What people ask before using the rhyme search

What counts as a rhyme here?

A rhyme is an overlap in sound at the end of two words, not an overlap in spelling. English is full of pairs that look alike and sound nothing alike — 'tough' and 'though' share four letters and rhyme with neither each other nor anything you would expect.

So every word is converted to its pronunciation first, and the match is made on that. This is why results sometimes look nothing like the word you typed.

How does the search work under the hood?

Four stages. The word is converted to a pronunciation first: Korean runs through phonological rules for liaison, nasal assimilation and h-deletion; English comes from the CMU Pronouncing Dictionary with stress marks intact; Japanese resolves kanji to readings and splits them into morae.

Then the ending is extracted as several keys at once — the last unit, the last two and the last three, stored separately so that a one-syllable rhyme and a three-syllable rhyme can be told apart. A vowels-only key and a consonants-only key are built alongside them, which is what assonance and consonance searches use.

Third, candidates sharing a key are pulled from the 4.2 million entries. That step hits a prepared index rather than scanning the table, which is why a search normally returns in under 0.3 seconds.

Finally each candidate is scored and ordered: how much of the pronunciation overlaps, whether the syllable count is comparable, and how common the word actually is. Cross-language candidates come from running loanword sound shifts in reverse.

• 1. Spelling to pronunciation, with per-language phonological rules
• 2. Endings split into one-, two- and three-unit keys plus vowel and consonant keys
• 3. Indexed lookup of candidates sharing a key
• 4. Scoring by phonetic overlap and word frequency

How do perfect and near rhymes differ?

A perfect rhyme matches everything from the last stressed vowel onward: 'baby' and 'maybe'. It lands hardest, which is why it belongs at the end of a hook.

A near rhyme shares most of that ending but not all of it. Writers lean on near rhymes far more than beginners expect, because a verse built entirely from perfect rhymes starts to sound like a nursery rhyme.

• Perfect: the line you want to land
• Near: carrying the flow mid-verse
• Assonance and consonance: when the rhyme should stay hidden

Why does stress matter so much in English?

English rhyme is measured from the last stressed vowel, not from the last letter. 'Comply' and 'supply' rhyme; 'comply' and 'simply' do not, because the stress falls in a different place and the ear hears that.

The pronunciations come from the CMU Pronouncing Dictionary, stress marks included, so candidates that only look similar are filtered out.

Why do Korean results depend on the final consonant?

Korean syllables end in a final consonant called batchim, and it carries the rhyme. Two words with the same vowel but different finals feel unrelated to a Korean ear.

Korean spelling also diverges from speech: assimilation, liaison and h-deletion all change the sound before it reaches the listener. The engine applies those rules first and matches on the result.

Why is Japanese counted in morae?

Japanese rhythm is measured in morae, not syllables. The small tsu, the final n and a long vowel each take a full beat, so a word that looks short on the page can take three beats to sing.

Since that count becomes the length of your line, results show morae rather than syllables. Kanji are resolved to their readings first.

How does cross-language rhyming work?

Loanwords shift predictably as they cross languages: English 'baby' becomes 베이비 in Korean and ベイビー in Japanese. Running those rules in reverse lets words from different languages be matched on sound.

Mixing English into Korean or Japanese lyrics is common, so the search checks whether a rhyme survives the switch.

What decides the order of results?

Candidates are scored on how closely the pronunciation matches, then sorted so that words people actually use come first. A flawless rhyme nobody recognises is useless in a lyric.

The style selector shifts that balance: 'practical' favours familiar vocabulary, 'discovery' surfaces rarer options.

What if my word is not in the dictionary?

The word lists come from open dictionaries, so new coinages and proper nouns may be missing. Searching a word that sounds similar will usually surface the same ending group.

The feedback buttons under each result feed back into the ranking.

Are my lyrics stored?

No. Text entered into the multi-line analyser stays in your browser and is never written to the server database, and the same is true of saved candidates.

Search terms are counted in aggregate to see which words are popular. No name, email address or raw IP is kept.

Where does the data come from?

English from the CMU Pronouncing Dictionary and OpenEPD, Korean from Open Korean Text, Japanese from SudachiDict — all openly licensed, with the licences kept alongside the data.

Frequency scores are calibrated with wordfreq.