GenBench
Organizers: Dieuwke Hupkes, Verna Dankers, Khuyagbaatar Batsuren, Koustuv Sinha, Amirhossein Kazemnejad, Christos Christodoulopoulos, Ryan Cotterell, Elia Bruni
"The ability to generalise well is often mentioned as one of the primary desiderata for models of natural language processing (NLP). However, how generalisation should be defined and evaluated, or when it is particularly important, is a far from trivial question. The GenBench workshop on generalisation (benchmarking) in NLP aims to provide a platform to discuss challenging questions related to generalisation in NLP and establish a shared platform for state-of-the-art generalisation testing. We invited submitters to contribute work discussing generalisation in NLP and also held a collaborative benchmarking task, for which we called for submissions of challenging generalisation tests. \\ \\
In this first edition of the workshop, we have 10 archival papers in our main track, 7 archival papers for our collaborative benchmarking track, and 6 extended abstracts. The workshop also provides a platform for the authors of 29 EMNLP findings paper related the workshop's topic to present their work as a poster at the workshop. In addition to poster sessions, we furthermore have three exciting invited speakers -- Adina Williams, Anna Rogers and Tatsunori Hashimoto. They will talk about challenges in evaluating LLMs, how to consider emergent properties from the perspective of generalisation, and evaluating generalisation in the era of instruction tuning, respectively. We will end the day with an exciting panel in which we discuss challenging questions related to generalisation.
The workshop would not have been possible without the dedication of the programme committee, whom we would like to thank for their contributions. We would also like to thank Amazon for their sponsorship of 5000 dollars, which we used to fund one of our invited speakers, to grant travel awards to allow participants that could otherwise not have attended to participate in the workshop, and to grant two awards, to the best submitted paper and best submitted benchmark. Lastly, we are grateful to our invited speakers, Adina Williams, Anna Rogers, and Tatsunori Hashimoto, for contributing to our programme."