How often will I meet this word again?

Why not just show “per million”?

Per million is a good lab number. It says: if you read or hear one million words from this source, how often does this meaning appear?

But learners do not live inside one million words. They live inside days. TV in the evening. Messages on a phone. News. Books. A conversation with a colleague.

So Glite keeps the lab number, then translates it into a more human one: estimated encounters per year.

Where do the counts come from?

The counts come from large corpora of English examples, split into different source types: TV and movies, spoken English, fiction, magazines, news, academic writing, web pages, and blogs.

For each dictionary sense, the pipeline reads three facts from each source row: how many times the sense appeared, how many total words that source had, and the resulting frequency per million words.

No magic. Just counts, source by source.

Why count meanings, not headwords?

Frequency belongs to a meaning. A headword can be common while one of its meanings is rare.

If Glite gave one frequency number to the whole word, the common meaning would hide the rare one. That would make some senses look more useful than they really are.

• Mean as "to intend" is everyday English. Mean as "the average" is much rarer.
• Bank as in a bank account appears far more than bank as the edge of a river.
• Husband as "a married man" is common. Husband as "to save carefully" is not.

So the dictionary asks a narrower question: if I learn this exact meaning, how often will English give me another chance to meet it?

How does one number come out?

A dictionary page needs one small badge. The pipeline still keeps several numbers behind it.

• Combined frequency adds all matched source counts and all source word totals.
• Mean frequency averages the per-million values across source types.
• Median frequency takes the middle source value, which makes a single loud source less dominant.
• UI frequency weights sources by a rough model of daily American media use.

The badge uses that final UI estimate. The page then turns estimated encounters per day into estimated encounters per year.

What does “media use” mean here?

If a source is technically huge but most people barely touch it, it should not dominate the learner number. A person does not spend the same number of minutes each day reading academic papers, watching TV, scrolling social posts, and talking.

So the pipeline uses a simple daily media model. TV and video-like time map to TV and movie sources. Radio, podcasts, and conversation map to spoken English. Books and fiction map to fiction. News, magazines, web pages, and blog-style text map to their matching source types.

Then the pipeline asks a practical question: with this mix of sources, how many times per day might an average American adult meet this meaning?

Why does the tooltip show four rows?

The raw data has more source types than the page can comfortably show. So the website groups them into four learner-friendly rows.

• TV includes TV and movie sources.
• Reading includes fiction, magazines, academic writing, and news.
• Social Media groups web and blog-style sources under one short label.
• Talking uses spoken English.

The source rows help answer a second question: not only “is this common?” but “where is this common?”

Is the number exact?

No. It is an estimate. A useful one, but still an estimate.

Your year is not the same as another person's year. A lawyer sees plaintiff more than a chef does. A teenager may hear dude more than a retired judge does. Language is not a train timetable.

But an estimate can still help you choose. Common and easy? Learn it early. Common and hard? Pay attention. Rare and hard? Maybe save it for later.