Suggestion: Adjust Skill and Word Strength Decay Time by Lexical Similarity
For example, if two things sound alike, for example, a refrigerator is refrigerador in Spanish. The only difference is the t and the d. The word shouldn't decay that soon. Socks are calcetines in Spanish. You can't tell when you see calcetines. That word should decay faster than others.
I like the idea, but there are a few problems: a lot of people who for example learn "Spanish from English" might have a different native-language and therefore find other words easy and hard to remember. This might of course not be a problem anymore when the new incubator courses arrive. Also I think there's no easy way to implement /program this without having to go through all of the 1500+ words of a language manually. I think the latter is potentially going to be the bigger challenge, but maybe there is a way to solve it by comparing the beginnings of the words or calculate how many replacements would have to be made to make the words match...? Anyways I like the idea :)
I believe how it's setup now is fine because you're more likely to always get words like "refrigerador" correct and never have to peek so when you practice your weakest words "refrigerador" will probably never show up.
Also there is a lot of lexical similarity between multiple languages (not just two), so you're actually better off practicing "soup" "sopa" "suppe" and "soupe" so you know which goes with English, Spanish, French, German, Portuguese, etc.
Any spaced repetition system modifies the decay rate based on your measured performance. So "refrigerador"'s decay rate should get a lot slower once you've got it right a few times. This is a far more effective and simple method than trying to construct a software model for how easily a user might be able to guess words. As FrankySka points out, the first problem with that would be that you'd have to assess the user's knowledge of any related languages first. A lot of Duolingo's English-to-Spanish learners don't speak native-level English. On the other hand, a lot of them have previous experience with other romance languages.