{"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/hierarchical-pre-training-for-sequence","title":"Hierarchical Pre-training for Sequence Labelling in Spoken Dialog","arxiv_id":"2009.11152","date":"2020-09-23","proceeding":"Findings of the Association for Computational Linguistics 2020","authors":["Emile Chapuis","Pierre Colombo","Matteo Manica","Matthieu Labeau","Chloe Clavel"],"abstract":"Sequence labelling tasks like Dialog Act and Emotion/Sentiment identification are a key component of spoken dialog systems. In this work, we propose a new approach to learn generic representations adapted to spoken dialog, which we evaluate on a new benchmark we call Sequence labellIng evaLuatIon benChmark fOr spoken laNguagE benchmark (\\texttt{SILICONE}). \\texttt{SILICONE} is model-agnostic and contains 10 different datasets of various sizes. We obtain our representations with a hierarchical encoder based on transformer architectures, for which we extend two well-known pre-training objectives. Pre-training is performed on OpenSubtitles: a large corpus of spoken dialog containing over $2.3$ billion of tokens. We demonstrate how hierarchical encoders achieve competitive results with consistently fewer parameters compared to state-of-the-art models and we show their importance for both pre-training and fine-tuning.","url_abs":"https://arxiv.org/abs/2009.11152v3","url_pdf":"https://arxiv.org/pdf/2009.11152v3.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":[],"tasks":[{"task_slug":"dialogue-act-classification","task_name":"Dialogue Act Classification"},{"task_slug":"emotion-recognition-in-conversation","task_name":"Emotion Recognition in Conversation"},{"task_slug":"text-classification","task_name":"Text Classification"}],"methods":[],"datasets_introduced":[{"slug":"silicone-benchmark","name":"SILICONE Benchmark","full_name":"SILICONE"}],"methods_introduced":[],"results":[{"leaderboard":"/sota/dialogue-act-classification-on-icsi-meeting","task":"Dialogue Act Classification","dataset":"ICSI Meeting Recorder Dialog Act (MRDA) corpus","model":"Pretrained Hierarchical Transformer","rank_in_archive_order":1,"of":8,"metrics":{"Accuracy":"92.4"},"uses_additional_data":false},{"leaderboard":"/sota/dialogue-act-classification-on-switchboard","task":"Dialogue Act Classification","dataset":"Switchboard corpus","model":"Pretrained Hierarchical Transformer","rank_in_archive_order":8,"of":11,"metrics":{"Accuracy":"79.2"},"uses_additional_data":false},{"leaderboard":"/sota/emotion-recognition-in-conversation-on-3","task":"Emotion Recognition in Conversation","dataset":"DailyDialog","model":"Pretrained Hierarchical Transformer","rank_in_archive_order":6,"of":22,"metrics":{"Micro-F1":"60.14"},"uses_additional_data":false},{"leaderboard":"/sota/emotion-recognition-in-conversation-on","task":"Emotion Recognition in Conversation","dataset":"IEMOCAP","model":"Pretrained Hierarchical Transformer","rank_in_archive_order":42,"of":59,"metrics":{"Accuracy":"66.05","Weighted-F1":"65.37"},"uses_additional_data":false},{"leaderboard":"/sota/emotion-recognition-in-conversation-on-meld","task":"Emotion Recognition in Conversation","dataset":"MELD","model":"Pretrained Hierarchical Transformer","rank_in_archive_order":51,"of":68,"metrics":{"Weighted-F1":"61.90"},"uses_additional_data":false},{"leaderboard":"/sota/emotion-recognition-in-conversation-on-2","task":"Emotion Recognition in Conversation","dataset":"SEMAINE","model":"Pretrained Hierarchical Transformer","rank_in_archive_order":2,"of":3,"metrics":{"MAE (Arousal)":"0.16","MAE (Expectancy)":"0.16","MAE (Power)":"7.70","MAE (Valence)":"0.16"},"uses_additional_data":false},{"leaderboard":"/sota/text-classification-on-silicone-benchmark","task":"Text Classification","dataset":"SILICONE Benchmark","model":"Pretrained Hierarchical Transformer","rank_in_archive_order":1,"of":1,"metrics":{"1:1 Accuracy":"71.25"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2009.11152","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}