Why I stopped studying grammar (and what I do instead) - Steve Kaufmann

In this video I talk about why grammar instruction is not as important as we are taught to think and why you’re better off focusing on reading, listening, and speaking.


Thanks, Steve! This is is probably the most important SLA insight of the last 50 years.

Unfortunately, I know many people (both teachers and students) who
don’t understand that grammar is only (post-festum) meta-knowledge (i.e. knowledge about
a language), which is not identical to competence in a language

That’s somehow depressing bc. sometimes 2023 feels like pre-1975 (pre-Krashen) :frowning:


Ah, this naive pre-Krashen Paleolithic age! Theoretical savagery without sufficient data, grammatical torturing, premature speaking…

Btw, it has to be 48 P.K now.


Thanks for the video Steve. I hear you when you say, If I just learn this then I can do x…,

In my experience the grammar has only helped me understand when spoken to and when reading. In Finnish I could see/hear the suffix -maa, -mis and know it’s 3rd or 4th infinite. If I hear -miseksi, I could think this is 4th infinite translative in written variation or because I studied it I know that means (in order to…)

Comprehending for me is a combination of so many things but speaking comes much easier. I’ve learned enough spoken phrases to communicate fairly fluently, although it doesn’t help much if only understand 50% of what they said or if I just pick a couple of words from their sentence.

It seems so different than learning as a child. Everything I need to know is already engrained in my brain and I’m just rewiring or something. At least the embarrassment is the same, never make the same mistake twice!

1 Like

The comparison is always on how grammar was taught and it is still taught in the vast majority of schools. However, it doesn’t mean that if improve the system grammar is useless.

Nobody cares about focusing on writing, but it is not so easy to write decently if you don’t know anything about grammar. You can see the vast majority of native speakers in any country, how much poorly they write in their own language. But they have been exposed all the time to their own language, and they haven’t developed any easy capability to know rules and understand patterns!

If the focus is reading, or listening, there is no problem with grammar. If the focus is speaking we definitely need to start doing some work about it. If the focus is writing…

The reality is that “we” have written grammar books when we didn’t know a lot of things, and we have continued to repeat the same principles adding rules, and exceptions, and more and more exceptions.

Most of the grammar we study is absurd, there are more exceptions than correct rules, and that’s a sign of how much wrong are those rules. In Italian some expressions are considered wrong, and in Spanish the same expressions are considered correct!

Those rules are just opinions, not real knowledge! Probably the real knowledge comes from the sound system, and how we communicate at the core of our brain functionalities.

That is the main problem, we should reform grammar, and we should reform when and how we introduce grammar studies in the language learning process. But of course, nobody cares because schools are not meant to do so!


I’d say it’s enough to use the existing grammar differently
(e.g. in grammar light approaches or as meta-knowledge used on
an as-needed basis in combination with a lot of listening / speaking /
reading and writing).

The “real game changer” could then be LLMs with which we can interact
continually in the oral and written dimensions…

Apart from that, I don’t see that humans learn languages by memorizing
language rules. Instead, we usually use statistics ourselves . And this works
because of the vast number of conventionalized word groups (aka “collocations”)
in our natural languages.

Maybe the only difference between generative AIs and us is this:

  • The former are stochastic parrots without meaning that “simulate” human communication.
  • We humans, on the other hand, are stochastic parrots with meaning :slight_smile:
    In both cases, statistics rule in processing natural languages.

Compared to this, grammar rules seem to be a completely artificial crutch with limited use cases…


I haven’t studied grammar for years now, at least not much. I had to look at certain points that I didn’t understand, but for the most part, I’m just reading and listening. I studied grammar for years in Spanish and got nowhere except for understanding the structure of the language, then 20 years later I found Stephen Krashen, Steve Kaufmann, Olly Richards as well as others and I actually started acquiring the language. So while learning French, I don’t plan on studying grammar. So far I haven’t and I have learnt a lot within the last 3 weeks :slight_smile:


I agree that the ideas popularised by Krashen provide a much better description of language learning than traditional class room based methods based on rote learning tables of grammar. I doubt there is any dispute there. (1) Where I differ is in the details.

With French I find the grammar mostly straightforward. When I come across an expression I don’t understand, I just look it up. So aller en justice and Il lui manque deux doigts à la main gauche are simple examples. For me French almost seems like a close cousin to English, so many words are shared, and the sentence structure and phraseology are similar.

However, I find German impenetrable without recourse to a grammar guide. It is so alien. I’m not learning tables of adjective case endings, but I am learning example sentences, and I do occasionally refer to a book when something is unclear. The focus is on reading and listening, but I cannot pick up the grammar passively.

A child is exposed to massive amounts of input which allows them to construct the rules subconsciously. Use of a grammar guide can help accelerate that process in the case of an adult who already has at least one language.

(1) Obviously Steve Kaufmann is creating videos to publicise LingQ, hence he has to address traditional methods, and give the big picture i.e. LingQ is better than traditional methods.


Hi Leif,

I’d say it’s not about avoiding grammar completely (bc. there are certain specific use cases where it makes perfect sense to resort to grammar books / exercises, use flashcards to learn grammar concepts / conjugation patterns, etc.).

However, the criticism relates to the idea that you first need grammar (rules) in order to be able to speak/write (= the idea that you need abstract meta-knowledge first to develop practical competence). That’s not how first or second language acquisition works because this is just a variant of essentialist thinking with regard to the relationship between the abstract and its empirical realizations.
Besides, I don’t think that native speakers construct “rules” (and exceptions to rules and exceptions to exceptions to rules, etc.) consciously or subconsciously either because that’s also based on “essentialist” thinking (see,for ex., in the context of complexity research: Part 4-(Social) Complexity Basics: Features of Complexity and the Scalability Problem | by Peter Bormann | (Social) Complexity | Medium).

It (probably) makes more sense to try to explain language acquisition processes using statistics (statistical inference) based on massive exposure to a language plus non-essentialist strategies.

Apart from that, I don’t think that Krashen is able to explain how language acquisition actually works (using Chomsky and an input-output model of communication). However, the (anti-Chomsky) idea of bottom-up, usage-based language development is awesome!

But that’s no longer common-sense territory (in and beyond SLA) :slight_smile:


Well, there are some things in German that are much easier than in English or French.
For example:

  1. The tenses (esp. of the past)
  2. The Konjunktiv, which is simpler in German than the equivalent “subjonctive” mode in Romance languages such as French.
  3. There is no difference between “simple” and “progressive” verb forms.

On the other hand, what is difficult in German is, for instance:

  1. The case system
  2. The problem of “trennbare Verben”
  3. The genders (der, die, das)
    and many nuances.

Ultimately, however, it boils down to knowing (tens of) thousands of conventionalized word groups (aka “collocations”). And this isn’t a problem specific to German, but to all languages because native speakers simplify communication processes in this way. And this is also one of the reasons why statistics works so well in language usage…

1 Like

Well, in German the verb likes to go walkabout, it can swap with the subject, or scoot off to the end of the sentence. Sometimes two verbs can swap round in a manner that I still do not understand. I asked a German why, and came away none the wiser. Apparently it is obvious. Then the vocabulary is so different. Anfangen, fahren, Ernährung bear no relation to English, unlike commencer, conduire and regime. About 60% of English comes from French and Latin. And French sentence structure is much closer to English. Then there is use of weird impersonal constructions such as Es tur mir leid.

In its defence, the meaning of many words such as vorstellen, vorteil and unternehman, can be guessed from the component parts. The case system is not so hard. I find the genders no harder than French, and about 80% are easy to work out. And the pronunciation and the timing (stress based) are much closer to English. And as you say, in all languages there are thousands of constructions to learn that do not come under the term grammar.


It’s great to look at grammar once you have a good or at least fair amount of vocabulary and feeling of the language. When you learn by absorption and massive input, you will get some feeling for some of the grammar rules. This depends a lot on the person. What Steve calls “your ability to notice” is vastly different for different people and different niches of what they are learning. After you have developed some vocabulary and some (perhaps vague) ideas of the grammar, learning grammar is a great way to sharpen this understanding and solidify what you have learned.

It’s terrible to begin with grammar and learn rules about words you really don’t know at all. It’s kind of like learning traffic rules without knowing what roads or vehicles are.


As I indicated, I think SK is simplifying as his videos are created essentially for marketing purposes. That’s not intended as a criticism.

Regarding the idea that you first need grammar rules in order to be able to speak / write, you say that is not how second language acquisition works. In which case, how does it work? In order for me to make sense of a simple sentence such as Ich habe dem Mann 5 Euro gegeben, I need to know the meaning of the components and their relationships. I could (according to many experts) listen intensively to massive amounts of comprehensible input (CI) and eventually work out the meaning. I would argue that that would be impractical for an adult, and perhaps impossible without complete immersion with suitable input which might not even be possible. (A Korean friend who lived in England for 30 years still does not understand aspects of English that are obvious to a 16 year old English speaker.) In practice it is far more usual for someone to explain the meaning, and then the learner listens to huge amounts of CI to enforce the meaning and usage. And they will in so doing gain a deeper insight, going beyond the simplistic grammar level description. Thus in French I know when to use certain words, but I don’t know why. In other words, I have acquired implicit knowledge. That in fact is how LingQ works. As far as I know no-one as an adult learns without first learning some grammar (in the loose sense of the word). The grammar lesson does not teach the language, it acts as a guide to the meaning, a framework. And learning tables of grammar is of course totally pointless.

Regarding how we learn a language, ChapGPT and others provide circumstantial evidence for the idea that the brain is a massively connected neural network, which is trained by exposure to huge amounts of input. And hence that Chomsky’s idea that the brain has language specific mechanisms is simply wrong. But that does not rule out the use of some grammar instruction as a guide for language learners.

As an aside, the way that ChapGPT works is incredible, I never thought we would have such capabilities so soon.


This issue is one I am thinking about (without firm results, necessarily), given that I am an English teacher in training, within a school system that requires students to master this-and-that bit of grammar by the end.

Introspectively, the way I personally learn today is with a lot of input (mostly via LingQ these days, especially in languages I am not all too great at yet), but my natural “nerdy” fascination almost automatically leads me to look up and research grammar and other theoretical aspects of the language such as phonology, the differences between printed and handwritten script etc. I think this is in some ways close to Steve’s approach, although he seems to have more tolerance for repetition in his input, whereas I tend to move on to harder material and look up things early on out of curiosity.

In school, that is not how I get to teach the students; I only get 3-4 lessons with them per week, and while I have great freedom in how I design my lessons, I still am expected to “deliver” students who know and can apply specific rules of grammar etc.

I think I’m somewhat in luck as an English teacher because unlike with just about any other foreign language, the majority of students come in with a huge amount of passive knowledge. Essentially, the whole world (and particularly the internet) is one massive bit of comprehensible English input. That’s not true of all students to the same extent, but it’s a massive help that my colleagues teaching French, Italian, Spanish, Russian, etc. don’t have. (There are tons of content available in those languages, of course, but only English is so omnipresent that the students cannot help but learn from exposure.)

So I get to frame grammar instruction at least a little like first-language grammar instruction: “Here’s the explanation behind what you’ve heard and read many times before, here’s how you can recognize and use it yourself consistently.” Still, grammar as mere auxiliary explanation doesn’t get all but the most invested students as far as quickly as is required of them. So memorization of rules, drill and other exercises remain a necessary part of class, and it’s very time-consuming, taking away time from more content-focused reading, conversation practice etc.


Replying to myself for a further thought:

My instructor in English Didactic Methods has an “extensive reading” system that really impressed me: she has a literature list (to which the students are free to add other books) and requires them to read so-and-so many entries per semester, in addition to the intensive reading in class. It’s quite an elaborate system and to be honest a little intimidating to organize, but I hope to be able to implement something like that once I get to teach longer-term classes!

(I have even toyed with the idea of getting students to use LingQ, but the expense is an issue there, and I would probably need approval from the school to use a third-party tool like that. Still, it’s an idea in the back of my mind…)

I’ve been studying French for almost a year now, ten months with LingQ, so I am junior to many here.

My two bits – I agree that the grammar-first approach ineffective. A key insight Steve offers is the link that grammar leads to testing.

I’d also switch that around and say that conventional education’s priority is to grade students via tests, which leads to emphasizing grammar since it is far more testable than the sort of immersion Steve recommends – even if grammar doesn’t lead to functional reading, listening and speaking skills.

That said, after six months of French, I felt the need to familiarize myself with verb conjugations, rather than crossing my fingers and waiting for exposure to enlighten me. I made out flashcards for the standard endings plus the Big Three Irregular Verbs – avoir, être, aller,

I didn’t try to hard-memorize them, but at times when I’m puzzled, I’ll pull a card out and do a quick review. I found this gave me more confidence and relaxed me in my learning.

I find I’m also consulting ChatGPT when I have more specific questions that bother me. It’s a great resource.

Same for me with Spanish. The issue arises with languages that are much further removed from English (assuming that’s your native language) than French and Spanish are. Honestly, for me, Spanish now feels like a dialect of English (it’s probably more accurate to say that English is a dialect of Spanish); there are soooooooo many cognates that it’s almost like cheating.

With some languages, there are NO cognates, and every single word has to be learned from scratch -there’s zero crutch. Add in “unusual” grammar structures, and you’re looking at an entirely different situation. You can probably still learn those languages without instruction/grammar explanations, but it’s going to take 3-4 times longer than the already REALLY long process with closely related languages to your mother tongue.

I feel like, with those “harder” languages, at some point along that looooong road, people just naturally give up on the ‘natural’ approach, which is completely understandable.


Sorry, I’m hard pressed for time at the moment. Therefore just a few hints:

  1. The “type (= proto-form) - token” logic that underlies traditional notions of rules (and their empirical applications) has probably been completely destroyed, esp. by Wittgenstein and Derrida.

This starts with basic processes of “categorization” (see e.g. Prototype theory - Wikipedia) and ends with all kinds of relationships between the abstract and its empirical variants. Today we tend to use paradoxical recursivity instead (Derrida, Niklas Luhmann, George Spencer Brown, Dirk Baecker, etc.), which leads to non-essentialist approaches.

See, as an appetizer, this post on Medium: Part 4-(Social) Complexity Basics: Features of Complexity and the Scalability Problem | by Peter Bormann | (Social) Complexity | Medium

  1. The role of statistics in (first / second) language acquisition: Statistical learning in language acquisition - Wikipedia

  2. Usage based model of language (which want to overcome Chomsky’s “universal grammar”) : https://en.wikipedia.org/wiki/Usage-based_models_of_languageapproach, esp. H-J. Schmidt’s entrenchment model based on “complex adaptive systems” (CAS) theory.

Insofar as CAS doesn’t make sense when applied to language (see, for ex., the section about the “mystery of the social (as the coordination of human behavior)” in my recent Medium post: Organizational Illusions. Part III of the series on… | by WAITS Software- und Prozessberatungsgesellsch. mbH | WAITS on Business Process Management | Medium), I’d use a socio-emergent approach (inspired by Niklas Luhmann and Peter Fuchs) reg. communication processes.

The latter also replaces traditional models of communication - in particular:

  • Concepts of “social mind” or other collective phenomena (such as “collective consciouness”, “team or group consciousness”, “collective memory”, “team / group memory”, etc.).
  • Tne sender-receiver / input-output model, which is appropriate for describing the signal-/data-processing of machines, but is too simplistic for human communication processes (For example, there is no “passive” reading, no “passive” watching of movies, no “passive” listening, etc., as our mind is always active in these cases → semper operantes sumus).
  • Approaches based on "inter"dimensions, esp. “intersubjectivity”: we’ve been waiting for more than 100 years (since Edmund Husserl) for a good explanation of what “inter” could mean applied to the social dimension. AFAIK, there is none.

Therefore, I’d say “socioemergent” approaches (Derrida, Niklas Luhmann, Peter Fuchs, Dirk Baecker, Ernesto Laclau, the theory of complex adaptivee systems - also in SLA research -, etc.) are more plausible at the moment when it comes to conceptualizing human communication.

In short, I’m trying to combine these four approaches:

  • non-essentialism / auto-recursivity (based on distinctions)
  • statistical learning
  • the entrenchment model of language (H.J. Schmidt)
  • socio-emergence (Luhmann / Fuchs)

And I’d like to compare this mix with statistical approaches in Machine Learning, esp. deep learning.

I’ve already written ca. 150 pages about this mix, but there’s still a long way to go.
Hopefully, I can finish this text in 2024 and present both an English and a German version.

Sorry if this sounds like gibberish, but that’s all I can say at the moment…

Happy holidays to you and the whole LingQ community,
Peter (out) :slight_smile:

Great idea!
However, why don’t you recommend the free version of ReadLang or
use LWT / Lute?

@PeterBormann well, Lute seems to be discontinued, LWT is too unreliable for a school without an inside developer, last update was 2 years ago.
ReadLang is not really like LingQ, it seems to me more based on flashcards, rather than LingQ more effectively based on extensive reading and listening. I don’t really understand what ReadLang could be useful for if one uses already LingQ extensively. It could be an add-on for something here and there, but not a substitute.

If LingQ’s team were more precise in developing this platform, the opened door for encouraging a business inside schools would be incredible.
They have, if I’m not wrong, a program for schools as well, and one could eventually ask for a massive discount for students, which is something I believe they already offer.
A teacher could verify the progress of each student within the system, and it would definitely be a vast better experience than what generally students are going through right now.