markok say that [[his]*i/k professor]i would publish
tj?'
Here, the crucial fact concerns the interpretation of the subject oriented possessive svoj,
which is always interpreted as co-referential with the subject NP of the clause it occurs
in. In (22), svoj has to be interpreted as co-referential with the subject of the embedded
clause which it was extracted from, namely njegov profesor, and it cannot be interpreted
as co-referential with the matrix clause subject Marko. In other words, svoj is interpreted
as if it were still in the downstairs clause, and these reconstruction effects are usually
taken to point to syntactic extraction.
Another question left open is how exactly to reconcile the data presented in this paper on
the one hand, with the data that Bošković (1997, 2002) and Stjepanović (2003) present on
the other. An attractive interpretation of the data relies on the claim that Superiority
effects do not arise because of any grammatical constraint, but are a result of an increased
processing load associated with questions that contain so-called Superiority violations
(see Sag et al (under review), Hofmeister et al (2007)).
The processing factors that are argued to influence the degree of acceptability of multiple
wh questions are the distance between the filler and the gap and the accessibility of the
wh fillers and interveners. In questions with a Superiority violation, the distance between
the left most filler and its gap is greater than in questions without a Superiority violation.
And fillers and interveners that are which X wh phrases are more accessible than bare wh
words, so the presence of more accessible wh phrases should facilitate the processing of
questions with Superiority violations and make them more acceptable to speakers. For
empirical evidence supporting the claim that the manipulation of these two factors
(distance and acceptability) influences the speakers' acceptability judgments as predicted
see Featherston (2005) (for German and English) and Hofmeister et al (2007) (for
English).
The processing account of Superiority effects predicts overall gradience in acceptability
judgments and is more compatible with speaker-to-speaker variation in judgments, than a
grammatical account of Superiority effects. Recall that Bošković (1997, 2002) and
Stjepanović (2003) do not categorically exclude all the examples they wish to exclude,
but seem to at least sometimes only mark them as (severely) degraded. Also,
counterexamples to Bošković's (1997, 2002) empirical generalizations abound, as shown
in Section 2.2. This state of affairs is perfectly consistent with the processing account, but
it is not at all obvious how it could be reconciled with the view of Superiority as a
grammatical principle. If the processing account is right, it may well be the case that
some speakers such as Bošković simply have a higher acceptability threshold than other
speakers.
14
Further, notice that the processing account explains away the effects of D-linking
(Pesetsky 1987) and obviates the need for a distinct grammatical mechanism of D-
linking. Which X fillers and interveners facilitate the integration of the filler at the gap
site, so they are predicted to ameliorate questions with so-called Superiority violations
because of processing, not grammatical, factors. The consequence is that the grammar
itself is simplified.
In conclusion, to account for multiple wh extraction in Serbo-Croatian, only one,
probably syntactic, mechanism is needed. Gradience and speaker-to-speaker variation in
acceptability judgments are most likely a consequence of extragrammatical processing
factors.
15
REFERENCES
Bošković, Ž. (1997). Superiority effects with multiple wh fronting in Serbo-Croatian. Lingua 102:
1-20.
Bošković, Ž. (2002). On multiple wh fronting. Linguistic Inquiry 33: 351-383.
Chomsky, N. (1973). Conditions on transformations. In Anderson, S. R. and P. Kiparsky, eds. A
Festschrift for Morris Halle. New York: Holt, Reinhart and Winston.
Featherston, S. (2005). Universals and grammaticality: Wh-constraints in German and English.
Linguistics 43: 667-711.
Godjevac, S. (2000). Intonation, Focus Projection and Word Order in Serbo-Croatian. PhD
dissertation, The Ohio State University.
Hofmeister, P., T.F. Jaeger, I.A. Sag, I. Arnon and N Snider. (2007). Locality and accessibility in
wh-questions. In Featherston, S. and W. Sternefeld, eds. Roots: Linguistics in Search of
its Evidential Base. Berlin: Mouton de Gruyter.
Pesetsky, D. (1987). Wh-in-situ: Movement and unselective binding. In Reuland, E. and A. ter
Meulen, eds. The Representation if (In)Definiteness. Cambridge, MA: MIT Press.
Progovac, Lj. (1996). Clitics in Serbian/Croatian: Comp as the second position. In Halpern, A.
and A. Zwicky, eds. Approaching Second: Second Position Clitics and Related
Phenomena. Stanford: CSLI Publications.
Rudin, C. (1988a). On multiple questions and multiple wh fronting. Natural Language and
Linguistic Theory 6: 445-501.
Rudin, C. (1988b). Multiple questions in South Slavic, West Slavic and Romanian. The Slavic
and East European Journal 32: 1-24.
Sag, I. A., I. Arnon, B. Estigarribia, P. Hofmeister, T. F. Jaeger, J. Pettibone and N.
Snider. (under review). Processing Accounts for Superiority Effects.
Stjepanović, S. (2003). Multiple Wh-fronting in Serbo-Croatian matrix questions and the matrix
sluicing construction. In Boecx, C. and K. Grohmann, eds. Multiple Wh Fronting.
Philadelphia: John Benjamins.
16