Willem Zuidema — accepted AI-conference papers

2025

PolyPythias: Stability and Outliers across Fifty Language Model Pre-Training Runs

Oskar van der Wal, Pietro Lesci, Max Müller-Eberstein, Naomi Saphra, Hailey Schoelkopf, Willem Zuidema +1

ICLR 2025poster

The stability of language model pre-training and its effects on downstream performance are still understudied. Prior work shows that the training process can yield significantly different results in response to slight variations in initial conditions, e.g., the random seed. Crucially, the research c…

Cited by 0 Source ↗PDF ↗code ↗Scholar ↗

2024

DecoderLens: Layerwise Interpretation of Encoder-Decoder Transformers

Anna Langedijk, Hosein Mohebbi, Gabriele Sarti, Willem Zuidema, Jaap Jumelet

NAACL 2024findings

In recent years, several interpretability methods have been proposed to interpret the inner workings of Transformer models at different levels of precision and complexity.In this work, we propose a simple but effective technique to analyze encoder-decoder Transformers. Our method, which we name Deco…

Cited by 14 Source ↗PDF ↗code ↗Scholar ↗

2024

Do Language Models Exhibit Human-like Structural Priming Effects?

Jaap Jumelet, Willem Zuidema, Arabella Sinclair

ACL 2024findings

We explore which linguistic factors—at the sentence and token level—play an important role in influencing language model predictions, and investigate whether these are reflective of results found in humans and human corpora (Gries and Kootstra, 2017). We make use of the structural priming paradigm—w…

Cited by 6 Source ↗PDF ↗code ↗Scholar ↗

2023

Feature Interactions Reveal Linguistic Structure in Language Models

Jaap Jumelet, Willem Zuidema

ACL 2023findings

We study feature interactions in the context of feature attribution methods for post-hoc interpretability. In interpretability research, getting to grips with feature interactions is increasingly recognised as an important challenge, because interacting features are key to the success of neural netw…

Cited by 10 Source ↗PDF ↗code ↗Scholar ↗

2023

Homophone Disambiguation Reveals Patterns of Context Mixing in Speech Transformers

Hosein Mohebbi, Grzegorz Chrupała, Willem Zuidema, Afra Alishahi

EMNLP 2023long main

Transformers have become a key architecture in speech processing, but our understanding of how they build up representations of acoustic and linguistic structure is limited. In this study, we address this gap by investigating how measures of 'context-mixing' developed for text models can be adapted…

Cited by 0 Source ↗code ↗Scholar ↗

2023

Transparency at the Source: Evaluating and Interpreting Language Models With Access to the True Distribution

Jaap Jumelet, Willem Zuidema

EMNLP 2023long findings

We present a setup for training, evaluating and interpreting neural language models, that uses artificial, language-like data. The data is generated using a massive probabilistic grammar (based on state-split PCFGs), that is itself derived from a large natural language corpus, but also provides us c…

Cited by 0 Source ↗code ↗Scholar ↗

2020

DoLFIn: Distributions over Latent Features for Interpretability

Phong Le, Willem Zuidema

COLING 2020main

Interpreting the inner workings of neural models is a key step in ensuring the robustness and trustworthiness of the models, but work on neural network interpretability typically faces a trade-off: either the models are too constrained to be very useful, or the solutions found by the models are too…

Cited by 0 Source ↗PDF ↗code ↗Scholar ↗