MILIE: Modular & Iterative Multilingual Open Information Extraction

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 6939 - 6950 May 22-27, 2022 c⃝2022 Association for Computational Linguistics MILIE: Modular & Iterative Multilingual Open Information Extraction Bhushan Kotnis1, Kiril Gashteovski1, Daniel Oñoro-Rubio1, Ammar Shaker1, Vanesa Rodriguez-Tembras2, Makoto Takamoto1, Mathias Niepert1,3, Carolin Lawrence1 1NEC Laboratories Europe, Heidelberg, Germany. firstname.lastname@neclab.eu 2Heidelberg University, Center for Iberoamerican Studies, Germany. 3University of Stuttgart, Germany Abstract Open Information Extraction (OpenIE) is the task of extracting (subject, predicate, object) triples from natural language sentences. Cur- rent OpenIE systems extract all triple slots in- dependently. In contrast, we explore the hy- pothesis that it may be beneficial to extract triple slots iteratively: first extract easy slots, followed by the difficult ones by conditioning on the easy slots, and therefore achieve a better overall extraction. Based on this hypothesis, we propose a neural OpenIE system, MILIE, that operates in an it- erative fashion. Due to the iterative nature, the system is also modular—it is possible to seam- lessly integrate rule based extraction systems with a neural end-to-end system, thereby al- lowing rule based systems to supply extraction slots which MILIE can leverage for extracting the remaining slots. We confirm our hypothe- sis empirically: MILIE outperforms SOTA sys- tems on multiple languages ranging from Chi- nese to Arabic. Additionally, we are the first to provide an OpenIE test dataset for Arabic and Galician. 1 Introduction Open Information Extraction (OpenIE) aims to ex- tract structured facts in the form of (subject, re- lation, object)-triples from natural language sen- tences (Etzioni et al., 2008). For example, given a sentence, "Barrack Obama became the US Pres- ident in the year 2008", an OpenIE system is expected to extract the following triples: (Bar- rack Obama; became; US President) and (Barrack Obama; became US President in; 2008). We refer to subject, predicate and the object of the triple as slots of a triple. OpenIE extractions are schema- free, human understandable intermediate represen- tations of facts in source texts (Mausam, 2016). They are useful in a variety of information extrac- tion end tasks such as summarization (Xu and Lap- ata, 2021), question answering (Khot et al., 2017; Yan et al., 2018) and automated schema extraction (Nimishakavi et al., 2016). The various slots of a triple are dependent on each other and hence an error in one slot renders the entire extraction unusable. We hypothesize that triple extraction errors largely stem from the diffi- culty of extracting certain slots of a triple and said difficulty may depend on the sentence construction and the language. For example, "Barrack Obama became the US President in the year 2008" con- tains two triples (Barrack Obama; became; US President) and (Barrack Obama; became US Presi- dent in; 2008). Extracting the predicate, "became US President in", for the second triple is tricky, because the object of the first triple (US President) overlaps with the predicate of the second triple. But if the extraction system was provided with the object, (2008), and then …

Content truncated to 500 words