{"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/lost-in-machine-translation-a-method-to","title":"Lost in Machine Translation: A Method to Reduce Meaning Loss","arxiv_id":"1902.09514","date":"2019-02-25","proceeding":"NAACL 2019 6","authors":["Reuben Cohn-Gordon","Noah Goodman"],"abstract":"A desideratum of high-quality translation systems is that they preserve\nmeaning, in the sense that two sentences with different meanings should not\ntranslate to one and the same sentence in another language. However,\nstate-of-the-art systems often fail in this regard, particularly in cases where\nthe source and target languages partition the \"meaning space\" in different\nways. For instance, \"I cut my finger.\" and \"I cut my finger off.\" describe\ndifferent states of the world but are translated to French (by both Fairseq and\nGoogle Translate) as \"Je me suis coupe le doigt.\", which is ambiguous as to\nwhether the finger is detached. More generally, translation systems are\ntypically many-to-one (non-injective) functions from source to target language,\nwhich in many cases results in important distinctions in meaning being lost in\ntranslation. Building on Bayesian models of informative utterance production,\nwe present a method to define a less ambiguous translation system in terms of\nan underlying pre-trained neural sequence-to-sequence model. This method\nincreases injectivity, resulting in greater preservation of meaning as measured\nby improvement in cycle-consistency, without impeding translation quality\n(measured by BLEU score).","url_abs":"http://arxiv.org/abs/1902.09514v4","url_pdf":"http://arxiv.org/pdf/1902.09514v4.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":"lost-in-machine-translation-a-method-to","repo_url":"https://github.com/reubenharry/pragmatic-translation","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"machine-translation","task_name":"Machine Translation"},{"task_slug":"sentence","task_name":"Sentence"},{"task_slug":"translation","task_name":"Translation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1902.09514","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}