Today’s learners have more resources than any generation before them, and often move more slowly than someone a century ago with a single book. This piece explains why as arithmetic rather than as a lecture about focus: vocabulary is a frequency game, and variety flattens the frequency curve.
Look at the phone of anyone learning a language this year. You will likely find three vocabulary apps, two subscribed channels, an online course bought last year and abandoned at lesson three, four groups, a downloaded roadmap, and two books read to chapter three.
That person has more resources than any learner in history. And they are quite possibly moving more slowly than someone a century ago with one coursebook and one novel.
This is not a moralising observation about focus. It is an arithmetic problem, and this piece sets out the arithmetic.
Start with a question: how many times must you meet a word before you actually know it?
Rott (1999) had learners read controlled texts and varied the number of occurrences. Two encounters produced almost nothing; six encounters produced substantially more, and the effect held on a delayed test weeks later.
Webb (2007) measured more finely, separating the aspects of word knowledge, and found that around ten encounters were needed for meaningful growth across several aspects at once.
And the largest synthesis of the question — Uchihara, Webb and Yanagisawa (2019) — confirms a consistent relationship between repetition and learning, while noting that its strength varies with context and test type.
The exact number matters less than the shape. Learning a word is not an event. It is a process requiring many encounters spread across time. Which brings us to the sum.
Suppose you have a hundred pages to read this month. There are two ways to allocate them.
Option one: ten different sources, ten pages each. A coursebook chapter, some blog posts, the first two chapters of a novel, an article, a short story.
Option two: a hundred continuous pages of the same author, or the same series, or the same subject.
Both are a hundred pages. Both cost the same hours. But they are not equivalent, because the repetition rate of the vocabulary differs enormously.
Under option one, each source has its own lexical stock. A word appearing three times in author A’s ten pages will scarcely appear in the other nine sources. You meet it three times and stop — below Rott’s threshold.
Under option two, the same author recycles their own vocabulary. The same subject drags in the same semantic field. The same characters appear in the same kinds of situation. That word may appear ten or fifteen times across a hundred pages.
Same hours. Four or five times the encounters.
This has a name and a literature. It is called narrow reading, and Krashen (2004) has argued for it hardest, against the conventional advice to read widely.
His argument is simple: read many works by the same author, or many books in one series, or many articles on one topic, and each successive item is easier than the last — because you already have the vocabulary, the context, the voice.
Schmitt and Carter (2000) set out the lexical advantages of narrow reading systematically. Gardner (2008) supplied hard corpus evidence: analysing authentic children’s reading materials and measuring vocabulary recycling, he found that thematically related texts recycle vocabulary at markedly higher rates than unrelated texts.
And there is a classic case study worth naming because it is memorably concrete. Cho and Krashen (1994) followed adult English learners reading the Sweet Valley Kids series — books of no literary pretension whatever. The learners read one after another in the same series. They acquired substantial vocabulary, and more importantly, they kept reading, because each book was easier than the last.
That is the crux. Narrow reading does not only raise vocabulary frequency. It also manufactures the sensation of progress — and the sensation of progress is what keeps people going.
It also fits Hu and Nation’s (2000) coverage threshold: roughly 98% known words for comfortable reading. Jumping between sources means falling below the threshold at every jump. Staying inside one means climbing above it.
Here most articles would cite the jam study. I will cite it, and also cite the rebuttal.
Iyengar and Lepper (2000) set up a jam tasting stall in a supermarket, sometimes with twenty-four varieties, sometimes with six. More people stopped at the twenty-four, but the purchase rate at the six-jam display was many times higher. The study became the foundation of an entire popular genre about the paradox of choice.
However:
Scheibehenne, Greifeneder and Todd (2010) meta-analysed the choice-overload literature and found a mean effect size close to zero. In many settings, more options do no harm at all.
Does the argument collapse? No — but it has to be stated more precisely.
Chernev, Böckenholt and Goodman (2015) ran the follow-up meta-analysis and identified the conditions under which overload actually bites. Four moderators:
Choice set complexity. When no option clearly dominates.
Decision task difficulty. When there is time pressure, or too little information to compare.
Preference uncertainty. When the chooser does not yet know what they want.
Decision goal. When the aim is to decide correctly rather than merely to browse.
Now hold that against a beginner in a new language. No app clearly dominates. They lack the knowledge to compare quality. They do not know what they need, because they do not know the language. And they are trying to choose correctly, not to browse.
The beginning language learner hits all four conditions. That is why the argument holds here even though the general effect is weak.
There is a second mechanism, subtler and less known.
Kruglanski and colleagues’ goal systems theory (2002) describes what happens when one goal is connected to many means. Intuition says more routes are better. The experiments say otherwise.
Zhang, Fishbach and Kruglanski (2007) set out what they call the dilution model: when a goal is attached to multiple means, each means is perceived as less instrumental for that goal. The association gets divided.
Applied to language learning this is rather sharp. When you have one app, that app is your route to English. When you have six, none of them is the route. Each becomes one of six — and when you skip a session on one, the other five are still there to reassure you.
That is how abundance corrodes commitment. Not through confusion, but because nothing becomes essential.
There is a phenomenon everyone recognises once it is pointed out: the satisfaction of saving a resource feels remarkably like the satisfaction of having learned from it.
Sparrow, Liu and Wegner (2011), in Science, found that when people believe information will be stored and accessible later, they remember the information less well — but remember where to find it better. Memory outsources.
An honest note: replication attempts on this have been mixed, so it should not be treated as law. But the surface phenomenon is recognisable to anyone with an overflowing bookmarks folder.
Bjork, Dunlosky and Kornell (2013) describe the underlying mechanism: learners systematically misjudge what is working, and ease gets mistaken for progress. Downloading a course feels like a good decision made. It is not learning; it is the precondition for learning, and preconditions are free.
This is the strongest objection to the whole article, and it deserves a proper answer.
The interleaving research — Rohrer and Taylor (2007) and the desirable-difficulties work — shows that mixing problem types produces better long-term results than working in blocks. So is variety not good?
The distinction is in what is being varied.
Interleaving means varying problem types within one body of material: mixing three kinds of question inside the same chapter. It does not mean using five different textbooks.
Carvalho and Goldstone (2015) clarify precisely this: interleaving helps learners discriminate between similar categories, while blocking helps them find what is common within a category. Two different jobs, two different arrangements.
For language learning that reduces to a compact rule: interleave the skills, do not interleave the sources.
Within a session, moving between listening, reading and self-testing is good. That is interleaving. Using app A this week, switching to app B next week, buying a new course next month — that is not interleaving. That is starting over.
People often invoke task-switching costs here. I will not, because those studies measure millisecond switches in a laboratory, and extrapolating them to changing apps monthly stretches the evidence too far.
The real cost is much more concrete, and needs no psychology to see.
Re-orientation cost. Every new source has its own organisation, terminology and assumptions about what you already know. The first hours with a new resource are hours spent learning the resource, not the language.
Re-covered ground cost. Almost every course starts from the beginning. Switching at lesson twelve means redoing eleven lessons in a different format.
Lost-thread cost. Cepeda and colleagues (2006) show that spaced review works by returning to the same material. If you never return because you have moved on, you forfeit the spacing effect — the best-evidenced finding in learning science.
These three are not subtle. They are merely invisible, because every switch feels like a promising fresh start.
Here is the prescription, and it is short.
One main source per skill, at a time. One coursebook or course. One listening source. One reading source. Three, not thirteen.
Narrow reading rather than wide, in the early stages. Same author, same series, same subject. Wide reading is the reward of a later stage, once the vocabulary base is thick enough to absorb the jumps.
Finish one thing before buying the next. Not out of discipline, but because the most valuable part of a course is its second half, where it starts reusing what it taught.
Keep a later list, and do not open it. Encountering a good resource does not mean starting it today. Note it, close the tab, return to what you are using.
It would be dishonest to say only do not switch. There are cases where switching is right, and they are specific.
When the level is wrong. If you cannot understand it, that is not a willpower problem. Per Nation (2006), material well above your level teaches nothing; it teaches discouragement.
When you have exhausted the useful part. An A2 coursebook for someone at B1 is revision, not learning.
When it trains the wrong skill. If the goal is speaking and your source is entirely reading comprehension, that is a legitimate reason.
All three rest on a judgement about content. What is not a legitimate reason is boredom, or seeing someone online recommend something new.
One image gathers up the whole argument.
Earlier learners, as the polyglots piece on this site discusses, often had one or two books. They read them repeatedly. They memorised passages. They knew those books far too well.
We look at that and see deprivation. But by the arithmetic at the top of this article, it was a favourable condition: every word in that book was met dozens of times, and every structure repeated until reflexive.
Today’s abundance does not give us more. It gives us the same hours, spread thinner — and because vocabulary is a frequency game, thinner means less.
What you lose with twenty resources is not focus. It is the seventh encounter — the one where a word would have moved from recognised to usable, and which never arrives, because you left for another source after the third.
The same article, told in plain words — for younger readers, or for anyone who wants the point quickly.
Look at the phone of anyone learning a language this year. You will likely find three vocabulary apps, two subscribed channels, an online course bought last year and abandoned at lesson three, four groups, a downloaded roadmap, and two books read to chapter three.
That person has more resources than any learner in history. And they are quite possibly moving more slowly than someone a century ago with one coursebook and one novel.
This is not a moralising observation about focus. It is an arithmetic problem, and this piece sets out the arithmetic.
Start with a question: how many times must you meet a word before you actually know it?
Rott (1999) had learners read controlled texts and varied the number of occurrences. Two encounters produced almost nothing; six encounters produced substantially more, and the effect held on a delayed test weeks later.
Webb (2007) measured more finely, separating the aspects of word knowledge, and found that around ten encounters were needed for meaningful growth across several aspects at once.
The largest synthesis — Uchihara, Webb and Yanagisawa (2019) — confirms a consistent relationship between repetition and learning, while noting that its strength varies with context and test type.
The exact number matters less than the shape. Learning a word is not an event. It is a process requiring many encounters spread across time.
Suppose you have a hundred pages to read this month. There are two ways to allocate them.
Option one: ten different sources, ten pages each. A coursebook chapter, some blog posts, the first two chapters of a novel, an article, a short story.
Option two: a hundred continuous pages of the same author, or the same series, or the same subject.
Both are a hundred pages. Both cost the same hours. But they are not equivalent, because the repetition rate of the vocabulary differs enormously.
Under option one, each source has its own lexical stock. A word appearing three times in author A's ten pages will scarcely appear in the other nine sources. You meet it three times and stop — below Rott's threshold.
Under option two, the same author recycles their own vocabulary. The same subject drags in the same semantic field. The same characters appear in the same kinds of situation. That word may appear ten or fifteen times across a hundred pages.
Same hours. Four or five times the encounters.
This has a name and a literature. It is called narrow reading, and Krashen (2004) has argued for it hardest, against the conventional advice to read widely.
His argument is simple: read many works by the same author, or many books in one series, or many articles on one topic, and each successive item is easier than the last — because you already have the vocabulary, the context, the voice.
Schmitt and Carter (2000) set out the lexical advantages systematically. Gardner (2008) supplied hard corpus evidence: analysing authentic children's reading materials and measuring vocabulary recycling, he found that thematically related texts recycle vocabulary at markedly higher rates than unrelated texts.
And there is a classic case study worth naming because it is memorably concrete. Cho and Krashen (1994) followed adult English learners reading the Sweet Valley Kids series — books of no literary pretension whatever. The learners read one after another in the same series. They acquired substantial vocabulary, and more importantly, they kept reading, because each book was easier than the last.
That is the crux. Narrow reading does not only raise vocabulary frequency. It also manufactures the sensation of progress — and the sensation of progress is what keeps people going.
It also fits Hu and Nation's (2000) coverage threshold: roughly 98% known words for comfortable reading. Jumping between sources means falling below the threshold at every jump. Staying inside one means climbing above it.
Here most articles would cite the jam study. This one will cite it, and also cite the rebuttal.
Iyengar and Lepper (2000) set up a jam tasting stall in a supermarket, sometimes with twenty-four varieties, sometimes with six. More people stopped at the twenty-four, but the purchase rate at the six-jam display was many times higher. The study became the foundation of an entire popular genre about the paradox of choice.
However: Scheibehenne, Greifeneder and Todd (2010) meta-analysed the choice-overload literature and found a mean effect size close to zero. In many settings, more options do no harm at all.
Does the argument collapse? No — but it has to be stated more precisely.
Chernev, Böckenholt and Goodman (2015) ran the follow-up meta-analysis and identified the conditions under which overload actually bites. Four moderators: choice set complexity, when no option clearly dominates; decision task difficulty, when there is time pressure or too little information to compare; preference uncertainty, when the chooser does not yet know what they want; and decision goal, when the aim is to decide correctly rather than merely to browse.
Now hold that against a beginner in a new language. No app clearly dominates. They lack the knowledge to compare quality. They do not know what they need, because they do not know the language. And they are trying to choose correctly, not to browse.
The beginning language learner hits all four conditions. That is why the argument holds here even though the general effect is weak.
There is a second mechanism, subtler and less known.
Kruglanski and colleagues' goal systems theory (2002) describes what happens when one goal is connected to many means. Intuition says more routes are better. The experiments say otherwise.
Zhang, Fishbach and Kruglanski (2007) set out what they call the dilution model: when a goal is attached to multiple means, each means is perceived as less instrumental for that goal. The association gets divided.
Applied to language learning this is rather sharp. When you have one app, that app is your route to English. When you have six, none of them is the route. Each becomes one of six — and when you skip a session on one, the other five are still there to reassure you.
That is how abundance corrodes commitment. Not through confusion, but because nothing becomes essential.
There is a phenomenon everyone recognises once it is pointed out: the satisfaction of saving a resource feels remarkably like the satisfaction of having learned from it.
Sparrow, Liu and Wegner (2011), in Science, found that when people believe information will be stored and accessible later, they remember the information less well — but remember where to find it better. Memory outsources.
An honest note: replication attempts on this have been mixed, so it should not be treated as law. But the surface phenomenon is recognisable to anyone with an overflowing bookmarks folder.
Bjork, Dunlosky and Kornell (2013) describe the underlying mechanism: learners systematically misjudge what is working, and ease gets mistaken for progress. Downloading a course feels like a good decision made. It is not learning; it is the precondition for learning, and preconditions are free.
This is the strongest objection to the whole article, and it deserves a proper answer.
The interleaving research — Rohrer and Taylor (2007) and the desirable-difficulties work — shows that mixing problem types produces better long-term results than working in blocks. So is variety not good?
The distinction is in what is being varied.
Interleaving means varying problem types within one body of material: mixing three kinds of question inside the same chapter. It does not mean using five different textbooks.
Carvalho and Goldstone (2015) clarify precisely this: interleaving helps learners discriminate between similar categories, while blocking helps them find what is common within a category. Two different jobs, two different arrangements.
For language learning that reduces to a compact rule: interleave the skills, do not interleave the sources.
Within a session, moving between listening, reading and self-testing is good. That is interleaving. Using app A this week, switching to app B next week, buying a new course next month — that is not interleaving. That is starting over.
People often invoke task-switching costs here. This article will not, because those studies measure millisecond switches in a laboratory, and extrapolating them to changing apps monthly stretches the evidence too far.
The real cost is much more concrete, and needs no psychology to see.
Re-orientation cost. Every new source has its own organisation, terminology and assumptions about what you already know. The first hours with a new resource are hours spent learning the resource, not the language.
Re-covered ground cost. Almost every course starts from the beginning. Switching at lesson twelve means redoing eleven lessons in a different format.
Lost-thread cost. Cepeda and colleagues (2006) show that spaced review works by returning to the same material. If you never return because you have moved on, you forfeit the spacing effect — the best-evidenced finding in learning science.
These three are not subtle. They are merely invisible, because every switch feels like a promising fresh start.
One main source per skill, at a time. One coursebook or course. One listening source. One reading source. Three, not thirteen.
Narrow reading rather than wide, in the early stages. Same author, same series, same subject. Wide reading is the reward of a later stage, once the vocabulary base is thick enough to absorb the jumps.
Finish one thing before buying the next. Not out of discipline, but because the most valuable part of a course is its second half, where it starts reusing what it taught.
Keep a later list, and do not open it. Encountering a good resource does not mean starting it today. Note it, close the tab, return to what you are using.
It would be dishonest to say only do not switch. There are cases where switching is right, and they are specific.
When the level is wrong. If you cannot understand it, that is not a willpower problem. Per Nation (2006), material well above your level teaches nothing; it teaches discouragement.
When you have exhausted the useful part. An A2 coursebook for someone at B1 is revision, not learning.
When it trains the wrong skill. If the goal is speaking and your source is entirely reading comprehension, that is a legitimate reason.
All three rest on a judgement about content. What is not a legitimate reason is boredom, or seeing someone online recommend something new.
One image gathers up the whole argument.
Earlier learners often had one or two books. They read them repeatedly. They memorised passages. They knew those books far too well.
We look at that and see deprivation. But by the arithmetic at the top of this article, it was a favourable condition: every word in that book was met dozens of times, and every structure repeated until reflexive.
Today's abundance does not give us more. It gives us the same hours, spread thinner — and because vocabulary is a frequency game, thinner means less.
What you lose with twenty resources is not focus. It is the seventh encounter — the one where a word would have moved from recognised to usable, and which never arrives, because you left for another source after the third.
These articles summarize well-established research in learning science and linguistics. Key sources and further reading:
The question assumes focus is a muscle you train. The strongest evidence says the training barely carries over, and the...
No time, no talent, too old — those are explanations, not causes. What can be measured is duller: one sentence with a ti...
Not one event but four, running on clocks a billion-fold apart: a wave at 400 milliseconds, a word locked in overnight,...
