Composition of Embeddings : Lessons from Statistical Relational Learning

Damien Sileo (1, 2) , Tim van de Cruys (1) , Camille Pradel (2) , Philippe Muller (1)

1 IRIT-MELODI - MEthodes et ingénierie des Langues, des Ontologies et du DIscours
2 Synapse Développement

Various NLP problems -- such as the prediction of sentence similarity, entailment, and discourse relations -- are all instances of the same general task: the modeling of semantic relations between a pair of textual elements. A popular model for such problems is to embed sentences into fixed size vectors, and use composition functions (e.g. concatenation or sum) of those vectors as features for the prediction. At the same time, composition of embeddings has been a main focus within the field of Statistical Relational Learning (SRL) whose goal is to predict relations between entities (typically from knowledge base triples). In this article, we show that previous work on relation prediction between texts implicitly uses compositions from baseline SRL models. We show that such compositions are not expressive enough for several tasks (e.g. natural language inference). We build on recent SRL models to address textual relational problems, showing that they are more expressive, and can alleviate issues from simpler compositions. The resulting models significantly improve the state of the art in both transferable sentence representation learning and relation prediction.

Mots clés: Relational learning; Embeddings

Domaines: Informatique et langage [cs.CL]

Damien Sileo, Tim van de Cruys, Camille Pradel, Philippe Muller. Composition of Embeddings : Lessons from Statistical Relational Learning. 8th Joint Conference on Lexical and Computational Semantics (SEM 2019), Jun 2019, Minneapolis, United States. pp.33-43. ⟨hal-02397476⟩