VerbamorNotes on remembering.
Tactics for tutored learners, the memory science we build on, and the odd look behind the curtain.
The six days after the lesson are where the money goes
Your lesson has a price. Your retention has a price too, and it's the one that matters. Real per-lesson figures, divided by what you actually keep.
What Rosetta Stone costs, as of 21 September 2026
The ranking pages get the lifetime plan, the annual price, the guarantee and the trial wrong. Here are the live figures, with the receipts.
Anki for language learning: the bill arrives in month three
Anki's own manual puts 20 new cards a day at roughly 200 daily reviews. None of the three pages ranking for this query mentions it, or that FSRS ships switched off.
What to tell someone starting from zero
A friend asks how to start learning Spanish. Here is the advice worth giving, the numbers to quote, and a four-week plan you can send them today.
Is Rosetta Stone worth it? The price every review gets wrong
Rosetta Stone sells two plans: $19.99 a month or $159 a year. No lifetime tier, no free trial, 14 days to get your money back. Read off the checkout page.
Is Babbel worth it? I read the study it advertises
Babbel says a semester of Spanish in 15 hours. Its own report says 21 for ordinary users, on a reading and grammar test. Here is what the research does and doesn't show.
The FSRS benchmark does not contain SM-2
The most-quoted spaced repetition benchmark has no SM-2 row. I computed it from the repository files, and worked out what the gap is and isn't worth.
What Anki costs, checked today
Free on desktop, Android and web. US$24.99 once on iOS. Every price read off Apple's own store on 21 September 2026, including the regions everyone gets wrong.
Preply reviews: the billing rules behind the 1-star score
13% of Preply's Trustpilot reviews are 1-star. Preply's own help centre publishes the billing rules a buyer should know first.
Every Spanish, French, Italian, Portuguese word list
SUBTLEX-ESP is noncommercial. CEFRLex is noncommercial. wordfreq stopped in 2021. The licence column app builders keep getting wrong, checked at source.
What Pimsleur costs, including the year-2 price
The annual plan renews $33 higher than it starts. Here is every price, read off Pimsleur's own pages today, divided by the hours you actually get.
The three top Lingopie reviews contradict each other
10, 15 or 17 languages. $144, $168 or $83.88 a year. I read Lingopie's own help centre instead, and it publishes no dollar figure anywhere.
Auditing your cards and deleting the bad ones
A 20 minute audit for a deck that has stopped working, with four tests for a bad card and rules for when to fix, split or delete it.
Pimsleur review: the money math nobody does
Four reviews quote ten different Pimsleur prices and none tells you what to pay. I read Pimsleur's own catalogue for all 52 languages instead.
How much is Super Duolingo? Nobody can tell you, including Duolingo
Six Super Duolingo prices are listed on Apple's App Store with no billing term attached to any of them. That is why three top results give you three different answers.
How to record a language lesson
Where to put the phone, what to say to your tutor, the length and file limits, and how to tell within a minute whether a recording worked.
What Duolingo costs, according to Duolingo
Two Duolingo-owned pages disagree on the trial length and on what the free tier limits. Neither publishes a price. Here is what I could actually verify.
The 90% retention target is an artifact of the objective
Anki's manual says the tool that computed your optimal retention was removed in 25.07. Swap the objective function and the answer moves from 0.70 to 0.95.
The ten-minute post-lesson review, and why yours never happens
The standard post-lesson review asks for seven decisions before you review a single word. The top-ranked guide blames your method. It has that backwards.
Anki's actual default settings, from the source
FSRS is off by default. Leeches are tagged, not suspended. Every deck option default in Anki 26.08.1, with the line of source code each one comes from.
Study abroad has never had a randomised trial
Moore 2021 screened 2,548 records and found 6 eligible studies, all quasi-experimental. Yang found 11. Kim found 14. Not one randomised trial among them.
Day 28, 192 cards, and the day the queue peaks
A 28-day run of a Verbamor deck simulated against the real scheduler: card counts, the peak day, and what you can check on day 28.
The evidence says translate. Your teacher says don't
Glossed reading beats unglossed 33.4% to 19.8% a week later. And the famous translation study lost most of its edge when someone re-ran it properly.
FSI's Category IV hours don't divide by FSI's own week
FSI states a 23-hour class week. Three of its four categories divide by 23 exactly. Category IV divides by 25, and the two figures differ by 176 hours.
Spanish, French, Italian or Portuguese: picking one
FSI puts Spanish and French at 30 weeks and Italian and Portuguese at 24. Here is what that six-week gap is made of, and how to choose.
One research team tested 10 rival language apps
Duolingo, Babbel, Busuu, Rosetta Stone, Pimsleur, italki and more all cite the same statistician. Every report names its funder. The comparison table drops that column.
A semester in Peru buys about 300 hours of Spanish
Immersion is four separate ingredients. Three of them are purchasable at home. Here is the research, the arithmetic, and what to do tonight.
Why you forget words after a language lesson
You understood every word in the lesson and lost most of them by Thursday. Three different failures cause that, and they need different fixes.
The input hypothesis still cannot be tested
Free choice d = 0.22, limited choice d = 0.73. No accountability d = 0.01. The doctrine's two purest ingredients are the two the data likes least.
Is Anki worth it? A straight answer for Spanish and French
Every page tells you Anki is powerful and then hands the decision back. Here's the actual cost in dollars and in hours, and who should buy it.
Every FSRS parameter, w0 to w20, in plain English
All 21 FSRS-6 parameters: default value, legal clamp range, and what each one does to your schedule, with the arithmetic worked out.
"8,000 words" means 8,000 word families
Nation's 8,000-9,000 figure counts families, not forms. Running the arithmetic on his own type counts gives 34,660 to 37,713 distinct written forms.
How to learn Spanish with your child
Two separate systems, one shared song, and the exact rules. Prices, runnable FSRS code, and a note you can hand to the other adult tonight.
Ebbinghaus never said you forget 90% in a week
The savings table from Chapter VII, the four replication subjects, and what Ebbinghaus's own equation returns when you solve it for 90 percent forgotten.
Missed a week of flashcards: a 4 day recovery
The four-day plan for a flashcard backlog, with the numbers to check each morning and hard rules for the cards that were already failing.
Duolingo's 10-K defines the numbers blogs get wrong
133.1M monthly active users, 52.7M daily. The 10-K defines both terms in a sentence each. Four years of filings, and the arithmetic nobody publishes.
The dementia claim we could market, and will not
Brini 2020: dementia diagnosed 3.3 years later in bilinguals, prediction interval -1.9 to 8.5, and no reduced risk of developing it. The inference to app users fails.
Interleaving has a negative effect size on words
The interleaving meta-analysis reports 0.42 overall and -0.39 for words. Blocking wins on the one material a vocabulary learner cares about.
The 66-day habit number is a median of 39 people
96 recruited, 82 analysable, 39 kept. The famous range runs 170 days past the last day anyone was observed. Read from the paper itself.
Verbamor iPhone requirements and storage
The minimum iOS version, the oldest iPhone that runs it, what a deck actually costs in storage, and the two permission prompts you will see.
Anki alternatives on iOS, honestly compared
AnkiMobile is $24.99 once and free on every other platform. The honest case for leaving, and the bigger case for staying.
Cloze deletion flashcards with audio
How a cloze card with a tone in place of the answer trains spoken recall, with the studies behind it and a four-week review schedule you can check.
The cheapest way to learn Spanish, per hour
Tutors, apps, Anki and courses, with the hours priced beside the dollars. Cheap per month and cheap per hour are different numbers.
Why you cannot speak after 900 days of Duolingo
A 900-day app streak is roughly a seventh of what FSI spends on Spanish, and the published outcome research only measured reading and listening.
How to explain spaced repetition in 2 minutes
A word-for-word script for explaining spaced repetition to a friend, the five questions they ask back, and how to learn alongside them.
Capture habits after your first 500 cards
Past 500 cards, saving more words makes your deck worse. Four capture rules, a weekly audit, and the settings that decide your daily load.
Picking between the monthly and annual plan
The break-even is 6.5 months. Here is the rule for choosing monthly or annual, what switching actually costs, and what you give up by starting monthly.
How to remember words from a conversation
A recorded lesson holds the words you failed at. Here is how to pull them out, which ones to card, and what the deck looks like after four weeks.
Anki and language apps: what each one costs you
AnkiMobile costs $24.99 once. Verbamor's annual plan works out to $10.83 a month. Here's the 10-card test that tells you which bill is worth paying.
The age 17.4 finding does not mean what you were told
Hartshorne studied 669,498 people and found a rate change at 17.4. The number does not mean what the internet says it means.
Your tutor taught you 15 words. You kept 3.
italki admits the retention half of your lesson is your unpaid job, then hands you a tool that shut down in 2016. Here's the method it skips.
Why you understand the show but not the person at the counter
You follow 50 minutes of television, then a barista says six words and you're gone. Conversation runs 222 words a minute against a lecture's 173, and the words lose their edges.
When you are actually ready for your first novel
The test takes one page and a tally. Nation put a number on the answer: 8,000 to 9,000 word families reads a novel, and even then 1 word in 50 is new.
How many flashcards a day should you actually do?
Set 50 new cards in week 1 and by week 6 you open the app to 400 reviews and close it again. The new-card number is a standing order, not a daily choice.
Is it too late to learn a language at 40?
669,498 people gave the famous answer: a change at 17.4 years. Read the paper and it measures how fast you learn grammar, not whether you still can.
How many words you actually need before the trip
The counts people quote are for reading. Nation measured speech separately, and it is cheaper: the first 1,000 word families cover more of a conversation than of a page.
Why you know the word in the app and not at the restaurant
You knew servilleta at 8am on the sofa. At 9pm with a waiter waiting, nothing. The word is not missing, it is filed somewhere the restaurant cannot reach.
How to pick a first show that is actually at your level
Every list sends beginners to Narcos. Researchers counted the vocabulary each genre demands, and the gap between the easiest and hardest runs to thousands of words.
The production effect: why saying it out loud beats reading it
Reading a word aloud made it far easier to recognise later. Then MacLeod ran the version where everything was read aloud, and the whole advantage disappeared.
What to ask for in your first lesson so lesson 2 is better
Researchers coded 686 corrections in real classrooms. The one teachers reach for most, repeating your sentence back fixed, worked the worst. Ask for the other one.
Subtitles: target language, native, or none
Dutch viewers who read native-language subtitles got 17% worse at hearing new speech. Viewers reading target-language subtitles got 19% better.
Graded readers work, for the opposite reason they are sold
25 new words went into one graded reader. About 1 was still there 3 months later. The books still earn their place, just not for the reason on the back cover.
Do bilingual brains age slower? What 86,149 people showed
A study across 27 European countries found multilingual speakers were 2.17 times less likely to show accelerated brain aging. The effect scaled with the number of languages.
The best apps for learning Spanish, by what you're actually doing
Duolingo, Babbel, Anki, Pimsleur, italki and Verbamor, sorted by which one you should buy. For most people reading this, it isn't Verbamor.
The best apps for learning French, by what you're actually doing
Pimsleur, Babbel, Lingopie, italki, Anki and Verbamor, sorted by the real French problem: you can read it and you can't hear it.
The best spaced repetition apps, and how their schedulers differ
Most roundups on this query can't tell SM-2 from FSRS. Here is what each app's scheduler actually does, sourced, with Anki named first.
If you already have a tutor, most language apps are the wrong tool
Duolingo, Babbel, Pimsleur, Anki and Verbamor, judged on one question: does this help with the lesson you already paid for?
The pretesting effect: why guessing before you learn works
Duolingo shows you a word you have never seen and asks you to pick its picture. You get it wrong. That wrong guess teaches you more than reading the answer would have.
The testing effect: why recalling a word beats rereading it
Rereading is the world's favorite study method, and one of its weakest. Pulling the answer out of your own head, even clumsily, does more for a memory than 10 more looks at the page.
Sleep and memory: why your vocab consolidates overnight
The word you half-learned tonight gets replayed, filed, and judged somewhere around 3am. Skip the sleep and you skip the filing.
Interleaving: why mixing your words beats blocking them
Studying one theme at a time feels tidy. Shuffling everything together feels like a mess, and it doubled test scores in Rohrer's classrooms.
Dual coding: why images make vocabulary stick
A word stored with an image lives in two memory systems, and either one can rescue the other mid-sentence. The research, plus one wry catch.
The 20-minute rule: why cramming before a lesson backfires
You block out an hour before your Tuesday session and drill the whole deck. It feels productive. It's quietly the least efficient thing you could be doing.
The word-frequency curve: why 650 words does so much
Spanish has maybe 100,000 words. You'll use about 1,000 of them for nearly everything you say. Zipf's law is the best deal in language learning.
Why your brain forgets Tuesday's vocab by Friday
The forgetting curve is old, brutal, and remarkably precise. Here's what Ebbinghaus figured out in 1885, and why it still runs your review schedule.
Comprehensible input vs. flashcards: do you need both?
One camp says just watch shows. The other grinds decks like a part-time job. The research has an answer, and it'll mildly annoy both camps.
How to grade yourself honestly on a flashcard
The scheduler believes every grade you give it, and a deck full of generous grades quietly falls apart. A rubric for the moment after the flip.
Desirable difficulty: why the struggle is the point
Spacing, testing, interleaving: every method this blog recommends feels worse than its alternative. One framework explains why the worse-feeling one keeps winning.
Cognates and false friends: your unfair head start
You walked into Spanish already knowing thousands of words. The research says they're cheaper to learn and slower to forget. A few of them are lying to you.
Sound before spelling: why native audio matters
A word is a sound long before it's a spelling. Learn it from text first and your English reading habits will pronounce it for you, badly, forever.
Speaking anxiety: getting through your first tutor sessions
Your heart rate has no business being this high over a video call about the weather. The dread is real and measurable, and specific things shrink it.
How FSRS schedules your reviews (under the hood)
Every card in your deck is secretly three numbers. What they mean, how your grades move them, and why this algorithm earned the job on a billion reviews.
Learning plateaus: why progress stalls and how to push through
Month 2, every week felt like a level-up. Month 8, the needle just sits there. You haven't stopped improving; you've stopped being able to see it.
Nothing here yet. Try another topic, or clear the search and start from all of them.