Pre-training models with large crawled corpora can lead to issues such as toxicity and bias, as well as copyright and privacy concerns. A promising way of alleviating such concerns is to conduct pre-training with synthetic tasks and data, since no real-world information is ingested by the model. Our goal in this paper is to understand the factors that contribute to the effectiveness of pre-training models when using synthetic resources, particularly in the context of neural machine translation. We propose several novel approaches to pre-training translation models that involve different levels of lexical and structural knowledge, including: 1) generating obfuscated data from a large parallel corpus 2) concatenating phrase pairs extracted from a small word-aligned corpus, and 3) generating synthetic parallel data without real human language corpora. Our experiments on multiple language pairs reveal that pre-training benefits can be realized even with high levels of obfuscation or purely synthetic parallel data. We hope the findings from our comprehensive empirical analysis will shed light on understanding what matters for NMT pre-training, as well as pave the way for the development of more efficient and less toxic models.

Multilingual machine translation addresses the task of translating between multiple source and target languages. We propose task-specific attention models, a simple but effective technique for improving the quality of sequence-to-sequence neural multilingual translation. Our approach seeks to retain as much of the parameter sharing generalization of NMT models as possible, while still allowing for language-specific specialization of the attention model to a particular language-pair or task. Our experiments on four languages of the Europarl corpus show that using a target-specific model of attention provides consistent gains in translation quality for all possible translation directions, compared to a model in which all parameters are shared. We observe improved translation quality even in the (extreme) low-resource zero-shot translation directions for which the model never saw explicitly paired parallel data.

The training data for statistical machine translation are gathered from various sources representing a mixture of domains. In this work, we argue that when translating dialects representing varieties of the same language, a manually assigned data source is not a reliable indicator of the dialect. We resort to automatic dialect classification to refine the training corpora according to the different dialects and build improved dialect specific systems. A fairly standard classifier for Arabic developed within this work achieves state-of-the-art performance, with classification precision above 90%, making it usefully accurate for our application. The classification of the data is then used to distinguish between the different dialects, split the data accordingly, and utilize the new splits for several adaptation techniques. Performing translation experiments on a large scale dialectal Arabic to English translation task, our results show that the classifier generates better contrast between the dialects and achieves superior translation quality than using the original manual corpora splits.