Status - Automatic Translation project

[Felipe G]

Hey everyone, if you are interested in the progress and current status of
our Automatic Translation project, here's an update describing the various
core pieces involved in this feature, their current state, and what are our
next steps.


The current parts are:

- Language detection library (done): to offer language translation, the
first step is detect the language of the pages visited (so we know if it's
in the user's known languages or not). HTTP and HTML offer optional headers
specifying the page's language, but that is vastly unreliable due to many
servers just not being configured properly (left on default settings,
shared hosts on different countries, etc). So we must perform language
detection on the actual content served to be accurate. To do that we have
imported a library from Chromium called CLD2, which we have emscripted to
asm.js and are running it from a Worker thread. (bug 971047)

- Webpage parsing (done): When the user chooses to translate a page, we
must parse it to find all the text content that is relevant for
translation, in a structured way, and hopefully not much else. The goal is
to strip as much of the page's HTML structure as possible (think nested
divs and tables for page layout), while keeping the text content with its
surrounding HTML preserved in order to produce a quality translation. I've
played quite a bit on different approaches for this, and it was implemented
in bug 971043.

- Translation module (in progress): This is the module that is responsible
for getting the parsed data from the page and to format and send it to the
translation service. We're still not certain of which provider we'll be
using for translation, but this module should be mostly independent of
provider. We want to make it easy to swap in providers or to let add-ons to
add new providers that covers different pairs of languages too. (bug 971054)

- Notification infobar (done): This piece consists of the infobar + a URL
bar icon that is displayed when a translation offer occurs and while
translation is being performed. It's a bit more detailed than a normal
infobar since it persists while being used, and can be dismissed and
retrieved back from the URL bar, similar to how doorhangers work. This was
done in bug 979424.

Next steps:

- Finish translation module (bug 971054), hook up the language detection
(bug 971048) and parse/display the translation result from the provider
(bug 976556 and bug 976554)

- Style the infobar
  - while the notification bar is functionally working, its style is still
not fully implemented according to UX's specs. OSX - bug 988480, Windows -
bug 988481, Linux - bug 988482

- Verify performance characteristics to enable building the feature
  - While all of this code has been landing on m-c (and we're running its
test suite locally), it's not being built by default yet. Before doing that
we need to ensure that enabling it won't cause any performance problems.
Bug 996119 was filed to track that.


Felipe
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.mozilla.org/pipermail/firefox-dev/attachments/20140414/94596b7f/attachment.html>