SpanNER: Named Entity Re-/Recognition as Span Prediction

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 7183–7195 August 1–6, 2021. ©2021 Association for Computational Linguistics 7183 SPANNER: Named Entity Re-/Recognition as Span Prediction Jinlan Fu ∗ Fudan University fujl16@fudan.edu.cn Xuanjing Huang Fudan University xjhuang@fudan.edu.cn Pengfei Liu † Carnegie Mellon University pliu3@cs.cmu.edu Abstract Recent years have seen the paradigm shift of Named Entity Recognition (NER) systems from sequence labeling to span prediction. Despite its preliminary effectiveness, the span prediction model’s architectural bias has not been fully understood. In this paper, we first investigate the strengths and weaknesses when the span prediction model is used for named entity recognition compared with the sequence labeling framework and how to fur- ther improve it, which motivates us to make complementary advantages of systems based on different paradigms. We then reveal that span prediction, simultaneously, can serve as a system combiner to re-recognize named en- tities from different systems’ outputs. We experimentally implement 154 systems on 11 datasets, covering three languages, compre- hensive results show the effectiveness of span prediction models that both serve as base NER systems and system combiners. We make all code and datasets available: https:// github.com/neulab/spanner, as well as an online system demo: http://spanner. sh. Our model also has been deployed into the EXPLAINABOARD (Liu et al., 2021) platform, which allows users to flexibly perform the sys- tem combination of top-scoring systems in an interactive way: http://explainaboard. nlpedia.ai/leaderboard/task-ner/. 1 Introduction The rapid evolution of neural architectures (Kalch- brenner et al., 2014a; Kim, 2014; Hochreiter and Schmidhuber, 1997) and large pre-trained models (Devlin et al., 2019; Lewis et al., 2020) not only drive the state-of-the-art performance of many NLP tasks (Devlin et al., 2019; Liu and Lapata, 2019) to a new level but also change the way ∗This work is done when Jinlan visited CMU remotely. †Corresponding author. ... Sysm <latexit sha1_base64="AP6Bg0eKen BqGfb7o1M5nHWmus=">AC0XicjVHLSsNAFD2Nr7a+qi7dBIsoLkriRpcFNy4r tQ9oa0nSaRuaF5OJUEpB3PoDbvWnxD/QpX/gnWkKahG9IcmZc+85M3euHXluLAz jNaMtLa+srmVz+fWNza3tws5uPQ4T7rCaE3ohb9pWzDw3YDXhCo81I84s3/ZYwx 5dyHzjlvHYDYNrMY5Yx7cGgdt3HUsQdP2LTHk/qQ6jqdv1soGiVDhb4IzBQUy7 mPI1BUwsIL2ughIMEPhgCMIeLMT0tGDCQERcBxPiOCFX5RmyJM2oSpGFRaxI /oOaNVK2YDW0jNWaod28ejlpNRxSJqQ6jhuZu8olyluxv3hPlKc82pr+devnEC gyJ/Us3r/yvTvYi0Me56sGlniLFyO6c1CVRtyJPrn/pSpBDRJzEPcpzwo5Szu9Z V5pY9S7v1lL5N1UpWbl20toE7/KUNGDz5zgXQf20ZBol84omfYJZLGPAxzTPM9 QxiUqJE3xyOe8KxVtbF2p93PSrVMqtnDt9AePgFrq5a0</latexit> <latexit sha1_base64="69ProrCO9h jMJpU8V6+Oy7dSFLg=">AC0XicjVHLSsNAFD3GV31XboJiIuSuJGlwU3Liu1 KvhiEqd1MC9mJmIpBXHrD7jVbxLEP7BLf0C8M43gA9EJSc6ce8+ZufcGWSU9rz nAWdwaHhktDQ2PjE5NT1Tnp3bU2kuQ94I0yiVBwFTPBIJb2ihI36QSc7iIOL7wc Wie9fcqlEmuzqdsaPY9ZKRFOETBN1chQzfS7jTr2tuqfxaXnJq3h2uT+BX4Cl6t jrSu/q8a2Wlp9whDOkCJEjBkcCTgCg6LnED48ZMQdo0OcJCRsnKOLcdLmlMUpg xF7Qd8W7Q4LNqG98VRWHdIpEb2SlC6WSZNSniRsTnNtPLfOhv3Nu2M9zd3a9A8Kr 5hYjXNi/9J9ZP5XZ2rRaGLT1iCopswyprqwcMltV8zN3U9VaXLIiDP4jOKScGiV H312rUbZ2k1vmY2/2EzDmn1Y5ObomVvSgP3v4/wJ9tYrvlfxd2jSa+ivEhawiFW a5waq2EYNDfKWuM9Hpy603aunZt+qjNQaObxZTm373EhmeY=</latexit> <latexit sha1_base64="69ProrCO9h jMJpU8V6+Oy7dSFLg=">AC0XicjVHLSsNAFD3GV31XboJiIuSuJGlwU3Liu1 KvhiEqd1MC9mJmIpBXHrD7jVbxLEP7BLf0C8M43gA9EJSc6ce8+ZufcGWSU9rz nAWdwaHhktDQ2PjE5NT1Tnp3bU2kuQ94I0yiVBwFTPBIJb2ihI36QSc7iIOL7wc Wie9fcqlEmuzqdsaPY9ZKRFOETBN1chQzfS7jTr2tuqfxaXnJq3h2uT+BX4Cl6t jrSu/q8a2Wlp9whDOkCJEjBkcCTgCg6LnED48ZMQdo0OcJCRsnKOLcdLmlMUpg xF7Qd8W7Q4LNqG98VRWHdIpEb2SlC6WSZNSniRsTnNtPLfOhv3Nu2M9zd3a9A8Kr 5hYjXNi/9J9ZP5XZ2rRaGLT1iCopswyprqwcMltV8zN3U9VaXLIiDP4jOKScGiV H312rUbZ2k1vmY2/2EzDmn1Y5ObomVvSgP3v4/wJ9tYrvlfxd2jSa+ivEhawiFW a5waq2EYNDfKWuM9Hpy603aunZt+qjNQaObxZTm373EhmeY=</latexit> Sys1 <latexit sha1_base64="2FlUfRDU5n y15AxB8oMK/71n/I=">AC0XicjVHLSsNAFD2Nr7a+qi7dBIsoLkriRpcFNy4r tQ9oa0nSaRuaF5OJUEpB3PoDbvWnxD/QpX/gnWkKahG9IcmZc+85M3euHXluLAz jNaMtLa+srmVz+fWNza3tws5uPQ4T7rCaE3ohb9pWzDw3YDXhCo81I84s3/ZYwx 5dyHzjlvHYDYNrMY5Yx7cGgdt3HUsQdP2LTHk/qQ6jqds1soGiVDhb4IzBQUy7 mPI1BUwsIL2ughIMEPhgCMIeLMT0tGDCQERcBxPiOCFX5RmyJM2oSpGFRaxI /oOaNVK2YDW0jNWaod28ejlpNRxSJqQ6jhuZu8olyluxv3hPlKc82pr+devnEC gyJ/Us3r/yvTvYi0Me56sGlniLFyO6c1CVRtyJPrn/pSpBDRJzEPcpzwo5Szu9Z V5pY9S7v1lL5N1UpWbl20toE7/KUNGDz5zgXQf20ZBol84omfYJZLGPAxzTPM9 QxiUqJE3xyOe8KxVtbF2p93PSrVMqtnDt9AePgHdHJZ4</latexit> <latexit sha1_base64="VoVbWLD8YK mXrBusN28oEof1Y=">AC0XicjVHLSsNAFD3GV31XboJiIuSuJGlwU3Liu1 KvhiEqd1MC9mJmIpBXHrD7jVbxLEP7BLf0C8M43gA9EJSc6ce8+ZufcGWSU9rz nAWdwaHhktDQ2PjE5NT1Tnp3bU2kuQ94I0yiVBwFTPBIJb2ihI36QSc7iIOL7wc Wie9fcqlEmuzqdsaPY9ZKRFOETBN1chQzfS7jTr2tuqf+aXnJq3h2uT+BX4Cl6t jrSu/q8a2Wlp9whDOkCJEjBkcCTgCg6LnED48ZMQdo0OcJCRsnKOLcdLmlMUpg xF7Qd8W7Q4LNqG98VRWHdIpEb2SlC6WSZNSniRsTnNtPLfOhv3Nu2M9zd3a9A8Kr 5hYjXNi/9J9ZP5XZ2rRaGLT1iCopswyprqwcMltV8zN3U9VaXLIiDP4jOKScGiV H312rUbZ2k1vmY2/2EzDmn1Y5ObomVvSgP3v4/wJ9tYrvlfxd2jSa+ivEhawiFW a5waq2EYNDfKWuM9Hpy603aunZt+qjNQaObxZTm37+KSmao=</latexit> <latexit sha1_base64="VoVbWLD8YK mXrBusN28oEof1Y=">AC0XicjVHLSsNAFD3GV31XboJiIuSuJGlwU3Liu1 KvhiEqd1MC9mJmIpBXHrD7jVbxLEP7BLf0C8M43gA9EJSc6ce8+ZufcGWSU9rz nAWdwaHhktDQ2PjE5NT1Tnp3bU2kuQ94I0yiVBwFTPBIJb2ihI36QSc7iIOL7wc Wie9fcqlEmuzqdsaPY9ZKRFOETBN1chQzfS7jTr2tuqf+aXnJq3h2uT+BX4Cl6t jrSu/q8a2Wlp9whDOkCJEjBkcCTgCg6LnED48ZMQdo0OcJCRsnKOLcdLmlMUpg xF7Qd8W7Q4LNqG98VRWHdIpEb2SlC6WSZNSniRsTnNtPLfOhv3Nu2M9zd3a9A8Kr 5hYjXNi/9J9ZP5XZ2rRaGLT1iCopswyprqwcMltV8zN3U9VaXLIiDP4jOKScGiV H312rUbZ2k1vmY2/2EzDmn1Y5ObomVvSgP3v4/wJ9tYrvlfxd2jSa+ivEhawiFW a5waq2EYNDfKWuM9Hpy603aunZt+qjNQaObxZTm37+KSmao=</latexit> Sys2 <latexit sha1_base64="q1Ma2lZE5+ EBD4YwCRZiIOUT5pY=">AC0XicjVHLSsNAFD3GV1tfVZdugkUFyXpRpcFNy4r tQ9oa0nSaRuaF5OJUEpB3PoDbvWnxD/QpX/gnWkKahG9IcmZc+85M3euHXluLAz jdUlbXldW89kcxubW9s7+d29ehwm3GE1J/RC3rStmHluwGrCFR5rRpxZvu2xhj 26kPnGLeOxGwbXYhyxjm8NArfvOpYg6qbtW2LI/Ul1HE+7pW6+YBQNFfoiMFNQKG c/jkFRCfMvaKOHEA4S+GAIAh7sBDT04IJAxFxHUyI4RclWeYIkfahKoYVjEj ug7oFUrZQNaS89YqR3axaOXk1LHEWlCquOE5W6yifKWbK/eU+UpzbmP526uUTK zAk9i/dvPK/OtmLQB/nqgeXeoUI7tzUpdE3Yo8uf6lK0EOEXES9yjPCTtKOb9n XWli1bu8W0vl31SlZOXaSWsTvMtT0oDNn+NcBPVS0TSK5hVN+hSzyOAhziheZ6 hjEtUCNvjkc84VmramPtTruflWpLqWYf30J7+ATfJZ5</latexit> <latexit sha1_base64="CvJCMPVwdV Ti+QwayXJ4kUndUbo=">AC0XicjVHLSsNAFD3GV+uz6tJNUBRxUZJudCm4cVmp 1UJ9MImjDubFzEQsRC3/oBb/SZB/AO79AfEO9MIPhCdkOTMufecmXtvkEVCac9 7HnAGh4ZHRkvlsfGJyanpyszsrkpzGfJmEapbAVM8UgkvKmFjngrk5zFQcT3gv NE9+74FKJNnRnYwfxOw0ESciZJqow/2Y6TMZdxsdXVUO6oselXPLvcn8AuwuF F+Xe5dPr7V08oT9nGMFCFyxOBIoAlHYFD0tOHDQ0bcAbrESULCxjmuMEbanLI4Z TBiz+l7Srt2wSa0N57KqkM6JaJXktLFEmlSypOEzWmujefW2bC/eXetp7lbh/5B4 RUTq3FG7F+6j8z/6kwtGidYtzUIqimzjKkuLFxy2xVzc/dTVZocMuIMPqa4JBxa 5UefXatRtnbTW2bjLzbTsGYfFrk5euaWNGD/+zh/gt1a1feq/jZNehX9VcI8FrB C81zDBrZQR5O8Je5wjwen4XSca+emn+oMFJo5fFnO7Tvk8pmr</latexit> <latexit sha1_base64="CvJCMPVwdV Ti+QwayXJ4kUndUbo=">AC0XicjVHLSsNAFD3GV+uz6tJNUBRxUZJudCm4cVmp 1UJ9MImjDubFzEQsRC3/oBb/SZB/AO79AfEO9MIPhCdkOTMufecmXtvkEVCac9 7HnAGh4ZHRkvlsfGJyanpyszsrkpzGfJmEapbAVM8UgkvKmFjngrk5zFQcT3gv NE9+74FKJNnRnYwfxOw0ESciZJqow/2Y6TMZdxsdXVUO6oselXPLvcn8AuwuF F+Xe5dPr7V08oT9nGMFCFyxOBIoAlHYFD0tOHDQ0bcAbrESULCxjmuMEbanLI4Z TBiz+l7Srt2wSa0N57KqkM6JaJXktLFEmlSypOEzWmujefW2bC/eXetp7lbh/5B4 RUTq3FG7F+6j8z/6kwtGidYtzUIqimzjKkuLFxy2xVzc/dTVZocMuIMPqa4JBxa 5UefXatRtnbTW2bjLzbTsGYfFrk5euaWNGD/+zh/gt1a1feq/jZNehX9VcI8FrB C81zDBrZQR5O8Je5wjwen4XSca+emn+oMFJo5fFnO7Tvk8pmr</latexit> SeqLab SeqLab SpanNer <latexit sha1_base64="DZv0jzXJT2 tUzVZoBGtnbFfy+Jg=">AC1XicjVHLSsNAFD2Nr1pfUZdugkXoqiTd6LgxpVU tA+opSTaQ1NkzCZFEvpTtz6A271l8Q/0L/wzpiCWkQnJDlz7j1n5t7rxYGfSNt +zRlLyura/n1wsbm1vaOubvXSKJUMF5nURCJlucmPBDXpe+DHgrFtwdeQFves NTFW+OuUj8KLySk5h3Ru4g9Ps+cyVRXdO8lvxWJmx6GbvhORezrlm0y7Ze1iJwMl BEtmqR+YJr9BCBIcUIHCEk4QAuEnracGAjJq6DKXGCkK/jHDMUSJtSFqcMl9ghf Qe0a2dsSHvlmWg1o1MCegUpLRyRJqI8QVidZul4qp0V+5v3VHuqu03o72VeI2Ilb oj9SzfP/K9O1SLRx4muwaeaYs2o6ljmkuquqJtbX6qS5BATp3CP4oIw08p5ny2t SXTtqreujr/pTMWqPctyU7yrW9KAnZ/jXASNStmxy85FpVgtZaPO4wCHKNE8j1H FGWqok/cYj3jCs9E0Zsadcf+ZauQyzT6+LePhAwx0lY=</latexit> <latexit sha1_base64="DZv0jzXJT2 tUzVZoBGtnbFfy+Jg=">AC1XicjVHLSsNAFD2Nr1pfUZdugkXoqiTd6LgxpVU tA+opSTaQ1NkzCZFEvpTtz6A271l8Q/0L/wzpiCWkQnJDlz7j1n5t7rxYGfSNt +zRlLyura/n1wsbm1vaOubvXSKJUMF5nURCJlucmPBDXpe+DHgrFtwdeQFves NTFW+OuUj8KLySk5h3Ru4g9Ps+cyVRXdO8lvxWJmx6GbvhORezrlm0y7Ze1iJwMl BEtmqR+YJr9BCBIcUIHCEk4QAuEnracGAjJq6DKXGCkK/jHDMUSJtSFqcMl9ghf Qe0a2dsSHvlmWg1o1MCegUpLRyRJqI8QVidZul4qp0V+5v3VHuqu03o72VeI2Ilb oj9SzfP/K9O1SLRx4muwaeaYs2o6ljmkuquqJtbX6qS5BATp3CP4oIw08p5ny2t SXTtqreujr/pTMWqPctyU7yrW9KAnZ/jXASNStmxy85FpVgtZaPO4wCHKNE8j1H FGWqok/cYj3jCs9E0Zsadcf+ZauQyzT6+LePhAwx0lY=</latexit> <latexit sha1_base64="DZv0jzXJT2 tUzVZoBGtnbFfy+Jg=">AC1XicjVHLSsNAFD2Nr1pfUZdugkXoqiTd6LgxpVU tA+opSTaQ1NkzCZFEvpTtz6A271l8Q/0L/wzpiCWkQnJDlz7j1n5t7rxYGfSNt +zRlLyura/n1wsbm1vaOubvXSKJUMF5nURCJlucmPBDXpe+DHgrFtwdeQFves NTFW+OuUj8KLySk5h3Ru4g9Ps+cyVRXdO8lvxWJmx6GbvhORezrlm0y7Ze1iJwMl BEtmqR+YJr9BCBIcUIHCEk4QAuEnracGAjJq6DKXGCkK/jHDMUSJtSFqcMl9ghf Qe0a2dsSHvlmWg1o1MCegUpLRyRJqI8QVidZul4qp0V+5v3VHuqu03o72VeI2Ilb oj9SzfP/K9O1SLRx4muwaeaYs2o6ljmkuquqJtbX6qS5BATp3CP4oIw08p5ny2t SXTtqreujr/pTMWqPctyU7yrW9KAnZ/jXASNStmxy85FpVgtZaPO4wCHKNE8j1H FGWqok/cYj3jCs9E0Zsadcf+ZauQyzT6+LePhAwx0lY=</latexit> SpanNer <latexit sha1_base64="DZv0jzXJT2 tUzVZoBGtnbFfy+Jg=">AC1XicjVHLSsNAFD2Nr1pfUZdugkXoqiTd6LgxpVU tA+opSTaQ1NkzCZFEvpTtz6A271l8Q/0L/wzpiCWkQnJDlz7j1n5t7rxYGfSNt +zRlLyura/n1wsbm1vaOubvXSKJUMF5nURCJlucmPBDXpe+DHgrFtwdeQFves NTFW+OuUj8KLySk5h3Ru4g9Ps+cyVRXdO8lvxWJmx6GbvhORezrlm0y7Ze1iJwMl BEtmqR+YJr9BCBIcUIHCEk4QAuEnracGAjJq6DKXGCkK/jHDMUSJtSFqcMl9ghf Qe0a2dsSHvlmWg1o1MCegUpLRyRJqI8QVidZul4qp0V+5v3VHuqu03o72VeI2Ilb oj9SzfP/K9O1SLRx4muwaeaYs2o6ljmkuquqJtbX6qS5BATp3CP4oIw08p5ny2t SXTtqreujr/pTMWqPctyU7yrW9KAnZ/jXASNStmxy85FpVgtZaPO4wCHKNE8j1H FGWqok/cYj3jCs9E0Zsadcf+ZauQyzT6+LePhAwx0lY=</latexit> <latexit sha1_base64="DZv0jzXJT2 tUzVZoBGtnbFfy+Jg=">AC1XicjVHLSsNAFD2Nr1pfUZdugkXoqiTd6LgxpVU tA+opSTaQ1NkzCZFEvpTtz6A271l8Q/0L/wzpiCWkQnJDlz7j1n5t7rxYGfSNt +zRlLyura/n1wsbm1vaOubvXSKJUMF5nURCJlucmPBDXpe+DHgrFtwdeQFves NTFW+OuUj8KLySk5h3Ru4g9Ps+cyVRXdO8lvxWJmx6GbvhORezrlm0y7Ze1iJwMl BEtmqR+YJr9BCBIcUIHCEk4QAuEnracGAjJq6DKXGCkK/jHDMUSJtSFqcMl9ghf Qe0a2dsSHvlmWg1o1MCegUpLRyRJqI8QVidZul4qp0V+5v3VHuqu03o72VeI2Ilb oj9SzfP/K9O1SLRx4muwaeaYs2o6ljmkuquqJtbX6qS5BATp3CP4oIw08p5ny2t SXTtqreujr/pTMWqPctyU7yrW9KAnZ/jXASNStmxy85FpVgtZaPO4wCHKNE8j1H FGWqok/cYj3jCs9E0Zsadcf+ZauQyzT6+LePhAwx0lY=</latexit> <latexit sha1_base64="DZv0jzXJT2 tUzVZoBGtnbFfy+Jg=">AC1XicjVHLSsNAFD2Nr1pfUZdugkXoqiTd6LgxpVU tA+opSTaQ1NkzCZFEvpTtz6A271l8Q/0L/wzpiCWkQnJDlz7j1n5t7rxYGfSNt +zRlLyura/n1wsbm1vaOubvXSKJUMF5nURCJlucmPBDXpe+DHgrFtwdeQFves NTFW+OuUj8KLySk5h3Ru4g9Ps+cyVRXdO8lvxWJmx6GbvhORezrlm0y7Ze1iJwMl BEtmqR+YJr9BCBIcUIHCEk4QAuEnracGAjJq6DKXGCkK/jHDMUSJtSFqcMl9ghf Qe0a2dsSHvlmWg1o1MCegUpLRyRJqI8QVidZul4qp0V+5v3VHuqu03o72VeI2Ilb oj9SzfP/K9O1SLRx4muwaeaYs2o6ljmkuquqJtbX6qS5BATp3CP4oIw08p5ny2t SXTtqreujr/pTMWqPctyU7yrW9KAnZ/jXASNStmxy85FpVgtZaPO4wCHKNE8j1H FGWqok/cYj3jCs9E0Zsadcf+ZauQyzT6+LePhAwx0lY=</latexit> Figure 1: ONE span prediction model (SPANNER) finishes TWO things: (1) named entity recognition …

Content truncated to 500 words