ConcretePlanning

Alex discussing with Mike about the system and the LREC paper we want to write about it.

these things have to be pluggable for different language pairs

• input verification: is the user putting in obviously wrong text?
  • character set identification
  • unicode checks
  • langid -- maybe with langid.py by default?
• sentence segmenters for your language: this should be pluggable.
  • check out OmegaT's segmenters and how they work.
• Also text normalization routines. Let's write about GuaraniTextProcessing in a separate page...

how does the comment system work?

how does logging in work?

• can we just use oauth or whatever? how does it work to log in to, eg, AskBot?

accounts: what are they for?

• Will they prevent spam?
  • devil's advocate: could we just include a captcha?
• what do we want to keep track of, for each user?

thinking about what kind of content we want to allow people to post and privacy concerns

• Can you anonymously upload documents?
• Can you anonymously provide translations?
• Other sites (installs of Guampa) might want to do political things. There's that site for leftist translators already -- what is it? Does Mike have the link?

documents should have some tags so we/users know what they're for

• tag: who's the intended audience?
• tag: what is the genre of this document?

interface i18n

• maybe the interface should be in the document target language
• this should be configurable for sure. Is there support for l10n in angular builtin?
• For me and Mike, for example, it would be easy to translate Spanish articles into English, and easiest for us if the interface is in English.

Should the owner of a document be able to lock a sentence?

• How do we prevent (or at least channel) edit wars?

what's novel, that we can write about in the paper?

• Well, one thing is Guampa is novel in part because it's FOSS and there doesn't seem to be a good FOSS tool for this.
• unlike tatoeba:
  • we are for TM, MT, and CAT (in the long run). But we're explicitly about MT.
  • We are for translating documents.
• unlike traduwiki:
  • we are for MT and have proper sentence segmentation
  • traduwiki is for translating a document too.
• deep question: what's the benefit for volunteer translators?
  • Well, we do have activists and students...
• why is this different than Pootle?
  • Pootle's interface is kind of nonobvious, and it's meant for UI strings, it seems like. Not documents? And it's not for reading.
  • but it does have terminology come up...

think about how to demo the thing

• canned data?
• maybe we should use it ourselves, translate some wikipedia documents es->en maybe.