Seems like there are two different ideas of what is wanted/needed. Some people are looking for an expedient solution that just pulls results from an existing service (via API or redirect). Others are thinking of an actual translation engine.

This is in my area of expertise, and I would be willing to take on a translation engine project if there were:
- good community support for the idea
- some clear use cases
- at least one compelling reason not to use when a person wanted translation

Google's translation engine is quite good - not perfect, but quite good. Due to the significant manpower that goes into such a project, there should be very clear goals and a strong impetus for the project.

With that said, Apertium is not really suitable for a translation engine with global relevance. It is, as Joop pointed out, good for syntactically similar languages, but this condition is probably not common to the most common uses.

The problem of dealing with dissimilar languages is extremely complex. But, an incremental approach like Shaquil suggested isn't likely to be an ultimately successful strategy. There are a few fundamental issues that need to be worked on before doing anything language-specific. The first is to decided on a syntax-centric approach or a semantics-centric approach. The former is much easier but works much less effectively with dissimilar languages. The second is much more useful but exponentially more difficult. A semantics-centered approach best involves glossing the original text into a language-independent semantic representation.

When this basic approach is established, then you can start to work on mapping individual languages.

I am most interested in working on a semantics-centered approach, as this is where I've done much of my past work. And, I think it's the most useful. The basic approach is not to try to translate from Language A --> Language B, but to to map from Language A --> a gloss (language-independent representation of the meaning) --> Language B. There's a long list of reasons for this.
posted by [UserVoice Brent] • 5 years and 3 months ago Link