While learning Thai -- which is a difficult language already -- I continually struggled with word definitions that should have been easy. How hard is it to remember the words for the various parts of your head and body? With just word definitions, apparently, it's really, really hard.
But then I supplemented this with other tools. One teaches a few words, then throw those words into full phrases in little quizlets. You pick out the written phrase which you think matches the spoken phrase. It always includes one or some of the words you've learned. It puts the words in context, cementing the definition, teaching you grammar, and "foreshadowing" new words. It's skyrocketed my retention and comprehension.
Pictures also help. Children's books with audio narrating are huge.
I guess my point is, Anki with word definitions alone is definitely garbage, or less than ideal at best. Which is why most language learning resources stress that it alone isn't enough - and most learning resources who push Anki (or SRS in general) don't say "do only this." It's part of a balanced breakfast, so to speak. Incorporate it into your workflow, but don't make it the workflow.
So, the flashcard group was just using flashcards as a supplement to extensive reading.
Well, at the end of the treatment period, both groups had the same vocabulary retention. But a couple months later they tested again, and the "no flashcards" group had better retention of the target vocabulary. And not by a small amount, either.
It's not terribly hard to find examples of second language acquisition researchers expressing skepticism about flashcarding, though. I don't necessarily find all of it universally convincing (if I did, I wouldn't have 99 reviews left to do today in my 中文 deck), but I do have to admit that a lot of the evidence people claim in support of using spaced repetition systems for language learning has the look of pseudoscience. It largely consists of taking some very limited, tightly focused lab studies, and massively over-extrapolating from them.
But I wouldn't want to over-extrapolate from the studies that claim flashcarding doesn't help, either. The protocols often focus on a very minimalist implementation of flashcarding that's easy to study. Or the protocol is really designed to test some theory about how memory works and not some theory about study strategy. (This second example describes basically all of Ebbinghaus's work, by the way.) One could easily argue, for example, that using a flashcarding application to rote memorize a pre-made deck of words and phrases that somebody else chose is a fundamentally different practice from sentence mining your input materials.
What does this refer to? Generally I've referenced the Improving Students’ Learning review by Dunlosky et. al. [0] to see what is supported by the literature. The two techniques listed there as having 'high' utility -- Practice testing and Distributed practice -- are most easily accessible through usage of SRS-based flashcards.
[0] https://pcl.sitehost.iu.edu/rgoldsto/courses/dunloskyimprovi...
Like, imagine if someone learning to program put a huge amount of their effort into using flashcards to memorize the functions in their preferred language's standard library. I think most people would agree that this person is wasting their time because IDE tooling and Internet searches make it really easy to look up that information on an as-needed basis. So the real effort should be put into learning skills like domain modeling or how to read and debug code.
I think that the same principle might apply to most subjects. Even in language learning, one of the places where SRS is most popular. Academics seem to be pretty skeptical of using any sort of flashcarding to learn vocabulary. In part because the empirical evidence doesn't actually seem to support it, and in part on theoretical grounds. The contemporary model for how the brain models language is fundamentally connectionist. Definitions and lemmas aren't something we naturally memorize as a list of facts and then plug into a grammar, which is why native speakers often have difficulty reciting conjugation tables or defining words.
I think that formal education's tendency toward representing subjects as lists of factoids is fundamentally another manifestation of the McNamara Fallacy (https://en.wikipedia.org/wiki/McNamara_fallacy). As long as educators have a mandate to (ostensibly) objectively quantify students' progress with grades and scores, they will have a need to represent the subject manner in a manner that permits easy quantification. And they will have to do so regardless of whether it actually makes sense to do so for whatever subject they're teaching.