If you've ever confidently written puerto when you meant puerta, or frozen mid-sentence because pero and perro both surfaced at once, you've experienced one of the most reliable effects in memory research. Smart Blend and the spaced-repetition games now actively work around it, using a map of which of your words look, sound, or mean too much alike: the same vocabulary graph behind mastery-aware blending.
The interference problem is nearly a century old#
Psychologists have been documenting this since the 1930s, under the banner of interference theory: when two memories are similar, storing or retrieving one makes the other harder to keep straight. Language vocabulary turned out to be a textbook case. Two studies are cited constantly in the vocabulary-teaching literature: Tinkham (1993) and Waring (1997) both tested learners on semantically clustered word sets — all the colors together, all the fruit together — against the same words presented in unrelated, mixed sets. In both studies, the clustered sets took measurably longer to learn and produced more lasting confusions. Paul Nation's widely used vocabulary-teaching guidelines draw the direct conclusion: don't teach related or confusable words in the same sitting.
It isn't only meaning that triggers it. Words that merely look alike (puerto/puerta), sound alike (pero/perro), or happen to share a translation (two target words that both mean "big") interfere just as reliably as words that mean the same thing. Any of these is enough for two words to compete for the same slot in memory — and the more recently and closely together you met them, the worse the competition.
Three ways the app now acts on it#
None of this changes when a word comes up for review — your spaced-repetition schedule is exactly as sensitive to memory strength as it was before. What changed is which words end up sharing a session with which other words, and in what order new words arrive.
| What the app does | When it applies | Why |
|---|---|---|
| Interference-aware sessions | Two confusable words drawn into the same round | One is swapped for the next word in line; the round stays full length |
| Tricky Pairs | Both words in a pair already have practice history | Deliberate side-by-side contrast — the opposite move, and just as evidence-backed |
| Connectivity-first new words | A new word is chosen for a session | Words connected to vocabulary you already know go first |
Interference-aware sessions: keeping strangers apart.
Every review session — flashcards, matching, quiz, fill-in-the-blank — draws from a pool of words that are due. Until now, that pool was built without checking whether any two selections were confusable. Now, before a session is finalized, the app checks the drawn words against the vocabulary graph's confusable edges — pairs curated per language, like puerto/puerta in Spanish or hombre/hambre — plus words that happen to translate to the same word in your known language. If two land in the same round, one is swapped out for the next eligible word in the queue. Nothing about your schedule moves: the swapped-out word is still exactly as due as it was a second ago, and it surfaces in a different round, not later or earlier than planned.
This matters most for genuinely new words, where there's no existing memory to anchor either one — introducing puerto and puerta on the same day, before either has a foothold, is close to the worst possible order. Once a word has real history behind it, a single co-appearance with its confusable twin is far less damaging, which is exactly the condition the next feature waits for.
Tricky Pairs: the deliberate-contrast half of the research.
Keeping confusable words apart is only the first half of what the literature recommends. The second half — once a learner keeps mixing two words up anyway — is to put them side by side on purpose and make the contrast explicit. That's Tricky Pairs, a new section on the Practice screen: up to 10 word pairs the app flags as likely to confuse you, built only from words you've actually practiced (no pair appears until both sides have review history). Tap one and you get a side-by-side compare view — both words, both translations, both in view at once, so the difference is impossible to skim past.
Today the pairs are selected from linguistic signals in the vocabulary graph: form similarity, shared translations, and curated confusable edges per language. The app has also started quietly logging which specific words you mix up against each other inside the Matching game — a real signal about your actual confusions, not just what looks confusable on paper. That data collection is live now; using it to build Tricky Pairs from your real mistakes, rather than predicted ones, is groundwork for a future version, not a claim we're making about what ships today.
New words, ordered by what they connect to.
The third change touches new words specifically. Previously, when a session had room for unreviewed vocabulary, which new word appeared was essentially arbitrary. Now new words are ordered by their connections in the vocabulary graph to words you already know — word family, topic, or meaning links to your existing vocabulary push a word toward the front of the queue. This is the same frontier idea behind why your vocabulary is a network, not a list: a brand-new word with no connections has to be memorized from nothing, while one linked to something you already hold has somewhere to attach immediately, which is a genuinely easier learning problem, not just a nicer-feeling one.
What didn't change#
It's worth saying plainly: none of this touches the spaced-repetition math. Interval calculation, ease factors, when a word graduates from learning to review, how a lapse is handled — all of it is exactly what it was before. We changed which words share a room together and which new word walks in first. We did not change when any individual word gets reviewed. That distinction is deliberate: the scheduling algorithm is already well-supported by the same body of memory research; the part that was missing was awareness of which words fight each other once they're in the room.