Conclusions: This paper challenges previous empirical generalizations surrounding bare

cardinal partitives. The inner cardinal need not be a specific indefinite. New data like that in (3-5) suggest that bare cardinal partitives deserve more detailed attention in the semantics and pragmatics literature. The theoretical account suggests that both ontological pluralities (sums) and evaluation pluralities (the multiple values a variable takes) have a role to play in the semantics of partitives.

References:Barker, C. 1998. Partitives, double genitives and anti-uniqueness. Natural Language & Linguistic Theory 16, 679–7171. ♣ Brasoveanu, A. & Farkas, D. 2011. How indefinites choose their scope. Linguistics and Philosophy 34, 1–55. ♣ de Hoop, H. 1997. A semantic reanalysis of the partitive constraint. Lingua 103, 151–174. ♣ Ladusaw, W. 1982. Semantic constraints on the English partitive construction. In Proceeding of WCCFL 1, 231–242. ♣ Link, G. 1983/2002. The logical analysis of plurals and mass terms: a lattice-theoretical approach. In Paul Portner & Barbara Partee (eds.), Formal semantics: the essential readings, 127–146. Malden, MA: Blackwell. ♣ Winter, Y. 2000. What makes choice natural? In Reference

and anaphoric relations, 229–245. Springer. ♣ Winter, Y. 2005. On some problems of (in)definiteness within flexible semantics. Lingua 115, 767–786.

2

Cognitive vs. Emotive Factives: An Experimental Differentiation

Kajsa Djärv, Jérémy Zehr, Florian Schwarz University of Pennsylvania A current question in presupposition theory is whether and how to differentiate (sub-)classes of triggers (Abusch 2002, Sudo 2012, Tonhauser et al. 2013, Romoli 2014, Abrusan to appear). In the realm of factives – i.e. clause-embedding predicates that presuppose the truth of their complements –, the distinction between semi-factives (e.g. discover ) and ‘full’ factives (e.g. regret) has been present right from the start (Karttunen 1971), but at the same time has remained elusive and hard to substantiate experimentally (e.g. Jayez et al. 2015). We present experimental evidence revealing differences between them using a yes/no- continuation task (Cummins et al. 2013), and explore a theoretical explanation in terms of a distinction between triggers that do vs. do not entail their presupposition (Sudo 2012).

Components of Factives. Two crucial factors contribute to the meaning of factives: the

factive component (that the complement clause is true) and the attitude component (relating the matrix subject’s mental state to the described state of affairs). Arguably, the relationship between the two differs for cognitive and emotive factives, as illustrated below:

(1) a. Cognitive: “Mark has found out that his parents are visiting him.” b. Emotive: “Mark is surprised that his parents are visiting him.” (2) a. Factive part: Mark’s parents are in fact visiting him.

b. Attitude part: Mark has become aware of this fact. (1a)

The prospect of their visit surprised Mark. (1b) Intuitively, finding out about a circumstance seems to strictly require that this circumstance is actual, tying the attitude and factive parts together. However, it is intuitively conceivable that one can experience surprise towards a circumstance that could turn out to be non-actual, which leaves some room for disentangling the attitude part from the factive part. We argue that such a difference should have an impact on the availability of yes/no-continuations that deny the presupposition and test the corresponding predictions experimentally.

Methods & Design. We adapt the yes/no-continuation task previously employed by Cum-

mins et al. (2013) to explore differences between types of triggers such as stop and again. The task pairs a presuppositional question with a response starting either with ‘yes, although. . . ’ or ‘no, because. . . ’, followed by a denial of the presupposition:

(3) Q. {Did Mark find out/Was M. surprised} that [p his parents are visiting him]? A1. Yes, although they had to cancel because of the weather.

A2. No, because they had to cancel because of the weather.

Cummins et al. interpret rating-differences between their triggers for the no responses of this sort as reflecting varying degrees of availability of local accommodation (making the presupposition part of what is negated by no). But in light of our distinction between factives above, we could also locate a potential difference in the availability of yes responses: if the factive part is entirely entangled in the attitude part, as suggested for cognitives, saying yes and then denying the presupposition should be contradictory. For emotives, on the other hand, where the attitude part is more independent from the factive part, it would seem possible to just endorse the former by using yes even if qualifying immediately that the latter does not hold. Thus, our hypothesized difference between cognitives and emotives predicts the cognitives to exhibit a stronger bias towards no answers than the emotives. We tested this

prediction through a web-based experiment on Prolific.ac with 36 English speakers who were presented with questions containing a presupposition introduced by either a cognitive (find

out, realize) or an emotive (be disappointed, be surprised). The two factive types were divided

into separate blocks, with block-order counter-balanced between subjects. Participants had to choose between yes and no-continuations (A1 & A2). They could also indicate a lack of preference by saying that both responses were equally good or bad.

Results & Discussion. Response patterns in the first block exhibited a clear difference

between cognitive and emotive factives, which were revealed to be statistically significant in mixed effect regression analyses, in that no answers were more frequent for cognitives (in addition, more emotive trials rendered both bad responses).

0% 20% 40% 60%

Yes No BothGood BothBad

Response Choice proportions Factive Cognitive Emotive * * *

Continuation Choices in 1st block.

Interestingly, these differences virtually disappeared in the sec- ond block, suggesting that exposure to one type of factive quite strongly influenced responses to the other, possibly through priming of one type of interpretation (with an increase in no responses for emotives) or through adjustment of standards (with an increase in both bad responses for cognitives). In sum,

while the contrasts are subtle and subject to influencing one another across blocks, the clear contrast in the first block supports the notion that the two aspects of meaning (Attitude & Factive) relate to one another differently in emotives and cognitives.

Theoretical Implications. Establishing differences between cognitives and emotives con-

stitutes a substantive empirical contribution, but the data also call for a more detailed theo- retical explanation. We propose that the intuitive difference rendered descriptively above can be straightforwardly captured in terms of a theoretical distinction between triggers that do vs. do not make their presupposition part of the entailed content, following Sudo (2012), who distinguishes pronouns and triggers like again from other triggers such as stop along these lines. In light of the present data, we argue that yes/no-continuations of the kind employed here constitute an appropriate test for such a difference. In particular, we argue that cogni- tives do, but emotives do not entail their presupposition. Schematically, the entailed content of a cognitive factive as illustrated in (1a) could be modeled as Factive+Attitude (2a+b) while that of an emotive (1b) would simply be Attitude (2a). This difference is largely un- noticeable in contexts where the factive presupposition holds, but in the present paradigm, it shows up in the relative goodness of yes answers (compared to no) with subsequent pre- supposition denials for emotives. These are possible because an aspect of the meaning (the entailed content: Attitude) can be singled out and then affirmed by yes independently of the factive presupposition. For cognitives, this is not possible, as the presupposition is also entailed, leaving no as the only clearly viable response.

This proposal also has the potential to inform other long standing observations about differ- ences between types of factives, such as Karttunen’s data on local readings in the antecedent in conditionals and claims about the greater ability of cognitives to have their complement serve as the main point of utterance (Simons 2007), which could be viewed as resulting from the presupposition also being present at the level of entailment. It could also bear on Hooper & Thompson’s (1973) observation that only cognitives (‘semifactives’) cross-linguistically al- low embedded main clause phenomena, such as Verb Second.

Selected References. Cummins, Amaral & Katsos. (2013). Backgrounding and accommodation of presuppositions: An

experimental approach. SuB 17: 201-218.Simons. (2007). Observations on embedding verbs, evidentiality, and presupposition.

Lingua 117. Sudo. (2012). On the Semantics of Phi-Features on Pronouns. PhD thesis, M.I.T.

2

When is not-believing believing that not? Mojmír Doˇcekal and Jakub Dotlaˇcil

Introduction. Despite its long research tradition, neg-raising (NR) still lacks an empirically adequate theory which would explain the phenomenon, its varied behaviour in different embed- dings, its occasional suspension and its cross-linguistic variation. In this study we bring new data from Slavic which can help us understand more about the nature of neg-raising (NR), its interaction with other processes at the syntax-semantics interface and which can help us decide which of the recent competing NR theories (Gajewski 2007, Romoli 2013, Collins & Postal 2014) approximates Slavic NR better.

Experiment. The experiment studied the interaction between NR and subjunctive/indicative mood in Czech. Participants of the experiment had to judge the acceptability of sentences with strong NPIs, ani jeden ‘not even one’ (NPI1) and až do ‘until’ + time expression (NPI2), using the 5-point Likert scale (5=best, 1=worst). The strong NPIs appeared in clauses embedded under three predicate types, shown in (1) – (3): (1) opinion class of NRs, (2) probability class of NRs, (3) non-NR communication predicates. Testing the acceptability of strong NPIs served as a test of NR-hood, since only NR predicates can license such NPIs in embedded clause. One account why NR predicates can license strong NPIs states that strong NPIs are grammatical only if their environment is downward-entailing in its at-issue, as well as implicated and presupposed content (see Gajewski 2011, Chierchia, 2013, a.o., and below). Each environment was varied for the mood of the predicate in the embedded clause (indicative mood vs. subjunctive mood). There were 18 exp. items in 2x3 (=6) conditions. The experiment also included 36 fillers. 60 Czech native speakers took part in the experiment. The experiment was run online in Ibex.

(1) Nemyslím, do-not-think-I že that 0/by IND/SUBJ ani even jeden one z of bˇežc˚u runners m˚uže/mohl can/could ten the závod race vyhrát. win ‘I don’t think that even one of the runners can/could win the race.’

(2) Není it’s-not možné, possible že that 0/by IND/SUBJ ani even jeden one z of bˇežc˚u runners m˚uže/mohl can/could ten the závod race vyhrát. win ‘It’s not possible that even one of the runners can/could win the race.’

(3) Netvrdím, do-not-say-I že that 0/by IND/SUBJ ani even jeden one z of bˇežc˚u runners m˚uže/mohl can/could ten the závod race vyhrát. win ‘I don’t say that even one of the runners can/could win the race.’

Results. All participants but 3 had no problems to pass control fillers (uncontroversially gram- matical/ungrammatical sentences), so the final number of participants: 57. Using random-effects probit models to analyze the data with mood (subjunctive vs. indicative), predicate type (opinion, probability, communication) and their interaction, we found that NR predicates ((1) and (2) above) were judged as significantly better than non-NR communication predicates ((3) above): z = -2.51, p = 0.012. No difference between opinion and probability NR predicates was de- tected. The difference between NR and non-NR predicates in their ability to license strong NPIs supports the existence of NR in Czech (as a representative of Slavic languages) (see also Doˇcekal & Dotlaˇcil, 2015, for further experimental support). Furthermore, subjunctive mood was a statistically significant factor (z = 2.39, p = 0.017), increasing the acceptability of strong NPIs of both types, NPI1 and NPI2. To our knowledge, this is a new finding in the research on NR and promises some insights into the NR suspension and cross-linguistic variation. Finally,

we split items into those using NPI1 and those using NPI2 and used the type of NPI as an extra predictor. The difference between NPIs was significant (z = 2.65, p = 0.008): NPI1 (ani jeden ‘not even one’) was judged as worse than NPI2 (až do ‘until’). Such a difference between NPIs

was not observed before, as far as we know.

0 50 100 1 2 3 4 5 Acceptability count Condition belI belS comI comS posI posS

Acceptability task Figure: counts of responses, 1-5, in Conditions

(1) – (3) (and varied mood)

Analysis. We explain the effect of mood and we offer some speculatory remarks as to why strong NPIs differ in their acceptabil- ity in negated embedded clauses. We follow Villalta (2008) who argues that subjunctive mood in embedded clauses allows the trans- fer of alternatives from the embedded sen- tence to its embedding predicate (unlike in- dicative mood which stops the alternatives transfer). This reasoning fits well the scalar approach to NR (esp. Romoli 2013) where NR predicates (beside the assertion) con- tribute the excluded middle alternative (EM) to the semantic composition: Alt(N R) = {λpλx.x[p], λpλx.[x[p] ∨ x[¬p]]}. When a sentence like (1) or (2) containing NR in the scope of negation is asserted, the exhaustification of the EM alternative leads to a schematical interpretation like ¬_{xp ∧ ¬¬[}xp ∨ x¬p] resulting in a deductively valid NR inference x¬p (the DE environment both in the at-issue and implicated meaning → strong NPIs licensed). But this can happen only if NR predicates can evaluate the alternatives of the embedded sentence, therefore in languages with the grammaticalized indicative/subjunctive distinction the NR com- putation can proceed only if the embedded clause is in subjunctive mood. We assume that Slavic languages (unlike English where the mood distinction is nowadays moot) reveal the dependence of NR interpretation on the availability of alternatives, which supports the scalar approach to NR (the presuppositional approach to NR would have to use some further assumptions to describe this dependency). However, we note that the effect of mood could also be captured by syntactic accounts of NR-hood (Collins & Postal 2014), given that subjunctives, but not indicatives, are generally more transparent for movement (cf. Progovac, 1996). Finally, the difference between two types of NPIs suggests that, contrary to standard claims, strong NPIs do not form one coherent class. One possibility is that only ani jeden ‘not even one’ is subject to conditions listed above, while až do ‘until’ can also appear in atelic environments (cf. Gajewski 2005). This would be relevant in our experiment since negative predicates are atelic, licensing ‘until’. Selected References

Ž. Boškovi´c & Gajewski, J., ‘Semantic correlates of the NP/DP parameter,’ in Proceedings of NELS 39, 2009. • C. Collins & M. Postal, Classical NEG Raising. Cambridge, MA: MIT Press, 2014 • M. Doˇcekal & J. Dotlaˇcil, ‘Experimental evidence for neg-raising in Slavic’, SinFonIJA 8, 2015 • J. R. Gajewski, ‘Licensing strong NPIs,’ NLS, 19, pp. 109–148, 2011 • J. R. Gajewski, ‘Neg-raising and polarity,’ L&P, vol. 30, no. 3, pp. 289–328, 2007 • L. Horn, A natural history

of negation. Chicago: UoCh Press, 1989 • G. Chierchia, Logic in Grammar. Oxford: OUP 2013 • J. Romoli, ‘A scalar implicature-based approach to neg-raising,’ L&P, vol. 36, no. 4, pp. 291–353, 2013 • E. Villalta, ‘Mood and gradability’, L&P, 31, pp. 467-522.

Explaining explain:

embedded declaratives as modifiers

Overview: Taking Pietroski’s (2000) observations concerning the difference between x ex- plained that P vs. x explained the fact that P as a starting point, we argue for an elimination of the basic type-distinction between individuals and events. This is replaced by a sortal dis- tinction within the domain of entities. Furthermore, we follow Kratzer (2006) and others in adopting a predicative semantics for embedded declarative clauses. The linguistic pay-off is that that-clauses are permitted to combine with the verbs that embed them via simple intersec- tive modification, rather than being introduced via a neo-Davidsonian thematic function. This is argued to provide an account for a range of substitution problems, as well as further reducing the role of syntax in accounting for argument structure.

Explain and the DP/CP distinction: Pietroski (2000, p. 1) shows that the interpretation of a sentence with explain depends on the syntactic category of its complement.

(1) a. Nora explained [CPwhy Fido barked]. explanandum reading

b. Nora explained [DPthe fact that Fido barked]. explanandum reading

c. Nora explained [CPthat Fido barked]. explanansreading

In (1-a) and (1-b), the complement specifies the content of what is explained, whereas in (1-c), the complement specifies the content of the explanation. The contrast is problematic on the view that facts are true propositions, meaning that the fact that Fido barked and that Fido barkedshould be inter-substitutable if the latter is true.

Pietroski’s account: Pietroski captures the the distinction by assigning (1-b) and (1-c) the Logical Forms in (2-a) and (2-b), in the framework of neo-Davidsonian event semantics.

(2) a. _{J(1-b)K = ∃e[}AGENT(e, Nora) ∧ explaining(e) ∧THEME(e, the fact that Fido barked)]

b. _{J(1-c)K = ∃e[}AGENT(e, Nora) ∧ explaining(e) ∧CONTENT(e, that Fido barked)]

Pietroski does not consider how the two Logical Forms are composed. In both (1-b) and (1-c) the that-clause fulfills theCONTENTrole – in (1-c) it specifies the content of the explaining, and

in (1-b) it specifies the content of the fact. It is implausible that the CONTENTrole is specified as part of the lexical semantics of the verb; rather that-clauses are inherently content-providers. Generality: As discussed by e.g., King (2002) and Uegaki (2015), Pietroski’s data involving explain is but one instance of a more widespread substitution problem involving embedded declarative and content nouns. Notice that there is a lack of entailment between (3-a) and (3-b). Intuitively, in (3-a) the embedded declarative specifies the propositional content of the remembering event, whereas in (3-b) this is not the case.

(3) a. John remembered that Mary is a fraud.

b. John remembered the rumour that Mary is fraud.

It is mysterious under Pietroski’s account why clausal complements quite generally seem to provide the propositional content of the event/state specified by the embedding verb.

Special quantifiers and c-selection: Most importantly, Pietroski implicitly assumes that DP complements are always THEMEs, whereas CP complements are alwaysCONTENTs. This can-

not be correct, based on evidence from Moltmann’s (2013) class of “special quantifiers”, which include something, everything, nothing and several things. These expressions may subsitute for clausal complements.

(4) a. John {thinks|believes} [CP that Mary is moving to Germany].

b. John {thinks|believes} [DPsomething].

c. John {*thinks|believes} [DPthe rumour that Mary is moving to Germany].

e. It is {*thought|*believed} [DP something].

Special quantifiers cannot be CPs, since they cannot appear in all CP positions: they do not survive when the verb is passivised, as shown in (4-d) and (4-e). Furthermore, they appear in positions in which CPs are disallowed, such as the complement to P: John hopes for something – namely that Mary will leave soon. Now, consider what happens when explain takes a special quantifier as its complement.

(5) John explained [DPsomething] – namely,

{[DP the fact that Fido barked]|[CPthat Fido barked]}

(5) shows that a special quantifier allows both the explanandum and explanans readings. The correct generalization is therefore that CPs receive explanans readings whereas DPs can receive either.

Analysis: Following Moulton (2015), we propose that that-clauses are of type he,ti – they denote properties of entities with propositional content. Departing from Moulton, we imple- ment sortal types as presuppositions. This captures the intuition that that-clauses are inherently content providers.

(6) _{Jthat mary leftK = λ x}e: ∃qst[CONT(x, q) = 1].CONT(x, λ w. Mary leftw) = 1

We further depart from Moulton in assuming no basic type distinction between individuals and events (see Bach et al. 1995) – rather, this also represents a sortal distinction within De.

Verbs denote properties of events. Certain verbs, like explain, denote properties of events with propositional content. When a CP (type he,ti) merges with a verb (type he,ti), they combine via Predicate Modification (PM).

(7) a. _{JexplainK = λ x :}EVENT(x) = 1 ∧ ∃qst[CONT(x, q) = 1]. explainingw(x) = 1

b. _{Jexplain that Mary leftK = λ x :}EVENT(x) = 1.

explaining_w(x) ∧CONT(x, λ w. Mary left_w)

Moulton conversely assumes that that-clauses combine with CPs via a complex remnant move- ment derivation. In the talk, we provide arguments for the simpler approach. That-clauses also combine with nouns via PM. The fact that Q denotes the unique fact with propositional con- tent specified by Q. The result of combining this directly with a verb is undefined, since verbs presuppose that the entity they combine with is an event. The DP integrated into the argument structure via merger of a thematic predicate THEME:

(8) _{Jexplain the fact that Mary leftK = λ x :}EVENT(x) = 1. explaining_w(x)

∧THEME(x, ιy[fact(y) ∧CONT(y, λ w.Mary left_w)]

We derive the generalization that that-clauses receive an explanans reading, whereas DPs headed by content nouns receive an explanandum reading, while maintaining one simple meaning for explain. As for special quantifiers, we assume they are generalized quantifiers of type het,ti. They are special, because they place no sortal restriction on the individuals they quantify over. Hence, special quantifiers are compatible explanandum and explanans readings. When some- thingranges over events with propositional content, its trace combines with the verb via Func- tional Application. When something ranges over individuals with propositional content, the result of applying the trace to the verb is undefined. The trace is therefore integrated into the structure viaTHEME.

References: BACH, EMMON ET AL. 1995. Introduction. In Quantification in Natural Lan- guages. Dordrecht: Kluwer. KRATZER, ANGELIKA. 2006. Decomposing attitude verbs. Talk

in honor of A. Mittwoch.. LAHIRI, UTPAL. 2002. Questions and Answers in Embedded Con- texts. OUP. MOLTMANN, FRIEDERIKE. 2013. Abstract Objects and the Semantics of Natural Language. OUP. MOULTON, KEIR. 2015. Copies and compositionality. Linguistic Inquiry.