How we make Palabra's word content

Palabra helps you remember Spanish and Greek words through sound-alikes: each word gets an English word that sounds like it plus a short scene linking the sound to the meaning. This page explains which parts come from dictionaries, which are written by our language model (AI), and how we check them.

From dictionaries and open data

• Verb forms and conjugation charts — Wiktionary (via kaikki.org) and UniMorph, CC BY-SA. A chart goes live only if the sources agree form for form.
• Noun gender and plural, IPA transcription — Wiktionary.
• Word frequency — subtitle frequency lists.

What the model writes

• sound-alikes and scenes, word webs, example sentences and translations in the shared word base;
• notes on common mistakes and the examples on the conjugation pages;
• first drafts of the blog posts.

How we check it

• New and rewritten associations go through automatic checks — the keyword is a real word that actually sounds like the foreign word, with no brand names and nothing crude or inappropriate — and then a second, independent review pass. The older base is re-checked against the same rules every night, a batch at a time, and weak ones are rewritten.
• Conjugation pages: every form used in the examples is matched against the chart, and the notes are proofread separately from writing.
• The Anki decks contain only cards that were read one by one: translation, sound-alike, scene and example.
• Blog posts are checked against the sources listed at the end of each post.

The base holds more than ten thousand associations, and not every one has been reviewed individually. If a sound-alike does not work, a scene feels off or a translation is wrong, tell us through the support form and we will fix it.

Images

There are no AI-generated images on the site: page previews are drawn by software from the page text.

Last updated October 2, 2026.