Chat with Macduff Hughes (100 Days of Google Dev) - Coffee with a Googler Video Lecture | Coffee with a Googler - IT & Software

welcome to the Googleplex this is an
incredible place with lots of great
stuff being worked on every single day
before I worked here I always wondered
what it'd be like to come to the
Googleplex meet up with a Googler I have
coffee with them I just chat about what
they do how they do it and why they do
it and today we're going to do exactly
welcome to this episode of coffee with a
Googler today I'm speaking with Macduff
from the google translate team and we're
going to be learning all about some of
the cool things are available in google
translate and it's one of my favorite
technologies because I'm a big star trek
fan and it's one of the things that
brings like the universal communicator
from Star Trek much closer to reality so
we hope they up here so welcome Macduff
and tell us all about machine
translation what's it all about well
machine translation is using computers
to solve problems of understanding and
communication across languages okay it's
based on a technology called statistical
machine translation because most of how
it works is based on finding very large
numbers of examples of translations
mostly on the world wide web and using
statistics to see how various words and
groups of words and phrases tend to be
translated from one language to another
so that it can make reasonable guesses
about how to translate a sentence it's
never seen before I see so it's not a
straight like lookup table of this word
is this there's no yes no it has a lot
of statistics about possible
translations and some pretty advanced
algorithms to pick the best one and how
how to combine them to make something
natural sounding in the resulting
language cool so sometimes like a phrase
might be translated one way in one
document and translated another way in
another dog yes for for common phrases
will have many candidate translations
cool roiland for our language net now
what kind of like volume we do you get
in this how big is google translate well
we translate about a billion characters
a day about a hundred million words or
so for around a hundred million upwards
of that users per day Wow okay so now
there's a lot of languages and less
support to get that kind of volume well
the volume doesn't really come from a
lot of languages a small number of
languages account for most of the volume
I think God we do support a lot of
languages we worked very hard to make
sure we support even languages with
relatively low usage so we've been a big
push over the last few years to
up to 90 languages supported 90 Wow yes
so like what are some of the like less
common languages that are supported do
you have a favor do I have a favorite
well my personal favorite is well should
I learned a little bit a few years ago
right so crazy recently coffee with a
Googler and we're gonna need translation
for that underneath deal from valley on
yeah I started taking some courses a
couple a couple years ago and learned it
so we we do support that and translate
have for several years it doesn't get a
tremendous amount of traffic but one
thing you learned working on translation
is people are very proud of their
languages it's a very important cultural
thing to see their language represented
on the internet to have tools available
to use it and especially for minority
languages languages are threatened
people feel very strongly about them
yeah and well she's a great example i
lived in Wales for three years and the
pride that they take it's a beautiful
language to you that the bright they
taken in so one of the things that will
he do isn't is like an OC or translation
so you can hold up a camera and read
text yes this is something we launched
earlier this year was instant camera
translations so you can just open
translate and open your camera and
pointed a text and it translated
instantly on the screen right so like
for wash for the road signs in Cardiff
that we don't do well yet we we launched
earlier this year with support for
English to and from six languages French
Spanish Italian German Russian and
Portuguese but just the other day we
launched support for downloadable packs
for an additional 20 languages oh nice
yeah it's cool so I look forward to
having a play with that now one of the
things with machine translation is that
it's not always accurate it is not
always accurate as I'm sure there's some
funny things that can go around and
there are a lot of translation yeah
there are a lot of funny things Latin
produces a lot of hilarious examples
because it it's used in funny ways on
the web not the way people normally talk
people put Latin quotations around in
strange places or there's this thing
called lorem ipsum text that printers
used to fill out web pages there's a
conspiracy theory going around round to
the Internet earlier this year that
there were coded messages in our
translation hidden is lorem ipsum yes
yes there were any web developers you
probably have seen lorem ipsum like an
yeah I put like a text block on your
pages that that's the the text that they
auto populates you can see layout right
yes
yes some of the more straightforward
funny examples are we get the sense of a
word wrong a lot of words and languages
have many possible meanings but 11
favorite that we've seen is bedbugs okay
the word bugs in the number of languages
got and translated word for software
bugs so bug software defects so there's
a software problem with my bed yes okay
yes Wow or it can get a little edgy with
are you free today if the word free gets
translated word meaning no cost no price
no money um we gotta be careful what you
say yeah absolutely you freeze dunno
look I'm free today yes there's one that
I personally love and it was like this
this girl on YouTube and she will like
take song yeah right and she will
translate like the song into one
language and then into another and then
into another and then into another and
come like on a roundabout way through
about 20 languages back into English and
then she will she's a great singer and
she will sing the the translation have
you seen it those things are hilarious
yeah I think they really show how much
is always lost in translation get it
gets lost even faster with machine
translation but it's very interesting
because so many things are hard to
express especially song there are some
poetry even professional translators
really struggle to get get those
translated to know standable way yeah
but what struck me was remarkable was
that even though it was wrong you got
the gist of it yes and I went through
all these languages and I think it was
one like bohemian rhapsody and yeah the
opening line of bohemian rhapsody came
out it like you know like a mother I
shot a guy or something like that after
going through 20 different translations
and yes it was really cool well that's
what's amazing about it is even though
we do make big mistakes in translation
and a lot of times you might translate a
web page or a new story maybe not one
sentence is translated absolutely
perfectly but over the course of a long
enough document you can get the sense of
what it's about and the difference
between seeing a wall of something you
don't understand at all and knowing the
general topic of an article going to be
a big difference to people yeah and
that's one of the things that the
statistical base translation helps you
get right yes cool that's the best thing
for it is getting the gist of an out of
an article yeah now one thing I'd also
noticed that you've been working on
recently is that bringing people who are
expert speakers of the language into the
fold
so giving people the ability to feed
back into the statistical base
translation that says you know what in
my language we don't really say it like
this we say it like this yeah so you're
talking about translate community yes
that's something we introduced last year
a great thing about that as you said
experts generally that even have to be
an expert anyone who is reasonably
competent in two languages and wants to
improve the translation can go to
translate community okay and and help
this is something that kind of surprised
us because we're people are just
volunteering to help make our product
better but I mentioned earlier how
passionate people are about their
languages and especially minority
languages or things that are
underrepresented in the world and for
years people were saying either Google
why don't you support my language or why
aren't you better at supporting my
language and for years we had to say
well there's just not enough data out on
the web for us to train our models on
but eventually we just tried asking
people hey will you come and help and
there was a lot of enthusiasm especially
from these minority language communities
so that's kind of exciting though it's
like getting people involved so it
really is exciting they're trying to
make it more fun and engaging for people
and let them use it as a way to connect
with each other and to feel some
satisfaction what they're doing and it
becomes less about just pure machine
translation and people are getting
involved yes as well which is really
neatly keeping the language and I yeah
so increasingly for very common queries
and casual and conversational queries it
becomes more and more likely that when
you ask for a translation the
translation you see from google
translate will have been fri provided by
members of the translate community which
is very satisfying yeah very cool so now
one of the things you've just recently
brought out a new app right though
there's enough yes we did we just
updated the the mobile apps like I
mentioned that it adds 20 new languages
to the instant camera mode oh cool and
we've also made the speech and
conversation translation much faster
than was in the previous release so the
big feature from the previous release
was you could set up a conversation and
two people could speak and it would
automatically detect which language
people are speaking in and switch back
and forth so that's happening much
faster now so you don't need as fast a
network for it to cowork responsively we
also improved the accessibility of the
app for visually impaired
users so this this launched just the
other day on Android and iOS thanks I'll
have to download never blame these do
now if I'm a developer and I want to
start building translation into my apps
or using translation my how do I get
started so we have a google translate
API okay it's on the cloud platform
cloud google com and yes we pay we
charge by by the character for poor
translation and we support all 90
languages and we have a big developer
community developing applications
without so if I'm if I'm a cloud
developer using the cloud google com
console and developers com yes I will
see translator there is an option okay
cool so thank you so much Macduff that
was really awesome and it's great to
catch up and learn about some of the
things and find somebody else who speaks
a few words of Welsh okay so and thank
you everybody for watching this episode
of coffee with a Googler I hope we
learned as much as I did today it was a
to me it was really interesting to learn
about the great things that are
happening and translate and the great
things that are happening when the
community gets involved and as Macduff
said go on to the cloud developers
console and to be able to get access to
the api's if you want to build
translation into your own apps or you
can download the new translation happen
like play with some the new languages
that we spoke about it again thanks
everybody and be sure to tune to the
Google developers channel for more great
videos