You acquire a language by understanding a lot of it, repeatedly, over a long time. Grammar rules, drills, and conjugation tables are support staff. They are not the thing that turns a language into something you can follow without effort.

The usual objection is that only children get to learn this way. That is not quite the problem. The problem is supply. A child hears thousands of hours of speech deliberately tuned to what they can follow. An adult gets a textbook that is too easy and a Netflix show that is too fast, with almost nothing in between. The bottleneck was never adult brains. It was the shortage of material you can understand and still want to read. That shortage is the part that changed.

Acquisition and study are two different systems

There is a real distinction between knowing a rule and being able to use it in real time. You can recite the rule for Spanish ser versus estar and still hesitate for two seconds in a live conversation. Those are different kinds of knowledge, usually called explicit and implicit, and only the second one produces fluent use.

Forty years of argument about this has not fully settled, and it is worth being honest about that. Lichtman and VanPatten's review, Was Krashen right? Forty years later, concludes that the core claims held up well: implicit acquisition is driven by input you understand. In the same journal issue, Loewen's instructed second language acquisition perspective reaches a more critical verdict, arguing that interaction, corrective feedback, and output do more than Krashen allowed.

The practical reading of both papers is the same: input is the engine, and explicit study is a useful accelerator that cannot replace it. If you spend 90% of your time on rules and drills and 10% on contact with the language, you have the ratio backwards. It is also the main limit of app-based courses, which are good at the rules and drills half and cannot supply the other half.

The supply problem

Extensive reading works. A meta-analysis of 34 studies covering 3,942 learners found a medium overall effect (Nakanishi, 2015, d = 0.46 for group contrasts, 0.71 for pre-post). Nobody seriously disputes that reading a lot in your target language helps.

The catch is the entry fee. To read comfortably without help, learners generally need to know something like 95 to 98% of the words on the page, a threshold that traces back to Hu and Nation in 2000 and that a 2023 replication found less clean than originally reported. Nation later estimated that reaching 98% coverage of ordinary written text takes something like 8,000 to 9,000 word families. At month two of a new language, essentially nothing written for adults sits in that band.

Graded readers were the traditional answer, and they are good. They are also finite, mostly written for English learners, thin on the ground for smaller languages, and rarely about anything you personally care about. You finish the useful ones in a few weeks and then fall off a cliff.

What actually changed

You can now produce text at your level, about your subject, on demand.

That is the whole shift, and it is easy to undersell. In LingoUnify you pick a language, pick a CEFR level from A0 through C2, pick a length, and describe what you want to read about. A0 is a custom Starter level below A1, kept deliberately short, concrete, and repetitive for people in their first weeks. This works across 80 languages and 90 regional varieties.

Two honest caveats. First, this is AI-written text, not authentic native writing, so it should be a bridge to real material rather than a permanent replacement for it. Second, generation quality is not equal across every language on the list, which is why LingoUnify shows a quality rating on languages where the AI is weaker instead of pretending they are all the same.

Look words up. The research does not say guess

There is a strand of immersion advice that says never use a dictionary, always infer from context. It sounds principled. The evidence does not support it.

A meta-analysis of glossing (short definitions supplied next to the text) pooled 359 effect sizes from 42 studies with 3,802 learners. Reading with glosses produced word learning of 45.3% immediately and 33.4% on delayed tests, against 26.6% and 19.8% for reading without them (Yanagisawa, Webb and Uchihara, 2020). Glosses in the learner's own first language beat glosses in the target language.

Reading with lookups available produced roughly 70% more vocabulary learning than reading without them. Guessing from context is a real skill and worth practicing, but as a default policy it is expensive and often wrong.

This is exactly what the pen tool in LingoUnify is for. Tap a word and you get its translation, a definition, its grammar and word forms, its etymology, and sometimes an image, without leaving the page. The lookup also sends the surrounding words with the request, so an ambiguous word gets resolved against the sentence you are actually reading rather than returning a dictionary list of nine possible meanings.

One meeting with a word is not enough

Here is the part that keeps pure reading from being a complete system.

Repetition does drive incidental vocabulary learning, but the relationship is moderate, not overwhelming: a meta-analysis of 26 studies with 1,918 learners put it at r = 0.34 (Uchihara, Webb and Yanagisawa, 2019). Meeting a word once in a text usually does nothing durable. Meeting it eight or ten times across different texts is a different story.

Reading alone will not schedule those meetings for the words you specifically need. That is what spaced repetition is for, and it is why LingoUnify pairs reading with review decks rather than picking a side in the flashcards versus immersion argument. A word you save goes into your word bank and is then highlighted everywhere it appears in later texts, so you can see your own re-encounters happening. Adding it to a deck is a separate, deliberate step, which is the right default: not every saved word deserves a review schedule.

Be selective about what you save. Not every unknown word deserves a card, and every card you create schedules a tail of future reviews.

A worked example

Here is an A1 Spanish text of the kind the generator produces:

Marta llega tarde al trabajo. El autobús no viene. Ella espera en la parada. Hace frío y no tiene abrigo. Un vecino la ve y le ofrece llevarla en su carro. Marta dice gracias y sube. El viaje es corto.

Forty words. Say you know all of them except parada and abrigo. That is 95% coverage, which is the low end of the comfortable band.

The loop takes about four minutes:

  • Read it once without touching anything, for the story, not for study.
  • Tap the two blockers. parada is the bus stop, abrigo is a coat. Two seconds each.
  • Save only what earns it. abrigo is common and concrete, so save it. parada you can leave for now, because you will meet it again in almost any text about getting around.
  • Listen to it once with the text in front of you, so the words attach to a sound and not just a shape.
  • Read it again. It should now feel like reading, not decoding.

Now the important part. If you actually did not know six of those forty words, the text was too hard and no amount of discipline fixes that. Drop a level. A0 with full comprehension beats B1 with a lookup every second line.

Where this goes wrong

Comprehension is not acquisition. Understanding a text once feels like progress and mostly is not. The gain comes from the twentieth text, not the first.

Looking up everything destroys the flow that makes input work. If you are tapping more than roughly one word in twenty, you are studying a text rather than reading one. That is a level problem, not a willpower problem.

Input does not automatically produce speech. Reading builds the model your speaking will draw on, but the first time you try to say something you will still be slow and clumsy. That is normal, and it is not a sign the reading failed. You have to actually speak to get fluent at speaking.

AI text is a bridge, not a destination. Once you can follow B1 material, move toward real content: news, Wikipedia, imported articles, EPUB and PDF books, podcasts, radio. LingoUnify supports that transition (imports, a browse feed, 30,000 plus radio stations, and the browser extension for any page on the web), but plenty of it is available free elsewhere, and you should use whatever gets you the hours.

No tool removes the time requirement. The volume needed is measured in hundreds of hours. Getting enough comprehensible input is a supply problem you have now solved, but it is still an attendance problem you have not.

What to do this week

  1. Pick one level below where your ego wants to be. If A2 feels like work, read A1.
  2. Generate or find one text a day, on something you would read in your own language anyway. Interest is not a luxury here; it is what keeps you coming back at day 40.
  3. Read for meaning first, then look up only what blocked you.
  4. Save two to five words a day, not twenty. Then actually review them.
  5. Listen to every text you read, at least once.

Do that for a month and you will not feel fluent. You will notice something better: words you never deliberately memorized turning up in a text and simply making sense. That is what acquisition feels like from the inside, and it is the only reliable signal that the process is working.