Domain Adapted Distant Supervision for Pedagogically Motivated Relation Extraction

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 2213–2222 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 2213 Domain Adapted Distant Supervision for Pedagogically Motivated Relation Extraction Oscar Sainz, Oier Lopez de Lacalle, Itziar Aldabe, Montse Maritxalar University of the Basque Country (UPV/EHU) {osainz006, oier.lopezdelacalle,itziar.aldabe,montse.maritxalar}@ehu.eus Abstract In this paper we present a relation extraction system that given a text extracts pedagogically motivated relation types, as a previous step to obtaining a semantic representation of the text which will make possible to automatically generate questions for reading comprehension. The system maps pedagogically motivated relations with relations from ConceptNet and deploys Distant Supervision for relation extraction. We run a study on a subset of those relationships in order to analyse the viability of our approach. For that, we build a domain-specific relation extraction system and explore two relation extraction models: a state-of-the-art model based on transfer learning and a discrete feature based machine learning model. Experiments show that the neural model obtains better results in terms of F-score and we yield promising results on the subset of relations suitable for pedagogical purposes. We thus consider that distant supervision for relation extraction is a valid approach in our target domain, i.e. biology. Keywords: distant supervision, relation extraction, domain adaptation, pedagogically motivated relations 1. Introduction Question generation (QG) has been traditionally linked to the education and psychology domains. However, during the last decade there has also been a growth of interest on QG in the area of computing (Rus and Graesser, 2009) thanks to advances in Artificial Intelligence and Natural Language Processing (NLP). The first published systems on QG (Heilman and Smith, 2010) limited the generation of questions to the sentence level and the generation was mainly based on shallow linguistic information at text level. Then, some QG systems based the generation process on concept maps (Olney et al., 2012; Jouault et al., 2016) and aimed at building a semantic representation of text in or- der to generate relevant questions. For example, Olney et al. (2012) build a concept map from a biology textbook based on NLP techniques and heuristics, and Jouault et al. (2016) used linked data technologies to extract the seman- tic information from Wikipedia, Freebase and DBpedia in order to generate meaningful questions for history learning. Both approaches generate questions following Graesser and Person’s question taxonomy (Graesser and Person, 1994), a taxonomy that describes domain-independent meaningful question types to support learning. Interestingly, the concept map extracted in (Olney et al., 2012) contains a set of edges that facilitates link- ages between the graph representations and question ask- ing/answering. The main goal of our work is to continue the process made by (Olney et al., 2012), but building the se- mantic representation based on a Relation Extraction (RE) system. In this paper we focus on the process to automatically build the semantic representation. For that, we present a RE sys- tem that given a text extracts different relation types fol- lowing (Olney …

Content truncated to 500 words