Leveraging Monolingual Data for Crosslingual Compositional Word Representations
This work addresses the challenge of crosslingual document classification by enabling more effective use of monolingual data, though it appears incremental as it builds on existing methods with specific enhancements.
The authors tackled the problem of inducing compositional crosslingual word representations by proposing a neural network architecture that leverages both bilingual and monolingual data, achieving state-of-the-art results with 92.7% accuracy for English to German and 84.4% for German to English, improving by 0.9 and 7.7 percentage points respectively.
In this work, we present a novel neural network based architecture for inducing compositional crosslingual word representations. Unlike previously proposed methods, our method fulfills the following three criteria; it constrains the word-level representations to be compositional, it is capable of leveraging both bilingual and monolingual data, and it is scalable to large vocabularies and large quantities of data. The key component of our approach is what we refer to as a monolingual inclusion criterion, that exploits the observation that phrases are more closely semantically related to their sub-phrases than to other randomly sampled phrases. We evaluate our method on a well-established crosslingual document classification task and achieve results that are either comparable, or greatly improve upon previous state-of-the-art methods. Concretely, our method reaches a level of 92.7% and 84.4% accuracy for the English to German and German to English sub-tasks respectively. The former advances the state of the art by 0.9% points of accuracy, the latter is an absolute improvement upon the previous state of the art by 7.7% points of accuracy and an improvement of 33.0% in error reduction.