In the translation industry, improving machine translation (MT) output
by post-editing is commonly used to increase productivity and
consistency. However, some research showed that post-editing leads
to homogenization and normalization. And, although the application
of MT to literary translation has been limited, with the increased MT
quality due to neural MT, literary translators could also benefit from
this technology. Current translation environments allow users to
personalise MT systems further by training them on their previously
made translations or by adapting to changes a translator makes
during translation.
Considering that literary translators are under a lot of time pressure
(sometimes to the extent that more than one translator has to work
on the same text) and that the translator who started the translation
is not always the person to complete the work (e.g., illness, death),
MT could help by increasing speed and preserving the original
translator's style.
This research proposal uses techniques from stylometry (i.e.,
identifying textual features to classify texts) to examine (i) whether a
personalised MT system can preserve a translator's style better than
a generic MT system, and this (ii) even to the extent that it can be
used to ensure uniformity in translation projects with more than one
translator, and (iii) whether an adaptive MT system adapts to a
translator's style or whether it is the translator adapting to the MT
system's output.