Your student doesn't have a bad accent. They have a different language.
Your student's accent is not a failure to reproduce the target language. It is an independent system under construction — and that changes everything about how you correct it.
I spent years working as a conference interpreter.
Community meetings. Business negotiations. Training sessions where the stakes were real, if not always high. Rooms where I had to listen, closely and continuously, to non-native speakers working hard to be understood.
And what I noticed, long before I had the research to explain it, was this: the best language learners were never the ones who had eliminated their accent. They were the ones whose accent had become consistent. Structured. Predictable. Systematic.
Something was being built. You could hear the architecture.
I didn't have a name for it then. Second language phonology research does.
It's called the interlanguage. Understanding it will change how you listen to your learners, how you design their practice, and what you decide to correct.
The Native Language Transfer Myth
Ask most language teachers why a student has a strong accent and the answer comes quickly: interference from the mother tongue.
The Spanish speaker who devoices English consonants. The Italian who adds vowels to the ends of words. The French speaker who drops the h. The explanation is tidy: the native language is bleeding into the target language, and the job of the teacher is to stop the bleeding.
This explanation is not wrong. It is just incomplete. And the part it leaves out changes everything.
Research in second language phonology, the study of how learners acquire the sound systems of new languages, has been building a different picture for the last five decades. The finding is this: the sounds a learner produces are not simply a mix of their native language and the target language. They are something else entirely.
Something the field calls the interlanguage.
What the Interlanguage Actually Is
The interlanguage is the mental phonological system that every L2 learner constructs. Not borrows. Not approximates. Constructs.
It is distinct from the learner's native language. It is distinct from the target language. It has its own rules, its own internal logic, and its own developmental trajectory. And here is the part that stops most practitioners mid-thought: it produces pronunciation patterns that exist in neither language the learner knows.
This is not a theoretical claim. It is an empirical one, documented in study after study across learner populations from different native language backgrounds, learning different target languages, in different instructional contexts.
The clearest evidence comes from cases where neither native language transfer nor target language input can explain what the learner is producing.
Researchers studying native speakers of Hungarian learning English found something that stopped them. Their subjects were producing a consistent pattern of word-final devoicing: a rule by which voiced consonants become voiceless at the end of words. Think the difference between the d in bad and the t in bat. The Hungarian learners were systematically moving toward the voiceless version at the end of words.
The striking thing was not the pattern itself. It was the source.
Hungarian doesn't do this. English doesn't do this. The pattern wasn't imported from the native language. It wasn't modelled by the target language. It came from neither direction.
But it exists. In Catalan. In German. In Polish. In Russian.
The learners had, independently, converged on a phonological solution that other languages of the world use. Not because they had been exposed to those languages. Because the interlanguage is a genuine linguistic system, and genuine linguistic systems follow universal principles. The learner wasn't making a mistake. The learner's system was making a decision.
That distinction is everything.
What the Accent Is Actually Telling You
If the interlanguage is a rule-governed system, not a failure to reproduce the target language but an independent grammar under active construction, then the accent is not evidence of what has gone wrong. It is evidence of what is being built.
Consider what it means to produce a pattern that exists in German but not in your native Spanish and not in your target English. That is not an error. That is a system making a principled decision. The interlanguage found a phonological solution, and it found one that real languages use.
The learner is not failing to sound native. The learner is operating a system that hasn't yet arrived at the target configuration. And that system is doing exactly what language systems do: generating structured, consistent output based on the rules it has developed so far.
This is the difference between a construction site and a demolition site. The accent is not rubble. It is scaffolding.
Why Some Sounds Are Hard for Everyone
If the interlanguage were simply a blend of native and target language, we would expect learners from different native language backgrounds to produce different errors when learning the same target language. Sometimes that's true. But sometimes it isn't. And those cases are the most revealing.
Certain pronunciation patterns are harder to acquire regardless of where the learner comes from. This is not random. It follows a principle called typological markedness: some sounds and structures are rare across the world's languages, and rare structures are harder to acquire. More common structures, those found in the widest range of languages, are acquired earlier and held more robustly.
An example. Voiced obstruents in word-final position, think the z at the end of jazz or the v at the end of love in English, are typologically marked. Many languages don't allow them. Languages that have voiced word-final obstruents also tend to have voiceless ones, but not the reverse. This asymmetry is not a coincidence. It reflects something about how the human phonological system is organised.
Learners whose native languages lack this structure, regardless of whether they are Japanese, Arabic, or Turkish speakers, will find it harder to acquire than learners whose native languages have it. The difficulty is not about interference. It is about markedness. About where a given structure sits in the hierarchy of cross-linguistic frequency.
What this means practically: some pronunciation problems are not language-specific. They are human. They will appear across learner populations, and they will require a particular kind of attention, not because the learner is confused between two systems, but because the target structure sits at a difficult position in the universal hierarchy.
The Interlanguage Has Two Frontiers: The Ear and the Mouth
The interlanguage is not a single system. It has two faces.
There is the system that perceives: that receives incoming speech, processes it, and maps it onto existing phonological categories. And there is the system that produces: that generates output based on the rules the learner has developed so far. These two faces belong to the same interlanguage. But they develop on different timelines, through different mechanisms, and they can fail independently.
This is not a minor technical detail. It is one of the most practically consequential findings in second language phonology, and one of the least discussed in teacher training.
A learner may be able to distinguish between two sounds in perception long before they can produce that distinction reliably. They hear that ship and sheep are different. They cannot yet make that difference themselves. The perceptual category exists. The articulatory habit does not. These are two separate acquisitions, and treating them as one is a design error.
The reverse is equally possible, and equally misread. A learner may produce something approximating the target sound before they can consistently perceive it in native speaker speech. In a slow, modelled drill, they reproduce it accurately. In connected speech at natural speed, the same sound washes over them unrecognised. The mouth arrived before the ear. The production was real. The perception wasn't there yet to support it.
What this means for practice is precise.
Drilling production before the perceptual category exists in the interlanguage builds muscle memory with no anchor. The learner produces the sound in a controlled exercise and cannot access it when it matters: when the language is moving at speed, when attention is split, when meaning and form are competing simultaneously for the same cognitive resources.
And asking learners to discriminate sounds before their interlanguage has built a slot for the contrast is asking them to hear a distinction that, for them, does not yet exist. It is not an attention problem. It is an architecture problem.
Perception and production tasks are not interchangeable. They are sequential. The interlanguage builds the perceptual category first. Then the articulatory habit has somewhere to land.
Three Things the Research Asks You to Do Differently
The interlanguage framework is not just a theoretical reframe. It has direct consequences for how practice is designed. Three principles follow from the research, each one a departure from standard classroom habit.
Principle 1: Treat the interlanguage as a system, not a symptom.
When a learner produces a consistent pattern, even one that sounds wrong, that consistency is data.
It means the interlanguage has a rule. Something is generating that output reliably. And a rule can be targeted precisely. A deficiency can only be filled.
The difference matters in practice. If a learner consistently devoices word-final consonants, the question is not why can't they hear the difference? The question is what rule is their system applying, and what input does it need to revise it? Those are different questions. They lead to different tasks.
Marking the output incorrect and modelling the target form again is not wrong. It is just the beginning of the work, not the work itself.
Principle 2: Learn to distinguish between transfer errors and universal difficulty.
Not all pronunciation problems have the same source. And they do not respond to the same instruction.
A transfer error is a native language rule applied to the target language. The Italian speaker who adds a vowel to the end of English words is not failing to listen. They are applying a phonological habit that has worked flawlessly for decades. That habit can be addressed directly, with targeted contrast practice and explicit attention to the difference.
A markedness problem is different. When the difficulty is structural and cross-linguistic, when the target sound is rare across the world's languages and therefore harder for any learner regardless of background, the source is not the native language. It is the universal hierarchy of phonological difficulty. The instruction required is more exposure, more repetition, more time. Not more contrast.
Conflating the two is one of the most common and costly errors in pronunciation teaching. It gives the learner a misleading map of what they are working with, and wastes instructional time on the wrong intervention.
Principle 3: Sequence perception before production for every new contrast.
The ear and the mouth are on different timelines. They always have been.
Before a learner can produce a sound distinction reliably, they need to hear it repeatedly, in context, at varying speeds, embedded in connected speech. The interlanguage needs to build the perceptual category first. Without it, production practice builds muscle memory with no anchor. The learner can reproduce a modelled sound in a controlled drill and still fail to recognise it when it arrives at native speaker speed, inside a sentence they are simultaneously trying to understand.
The reverse risk is equally real. Asking learners to discriminate sounds before their interlanguage has carved out a category for them is asking them to hear a distinction that, for them, does not yet exist. The sound washes over them, not because they are not paying attention, but because the system has not built the slot.
Perception tasks and production tasks are not interchangeable. They train different things. Sequencing them correctly is not a refinement. It is the design.
The Interlanguage Is Not a Phase to Rush Through
Perhaps the most important shift the research asks of practitioners is a change in posture toward the in-between state.
The interlanguage is not a waiting room. It is not a misconfigured version of the target language that needs to be corrected into shape as quickly as possible. It is a linguistic system in active development, one that obeys universal principles, draws on what the learner already knows, and produces genuinely systematic output.
Learners who are treated as builders rather than imitators engage differently. They bring different attention to the task. They notice more. They are less destabilised by the gap between where they are and where they are going, because they understand that the gap is not failure. It is the architecture of acquisition.
The accent is not the problem. It is the progress report. And reading it accurately is one of the most useful things a teacher can do.
The Objection Worth Taking Seriously
At this point, a reasonable practitioner pushes back.
Understanding the interlanguage is useful. But the goal is still target-like pronunciation. Students need to close the gap, not simply understand why it exists. And a framework that reframes the accent as a construction site doesn't tell you how to finish the building.
It's a fair objection. Here is the direct answer.
Understanding the interlanguage does not replace correction. It makes correction more precise.
The teacher who knows that word-final devoicing is an interlanguage rule, not random error, not laziness, not a failure of attention, knows exactly where to intervene and what kind of intervention to use. They are not correcting a deficiency. They are updating a rule. Those are different operations. They require different tasks, different sequencing, different feedback.
The teacher who conflates transfer errors with markedness problems will apply the same correction to both. Sometimes it will work. Often it won't. And when it doesn't, there is no diagnostic available, just the sense that the student isn't trying hard enough, or isn't talented enough, or needs more of the same thing that already isn't working.
Knowing the source of a pronunciation pattern is not academic. It is the difference between targeted intervention and repetitive correction that produces frustration on both sides of the desk.
The interlanguage framework does not ask you to stop correcting. It asks you to correct the right thing, for the right reason, at the right moment.
That is not a softer version of pronunciation teaching. It is a more accurate one.
If You Read Nothing Else, Read This
For the practitioner who has skimmed to this point, here is what the research asks of you. Three shifts. One paragraph each.
The accent is a progress report, not a problem report. Your student's consistent pronunciation patterns are not random error. They are the output of a rule-governed system under active construction. Read them as data. The consistency is the signal.
Not all pronunciation problems have the same source. Some errors come from the native language. Some come from universal phonological difficulty that affects learners regardless of background. They look similar on the surface. They require different interventions. Knowing the difference is the diagnostic skill the interlanguage framework gives you.
Perception before production. Always. The ear builds the category. The mouth follows. Drilling production before the perceptual category exists builds habit without anchor. Sequence accordingly: hear it first, in context, at speed, repeatedly. Then produce.
These are not refinements to standard practice. They are corrections to it.
The Question That Changes Everything
There is a version of language teaching that treats the learner as a recording that hasn't been corrected yet.
Fix the interference. Close the gap. Get them to sound like a native speaker. The work is subtraction, removing everything that isn't the target language until what's left is clean.
That version of teaching is built on the wrong model.
The interlanguage is not interference. It is construction. And construction doesn't need to be corrected. It needs to be read, understood, and built upon. The learner standing in front of you with a strong accent and consistent patterns is not behind. They are mid-build. The scaffolding is up. The architecture is visible.
What changes when you understand this is not your syllabus. It's your posture.
You stop asking: why can't they sound native? You start asking: what system are they operating, and what does it need next?
That is a different question. It produces different attention, different tasks, and different results. It also produces something harder to measure but easier to feel in a classroom: learners who are not ashamed of where they are, because they understand that where they are is not a failure state. It is a construction site.
Every construction site, read correctly, tells you exactly what to build next.
Source Eckman, F.R. — Second Language Phonology in The Handbook of Second Language Acquisition (Blackwell Publishing)
Further reading Corder, S.P. (1971) — Idiosyncratic Dialects and Error Analysis Selinker, L. (1972) — Interlanguage Nemser, W. (1971) — Approximate Systems of Foreign Language Learners
Related topics on this site Chunk-based listening and prosodic acquisition Perception-production asymmetry in L2 listening design