How logical and pragmatic inferences
determine acceptability judgements
FDSL 12.5, Nova Gorica, 7.12.2017
Mojmír Dočekal
Intro
1/79
N-words vs. NPIs
• many languages distinguish
1) syntactically negative dependentent expressions: n-words
2) semantically negative dependent expressions: N(egative)
P(olarity) I(tems)
(1) a. Nimeni
n-person
nu
not
a
has
venit.
come
‘Nobody came.’
b. *Vreun
NPI
student
student
nu
not
a
has
venit.
come
‘*Any student didn’t come.’ [Ro]
2/79
(2) a. Dhen
not
idhe
saw
kanenan
NPI-person
o
the
Janis.
John
‘John didn’t see anybody’
b. Dhen
not
idhe
saw
KANENAN
n-person
o
the
Janis.
John
‘John didn’t see anybody at all.’ [Gr]
• agreed criterion (recently Giannakidou and Zeijlstra (2017)):
(3) X qualiﬁes as an n-word iﬀ:
a. X can be used with structures with sentential negation or
other X with meaning equivalent to one ¬
b. X provides a negative fragment answer.
3/79
Outline
joint work with Jakub Dotlačil
• slides: http://bit.ly/2AYE6Yl
1) NPIs vs. n-words: theory
• n-words
• NPIs: types
• Czech
2) Experimental evidence
• NR
• Fragment answers
• Likelihood
4/79
NPIs vs. n-words: theory
5/79
Sorting the data
• Germanic negative quantiﬁers ∈ n-words:
(4) a. John didn’t see nobody. [En]
¬∃x[Person (x) ∧ ¬See (John, x)]
b. John nikoho neviděl. [Cz]
¬∃x[Person (x) ∧ See (John, x)]
• NPIs = n-words:
(5) a. Whom did you talk to?
b. *Anybody.
6/79
n-words:
1) non-negative indeﬁnites (predicate at type 〈e, t〉) plus rooﬁng
requirement (Ladusaw (1992), Giannakidou (1997) a.o.)
2) n-words as agreement markers ↔ locality, licensed in syntax
(Zeijlstra (2004), Zeijlstra (2008))
• n-words (unlike indeﬁnites or NPIs) need local negation
(Giannakidou and Zeijlstra (2017)):
(6) a. Dhen
Not
prodhosa
betrayed.1sg
mistika
secrets
[pu
that
eksethesan
exposed.3pl
[kanenan/*KANENAN]]
anybody/n-body
‘I didn’t revela secrets that exposed anybody.’
7/79
• n-words in Slavic languages: locality is very strict (Progovac
(1993)), unlike in Spanish, Italian, Greek
(7) a. *Petr
Petr
neřekl,
neg.said
že
that
nikdo
n-body
přišel.
came
‘Petr didn’t say that anybody came.’
b. *Petr
Petr
nechce,
neg.wants
aby
C.subj
tu
here
nikdo
n-body
byl.
were
‘Petr doesn’t want anybody were here.’
8/79
NPIs
• any – Negative Polarity Item (NPI)
(8) *Peter visited anyone.
(9) Petr didn’t visit anyone.
a. ¬∃x[Person (x) ∧ Visit (Peter, x)]
9/79
• NPI licensing expressions share the property of reversing the
direction of entailment in their argument
• negation reverses entailment:
p q (p ∧ q) → (p ∨ q) ¬(p ∨ q) → ¬(p ∧ q)
1 1 1 1
0 1 1 1
1 0 1 1
0 0 1 1
10/79
• natural language example:
(10) red wine → wine
a. John likes red wine. → John likes wine.
b. John doesn’t like red wine. → John doesn’t like wine.
c. John doesn’t like wine. → John doesn’t like red wine.
• general term: Downward Entailing (DE)
(11) Fauconnier-Ladusaw’s Licensing Condition: An NPI is only
grammatical if it is in the scope of an α such that α is DE.
11/79
• upward monotonic quantiﬁer: subsets → supersets
• downward monotonic quantiﬁer: supersets → subsets
(12) a. det A: upward entailing iﬀ for any B, C (B ⊆ C)
Det A B ⇒ Det A C
b. det A: downward entailing iﬀ for any B, C (B ⊆ C)
Det A C ⇒ Det A B
c. if not upward or downward monotonic →
non-monotonic
12/79
(13) Upward/Downward entailing and Non-monotonic
determiners:
a. some: Some toys are blue ⇒ Some toys are colored
b. few: Few toys are colored ⇒ Few toys are blue
c. exactly n: Exactly three toys are blue ↔ Exactly three
toys are colored
13/79
• monotonicity properties of a position in a sentence are
computed compositionally:
(14) a. [↓ At most three detectives arrested ↓[fewer than four
↑[criminals]]]
b. ⇒[↓ At most three detectives arrested ↓[fewer than
four ↑[humans]]]
14/79
Weak NPIs
• Downward Entailing scope/environment
• weak NPIs: any, ever, ...
(15) a. Bill didn’t ever say anything.
b. No student ever said anything.
c. Few students ever said anything.
d. At most 5 students ever said anything.
e. *Between 5 and 10 students ever said anything.
f. *Some/*all/*most students ever said anything.
15/79
Strong NPIs
• in weeks, additive either, and punctual until
(16) a. Bill didn’t leave until his birthday.
b. No student left until his birthday.
c. ??Few students left until their birthdays.
d. *At most 5 students left until their birthdays.
e. *Between 5 and 10 students left until their birthdays.
f. *Some/*most/*all students left until their birthdays.
16/79
(17) Anti-additive function: F(x ∨ y) ↔ F(x) ∧ F(y)
(18) No student smokes or drinks ↔ No student smokes and no
student drinks.
(19) Few students smoke or drink ↔ Few students smoke and
few students drink
• anti-additivity as necessary for Strong NPIs: Zwarts (1998)
• popular alternative epxlanation: Gajewski (2011)
17/79
NPIs vs. n-words: modularity
• n-words vs. NPIs: syntax (agreement) vs. semantics
(monotonicity, ...)
• goes well with modularity: distinguishing diﬀerent forms of
ill-formedness (syntactic, semantic, ...)
• logical properties correlating with syntactic acceptability (NPIs)
→ linking the domains
• some theories (Heim/Crnič) of NPIs licensing: via presupposition
(→ linking pragmatics and syntactic acceptability)
18/79
Czech: “= 0” vs. “< 1”
• in Czech: two candidates both for NPI and n-word status:
(20) a. Petr
Petr
neviděl
neg.saw
ani
even
jednoho/žádného
one/any
studenta.
student
‘Petr didn’t see any student.’
b. *Ani jeden/*žádný student přišel.
Not even one/any student came.
c. Petr neslyšel, že *ani jeden/*žádný student přišel.
Petr didn’t hear that even one/any student came.
19/79
• the meaning (natural numbers) at ﬁrst sight identical
(21) a. ani jeden N ≈ #( N ) < 1
b. žádný N ≈ #( N ) = 0/¬( N )
• all four possibilities reasonable, Czech tradition (Havránek and
others (1960)):
item/status NPIs n-words
ani jeden X
žádný X
(22) Research question: do strict neg-concord languages even
allow grammaticalization of Strong NPIs?
20/79
Experimental evidence
21/79
Neg-raising
• long distance licencing should be possible for NR (unlike for
non-NR) in case of NPIs
• n-words in Slavic languages obey strict locality (syntax)
• predictions:
environment/status NPIs n-words
NR embedded X
non NR embedded X X
• experiment 1: Dočekal and Dotlačil (2016)
22/79
(23) a. Ztratila
Lost
se
SE
ani
not-even
jedna
one
ovce.
sheep
‘A single sheep is missing.’
b. Neztratila
neg-lost
se
SE
ani
not-even
jedna
one
ovce.
sheep
‘Not a single sheep is missing.’
c. Nový bača v Tatrách nechce, aby se ztratila ani jedna
ovce.
new shepherd in Tatras neg-wants C SE lost not-even
one sheep.
d. Nový bača v Tatrách si nemyslí, že se ztratila ani jedna
ovce.
new shepherd in Tatras SI neg-think C SE lost not-even
one sheep
e. Nový bača v Tatrách neříká, že se ztratila ani jedna ovce.
new shepherd in Tatras neg-say C SE lost not-even one
sheep
23/79
• 5 environments
(A) a positive sentence (A)
(B) a negative sentence (B)
(C) a clause embedded under negated NR predicates of intention
and judgement/obligation (e.g. want, advise) (C)
(D) a clause embedded under negated NR predicates of opinion
(believe) (D)
(E) non-NR predicates (E)
• only ani jeden
24/79
Figure 1: Experiment 1
25/79
• the scalar approach to NR (Horn (1973), Romoli (2012), Romoli
(2013))
• NR predicates (beside the assertion – (24-a)) contribute the
excluded middle (EM) to the semantic composition ((24-b)):
• alternatives are exhaustiﬁed by EXH – (25)
(24) a. P = λpλx. x[p]
b. Alt(NR) = {λpλx. x[p], λpλx.[ x[p] ∨ x[¬p]]}
(25) EXH(Alt(p))(p)(w) = p(w)∧∀q ∈ Excl(p, Alt(p))[¬q(w)]
26/79
(26) ‘A new shepherd in Tatra mountains doesn’t want even one
sheep to be missing.’ ¬wants[p] .
(27) a. Alt(¬wants[p]) =
{¬wants[p], ¬(wants[p] ∨ wants[¬p])}
b. EXH (¬wants[p]) = ¬wants[p] ∧ ¬¬(wants[p] ∨
wants[¬p]) |= wants[¬p]
• consequence of exhaustiﬁcation of NR: negation is interpreted
as having low scope (semantically)
27/79
• recall: strong NPIs are licensed by anti-additive functions
(28) a. It didn’t rain and it didn’t snow.
b. It didn’t rain or snow.
c. ¬p ∧ ¬q
d. ¬[p ∨ q]
28/79
• the same for NR predicates (like want)
(29) a. Susan does not want to sleep and she does not want to
dance.
b. Susan does not want to sleep or dance.
c. ¬p ∧ ¬q ↔
d. ¬(p ∨ q)
world/proposition p q
w1 0 0
w2 0 0
29/79
• not NR predicates (like say): (30-b) does not follow from (30-a)
• not-NR are not antiadditive and not able to license strong NPIs
(30) a. Susan didn’t say that she will sleep and she didn’t say
that she will dance.
b. Susan didn’t say that she will sleep or dance.
(31) a. ¬ p ∧ ¬ q (true in the table)
b. ¬ [p ∨ q] (false in the table)
world/proposition p q
w1 0 1
w2 1 0
30/79
• initial predictions
environment/status NPIs n-words
NR embedded X
non NR embedded X X
• experimental support of clasiﬀying ani jeden as strong NPI
31/79
Fragment answers
• the distinction between n-words and Strong NPIs
• in the experiment 3 we observed negative interaction of ani and
ellipsis in non-negative questions:
(32) Kdo odešel z hospody?
who left from pub?
a. Žádný student.
n-ADJ student
b. ??Ani jeden student.
NPI one student
32/79
Figure 2: Experiment 3
33/79
• but in Experiment 4 (more context informations) the correlation
disappeared:
(33) Koho vyhodil profesor Palný včera ze zkoušky?
whom ﬁred prof Palný yesterday from exam?
a. Žádného studenta.
n-ADJ student
b. Ani jednoho studenta.
NPI one student
34/79
Figure 3: Experiment 4 35/79
• usually taken as the standard test of being n-word (vs. NPIs)
• Fălăus, and Nicolae (2016) observes that in strict neg-concord
languages n-word answers to negative questions can have
(surprisingly) Double Negation (DN) reading:
• against the n-words vs. NPIs criterion (two negations)
(34) Kdo nepřišel na party? Nikdo.
who neg.came to party n-person
a. NC-reading: ¬∃x[Person (x) ∧ ComeParty (x)] ≡
∀x[Person (x) → ¬ComeParty (x)]
b. DN-reading: ¬∃x[Person (x) ∧ ¬ComeParty (x)] ≡
∀x[Person (x) → ComeParty (x)]
36/79
• Slavic speakers (Czech, Slovak, Russian) judgments conﬁrm this
(small survey: 10 students)
• slight preference (7/10) for NC reading
• new data: DN is strengthened by pressence of other n-word(s)
(35) Kdo nepřečetl žádný článek? Nikdo.
who neg.read n-ADJ article n-person
a. NC (2/10):
∀x[Person (x) → ¬∃y[Article (y) ∧ Read (x, y)]]
b. DN (8/10):
∀x[Person (x) → ∃y[Article (y) ∧ Read (x, y)]]
37/79
• the same pattern is observed even in aﬃrmative sentences
(new data in Slavic languages):
(36) Nikdo ničemu nevěří.
n-person n-thing neg.believes
a. NC:
∀x[Person (x) → ¬∃[Entity (y) ∧ Believes (x, y)]]
b. *DN: ∀x[Person (x) → ∃[Entity (y) ∧ Believes (x, y)]]
(37) V nic nikdo nevěří.
in n-thing n-person believes
a. NC (0/10):
∀x[Person (x) → ¬∃[Entity (y) ∧ Believes (x, y)]]
b. DN (10/10):
∀x[Person (x) → ∃[Entity (y) ∧ Believes (x, y)]]
38/79
• similarly:
(38) Nikdo při té zkoušce nic nenapsal.
n-person at the exam n-thing neg.wrote
a. NC (10/10): ∀[...¬∃...]
b. DN (0/10): ∀[...∃...]
(39) Nic při té zkoušce nikdo nenapsal.
n-thing at the exam n-person neg.wrote
a. NC (0/10): ∀[...¬∃...]
b. DN (10/10): ∀[...∃...]
39/79
• assumption (after Fălăus, and Nicolae (2016)): such data →
evidence for a focus position in the left-periphery of Slavic
clause where n-words can license Covert Negation (CN)
• constraints: presence of another n-word(s) plus in same cases
ellpisis (but not the sine qua non condition)
→ double negation reading: CN = ¬ + SN = ¬
(40)
CN
FOC
n-word
TP
NegP
SN
VP
V n-word
40/79
• inconclusive evidence: both from experiments and DN data
• some other factor: left peripheral position licensing second
negation but parasitic on SN?
• have to be constrained:
(41) *V nic Petr věří.
in n-thing Petr believes
• plan: investigate DN readings w.r.t. n-words/Strong NPIs
• prediction: n-words should be preferred
41/79
Likelihood scenarios
• n-words vs. NPIs w.r.t. semantic properties
• predictions:
1) n-words (syntax) shouldn’t be sensitive to logical properties of
their environment (just sentential/verbal negation)
2) NPIs licensed in semantics by deﬁnition are
42/79
• very inﬂuental current theory of NPI licensing – simple even
hypothesis of NPI licensing (Heim (1984), Krifka (1995), Crnič
(2014)):
– NPIs associate with covert even
– NPIs (and focus) generate sets of possible alternatives
– covert even associates with the alternatives and generates
presupposition of its prejacent being the least probable member
of the set of alternatives
• predictions of Heim/Crnič theory: NPIs should be sensitive to
probability
property/item probability
n-words *
NPIs
43/79
• we tested exactly this prediction in Experiment 3 and
Experiment 4
• in both we found strong correlation of ani and probability
• side-note: all natural examples of ani: most likely situations
(ČNK) in the un-negated form and least probable in the negated
form:
(42) tento nyní úspěšný podnikatel [...] v prvním měsíci neměl ani
jednoho zákazníka
this now very succesfull businesman [...] in ﬁrst month didn’t
have [NPI one customer]
44/79
• Experiment 3: ani/žádný in likely (negated) sentences
(43) (...) nestal se ani/žádným kardinálem
‘He didn’t become even a cardinal.’
• people strongly preferred žádný (n-word)
• it doesn’t clash with the least likely presupposition of ani
45/79
Figure 4: Experiment 3
46/79
• Experiment 4: ellaboration
• truth value judgment task
• example item (B.A. pass the exam: most unlikely, B.A. fail the
exam most likely – negated ne-složili):
(44) Scenario: prof. Novák yesterday examined an easy course
which B.A., M.A. and Ph.D. students attend. Ph.D. students
pass the exam always, M.A. in most cases but B.A. only
seldomly.
a. Včerejší zkoušku u prof. Nováka nesložili ani/žádní
bakaláři.
yesterday exam at prof. Novák neg.passed NPI/n-Adj
B.A.-students
47/79
Figure 5: Experiment 4 48/79
• and again we found strong preference for žádný (n-word) in
most likely scenarios (collision with least likely ani (NPI)
presupposition)
• conclusion:
property/item probability
žádný *
ani
49/79
• theoretical explanation (least probability):
(45) even w(C)(p) is deﬁned only if ∀q ∈ C[q = p → q >likely p]
• scope of covert even over ani (plus anti-additivity requirement):
(46) [even C] [↓ ¬ [↑ ... ani ...]]
(47) [even C] [↓ ¬ [↑ ... ani BA passed ...]]
50/79
Correlations and dialectal variation
• the evidence to consider ani a strong NPI is very limited
– the only positive evidence: it might appear under negated NR
predicates
• Question 2:
(48) Could speakers diﬀer w.r.t. their categorisation of ani?
51/79
• Experiment 3: we checked whether ani correlates with
Likelihood/NR conditions
• some speakers might accept likelihood if ani is an n-word for
them;
• the same speakers should reject NR with ani
• a nearly signiﬁcant negative correlation between
such-constructed Likelihood and NR (t = −1.9, p = .065):
dialectal variation condition preference
correlation A NR ani ≈ n-words
correlation A likelihood ani ≈ n-words
correlation B NR ani > n-words
correlation B likelihood ani < n-words
• correlation B: ani strong NPI, correlation A: ani n-word 52/79
Experiment 4:
• more ellaborated context (B.A. vs M.A. & Ph.D.)
• scales: logically stronger → least likely (Kolmogorov)
• plus logically independent but contextually manipulated
propositions: restaurant and customers (tourists >likely
bureaucrats >likely students): clash of most likely in the context
plus ani least likely presupposition
(49) nenavštívili ani studenti
neg.visited ani students
• a strong eﬀect: negative correlation of (z-transformed) ani
acceptability in NR with likelihood (t=-3.2, p=.003)
53/79
54/79
• points: mean value of each subject answers
• blue line: regression line
• NR as a function of Likelihood minimizing residuals
• subjects either:
– observed ani presupposition and allowed NR with ani
– allowed presupposition failure and didn’t accept NR with ani
– (no top right corner): no subject allowed presupposition failure
of ani and allowed NR with ani
• stronger eﬀect in experiment 4 than in experiment 3
55/79
Correlations and dialectal variation summary
1) Speakers more aware of likelihood presupposition of ani prefer
it to n-words in NR (negative correlation of likelihood and NR)
2) Speakers more relaxed with ani presupposition failure do not
diﬀerentiate it from n-words
3) Strong NPIs are licensed semantically via likelihood
presupposition
56/79
Summary
• n-words and strong NPIs co-exist in natural language
• even strict neg-concord languages distinguish Strong NPIs and
n-words
• n-words are licensed syntactically
• NPIs in semantics/pragmatics: antiaditivity + likelihood
presupposition (direct evidence for Heim/Crnič type of theory)
• the domains are more linked then strict modularity predicts
• such subtle date can only be dealt with experimental methods
57/79
Thanks!
58/79
Appendix
59/79
Experiment 1 (NR)
• 40 exp. items in part 1 and 20 exp. items in part 2: 60 tested
sentences
• each part 30 ﬁllers, 60 Czech native speakers, ≈ 1 hour
• the experiment online in Ibex: link
60/79
Results of acceptability task
• all participants passed control ﬁllers (uncontroversially
grammatical/ungrammatical)
• acceptability task: modeled by mixed-eﬀects ordered probit
regression
• Condition C as the reference level
• negated sentences, Condition B, were judged as better than
NRs (β = 1.84, z = 23, p < .001)
• positive sentences, Condition A, were judged as worse than NRs
(β = −1.1, z = −15, p < .001)
• sentences with negated non-NR (E) predicates worse than any
NR (β = −0.65, z = −9, p < .001)
• evidence for:
1) treating ani ‘not even’ and až do ‘until’ as a strict NPIs
2) Czech has a class of NR verbs.
61/79
Experiment 2
Mixed-eﬀects probit models to analyze the data with mood
(subjunctive vs. indicative), predicate type (opinion, probability,
communication) and their interaction as ﬁxed eﬀects
• NR predicates ((55) and (56)) judged as signiﬁcantly better than
non-NR communication predicates (57) – z = -2.51, p = 0.012
• no diﬀerence between opinion and probability NR predicates
• subjunctive mood better than indicative mood (z = 2.39, p =
0.017)
• strict NPIs (ani jeden ‘not even one’) judged as worse than HSEs
(až do ‘until’) with NR predicates (z = 2.65, p = 0.008)
62/79
Post-hoc analysis of two types of strict NPIs
• ‘even one’ was fully acceptable in sentences with clause-mate
negation but degraded with negated NR predicates
(β = −4.7, z = 10.4, p < .001) – (50-a) (one of the items in two
conditions)
• with non-NR predicates ((50-b)) ‘even one’ was judged as worse
than with NR predicates (β = −1.1, z = 5.7, p < .001).
(50) a. Náš
our
nový
new
knihovník
librarian
si
SE
nepřeje,
neg-wishes
aby
C
zmizela
lost
ani
even
jedna
one
kniha.
book
‘Our new librarian doesn’t wish even one book to be
missing.’
b. Náš nový knihovník neslyšel, že zmizela ani jedna kniha.
‘Our new librarian didn’t hear that even one book was
lost.’
63/79
• ‘until’ behaved strangely diﬀerently: signiﬁcantly less
acceptable in sentences with clause-mate negation compared
to ‘even one’ (β = −3.2, z = −6.4, p < .001)
• even more surprisingly ‘until’ was more acceptable than ‘even
one’ with NR predicates (β = 0.6, z = 2.6, p < .01) – (51-a)
64/79
Experiment 3
• We analyzed the data in a mixed-eﬀects linear model with
subject and item intercept+slope random eﬀects.
• The dependent variable was a by-subject z-transformed
response.
• The independent variables were environment (ref-level:
without), expression, (ref-level: žádný) and their interaction.
65/79
• The model also revealed a negative interaction of ani by ellipsis
and likelihood (t = −2.6, p < .05, t = −4.7, p < .001)
• The interactions show that ani is worse in ellipses/the likelihood
constructions than n-words, but it is better under NR predicates
than n-words
schematic structure preference condition
A: Subj V? B: ... n-word > ani jeden ellipsis
Subj V ... n-word > ani jeden likelihood
• We found a negative main eﬀect of NR (t = −4.1, p < .001):
n-words in NR are signiﬁcantly worse than in the reference level
condition, without
66/79
• and a positive interaction of ani by nr (t = 2.4, p < .05): ani
(jeden) is far more acceptable than n-words in:
schematic structure preference condition
[Subj ¬NR-V [emb-V ...] ani jeden > n-word NR
67/79
Summary
• The results taken together strongly support the position that:
1) ani (jeden) ‘even (one)’ is a strong NPI and not an n-word
2) strong NPIs co-exist with n-words in Czech.
68/79
Predicative position
(51) a. Petr není žádný lékař.
b. *Petr není ani jeden lékař.
• ¬doctor(petr)
(52) Petr a Marie jsou 2 lékaři.
X = ∗doctor ∧ #(X) = 2 ∧ X(Petr Marie)
69/79
• ani jeden vs. žádný in predicative (and possibly idiomatic)
positions:
(53) *Petr není ani jeden lékař.
X = ∗doctor ∧ #(X) < 1 ∧ X(Petr)
(54) Petr nekupoval žádného/???ani jednoho zajíce v pytli.
a. ¬BuyRabitInSack(Petr)
b. ¬K(P, X) ∧ X = RabitInSack ∧ #(X) < 1
70/79
Strong NPIs are not one homogenous class
• ani jeden vs. až do
• až do better unlike ani jeden acceptability of sentences with
strict NPIs, ani jeden ‘even one’ and HSE až do ‘until’ + time
expression
• three predicate types: (55) opinion class of NRs, (56) probability
class of NRs, (57) non-NR communication predicates
• each environment was varied for the mood of the predicate in
the embedded clause (indicative mood vs. subjunctive mood)
• 36 exp. items in 2x3 (=6) conditions + 36 ﬁllers
71/79
(55) Nemyslím,
do-not-think-I
že
that
0/by
IND/SUBJ
ani
even
jeden
one
z
of
běžců
runners
může/mohl
can/could
ten
the
závod
race
vyhrát.
win
‘I don’t think that even one of the runners can/could win the
race.’
(56) Není
it’s-not
možné,
possible
že
that
0/by
IND/SUBJ
ani
even
jeden
one
z
of
běžců
runners
může/mohl
can/could
ten
the
závod
race
vyhrát.
win
‘It’s not possible that even one of the runners can/could win
the race.’
(57) Netvrdím,
do-not-say-I
že
that
0/by
IND/SUBJ
ani
even
jeden
one
z
of
běžců
runners
může/mohl
can/could
ten
the
závod
race
vyhrát.
win 72/79
• ‘even one’ behaved like a strict NPI: fully licensed by the
clause-mate negation
• licensed by a non-local NR negation but hardly acceptable with
a non-local non-NR negation (not anti-additive)
• ‘until’ behaved strangely diﬀerently: signiﬁcantly less acceptable
in sentences with clause-mate negation compared to ‘even one’
73/79
74/79
Conclusion:
• ani jeden is strong NPI
• až do aspectually sensitive expression which requires its local
predicate to be homogeneous
– most frequent occurrences of Czech ‘until’ in the SYN2010
corpus: upward entailing with imperfective verbs
75/79
Experiment 4
• Cumulative Link Mixed Model ﬁtted with the Laplace
approximation
formula: as.factor(Answer) ~ Condition + (1 | Subject) + (1 | Item)
data: datalik
link threshold nobs logLik AIC niter max.grad cond.H
probit flexible 320 -403.53 821.07 459(1422) 5.97e-04 1.3e+02
Random effects:
Groups Name Variance Std.Dev.
Subject (Intercept) 0.3482 0.5901
Item (Intercept) 0.1229 0.3505
Number of groups: Subject 40, Item 16
Coefficients:
Estimate Std. Error z value Pr(>|z|)
Conditionlik-a -1.6706 0.2313 -7.221 5.14e-13 ***
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
76/79
References I
Crnič, Luka. 2014. “Against a Dogma on Npi Licensing.” The Art and Craft of Semantics: A
Festschrift for Irene Heim 1:117–45.
Dočekal, Mojmír, and Jakub Dotlačil. 2016. “Experimental Evidence for Neg-Raising in Slavic.”
Linguistica 56 (1). University of Ljubljana, Faculty of Arts:93.
Fălăus, , Anamaria, and Andreea Nicolae. 2016. “Fragment Answers and Double Negation in
Strict Negative Concord Languages.” In Semantics and Linguistic Theory, 26:584–600.
Gajewski, Jon R. 2011. “Licensing Strong Npis.” Natural Language Semantics 19 (2).
Springer:109–48.
Giannakidou, Anastasia. 1997. The Landscape of Polarity Items. Rijksuniversiteit.
Giannakidou, Anastasia, and Hedde Zeijlstra. 2017. “The Landscape of Negative
Dependencies: Negative Concord and N-Words.” Linguistics Companion, 2nd Edition. New
York: Blackwell.
Havránek, Bohuslav, and others. 1960. Slovník Spisovného Jazyka českého. Nakl.
Československé akademie věd.
77/79
References II
Heim, Irene. 1984. “A Note on Negative Polarity and Downward Entailingness.” In
Proceedings of Nels, 14:98–107. 1983.
Horn, Laurence Robert. 1973. “On the Semantic Properties of Logical Operators in English.”
Krifka, Manfred. 1995. “The Semantics and Pragmatics of Polarity Items.” Linguistic Analysis
25 (3-4):209–57.
Ladusaw, William A. 1992. “Expressing Negation.” In Semantics and Linguistic Theory,
2:237–60.
Progovac, Ljiljana. 1993. “Negative Polarity: Entailment and Binding.” Linguistics and
Philosophy 16 (2). Springer:149–80.
Romoli, Jacopo. 2012. “Soft but Strong. Neg-Raising, Soft Triggers, and Exhaustiﬁcation.” PhD
thesis, Harvard University.
———. 2013. “A Scalar Implicature-Based Approach to Neg-Raising.” Linguistics and
Philosophy 36 (4). Springer:291–353.
Zeijlstra, Hedde. 2004. Sentential Negation and Negative Concord. LOT/ACLC.
78/79
References III
———. 2008. “Negative Concord Is Syntactic Agreement.” Ms., University of Amsterdam
5:113.
Zwarts, Frans. 1998. “Three Types of Polarity.” In Plurality and Quantiﬁcation, 177–238.
Springer.
79/79