Frequency lists: learn the words that matter

Why do so few words cover so much text?

Word frequencies follow a steep curve known as Zipf's law: the most common word turns up roughly twice as often as the second, three times as often as the third, and so on. In the Brown Corpus of about a million words of American English, the alone accounts for nearly 7% of all the words, and just 135 different words make up half the corpus. Steven Piantadosi's review finds the pattern across languages, while stressing that real distributions are only near-Zipfian and that no theory yet explains all of it.

Paul Nation and Robert Waring set out what the curve means for learners:

The first thousand words do more work than the next fourteen thousand put together. Short stories for learners lean on the top of the list even harder. In the free chapters of the Contos library (October 2026), the most frequent word forms cover this much of the text:

The ten most frequent Spanish forms are all small grammar words:

• lathe (feminine)
• deof, from
• enin, on
• elthe (masculine)
• quethat, which
• yand
• ato, at
• nono, not
• suhis, her, their
• una, an (masculine)

Which frequency list should you use?

A list can only be as good as the texts it was counted from. Marc Brysbaert and Boris New showed that word frequencies taken from film and television subtitles predict how quickly people recognise English words better than counts from written sources, and that the long-used Kučera and Francis count rested on only about a million words. For words rarer than ten per million, they estimate a corpus of at least 16 million words is needed. Subtitle-based lists now exist for other languages, including SUBTLEX-ESP for Spanish, built from 41 million words of film and TV subtitles.

For learners there are also printed frequency dictionaries. Mark Davies's A Frequency Dictionary of Spanish lists the 5,000 most frequent words; its first edition drew on a 20-million-word corpus split between speech, fiction and non-fiction, and the second edition on a web corpus of more than two billion words from 21 Spanish-speaking countries, with an example sentence for each entry.

The texts behind a list matter more than its size. To show this, we built a list from half of the Spanish series in the Contos library and tested it on the other half:

Same writers, same levels, same kind of story, and the 1,000-form list still lost about nine points on texts it hadn't seen, mostly because each story brings its own names and topic words. A general list loses more on texts unlike its sources, which is the case for choosing one built from the kind of language you want to understand.

What counts as one word on a list?

Lists count in different units, and the unit changes the numbers:

Nation's coverage research uses word families for reading, since a reader who knows nation can usually handle national, but he suggests the lemma is the largest sensible unit when you need to produce a word. Brysbaert and New found that, for English at least, lemma frequencies were no better than word-form frequencies at predicting how fast words are recognised. The Contos figures above count word forms, so a list of lemmas would cover more with fewer entries.

Names are another catch. They rarely appear on general lists, yet in fiction they come back constantly: in the free Italian chapters of the Contos library, the name Marco is the seventh most frequent word form. Coverage studies count names as known, because readers pick them up on sight.

How should you learn from a frequency list?

Nation and Waring are direct about the top of the list: learners need the 3,000 or so high-frequency words, they are an immediate priority, and it makes little sense to focus on rarer words until these are known. They also note that there is no research showing that learning from context gives better results than learning from word cards.

Lower down, the return per word shrinks fast. In the Brown Corpus figures, the step from 5,000 to 6,000 words adds 1.2 points of coverage, compared with almost 8 for the step from 1,000 to 2,000. Nation and Waring argue that low-frequency words give too little coverage to be worth teaching one by one; time is better spent on strategies for handling them, such as guessing from context, using word parts and learning with word cards.

• Use a list as a map, not a script. Make sure the top 1,000–3,000 words are covered, but learn them in sentences and stories rather than strictly in order.
• Pair cards with reading. Cards make the first meeting stick; reading brings each word back in new sentences. Our page on spaced repetition explains how to schedule the cards.
• Skip what you already know. A list is most useful for spotting gaps, especially among common words that are easy to misread.
• Switch to reading later on. Past a few thousand words, each new item covers so little text that broad reading at the right coverage is the more efficient way to grow.

Contos follows the same logic from the other end: chapters are written for a level, tapped words become flashcards, and moving up is gated by known words, 500 for A2, 1,500 for B1 and 4,000 for B2.