Abstract:
This talk focuses on the intersection of linguistics and new approaches to translation. Using computational techniques to work with large amounts of data and automatically generate translations poses a series of questions around choice in language use. These include a range of statistical models, as well as their strengths and limitations in capturing linguistic patterns. Even if statistical models are quickly learning from linguistic patterns in training data (i.e. syntactic structures, collocations, and discourse markers) and use them to make translation choices, idiolects are one of the key issues to address at this point. How can these models be adapted to handle idiolectal variation? Translation is a process of making choices among multiple possible versions, so this talk explores how statistical translation models navigate between linguistic patterns and idiolectal variation, and how the concept of choice plays an essential role in translation when using such models.