Cross-lingual document similarity estimation and dictionary generation with comparable corpora

This paper proposes an approach for performing bilingual dictionary generation even when trained on widely available comparable bilingual corpora. We also show its capability to provide cross-lingual similarity estimates that correlate well with human judgments. We implement an approach using a nonlinear bilingual translation model that we train using comparable corpora. We propose a method using word embeddings and kernel approximation to train scalable nonlinear transformations. We demonstrate that this novel method works better on a majority of evaluated language pairs.