What you can cram into a single $&!#* vector: Probing sentence embeddings for linguistic properties}. Conneau, A., Kruszewski, G., Lample, G., Barrault, L., & Baroni, M. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), volume 1, pages 2126–2136, Stroudsburg, PA, USA, 2018. Association for Computational Linguistics. ![What you can cram into a single $&!#* vector: Probing sentence embeddings for linguistic properties} [link]](https://bibbase.org/img/filetypes/link.svg "link")
Paper doi abstract bibtex Although much effort has recently been devoted to training high-quality sentence embeddings, we still have a poor understanding of what they are capturing. "Downstream" tasks, often based on sentence classification, are commonly used to evaluate the quality of sentence representations. The complexity of the tasks makes it however difficult to infer what kind of information is present in the representations. We introduce here 10 probing tasks designed to capture simple linguistic features of sentences, and we use them to study embeddings generated by three different encoders trained in eight distinct ways, uncovering intriguing properties of both encoders and training methods.
@inproceedings{Conneau2018,
abstract = {Although much effort has recently been devoted to training high-quality sentence embeddings, we still have a poor understanding of what they are capturing. "Downstream" tasks, often based on sentence classification, are commonly used to evaluate the quality of sentence representations. The complexity of the tasks makes it however difficult to infer what kind of information is present in the representations. We introduce here 10 probing tasks designed to capture simple linguistic features of sentences, and we use them to study embeddings generated by three different encoders trained in eight distinct ways, uncovering intriguing properties of both encoders and training methods.},
address = {Stroudsburg, PA, USA},
author = {Conneau, Alexis and Kruszewski, German and Lample, Guillaume and Barrault, Lo{\"{i}}c and Baroni, Marco},
booktitle = {Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
doi = {10.18653/v1/P18-1198},
file = {:Users/shanest/Documents/Library/Conneau et al/Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1 Long Papers)/Conneau et al. - 2018 - What you can cram into a single $&!# vector Probing sentence embeddings for linguistic properties.pdf:pdf},
isbn = {9781948087322},
keywords = {method: diagnostic classifier,method: pre-training task comparison,method: sentence-level,phenomenon: various},
pages = {2126--2136},
publisher = {Association for Computational Linguistics},
title = {{What you can cram into a single $&!#* vector: Probing sentence embeddings for linguistic properties}},
url = {http://aclweb.org/anthology/P18-1198},
volume = {1},
year = {2018}
}
Downloads: 0
{"_id":"vc22AQuqerxcJJyCy","bibbaseid":"conneau-kruszewski-lample-barrault-baroni-whatyoucancramintoasinglevectorprobingsentenceembeddingsforlinguisticproperties-2018","authorIDs":[],"author_short":["Conneau, A.","Kruszewski, G.","Lample, G.","Barrault, L.","Baroni, M."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","abstract":"Although much effort has recently been devoted to training high-quality sentence embeddings, we still have a poor understanding of what they are capturing. \"Downstream\" tasks, often based on sentence classification, are commonly used to evaluate the quality of sentence representations. The complexity of the tasks makes it however difficult to infer what kind of information is present in the representations. We introduce here 10 probing tasks designed to capture simple linguistic features of sentences, and we use them to study embeddings generated by three different encoders trained in eight distinct ways, uncovering intriguing properties of both encoders and training methods.","address":"Stroudsburg, PA, USA","author":[{"propositions":[],"lastnames":["Conneau"],"firstnames":["Alexis"],"suffixes":[]},{"propositions":[],"lastnames":["Kruszewski"],"firstnames":["German"],"suffixes":[]},{"propositions":[],"lastnames":["Lample"],"firstnames":["Guillaume"],"suffixes":[]},{"propositions":[],"lastnames":["Barrault"],"firstnames":["Loïc"],"suffixes":[]},{"propositions":[],"lastnames":["Baroni"],"firstnames":["Marco"],"suffixes":[]}],"booktitle":"Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","doi":"10.18653/v1/P18-1198","file":":Users/shanest/Documents/Library/Conneau et al/Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1 Long Papers)/Conneau et al. - 2018 - What you can cram into a single $&!# vector Probing sentence embeddings for linguistic properties.pdf:pdf","isbn":"9781948087322","keywords":"method: diagnostic classifier,method: pre-training task comparison,method: sentence-level,phenomenon: various","pages":"2126–2136","publisher":"Association for Computational Linguistics","title":"What you can cram into a single $&!#* vector: Probing sentence embeddings for linguistic properties}","url":"http://aclweb.org/anthology/P18-1198","volume":"1","year":"2018","bibtex":"@inproceedings{Conneau2018,\nabstract = {Although much effort has recently been devoted to training high-quality sentence embeddings, we still have a poor understanding of what they are capturing. \"Downstream\" tasks, often based on sentence classification, are commonly used to evaluate the quality of sentence representations. The complexity of the tasks makes it however difficult to infer what kind of information is present in the representations. We introduce here 10 probing tasks designed to capture simple linguistic features of sentences, and we use them to study embeddings generated by three different encoders trained in eight distinct ways, uncovering intriguing properties of both encoders and training methods.},\naddress = {Stroudsburg, PA, USA},\nauthor = {Conneau, Alexis and Kruszewski, German and Lample, Guillaume and Barrault, Lo{\\\"{i}}c and Baroni, Marco},\nbooktitle = {Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},\ndoi = {10.18653/v1/P18-1198},\nfile = {:Users/shanest/Documents/Library/Conneau et al/Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1 Long Papers)/Conneau et al. - 2018 - What you can cram into a single $&!# vector Probing sentence embeddings for linguistic properties.pdf:pdf},\nisbn = {9781948087322},\nkeywords = {method: diagnostic classifier,method: pre-training task comparison,method: sentence-level,phenomenon: various},\npages = {2126--2136},\npublisher = {Association for Computational Linguistics},\ntitle = {{What you can cram into a single $&!#* vector: Probing sentence embeddings for linguistic properties}},\nurl = {http://aclweb.org/anthology/P18-1198},\nvolume = {1},\nyear = {2018}\n}\n","author_short":["Conneau, A.","Kruszewski, G.","Lample, G.","Barrault, L.","Baroni, M."],"key":"Conneau2018","id":"Conneau2018","bibbaseid":"conneau-kruszewski-lample-barrault-baroni-whatyoucancramintoasinglevectorprobingsentenceembeddingsforlinguisticproperties-2018","role":"author","urls":{"Paper":"http://aclweb.org/anthology/P18-1198"},"keyword":["method: diagnostic classifier","method: pre-training task comparison","method: sentence-level","phenomenon: various"],"metadata":{"authorlinks":{}},"downloads":0},"bibtype":"inproceedings","biburl":"https://www.shane.st/teaching/575/win20/MachineLearning-interpretability.bib","creationDate":"2020-01-05T04:04:02.870Z","downloads":0,"keywords":["method: diagnostic classifier","method: pre-training task comparison","method: sentence-level","phenomenon: various"],"search_terms":["cram","single","vector","probing","sentence","embeddings","linguistic","properties","conneau","kruszewski","lample","barrault","baroni"],"title":"What you can cram into a single $&!#* vector: Probing sentence embeddings for linguistic properties}","year":2018,"dataSources":["okYcdTpf4JJ2zkj7A"]}