Syntactically Rich Discriminative Training: An Effective Method for Open Information Extraction

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 5972 - 5987 December 7-11, 2022 ©2022 Association for Computational Linguistics Syntactically Rich Discriminative Training: An Effective Method for Open Information Extraction Frank Mtumbuka3,1 Thomas Lukasiewicz2,1 1 Department of Computer Science, University of Oxford, UK 2 Institute of Logic and Computation, TU Wien, Austria 3 Cardiff University, Cardiff, UK firstname.lastname@cs.ox.ac.uk Abstract Open information extraction (OIE) is the task of extracting facts “(Subject, Relation, Object)” from natural language text. We propose several new methods for training neural OIE models in this paper. First, we propose a novel method for computing syntactically rich text embeddings using the structure of dependency trees. Sec- ond, we propose a new discriminative training approach to OIE in which tokens in the gen- erated fact are classified as “real” or “fake”, i.e., those tokens that are in both the generated and gold tuples, and those that are only in the generated tuple but not in the gold tuple. We also address the issue of repetitive tokens in generated facts and improve the models’ abil- ity to generate implicit facts. Our approach reduces repetitive tokens by a factor of 23%. Finally, we present paraphrased versions of the CaRB, OIE2016, and LSOIE datasets, and show that the models’ performance substantially im- proves when trained on datasets augmented by such data. Our best model beats the SOTA of IMoJIE on the recent CaRB dataset, with an improvement of 39.63% in F1 score. 1 Introduction OIE (Banko et al., 2007) is a branch of informa- tion extraction (IE) that focuses on extracting struc- tured information (Niklaus et al., 2018) from un- structured natural language text. This structured information is a set of tuples of the form “(Sub- ject, Relation, Object)”, also called facts in OIE. For instance, given the sentence “Machine learn- ing is a subfield of AI.”, the tuple ⟨machine learn- ing, is a subfield of, AI⟩ can be extracted, where the relation phrase “is a subfield of” indicates the semantic relationship between the subject “Ma- chine learning” and the object “AI”. OIE is use- ful in many downstream natural language process- ing (NLP) tasks like natural language understand- ing (Mausam, 2016), multi-document question an- swering and summarization (Fan et al., 2019), and knowledge base construction from text (Soderland et al., 2010). Several neural OIE methods in the litera- ture approach OIE as either a sequence label- ing (Stanovsky et al., 2018; Roy et al., 2019; Jiang et al., 2019; Zhan and Zhao, 2019; Hohenecker et al., 2020) or a sequence generation problem (Cui et al., 2018; Sun et al., 2018; Kolluru et al., 2020). Sequence labeling approaches label each token in the input text as either belonging to the subject, relation, or object, while sequence generation ap- proaches generate facts one word at a time given the input text. For the task of OIE, it is a common practice to make use of part-of-speech (PoS) and dependency tags in addition to the actual input text as …

Content truncated to 500 words