Put mitigate on the front of a flashcard and to make less severe on the back. A student answers it correctly. The flashcard goes back into the deck and returns a few days later, and each correct answer pushes the next showing further out. That is spaced retrieval, and it is the mechanic underneath most of the vocabulary work in our product. Assume for a moment that it does exactly what it was designed to do, and that weeks later she can still give the definition without pausing. The question worth asking is what she has then, and what she does not.
Asked that directly, our answer was that it leaves her with just word association to an extent. That was not modesty for effect. A definition is an abstract statement about a word. For it to become usable it has to come down to a particular sentence in a particular piece of writing, and then travel back up, so that the abstract version has somewhere to sit in what the student already knows. A flashcard makes neither journey. It states the abstraction and stops, and it stops in the same place for every student holding it. What she ends with is a dependable link between two strings of characters. That is a real thing to have. It is not the same as knowing where the word goes and what tends to sit next to it.
A student who had been using the product for two days wrote to us unprompted about this. She is Chinese and had scored 6.5 on IELTS. Explaining why she found the spoken roleplay the most useful part of the product, she wrote: "For Chinese students, we only remember and spell but don't know how to use it". Nobody had asked her about theories of vocabulary. She was describing what learning English had been like for her and for the people she studied alongside. That is one student, two days of use, volunteered rather than solicited, and it is worth exactly what a single observation of that kind is worth.
What makes it matter here is that her complaint and our own account of meaning arrive at the same defect from opposite directions. We got there by thinking about what a definition is. She got there by having done it for years.
Where applied linguistics puts the unit
The position we had reached clumsily is ordinary inside applied linguistics, and it does not need philosophy propping it up. Usage-based accounts of language, associated with Nick Ellis and Michael Tomasello, hold that what a speaker knows is an inventory of constructions: form-meaning pairings at any size, from a single morpheme up to a whole clause frame, learned by abstraction from encountered use. On that account the orthographic word is not the unit of knowledge. The chunk is.
What someone acquires when they acquire mitigate is a distribution rather than a gloss: that it takes objects like effects and risks, that it belongs to the register of a written report rather than to a conversation in a corridor. A flashcard reading mitigate against to make less severe has removed precisely the information the theory says is being learned. On that reading the front of the flashcard should say mitigate the effects of.
We had said something adjacent to this ourselves, without knowing there was a literature attached to it. Asked whether there is such a thing as knowing a word, our answer was that meaning is completely contextual, and that there is frequently no direct translation from a word in one language to a word in another. It works phrase to phrase or sentence to sentence. Sometimes no equivalent exists at all, and that is roughly the way it is. Whatever else that is, it is a claim about units. If the smallest reliable unit of mapping between two languages is the phrase, then a product whose smallest unit is the word has picked the wrong size, and a flashcard setting one word against one gloss presents a tidier correspondence than we ourselves believe exists.
So, plainly: our core mechanic is built on a theory of meaning we do not hold.
How far the flashcard evidence reaches
The flashcard system rests on the psychology of memory, on spacing and retrieval. That is a serious body of evidence and we are not embarrassed to be using it. But the psychology of memory is not a theory of language. The theory of language we do appear to hold, which we stated in three different ways across three separate questions before anybody pointed out that we were repeating ourselves, says the unit we built the product on is too small to carry meaning.
None of which is an argument for deleting flashcards, and anyone who claims applied linguistics rejects them is misrepresenting the field. Paul Nation is explicit that flashcards are among the most effective techniques available for deliberate vocabulary learning. His Four Strands paper, in Innovation in Language Learning and Teaching in 2007, also attaches a constraint to that endorsement which we would rather state than hide. Deliberate, language-focused study is roughly one quarter of a learner's time. The rest goes to meaning-focused input and output and to fluency development. Our vocabulary work sits inside that one quarter, and a reader should not assume it covers the other three.
Elgort (2011), in Language Learning, found that deliberately learned words produced both form priming and semantic priming, which indicates they become genuine lexical knowledge rather than inert labels, a finding that critics of flashcards routinely leave out. Kim and Webb (2022), also in Language Learning, is the meta-analysis showing the spacing effect replicating in second language learning specifically.
What that work supports is narrower than a marketing page would like it to be. Deliberate spaced retrieval efficiently builds receptive knowledge of the form and meaning of items the learner has selected, which raises how much of a text they can follow. It does not by itself produce writing that puts the right words next to each other in the register a marker expects, and we do not claim that it does.
One further finding costs us a design we are fond of. Nakata and Suzuki (2018), in Studies in Second Language Acquisition, report that grouping semantically related words together in the same set impedes learning. Our topic-based decks are assembled the way that evidence says not to assemble them. We have not fixed it.
The extraction holds, the unit does not
The part of the product that survives this argument best is the extraction. A student uploads their own course material and the vocabulary comes out of that, so the words are the ones their own lecturers chose, in their own department. Hyland and Tse (2007), in TESOL Quarterly, argued that academic vocabulary is discipline-specific and that general lists mislead, a position contested since by Durrant (2016) among others. Building the list from a student's own course is our answer to that question, and it was the founding idea rather than a retrofit. Whatever is wrong with the flashcard, the words on it are at least the right words.
What has not changed is the unit. The flashcard still shows a word against a gloss. Asked what it would cost us to move to the chunk, our answer was that it would be hard but we could probably do it, and that the product would then look nothing like other language apps. That cost is not engineering. It is recognisability, which is a real cost in a market where a user decides what an app is in about three seconds.
There is one more thing to put on the record, because it bears on everything above. The place in the product where use actually happens is the spoken roleplay, and it is the feature that student named as most useful, for exactly the reason she gave. The roleplay does not currently run on her uploaded course material. Vocabulary extraction does. The voice tutor and the writing tutor do not. That is buildable and it is not built, and until it is, the claim that the product teaches the language of a student's own course holds for one feature rather than for the product.
The trade we cannot settle
The same student who told us that Chinese students only remember and spell also asked us to let her create flashcards with a single click, because making them is troublesome when you are not very motivated. In the sentences just before that request she had explained that the making is the useful part, since assembling the flashcard forces her to recall and rehearse the word. On her account, then, the effortful step is the one that does the work, and it is also the step she wants taken away. We have no data of our own on either half of that. She set the trade out more clearly than we would have.
That is the shape of the thing we cannot resolve. Moving the unit from the word to the chunk is the more defensible design, and it makes the product harder to recognise, and a mechanic nobody opens teaches nobody anything. Making the flashcard easier to build is the change she asked for, and by her own description it removes the part that was working. We do not know how to weigh the version that is right against the version that gets used, and we are not confident the answer is the same for every student who opens it.
