What is "Code-Switching"?

Discussions around Aviation English Language Proficiency testing often use the term code-switching. In this article, we’d like to consider this term and the implications of this phenomenon for language testing.

Code-switching is defined (by Barbara Bullock and Almeida Jacqueline Toribio in The Cambridge Handbook of Linguistic Code-Switching) as “the ability on the part of bilinguals to alternate effortlessly between their two languages.” Speakers are able to switch, sometimes mid-sentence, between the two “codes” with which they are familiar, depending on the situation, their listeners and many other things.

But, as the authors go on to point out, this phenomenon is not restricted to different languages. We all code switch to some extent when talking with different people in different situations: different registers, dialects or styles may come out of our mouths, mostly subconsciously.

In aviation the term relates to switching between phraseology / RT and plain English, which are each considered as separate “codes”. Document 9835 uses the term “code-switching” to describe the use of these two registers. When this code-switching is employed inappropriately (with plain language creeping into the deployment of phraseology) it can lead to confusion.  

To consider the use of phraseology and plain language as two distinct codes makes sense to us, and it seems to have implications for unravelling the knot of how best to test people’s aviation English language proficiency.

What we want to know, essentially, is how good the candidate is likely to be at dealing with this code-switching when it occurs over the radio, for whatever reason. 

We need to be able to make a valid inference about this within the confines of a test.

Code Switching

ICAEA seem to have decided, reasonably enough, that the "construct" of aviation English is at least partially synonymous with the ability to code switch.

However, they seem to infer from this that the code-switching itself is what needs to be tested. This is the reason that they insist upon role plays that engineer such situations in English language tests. 

We have spoken about our reservations with this approach elsewhere, but let’s just recap some of our issues with test tasks that try to engineer a situation in which we simulate a “real-life” situation of code-switching between phraseology and plain English:

  1. Such test items rely on “realism” in order to meet their own claims, yet they are by design highly unrealistic and artificial.
  2. The design of such items is horrendously difficult to standardise across test versions.
  3. The representation of the plain English construct in such tasks is typically very poor, especially because the task must be heavily scripted in order to ensure it stays “on track”.
  4. Most compellingly of all, in our view, interpreting poor performance in such a task is unnecessarily complicated. If someone does badly, is it because of their phraseology or their plain English, both or neither?

Fortunately, however, we do not believe it is necessary to test whether someone can code-switch between two languages or genres in order to know whether or not they can. We only need to test that they can speak each of the languages or genres separately.

To say otherwise is to claim the existence of a third factor, something that itself needs to be tested: a code-switching ability that is gained in addition to the abilities in the two separate codes.

This, it seems to us, is to make a claim that goes far beyond any empirical evidence, and also beyond common day to day experience, since we are all code switchers to some extent.

For example, I cannot code switch between English and Cantonese because I do not speak Cantonese to any level of fluency.

I can code switch between formal English and my local vernacular because I am adept at both of them (and if you say I ain't , I'll flippin’ well 'ave ya). I have never had to practise this skill in itself – it arises naturally out of the familiarity I have with the two codes.

If you want to know whether or not I can code-switch effectively between Cantonese and English, you can either:

a) try to invent a convoluted test that attempts to put me in some strange situation wherein I need to speak both Cantonese and English at the same time.


b) test if I can speak English, then test if I can speak Cantonese, using more established methods of language testing.

If I can’t speak one of them, we can rule out my being able to code-switch. If I can speak both of them to a standard that approaches fluency, we can without controversy assume I can also code-switch between them, should I need to.

All of the complication and confusion in Aviation English language testing seems to arise (from our point of view) because one of the two codes - spoken ability in RT - is seldom tested at all. Continuing to avoid testing it, and instead insisting on more Byzantine methods, will not reduce the confusion.

Indeed, testing both codes separately actually allows us to make much more fine-grained analyses of the problems:


Plain english test

RT Test

Inferred ability to code-switch


















Alice is a native plain English speaker with basically no phraseology training at all. Belle knows English RT very well, but is unable to speak English beyond simple sentences otherwise. Carlos is a near-native user of plain English but is also thoroughly adept at using English language RT. David knows very little English, RT or otherwise.

If we try to test their code-switching at the same time we get into a muddle, since differentiating between Alice, Belle and David becomes problematic, yet each of them has a very different set of training needs.

Testing these skills separately is clean, relatively simple and avoids the muddling of constructs, yet still allows us to make valid inferences about code-switching.

That is why we are proposing a way ahead for ELP proficiency that includes both plain English and English RT testing (and thus, by extension, code-switching ability).

The way to prove us wrong would be to demonstrate that ability in both of the codes separately does not imply the ability to switch between them. That is to say: demonstrate the existence of some additional “X-Factor” that enters the mix – an ability to code-switch that perhaps even Carlos in the example above lacks.

Even if so (and we have not been able to find evidence for this X-Factor in the literature to date), it is still not clear what is the best way to test for this ability would be.

In the absence of such data in the meantime, we continue to argue for the validity of a model that tests each of the two codes separately.

Notify of

Inline Feedbacks
View all comments
{"email":"Email address invalid","url":"Website address invalid","required":"Required field missing"}
Would love your thoughts, please comment.x