I'm not sure that's true.

The founder seems to genuinely believe in what he's doing, I just think they've painted themselves into a corner. It's hard to accept your mistakes, even when they're easy to fix. It's easier to justify yourself as having done the best you could and brushing off the problems as "a price worth paying" for all the good you're doing.

But in this case, change probably wouldn't be easy at all. They've got a system that has no concept of meaning, and that's something that's hard to retrofit as an afterthought. Depending on how the main code was designed, it's quite possible that any real improvements to the system would require them ripping everything up and starting from zero. There have been lots of computer language-learning products over the years, but I've never seen any of them change significantly from one version to the next, which to me suggests that the code was written with the assumption that the method they had decided to use was correct, before they'd even had the chance to test it out themselves.

I'm writing some language software myself right now, and it takes a lot of work to keep your options open for a later change of direction, and sometimes it feels like it's a sign you don't trust yourself.

_________________
A language belongs to its native speakers, and when you speak it, you are a guest in their homes.
If you are not a good guest, you have no right to complain about receiving poor hospitality.

What he was describing seemed to me to be very random and piecemeal - I teach English to French-speakers, but I make sure that I offer them different forms of teaching materials and approaches that I can adapt to their needs, where you can combine conversation skills and grammar for example, and above all, that what you use is authentic language and that the subject matter interests them. They seem to have set up a system that applies the same formula for each language.

I think that one way they could do this would be to take the crowd sourced model further. Once people have achieved a certain amount of "reputation" in a language they could be able to "upvote" good material and "downvote" bad material. This would allow things that people don't feel make any sense to be weighted differently and eventually be taken out of the pool of available sentences.

The problem isn't really a technological one in this respect as a system like that could be easily bolted onto their existing data model. The real problem is a human one. Getting Duolingo to see that this is in fact an issue and is as important as addressing a software bug. I think that the argument could actually be made that these sentences are bugs in the system and that the software should not be allowed to generate examples that one could argue are grammatically correct but are at best awkward and unnatural because it causes confusion and raises cognitive friction. If it's not an out-right bug, it's bad UX period.

It all sounds like the usual Web 2.0+ business model of acting all 'passionate' about something, getting gawms to support it and then flogging it to a bigger piranha

_________________
__̴ı̴̴̡̡̡ ̡͌l̡̡̡ ̡͌l̡*̡̡ ̴̡ı̴̴̡ ̡̡͡|̲̲̲͡͡͡ ̲▫̲͡ ̲̲̲͡͡π̲̲͡͡ ̲̲͡▫̲̲͡͡ ̲|̡̡̡ ̡ ̴̡ı̴̡̡ ̡͌l̡̡̡̡.___

Thanks everyone for all the helpful comments, and Redwolf that sounds particularly sensible, have started a piggy bank with Gaeltacht on top of the wish list.

JulieA