We are happy to announce that Sara Papi is attending ICLR 2026 in Rio de Janeiro, Brazil ๐ง๐ท.
She will present our latest work, MCIF: the first human-annotated crosslingual + multimodal benchmark for instruction following in the scientific domain.
In this paper, we evaluate how current LLMs, SpeechLLMs, VideoLLMs, and MLLMs handle instructions across:
๐น Text, speech, and video
๐น English, German, Italian, and Chinese
๐น Recognition, translation, QA, and summarization
๐น Short-form and long-form inputs
This is the result of a collaboration with Maike Zรผfle, Marco Gaido, Beatrice Savoldi, Danni Liu, Ioannis Douros, Luisa Bentivogli, and Jan Niehues from Machine Translation at FBK, Artificial Intelligence for Language Technologies (AI4LT), and Translated within the Meetween project!
More details about the paper and the poster session can be found here
