EVENING LECTURES

Short Bio: Gemma Boleda is an ICREA Research Professor in the Department of Translation and Language Sciences of the Universitat Pompeu Fabra, where she co-leads the Computational Linguistics and Linguistic Theory (COLT) research group. She previously held post-doctoral positions at the University of Trento (Italy), The University of Texas at Austin (USA), and Universitat Politècnica de Catalunya (Spain). Before that, she graduated in Spanish Philology at Universitat Autònoma de Barcelona and obtained her Ph.D. in Cognitive Science and Language at Universitat Pompeu Fabra (both in Spain). She was also a visiting researcher at the Computational Linguistics & Phonetics department (CoLi) of Saarland University and the Institute for Natural Language Processing (IMS) of the University of Stuttgart, both in Germany. In her research, Prof. Boleda uses quantitative and computational methods to better understand how natural languages convey meaning, and why they do it like they do.

DATE: August 5, 2025

TOPIC: Why are deep learning models so good at language?

The advent of deep learning has been a revolution in society, in general and also in particular regarding language, with billions of people now using generative AI systems in their daily lives. It is also being a revolution in science, again in general and in particular regarding language: for the first time in history, we have artificial models that reproduce linguistic behavior with a high degree of accuracy. This raises the question, why are they so good? What do they have that was lacking before? In my talk, I will provide theoretical arguments and empirical evidence for the view that deep learning models are successful because they natively support two crucial properties of language: systematicity and messiness. They are thus a synthesis between previous approaches that focused on one at the expense of the other, like formal vs distributional approaches to semantics. I will finish with some thoughts on what this implies for linguistic theory.

Short Bio: Regine Eckardt works in formal semantics and pragmatics. Her writings include monographs on formal semantics in grammaticalization (2006), and on free indirect speech and its representation in formal semantics (2015). In recent years, she studied the pragmatics of non-canonical questions, such as conjectural, rhetorical or pedagogical questions. She passed her PhD at the IMS, Stuttgart with Ede Zimmermann and Hans Kamp. After positions at Berlin and Göttingen, Eckardt has been working at the University of Konstanz since 2015.

DATE: July 29, 2025

TOPIC: What metaphors and axiomatic theories (may) have in common

Nonliteral, metaphorical language poses a notorious challenge to formal semantic theory. Dedre Gentner and co-authors (1983a, 1983b, 2005) suggest that metaphors rest on a structure-preserving mapping between ontological domains. I propose that Gentner’s approach to metaphor should be reconceived in terms of axioms in model theory. Axioms are first principles that define the essence of a type of structure. For instance, the axioms of group theory G capture the essence of addition, subtraction and zero (+, -, 0). Group theory, hence, doesn't ONLY study the natural numbers N but any set with a binary function that supports the axioms G. Conversely, a structure isn’t “a group” born by nature — any structure that supports the axioms G can be conceived as a group. This allows easy transfer of group theorems into new domains.  

I propose that metaphors rest on a similar transfer of first principles from one domain to another. Instead of literally mapping one ontological domain onto another (which often doesn’t make mathematical sense simply because one domain may be larger in size than the other), the essential transfer consists in (a) implicit reference to a set of first principles M in the source domain and (b) the interpretation of M in the target domain. For instance, the metaphor “the boat is full” intuitively likens states to sea vessels. This, however, doesn’t mean that we should literally map all states to boats, or all members of one state to the crew of one boat (which one?). Instead, the speaker alludes to an abstract “boat theory” with axioms like, e.g., ’the vessel carries persons’, ‘persons divide into passengers and crew’, ’the captain is part of the crew’, ’the captain is responsible for all persons carried by the vessel’, etc. The metaphor triggers an interpretation of these first principles in the domain of a (legal) state. Some terms are interpreted against their literal meaning (the captain = the president), but at an abstract level, boat governance may literally be organised like state administration.

Unlike mathematical axioms, these first principles in metaphors don’t entail consequences with formal rigour. Yet, metaphors also highlight statements that are natural consequences in the source domain (“a boat can be so full that it no longer fulfils its function to serve and save its population”), and strongly suggest that they also hold true in the target domain (“a state can be so full that it no longer fulfils its function to serve and save its population”). 

My talk explores this way of conceiving of metaphors, which opens new avenues for integrating non-literal language use and formal semantics.

Short Bio: Phokion Kolaitis is a Distinguished Research Professor at the University of California Santa Cruz and a Principal Research Staff Member at the IBM Almaden Research Center. His research interests include principles of database systems, logic in computer science, and computational complexity. He is a Fellow of the AAAS and the ACM, a Foreign Member of the Finnish Academy of Science and Letters as well as the Academia Europaea. Among other distinctions, he was the recipient of a Guggenheim Fellowship, two IBM Research Division Outstanding Innovation Awards, an IBM Research Division Outstanding Technical Achievement Award, two ACM PODS Alberto O. Mendelzon Test-of-Time Awards,  an ICDT Test-of-Time Award, and an Alonzo Church Award for Outstanding Contributions to Logic and Computation.

DATE: August 7, 2025

TOPIC: Possible Worlds and Certain Answers

Databases are typically assumed to have definite content so that users can pose queries and retrieve unambiguous answers. It is often the case, however, that a database may contain information that is incomplete, inconsistent, or uncertain. Possible world semantics provides meaning to queries on databases suffering from these deficiencies. Such databases are viewed as compact representations of all their possible rectifications; by definition, the certain answers are the query answers that hold true in every possible rectification of a deficient database. 

The goal of this lecture is to provide an overview of some of the work on certain answers as a unifying framework for coping with incompleteness, inconsistency, and uncertainty in databases. Case studies include data exchange, probabilistic databases, inconsistent databases, and election databases in social choice theory.

Short Bio: Kristina Liefke is a Junior Professor at Ruhr-Universität’s Department of Philosophy II, where she leads the research group ‘Philosophy of Information and Communication’. Her research lies at the intersection of formal semantics, philosophy of language, and philosophy of mind. She has worked extensively on intensionality, non-propositional attitudes, type-logical semantics, and natural language ontology. In her most recent work, she tries to apply formal semantic tools to mental representations. Kristina is a PI in the research unit FOR 2812: Constructing Scenarios of the Past (Bochum), an associate member of the Centre for Philosophy of Memory (Université Grenoble Alpes), and an external member of the Munich Center for Mathematical Philosophy (LMU Munich). She obtained her PhD from the Tilburg Center for Logic and Philosophy of Science (the Netherlands).

DATE: July 31, 2025

TOPIC: Mnemic meaning: The super semantics of episodic memory

‘Montague’s thesis’ (Bach, 1986) holds that natural languages can be described as interpreted formal systems. Its extension to other representational media (like pictures and film) claims that all public, systematic, and conventional representations can be described as interpreted formal systems (Greenberg, 2011). My talk argues that Montague’s thesis can be further extended to mental representations — esp. to episodic memory representations (see Addis, 2020; Michaelian, 2016). This argument is based on the observation that mnemic and pictorial representations share many semantic properties (incl. reference, truth/accuracy, compositionality, iconicity, perspectivity). As a result, the familiar tools from picture semantics (e.g. situation semantics, geometrical projection) can be fruitfully applied to mnemic representations. My talk illustrates the fruitfulness of this application by showing how these tools make precise existing views and concepts from the memory sciences (e.g. mnemic contents, intentional objects, episodic recombination). Inversely, it shows how memory research can enrich state-of-the-art picture semantics (e.g. by demonstrating the importance of metarepresentation, and by identifying new accuracy concepts).