TraMOOC was present at the 27th Meeting of Computational Linguistics in The Netherlands (CLIN 27), which was organised by the Centre for Computational Linguistics group of KU Leuven and took place at the Faculty of Arts in Leuven on February 10, 2017. The poster entitled "Wikification for Implicit MT evaluation, First results for Dutch" was displayed by Iris Hendrickx (Radboud University). Based on the authors of this work machine translation evaluation requires a human in the loop: either for providing gold-standard written translations or for human judgements. As such human effort is expensive and time-consuming, the authors aim to look at methods to evaluate translation quality without human supervision. They focus on topics and entities in the text and make smart use of the inter-lingual connections in Wikipedia to determine whether the MT system managed to translate the entity correctly. They use a Wikifier (Ratinov et al, 2011) to automatically find entities with Wikipedia links in the English text and compare them against the linked entities in the target language's equivalent Wikipedia page. We present the first outcomes of this method on Dutch MOOC data and discuss the limits of this approach such as Wikipedia coverage and Wikifier errors.
More information about the progress of the project may be accessed via the related section of the multilingual project website at http://tramooc.eu/content/progress-and-latest-results.