Purpose of Review

Owen’s recent post brought up the topic of optimizing education. One particular approach, Direct Instruction (Misha’s better explanation), claims to have essentially solved the problem. In particular, Direct Instruction (DI) does allegedly not only work for basic reading skills, but any teaching task. Owen brought up the Michel Thomas language courses as a good application. Language learning is one of my main interests, so I gave the French Foundation course a try.

The main point of the review is to summarize what Michel Thomas actually does, how it differs from other common paradigms and how effective it seems to me.

Summary: Nice for beginners and people with bad learning experiences; limited use afterwards. The audio-only aspect is very convenient. It complements other strategies well and I see it as a good proof-of-concept of DI-like methods for language learning.


Let’s start with a disclaimer. Michel Thomas (MT) is not officially a DI course and as far as I could google, Thomas propbably wasn’t aware of DI at all. However, according to Solity's The Language Revolution and Owen, the reason MT works so well is that it applies (an approximation of) DI techniques. It is right now the best realistic example beyond the grade school level, so it’ll have to do.

I had some French in high school and thanks to fluency in German and English, I can read some French, but I have no active skill at all, nor have I ever used French in a serious way.

I have now completed the first half of the French Foundation course and skimmed many other courses. You can listen to the first 20 minutes here; they are very representative. Furthermore you can read the booklets to get an idea of the material covered in each course. The whole course is audio-only, consists of 8 CDs (and 2 review CDs) and is intended to be listened to only once.1 There are several advanced courses which merely cover more grammar points and vocabulary. Structurally, they are all the same.


MT teaches the course to two new students2. You’re supposed to take the role of a third student, pausing the recording whenever MT asks a question so that you can say your own answer. One of the two students also answers and you can compare your reply and listen to MT’s advice and error correction. Both students are beginners, so most of your mistakes will be covered that way.

MT introduces one language component at a time and makes you use it in a given sentence. He provides a short explanation first and then lets the students answer a couple of examples by giving them an English sentence and asking them to translate it into French. Each component is thus reinforced through many examples.

MT also tries to combine the translation tasks over time by re-using partial sentences. This way, sentence quickly look complex, but always stay easy. (“What impression do you have of the political and economical situation in France at the present time?” is used about one hour in!)

Vocabulary is only introduced as necessary and relies heavily on cognates. The primary focus is on teaching structure. MT strongly emphasizes not to guess or try to remember anything, but instead to rely on induction (“Do not guess, but think it out!”). This works because the examples are carefully chosen to be as obvious as possible. All translation tasks have only one correct answer. All production is tightly controlled. MT relies on the constant tests to see that the students are successfully keeping up. He is never unsure if some concept has been understood or not.

Complex rules that might thematically belong together (like verb conjugations) are broken apart so that each individual new form or word is learned on its own. Similar rules that might be confused are deliberately spread out.

MT stresses that you aren’t supposed to try to remember anything. If you don’t know something, then he has not succeeded as a teacher yet and he will take care of it, not you. He does this by doing manual spaced repetition, i.e. he repeats previous questions (or similar ones) over time and tests the students constantly. If they have trouble answering, then he quickly goes back to the relevant lesson. This is of course how most language textbooks are supposed to be used, but they rely entirely on the student doing the testing themselves. Instead, MT provides the complete lesson including all necessary repetitions so the student doesn’t have to do anything at all except answer MT’s constant stream of translation tasks. (As a programmer, I’m strongly reminded of loop unrolling.)

What I stood out for me was the reward structure. Students rarely make big mistakes and actual correction is mostly needed for pronunciation issues. The major way students do fail is by simply not remembering something, which MT easily fixes by reminding them again. The students have good confidence in their answers and don’t have to guess. The lessons are fast-paced and consist mostly of tests. MT is constantly positively reinforcing the students, rarely correcting them. The whole lesson looks a lot more like an Anki session than a class room or a traditional textbook.

Comparison to other methods

The course is basically a (minimally edited) live class MT teaches. The result is a very natural pacing. This has the major advantage that it never goes too fast. Most other courses edit out mistakes or necessary repetitions out of fear they might be too boring, but by doing so, no student can actually keep up. This can’t happen with MT’s untrained live students. (Unfortunately, MT’s courses are also unscripted, so he does make a few organizational mistakes and the later courses don’t exactly fit together. Fortunately this is not a big issue due to MT’s large experience.)

A major difference to most other approaches is that MT actively implements what Krashen calls “i+1”, where i is the current level of a learner, meaning that concepts are taught in the order of minimal effort. Each new step contains exactly one new rule. Most language courses group rules according to some underlying pattern, like tenses, and expect you to learn a whole group at once.

MT focuses entirely on production, both by using only translation tasks and by teaching only useful components, i.e. parts of the language you need for a wide variety of contexts. No lesson has only one narrow use. This creates a very active learning experience. I fully agree with this early focus on grammar (but not grammar theory!). Once you’re done with that, you can go more-or-less monolingual and immerse yourself in the target language, relying on spaced repetition software to rapidly build your vocabulary.

Furthermore, MT’s course is very engaging. There is little downtime where you merely listen. It consists almost exclusively of quick tests. Thanks to i+1, you never have to juggle more than one new rule at a time. The subject matter does not get repetitive and MT is a very enthusiastic teacher. This can be a major problem with other language courses.

My main criticism, especially as an autodidact, would be that MT never makes his methods explicit. You entirely rely on him. He may have an awesome lesson plan, but you’re never taught how he arrived at it or how to continue beyond that.3 Hopefully that’s not a general problem with DI. In particular, any language course should teach you how to use spaced repetition. It’s the only sane way to handle vocabulary and prevent unnecessary review sessions.4

For contrast, look at the (excellent) Remembering The Kanji, which similarly teaches Japanese characters through decomposition, logical ordering and the use of mnemonics. However, much of the book focuses on teaching the method and the logic behind it, so that you can use it for any amount of characters you want. It is very simple to move beyond the scope of the book. I wish every textbook worked like this.


I’m quite impressed by the course design. It’s really effective at building a solid speaking foundation. It won’t get you anywhere near fluency and, being audio-only, totally ignores literacy, but by the end of the course you should have enough skill to actively engage the language.

After finishing MT, you should have a good grasp of the grammar. A good follow-up course might be something like Assimil (video overview), which would take care of literacy and fill in any remaining grammar gaps. After that, the only thing missing is vocabulary and general practice. This is the point where traditional language teaching ends, but graded readers, parallel texts and so on, combined with spaced repetition, solve this problem nicely. Or, you know, start talking, maybe on lang-8.

Personally, I plan to work through the full French, Spanish and Italian courses, and would recommend checking them out. Again, try listening to the preview to see if this approach appeals to you.


  1. MT recorded the whole course on one weekend, so listening once might work, but I find it too overwhelming. Spreading it out over a few weeks is probably the way to go.

  2. The students have quite a different aptitude for the language. <harsh>I like that one of them sucks; it makes me feel superior. I suspect this is intentional, but regardless, it certainly is rewarding. You don’t feel so bad about making minor mistakes or for forgetting something.</harsh>

  3. Further evidence for MT’s lack of meta-teaching is the poor quality of the courses produced after his death. They strongly diverge from his method and outright remove crucial features like the natural pacing.

  4. I converted the French Foundation course into an Anki deck based on the official booklet. It’s available as a shared deck in Anki (search for Michel Thomas) or as a tab-separated text file.

New Comment
17 comments, sorted by Click to highlight new comments since:

I should also note that MT was not merely poor at meta-teaching how to use his eponymous 'Method' itself; He was actively secretive about it.

I believe the reason he claimed was something like fear of the establishment stealing it and claiming credit? That doesn't seem to make much sense to me. Was that the 'real' reason, or a rationalization for some traumatic after-effect of his war experience, or what? Not really a question I'd expect a high chance of success or high returns on answering.


I assumed an active failure on MT's side because of comments like these (comment 10) by Cainntear, who is much more knowledgeable about MT than me. Quote:

[The new course design] is not the same method, although it looks deceptively like it.

It's not a matter of Thomas being a magician, but I think Splog's use of the word "tricks" is more relevant than even he realises.

I suggest that Thomas had a core method, and that he augments that with a couple of tricks, most notably mnemonics and cognates. Sadly, the subsequent teachers (and therefore most likely Thomas himself, given that he explicitly taught Goodman how to teach) appear to have mistaken the tricks for the method.

For one thing, in the Russian, Japanese and Polish courses, you'll hear more mnemonics in the first CD or 2 than you will hear in any of Thomas's own courses. It's not their fault -- Thomas did speak about mnemonics as being a core part of his method on the TV documentary The Language Master. The particular mnemonic he used there was for the French faire: to make or to do. "It's a fair thing to do," he said. But it is easily demonstrated that this was demonstrably unnecessary: I don't recall ever hearing him use an equivalent mnemonic in any of his other courses, but he still managed to teach tun, fare and hacer without problems.

Yeah, and all of what Cainntear was talking about later in that post with "teach confusable things separately" is covered in Chapter 10 of ToI, "Introducing coordinate members to a set".

[And yes, there's a typo in the table on the first page. Awareness of abstract feature "C" rules out example 6. It doesn't have to wait until feature "(D)" is brought to attention.]

I really think both the places where MT follows DI principles, and the places where he fails, should jump out to anyone familiar with both.

Which makes me wonder about how familiar Solity was with ToI, because I don't remember his exploration of the samenesses and differences as very good. How much of that recollection is due to my lack of familiarity at the time, I'm not sure, but I don't think all of it. Don't think I'm gonna reread it any time soon, though.

However, according to several DI proponents the reason MT works so well is that it applies (an approximation of) DI techniques.

Wait, 'several DI proponents'? Are you sure? Because I know of no-one in the DI world who is aware of MT (unless I were to count myself as properly 'in the DI world', which I do not yet).

The only place I found the connection was in the book "The Learning Revolution" by Jonathan Solity (2008). This was where I found the first reference to DI period, setting me off down this long path. However, it isn't really the focus of Solity's book itself, and I wouldn't recommend the book as useful to anyone who already knows about DI, especially not if they already know about MT.


You're right, I seem to have miscounted the proponents there, so it's just Solity's book and you. I edited the post.

Yes, and I should make clear that Solity didn't say, 'The MT courses work well, wherever they work well, due to approximating DI'. He presented DI more as one of many interesting little connected pieces (many of which were pretty much fluff), rather than as an overarching explanation.

The interpretation that, "If dalmatians are metaphorically the gold standard, then the MT courses are mangy mutts in an industry where everyone else is painting black spots and pinning floppy ears on chickens," is mine.

Are you and misha working on a top level DI post?

No, although I'd like to. I've just been really inactive the past couple weeks. Settling in to the internship and making sure I'm actually learning what I'm supposed to be learning there is still taking most of my energy during the week, and then I found out my mom had cancer (she just had a little bit, they got it out completely, and the chances of it coming back are apparently 'virtually nil' with just five weeks of radiation... but still, totally killed my productivity for one weekend), and then lazing around with a cold the next weekend. Yeesh.

The most I've been doing is poking at a post tentatively titled "A dry introduction to the empirical evidence on DI's effectiveness", essentially a summary of "Research on Direct Instruction", since I was feeling like maybe the best thing to do would be to take a step back and present a better explanation of why people should be interested in the theory before explaining the theory itself. (Yes? No? Maybe?)

MT was never supposed to be presented as evidence itself, but as an explanation for the initial inspiration for strategies for the "what we can do for DI / what DI can do for us" thing. Obviously, it would be way better to wait until many other things are explained before I tackle that one, but I might have to try a bit anyway just to keep the confusion down since it's already out there...

Anyway, I'll go ask Misha if he wants to work on the 'top-level post on DI' project now, and if he could use me for it.


The most I've been doing is poking at a post tentatively titled "A dry introduction to the empirical evidence on DI's effectiveness", essentially a summary of "Research on Direct Instruction", since I was feeling like maybe the best thing to do would be to take a step back and present a better explanation of why people should be interested in the theory before explaining the theory itself. (Yes? No? Maybe?)

Personally, I'm eager to actually use DI more in my own learning, so I'm currently working through Theory of Instruction. But some better evidence than PFT would be nice, yes. Especially if it isn't always about basic skills. (Because otherwise, no matter how good the technique, I won't benefit from it.)

Interestingly enough, the study with the highest effect size in the meta-analysis (2.44) involved non-basic skills. Actually I think I'll just type up the summary:

This study analyzes the use of the Earth Science videodisc program with elementary education majors who traditionally have had negative attitudes towards science teaching. One group received the DI program and the other group received the traditional approach [random assignment, of course] during a one-semester science course. The DI group had significantly higher posttest knowledge scores (91%) and higher confidence in their understanding of science knowledge and ability to teach science.

Cited as:

"Vitale, M. & Romance, N. (1992). Using videodisc instruction in an elementary science methods course: Remediating science knowledge deficiencies and facilitating science teaching. Journal of Research in Science teaching, 29, 915-928."

Not that I've dug up the original paper myself yet.

But one of my favorites was a study that didn't use random assignment, but actually compared the performance of two groups of high school students: AP kids (doing whatever they normally do to study), and kids with performance previously in the lower two quartiles (taught through a videodisc course on "Chemistry and energy"). Both groups then took the same test.

Results as a researcher reported informally outside the study: “The experimentals whumped the AP students on all topics related to what was covered by the videodiscs of our course.”

(This one wasn't included in the meta-analysis, so I'll have to try to dig up the reference later.)

My best wishes to your mom. I can sympathize with you a lot there.

If you want someone to read over material you have and give you critical feedback at any stage of the process, I am eager to help you. For example, if you have an outline or are planning on basing your writing on the previous post, I'd be happy to give you feedback on those. Also, if you need access to any papers, I have access to a university library account, so I can get you those. The same goes for Misha. Don't be at all shy about asking me for help. You can either contact me through PM, email me (username at gmail), or make a discussion post.

I suspect the motivation for studying DI shouldn't take more than a paragraph.

Hope your internship is starting off well. Who are you interning with/what are you doing?

Thank you for your offer of help with feedback (I'll def take you up on that) and papers (there are some papers referenced in "Research on Direct Instruction" I might like to get my hands on), and the sympathy on my ma.

I'm interning at a DI school in Baltimore (City Springs). Currently working with the kindergarteners on the language program (I'm supposed to move on to also doing math and reading soon, and teach older kids as well).

The National Institute For Direct Instruction (NIFDI) placed me here. It usually takes a minimum of two years for someone to get really good at presentation, but they figure I should be able to do it in one.

They're just setting up a program for talented DI teachers to learn design by becoming coauthors on new programs, and that's obviously where I'm aiming to go next year right after the annual summer DI conference.

Anyway, thing about the internship is that they've never had an unpaid foreign intern floating around before, so I end up as the third teacher in the room (the usual set-up at City Springs is a two teacher team. One of them is technically just a 'paraprofessional', but their instructional responsibilities are the same at a DI charter school). So I have to make sure I'm actually working on the things I need to be working on rather than getting side-tracked into some not really DI-relevant task.

I look forward to seeing your drafts.

Good luck with your DI internship!

Summary: Nice for beginners and people with bad learning experiences ...

Actually, this strikes me as a bit weakly worded. I think the MT courses are the best resource currently available for an English speaker looking to start learning French or Spanish, by a significant margin.

Unfortunately there are no scientific studies comparing the effectiveness of various different 'teach yourself' programs and traditional classroom instruction, so I can't find any direct evidence on that question beyond my own anecdotal experience.

But still, what with that and the much more indirect evidence available, I'd still be very surprised if this wasn't true.


Unfortunately there are no scientific studies comparing the effectiveness of various different 'teach yourself' programs and traditional classroom instruction, so I can't find any direct evidence on that question beyond my own anecdotal experience.

This is the reason I don't give a glowing recommendation. Research about long-term language learning is pretty lacking (and difficult to do, obviously), so this is a problem of pretty much any approach. I would agree that MT is the best starting material for these European languages I've seen so far.

Yes, but we're not talking about long-term research here. It wouldn't be hard at all to get a bunch of volunteers interested in learning a language, and randomly assign them to one of the various different treatments popular in the industry (MT, Pimsleur, Rosetta, traditional classroom instruction, whatever). It would take less than a year, probably.

(Various choices would be up to the experimenter, like whether they wanted to control time so all groups spent, like, an hour a day or whatever, or examine the time students chose to spend themselves as one of the dependent variables, or whatever. Obviously it would be best to do it both ways.)

Obvious, easy, and valuable. Just nobody with the resources has the incentive or the perspective to see that it should be done.

Anyway, I'd expect to see a pretty high failure rate even for the MT courses. Just a significantly lower failure rate than any of the other courses.

Wow, nice work!

Just one note for now: On the MT courses being approximations: Yeah, the way I usually think of it now references an article by Zig Engelmann called "The Dalmatian and Its Spots" (contextual prologue here).

To summarize the most here-relevant message of the article:

Thinking that programs with certain features [eg, some focus on 'phonemic awareness' and 'phonics' in a reading program] will be successful because research shows successful programs have those features is like thinking something with spots must be a dalmatian because research has shown dalmatians have spots.

So I would say that the MT courses are, "Essentially dalmatians, not just spotted. Some mutt, some mange, but dalmatian enough to suffice for many practical purposes”.

(Although this metaphor obviously shouldn't be phored too meta :P)