Pivot-Based Topic Models for Low-Resource Lexicon Extraction

This paper proposes a range of solutions to the challenges of extracting large and highquality bilingual lexicons for low-resource language pairs. In such scenarios there is often no parallel or even comparable data available. We design three effective pivotbased approaches inspired by the state-ofthe-art technique of bilingual topic modelling, extending previous work to take advantage of trilingual data. The proposed models are shown to outperform traditional methods significantly and can be adapted based upon the nature of available training data. We demonstrate the accuracy of these pivot-based approaches in a realistic scenario generating an IcelandicKorean lexicon from Wikipedia.