What question did this study set out to answer?

To introduce and evaluate a label-conditioned data augmentation framework that resolves token-label misalignment in low-resource named entity recognition.

January 1, 2022Open Access

MELM: Data Augmentation with Masked Entity Language Modeling for Low-Resource NER

View Full Paper

Ask AI

Bookmark

Authors

ZRZhou RanTianjin Research Institute of Water Transport Engineering XLXin LiJohns Hopkins University RHRuidan HeAlibaba Group (Cayman Islands)

Discussion

Member takes

Implication

Computational study demonstrates improved named entity recognition across low-resource languages via label-conditioned augmentation, indicating reduced token-label misalignment errors.

Key Points

To introduce and evaluate a label-conditioned data augmentation framework that resolves token-label misalignment in low-resource named entity recognition.
Developed Masked Entity Language Modeling (MELM) to inject entity category labels directly into sentence context, enabling models to predict masked tokens conditioned on their labels.
Combined MELM with cross-lingual codemixing when multilingual source data was accessible.
Evaluated the framework on monolingual, cross-lingual, and multilingual named entity recognition benchmarks across multiple low-resource sample levels.
MELM produced synthetic training text with novel, properly aligned entities that provided rich entity regularity knowledge.
Demonstrated substantial performance improvements across monolingual, cross-lingual, and multilingual named entity recognition tasks relative to baseline augmentation techniques.

Cite This Study

Ran et al. (2022) studied this question.

synapsesocial.com/papers/6a168638ba073c4305bc0627 https://doi.org/10.18653/v1/2022.acl-long.160

View Full Paper

Ask AI

Bookmark

Implication

Computational study demonstrates improved named entity recognition across low-resource languages via label-conditioned augmentation, indicating reduced token-label misalignment errors.

MELM: Data Augmentation with Masked Entity Language Modeling for Low-Resource NER

Discussion

Member takes

Implication

Key Points

Cite This Study

MELM: Data Augmentation with Masked Entity Language Modeling for Low-Resource NER

Discussion

Member takes

Implication

Key Points

Cite This Study

Also Consider

Also Consider