This page displays the upcoming and past meetings of the NihiL seminar. The meetings take place usually on Friday at 16:00 in the ILLC seminar room F1.15 in Science Park 107 and online. If you would like to attend a particular session online or to be added to the mailing list please contact Sonia Ramotowska. The seminar meetings alternate with the meetings of the Discourse in Philosophy (DIP) Colloquium lecture series. To subscribe to the mailing list of the DIP Colloquium please visit this website.
TBA
The interplay between language and reasoning. (Joint talk with CLA and DIP)
Abstract: Large Language Models, and ChatGPT in particular, have recently grabbed the attention of the community and the media. Having reached high language proficiency, attention has been shifting toward its reasoning capabilities. It has been shown that ChatGPT can carry out some simple deductive reasoning steps when provided with a series of facts out of which it is tasked to draw some inferences. In this talk, I will argue for the need for models whose language generation is driven by an implicit reasoning process and a communication goal. To support my claim, I will present two papers recently produced within my group: one evaluates LLMs’ formal reasoning skills and the other focuses on LLMs’ information-seeking strategies; to this end, we take syllogisms and the 20-Questions game as test beds. These tasks have been used extensively in cognitive sciences to study human reasoning skills, hence they provide us with a variety of experiments to inspect the language and reasoning interplay in LLMs.
Abstract: In this talk, I survey certain underappreciated connections between intuitionistic logic (and other broadly constructive logics) and relevance logic. I am particularly interested in a family of heterodox relevance logics whose operational semantical characterizations have an affinity to the informal BHK semantics. Various problems and results connected with these systems are examined, including the disjunction property, embeddings of constructive logics into them, and (un)decidability. On the more philosophical side, I discuss how constructive and relevance logics might inform, and motivate, one another. (This is an updated version of a talk I gave at the PLEXUS Inaugural Conference in 2023 which takes account of several recent results obtained by various researchers.)
Abstract: This talk is based on a joint work with Yanjing Wang and Yunsong Wang. We would like to propose an epistemic interpretation of the tensor disjunction in dependence logic inspired by its connection with the weak disjunction in Medvedev’s early work on intermediate logic based on the BHK-like problem semantics. We expose this rather surprising connection in the setting of propositional inquisitive logic with tensor disjunction discussed in the literature with a powerful epistemic language of knowing how. We give a complete axiomatization of the full logic of this language based on Fine’s axiomatization of propositionally quantified S5 modal logic. In this way, we can obtain an intuitive epistemic understanding of the tensor operator. In the second part of the talk, I will discuss a set of generalized tensor operators with the parameters 𝑘 and 𝑛. We show that adding the generalized tensor operators does not increase the expressive power of our dynamic epistemic logic of know-how, the inquisitive logic, and the propositional dependence logic. However, we show that all the non-trivial generalized tensors are not uniformly definable in those languages with the original tensor operator. We also give the complete axiomatizations of extensions of our dynamic epistemic logic and the inquisitive logic with the general tensors.
Abstract: Click here.
Abstract: In this paper, we propose Point-set Neighborhood Logic (PSNL) to reason about neighborhood structures. The bimodal language of PSNL is defined via a mutual induction of point-formulas and set-formulas. We show that this simple language is equally expressive as the language of Instantial Neighborhood Logic (INL) proposed by van Benthem et al. (2017). As the main results, we first give two complete proof systems, one in Hilbert-style and one in Gentzen sequent-style, each featuring two intertwined K-like systems. The proof of strong completeness of the Hilbert-style system is based on a direct canonical model construction without relying on a normal form. Based on the sequent calculus, we establish constructively the uniform interpolation property of PSNL, from which that of INL follows. (Joint work with Junhua Yu)
Abstract: I develop a uniform treatment of free-choice effects with English ‘any’, ‘or’, and wh-ever free relatives within different environments in terms of the notion of an arbitrary object (Fine 1983, 1985, Horsten 2019). In the resulting theory, there are two kinds of indefiniteness or variableness in natural language semantics. There is the indefiniteness of a Hamblin-style alternative set (i.e., variable reference to a definite thing), which behaves in a disjunctive manner due to its being the input to a pointwise compositional principle. And there is the indefiniteness of an arbitrary object (i.e., definite reference to a variable thing), which may be regarded as a peg for hanging properties common to a range of things, with resulting law-like universal or conjunctive effects.
Abstract: The main topic of this talk is logic and semantic change. I will discuss episodes of semantic change that have to do with the meaning(s) of negation and show how to model them in Bilateral State-based Modal Logic. The discussion will connect with work I have done on lexicalizability (the NAND problem), with ongoing work on homogeneity and definite plurals, and with future work on Negative Polarity Items (the meaning of “any”). Along the way, I will argue that we can use logic to study linguistic fossils, that “not even” is the non-classical negation of “even”, that “any” is a non-harmonious quantifier (in the sense of Dummett), and much more.
Abstract: It’s debated whether periphrastic causatives contain a modal component (Ilič 2013, 2014; Hill 2023) or not (Copley & Harley 2015, Martin 2018). Any semantic theory must capture the fact that deontic modal interpretations are associated with causative constructions. This is best observed when we attempt to cancel obligation after a causative sentence, “Giulia made Giorgio go to the store, ??but he didn’t have to.” To account for this Hill (2023) provides a counterfactual analysis of causatives in which made universally quantifies over possible situations, entailing necessity, which is also entailed by deontic modal expressions, such as had to. Assuming Hill (2023), we tested the hypothesis that strong causatives and deontic necessity modals in English share a necessity entailment using the psycholinguistic tool of semantic priming in a sentence recall task. Our results show that the utterance of had to can indeed by primed by the utterance of made suggesting that priming can target shared core meaning between two syntactically dissimilar constructions. Moreover our results provide support for semantic accounts which propose similar analyses for both periphrastic causatives and deontic modals.
Abstract: In this talk I will extend Bilateral State-based Modal Logic from Aloni (2022) with an exclusive disjunction (ED). ED will be “classically” defined as supporting the disjunction of its disjuncts and rejecting their conjunction, with a caveat: conjunction rejection will be encoded as a presupposition, in order to account for the behavior of disjunction under negation. I will claim that implementing ED in a system that rules out zero models by default (i.e. BSML* from Aloni, 2022) is a better predictor of assertability conditions of disjunction in natural language compared to the standard BSML definition of disjunction and it bears welcomed repercussions for various linguistic phenomena, the foremost being Hurford Disjunctions (Hurford, 1974). I will show how equipping BSML* with ED provides a unified and direct semantic explanation to both standard Hurford Disjunctions and “disjoint Hurford Disjunctions”, a novel class of examples of deviant disjunctions in which there is no entailment relation between the two disjuncts: # Either John lives in Paris and he’s married, or he lives in France and he’s single. (Amir Anvari, p.c.)) Two other felicitous predictions made by ED in BSML* will then be presented: the “uniqueness” interpretation of sentences with multiple disjuncts and the “validity” of a common fallacy in disjunctive reasoning discussed in Johnson-Laird et al. (1992) and Mascarenhas (2014). I will conclude the talk by showing that assuming ED as the standard meaning for disjunction actually clashes with some results in Aloni (2022) (wide scope free choice) and Degano et al. (2023) (absence of exclusivity in production tasks), and by briefly discussing an in-progress solution to solve the conflicts.
Abstract: Homogeneity appears in sentences containing plural definite descriptions exemplified by statements such as “The books are written in Dutch.” This sentence seems to be clearly true if all of the books referred to are written in Dutch and clearly false if none of them are. Despite the homogeneity requirement, sentences with plural definites can have contextually motivated non-maximal interpretations. Kriz and Spector (2020) present an underspecification account that explains how homogeneity and non-maximality emerge with plural definites. They introduce a concept of strong relevance that filters candidate interpretations based on a QUD-derived partition of the common ground. In this talk, I propose an extension of their framework to encompass generics, which are another structurally complex linguistic construct known to display homogeneity and non-maximality. Furthermore, I briefly discuss to what extend plural predication may represent a cognitive default, particularly in its connection to neglect-zero reasonings.
Abstract: In propositional inquisitive logic, formulas are evaluated in terms of support at information states, and the maximal information states supporting a formula are the ‘alternatives’ for the formula. In the first part of the talk, I present an extension of propositional inquisitive logic with a modal operator that roughly expresses that each alternative for its argument formula is consistent with the available information. The resulting logic can be used to model some properties of deontic free-choice inferences, as well as question-directed ignorance. In the second part of the talk, I focus on how to axiomatize the logic by extending it with a version of the global/universal modality. I will also discuss how the technique used to prove completeness can be extended to other types of modal operators with semantics defined in terms of quantification over alternatives, in particular the `minimal cover’ modality introduced by Booth (2022, Independent alternatives, Philosophical Studies, 179:1241–1273).
Abstract: In traditional update semantics, there are two commonly employed notions of discourse consistency, both of which are defined in terms of the absurd state, represented by the empty set. A sentence is consistent iff an update with it does not alway results in the absurd state; a sentence is coherent iff updating with it can reach a fixed point that is not the absurd state. However, it has been noted that both notions fall short of capturing all intuitively contradictory sentences in the face of epistemic modal claims. In this talk, I present a novel solution to this puzzle by providing an analysis of epistemic modals in commitment space semantics. Epistemic modals can have a `meta speech-act’ effect on commitment space by delimiting possible developments of the discourse without altering the current common ground. This picture in turn enables a new conception of discourse inconsistency. A non-empty commitment space can still be deemed inconsistent if all possible developments of the current common ground have been ruled out. In addition, I will explore the prospect of providing an axiomatization of the current framework via a translation into possibility semantics.
Abstract: In this talk, we will introduce a new semantic approach to deontic logic based on the so-called bundled modalities, which essentially pack a quantifier and a modality together. Our starting point is the observation that many “strange” logical behaviors of modalities and logical connectives in natural language are due to the fact that they have more complicated inner logical structures in the semantics. Many examples can be found in epistemic logics of know-wh, where the know-wh modalities often have the implicit ∃x structure based on the mention-some interpretation. As the logical puzzles are abundant in deontic logic, a natural question arises for us: are there also some bundles hidden in the deontic modalities? In fact, the possibilities of viewing permissions and obligations as bundles were informally discussed by Hintikka in the early days of deontic logic. For example, Hintikka proposed to understand permission as a bundle of ∀x◇, i.e., an action type α is permitted iff every token of α is executable on some deontically ideal world. Given the techniques of the bundled modalities, we can flesh out this proposal formally, which results in a desirable logic of free-choice permission satisfying most of the intuitive properties. Moreover, this semantics also predicts new logical behaviors not yet discussed in the literature. For example, according to our semantics, one of the four distributive laws of conjunction and disjunction is invalid, which aligns with our linguistic intuition. Besides the bundled modalities, our approach also features the Brouwer-Heyting-Kolmogorov (BHK) style treatment of propositions as action types inspired by intuitionistic logic. This opens the possibility of fine-grained control of the composition of action types in terms of non-classical connectives. It also reveals the subtleties behind the negation, conjunction, and implication in deontic logic. We will end with a discussion on higher-order permissions that can be treated in our approach by allowing nested permission modalities in the language.
Abstract: Wh-indefinites are indefinites having also the interrogative use. The talk will be on the semantics of wh-indefinites in Mandarin with a special focus on a particular Mandarin wh-indefinite “shenme”. In its indefinite use, “shenme” behaves like an epistemic indefinite triggering an obligatory ignorance inference when unembedded. Additionally “shenme” displays a form distinction with its two forms – bare and non-bare “shenme” – having slightly different distributions with respect to the uses that epistemic indefinites may possibly license. Using the team semantics framework (Aloni & Degano 2022, 2023), I propose that “shenme” is a strict existential with additionally the conditions of variation and maximality, and develop a uniform account of the dual use of “shenme” as either an epistemic indefinite or an interrogative word. The work is based on my master’s thesis supervised by Maria Aloni and Marco Degano.
Abstract: Within team semantics, a focal point of study has been that of expressive power (what properties can a given logic express). One such team logic is BSML, a modal team logic designed for modeling free choice inferences and related linguistic phenomena. In recent work, Aloni et al. (2023) present two extensions of BSML, demonstrating their expressive completeness for all properties [invariant under bounded bisimulation] and all union-closed properties, respectively, and leave open the problem of characterizing the expressive power of BSML. Continuing this line of work, we solve this problem by showing that BSML is expressively complete for all convex, union-closed properties. This leads us to ponder a logic that is expressively complete for all convex properties simpliciter. We introduce a logic which accomplishes precisely that.
Abstract: Polar questions like “May I go to the park or to the beach?” seem to give rise to Free Choice-like inferences. The “yes” answer to such question seem to correspond to the permission to freely choose between going to the park and going to the beach. Similarly, the answer “No” seem to correspond to Dual Prohibition, i.e., prohibition to go to either place. We conducted an experiment to empirically test whether these intuitions are true. In the talk I will indicate how the collected data can allow to establish the source of these inferences. Moreover, I will compare the results to the predictions of the current theories of Free Choice extended with question semantics.
Abstract: The goal of this work is to explore the properties of epistemic might and its interaction with knowledge and belief in multi-agent scenarios. Analogous to the predicate of taste, epistemic might is argued to be sensitive to an agent’s perspectives. In a multi-agent setting, the perspective-sensitivity could potentially reshape our conventional understanding of epistemic might. For example, epistemic contradictions may disappear in a dialogue environment, and factive inferences for know-might sentences are valid only under specific constraints. To tackle these issues, we propose a framework that merges BSEL models—an epistemic variant of BSML—and a two-dimensional semantics for Epistemic Friendship Logic (Seligman et al., 2011, 2013). The resulting model bring the locality of epistemic might and a mechanism of perspective shifting together to model the perspective sensitivity. In addition, the knower(s) of an might-claim can be explicitly represented in the model. Through this point, we can more deeply distinguish between a believe-might sentence and a know-might sentence. This is a joint work with Maria Aloni and Fenrong Liu.
Abstract: According to a recent cognitively motivated theory of natural language quantification, so-called empty-set quantifiers are extraordinarily difficult to interpret because their verification algorithm involves one specific rule intended to handle empty-set situations. The rule relies on the explicit representation of objects in the restrictor set that are not in the scope set. This type of encoding is assumed to be cognitively costly because the cognitive system is tuned to encoding the presence of properties rather than their absence, a strategy viable for non-empty-set quantifiers but doomed to failure for empty-set quantifiers. In this talk, I give an overview of the proposal’s theoretical merits as well as experimental results that support it and rule out a number of alternative accounts. The experimental results include data from the comprehension and verification of German simply and doubly quantified sentences as well as the verification of empty-set quantifiers in außer’-(
besides’)-phrases. In addition, I provide an outlook covering potential future directions for empirical research and theoretical extensions.
Abstract: The disjunction embedded in the scope of a universal quantifier, (e.g., Every X is A or B) gives raise to so-called ‘Diversity’ inferences (D-inferences, e.g., Some Xs are A and Some Xs are B). According to the traditional implicature approach, D-inferences are implicatures that arise via the ‘Negative Universal’ inferences (NU-inferences, the negations of Every X is A and Every X is B). Crnic et al. (2015) provided experimental evidence (i) that D-inferences are independent from NU-inferences, however, they argued (ii) that for disjunction in the scope of a universal modal the D-inferences cannot be observed independently of the NU-inferences.
In this talk, I will present two experiments that have tested the availability of D-inferences in the absence of NU-inferences for the determiner every and the epistemic (Experiment 1) and deontic (Experiment 2) modals must. The results show that, for both types of quantifiers, D-inferences could be derived independently of NU-inferences. The findings make a two-fold contribution: the results for the modal cases go against the predictions (ii); and the response time results from both experiments challenge the implicature-based approach to D-inferences.
Abstract: I will discuss some consequences of the combination of Neglect Zero with a non-classical definition of rejection, which allows us to derive contradictory and contrary readings of negation as illocutionary effects. I will focus on three applications: an account of homogeneity inferences, an hypothesis about lexical simplicity, and an account of lexication patterns of inclusive and exclusive operators (words like even and only).
Abstract: My talk will be on an application of BSML (or in fact of a variant) in natural language interpretation and my focus will be on the possibility to interpret natural language sentences “from left to right”, or, more accurately, following the linguistic precedence order. Left-to-right interpretation seems key in understanding how semantics and pragmatics interact, as Stalnaker and Karttunen already saw in the 1970s.
Officially, the logic I use is classical type theory, but I’ll show that the BSML ideas can be reproduced within a certain domain of that logic, the domain of functions from states to functions from truth-values to truth-values. The reason why these ideas help with modelling left-to-right interpretation has to do with the fact that bilateral negation and split disjunction each in their own way can be viewed as mechanisms that push essential information to the atomic level. Another mechanism, available in classical logic, that does this is Skolemisation. With the help of these a refutational tableau calculus for natural language can be set up that has the property that the order of the branches of its trees is closely aligned with the precedence order of elements in the sentence under interpretation. Semantics-pragmatics interaction will be modelled as the combination of interleaved processes of developing such a tree, abductively contradicting its open branches from left to right (taking context into consideration), and adding the contradicting information to context. The talk will mainly be concerned with the first of these aspects and the role of BSML in it.
Abstract: The game-theoretical or so-called dual negation of independence-friendly logic and dependence logic does not correspond to any well-defined semantic operation: the class of models of a sentence does not determine the class of models of its dual negation, so the dual negations of two equivalent sentences need not be equivalent. Burgess showed (in the equivalent context of Henkin sentences) that this lack of determination is extreme in the sense that for any pair of sentences A and B of dependence logic which share no models, there is a sentence C such that A is equivalent to C and the dual negation of C is equivalent to B. So given only the class of models X of a sentence, all we know of the class of models Y of its negation is that Y is the class of models of some sentence of dependence logic, and that X and Y share no models. Bilateral State-based Modal Logic (BSML) is a modal logic employing team semantics developed to model natural language phenomena such as free choice inference. BSML makes use of a bilateral negation which is essentially the same notion as the dual negation. We show that Burgess’ result also holds for BSML.
Abstract: Indefinites are known to give rise to different scopal (specific vs non-specific) and epistemic (known vs unknown) uses. Farkas & Brasoveanu (2020) explained these specificity distinctions in terms of stability vs. variability in value assignments of the variable introduced by the indefinite. Typological research (Haspelmath 1997) showed that indefinites have different functional distributions with respect to these uses. In this work, we present a formal framework where Farkas & Brasoveanu (2020)’s ideas are rigorously formalized. We develop a two-sorted team semantics using tools from dependence logic (Väänänen 2007) and model these uses by means of dependence and non-dependence atoms. We apply the framework to explain typological variety of indefinites, their restricted distribution and licensing conditions, as well as some diachronic developments of indefinite forms. We also use the inclusion atom to capture the difference between epistemic and deontic modals. We conclude by offering a novel notion of negation, intensional negation.
Abstract: Bilateral State-based Modal Logic (BSML) is a modal logic employing team/state-based semantics which can be used to model free choice inference and other natural language phenomena. We introduce a natural deduction system for BSML as well as for two extensions: BSML with the inquisitive disjunction and BSML with a novel emptiness operator. We also study the expressive power of these logics—we show that the two extensions are expressively complete.