Improving Resource-constrained Machine Translation and Text generation Using Knowledge Transition

Despite the significant improvements of Neural Text Generation (NTG) systems such as Neural Machine Translation and Natural Language Generation, there are still some open challenges in this domain due to constraints on resources. The resource limitation is sometimes rooted in the model’s incapability to use maximum resources (e.g., processing full document instead of processing sentences independently), or more typically because of the lack of annotated training data. This research addresses these limitations utilizing knowledge transition from high-resource NTG models to low-resource ones.