PreprintPDF Available

Extraction from clausal adjuncts in Czech: A rating experiment

Preprints and early-stage research may not have been peer reviewed yet.
Preprint

Extraction from clausal adjuncts in Czech: A rating experiment

Abstract

The paper provides experimental evidence (naturalness rating paradigm) that wh-extraction from clausal adjuncts (conditionals, temporal clauses, purpose clauses) is grammatical in Czech, although it is fully natural only in one specific condition: if the adjunct is left-peripheral (vs. right-peripheral/central), if the adjunct is not headed by nominal material (the case if..., the moment when...), and if the extracted wh-word is a relative pronoun (vs. an interrogative phrase). The evidence supports the analysis proposed in Biskup & Šimík (2019; https://ling.auf.net/lingbuzz/004573), whereby left adjuncts are proposition-denoting clauses/CPs (and hence transparent), while right adjuncts are entity-denoting free relatives/DPs (and hence islands). The evidence is incompatible with a discourse-based analysis of islands (e.g. Abeille et al. 2020; https://doi.org/10.1016/j.cognition.2020.104293), which predicts right (focused) material to be more transparent than left (backgrounded) material.
Extraction from clausal adjuncts in Czech: A rating experiment*
Radek ˇ
Sim´
ık1, Petr Biskup2, Kateˇ
rina Bartasov´
a1, Mark´
eta Danˇ
cov´
a3, Eliˇ
ska
Dost´
alkov´
a1, Kateˇ
rina Hrdinkov´
a1, Gabriela Koskov´
a1, Jarom´
ır Koz´
ak1, Kl´
ara
Lupomˇ
esk´
a1, Albert Marˇ
s´
ık1, Edita Schejbalov´
a1& Illia Yekimov1
1Charles University, 2University of Leipzig, 3Palack´
y University
1. Introduction
This paper focuses on the issue of wh-extraction from clausal (tensed) adjuncts in Czech.
Building on the observation that Czech clausal adjuncts are in some cases transparent
for extraction (Leˇ
snerov´
a and Oliva 2003, Biskup and ˇ
Sim´
ık 2019), we have designed a
naturalness-rating experiment with the aim to evaluate the adequacy of two competing ap-
proaches: the syntactic/semantic approach of Biskup and ˇ
Sim´
ık (2019) and the discourse-
based approach of Abeill´
e et al. (2020).
According to Biskup and ˇ
Sim´
ık (2019), adjuncts in Czech are transparent for extrac-
tion if they are proposition-denoting clauses (CPs), syntactically and semantically akin
to (embedded) questions. They are islands if they are entity-denoting complex nominals
(often prepositional), akin to (free) relative clauses. The former scenario obtains if the
adjunct is left-peripheral, the latter if it is right-peripheral or if the nominal structure is
overtly represented. According to the discourse-based approach, left-peripheral clauses are
informationally backgrounded, which makes them less transparent than the corresponding
right-peripheral ones, which are focused. Nevertheless, as recently argued by Abeill´
e et al.
(2020), even backgrounded constituents can be transparent as long as the extracted ele-
ment is also backgrounded. We test this prediction by looking at the difference between the
extraction of relative pronouns (backgrounded) vs. interrogative phrases (focused).
Our experimental results largely support the syntactic/semantic analysis of Biskup and
ˇ
Sim´
ık (2019) and cannot be easily reconciled with the discourse-based approach of Abeill´
e
et al. (2020). In addition, we discover effects not predicted by neither account.
*Besides NELS 52, this work was presented in the ERCEL research group (June 2021; Charles University)
and at the workshop Adverbial clauses: between subordination and coordination (May 2022; University of
Cologne). We received valuable feedback from James Brand, Sam Featherston, Jan Chrom´
y, Radim Lacina,
and Mikul´
aˇ
s Preininger.
ˇ
Sim´
ık & Biskup et al.
The paper is structured as follows. In section 2 we provide some background and
the relevant empirical pattern as reported in the literature. In 3 we discuss theoretical ap-
proaches applicable to the pattern, with special attention paid to Biskup and ˇ
Sim´
ık (2019)
and Abeill´
e et al. (2020). Section 4 reports on our naturalness rating experiment. Finally,
section 5 discusses the results and concludes.
2. Background
Clausal adjuncts have traditionally been considered islands for extraction because of the
ungrammaticality of sentences like (1) (e.g. Huang 1982, Cinque 1990).
(1) *Who1did Mary cry [after John hit t1]?
Yet over the years exceptions have frequently been reported (Grosu 1981, Deane 1991,
Kluender 1998); for more examples and references see Chaves’s (2021) survey, which is
also the source of the examples below (p. 691).
(2) a. These are the pills OP1that Mary died [before she could take t1].
b. The person who1I would kill myself [if I couldn’t marry t1] is Jane.
c. Which book1will Kim understand linguistics better [if she reads t1]?
There is also a growing body of experimental evidence supporting the stance that at least
some clausal adjuncts in some languages are indeed transparent for extraction (Dal Farra
2019, Kush et al. 2019, Bondevik et al. 2021). Below are two examples from Bondevik
et al.’s (2021) experiment, involving the extraction of a contrastive topic (CT; manipulated
by a preamble) out of a clausal adjunct. Example (3a) (if-clause) was rated as rather “good”
while (3b) (because-clause) as “bad” (here *), which demonstrates that extraction from
clausal adjuncts is in principle possible, but also that not all adjuncts are alike.
(3) Norwegian (Bondevik et al. 2021:229)
a. [. . . ] men
but
[CT takkekortene1]
thank.you.cards.DEF
blir
becomes
hun
she
skuffet
disappointed
[om
if
de
they
glemmer
forget
˚
a
to
sende
send
ut
out
t1med
with
en
one
gang].
time
Literally: ‘[. . . ] but the thank-you cards she will be disappointed if they forget
to send out right away.’
b. *[. . . ] men
but
[CT vintertemperaturene1]
winter.temperatures.DEF
blir
becomes
hun
she
boende
living
[fordi
because
hun
she
liker
likes
t1].
Literally: ‘[. . . ] but the winter temperatures she stays there because she likes.
Extraction from clausal adjuncts in Czech: A rating experiment
Extraction from clausal adjuncts is also attested in Czech, as first noticed and discussed –
based on corpus evidence – by Leˇ
snerov´
a and Oliva (2003).
(4) Leˇ
snerov´
a and Oliva (2003:241)
Na
on
kaˇ
zd´
e
every
zakoupen´
e
purchased
plyˇ
sov´
e
plush
hraˇ
cce
toy
je
is
n´
alepka,
sticker
kterou1
which
[kdyˇ
z
when
d´
ıtˇ
e
child
odevzd´
a
hands.in.3SG
t1v
in
ZOO],
ZOO
obdrˇ
z´
ı
receive.3SG
nav´
ıc
in.addition
drobn´
y
small
d´
arek.
gift
‘On every purchased plush toy there’s a sticker such that when a child hands in the
sticker in the ZOO, the child will receive a small gift in addition.
Biskup and ˇ
Sim´
ık (2019), who offered a formal analysis of this type of extraction (see be-
low), noticed that extraction is possible not just from conditional adjuncts, but also purpose
clauses or correlatives. Crucially, however, extraction is only claimed to be grammatical
if the adjunct appears to the left of the clause it modifies; compare the grammatical (a)
examples with the ungrammatical (b) examples below.
(5) a. To
it
je
is
ten
that
chlap,
man
kter´
emu1
which.DAT
[co(koliv)2
what(ever).ACC
d´
aˇ
s
give.2SG
t2t1], to
it
ztrat´
ı.
lose.3
b. *To
it
je
is
ten
that
chlap,
man
kter´
emu1
which.DAT
ztrat´
ı
lose.3
[co(koliv)2
what(ever).ACC
d´
aˇ
s
give.2SG
t2t1].
(Intended:) ‘This is the man such that he will lose whatever you give him.’
(6) a. To
it
je
is
ten
the
ˇ
reˇ
cn´
ık,
speaker
kter´
eho1
which.ACC
[aby
in.order.SBJV.3
nal´
akali
attract.PL
t1], museli
must.PL
by
SB JV.3
m´
ıt
have
pen´
ıze.
money.
b. *To
it
je
is
ten
the
ˇ
reˇ
cn´
ık,
speaker
kter´
eho1
which.ACC
museli
must.PL
by
SB JV.3
m´
ıt
have
pen´
ıze
money
[aby
in.order.SBJV.3
nal´
akali
attract.PL
t1].
(Intended:) ‘This is a speaker such that they need money in order to attract
him.
Biskup and ˇ
Sim´
ık (2019) further observed that when the adjunct clause has the form of a
complex nominal, extraction is blocked. This is the case for the adjunct types mentioned
above, as illustrated in (7) for the purpose adjunct, as well as for adjuncts that are introduced
by complementizers whose internal structure involves a nominal layer. This is illustrated
in (8) for a causal adjunct, whereby the Czech causal complementizer protoˇ
ze ‘because’ is
composed of pro ‘for’, to ‘that’ (demonstrative), and ˇ
ze ‘that’ (declarative complementizer),
suggesting that protoˇ
ze-clauses in Czech are in fact complex nominals embedded in a PP.
ˇ
Sim´
ık & Biskup et al.
(7) a. *To
it
je
is
ten
the
ˇ
reˇ
cn´
ık,
speaker
kter´
eho1
which.ACC
[na
for
to,
that.DEM
aby
in.order.SBJV.3
nal´
akali
attract.PL
t1],
museli
must.PL
by
SB JV.3
m´
ıt
have
pen´
ıze.
money.
Intended: ‘This is a speaker such that they need money in order to attract him.
b. *To
it
je
is
ta
that
kn´
ıˇ
zka,
book
kterou1
which.ACC
[protoˇ
ze
because
koupil
bought..SG .M
t1], uˇ
z
anymore
nem´
a
NE G.has.3S G
pen´
ıze.
money
Intended: ‘This is a book such that he doesn’t have any more money because
he bought that book.
3. Competing theories
The reported empirical pattern, or more precisely the correlation between adjunct posi-
tion and its transparency, might strike an expert in island syntax as puzzling. If there is
a left–right asymmetry, then it should go in the opposite direction: one would expect a
right-peripheral structure to be more transparent for extraction than the corresponding left-
peripheral structure, not conversely.
The asymmetry has received significant attention in the literature. For instance, Haege-
man (2003, 2010) has accummulated plenty of evidence that left (called “peripheral”) ad-
juncts are syntactically more complex than right (“central”) adjuncts. If syntactic complex-
ity positively correlates with opacity for extraction (the more complex a structure, the more
opaque for extraction; e.g. Starke 2001, Abels 2012), the reported pattern is unexpected.
The same holds for analyses based on movement: while in-situ structures may be transpar-
ent for extraction, they become islands once they move. This type of analysis, originally
devised to account for the subject–object island asymmetry or in-situ vs. scrambled object
asymmetry (so-called condition on extraction domains/CED; Huang 1982, M¨
uller 2010),
could in principle be applied in our case, too, but – alas – with the wrong result. Finally, it
has been argued that informationally backgrounded (given, presupposed, topicalized) con-
stituents are less transparent for extraction than focused constituents (Erteschik-Shir 1973,
2007, Goldberg 2006, Abeill´
e et al. 2020). The backgrounded vs. focused nature of left vs.
right adjunct clauses makes sense for Czech, a language that heavily relies on word order
in the expression of information structure. Basically all the existing theories of Czech in-
formation structure (see, e.g., Mathesius 1941, 1947, Daneˇ
s 1964, Sgall et al. 1986, Firbas
1992, Kuˇ
cerov´
a 2007, Biskup 2011) rely on the idea the backgrounded or given is placed
before focused or new.1This is considered to hold for non-clausal and clausal constituents
alike (Mathesius 1947, Chudobov´
a 2011).
1See ˇ
Sim´
ık and Wierzba (2015, 2017) for a dissenting view, although one that does not make much of a
difference for the case at hand.
Extraction from clausal adjuncts in Czech: A rating experiment
3.1 Biskup and ˇ
Sim´
ık (2019)
Biskup and ˇ
Sim´
ık (2019) propose to account for the left-right asymmtry in the following
way. There is no adjunct island condition per se (in Czech). What seems like adjunct islands
are in fact complex nominal islands. If adjuncts are placed right-peripherally (or “centrally”
in terms of Haegeman 2010), they are always nominals. More particularly, they are like
free relatives in that they have an external nominal syntax as well as semantics – they
denote entities of the relevant sort, e.g. a particular time interval in the case of temporal
adjuncts (Hall and Caponigro 2011) or a particular (maximal) situation/world in the case
of conditionals (Schlenker 2004). The referential semantics facilitates their integration into
the event structure of the host clause, but the side effect is that their effectively nominal
nature renders the adjunct clause a (complex NP) island. Biskup and ˇ
Sim´
ık (2019) propose
a phase-based syntactic analysis of the complex nominal island fine-tuned for the purpose
of NP languages, but, in principle, any complex nominal island analysis will do the job of
deriving the observed empirical pattern.
Left adjuncts (called “peripheral” by Haegeman), on the other hand, are syntactically
clausal/CPs. Semantically, they are not entities, but rather propositions, as in the standard
Kratzerian account of conditionals (Kratzer 2012), in proposition-based analyses of correl-
atives (Bittner 2001, Brasoveanu 2008), or unconditionals (Rawlins 2013). The proposition
restricts a potential modal or quantificational operator in the host clause. At the same time,
it introduces the relevant kind of variable (e.g. time interval in temporal clauses) which is
then referentially related to the corresponding participant in the event structure of the host
clause (the event time).
If a clausal adjunct is headed by a nominal – possibly integrated in the complex adver-
bial complementizer (see (7)), it is predicted to be an island independently of its position.
3.2 Abeill´
e et al. (2020)
Abeill´
e et al. (2020) recently proposed a refinement of the discourse-based theory (going
back to Erteschik-Shir 1973). They postulate the focus-background conflict (FBC) con-
straint, which states that “a focused element should not be part of a backgrounded con-
stituent” (p. 3). In their experiment, they compare the extraction from subjects and objects
on the one hand and the extraction of relative vs. interrogative pronouns on the other. They
find that it is more acceptable to extract relative than interrogative pronouns from subjects.
This is claimed to be predicted by the FBC constraint. Since relative pronouns, as opposed
to interrogative ones, are backgrounded, they do not give rise to a violation of the FBC
constraint.2
2There are some caveats in their theory. First, it is not clear why the grammatical categories of subject
vs. object should correspond to the discourse categories backgrounded vs. focused (obviously, subjects can
be focused and objects backgrounded). Second, it is unclear why the FBC constraint should be violated by
an interrogative pronoun (focus) that is extracted from the subject (background). When it is extracted, it is
not really a part of the extraction site, unless one assumes a copy or trace theory of filler-gap dependencies,
which the authors do not (p. 1). We leave these caveats aside and concentrate on the predictions stated.
ˇ
Sim´
ık & Biskup et al.
The FBC-based theory correctly predicts the acceptability of the extractions from the
left adjuncts observed by Biskup and ˇ
Sim´
ık (2019). This is because all of them involve
relative pronouns, whose backgrounded nature makes the extraction from a backgrounded
clause possible. The theory further makes the prediction that extracting an interrogative
pronoun/phrase should be less acceptable because it would consitute a violation of the
FBC constraint. However, it also expects that the extraction from right (focused) adjuncts
should be acceptable, irrespective of the relative vs. interrogative nature of the extracted
element. This prediction is clearly problematic in the light of the data reported in Biskup
and ˇ
Sim´
ık (2019). However, it is good to keep in mind that Biskup and ˇ
Sim´
ık’s data rely
on their own individual judgments, which, of course, might be wrong.
3.3 Filling in the gaps
We have so far identified three factors relevant for the extraction out of clausal islands in
Czech: (i) the adjunct position, (ii) the presence of a nominal layer, and (ii) the nature of
the extracted element. So far, the syntactic/semantic theory makes predictions about (i) and
(ii) (but not (iii)) and the discourse-based theory about (i) and (iii) (but not (ii)). Is there a
natural way to extend the theories and fill in the missing predictions?
Consider first the syntactic/semantic theory and factor (iii), i.e., the relative vs. interrog-
ative contrast. Following Starke (2001), Abels (2012) one could work on the assumption
that the larger (featurally more specific) a structure is, the more mobile it is. There are good
reasons to assume that relative pronouns are structurally more specific than interrogative
pronouns. It holds that a relative pronoun can be a structural superset of the interroga-
tive, but not conversely. A paradigmatic example is the Bulgarian relative pronoun kojto
‘which.REL’, which is based on the interrogative koj ‘which.INTER’ (Rudin 2009; see also
Mitrovi´
c 2016 for Slovenian or Daskalaki 2020 for Greek). To the best of our knowledge,
the opposite pattern is not documented. A syntactic (structure-based) analysis would there-
fore expect relative pronouns to be more mobile (to be able to escape wh-/adjunct-islands)
than interrogative ones.
Let us now turn to the discourse-based theory and factor (ii), i.e. the presence/absence
of a nominal layer on the adjunct. It seems natural to assume that if an adjunct is further
embedded in a nominal phrase (e.g. when. . . vs. the time when. . . ), it becomes more back-
grounded and therefore more difficult to extract from. This aligns with Erteschik-Shir and
Lappin’s (1979) assumptions, who consider headed relative clauses islands for extraction,
unless they occur in specific syntactic environments, as in pivots of existential construc-
tions.
4. Experiment
4.1 Questions and hypotheses
In our experiment, we investigated the naturalness of A0-extraction from clausal adjuncts
in Czech. Below we list the main research questions (Q) and the corresponding hypotheses
associated with the two types of theories under discussion: syntactic/semantic (Biskup and
Extraction from clausal adjuncts in Czech: A rating experiment
ˇ
Sim´
ık 2019, complemented by Starke 2001/Abels 2012-style approach to locality) and
discourse-based (Abeill´
e et al. 2020).
Q1 What is the difference between the extraction from left (peripheral) vs. right (central)
adjuncts?
Q2 What is the difference between the extraction of relative vs. interrogative pronouns
or phrases?
Q3 What is the difference between the extraction from clausal vs. complex nominal ad-
juncts?
The syntactic/semantic theory predicts that the only completely natural (grammatical)
extraction should be the one of relative pronouns from left clausal adjuncts without a nom-
inal layer. Extraction from right and/or nominally headed adjuncts is predicted to be unnat-
ural. Extraction of interrogative phrases could be marginally acceptable from left adjuncts
without a nominal layer, corresponding to a weak island violation.
The discourse-based theory predicts that the extraction of relative pronouns and inter-
rogative phrases from right adjuncts (without a nominal layer) should be at least as accept-
able as the extraction of relative pronouns from left adjuncts (without a nominal layer).
Extraction of interrogative phrases from left adjuncts is expected to be less natural, just as
extraction from nominally headed adjuncts.
4.2 Design and materials
In line with the above questions and hypotheses, we manipulated three variables in a fully
crossed 2 ×2×2 design: the POSITION of the adjunct (left vs. right), the EXTRACTE D ELE -
MENT (relative vs. interrogative), and the NOMI NA L layer on top of the adjunct (absent vs.
present). Table 1 provides schematic stimuli in all the conditions. Example (8) exemplifies
the actual stimuli (item 11), where (a)/(b) formed a sentence together with (i)/(ii).3
(8) a. rel
Zn´
am
know.1SG
p´
ısniˇ
cku,
song
kterou...
which.ACC
‘I know a song such that. . .
b. inter
Nev´
ım,
NE G.know.1SG
kterou
which
p´
ısniˇ
cku
song.ACC
‘I don’t know which song is such that.. .
(i) left+abs/pres
[{kdyˇ
z
when
/ ve
at
chv´
ıli,
time
kdy}
when
poslouch´
aˇ
s
listen.2SG
t], l´
epe
better
se
RE FL
soustˇ
red´
ıˇ
s.
concentrate.2SG
3All the materials and a preregistration can be accessed here: https://osf.io/zu9d4/?view_
only=e813eb862f4b4052ad55c2f91659048d.
ˇ
Sim´
ık & Biskup et al.
(ii) right+abs/pres
se
RE FL
l´
epe
better
soustˇ
red´
ıˇ
s
concentrate.2SG
[{kdyˇ
z
when
/ ve
at
chv´
ıli,
time
kdy}
when
poslouch´
aˇ
s
listen.2SG
t].
‘you can concentrate better when you listen to it.
stimulus (schematic) POSITION EXTRACTED NOMINAL
song which [when hear t] relax left rel abs
song which [at time when hear t] relax left rel pres
wonder which song [when hear t] relax left inter abs
wonder which song [at time when hear t] relax left inter pres
song which relax [when hear t]right rel abs
song which relax [at time when hear t]right rel pres
wonder which song relax [when hear t]right inter abs
wonder which song relax [at time when hear t]right inter pres
Table 1: Set of schematic stimuli for a single item
We created 48 experimental items and complemented them with 64 fillers (which con-
tained a number of smaller experiments manipulating factors that could potentially be rel-
evant or confounding for the critical manipulations; see discussion). Of the 48 items, all
involved the wh-word kter´
y‘which’, which is very common in both relatives and inter-
rogatives. In interrogatives, kter´
ywas accompanied by a nominal which corresponded to
the relative clause nominal head. Given the growing and converging evidence for the so-
called matching analysis of relative clauses, whereby the relative wh-word is followed by
an elided nominal under identity with the head, the interrogative and relative wh-clauses
could even be considered string-identical, differing only in the critical manipulation. Fi-
nally, we used embedded interrogatives, to minimize the difference between the relative
vs. interrogative condition – both in terms of embedding and the overall illocutionary force
(all stimuli were statements). Many other properties not central to our research questions
varied across the items, to cover more empirical ground and to make the stimuli less tiring
for the participant. The extracted element was mostly an object (33 items), sometimes an
adverbial (10 items) or a subject (5 items). It was mostly in singular (41 items). The ad-
junct was mostly a conditional or temporal clause, using a variety of complementizers, in
9 cases it was a purpose clause. The nominal used in the pres condition of the NOMINAL
factor was mostly a suitable noun (‘moment’, ‘case’, etc.) and sometimes a demonstrative
(‘this/that’). In the majority of cases (43 items) the clause modified by the adjunct clause
contained no pronominal element (or a subject pro) bound by the extracted wh-expression,
mainly to avoid a potential weak crossover effects in the right adjunct condition.
4.3 Procedure and participants
The experiment was coded, pseudo-randomized, distributed on lists using the Latin Square
design, and eventually administered with the help of the L-Rex software (Starschenko and
Extraction from clausal adjuncts in Czech: A rating experiment
Wierzba 2021). The participants were instructed to rate the naturalness of the stimuli (al-
ways a single complex sentence) on the scale 1 (completely unnatural) to 7 (completely
natural). Only the extremes of the scale were labeled. The procedure was exemplified by
the means of two examples – one natural, the other unnatural – where the naturalness
contrast was qualitatively unrelated to the main manipulations. Each participant rated 112
stimuli (48 experimental and 64 filler items). The pace was individual and most participants
completed the experiments in 20 to 30 minutes. 96 participants (non-experts) completed the
experiment, amounting to 576 ratings per each unique condition. They were recruited on
an individual basis by the coauthors of this paper and completed the experiment in their
homes at their own computers.
4.4 Results
The results are visualized by the bar plots in Figure 1. Each bar plot indicates the propor-
tional distribution of the individual ratings for one of the eight conditions (see section 4.2).
The darker the plot, the more natural the condition. The 50% line cuts through the median
rating and the tallest shade in each bar indicates the modus (the most frequent rating in the
condition). For instance, for the extraction of a relative pronoun from a left clausal adjunct
(nominal absent) – the top left bar – the median rating was 6 and modus 7. That means
that it was rated as very natural. Compare this with the corresponding extraction from right
adjunct (the bar next to it), where the median was 2 and modus 1. That condition was rated
as very unnatural.
We fitted a cumulative link mixed model with random intercepts for participants and
items, using the clmm function of the ordinal package (Christensen 2013) of R (R Core
Team 2021), to estimate the effects of the manipulated variables and their interactions on
the dependent variable (rating). All three variables were sum coded, allowing us to estimate
their main effects.
The main effects of all the variables as well as all their mutual interactions (two-way
and three-way) turned out significant. Extracting a relative pronoun is more natural than
extracting an interrogative phrase (z=23.125,p< .0001); extraction from a left adjunct is
more natural than from a right adjunct (z=38.204,p< .0001); finally, extracting from a
clause (abs) is more natural than from a complex nominal (pres) (z=12.874,p< .0001).
The three-way interaction (z=3.936,p< .0001), which in a way subsumes all the two-way
interactions, can be formulated as follows. Extraction from right adjuncts is very clearly
the least natural condition in all four subcases, with the modus or even median rating at 1.
How much better the extraction from the left adjunct is depends on the other two factors
in the following way: the naturalness increase is more pronounced for relative than for
interrogative elements and it is more pronounced if the nominal layer is absent than when
it is present.
5. Discussion and conclusion
Overall, the syntactic/semantic approach provides a better fit of the data than the discourse-
based approach. It correctly predicts that the extraction of relative pronouns from the left
ˇ
Sim´
ık & Biskup et al.
inter
abs
inter
pres
rel
abs
rel
pres
left right left right
0.00
0.25
0.50
0.75
1.00
0.00
0.25
0.50
0.75
1.00
position
proportion
rating
7
6
5
4
3
2
1
Figure 1: Proportions of naturalness ratings on the scale 1 = unnatural to 7 = natural
adjuncts without a nominal layer is the only completely natural condition (cf. the afore-
mentioend three-way interaction in the expected form). With the median of 6 (modus 7), it
is comparable to extraction from complement clauses (filler exp.; median 5, modus 6). The
prediction of the discourse-based approach, on the other hand, is not borne out. Extraction
from right adjuncts is generally rated as very unnatural (median 1 or 2, modus 1) and di-
rectly comparable to the extraction from another type of strong island (filler exp.; median
2, modus 1).4This is at odds with the prediction that the extraction from right adjuncts
should be at least as natural as the extraction of relative pronouns from left adjuncts.
It is good to notice that the effect of POSITION is very close to being categorical: the
rating of the right condition is at or very close to floor. This interpretation is supported by
the results of our filler experiments, where (i) extractions from another type of strong island
received comparable ratings and (ii) linearly long extractions from non-islands imposed
only a very minor penalty. The categorical effect can be considered to reflect the violation
of a grammatical constraint, in line with the syntactic/semantic analysis, and not just a
processing issue.
4The strong island used in the fillers was a complement clause headed by a demonstrative (e.g. ‘I’d like
to know which room1she claimed (*that.DE M) that.COMP her boss redecorated t1.’).
Extraction from clausal adjuncts in Czech: A rating experiment
Both approaches correctly capture the decreased naturalness of the extraction of in-
terrogative phrases (effect of E XTR AC TED element). The fact that this effect is less pro-
nounced than the effect of POSITION follows naturally from both approaches. The syntactic
one can resort to the notion of a weak island; the discourse-based one to processing issues.
Both approaches also correctly capture the decreased naturalness of extractions from
complex nominals (effect of NOMI NA L layer). What is unexpected for the syntactic ap-
proach is the relative naturalness of the extraction from overt complex nominals: extraction
of relative pronouns from left-positioned complex nominal adjuncts reaches the median
of 4 (modus 3), which is clearly a better rating than the observed extraction from right
adjuncts (median 2, modus 1), which Biskup and ˇ
Sim´
ık (2019) also consider to be com-
plex nominals. This is even more pressing in the light of a filler experiment, which showed
that left-positioned complex nominal adjuncts are less natural (than clausal ones) indepen-
dently of extraction. The syntactic/semantic analysis could in principle be upheld if what
mattered was not just the nominal nature, but the particular kind of nominal: while right
adjuncts are, by hypothesis, donkey-anaphoric definites (Hirsch 2016, ˇ
Sim´
ık 2018), many
of the overt nominals we used in our items might well have been interpreted as indefinite.
If definiteness matters for extraction (see e.g. Kluender 1998), we might have a handle on
the unexpected difference between extraction from left complex nominals and right clausal
adjuncts.
To conclude, we have demonstrated that wh-extraction from clausal adjunct is natu-
ral in Czech, adding to the recent literature (e.g. Bondevik et al. 2021), but only in very
specific conditions, namely when a relative pronoun is extracted from a left adjunct that
lacks an overt nominal layer, i.e., when it is (by assumption) not a complex nominal. This
overall naturalness profile is predicted by the syntactic/semantic approach of Biskup and
ˇ
Sim´
ık (2019), but not by the discourse-based approach of Abeill´
e et al. (2020). Our exper-
iment further revealed significant distinctions among the other conditions, which could all
be considered unnatural. The gradient character of these distinctions points to processing
rather than grammatical issues and is thus more in line with the discourse-based account.
More research is needed to disentangle the sources of the observed naturalness distinctions.
References
Abeill´
e, Anne, Barbara Hemforth, Elodie Winckel, and Edward Gibson. 2020. Extrac-
tion from subjects: Differences in acceptability depend on the discourse function of
the construction. Cognition 204:104293. URL https://doi.org/10.1016/j.
cognition.2020.104293.
Abels, Klaus. 2012. The Italian left periphery: A view from locality. Linguistic Inquiry
43:229–254. URL https://doi.org/10.1162/LING_a_00084.
Biskup, Petr. 2011. Adverbials and the phase model. Amsterdam: John Benjamins. URL
https://doi.org/10.1075/la.177.
Biskup, Petr, and Radek ˇ
Sim´
ık. 2019. Structure of conditional and (cor)relative clauses:
New evidence from locality. In NELS 49: Proceedings of the 49th Annual Meeting of
ˇ
Sim´
ık & Biskup et al.
the North East Linguistic Society, Volume 1, ed. by Maggie Baird and Jonathan Peset-
sky, 135–144. Amherst, MA: GLSA Publications.
Bittner, Maria. 2001. Topical referents for individuals and possibilities. In SALT 11: Pro-
ceedings from the 11th Conference on Semantics and Linguistic Theory, ed. by Rachel
Hastings, Brandon Jackson, and Zsofia Zvolenszky, 36–55. Ithaca, NY: Cornell Uni-
versity. URL https://doi.org/10.3765/salt.v11i0.2854.
Bondevik, Ingrid, Dave Kush, and Terje Lohndal. 2021. Variation in adjunct islands: The
case of Norwegian. Nordic Journal of Linguistics 44:1–32. URL https://doi.
org/10.1017/S0332586520000207.
Brasoveanu, Adrian. 2008. Uniqueness effects in correlatives. In Proceedings of Sinn und
Bedeutung 12, ed. by Atle Grønn, 47–65. Oslo: ILOS.
Chaves, Rui P. 2021. Island phenomena and related matters. In Head-Driven Phrase
Structure Grammar: The handbook, ed. by Stefan M¨
uller, Anne Abeill´
e, Robert D.
Borsley, and Jean-Pierre Koenig. Berlin: Language Science Press. URL https://
doi.org/10.5281/zenodo.5599846.
Christensen, Rune Haubo Bojesen. 2013. Ordinal: Regression models for ordinal data.
URL http://www.cran.r-project.org/package=ordinal/, r package
version 2013.9-30.
Chudobov´
a, Martina. 2011. Vˇ
etosled v souˇ
casn´
eˇ
ceˇ
stinˇ
e [Clause order in contemporary
Czech]. Master’s thesis, Charles University, Praha. URL http://hdl.handle.
net/20.500.11956/33727.
Cinque, Guglielmo. 1990. Types of A0-dependencies. Cambridge, MA: MIT Press.
Dal Farra, Chiara. 2019. Variability in adjunct islands: An experimental study in Italian.
In WCCFL 36: Proceedings of the 36th West Coast Conference on Formal Linguistics,
ed. by Richard Stockwell, Maura O’Leary, Zhongshi Xu, and Z. L. Zhou, 115–120.
Somerville, MA: Cascadilla Proceedings Project. URL http://www.lingref.
com/cpp/wccfl/36/paper3453.pdf.
Daneˇ
s, Frantiˇ
sek. 1964. A three-level approach to syntax. In Travaux Linguistiques de
Prague 1: L’ ´
Ecole de Prague d’aujourd’hui, ed. by Josef Vachek, 225–240. Praha:
´
Editions de l’Acad´
emie Tch´
ecoslovaque des Sciences.
Daskalaki, Evangelia. 2020. Types of relative pronouns. In Syntactic architecture and
its consequences II: Between syntax and morphology, ed. by Andr´
as B´
ar´
any, Theresa
Biberauer, Jamie Douglas, and Sten Vikner, 277–296. Berlin: Language Science Press.
URL https://doi.org/10.5281/zenodo.4280653.
Deane, Paul D. 1991. Limits to attention: A cognitive theory of island phenomena. Cog-
nitive Linguistics 2:1–63. URL https://doi.org/10.1515/cogl.1991.2.
1.1.
Erteschik-Shir, Nomi. 1973. On the nature of island constraints. Doctoral dissertation, MIT,
Cambridge, MA. URL http://dspace.mit.edu/handle/1721.1/7582.
Erteschik-Shir, Nomi. 2007. Information structure: The syntax-discourse interface. Ox-
ford: Oxford University Press.
Erteschik-Shir, Nomi, and Shalom Lappin. 1979. Dominance and the functional ex-
planation of island phenomena. Theoretical Linguistics 6:41–85. URL https:
//doi.org/10.1515/thli.1979.6.1-3.41.
Extraction from clausal adjuncts in Czech: A rating experiment
Firbas, Jan. 1992. Functional sentence perspective in written and spoken communication.
Cambridge: Cambridge University Press.
Goldberg, Adele. 2006. Constructions at work. Oxford: Oxford University Press.
Grosu, Alexander. 1981. Approaches to island phenomena. Amsterdam: North-Holland.
Haegeman, Liliane. 2003. Conditional clauses: External and internal syntax. Mind & Lan-
guage 18:317–339. URL https://doi.org/10.1111/1468-0017.00230.
Haegeman, Liliane. 2010. The internal syntax of adverbial clauses. Lingua 120:628–648.
URL https://doi.org/10.1016/j.lingua.2008.07.007.
Hall, David P., and Ivano Caponigro. 2011. On the semantics of temporal when-clauses. In
SALT 20: Proceedings from the 20th Conference on Semantics and Linguistic Theory,
ed. by Nan Li and David Lutz, 544–563. Ithaca, NY: CLC Publications. URL https:
//doi.org/10.3765/salt.v20i0.2566.
Hirsch, Aron. 2016. A compositional semantics for wh-ever free relatives. In Pro-
ceedings of Sinn und Bedeutung 20, ed. by Nadine Bade, Polina Berezovskaya,
and Anthea Sch¨
oller, 341–358. URL http://semanticsarchive.net/
sub2015/SeparateArticles/Hirsch-SuB20.pdf.
Huang, Cheng-Teh James. 1982. Logical relations in Chinese and the theory of grammars.
Doctoral dissertation, MIT, Cambridge, MA. URL http://dspace.mit.edu/
handle/1721.1/15215.
Kluender, Robert. 1998. On the distinction between strong and weak islands: A processing
perspective. In Syntax and semantics 29: The limits of syntax, ed. by Peter Culicover
and Louise McNally, 241–279. San Diego, CA: Academic Press. URL https://
doi.org/10.1163/9789004373167_010.
Kratzer, Angelika. 2012. Modals and conditionals: New and revised perspectives. Oxford:
Oxford University Press.
Kush, Dave, Terje Lohndal, and Jon Sprouse. 2019. On the island sensitivity of topical-
ization in Norwegian: An experimental investigation. Language 95:393–420. URL
https://doi.org/10.1353/lan.2019.0051.
Kuˇ
cerov´
a, Ivona. 2007. The syntax of givenness. Doctoral dissertation, MIT, Cambridge,
MA. URL http://hdl.handle.net/1721.1/39580.
Leˇ
snerov´
a, ˇ
S´
arka, and Karel Oliva. 2003. ˇ
Cesk´
a vztaˇ
zn´
a souvˇ
et´
ı s nestandardn´
ı strukturou.
Slovo a slovesnost 64:241–252. URL http://sas.ujc.cas.cz/archiv.
php?art=4129.
Mathesius, Vil´
em. 1941. Z´
akladn´
ı funkce poˇ
r´
adku slov v ˇ
ceˇ
stinˇ
e. Slovo a slovesnost 7:169–
180. URL http://sas.ujc.cas.cz/archiv.php?lang=en&art=413.
Mathesius, Vil´
em. 1947. ˇ
Ceˇ
stina a obecn´
y jazykozpyt. Praha: Melantrich.
Mitrovi´
c, Moreno. 2016. A relative syntax and semantics for Slovenian. In Formal studies
in Slovenian syntax: In honor of Janez Oreˇ
snik, ed. by Franc Maruˇ
siˇ
c and Rok ˇ
Zaucer,
221–252. Amsterdam: John Benjamins. URL https://doi.org/10.1075/la.
236.10mit.
M¨
uller, Gereon. 2010. On deriving the CED from the PIC. Linguistic Inquiry 41:35–82.
URL http://www.jstor.org/stable/40606828.
R Core Team. 2021. R: A language and environment for statistical computing. URL
https://www.r-project.org, vienna: R Foundation for Statistical Computing.
ˇ
Sim´
ık & Biskup et al.
Rawlins, Kyle. 2013. (Un)conditionals. Natural Language Semantics 40:111–178. URL
https://doi.org/10.1007/s11050-012-9087-0.
Rudin, Catherine. 2009. The Bulgarian relative marker -to. In A linguist’s linguist: Studies
in South Slavic linguistics in honor of E. Wayles Browne, ed. by Steven Franks, Vrinda
Chidambaram, and Brian Joseph, 403–422. Bloomington, IN: Slavica Publishers.
Schlenker, Philippe. 2004. Conditionals as definite descriptions (A referential analysis).
Research on Language and Computation 2:417–462. URL https://doi.org/
10.1007/s11168-004-0908-2.
Sgall, Petr, Eva Hajiˇ
cov´
a, and Jarmila Panevov´
a. 1986. The meaning of the sentence and
its semantic and pragmatic aspects. Dordrecht: Reidel.
ˇ
Sim´
ık, Radek. 2018. Ever free relatives crosslinguistically. In Proceedings of Sinn und
Bedeutung 22, ed. by Uli Sauerland and Stephanie Solt, volume 2, 375–392. Berlin:
Leibniz-Center General Linguistics. URL https://ojs.ub.uni-konstanz.
de/sub/index.php/sub/article/view/112.
ˇ
Sim´
ık, Radek, and Marta Wierzba. 2015. The role of givenness, presupposition, and
prosody in Czech word order: An experimental study. Semantics & Pragmatics 8:1–
103. URL https://doi.org/10.3765/sp.8.3.
ˇ
Sim´
ık, Radek, and Marta Wierzba. 2017. Expression of information structure in West
Slavic: Modeling the impact of prosodic and word-order factors. Language 93:671–
709. URL https://doi.org/10.1353/lan.2017.0040.
Starke, Michal. 2001. Move dissolves into merge: A theory of locality. Doctoral disserta-
tion, University of Geneva, Geneva. URL http://ling.auf.net/lingbuzz/
000002.
Starschenko, Alexej, and Marta Wierzba. 2021. L-Rex (beta). URL https://github.
com/2e2a/l-rex/, computer software.
Radek ˇ
Sim´
ık, Petr Biskup, Kateˇ
rina Bartasov´
a, Mark´
eta Danˇ
cov´
a, Eliˇ
ska Dost´
alkov´
a, Kateˇ
rina
Hrdinkov´
a, Gabriela Koskov´
a, Jarom´
ır Koz´
ak, Kl´
ara Lupomˇ
esk´
a, Albert Marˇ
s´
ık, Edita
Schejbalov´
a & Illia Yekimov
radek.simik@ff.cuni.cz, biskup@rz.uni-leipzig.de
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Finite adjunct clauses are often assumed to be among the strongest islands for filler–gap dependency creation cross-linguistically, but Kush, Lohndal & Sprouse (2019) found experimental evidence suggesting that finite conditional om -adjunct clauses are not islands for topicalization in Norwegian. To investigate the generality of these findings, we ran three acceptability judgment experiments testing topicalization out of three adjunct clause types: om ‘if’, når ‘when’ and fordi ‘because’ in Norwegian. Largely replicating Kush et al. (2019), we find evidence for the absence of strong island effects with topicalization from om -adjuncts in all three experiments. We find island effects for når - and fordi -adjuncts, but the size of the effects and the underlying judgment distributions that produce those effects differ greatly by island type. Our results suggest that the syntactic category ‘adjunct’ may not constitute a suitably fine-grained grouping to explain variation in island effects.
Conference Paper
Full-text available
Based on a sample of seven languages, I show that the so called modal inferences in ever free relatives (ignorance and indifference) are not universally available. The primary reading of eFRs crosslinguistically turns out to be a "non-modal" one, which is available to all languages under investigation. The implication is that if there is a modal inference triggered by the use of the ever-morpheme in FRs, the inference is likely to have a source external to the eFR (Lauer 2009, Condoravdi 2015, Hirsch 2016). In line with this conclusion, I propose to generalize Hirsch's (2016) analysis of ignorance eFRs, suggesting that all eFRs, no matter what their interpretation ends up being, are instances of (un)conditionals + donkey-anaphoric definite descriptions.
Article
Full-text available
The received wisdom is that word order alternations in Slavic languages arise as a direct consequence of word order-related information structure constraints such as 'place given expressions before new ones'. In this paper, we compare the word order hypothesis with a competing one, according to which word order alternations arise as a consequence of a prosodic constraint – 'avoid stress on given expressions'. Based on novel experimental and modeling data, we conclude that the prosodic hypothesis is more adequate than the word order hypothesis. Yet, we also show that combining the strengths of both hypotheses provides the best fit for the data. Methodologically, our paper is based on gradient acceptability judgments and multiple regression, which allows us to evaluate whether violations of generalizations like 'given precedes new' or 'given lacks stress' lead to a consistent decrease in acceptability and to quantify the size of their respective effects. Focusing on the empirical adequacy of such generalizations rather than specific theoretical implementations also makes it possible to bridge the gap between different linguistic traditions, and to directly compare predictions emerging from formal and functional approaches.
Article
In order to explain the unacceptability of certain long-distance dependencies – termed syntactic islands by Ross (1967) – syntacticians proposed constraints on long-distance dependencies which are universal and purely syntactic and thus not dependent on the meaning of the construction (Chomsky, 1977; Chomsky, 1995 a.o.). This predicts that these constraints should hold across constructions and languages. In this paper, we investigate the “subject island” constraint across constructions in English and French, a constraint that blocks extraction out of subjects. In particular, we compare extraction out of nominal subjects with extraction out of nominal objects, in relative clauses and wh-questions, using similar materials across constructions and languages. Contrary to the syntactic accounts, we find that unacceptable extractions from subjects involve (a) extraction in wh-questions (in both languages); or (b) preposition stranding (in English). But the extraction of a whole prepositional phrase from subjects in a relative clause, in both languages, is as good or better than a similar extraction from objects. Following Erteschik-Shir (1973) and Kuno (1987) among others, we propose a theory that takes into account the discourse status of the extracted element in the construction at hand: the extracted element is a focus (corresponding to new information) in wh-questions, but not in relative clauses. The focus status conflicts with the non-focal status of a subject (usually given or discourse-old). These results suggest that most previous discussions of islands may rely on the wrong premise that all extraction types behave alike. Once different extraction types are recognized as different constructions (Croft, 2001; Ginzburg & Sag, 2000; Goldberg, 2006; Sag, 2010), with their own discourse functions, one can explain different extraction patterns depending on the construction.