Sentence sampling based on frequency

Question

I have a database with 300k+ Russian sentences and their English translation. My goal is to use these sentences as flashcards, so the users can learn the top N most frequent Russian words (let's assume N = 10k).

A requirement is that the easiest sentences are shown first, and more complex sentences get slowly introduced as you progress. The notion of easy/complex relates only to word frequency and maybe sentence size to break ties.

I thought on doing the following:

1. Get the word frequency from the entire corpus
2. Rate each sentence

What's a good and simple way to achieve step 2? A simple average would work, or it's better to multiply each word frequency?

I appreciate any suggestions on this, thanks!

Note

Since Russian has declensions (noums/adjectives/pronoums change their endings according to grammatical function), that bring's another problem, but maybe I could ignore it.

Answer 1

One option is to take your 10k words and divide them into e.g. 100 percentiles, each with 100 words. The 101th group would be the rest of words. You can rank your sentences by the most complex word in it. For example if you have sentence with 5 words and 4 of them belong to 1st percentile and the last one to the 7th percentile, the sentence is in the 7th percentile. This ranking would guarantee that the first group of sentences will be from the most simple words only.

Regarding the flection in Russian, it may be fine to consider different forms as different words. The learner will get them gradually. This approach is valid especially if you don't provide explanation of grammar.