3 ms·
Yes, the huge repository of raw materials is likely the hardest part. You can try crowdsourced collections ( https://tatoeba.org https://tatoeba.org , https://d
by yorwba 11mo ago
Yes, the huge repository of raw materials is likely the hardest part. You can try crowdsourced collections ( https://tatoeba.org https://tatoeba.org , https://datacollective.mozillafoundation.org/datasets?q=common+voice https://datacollective.mozillafoundation.org/datasets?q=comm... , https://opus.nlpl.eu/OpenSubtitles/corpus/version/OpenSubtitles https://opus.nlpl.eu/OpenSubtitles/corpus/version/OpenSubtit... ) but you'll quickly run into data quality issues. My personal solution is to do manual data curation on the fly, but I think an app that occasionally throws up garbage and asks its users to pick out the good parts is unlikely to get popular.
- amelius 11mo agoMaybe the free version of the app could do the collaborative filtering part. And in the paid version you'd get the high quality content.
- yorwba 11mo agoThat lets you turn the problem of figuring out which part of your content is badly machine-translated into the problem of figuring out which of your users have enough attention to detail to spot badly machine-translated content in a language they themselves are still learning. Though I guess if you show paying users only content that lots of people think is good, at least it reduces the chances that a paying user notices an issue and complains, so it could still work in that sense.