Can we “prove” Krashen’s comprehensible input hypothesis?

Steve Smith tells me on Twitter that nobody can “prove” Krashen’s hypothesis that languages are acquired by getting lots of comprehensible input. Clearly, as Krashen himself recently said, “we need to talk about science.”  Specifically, today’s question:

Can science prove anything, and can we “prove” the comprehensible input hypothesis?

The answer: science can never prove anything. Truth, technically, is a property of closed symbolic systems (e.g. logic, math).  So, why– how?— is science useful?  It’s pretty simple.  All science does is make testable predictions about causes of phenomena.  Sometimes, scientists will also propose an actual mechanism.

Scientists:

1. observe a phenomenon (e.g. people acquiring languages)
2. make a prediction/guess about how this happens (e.g. via comprehensible input). This is called forming a hypothesis
3. test via experiment your hypothesis to confirm it (e.g. expose people to comprehensible input and see whether or not they learn the language; expose them to grammar lessons and see whether/how well they acquire)
4. At the end of your experiment, you will know whether  or not X causes Y.
5. Investigate confounds (potential alternative explanations for phenomenae)
6. For it to qualify as science, an experiment has to repeatedly generate the same results.

Krashen’s hypothesis is simple: if people are exposed to comprehensible input in the target language, they will acquire the language.  (Technically, Krashen’s hypothesis– which has been tested and confirmed– is now a theory.)

Steve Smith has two objections to Krashen’s hypothesis.  First, he says this:

Strictly speaking, no hypothesis can ever be “proven” true. All you can do is test the hypothesis and see whether data confirms it (aligns with its predictions).  With language acquisition, the research is clear: people who get comprehensible input acquire languages; people who get incomprehensible input, grammar practice, too much output “practice,” or a mix of all acquire no (or very little) language, and always much less than those exposed to comprehensible input.

Is the hypothesis testable? Yes.  Has it been tested, and its predictions confirmed? Yes.  Has anything else come along to provide a better explanation?  No(t yet).  Karl Popper reminds us that good science isn’t true.  He notes that good science has only two properties:

• it’s just not wrong (yet)
• it makes accurate, testable predictions.

While Smith is technically correct, he misses the point.  Suppose we hypothesise that an analgesic such as Ibuprofene reduces pain.  On testing our hypothesis, we find that it does indeed reduce pain.  The confirmed hypothesis is thus useful and accurate, but, technically, it’s not “true.” It “does the job” of explaining and predicting.  Hypotheses aren’t true— they work, or they don’t.

Smith’s second claim is this:

This misses the point entirely. First, Krashen does not propose an explanation for language acquisition on the neurological level, nor does he need to.  His hypothesis only involves comprehensible input and acquisition (both of which he defines).

An analogy may be of service here.  Imagine: we bring a preindustrial tribesman into the modern world and he observes cars.  He forms a hypothesis– gas makes cars go– and predicts that, ceteris paribus, a car with a full gas tank will go further than a car with an empty tank, and tests this hypothesis.  The car’s performance obviously substantiates his hypothesis.  Now, the fact that the tribesman doesn’t know anything about internal combustion engines, energy efficiency, math, etc, while true, is irrelevant and does not discredit his hypothesis.

Similarly, the fact that Krashen (and Chomsky, and VanPatten, and Lightbrown, and every other person who investigates S.L.A.) do not propose a neurological explanation for language acquisition is irrelevant.  What counts is whether or not the hypothesis holds up under experimental scrutiny (i.e. whether or not people acquire language through comprehensible input).

Somebody could come along with a better explanation (in which case the comprehension hypothesis, as Krashen notes, gets tossed).  Or, somebody could get right down to the neuronal level and explain the acquisitional mechanism.  If this “neuronal explanation” showed that something other than c.i. accounted for SLA, the hypothesis would again get the boot.  Or, it might simply show us the mechanism by which comprehensible input becomes acquisition.  (This would be something like how Einstein updated Newton: relativity doesn’t invalidate Newtonian mechanics, rather, it just applies on a different level).

Second, Smith is wrong when he says there is no way to say whether or not the use of comprehensible input, focus on form (grammar instruction and/or practice) or a mix of the two are best practice in the language class.  First, we know what works (comprehensible input) from research.  Second, we know– again from research– what has no (or very limited, conscious-mind-only, and short term) effects: grammar teaching and practice, and output. Unless you want to advocate doing something that we know doesn’t work very well, the conclusion is obvious: the more c.i. learners get, the better off they will be, and the best mix is probably as little grammar talk as possible.  VanPatten has also weighed in here, saying that traditional practice and grammar explanations do “very little” for acquisition.

Again, we don’t know for sure how much grammar instruction and how much input learners should be getting. There are a lot of suggestions, though.  In the New Brunswick E.L.L. study (Lightbrown et al), French-speaking students who received only comprehensible input (by reading and listening)  without a teacher did almost as well as students who were taught English and tutored in writing.  In other words, 90+% of the work was done by input.  Beniko Mason (1997) found that Japanese college students who simply read in English far outperformed students who had writing practice and direct grammar instruction in vocabulary recognition.  In both first and second languages, free voluntary reading (teacherless comprehensible input, as it were) has overwhelmingly and repeatedly outperformed any other method of teaching vocabulary, grammar, style, etc (Krashen’s site has all the data).

[real-life digression: Blaine Ray told me the following:  when T.P.R.S. was being developed in the late 1980s, Ray called Krashen– who was then with Tracy Terrell testing the “Natural Approach”– and asked, “how much grammar homework should I be doing?”  Krashen, skeptical of grammar practice from his linguistics research but aware that there were also gaps in said research, told Ray “well, get them to do some grammar practice just for homework.”  So, Ray– whose Bakersfield school district mandated grammar teaching– had his kids do the stupid fill-in-the-blanks stuff that comes with the ¡Díme! program– the cuaderno exercises.  At the end of the year of grammar homework, Ray found the kids writing and saying basic errors like *Yo quiero juego fútbol americano (I want I play football– the sentence should read yo quiero jugar fútbol americano).  Exasperated, thinking “why waste time?”, he ditched all the grammar homework, and next year, in class, while announcements blared and he had to take attendance, Ray had a kid stand at the front of the room, read the  ¡Díme! grammar explanations aloud to fulfill District requirements (nobody listened), assigned reading for homework, and found the kids at the end of the next year making fewer mistakes.

This is an experience that every stick-to-your-guns T.P.R.S. teacher has had or will have.  You will doubt the power of comprehensible input, you will assign grammar homework (or “conversation practice” or whatever legacy method), your kids will dutifully do this, and it won’t work.]

Third, Smith is also wrong when he says that because we cannot “see into [the] brain,” there is no way to decide what language class activities are best.  We don’t need to “see into [the] brain” to know what works.  You probably can’t explain on a chemical level what happens when your car burns gas.  Do you need to in order to drive? I’d say, if you know enough to put the right fuel in, and you do put the right fuel in, you’re all set.  And if you did know a chemical explanation for combustion, would that help you drive?

Smith also says this: 

I didn’t say this, and the research flatly contradicts it.  Krashen (2003) in “Explorations in Language Acquisition” notes that all the research on grammar-focused teaching shows positive effects only when assessment is done under Monitor-use conditions.  

In other words, grammar teaches you…grammar.  VanPatten comments that “what we call grammar rules are what we end up with, and are not how we learn or what the brain actually does” (MIWLA presentation, 2013), and the rest of the research is here.  

Grammar-focused teaching works when

• grammar items are either elicited and/or “overloaded” in the input
• learners have time to think of and plan for responses
• learners know, know how to apply, and have time to apply, the grammar rules

Krashen proposes a much higher standard for testing whether or not grammar teaching becomes implicit (automatic) learning (i.e. whether people have acquired the item in question), with broadly two criteria:

a) a three-months-delayed post-test.  Most of the research will do an immediate treatment post-test (i.e. they will see if people can do/use grammar rule ______ right after the experiment) and a slightly delayed post test (e.g. two weeks later).  However, if we waited three months, and grammar rule ____ was still recognised or put into use, then we would have much stronger evidence that explicit teaching can become implicit knowledge.

b) Monitor-free testing.  This just means that you see whether people have picked up ______ without making them consciously aware that they have learned, should use, etc ____.

Say your treatment was teaching English speakers Spanish pronoun placement.  Pronoun goes before one verb, or before or after verb clause w/ some exceptions, bla bla.  This is a classic S.L.A. research area, because Spanish pronoun location is different from English, so it’s brainwork to acquire this new rule.  Now, when you do your post-test, here are two possible scenarios:

1.  You tell your test subjects “OK, we are gonna ask you some questions.  A question might be ¿Conoces a George Clooney? and you could answer Sí, lo conozco or No, no lo conozco.”  You could also (or instead) tell them “we would like you to answer using pronouns, like lo or la etc.”

2.  You tell your test subjects “OK, we are going to ask you some questions, just answer.”

Under (1), we are modeling specific behaviours, reminding people about expectations and grammar rules, pronouns, etc.  We are bringing grammar knowledge to conscious awareness.  Under (2), we just see what they do.  They might use pronouns, or not, or sometimes, or use them in a mix of properly and not, etc.  Krashen’s point is very simple:  if we do anything like (1), we are not necessarily seeing what people have acquired.  We are seeing what people can do with conscious knowledge and/or modeling.  This is what Krashen calls “Monitor use.”

Why do we want to have Monitor-free assessment of instructional treatment?  Because, in the real world, we simply do not have time to think, rule-remember, edit, etc.  Good language teaching will “wire the language in” below the level of conscious awareness.  If I teach rock climbing, I don’t want you to be able to tell me how to tie a figure 8, or how to do a drop knee and lock-off; I want you to tie a figure 8, and automatically do a drop knee with lock-off when you need it.  When I am at the Paris Metro and a smoking hot Parisienne is flirting with me, I need to be able to spit out, without thinking, right away, j’aimerais vous inviter à manger avec moi, parce-que vous êtes une femme incroyablement interesante or whatever.  If I am standing there going “OK, do I put vous in front of or behind the inviter?” I am not going to have even a shot at the lady’s company.

OK, back to Steve Smith:

Smith also commits a few logical fallacies here.

First, the appeal to authority and mass opinion– that people “feel” something works–  does not qualify as evidence that it does.  I “feel” that the Moon is made of cheese.  Is it?

Second, it’s also post hoc, ergo propter hoc— after this, because of this.  You teach French grammar (and whatever else), and after that, your kids acquire some French.  Was it the grammar, the “whatever else,” or both that got them to learn?  Eric Herman and I have discussed what he calls “incidental learning,”, and we concluded this: even horrible languagen teaching– what I did for the first 12 years of my career– “works” because even if you are doing forced output, grammar worksheets, bla bla bla, the kids are getting comprehensible input.  Boring, impoverished, low quantities, etc, but c.i.  So…do Steve Smith (or whoever)’s kids acquire because of grammar, or because grammar contains some c.i.?

Third, Smith says that “learners” feel “conscious learning” can “become acquired.”  Really?  We’d need some evidence for this– i.e. Smith would have to ask say 100 students how well they felt that grammar teaching and practice was helping them, and then compare those statements with results, and show us that the students who liked their grammar teaching did significantly better (than controls) as a result, etc.  Any T.P.R.S. teacher would respond to this by saying “we don’t spend more than 20 sec/class on grammar, and our kids feel that comprehensible input stories are the most effective way to learn ____.”  Again…we’d need evidence from TPRS kids.

Another problem here: even if you “feel” grammar teaching helps, how do you know it does?  This is much like the “noticing” argument that Swain developed and Truscott dismissed: the fact that you are aware of a form-meaning connection (a grammar point) which you’ve acquired does not mean that you acquired it because of this awareness.  (In my experience, it’s the opposite: I “notice” grammar awareness once I have acquired it– your mileage may vary.)

The question of whether or not one could ever deliver “pure” grammar instruction is up in the air.  I have said this before and I think Smith may be referring to that statement.  Even T.P.R.S. is technically not 100% input– because we do occasionally say “-s means you in Spanish.”

Suppose you have a terrible book– Avancemos, say– where the kids have to conjugate “to go” in Spanish.  So they are writing Yo voy al cine, ella va a la escuela, etc.  Boring & dumb, and output as VanPatten reminds us is useless, etc.  BUT…if the kids actually understand what they are reading, it is still (tedious, two-dimensional, impoverished) input.  So, you could get them to pick up some Spanish that way.  I guess.  If you wanted to totally suck, and make your kids hate Spanish, and make them learn slowly, and check out emotionally…