Distant Learning for Entity Linking with Automatic Noise Detection

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4081–4090 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 4081 Distant Learning for Entity Linking with Automatic Noise Detection Phong Le1 and Ivan Titov1,2 1University of Edinburgh 2University of Amsterdam lephong.xyz@gmail.com ititov@inf.ed.ac.uk Abstract Accurate entity linkers have been produced for domains and languages where annotated data (i.e., texts linked to a knowledge base) is available. However, little progress has been made for the settings where no or very limited amounts of labeled data are present (e.g., le- gal or most scientific domains). In this work, we show how we can learn to link mentions without having any labeled examples, only a knowledge base and a collection of unanno- tated texts from the corresponding domain. In order to achieve this, we frame the task as a multi-instance learning problem and rely on surface matching to create initial noisy la- bels. As the learning signal is weak and our surrogate labels are noisy, we introduce a noise detection component in our model: it lets the model detect and disregard examples which are likely to be noisy. Our method, jointly learning to detect noise and link en- tities, greatly outperforms the surface match- ing baseline. For a subset of entity categories, it even approaches the performance of super- vised learning. 1 Introduction Entity linking (EL) is the task of linking poten- tially ambiguous textual mentions to the corre- sponding entities in a knowledge base. Accurate entity linking is crucial in many natural language processing tasks, including information extraction (Hoffart et al., 2011) and question answering (Yih et al., 2015). Though there has been significant progress in entity linking recently (Ratinov et al., 2011; Hoffart et al., 2011; Chisholm and Hachey, 2015; Globerson et al., 2016; Yamada et al., 2017; Ganea and Hofmann, 2017; Le and Titov, 2018), previous work has focused on supervised learning. Annotated data necessary for supervised learning is available for certain knowledge bases and do- mains. For example, one can directly use web- pages linking to Wikipedia to learn a Wikipedia linker. Similarly, there exist domain-specific sets of manually annotated documents (e.g., AIDA- CoNLL news dataset for YAGO (Hoffart et al., 2011)). However, for many ontologies and do- mains annotation is not available or limited (e.g., law). Our goal is to develop a method which does not rely on any training data besides unlabeled texts and a knowledge base. In order to construct such a method, we use an insight from simple surface matching heuris- tics (e.g., Riedel et al. (2010)). Such heuristics choose entities from a knowledge base by measur- ing the overlap between the sets of content words in the mention and in the entity name. For ex- ample, in Figure 1, the entities BILL CLINTON (PRESIDENT) and PRESIDENCY OF BILL CLIN- TON both have two matching words with the men- tion Bill Clinton. Whereas we will see in our ex- periments that this method alone is not particularly accurate …

Content truncated to 500 words