{"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/girnet-interleaved-multi-task-recurrent-state","title":"GIRNet: Interleaved Multi-Task Recurrent State Sequence Models","arxiv_id":"1811.11456","date":"2018-11-28","proceeding":null,"authors":["Divam Gupta","Tanmoy Chakraborty","Soumen Chakrabarti"],"abstract":"In several natural language tasks, labeled sequences are available in\nseparate domains (say, languages), but the goal is to label sequences with\nmixed domain (such as code-switched text). Or, we may have available models for\nlabeling whole passages (say, with sentiments), which we would like to exploit\ntoward better position-specific label inference (say, target-dependent\nsentiment annotation). A key characteristic shared across such tasks is that\ndifferent positions in a primary instance can benefit from different `experts'\ntrained from auxiliary data, but labeled primary instances are scarce, and\nlabeling the best expert for each position entails unacceptable cognitive\nburden. We propose GITNet, a unified position-sensitive multi-task recurrent\nneural network (RNN) architecture for such applications. Auxiliary and primary\ntasks need not share training instances. Auxiliary RNNs are trained over\nauxiliary instances. A primary instance is also submitted to each auxiliary\nRNN, but their state sequences are gated and merged into a novel composite\nstate sequence tailored to the primary inference task. Our approach is in sharp\ncontrast to recent multi-task networks like the cross-stitch and sluice\nnetwork, which do not control state transfer at such fine granularity. We\ndemonstrate the superiority of GIRNet using three applications: sentiment\nclassification of code-switched passages, part-of-speech tagging of\ncode-switched text, and target position-sensitive annotation of sentiment in\nmonolingual passages. In all cases, we establish new state-of-the-art\nperformance beyond recent competitive baselines.","url_abs":"http://arxiv.org/abs/1811.11456v2","url_pdf":"http://arxiv.org/pdf/1811.11456v2.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":"girnet-interleaved-multi-task-recurrent-state","repo_url":"https://github.com/divamgupta/mtl_girnet","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null}],"tasks":[{"task_slug":"part-of-speech-tagging","task_name":"Part-Of-Speech Tagging"},{"task_slug":null,"task_name":"Position"},{"task_slug":"sentiment-analysis","task_name":"Sentiment Analysis"},{"task_slug":"sentiment-classification","task_name":"Sentiment Classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}