We recently wrote about why your vocabulary is a network, not a list — the fifty-year-old research consensus that words live in your head as a connected web. This post is the other half: what we actually built to match that shape, and the concrete decisions behind it.
Which words made the cut — and why frequency beats textbooks#
Each language's graph holds roughly 3,000 words. That number is not arbitrary: vocabulary-coverage research (Paul Nation's work most famously) puts the most common ~3,000 word families at about 95% of the running words in typical text — the threshold where reading starts to flow instead of feeling like decoding. A graph of that size covers the vocabulary a learner from A1 up through B2 actually meets.
The selection comes from real usage — how often words actually occur in genuine, everyday language — refined with our own curated vocabulary. Frequency data from real speech has a property textbook lists lack: it reflects what people actually say, so conseguir (to manage to, to get) ranks where it belongs — far above tidy textbook staples that rarely appear in real conversation. Real language is messy, though: it arrives inflected (necesito, not necesitar) and full of names and noise, so a great deal of the work went into normalizing every entry to its dictionary form and checking its quality before it earned a place in the graph.
The connections between words#
Each word connects to its neighbors in several ways, each earning its place with a different job:
| Connection | Example | What it's for |
|---|---|---|
| Word family | comer → comida, comedor | Knowledge transfers along roots; families make new members cheap to learn |
| Synonym | feliz → contento | Meaning alternatives; depth of knowledge |
| Antonym | feliz → triste | Contrast pairs, natural quiz material |
| Confusable | puerto ↔ puerta | The pairs learners mix up — to keep apart early, contrast later |
| Topic | comer → cocina, almuerzo | Words that occur in the same situations |
| Similar | comer → cenar, almorzar | Words close in meaning without being strict synonyms |
The confusable edges deserve a special note because they encode the interference research directly: words that look or sound alike (pero/perro, hombre/hambre) are precisely the ones that should not be introduced in the same session, and should be deliberately contrasted once you keep confusing them. Mapping those pairs per language is what makes both behaviors possible, and both now run: review sessions keep fresh confusable pairs apart, and Tricky Pairs deliberately contrasts the ones you keep missing.
Why despierta now counts toward despertar#
Real learners meet real text, and real text is inflected. You save despierta from an article; the dictionary form is despertar. Treat those as unrelated entries — which is what a flat word list does — and your knowledge fragments across rows.
So the graph resolves the forms a learner actually meets to the dictionary words they belong to — and doing that across 17 languages is a tour of world morphology:
| Language | What form resolution has to handle |
|---|---|
| Finnish | Fifteen noun cases |
| Serbian | Full Slavic declension |
| Russian | Declension plus verbal aspect pairs |
| Turkish | Agglutinative suffix chains |
| Spanish | Rich verb conjugation |
| English | Plurals, -ing, -ed — comparatively tame |
| Japanese | Verb forms; much meaning carried by particles |
| Chinese | Barely inflects at all |
This is not a spelling-rule stemmer, which is why it catches the irregulars that break every rule-based approach: Italian va → andare and detto → dire, Spanish fue, French eu. Mastery of any form credits the dictionary word — the "parent terms" behavior that power users of jpdb and LingQ-style readers have long asked for.
Your layer: mastery, computed on your device#
The shared graph is the map; your knowledge is the overlay. LingoBlend already records how every review goes in the spaced-repetition games — that history tells the app how well you actually know each saved word (brand new, still being learned, solidly mastered, or slipping), using the same data the games run on.
Three practical properties of that design are worth stating, because they are the opposite of how "AI features" usually work:
- No AI at runtime. The intelligence went into building the graphs, not into calling a model every time you open the app. Nothing here waits on a server.
- On-device and offline. Your knowledge overlay is computed on your phone from data already there, and the shared map lives on your device too.
- Zero added cost. Nothing about the graph makes the app slower or more expensive to run — for you or for us.
What it unlocks#
Today the graph works behind the scenes — you will not find a "graph" screen in the app, and we think that is the right call (learning platforms that shipped literal graph visualizations, like Khan Academy's old knowledge map, ended up retiring them). What you see instead are the features built on it:
- Blends that know you — shipped. Text blending spends a limited budget of translated words per text; mastered words are wasted slots. With Blend Engine v2, the graph's mastery layer now skips what you already know, resurfaces words due for review, and prefers new words on your frontier.
- Level-aware selection — shipped. The graph knows how common every word is in real usage, so blends can target words just beyond your level.
- Your dictionary, browsable by connection — shipped. With the latest version of the app, Related Words on every saved word's detail page surfaces its family, synonyms, antonyms, confusables and topic neighbors, saved words group into families instead of alphabetized rows, and the Dictionary home recommends "because you know X" words pulled from your own frontier.
- Interference-aware review, and coverage you can see — shipped. Review sessions now keep freshly-introduced confusable pairs apart, and a Tricky Pairs view contrasts the likeliest mix-ups side by side; stories and books show a "% known" coverage badge before you start reading; and the graph also powers a standing estimate of your vocabulary size.
We write about each of these as they ship. For the research grounding behind the whole approach, see the science page, or start with the network model of vocabulary that motivated all of it.