Hi funcDropShadow, that's a good point. We really want to expand to other language pairs (e.g. German-Spanish). In fact we are also not native English speakers :)
Hi James, thank you for your comment. It's great to hear you say that because it is exactly what we had in mind when we started the project -- we were also pretty tired of the subscription model in language learning apps :)
Indeed, using frequency alone will lead to a suboptimal set of sentences. Our approach was to choose sentences that increase smoothly in complexity, where we estimate complexity through a combination of criteria: frequency of vocab, tenses, number of clauses, etc.
I wasn’t sure how many sentences were included but bought anyway (you might want to add this to the FAQ unless somehow I missed it elsewhere). Seems to be about 2000 sentences per deck. Learning Spanish so this gives me access to 6000 sentences.
Hey Trias!
Yes, our flashcards run on top of Anki, so they can be used completely offline - that was of our main intentions when making this product, as many apps require internet access.
A little of backstory about this project: my co-founder and I are big fans of Anki, which is an awesome flashcard software and a really effective tool for language learning.
However, it's hard to find quality decks online. There are some apps with good flashcards, but they tend to be expensive. You certainly can make your own, but that's really time consuming. That's why we decided to build Deckmill, a repository of high quality, but affordable flashcards.
For those are interested in the tech, this is how we did it:
1. Used a mix of ML + hand curation to select 3000 sentences from tatoeba.org (a massive dataset)
2. Then machine translation to translate the sentences
3. Then we had native speakers review the translations
4. Wavenet to generate audio
5. genanki to put together the cards
Fluent speaker of Spanish as second language. The audio here is terrible for learning the language, "Mi hermano es medico" is pronounced as though two Venezuelans are talking at a dinner party. It needs to be slower and more distinctly enunciated for a learner.
Hi spoonjim! In our experience it's important to train your ears to listen to native speakers as early as possible in the learning process. However, you are correct that listening to slow speaking is also very pedagogical -- this is exactly why we provide a 0.5x speed version of the audio, alongside the 1x (native-like) version.
Yes, I tried the 0.5x version, but it's not the way that a Spanish teacher talks. A good Spanish teacher emphasizes the parts of the sounds that are different from English, and doesn't slow down even close to 0.5x -- it's more like 10%, but a much clearer 10%.
In this example, a Spanish teacher would have put more of a pause between "Mi" and "hermano", not such a weird elongation of the "o" in "hermano", and less emphasis on the "es" (even a native conversation would not have that kind of stress on the "es").
Hi ngokevin! The Anki community is fantastic and there are some great decks out there. However, we found that the quality, format and even difficulty tend to vary wildly across decks. For example, many decks have low quality audio or don't include it at all; others have poor sentence selection, etc
Thanks for buying our decks! Each deck has 1000 cards, so each language has 3000 cards total across levels.
Perhaps you only downloaded the sample decks? Let us know if you have any further questions by dropping an email at [email protected].