Lecture

Machine translation (MT), like other tasks in Natural Language Processing,
has  progressed tremendously since the adoption of neural networks around a
decade ago. As a consequence, this technology is being widely adopted, both
by language professionals such as translators and by society at large. This,
however, can have negative implications. For example, it has been shown that
translations produced by MT tend to have low lexical variety, low creativity,
and exhibit syntactic calques. In this talk I will discuss these issues and
why they matter, linking them to the notion of algorithmic biases. Subsequently
I will describe methods that are being proposed to improve MT in this regard,
including our own work.

Antonio Toral is associate professor and coordinator of the Computational
Linguistics group of the Center for Language and Cognition at University of
Groningen, Netherlands.