The Research Finding Some of Us Didn't Want to Believe

There is a very old study that was done with bilinguals in Canada and whose results I couldn't quite believe at first. McGill researchers John Macnamara and Seymour Kushnir, back in 1971, asked French-English bilinguals to listen to short sentences containing code-switches. These are complete shifts to the other language for a word, a phrase or a sentence, before coming back to the base language—that is, the language of the interaction (see here). The researchers presented statements such as, "A monkey can drink eau (water)" and recorded the time it took the participants to say whether they were true or false. They compared the results they obtained with those found for statements that contained no code-switches and found that those containing switches took about a quarter of a second longer to process. If there were two switches, the delay approached half a second.

Macnamara and Kushnir's methodology was questioned by a number of researchers, and I was one of them. Their code-switches did not follow the precise grammatical constraints of natural code-switching, there were grammatical problems with the French segments (e.g. one would say, "de l'eau" in the example above), it was unclear whether the participants were regular code-switchers themselves, etc. So a few years later, my colleague Carlos Soares and I, both active code-switchers in our everyday bilingual lives, undertook a study where we did away with these potential problems. We also made sure our bilingual participants were as fluent in the one as in the other language, and that they were indeed code-switchers when the situation and the interlocutor permitted it. We also told them that they would be hearing sentences with and without code-switches thereby removing, we thought, the surprise component of code-switching.

Much to our amazement, given all the precautions we had taken, we again found that the processing of code-switches took more time than that of base-language words. The difference was 152 milliseconds. Since then, other studies examining reaction time but also electrical activity of the brain (EEG) have replicated this finding. If one calculates the mean delay time for the published speech perception code-switching studies over the last 45 years, it is 133 ms. This is not a particularly long time and does not deserve the label "switching cost" that some researchers have used since, but it is nevertheless present.

Macnamara and Kushnir proposed an interesting explanation for the phenomenon. They hypothesized that as listeners, we have certain expectations and that one of them is that all the words in a sentence should be in a single language. We now talk about a "base-language effect," that is, the fact that in normal bilingual speech, elements belonging to the language being spoken—the base language—are favored over guest-language elements. This is because the base language is being processed primarily and is the most active.

A number of studies have gone beyond just finding a switching delay and have examined the factors that modulate it or remove it. Here are a few that have emerged over time. First, the amount of code-switching that takes place before the point at which code-switching processing is measured seems to play a role. The greater the amount of code-switching, the more the guest language is activated, and hence the more readily a code-switch is processed. A second factor concerns the situational context bilinguals are in. Yu-Lin Cheng and David Howard showed quite convincingly that bilinguals can process mixed-language utterance with no significant processing delay when they are in a situation where both languages are used interchangeably and frequently. Other factors that modulate the findings concern one-word switches—their frequency, their syllabic configuration, the way they are pronounced, the presence of a near homophone in the other language, etc.

Since many studies have shown that the perception of code-switches does take extra time although, as we have seen, various factors can affect the delay and may even make it disappear, researchers have asked how long the delay lasts after the code-switch. If it is carried over to the next word(s), then the bilingual listener may start falling behind the speaker—especially if the latter is code-switching a lot—something that seems quite counter-intuitive to all those who practice code-switching on a daily basis. In my laboratory, with two master's students, Corinna Domenighetti and Dolorès Caldognetto, we showed that the switching delay appears to be short-lived. By the time the next words arrive, any delay that might have occurred has been made up. Other speech studies have shown that the persistence of the delay depends on the switching direction—is it a switch into the first or into the second language?—as well as on the proficiency one has in the switch language.

When John Macnamara and Seymour Kushnir undertook that first speech delay study many years ago, they could not have imagined that the line of research they inaugurated would still be alive and well almost 50 years later. As researchers, we should salute their seminal work even though many of us didn't want to believe their results at first. But then, French medieval scholar Pierre Abélard did write, "It is by doubting that we come to investigate, and by investigating that we recognize the truth."