Vocabulary Crossword Puzzles That Teach by Retrieval
September 30, 2026
A normal crossword rewards recognition: a clue points sideways at a word you may already know, and crossing letters finish the job. That is good entertainment, but weak vocabulary training, because the puzzle has no memory of what you missed and no obligation to bring the word back at the right time. Vocabulary crossword puzzles become a learning system only when the crossword is the interface for repeated contextual retrieval.
Vocabulary crossword puzzles need memory
A printed crossword can be clever. It cannot know that you hesitated on obdurate yesterday, solved pellucid instantly, and have not seen trenchant for nine days.
That missing learner state is the difference between a word game and a vocabulary system. A vocabulary crossword needs one schedule behind the grid: each word returns when it is due, not when an editor happens to reuse it.
Lexi is built on that distinction. It teaches vocabulary through spaced repetition delivered as dynamically generated crossword puzzles. The grid is not the curriculum. The word schedule is.
In cognitive-science terms, this combines spaced repetition, which spaces reviews over time, with retrieval practice, which strengthens memory by forcing recall. The crossword provides the constraint: you have to produce the word, not merely recognize it.
Memorizing definitions trains the wrong association
A definition is useful once. It establishes what the word points at.
Memorizing that definition is a different skill. It trains recall of a sentence about the word. Worse, if the same definition appears every time, you can overfit to its phrasing: the wording becomes the cue, and the word itself stays fragile.
That is why Lexi does not treat definitions as the main review surface. Definitions live on the word card. Retrieval happens later inside a sentence.
The distinction matters because recognition is cheap. You can recognize a word in a paragraph and still fail to produce it when speaking or writing. I wrote more about that failure mode in Why you recognize a word but can’t find it when you need it.
Context clues force the word to do real work
Lexi clues are fill-in-the-blank sentences plus a meaning-evoking hint. The blank forces the word to behave grammatically. The hint narrows the intended meaning. Crossing letters help, but they do not replace semantic recall.
That is stricter than a definition prompt. If the answer is laconic, the clue should make you retrieve terse expression inside a usable context, not repeat a dictionary gloss from memory.
This is where crossword form earns its keep. A crossword answer has length, letters, intersections, and a sentence. Constraint comes from multiple directions. Guessing is possible; stable recall is faster.
The grid should never quiz you cold
Cold quizzing wastes time. If you have never seen the word, failure proves only that the app selected a word you do not know.
Lexi introduces every unseen word before it can appear in a puzzle. Word cards are styled like collectible game cards: an image designed to evoke the meaning, a definition, an example sentence matching the image, and connotation tags. The first roughly 1,500 words have illustrated word cards.
The dataset contains 20,000 words ordered roughly by rarity for maximal real-world utility per unit time spent. About 3,000 words currently have pipeline-grade clues.
Show the word first. Attach meaning. Then retrieve it later.
Repeated clue variation prevents phrasing overfit
If a word always appears with the same clue, the clue becomes a password prompt. You learn the pairing, not the word.
Lexi currently cycles each word through three differently phrased clues under one spaced-repetition schedule per word. Those are different surface forms of a single underlying item. The schedule tracks the word; the clue phrasing changes.
That detail matters more than it sounds. Multiple clues make prompt memorization harder. One schedule keeps the learning signal coherent. If you are slow on one phrasing and fast on another, the system is still training the same word, not three disconnected flashcards.
For the deeper design rationale, see A Spaced Repetition Vocabulary App Without Flashcards.
Generated puzzles only work if the clues are policed hard
Dynamic generation creates a quality problem. A bad clue can teach the wrong nuance, cue the wrong word, or become impossible without crossing letters.
I built Lexi’s clue pipeline as Creator → Validator → Guesser → Reviser. Frontier language models produce and test clues through that loop; a clue can undergo up to seven revisions before acceptance, against quality criteria refined through repeated in-app revision rounds.
That is the point of generation here: not infinite low-effort content, but content under a revision loop. Definitions, clues, and images are continuously revised from in-app user feedback through regular updates.
Timing should grade the review without making you grade yourself
Manual flashcard grading adds a second task: after recalling, you must decide whether that recall was Again, Hard, Good, or Easy. That decision is noisy, and it interrupts the thing you came to do.
Lexi infers SRS grading automatically from solve timing: fast, normal, or slow, normalized for clue and word length. The puzzle already contains the measurement. A hesitant solve is not the same signal as an instant solve.
I designed it this way because I wanted the review loop to disappear under the puzzle surface. I have personally learned over 1600 words with Lexi. I was surprised to then encounter these words countless times in daily life, especially in books but also just in ordinary life! You really don't realize how many words you just filter out that you don't know until you learn these words and then it makes sense when you hear them. For example, in Curb Your Enthusiasm, Larry David used "extricate" to describe the action of freeing himself from an emotional, bereaved family. And then in books the gains are just ridiculous. I started reading The Brothers Karamazov and each page contains at least one of these words, and within the first 20 pages I've seen a single sentence even use two of them that I hadn't known prior to Lexi.
Lexi is narrow by design: dynamically generated vocabulary crosswords with contextual clues, varied retrieval, and one spaced-repetition schedule underneath. That is a different tool from crossword entertainment with hard words sprinkled in, and a different tool from flashcards that ask you to memorize a definition.
Skip Lexi if you want Android, another language, or a system that trains multiple senses per word today; Lexi is iPhone/iPad only, English only, and currently one sense per word.
Try Lexi on iPhone or iPad. It is free, ad-free, and built around dynamically generated vocabulary crossword puzzles with spaced repetition under the grid: Download Lexi on the App Store.