A Study on the Impact of Intradomain Finetuning of Deep Language Models for Legal Named Entity Recognition in Portuguese

Publikation: Beiträge in SammelwerkenAufsätze in KonferenzbändenForschungbegutachtet

Standard

A Study on the Impact of Intradomain Finetuning of Deep Language Models for Legal Named Entity Recognition in Portuguese. / Bonifacio, Luiz Henrique; Vilela, Paulo Arantes; Lobato, Gustavo Rocha et al.
Intelligent Systems: 9th Brazilian Conference, BRACIS 2020, Rio Grande, Brazil, October 20–23, 2020, Proceedings, Part I. Hrsg. / Ricardo Cerri; Ronaldo C. Prati. Cham: Springer Nature Switzerland AG, 2020. S. 648-662 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Band 12319 LNAI).

Publikation: Beiträge in SammelwerkenAufsätze in KonferenzbändenForschungbegutachtet

Harvard

Bonifacio, LH, Vilela, PA, Lobato, GR & Fernandes, ER 2020, A Study on the Impact of Intradomain Finetuning of Deep Language Models for Legal Named Entity Recognition in Portuguese. in R Cerri & RC Prati (Hrsg.), Intelligent Systems: 9th Brazilian Conference, BRACIS 2020, Rio Grande, Brazil, October 20–23, 2020, Proceedings, Part I. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Bd. 12319 LNAI, Springer Nature Switzerland AG, Cham, S. 648-662, Brazilian Conference on Intelligent Systems - BRACIS 2020, Rio Grande, Brasilien, 20.10.20. https://doi.org/10.1007/978-3-030-61377-8_46

APA

Bonifacio, L. H., Vilela, P. A., Lobato, G. R., & Fernandes, E. R. (2020). A Study on the Impact of Intradomain Finetuning of Deep Language Models for Legal Named Entity Recognition in Portuguese. In R. Cerri, & R. C. Prati (Hrsg.), Intelligent Systems: 9th Brazilian Conference, BRACIS 2020, Rio Grande, Brazil, October 20–23, 2020, Proceedings, Part I (S. 648-662). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Band 12319 LNAI). Springer Nature Switzerland AG. https://doi.org/10.1007/978-3-030-61377-8_46

Vancouver

Bonifacio LH, Vilela PA, Lobato GR, Fernandes ER. A Study on the Impact of Intradomain Finetuning of Deep Language Models for Legal Named Entity Recognition in Portuguese. in Cerri R, Prati RC, Hrsg., Intelligent Systems: 9th Brazilian Conference, BRACIS 2020, Rio Grande, Brazil, October 20–23, 2020, Proceedings, Part I. Cham: Springer Nature Switzerland AG. 2020. S. 648-662. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-61377-8_46

Bibtex

@inbook{aefc1e9b06a54d238646d433c41dcb30,
title = "A Study on the Impact of Intradomain Finetuning of Deep Language Models for Legal Named Entity Recognition in Portuguese",
abstract = "Deep language models, like ELMo, BERT and GPT, have achieved impressive results on several natural language tasks. These models are pretrained on large corpora of unlabeled general domain text and later supervisedly trained on downstream tasks. An optional step consists of finetuning the language model on a large intradomain corpus of unlabeled text, before training it on the final task. This aspect is not well explored in the current literature. In this work, we investigate the impact of this step on named entity recognition (NER) for Portuguese legal documents. We explore different scenarios considering two deep language architectures (ELMo and BERT), four unlabeled corpora and three legal NER tasks for the Portuguese language. Experimental findings show a significant improvement on performance due to language model finetuning on intradomain text. We also evaluate the finetuned models on two general-domain NER tasks, in order to understand whether the aforementioned improvements were really due to domain similarity or simply due to more training data. The achieved results also indicate that finetuning on a legal domain corpus hurts performance on the general-domain NER tasks. Additionally, our BERT model, finetuned on a legal corpus, significantly improves on the state-of-the-art performance on the LeNER-Br corpus, a Portuguese language NER corpus for the legal domain.",
keywords = "Deep learning, Named entity recognition, Natural language processing, Informatics, Business informatics",
author = "Bonifacio, {Luiz Henrique} and Vilela, {Paulo Arantes} and Lobato, {Gustavo Rocha} and Fernandes, {Eraldo Rezende}",
year = "2020",
doi = "10.1007/978-3-030-61377-8_46",
language = "English",
isbn = "978-3-030-61376-1",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Nature Switzerland AG",
pages = "648--662",
editor = "Ricardo Cerri and Prati, {Ronaldo C.}",
booktitle = "Intelligent Systems",
address = "Switzerland",
note = "Brazilian Conference on Intelligent Systems - BRACIS 2020 ; Conference date: 20-10-2020 Through 23-10-2020",
url = "http://www2.sbc.org.br/bracis2020/#:~:text=The%209th%20Brazilian%20Conference%20on,%2C%2020%20to%2023%2C%202020.",

}

RIS

TY - CHAP

T1 - A Study on the Impact of Intradomain Finetuning of Deep Language Models for Legal Named Entity Recognition in Portuguese

AU - Bonifacio, Luiz Henrique

AU - Vilela, Paulo Arantes

AU - Lobato, Gustavo Rocha

AU - Fernandes, Eraldo Rezende

N1 - Conference code: 9

PY - 2020

Y1 - 2020

N2 - Deep language models, like ELMo, BERT and GPT, have achieved impressive results on several natural language tasks. These models are pretrained on large corpora of unlabeled general domain text and later supervisedly trained on downstream tasks. An optional step consists of finetuning the language model on a large intradomain corpus of unlabeled text, before training it on the final task. This aspect is not well explored in the current literature. In this work, we investigate the impact of this step on named entity recognition (NER) for Portuguese legal documents. We explore different scenarios considering two deep language architectures (ELMo and BERT), four unlabeled corpora and three legal NER tasks for the Portuguese language. Experimental findings show a significant improvement on performance due to language model finetuning on intradomain text. We also evaluate the finetuned models on two general-domain NER tasks, in order to understand whether the aforementioned improvements were really due to domain similarity or simply due to more training data. The achieved results also indicate that finetuning on a legal domain corpus hurts performance on the general-domain NER tasks. Additionally, our BERT model, finetuned on a legal corpus, significantly improves on the state-of-the-art performance on the LeNER-Br corpus, a Portuguese language NER corpus for the legal domain.

AB - Deep language models, like ELMo, BERT and GPT, have achieved impressive results on several natural language tasks. These models are pretrained on large corpora of unlabeled general domain text and later supervisedly trained on downstream tasks. An optional step consists of finetuning the language model on a large intradomain corpus of unlabeled text, before training it on the final task. This aspect is not well explored in the current literature. In this work, we investigate the impact of this step on named entity recognition (NER) for Portuguese legal documents. We explore different scenarios considering two deep language architectures (ELMo and BERT), four unlabeled corpora and three legal NER tasks for the Portuguese language. Experimental findings show a significant improvement on performance due to language model finetuning on intradomain text. We also evaluate the finetuned models on two general-domain NER tasks, in order to understand whether the aforementioned improvements were really due to domain similarity or simply due to more training data. The achieved results also indicate that finetuning on a legal domain corpus hurts performance on the general-domain NER tasks. Additionally, our BERT model, finetuned on a legal corpus, significantly improves on the state-of-the-art performance on the LeNER-Br corpus, a Portuguese language NER corpus for the legal domain.

KW - Deep learning

KW - Named entity recognition

KW - Natural language processing

KW - Informatics

KW - Business informatics

UR - http://www.scopus.com/inward/record.url?scp=85094121387&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-61377-8_46

DO - 10.1007/978-3-030-61377-8_46

M3 - Article in conference proceedings

AN - SCOPUS:85094121387

SN - 978-3-030-61376-1

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 648

EP - 662

BT - Intelligent Systems

A2 - Cerri, Ricardo

A2 - Prati, Ronaldo C.

PB - Springer Nature Switzerland AG

CY - Cham

T2 - Brazilian Conference on Intelligent Systems - BRACIS 2020

Y2 - 20 October 2020 through 23 October 2020

ER -

DOI

Zuletzt angesehen

Publikationen

  1. The State and Healthcare
  2. The Multiple Self Objection to the Prudential Lifespan Account
  3. The new European database for chemicals of concern
  4. Inequality in the Transition from Primary to Secondary School
  5. Energy-aware system design for autonomous wireless sensor nodes
  6. Discussion report part 1
  7. Keep calm and follow the news
  8. Organisationen hacken
  9. Evidence-Based Entrepreneurship
  10. Report on the relative strengths and weaknesses of the United States in PISA 2012 mathematics
  11. Integrated Concept for the Selection of Process-improving and Competence-increasing Methods for the Shopfloor
  12. Introduction to the Handbook on the Politics of Taxation
  13. The challenge of managing multiple species at multiple scales
  14. Herrmann Bahr - Renaissance
  15. Effect of laser peening process parameters and sequences on residual stress profiles
  16. iTaukei ways of knowing and managing mangroves for ecosystem-based adaptation
  17. [U]topische Körper in der Adoleszenz
  18. Tourists’ Weather Perceptions and Weather Related Behavior
  19. Earnings less risk-free interest charge (ERIC) and stock returns: ERIC’s relative and incremental information content in a European sample
  20. Investigation of temperature evolution and flash formation at AA5083 studs during friction surfacing
  21. Sustainability-oriented technology exploration: managerial values, ambidextrous design, and separation drift
  22. Mindfulness-based interventions in the workplace
  23. Global Tax Governance: What is Wrong With It and How to Fix It by Peter Dietsch and Thomas Rixen (eds). Colchester: ECPR Press, 2016
  24. The continued relevance of compromissory clauses as a source of ICJ jurisdiction
  25. unplugged
  26. Consequence evaluations and moral concerns about climate change
  27. Corporate hedging for different production cycles with the wavelet-approach
  28. Goffman’s Return to Las Vegas
  29. Cross-Channel Real-Time Response Analysis
  30. Die Unternehmergesellschaft
  31. Differences in labor supply to monopsonistic firms and the gender pay gap
  32. Individuelle Beteiligung am Unterrichtsgespräch in Grundschulklassen
  33. Ballonflüge