no code implementations • NAACL 2022 • Pieter Delobelle, Ewoenam Tokpo, Toon Calders, Bettina Berendt
We survey the literature on fairness metrics for pre-trained language models and experimentally evaluate compatibility, including both biases in language models and in their downstream tasks.
1 code implementation • 2 May 2024 • Dimitri Staufer, Frank Pallas, Bettina Berendt
Such an approach, however, is inadequate for the whistleblowing scenario since it neglects further re-identification potential in textual features, including writing style.
no code implementations • 26 Feb 2024 • David Hartmann, José Renato Laranjeira de Pereira, Chiara Streitbörger, Bettina Berendt
By considering the value of third-party audits and third-party data access in an audit ecosystem, we identify a regulatory gap in that the Artificial Intelligence Act does not provide access to data for researchers and civil society.
no code implementations • 5 Oct 2023 • François Remy, Pieter Delobelle, Bettina Berendt, Kris Demuynck, Thomas Demeester
This one-to-many token mapping improves tremendously the initialization of the embedding table for the target language.
no code implementations • 7 Mar 2023 • Bettina Berendt, Özgür Karadeniz, Sercan Kiyak, Stefan Mertens, Leen d'Haenens
Libraries are increasingly relying on computational methods, including methods from Artificial Intelligence (AI).
1 code implementation • 27 Feb 2023 • Jose M. Alvarez, Kristen M. Scott, Salvatore Ruggieri, Bettina Berendt
In uses of pre-trained machine learning models, it is a known issue that the target population in which the model is being deployed may not have been reflected in the source population with which the model was trained.
1 code implementation • 30 Jan 2023 • Ewoenam Tokpo, Pieter Delobelle, Bettina Berendt, Toon Calders
Considering that the end use of these language models is for downstream tasks like text classification, it is important to understand how these intrinsic bias mitigation strategies actually translate to fairness in downstream tasks and the extent of this.
no code implementations • 29 Dec 2022 • Ozgur Karadeniz, Bettina Berendt, Sercan Kiyak, Stefan Mertens, Leen d'Haenens
We describe a case study of the relative over- or under-representation of Belgian political parties between 1990 and 2020 in the English-language DBpedia, the Dutch-language DBpedia, and Wikidata, and highlight the many decisions needed with regard to the design of this data analysis and the assumptions behind it, as well as implications from the results.
no code implementations • 15 Nov 2022 • Pieter Delobelle, Thomas Winters, Bettina Berendt
To evaluate if our new model is a plug-in replacement for RobBERT, we introduce two additional criteria based on concept drift of existing tokens and alignment for novel tokens. We found that for certain language tasks this update results in a significant performance increase.
1 code implementation • 10 Jul 2022 • Pieter Delobelle, Bettina Berendt
Large pre-trained language models are successfully being used in a variety of tasks, across many languages.
no code implementations • 28 Apr 2022 • Pieter Delobelle, Thomas Winters, Bettina Berendt
We found that the performance of the models using the shuffled versus non-shuffled datasets is similar for most tasks and that randomly merging subsequent sentences in a corpus creates models that train faster and perform better on tasks with long sequences.
1 code implementation • 14 Dec 2021 • Pieter Delobelle, Ewoenam Kwaku Tokpo, Toon Calders, Bettina Berendt
We survey the existing literature on fairness metrics for pretrained language models and experimentally evaluate compatibility, including both biases in language models as in their downstream tasks.
no code implementations • 4 Nov 2021 • Bettina Berendt, Stefan Schiffner
When technology enters applications and processes with a long tradition of controversial societal debate, multi-faceted new ethical and legal questions arise.
1 code implementation • 20 Apr 2021 • Kristen Scott, Pieter Delobelle, Bettina Berendt
We classify seven months' worth of Belgian COVID-related Tweets using multilingual BERT and relate them to their governments' COVID measures.
1 code implementation • 14 May 2020 • Pieter Delobelle, Paul Temple, Gilles Perrouin, Benoît Frénay, Patrick Heymans, Bettina Berendt
These new examples are then used to retrain and improve the model in the first step.
1 code implementation • Findings of the Association for Computational Linguistics 2020 • Pieter Delobelle, Thomas Winters, Bettina Berendt
Training a Dutch BERT model thus has a lot of potential for a wide range of Dutch NLP tasks.
Ranked #1 on Sentiment Analysis on DBRD
no code implementations • 30 Oct 2019 • Pieter Delobelle, Bettina Berendt
Graphical emoji are ubiquitous in modern-day online conversations.
1 code implementation • ACL 2019 • Pieter Delobelle, Murilo Cunha, Eric Massip Cano, Jeroen Peperkamp, Bettina Berendt
Fallacies like the personal attack{---}also known as the ad hominem attack{---}are introduced in debates as an easy win, even though they provide no rhetorical contribution.
no code implementations • 30 Oct 2018 • Bettina Berendt
But what is the Common Good, and is it enough to want to be good?