23 Apr 2026

Patrick Elliott (Heinrich Heine University Düsseldorf)

Explorations in the negative zone

Classically, the core contribution of negation is taken to be boolean. In this talk, I explore the possibilities afforded by an individual-level negation, building on recent philosophical work by Akiba (2009), Fine (2017), and Bledin (2024). In other words, alongside individuals such as 'Andreea' and 'Yasu', we may posit negative counterparts, 'not Andreea' and 'not Yasu', which behave, in all relevant respects, as individuals. I show how this ontological-enrichment allows for a uniform treatment of quantificational determiners as predicates of (plural) individuals, constituting a predicative alternative to the prevailing approach based on Generalized Quantifier theory. In the presentation, I'll primarily focus on exploring some new applications of the predicative theory to the following phenomena: exceptive constructions, exceptional scope phenomena, and the definiteness effect.

21 May 2026

Luca Molinari (University of Wrocław)

A Nanosyntactic approach to the singulative suffix -at in Hasawi Arabic

18 Jun 2026 (TBC)

Danfeng Wu (University of Geneva)

Title TBA

20 Nov 2026 (TBC)

Anne Abeillé (Université Paris Cité)

Title TBA

9 Apr 2026

Justyna Grudzińska (University of Warsaw)

Disentangling Form and World Knowledge in LLM Interpretation: Evidence from Quantifier Scope Disambiguation

(Joint work with Jakub Kosterna, Maciej Miecznikowski, Wojciech Borysewicz, Julia Poteralska, Kacper Rutkowski and Jan Henryk Kwapisz)

We investigate how large language models (LLMs) construct meaning using Quantifier Scope Disambiguation (QSD) as a controlled probe. We introduce two complementary resources: (i) a balanced English corpus designed to neutralize classical surface heuristics in scope resolution, and (ii) a pseudosentence dataset that removes real-world referents to isolate formal cues.

In a zero-shot question-answering setup, we evaluate a range of LLMs and compare them to human baselines. While models achieve high accuracy on the balanced corpus, performance drops substantially on pseudosentences, with the largest degradation observed for inverse-scope readings. This pattern indicates that surface-level cues alone cannot account for model behavior and points to a crucial role of world knowledge in sustaining non-surface interpretations.

To test the causal contribution of external knowledge, we further equip the same models with retrieval-augmented generation (RAG) while keeping the task and prompt fixed. RAG yields only limited gains in overall accuracy, but these gains are highly selective: they primarily affect inverse-scope readings, especially in configurations where classical surface predictors conflict with the preferred interpretation.

Taken together, our results show that LLM scope interpretation is best understood as emerging from an interaction between formal interpretive pressures and world knowledge. World knowledge – implicit or retrieved – does not replace formal constraints, but selectively modulates interpretation when such constraints are insufficient or misleading.

9 Oct 2025

Nina Haslinger (Leibniz-Zentrum Allgemeine Sprachwissenschaft – ZAS)

Morphosemantic primitives in universal quantification

(Joint work with Alain Noindonmon Hien, Emil Eva Rosina, Viola Schmitt and Valerie Wurm)

The quantifier systems of natural languages encode distinctions that have no counterpart in classical first-order logic, such as the distinction between what I’ll call "distributive" universal quantifiers (English *every*, *each*) and "maximizing" universal quantifiers (English *all*). Distributive universal quantifiers systematically disallow collective and cumulative interpretations wrt. their nuclear scope, whereas maximizing ones permit them to a limited extent and seem to have the effect of picking out a maximal element with respect to a part-whole relation. The standard view has been that this reflects two distinct primitive quantifier meanings, in line with the distinct lexicalizations generally found in European languages.

In this talk, we argue based on data from Keenan & Paperno (2012, 2017) and new data from Mabia languages that the standard view fails to derive a number of typological facts. First, some languages use the same form for distributive and maximizing universal quantifiers, with the choice of interpretation seemingly determined by properties of the complement, such as number and definiteness (cf. Gil 1995, Winter 2001, Fassi Fehri 2020). Second, in languages with distinct distributive and maximizing forms, the same properties of the complement correlate with the choice between quantifier forms in a non-arbitrary way that does not reduce to syntactic agreement and does not have a straightforward account on a standard theory with two lexical primitives. Third, distributive forms are often internally complex and formally related to the numeral *one*.

We propose an alternative account in which there is a single primitive for universal quantification, which contributes either maximization or distributivity depending on the algebraic properties of the restrictor predicate it combines with. "Distributive" forms like English *every* and *each* are analyzed as portmanteau forms that jointly realize the quantifier itself and elements that add a non-overlap or atomicity presupposition concerning the restrictor predicate. We implement this idea within a spanning approach to lexicalization in which portmanteau forms are not exceptional, but reflect the default mechanism for exponents that realize more than one feature. The resulting system derives the correlation between form choice and semantic properties of the restrictor predicate and also allows us to make sense of apparent mismatches between distributivity and complement number, e.g. uses of maximizing forms with definite singular complements to express roughly the meaning of *whole*, and uses of distributive forms with degree-interval expressions as in *every ten minutes*.

This talk is based on our recent NLLT paper, but in some places analytical choices will be made that deviate slightly from the "official" proposal adopted in the paper.

26 Jun 2025

Adam Szczegielniak (University of Gdańsk)

Multiple remnant sluicing

This talk explores the properties of a type of ellipsis called sluicing: ‘Russel claimed some barber shaved himself, but I do not know [which barber]₁ [Russel claimed t₁ shaved himself]’ vis a vis the properties of multiple remnant sluicing: ‘Russel claimed he introduced some barber to some philosopher, but I do not know which barber to which philosopher [Russel claimed he introduced]’. Both types of sluicing are argued to involve wh-movement evacuation of the remnant out of a complete syntactic structure whose PF material is then subsequently deleted (Merchant 2001). However, multiple remnant sluicing has been shown to differ from single remnant sluicing as far extraction restrictions of the wh-remnant (Lasnik 2014, Abels & Dayal 2013, Citko & Gračanin-Yuksek 2020, Barros & Frank 2023). Based on data from island alleviation, I propose a phase based linearization approach (Fox & Pesetsky 2005) that complements existing approaches to how sluices are derived. Furthermore, it will be argued that evacuation of the remnants via wh-movement is best analyzed as prosody driven wh-movement (Richards 2010).

Abels, K. and Dayal, V., 2023. On the syntax of multiple sluicing and what it tells us about wh-scope taking. Linguistic Inquiry, 54(3), pp.429-477.

Barros, M. and Kotek, H., 2019. Ellipsis licensing and redundancy reduction: A focus-based approach. Glossa: A Journal of General Linguistics, 4(1).

Barros, M. and Frank, R., 2023. Attention and locality: On clause-boundedness and its exceptions in multiple sluicing. Linguistic Inquiry, 54(4), pp.649-684.

Citko, B. and Gračanin-Yuksek, M., 2020. Conjunction saves multiple sluicing: How *(and) why?. Glossa: A Journal of General Linguistics, 5(1).

Fox, D. and Pesetsky, D., 2005. Cyclic Linearization of Syntactic Structure. Theoretical Linguistics, 31(1-2), pp.1-45.

Kotek, H. and Barros, M., 2018. Multiple sluicing, scope, and superiority: Consequences for ellipsis identity. Linguistic Inquiry, 49(4), pp.781-812.

Lasnik, H., 2014. Multiple Sluicing in English?. Syntax, 17(1), pp.1-20.

Merchant, J., 2001. The Syntax of Silence: Sluicing, Islands, and the Theory of Ellipsis. Oxford University Press.

Richards, N., 2010. Uttering Trees. MIT Press.

5 Jun 2025

Sebastian Zawada (IPI PAN)

Agreement in Polish revisited: Evidence from copular clauses

In this talk, I examine some intriguing types of Polish copular constructions in which nominative nominal predicates control agreement – (i) Z Marysi był dobry człowiek ‘Mary (lit. of Mary) was a good person’, (ii) To był skandal ‘This was a scandal’, (iii) Marysia to był dobry człowiek ‘Mary was a good person’ (the underlined phrases in the translations unambiguously agree in gender in the Polish examples). These constructions challenge the traditional view that Polish verbs agree with their subjects. Through a detailed analysis of the syntactic items involved, I show that the observed agreement pattern (as well as canonical subject-verb agreement) is best accounted for by the influential cross-linguistic view that verbs agree with the highest accessible phrase in the domain of agreement. Particular attention is given to constructions such as (iii): I evaluate various syntactic analyses thereof and present novel data that undermine some existing accounts and support a revised version of another.

8 May 2025

Sonia Ramotowska (University of Amsterdam, ILLC)

Individual differences in the computation and processing of scalar implicatures

Sentences like “Some elephants are mammals” are ambiguous between a literal reading, compatible with all elephants being mammals, and a reading with scalar implicature (SI), implying that not all elephants are. Thus, when providing truth-value judgements of these sentences, the participants may respond either way, depending on which reading they get. Experimental studies show considerable individual differences in the choice of reading. While some individuals accept literally true sentences with false SI, others systematically reject them. The factors driving these differences are crucial to understanding the mechanisms involved in processing SIs. In this talk, I will present a rigorous way of quantifying individual differences in the computation of SIs using a hierarchical Bayesian model with latent group classifications. I will demonstrate two applications of this model. The first study will be about individual variability in the rates of upper-bounding and lower-bounding SIs associated with the <some, all>-scale. I will argue that the robustness of an SI is modulated within individuals by certain linguistic features, such as the presence of negation. The second study will be a reanalysis of the previously published data by Van Tiel, Pankratz, & Sun (2019) and Van Tiel and Pankratz (2021) involving multiple scales. Based on this analysis, I will argue that individual differences in reading preferences modulate the processing of SIs.

5 Dec 2024

Berke Şenşekerci (University of Warsaw)

Towards gradient grammaticality

Linguistic theories typically assume that grammaticality is a binary concept, i.e., a given sentence is either licensed by the grammar or it is not. However, acceptability judgment data, which form the empirical foundation of these theories, consistently exhibit gradient patterns. The interpretation of this gradience remains a subject of debate. Some view it as evidence that mental grammars are inherently gradient (i.e., gradient "grammaticality"), while others attribute it to performance-related confounds, maintaining a strict distinction between grammaticality and acceptability. In this talk, I will explore arguments for and against the notion of gradient "grammaticality" and propose a series of cross-linguistic experiments aimed at investigating whether mental grammars are the source of this gradience. Finally, I will discuss the implications of gradient "grammaticality" for formal theories of grammar that assume binary grammaticality.

14 Nov 2024

Andrés Soria Ruiz (University of Barcelona)

Expressive inferences & mighty obstacles

Subjective predicates (i.e. ‘tasty’) carry so-called acquaintance inferences (AI), to the effect that the speaker has first-hand experience of whatever they are tasting. E.g., an utterance of 'pizza is tasty' implies that the speaker has tried pizza. Similarly, moral predicates (i.e., 'okay') carry so-called practical inferences (PI), to the effect that the speaker has a practical stance towards whatever they are judging. E.g., 'tax fraud is okay' implies that the speaker has a practical stance regarding tax fraud. AI & PI have similar properties (Willer & Kennedy 2020): they project under negation ('pizza is not tasty' implies acquaintance; 'tax fraud is not okay' implies a practical stance), but "flip" under certain operators ('pizza might be tasty' implies unacquaintance with pizza; 'tax fraud might be okay' suggests no definite stance towards tax fraud). Previous accounts have failed to characterize the exact empirical profile of this inference—by treating modals and other operators merely as blocking, rather than "flipping" AI/PI. Moreover, they provide either a principled but limited account of these inferences (e.g., Anand & Korotkova 2018; Ninan 2024 on the AI) or a broader but ad-hoc account of its projection profile (Willer & Kennedy 2020). I offer a new proposal, where the correct predictions fall out of the independent meaning of all expressions involved.

3 Oct 2024

Jan Wiślicki (University of Warsaw)

Verbs' hierarchy regulates the nouniness of CPs

In this talk, I discuss the problem of nominal properties of complementizer phrases (CPs) in both direct and indirect discourse. I use data from nominalization of CPs in Japanese to show that the availability and formal properties of such structures are not uniform across types of matrix verbs. Then I argue that these properties are regulated by a hierarchy of verbs recently developed by Susi Wurmbrand and her colleagues. Based on these findings, I propose new formal accounts of matrix verbs capturing the effects observed for nominalised CPs. I conclude by drawing more general prospects for deriving the reportative character of matrix verbs, regulated by the abovementioned hierarchy.

26 Sep 2024 (Joint IPI Seminar & Formal Linguistics Seminar)

Daniel Rothschild (University College London)

Language and Thought: Evidence from LLMs

Daniel Dennett speculated in Kinds of Minds (1996):

“Perhaps the kind of mind you get when you add language to it is so different from the kind of mind you can have without language that calling them both minds is a mistake.”

Recent work in AI can be seen as testing Dennett’s thesis by exploring the performance of AI systems with and without linguistic training. I will argue that the success of Large Language Models at inferential reasoning, limited though it is, supports Dennett’s radical view about the effect of language on thought. I suggest it is the abstractness and efficiency of linguistic encoding that lies behind the capacity of LLMs to perform inferences across a wide range of domains. In a slogan, language makes inference computationally tractable. I will assess what these results in AI indicate about the role of language in the workings of our own biological minds.

12 Sep 2024

Jacopo Romoli (Heinrich Heine University Düsseldorf)

The ups and downs of ignorance (and distributivity)

Plain disjunctive sentences, such as The mystery box contains a blue ball or a yellow ball, typically imply that the speaker does not know which of the two disjuncts is true. This is known as an 'ignorance' inference. We can distinguish between two aspects of this inference: the negated universal upper bound part (i.e., the speaker is uncertain about each disjunct), which we call 'uncertainty', and the existential lower bound part (i.e., the speaker considers each disjunct possible), which we call 'possibility'. In the traditional approach, uncertainty is derived as an implicature, from which possibility follows. When disjunctions are embedded under a universal operator, such as It is certain that the mystery box contains a blue ball or a yellow ball, an inference analogous to possibility arises (i.e., it's possible that the mystery box contains a blue ball and it's possible that it contains a yellow ball). This inference is generally called a 'distributive' inference and the traditional approach derives it in the same way as in the unembedded case, from the corresponding negated universal inference (i.e. it is uncertain that the mystery box contains a blue ball and it is uncertain that it contains a yellow ball). In both the simple and the embedded cases, the traditional implicature approach predicts that the derived lower bound inference should never arise without the corresponding negated universal one. In this talk, we report on two experiments using a sentence-picture verification task based on the mystery box paradigm, the results of which challenge the traditional approach to ignorance and distributive inferences. Our findings show that possibility can arise without uncertainty, and that, in the same way, the distributive inference can arise without the corresponding negated universal one, thus calling for a reevaluation of the traditional view of disjunction and its inferences. We discuss how alternative theories can account for the observed patterns of inference derivation in a unified fashion and the open challenges that remain.

13 June 2024

Janek Guerrini (École Normale Supérieure)

Revisiting kind predication

English bare plurals and Italian definite plurals are thought to be kind-denoting, as they support kind predication (Carlson, 1977), such as e.g. ‘Lions are extinct’. Kinds are standardly seen as intensional sums. In this work, I argue that, if we extend to kind-denoting plurals tools independently motivated by the treatment of referential plurals, a number of puzzles, both old and novel, fall in line.

Optional reading:

https://lingbuzz.net/lingbuzz/008172

22 May 2024 (Wednesday!)

Sebastian Zawada (IPI PAN)

Polish demonstrative TO, copular clauses and agreement

In the first part of the talk, I present a syntactico-semantic analysis of the Polish demonstrative word TO ‘this’, which builds on the known observation that it can be used in two different syntactic environments: (i) in typical nominal positions and (ii) in a unique copular structure [TO + BYĆ ‘to be’ + NP], in which the right-hand NP appears in the nominative case and triggers agreement with the verb. I propose an analysis assuming one demonstrative root for TO, which gets different grammatical categories in (i) and (ii). The analysis, couched in the LrFG framework, accounts not only for the different syntactic behaviour of TO in (i) and (ii), but also for its different anaphoric possibilities in the two uses. In the second part of the talk, I present some peculiarities related to agreement facts in Polish copular constructions in general, some of which have attracted little attention so far, and consider possibilities of a unified syntactic analysis thereof.

9 May 2024

Justin Bledin (Johns Hopkins University)

Free Choice with Arbitrary Objects

I develop a uniform treatment of free-choice effects with English 'any' and 'or' 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.

11 Apr 2024

Dariusz Kalociński (IPI PAN)

Simulating the Emergence of Semantic Universals in Populations of Cognitive Agents

Natural languages vary in their quantity expressions, but the variation seems to be constrained by general properties, so-called universals. In the talk I will show an attempt of explaining the universals of monotonicity and convexity as emergent properties arising through language coordination processes, and further reinforced by approximate number sense - a well known constraint of (not only) human perception. The talk will be based on a joint work with Gierasimczuk, Rakowski and Uszyński (Gierasimczuk et al. 2023, Journal of Logic, Language and Information).

20 Mar 2024 (Wednesday!)

Adam Przepiórkowski (IPI PAN / University of Warsaw)

Symmetry in Coordination

Linguists do not agree on the basics of the syntactic structure of coordination: is John and Mary headed by the conjunction (and), by one of the conjuncts (e.g., John), or perhaps by both conjuncts simultaneously (John and Mary)? Important for the debate is the phenomenon of coordination of unlike grammatical categories, as in Pat is a Republican and proud of it, where the first conjunct (a Republican) is a noun phrase, and the second (proud of it) is an adjectival phrase. According to asymmetric approaches to coordination, the first conjunct is the head of the whole structure, so a Republican and proud of it is effectively a noun phrase, while on symmetric approaches all conjuncts contribute their categorial features equally, so a Republican and proud of it has both nominal and adjectival features. The aim of this talk is 1) to re-examine arguments – based on unlike category coordination – for the two classes of approaches to the syntax of coordination and 2) to present the design of acceptability judgement experiments which, if successful, will provide new evidence against asymmetric approaches to coordination.

29 Feb 2024

Diego Feinmann (IPI PAN)

Characterising Oddness

There’s something odd or deviant about (1) and (2).

(1) John lives in Łódź, and he lives in Poland.

(2) John lives in Łódź, and/but he doesn’t live in Poland.

(1) is a redundant sentence (intuitively, the last conjunct isn’t needed); (2), by contrast, is a contradictory, or incoherent, sentence (given what is known, it’s just not possible for someone to live in Łódź and not live in Poland).

In this talk, I’ll present a set of minimal pairs that call into question all current theories of redundancy and contradictoriness. In addition, I’ll put forth two generalisations—grounded in the novel notion of predicate connection (Feinmann 2022)—that provide a fresh perspective on the nature of oddness phenomena.

Optional background readings:

https://semprag.org/index.php/sp/article/view/sp.9.7

https://lingbuzz.net/lingbuzz/006989