Wafaa Mohammed

Language consists of collocated, structured, coherent groups of sentences referred to as a discourse. While for some languages, sentences provide a base unit of meaning, there are many sentences that contain discourse phenomena that are difficult to disambiguate at the sentence level. The goal of the project is to study the context utilization abilites of context-aware machine translation systems. We aim to evaluate models based on the ability to use the correct context, and not only the ability to properly generate text without necessarily utilizing the context. Moreover, we aim to improve context-aware translation models in their ability to modeling context-dependent discourse phenomena while not sacrificing translation performance.