Fraunhofer-Gesellschaft

Publica

Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

NAT: Noise-Aware Training for Robust Neural Sequence Labeling

 
: Namysl, Marcin; Behnke, Sven; Köhler, Joachim

:
Fulltext urn:nbn:de:0011-n-5969782 (1.2 MByte PDF)
MD5 Fingerprint: f1773b09d7a9a9bd23593f3ef15f2783
(CC) by
Created on: 30.7.2020


Association for Computational Linguistics -ACL-:
ACL 2020, 58th Annual Meeting of the Association for Computational Linguistics. Proceedings. Online resource : July 5 - 10, 2020, virtual conference
Stroudsburg/Pa.: ACL, 2020
https://www.aclweb.org/anthology/volumes/2020.acl-main/
ISBN: 978-1-952148-25-5
pp.1501-1517
Association for Computational Linguistics (ACL Annual Meeting) <58, 2020, Online>
English
Conference Paper, Electronic Publication
Fraunhofer IAIS ()
robustness; sequence labeling; data augmentation; stability training; Named Entity Recognition (NER); Optical Character Recognition (OCR); information extraction; Natural Language Processing (NLP)

Abstract
Sequence labeling systems should perform reliably not only under ideal conditions but also with corrupted inputs as these systems often process user-generated text or follow an error-prone upstream component. To this end, we formulate the noisy sequence labeling problem, where the input may undergo an unknown noising process and propose two Noise-Aware Training (NAT) objectives that improve robustness of sequence labeling performed on perturbed input: Our data augmentation method trains a neural model using a mixture of clean and noisy samples, whereas our stability training algorithm encourages the model to create a noise-invariant latent representation. We employ a vanilla noise model at training time. For evaluation, we use both the original data and its variants perturbed with real O CR errors and misspellings. Extensive experiments on English and German named entity recognition benchmarks confirmed that NAT consistently improved robustness of popular sequence labeling models, preserving accuracy on the original input. We make our code and data publicly available for the research community.

: http://publica.fraunhofer.de/documents/N-596978.html