Retrieval-based approaches for natural language processing (NLP) aim to augment neural models by retrieving similar texts to a given input during training and/or inference times. In the context of neural machine translation (NMT), retrieval-based methods have led to impressive gains in translation quality by integrating similar translations into the NMT architecture (i) through modifying the attention mechanism, (ii) in the form of additional network components or (iii) via a data augmentation method. 

Due to their ability to effectively utilize similar translations, RBNMT can be seen as a leap forward in the integration of NMT and translation memory (TM) systems, which are commonly used in the computer-assisted translation workflows, while blurring the distinction between the two technologies. This project aims to use state-of-the-art NLP methodologies to further improve NMT systems and TM-NMT integration strategies.