MIT Workshop on Exhaustivity 2016

Date: Saturday, September 10th 2016
Location: The Stata Center32-141.

The information about the posters in the poster session can be found below.


 Time  Title and speaker  Commentator
8:30-9:00  Breakfast
Marie-Christine Meyer
Danny Fox [handout]
Andreas Haida, Tue Trinh
 Jacopo Romoli [handout]
10:40-11:10  Break
Yimei Xiang
Gennaro Chierchia [slides]
Giorgio Magri
Paolo Santorio [slides]
12:40-14:10 Lunch and poster session
14:10-15:00 Uli Sauerland [slides]
Wataru Uegaki
Yael Sharvit [handout]
15:50-16:20 Break
Kathryn Davidson
Yosef Grodzinsky
Leon Bergen, Noah Goodman, Roger Levy
 Benjamin Spector

Reception and dinner: There will be a reception, at 18:30, and dinner, at 19:30, after the workshop (registration required for dinner). These will take place in the R&D space on Stata Center's 4th floor.  

Abstracts of the talks

Grice and Grammar: How cooperative are weak sentences?, Marie-Christine Meyer
In this talk we are going to explore the relation between grammatical ignorance implicatures and Gricean pragmatics. We argue that the maxims of Brevity and/or Quantity+Relation force us to assume that language, i.e, grammar, can encode not only lower-bounded meanings like "at least some", but also upper-bounded meanings like "some but not all". Crucially, grammar must even be able to encode the speaker's ignorance about the upper bound if Grice's maxims are valid at all. We arrive at the conclusion that syntactic operators like exh and K (be certain) make language highly suitable for cooperative communication in Grice's sense -- more suitable than a language without these operators. [slides]

A plea for (no) monsters, Andreas Haida and Tue Trinh
Fox & Hackl (2006) proposes a theory of measurement from which it follows that "John has 3 children" is asymmetrically entailed by "John has 3.1 children," or more generally by any proposition of the form "John has n children" where n is a rational number greater than 3. However, these authors did not discuss explicitly the semantics of phrases such as "3.1 children." In this talk, we provide an analysis of numerals which validates such entailment relationships as exemplified above and, at the same time, does justice to the intuition that a concept, such as "children," can be identified with a set whose cardinality, by logical necessity, is a natural number. [handout]

Mandarin particle "dou": A pre-exhaustification exhaustifier,Yimei Xiang
This paper provides a uniform semantics to capture various functions of Mandarin particle "dou", including the quantifier-distributor use, the FCI-licenser use, and the scalar marker use. I argue that "dou" is a special exhaustifier with the following three features: (i) it triggers an additive presupposition, just like any other overt exhaustifier (e.g., 'only'); (ii) it operates on only sub-alternatives (namely, alternatives that are not innocently excludable and distinct from the prejacent); and (iii) it has a pre-exhaustification effect. [handout]

Blindness and Hirschberg’s contextually ordered alternatives, Giorgio Magri
Magri (2009, 2011) argues that scalar implicatures are blind to any contextual information. Schlenker (2012) objects that contextual Blindness is inconsistent with the contextually ordered alternatives documented in Hirschberg (1991). This talk tries to resolve this challenge by providing some initial evidence in favor of the following conjecture. Whenever contextual ordering seems to drive a certain pattern of scalar behavior, there is actually more logical structure than meets the eye and provides a logical ordering of the alternatives which happens to be homomorphic to the contextual ordering. It is thus conceivable that it is the logical ordering which drives the scalar behavior, while the contextual ordering plays no role, as indeed predicted by Blindness. [slides]

Reducing the locality of PPI anti-licensing to an instance of PPI shielding, Andreea Nicolae
In recent work, Nicolae (2015) argues that what distinguishes PPI disjunctions, such as French 'ou', from polarity insensitive disjunctions, such as English 'or', is the fact that PPI-disjunctions obligatorily trigger epistemic inferences (building on the work of Spector 2014). This analysis explains the first two PPI properties (Szabolcsi 2002), namely anti-licensing by a negative operator and rescuing, but in its current form, wrongly predicts that PPI disjunctions should be unacceptable in the scope of a negative operator, regardless of its locality wrt the disjunction. In this talk I will propose an extension of this proposal which takes the acceptability of PPIs under non-local negation to correlate with their acceptability under negation in the presence of a shielding operator. I will not only explain why this correlation should hold, but also illustrate that global PPIs, such as French 'soit soit', which are known to be insensitive to the locality of negation (Spector, 2014), are also insensitive to shielding. [handout]

'Wonder' and embedded exhaustivity, Wataru Uegaki (based on joint work with Floris Roelofsen)
A sentence such as 'John wonders whether Ann, Bill or Carol arrived' gives rise to an implication that John is ignorant about each of the answers to the complement question, i.e., he is ignorant about whether Ann arrived, whether Bill arrived and whether Carol arrived. I will argue that this implication is a result of an exhaustification sensitive to structural alternatives (Katzir 2007, Fox & Katzir 2011), and discuss evidence showing that the exhaustification has to be obligatory and local. The analysis is implemented with a built-in exhaustification operator in the lexical semantics of 'wonder'. I will also discuss the general implication of the analysis with respect to other items that lexically mark the presence of the EXH-operator (Spector 2014, Mekik & Singh 2016). [handout]

The contribution of sign languages toward a new unified picture of conjunction and disjunction, Kathryn Davidson
In many contexts and in many languages, logical conjunction and logical disjunction are expressed by separate lexical items ('and' and 'or'), but there has been increasing interest in apparent exceptions to this, including American Sign Language, Warlpiri, Cheyenne, child language, and free choice contexts. In this talk I'll focus on this question through the lens of ASL, comparing with other cases both in terms of the kind of empirical data that have been highlighted for each case, and in theoretical analyses that have been argued for them, which include underspecification and a lack of scalar alternatives for exhaustification. [slides]

Three ways to solve the symmetry problem, Leon Bergen, Noah Goodman and Roger Levy
The symmetry problem (Fox 2007, Katzir 2007) poses a challenge for a broad range of pragmatic theories. Several groups have proposed related solutions to the problem: total (Katzir 2007, Fox & Katzir 2011) and graded (Bergen, Levy, & Goodman 2016) elimination of alternatives on the basis of complexity or cost. Recently, Romoli (2013, personal communication) has introduced several instances of the symmetry problem which cannot be solved through complexity or cost alone. We show that these apparent counterexamples can be resolved through two symmetry-breaking principles, informativity and Bayesian Occam's razor (MacKay 2003). These principles follow automatically from the model proposed in Bergen et al. (2016), requiring no changes to this model. [slides]

Abstracts of the posters

Scalar NPIs in embedded questions, Maayan Abenina-Adar
It’s well-known that NPIs can appear in questions, as in Does Bill have any change? and Who here has ever been abroad? This work presents a puzzle in the distribution of a particular type of NPI in embedded questions -- namely, scalar NPIs like lift a finger (AKA, ‘minimizers’) and even ONE (AKA, ‘weak even’). It is shown that the licensing of scalar NPIs in embedded questions depends not only on whether the question is strongly- or weakly-exhaustive but also on what the embedding predicate is; in this respect, scalar NPIs differ from weak NPIs like any and ever, which appear to be sensitive only to the question’s exhaustive-strength (Guerzoni and Sharvit 2007, 2014). The difference between these classes of NPIs is illustrated in (1)-(2). (1) shows that scalar NPIs are in principle embeddable, while (2) shows that only the weak NPI any is licensed in a strongly-exhaustive embedded polar question under the verb tell (and more generally, under episodic responsive verbs). 
  1. Sue asked Mary [whether Bill had read {any articles / even ONE article} from the reading list] 
  2. Mary told Sue [whether Bill had read {any articles / *even ONE article} from the reading list]  
I speculate about how to explain the contrast between scalar and weak NPIs under a theory where their distribution is constrained by the meaning contributed by implicit or explicit exhaustivity and scalar operators (Krifka 1995, Lahiri 1998, Chierchia 2013 et al.).

At least as nary disjunction: Scales, context and exhaustification, Peter Alrenga
A hallmark feature of the scalar focus operator at least is its capacity to convey speaker ignorance: from an utterance of Grover ate at least [some]F of his dinner, a listener will typically infer that the speaker does not know the exact quantity that was consumed. Following Büring (2008), much recent work has explored the view that these ignorance inferences are conversational implicatures due to the interaction of at least's core semantic properties with general pragmatic principles. A common impulse of such pragmatic approaches is to draw an analogy to ordinary disjunction (e.g., Grover ate tuna, chicken, or duck for dinner). But capitalizing on this analogy has proven surprisingly difficult--while the simplest version of this view correctly captures at least's truth-conditional effects, it appears to mischaracterize at least's pragmatic contributions. I argue that this simple view can indeed be maintained, once it is recognized (i) that the scales that at least operates over are fundamentally pragmatic/contextual in nature, and (ii) that these scales are never ordered by entailment. Re: (ii), I assume that in all putative cases where at least operates over an entailment scale, a covert exhaustification operator occurs in the prejacent, which disrupts any entailments amongst its scalar alternatives. I further show how the presence of this operator can account for certain restrictions on F-marking with at least observed by Spector (2014) (cf. At the very least, #John is [more than six feet]F tall vs. John has to be [more than six feet]F tall).

*onlyonly, Sam Alxatib
Why can Exh(Exh...) produce FC inferences, but not only(only...)? In this poster I attempt to provide an answer. The reason, I propose, has to do with (i) a difference in prejacent status, which is presuppositional for only but assertoric for Exh, and (ii) a difference in distributional constraints---occurrences of Exh can sometimes be vacuous, but occurrences of only never can. The observation in (i) is relevant, because the difference makes ¬only(p) stronger than ¬Exh(p), and produces an asymmetry that is critical for FC-disjunctions: the negations of the alternatives that generate FC (under Exh) are weak enough to be consistent with Exh's prejacent, but the negations of the parallel alternatives (under only) are too strong to be consistent, and are therefore not excluded. We show moreover that the remaining alternatives in the case of only(only...) are either non-excludable, or are excludable but contribute information that is already presupposed, hence violating the ban on vacuous use (ii).

On the global calculation of embedded implicatures, Moshe E. Bar-Lev and Danny Fox
Sentences of the form `every A is P or Q' can admit two kinds of inferences at the same time: an embedded implicature (`every A is P or Q but not both'), and distributive inferences (`some A is P' and `some A is Q'). Since the Neo-Gricean theory (NG) of scalar implicatures allows only for a global computation, it requires admitting `some'-alternatives to derive the embedded implicature (Chemla & Spector 2011). However, we show that this assumption rules out the distributive inferences due to symmetry. Moreover, within NG, symmetry yields ignorance inferences, so embedded implicatures are incorrectly predicted to be obligatorily accompanied by inferences that the speaker does not know whether or not some A is P (and likewise for Q). The grammatical theory (GT) of scalar implicatures, by contrast, can derive both the embedded implicature and the distributive inferences even when admitting `some'-alternatives, with the application of recursive exhaustification. We extend this point of view to globally derive embedded free choice for `every A is allowed P or Q' along the lines of Fox (2007), following Buccola & Haida's (2016) assumption that the set of alternatives is closed under disjunction.

van Benthem's problem, exhaustification, and distributivity, Natasha Ivlieva and Sam Alxatib
The sentence Fewer than 4 men arrived has the upper–bounded interpretation that can be paraphrased in the following way “It’s not true that more than 3 men arrived”. However, under the assumption that few has an adjectival meaning, and that its quantificational meaning arises from existential closure, we derive uninformative truth conditions that make the sentence equivalent to someone arrived (van Benthem’s problem). We show that the right truth conditions can be derived by applying an exhaustification operator above existential closure. Importantly, we make the following two assumptions: (i) the mechanism that calculates scalar implicatures is only sensitive to logical information (cf. Magri 2009, 2011, Fox & Hackl 2006), and more specifically (ii) information about distributivity is not part of logical information. We show that the Exh operator that is “blind” to distributivity derives the desired upper-bounded interpretation. We also point out some similarities and differences between our analysis and Buccola & Spector’s (2016) approach to numerals, which utilizes a maximality operator instead of Exh.

Untangling Tanglewood using covert focus movement, Hadas Kotek (joint work with Michael Yoshitaka Erlewine)
We argue for the existence of covert focus movement in English focus association. Our evidence comes from Tanglewood configurations of the form in Kratzer 1991. We show that Tanglewood configurations are sensitive to syntactic islands, contrary to Kratzer’s claims and predictions. We propose that Tanglewood configurations always involve covert movement of the focused constituent––possibly with covert pied-piping (Drubig 1994; Krifka 1996, 2006; Tancredi 1997, 2004; Wagner 2006; Erlewine and Kotek 2014)––to bind a bound variable in the ellipsis site. This availability of covert pied-piping explains examples such as Kratzer’s which are apparently not island-sensitive. We show that covert focus movement is long-distance and not simply QR. Kratzer’s proposal that ellipsis enforces the identity of focus indices and other previous approaches will be argued to overgenerate Tanglewood readings.

What is at issue? Exhaustivity of structural and morphological DP focus in Telugu, Lilla Magyar
The most important property which has been claimed to differentiate between the type of exhaustivity associated with preverbal foci (and it-clefts) and foci in the scope of only is the status - the at-issueness - of the inference (Velleman et al. (2012)).The main goal of this study is to investigate the relationship between at-issueness, exhaustivity and focus in Telugu, a Dravidian language, which, like Hungarian, has a designated preverbal focus position, but also has postverbal and morphologically marked focus which seem to be interpreted similarly to foci in the scope of only. With the help of "yes, but" tests (similarly to (Destruel et al. (2015)), we diagnose the status of the preverbal focus construction of Telugu in comparison with the parallel construction in Hungarian. We also compare the status of postverbal and morphologically marked focus with that of only-exclusives. Our results suggest that Telugu preverbal focus is similar to Hungarian preverbal foci and English it-clefts with respect to the status of the exhaustive inference, that is, the exhaustivity associated with it is not at issue, whereas postverbal and morphologically marked foci have the same interpretation as foci in the scope of only, that is, the exhaustivity associated with them is at issue. This is consistent with the crosslinguistic picture and provides further evidence that "yes, but" tests are a good diagnostic of the status of exhaustive inferences.

Towards computer implementations of exh, Can Mekik and Michael Vertolli
We discuss computational properties of an implemented algorithm for computing innocent exclusion (Fox, 2007). The computation is indeed costly (Spector, 2015). We discuss sources of complexity, and discuss avenues for reducing the complexity.

Deriving the strong-reading of Imperatives via Exhaustification, Despoina Oikonomou 
The dual character of Imperatives with respect to their quantificational force has been a long-lasting puzzle in the literature (Han 2000, Schwager 2006 / Kaufmann 2012, Portner 2007, Grosz 2009, Condoravdi & Lauer 2012, von Fintel & Iatridou 2015). The sentence in (1) can be interpreted as permission in a context where the Addressee wants to open the window or as command/request in an out-of-the-blue context where a Professor asks a student to open the window: 
  1. Open the window. 
In this poster I argue that Imperatives involve an existential modal. The universal reading is explained on the basis of two factors; i) lack of a scalar counterpart as opposed to overt modals (cf. Deal 2011) ii) strengthening via an Implicature derived in the presence of certain Focus Alternatives (cf. Schwager 2005). I present evidence for the existential analysis of the Imperative operator from scopal ambiguities with only. Then, I provide an analysis for the emergrence of the strong-reading as an Implicature. 

A subject-object asymmetry in the online processing of only: evidence from eye-tracking, Pooja Paul, Tanya Levari, Dylan Hardenbergh and Jesse Snedeker
Rooth's (1985, 1992) account of only predicts the integration of contextual information with the linguistic representation, making the online processing of only potentially insightful for understanding the nature of the interface between the formal alternative-generating machinery and the context-sensitive processes involved in restricting the ALT-set. Recent work investigating the real-time processing of only (Kim et al. 2015; Romoli et al. 2015) has suggested processing differences based on whether only associates with the subject argument ('Only [John]F bought an apple') or the object-argument ('John only bought [an apple]F'). Our findings from two Visual World eye-tracking studies (n = 32, 33) show that the position of only, i.e., whether it associates with the subject or object, interacts differently with discourse context in guiding online comprehension. Our results support a more nuanced picture of the syntax-semantics-pragmatics interface than standardly assumed, in which syntactic and pragmatic constraints interact dynamically with lexically encoded meaning to guide online comprehension. 

Russian particle li and exhaustivity effects, Philip Shushurin
Russian li is a focus-sensitive particle used to form polar questions (King 1993). One remarkable contrast is the following: while modifiers of universal quantification appear, modifiers of existential quantification are banned from doing so, see the contrast between (1) and (2).
  1. …vsex li detej on vstretil 
    all.ACC LI kids.ACC he met
    ‘whether he met all kids’
  2. …*nekotoryx li detej on vstretil 
    some.ACC LI kids.ACC he meu
    int. ‘whether he met some kids’
I want to argue that this contrast arises from the fact that li is associated with an exh operator. While both some and all form the same alternative set (some; all), the latter element is deleted once exh is applied in (2) (but not in (1)). Furthermore, I hypothesize that li takes focus value of a constituent in its scope and transforms it into the set of possible answers (see Uegaki 2016) for a similar proposal for Japanese ka). Since at least two alternatives are needed to form a focus value (Rooth 1996), li fails to apply, and 2. results in ungrammaticality.

Organizers: Luka Crnič, Roni Katzir, Raj Singh
with the generous support of the MIT Department of Linguistics and Philosophy