Unsupervised Relation Extraction for Automatic Generation of Multiple-Choice Questions

International Conference RANLP 2009 - Borovets, Bulgaria, pages 1–5 Unsupervised Relation Extraction for Automatic Generation of Multiple-Choice Questions Naveed Afzal Research Institute for Information and Language Processing University of Wolverhampton Wolverhampton, UK n.afzal@wlv.ac.uk Viktor Pekar Oxford University Press Great Clarendon St. Oxford, OX2 6DP, UK viktor.pekar@oup.com Abstract In this paper, we investigate an unsupervised approach to Relation Extraction to be applied in the context of automatic generation of multiple-choice questions (MCQs). The approach aims to identify the most important semantic relations in a document without assigning explicit labels to them in order to ensure broad coverage, unrestricted to predefined types of relations. The paper examines three different surface pattern types, each implementing different assumptions about linguistic expression of semantic relations between named entities. Our main findings indicate that the approach is capable of achieving high precision rates and its enhancement with linguistic knowledge helps to produce significantly better patterns. The intended application for the method is an e-learning system for automatic assessment of students’ comprehension of training texts; however it can also be applied to other NLP scenarios, where it is necessary to recognise important semantic relations without any prior knowledge as to their types. Keywords Information Extraction, Relation Extraction, Biomedical domain, MCQ generation. 1. Introduction Information Extraction (IE) is an important problem in many information access applications. The goal is to identify instances of specific semantic relations between named entities of interest in the text. As is known from the literature, Relation Extraction in the biomedical domain is quite difficult compared to other domains, such as news domain, due to the inherently complex nature of its texts: biomedical Named Entities (NEs) are expressed in various linguistic forms such as abbreviations, plurals, compounds, coordination, cascades, acronyms and apposition. Sentences in such texts are syntactically complex as the subsequent Relation Extraction phase depends upon the correct identification of the named entities and correct analysis of linguistic constructions expressing relations between them (e.g., [3, 21]). The main advantage of the approach presented in this paper is that it can cover a potentially unrestricted range of semantic relations while most supervised and semi- supervised approaches can learn to extract only those relations that have been exemplified in annotated text, seed patterns or seed named entities. Moreover, our approach is suitable in situations where a lot of unannotated text is available as it does not require manually annotated text or seeds. These properties of the method can be useful, specifically, in such applications as Multiple-Choice Question generation [12] or a pre- emptive approach in which viable IE patterns are created in advance without human intervention [20,15]. In the future, we plan to employ the Relation Extraction method for automatic MCQ generation, where it will be used to find relations and named entities in educational texts that are important for testing students’ familiarity with key facts contained in the texts. In order to achieve this, we need an IE method that has a high precision and at the same time works with unrestricted semantic types …

Content truncated to 500 words