A dictionary of saved words has always been a list: one row per word, alphabetized, a translation next to each. That's an honest record of what you clicked "save" on, but it's a poor model of what's actually in your head — nobody's vocabulary lives as a sorted column. The app now reads that list the way your brain does: as a network of connected words, and it surfaces the connections in three places.
From two thousand rows to a few hundred families#
Open All Saved Words and there's a new toggle next to the search bar: A–Z or Families. Switch to Families and comer, comí, comida, comedor collapse into one card — a root with its members listed underneath, instead of four unrelated rows scattered across the alphabet.
The grouping runs two of the graph's structures at once. Inflections — comí being a past-tense form of comer — resolve through the same form index that already lets Smart Blend recognize a mastered word in any of its conjugated forms. Derivational relatives — comedor (dining room) sharing a root with comer but not a paradigm — connect through the graph's family edges instead, which is a different kind of link built to catch exactly this case.
The number this produces is usually smaller than you'd guess, and that's the point. Two hundred saved words might genuinely be closer to a hundred and twenty families. That isn't fewer words than you thought you had — every one of the two hundred is still there, still reviewable, still counted in your stats. It's a more honest shape for the same pile: it tells you where you've been building depth on a root (several forms of comer saved) versus where you've only touched the surface (one isolated word with no relatives saved at all), which is a genuinely different kind of progress and worth being able to see. Families view is additive — A–Z is still there for anyone who just wants to scan or search alphabetically.
Every word's neighborhood, one tap away#
Tap into any saved word's detail page and there's a new section: Related Words. It's chips, grouped by relation type, pulled straight from the word's node in the graph:
| Chip | Example (from comer) | What tapping it does |
|---|---|---|
| Family | comida, comedor | Not saved → one-tap save (auto-translated). Already saved → jumps to that word's own page |
| Synonym | cenar (to dine) | Same |
| Antonym | ayunar (to fast) | Same |
| Confusable | — (see below) | Same, plus a note on what distinguishes the pair |
| Topic | cocina, almuerzo | Same |
This is, concretely, what it means to browse your mental lexicon's neighborhood instead of scrolling a list. You don't have to already know a word exists to find it — you find it by standing next to a word you do know and looking at what's connected. A synonym chip shows you cenar sitting beside comer, which is a different and more useful kind of exposure than meeting cenar cold in a frequency list ranked #340.
The confusable chip is worth a specific mention, because it's the one relation type that isn't primarily about adding vocabulary — it's about a pair you already have. When two saved words are known confusables (puerto/puerta, pero/perro), the detail page flags the pairing directly rather than waiting for you to mix them up in a game first. The review-side handling of confusable pairs — keeping them apart while new, contrasting them once one is solid — is its own feature; see how LingoBlend keeps similar words apart.
Not every word has every chip. A rarely-connected word might show only two or three; a hub word like comer can show all five. Empty categories are simply omitted rather than shown as blank — a page padded with "no synonyms found" for a dozen relation types would bury the connections that do exist.
Because you know comer: recommendations that start from your strongest words#
The Dictionary home has carried a recommendations row for a while; what changed is where the suggestions come from. Instead of a generic frequency list handed to every learner regardless of what they've saved, the row now reads "Because you know comer" and suggests comida, cocina, almuerzo — words sitting one connection away from vocabulary you've already made strong.
The logic is the same insight Blend Engine v2 uses to pick which words to translate: the cheapest place to learn something new is adjacent to something solid. A word connected to one you know well inherits part of its context for free — you already have the root, the topic, or the situation it lives in, so there's less to learn from scratch. A recommendation pulled from a distant, unconnected corner of the frequency list has none of that scaffolding, even if it's technically just as common.
New users aren't left out of this, because there's a fallback with no cold-start gap. Before you've saved enough words and played enough review games to have a real "strongest vocabulary" to branch from, the row suggests curated frequency-ranked starter words for your level instead — the same words a level-aware blend would introduce first. As your dictionary grows, the graph has more of your words to connect from, and the row quietly shifts from level-based to connection-based without any setting to find or toggle. If you're curious what "your level" is actually derived from, we cover the vocabulary-size estimate the app now shows separately.
One more thing worth stating plainly: recommendations aren't a queue you have to clear. Save a suggested word and it drops out of the row on its own — there's no "dismiss" needed, and nothing re-suggests a word you've already added.