How Orablo is made

Each set of ten words goes through three steps before it reaches you.

Choosing the words

We start with a longer list of candidates for the day or situation. Each candidate is checked against dictionary data to confirm it’s a real, everyday word, what part of speech it is and which senses it has. Candidates that fail these checks are dropped before any writing starts.

Pronunciation

Every transcription comes from a published pronunciation dictionary: the CMU Pronouncing Dictionary for English, and Wiktionary and the Montreal Forced Aligner dictionaries for Spanish, Portuguese, Italian, French and German. Spanish is also transcribed by its spelling rules, with the dictionaries as a check.

When two sources disagree, we use one of their transcriptions and never write our own.

Writing and checking

Definitions, pronunciation notes and example sentences are drafted by large language models (LLMs) from the senses and pronunciations found in the first two steps.

Each one is then checked: definitions against the dictionary senses, notes against the pronunciation data, and sentences against the scorer the app uses when you practise. Anything that fails is rewritten once and checked again. If it still fails, it isn’t published.

Sources

The dictionaries and tools we rely on, with their licences, are listed in Credits.