The method

01 · The on-ramp

The part everyone skips

The comprehensible input hypothesis is straightforward: you acquire a language by understanding messages in it, slightly beyond your current level.¹ It has been argued about for forty years, and the version that survived the argument is narrower than the original. Understanding a lot of Spanish is what builds comprehension, but it does not, on its own, teach you to speak.⁸ That narrower claim is the one this method rests on, and it is well supported. It is also enough. §06 deals with what gets left out.

There is a gap in how it gets applied. Input has to be comprehensible. At the very beginning, almost nothing is. A learner with no Spanish opening a Spanish article is not receiving input at all. They are looking at noise. The hypothesis describes what happens once understanding is possible. It does not tell you how to make the first thousand sentences understandable.

Tools built around comprehensible input tend to assume the problem away. They hand you native material and a dictionary, and the first month is spent clicking every third word. That is not acquisition. It is translation with extra steps, and it is why most people stop.

The on-ramp is a separate problem with a separate solution, and it is finite in a way the rest of the process is not.

02 · Core vocabulary

The deck, and why it ends

Word frequency in natural language is extremely uneven. In Spanish, the 2,000 most common word families account for roughly 80 percent of the words in ordinary written text — journalism, fiction, conversation.² The next 2,000 add something closer to five percent. The curve flattens fast.

This is the one place where deliberate memorisation earns its keep. Two thousand items is a knowable quantity. At twenty new cards a day it is about fourteen weeks. There is no version of it that feels like reading, and nothing about it is enjoyable. But at the end, most words in most sentences are already yours.

Then there is a gap. Eighty percent coverage is not enough to read native material, but it is enough to read something written for it. Lemmesa's seventy graded stories exist to cover that stretch — they start at the edge of the deck and end where ordinary Spanish begins. This is the shortest stage and the one most tools leave out, which is why so many learners finish a deck and find they still cannot read anything.

What matters is that the drilling terminates. The deck is an on-ramp, not a practice. If a tool has you adding cards indefinitely, it has confused the on-ramp with the road.

Eighty percent coverage is not comprehension. It is the point at which reading starts to teach you more per hour than drilling does.

03 · Counting

What counts as a word

Hablar, hablo, hablé, hablando, hablaría. Five strings. One thing the brain has learned. Once the verb and the conjugation system are in place, the inflected forms are not separate acquisitions — they are the same knowledge applied.³

Spanish inflects heavily, so the two ways of counting diverge sharply. A single common verb can surface as forty or more distinct forms across the indicative, subjunctive and imperative. Count surface forms and a learner appears to know 9,000 words after a year. Count families and the same learner knows 2,600.

The second number is the useful one, because it predicts coverage. It is also the smaller and less flattering one, which is presumably why it is rarer.

Lemmesa counts families. Progress figures look lower here than elsewhere; they refer to something different.

04 · Difficulty

i+1, made literal

The i+1 formulation describes input one step past your level. It is usually left as a metaphor, but it can be computed. Given a model of which word families a reader knows, any text has a known-word coverage figure, and that figure predicts whether the text will be comprehensible.

Below about 95 percent, comprehension falls off sharply rather than gradually: unknown words start appearing near each other, and there is not enough surrounding context left to infer from.⁴ Above about 98 percent the text is comfortable and teaches almost nothing new — that is the coverage figure associated with reading for pleasure without support.⁵

Lemmesa targets 92 to 96 percent for reading, slightly below the classic threshold on purpose: the reader has a gloss one tap away, so a little more friction is affordable and productive. In practice that is about one unknown word every two lines — frequent enough that you are meeting new language continuously, sparse enough that each one sits in a sentence you otherwise understand. Words acquired that way are learned in context, with a situation attached, which is a materially different kind of knowing than a card produces.⁶

Audio is scored the same way and targeted higher. The known-word model runs over a podcast transcript exactly as it runs over an article, but the same coverage is harder to use in speech: you cannot re-read a sentence that has already gone past, word boundaries are not marked, and connected speech blurs the endings that carry most of Spanish's grammar. A transcript at 92 percent reads comfortably and listens badly. Lemmesa aims listening at 95 to 98.

Every text and every episode is scored against your model, continuously. The bands stay fixed; what qualifies for them moves as you do.

05 · Scheduling

Why your deck stops growing

Review scheduling is a solved problem in the narrow sense: FSRS models the probability that you will recall a given item on a given day and schedules the review at the point where recall is about to fail.⁷ It is better than the older interval heuristics, and Lemmesa uses it.

The more consequential rule sits on top of it. A card exists to keep a word alive until real input can do that job. Once a word has been met enough times in real material — understood, not looked up — the encounter itself is the review, and the card is redundant. Lemmesa retires it.

An encounter is an encounter regardless of which sense it arrived through. A word met in a podcast counts the same as a word met in an article, because both are the same family in the same record. This is the practical reason the two currents share one vocabulary model rather than two: an hour of listening maintains the words you read last week.

So two curves separate. Vocabulary keeps climbing, mostly through input. The drill pile rises during the on-ramp, peaks, and then settles somewhere around 200 to 300 active cards and stays there — new words entering at roughly the rate that mature ones leave.

The daily review load therefore does not scale with how much Spanish you know. That is the intended behaviour, not an optimisation.

06 · Limits

The part input doesn't do for you

Input builds comprehension. It does not, by itself, build fluent speech. A learner with several hundred hours of reading and listening can follow a podcast and still stall badly in a conversation, and this is the normal outcome rather than a sign that something went wrong.

Production is a motor and retrieval skill. Recognising a word in context is easier than producing it under time pressure, and the two are trained separately. But the gap narrows sharply with volume. At two hundred hours, learning to speak is a long grind. At fifteen hundred, learners commonly report it arriving in weeks — and arriving better, with phrasing and an accent that people who started talking early spend years unlearning.⁸ The output work still has to happen. It just arrives faster, and hurts less, when the understanding is already there. The evidence here is observational rather than experimental, like the hour figures above.

One further limit worth stating plainly. Explicit grammar study is not the engine of acquisition, but it is a reasonable shortcut for a handful of high-frequency structures — the subjunctive being the obvious case. Refusing to look anything up is a purity position, not a method.

Lemmesa is a comprehension tool. It takes you to the point where you understand the language, which is the long part, and the part that makes speaking cheap when you get to it. The speaking itself is still yours to do.

07 · Measurement

How to tell if it's working

Both currents measure the same thing: how much comprehended Spanish went in. They measure it differently because the two are paced differently.

Reading is self-paced. An hour at the page could be four hundred words or four thousand, depending on the reader, the text, and how much of the hour went on lookups — and a learner's own reading speed changes by an order of magnitude across the first year. Time in the chair therefore says almost nothing about how much language was met, which is why Lemmesa counts words read.

Listening is paced by the speaker. An hour of Spanish at conversational speed is seven to nine thousand words whether you followed all of it or drifted, so the clock is already a volume measure. Lemmesa counts hours listened.

Different units, one quantity. Neither is a proxy for the other, and neither is a proxy for effort.

Below is the shape of the whole road. The figures are targets, not thresholds, so nothing unlocks and the levels are descriptive.

Reading figures follow Nation's input estimates for frequency-band coverage.⁹ Hour figures are listening-derived and come from a community roadmap and aggregated learner reports.¹⁰ Both are descriptive rather than experimental, and individual variation is wide.

The columns are not interchangeable. Read 1.5 million words and nothing else and you have B2 reading, but you will still stall on a podcast. Listen nine hundred hours and nothing else and you follow the radio easily, then find a novel slow going. Neither is a failure. Skip a column and you skip the skill.

Notice the asymmetry. A million and a half words is roughly two hundred hours at the page for a reader at that level, while the listening column asks for nine hundred. Reading is the cheaper way to buy vocabulary. It is denser, self-paced, and the spelling preserves the endings connected speech swallows. Listening costs more because it is building something else: the real-time parsing that lets you use the vocabulary when it arrives at speed and does not wait. There is no researched optimal ratio, and anyone quoting one is guessing. Weight the mix toward the skill you actually want.

At an hour a day of mixed reading and listening, B2 is about two years. That is the honest timescale, and it is roughly the same whatever tool you use.

08 · The product

Where Lemmesa fits

Lemmesa runs the on-ramp and then gets out of the way. It builds the 2,000-family deck, walks you across the seventy graded stories, retires cards as real input takes over, and keeps a library of texts and episodes scored to sit in your band, including anything you bring in yourself.

None of the above is proprietary. It is the published research applied consistently, which is a lower bar than it sounds and one most tools do not clear.

Here is what that buys you. In two years, at an hour a night, you finish a novel in Spanish and realise you forgot it was in Spanish. That is the whole promise. No streaks, no levels, no one congratulating you for showing up. Just the thing you actually wanted, on a timeline someone finally told you the truth about.

Sources

1. Krashen, S. (1982). Principles and Practice in Second Language Acquisition. Pergamon.
2. Davies, M. (2006). A Frequency Dictionary of Spanish. Routledge — coverage figures by frequency band.
3. Bauer, L. & Nation, I.S.P. (1993). Word families. International Journal of Lexicography, 6(4).
4. Laufer, B. (1989). What percentage of text-lexis is essential for comprehension? In Special Language: From Humans Thinking to Thinking Machines.
5. Hu, M. & Nation, I.S.P. (2000). Unknown vocabulary density and reading comprehension. Reading in a Foreign Language, 13(1).
6. Webb, S. (2008). The effects of context on incidental vocabulary learning. Reading in a Foreign Language, 20(2).
7. Ye, J. et al. (2022). Free Spaced Repetition Scheduler — a DSR-based memory model. Open-source implementation and evaluation.
8. Swain, M. (1985). Communicative competence: the role of comprehensible output. In Input in Second Language Acquisition.
9. Nation, I.S.P. (2014). How much input do you need to learn the most frequent 9,000 words? Reading in a Foreign Language, 26(2) — words-read figures.
10. Dreaming Spanish level roadmap and aggregated learner hour reports (2020–2024). Observational, listening-derived.