Comprehensible input, what Krashen claimed and what the numbers say

What comprehensible input means in Krashen's theory

Comprehensible input is language you can understand, heard or read, that sits a little above what you could produce yourself. The idea comes from Stephen Krashen, an American linguist who set it out in his 1982 book Principles and Practice in Second Language Acquisition. It has shaped language teaching ever since.

Krashen split language skill into two parts. Learning, in his terms, is conscious knowledge about a language, the rules you can recite and check yourself against. Acquisition is the unconscious ability that lets you understand and speak without thinking about the rules, and he argued that acquisition is what actually drives fluency.

So how does acquisition happen? His answer was the input hypothesis. If your current level is i, you reach the next stage, which he wrote as i+1, by understanding messages that contain it. The condition is that your attention stays on the meaning rather than the form. You understand the new parts thanks to context, your knowledge of the world and everything around the words that you already know.

Two details are easy to miss. Krashen called comprehensible input a necessary but not a sufficient condition, which means he never claimed it was the whole story. He also argued that speaking cannot be taught directly and instead emerges over time, once a learner has understood enough, and that early speech is usually inaccurate.

This flipped the usual order of a language class. Teachers had long taught structures first and then asked students to use them. Krashen proposed the opposite, going for the meaning first and picking up the structures along the way. Much of today's comprehensible input language learning, from graded readers to story-based classes, follows from that reversal.

Where critics push back

Krashen's ideas became popular with teachers faster than they won over researchers, and the criticism is worth knowing because it changes how you should use them. The most common objection is that the theory is hard to test. Barry McLaughlin argued in 1987 that nobody can measure a learner's level i, let alone define exactly what counts as i+1. As a result the hypothesis makes few predictions that an experiment could prove wrong.

The second objection comes from Canada. Merrill Swain studied children in French immersion programmes, who had spent years in classes taught through French and so had received a great deal of comprehensible input. They understood French very well, yet their speaking and writing still differed noticeably from those of native-speaking peers.

Swain concluded in 1985 that input alone had not been enough. The students rarely had to produce French precisely, because their teachers understood them anyway. Swain argued that being pushed to speak and write accurately makes learners notice gaps that listening lets them skip. Her idea became known as the output hypothesis, and it is usually read as a complement to Krashen's idea rather than a replacement.

There is also a practical gap in the original theory. It says that input should be understood, but it does not say how much of it you must understand for acquisition to work. That question turned out to have a fairly precise answer, and it came from reading research rather than from Krashen himself.

None of this explains away why the idea stuck. It matches the experience of many people who picked up English from films, games and forums long before they could name a tense. It also gives teachers a clear rule for their classes. The disagreement is less about whether understanding matters and more about whether understanding is enough.

So the fair summary is narrower than the slogans. Understanding plenty of real language is clearly central, and nobody serious argues for learning a language without it. What the evidence does not support is the stronger claim that input is all you need, or that grammar teaching and speaking practice are wasted time.

How much of a text you need to know

Hu and Nation (2000) tested what comprehensible means in numbers. They gave learners the same fiction text in versions where 80%, 90%, 95% or 100% of the running words were familiar. The rest were replaced with words the readers could not know, and comprehension was then checked in two ways.

At 80% none of the learners understood the text adequately on either measure. At 90% and 95% some of them did, although most still did not. From the pattern in their data the authors concluded that reading fiction without help needs about 98% of the words known, or one unknown word in fifty.

Share of known words in the text and adequate comprehension. Hu & Nation, 2000.

That threshold says something unwelcome about i+1. The plus one has to be very small, since a page where one word in five is new is no longer input you can understand, however good the context. It becomes a puzzle you solve with a dictionary, and solving puzzles is the conscious learning that Krashen set apart from acquisition.

The threshold also shows why ordinary texts are out of reach for beginners. Davies (2005) found that the 1,000 most frequent Spanish lemmas cover 76% of non-fiction, 79.6% of fiction and 87.8% of speech. A learner who knows exactly those words is therefore below the 80% line on a typical article, where Hu and Nation's readers all failed.

Even 3,000 words only bring a reader to 88.2% of non-fiction and 94% of speech. Nation (2006) estimated that 98% of an English text takes 8,000 to 9,000 word families, so a native-level newspaper stays out of comprehensible range for a long time. Most of the work of finding comprehensible input is really the work of finding or making texts that match a small vocabulary.

Finding Spanish comprehensible input at your level

For Spanish comprehensible input the useful question is never whether a source is authentic or easy in general. It is whether you know about 49 of every 50 words in it, and that depends entirely on your level.

A quick test works for any page. Take a passage of about 250 words, which is roughly a page of a paperback, and count the words you do not know. Five or fewer puts you near 98%, so you can read it for meaning. Twenty or more puts you at about 92% or below, where most learners lose the thread and reach for a dictionary.

At the beginning very little real Spanish passes that test, which is why graded readers, children's books and slow podcasts for learners exist. They are written with a small vocabulary on purpose, and the best of them tell you which level they target. Videos with a strong visual story also help, because the images supply the context that makes new words guessable.

Levels matter because the order of words matters. In NativeLingo's Spanish dictionary ventana, window, is an A2 word at rank 1,140, while risa, laughter, is a B2 word at rank 4,014. For an A2 learner a page with ventana is comprehensible input, whereas a page full of words like risa is mostly noise.

Listening has an advantage here that surprises many learners. Speech repeats a small set of words much more than writing does, so the same 1,000 lemmas cover 87.8% of spoken Spanish against 76% of non-fiction. A slow conversation or a podcast for learners therefore becomes comprehensible earlier than a newspaper, even though listening feels harder at first because you cannot pause on a word.

As your vocabulary grows, the range of comprehensible material grows with it, slowly at first and then faster. Around the 3,000 mark everyday speech gets much easier to follow. From there native podcasts and series start to work, especially on topics you already know well in your own language.

How NativeLingo keeps your reading comprehensible

NativeLingo approaches the problem from the other side. Instead of looking for Spanish texts simple enough for your level, it takes the English pages you already read and swaps about one word in a hundred into Spanish. The hard limit is two in a hundred on a page and 5% within a single paragraph. Because you understand the rest of the page completely, you stay well above the 98% line.

In Krashen's terms, the swapped word is a tiny i+1 inside input that is otherwise fully comprehensible. The sentence around it does what context is supposed to do, so you can usually guess the meaning before you hover over the word to check it.

The form of that word matters as much as the word itself. Comprehensible input is supposed to show you how the language really works. A swap that puts a dictionary form into a sentence that needs a plural therefore quietly teaches the wrong grammar. NativeLingo matches number, tense and gender to the sentence, and when the right form is missing from its tables it uses the dictionary form rather than inventing one.

Why not swap more? Because the numbers from Hu and Nation work in both directions. A page with one Spanish word in five would sit at the 80% line where nobody understood the story, and you would stop reading for meaning and start decoding. One word in a hundred keeps the new word rare enough to notice and the page easy enough to keep reading.

The swaps also follow your level. After a three-minute test of 12 questions, NativeLingo picks words from your band in its frequency list. The same word then appears at most once a page and twice a day. Spacing the meetings out points in the direction memory research favours, and it keeps any single word from taking over your reading.

There is a clear limit to this approach as well. You read in English, so the Spanish you get is single words and their forms, not full Spanish sentences. That makes it a way to build vocabulary for later input, the words that will lift your Spanish reading over the 98% line. It does not replace listening to and reading Spanish itself.

What input alone will not do

Even inside its strengths, input builds one kind of knowledge faster than another. Pellicer-Sánchez and Schmitt (2010) followed learners through a whole novel and found that after ten or more meetings readers recognised the meaning of 84% of the words. They could recall the meaning of only 55%, so reading had taught them to recognise words well before they could produce them.

Input also works slowly for any single word. Webb (2007) found that clear gains begin at around ten encounters. A word from the B1 or B2 band, however, may turn up only now and then in what you read. Waiting for ten natural meetings with every useful word would take years, so most learners speed up the process with deliberate review.

Getting from recognition to recall takes retrieval, which means pulling the word out of memory rather than seeing it again. Karpicke and Roediger (2008) showed how much this matters. A week after learning, students who kept testing themselves recalled about 80% of the words, while those who stopped being tested and only studied recalled 36% and 33%.

That is why NativeLingo pairs the swaps with a short daily review in which you type the Spanish word, and FSRS schedules the next review. The swaps stay free, while the reviews come with a 7-day free trial and then need Pro.

Speaking is the other gap, and here Swain's results still apply. Hearing and reading a lot will make speech easier. Still, it is producing Spanish and being understood, or not, that shows you which forms you have not mastered yet. A tutor, a class or a patient friend covers that part, and none of them is replaced by an extension.

These limits do not make input any less important. They mean that hours spent understanding Spanish work best when something else turns what you understood into what you can say. Usually that something is active recall and real conversation.

Put together, the practical version of Krashen's idea is fairly modest. Spend a lot of time with Spanish you can almost fully understand, and keep the unknown share near one word in fifty. Review the new words actively, and start speaking as soon as you have something to say.

Questions