Colloquium

Distinguished Speakers in Language Science, the joint Colloquium of the Department of Language Science & Technology and CLaCC, takes place Thursdays, 16:15, Building A2.2, room 1.20.2, unless indicated otherwise. An overview of previous talks can be found here.

Links to LangSci calendar: iCal  HTML

Winter Semester 2026/27

19.11.2026 - Ellen Lau

Ellen Lau

Against isolated items and individuals: structured mappings in production, comprehension, and non-linguistic knowledge

Since the early 20th century, both number theory and neural network theory have inspired cognitive scientists to conceive perception and cognition in terms of isolated items and individuals--individual feature detectors which can be activated in different patterns to yield different results; individual lexical items which are activated in the course of language production and comprehension; the domain of individuals which sentences can assert true or false statements about. In this talk I will argue that moving away from the isolated item metaphor can yield better models of language processing and interpretation. Recognizing that language production and comprehension require distinct, non-invertible mappings between complex data structures rather than activating a common set of bundled 'items' can allow us to make better sense of cross-linguistic diversity, the prevalence of so-called 'idiomatic' non 1-to-1 mappings, and dissociations between production and comprehension in behavioral and neural responses (e.g., Krauska & Lau, 2023). With respect to non-linguistic knowledge, I adopt Sandeep Prasada's (2026) arguments that humans do not mentally represent individuals as isolated entities or lists of features, but rather make use of instance-of-kind conceptual structures that inherently relate instances to a kind. This kind of data structure explains how humans reason about and organize long-term knowledge of individuals, and motivates new approaches to the semantics of common nouns, quantification, and adjectival modification. 

10.12.2026 - Trevor Brothers

What sets bilinguals apart? Neural signatures of prediction and reprocessing in second-language reading

Do skilled bilinguals process sentences just like monolinguals do? Or, does a dual-language system require additional adaptations or processing strategies to support comprehension? In a series of eye-tracking and event-related potential (ERP) experiments we found both quantitative and qualitative differences in bilingual reading. First, low-proficiency bilinguals had difficulty predicting upcoming determiners (“the” vs. “a”), when this feature (definiteness marking) was absent in their native language. Critically, these reductions in predictive processing were quantitative, and disappeared with increasing levels of L2 proficiency. In contrast, when encountering a rare or unfamiliar word in an L2 (“algebra”, “vendor”) bilinguals engaged in prolonged rereading and showed a distinct neural response (a P600) that was not present in monolinguals. This unique ERP signature was present even in bilinguals with “native-like” proficiency, which may reflect a qualitative shift in how bilinguals resolve cross-language interference or support L2 learning. These findings help clarify the interactions between language experience, typological differences, and cognitive control. They also offer new theoretical implications for models of predictive processing and bilingual reading.

07.01.2027 - Anita Körner

Anita Körner

Gender Form Influences Gender Representations: Evidence From Psycholinguistic Experiments

Do people predominantly think of men when reading person references in the generic masculine? And does the use of so-called gender-sensitive language lead to different gender representations? These questions, that is, questions about gender representations elicited by linguistic gender forms, are a central point of contention in the public debate about gender-sensitive language. This talk will discuss experimental evidence on these questions, specifically experiments examining how reading or hearing different linguistic forms affects cognitive gender representations and influences language processing. I will present studies that employ different measures, such as, participant judgments, memory retrieval, response times, eye tracking and EEG to assess gender representations and discuss their strengths and limitations. The results yield converging evidence demonstrating that linguistic gender form influences cognitive gender representations, both for explicit and implicit measures, for sentence processing as well as text processing, and for participants with different socio-demographic characteristics. Open questions as well as criticisms of these studies raised in public discourse will also be discussed and evaluated in the light of experimental results.

 

04.02.2027 - Linda Drijvers

TBA

Summer Semester 2026

16.07.2026 - Louis ten Bosch

Louis ten Bosch

AI, Automatic Speech Recognition and phonetics

Over the last couple of years, we witness the enormous impact of AI-inspired end-to-end approaches for Automatic Speech Recognition. In this talk I will discuss the progress in the field, by sketching a number of experiments that aim to find phonetic relations in the latent representations of Wav2vec2.0. Also the use of end-to-end audio decoding will be discussed for the early detection of neurodegenerative diseases (Alzheimer, Parkinson etc.) from the speech signal. Finally, we will address the topic of interpretability of a foundation model, by discussing various techniques that can be used to ‚crack‘ the black box. A nice discussion point is insofar/whether the old ‚classical‘ knowledge about speech is represented in some way in the parameters of a trained end-to-end model, and how.
 

20.08.2026 - Tiago Pimentel

Tiago Pimentel

Promises and Limitations of Causality for Machine Learning Interpretability

How can we move from observing what a model does to understanding why it does it? In this talk, I argue that causality is necessary but not sufficient to uncover the mechanisms underlying model predictions. First, I examine a “macro” view of model analysis, showing how econometric tools—such as regression discontinuity or difference-in-differences—can isolate the causal impact of specific design choices, like tokenisation and training-data selection, on a model’s outputs. Second, I turn to a “micro” view of mechanistic interpretability, focusing on causal abstraction as a method to establish whether a model implements a high-level algorithm. I demonstrate that this approach faces a critical limitation: without strict assumptions about how models encode information, the framework becomes vacuous, implying that any model implements any algorithm. This reveals that the ability to make counterfactual predictions about a model is not, on its own, sufficient to guarantee that we understand it. I conclude with a short discussion of how causality can be used to develop more principled interpretability methods, and with an open question about what interpretability is currently missing.

17.09.2026 - Xing Han Lu

Xing Han Lu

From a dozen clicks to a thousand steps: The evolution of digital agents and the future of work

A year ago, mentioning the word “agent” in a conversation would’ve yielded a confused stare. So how come everyone is now talking about agents? In this talk, we take a look at earlier developments of agents designed for navigating websites, which can be controlled through a conversational UI. To understand how to train those models, we examine gym environments designed for web agents, with a focus on LLM judges, which are the keys for unlocking environment-based training (such as Reinforcement Learning). We look at how those environments and judges can be combined to design a simple but effectively pipeline to distill frontier models into smaller models, and discuss how to extend them to long-horizon computer-use tasks. Finally, we conclude this talk with a discussion about the pace of improvements of agents capable of doing increasingly more complex work in browsers and on desktops, and muse about how the future of knowledge work might look like.