Linguistic Potluck: Crowdsourcing Localization in Rails
From Gogaruco 2012
Presenter: Heather Rivers
Bridge gap between computers language and ours
- locales
- i18n
- localization or l10n
- token
- placeholder in a string
- hardest part of i18n
- should always store the highest level translation of text (store meanings)
- let your users help with translation
People will only fall in love with your app until they're able to make decisions in their native lanuage.
With a secondary language, they aren't as emotionally involved.
-
Design good software through separation of concerns
Storage > Selection > Transformation > Presentation
i18n enforces separating these concerns
(see slides for complete details on separation)Only intent is stored in template!
-
Simple UI
Localizations increases cost of every word
Well placed, widely understood symbol is awesome
-
Encourages you to ditch images for Semantic Markup and CSS
-
Forces flexible page layout to support different sizes and shapes of content
-
Make sure of your encodings
Content-Types consistent
MOJIBAKE will result
MOJIBAKE means "a terrible thing" (encoding error)
- HTTP Headers
- Google Analytics
morpheme - smallest chunk of a language which means something
cat has one morpheme
I saw the cats running away
6 words, 9 morphemes
(see slides)
Ratio can be calculated for any language from Analytic to Synthetic
Analytic is typically 1 to 1 ratio
Synthetic is typically much higher like 2 to 9
A word in a Synthetic language is a whole phrase in an Analytic language
Ratio predicts how much agreement there is in languages
Agreement is very WET
Synthetic is very DRY
Agreement categories
- Pluralization 2 cases for numbers in English Tons in Arabic
- Gender
Machine, Crowdsourcing or Professional Translators
Client-side solution from Google
Users have the ability to edit, approve and reject translations
(see slides for path through storage > selection > transformation >
presentation)
PROS: relying on 3rd party API and bad user experience
Storage is our boxes
Selection establish locale in path
Transformation apply stored rules to data in request
Presentation render result
I18n.translate
I18n.localize
I18n.transliterate
i18n activerecord gem
globalize3 gem
tr8n (pronounced tron, used by Yammer)
YAML sucks, so store translations into persistent key/value store
Heavily cache translations!
PROS: almost free, progressive enhancement and totally self monitoring
CONS: Delays so make everything as automatic as possible
PROS: Good for landing page copy
Turn around time can be reduced by using 3rd parties such as Gengo
5 to 15 cents a word
ReST API
Ruby client
