Popular in Social Sciences

Books

City Time: On Being Sentenced to Rikers Island

Independent Scholar David A. Campbell

Culture and Society

The City as a Growth Machine: Toward a Political Economy of Place

New York University Harvey Moltoch

Economics

Gender and Economics

University of California Santa Barbara Shelly Lundberg

American Politics

Keeping the Republic

Boston College Marc Landy

Banks and Policy Organisations

Greening the financial sector

European Investment Bank Matteo Ferrazzi

British Politics

Entrepreneurial thinking: the politics and practice of policy impact

University of Sheffield Matthew Flinders

Culture and Society

Carceral Entrapments: Views from the Prison/Street Interface in India

Indian Institute of Technology Delhi Mahuya Bandyopadhya

Communication Effort and the Cost of Language

View more insights in Social Sciences - Economics - Microeconomics

View more figures from University of Liverpool

View the page of Jacopo Bregolin

The transmission of information is crucial for productivity and growth, but language differences may limit its effectiveness. Jacopo Bregolin discusses how the exogenous cost of language affects communication effort, and the trade-off faced by knowledge platforms in implementing their website in multiple languages.

Image courtesy of interviewee. May 3, 2024

Generating Workspace

Export as PDF

This video video discusses a research paper that investigates how communication friction affects the quality of information transmission, focusing on two primary channels: incentives and exogenous constraints.

The paper highlights that communication friction can arise when information holders lack incentives to share information optimally or face external constraints—like language barriers—that hinder effective communication. The study specifically examines these issues in the context of knowledge platforms, particularly Stack Overflow, which has introduced versions in various languages to accommodate non-English speakers.

Key findings include:
1. Allowing users to communicate in their native language significantly improves the quality of their contributions, increasing coding snippets in responses by 21%.
2. When quality incentives are simultaneously present, the communication quality can further increase to 118%.
3. New users who participate only when their native language is available tend to produce lower quality answers compared to users who were already active in English, suggesting that these less-expert participants may dilute overall content quality.

The implications suggest that platforms should allow users to communicate in their native languages to enhance information quality. However, a balance must be struck, as this openness may invite less knowledgeable users, requiring effective moderation to maintain content quality. The findings underscore the importance for organizations to consider communication frictions when forming multinational teams, highlighting the need for strategies that accommodate diverse language capabilities.

Language:

Export as PDF

Export as PDF
Here are some articles that are related to the themes of communication friction, information transmission quality, and the implications of language barriers in collaborative environments:

1. **"The Impact of Language on Knowledge Sharing in Multilingual Teams"**
- This article explores how language diversity in teams affects collaboration and knowledge sharing. It analyzes the challenges and benefits of having team members communicate in multiple languages.

2. **"Incentives and Information Sharing: Understanding Communication Friction in Organizations"**
- This paper discusses how incentives influence information sharing and the role of communication barriers in organizational settings, emphasizing strategies to enhance information flow.

3. **"Language Barriers and Their Effect on Team Dynamics in Global Organizations"**
- This study examines how language barriers impact team dynamics and communication effectiveness in multinational corporations, providing recommendations for overcoming these challenges.

4. **"Enhancing Online Communication through Language Accommodations: A Study of Online Platforms"**
- This article investigates how online platforms can improve user interaction by allowing communication in native languages, similar to the implications discussed in your video.

5. **"Quality of Contributions in Online Knowledge Platforms: Analyzing User Behavior"**
- This research focuses on the factors that influence the quality of user-generated content on online Q&A platforms, including the relationship between user expertise and the language used for contributions.

6. **"The Role of Language in Collaboration: Mitigating Communication Friction in International Teams"**
- This paper looks into different strategies that organizations can employ to minimize communication friction due to language differences, thus improving collaboration and information exchange.

7. **"Communication Friction in Knowledge Sharing: How External Constraints Affect Team Performance"**
- This article delves into how external constraints—such as cultural differences and language issues—affect the performance of teams in knowledge-intensive industries.

These articles all touch on aspects related to your original video, including communication friction, language barriers, information transmission quality, and the dynamics of knowledge sharing in diverse environments.

X-TIMESTAMP-MAP:LOCAL=00:00:00.000,MPEGTS=900000

00:00.075 --> 00:01.965
This paper wants to study

00:02.785 --> 00:05.765
how communication friction may affect the quality

00:05.825 --> 00:07.285
of information transmissions.

00:07.865 --> 00:10.605
Uh, and, there are two possible channels

00:10.605 --> 00:14.045
that communication over which should communication friction

00:14.145 --> 00:16.365
may affect communication outcomes.

00:16.545 --> 00:18.045
One channel is incentives,

00:18.505 --> 00:21.485
and the other channel is exogenous constraints.

00:21.945 --> 00:25.645
Let me just lead you through this On one side, you know,

00:25.795 --> 00:27.845
potentially if, in the company,

00:28.145 --> 00:31.685
in the information holder, needs to transmit information

00:31.705 --> 00:33.725
to the actual person that will make the action,

00:33.905 --> 00:35.645
if there are no incentives, means

00:35.645 --> 00:39.405
that this information holder may have interest

00:39.865 --> 00:42.445
or of not disclosing that information

00:42.785 --> 00:45.205
or to disclose it in a non-optimal way

00:45.635 --> 00:48.165
that creates the information reduction

00:48.225 --> 00:49.805
and the, the information transmission is

00:49.805 --> 00:50.845
not effective, right?

00:50.905 --> 00:53.845
So that's incentives problems that are very strong and,

00:53.905 --> 00:56.205
and the exist in many different environments

00:57.025 --> 00:59.725
at the same time, there could be situation in which,

00:59.905 --> 01:02.485
the information holder has perfect incentive

01:02.585 --> 01:03.685
to transmit information.

01:03.705 --> 01:06.125
So it really wants to transmit information as much

01:06.125 --> 01:10.205
as possible, but cannot, or cannot in an optimal way

01:10.205 --> 01:11.845
because of exogenous constraints.

01:12.225 --> 01:14.325
So one example of exogenous constraints

01:14.785 --> 01:15.965
are language friction.

01:15.985 --> 01:18.085
So if the person all the information needs

01:18.085 --> 01:19.805
to communicate in a foreign language

01:20.225 --> 01:22.245
and is not fluent in that language,

01:22.315 --> 01:23.565
that may prevent the person

01:23.745 --> 01:25.205
to actually communicate fluently.

01:25.265 --> 01:27.405
So that's kind of the main friction

01:27.405 --> 01:32.365
that I address in this paper, which studies exactly how,

01:32.705 --> 01:35.445
the use of offering language may prevent the

01:36.125 --> 01:37.565
optimal transmission of information.

01:38.765 --> 01:42.545
And, I do this in the context of, of online platforms

01:43.005 --> 01:45.865
and the platform that I call on knowledge platforms.

01:46.175 --> 01:47.825
This means that our platform that

01:49.055 --> 01:51.945
rely on voluntary participation from the internet

01:52.045 --> 01:54.785
and aggregate information from around the world

01:55.085 --> 01:57.745
to build some knowledge, that,

01:57.745 --> 01:59.145
that people can then browse

01:59.145 --> 02:01.145
and use for their, for their product,

02:01.245 --> 02:02.745
for their activity and, and work.

02:03.365 --> 02:06.505
Um, so this platform really rely on high quality content

02:06.615 --> 02:08.545
that comes from very different countries, right?

02:08.545 --> 02:11.385
So they are particularly affected by this type of friction

02:11.385 --> 02:15.025
because now I have people, users from different parts

02:15.025 --> 02:18.065
of the world that, will, will bring in content

02:18.245 --> 02:21.785
and information that everyone wants to, brands and,

02:21.785 --> 02:22.945
and, and, and see.

02:23.405 --> 02:25.545
And so I believe that in this context, this type

02:25.545 --> 02:27.105
of friction is particularly important

02:27.175 --> 02:30.665
because, the quality of the information that the,

02:30.775 --> 02:33.825
this information aggregators collect becomes really

02:34.625 --> 02:36.665
determinant for the success of these platforms.

02:37.205 --> 02:39.055
Yeah, so, so as I mentioned, the,

02:39.055 --> 02:41.335
the setting I'm using is the one on Stack overflow.

02:41.355 --> 02:43.455
So Stack Overflow, just to give a quick introduction

02:43.555 --> 02:47.015
to people that may not be familiar, this is a website that,

02:47.075 --> 02:49.175
provides a question answers

02:49.555 --> 02:52.895
so people from the community can ask question freely

02:53.235 --> 02:55.655
and then other members provide answers.

02:55.755 --> 02:58.335
And this is all based on voluntary participation,

02:58.335 --> 03:02.205
so there is no remuneration of such of sexual interactions.

03:02.825 --> 03:07.005
Um, and so these websites were first,

03:07.005 --> 03:08.565
first introduced in English

03:08.565 --> 03:11.685
because it's a English company, American company actually.

03:11.825 --> 03:12.925
Uh, and then,

03:13.465 --> 03:16.445
and then all communication in this website was constrained

03:16.445 --> 03:17.685
to be in, in English.

03:18.185 --> 03:21.485
Um, so what is interesting is that the platform has decided

03:21.585 --> 03:23.725
to introduce clone versions

03:23.725 --> 03:25.925
of their website in different languages.

03:26.305 --> 03:29.165
And in particular, they, they introduced at different point

03:29.165 --> 03:32.965
in time, our website in Portuguese, in Spanish,

03:33.025 --> 03:34.245
in Japanese and Russian.

03:35.315 --> 03:38.295
And this was really driven by the fact that they wanted to,

03:38.515 --> 03:41.455
increase potentially the reach of the platform

03:41.475 --> 03:43.015
to also communities that are not

03:43.015 --> 03:44.375
proficient in English, right?

03:44.755 --> 03:46.015
So what is most interesting

03:46.015 --> 03:47.495
and useful for my research is

03:47.495 --> 03:49.855
that I can track empirically users

03:49.925 --> 03:53.935
that are participating in English also on the other side.

03:54.195 --> 03:57.775
So basically if there is a native Portuguese speaker,

03:58.295 --> 04:01.895
I can see what this person was producing

04:01.955 --> 04:05.055
and contributing when only English was available,

04:05.795 --> 04:09.575
and then I can track their participation once their

04:09.575 --> 04:11.015
native language became available.

04:11.195 --> 04:14.255
So basically, again, my Portuguese user

04:14.555 --> 04:17.655
before who only write in English now, is also able

04:17.655 --> 04:18.775
to write in Portuguese,

04:19.075 --> 04:20.575
and they can compare the content

04:20.575 --> 04:23.655
that they produced in English and the Portuguese.

04:24.315 --> 04:27.415
So this allows me really to compare their participation and,

04:27.415 --> 04:29.455
and see if the quality of their communication

04:29.595 --> 04:30.855
as change it, right?

04:30.875 --> 04:34.495
So that's the kind of the main, the main empirical setting

04:34.525 --> 04:35.535
that I exploit.

04:35.915 --> 04:38.135
And I will also explore the fact that,

04:38.495 --> 04:42.375
I can observe users that were not active in English before,

04:42.515 --> 04:44.095
but now were able

04:44.155 --> 04:47.135
to start participating once their native language

04:47.135 --> 04:48.415
became available, right?

04:48.435 --> 04:52.335
So these kind of the, things I can observe in the data,

04:52.335 --> 04:54.215
which allow me to carry this research.

04:54.755 --> 04:59.055
Um, and then normally at this point, most people ask me,

04:59.485 --> 05:02.175
what is the measure of quality that, that you're using?

05:02.175 --> 05:03.855
Because, you know, like as I said

05:03.855 --> 05:05.975
before, I think one big contribution of this paper is

05:05.975 --> 05:08.535
that I actually can, can see in the text

05:08.835 --> 05:11.535
of the communication what is the quality,

05:11.755 --> 05:12.815
but then the question is

05:12.835 --> 05:14.855
how I can actually quantify that quality.

05:15.555 --> 05:19.255
Um, so just to give a word about this,

05:19.675 --> 05:20.815
of course I wanted a measure

05:20.815 --> 05:22.815
that is comparable across different

05:22.815 --> 05:24.055
sites and different languages.

05:24.155 --> 05:25.855
So I could not be just a text measure

05:25.885 --> 05:27.655
that produce language specific.

05:28.195 --> 05:30.055
So what I used, I rely on the fact

05:30.055 --> 05:32.495
that Stack Workflow is specifically about programming

05:32.735 --> 05:36.375
languages and coding, and then I found the number

05:36.435 --> 05:39.815
of separate snippets of code that appear in an answer.

05:40.195 --> 05:42.735
So basically, people normally as a,

05:42.735 --> 05:45.095
when they answer a question about programming, they tend

05:45.095 --> 05:46.415
to write some explanation

05:46.475 --> 05:49.535
and then introduce some code that, that people can use.

05:49.995 --> 05:52.895
Um, so they, they, the hypothesis here is,

05:52.895 --> 05:55.295
or the assumption if you want, is that,

05:55.475 --> 05:58.325
if I have text code, text code multiple times,

05:59.025 --> 06:02.285
I'm either guiding you a clear step-by-step procedure.

06:02.385 --> 06:05.565
So more, more clear clarity on the, on the process that the,

06:05.665 --> 06:07.245
the questionnaire should implement

06:07.425 --> 06:09.645
to solve the answer the problem,

06:10.755 --> 06:12.855
or I'm just giving you more information, right?

06:12.875 --> 06:15.015
So I'm giving you different pieces of code

06:15.015 --> 06:16.495
because there are complementary inform.

06:17.315 --> 06:20.775
So either way I'm giving you a higher quality answer

06:20.775 --> 06:23.935
because it provides either more clarity or more information.

06:24.355 --> 06:27.815
But I'm using the number of separate snippet of code

06:27.995 --> 06:29.935
as a way to measure quality of communication.

06:30.195 --> 06:32.095
And I like this measure because it really based,

06:32.095 --> 06:34.535
based on the choice of communication

06:34.565 --> 06:36.335
that the answer has made.

06:36.795 --> 06:39.095
Um, of course, I will compare also these measure

06:39.245 --> 06:41.215
with measure based on outcomes

06:41.675 --> 06:42.895
and in particular on,

06:43.075 --> 06:46.575
on whether the answer has been selected as best answer.

06:46.795 --> 06:49.655
So, as in many of these platforms,

06:49.675 --> 06:52.815
so the question I can then pick which answer actually

06:52.855 --> 06:53.895
solved the problem, right?

06:53.915 --> 06:56.535
So that's also sort of measure of, of, of quality

06:57.065 --> 07:00.215
based on whether actually the answer was successful in,

07:00.215 --> 07:01.775
in helping the questionnaire.

07:02.115 --> 07:06.135
So, yeah, so, the, so I carry two main analysis.

07:06.135 --> 07:09.655
The first analysis is focused on understanding whether the

07:09.655 --> 07:13.375
use of a language did impact communication quality.

07:13.635 --> 07:17.095
And what I find is that being able

07:17.115 --> 07:21.255
to use the native language rather than English allow leads

07:21.275 --> 07:24.855
people to increase their communication quality by 21%.

07:25.075 --> 07:27.695
So this means that the, the number of pieces of code

07:27.695 --> 07:31.175
that the, the, the answer contain increases on average

07:31.395 --> 07:33.535
by 21% percentage points.

07:35.235 --> 07:39.055
So, so this is a substantial, that substantial effect.

07:39.195 --> 07:41.855
And what is interesting is that this effect is complimentary

07:41.855 --> 07:42.895
to other dimensions.

07:42.915 --> 07:46.295
So as I mentioned at the beginning of the, I, I said

07:46.295 --> 07:47.295
that,

07:47.295 --> 07:50.095
incentives also may matter in communication friction.

07:50.315 --> 07:52.015
So what I also do in this paper is

07:52.015 --> 07:53.375
to look at incentives as well.

07:53.795 --> 07:56.615
And theoretically speaking, we should expect

07:56.615 --> 07:58.415
that incentives may also play a role.

07:58.675 --> 08:01.375
And what I find is that these exogenous cost

08:01.375 --> 08:04.215
of language are complimentary to incentives, meaning

08:04.215 --> 08:07.815
that if I allow a person to speak their native language,

08:07.875 --> 08:11.255
but I also incentivize this person to provide high effort,

08:11.685 --> 08:15.015
then the overall increasing quality is much higher.

08:15.155 --> 08:18.815
So I find an increase by 118 percent,

08:19.515 --> 08:22.655
and also there is a complementarity with the questionnaire

08:23.375 --> 08:24.695
que, effort.

08:24.875 --> 08:26.135
So the question quality.

08:26.595 --> 08:30.895
So if, the question quality is higher, the impact

08:30.915 --> 08:33.375
of being able to use the native language

08:33.515 --> 08:34.655
is actually even larger.

08:36.235 --> 08:39.775
So this shows overall that there is a positive effect

08:39.995 --> 08:41.815
of allowing people,

08:41.815 --> 08:44.655
to use the native language in their communication quality.

08:46.085 --> 08:48.505
On the, on the downside, what I find is

08:48.505 --> 08:52.705
that the new joiners, so join, so users that started

08:53.225 --> 08:55.325
participating only when the native

08:55.685 --> 08:56.725
language became available.

08:56.985 --> 08:59.645
So these that were not able to participate in English,

08:59.745 --> 09:02.925
but were able to participate in their native language,

09:03.515 --> 09:04.725
this user, you know,

09:04.785 --> 09:08.685
in fact produce lower quality answers compared to users

09:08.685 --> 09:10.565
that were already active in English.

09:11.645 --> 09:16.425
So what the, the kind of mechanism that I, I, I suggest

09:16.525 --> 09:18.265
to explain this fact is that

09:19.145 --> 09:22.105
whenever users are only able to use English,

09:22.815 --> 09:25.825
they face the cost of using a foreign language.

09:26.285 --> 09:29.865
But if they are very highly expert of their, of their

09:30.455 --> 09:33.265
subject, then they will still participate.

09:33.835 --> 09:38.745
While, while if users are not native in English,

09:38.775 --> 09:42.545
then they are not expert, then they will not participate.

09:42.825 --> 09:46.945
Nevertheless, these users that are not expert are now able

09:46.945 --> 09:49.745
to participate because they can use their native language.

09:50.405 --> 09:53.385
So to some, to some extent, what I'm saying is

09:53.385 --> 09:57.025
that the language cost, the language friction act as a,

09:57.145 --> 10:01.065
a selection device or as a barrier basically to entry.

10:01.125 --> 10:04.185
So users that, have higher expertise,

10:04.695 --> 10:07.025
even if they have a high cost of language,

10:07.095 --> 10:09.825
will still participate in English while user

10:09.825 --> 10:11.305
with low expertise will not.

10:11.685 --> 10:15.265
So basically allowing users to use their native language

10:15.935 --> 10:19.065
will induce low, less, less expert user

10:19.245 --> 10:20.585
to start participating.

10:21.045 --> 10:23.385
So this is kind of a downside for the platform

10:23.385 --> 10:26.625
because, while improved communication quality from the

10:26.625 --> 10:30.065
first set of users now is bringing in more users

10:30.175 --> 10:31.785
that are less expert, right?

10:31.845 --> 10:33.025
So first of all,

10:33.025 --> 10:35.185
there are policy implication for the platform.

10:35.485 --> 10:37.825
So the platform now platform, no,

10:38.015 --> 10:39.705
that I call knowledge platform platform

10:39.705 --> 10:41.825
that rely on on users that come from all,

10:41.825 --> 10:44.745
all over the world, that face this, this problem of having

10:45.655 --> 10:48.385
information aggregated from, from the people

10:48.385 --> 10:50.985
that speak different languages, you know,

10:51.095 --> 10:52.105
face this challenge,

10:52.245 --> 10:54.185
and the main policy implications that yes,

10:54.335 --> 10:55.585
that affects the quality.

10:55.985 --> 10:58.185
I mean, the fact that these users cannot use the native

10:58.505 --> 11:00.505
language affects the quality of their communication

11:00.525 --> 11:01.705
and the information they provide.

11:01.965 --> 11:06.225
So indeed, if the quality is really one of the major concern

11:06.245 --> 11:08.025
of the platform, they should allow users

11:08.165 --> 11:09.625
to use the native languages.

11:09.845 --> 11:12.225
And this is a bit what we see in Wikipedia, which,

11:12.225 --> 11:15.665
finally has allowed all other language to appear out there,

11:16.125 --> 11:18.865
because other, they care a lot about quality

11:19.005 --> 11:21.905
and they have less, problem other dimensions.

11:22.565 --> 11:26.425
At the same time, we have this problem that, the,

11:26.525 --> 11:30.425
they allow platforms to increase, to, to,

11:30.425 --> 11:32.425
to introduce different languages,

11:32.855 --> 11:35.745
will lead the less e user to participate.

11:36.045 --> 11:38.505
So then the policy implication here is

11:38.505 --> 11:43.185
that if the platform has in place a very strong moderation,

11:43.965 --> 11:46.545
program, then, then

11:46.615 --> 11:48.465
that should not be a big problem, right?

11:48.525 --> 11:49.705
And then should compensate.

11:49.705 --> 11:52.865
And so the platform may still want to have as many

11:53.165 --> 11:56.525
Languages as possible and then moderate very well, so

11:56.525 --> 11:58.805
that all these type of lower quality content

11:59.105 --> 12:00.405
are, are faded out.

12:01.025 --> 12:03.405
Uh, but if this is not the case,

12:03.875 --> 12:06.005
then potentially the platform may prefer

12:06.065 --> 12:09.725
to keep fewer languages so that the barriers will still

12:10.275 --> 12:14.365
prevent not too much participation from less expert users.

12:15.185 --> 12:17.005
So these are kind of main policy implications

12:17.025 --> 12:18.525
for this specific platform.

12:18.875 --> 12:21.845
Then of course, you know, it's always the kind

12:21.845 --> 12:23.325
of the big challenge for economists

12:23.385 --> 12:25.925
to draw more general conclusions.

12:25.985 --> 12:28.005
And of course, the settings may not allow

12:28.845 --> 12:31.525
complete external validity to other settings, but I will.

12:31.785 --> 12:33.885
But overall, this paper suggests that

12:34.605 --> 12:37.125
whenever firms collaborate across countries,

12:37.225 --> 12:39.285
or when teams are multinational

12:39.305 --> 12:41.365
and there are different people working from different

12:41.645 --> 12:43.645
countries and so on, then these are factors

12:43.645 --> 12:45.165
that should be taken into account, right?

12:45.225 --> 12:48.045
So, there are communication friction, they exist

12:48.265 --> 12:50.805
and they provide the potential,

12:50.915 --> 12:52.765
information quality effects.

12:53.345 --> 12:54.885
So the, the companies

12:54.885 --> 12:56.245
and organizations should take

12:56.245 --> 12:57.925
that into account when they build teams

12:58.305 --> 13:01.005
and multinational, multinational, interactions.

00:00 - This paper wants to study how communication friction may affect the quality of information transmissions. and, there are two possible channels that communication over which should communication friction may affect communication outcomes.

00:16 - One channel is incentives, and the other channel is exogenous constraints. Let me just lead you through this On one side, you know, potentially if, in the company, in the information holder, needs to transmit information to the actual person that will make the action, if there are no incentives, means that this information holder may have interest or of not disclosing that information or to disclose it in a non-optimal way

00:45 - that creates the information reduction and the, the information transmission is not effective, right? So that's incentives problems that are very strong and, and the exist in many different environments at the same time, there could be situation in which, the information holder has perfect incentive to transmit information. So it really wants to transmit information as much as possible, but cannot, or cannot in an optimal way

01:10 - because of exogenous constraints. So one example of exogenous constraints are language friction. So if the person all the information needs to communicate in a foreign language and is not fluent in that language, that may prevent the person to actually communicate fluently. So that's kind of the main friction that I address in this paper, which studies exactly how,

01:32 - the use of offering language may prevent the optimal transmission of information. And, I do this in the context of, of online platforms and the platform that I call on knowledge platforms. This means that our platform that rely on voluntary participation from the internet and aggregate information from around the world to build some knowledge, that, that people can then browse and use for their, for their product,

02:01 - for their activity and, and work. so this platform really rely on high quality content that comes from very different countries, right? So they are particularly affected by this type of friction because now I have people, users from different parts of the world that, will, will bring in content and information that everyone wants to, brands and, and, and, and see. And so I believe that in this context, this type of friction is particularly important

02:27 - because, the quality of the information that the, this information aggregators collect becomes really determinant for the success of these platforms. Yeah, so, so as I mentioned, the, the setting I'm using is the one on Stack overflow. So Stack Overflow, just to give a quick introduction to people that may not be familiar, this is a website that, provides a question answers so people from the community can ask question freely and then other members provide answers.

02:55 - And this is all based on voluntary participation, so there is no remuneration of such of sexual interactions. and so these websites were first, first introduced in English because it's a English company, American company actually. and then, and then all communication in this website was constrained to be in, in English. so what is interesting is that the platform has decided to introduce clone versions

03:23 - of their website in different languages. And in particular, they, they introduced at different point in time, our website in Portuguese, in Spanish, in Japanese and Russian. And this was really driven by the fact that they wanted to, increase potentially the reach of the platform to also communities that are not proficient in English, right? So what is most interesting and useful for my research is

03:47 - that I can track empirically users that are participating in English also on the other side. So basically if there is a native Portuguese speaker, I can see what this person was producing and contributing when only English was available, and then I can track their participation once their native language became available. So basically, again, my Portuguese user before who only write in English now, is also able to write in Portuguese,

04:19 - and they can compare the content that they produced in English and the Portuguese. So this allows me really to compare their participation and, and see if the quality of their communication as change it, right? So that's the kind of the main, the main empirical setting that I exploit. And I will also explore the fact that, I can observe users that were not active in English before, but now were able

04:44 - to start participating once their native language became available, right? So these kind of the, things I can observe in the data, which allow me to carry this research. and then normally at this point, most people ask me, what is the measure of quality that, that you're using? Because, you know, like as I said before, I think one big contribution of this paper is that I actually can, can see in the text of the communication what is the quality,

05:11 - but then the question is how I can actually quantify that quality. so just to give a word about this, of course I wanted a measure that is comparable across different sites and different languages. So I could not be just a text measure that produce language specific. So what I used, I rely on the fact that Stack Workflow is specifically about programming

05:32 - languages and coding, and then I found the number of separate snippets of code that appear in an answer. So basically, people normally as a, when they answer a question about programming, they tend to write some explanation and then introduce some code that, that people can use. so they, they, the hypothesis here is, or the assumption if you want, is that, if I have text code, text code multiple times, I'm either guiding you a clear step-by-step procedure.

06:02 - So more, more clear clarity on the, on the process that the, the questionnaire should implement to solve the answer the problem, or I'm just giving you more information, right? So I'm giving you different pieces of code because there are complementary inform. So either way I'm giving you a higher quality answer because it provides either more clarity or more information. But I'm using the number of separate snippet of code as a way to measure quality of communication.

06:30 - And I like this measure because it really based, based on the choice of communication that the answer has made. of course, I will compare also these measure with measure based on outcomes and in particular on, on whether the answer has been selected as best answer. So, as in many of these platforms, so the question I can then pick which answer actually solved the problem, right?

06:53 - So that's also sort of measure of, of, of quality based on whether actually the answer was successful in, in helping the questionnaire. So, yeah, so, the, so I carry two main analysis. The first analysis is focused on understanding whether the use of a language did impact communication quality. And what I find is that being able to use the native language rather than English allow leads people to increase their communication quality by 21%. So this means that the, the number of pieces of code

07:27 - that the, the, the answer contain increases on average by 21% percentage points. So, so this is a substantial, that substantial effect. And what is interesting is that this effect is complimentary to other dimensions. So as I mentioned at the beginning of the, I, I said that, incentives also may matter in communication friction. So what I also do in this paper is to look at incentives as well.

07:53 - And theoretically speaking, we should expect that incentives may also play a role. And what I find is that these exogenous cost of language are complimentary to incentives, meaning that if I allow a person to speak their native language, but I also incentivize this person to provide high effort, then the overall increasing quality is much higher. So I find an increase by 118 percent, and also there is a complementarity with the questionnaire que, effort.

08:24 - So the question quality. So if, the question quality is higher, the impact of being able to use the native language is actually even larger. So this shows overall that there is a positive effect of allowing people, to use the native language in their communication quality. On the, on the downside, what I find is that the new joiners, so join, so users that started participating only when the native

08:55 - language became available. So these that were not able to participate in English, but were able to participate in their native language, this user, you know, in fact produce lower quality answers compared to users that were already active in English. So what the, the kind of mechanism that I, I, I suggest to explain this fact is that whenever users are only able to use English, they face the cost of using a foreign language.

09:26 - But if they are very highly expert of their, of their subject, then they will still participate. While, while if users are not native in English, then they are not expert, then they will not participate. Nevertheless, these users that are not expert are now able to participate because they can use their native language. So to some, to some extent, what I'm saying is that the language cost, the language friction act as a, a selection device or as a barrier basically to entry. So users that, have higher expertise,

10:04 - even if they have a high cost of language, will still participate in English while user with low expertise will not. So basically allowing users to use their native language will induce low, less, less expert user to start participating. So this is kind of a downside for the platform because, while improved communication quality from the first set of users now is bringing in more users that are less expert, right?

10:31 - So first of all, there are policy implication for the platform. So the platform now platform, no, that I call knowledge platform platform that rely on on users that come from all, all over the world, that face this, this problem of having information aggregated from, from the people that speak different languages, you know, face this challenge, and the main policy implications that yes,

10:54 - that affects the quality. I mean, the fact that these users cannot use the native language affects the quality of their communication and the information they provide. So indeed, if the quality is really one of the major concern of the platform, they should allow users to use the native languages. And this is a bit what we see in Wikipedia, which, finally has allowed all other language to appear out there, because other, they care a lot about quality

11:19 - and they have less, problem other dimensions. At the same time, we have this problem that, the, they allow platforms to increase, to, to, to introduce different languages, will lead the less e user to participate. So then the policy implication here is that if the platform has in place a very strong moderation, program, then, then that should not be a big problem, right? And then should compensate.

11:49 - And so the platform may still want to have as many Languages as possible and then moderate very well, so that all these type of lower quality content are, are faded out. but if this is not the case, then potentially the platform may prefer to keep fewer languages so that the barriers will still prevent not too much participation from less expert users. So these are kind of main policy implications for this specific platform.

12:18 - Then of course, you know, it's always the kind of the big challenge for economists to draw more general conclusions. And of course, the settings may not allow complete external validity to other settings, but I will. But overall, this paper suggests that whenever firms collaborate across countries, or when teams are multinational and there are different people working from different countries and so on, then these are factors

12:43 - that should be taken into account, right? So, there are communication friction, they exist and they provide the potential, information quality effects. So the, the companies and organizations should take that into account when they build teams and multinational, multinational, interactions.

Faculti AI can make mistakes