v-trel: Vocabulary Trainer for Tracing Word Relations - An Implicit Crowdsourcing Approach

Proceedings of Recent Advances in Natural Language Processing, pages 674–683, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_079 674 v-trel: Vocabulary Trainer for Tracing Word Relations - An Implicit Crowdsourcing Approach Verena Lyding Institute for Applied Linguistics, Eurac Research Bolzano/Bozen verena.lyding@eurac.edu Christos T. Rodosthenous Open University of Cyprus christos.rodosthenous@ouc.ac.cy Federico Sangati Orientale University of Naples federico.sangati@gmail.com Umair ul Hassan Insight Centre for Data Analytics, National University of Ireland Galway umair.ulhassan@insight-centre.org Lionel Nicolas Institute for Applied Linguistics, Eurac Research Bolzano/Bozen lionel.nicolas@eurac.edu Alexander König Institute for Applied Linguistics, Eurac Research Bolzano/Bozen alexander.koenig@eurac.edu Jolita Horbacauskiene Kaunas University of Technology jolita.horbacauskiene@ktu.lt Anisia Katinskaia Department of Computer Science, University of Helsinki anisia.katinskaia@helsinki.fi Abstract In this paper, we present our work on de- veloping a vocabulary trainer that uses ex- ercises generated from language resources such as ConceptNet and crowdsources the responses of the learners to enrich the lan- guage resource. We performed an empir- ical evaluation of our approach with 60 non-native speakers over two days, which shows that new entries to expand Concept- Net can efficiently be gathered through vo- cabulary exercises on word relations. We also report on the feedback gathered from the users and an expert from lan- guage teaching, and discuss the potential of the vocabulary trainer application from the user and language learner perspective. The feedback suggests that v-trel has edu- cational potential, while in its current state some shortcomings could be identified. 1 Introduction One of the major challenges for the NLP commu- nity is the continuing lack of comprehensive and high-quality language resources (LRs) for most languages. While LR creation can partly be ap- proached in an automatic fashion (e.g., by har- vesting a vocabulary from existing corpora), it of- ten requires human intervention to reach a high level of quality and coverage. Crowdsourcing (Howe, 2006) is one promising approach that can be leveraged for the task of LR creation. How- ever, any crowdsourcing ambition is faced with the challenge of attracting and retaining crowd- workers and with safeguarding the quality of re- sults produced by the crowd (Daniel et al., 2018). EnetCollect1 (Lyding et al., 2018; Nicolas et al., 2018) aims at exploring a solution to this challenge by combining the activities performed in lan- guage learning with approaches for crowdsourcing language-related datasets. Thus exploring a new path to address the NLP bottleneck of high-quality language resource creation. In this paper, we present an application for vo- cabulary training that is specifically designed to crowdsource learners’ answers to improve LRs like concept networks. The application builds on top of an architecture for crowdsourcing of language resources (Rodosthenous et al., 2019), which instantiates one of the core ideas of enet- Collect: the implicit crowdsourcing paradigm (Section 2). Accordingly, the vocabulary trainer aims at a two-fold purpose: serving automatically 1COST Action enetCollect: European Network for Com- bining Language Learning and Crowdsourcing Techniques, http://enetcollect.eurac.edu/ 675 generated vocabulary exercises and gaining con- tinuous input from the learners to improve LRs. In the remainder of the paper, we first intro- duce the vocabulary trainer …

Content truncated to 500 words