{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/crosslingual-document-embedding-as-reduced","title":"Crosslingual Document Embedding as Reduced-Rank Ridge Regression","arxiv_id":"1904.03922","date":"2019-04-08","proceeding":null,"authors":["Martin Josifoski","Ivan S. Paskov","Hristo S. Paskov","Martin Jaggi","Robert West"],"abstract":"There has recently been much interest in extending vector-based word\nrepresentations to multiple languages, such that words can be compared across\nlanguages. In this paper, we shift the focus from words to documents and\nintroduce a method for embedding documents written in any language into a\nsingle, language-independent vector space. For training, our approach leverages\na multilingual corpus where the same concept is covered in multiple languages\n(but not necessarily via exact translations), such as Wikipedia. Our method,\nCr5 (Crosslingual reduced-rank ridge regression), starts by training a\nridge-regression-based classifier that uses language-specific bag-of-word\nfeatures in order to predict the concept that a given document is about. We\nshow that, when constraining the learned weight matrix to be of low rank, it\ncan be factored to obtain the desired mappings from language-specific\nbags-of-words to language-independent embeddings. As opposed to most prior\nmethods, which use pretrained monolingual word vectors, postprocess them to\nmake them crosslingual, and finally average word vectors to obtain document\nvectors, Cr5 is trained end-to-end and is thus natively crosslingual as well as\ndocument-level. Moreover, since our algorithm uses the singular value\ndecomposition as its core operation, it is highly scalable. Experiments show\nthat our method achieves state-of-the-art performance on a crosslingual\ndocument retrieval task. Finally, although not trained for embedding sentences\nand words, it also achieves competitive performance on crosslingual sentence\nand word retrieval tasks.","url_abs":"http://arxiv.org/abs/1904.03922v1","url_pdf":"http://arxiv.org/pdf/1904.03922v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"crosslingual-document-embedding-as-reduced","repo_url":"https://github.com/epfl-dlab/Cr5","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"document-embedding","task_name":"Document Embedding"},{"task_slug":"retrieval","task_name":"Retrieval"},{"task_slug":"sentence","task_name":"Sentence"},{"task_slug":"regression-1","task_name":"regression"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1904.03922","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}