Cross-lingual Entity Alignment with Incidental Supervision

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 645–658 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 645 Cross-lingual Entity Alignment with Incidental Supervision Muhao Chen1,2∗, Weijia Shi3∗, Ben Zhou1, Dan Roth1 1Department of Computer and Information Science, UPenn 2Viterbi School of Engineering, USC 3Department of Computer Science, UCLA {muhao,xyzhou,danroth}@seas.upenn.edu; swj0419@cs.ucla.edu Abstract Much research effort has been put to multilin- gual knowledge graph (KG) embedding meth- ods to address the entity alignment task, which seeks to match entities in different language- specific KGs that refer to the same real-world object. Such methods are often hindered by the insufficiency of seed alignment provided between KGs. Therefore, we propose an inci- dentally supervised model, JEANS , which jointly represents multilingual KGs and text corpora in a shared embedding scheme, and seeks to improve entity alignment with inci- dental supervision signals from text. JEANS first deploys an entity grounding process to combine each KG with the monolingual text corpus. Then, two learning processes are conducted: (i) an embedding learning pro- cess to encode the KG and text of each lan- guage in one embedding space, and (ii) a self- learning based alignment learning process to iteratively induce the matching of entities and that of lexemes between embeddings. Ex- periments on benchmark datasets show that JEANS leads to promising improvement on entity alignment with incidental supervision, and significantly outperforms state-of-the-art methods that solely rely on internal informa- tion of KGs.1 1 Introduction A multilingual knowledge base (KB) such as DB- pedia (Lehmann et al., 2015), ConceptNet (Speer et al., 2017) and Yago (Mahdisoltani et al., 2015) stores multiple language-specific knowl- edge graphs (KGs) that express relations of many concepts and real-world entities. As each KG thereof is either extracted independently from monolingual corpora (Lehmann et al., 2015; Mahdisoltani et al., 2015) or contributed by speakers of the language (Speer et al., 2017; ∗ Indicating equal contributions. 1Software and resources are available at http:// cogcomp.org/page/publication_view/929. KG of KG of LanguageL1 L2 Language Entity Alignment (Noisy) Grounding (Noisy) Grounding Text Corpus of Induced Lexical Alignment Incidental Supervision L1 L2 L1 Text Corpus of L2 Figure 1: The learning framework of JEANS. Mitchell et al., 2018), it is common for dif- ferent KGs to constitute complementary knowl- edge (Bleiholder and Naumann, 2009; Bryl and Bizer, 2014). Hence, aligning and synchroniz- ing language-specific KGs support AI systems with more comprehensive commonsense reason- ing (Lin et al., 2019; Li et al., 2019b; Yeo et al., 2018), and benefit various knowledge-driven NLP tasks, including machine translation (Moussallem et al., 2018), narrative prediction (Chen et al., 2019) and dialogue agents (Sun et al., 2019a). Learning to align multilingual KGs is a non- trivial task, as KGs with distinct surface forms, heterogeneous schemata and inconsistent struc- tures easily cause traditional symbolic methods to fall short (Suchanek et al., 2011; Wijaya et al., 2013; Jim´enez-Ruiz et al., 2012). Recently, much attention has been paid to methods based on mul- tilingual KG embeddings (Chen …

Content truncated to 500 words