Hacker News Reader

Comment by yorwba | Hacker News Reader

One way to get content would be https://tatoeba.org For example, to teach "oso" = "bear", there are already ten different example sentences with audio recordings: https://tatoeba.org/eng/sentences/search?query=oso&from=spa&... (Okay, one of them is actually "oso" = "dare".)

View on HN

The Nordic Language Processing Laboratory has a sizeable collection of open source parallel texts: http://opus.nlpl.eu

Click on any of the headings in [brackets] on the top of the page to see the language pairs available for download for that collection.

The problem with tatoeba is that the content doesn’t seem to be curated. I checked the translation of I love you in Japanese recently and more than half of the listing is hot garbage that would never be uttered by a native speaker. It’s just like these sentences where contributed by learners who just finished their first lesson.

Japanese is a special case, because most of the Japanese sentences come from the Tanaks corpus, which... consists of sentences written by learners of either English or Japanese who likely weren't all that qualified for the task.

However, the content is actively getting curated. E.g. https://tatoeba.org/eng/sentences/show/137571 is a translation of "I love you." owned by a native speaker. The search form doesn't make it easy to find native-speaker translations, though. The search option to limit results to native-speaker sentences only works if you're searching in the target language. E.g. if you want usage examples for "愛": https://tatoeba.org/eng/sentences/search?query=%E6%84%9B&fro...

If you're only using Tatoeba as a data source for automatic creation of course material, though, you can filter using whatever criteria you want.

Reply on news.ycombinator.com