I'm an absolute rookie with this stuff, but I'm building a translation program, which will hopefully enable people to create their own translators to share with the world. I'm particularly excited for its use by conlang enthusiasts, but I want to make sure it's got the features that would be needed. At the moment it can swap phrases, words, 'intrawords' (I don't know correct name) along with prefixes and suffixes.

So my question is: What features would be fundamental to a program like this? I think it would depend on the complexity of the language, and how close it is to English communication, but I'm sure there are some fundamentals that I'm missing - it just feels a little bit primitive at the moment.

Thanks for you time in helping me with this!

  • There is a project that uses esperanto as the middle language in an attempt to create a universal dictionary. dicts.info/uddl.php Google translate works by matching collocations (groups of words that are equivalent by comparison in a large corpus). Old fashioned attempts tried to grammatically parse the syntax and morphology and then map that to the equivalent structure in the other language, but that was hard and didn't work well. – MatthewMartin Jun 27 '13 at 16:13
  • Also, if you are doing this recreationally, I'd recommend starting out making a intralinear glosser, a gloss being a sort of marked-up, awkward, word for word translation. Professional linguists use glosses all the time, since not all readers of a linguistic paper can be expected to be fluent in all languages. A translation is an incredibly smoothly reading gloss, something hard to do algorithmicly. – MatthewMartin Jun 27 '13 at 16:15
  • Okay, so it sounds like currently the website (see link in question) only supports making these word for word gloss translations. If I wanted to allow people to use google's method, that would require a massive amount of text stored, right? Also, from what you were saying it sounds like actually mapping the syntax of one language to the syntax of another is too complex, right? If so, that's sort of a relief, but also a worry - if you wanted to create a translator for a conlang now, how would you do the mapping do you think? – user2198 Jun 27 '13 at 17:17
  • Conlang to conlang or natlang to conlang? Conlang to conlang, I imagine one could machine translate among languages of the same sort, i.e. variants of esperanto work the same way, machine translation is probably algorithmic. Variants on toki pona would be machine translatable. Relexes (pig latin & the like) are by definition easy to translate algorithmically. If you are making the conlang, you could pick a syntax that makes it easy to target into a certain natlang (just need to mark the same things, e.g. time, gender, number) – MatthewMartin Jun 27 '13 at 19:18
  • Mh, no offense, but if you're a rookie, why are you attempting to solve the (to my knowledge) hardest problem in all of AI? – Fryie Jul 19 '13 at 1:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy