Cross-lingual Multi-Level Adversarial Transfer to Enhance Low-Resource Name Tagging

Cross-lingual Multi-Level Adversarial Transfer to Enhance Low-Resource Name Tagging. Huang, L., Ji, H., & May, J. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 3823–3833, Minneapolis, Minnesota, June, 2019. Association for Computational Linguistics.

Paper abstract bibtex

We focus on improving name tagging for low-resource languages using annotations from related languages. Previous studies either directly project annotations from a source language to a target language using cross-lingual representations or use a shared encoder in a multitask network to transfer knowledge. These approaches inevitably introduce noise to the target language annotation due to mismatched source-target sentence structures. To effectively transfer the resources, we develop a new neural architecture that leverages multi-level adversarial transfer: (1) word-level adversarial training, which projects source language words into the same semantic space as those of the target language without using any parallel corpora or bilingual gazetteers, and (2) sentence-level adversarial training, which yields language-agnostic sequential features. Our neural architecture outperforms previous approaches on CoNLL data sets. Moreover, on 10 low-resource languages, our approach achieves up to 16% absolute F-score gain over all high-performing baselines on cross-lingual transfer without using any target-language resources.

@InProceedings{huang-ji-may:2019:N19-1,
  author    = {Huang, Lifu  and  Ji, Heng  and  May, Jonathan},
  title     = {Cross-lingual Multi-Level Adversarial Transfer to Enhance Low-Resource Name Tagging},
  booktitle = {Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)},
  month     = {June},
  year      = {2019},
  address   = {Minneapolis, Minnesota},
  publisher = {Association for Computational Linguistics},
  pages     = {3823--3833},
  abstract  = {We focus on improving name tagging for low-resource languages using annotations from related languages. Previous studies either directly project annotations from a source language to a target language using cross-lingual representations or use a shared encoder in a multitask network to transfer knowledge. These approaches inevitably introduce noise to the target language annotation due to mismatched source-target sentence structures. To effectively transfer the resources, we develop a new neural architecture that leverages multi-level adversarial transfer: (1) word-level adversarial training, which projects source language words into the same semantic space as those of the target language without using any parallel corpora or bilingual gazetteers, and (2) sentence-level adversarial training, which yields language-agnostic sequential features. Our neural architecture outperforms previous approaches on CoNLL data sets. Moreover, on 10 low-resource languages, our approach achieves up to 16\% absolute F-score gain over all high-performing baselines on cross-lingual transfer without using any target-language resources.},
  url       = {http://www.aclweb.org/anthology/N19-1383}
}

Downloads: 0

{"_id":"cBqfhFW8XR3Ki3ou7","bibbaseid":"huang-ji-may-crosslingualmultileveladversarialtransfertoenhancelowresourcenametagging-2019","author_short":["Huang, L.","Ji, H.","May, J."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","author":[{"propositions":[],"lastnames":["Huang"],"firstnames":["Lifu"],"suffixes":[]},{"propositions":[],"lastnames":["Ji"],"firstnames":["Heng"],"suffixes":[]},{"propositions":[],"lastnames":["May"],"firstnames":["Jonathan"],"suffixes":[]}],"title":"Cross-lingual Multi-Level Adversarial Transfer to Enhance Low-Resource Name Tagging","booktitle":"Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)","month":"June","year":"2019","address":"Minneapolis, Minnesota","publisher":"Association for Computational Linguistics","pages":"3823–3833","abstract":"We focus on improving name tagging for low-resource languages using annotations from related languages. Previous studies either directly project annotations from a source language to a target language using cross-lingual representations or use a shared encoder in a multitask network to transfer knowledge. These approaches inevitably introduce noise to the target language annotation due to mismatched source-target sentence structures. To effectively transfer the resources, we develop a new neural architecture that leverages multi-level adversarial transfer: (1) word-level adversarial training, which projects source language words into the same semantic space as those of the target language without using any parallel corpora or bilingual gazetteers, and (2) sentence-level adversarial training, which yields language-agnostic sequential features. Our neural architecture outperforms previous approaches on CoNLL data sets. Moreover, on 10 low-resource languages, our approach achieves up to 16% absolute F-score gain over all high-performing baselines on cross-lingual transfer without using any target-language resources.","url":"http://www.aclweb.org/anthology/N19-1383","bibtex":"@InProceedings{huang-ji-may:2019:N19-1,\n author = {Huang, Lifu and Ji, Heng and May, Jonathan},\n title = {Cross-lingual Multi-Level Adversarial Transfer to Enhance Low-Resource Name Tagging},\n booktitle = {Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers)},\n month = {June},\n year = {2019},\n address = {Minneapolis, Minnesota},\n publisher = {Association for Computational Linguistics},\n pages = {3823--3833},\n abstract = {We focus on improving name tagging for low-resource languages using annotations from related languages. Previous studies either directly project annotations from a source language to a target language using cross-lingual representations or use a shared encoder in a multitask network to transfer knowledge. These approaches inevitably introduce noise to the target language annotation due to mismatched source-target sentence structures. To effectively transfer the resources, we develop a new neural architecture that leverages multi-level adversarial transfer: (1) word-level adversarial training, which projects source language words into the same semantic space as those of the target language without using any parallel corpora or bilingual gazetteers, and (2) sentence-level adversarial training, which yields language-agnostic sequential features. Our neural architecture outperforms previous approaches on CoNLL data sets. Moreover, on 10 low-resource languages, our approach achieves up to 16\\% absolute F-score gain over all high-performing baselines on cross-lingual transfer without using any target-language resources.},\n url = {http://www.aclweb.org/anthology/N19-1383}\n}\n\n","author_short":["Huang, L.","Ji, H.","May, J."],"key":"huang-ji-may:2019:N19-1","id":"huang-ji-may:2019:N19-1","bibbaseid":"huang-ji-may-crosslingualmultileveladversarialtransfertoenhancelowresourcenametagging-2019","role":"author","urls":{"Paper":"http://www.aclweb.org/anthology/N19-1383"},"metadata":{"authorlinks":{}}},"bibtype":"inproceedings","biburl":"https://jonmay.github.io/webpage/cutelabname/cutelabname.bib","dataSources":["ZdhKtP2cSp3Aki2ge","X5WBAKQabka5TW5z7","BnZgtH7HDESgbxKxt","hbZSwot2msWk92m5B","fcWjcoAgajPvXWcp7","GvHfaAWP6AfN6oLQE","j3Qzx9HAAC6WtJDHS","5eM3sAccSEpjSDHHQ"],"keywords":[],"search_terms":["cross","lingual","multi","level","adversarial","transfer","enhance","low","resource","name","tagging","huang","ji","may"],"title":"Cross-lingual Multi-Level Adversarial Transfer to Enhance Low-Resource Name Tagging","year":2019}