Many machine translation toolkits make use of a data preparation step wherein raw data is transformed into a tensor format that can be used directly by the trainer. This preparation step is increasingly at odds with modern research and development practices because this process produces a static, unchangeable version of the training data, making common training-time needs difficult (e.g., subword sampling), time-consuming (preprocessing with large data can take days), expensive (e.g., disk space), and cumbersome (managing experiment combinatorics). We propose an alternative approach that separates the generation of data from the consumption of that data. In this approach, there is no separate pre-processing step; data generation produces an infinite stream of permutations of the raw training data, which the trainer tensorizes and batches as it is consumed. Additionally, this data stream can be manipulated by a set of user-definable operators that provide on-the-fly modifications, such as data normalization, augmentation or filtering. We release an open-source toolkit, SOTASTREAM, that implements this approach: https://github.com/marian-nmt/sotastream. We show that it cuts training time, adds flexibility, reduces experiment management complexity, and reduces disk space, all without affecting the accuracy of the trained models.

Discourse parsing is a challenging task in NLP and plays a crucial role in discourse analysis. To enable discourse analysis for Hindi, Hindi Discourse Relations Bank was created on a subset of Hindi TreeBank. The benefits of a discourse analyzer in automated discourse analysis, question summarization and question answering domains has motivated us to begin work on a discourse analyzer for Hindi. In this paper, we focus on discourse connective identification for Hindi. We explore various available syntactic features for this task. We also explore the use of dependency tree parses present in the Hindi TreeBank and study the impact of the same on the performance of the system. We report that the novel dependency features introduced have a higher impact on precision, in comparison to the syntactic features previously used for this task. In addition, we report a high accuracy of 96% for this task.