Computers have been teaching themselves to
translate text for some time now,
but most methods are concerned with
translations on entire sentences.
We address the problem of simultaneous
machine translation for distant
language pairs: from verb-final (SOV) to verb-medial
Simultaneous translation is the process of translating
before a sentence
is complete. When humans do this, it is
called simultaneous interpretation.
Much of the prior work in this area has focused on using
We use machine learning to allow the system to teach
itself how to
create better simultaneous translations.
We do not program translation strategies into the
Instead, we provide our imitation learning system with
On the basis of these features, the system decides when
to translate sections of a sentence. Does it WAIT
for more words
or COMMIT to a partial translation?
We use imitation learning to allow the system to produce
simultaneous translations on the basis of its
features. The focus
is then shifted to better features, not programming
Every time the system makes a decision, it compares it to
decision that it could have made. In this
way, the system
learns from its mistakes and its successes, much in the
a professional simultaneous interpreter would.
Some languages, such as German and Japanese, have
constructions, in which the main verb appears at the end of
How does one translate something that hasn't yet been
One way is to guess what the speaker will say based on
speaker has already said. We integrate a verb
to allow for translation from SOV to SVO languages.
Through reinforcement learning, the system learns when
trust these predictions.
A Cumulative Metric
BLEU has been the
standard metric for most languages in machine
LBLEU sums the BLEU scores of the
incremental translations, while
translation. We introduce latency BLEU
(LBLEU), which takes into
account the expeditiousness of the simultaneous
weighting the final translation in proportion to the
As a result, translations that are fast and accurate, as
opposed to merely fast or accurate, achieve higher scores.