Is it eas­i­er to use one lan­guage vari­ety at a time or mix them? An online vol­un­tary lan­guage switch­ing study with “bidi­alec­tals”

Jo Ever­shed 0:00
We’ve got Neil, Neil do come across and share your slides. Neil and I first met, oh, gosh, a few years ago, Neil has been up in Dundee study­ing dif­fer­ent dialects of dif­fer­ent lan­guages. He’s done all sorts of amaz­ing things, includ­ing using the local radio sta­tions to recruit his par­tic­i­pants, which is one of the rea­sons I’ve asked him to speak today. So don’t let him leave the stage with­out shar­ing how he got that to work and what impact it had on his research. Neil, over to you.

Neil Kirk 0:30
And I just check that you’re see­ing my title slide only? Yeah, yeah, no, no. Per­fect. Because, yeah, I’ve not been using zoom that much now that we’re able to get out the house again. So bit rusty. Yeah, thanks, Jo. I’m going to talk today about some work that we’ve been run­ning with. I’ve explained this term in a minute, but what we refer to as bidi­alec­tels, and we’re run­ning lan­guage switch­ing par­a­digms, essen­tial­ly try­ing to inves­ti­gate the cog­ni­tive process­es that are involved in using more than one lan­guage vari­ety. And the work I’m going to present today has been pub­lished as a stage one reg­is­tered report.

So as some­one who was allud­ing to the things like the design hypoth­e­sis, or analy­sis plan or peer reviewed, I’m not going to go too heavy into sort of the­o­ry or any­thing in my talk, this Papers, Please, if you’re inter­est­ed in more, it’s a good start­ing point. So maybe I will go into a lit­tle bit of the­o­ry. So mod­els of bilin­gual lan­guage con­trol pro­pose that a bilin­guals lan­guages are always active, and they’re always com­pet­ing for selec­tion to some extent. So when a bilin­gual plans to speak in that par­tic­u­lar tar­get lan­guage, they have to inhib­it or sup­press all the items from the oth­er non tar­get lan­guage from…they have to inhib­it them to pre­vent any cross lan­guage interference.

So in this exam­ple, if you want to name this ani­mal, the bilin­guals have to inhib­it the Ger­man word to allow them to select the Eng­lish word dog. And the amount of inhi­bi­tion that’s required to do this is rel­a­tive to how strong­ly rep­re­sent­ed or dom­i­nant that lan­guage is. So in this case, let’s say Eng­lish is the more dom­i­nant lan­guage, it would take more ener­gy, in a sense to sup­press this to allow them to speak the Ger­man word. How­ev­er, it can take a while for that inhi­bi­tion to wear off, mean­ing that in rel­a­tive terms, it can take longer to switch back to your dom­i­nant lan­guage, com­pared to switch­ing into your less dom­i­nant language.

So these effects are com­mon­ly mea­sured using lan­guage switch­ing par­a­digms. And a typ­i­cal set­up here is that a par­tic­i­pant is asked to name pic­tures of objects, and usu­al­ly giv­en some kind of cue as to which lan­guage vari­ety they should name the object in. We’re inter­est­ed in how accu­rate they are, but pre­dom­i­nant­ly, we’re inter­est­ed in how quick­ly they can do this. And by mea­sur­ing the response times, we can observe the effects of things like nam­ing in your dom­i­nant ver­sus your non dom­i­nant lan­guage, and the effects of switch­ing from lan­guage A to B or from lan­guage B to A and so on.

Gen­er­al­ly, what you find in these par­a­digms is that switch­ing between vari­eties takes longer to name the object than if you had already sort of remained with­in the same lan­guage set­ting. So, there are dif­fer­ent types of lan­guage envi­ron­ments a bilin­gual could find them­selves in which might require dif­fer­ent appli­ca­tions of lan­guage con­trol. So for exam­ple, a bilin­gual could find them­selves in a sin­gle lan­guage envi­ron­ment. And what they can do here is proac­tive­ly apply lan­guage con­trol to keep a high­er acti­va­tion lev­el of one lan­guage, because they might have no expec­ta­tion that they will require the use of the oth­er language.

And in lan­guage switch­ing par­a­digms, we can tap into this by using sin­gle lan­guage blocks. So this is where the par­tic­i­pant would be informed that they will only name in one lan­guage have it throat and again, they would have no expec­ta­tion of hav­ing to shift gears and to use a dif­fer­ent variety.

A bilin­gual could also find them­selves in a mixed lan­guage envi­ron­ment where they have to reac­tive­ly switch lan­guages on the fly depend­ing on who they are com­mu­ni­cat­ing with at any giv­en moment. And we can mea­sure this in lan­guage switch­ing par­a­digms by using mixed lan­guage blocks. Usu­al­ly this involves the par­tic­i­pant being cued are still which lan­guage vari­ety the to name, the object, then often that’s a colour cue.

4:48
And these are often pre­sent­ed in an unpre­dictable sequence so the bilin­gual can’t antic­i­pate what lan­guage we’ll have to use on the next tri­al until that pic­ture, and its cor­re­spond­ing bor­der appear. These kinds of blocks give us dif­fer­ent types of tri­als. So we have with­in these, what we’d call rep­e­ti­tion tri­als, which have tri­als in which you have named in a par­tic­u­lar lan­guage hav­ing stayed with­in the same vari­ety com­pared to the pre­vi­ous tri­al. And we also have switch tri­als that require you to switch from one vari­ety to the oth­er, and basi­cal­ly shift­ing these dif­fer­ent men­tal set­tings. And we can also, this can involve switch­ing from lan­guage A to B, or from lan­guage B to A, and again, tri­als that involve switch­ing tend to take longer than tri­als that require you to remain with­in the same variety.

Now, a com­mon find­ing in bilin­gual research is that there is a mix­ing cost asso­ci­at­ed with hav­ing to use more than one vari­ety. And what this… basi­cal­ly what we gen­er­al­ly find is that tri­als in sin­gle lan­guage blocks are named faster than rep­e­ti­tion tri­als from mixed lan­guage blocks, despite both types of tri­als basi­cal­ly involv­ing hav­ing named with­in the same vari­ety from one tri­al to the next. But basi­cal­ly, in a mixed lan­guage block the antic­i­pa­tion of poten­tial­ly hav­ing to use your oth­er vari­ety. And to switch to that means that both lan­guages main­tain a high­er lev­el of acti­va­tion, which is caus­ing these slow­er respons­es. So that’s the mix­ing cost, you’re slow­er when you’re hav­ing to mix your languages.

There’s also anoth­er type of lan­guage set­ting a bilin­gual could find them­selves in where they’re pre­sent­ed with speak­ers who speak both their lan­guages. So they’re basi­cal­ly able to use the lan­guages freely and mix them, some­times even with­in the same sen­tence. So in a sense, this kind of set­ting might not require any kind of lan­guage con­trol, because it’s just a case of an any­thing goes, it does­n’t mat­ter what word I use, from what lan­guage, who I’m com­mu­ni­cat­ing with will understand.

And in terms of lan­guage switch­ing, we can inves­ti­gate this using vol­un­tary lan­guage switch­ing blocks. So this is where the bilin­gual be told to name, each pic­ture with what­ev­er comes to mind does­n’t mat­ter which lan­guage just what­ev­er, what­ev­er names comes to mind first. So in this case, they might name that first pic­ture in lan­guage A, they might then stay with­in that lan­guage and name the next pic­ture with­in lan­guages as well, they might then switch into lan­guage B. And again, that could then remain with­in that lan­guage. What we then do is sub­se­quent­ly code, what lan­guage they used for each tri­al, and whether that tri­al forms a rep­e­ti­tion or a switch.

And what we find here is a mix­ing ben­e­fit of in some stud­ies that we saw pre­vi­ous stud­ies by by lin­guist find a mix­ing ben­e­fit when you are able to vol­un­tar­i­ly choose the lan­guage. And so basi­cal­ly, in this case, the bilin­guals are often faster to name rep­e­ti­tion tri­als from the mixed block than they are to name their sin­gle lan­guage tri­als from this sin­gle lan­guage block, pre­sum­ably because they’re not hav­ing to employ any lan­guage con­trol in the vol­un­tary block. And in the sin­gle lan­guage block, they are hav­ing to sup­press the oth­er vari­ety, because they’re, they know they won’t be using it. Where­as in the mixed vol­un­tary block, they’re free to use what­ev­er comes to mind.

8:32
So we’ve pre­vi­ous­ly inves­ti­gat­ed whether sim­i­lar effects exist for, as have been found for bilin­guals, for speak­ers of close­ly relat­ed lan­guage vari­eties. And in our research, we’ve used speak­ers of Scots, which is a Ger­man­ic vari­ety, close­ly relat­ed to Eng­lish it has from the last cen­sus, some­thing like 1.5 mil­lion speak­ers. It’s recog­nised by the Euro­pean Char­ter for region­al or minor­i­ty lan­guages as a minor­i­ty lan­guage. But its sta­tus as a lan­guage is often ques­tioned and even the speak­ers them­selves might not regard them­selves as bilingual.

In our pre­vi­ous stud­ies, using adap­ta­tions of lan­guage switch­ing par­a­digms, we found lots of sim­i­lar effects as are found for bilin­guals. So essen­tial­ly, we’re see­ing in these par­tic­i­pants bilin­gual effects, even though they would regard them­selves as mono­lin­gual. And why why do they do that? Well, Scott is a sort of stig­ma­tised vari­ety, it’s often con­sid­ered quite low sta­tus. I’ve just thrown up a cou­ple of exam­ples of tweets where in response to some­body who was speak­ing Scott. So you can see you know, it’s not con­sid­ered you know, peo­ple don’t are not very favourable towards it sometimes.

And even speak­ers them­selves, don’t nec­es­sar­i­ly regard it as a dif­fer­ent lan­guage and this gov­ern­ment report from a few years back so 64% of respon­dents said that they did­n’t think of Scots as a lan­guage, it was just a way of speak­ing. And in a sense, we can think of it more like a reg­is­ter, you know, it’s kind of like a for­mal or infor­mal way of talk­ing, in that sense more of an infor­mal way of talking.

So in the present study, we want­ed to inves­ti­gate whether our so called bidi­alec­tal speak­ers would dis­play this mix­ing ben­e­fit that has been found for more tra­di­tion­al­ly recog­nised bilin­guals. So to do this, we used speak­ers of Scot­tish Stan­dard Eng­lish, which is essen­tial­ly what I’m speak­ing though, and dun­don­ian Scots, which is the the Scots dialect spo­ken in Dundee.

And, as I men­tioned, this was a reg­is­tered report, we end­ed up with 46 par­tic­i­pants, which was our tar­get. They report­ed using dun­don­ian around 26% of the time. As Jo has said, I’ve used the local media, radio, news­pa­pers, to and social media to recruit our par­tic­i­pants, and we paid them 15 points for par­tic­i­pat­ing. I should say that recruit­ment plat­forms like pro­lif­ic do actu­al­ly have ques­tions in the demo­graph­ics about Scots, we decid­ed not to use that because from our expe­ri­ence, a lot of speak­ers them­selves don’t even know what Scots is, they might think of it as some­thing that oth­er peo­ple in oth­er parts of Scot­land speak. And because we were using stim­uli that was relat­ed to dun­don­ian Scots, it does­n’t, pro­lif­ic does­n’t drill that deep. So you know, you don’t have to trav­el very far for peo­ple to start using dif­fer­ent names and stuff. So to keep it you know, with­in this dialect region we’ve recruit­ed ourselves.

One draw­back to using the local media is that you can nev­er be sure of what head­lines are going to put your pic­ture next to. So just a cou­ple of unfor­tu­nate inci­dents where my face has appeared next to these head­lines, I real­ly should get lawyers involved at some point.

So we use goril­la to run this lan­guage switch­ing par­a­digm and mate­ri­als are avail­able as open mate­ri­als. And this con­sist­ed of par­tic­i­pants com­plet­ing a sin­gle lan­guage block in one vari­ety, they then moved on to com­plet­ing four vol­un­tary blocks with they could pick whichev­er item came to mind. And then they would fin­ish by com­plet­ing anoth­er sin­gle lan­guage book in the oth­er vari­ety, and we coun­ter­bal­ance the order of those across the participants.

We used in this study, only used non-cog­nate items. So these items that don’t have any phono­log­i­cal over­lap with each oth­er. So that’d be things like girl and Lassie, ears and lugs, ear­ache and forky­tai­ly, and sand­wich and piece. And par­tic­i­pants com­plet­ed this at home. It only required the use of a device that had a micro­phone and most­ly desk­top PCs and lap­tops. And because we’re cap­tur­ing audio, we required a qui­et environment.

13:09
So what we then did with the audio tri­als is we took them and re uploaded them into a new task so that we could code their accu­ra­cy for the sin­gle lan­guage blocks where they were instruct­ed what they had to name. And for the vol­un­tary blocks, we could then code which vari­ety they had used in that, for that tri­al. And we also uploaded the audio files to chron­set which extracts their response times.

And com­bin­ing these togeth­er basi­cal­ly gives us what we need for our analy­sis. So what did we find? In terms of this mix­ing analy­sis, we are com­par­ing here, the tri­als from the sin­gle lan­guage blocks with the rep­e­ti­tion tri­als from the vol­un­tary blocks, and we use lin­ear mixed effects mod­els to do this. We find a main effect of lan­guage vari­ety in that the stan­dard Eng­lish items were named more slow­ly than the dialect items. Bear­ing in mind, pri­or to the task, we’ve con­trolled for things like word length, and things like that. So so peo­ple were slow­er to name in the stan­dard than they were the dialect.

We did­n’t find any effects of block type. So the sin­gle lan­guage tri­als and the vol­un­tary rep­e­ti­tion tri­als were approx­i­mate­ly pret­ty much the same. And we did­n’t find an inter­ac­tion between lan­guage vari­ety and block type either.

So our take home here is that we did­n’t find any evi­dence for a mix­ing ben­e­fit in this group of speak­ers. But nor did we find any mix­ing cost. I haven’t report­ed it here but we also mea­sured the switch­ing costs in the vol­un­tary blocks and we did find the same effect that you would find for bilin­guals. So You know, some some aspects repli­cat­ed, but we did­n’t observe any mix­ing ben­e­fit or cost here. And although all our pre­vi­ous stud­ies that we’ve done with these types of bidilect­cals have shown effects that we find for bilin­guals, this is the first time we’ve not found a sim­i­lar effect. So it sug­gests that poten­tial­ly the by dialec­tal, and bilin­gual lan­guage con­trol process­es are not nec­es­sar­i­ly iden­ti­cal. But this has giv­en us more work for us to inves­ti­gate this fur­ther. So I just like to thank my col­lab­o­ra­tor, Dr. Math­ieu Decler­ck, and the Carnegie trust, who fund­ed this research. And that’s me.

Jo Ever­shed 15:44
Neil, that’s fab­u­lous. You snuck it in there in the mid­dle, did I get it right that you got these audi­to­ry record­ings online? And then you extract­ed those audi­to­ry record­ings, and then you put those back on pro­lif­ic and got pro­lif­ic work­ers to score them from

Neil Kirk 15:57
No, no, no. In this case? I mean, we could do that. It’s, it’s crossed my mind to do that. No, we for this pur­pose cod­ed them our­selves. But used a task in goril­la to do that, because then it’s easy to share amongst peo­ple. And it’s Yeah,

Jo Ever­shed 16:16
and you can get three rat­ings for each one and see how that

Neil Kirk 16:18
Exact­ly, yeah, so I’ve done that. We’ve done that in all our exper­i­ments, basi­cal­ly cre­ate a, it’s also avail­able as open mate­ri­als. Yeah, like an accu­ra­cy cod­ing task that, again, just uses the plat­form, because, you know, goril­la is so easy to use.

Jo Ever­shed 16:34
Ah, thanks, Neil. No, I do know oth­er researchers who do put their accu­ra­cy cod­ing back onto pro­lif­ic and get their pro­lif­ic work­ers to do it for them, which seems to work real­ly well as well. And then they are the ones where they, they sit togeth­er, they let them through. And then when they only when there’s a dis­par­i­ty, they go and check them man­u­al­ly. They do them them­selves. So that works quite well.

Now, you did­n’t say a lot about what it was like recruit­ing with. Using media, I know you’ve recruit­ed in the news­pa­pers and using radio, can you just say a lit­tle bit more about that. And while Neil is chat­ting, if you’ve got oth­er ques­tions, put them into the q&a. And hope­ful­ly, I will get around to them.

Neil Kirk 17:09
So one aspect of this work is it seems to cap­ture sort of the pub­lic’s inter­est. And any­time we put a press release about this, we always get some kind of cov­er­age. So it always makes the sort of, I guess, a bit of a quirky new sto­ry some­times. So there’s been a cou­ple of expe­ri­ences when I first got this fund­ing and launched my lab, and we were run­ning some stud­ies, I did some radio inter­views on nation­al radio, which led to, I think, two par­tic­i­pants respond­ing and nei­ther had usable data. So almost like the big­ger that we the the sort of broad­er the reach, the less ben­e­fi­cial it was because you know, then it was­n’t so rel­e­vant to everyone.

When we’ve done this, for, we actu­al­ly run two projects at the same time. So we recruit­ed for both of them this way. I cer­tain­ly found a lot of inter­est, a lot of emails com­ing from, from peo­ple who are inter­est­ed, but actu­al­ly the fol­low through of peo­ple to to com­plete it was it was maybe only about 50%. So I’m here today to hear oth­er peo­ple’s tips for recruit­ment, because I’m find­ing when you when I don’t use a recruit­ment plat­form, even when I’m pay­ing, I have to say quite gen­er­ous­ly par­tic­i­pa­tion. Even using media, some­times there are still chal­lenges with get­ting peo­ple who have shown an inter­est to fol­low through to com­ple­tion. So I tried a few lit­tle tips like bit of nudg­ing a bit of you know, this. You know, that might only be the study might only run until we’ve got an x number.

18:50
Yeah, kind of did­n’t work. I think all it did was back­fire slight­ly, and peo­ple thought, Oh, I’ve missed, I’ve missed it. I should have done it yes­ter­day. So it’s, it has its chal­lenges, but cer­tain­ly we would­n’t have I think recruit­ed in the time peri­od that we did with­out that media cov­er­age. It was a huge help.

Jo Ever­shed 19:07
Okay, so it helps but it’s not it’s not the the Nir­vana that we’re all hop­ing for.

Neil Kirk 19:14
Poten­tial. I think the oth­er aspect is what I’ve noticed, with this type of speech pro­duc­tion research, I think peo­ple are just that bit more hes­i­tant about doing I think peo­ple can be slight­ly more self con­scious about doing this type of work. It also requires that they have, you know, a par­tic­u­lar type of set­up, and that they can do it in a place where there’s no dis­trac­tions and you know, often at home, that’s that isn’t the case. And so that can some­times I think, be a slight bar­ri­er. It’s yeah, I think it’s with more com­pre­hen­sion based or where they weren’t hav­ing to pro­duce I think we would have got a, been more successful,

Jo Ever­shed 19:52
I think, yeah, it’s sort of the death by 1000 cuts, isn’t it that you lose a few because of this and a few because of that, and each time you’re like, oh, so I’ve only got a few. Have you ever tried recruit­ing with char­i­ty part­ners like are there oth­er Scot­tish char­i­ties that focus par­tic­u­lar­ly on regions?

Neil Kirk 20:07
No, that’s real­ly a good shout. I’ve got a few con­nec­tions because because this this lan­guage vari­ety forms part of the Scots lan­guage, I’ve got a few con­nec­tions like that would have been able to recruit. Basi­cal­ly, with any of these tasks, we use every tac­tic we can think of. So there’s a few things like that. But yeah, that’s cer­tain­ly some­thing, espe­cial­ly now that we might be able to not every­thing has to be done online. In terms of how you reach peo­ple in the first place. Yeah, some­thing to think about for sure.

Jo Ever­shed 20:40
There’s a ques­tion for you in the q&a from Sachin to ques­tions about the pro­ce­dure, did you record a par­tic­i­pants respons­es in a block from the first or last tri­al? And how did you decide the onset of each vocal response?

Neil Kirk 20:52
So what the audio, was it called the audio record­ing zone does is it cap­tures every response on a tri­al by tri­al basis. So we did­n’t have to then splice them into indi­vid­ual tri­als, we get an out­put that is basi­cal­ly one an audio out­put for every sin­gle tri­al. And we use the chron­set sys­tem, which is an online serv­er, you can upload your files into an extracts the response times, in some of our pre­vi­ous stud­ies, what we’ve done is tak­en a sam­ple of the audio and we’ve man­u­al­ly mea­sured them as well. And we’ve shown that, you know, they’re pret­ty much bang on. So I’m quite trust­ing of the chron­set sys­tem to give me some accu­rate respons­es. In our accu­ra­cy cod­ing, what we also have is if we can detect extra­ne­ous noise, we’ll code that as a, basi­cal­ly a no or oth­er signs so we know to exclude it because the reac­tion time mea­sure­ment might have been messed up. So we have that kind of extra qual­i­ty check when we’re doing an accu­ra­cy coding.

Jo Ever­shed 21:59
Bril­liant. I’m sure there are lots of audi­to­ry resources here who are going to be delight­ed to hear about con­cepts. So audi­ence in the chat if you can share what was the one mes­sage you took away from Neil’s talk today that you’re going to take back to your research what was help­ful, that would be fan­tas­tic. That’s your thank you to Neil. Neil, you’ve been great.