AdvPicker: Effectively Leveraging Unlabeled Data via Adversarial Discriminator for Cross-Lingual NER

Chen, Weile; Jiang, Huiqiang; Wu, Qianhui; Karlsson, Börje F.; Guan, Yi

doi:10.18653/v1/2021.acl-long.61

Cited by 19 publications

(18 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Most recent studies focus on the bilingual transfer case and can be grouped into three categories: (1) Data-based approaches utilize machine translation and label projection (Jain et al 2019;Yang et al 2022) to create pseudo-training data for the target language. (2) Feature-based approaches rely on features alignment to diminish the language shift (Chen et al 2021;Ge et al 2023). (3) The distillation-based methods (Wu et al 2020b;Liang et al 2021;Ma et al 2022) enable the student network to gain task knowledge from soft labels predicted by the teacher network on the target language.…”

Section: Related Workmentioning

confidence: 99%

“…Recent methods, driven by the weighted combining rule, learn language-specific classifiers and obtain a weighted ensemble prediction for target samples. For instance, MulTS (Chen et al 2021) trains one specific network for each source language separately to derive the combined final prediction, which causes a substantial computational burden. MAN (Chen et al 2019) and G-MOE (Jin et al 2022) force multiple source languages to share one same encoder, avoiding the huge parameter computation problem.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

DA-Net: A Disentangled and Adaptive Network for Multi-Source Cross-Lingual Transfer Learning

Ge,

Hu,

et al. 2024

AAAI

View full text Add to dashboard Cite

Multi-Source cross-lingual transfer learning deals with the transfer of task knowledge from multiple labelled source languages to an unlabeled target language under the language shift. Existing methods typically focus on weighting the predictions produced by language-specific classifiers of different sources that follow a shared encoder. However, all source languages share the same encoder, which is updated by all these languages. The extracted representations inevitably contain different source languages' information, which may disturb the learning of the language-specific classifiers. Additionally, due to the language gap, language-specific classifiers trained with source labels are unable to make accurate predictions for the target language. Both facts impair the model's performance. To address these challenges, we propose a Disentangled and Adaptive Network ~(DA-Net). Firstly, we devise a feedback-guided collaborative disentanglement method that seeks to purify input representations of classifiers, thereby mitigating mutual interference from multiple sources. Secondly, we propose a class-aware parallel adaptation method that aligns class-level distributions for each source-target language pair, thereby alleviating the language pairs' language gap. Experimental results on three different tasks involving 38 languages validate the effectiveness of our approach.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

DA-Net: A Disentangled and Adaptive Network for Multi-Source Cross-Lingual Transfer Learning

Ge,

Hu,

et al. 2024

AAAI

View full text Add to dashboard Cite

show abstract

“…Extensive works [29][30][31][32] have shown that dual-branch network can not only reduce computational cost but also have the potential to improve task performance. Chen et al [33] proposed a dual-branch adversarial learning framework in the task of named entity recognition, where one branch learned entity domain knowledge, and the other branch captured better features, the two aspects worked together to improve the results. Chen et al [34] combined a Bayesian network based on medical knowledge graph augmentation with a multibranch entity-aware convolutional neural network.…”

Section: Related Workmentioning

confidence: 99%

TCM Prescription Generation via Knowledge Source Guidance Network Combined with Herbal Candidate Mechanism

Hou

Song

Zhao

et al. 2023

Computational and Mathematical Methods in Medicine

View full text Add to dashboard Cite

Traditional Chinese medicine (TCM) prescriptions have made great contributions to the treatment of diseases and health preservation. To alleviate the shortage of TCM resources and improve the professionalism of automatically generated prescriptions, this paper deeply explores the connection between symptoms and herbs through deep learning technology, and realizes the automatic generation of TCM prescriptions. Particularly, this paper considers the significance of referring to similar underlying prescriptions as herbal candidates in the TCM prescribing process. Moreover, this paper incorporates the idea of referring to the potential guidance information of corresponding prescriptions when model extracts symptoms representations. To provide a reference for inexperienced young TCM doctors when they prescribe, this paper proposes a dual-branch guidance strategy combined with candidate attention model (DGSCAM) to automatically generate TCM prescriptions based on symptoms text. The format of the data used this paper is the “symptoms-prescription” data pair. The specific method is as follows. First, DGSCAM realizes the extraction of key information of prescription-guided symptoms through a dual-branch network. Then, herbal candidates in the form of prescriptions that can treat symptoms are proposed in view of the compatibility knowledge of TCM prescriptions. To our knowledge, this is the first attempt to use prescriptions as herbal candidates in the prescription generation process. We conduct extensive experiments on a mixed public and clinical dataset, achieving 37.39% precision, 25.04% recall, and 29.99% F1 score, with an average doctor score of 7.7 out of 10. The experimental results show that our proposed model is valid and can generate more specialized TCM prescriptions than the baseline models. The DGSCAM developed by us has broad application scenarios and greatly promotes the research on intelligent TCM prescribing.

show abstract

“…In other words, the adversarial techniques encourage the model to align the representations of English documents and their translations of target languages. Furthermore, the adversarial learning framework can leverage the unlabeled data and improve the result [135], where a discriminator picks less language-dependent target language data according to the similarity to the source language. • Multilingual Pretrained Language Model.…”

Section: Future Directionmentioning

confidence: 99%

A Survey on Arabic Named Entity Recognition: Past, Recent Advances, and Future Trends

Qu¹,

Gu²,

Xia³

et al. 2023

Preprint

View full text Add to dashboard Cite

As more and more Arabic texts emerged on the Internet, extracting important information from these Arabic texts is especially useful. As a fundamental technology, Named entity recognition (NER) serves as the core component in information extraction technology, while also playing a critical role in many other Natural Language Processing (NLP) systems, such as question answering and knowledge graph building. In this paper, we provide a comprehensive review of the development of Arabic NER, especially the recent advances in deep learning and pre-trained language model. Specifically, we first introduce the background of Arabic NER, including the characteristics of Arabic and existing resources for Arabic NER. Then, we systematically review the development of Arabic NER methods. Traditional Arabic NER systems focus on feature engineering and designing domain-specific rules. In recent years, deep learning methods achieve significant progress by representing texts via continuous vector representations. With the growth of pre-trained language model, Arabic NER yields better performance. Finally, we conclude the method gap between Arabic NER and NER methods from other languages, which helps outline future directions for Arabic NER.

show abstract

AdvPicker: Effectively Leveraging Unlabeled Data via Adversarial Discriminator for Cross-Lingual NER

Cited by 19 publications

References 27 publications

DA-Net: A Disentangled and Adaptive Network for Multi-Source Cross-Lingual Transfer Learning

DA-Net: A Disentangled and Adaptive Network for Multi-Source Cross-Lingual Transfer Learning

TCM Prescription Generation via Knowledge Source Guidance Network Combined with Herbal Candidate Mechanism

A Survey on Arabic Named Entity Recognition: Past, Recent Advances, and Future Trends

Contact Info

Product

Resources

About