Do Sentence Embeddings Capture Discourse Properties of Sentences from Scientific Abstracts ?

Laurine Huber (1) , Chaker Memmadi (2) , Mathilde Dargnat (3, 4) , Yannick Toussaint (1)

1 ORPAILLEUR - Knowledge representation, reasonning
2 LORIA - NLPKD - Department of Natural Language Processing & Knowledge Discovery
3 ATILF - Analyse et Traitement Informatique de la Langue Française
4 L2C2 - Institut des Sciences cognitives Marc Jeannerod - Laboratoire sur le langage, le cerveau et la cognition

We introduce four tasks designed to determine which sentence encoders best capture discourse properties of sentences from scientific abstracts, namely coherence between clauses of a sentence, and discourse relations within sentences. We show that even if contextual en-coders such as BERT or SciBERT encodes the coherence in discourse units, they do not help to predict three discourse relations commonly used in scientific abstracts. We discuss what these results underline, namely that these discourse relations are based on particular phrasing that allow non-contextual encoders to perform well.

Domaines: Traitement du texte et du document; Intelligence artificielle [cs.AI]

Laurine Huber, Chaker Memmadi, Mathilde Dargnat, Yannick Toussaint. Do Sentence Embeddings Capture Discourse Properties of Sentences from Scientific Abstracts ?. CODI 2020 - EMNLP 1st Workshop on Computational Approaches to Discourse, Nov 2020, Punta Cana, Dominican Republic. ⟨hal-03000308⟩