English irregular verbs: how many do you need?

You need about 45 irregular verbs. In 2.1 million words of English novels, 10 verbs account for 74 % of all irregular past forms, 30 verbs for 90 % and 45 verbs for 95 %; the remaining 150 or so verbs on a typical long list share the last few per cent between them.

That is a measurement, not a rule of thumb. Below is how we counted, the ranking it produced, and what it means for the order in which you learn them.

How many irregular verbs does English have?

It depends on who is counting. Reference lists usually run from about 150 to more than 200 verbs, because they include archaic forms such as smite, smote, smitten and prefixed compounds such as undertake, undertook, undertaken.

Our reference table holds 213 irregular verbs with their past simple and past participle. Where a verb also has a regular form, such as learned beside learnt, only the irregular form counts as irregular. The question is not how long the list is, but how much of it you meet.

How did we measure which ones matter?

We built an English corpus of 21 novels from Project Gutenberg, originally written in English between 1813 and 1925: Austen, the Brontës, Dickens, Twain, Conan Doyle, Stoker, Joyce and Fitzgerald among them. That is 2,138,259 words.

Every word was labelled with an automatic part-of-speech tagger from the NLTK library. We kept only the words tagged as past simple or past participle. That step matters: it counts put and read when they are past forms, and ignores saw the tool or left the direction.

The tagger found 201,391 past forms. Of those, 121,775 (60.5 %) were irregular. English has far more regular verbs, but the irregular ones are the verbs people use most, so they dominate real text.

How many verbs cover 80, 90 and 95 per cent?

Here is the cumulative coverage, with and without be, have and do. Those three are usually learned first, as auxiliaries, and on their own they account for more than half of all irregular forms:

The curve flattens fast. The first 10 verbs cover 73.8 %; the first 25 cover 88.1 %; the first 50 cover 95.9 %. After that, each new verb adds less than a fifth of a per cent.

At the far end, 54 of the verbs in our table appeared fewer than ten times in 2.1 million words, and 23 never appeared at all, including forgo, oversleep and slink.

Which 25 irregular verbs should you learn first?

Leaving be, have and do aside, these are the 25 most frequent irregular verbs in the corpus, with the share of irregular forms they cover together:

Said alone is 15.4 % of everything outside the three auxiliaries, because every line of dialogue in a story needs it: “Come in,” she said. Add be, have and do back in, and these 28 verbs cover almost nine irregular forms in ten.

Does the ranking hold outside Victorian novels?

Mostly, and we tested it two ways, because a corpus of novels over-represents storytelling in the past tense.

Book by book. The number of verbs needed for 90 % coverage (without be, have and do) ranged from 25 in Alice’s Adventures in Wonderland to 51 in The Call of the Wild. Of the top 25 verbs, 24 appear in all 21 books.

On modern web English. We reran the ranking on Peter Norvig’s word frequency list, built from the Google Web Trillion Word Corpus. That list has no grammatical labels, so we could only count forms with no common homograph, such as went or taken. On those forms, 19 of the top 25 verbs are the same in both corpora, and 7 of the top 10. The web promotes write, show and build (written, shown, built), which suit instructions and product descriptions, and demotes stand and fall, which suit narrative.

The conclusion survives both tests: the core is a few dozen verbs, and it is roughly the same few dozen whatever you read.

Which forms deserve extra attention?

The past participle, for a handful of verbs. For most verbs our novels use the past simple more, but lose appears as the participle lost more than twice as often as the past simple (324 against 155), and show is only irregular in its participle, shown.

Participles carry the present perfect and the passive: I’ve lost my phone, Nothing has been decided, The house was built in 1920. Learning go, went, gone as a set of three, rather than went alone, is what lets you use the present perfect correctly when you reach the A2 level.

A third group hides in plain sight: verbs whose three forms are identical. Put, put, put appeared 898 times as a past form in our corpus, set 476 times, cut 201 times. Nothing in the spelling tells you the sentence is in the past; only the context does, as in Yesterday she put the letter on the table. With read, the spelling stays the same but the pronunciation changes, from “reed” in the present to “red” in the past.

These verbs are easy to memorise and hard to notice, which is why they are worth practising in full sentences rather than in a table.

Two more traps: lay is the past of lie (to recline), and found is the past of find, not a separate verb, in almost every sentence you will read.

Where does Leo fit with irregular verbs?

Leo’s curriculum introduces irregular verbs in two waves, and we measured both against the corpus. The A1 level covers be, do and eight verbs (go, see, have, say, come, take, get, make): those ten are exactly the ten most frequent, and cover 73.8 % of irregular past forms. The A2 second wave adds 18 more, bringing the total to 28 verbs and 87.2 %.

The measurement also shows what those lists leave out: begin, stand, fall, speak, become and hold are in the top 25 but not in the first two waves. In Leo, they arrive the way the rest of the long tail should, inside graded stories where the context makes them guessable. The app is not published yet; the waitlist is below, and the full progression is on the English levels page.

What is the bottom line on irregular verbs?

Learn about 45 irregular verbs well, by frequency, as sets of three forms. That covers 95 % of the irregular past forms in our corpus. The remaining verbs are better met in reading than memorised from a list.

The counts come from our own script over 21 Project Gutenberg novels (2,138,259 words), tagged automatically, with a cross-check on the Norvig web frequency list. Automatic tagging makes some mistakes, so treat the decimals as measurements of this corpus, and the shape of the curve as the finding.