Powered by OpenAIRE graph
Found an issue? Give us feedback
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/ HAL Sorbonne Univers...arrow_drop_down
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/
HAL Sorbonne Université
Doctoral thesis . 2022
addClaim

This Research product is the result of merged Research products in OpenAIRE.

You have already added 0 works in your ORCID record related to the merged Research product.

Contextualization and Generalization in Entity and Relation Extraction

Authors: Taillé, Bruno;

Contextualization and Generalization in Entity and Relation Extraction

Abstract

Depuis 2018, le transfert de modèles de langue pré-entraînés et la préservation de leurs capacités de contextualisation ont permis d'atteindre des performances sans précédent sur les benchmarks de Traitement Automatique des Langues. Cependant, alors que ces modèles atteignent des scores impressionnants, leurs capacités de compréhension apparaissent assez peu développées, révélant les limites des jeux de données de référence pour identifier leurs facteurs de performance et pour mesurer précisément leur capacité de compréhension. Dans cette thèse, nous étudions la généralisation à des faits inconnus par des modèles état de l'art en Extraction d'Entités Nommées et de Relations. En effet, les benchmarks traditionnels présentent un recoupement lexical important entre les mentions et les relations utilisées pour l'entraînement et l'évaluation des modèles. Au contraire, l'intérêt principal de l'Extraction d'Information est d'extraire des informations inconnues jusqu'alors. Nous proposons plusieurs études pour séparer les performances selon le recoupement des mentions et des relations avec le jeu d'entraînement. Nous constatons que les modèles de langage pré-entraînés sont principalement bénéfiques pour détecter les mentions non connues, en particulier dans des genres de textes nouveaux. Bien que cela les rende adaptés à des cas d'utilisation concrets, il existe toujours un écart de performance important entre les mentions connues et inconnues. En particulier, même les modèles d'Extraction d'Entités et de Relations les plus récents reposent sur une heuristique de rétention superficielle, basant plus leur prédiction sur les arguments des relations que sur leur contexte.

Since 2018, the transfer of entire pretrained Language Models and the preservation of their contextualization capacities enabled to reach unprecedented performance on virtually every Natural Language Processing benchmark. However, as models reach such impressive scores, their comprehension abilities still appear as shallow, which reveals limitations of benchmarks to provide useful insights on their factors of performance and to accurately measure understanding capabilities. In this thesis, we study the behaviour of state-of-the-art models regarding generalization to facts unseen during training in Entity and Relation Extraction. Indeed, traditional benchmarks present important lexical overlap between mentions and relations used for training and evaluating models, whereas the main interest of Information Extraction is to extract previously unknown information. We propose studies to separate performance based on mention and relation overlap with the training set and find that pretrained Language Models are mainly beneficial to detect unseen mentions, in particular out-of-domain. While this makes them suited for real use cases, there is still a gap in performance between seen and unseen mentions that hurts generalization to new facts. In particular, even state-of-the-art ERE models rely on a shallow retention heuristic, basing their prediction more on arguments surface forms than context.

Country
France
Keywords

[INFO.INFO-AI] Computer Science [cs]/Artificial Intelligence [cs.AI], Apprentissage profond, Reconnaissance d'entités nommées, Information extraction, Natural language processing, Deep learning, Extraction de relations, Traitement automatique des langues, Extraction d'information, Généralisation

  • BIP!
    Impact byBIP!
    selected citations
    These citations are derived from selected sources.
    This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    0
    popularity
    This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
    Average
    influence
    This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    Average
    impulse
    This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
    Average
Powered by OpenAIRE graph
Found an issue? Give us feedback
selected citations
These citations are derived from selected sources.
This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Citations provided by BIP!
popularity
This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
BIP!Popularity provided by BIP!
influence
This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Influence provided by BIP!
impulse
This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
BIP!Impulse provided by BIP!
0
Average
Average
Average
Green
Related to Research communities