as good as human translators,” said Tom Simonite
in TechnologyReview.com. The search
giant is replacing its existing translation system
with a new one built around “deep learning,”
which uses math functions based loosely on
studies of the brain. To test its new software,
which has already been rolled out for translations
from Chinese to English, Google asked
fluent speakers to rank its translation results
for random Wikipedia entries and news articles
against work by human translators.
For English to Spanish, Google’s new system
scored 5.43 out of 6 on average, compared to
the 5.55 average score for human translations.
Google’s new system scored even closer to
human translators for French to English.
