From Ultra-Fine to Fine: Fine-tuning Ultra-Fine Entity Typing Models to Fine-grained

Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 2259–2270 July 9-14, 2023 ©2023 Association for Computational Linguistics From Ultra-Fine to Fine: Fine-tuning Ultra-Fine Entity Typing Models to Fine-grained Hongliang Dai1 and Ziqian Zeng2 1College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics hongldai@nuaa.edu.cn 2Shien-Ming Wu School of Intelligent Engineering, South China University of Technology zqzeng@scut.edu.cn Abstract For the task of fine-grained entity typing (FET), due to the use of a large number of entity types, it is usually considered too costly to manu- ally annotate a training dataset that contains an ample number of examples for each type. A common way to address this problem is to use distantly annotated training examples that con- tains incorrect labels. But the errors in the au- tomatic annotation may limit the performance of trained models. Recently, there are a few ap- proaches that no longer depend on such weak training data. However, without using sufficient direct entity typing supervision may also cause them to yield inferior performance. In this pa- per, we propose a new approach that can avoid the need of creating distantly labeled data. We first train an entity typing model that have an ex- tremely broad type coverage by using the ultra- fine entity typing data. Then, when there is a need to produce a model for a newly designed fine-grained entity type schema, we can simply fine-tune the previously trained model with a small number of corresponding annotated ex- amples. Experimental results show that our approach achieves outstanding performance for FET under the few-shot setting. It can also outperform state-of-the-art weak supervision based methods after fine-tuning the model with only a small-size manually annotated training set. 1 Introduction Entity Typing is the task of assigning type labels to entity mentions in texts. Its results have been shown to be beneficial to downstream tasks such as Entity Linking (Ling et al., 2015; Vashishth et al., 2021), Coreference Resolution (Onoe and Durrett, 2020), etc. Currently, there are mainly two forms of entity typing tasks: Fine-grained Entity Typing (FET) (Ling and Weld, 2012) and Ultra-fine Entity Typing (UFET) (Choi et al., 2018; Lee et al., 2020). Table 1 and Table 2 provide a few examples for them. Sentence with Entity Mention Labels Police said he had been kidnapped from his home on Tuesday. person, victim, man, male He competed at the 2008 Summer Olympics, where despite missing the finals by .13 second, he posted a personal best time. event, match Embassy Suites was owned by Promus Hotel Corporation , a hotel management and franchise company from Memphis, Tennessee. company, busi- ness, corpora- tion, organiza- tion Table 1: Examples of Ultra-Fine Entity Typing. Target entity mentions are highlighted with yellow background. Sentence with Entity Mention Labels In the first RTC transaction with a foreign buyer, Royal Trustco Ltd., Toronto , will acquire Pacific Sav- ings Bank, Costa Mesa, Calif. /location, /loca- tion/city The Fiero plant was viewed as a model of …

Content truncated to 500 words