Here’s how they compare, what they cost in October 2026, and who each one suits.

What makes a good reading app

Reading works for language learning because it gives you lots of input at your own pace. Research on lexical coverage suggests you need to know about 98% of the words in a text to read it comfortably without help (Hu and Nation, 2000, Reading in a Foreign Language), and that comfortable reading across a wide range of texts takes a vocabulary of roughly 8,000 to 9,000 word families (Nation, 2006, Canadian Modern Language Review). A good app helps you close that gap. It should do four things:

1. Make unknown words cheap. A tap should give you the meaning in context, fast enough that you stay in the story.
2. Help you find text at the right level. Either by grading content, or by telling you how hard a text is for you.
3. Remember what you’ve learned. Save words with their sentence and bring them back.
4. Let you read things you care about. Interest keeps you reading, and volume is what counts.

Our post on the 98% rule explains the coverage research in more depth.

The best apps to learn a language by reading, compared

Prices are US prices from each product’s official site at the time of writing (October 2026), and they can change.

LingQ: best all-round reading system

LingQ is the best-known app in this space, for good reason. It has a large library of lessons with audio across 50+ languages, and it imports almost anything: web pages through the LingQ Importer extension (Chrome, Firefox and Safari), ebooks, podcasts, YouTube and Netflix.

Its signature feature is word tracking. Every word on the page has a status: new words are blue, saved words are yellow, and known words are plain. Turn the page and untouched new words become known, so your known-word count rises as you read. It’s motivating, and it shows you at a glance how hard a lesson is.

Downsides: it’s one of the pricier options at $14.99 a month, or $10 a month on the annual plan, and the free plan (20 saved words in total) is really just a demo. The interface has a lot going on.

Best for: learners who want one app for reading and listening, with a library to fall back on and a clear measure of progress. See Readlang vs LingQ for a closer look.

Readlang: best value for reading your own texts

Readlang does the core job, click a word and see its translation, with less clutter and a much lower price. A Chrome extension and bookmarklet work on any web page, so you can read the news without importing it. You can also upload texts and EPUB books, and it can sync YouTube videos with their transcripts. Saved words go into spaced repetition flashcards.

The free plan is genuinely usable: unlimited single-word translations, plus 10 phrase translations and 10 AI context explanations a day. Premium ($6 a month or $48 a year) removes the limits. It supports over 100 languages, more than anything else here.

Downsides: a smaller library, and at the time of writing no LingQ-style known-word count.

Best for: budget-conscious readers, learners of less common languages, and anyone who already knows what they want to read.

Satori Reader: best for intermediate Japanese

Satori Reader has more than 1,650 Japanese episodes across many genres. Every story has native-speaker audio, inline grammar explanations and definitions that fit the context. You can read in kanji, kana, or kanji with furigana, which makes it unusually flexible as your kanji knowledge grows. Notes and SRS flashcards are built in.

Selected episodes are free, and Pro is $9 a month or $89 a year at the time of writing.

Downsides: Japanese only, aimed at intermediate learners, and you can’t import your own texts.

Best for: Japanese learners past the beginner stage who want trustworthy, well-explained material.

Du Chinese: best for graded Mandarin

Du Chinese offers more than 3,000 readings and 80+ graded stories, from Newbie to Master, with native-speaker audio, a pop-up dictionary and grammar explanations. It runs on iOS, Android and the web, and saved words go into its flashcards.

It’s $14.99 a month, $79.99 for six months or $119.99 a year, with some free lessons.

Downsides: Mandarin only, no importing, and the monthly price is high if you don’t commit.

Best for: Mandarin learners who want reading practice pitched at their level, from the very start.

Lute: best free option

Lute (Learning Using Texts) is free and open source under the MIT license. You install it on your own computer, with Python or Docker, and use it in your browser. You paste or import texts, configure dictionaries for your language, and mark words with statuses as you read. It supports multi-word terms, and “parent” terms so one definition covers several forms of a word.

Downsides: you have to install and configure it, there’s no official mobile app, and there’s no built-in content.

Best for: learners who want LingQ-style word tracking without a subscription, and who don’t mind some setup.

Migaku: best for reading subtitles and the web

Migaku isn’t a book reader, but if most of your reading happens in subtitles, it’s excellent. Its Chrome extension makes subtitles clickable on Netflix, YouTube, Disney+, Rakuten Viki and Animelon, as well as text on ordinary web pages. One click makes a flashcard with the sentence, audio and a screenshot. It also estimates how hard a video or page will be from the words you know.

Downsides: 11 languages, no free plan (only a free trial, 10 days at the time of writing), and its FAQ lists full ebook support as “coming soon.”

Best for: video-first learners. Read our full Migaku review.

Language Reactor: best free add-on for video

Language Reactor is a Chrome extension with about 2 million users on the Chrome Web Store. It adds dual subtitles and a pop-up dictionary to Netflix and YouTube, and can import text, books and websites with machine translation and text-to-speech. The free version is generous, and a paid Pro tier adds extras such as generated subtitles, saved-word highlighting and export.

Downsides: desktop only, and reading tools are basic compared with a dedicated reader.

Best for: learners who want help with the shows they already watch, at little or no cost.

A word on Anki

Anki isn’t a reading app, but many readers use it alongside one. It’s free on desktop, the web and Android (AnkiDroid), and AnkiMobile on iOS is a paid app that funds development (apps.ankiweb.net). If you already review in Anki, check that your reader can export cards before you commit.

Lumo: coming soon

We’re building Lumo, so read this with that in mind. It isn’t available yet, but it’s coming soon. It’s a reading and listening app aimed at the “help me find the right level” problem. Every text gets a difficulty label from your known-word coverage: Easy at 98% and up, Just right from 90%, Challenging from 80%, Hard below that. You’ll be able to import pasted text, web articles, EPUBs, YouTube videos with an interactive transcript and dual subtitles, and audio or video with subtitles. There will also be short original graded stories in 8 languages (A1–B1). Saved words become flashcards scheduled with FSRS, and everything is processed on your device, so it works offline.

It won’t have a browser extension, Netflix support, a big community library or tutors at launch. Planned pricing is a free tier and Pro at $5.99 a month or $49.99 a year.

How to choose

Start with your language, then your level.

• Japanese, intermediate: Satori Reader, plus a general reader for native content.
• Mandarin, any level: Du Chinese, then a general reader as you move to native texts.
• A common European language: LingQ if you want a library, Readlang if you have your own material.
• A less common language: Readlang (100+ languages) or Lute.
• Mostly video: Migaku or Language Reactor.
• No budget: Readlang’s free plan or Lute.

Then stick with one app for a month before judging it. The habit matters more than the features.

FAQ

Can you really learn a language by reading?

Reading is one of the best-researched ways to build vocabulary and grammar, as long as the text is mostly comprehensible. In one well-known case study, Pigada and Schmitt (2006, Reading in a Foreign Language) followed one learner of French through a month of reading graded readers and found that knowledge of about two-thirds of the 133 target words improved in some way, most strongly spelling, from reading alone. Most people combine it with listening and, later, speaking.

What is the best free app for learning a language by reading?

Readlang’s free plan is the most usable free option for most people, with unlimited single-word lookups. Lute is completely free and open source, but you install it yourself. Satori Reader and Du Chinese both have free lessons to try.

Is LingQ better than Readlang?

LingQ is better if you want a large library, audio and a known-word count. Readlang is better if you want a cheaper, simpler tool for your own texts, or you’re learning one of the many languages LingQ doesn’t support.

What level do I need to start reading in a new language?

You can start with graded stories almost from day one. Native content gets comfortable once you know most of the words on the page, which for general text means thousands of word families. Our post on graded readers vs native content covers when to make the switch.

LingQ is still a good tool. It’s big, it covers more than 50 languages, and its import system is mature. But it isn’t cheap, the free plan is very small, and some people find it busy. Here’s where to go depending on why you’re leaving, what each option costs in October 2026, and what you give up by switching.

Why people look for LingQ alternatives

When learners say they’re leaving LingQ, it usually comes down to one of four things.

• Price. Premium is $14.99 a month billed monthly, or $10 a month on the annual plan, according to LingQ’s own comparison of its plans (August 2026). Premium Plus, which adds AI voices and AI-simplified lessons, starts at about $22.50 a month on the annual plan.
• The free plan. Free accounts get 20 saved words or phrases in total, 5 imported lessons and 1 playlist. That’s enough to see how the reader works, not enough to use it.
• Clutter. LingQ has grown a lot of features: lessons, playlists, challenges, AI chat, tutors. Some people love that. Others just want a page of text and a dictionary.
• Video. LingQ can import YouTube and Netflix content, but it turns them into lessons. If you’d rather stay on the video with subtitles you can click, a browser extension built for video may fit better.

None of these means LingQ is bad. It means a different tool might suit you better.

What you give up by leaving LingQ

Be honest with yourself about what you’ll miss before you cancel. Most alternatives below are better at one thing and thinner at the rest.

• The known-word count. LingQ’s running total, and the blue page that shows every word you haven’t met yet, is rare. Readlang doesn’t have it, and Lute and Migaku track word statuses in their own ways.
• The library. Few tools come with thousands of lessons with audio in dozens of languages. With most alternatives, you bring your own material.
• One place for everything. LingQ handles reading, audio, imports and review in one app. Switching often means pairing two tools, such as a reader plus Anki.
• Your history. Your saved words and known words don’t move automatically. See the FAQ below for workarounds.

The short list

Prices are US prices in October 2026, taken from each product’s official site where it publishes them. Prices change and some tools run sales, so check before you pay.

Readlang: the closest like-for-like swap

Readlang does the core LingQ job: open a text, click a word, see a translation, and save it as a flashcard. It’s lighter and much cheaper.

What it does well

• The free plan is usable. Single-word translations are unlimited. Phrase translations and AI context explanations are capped at 10 a day each.
• It reads the web. A Chrome extension and a bookmarklet let you translate words on any page, so you don’t have to import first.
• It handles books and video. You can upload your own texts and DRM-free EPUBs, and sync YouTube videos with their transcripts. It doesn’t support Netflix.
• Price. Premium is $6 a month or $48 a year, with unlimited translations. Premium Plus ($15 a month or $120 a year) adds high-quality voices for whole texts and stronger AI features.

Where it falls short

Readlang’s built-in library is smaller than LingQ’s. It can underline words you’re already studying, but at the time of writing it has no LingQ-style known-word count, and it doesn’t mark every untouched word on the page as new. If that color-coded “how much of this do I know?” view is what you love about LingQ, you’ll miss it.

Best for: learners who want cheap, unfussy click-to-translate reading on the web and in their own books. We go deeper in Readlang vs LingQ.

Migaku: for learners who live in video

Migaku is built around native media. Its browser extension adds clickable subtitles to Netflix, YouTube, Disney+, Rakuten Viki and Animelon, and it works on ordinary web pages and on video files you own. One click turns a sentence into a flashcard with the word, the sentence, audio, a screenshot and a definition.

What it does well

• Sentence mining is fast. If you already make cards from shows, this is the smoothest workflow on the list.
• It estimates difficulty. Migaku tracks what you know and estimates how hard a video or page will be.
• There’s a course, too. Migaku Fundamentals covers writing systems and pronunciation, and Migaku Academy covers roughly the first 1,500–1,700 words and a few hundred grammar points, depending on the language.
• Mobile apps cover flashcards, YouTube, web browsing and OCR for printed text.

Where it falls short

Migaku supports 11 languages: Japanese, Mandarin, Cantonese, Korean, Spanish, English, German, Portuguese, Vietnamese, Italian and French. There’s no free plan, only a free trial (10 days at the time of writing), and Migaku offers a full refund within 14 days of purchase. Standard is $10 a month or $96 a year, Early Access is $15 a month or $144 a year, and lifetime is $499. It also asks for more setup than a reader like Readlang.

Best for: anime, drama and YouTube learners, especially in Japanese, Chinese or Korean, who like building their own cards. Our Migaku review and LingQ vs Migaku go further.

Language Reactor: the cheapest way into dual subtitles

Language Reactor is a Chrome extension with about 2 million users on the Chrome Web Store. It adds dual-language subtitles, a pop-up dictionary and fine playback controls to Netflix and YouTube. You can also import text and get a machine translation plus text-to-speech.

It’s the lowest-friction way to make a show you’re already watching more useful, and the free version goes a long way. A paid Pro tier adds extras such as generated subtitles for videos that lack them, saved-word highlighting and export. Language Reactor doesn’t publish Pro pricing on a public page, so check the extension for the current price.

The trade-off: it’s a desktop extension (Chrome, and it also runs in Edge), and it’s a watching tool first. Reading long texts and reviewing vocabulary are thinner here than in LingQ.

Best for: people who watch a lot of Netflix and YouTube and want help without leaving the player.

Satori Reader and Du Chinese: depth in one language

If you’re learning Japanese or Mandarin, a single-language app can beat any general tool, because humans wrote and annotated every line.

Satori Reader has more than 1,650 Japanese episodes with native-speaker audio, inline grammar explanations and word definitions in context. You can read in kanji, kana, or kanji with furigana. Pro is $9 a month or $89 a year, and selected episodes are free. It’s aimed at intermediate learners.

Du Chinese has more than 3,000 readings and 80+ graded stories, from Newbie to Master, with native audio, a pop-up dictionary and grammar notes. It’s $14.99 a month, $79.99 for six months or $119.99 a year, with some lessons free.

The trade-off for both: you read their content, not yours. They don’t import articles or books.

Best for: Japanese or Mandarin learners who want reliable, level-appropriate material rather than wrestling native content too early.

Lute: free and open source

Lute (Learning Using Texts) is a free, MIT-licensed app that you install and run on your own computer. It’s the modern Python rewrite of the older LWT project. You paste or import texts, set up your own dictionaries, and mark words with statuses as you read. It handles multi-word terms and “parent” terms, so the definition for hablar can cover habló.

There’s no subscription, no account and no company that can change the price. The cost is your time: you install it with Python (pip install lute3) or Docker, configure dictionaries per language, and there’s no official mobile app.

Best for: tinkerers, privacy-minded learners, and anyone who wants LingQ-style word tracking for free and doesn’t mind setup.

What about Anki?

Anki isn’t a LingQ alternative on its own. It’s a flashcard app, not a reader. But many people pair a reader with it. The desktop version and AnkiDroid on Android are free, and AnkiMobile on iOS is a one-time purchase that funds development (see apps.ankiweb.net). If you already have a big Anki collection, check which readers can export to it before you switch.

Lumo: coming soon

We’re building Lumo, so weigh this section accordingly. It isn’t available yet. It’s a reading and listening app that keeps LingQ’s word-level idea and adds a difficulty label to every text, based on how many of its words you already know: Easy at 98% and above, Just right from 90%, Challenging from 80%, and Hard below that. You can import pasted text, web articles, YouTube videos with an interactive transcript, EPUBs, and audio or video files with subtitles. Saved words become flashcards scheduled with FSRS, and it works offline.

What it doesn’t have: no browser extension, no Netflix support, and no community or tutors. If those matter to you, the tools above are the better pick today. Planned pricing is a free tier (one language, 5 imports a week) and Pro at $5.99 a month or $49.99 a year.

How to choose

Ask yourself one question: what will you actually spend your hours on?

• Your own articles and books, in any language: Readlang, or Lute if you want free.
• Netflix and YouTube: Language Reactor to start, Migaku if you want serious card-making.
• Japanese or Mandarin, with trustworthy graded material: Satori Reader or Du Chinese.
• A huge built-in library plus imports, and you don’t mind the price: stay with LingQ.

Whatever you choose, the tool matters less than the hours. Most of the benefit comes from reading and listening to lots of material you mostly understand. The research on lexical coverage (Hu and Nation, 2000, Reading in a Foreign Language) suggests you need to know about 98% of the words in a text to read it comfortably without help. Pick the tool that gets you to that kind of text quickest, and makes you want to open it tomorrow. Our piece on the 98% rule explains why.

FAQ

What is the best free LingQ alternative?

For most people it’s Readlang’s free plan, which gives you unlimited single-word translations and 10 phrase translations a day. If you want something completely free with no limits and you’re comfortable installing software, Lute is open source and runs on your own computer.

Is there a cheaper alternative to LingQ?

Yes. Readlang Premium is $6 a month or $48 a year, against LingQ Premium’s $14.99 a month or $10 a month billed yearly. Language Reactor has a free version too, but it’s mainly for video rather than reading.

Is Migaku better than LingQ?

It’s better for some people. Migaku is stronger if you learn mostly from Netflix, YouTube and anime and like making sentence cards. LingQ is stronger for reading, supports far more languages (50+ against Migaku’s 11), and has a much bigger library.

Can I move my known words from LingQ to another app?

Not directly. There’s no standard format for known words, and at the time of writing LingQ doesn’t offer a one-click export of them. Community-made browser scripts can download LingQ vocabulary as a CSV file, and a few tools can take such a list in. Otherwise, the practical workaround is to read easy texts in the new app and mark words known as you go. Expect some manual work.

Does Lumo replace LingQ?

Not yet, because Lumo isn’t available. When it launches it will cover the core reading workflow, with imports, word levels, difficulty labels and FSRS flashcards. But it won’t have a browser extension, Netflix support or tutors at first.

This guide uses one very ambitious first novel as its running example: Gabriel García Márquez’s Cien años de soledad. Plenty of learners put it at the top of their list, and plenty of them stall on page 20. Looking at exactly why it’s hard tells you how to prepare for any novel, and how to get through this one when you’re ready.

Why your first novel feels so different

Short texts forgive a lot. If a news article is hard, it’s over in five minutes. A novel asks you to keep a world in your head for 300 pages: who is related to whom, what happened three chapters ago, why a character is angry. When your Spanish is still taking up most of your attention, there isn’t much left over for that.

The other issue is coverage, the share of words on the page you already know. Research suggests you need around 98% coverage to read comfortably without help, and Paul Nation estimated that for general written text that takes a vocabulary of roughly 8,000 to 9,000 word families (Nation, 2006, Canadian Modern Language Review). Most learners pick up a first novel well before that. That’s fine, but it means you need support, and you need a book that isn’t fighting you on every other front. We go deeper on the numbers in the 98% rule.

The running example: Cien años de soledad

Cien años de soledad (One Hundred Years of Solitude) was published in 1967 by Editorial Sudamericana in Buenos Aires. It follows seven generations of the Buendía family in the fictional town of Macondo, across 20 untitled chapters. It is one of the most widely read novels in Spanish, and its opening line is famous for a reason:

“Muchos años después, frente al pelotón de fusilamiento, el coronel Aureliano Buendía había de recordar aquella tarde remota en que su padre lo llevó a conocer el hielo.” (Gabriel García Márquez, Cien años de soledad, 1967)

That single sentence holds most of what makes the book hard for learners. Let’s take the difficulties one at a time.

Long sentences

The opening is actually short by the book’s standards. García Márquez often writes sentences that run for half a page, stacking clauses with commas and que until you’ve forgotten where the subject was. Learners usually understand each piece and lose the whole.

The fix is mechanical. When a sentence gets away from you, find the main verb first, then the subject, and treat everything else as decoration you can fill in on a second pass. In the opening line, the skeleton is el coronel Aureliano Buendía había de recordar aquella tarde: the colonel would remember that afternoon. Everything else (the firing squad, the father, the ice) hangs off that.

The same names, over and over

This is the problem everyone mentions. The family reuses names across generations, so you meet several characters called José Arcadio and many more called Aureliano, plus repeated Remedios, Amaranta and Úrsula. The repetition is deliberate: the book is about a family that keeps repeating itself. For a learner who’s already working hard at the sentence level, though, “which Aureliano is this?” can be the thing that makes you quit.

You solve this one before you start (more below).

Magical realism vocabulary

The book describes extraordinary things in a perfectly calm voice, and the calm is part of the effect. For you, it means a lot of vocabulary that no textbook prepared you for. The first chapter alone has gypsies bringing magnets, magnifying glasses and alchemy equipment to the village, and you’ll meet words for old scientific instruments, tropical plants, household objects from another century, and military ranks from Colombia’s civil wars.

The good news is that much of this vocabulary is concrete and repeats within the book. Once you’ve looked up alquimia or imán a couple of times, it’s yours for the rest of the novel.

The past tenses

Nearly the whole novel is narrated in the past, and it uses the full range of past tenses:

If you learned the preterite and imperfect as two lists of endings, a novel is where they finally start to make sense. The preterite is the plot line; the imperfect is the painted backdrop behind it. The pluperfect, which this book leans on heavily because it keeps jumping between moments in time, tells you a flashback has started.

The opening line jumps forward to the firing squad and then back to the afternoon with the ice, all within one sentence. The novel keeps doing this. Knowing that in advance turns confusion into “ah, it’s doing the time thing again.”

How to read a novel in Spanish without stalling

Here’s the plan, in roughly the order you’d use it.

1. Read an easier novel first

If Cien años de soledad is your goal, don’t make it your first book. Make it your third or fourth. A first novel should be short, plot-driven and set in a world you can picture without effort. Contemporary fiction, young adult novels and translations of books you’ve already read in English are all good choices. Reading Harry Potter in Spanish is a cliché because it works: you already know the plot, so your attention goes to the language.

García Márquez himself offers a useful stepping stone. Crónica de una muerte anunciada is short, tightly plotted and written in a plainer style. If you enjoy it, you’ll arrive at Macondo already used to his voice. For more options at different levels, see our list of the best Spanish books for learners. And if you’re still choosing between learner material and the real thing, graded readers vs native content covers when to switch.

A simple test before you commit: read the first two pages. If you’re looking up more than about one word per line, put the book aside for a few months. It’ll still be there.

2. Get the family tree before page one

For a book with a big cast, and especially for this one, a family tree is not cheating. Many editions print one at the front, and plenty of versions are online. Keep it open next to you. When a new Aureliano appears, find him on the tree and note his parents in the margin.

If you’d rather build your own, do it as you read: one line per character, with a few words on who they are. It doubles as a summary of the book when you’re done.

3. Read a chapter summary first (yes, really)

This feels like spoiling the book. For a first novel in a new language, it isn’t. Knowing roughly what a chapter is about gives you the background knowledge that makes lower coverage manageable. You’ll recognize events as they happen instead of spending half your effort figuring out what’s happening at all.

Read a summary of the chapter, in English or easy Spanish, just before you read it. Keep it to the chapter you’re on. Once you’re comfortable, stop using summaries and see how you do without them.

4. Listen while you read

An audiobook carries you through long sentences at a steady pace, gives you the rhythm of the clauses, and stops you from getting stuck on one hard word. Unabridged Spanish audiobooks of Cien años de soledad exist from more than one publisher; library catalogs list one at around 17 and a half hours. Check the narrator’s accent and speed on a sample before you buy.

There’s research behind pairing text and audio. In a study of learners reading graded stories, Ronan Brown, Rob Waring and Sangrawee Donkaewbua (2008) found that new words were picked up incidentally from reading, from reading while listening, and from listening alone, with reading-based modes retaining more than listening alone (Reading in a Foreign Language). The same study is a useful reality check: most new words were not learned from a single book, and words that came up more often were the ones that stuck. Reading lots, with support, is what moves the needle.

The 2024 Netflix series, filmed in Colombia in Spanish, is another way in. Watching an episode before you read the matching chapters gives you faces for the names and a picture of Macondo.

5. Read the first chapter intensively

The first chapter of any novel is the hardest. You don’t know the characters, the setting or the author’s style yet. So slow down on purpose. Read it with a dictionary, look up everything that seems important, reread paragraphs that lost you, and listen to the audio more than once if it helps.

This might take you a week. That’s fine. You’re building the vocabulary and the mental map that the rest of the book will reuse.

6. Then switch to extensive reading

From chapter two onward, change your rules. Read for the story. Look up a word only when it blocks your understanding of what’s happening, or when you’ve seen it three or four times and still don’t know it. Let the rest go. You’ll miss details, and that’s the trade you’re making for momentum.

This combination of a short intensive start and a long extensive middle is the core of the plan. Intensive reading teaches you the book’s language; extensive reading is what gets you to the last page. We explain the difference properly in extensive vs intensive reading.

7. Set a pace you can keep

Pick a small, daily target: ten pages, or one chapter section, or 20 minutes. Cien años de soledad has 20 chapters, so a chapter every three or four days gets you through in about two to three months. That’s a realistic pace for a first big novel, and finishing slowly beats quitting fast.

What to do when you get lost

You will get lost. Every learner does, and so do plenty of native readers with this particular book. When it happens:

• Reread the last page you understood, then keep going.
• Check the family tree. In this novel, confusion is usually about who, not what.
• Read the summary for that section and go back in.
• Keep reading anyway. Often the next page explains what you missed. Novels repeat their important information.

What you shouldn’t do is go back and reread the whole chapter with a dictionary. That’s how a two-month project becomes a six-month one, and then an abandoned one.

How Lumo helps

Lumo is a reading and listening app we’re building for exactly this kind of project, and it’s coming soon. You’ll be able to import an EPUB of a book you own, tap any word for its meaning in context, and save it with its sentence, so the words you look up in chapter one turn into flashcards you’ll review while you read chapter five. Every text gets a difficulty label based on how many of its words you already know, which makes the “is this book too hard yet?” question much easier to answer.

But the plan above works with a paperback and a pencil too. The main thing is to pick the right first book, prepare for what’s hard about it, and keep reading.

FAQ

Is Cien años de soledad too hard for a first Spanish novel?

For most learners, yes. The long sentences, repeated names, unusual vocabulary and constant jumps in time add up. It’s a great third or fourth novel, once you’ve read a couple of easier books and are comfortable with the past tenses.

What level do I need to read a novel in Spanish?

There’s no fixed level, because it depends on the book. Many learners start an easier contemporary novel at around B1 with a dictionary handy, and literary novels usually feel more comfortable from B2 upward. A better test than your level is the book itself: if the first two pages need more than about one lookup per line, try something easier.

Should I look up every word I don’t know?

No. Look up words that block your understanding of what’s happening, and words you keep seeing. Skip the rest. Looking up everything turns a novel into a translation exercise and makes it much more likely you’ll quit.

Is it okay to read a summary or watch the adaptation first?

Yes. Background knowledge makes difficult text far easier to follow, and knowing the plot frees your attention for the language. For a first novel in a new language, understanding the story is more important than being surprised by it.

How long does it take to read a first novel in Spanish?

It varies a lot with the book and your level, but two to three months for a long novel at a steady daily pace is realistic. A short novel might take a few weeks. Your second novel will go noticeably faster than your first.

Both tools rest on the same idea: read real text, tap the words you don’t know, and let the app keep track. The difference is in how much they do around that core, and what that costs.

Readlang vs LingQ at a glance

Prices are US prices from each company’s official pages in October 2026. Prices change, so check before you pay.

Sources: Readlang pricing, LingQ free vs Premium, LingQ homepage.

Price and free plans

This is the biggest practical difference.

Readlang’s free plan is a real plan. Single-word lookups are unlimited, so you can read as much as you like. The caps apply to phrase translations (10 a day, up to 6 words each) and AI context explanations (10 a day). Plenty of people use it for months without paying.

LingQ’s free plan is a trial in all but name. Twenty saved words in total goes in a single sitting. LingQ is upfront about this: its own review calls the free plan very limited.

On paid plans, Readlang Premium costs $48 a year. LingQ Premium costs about $120 a year on the annual plan, or $14.99 a month if you pay monthly. LingQ also offers a 24-month plan at $8.99 a month, paid upfront.

The top tiers are both about AI. Readlang Premium Plus ($15 a month) adds high-quality voices for whole texts, more audio storage and a stronger AI model. LingQ Premium Plus adds AI-generated voices, a more advanced AI chat and AI-simplified lessons. Most learners don’t need either.

How reading feels

LingQ

On LingQ, every word on the page has a status. New words are highlighted blue. Words you’ve saved (LingQs) are yellow, fading as you move them through levels. Known words are plain. When you finish a page, the blue words you didn’t touch become known, so your known-word count climbs as you read.

That count is one of the most motivating things about LingQ. It turns “am I getting anywhere?” into a number that goes up every day. The page also tells you at a glance how hard a lesson is: lots of blue means lots of new words.

Readlang

Readlang is quieter. The text looks like a normal page until you click a word, and then you get a translation in place. Click and drag across several words for a phrase. Turn on auto-highlight and the words you’re studying are underlined when they come back.

What Readlang doesn’t do, at the time of writing, is mark every untouched word as new, or give you a running total of known words. Some learners find that calmer. Others miss the feedback.

Our take: if you’re motivated by visible progress, LingQ’s system is better. If you find a page full of blue stressful, Readlang will feel lighter.

Content and importing

LingQ has the library

LingQ comes with a large library of lessons, many with audio, across its 50+ languages. For beginners that matters: you can find something at your level without hunting. It also imports almost anything, including web pages through the LingQ Importer extension, YouTube, Netflix, podcasts and ebooks. According to LingQ’s own review, imported audio is capped at 60 minutes and 60 MB per file, and longer content is split into separate lessons.

Readlang reads where you already are

Readlang’s strength is that it comes to the text. The Chrome extension and bookmarklet let you click words on any web page without importing it. You can also upload texts and DRM-free EPUB books, read suggested articles and stories, and sync YouTube videos with their transcripts. It doesn’t support Netflix.

If you already know what you want to read, such as a news site, a novel or a blog, Readlang has less friction. If you need the app to find you material, LingQ is stronger.

Listening and audio

LingQ is built around reading and listening. Many library lessons have real recorded audio, and its playlist feature lets you listen on the go.

Readlang leans on text-to-speech. Premium adds high-quality voices for words and phrases, plus audio upload and transcription in 61 languages, with 10 hours of storage. Premium Plus raises that to 200 hours and adds high-quality voices for whole texts.

If listening matters to you as much as reading, LingQ’s library audio is the stronger offer. For more on why pairing the two helps, see how to learn a language from YouTube.

Vocabulary review

Both tools save the words you look up, with their sentence, and let you review them.

• Readlang uses spaced repetition flashcards and lets you export your words. Premium adds practice and export from specific books.
• LingQ offers several review types, including multiple choice, fill-in-the-blank and dictation. Its own review admits a common problem: save words fast and the review queue grows faster than you can clear it.

Neither is the main event. In both apps the reading does most of the work, and the review just catches words that haven’t come back yet. If you take reviewing seriously, pick whichever exports cleanly to the flashcard app you already use. Our FSRS explainer covers how modern schedulers decide when you see a card.

Who should pick Readlang

• You want a free plan you can live on, or the cheapest paid plan.
• You read web articles and your own books and want to click words in place.
• You’re learning a less common language. Readlang supports 100+.
• You prefer a quiet page to color-coded word statuses.

Who should pick LingQ

• You want a built-in library with audio, especially as a beginner.
• You’re motivated by a known-word count and seeing every word’s status.
• You want native mobile apps for reading and listening on the go. Readlang runs in a mobile browser but has no app.
• You don’t mind paying around $120 a year for an all-in-one system.

Is there something in between?

That’s roughly where we’re aiming with Lumo, which isn’t available yet. Like LingQ, it tracks every word: new words are blue, saved words move through levels and their yellow fades, and when you turn the page the new words you didn’t tap become known. Like Readlang, it’s built around reading your own material: pasted text, web articles, EPUBs, YouTube transcripts, and audio or video with subtitles. Each text also gets a difficulty label from your known-word coverage.

It doesn’t have a browser extension yet, or LingQ’s library. Planned pricing is a free tier and Pro at $5.99 a month or $49.99 a year.

The verdict

Both are good tools with long track records, and you can’t really go wrong. Choose Readlang if price and simplicity come first, or you read mostly on the web. Choose LingQ if you want the library, the known-word count and the apps, and you’ll use them enough to justify the cost.

Whichever you pick, aim for texts where you know most of the words. You’ll read more, enjoy it more, and learn more words along the way. Research by Hu and Nation (2000, Reading in a Foreign Language) suggests about 98% known-word coverage is where unassisted reading becomes comfortable. Our post on the 98% rule has the details.

FAQ

Is Readlang free?

Yes, Readlang has a free plan with unlimited single-word translations, 10 phrase translations a day and 10 AI context explanations a day. Premium, which removes the limits, costs $6 a month or $48 a year.

Is LingQ worth the money?

For learners who use the library, the audio and the known-word tracking every day, many find it is. At about $120 a year on the annual plan it costs more than most alternatives, so try it properly during the free plan or first month before committing to a long plan.

Which has more languages, Readlang or LingQ?

Readlang lists over 100 languages, and LingQ supports more than 50. LingQ’s library content and audio are richer in its major languages, so check what’s available for yours before choosing.

Can I use Readlang and LingQ together?

You can. Some learners use LingQ for library lessons and listening, and the Readlang extension for clicking words while browsing the web. Your vocabulary won’t sync between them, though, so you’ll be keeping two word lists.

Below is a level-by-level list. The levels are our estimates for a typical adult learner, not publisher labels, and your own coverage will depend on your vocabulary and what you already know about the story.

How to use this list

A level label on a cover is a rough guide. What actually decides whether a book feels good is how many words per page you don’t know. The research behind that is summed up in the 98% rule: Hu and Nation (2000, Reading in a Foreign Language) found that most readers needed about 98% of the words in a fiction text to understand it well without help. A later replication by Kremmel and colleagues (2023, Language Learning) couldn’t fully reproduce the original numbers, so treat 98% as a guideline rather than a hard line. It’s still a very good guideline.

Two practical rules follow from it:

• If you’re looking up a word in every line, step down. That’s around 90% coverage. You can push through a short text that way, but not a whole book.
• If a book feels easy, keep reading it. Easy books are where the volume comes from, and reading in volume is how you get to the hard books.

If you’re unsure when to stick with learner books and when to switch, we’ve covered that in graded readers vs native content.

The best Spanish books for learners at a glance

A1: your first stories

At A1 you know a few hundred words, and almost no real book will be comfortable. That’s fine. A1 readers are short, repetitive and built around the most common words, and their job is to show you that reading in Spanish is possible at all.

Lola Lago, detective (Difusión)

This is the series many teachers hand to beginners first. It was written by Lourdes Miquel and Neus Sans and is published by Difusión in Barcelona. Lola Lago runs a small detective agency in Madrid, and every book is a self-contained case. There are seven titles in the original series, from A1 to B1: Vacaciones al sol (A1); Una nota falsa, Poderoso caballero and Por amor al arte (A2); La llamada de La Habana and Lejos de casa (A2+); and ¿Eres tú, María? (B1). Each is short enough to finish in a sitting or two, and current editions come with audio.

Difusión has since added a newer series, Lola Lago & asociados, by the same authors. It starts at A1 with titles such as Sin noticias and Enredados, and each book comes with downloadable MP3 audio for every chapter, cultural notes, activities and a multilingual glossary.

Why it works: the same characters and setting come back, so vocabulary repeats from book to book.

Paco Ardit’s Spanish Novels

Paco Ardit’s self-published Spanish Novels series has 26 short readers across six levels, from A1 to C2, available as ebooks and, for many titles, audiobooks. They’re cheap and there are plenty of them, and that’s the point. Some reviewers find the plots thin and the writing a little stiff at the lower levels. For a beginner who needs pages of easy text more than great literature, that’s a fair trade.

A2: more words, longer stories

At A2 you can follow a simple plot with past tenses, and a 100-page book stops being scary.

Olly Richards, Short Stories in Spanish for Beginners

Published by Teach Yourself in 2018, this book collects eight stories in different genres (crime, science fiction, fantasy and others). The publisher maps it to A2–B1. Each chapter comes with a glossary of bolded words and comprehension questions, and there’s an audiobook edition narrated by Javier Marzán. It’s a good bridge book: the stories are longer than in most A2 readers, but the support is generous.

SGEL, Lee y disfruta

SGEL’s Lee y disfruta readers come in levels such as elemental (A2) and medio, with footnotes on cultural points, exercises and a chapter-by-chapter multilingual glossary. They’re more classroom-flavored than Lola Lago, which suits some readers and bores others.

Easy Spanish Reader (William Tardy)

This one is an old American classic, published by McGraw-Hill and now in a fourth “Premium” edition with an app. It’s in three parts: simple stories about two students, Enrique and María; short readings on Mexican history; and an adapted Lazarillo de Tormes. It’s dated in places, but the difficulty rises steadily.

Your first real book: El principito

Antoine de Saint-Exupéry’s Le Petit Prince has been read in Spanish since 1951, in the translation by Bonifacio del Carril. Salamandra publishes it in Spain, including a bilingual English–Spanish edition. The sentences are short and the vocabulary is concrete. Many people already know the story, and that background knowledge carries you through the trickier passages. At A2 it’s a stretch. At B1 it’s very manageable.

B1: where real books begin

B1 is the turning point. You know maybe 2,000 to 3,000 words, and with a look-up tool you can read books written for native speakers, as long as you pick them well. If that’s your goal, our guide to reading your first novel in Spanish goes step by step.

Graded readers that still help

• ¿Eres tú, María? is the B1 finale of the Lola Lago series, and a satisfying one if you’ve read the others.
• Olly Richards, Short Stories in Spanish for Intermediate Learners (Teach Yourself, 2019) has eight more stories, mapped to B1 going on B2, with the same glossary-and-questions format and an audiobook.
• Edinumen’s Lecturas de Español are original stories in six levels, from elemental through intermedio to superior, with vocabulary notes in the margins and activities. Titles include Carnaval (elemental) and A los muertos no les gusta la fotografía (superior).
• SGEL’s Historias breves para leer come in three levels (elementary, intermediate, advanced). Each is a set of short texts with exercises and a Spanish–English glossary, which is handy if you only have ten minutes a day.

Real books for B1

Harry Potter y la piedra filosofal (translated by Alicia Dellepiane, Salamandra). If you’ve read it in English, this is one of the easiest long books you’ll ever read in Spanish. You’ll guess most new words from a plot you already know, and the long, steady stretch of one story is exactly what B1 readers need.

Manolito Gafotas by Elvira Lindo (Alfaguara, 1994). Manolito is a boy from the working-class Madrid neighborhood of Carabanchel, and he tells his own story in a rambling, very funny voice. The character began on Lindo’s radio scripts, and she won Spain’s Premio Nacional de Literatura Infantil y Juvenil in 1998. It’s a children’s book, but don’t let that fool you. It’s full of colloquial Madrid Spanish and running jokes. That makes it wonderful input, but it reads harder than its length suggests, so read it at a strong B1, not a weak one.

B2: books you’ll actually talk about

At B2 you can read most contemporary fiction with occasional look-ups. The challenge is less vocabulary and more stamina, style and cultural references.

Marina, Carlos Ruiz Zafón

Zafón wrote Marina as a young adult novel, and Edebé published it in Barcelona in 1999, two years before he became world-famous. A teenager at a boarding school in Barcelona meets a girl called Marina, and the two of them get pulled into an old, dark mystery. It has all the atmosphere of his later books in simpler language and at half the length. If you plan to read La sombra del viento eventually, read this first.

Como agua para chocolate, Laura Esquivel

The full Spanish title calls it a novela de entregas mensuales con recetas, amores y remedios caseros, a novel in monthly installments with recipes, love stories and home remedies. It has twelve chapters, each named for a month and opening with a recipe. That structure is a gift to a learner. Each chapter is a natural stopping point, and kitchen vocabulary (ingredients, verbs like picar, freír and batir) repeats constantly. The Mexican setting also gives you a break from Peninsular Spanish if most of your input so far has come from Spain.

Crónica de una muerte anunciada, Gabriel García Márquez

The narrator tells you who dies on the first page, and the rest of the book reconstructs how a whole town let it happen. Because you know the outcome, you can focus on the language. It’s the most approachable García Márquez.

Graded material at B2

At this level, graded readers are optional. One useful exception is SGEL’s Cuentos, cuentos, cuentos, which collects unadapted short stories from Spanish and Latin American writers for intermediate and advanced students, with notes. It’s a good way to sample authors before you commit to a whole novel by one of them.

C1: the books people are reading in Spanish

By C1 the limit is mostly your patience. These books are long or dense or both, and they’re worth it.

La sombra del viento, Carlos Ruiz Zafón

Published by Planeta in 2001, this is the first book in the Cementerio de los Libros Olvidados series. In 1945 Barcelona, a boy named Daniel Sempere is taken to a hidden library and chooses a book that takes over his life. It’s long (the Planeta paperback runs to almost 600 pages) and the prose is ornate, but the plot keeps you turning pages. A long, gripping book is exactly what extensive reading needs.

Nada, Carmen Laforet

Laforet wrote Nada in 1944, when she was in her early twenties. It won the first Premio Nadal and was published in 1945. A young woman named Andrea arrives in postwar Barcelona to study and moves in with her strange, bitter family. It’s a classic of twentieth-century Spanish literature. The vocabulary is literary and the mood is heavy, so it’s a C1 book, but a short one.

Pedro Páramo, Juan Rulfo

It’s only around 130 pages, but don’t let the length fool you. Rulfo’s 1955 novel jumps between voices and times without warning, and some of the voices belong to the dead. The words aren’t the hard part. Following the structure is. Read it once for the atmosphere, then again with a summary at hand.

Reading these books well

How you read matters as much as what you read. A few habits that pay off:

• Mostly read extensively. Read fast, look up only the words that block the story, and keep going. Save slow, line-by-line study for a page or two a day. We explain the difference in extensive vs intensive reading.
• Use the audio. Lola Lago, the Olly Richards books and many of the novels above have audiobooks. Listening while you read helps with pronunciation and keeps your pace steady.
• Read narrowly. Three Lola Lago books in a row, or Marina followed by La sombra del viento, is easier than jumping between authors. The vocabulary and style carry over.

If you read on a screen, a reader with tap-to-translate helps a lot at B1 and B2. Lumo, the reading app we’re building, will do this when it launches (it’s coming soon). You’ll be able to import an EPUB you own and tap any word for its meaning in context. Each book also gets a difficulty label based on how many of its words you already know, so you can check whether Marina is Just right for you before you start. For A1 to B1, Lumo Stories adds short original graded stories in Spanish.

FAQ

What is the best first book to read in Spanish?

For a true beginner, a graded reader such as Vacaciones al sol from the Lola Lago, detective series is the safest start. It’s short, it has audio and it’s written for A1. If you’re already around B1 and want a real book, El principito or Harry Potter y la piedra filosofal are good first picks, because many readers already know the story.

Are graded readers worth it, or should I read real books?

Both have a place. Graded readers let you read a lot at high coverage early on, when real books would mean a look-up in every line. Most learners can start mixing in real books around B1 and drop graded readers at B2.

Is Manolito Gafotas easy?

It’s a children’s book, but it isn’t easy. Manolito’s voice is full of colloquial Madrid Spanish and slang, so it reads closer to B1+ or B2 than its length suggests. It’s great input once you’re there.

Should I read Spanish books from Spain or Latin America?

Read whatever you’ll enjoy. The written language is far more uniform than the spoken one, and a B1 reader can handle either. If you want variety, pair a book from Spain such as Marina with a Mexican one such as Como agua para chocolate.

How many books should I read to reach B2?

There’s no fixed number, but volume is what counts. Reading even 15 to 20 pages a day of material at the right level adds up to dozens of books a year, and that’s the kind of exposure that moves people from B1 to B2.

Both come from the same comprehensible-input school of thought, and both work. They just start from opposite ends.

LingQ vs Migaku at a glance

Prices are US prices from official pages in October 2026. Both change over time, and Migaku has discounted its lifetime plan during sales.

Sources: LingQ free vs Premium, LingQ homepage, Migaku pricing, Migaku FAQ.

Languages

This settles it for many people.

Migaku supports 11 languages: Japanese, Mandarin Chinese, Cantonese, Korean, Spanish, English, German, Portuguese, Vietnamese, Italian and French. One subscription covers all of them, though its tools and courses are deepest for Japanese and Chinese.

LingQ supports more than 50, including many that most apps skip. If you’re learning Swedish, Greek, Ukrainian or Persian, Migaku isn’t an option, and LingQ is.

Price

Migaku is cheaper month to month: Standard is $10 a month against LingQ Premium’s $14.99. On yearly plans they’re close: $96 for Migaku Standard, about $120 for LingQ Premium. LingQ’s 24-month plan brings it down to $8.99 a month, paid upfront.

The real difference is in how you try them.

• LingQ has a permanent free plan, but it’s tiny: 20 saved words in total and 5 imported lessons. It shows you how the reader works, and that’s all.
• Migaku has no free plan. You get a free trial (10 days at the time of writing), then pay. Migaku offers a full refund within 14 days of purchase.

Migaku also sells a lifetime plan at $499. At the time of writing it includes Early Access features. If you know you’ll use it for years, it can pay off. It’s been discounted during sales, so if you’re considering it, it may be worth waiting for one.

Reading

LingQ is built for it

Reading is LingQ’s home turf. You open a lesson, and every word has a status. New words are blue. Saved words are yellow and fade as you learn them. Known words are plain. When you turn a page, the new words you didn’t tap become known, and your known-word total climbs.

You can import nearly anything: web pages through the LingQ Importer extension (Chrome, Firefox and Safari), ebooks, podcasts, YouTube and Netflix. Long texts are split into lessons. And the library means you can always find something at your level.

Migaku reads the web

Migaku’s browser extension works on ordinary web pages, too. Click a word for a definition and an AI sentence breakdown, then make a card. It estimates how hard a page will be from what you know. On mobile, the app has a built-in browser and OCR, so you can look up words in a printed textbook.

But Migaku has no lesson library in LingQ’s sense, and the experience is designed around grabbing sentences rather than reading long texts from start to finish.

For long-form reading, LingQ wins clearly. If you want to read a novel, LingQ (or another dedicated reader) will feel more natural. See how to import EPUB books for language learning for options.

Video and listening

Migaku is built for it

This is Migaku’s home turf. Its extension adds clickable subtitles to Netflix, YouTube, Disney+, Rakuten Viki and Animelon, and it plays video files you own. Click any word in the subtitle to look it up. Click again to make a card that includes the sentence, the audio clip and a screenshot from the scene.

For anime and drama learners, that’s a huge deal. Cards with the actor’s voice and the scene image are far more memorable than a word and a translation.

LingQ imports video

LingQ can import YouTube videos and Netflix subtitles, but it turns them into lessons: you read the transcript with the audio, rather than watching with clickable subtitles. That’s good for intensive study of a clip, and it’s different from watching a show.

LingQ’s library lessons usually come with recorded audio, and playlists let you listen on the go. That’s real strength for listening practice, just not for video.

For video, Migaku wins clearly. We cover more of this in learning from Netflix and YouTube subtitles.

Flashcards and review

Here the philosophies differ most.

Migaku treats cards as a core part of the method. Its FAQ recommends cards built from sentences with a single unknown word, and a pace of about 10 new cards a day. Cards are rich, with audio and images, and you review them in Migaku Memory on the web or in the app. If you’ve done Refold-style sentence mining, this will feel familiar.

LingQ treats review as support. Every word you save becomes a card with its sentence, and you can review it with multiple choice, fill-in-the-blank or dictation. But LingQ’s stance is that reading is the main event and review is a top-up. Its own review admits that if you save words quickly, the queue can outgrow your time.

Neither is right. If you like making and reviewing cards, Migaku’s are better. If you’d rather spend almost all your time reading, LingQ asks less of you.

Setup and learning curve

LingQ works the moment you sign up. Pick a language, open a lesson and start tapping. There’s a lot of interface to learn later, but the first session is easy.

Migaku asks more. You install a browser extension (designed for Chrome) and the mobile app, set up your language, and learn the card workflow. It’s not hard, but the first hour is more setup than study. In return you get a toolkit that works across many sites.

For beginners, Migaku also offers structured courses: Fundamentals (writing system and pronunciation), Academy (roughly the first 1,500–1,700 words and a few hundred grammar points, depending on the language) and a Kanji course for Japanese. LingQ’s library includes beginner lessons, but no course in the same sense.

Who should pick LingQ

• You learn mainly by reading, and want long-form texts and books.
• Your language isn’t one of Migaku’s 11.
• You want a library that always has something at your level.
• You’re motivated by a known-word count.

Who should pick Migaku

• You learn mainly by watching: Netflix, YouTube, anime, drama.
• You’re learning Japanese, Chinese or Korean, where its tools are most developed.
• You like sentence mining and want rich cards with audio and screenshots.
• You’d like a structured course alongside native media.

Can you use both?

Plenty of people do: LingQ for reading, Migaku for video. The catch is that your known words don’t move between them, so each app’s idea of what you know will be incomplete. If you go this way, pick one as your main flashcard home so you’re not reviewing the same word in two places.

Where Lumo fits

We’re building Lumo, which isn’t available yet, so take this as context rather than a recommendation. It’s closer to LingQ in spirit: reading first, with every word tracked and a difficulty label on each text. It also covers YouTube with an interactive transcript and dual subtitles, plus audio and video files with SRT or VTT subtitles, and saved words become FSRS flashcards. It has no Netflix support or browser extension, so for streaming-heavy learners Migaku remains the better fit.

The verdict

LingQ is the better reading app, covers far more languages and has the library. Migaku is the better video and sentence-mining tool, especially for East Asian languages. Pick the one that matches the hours you’ll actually put in, because the input itself is what matters. The research on incidental vocabulary learning, such as Pigada and Schmitt’s 2006 case study in Reading in a Foreign Language, points to lots of encounters with words in context. In that study, a month of extensive reading improved one learner’s knowledge of 65% of the 133 target words in some way. Both apps can give you that.

FAQ

Is Migaku better than LingQ for Japanese?

For most Japanese learners who watch anime, drama or YouTube, yes. Migaku’s subtitle tools, sentence cards and Kanji course give Japanese learners a lot in one place. If you mainly want to read Japanese novels and articles, LingQ or a dedicated reader like Satori Reader may suit you better.

Does Migaku have a free version?

No. Migaku offers a free trial (10 days at the time of writing) and a full refund within 14 days of purchase, but no permanent free tier. Standard costs $10 a month or $96 a year in the US.

Does LingQ work with Netflix?

Yes, through import. LingQ turns Netflix and YouTube content into lessons, so you read the subtitles as a text with the audio. If you want clickable subtitles while you watch in the player itself, Migaku and Language Reactor are built for that.

Which is cheaper, LingQ or Migaku?

Migaku is cheaper monthly ($10 against $14.99) and slightly cheaper yearly ($96 against about $120). LingQ’s 24-month plan ($8.99 a month) and Migaku’s $499 lifetime plan change the math if you’re committing long term.

Japanese looks like the hardest language to read your way into. It’s also the language where reading pays off most, because the writing system is something you can only really learn by seeing it thousands of times. Here’s how to get from your first hiragana to your first novel.

Why reading works so well for Japanese

Two things make Japanese different from, say, Spanish.

First, there’s a long distance between the language and English. You can’t lean on shared vocabulary the way an English speaker can in French. Every word has to be met, recognized and met again. That takes a lot of input, and reading is the fastest input you can control: you set the pace, you can re-read, and you can look things up.

Second, the writing system rewards exposure. Kanji are much easier to remember as parts of words you keep seeing than as isolated characters on a list. Many learners find that a few hundred graded stories teach them more kanji, in context, than the same hours spent drilling a list.

The same coverage research that applies everywhere applies here. Reading is comfortable when you know about 98% of the running words on the page, and it gets hard fast below 95%. (We dig into the numbers in the 98% rule.) In Japanese, the trick is that “knowing a word” includes being able to read how it’s written. That’s where furigana comes in.

The three scripts, briefly

Written Japanese mixes three scripts in the same sentence.

• Hiragana (46 basic characters) writes grammar: particles, verb endings, and words that have no common kanji.
• Katakana (also 46 basic characters) writes loanwords, foreign names and emphasis: テレビ (terebi, TV), コーヒー (kōhī, coffee).
• Kanji write most nouns and the stems of verbs and adjectives.

Learn both kana sets before anything else. It takes most people one to three weeks of short daily practice, and it unlocks almost every beginner reader.

Kanji are the long project. Japan’s official list of characters for general use, the Jōyō kanji, was revised in 2010 and now has 2,136 characters. Schoolchildren learn just over a thousand of them in elementary school, and Japanese children’s books lean heavily on furigana until they do. You don’t need all 2,136 to start reading. You need enough of the common ones, plus readings handed to you for the rest.

Furigana, and why it matters so much

Furigana are small kana printed above (or beside) kanji to show how they’re read. 漢字 with furigana shows かんじ on top.

For a learner, furigana changes everything. Without it, an unknown kanji is a wall: you can’t sound it out, you can’t easily look it up, and you can’t connect it to a word you might already know by ear. With it, the kanji becomes just another word you can read aloud, hear in your head and look up in a second. Every time you see it, the shape and the sound get linked a little more.

That’s why the best path through Japanese reading is really a path through furigana:

1. Full furigana on every kanji (graded readers, children’s books, NHK Easy).
2. Furigana on harder kanji only (many shonen manga, some learner material).
3. Furigana only on rare or unusual readings (most novels, including light novels).
4. No furigana at all (newspapers, most adult nonfiction, the web).

A good rule: don’t leave a stage until reading at it feels easy. And don’t be embarrassed to read with furigana for a long time. Native readers do too.

Graded readers: where to start

Graded readers are books written with controlled vocabulary and grammar for learners. In Japanese, the main ecosystem is built around tadoku (多読, “reading a lot”), an extensive-reading approach promoted by the nonprofit NPO Tadoku Supporters (NPO多言語多読).

The Tadoku levels

The NPO grades books on a shared scale. According to its levels page:

The JLPT column is the NPO’s own loose guide, not a promise, and the page also lists an easier Pre-Starter level below Level 0. Level 0 and 1 books stick to polite present and past forms; plain forms, て-forms and conditionals arrive at Levels 2 and 3.

Free readers

The NPO hosts a free library of graded books you can read online or download as PDFs, many with furigana and some with audio narration. Many of the free titles sit at the lower levels, which is exactly where you need the most volume. They’re shared under a non-commercial, no-derivatives license, so read and print them freely, but don’t repost edited versions.

Paid series

The NPO’s graded readers page lists the main print series:

• Japanese Graded Readers (Ask Publishing): Levels 0–4, 86 books sold in packs. Includes abridged classics by authors such as Natsume Sōseki.
• Taishukan Japanese Readers: Levels 0–5, 55 books in packs, and the main source of Level 5 titles.
• Nihongo Tadoku Books (NPO Tadoku Supporters): Levels 0–3, sold individually, also available as an app from White Rabbit Press.

Packs are not cheap if you buy them new, so check your library first. University libraries with Japanese programs often carry them.

Let’s Read Japanese

A separate series, Let’s Read Japanese from Oxford Brookes University, is graded to the CEFR instead of the tadoku scale: six levels of ten books each, running from A1 to B2. Its levels don’t line up one-to-one with tadoku levels, so treat each series on its own terms and start one step easier than you think you need.

If you’re wondering when to move from graded material to native content, we compare the two in graded readers vs native content.

Beyond books: news, stories and websites

NHK News Web Easy

NHK, Japan’s public broadcaster, has published simplified news for foreign residents and children for more than a decade. NHK News Web Easy rewrites selected real news stories in easier Japanese, with furigana you can turn on and off and audio you can play along with the text. Some reviewers note that the audio is synthesized rather than read by a person.

It’s a natural step after Level 2 or 3 graded readers. The topics are real, the sentences are short, and each story is often a simplified version of a regular NHK article, so you can later read the full one as a challenge.

Satori Reader

Satori Reader is a paid reading app built only for Japanese. It offers a large library of serialized stories and articles (more than 1,650 episodes at the time of writing), with native-speaker audio, inline grammar and vocabulary notes, and built-in review cards. You can choose to read in kanji, kana, or kanji with furigana.

At the time of writing, its pricing page lists a free tier with selected episodes and a Pro plan at $9/month or $89/year. Satori is one of the best options for intermediate learners who want polished, explained Japanese with matching audio. If you want only Japanese and you value the explanations, it may be the better pick over a general reading app.

Watanoc

Watanoc (和タのC) is a free web magazine in easy Japanese, with articles tagged N5, N4 and N3 about food, travel and daily life. The newest dated articles we could find are from 2016, so think of it as a back catalog rather than a living site. The archive is still useful for beginners who want short, cheerful, real-world texts.

Manga and light novels

Manga

Manga is where many learners first read something “real.” Pictures carry a lot of meaning, dialogue is short, and many series aimed at younger readers print furigana over the kanji.

Yotsuba&! (よつばと!) is the classic first manga. It follows a five-year-old girl through everyday life, so the vocabulary is concrete and the situations are obvious from the art. Most kanji carry furigana. The catch is that the dialogue is very casual, with contractions and child speech you won’t meet in textbooks. Read it after a few weeks of graded readers, not instead of them.

Shonen manga (aimed at teenage boys) often include furigana, which is why series in magazines like Weekly Shōnen Jump are popular with learners. Check a sample page before you buy; furigana practice varies by publisher and series, and seinen and josei titles for adults often drop it.

Light novels

Light novels are short, dialogue-heavy novels, often with anime or manga adaptations. They’re a good bridge to full novels because the prose is plain and the stories are easy to follow, especially if you’ve watched the adaptation. Furigana usually appears only on rare kanji or unusual readings, so you’ll need a lookup tool. They suit learners around tadoku Level 4–5 or JLPT N3–N2.

A level-by-level path to learn Japanese by reading

Here’s a plan you can follow. Move on when the current stage feels easy, not when the calendar says so.

Stage 1: kana and first stories (weeks 1–6)

• Learn hiragana, then katakana.
• Read tadoku Pre-Starter and Level 0 free books, which are often hiragana-only or fully annotated.
• Re-read favorites aloud. Speed comes from repetition.

Stage 2: graded readers with full furigana (months 2–6)

• Read Levels 1–2, as many books as you can. Aim for volume over difficulty.
• Start learning common kanji alongside reading, but let reading do most of the work.
• Save a handful of new words per session for review.

Stage 3: news and learner stories (months 6–12)

• Move to Levels 3–4, NHK News Web Easy and learner apps such as Satori Reader.
• Try turning furigana off for kanji you know, and back on when you get stuck.
• Start an easy manga like Yotsuba&!.

Stage 4: native content with support (year 2)

• Manga with partial furigana, light novels, articles on topics you know well.
• A lookup tool becomes essential here, because most kanji are no longer annotated.
• Many learners start sentence mining at this point: saving whole sentences with one new word each.

Stage 5: reading freely

• Novels, essays, the web, no furigana.
• Re-read early graded books occasionally. You’ll be surprised how fast they go.

Lookup tools and the problem of no spaces

Japanese is written without spaces between words. 私は本を読みました (“I read a book”) has five words and zero gaps. For a native reader the kana-kanji rhythm shows the boundaries. For a learner, especially one facing unknown kanji, it’s often unclear where one word ends and the next begins.

That’s why lookup tools for Japanese do two jobs:

1. Segmentation. The tool has to split the text into words. Software uses dictionaries and statistical models (tools such as MeCab are common) to guess the boundaries. They’re good but not perfect, especially with names, slang and long compounds.
2. Deinflection. Japanese verbs and adjectives change form a lot. 読みました, 読んで and 読まなかった all come from 読む. A good tool traces the form back to the dictionary word.

Popular options include browser extensions like Yomitan (free, for Chrome, Firefox and Edge), reading apps with their own parsers, and plain dictionaries like Jisho for copy-and-paste lookups. Whatever you use, the goal is the same: an unknown word should cost you a second, not a minute, so you stay inside the story.

One caution: tap-to-translate tools make it easy to read text that’s far too hard. Lookup speed helps, but it doesn’t replace coverage. If you’re tapping more than one word per line, step down a level.

Where Lumo fits

Lumo is a reading and listening app we’re building. It isn’t available yet, but it’s coming soon. It splits Japanese text into words you can tap for a meaning in context, saves new words with their sentences, and turns them into flashcards scheduled with FSRS. Every text gets a difficulty label based on how many of its words you already know, which takes the guesswork out of “is this too hard yet?” It also includes Lumo Stories, short original graded stories in Japanese at A1 to B1, for the stretch between your first graded readers and native content. You can import your own texts, web articles, YouTube videos and EPUB books too.

If you’re learning more than one language, our guide to learning Chinese by reading covers the same ideas for a script with no kana to lean on.

FAQ

Can you really learn Japanese just by reading?

Reading can carry most of your vocabulary, kanji and grammar growth, but most learners combine it with listening and some speaking practice. Reading alone won’t train your ear or your pronunciation. A good balance is reading as the core, with audio versions of what you read whenever they exist.

How many kanji do I need to start reading?

None, if you start with furigana. Graded readers and NHK News Web Easy annotate every kanji, so you can begin right after learning kana. To read native material without furigana comfortably, you’ll eventually want most of the 2,136 Jōyō kanji, but you’ll learn many of them through reading itself.

Should I turn furigana off?

Turn it off for kanji you already know, if your tool allows it, and leave it on for the rest. Forcing yourself to read without furigana too early turns reading into a kanji quiz and slows you down. The aim is to read a lot, and comfort is what keeps you reading.

Is manga good for learning Japanese?

Yes, with two caveats. Manga dialogue is casual and full of slang, contractions and character voices, so pair it with graded readers or news for more standard Japanese. And pick series with furigana until your kanji knowledge can handle text without it.

What should I read after graded readers?

NHK News Web Easy, Satori Reader and easy manga are the usual next steps, followed by light novels and books whose stories you already know. Choose texts where you know about 95% or more of the words, and use a lookup tool for the rest.

Here’s where the idea comes from, what holds up, what’s debated, and how to use it without turning it into dogma.

Where the idea comes from

The term is most closely associated with the linguist Stephen Krashen, who developed it in the late 1970s and early 1980s, most famously in Principles and Practice in Second Language Acquisition (1982). Krashen proposed a set of five hypotheses about how people pick up a second language. Two of them matter most for everyday learners.

The first is the distinction between acquisition and learning. Acquisition is the subconscious process that gives you a feel for what sounds right, the way you know your first language. Learning is conscious knowledge about the language: rules, conjugation tables, exceptions. Krashen argued that fluent, spontaneous use comes mostly from acquisition.

The second is the input hypothesis itself: acquisition happens when we understand input that is slightly beyond our current level. Krashen wrote this as i+1, where i is what you can already handle and +1 is the next step. You understand the new part thanks to context, pictures, tone of voice and everything you already know.

He also described an affective filter: anxiety, boredom and low confidence act like a filter that keeps input from “getting in.” Relaxed, interested learners acquire more from the same material.

What “comprehensible” really means

Comprehensible doesn’t mean you understand every word. It means you understand the message. When you watch a cooking video and miss a few words but know exactly what’s happening to the onions, that’s comprehensible input. When you read a sentence where every second word is a mystery, it isn’t, no matter how good the content is.

A useful way to think about i+1 is as a feeling rather than a formula:

• i+0 feels effortless. Pleasant, great for fluency, but you’re not meeting much that’s new.
• i+1 feels absorbing. You’re following along and occasionally stretching to catch something.
• i+10 feels like work. You spend more time decoding than understanding, and it’s tiring fast.

Most of your time should be spent in that middle zone.

What the research supports, and what it doesn’t

The core claim that people need large amounts of meaningful input is about as uncontroversial as anything in language learning gets. Studies of extensive reading (reading lots of easy, enjoyable material) and of incidental vocabulary learning keep finding that learners pick up words and grammar from input they understand. The gains from any single encounter with a word are small, which is exactly why volume matters: you need to meet a word many times, in different contexts, before it sticks.

Where Krashen’s model draws criticism is in its stronger claims. Researchers such as Barry McLaughlin pointed out that “i” and “+1” were never defined precisely enough to be tested directly. Others argued that input, while necessary, isn’t the whole story:

• Merrill Swain’s output hypothesis grew out of her work with Canadian French immersion students, who understood French very well after years of input but whose speaking and writing were often less accurate. Producing language, she argued, pushes you to notice gaps you can skip over when you’re only listening.
• Michael Long’s interaction hypothesis emphasized the value of conversation, where you negotiate meaning and get feedback in real time.
• Richard Schmidt’s noticing hypothesis proposed that you learn what you pay attention to. Input helps most when something in it catches your eye.

Most researchers today would put it like this: comprehensible input is the foundation, and attention, output and feedback build on top of it. That’s a far less dramatic message than “just listen and you’ll be fluent,” but it’s a far more useful one.

Turning the idea into practice

You don’t need to pick a side in an academic debate to benefit from this. A practical, input-first routine looks something like this:

1. Choose material you mostly understand. If you’re looking up every other word, step down a level. Graded stories, content made for learners and native material on familiar topics are all good options.
2. Make unknown words cheap. The moment a word blocks your understanding, look it up and keep going. The goal is to stay inside the story.
3. Notice, lightly. Pay a little attention to the words and phrases you’re learning when they come back. You don’t need to study them; you just need to see them.
4. Go for volume, not intensity. Fifteen relaxed minutes every day beats a two-hour cram on Sunday.
5. Re-read and re-listen. The second time through, more of the text becomes comprehensible, and what was i+1 becomes i.
6. Read narrowly. Stick with one topic, author or creator for a while. Vocabulary repeats, and each new piece gets easier.
7. Add output when you’re ready. Writing a few sentences or chatting with someone will feel easier once you have a deep well of words to draw from.

How Lumo is built around it

We built Lumo to make that routine as easy as possible. Every text in your library shows how much of it you already know and is labeled Easy, Just right, Challenging or Hard, so you can find i+1 without guessing. Tapping a word gives you a meaning that fits the sentence, so unknown words cost seconds instead of minutes. Words you’re learning stay highlighted on every page, which helps with the noticing part. Read-along links what you read to what you hear, and a daily streak keeps the volume coming.

None of that replaces the core idea, which has always been the same: find things you want to understand, and spend lots of time understanding them. The rest is just making that easier to do every day.

We don’t make Migaku, and we’re building a different kind of app, so we’ve tried to stick to what Migaku itself documents and what’s widely agreed among users.

What Migaku is

Migaku is a set of tools for learning from content you’d watch or read anyway. The core pieces are:

• A Chrome extension that makes subtitles and web text clickable. It works on Netflix, YouTube, Disney+, Rakuten Viki and Animelon, on ordinary web pages, and on video files you own.
• Migaku Memory, its flashcard system, on the web and in the apps.
• iOS and Android apps for reviewing cards, watching YouTube, browsing the web and using OCR on printed text.
• Courses: Fundamentals for writing systems and pronunciation, Academy for roughly the first 1,500–1,700 words and a few hundred grammar points (it varies by language), and a Kanji course for Japanese.

One subscription covers all 11 languages: Japanese, Mandarin Chinese, Cantonese, Korean, Spanish, English, German, Portuguese, Vietnamese, Italian and French. Details here come from Migaku’s FAQ and language pages.

Migaku pricing in 2026

Migaku sells three plans. These are the US prices Migaku lists at the time of writing (October 2026). Its FAQ notes that prices will periodically increase, so check the current figures before you buy.

There’s no free plan, only a free trial (10 days at the time of writing), and Migaku offers a full refund within 14 days of purchase. Every plan includes all features. EA just gets new ones sooner.

Is lifetime worth it? At $96 a year, Standard would cost $499 after about five years. If you’re sure you’ll still use Migaku in five years, or you’ll want EA, lifetime pays off sooner. Migaku has discounted lifetime during sales, so if you’re considering it, it can pay to wait for one.

What Migaku does well

Clickable subtitles on real streaming sites

This is Migaku’s headline feature, and it’s very good. Turn on the extension, open an episode on Netflix, and every word in the subtitles becomes clickable. Look up a word without pausing for long, then carry on. You can use it on several streaming sites, not just Netflix and YouTube, which sets it apart from many rivals.

Sentence cards that carry the scene

Click to make a card and Migaku captures the word, the full sentence, the audio clip, a screenshot and a definition, with an AI explanation if you want it. When you review later, you hear the actor say the line and see the moment it happened.

That context helps memory. A word tied to a scene you enjoyed is easier to recall than one on a plain card. For anyone already doing sentence mining, Migaku cuts the time per card dramatically compared with building them by hand.

A clear method

Migaku doesn’t just hand you tools. Its FAQ lays out a plan: about 10 new cards a day, Fundamentals in roughly a month, then Academy over about six months, then mostly native content, mining sentences with a single unknown word. For learners who feel lost in “just immerse” advice, that structure is valuable.

Difficulty estimates

Migaku tracks the words you know and estimates how hard a video or web page will be. That helps you pick content you can actually follow, which is the whole point of comprehensible input.

Where Migaku struggles

Language coverage

Eleven languages is enough for many learners and a dealbreaker for others. If you’re learning Russian, Arabic, Dutch, Polish or any of dozens of others, Migaku isn’t an option. LingQ supports 50+ languages and Readlang over 100.

Long-form reading

Migaku reads web pages, but it isn’t designed for reading a novel from start to finish. Its FAQ lists ebook support as “coming soon.” If reading is a big part of your plan, you’ll want a dedicated reader alongside it, or instead. See our guide on importing EPUB books.

Browser and mobile limits

The extension is built for Chrome. Migaku says other Chromium browsers like Brave or Arc may work but aren’t officially supported. There’s no Firefox or Safari extension. On mobile, Migaku’s FAQ says the apps support YouTube, plus the built-in browser and OCR, but not Netflix or the other streaming sites.

Setup and card discipline

Migaku asks you to install an extension and an app and learn a workflow. That’s an hour, not a weekend, but it’s more than tapping a word in a reader. The bigger risk is the card habit. Sentence cards are seductive, and if you make too many, reviews pile up. Migaku’s own suggestion of about 10 a day is wise. Stick to it.

No free tier

Ten days is enough to test the workflow, not to know whether it’ll stick. If you’re on a tight budget, Language Reactor’s free version gives you clickable dual subtitles on Netflix and YouTube without paying.

Migaku vs the alternatives

Sources: LingQ, Readlang, Language Reactor. For a head-to-head, see LingQ vs Migaku.

Who Migaku suits best

Migaku is a great fit if:

• You learn mainly from Netflix, YouTube, anime or drama.
• You’re learning Japanese, Chinese or Korean, where much of its community and course material is focused.
• You like flashcards and want them rich and quick to make.
• You want a structured path from beginner courses into native content.

Look elsewhere if:

• Your language isn’t one of the 11.
• You mainly want to read books and long articles.
• You use Firefox or Safari, or study mostly on your phone with streaming apps.
• You want to try something properly for free first.

A note on Lumo

We’re building Lumo, which isn’t available yet. It’s a reading-first app, so it overlaps with Migaku less than you might think. When it launches, it will handle YouTube videos with an interactive transcript and dual subtitles, plus video or audio files with subtitles, EPUB books and web articles, and the words you save will become FSRS flashcards. It won’t have a browser extension or Netflix support. If streaming sites are where you learn, Migaku is the better tool.

Migaku review: the verdict

Migaku does one thing better than almost anyone: it turns the show you’re watching into a study session without breaking the fun. If that’s how you want to learn, the $10 a month is good value, and the course plus method gives you a clear path.

It’s not an all-in-one tool. Readers who want novels and articles, or learners of languages beyond its 11, will be better served by LingQ, Readlang or a dedicated reader. But for video-first learners, it’s one of the strongest picks out there. Just keep the content comprehensible. A show where you follow the story with a few lookups will teach you more than one where every line needs a click. Our explainer on comprehensible input covers why.

FAQ

Is Migaku worth it?

For learners who watch a lot of native video and like sentence cards, yes, especially in Japanese, Chinese or Korean. For learners who mostly read, or whose language isn’t supported, a reader like LingQ or Readlang is a better use of the money.

How much does Migaku cost?

At the time of writing, US prices are $10 a month or $96 a year for Standard, $15 a month or $144 a year for Early Access, and $499 once for lifetime. Prices change over time, and lifetime has been discounted during sales.

Is there a free version of Migaku?

No. There’s a free trial (10 days at the time of writing), and Migaku offers a full refund within 14 days of purchase. Language Reactor’s free version is the closest free option for clickable subtitles.

Does Migaku work on Firefox or Safari?

Not officially. Migaku’s extension is designed for Chrome, and it says some other Chromium browsers may work but aren’t supported. On iPhone and Android, the Migaku app supports YouTube, plus its own browser and OCR.

What languages does Migaku support?

Eleven: Japanese, Mandarin Chinese, Cantonese, Korean, Spanish, English, German, Portuguese, Vietnamese, Italian and French. One subscription covers all of them.

Here’s how Korean reading actually works, what’s tricky about it, and a stage-by-stage path with real materials at each level.

Why Korean is a good language to read in

Learners coming from Japanese or Chinese often brace for years of character study. Korean doesn’t ask for that. Modern Korean is written almost entirely in Hangul, an alphabet created under King Sejong in 1443 and officially published in 1446. Hanja (Chinese characters) still exist, but you’ll rarely meet them outside academic writing, some newspaper headlines and names.

That means the gap between “I can sound this out” and “I can read this” is mostly vocabulary and grammar, which are exactly the things reading teaches.

Hangul in an afternoon (or a week)

Hangul has 24 basic letters: 14 consonants and 10 vowels. Double consonants (like ㄲ and ㅃ) and combined vowels (like ㅘ and ㅢ) are built from those basics, so the real inventory to learn is small and logical. The letters were designed on purpose: several consonant shapes sketch the position of the mouth or tongue, and related sounds share related shapes.

The scholars who presented the script in the 1446 Hunminjeongeum Haerye famously claimed a clever person could learn it in a morning and a slow one in ten days. For an adult learner, a few focused sessions over a week is a fair expectation. You’ll be slow at first, but you’ll be able to sound out anything.

Syllable blocks

The one thing that looks strange at first is that letters aren’t written in a line. They’re stacked into syllable blocks, each with an initial consonant, a vowel and sometimes a final consonant (called batchim):

• ㅎ + ㅏ + ㄴ = 한
• ㄱ + ㅡ + ㄹ = 글
• Together: 한글, “Hangul”

Once you see that every block is one syllable, reading speeds up quickly. The bigger early challenge is sound change: final consonants shift depending on what follows, so 한국어 (“Korean language”) is pronounced roughly han-gu-geo, not han-guk-eo. This is one reason to read with audio from day one.

The tricky part: particles and verb endings

Korean puts spaces between words, which is a relief after Japanese or Chinese. But a “word” between spaces in Korean, called an eojeol (어절), is often a word plus several grammatical pieces glued on. Korean is agglutinative: it builds meaning by stacking suffixes.

Why this matters for looking words up

A dictionary lists 먹다 (to eat), not 먹었어요. It lists 학교 (school), not 학교에서. If you copy the whole chunk between spaces into a dictionary, you’ll often get nothing, or the wrong thing. To find the entry, you need the lemma, the dictionary form, and getting there is called lemmatization.

For regular verbs that’s easy once you know the pattern: strip the ending and add 다. Irregular verbs make it harder. 들어요 comes from 듣다 (“to listen”), 도와요 from 돕다 (“to help”), and 몰라요 from 모르다 (“to not know”). Early on, these will trip you up constantly.

A few practical tips:

• Learn the top particles first. 은/는, 이/가, 을/를, 에, 에서, 의, 도, 와/과, 하고, 한테. They appear in nearly every sentence, so they’re worth drilling until you strip them off without thinking.
• Learn the common endings as patterns. -아요/어요, -았어요/었어요, -고, -지만, -아서/어서, -(으)면. Each one you learn makes hundreds of verb forms readable.
• Use a learner dictionary. The National Institute of Korean Language’s Korean-English Learners’ Dictionary is free and written for learners, with translations into English and about ten other languages.
• Tap, don’t type. A reader that shows a word’s meaning in context saves you from guessing at lemmas for every lookup.

The good news is that this pain is front-loaded. The set of particles and endings is limited, and after a few months of reading, you’ll recognize the stems inside long words almost automatically.

Sino-Korean vocabulary: your secret multiplier

A large share of Korean vocabulary comes from Chinese. A commonly cited figure, based on the Standard Korean Language Dictionary, puts Sino-Korean words at about 57% of the vocabulary (Wikipedia’s overview of Sino-Korean vocabulary). Everyday speech leans more on native Korean words, but formal writing, news and academic text are heavy with Sino-Korean.

The useful thing is that these words are built from reusable syllables, each with a stable meaning:

• 학 (學, learning): 학교 school, 학생 student, 대학 university, 대학생 university student
• 전 (電, electricity): 전화 telephone, 전기 electricity
• 국 (國, country): 한국 Korea, 외국 foreign country, 외국어 foreign language

Once you notice the pattern, new words stop being random strings and start being combinations of parts you know. If you already speak Japanese or Chinese, you get a real head start here: 도서관 (library) maps onto 図書館 / 圖書館, and many others line up the same way. For more on the Japanese side, see our guide to learning Japanese by reading.

You don’t need to learn Hanja to benefit. Just pay attention when the same syllable shows up across words with related meanings.

TOPIK levels, roughly mapped to reading stages

The TOPIK (Test of Proficiency in Korean) is the standard Korean exam. It comes in two papers: TOPIK I covers Levels 1–2 and TOPIK II covers Levels 3–6, with your level set by your total score. People often map those levels loosely to CEFR (1–2 to A1–A2, 3–4 to B1–B2, 5–6 to C1–C2), but treat that as a rough guide, not an official equivalence.

A level-by-level path to learn Korean by reading

The principle behind every stage is the same: read things you mostly understand. Research on lexical coverage suggests reading becomes comfortable when you know around 98% of the words on the page, and gets hard below 90%. We go into the numbers in the 98% rule. In Korean, count a word as “known” only if you recognize it with its endings, not just its stem.

Stage 0: Hangul and sound (first 1–2 weeks)

• Learn the 24 basic letters, then double consonants and combined vowels.
• Practice reading blocks aloud until sounding out feels automatic, even if slow.
• Learn the basic sound-change rules as you meet them, with audio, not from a chart.
• Start a list of the most common particles.

Don’t wait for Hangul to feel fluent before moving on. Speed comes from reading.

Stage 1: very easy stories (TOPIK 1 territory, A1)

Your goal here is your first few hundred words and the polite -아요/어요 style that most learner material uses.

• Graded stories at the very lowest levels. Short texts with heavy repetition, audio and translations.
• Children’s picture books. Korean picture books use short sentences and lots of repetition, and the pictures carry the meaning. Be aware they often use the casual or storytelling (-다) style, which is a useful second style to recognize.
• Re-read. The same short story three times teaches more than three new ones you half understand.

Stage 2: beginner readers (TOPIK 2, A2)

• Talk To Me In Korean’s Easy Korean Reading for Beginners has 30 short passages on everyday topics like cooking, buses and pets, with English translations, vocabulary lists and audio. TTMIK pitches it at learners around its levels 1–3.
• TTMIK Stories, the same publisher’s graded reading app, lists over 1,000 original stories across 10 levels tied to its curriculum and TOPIK vocabulary, with tap-to-define and native audio. It’s a subscription app; check the App Store listing for current pricing.
• The first volumes of the Darakwon Korean Readers series. These adapt classic Korean tales into modern Korean, with audio and an English translation at the back. The Story of Kongjwi and Patjwi, a Cinderella-like folktale, is labeled A2 by retailers.

Most learners find this the stage where particles stop feeling like noise and start feeling like signposts.

Stage 3: intermediate readers and webtoons (TOPIK 3, B1)

• Talk To Me In Korean’s Korean Folktales in Everyday Conversation (2024) retells 15 well-known folktales for intermediate learners, with translations, vocabulary and grammar notes. Knowing these stories also helps with cultural references in dramas and webtoons.
• Higher Darakwon Korean Readers. The series as a whole runs from A1 to C2, with later volumes such as The Story of Chunhyang (B2) and The Story of Sim Cheong (C1). Read the level on each volume rather than assuming the order.
• Webtoons. Naver Webtoon is the big Korean platform, and its global sister service is Webtoon. Many series let you read a chapter free each day, with coins to unlock more, though access varies by series. Pick slice-of-life, school or romance stories: short speech bubbles, everyday vocabulary, and pictures that carry the plot. Avoid fantasy and historical series at first, which bring dense, specialized vocabulary.

Webtoons are also where you’ll meet real casual speech: 반말, dropped particles and slang. That’s a feature, not a bug, but expect it to slow you down for a while.

If you’re weighing learner books against native material, our post on graded readers vs native content walks through when to switch.

Stage 4: simplified news and longer texts (TOPIK 4, B2)

This is where Sino-Korean vocabulary jumps, and where it pays off.

• Simplified news. Apps such as Todaii: Easy Korean rewrite news stories in simpler Korean. Third-party options like this come and go, so check what’s current.
• Native news on familiar topics. Start with stories you already know from English news. Sports, weather and tech pieces are easier than politics. KBS and other broadcasters publish articles alongside video, so you can read and listen to the same story.
• YouTube with Korean subtitles. Vlogs and interviews on topics you follow give you reading and listening practice together. See how to learn a language from YouTube for a workflow.
• Read narrowly. Stick with one webtoon, one news topic or one YouTuber for weeks. Vocabulary repeats, and every episode gets easier.

Stage 5: native books and long-form (TOPIK 5–6, C1+)

• Contemporary novels and essay collections. Start with ones you’ve read in translation, or genres you know well.
• Long webtoon series you can binge, which give the repetition of a novel with visual support.
• Opinion writing and longer feature articles, where the grammar gets more formal and the written (-다) style dominates.

At this point, most of the work is volume. Keep a steady habit, keep looking up only what blocks understanding, and let the rest come from repetition.

Common mistakes when reading Korean

• Staying on romanization. Drop it within the first week. Romanization hides the syllable structure and makes sound changes harder to learn.
• Treating every eojeol as a new word. 먹어요, 먹었어요 and 먹고 are one word in three forms. Recognize the stem and the ending separately.
• Ignoring speech levels. Learner books favor the polite -아요/어요 style. Webtoons use casual speech, and books and news use the -다 style. Expect a few weeks of adjustment each time you switch.
• Jumping to native material too early. A webtoon you understand at 80% is a slog. A graded story at 95% is reading.

Where Lumo fits

Lumo, a reading and listening app we’re building (coming soon), is designed for this kind of path. Lumo Stories include short original graded stories in Korean from A1 to B1, so you can start at Stage 1 and keep going. When you’re ready, you’ll be able to paste in articles, import EPUB books or bring in a YouTube video with an interactive transcript, tap any word for its meaning in context, and hear it read aloud. Every text gets a difficulty label based on how many of its words you already know, so you can see whether a webtoon script or news story is “Just right” or still “Hard” before you start.

FAQ

Can you learn Korean just by reading?

Reading can carry a huge share of the load, especially for vocabulary and grammar recognition. But Korean pronunciation has sound changes the spelling doesn’t show, so pair reading with audio from the start. Most learners also add some speaking practice once they have a solid base of words.

How long does it take to learn to read Hangul?

Most adults can learn the 24 basic letters and sound out syllable blocks within a few days to a week of short daily sessions. Reading quickly and understanding what you read takes much longer, and that comes from vocabulary and practice, not more alphabet study.

What should a beginner read in Korean?

Start with graded stories and beginner readers written for learners, such as TTMIK’s Easy Korean Reading for Beginners or the lowest levels of a graded app. Children’s picture books work too. Save webtoons until you have a few hundred words and know the common particles.

Are webtoons good for learning Korean?

Yes, once you’re past the absolute beginner stage. Slice-of-life webtoons have short sentences, everyday vocabulary and pictures that carry the meaning. They use casual speech and slang, so expect a learning curve if you’ve only read polite textbook Korean.

Do I need to learn Hanja to read Korean?

No. Modern Korean is written in Hangul, and you can read news and novels without Hanja. Noticing Sino-Korean roots like 학 (learning) or 국 (country) does help you guess new words, and learners who know Chinese or Japanese have a head start there.

Done badly, it becomes a review pile that eats the time you meant to spend reading. Here’s how to do it well.

What sentence mining is

The idea spread through online language-learning communities in the 2000s. It was popularized by Khatzumoto’s All Japanese All The Time, and later by communities built around comprehensible-input methods such as Refold. The basic loop is simple:

1. Read or watch something in your target language.
2. Notice a sentence where you understand almost everything except one word or phrase.
3. Save that sentence with the new item marked.
4. Review it later with a spaced-repetition scheduler, so it comes back just before you’d forget it.
5. Meet it again in your reading, which locks it in.

The key constraint is in step 2. A good mined sentence has exactly one unknown thing in it. This is often called an i+1 sentence, borrowing Stephen Krashen’s shorthand for input just beyond your current level (more on that in comprehensible input, explained).

Why sentences beat isolated words

You could just save the word and its translation. Here’s why context is worth the extra effort.

Context carries meaning a translation can’t

A dictionary tells you quedar can mean “to remain,” “to be located,” “to fit,” “to arrange to meet” and more. The sentence ¿Quedamos el viernes? tells you which one you met, and how it’s used. Over time you collect the uses that matter.

Repetition in context is what builds a word

Research on vocabulary learning keeps finding that words are learned through repeated encounters, and that one meeting rarely does it. Stuart Webb’s experiments (”The effects of repetition on vocabulary knowledge,” Applied Linguistics 28(1), 2007) found that learners’ knowledge of new words grew on several measures as encounters went up from one to ten. A meta-analysis by Takumi Uchihara, Stuart Webb and Akifumi Yanagisawa (”The effects of repetition on incidental vocabulary learning,” Language Learning 69(3), 2019) confirmed that repetition helps, though it isn’t the only factor.

Flashcards give you some of those encounters on a schedule. Reading gives you the rest. A mined sentence is a bridge between them: it’s a review that still looks like real language.

Retrieval practice and spacing do the rest

Two of the most robust findings in learning research apply here. Retrieval practice: actively recalling something strengthens memory more than rereading it (Roediger & Karpicke, Psychological Science, 2006). And the spacing effect: reviews spread out over time beat reviews crammed together (Cepeda and colleagues’ meta-analysis, Psychological Bulletin, 2006). A spaced-repetition app does both. Our FSRS explainer covers how a modern scheduler picks the timing.

How to pick a good sentence to mine

Not every sentence with a new word is worth a card. Run each one through a few quick checks.

The checklist

• One unknown. You should understand every other word. Two unknowns make a confusing card. Three make a card you’ll fail forever.
• Short. Under about 15 words is ideal. Long sentences are slow to review and the target gets lost.
• Useful word. Prefer words you’ve seen before or expect to see again. A word that has come up three times this month is a better pick than a rare one from a poem.
• Clear context. The sentence should make the meaning guessable, or at least memorable. “He took the ___” teaches nothing. “He took the umbrella because it was raining” does.
• Natural language. Real sentences from things you enjoyed beat textbook examples. You’ll remember the scene.

What to skip

• Names, places and one-off technical terms.
• Words you can already guess correctly every time from context.
• Sentences that only make sense with the paragraph before them.
• Anything you saved just because it was unknown, not because you’ll need it.

How to build the card

There are two common formats. Both work. Pick one and stick with it.

A few details make cards much better:

• Highlight the target word so you know what you’re being tested on.
• Write the meaning in context, not every dictionary sense. “Arrange to meet (here)” beats a list of eight.
• Add audio if you can. Hearing the sentence connects the written word to its sound, which helps listening later.
• Keep the source. A note like “from chapter 3 of the book” brings back the scene, and the scene brings back the word.
• Grade on the target word only. If you understood the highlighted word, it’s a pass, even if you stumbled elsewhere in the sentence.

Some learners also add an image or a screenshot from the video. That helps if it’s quick. It isn’t worth five minutes per card.

How many sentences to mine per day

This is where most people go wrong. Every new card adds future reviews. Mine 50 sentences in an enthusiastic weekend and you’ll be paying for it for months.

As a rough guide:

• Beginners (under about 2,000 words): 5–10 new cards a day. Many common words will come back in your reading anyway.
• Intermediate learners: 10–20 a day is sustainable for most people with 20–30 minutes of review.
• Advanced learners: fewer, more carefully chosen cards. At this stage most unknown words are rare, and reading volume matters more than any one card.

A useful rule: your daily reviews should take less time than your reading. If flashcards start crowding out the input they were supposed to support, add fewer new cards for a while.

Common mistakes

1. Mining everything. If every unknown word becomes a card, your deck fills with words you’ll rarely see again. Be picky.
2. Mining from material that’s too hard. If a page has thirty unknown words, it won’t have many one-unknown sentences, and the ones it has will be shaky. Mine from text you mostly understand: around 95% coverage or above, the zone in the 98% rule.
3. Letting reviews pile up. A backlog of 400 due cards feels terrible and leads people to quit. Missing a day is fine. Then cut new cards until you’re caught up.
4. Treating the deck as the method. Mining supports reading. It doesn’t replace it. You’ll learn many more words from meeting them in stories than from cards.
5. Grading generously. Pressing “Good” when you half-remembered teaches the scheduler that the card is easier than it is, so you’ll see it too rarely.

Tools for sentence mining

The classic setup is manual: copy a sentence into Anki, add the meaning, and let Anki’s scheduler handle the rest. It’s flexible, free on desktop and Android, and fully under your control, but making each card by hand is slow.

Many tools now automate the capture step. Migaku can make a card from a word in a subtitle or web page with a keystroke, including the sentence, audio and a screenshot, and send it to its own flashcard app or to Anki. Language Reactor lets you save words and phrases from subtitles with their context and export them. These are especially good if most of your input is video.

If most of your input is reading, a reader with built-in saving keeps the whole loop in one place.

How Lumo handles it

Lumo, the reading app we’re building, isn’t available yet, but this is how it will work. Tapping a new word in any text saves it with the sentence it came from, and saved words become flashcards automatically, scheduled with FSRS. You can also save multi-word phrases. The words you save are highlighted when they turn up again in anything else you read, so each new encounter is easy to spot. It doesn’t replace careful card-building for people who love it, but it takes the copying and pasting out of the loop.

A simple daily routine

1. Read first, 20–30 minutes of something at a comfortable level.
2. Save as you go, but only sentences that pass the one-unknown test.
3. Review after, 10–15 minutes of due cards.
4. Watch your ratio. If reviews take longer than reading, mine less for a week.

That’s the whole method. The sentences come from your reading, the reviews keep them alive, and the reading brings them back again.

FAQ

What is sentence mining in language learning?

Sentence mining is collecting real sentences from books, articles or videos, each with one new word or structure, and reviewing them as flashcards with spaced repetition. It teaches words together with their context, which makes them easier to remember and use.

Is sentence mining effective?

It builds on well-supported findings: words are learned through repeated encounters in context, and retrieval practice with spaced reviews improves memory. It works best alongside lots of reading or listening, not as a replacement for it.

How many sentences should I mine per day?

Most learners do well with 5–20 new cards a day, depending on level and available time. Watch your review load: if reviews start taking longer than your reading, cut back on new cards.

Should I mine sentences or individual words?

Sentences usually make better cards because the context shows how the word is used and helps you remember it. A word-only card is faster to review, so some learners use word cards for very common, concrete words and sentence cards for everything else.

What is an i+1 sentence?

It’s a sentence where you understand everything except one new word or grammar point. The term borrows Krashen’s “i+1” idea of input just beyond your level. These are the best sentences to mine, because the context makes the new item clear.

Characters are not words

In Spanish or French, a word is a string of letters between two spaces. In Chinese there are no letters and no spaces. The unit you see is the character (字, zì), and most words (词, cí) are made of one, two or more characters.

Some examples:

• 人 (rén) is “person” on its own.
• 电 (diàn, electricity) plus 脑 (nǎo, brain) gives 电脑, “computer.”
• 东 (dōng, east) plus 西 (xī, west) gives 东西, “thing.” Neither half tells you that.

This has two practical consequences.

First, knowing a character is not the same as knowing every word it’s in. You can recognize both 东 and 西 and still not know 东西. Character knowledge gives you a head start on new words, often a big one, but it’s not the whole job.

Second, coverage figures for characters look better than coverage figures for words. When you hear that a few thousand characters cover nearly all of a text, that’s true, and it’s encouraging. But your word coverage of the same page will be lower, sometimes much lower. Keep both numbers in mind.

How many characters you need

The best-known frequency data comes from Jun Da’s Modern Chinese Character Frequency List (Middle Tennessee State University), which ranks characters by how often they appear in a large corpus of modern Chinese. Its cumulative figures show how steep the curve is:

Two things stand out. The first few hundred characters pay off enormously: learn 500 and three of every four characters on a page are familiar. And then the returns slow down. Going from 2,000 to 2,500 characters buys you about one and a half percentage points.

Why does that last point and a half matter? Because, as we explain in the 98% rule, comfortable unassisted reading tends to need around 98% coverage of words. At 95%, there’s still an unknown item every couple of lines. In Chinese, the rarer characters are often the ones carrying the meaning: names, places, technical terms and the vivid verbs in a story.

So a realistic target for reading native material with occasional lookups is around 2,500 to 3,000 characters and several thousand words. That’s a long road, but you don’t wait until the end to start reading. You read your way there.

What HSK levels mean for readers

HSK (汉语水平考试) is the standard proficiency test for Mandarin. It is in the middle of a big change.

The older system, often called HSK 2.0, has six levels, topping out at 5,000 words at HSK 6. In 2021 China’s Ministry of Education and State Language Commission issued a new framework, the Chinese Proficiency Grading Standards for International Chinese Language Education, usually called HSK 3.0. It has nine levels in three bands: elementary (1–3), intermediate (4–6) and advanced (7–9). Levels 7–9 are tested together in a single exam.

The 2021 standard set cumulative targets like these (summary on Wikipedia):

An official HSK 3.0 exam syllabus, released in November 2025, builds on that standard but lowers the vocabulary requirements at the beginner and intermediate levels. A global trial exam was held in January 2026, and at the time of writing the new exams are still being phased in. Check Chinese Testing International for current dates and lists before you plan around a specific exam.

For reading purposes, you don’t need to care much about the exact version. What matters is the shape: by the top of the intermediate band you know around 1,800 characters, which by Jun Da’s figures already covers well over 90% of running text. That’s the point where native material starts to open up.

Pinyin: a ladder, not a floor

Pinyin is the official romanization of Mandarin. It shows you how a character sounds, tones included: 你好 is nǐ hǎo. Most learner materials print it above the characters, and most reading apps let you show or hide it.

Pinyin is essential at the start. You need it to learn pronunciation, to look words up, and to type on a phone. The risk is that it becomes the thing you actually read. If pinyin is visible, your eyes drift to it, because it’s easier. You can read a whole story and barely look at a character.

A sensible pattern:

1. First weeks: read with pinyin on. Learn the sounds and tones properly.
2. Once you know a couple of hundred characters: turn pinyin off by default and reveal it only for words you don’t know.
3. From then on: treat pinyin as a lookup, like a dictionary entry, not as part of the text.

If you find you can read a passage with pinyin but not without it, that’s useful information. It means your listening vocabulary is ahead of your character knowledge, and the fix is more character reading at an easier level.

No spaces: word segmentation

Because written Chinese has no spaces, the reader has to work out where one word ends and the next begins. Native readers do it without thinking. Learners have to build the habit.

Take 我们在研究生命的起源 (“we are studying the origin of life”). Should you read 研究生 (yánjiūshēng, graduate student) or 研究 + 生命 (study + life)? Context settles it, but only if you know enough of the surrounding words.

Software has the same problem. Pop-up dictionaries and reading apps use segmentation algorithms to guess word boundaries, and they’re good but not perfect. When a lookup gives you something that makes no sense, try selecting one character fewer or more. And when you save words for review, save the word, not just the character: 起源 is more useful to know than 起 and 源 separately.

The same issue affects known-word counts. Apps that track your vocabulary in Chinese are counting whatever their segmenter decides is a word, so numbers aren’t directly comparable between tools. We go into this in what your known-words count really means.

Simplified or traditional?

Mainland China and Singapore use simplified characters. Taiwan, Hong Kong and Macau use traditional ones. Simplified characters often have fewer strokes (书 versus 書, “book”), but many common characters are identical in both systems.

Pick the one that matches where you want to go, the people you’ll talk to or the media you want to read. If you have no preference, simplified has more learner material, including most graded readers. Learning the second script later is far easier than learning the first: once you can read one comfortably, the other is a matter of weeks or months of exposure, not years.

Graded readers and reading apps

Graded readers are books written with a limited set of characters and words, so you can read at high coverage from early on. They matter more in Chinese than in most languages, because the gap between beginner and native text is so wide. (For the general case, see graded readers vs native content.)

Mandarin Companion

Mandarin Companion publishes graded novels, many adapted from Western classics and retold in a Chinese setting, plus original stories. Its levels are defined by unique characters: Breakthrough (about 150 characters), Level 1 (about 300) and Level 2 (about 450). New words are introduced gradually with footnotes. Finishing a whole novel at 300 characters is a real confidence boost.

Chinese Breeze

Chinese Breeze (汉语风), from Peking University Press, is graded by vocabulary rather than characters, with levels from 300 words up through 500, 750, 1,100 and beyond, with the full series running to eight levels. The stories are original Chinese fiction, and newer editions include audio. It’s a good next step after Mandarin Companion because it goes further up the ladder.

Du Chinese

Du Chinese is a reading app with graded lessons and serialized stories across six levels, from Newbie to Master, with native-speaker audio, a tap-to-look-up dictionary and flashcards. Some lessons are free. At the time of writing, Premium is listed at $14.99 a month, $79.99 for six months or $119.99 a year. It’s polished, the audio is excellent, and if you want a single app built specifically for Chinese reading, it’s one of the strongest choices.

The Chairman’s Bao

The Chairman’s Bao focuses on news-style articles graded across six levels, with a very large back catalog (over 11,000 lessons at the time of writing), human-recorded audio, a pop-up dictionary, flashcards and grammar notes. Sample lessons are free; full access is a paid subscription. It suits learners who want topical, adult content and a steady stream of new material.

Pleco

Pleco is the dictionary almost every Chinese learner ends up with. The free app includes a clipboard reader; the paid document reader add-on lets you open text, EPUB and PDF files (and web pages on iOS) and tap any word to look it up. If you’re ready for native text and want a lookup tool rather than a curated library, it’s hard to beat.

Quick comparison

Learn Chinese by reading: a level-by-level path

Here’s a path that works for most adult learners. The character counts are rough guides, not gates.

0 to 150 characters: sounds and first texts

Learn pinyin and tones properly. Study the first 100–150 high-frequency characters with a spaced-repetition tool. Read tiny dialogues and short stories with pinyin on. Re-read them until they feel easy.

150 to 500 characters: your first books

Read Mandarin Companion’s Breakthrough and Level 1 books, or the easiest stories in an app. Turn pinyin off by default. Aim for volume: a few pages every day, not one heroic session a week. Listen to the audio of what you’ve already read.

500 to 1,200 characters: reading becomes reading

Move to Mandarin Companion Level 2 and the lower Chinese Breeze levels. Use app content at the intermediate levels. You’ll start recognizing words you never studied, because you’ve seen their characters elsewhere. Save words, not characters, and let flashcards handle the review.

1,200 to 2,000 characters: bridging to native text

This is the long middle. Read higher Chinese Breeze levels, graded news, and native material on topics you already know well: a blog about a hobby, a children’s book, a web novel you’ve read in translation. Expect to look things up often. Choose texts where you can still follow the story.

2,000+ characters: native material

Read novels, articles and long-form fiction with a pop-up dictionary to hand. Web novels are a popular choice because they’re long, repetitive in a helpful way and easy to find. Your character coverage is now high; what you’re building is word coverage, and the only way to build that is lots of pages.

If you’re also learning Japanese, a lot of this carries over. Kanji and hanzi share roots, and the reading strategies are similar, as we cover in learning Japanese by reading.

Where Lumo fits

Lumo, a reading and listening app we’re building, includes short original Lumo Stories in Chinese from A1 to B1, alongside seven other languages. Chinese text is split into words you can tap for their meaning and hear spoken, and the words you save keep their sentences and become flashcards scheduled with FSRS. Every text gets a difficulty label based on how many of its words you already know: Easy, Just right, Challenging or Hard.

Lumo isn’t available yet, but it’s coming soon. Whatever you use, the method is the same: read a lot, at a level you can mostly follow, and let the characters pile up.

FAQ

Can you really learn Chinese just by reading?

Reading can carry a large share of the work, especially for vocabulary and character recognition. You’ll still want to learn pinyin and tones deliberately at the start, and to listen a lot, because Chinese pronunciation can’t be picked up from characters. Most successful learners combine reading with audio of the same texts.

How many characters do I need to read Chinese?

About 1,000 characters cover roughly 89% of typical modern text and about 2,500 cover around 98%, based on Jun Da’s frequency list. In practice, 2,500 to 3,000 characters plus several thousand words gets you to comfortable reading of most native material with occasional lookups. You can start reading graded books at 150–300 characters.

Should I learn characters or words first?

Learn words, and let characters come along with them. Studying the most common characters on their own helps early on, but words are what carry meaning in a sentence. When you save something for review, save the whole word with its sentence.

When should I stop using pinyin?

Once you know a couple of hundred characters, turn it off by default and reveal it only for words you don’t know. If pinyin stays visible, you’ll tend to read it instead of the characters. Keep using it for lookups and typing as long as you like.

Is HSK a good guide for reading level?

It’s a useful rough guide, especially the character and word targets, but it’s changing: HSK 3.0 has nine levels and revised word lists, and its new exams are still being phased in at the time of writing. For choosing what to read, your actual coverage of a given text matters more than any level label.

If you use Anki, you have probably seen the FSRS toggle in your deck options. If you are deciding whether to flip it, or just want to know what your flashcard app is doing behind the scenes, read on.

Where FSRS comes from

FSRS was created by Jarrett Ye (known online as LMSherlock) and is maintained by the open-spaced-repetition project on GitHub. It grew out of research he did at MaiMemo, a Chinese vocabulary app, which produced two peer-reviewed papers:

• Ye, Su & Cao (2022), “A Stochastic Shortest Path Algorithm for Optimizing Spaced Repetition Scheduling,” Proceedings of the 28th ACM SIGKDD Conference, pp. 4381–4390 (DOI). It trained a memory model on about 220 million review logs and treated scheduling as an optimization problem: minimize review cost while reaching a memory goal.
• Su, Ye, Nie, Cao & Chen (2023), “Optimizing Spaced Repetition Schedule by Capturing the Dynamics of Memory,” IEEE Transactions on Knowledge and Data Engineering 35(10) (DOI), an extended version of the same work.

FSRS took the memory model from that research, simplified it and made it trainable on one person’s data. Anki added it as a built-in option in version 23.10, and it is now used by a long list of apps. The versions have kept moving: FSRS-4.5 had 17 parameters, FSRS-5 had 19, and FSRS-6, the version most apps run today, has 21. The project’s benchmark already lists experimental successors.

The FSRS algorithm explained: three numbers per card

The heart of FSRS is what its authors call the DSR model, for Difficulty, Stability and Retrievability. It goes back to an idea Piotr Wozniak, the creator of SuperMemo, described decades ago: memory has two components, how strong a memory is and how retrievable it is at this moment.

Retrievability (R)

Retrievability is the probability that you can recall the card right now. It is 100% just after a successful review and declines as days pass. It is the only one of the three numbers that changes on its own, without you doing anything.

Stability (S)

Stability is measured in days. In FSRS it has a precise definition: the number of days it takes for retrievability to fall from 100% to 90%. A card with a stability of 30 days will, on average, be recalled 90% of the time if you see it 30 days after your last review.

Stability is the number you are really trying to grow. Each successful review multiplies it, and a forgotten card has it cut sharply.

Difficulty (D)

Difficulty runs from 1 to 10 and describes how hard it is to make this card’s memory more stable. A high-difficulty card gains less stability from each successful review than an easy one. Difficulty goes up when you press Again or Hard and down when you press Easy.

How the numbers update

You don’t need the formulas to use FSRS, but seeing their shape explains a lot of its behavior. The descriptions below follow the FSRS-6 reference implementation, py-fsrs, and the project’s algorithm notes. The w values are the trainable parameters.

The forgetting curve

FSRS-6 predicts retrievability with a power-law curve:

R(t, S) = (1 + factor * t / S) ^ (-w20)
factor  = 0.9 ^ (-1 / w20) - 1

Here t is the days since your last review. The factor is chosen so that R is exactly 90% when t equals S, which is what makes the definition of stability work. The decay parameter w20 was a fixed constant in FSRS-4.5 and FSRS-5 and became trainable in FSRS-6, so the curve’s shape can now fit the way you forget.

Why a power law and not the textbook exponential? Because your collection is a mix of memories of different strengths, and averaging many exponential curves with different rates gives a curve that falls fast at first and then flattens out. That’s a power law. With the FSRS-6 default decay, a card is at 90% after one stability-length, about 85% after two and about 76% after five. It drops slowly in the long tail.

After a successful review

When you press Hard, Good or Easy, the new stability is the old one multiplied by a growth factor. In words, that factor is larger when:

• difficulty is lower (the formula uses 11 - D),
• stability is lower (stable memories grow more slowly, a saturation effect),
• retrievability was lower at the moment you reviewed (recalling something you were about to forget strengthens it more, the spacing effect in one term).

Hard applies a penalty and Easy a bonus, both trainable.

After a lapse

When you press Again, stability is recalculated from difficulty, the old stability and the retrievability at the time. It usually ends up far lower than before, but not back at zero. Forgetting a card you once knew well leaves more behind than forgetting a card you barely learned.

Difficulty and same-day reviews

Each rating nudges difficulty up or down. The nudge shrinks as difficulty approaches 10 (linear damping), and a small mean reversion pulls difficulty back toward a default value so that cards don’t get stuck as permanently “hard.” FSRS-5 and FSRS-6 also model same-day reviews (learning steps), and FSRS-6 makes those short-term gains smaller for cards that are already stable.

Desired retention: the one setting that matters

Once FSRS knows a card’s stability, scheduling is just the forgetting curve run backwards: find the day when R will fall to your desired retention, and schedule the review then. At 90%, the interval equals the stability. Change the target and the interval changes with it. Here’s what that looks like with the FSRS-6 default decay:

So moving from 90% to 95% cuts every interval by more than half, and 97% cuts it to under a quarter. That’s why the Anki manual warns that workload “increases very quickly” above 90%, and advises keeping the value below 97%. Going the other way, below 80% or so, saves fewer reviews than you would expect, because you forget more cards and every lapse costs extra reviews to relearn.

For most language learners, 85–90% is a sensible range. If reviews are crowding out reading, lower it a little. If you have an exam soon, raise it for that deck.

One more distinction: desired retention is a target, not a measurement. Your true retention will be close to it only if FSRS’s predictions are well calibrated, which is what training the parameters is for.

Training the parameters on your history

FSRS ships with default parameters fitted on a large collection of real users’ reviews. Once you have a decent history (the more reviews, the better the fit), you can fit all 21 parameters to your own data. The optimizer adjusts them to minimize log loss: how far its predicted recall probabilities were from what actually happened when you pressed Again or Good.

This is the real break from older schedulers. FSRS is not a fixed set of rules for how intervals should grow. It’s a model of your memory that gets tested against your own answers.

FSRS vs SM-2

SM-2 is the algorithm Piotr Wozniak published for SuperMemo in the late 1980s. A variant of it was Anki’s scheduler for most of Anki’s life. It is simple:

• A new card is shown after 1 day, then after 6 days.
• After that, each interval is the previous one multiplied by the card’s ease factor, which starts at 2.5.
• Grading a card poorly lowers its ease (to a floor of 1.3) and, on a lapse, resets the interval.

The most practical difference is that SM-2 never says how likely you are to remember a card. It just multiplies intervals. FSRS predicts that probability, so it can schedule for a target you choose and adjust properly when you come back from a week off.

What the benchmark says

The open-spaced-repetition project maintains a public SRS benchmark that tests algorithms on about 10,000 Anki users’ collections, with roughly 350 million reviews used for evaluation in its main table. FSRS-6 reaches a log loss of about 0.346 there, against 0.395 for a baseline that always predicts the user’s average retention. Because SM-2 doesn’t output probabilities at all, the benchmark had to bolt extra formulas onto it before it could be compared. In the early-2025 edition of the table, those SM-2 adaptations scored worse than even the always-predict-the-average baseline, while FSRS-5 scored well ahead of it. The current table no longer lists SM-2.

Keep the benchmark in perspective. It mostly measures how accurately an algorithm predicts recall. Lower log loss means better-calibrated intervals, and in practice that usually means fewer reviews for the same retention. But it’s not a controlled trial of long-term learning outcomes. The benchmark also shows newer models predicting better still, from the experimental FSRS-7 to neural networks with millions of parameters. FSRS is a good, efficient, interpretable model, not the last word.

What FSRS can’t fix

A scheduler decides when you see a card, not whether the card is worth seeing. For language learners, the card itself matters more:

• Cards with context beat bare words. A word on its own is hard to recall and easy to forget. A word in a sentence you understood is easier. See our sentence mining guide for how to build good cards.
• Reading does reviews for free. Every time you meet a word in a story, you get a natural review that no scheduler counts. The more you read, the less your flashcards have to carry. That’s the argument of extensive vs intensive reading.
• Grade honestly. FSRS learns from your button presses. If you press Good when you half-remembered, it will learn that you remember better than you do, and space things too far apart.

How Lumo uses FSRS

Lumo schedules its flashcards with FSRS-6. Words you save while reading become cards automatically, along with the sentence you found them in. Deck packages you import (.apkg or .colpkg files) are scheduled the same way. The point is for reviews to back up your reading, not replace it. Lumo isn’t available yet; it’s coming soon.

FAQ

Is FSRS better than SM-2?

On the open-spaced-repetition benchmark, FSRS predicts recall far more accurately than SM-2. In the edition that still compared them, adapted SM-2 did worse than a baseline that just predicts your average retention. In practice, that usually means fewer reviews for the same retention, or higher retention for the same time. SM-2’s advantage is simplicity: you can predict its intervals in your head.

What desired retention should I use with FSRS?

90% is the default and a sound choice. Many language learners do well at 85–90%, and a lot of reading on the side helps cover the gap. Going above 95% makes your workload grow much faster than your retention.

What do stability and difficulty mean in FSRS?

Stability is the number of days until your chance of recalling a card drops to 90%. Difficulty, on a scale from 1 to 10, describes how slowly a card’s stability grows with each successful review. Retrievability is your current chance of recall, worked out from stability and the time since your last review.

Do I need to optimize FSRS parameters?

The defaults work reasonably well from day one. Optimizing on your own review history makes the predictions better calibrated. It’s worth doing once you have a solid review history, and again every month or so as that history grows.

Who created FSRS?

Jarrett Ye created FSRS, building on spaced-repetition research he published with colleagues at MaiMemo (KDD 2022 and IEEE TKDE 2023). It is developed in the open by the open-spaced-repetition community on GitHub.

It’s also easy to use badly. Putting on a video you understand a fifth of and letting it wash over you feels productive, but it mostly isn’t. The difference between “watching videos in Spanish” and “learning Spanish from videos” comes down to three choices: what you watch, how you use subtitles, and what you do after the first viewing.

1. Choose videos you can almost follow

The single most important decision is picking the right level. You want videos where you can follow the gist and catch a good share of the words, with a few gaps you can close by looking things up. Some things that help:

• Captions in your target language. Human-made captions are best, but decent automatic captions are often usable for clear speech.
• One or two speakers, clear audio. A vlogger talking to the camera is far easier than a noisy group conversation or a film with music under every line.
• Visual support. Cooking, travel, crafts and how-to videos show you what’s being said, which makes more of the speech comprehensible.
• Topics you’d watch anyway. If you’d happily watch it in your own language, you’ll stick with it in your target language.
• Content made for learners. Many creators make videos specifically for learners, often labeled with a level or as “comprehensible input.” They speak a little slower and use visuals deliberately. They’re an excellent bridge to native content.

If you need to look up more than a word or two in every sentence, the video is too hard for now. Save it for later. It will still be there.

2. Use target-language subtitles, not translations

It’s tempting to watch with subtitles in your native language, and it’s not useless: you’ll follow the story. But your attention goes to the text you can read effortlessly, and your ear does very little work.

There’s some neat evidence for this. In a 2009 study, Holger Mitterer and James McQueen had Dutch speakers watch English-language video in an unfamiliar regional accent. Those who watched with English subtitles got better at understanding that accent afterward. Those who watched with Dutch subtitles didn’t, and the paper’s title sums up the finding: foreign subtitles help, but native-language subtitles harm foreign speech perception. Seeing the words in the language you’re hearing helps your brain map sounds to words. Seeing a translation lets it skip that step.

A good rule of thumb:

• Target-language subtitles for most watching.
• A translation of one sentence when you genuinely can’t work it out, then straight back to the original.
• No subtitles for rewatching something you already know well.

3. Watch in three passes

For videos you want to really learn from, try this. It turns a ten-minute video into roughly half an hour of focused practice, and it’s far more effective than watching three new videos once each.

Pass one: for the story

Watch straight through with target-language subtitles. Don’t stop for every unknown word, only for the ones that block your understanding of what’s happening. The goal is to get the gist and enjoy it.

Pass two: line by line

Now go back through more slowly. Replay any line you didn’t catch. Look up the words you skipped. Notice useful phrases, the chunks people actually say, like “the thing is” or “I was about to,” and save a few. This is where most of the learning happens.

Pass three: just listen

Watch again with subtitles off, or play the audio while you’re cooking or walking. You’ll be surprised how much more you hear now that you know what’s being said. If you want to push further, pause after a few lines and repeat them aloud, copying the rhythm and intonation. This is often called shadowing, and it’s a gentle way to start speaking.

4. Make it stick

A single viewing, even a careful one, won’t make new words permanent. A few habits make a big difference:

• Follow creators, not just videos. Watching many videos by the same person is a kind of narrow listening. They reuse the same words, topics and speech patterns, so each video gets easier than the last.
• Rewatch favorites. Coming back to a video a week later is a great way to see how far you’ve come.
• Review the words you saved. A few minutes of flashcards keeps them from fading before you meet them again.
• Balance listening with reading. Reading lets you meet far more words per minute, and listening teaches you how they sound in real speech. They reinforce each other.

Common traps

• Watching content that’s far too hard. It feels like immersion but it’s mostly noise.
• Pausing for every word. If you look up everything, you lose the thread and the fun. Look up what matters.
• Never rewatching. New content is exciting, but rewatching is where comprehension jumps.
• Treating it as all-or-nothing. One short video, done properly, is a good day.

How Lumo helps

Paste a link to a captioned YouTube video into Lumo, and it shows an interactive transcript beside the player. The current line lights up as the video plays, and every word is one tap away from its meaning, so you can look things up without losing your place. Words you look up join the rest of your vocabulary, with the same highlights you’ll see when you read, and they can go straight into your flashcards.

The free plan includes two videos a week, which is plenty for the three-pass method. Pick one you love, watch it properly, and you’ll get more out of it than an evening of passive scrolling.

That’s the short version. The longer version is about the two things that make French different from other languages you might read in: an enormous head start on vocabulary, and a spelling system that hides its pronunciation. Here’s how to use the first and work around the second.

Why reading works so well for French

If your first language is English, you already know thousands of French words. After the Norman conquest of 1066, French was the language of England’s ruling class for centuries, and English absorbed a huge amount of its vocabulary: gouvernement, justice, nation, possible, important, question. A lot of the abstract, formal vocabulary of English came in through French or Latin.

On the page, that’s a gift. A French newspaper article is full of words you can recognize on sight, even at a low level. In speech the same words are much harder to catch, because the sounds have drifted far apart. Gouvernement is easy to read and surprisingly hard to hear.

That makes reading the natural place to build vocabulary in French. You can meet far more words per minute than you can by listening, and you can recognize many of them without a dictionary.

The vocabulary you need

Reading gets comfortable at around 98% coverage, meaning you know about 49 of every 50 words. (We go into the research in our post on the 98% rule.) For general written text in English, Paul Nation’s estimate is around 8,000 to 9,000 word families for that level.

You don’t need all of that to start. A Frequency Dictionary of French by Deryle Lonsdale and Yvon Le Bras (Routledge, 2009) lists the 5,000 most common words from a 23-million-word corpus of spoken and written French. Frequency lists like that show the same pattern in every language: a few thousand words do most of the work. The first thousand or so cover a large share of any page, and every word after that adds a little less.

The practical lesson is to read a lot of easy material early, because that’s where the frequent words repeat most.

The catch: French spelling hides the sound

French spelling is consistent in one direction and unhelpful in the other. Given a word you’ve heard, you can usually guess how it’s spelled. Given a word on the page, you often can’t guess how it sounds unless you’ve learned the rules.

A few examples of what reading alone won’t teach you:

• Silent endings. Ils parlent sounds the same as il parle. The -ent is silent.
• Liaison. Les amis is pronounced with a z sound linking the words, but les haricots isn’t.
• Nasal vowels. Vin, vent and vont differ only in their nasal vowel, and vingt sounds exactly like vin. You have to hear these to learn them.

If you only read, you’ll build a private pronunciation in your head that doesn’t match real French, and it can be stubborn to fix later. The answer isn’t to read less. It’s to make sure a good share of your reading has audio attached: graded readers with audiobooks, articles with recordings, or read-along text-to-speech.

Written French and spoken French are not quite the same language

The other surprise for French readers is how different the written register can be from everyday speech. Three things you’ll notice quickly:

• The passé simple. Novels and fairy tales tell stories in a past tense (il alla, elle fut, ils prirent) that almost nobody uses in conversation. Speech uses the passé composé instead (il est allé). You need to recognize the passé simple to read literature, but you don’t need to produce it.
• Negation. Writing keeps both parts of ne... pas. Speech often drops the ne: je sais pas.
• On for nous. In conversation, on (“one”) very often replaces nous (“we”). Formal writing keeps nous far more often.

None of this is a reason to avoid novels. It’s a reason to mix your diet. If you read only 19th-century classics, you’ll learn a French nobody speaks. If you read only chat messages, you’ll struggle with a newspaper. Read across registers and the differences start to feel natural.

What to read at each level to learn French by reading

Here’s a path that works for most adult learners. The levels are approximate and use the CEFR scale (A1 is beginner, C2 is mastery).

A1–A2: graded readers

Graded readers are books written or adapted with a limited vocabulary. Hachette FLE’s Lire en français facile series, for example, is organized into A1, A2, B1 and B2 levels, with word definitions at the foot of the page and comprehension activities at the back. Several other publishers have similar series, and many include an audio recording.

At this stage, read more than one book per level before moving up. Each book repeats the same core vocabulary, and the repetition is the point. (If you’re wondering how long to stay with learner material, our comparison of graded readers and native content goes into that.)

A2–B1: news written for young readers

Once graded readers feel easy, news for children and teenagers is a great bridge. 1jour1actu is a French news site for children, roughly 8 and up, from the children’s publisher Milan. It explains real current events in short, clear articles.

For listening and reading together, RFI’s Journal en français facile is a daily news bulletin in slower, simpler French, published with a transcript. Read the transcript, listen, then listen again without it.

B1: comics and books you already know

French has one of the world’s great comics traditions, and bandes dessinées are excellent reading practice. The pictures do half the work, the text is mostly dialogue, and you get a lot of everyday spoken French that novels leave out.

Reading a French translation of a book you’ve already read in English works for the same reason: you know the plot, so you can follow the story even when the vocabulary gets thin.

B1–B2: genre fiction

Crime novels, thrillers and romance are underrated for learners. They’re written to be page-turners, the vocabulary is mostly concrete, and the plot pulls you forward when a sentence gets hard. Pick a series and stay with it: the same characters, places and phrasings keep coming back, and each book gets easier than the last.

B2 and beyond: the classics, and where to find them free

French public-domain literature is easy to get legally. Project Gutenberg’s French collection has thousands of free books, from Jules Verne and Alexandre Dumas to Maupassant’s short stories. The French-language Wikisource and the Swiss Bibliothèque numérique romande also offer free, carefully prepared EPUB editions. (Our guide to importing EPUB books covers how to use them.)

Two warnings about classics. They’re heavy on the passé simple, and their vocabulary is older and more literary than modern French. Maupassant’s short stories are a good first step: short, gripping and fairly plain. Hugo and Proust can wait.

For audio, Littérature audio offers more than 10,000 free French audiobooks, mostly of public-domain works, recorded by volunteer readers. Pairing a free text with its free recording is one of the best deals in language learning.

How to read a French text

A good session looks like this:

1. Pick something at the right level. If you’re looking up more than a word or two per sentence, it’s too hard for now. Save it.
2. Read for the story first. Look up only the words that block your understanding. Guess the rest from context and keep going.
3. Use cognates, but check the suspicious ones. Most words that look like English mean what you’d expect. A handful don’t: actuellement means “currently,” librairie means “bookstore,” éventuellement means “possibly” and attendre means “to wait.” If a cognate makes the sentence strange, look it up.
4. Listen to the same text. Read along with audio, or listen afterward. This is where you fix the pronunciation that spelling hides.
5. Re-read. The second time through, more of the text is comprehensible, and words you looked up the first time start to stick.
6. Read narrowly. Stay with one author, series or topic for a while. The vocabulary repeats, and repetition is what turns a looked-up word into a known one.

Don’t stop to study every new verb form or grammar point you meet. Notice it, maybe look it up once, and keep reading. You’ll meet the passé simple of faire so many times in a novel that you’ll know it before you’ve ever looked at a conjugation table.

How much reading does it take?

There’s no fixed answer, but volume matters more than anything else. Fifteen to thirty minutes a day of comfortable reading adds up to millions of words over a few years. A typical novel runs somewhere between 60,000 and 100,000 words, so a reader who manages one short novel a month is meeting French at a scale no course can match.

Consistency beats intensity here. A short daily session is easier to keep up than a long weekly one, and the frequent words get more chances to repeat.

How Lumo fits in

Lumo, the reading and listening app we’re building, is designed for this kind of reading. Every text gets a difficulty label based on how many of its words you already know, so you can tell at a glance whether a French article is Easy, Just right, Challenging or Hard for you. Tapping a word shows its meaning in context and plays it aloud, which helps with all those silent letters, and read-along text-to-speech highlights each word as it’s spoken. Lumo Stories includes short original graded stories in French from A1 to B1, and you can import articles, YouTube videos and DRM-free EPUB books once you’re ready for native material.

Lumo isn’t available yet; it’s coming soon. Until then, the method works with any book and a good dictionary: read a lot, read easy, and listen as you go.

FAQ

Can you learn French just by reading?

Reading alone will give you a large vocabulary and strong grammar intuition, but it won’t teach you pronunciation or how to hold a conversation. French spelling hides too much of the sound. Pair reading with listening to the same texts, and add speaking practice when you’re ready.

What is the best first book to read in French?

For most learners, the best first book is a graded reader at your level, ideally one with audio. If you’d rather start with native material, a French translation of a book you know well, or a comic, is easier than an original French novel because the context carries you.

Should I learn the passé simple before reading French novels?

You only need to recognize it, not use it. Learn the common third-person endings (-a, -it, -ut, -èrent, -irent) and a few frequent irregular forms like fut, eut and fit. You’ll pick up the rest from reading.

Where can I find free French books to read?

Project Gutenberg, French Wikisource and the Bibliothèque numérique romande all offer free public-domain French books, many as EPUB files. For news, try 1jour1actu or RFI’s Journal en français facile. For free audiobooks, Littérature audio has thousands of recordings.

How long does it take to read a novel in French?

Most learners can read contemporary genre fiction with occasional lookups somewhere around B1 to B2, which typically takes a year or two of steady study. Classics come later. The fastest way there is lots of easy reading now.

How many known words you need and what the number actually means are two different questions. Here’s how to read your count without fooling yourself.

What a known words count measures

Reading apps that track vocabulary give every word you meet a status. Each app has its own scheme:

• LingQ marks unseen words blue, words you’re learning yellow at levels 1–4, and known words with no highlight. Its support article defines your known words count as the words you’ve marked Known plus those at level 4 (“Learned”).
• Migaku gives words statuses such as unknown, learning and known as you read and watch.
• Lumo, the app we’re building (coming soon), uses new, unknown, recognized, familiar and known, plus ignored for names and junk.

The count itself is just the number of distinct items in the “known” bucket. What goes into that bucket, and what counts as one item, is where it gets interesting.

How words get into the known bucket

There are two routes:

1. You mark them. You tap a word you already knew, or you graduate it from your flashcards.
2. You read past them. LingQ’s Reader is set by default to mark all remaining blue words as Known when you page. You can switch this off. Lumo will do the same: when you turn the page, new words you didn’t tap become known.

The second route is the one that makes counts climb fast, and it rests on an assumption: if you didn’t look it up, you knew it. Usually that’s true. Sometimes you skipped a word because the sentence made sense without it, or because you were tired. So the count drifts a little above your real vocabulary, more if you skim.

What counts as one “word”?

This is the big one. Most reading apps count each distinct written form as its own word. In Spanish, hablar, hablo, hablas, habla, hablamos, hablé, hablaba, hablaría and hablando might all count separately. You learned one verb and got credit for a handful of words.

Vocabulary research mostly counts word families: a base word plus its inflections and close derivations, so all the forms above plus hablante are one family. Sometimes it counts lemmas (a base form plus its inflections).

The more inflected the language, the bigger the gap. You can see this in LingQ’s own targets.

How many known words LingQ expects at each level

LingQ publishes a table of known-word thresholds for its six levels. The table dates from 2016, and LingQ still links to it. Here are its rows:

Notice that Russian and Korean need roughly twice as many known words as English for the same level, or more. That isn’t because Russians or Koreans need twice the vocabulary. It’s because Russian nouns and adjectives decline through six cases, and Korean attaches particles and verb endings that produce many surface forms per word. The thresholds scale the count to make up for how many forms each language produces.

These levels are LingQ’s own scale. They aren’t CEFR levels, and they aren’t based on a published study. They’re a reasonable rule of thumb from a company that has watched a lot of learners read.

Known words vs research vocabulary sizes

It’s tempting to set your count beside the research: “Nation says I need 8,000–9,000 words to read novels, and I have 15,000 known words, so I’m there.” Not quite.

The research figures, which we cover in how many words you need to be fluent, are in word families. In English, each family has several members, and an app counting forms will rack up many more “words” than families. In a heavily inflected language the ratio is larger still. And the count includes proper nouns, numbers and borrowed words the research handles separately.

So a known words count of 20,000 in Spanish might mean somewhere in the mid-thousands of word families. That’s a solid intermediate vocabulary, but it isn’t 20,000 of anything a researcher would recognize. The exact ratio depends on the language, the texts you read and how the app splits words, so treat any conversion as a rough guess.

Why the count is still worth tracking

None of this makes the number useless. It’s just a different unit. What makes it valuable:

• It goes up with real reading. You can’t raise it much without meeting words in context.
• It’s consistent with itself. Your count this month against last month, in the same language and app, is a fair comparison.
• It predicts what you can read. What matters is what share of the words in this text you already know. That’s lexical coverage, and it’s what decides whether reading feels easy or exhausting.

That last point is the most useful thing a known words list can do. A count of 12,000 is an abstraction. “You know 96% of the words in this article” tells you what tonight’s reading will feel like.

Ways the count misleads

A few patterns to watch for:

• Fast paging inflates it. If you page through texts you barely followed, untapped words become known anyway. If the count jumps after a hard text, be skeptical.
• Names and noise inflate it. Character names, place names, numbers and typos all count as words unless you mark them ignored.
• Recognition isn’t recall. A known word is one you understand when you read it. You might not be able to produce it in conversation.
• Comparisons across languages don’t work. 10,000 known words in Russian and 10,000 in English are very different achievements, as the LingQ table shows.
• Comparisons across apps don’t work either. Apps split and group words differently, especially in Japanese and Chinese, where text has no spaces and the app has to decide where words begin and end.

Making your count more honest

If you want the number to track your real vocabulary more closely:

1. Tap anything you’re unsure of. Tapping costs a second. A false “known” stays hidden for good.
2. Mark names as ignored so they don’t pad the count.
3. Don’t page through text you didn’t read. If you skimmed, don’t let the app credit you with every word on the page.
4. Watch coverage, not just the total. Watch how texts you used to find hard start to feel easier.

How Lumo handles known words

Lumo isn’t available yet, but here’s how it will work. It tracks every word you meet through the levels new, unknown, recognized, familiar and known. Turning a page marks the new words you didn’t tap as known, and words you save become flashcards with the sentence you found them in. Instead of making the raw count the headline, Lumo uses your known words to label every text by coverage: Easy at 98% and above, Just right at 90–98%, Challenging at 80–90% and Hard below that. The count is in the background. What you see is whether the next thing you read will be comfortable.

FAQ

What does known words mean on LingQ?

It’s the number of distinct words you’ve marked Known or moved to level 4 (“Learned”). By default, LingQ also marks the remaining blue words on a page as Known when you page forward. It counts distinct word forms, so different forms of the same verb count separately.

How many known words do I need to be fluent?

There’s no single number, and a known words count is the wrong unit to answer it with. On LingQ’s own scale, Advanced 1 starts at 17,000 known words in English and 22,000 in Spanish or French. The research figures are in word families, a different and smaller unit, and we cover them in how many words you need to be fluent.

Why does LingQ need more known words for Russian or Korean?

Russian and Korean produce many more distinct forms per word through cases, particles and verb endings. Since apps count forms, you have to “know” more of them to reach the same practical level. LingQ’s thresholds for those languages are roughly double the English ones, or more, for that reason.

Is a known words count accurate?

It’s a reasonable estimate that runs slightly high. Words you paged past without really reading, names and numbers all inflate it. It’s most useful for tracking your own progress over time in one language and one app.

Can I compare my known words count across apps?

Not reliably. Apps split text into words differently, especially in Japanese and Chinese, and handle names, numbers and inflected forms in their own ways. Compare your count only with your own past count in the same app.

So the usual advice holds up: target-language audio, target-language subtitles, and a translation only when you really need one. The rest of this post covers the evidence, the tools that make it practical on Netflix and YouTube, and a routine for getting more out of an episode than just the plot.

First, some terms

People use “subtitles” for everything, but researchers separate them. The difference matters here.

Most of the studies below compare captions with no on-screen text, and some compare captions with L1 or dual subtitles.

What the research says about subtitles

Captions help listening and vocabulary

The best-known synthesis is a 2013 meta-analysis by Maribel Montero Perez, Wim Van Den Noortgate and Piet Desmet in the journal System (doi:10.1016/j.system.2013.07.013). It pulled together 18 studies, with separate analyses for listening comprehension (15 studies) and vocabulary learning (10 studies). Captions had a large effect on both. The type of test mattered too: how much captions seemed to help depended partly on how learning was measured.

A larger and more recent meta-analysis by Satsuki Kurokawa, Aung Myo Hein and Takumi Uchihara in Language Learning (2025, doi:10.1111/lang.12697) focused on incidental vocabulary learning: picking up words while watching for meaning rather than studying them. Across 49 experiments, viewers who watched with captions learned more words than those who watched without, with a medium overall effect (g = 0.56). The effect varied with learners’ level, the intended audience of the video, and how the studies were run.

“Medium” is a fair summary. Captions don’t turn a TV show into a vocabulary course. Most studies find modest word gains from any single viewing. But they reliably beat watching the same video without text.

Native-language subtitles can get in the way of your ear

The most quoted study on L1 subtitles is by Holger Mitterer and James McQueen (2009, PLoS ONE, open access). Dutch speakers watched about 25 minutes of English-language video in an unfamiliar Scottish or Australian accent, with English subtitles, Dutch subtitles or none. Afterward they repeated back audio clips in the accent.

English subtitles improved their ability to understand the accent. With Dutch subtitles, they were worse at repeating new clips they hadn’t heard before than people who had watched with no subtitles at all. The authors’ explanation: seeing the words in the language you’re hearing tells your brain which words, and so which sounds, are being spoken. Seeing a translation competes with that.

That was a study of accent adaptation by fairly advanced listeners, not of beginners learning vocabulary. But it fits the broader pattern. In a 2016 study in System, Elke Peters, Eva Heynen and Eva Puimège found that captions helped learners of English in Flanders learn the form of new words better than L1 subtitles did, and L1 subtitles didn’t produce better results for word meaning either (doi:10.1016/j.system.2016.10.002). A 2022 scoping review and meta-analysis by Barry Lee Reynolds and colleagues in Systems found that every type of on-screen text had a positive effect on vocabulary, with intralingual captions showing the largest (open access).

Dual subtitles: better for meaning, worse for form

Dual subtitles are what most learners reach for on Netflix, so the evidence here is worth knowing in detail. Andi Wang and Ana Pellicer-Sánchez (2022, Language Learning, doi:10.1111/lang.12495) had 112 intermediate to advanced Chinese learners of English watch a documentary with dual subtitles, captions, Chinese subtitles or no subtitles, while an eye tracker recorded where they looked.

The results split neatly:

• Dual subtitles beat captions for recognizing what a new word meant, and beat L1 subtitles for recalling its meaning.
• Dual subtitles were worse than captions for recognizing the word’s form: what it looked like.
• Viewers with dual subtitles spent more time reading the Chinese translation than the English word.
• The time spent looking at the English word, not the translation, predicted how much they learned.

That last point is the key. Dual subtitles work when you actually look at the target-language line. If your eyes drift to the translation every time, you’re mostly watching with L1 subtitles.

What this means in practice

The best tools for Netflix subtitles

At the time of writing, Netflix’s own player shows one subtitle track at a time, and which subtitle languages are available varies by title and region. Dual subtitles and click-to-translate come from browser extensions and separate services.

Language Reactor

Language Reactor, formerly Language Learning with Netflix, is the best-known option. It’s a browser extension for desktop Chrome that adds dual subtitles, a pop-up dictionary and line-by-line playback controls to Netflix and YouTube. It can also handle imported text from books and websites. The Chrome Web Store listing shows about two million users and in-app purchases for its paid tier.

It’s the best pick if your main goal is Netflix specifically, you watch on a laptop, and you want something free to start. The trade-off is that it lives in a desktop browser, not on your TV or phone.

Migaku

Migaku’s Chrome extension works with Netflix, YouTube, Disney+, Rakuten Viki and Animelon, according to its Chrome Web Store listing. Its focus is sentence mining: making flashcards from subtitle lines, with screenshots and audio attached. It suits learners who want to build cards as they watch, and it requires a Migaku account (there’s a free trial, but no free plan). (We go deeper in our Migaku review.)

Lingopie

Lingopie takes a different approach. It’s mainly a subscription streaming service with its own catalog of shows, with dual subtitles and clickable words built in, in about 15 languages at the time of writing. Its subscription also includes a Chrome extension that works with a selection of Netflix titles. It’s a good fit if you’d rather pay one service than manage several extensions, but you’re mostly limited to what’s in its library.

Smaller extensions

There are also many small, free extensions that do only one thing, like adding a second subtitle track to Netflix. They can be handy, but they come and go and tend to break when Netflix updates its player, so check recent reviews before installing one.

How to learn a language with Netflix subtitles, not just watch TV

Watching a show with captions is good. A little structure makes it much better.

Pick shows made in your target language

Original productions beat dubbed ones. A show filmed in the language has natural speech, matching lip movements and subtitles that were written from the original script. Netflix carries many well-known non-English series, such as Lupin (French), La casa de papel (Spanish), Dark (German) and Squid Game (Korean).

If you do watch dubbed content, be aware that target-language subtitles on a dubbed show are often translated separately from the dub script, so the text and the audio won’t match. That’s confusing for learners.

Use episodes as narrow listening

A TV series is perfect for what researchers call narrow input: lots of material on the same topics with the same voices. By episode five you know the characters’ names, their speech habits and the show’s recurring vocabulary, and each episode gets easier. Stick with one series for a while instead of sampling.

Watch in passes

For one scene or episode a week, try three passes:

1. For the story. Captions on, minimal pausing.
2. Line by line. Replay what you missed, look up words, save useful phrases.
3. Just listening. Subtitles off, or audio only while you walk or cook.

We describe this routine in more detail in our guide to learning a language from YouTube, and it works the same way for Netflix.

Know your level

If you need to pause every line, the show is too hard for now. A comedy or drama with fast, overlapping dialogue is much harder than a documentary with a narrator. Content made for children or teenagers, cooking shows and reality TV are often easier starting points.

What about YouTube and your own videos?

YouTube is the other big source, and often a better one for learners: shorter videos, more one-on-one speech, and huge amounts of learner-oriented content with captions.

This is where Lumo comes in. Lumo has no Netflix integration, and we don’t plan to pretend otherwise. What it does is open YouTube videos with an interactive transcript and dual subtitles beside the player, where every word is one tap from its meaning and saved words become flashcards. You can also import your own video or audio files with SRT or VTT subtitle files, so a recording you own, with captions, gets the same treatment. Lumo isn’t available yet, but it’s coming soon.

Whatever you use, the research points in the same direction: put the target language on screen, look at it, and come back to the same material more than once.

FAQ

Is watching Netflix with subtitles good for learning a language?

Yes, especially with subtitles in the language you’re learning. Meta-analyses find that captions improve listening comprehension and vocabulary learning compared with watching without text. Native-language subtitles help you follow the story but do much less for listening.

Should I use English subtitles or target-language subtitles?

Use target-language subtitles whenever you can follow along. Research such as Mitterer and McQueen (2009) suggests native-language subtitles can interfere with tuning your ear to the new language. Switch to English for a single line when you’re stuck, then go back.

Do dual subtitles help or hurt?

Both. A 2022 eye-tracking study found that dual subtitles helped learners understand the meaning of new words, but they learned the words’ written form less well than with captions alone, and they spent more time reading the translation. Dual subtitles help most when you keep your eyes on the target-language line.

Can Netflix show two subtitles at once?

Not natively, at the time of writing. Netflix’s player shows one subtitle track at a time. Browser extensions such as Language Reactor add a second track on desktop.

Can you become fluent just by watching Netflix?

Watching alone builds listening comprehension and vocabulary, but it’s slow on its own and doesn’t train speaking. Most learners get further combining shows with reading, a bit of vocabulary review and, eventually, conversation.

Here’s what each one is, what the evidence says, and how to fit them into a week.

What extensive reading is

The standard description comes from Richard Day and Julian Bamford, whose book Extensive Reading in the Second Language Classroom (Cambridge University Press, 1998) and article ”Top ten principles for teaching extensive reading” (Reading in a Foreign Language 14(2), 2002) shaped how the field talks about it. Their principles, condensed:

• The material is easy.
• There’s a wide variety of material on many topics.
• Learners choose what they want to read, and can stop reading anything they don’t enjoy.
• Learners read as much as possible.
• The purpose is pleasure, information or general understanding.
• Reading is its own reward, with no comprehension quizzes required.
• Reading is usually fast rather than slow.
• Reading is individual and silent.

The first principle is the one people skip. “Easy” means really easy. Paul Nation and Rob Waring, in Teaching Extensive Reading in Another Language (Routledge, 2020), argue that extensive reading should be done at about 98% coverage: no more than one or two unknown words in every hundred. At that level you can read without a dictionary, at something close to your normal reading speed, and actually enjoy it.

What intensive reading is

Intensive reading is the familiar classroom kind. You take a short text, often above your comfortable level, and work through it in detail. You look up unknown words, parse tricky sentences, check the grammar, maybe translate a paragraph, and reread until you understand everything.

Done well, it’s how you notice things you’d otherwise skim past: a verb form you hadn’t met, a set phrase, how a connector works. Done badly, it’s an hour spent on a page of text you’ll never think about again.

Extensive vs intensive reading side by side

What the research says

Extensive reading works, but slowly per word

Meta-analyses have found positive effects for extensive reading on reading comprehension and reading speed. Eun-Young Jeon and Richard Day pooled 49 studies with 5,919 participants in ”The effectiveness of ER on reading proficiency: A meta-analysis” (Reading in a Foreign Language 28(2), 2016) and found a small-to-medium positive effect on reading proficiency. Effects were larger for adults than for children and teenagers, and larger in foreign-language settings than in second-language ones.

The catch is that you learn surprisingly little from any single encounter with a word. Rob Waring and Misako Takaki tested this directly (”At what rate do learners learn and retain new vocabulary from reading a graded reader?“, Reading in a Foreign Language 15(2), 2003). Fifteen intermediate learners read a 400-headword graded reader in which 25 target words, met anywhere from once to 18 times, had their spelling changed so nobody could know them in advance. Form recognition right after reading was respectable, but after three months, the data suggested only about one of the 25 words would be remembered by meaning. Words met more often were learned better and forgotten less, and none of the words met fewer than eight times was predicted to be remembered by meaning.

Their conclusion wasn’t that extensive reading fails. It was that you need a massive amount of it. Each encounter adds a little. The gains come from volume, from reading enough that the same words come back again and again.

Intensive work fills the gaps

This is why most researchers don’t argue for extensive reading alone. Nation’s four strands framework (2007, Innovation in Language Learning and Teaching) splits a balanced course into roughly equal parts:

1. Meaning-focused input: reading and listening for meaning, where extensive reading lives.
2. Meaning-focused output: speaking and writing to communicate.
3. Language-focused learning: deliberate study of words, grammar and pronunciation, where intensive reading and flashcards live.
4. Fluency development: working with very easy, familiar material to get faster.

Notice that extensive reading covers two strands. Reading easy material for meaning is input, and reading very easy material fast is fluency practice. Intensive reading belongs to language-focused learning, which on this view should take about a quarter of your time. Not zero, and not most of it.

Deliberate study speeds up what reading does slowly. Learning a word with a flashcard after meeting it in a story, and then meeting it again in the next story, is much more efficient than either on its own.

The trap most learners fall into

Self-directed learners, especially those using reading apps with a dictionary built in, often end up doing intensive reading by accident. You import a native article or novel that’s too hard, look up twenty words a page, crawl through two pages in half an hour, and call it reading.

That has value. But it’s not extensive reading, and it doesn’t build the same things. At 90% coverage, you’re decoding one unknown word per line. You won’t read fast, you won’t read much, and you probably won’t enjoy it enough to keep going every day.

The fix isn’t to stop reading native material. It’s to have two piles:

• an easy pile you read a lot of, quickly, mostly without looking anything up, and
• a study pile of short, harder texts you work through properly.

If you’re early on, the easy pile will be graded readers and learner stories. Our post on graded readers vs native content covers when to switch. Later, it’ll be native material on topics you know well.

How much of each should you do?

There’s no single correct ratio, but a few useful rules of thumb:

• Most of your reading time should be extensive. If you follow the four strands, intensive reading is one strand out of four, and extensive reading feeds two. A 70/30 or 80/20 split in favor of easy reading is a reasonable target for most learners.
• Volume targets help. A common recommendation in the extensive reading literature (Nation & Wang, 1999; Waring) is about a graded reader a week, more at higher levels. For self-learners reading native text, a daily target in pages or minutes works well.
• Keep intensive sessions short. Twenty focused minutes on one dense passage beats an hour of tired decoding.
• Reuse your intensive texts. After studying a passage, reread it a few days later. What took effort the first time becomes extensive-style reading the second time.

Picking texts for each mode

The simplest test is how many words per page you don’t know.

If you can’t tell at a glance, read a page and count the words you’d have to look up. Our post on the 98% rule explains why the difference between 98% and 90% feels so much bigger than it sounds.

Lumo, the reading app we’re building, will do this count for you when it launches. It labels every text by the share of words you already know: Easy (98% and above) for extensive reading, Just right (90–98%) for assisted reading, and Challenging or Hard for intensive work. Lumo Stories, its short graded stories in eight languages, fill the easy pile at A1–B1. And because untapped words become known when you turn the page, texts move from one pile to the other as your vocabulary grows.

A sample week

For an intermediate learner with about 45 minutes a day:

• Most days: 30 minutes of easy reading (a novel, a series of articles on one topic, graded stories) with only occasional lookups, then 10–15 minutes of flashcard review.
• Two or three days a week: swap 15 minutes of easy reading for one intensive session on a short, harder text. Look everything up, save the useful words and phrases, and reread it at the end.
• Once a week: reread something you studied intensively the week before, at normal speed.

The pattern is the point. Lots of easy reading, a little hard reading, and both feeding into each other.

FAQ

What is the difference between extensive and intensive reading?

Extensive reading is reading a lot of easy material quickly for meaning and enjoyment, with few lookups. Intensive reading is working slowly through a short, harder text to understand every word and structure. One builds fluency and volume. The other builds accuracy and noticing.

Is extensive reading better than intensive reading?

Neither replaces the other. Research supports extensive reading for reading speed, comprehension and gradual vocabulary growth, but individual words are learned slowly, so deliberate study helps. Most experts, including Nation, recommend spending more time on meaning-focused input than on language-focused study, while doing both.

How easy should extensive reading be?

Very easy: around 98% known words, or about one unknown word in every 50. You should be able to read without a dictionary, at close to your normal speed, and follow the story without effort. If you’re looking up words on every line, it’s intensive reading.

Does extensive reading improve vocabulary?

Yes, but gradually. Studies such as Waring and Takaki (2003) found that a single graded reader teaches only a few new words that last, so the benefit comes from reading a lot. Words you meet many times across many texts are the ones that stick.

How many books should I read for extensive reading?

A widely cited target from the extensive reading literature is about one graded reader a week, and more at higher levels, so that new words come back often enough. For native material, a steady daily habit of 20–30 minutes is a good target. Volume matters more than any single book.

Both camps are right, at different times. The useful question isn’t which one to choose, it’s when to move from one to the other.

What graded readers are

Graded readers are books written or adapted for learners, with controlled vocabulary and grammar. They’re usually organized into levels, either by the number of headwords (the distinct base words a book may use) or by CEFR levels from A1 upward. A beginner-level reader might draw on a few hundred common words; the top levels might use a few thousand.

They sit at the heart of extensive reading, the practice of reading lots of easy, enjoyable material quickly, for meaning, without stopping constantly. Richard Day and Julian Bamford’s work on extensive reading in the late 1990s helped popularize the approach in classrooms, and researchers such as Paul Nation and Rob Waring have long argued that graded readers are what make extensive reading possible for learners who don’t yet know enough words for native texts.

Why graded readers work

• High coverage. Because the vocabulary is controlled, you know almost every word on the page, which is the condition under which reading feels like reading instead of decoding.
• Repetition of common words. The most frequent few thousand words of a language make up the bulk of any text. Graded readers make you meet them again and again in context, which is how they become automatic.
• Speed. When you’re not stopping for every word, you read faster, and reading faster means reading more. That volume is where much of the learning happens.
• Confidence. Finishing a whole book in a new language, even a simple one, is a real milestone.

Why they’re not the destination

Graded readers have limits. The language can feel flattened, the topics are often safe and generic, and the idioms, slang and texture of real speech are smoothed away. At some point you’ll notice you’re reading fluently but not being surprised by anything. That’s a sign the readers have done their job.

Native content, meanwhile, offers everything graded material can’t: the stuff you actually care about, real usage, culture, humor, and the reason you started learning in the first place. Motivation is not a side issue. People keep reading what they enjoy.

Signs you’re ready to switch

There’s no exam for this, but a few signals are reliable:

1. The top levels feel easy. You read graded readers quickly, rarely look anything up, and could read for an hour without getting tired.
2. A coverage check passes. Take a native text on a topic you know well and estimate how many words you know. Research on lexical coverage suggests that around 98% known words makes unassisted reading comfortable, and that about 95% is workable with some help. With a tap-to-look-up tool you can push a little below that, at the cost of reading more slowly.
3. You’re bored. If graded stories feel like a chore, your motivation is telling you something.

If you’re not seeing these yet, that’s fine. Another month of easy, enjoyable reading is never wasted.

Don’t switch all at once: build a bridge

The jump from the top graded level to a contemporary novel is still big. Instead of leaping, try these stepping stones:

• Narrow reading. Pick one topic you love, such as football, cooking or a particular video game, and read lots about it. The same vocabulary keeps coming back, so coverage rises fast.
• Children’s and young-adult books. Written for native readers, but with simpler language and clear storytelling.
• Books you’ve read in your own language. If you already know the plot, you can follow far more than your vocabulary alone would allow.
• Short-form native content. Blog posts, recipes, product reviews and comics are short enough to finish even when they’re a stretch.
• Transcripts of shows you’ve watched. Reading the words of a scene you already know is a gentle way into natural dialogue.

A simple ladder looks like this:

Mix, don’t replace

Even once you’re reading native content, easy material still has a place. A good balance for many learners is mostly just-right reading, with occasional stretches into harder material for motivation, and easy reading when you’re tired. Drop back down a level without guilt whenever reading stops being enjoyable.

How Lumo helps you make the switch

Lumo Stories are short graded originals, written for learners from A1 up, with read-along audio. They’re there for the starting-out and building stages. When you’re ready for more, you can import native articles, ebooks, subtitles and YouTube videos, and Lumo shows how much of each one you already know before you start. Every text is labeled Easy, Just right, Challenging or Hard, based on your own known words.

That turns “am I ready?” from a guess into something you can check. Import a few things you’d love to read. If they come up Just right, it’s time. If they come up Hard, keep them in your library. One day soon, they won’t be.

Here’s what an EPUB actually is, what DRM blocks, where to find books you can use, and how language learning apps handle them.

What an EPUB is, and why it’s ideal for learners

An EPUB is, underneath, a zip file of web pages. Inside are the chapters as HTML, a table of contents, the book’s metadata (title, author, often its language) and any images. That structure is why EPUBs are so useful for language learners:

• The text is real text. Unlike a scanned PDF, every word can be tapped, looked up, highlighted or read aloud.
• It reflows. Text adapts to any screen size, so it works on a phone as well as a laptop.
• It has chapters. A good importer can split a 90,000-word novel into manageable sessions.

PDFs are fine for printing but awkward for reading on a screen, and scanned PDFs contain pictures of text rather than text. If you have a choice, take the EPUB.

The DRM problem

DRM (digital rights management) is copy protection. Most ebooks sold by large stores are encrypted so they open only in that store’s apps and devices. A DRM-protected EPUB is just scrambled data to any other app, including every language learning tool. That’s why importers ask for DRM-free files.

Here’s where the big stores stand at the time of writing:

A few details worth knowing:

• Amazon. In February 2025, Amazon removed the “Download & Transfer via USB” option that people used to get copies of their Kindle books. Then, from January 20, 2026, it started letting verified buyers download EPUB or PDF files of books that the publisher has marked DRM-free, according to Amazon’s KDP help page (also reported by heise). The option covers books published through KDP, Amazon’s self-publishing platform, whose authors or publishers have chosen to sell them without DRM. Books published before December 9, 2025 qualify only if the author opts in, and Kindle Unlimited borrows are excluded, so most Kindle books still can’t be downloaded this way.
• Kobo. Its help page explains that most books are protected and need Adobe Digital Editions, while DRM-free ones download as a normal file.
• Google Play Books. Google’s help page notes that some publishers don’t allow downloads at all.

What about removing DRM?

You’ll find tools online that strip DRM from books you’ve bought. We won’t link to them. In many countries, breaking copy protection is legally restricted even for books you’ve paid for: in the US, the Digital Millennium Copyright Act prohibits circumventing it, with narrow exceptions, and the EU has similar rules. We’re not lawyers, and the rules vary, but the safe approach is simple: use DRM-free files.

The good news is that you rarely need to. There’s far more legal, DRM-free reading in most languages than you’ll ever finish.

Where to find free, legal EPUBs

Project Gutenberg

Project Gutenberg is the oldest free ebook library, with books in dozens of languages, all available as EPUB. English dominates, but the other collections are substantial. Going by the Gutendex catalog at the time of writing, there are about 4,300 books in French, 2,500 in German, 1,100 in Italian, 900 in Spanish and 650 in Portuguese. You can browse by language at gutenberg.org/browse/languages.

One caution: Gutenberg checks copyright under US law. Its terms tell readers outside the US to check their own country’s rules, because a book that’s public domain in the US may still be under copyright elsewhere, and the other way round.

Wikisource

Wikisource is Wikipedia’s sister library of public-domain and freely licensed texts, with separate editions for many languages, such as French, German, Spanish, Chinese and more. Texts are often proofread against scans of the original printed book. Its WS Export tool, reachable from the “Download” link on a book’s page, turns a work into an EPUB.

National and regional libraries

Many languages have their own free digital libraries:

• French: the Bibliothèque numérique romande and Ebooks libres et gratuits publish carefully prepared, free EPUBs of public-domain French literature.
• Japanese: Aozora Bunko is a huge volunteer library of public-domain Japanese texts, including authors such as Natsume Sōseki and Ryūnosuke Akutagawa. It publishes plain text and HTML rather than EPUB, though free tools convert its files to EPUB.
• English: if English is your target language, Standard Ebooks publishes beautifully formatted, free editions of public-domain books.

DRM-free stores and publishers

Some publishers sell ebooks without DRM, and some stores label DRM-free books clearly. Self-published and small-press authors are especially likely to sell DRM-free. If you’re about to buy a book in your target language, check the listing for a DRM-free option first.

A word on old books

Public-domain books are old books. The French of Jules Verne or the Spanish of Benito Pérez Galdós is perfectly readable, but it’s more formal than modern speech and full of words you won’t hear today. Classics make a great part of your reading, but you’ll want some modern text too. (If you’re choosing between book types, see our post on graded readers versus native content.)

How to import an ebook into a language learning app

Most reading-focused language apps import EPUBs, but they handle them differently. Here’s what the main tools say about EPUB at the time of writing.

LingQ is a strong choice if you also import PDFs and other formats, and Readlang is a good lightweight option for reading in a browser. Whichever you use, the import step is similar: choose the file, check that the language was detected correctly, and start reading.

Lumo, the app we’re building, isn’t available yet, but we kept its EPUB import simple: you pick a DRM-free EPUB, and Lumo processes it on your device, detects the language, splits it into chapters and adds it to your library with a difficulty label. The book doesn’t need to be uploaded anywhere to be read.

When an import goes wrong

Even DRM-free EPUBs can be messy. The common problems, and fixes:

• “Not a valid EPUB.” Check that the file really is an EPUB. If it’s a different format (MOBI, AZW3 or a DOCX), the free tool Calibre can convert DRM-free files to EPUB.
• Garbled characters. Very old EPUBs sometimes use unusual text encodings. Converting the file to EPUB again in Calibre often fixes it.
• Pages of front matter. Copyright pages, tables of contents and introductions in your native language come along with the book. Skip ahead to chapter one.
• The wrong language detected. A book with a long English introduction can fool language detection. Correct it before you start.
• Huge chapters. Some books are one enormous file with no chapter breaks. That’s readable, but you’ll want to keep track of where you stopped.

Picking your first EPUB to read in another language

The format is the easy part. The hard part is choosing a book you can actually finish. A few rules of thumb:

1. Check the difficulty first. If you’re looking up more than one or two words per line, the book is too hard for now. Save it for later.
2. Prefer short. A 150-page novella you finish teaches more than a 600-page classic you abandon.
3. Choose something you know. A translation of a book you’ve already read is easier because you know the plot.
4. Read in sessions, not marathons. One chapter a day adds up fast.

If you’re aiming for your first full novel, our guides to reading your first novel in Spanish and learning French by reading go through what to read at each level.

FAQ

Can I import Kindle books into a language learning app?

Usually not. Kindle books are DRM-protected and can only be opened in Amazon’s own apps and devices. Since January 2026, Amazon lets verified buyers download EPUB or PDF files of self-published KDP titles that the author has made DRM-free, and those files can be imported.

Is it legal to remove DRM from ebooks I bought?

In many countries it’s restricted. In the US, the DMCA prohibits circumventing copy protection, with narrow exceptions, and the EU has similar rules. Laws vary and we’re not lawyers, so the safest choice is to use DRM-free files.

Where can I download free ebooks in other languages?

Project Gutenberg and Wikisource have free public-domain books in many languages, and both offer EPUB downloads. Many languages also have dedicated libraries, such as the Bibliothèque numérique romande for French and Aozora Bunko for Japanese. Check that a book is public domain in your country.

Is EPUB or PDF better for language learning?

EPUB, in almost every case. The text reflows to fit your screen, it’s split into chapters, and every word is real text you can tap and look up. PDFs keep a fixed page layout, and scanned PDFs contain images of text rather than text.

How do I know if an EPUB has DRM?

The simplest test is to try opening it in a different app. A DRM-protected file won’t open, or it will ask for an account or Adobe Digital Editions. Store listings sometimes say “DRM-free,” and Kobo, for example, lets you download DRM-free books directly from your account.

“Fluent” isn’t a technical term. So instead of one magic number, here is what the research measures, what the main studies found, and what that means for your own learning.

How many words do you need to be fluent? It depends on the goal

Vocabulary researchers don’t ask “how many words make you fluent?” They ask a narrower, testable question: how many words do you need to know to understand a given kind of text?

They answer it with lexical coverage, the share of running words in a text that you know. If you know 196 of the 200 words in a passage, your coverage is 98%. (We go deep on this in the 98% rule.) The research then has two steps:

1. Find out what coverage you need to understand a text comfortably.
2. Use large collections of real language (corpora) to work out how many words reach that coverage for different kinds of text.

Different goals, such as chatting, watching TV or reading literature, give different answers.

Words, word families and lemmas

Most of the classic figures count word families: a base word plus its inflections and common derivations. Decide, decides, decided, deciding, decision and undecided are one family. A lemma is narrower, just a base form and its inflections (decide, decides, decided, deciding).

This matters because the numbers aren’t interchangeable. 8,000 word families is many more than 8,000 distinct forms on the page. Keep that in mind when comparing research figures with the word counts apps show you. Our post on what a known words count means covers that in detail.

Step one: how much coverage do you need?

Laufer (1989): about 95%

Batia Laufer proposed that learners need about 95% coverage of a text to reach reasonable reading comprehension. That means roughly one unknown word in every 20.

Hu & Nation (2000): about 98%

Hu Hsueh-chao and Paul Nation tested this directly. In “Unknown vocabulary density and reading comprehension” (Reading in a Foreign Language 13(1), 2000, article), they replaced words in a short fiction text with nonsense words to create versions at 80%, 90%, 95% and 100% coverage. Their analysis suggested that most learners need about 98% coverage to understand fiction adequately on their own. At 95%, some did and many didn’t. At 80%, almost no one did.

That study was small, with 66 learners, and a 2023 replication with 104 learners in Sri Lanka by Kremmel and colleagues in Language Learning (doi:10.1111/lang.12622) couldn’t fully reproduce its results. Treat 98% as a well-supported guideline, not a hard line.

Laufer & Ravenhorst-Kalovski (2010): two thresholds

In “Lexical threshold revisited” (Reading in a Foreign Language 22(1), article), Laufer and Tamar Ravenhorst-Kalovski proposed two levels:

• a minimal threshold of about 4,000–5,000 word families for 95% coverage (including proper nouns), enough for reading with some help,
• an optimal threshold of about 8,000 word families for 98% coverage, enough for independent reading.

Schmitt, Jiang & Grabe (2011): no cliff edge

Norbert Schmitt, Xiangying Jiang and William Grabe (”The percentage of words known in a text and reading comprehension,” The Modern Language Journal 95(1), 2011) found that comprehension rises roughly in a straight line with coverage. With 661 participants from eight countries, they found no point where comprehension suddenly switches on, and they concluded that 98% is the more reasonable coverage target for academic texts. So every extra point of coverage helps, and there’s no cliff at 97.9%.

Step two: how many words reach that coverage?

Nation (2006): the headline figures

The most cited answer is Paul Nation’s “How large a vocabulary is needed for reading and listening?” (The Canadian Modern Language Review 63(1), 2006, PDF). Using fourteen 1,000-word-family lists built from the British National Corpus, he estimated that, assuming 98% coverage is needed:

• written text (novels, newspapers) needs about 8,000–9,000 word families,
• spoken text (conversation, films) needs about 6,000–7,000 word families.

These figures are for English, and they include proper nouns as known.

Films and TV: Webb & Rodgers (2009)

Stuart Webb and Michael Rodgers analyzed a large corpus of movie scripts (”The lexical coverage of movies,” Applied Linguistics 30(3), 2009). They found that 3,000 word families plus proper nouns and marginal words gave about 95% coverage, and about 6,000 gave about 98%. Their companion study of TV programs (Language Learning, 2009) found similar figures: about 3,000 families for 95% and about 7,000 for 98%.

Conversation and listening: Adolphs & Schmitt (2003), van Zeeland & Schmitt (2013)

Everyday speech uses a smaller vocabulary than writing. Svenja Adolphs and Norbert Schmitt (Applied Linguistics, 2003) found that it takes roughly 2,000–3,000 word families to reach about 95% coverage of spoken English, more than earlier estimates had suggested. Hilde van Zeeland and Schmitt (Applied Linguistics, 2013) then found that many learners understood informal spoken narratives well at 95% coverage, and estimated that 2,000–3,000 families would get you there. That puts the target for following ordinary conversation far below Nation’s 6,000–7,000.

The numbers in one table

For comparison, Goulden, Nation and Read (1990) estimated that university-educated native English speakers know around 17,000 base words. Marc Brysbaert and colleagues (Frontiers in Psychology, 2016), counting lemmas instead, estimated that an average 20-year-old American knows about 42,000 lemmas. Different units, different numbers. Either way, “fluent” adults know far more words than any of the thresholds above. You don’t need a native-sized vocabulary to use a language well.

Not all words are equal: frequency does the heavy lifting

Word frequency is extremely skewed. A few hundred words, like the, of, and, is, make up a huge share of everything anyone says or writes. In their reassessment of frequency bands (“A reassessment of frequency and vocabulary size in L2 vocabulary teaching,” Language Teaching 47(4), 2014), Norbert and Diane Schmitt argued that the first 3,000 word families should count as “high frequency.” These are the words every learner needs, and they deserve direct study. Beyond that, each extra thousand words adds less coverage than the one before.

That’s why progress feels fast early on and slow later. Going from 1,000 to 2,000 families adds a lot of coverage. In Nation’s analysis of novels, going from 7,000 to 8,000 adds well under one percentage point. But near the top, small gains matter: at 97% coverage you meet an unknown word about every three lines, and at 98% about every five.

It also explains why the last few thousand words are hard to learn from flashcards. They’re rare. You might meet a given 8,000th-level word only once every few books. The only realistic way to meet them often enough is to read and listen a lot.

What this means for other languages

Almost all of this research is on English. Some things carry over and some don’t:

• The coverage findings (around 95% for support, 98% for comfort) are about how reading works, and they likely apply broadly.
• The vocabulary sizes depend on how a language builds words. Languages with rich inflection, like Russian or Finnish, have many more distinct forms per family. Languages with productive compounding, like German, produce long words you can often work out from their parts.
• Writing systems matter. In Chinese and Japanese, you also need to know the characters, and the relationship between characters and words adds its own learning load.

So use the numbers as orders of magnitude: low thousands for conversation, mid-to-high thousands for comfortable reading.

How to get there

1. Learn the first 2,000–3,000 families deliberately. Graded readers, frequency-based decks and plenty of easy input all help. This is where a flashcard scheduler like FSRS earns its place.
2. Then let reading carry you. Past about 3,000 families, each word is rare enough that only volume will get you enough encounters. Lots of reading at 95–98% coverage is the most efficient way to build the long tail.
3. Pick texts by coverage, not by level labels. A text about a topic you know well can be comfortable long before a general-interest novel is.
4. Expect listening to lag. Listening coverage targets are lower, but listening is harder in other ways: speed, accents, no rereading. Reading along with audio helps bridge the gap.

Lumo, the reading app we’re building (it isn’t available yet), is designed around this idea. It tracks the words you know, and it labels every text by how much of it you’d already understand: Easy at 98% and above, Just right at 90–98%, Challenging at 80–90% and Hard below that. Each page you read moves your count along, so you can watch texts shift from Hard to Just right.

FAQ

How many words do you need to be conversational?

Research suggests around 2,000–3,000 word families get you to about 95% coverage of everyday spoken English, which is enough to follow most casual conversation. You’ll still miss words, but context fills in most of them. Topic-specific talk, like work or hobbies, needs its own vocabulary on top.

How many words does a native speaker know?

Estimates depend on how you count. Goulden, Nation and Read (1990) estimated about 17,000 base words for educated native English speakers, and Brysbaert and colleagues (2016) estimated about 42,000 lemmas for an average American 20-year-old. Either way, it’s well beyond what you need to read and speak comfortably.

Is 1,000 words enough to be fluent?

No, though 1,000 well-chosen words go a long way. The most frequent 1,000 word families cover a large share of everyday English, but well short of the 95% that research links to comfortable understanding, so you’ll still hit unknown words in almost every exchange. Think of 1,000 words as survival level and 3,000 as the start of real independence.

How many words do you need to read a novel?

Research points to about 8,000–9,000 word families for reading novels comfortably without a dictionary (Nation, 2006), or 4,000–5,000 if you’re happy to read with help (Laufer & Ravenhorst-Kalovski, 2010). Graded readers and a tap-to-translate reader let you start well before that.

Do these numbers apply to languages other than English?

The coverage targets of around 95% and 98% probably do, because they’re about how reading comprehension works. The vocabulary sizes are English-specific and will shift with how a language forms words, so treat them as a rough scale, not exact targets.

The input hypothesis is one of five that Stephen Krashen set out in the late 1970s and early 1980s, most fully in Principles and Practice in Second Language Acquisition (1982), which he now makes freely available. If you want the practical, day-to-day version of the idea, start with comprehensible input, explained. This post is the closer look: each hypothesis, the evidence for it, and the critics who pushed back.

The five hypotheses at a glance

1. The acquisition–learning hypothesis

Krashen distinguished acquisition, the subconscious process by which children pick up their first language, from learning, the conscious knowledge of rules you get from a textbook. His most controversial claim was that the two are separate: learned knowledge does not turn into acquired knowledge, no matter how much you practice it. This is often called the “non-interface” position.

What supports it

The broad distinction has aged well. Modern research talks about implicit and explicit knowledge, and most researchers accept that they’re different. You can know the rule for French gender agreement perfectly and still get it wrong in fast speech, and native speakers use rules they could never state. In their 2021 essay “Was Krashen right? Forty years later” in Foreign Language Annals (doi:10.1111/flan.12552), Karen Lichtman and Bill VanPatten argue that the acquisition–learning distinction lives on today as implicit versus explicit learning, often without credit.

What’s contested

The non-interface claim is where the disagreement lies. “Weak interface” positions hold that explicit knowledge can help acquisition indirectly, for example by helping you notice things in input. “Strong interface” positions, associated with skill-acquisition theory, hold that practice can make explicit knowledge automatic. Kevin Gregg’s early critique, “Krashen’s monitor and Occam’s razor” (Applied Linguistics, 1984, doi:10.1093/applin/5.2.79), argued that the two systems were defined in a way that made the claim impossible to test.

2. The monitor hypothesis

According to Krashen, learned knowledge has one job: acting as an editor, or monitor, that checks and corrects what the acquired system produces. And it only works when three conditions are met: you have time, you’re focused on form, and you know the rule.

This explains something real. People make fewer errors on a written test than in conversation, and grammar knowledge does seem most useful when you can slow down. But most researchers found the hypothesis too narrow and too hard to test directly. It’s rarely tested on its own terms today.

3. The natural order hypothesis

Krashen drew on the morpheme order studies of the 1970s, such as Heidi Dulay and Marina Burt’s “Natural sequences in child second language acquisition” (Language Learning, 1974). These found that learners of English acquired grammatical features like progressive -ing, plural -s and third-person -s in a broadly similar order, whatever their first language. Third-person -s (she walks) comes surprisingly late, even though it’s taught early.

What supports it

The idea of ordered development holds up well. Learners pass through similar stages for structures like negation and questions, and teaching a structure before the learner is ready often doesn’t stick. Lichtman and VanPatten count this as another of Krashen’s ideas that survived.

What’s contested

The original studies had methodological problems: they often measured accuracy at one point in time rather than tracking acquisition. And the order itself turns out to be largely explainable. A meta-analysis by Jennifer Goldschneider and Robert DeKeyser (Language Learning, 2001, doi:10.1111/1467-9922.00147) found that a combination of factors, including how salient a form is in speech, how frequent it is and how complex its meaning is, accounted for much of the order. That doesn’t contradict a natural order, but it suggests an explanation grounded in the input rather than a fixed internal program.

4. The input hypothesis

This is the center of the theory. Krashen argued that we acquire language in one way: by understanding messages containing structures slightly beyond our current level, which he wrote as i+1. Speaking, in this view, emerges from acquisition rather than causing it. And if learners get enough comprehensible input, the next structures they need will be in it automatically.

What supports it

The evidence that input drives acquisition is broad and consistent:

• Extensive reading. In a well-known “book flood” study in Fiji, Warwick Elley and Francis Mangubhai (Reading Research Quarterly, 1983, doi:10.2307/747337) gave primary schools a supply of high-interest illustrated storybooks in English. Children who read them made larger gains than those taught with the standard program. Takayuki Nakanishi’s 2015 meta-analysis in TESOL Quarterly (doi:10.1002/tesq.157) pooled 34 studies of extensive reading and found a medium effect in group comparisons (d = 0.46) and a larger one in pre–post comparisons (d = 0.71).
• Incidental vocabulary learning. Study after study finds that learners pick up words from reading and viewing, as long as they understand what’s going on. The gains from a single encounter are small, which is why volume matters.
• Coverage research. Work on lexical coverage, covered in our post on the 98% rule, shows how strongly comprehension depends on knowing nearly all the words, which fits Krashen’s emphasis on comprehensible.

No mainstream theory of second language acquisition today denies that input is essential. That’s a real win for Krashen.

Criticism: “i+1” can’t be measured

Barry McLaughlin, in Theories of Second-Language Learning (1987), made the most cited criticism: the theory’s key terms were never defined precisely enough to test. What exactly is a learner’s “i”? How would you know if a text is “+1” rather than “+2”? If any learning is explained after the fact by “there must have been i+1 in the input,” the claim can’t be proven wrong. Gregg made a similar point.

This is a fair criticism of i+1 as a scientific claim. It matters less for you as a learner, because “mostly understandable, slightly stretching” is a perfectly good practical guide even if it isn’t a measurable quantity.

Criticism: input isn’t enough (Swain)

Merrill Swain’s challenge came from Canadian French immersion programs, where English-speaking children studied school subjects in French for years. If input were sufficient, these students should have been in great shape. They understood French extremely well, but their speaking and writing remained noticeably less accurate than those of native speakers, even after years of input.

In her 1985 chapter “Communicative competence: Some roles of comprehensible input and comprehensible output in its development” (in Gass and Madden’s Input in Second Language Acquisition), Swain proposed the output hypothesis. When you have to produce language, you discover gaps you can skip over while listening, because comprehension lets you get the meaning from context without processing every grammatical detail. Speaking and writing push you to process the grammar.

Criticism: conversation matters (Long)

Michael Long’s interaction hypothesis didn’t reject input. It refined it. Long’s research in the 1980s looked at conversations between native and non-native speakers, and found that a great deal of input becomes comprehensible through negotiation of meaning: asking for clarification, checking understanding, rephrasing. In his updated version (in Ritchie and Bhatia’s Handbook of Second Language Acquisition, 1996), interaction helps because it connects input, the learner’s attention and output, and gives learners feedback right when they need it.

The interaction hypothesis is arguably closest to Krashen of all the critiques: comprehensible input is still the engine, but conversation is an especially good way to get it and fine-tune it.

Criticism: you learn what you notice (Schmidt)

Richard Schmidt’s “The role of consciousness in second language learning” (Applied Linguistics, 1990, doi:10.1093/applin/11.2.129) drew partly on his own experience learning Portuguese in Brazil. He noticed that forms he’d been hearing for months only started appearing in his speech after he had consciously noticed them in input. His noticing hypothesis claimed that input only becomes “intake” when you notice it. He later softened this to a weaker claim: people learn about what they attend to, and don’t learn much about what they don’t.

This directly contradicts the idea of purely subconscious acquisition. The strong version has been hard to test, because it’s hard to measure whether someone noticed something. But the weaker version, that attention matters, is now widely accepted.

Criticism: teaching grammar does help

Krashen argued that grammar instruction has little effect on acquisition. A large meta-analysis by John Norris and Lourdes Ortega (Language Learning, 2000, doi:10.1111/0023-8333.00136) of 49 studies found the opposite: targeted instruction produced large gains, explicit instruction outperformed implicit instruction, and the effects lasted.

There’s a caveat Krashen’s defenders raise, and Norris and Ortega noted it themselves: many studies used tests that favor explicit knowledge, such as grammar tests, so the effect on spontaneous use may be smaller. Still, the claim that instruction does nothing is hard to sustain.

5. The affective filter hypothesis

Krashen proposed that anxiety, low motivation and low self-confidence raise a mental “filter” that keeps input from reaching the parts of the brain that acquire language.

The evidence that emotions matter is solid. A meta-analysis by Yasser Teimouri, Julia Goetze and Luke Plonsky (Studies in Second Language Acquisition, 2019, doi:10.1017/S0272263118000311) covered almost 20,000 learners and found a moderate negative correlation between language anxiety and achievement (r = −.36).

But correlation runs both ways. Anxious learners may do worse, and learners who are doing worse may become anxious. And the “filter” itself is a metaphor rather than a mechanism anyone has identified. The practical advice survives, even if the model doesn’t: relaxed, interested learners tend to keep going, and keeping going is most of the battle.

Krashen’s input hypothesis today: was he right?

Partly, and in the ways that matter most to self-directed learners.

What holds up:

• Comprehensible input is necessary, and it’s the main source of acquisition.
• Large amounts of reading and listening produce real gains in vocabulary, grammar and fluency.
• Implicit knowledge is different from explicit knowledge, and implicit knowledge is what fluent use depends on.
• Learners develop in a broadly predictable order that teaching can’t easily override.
• Anxiety and boredom get in the way.

What doesn’t:

• That input is sufficient on its own for everyone, for every skill.
• That conscious learning and attention play no role.
• That grammar instruction is essentially useless.
• That “i+1” is a precise, testable quantity.

Most researchers today would describe input as the foundation, with attention, output, interaction and some explicit teaching each adding something on top. Lichtman and VanPatten’s conclusion is a good summary: many of Krashen’s ideas are alive, often under new names, even where his specific formulations were rejected.

What this means for how you learn

If you’re an input-first learner, the research suggests a few adjustments rather than a change of direction:

1. Make input the bulk of your time. Reading and listening at the right level is where most acquisition happens.
2. Pay attention, lightly. Schmidt’s point is easy to apply: notice the words and patterns you meet. Looking up a word, saving a phrase or highlighting a form you keep seeing all count.
3. Use a little explicit knowledge. A quick look at a grammar point can help you notice it in input. It doesn’t replace input.
4. Add output and conversation eventually. Swain and Long both suggest that speaking and writing push you further than input alone, especially for accuracy.
5. Keep it enjoyable. Whatever you call the affective filter, you’ll do more of something you like.

Lumo, the reading and listening app we’re building, is designed around the first two points. It labels every text by how much of it you already understand, so you can find input at the right level, and it keeps the words you’re learning highlighted so you notice them each time they come back. It isn’t available yet. Its job is to make the input part easy; the conversation part is up to you and whoever you talk to.

FAQ

What are Krashen’s five hypotheses?

The acquisition–learning hypothesis, the monitor hypothesis, the natural order hypothesis, the input hypothesis and the affective filter hypothesis. Together they’re often called Monitor Theory, and they were set out most fully in Krashen’s 1982 book Principles and Practice in Second Language Acquisition.

What is the main criticism of Krashen’s input hypothesis?

The most common criticism, from Barry McLaughlin and Kevin Gregg, is that the key terms, especially “i+1,” are too vague to test. Others, like Merrill Swain, argued that input alone isn’t sufficient, because learners with years of input in immersion programs still had gaps in accuracy.

What is the difference between Krashen’s input hypothesis and Swain’s output hypothesis?

Krashen argued that acquisition comes from understanding input, and that speaking is a result of acquisition, not a cause. Swain argued that producing language also drives learning, because it forces you to notice gaps in your knowledge and process grammar you can skip while listening.

Is comprehensible input enough to become fluent?

For comprehension, large amounts of input take you a very long way. For accurate speaking and writing, most research suggests you also need practice producing language and some feedback. Input is the foundation, not the whole building.

Is Krashen’s theory still accepted today?

Not as a complete theory, but many of its ideas are. The importance of comprehensible input, the difference between implicit and explicit knowledge, and ordered development are all part of mainstream research. The claims that input is sufficient and that instruction doesn’t help are widely rejected.

Researchers have a name for the share of words in a text that you know: lexical coverage. And one of the most useful findings in reading research is how sharply the experience of reading changes as coverage climbs from 90% to 95% to 98%.

What coverage means

Coverage is counted over running words, every word on the page in order, not over the dictionary. If a 200-word passage contains four words you don’t know, your coverage is 98%. If it’s twenty, it’s 90%.

That sounds like a small difference. On the page, it isn’t.

At 98%, an unknown word is an occasional bump. At 90%, there’s one in nearly every line, and you spend more energy decoding than reading.

What the research found

In 1989, Batia Laufer proposed that around 95% coverage was needed for reasonable reading comprehension. A decade later, Hu Hsueh-chao and Paul Nation (2000) tested this more directly. They took a short fiction text and replaced different proportions of its words with made-up words, creating versions at different coverage levels, and then measured how well learners understood each version. Their results suggested that most learners need about 98% coverage to understand a fiction text adequately without help, and that at 95% only some managed it.

Later work refined the picture. Laufer and Tamar Ravenhorst-Kalovski (2010) proposed two thresholds: around 95% as a minimal level for reading with some support, and 98% as an optimal level. Norbert Schmitt, Xiangying Jiang and William Grabe (2011) found that comprehension rises fairly steadily with coverage rather than switching on at a single magic number. Every extra percentage point helps.

So “the 98% rule” is best understood as a guideline, not a law: around 98% is where unassisted reading becomes comfortable for most people, and comprehension falls away gradually below it.

How many words is that?

In a well-known 2006 paper, Paul Nation used large collections of real text to estimate how big a vocabulary you need to reach 98% coverage. His estimate: roughly 8,000 to 9,000 word families for a wide range of written text, and around 6,000 to 7,000 for spoken language such as conversation and films.

A word family is a base word plus its related forms: read, reads, reading, reader, readable and so on. So the number of individual forms you’d recognize is larger, but you don’t learn each one from scratch.

That might sound daunting. But you don’t need 98% coverage of everything to enjoy reading something. Coverage depends on the text: a blog about a hobby you know inside out might hit 98% long before a literary novel does. And it grows with every page you read.

Why the last few percent matter so much

Three things make high coverage disproportionately valuable:

1. Context needs company. You can usually guess an unknown word from the words around it, but only if you know those words. At 98% the surroundings are solid; at 90% the clues themselves are often unknown.
2. Attention is limited. Every unknown word takes a little working memory. A few per page is fine. Thirty per page means you lose the thread of the sentence you’re in.
3. Enjoyment drives volume. At high coverage, reading is pleasant, so you read more, and reading more is how coverage rises. At low coverage it’s exhausting, so you stop.

Caveats worth knowing

• Background knowledge counts. Knowing the topic, the genre or the plot makes lower coverage much more manageable.
• Not all unknown words are equal. A name or a word whose meaning is obvious from the picture costs less than a key verb.
• Assisted reading changes the maths. If looking up a word takes a second, you can read comfortably at somewhat lower coverage. You’ll be slower, but you’ll understand.
• Listening is different. Research on listening is thinner, and some of it suggests that slightly lower coverage can be enough for informal speech.

Putting it to work

A practical approach is to match coverage to the kind of session you want:

• 95–98%+ for relaxed, extensive reading: lots of pages, few lookups, maximum enjoyment.
• 90–95% for focused sessions with a look-up tool, where you’re happy to slow down and learn new words.
• Below 90% only for short texts you’re very motivated to read.

How Lumo uses coverage

Lumo keeps track of the words you know, and for every text in your library it works out what share of the words on the page are already known. That becomes a simple label: Easy at 98% and above, Just right from 90% to 98%, Challenging from 80% to 90%, and Hard below that. When Lumo recommends what to read next, it aims for the sweet spot around 95%.

Because turning the page marks the words you didn’t tap as known, the estimate grows with you. Texts that came up Hard a few months ago quietly become Just right.

That’s the hopeful part of the 98% rule. The way to get to 98% is to keep reading at 95%.

Reading aloud is a close cousin with a different job. Below: what the research actually says about both, how to shadow well, and the mistakes that make it a waste of time.

What shadowing is, and what it isn’t

In shadowing, you speak while you listen. You don’t wait for the end of a sentence, and you usually don’t pause the audio. The point is to track the speech in real time, copying its sounds, rhythm and intonation as you go.

That makes it different from a few things it’s often confused with:

The technique comes from interpreter training, where trainees shadow speech to build the skill of listening and speaking at once. It was introduced into English teaching in Japan in the early 1990s, and Japanese researchers have done most of the published work on it since.

What the research says about shadowing in language learning

Kadota’s framework

The most complete account is Shuhei Kadota’s Shadowing as a Practice in Second Language Acquisition: Connecting Inputs and Outputs (Routledge, 2019). Kadota, a professor of applied linguistics at Kwansei Gakuin University, proposes that shadowing has four separate effects:

1. Input effect. Shadowing trains faster, more automatic perception of speech, which improves listening comprehension.
2. Practice effect. It exercises the “inner voice” that holds sounds in working memory, which helps you learn new words, phrases and constructions.
3. Output effect. It simulates some of the stages of producing speech, without the effort of deciding what to say.
4. Monitoring effect. It trains you to monitor and control your own processing, a metacognitive skill.

Not every part of that model has equal evidence behind it. The listening effect is the best studied.

Listening: the strongest evidence

Yo Hamada’s study “Shadowing: Who benefits and how?” (Language Teaching Research, 2016, doi:10.1177/1362168815597504) set out to test two common claims: that shadowing improves learners’ perception of individual sounds, and that it helps lower-level learners most. Forty-three Japanese university students did a series of shadowing lessons using material from their English course.

The results:

• Sound perception improved in both the lower- and intermediate-proficiency groups, as measured by a 20-item dictation test.
• Listening comprehension improved only for the lower-proficiency group, on the easier of the listening tests. Neither group improved on the harder one.

Hamada’s interpretation is that shadowing works mainly on bottom-up listening: hearing where words begin and end and recognizing them in fast speech. That’s the part of listening that lower-level learners struggle with most, so they have the most to gain. More advanced learners, who already hear the words, need other kinds of practice to improve comprehension.

In a later review, “Shadowing: What is it? How to use it. Where will it go?” (RELC Journal, 2019, doi:10.1177/0033688218771380), Hamada summarizes the research on listening and speaking and suggests that beginners start with shadowing for listening and move to shadowing for speaking later.

Speaking: promising, smaller evidence

Jennifer Foote and Kim McDonough’s study “Using shadowing with mobile technology to improve L2 pronunciation” (Journal of Second Language Pronunciation, 2017, doi:10.1075/jslp.3.1.02foo) is one of the few tests outside Japan. Sixteen learners shadowed short dialogues on iPods at least four times a week, for at least ten minutes a session, over eight weeks, recording themselves as they went. Twenty-two English speakers rated recordings made before, midway through and after.

The learners improved on comprehensibility (how easy they were to understand) and fluency. They did not become significantly less accented. That’s a useful result, because comprehensibility is what matters for communication, and accent is notoriously slow to change.

Conversational shadowing

Tim Murphey’s “Exploring conversational shadowing” (Language Teaching Research, 2001, doi:10.1177/136216880100500203) looked at a different version: quietly repeating some of what a conversation partner says, in real time. He described shadowing as varying along continuums: from silent to fully voiced, and from complete (repeating everything) through selective (repeating key words) to interactive (adding short questions and comments of your own). Selective and interactive shadowing turned out to be a natural way to show a partner you’re following, and to keep your own attention on what they’re saying.

The honest caveats

The research on shadowing has real limits:

• Small samples. Most studies have a few dozen participants or fewer.
• One context. Much of the work is with Japanese university students learning English.
• Few strong controls. Many studies compare learners before and after shadowing without a comparison group doing something else for the same amount of time.
• Mixed findings on individual sounds. Not every study finds that shadowing improves the perception of difficult sound contrasts.

So the best summary is: shadowing reliably helps learners hear words in connected speech, it probably helps fluency and comprehensibility, and the bigger claims are plausible but not yet proven.

What about reading aloud?

Reading aloud has had a bad reputation in language teaching, partly because reading aloud in class is slow and stressful and can get in the way of understanding. In “Reading aloud: a useful learning tool?” (ELT Journal, 2008, doi:10.1093/elt/ccm075), Sally Gibson argued that it has been unfairly dismissed, and that it can be useful for specific purposes such as pronunciation practice, as long as it’s used selectively.

There’s also good memory research behind it. Colin MacLeod and colleagues’ paper “The production effect: Delineation of a phenomenon” (Journal of Experimental Psychology: Learning, Memory, and Cognition, 2010, doi:10.1037/a0018785) found that people remember words they read aloud better than words they read silently. The catch is that the advantage appeared when people read some words aloud and others silently. Reading a whole list aloud didn’t beat reading a whole list silently. Saying a word aloud seems to make it distinctive, and distinctiveness only helps when not everything is distinctive.

For learners, that suggests a specific use: read aloud the new words and phrases you want to remember, and read everything else silently. Reading a whole novel aloud is slower and won’t make it more memorable.

How to shadow: a 15-minute routine

Choose the right audio

• Short. One to three minutes is plenty. You’ll go over it several times.
• Easy. Pick something you can mostly understand. If you can’t follow the meaning, you’re just making sounds.
• One clear speaker. A podcast host or audiobook narrator is easier than a group conversation.
• With a transcript. You’ll need it for the first few passes.
• A natural speaker. Human recordings are better than text-to-speech for copying rhythm and intonation.

The steps

1. Listen for meaning. Play the audio once or twice without speaking. Read the transcript and look up anything you don’t understand. Shadowing works best on material you already understand.
2. Mumble along. Play it again and murmur along quietly, without worrying about getting every sound. This is about keeping pace.
3. Parallel read. Read the transcript aloud along with the audio. This links the sounds to the written words.
4. Shadow without the text. Put the transcript away and shadow, a moment behind the speaker. Focus on rhythm and intonation first, then on individual sounds.
5. Record and compare. Record one pass, then listen to it next to the original. Pick one thing to fix: a sound, a stressed word, a run of words you blurred together.
6. Come back tomorrow. Shadow the same clip for a few days, then move on to a new one.

Ten to fifteen minutes a day is a good amount. Shadowing is tiring, and quality drops quickly when you’re worn out.

Common mistakes

• Shadowing material that’s too hard. If you don’t understand it, you can’t process it, and you’ll just be copying noise. Step down a level.
• Never looking at the meaning. Shadowing nonsense syllables trains your mouth, not your listening.
• Only ever using the transcript. Parallel reading is a stage, not the goal. The listening benefit comes from tracking speech without the text.
• Doing too much at once. An hour of shadowing a week will do less than ten minutes a day.
• Treating it as a replacement for input. Shadowing is a few minutes of focused practice. Most of your listening should still be relaxed listening to lots of material you understand. (Our post on comprehensible input explains why.)

Where shadowing fits in an input-based routine

Shadowing is a good companion to an input-heavy approach rather than an alternative to it. Your reading and listening build vocabulary and a feel for the language. Shadowing helps you hear the words you already know when they come at native speed, and it’s a gentle bridge to speaking for learners who aren’t ready for conversation.

A good source of material is a video or podcast you’ve already watched and understood. If you use the three-pass method from our guide to learning a language from YouTube, shadowing fits naturally as the final pass: pick a favorite minute and shadow it.

Lumo can help with the “understand it first” part. It imports YouTube videos with an interactive transcript and dual subtitles, so you can look up words in the clip before you shadow it, and you can import your own audio or video files with SRT or VTT subtitles. Read-along text-to-speech highlights each word as it’s spoken, which is handy for linking spelling to sound in parallel reading, though for shadowing itself we’d still suggest a human recording. Lumo isn’t available yet, but it’s coming soon.

FAQ

Does shadowing actually work?

For listening, the evidence is fairly good: studies such as Hamada (2016) find that shadowing improves learners’ ability to perceive words in speech, especially for lower-level learners. For speaking, smaller studies such as Foote and McDonough (2017) found gains in fluency and comprehensibility. Most research is small-scale, so the effects are encouraging rather than proven.

How long should I shadow each day?

Ten to fifteen minutes a day is a good target. Shadowing takes concentration, and short, regular sessions work better than long ones. Use the same short clip for several days before moving on.

Can beginners do shadowing?

Yes, with easy material. Research suggests lower-level learners may benefit most for listening, and Hamada recommends that beginners start with shadowing for listening before using it for speaking. Use slow, clear audio you already understand, and start with a transcript.

Is shadowing better than reading aloud?

They do different things. Shadowing trains listening and fluency by tracking real speech, so it needs audio. Reading aloud doesn’t train listening, but saying selected new words aloud can help you remember them better than reading them silently.

Should I use subtitles or a transcript when shadowing?

Use a transcript to understand the clip and for parallel reading at first. Then put it away. The listening benefit comes from tracking the speech by ear, and reading the text the whole time turns shadowing into reading aloud.

The difference is what the streak rewards. A streak for opening an app is noise. A streak for reading or listening to something you understand, every day rewards exactly the behavior that leads to fluency. Here’s why that works, and how to keep one going without letting it run your life.

Reason 1: spacing beats cramming

One of the oldest and most replicated findings in memory research is the spacing effect: information studied in several sessions spread over time is remembered far better than the same amount of study crammed into one sitting. Hermann Ebbinghaus documented it in his experiments on memory in the 1880s, and it has been confirmed many times since, including in a large 2006 review of the research by Nicholas Cepeda and colleagues.

Reading every day is spacing in its most natural form. The words you met on Monday show up again on Wednesday and the following week, in new sentences, just as they’re starting to fade. A three-hour Sunday session gives you lots of encounters, but they’re bunched together, and many of them are gone by the next weekend.

Reason 2: small numbers add up

Language learning is mostly a volume game, and daily volume compounds quietly. Say you read at a modest 100 words a minute, which is a comfortable pace for many intermediate learners. Fifteen minutes a day is around 1,500 words a day. Over a year, that’s more than half a million words, or several novels’ worth of reading.

You’d never sit down and decide to read five novels in a language you’re learning. But you might decide to read for fifteen minutes today. The streak is how the second decision turns into the first.

Reason 3: habits form through repetition

A well-known study led by Phillippa Lally at University College London followed volunteers as they tried to turn a new daily behavior, such as eating fruit at lunch or going for a short run, into a habit. The time it took for the behavior to feel automatic varied enormously from person to person, from under three weeks to well over six months, with a median of around 66 days.

Two lessons from that work are especially useful:

• Habits take longer than the internet says. “21 days” is a myth. If reading still takes willpower after a month, nothing’s wrong with you. Keep going.
• Missing a single day didn’t derail habit formation. What mattered was getting back on track, not perfection.

That second point is why a good streak system has to forgive the occasional bad day.

The danger: all-or-nothing thinking

Psychologists Janet Polivy and C. Peter Herman described what they called the “what-the-hell effect” in dieters: break a strict rule once, and it’s tempting to abandon the rule entirely. Streaks invite exactly this. You miss one day, the counter resets to zero, and suddenly the habit you built for three months feels pointless.

The fix isn’t to drop streaks, it’s to design them so a single slip isn’t a catastrophe. That’s what streak freezes are for: a freeze covers a missed day so your streak survives, and you can get straight back to it the next day.

How to keep a streak going

Make the daily goal small

Pick a goal you could hit on your worst day: five or ten minutes. On good days you’ll blow past it. On bad days, a small goal is the difference between doing something and doing nothing, and doing something is what keeps the habit alive.

Attach it to something you already do

Psychologist Peter Gollwitzer’s research on implementation intentions found that simple if-then plans make people much more likely to follow through: “When I pour my morning coffee, I’ll read one page.” Linking reading to an existing routine removes the daily decision about when to do it.

Queue up tomorrow’s reading

Half the friction of a daily habit is deciding what to do. Before you finish today, pick tomorrow’s story, article or video. Then all you have to do is open it.

Have a minimum viable day

Decide in advance what counts on a terrible day: one page of a story, one short video, one round of flashcards. Lower the bar rather than skipping.

Read things you actually like

The best protection for a streak is enjoying the time. If your reading feels like homework, switch to something easier or more interesting. There’s no prize for suffering.

Celebrate the milestones

A week, a month, a hundred days. These are real achievements. Notice them.

How streaks work in Lumo

In Lumo, you choose a daily goal of 5 to 30 minutes. Each day you hit it adds a sun to your week, and your streak counts the days in a row you’ve shown up. If life gets in the way, a streak freeze covers the gap so one missed day doesn’t erase months of work. The free plan lets you hold one freeze at a time; Pro holds up to five. Milestones along the way, from your first week to 100 days and beyond, mark the progress.

The streak is a means, not the goal

The number on the screen isn’t the point. The point is the half-million words, the hundreds of hours of listening, and the day you notice you understood a whole conversation without trying. A streak is just a friendly way to make sure you keep showing up long enough to get there.

Lumo opens to everyone soon. This guide walks you through your first week with it, one small step a day, so you know exactly where to start. None of the days take more than fifteen or twenty minutes, and you can do them in any order. If you only remember one thing, make it this: pick content you mostly understand, and come back tomorrow.

Day 1: Set up and read your first story

When you first open Lumo, you tell it three things: the language you’re learning, roughly where you are (beginner, intermediate or advanced), and a daily goal.

For the goal, choose less than you think you can do. Five or ten minutes is perfect. You can always read more, and a small goal is far easier to keep on a busy day. You can change it later.

Then open a Lumo Story at your level. These are short originals written for learners, and they’re the easiest way to see how Lumo works. As you read, you’ll notice the words are colored:

• Blue, underlined words are new: you haven’t met them in Lumo before.
• Yellow words are ones you’re learning. The yellow fades as you go from unknown to recognized to familiar.
• No color means known.

Tap any word to see what it means in this sentence and to hear it spoken. Then set how well you know it, using the buttons on the word card or the keys 1 to 4 on a keyboard. Press K to mark a word known straight away, or X to ignore it (handy for names).

One important habit: when you turn the page, any blue words you didn’t tap are marked as known. That’s how your known-word count grows without you having to click every word you already know. So if you don’t know a word, tap it before you move on.

Day 2: Read along

Open yesterday’s story again and turn on read-along. Lumo reads the text aloud and highlights each word as it’s spoken. Follow along with your eyes. Slow it down if you need to.

Then read the story once more on your own, without audio. You’ll find it’s noticeably easier the second time. That’s not a trick: re-reading is one of the most effective things you can do, because more of the text is now within reach.

Day 3: Import something you love

Now bring in something of your own. Paste some text, or paste a link to a web article on a topic you care about, whether that’s football, recipes or a band you like. Lumo turns it into a clean, readable page.

Before you start reading, look at the label. Lumo shows how much of the text you already know and calls it Easy, Just right, Challenging or Hard. Early on, almost everything native will come up Hard, and that’s completely normal. Keep it in your library and choose something easier for today. Those labels change faster than you’d expect.

Day 4: Watch a video

Find a YouTube video with captions in your language. Videos made for learners are a great place to start. Paste the link into Lumo and you’ll get the video with an interactive transcript beside it. The current line lights up as it plays, and you can tap any word without losing your place.

Try watching it twice: once for the gist, then once more slowly, looking up the words you skipped. The free plan is planned to include two videos a week, so pick one you’ll enjoy.

Day 5: Your first review session

By now you’ve saved a handful of words. Open your flashcards and do a short review. For each card, try to recall the meaning, flip it, then grade yourself honestly: Again, Hard, Good or Easy.

Lumo schedules every card with FSRS, a modern open-source scheduling algorithm. Words you find hard come back soon; words you know well come back after longer and longer gaps. Honest grading makes the schedule work, so don’t be shy about pressing Again.

Keep reviews short. A few minutes a day is plenty. Reading is still where most of your learning happens; flashcards just stop words from fading between encounters.

Day 6: Phrases and sentences

Some meanings live in groups of words, not single ones: by the way, to take part in, no matter what. When you spot one, select the whole phrase and save it. It’ll be highlighted as a unit wherever it appears again.

If a whole sentence doesn’t make sense even after looking up the words, use sentence translation. The free plan is planned to include fifteen a day. Treat them as a safety net rather than a crutch: try first, then check.

Day 7: Look back and plan ahead

Take a minute to look at your progress: your streak, your known words, and the stories and articles you’ve finished. A week of daily reading is a real achievement.

Then set yourself up for next week:

• Queue up a few things to read, so you never have to decide what to do when you open Lumo.
• Pick a time and a trigger. “After dinner, on the sofa” works better than “sometime today.”
• Know what happens on a bad day. One page or one short video still counts. If you miss a day entirely, a streak freeze can cover it, so you don’t lose your streak over one rough evening.

A few extra tips

• On your phone, add Lumo to your home screen so it opens like an app. In Safari, tap Share, then Add to Home Screen.
• Stay in the Just right zone most of the time. Easy reading builds fluency; harder reading builds vocabulary; you need both, but mostly the middle.
• Re-read and re-watch favorites. It’s not cheating. It’s how input becomes yours.

Where to go next

If you’d like to understand the ideas behind Lumo, start with Comprehensible input, explained and The 98% rule. When you’re ready to graduate from stories to native content, read Graded readers vs. native content.

Most of all, enjoy it. You’re not studying a language this week. You’re reading in one.