November 2nd, 2022

We are delighted to announce three new publications, by members of our team and collaborators, at AACL-IJCNLP, EMNLP Findings and the Eval4NLP Workshop @AACL-IJCNLP.

  • Juri Opitz and Anette Frank: SBERT studies SBERT studies MRMeaning Representations:
    Decomposing Sentence Embeddings into Explainable Semantic Features
    , to appear in AACL-IJCNLP 2022.

    We induce Semantically Structured Sentence BERT embeddings by isolating fine-grained meaning aspects in different sub-symbolic subspaces.
    This yields a highly efficient and explainable semantic similarity prediction system that preserves or outperforms SBERT's prediction accuracy.
  • Juri Opitz and Anette Frank: Better Smatch = Better Parser? AMR evaluation is not so simple anymore, to appear in the Eval4NLP Workshop @AACL-IJCNLP 2022:

    Powered by the development of large language models, semantic parsing systems have improved tremendously over the recent years. We show, however, that any further progress needs to be assessed more carefully and extensively, since evaluation with classic metrics cannot reliably detect parser quality differences anymore.
  • Constantin Eichenberg, Sidney Black, Samuel Weinbach, Letitia Parcalabescu and Anette Frank: MAGMA – Multimodal Augmentation of Generative Models through Adapter-based Finetuning.
    To appear in Findings of EMNLP.
    We present MAGMA, a vision & language model capable of open-ended text generation. We build MAGMA upon a frozen language model augmented with the visual modality using adapter-based finetuning.

Congratulations to all authors!