How differences in the information density of languages leads to different user experiences on Twitter is something I published on in 2015 with Han-Teng Liao and King-wa Fu and also a question that researchers at the Nara Institute of Science and Technology (NAIST) investigated from a different prospective.

Ostensibly Twitter launched with a 140-character limit to allow tweets to be “consumed easily anywhere, even via mobile text messages” (SMS). Of course, text messages have a byte limit, not a character limit; so, tweets only ever fit within one text message for a small number of languages. Furthermore, what seemed like a universal design choice—140 characters—actually has rather different effects in different languages. These differences are most pronounced for Japanese and Chinese, and Twitter has done particularly well in Japan, a success it wants to replicate in more locations.

There are a number of ways we can measure how much information is conveyed by one character. Regardless of the measure, however, it is clear that a language like Japanese contains more information per character than English (e.g., 政治 vs. politics). In our research, we used translations of TED talks to arrive at a figure of one Japanese character approximating 2.5 English characters (and one traditional Chinese character approximating 3.2 English characters). That isn’t the whole story, however, because the amount of information in a tweet is also a function of its length, and tweets in Japanese and Chinese tend to have fewer characters than English tweets. Nonetheless, we do find that tweets in Japanese and Chinese tend to have more information content than tweets in English on average. Twitter’s move to double the amount of characters for languages other than Chinese, Japanese, and Korean (CJK) while somewhat arbitrary should allow more than enough space to fit the information content of the average Japanese or Chinese tweet in English.

The big question for Twitter is whether the move will increase use of and engagement with the platform. Here, unfortunately, there is a lot more than the character limit at play. Japan has long been an outlier in technology use, leading to what the Japanese have dubbed the Galápagos effect (ガラパゴス化). Its mobile phones had email in 1999, cameras in 2000, 3G in 2001, music downloads in 2002, e-money in 2004, and digital TV in 2005—all long before other markets.

Twitter was embraced quickly in the country and Japanese users of Twitter form a dense, but isolated, portion of the Twitter network. Japanese also stands out as a unique language community on TripAdvisor, Wikipedia (in terms of article creation by anonymous users, bilingualism, and content differences [pdf]), and other platforms. Ultimately, the use of a given technology is a function not only of its user experience (which Twitter can shape as it has with the character limit), but also many elements of culture that are beyond the control of the company.

Similar tweets from @usembassytokyo in Japanese and English. The embassy often includes more detail/explanation in its Japanese-language tweets.

Even within a single language, Twitter is supporting a number of different use cases simultaneously (staying in touch with friends, following news, co-experiencing a live event or conference, etc.). More characters will enhance some of these, be unimportant to others, and perhaps detract from still others. There simply is not a single technological fix that will enhance everyone’s experience across languages and use cases. One possible character-related change is adding a “soft limit”—showing only the first X characters of a tweet by default and providing a “…more” link that can be clicked to show the rest of the tweet. This would balance the compact display of tweets, encourage users to strive for conciseness, and yet also accommodate use cases where more information is truly needed.

A figure from the book was also included in a tweet and blog post by The Economist:

How social media give voice and power to people who have neither (and to Donald Trump) https://t.co/7TLz4VmHI7 pic.twitter.com/QLoDFTCkz5

— The Economist (@TheEconomist) March 28, 2016

From the special report:

A new book entitled “Political Turbulence” come[s] to an intriguing conclusion: social media are making democracies more “pluralistic”, but not in the conventional sense of the word, involving diverse but stable groups. Instead, the authors see the emergence of a “chaotic pluralism”, in which mobilisations spring from the bottom up, often reacting to events. Online mobilisation can develop explosively and seemingly at random. …
Politics in the age of social media, the authors conclude, is better described by chaos theory than by conventional social science: “Tiny acts of political participation that take place via social media are the units of analysis, the equivalent of particles and atoms in a natural system, manifesting themselves in political turbulence.” One day, say the authors, it [might] be possible to predict and trigger such surges, in the same way that meteorologists have become good at forecasting the weather. …Read more (paywall).

… contributes an important series of creatively and rigorously researched insights into the social mechanics of Internet-based collective action, handing researchers a new toolbox of methods and techniques in the process. …Read more (paywall)

In a thought-provoking new book, Political Turbulence: How Social Media Shape Collective Action, Professor Helen Margetts and her colleagues at the Oxford Internet Institute provide empirical evidence that social media are starting to change our politics in ways not yet appreciated or understood. Platforms such as Twitter, they write, are providing “zero-touch co-ordination for micro-donations of time, effort, and money and are replacing organisations and institutions in some areas of political life. Indeed, organisations increasingly resemble social media platforms in the way they present themselves to the public, with facilities for commenting and encouraging the sharing of content.” …Read more

A few years back I was intrigued and captivated, as a largely analogue political animal, by Paul Mason’s Why It’s Kicking Off Everywhere on the revolutionary part that social media were playing in the Arab Spring and global politics. But for all the enthusiasm, anecdotes and insights, it was ultimately unsatisfying. Here now is a revelatory study, Political Turbulence, which looks more closely and systematically at why “it” – that is, significant collective action – “is kicking off”, and why, much more frequently, it doesn’t. …Read more

Scott Hale, a data scientist focused on bilingualism at the Oxford Internet Institute, told me that user interfaces could help break down language barriers by allowing users to interact … in multiple languages at once. Allowing people to easily toggle between languages is one way to break down the linguistic silos that online life creates. …Read more

Political Turbulence: we’re ‘dripping with data’ and it may make democracy better

Do social media shape collective action? Professor Helen Margetts, co-author of a new book called “Political Turbulence” says they do. By allowing us to make “micro-donations” it’s easy to join a cause. …Read more.

“Chaotic pluralism…a new kind of pluralism, highly decentred and chaotic” is what we’re living through, if we are to believe the authors of Political Turbulence. The authors, whose disciplinary backgrounds range across political science, computational science and physics, argue that this new status quo has resulted from the intrusion, if that’s the right word, of social media into the political sphere, an intrusion that they describe as “unstable, unpredictable and often unsustainable”. …Read more

On Start the Week Tom Sutcliffe talks to the American writer Jonathan Franzen about his latest novel, Purity. One of Franzen’s characters compares the Internet with the East German Republic and he satirises the utopian ideas of the apparatchik web-users. The head of the Oxford Internet Institute, Helen Margetts, counters with her research on the success and failure of political action via social media. The artist Tacita Dean laments the ubiquity of digital at the expense of film, and the financial journalist Gillian Tett roots out tunnel vision – both personal and business – in her new book on silos.…Listen now

Translation technologies offer one solution to bridging online language divides, while also opening up new markets for businesses. Although currently only available in a few languages, last year Microsoft launched the Skype translator, and both Facebook and Twitter have also paired up with Bing to offer users translation services.

Scott Hale, data scientist at the Oxford Internet Institute, argues that more could also be done to unlock the power of multilinguals online. Internet platforms he believes could be modified to make it easier for multilingual users to find content in other languages, as well as encourage them to contribute in more than one language. “Many review sites, such as TripAdvisor and Google Play, prioritise reviews in a person’s selected user-interface language or even completely hide reviews not in the user-interface language,” says Hale. Platforms like Wikipedia, he says, could allow you to search a topic in multiple language editions at the same time.

Hale also found that although only 11% of people are multilingual on Twitter, and 15% on Wikipedia, these multilingual individuals are more active, writing more tweets and creating and editing more Wikipedia content. These people, he believes, could potentially challenge the Balkanisation of information and discussion online. Whether it is translating and bringing foreign concepts into different language editions on Wikipedia, or moving breaking local news stories to new language communities and different geographies, they have the power to be influential.

Full article is available at http://labs.theguardian.com/digital-language-divide/. Some of my research is covered in the section on “Bridging the divide”.

consider bilingual and multilingual users when designing platforms

This seems super-simple and obvious as a bullet point on my slides. However, a long list of well-known platforms easily illustrates this insight is often overlooked. Since the confines of my presentation won’t allow me to look in-depth at this, I thought I would write here about one common product which I feel could be designed better for multilingual users: the Google Play Store. It is by far not the only platform, but it is a well-known platform that I use often.

Consider the app shown on the left, which I recently installed on my tablet. This is a screen shot of the app as it appears in the Google Play Store on my tablet with my user-interface language set to English.

Play tells me it has 10,000+ downloads and 3.6 stars based on 27 reviews (highlighted with a red box I added at the top). However, where are those 27 reviews? There is absolutely no indication of how I could read those 27 reviews. The web interface is even more confusing by inviting me to “Be the first to review this application.” How could I be the first to review the application if it already has 27 reviews?

As the astute reader might guess, the missing reviews are written in Japanese, which is perfectly fine since I read Japanese (and even if I didn’t I could try machine translation). However, in order to see these reviews I must switch the user-interface language of my entire tablet (or change the Language Accept parameter of my browser after logging out of my Google account and clearing all cookies). I also need to know somehow that the missing reviews are in Japanese. If I want to check if a Korean user has reviewed the app, I have to yet again change the entire user-interface language of my tablet. If I don’t want all my other apps and the system software itself to be in Japanese, I then have to yet again change the user-interface language back to English after reading the reviews.

The image below and to the right is screen shot of the app in the Google Play store after I’ve changed the user-interface language of my tablet to Japanese. Note that all the text that was previously in English (Apps, Open, Uninstall) is now in the new user-interface language of Japanese. The top area of the information is the same (3.6 stars, 10,000+ downloads), but now a new section with Reviews (レブュー) is given (which I’ve put a second red box around for emphasis).

Google Play is fully internationalized and localized on the one hand: menus are translated, the design accommodates right-to-left languages, etc. The difficulty is in how user-generated content is handled (specifically user reviews of apps). More niche apps are often reviewed in only one language in Google Play. However, reviews are grouped by the user interface language settings of users’ devices. This means that a user cannot see reviews of an app in another language without temporarily changing the user interface language of the entire Android operating system—a process that takes several minutes and affects all apps on the device.

Confusingly, apps are rated with a number of stars (1-4) averaged across reviews from all languages. This can lead to the rather odd case discussed above where the English interface shows an average rating of 3.6 stars from 27 user reviews, but then gives no indication of how the user could see these reviews. This can be particularly frustrating for multilingual users who could read the reviews in another language if the option were provided. It is further the case that some users write a review in a language different from their user interface language (e.g., a user with a Japanese UI language writing a review of an app in English). These users may think they are helping potential users in the other language, but in fact those users are very unlikely to see the review as it remains only accessible to users with the same user interface language selection as the author of the review.

So, while the bullet point on my CHI slides to “design with multilingual users in mind” seems obvious and simple, there are many common platforms that do not follow this advice (Google Play is certainly not alone). This is particularly surprising given “multilingualism…[is] the norm for most of the world’s societies,” with over half of Europe and over a fifth of the US multilingual. My work on language and Twitter and Wikipedia shows that a non-trivial percentage of users on both platforms engage in multiple languages.

As many human-computer interaction researchers gather this week for CHI in a country with a strong tradition of multilingualism, I hope that those envisioning new platforms or redesigning existing platforms will consider multilingual users specifically in their designs.

In collaboration with Helen Margetts and Taha Yasseri I have collected information on petitions and signature activity for various online platforms since 2009. The BBC covered some of our research about the previous UK government petition platform that was active from 2011-2015:

Online petitions need to attract large numbers of signatures on their first day if they are to stand any chance of success, researchers have said.

In a forthcoming book, a research team from Oxford University will show that 99.9% of e-petitions fail to reach the 100,000 signatures needed to trigger the prospect of a Commons debate.…Read more

It is important to simply interpret this as where Twitter users who geotag their tweets mention different UK football teams. Twitter users are not representative of all humanity and users who geotag are a small fraction (about 1%) of all Twitter users; so, caution is needed in interpreting the data too far. There was a very limited period of time to develop this, so some more advanced techniques to determine location and analyze sentiment were not employed, but I think the visualization suggests exciting possibilities for the future as users increase and data analysis techniques improve.

For those interested, the code behind the visualization is available freely on GitHub (CC BY-NC-SA license).

I’ve not blogged for a while on this site, because I’ve been doing lots of blogging on the InteractiveVis project site. InteractiveVis is a project to create easy to use tools to build HTML5 interactive visualizations. These tools will be public very soon, but in the meanwhile, please enjoy some demonstrations of the types of visualizations users of the tools can create. The last demo using mentions of US Presidential candidates on Twitter is particularly timely (it remains to be seen how accurate!).

Demo 1: Visualization of followers at @OIITwitter

Background information about this visualization

Demo 2: Visualization of UK Central Government

Demo 3: Visualization of Literacy and Gender

Background information about this visualization

Demo 4: US Elections on Twitter
Background information about this visualization

I wrote an article looking at the role of language online, specifically with an example of image search on the Free Speech Debate website. The article initially had several comparisons of queries in Google Image search, but all but one were edited out. I’m including all of the images below and encourage you to check out the post as well.

I also gave a very general, accessible talk (only 10 minutes!) on the idea at a panel discussion for St Antony’s International Review earlier in the year. With some help from Kdenlive, I have now been able to edit the video and place this online as well.

• Irene Eleta Mogollon at the University of Maryland College of Information Studies (iSchool) and the Human-Computer Interaction Lab (HCIL). Her work has looked at multilingual social tagging of museums’ image collections and her PhD research is about multilingual communication on Twitter.
• Chris Salzberg Chris is working on a tool, Cojiro, for cross-language curation that allows users to collect references/content around a particular topic and translate the relevant/interesting parts of that content. He’s focusing in the Japanese–English space, but the tool
• John Dalton, a masters student at the Computer Science Department of the University of Oxford supervised by Phil Blunsom is completing fascinating research trying to develop an approach to identify human translations of content using techniques and theory from machine translation. I think this is a huge area of future research. If a (good) human translation of content has already been made, I would like Google Translate/Chrome to be able to identify that translation and recommend it to me in place of a machine translation. The work is also important to be able to develop corpora of more informal writing to improve the performance of machine translation algorithms on general web text.
• Clare Wardle is at Storyful, which works with professional news clients to identify and verify legitimate news “from the noise of the real-time web, 24/7”. A key part of identifying and verifying content deals with language issues.
• An Xiao, last but not least, is a design strategist, researcher, and
  artist. She co-founded a Chinese-to-English Twitter translation site with nearly 10,000 followers and a dozen contributing members and also has some blog entries touching on cross-language issues such as this post noting similarities between topics trending in two different languages on Twitter.

Finally, a quick self-promotion (and thank you to those who voted!) to say that the Interactive Visualization project I blogged about earlier was successful in receiving funding and is plowing ahead. I hope to demo some very cool interactive visualizations very soon. We just need to get the final user-interface graphics into the demos.

Update: 27 March 2012 We’ve been successful in getting 150 votes to be considered for funding. Thanks to everyone for the votes and support.

Since getting involved in some visualization development and research, I’ve often thought that interactivity was the best way to create great inviting visualizations that can be grasped quickly, yet allow in-depth exploration of data by users. Some interactivity is possible with existing tools, but most of these rely on users having additional software (e.g. Flash, Java) that, while common for desktops, limits the wider dissemination of these visualizations.

As web technologies have developed and become better supported, I’ve been working on extremely early stage code with OII Information Officer Kunika Kono to create interactive visualizations that run with entirely with native web technologies (HTML5, CSS3, SVG). We’ve applied for funding to further develop our ideas and create ways for all users to easily make interactive visualisations for geospatial and network data. To get considered for funding, however, we need 150 votes from .ac.uk email addresses.

Please vote, find out more information, and see some alpha code demonstrations at:
http://elevator.jisc.ac.uk/ideas/interactive-visualisations-teaching-research-and-dissemination

This research analyzes linguistic barriers and cross-lingual interaction through link analysis of more than 100,000 blogs discussing the 2010 Haitian earthquake in English, Spanish, and Japanese. In addition, cross-lingual hyperlinks are qualitatively coded. This study finds English-language blogs are significantly less likely to link cross-lingually than Spanish or Japanese blogs. However, bloggers’ awareness of foreign language content increases over time. Personal blogs contain most cross-lingual links, and these links point to (primarily English-language) media. Finally, most cross-lingual links in the dataset signal a citation or reference relationship while a smaller number of cross-lingual links signal a translation. Although most bloggers link to other blogs in the same language, the dataset reveals a surprising level of human translation in the blogosphere.

Full paper…

In addition, a new publication examining the sharing of off-site links in Twitter and Wikipedia following the 2011 earthquake and tsunami in Tōhoku, Japan, has just been accepted to International Conference on Human Factors in Computing Systems, CHI ’12, ACM, which will be held in Austin, Texas, in May. I’ll be blogging more about this research in the future in order to expand upon data and details that did not fit within the page limit.

This paper describes two case studies examining the impact of platform design on cross-language communications. The sharing of off-site hyperlinks between language editions of Wikipedia and between users on Twitter with different languages in their user descriptions are analyzed and compared in the context of the 2011 Tohoku earthquake and tsunami in Japan. The paper finds that a greater number of links are shared across languages on Twitter, while a higher percentage of links are shared between Wikipedia articles. The higher percentage of links being shared on Wikipedia is attributed to the persistence of links and the ability for users to link articles on the same topic together across languages.

Pre-print copy of the paper…

Finally, I am very excited to announce I have written my first grant proposal and that the proposal was funded. The resulting project, Big Data: Demonstrating the Value of the UK Web Domain Dataset for Social Science Research, will perform research with a 30TB achieve of Web data of the .uk country-code top-level domain collected from 1996 to 2010. We are now looking to hire a Big Data Research Officer, who will contribute to this new project and to two other large-scale data projects (Leaders and Followers in Online Activism and The Internet, Political Science and Public Policy). If you have strong computer science skills and an interest in the social aspects of online technologies, please consider applying or sharing this announcement with others who might be interested. Applications close 16 March, and further information, contact details, and application information are available on the University of Oxford’s Job Search website.

I learned that it is possible, although not recommended, to teach until 12:30 in Oxford and then have a meeting in central London at 2:00 yesterday. Despite the travel challenges, I was happy to see a number of companies represented at my language session. I wish there had been more space for additional attendees and more time for discussion; however, I hope it was useful for those who attended.

The graphics we used in the language session at mindshare are listed below with links to PDFs and original sources. I’ve skipped analysis of them in this post: that has been done in some cases at the source, but please feel free to post comments or email questions and I’ll do my best to expand on anything.

Mentions of “beer” in various languages across Europe on Google Maps
Source: FloatingSheep

Geo-tagged photos on Flickr
Source: OII Visualization Gallery

Internet Penetration
Source: OII Visualization Gallery

Social Networking Sites over time
Source: Vincos Blog

Links between languages
Sources: Kovas Boguta (Twitter) and Scott Hale (Blogs)

More links between languages
Source: Hale, S. A. (Forthcoming) Net Increase? Cross-lingual Linking in the Blogosphere. Journal of Computer-Mediated Communication.

User-generated Content on Wikipedia and Google
Sources: OII Visualization Gallery (Wikipedia), OII Visualization Gallery (Google)

Top languages on the Internet
Source: Language Connect

Country code top-level domains and international top level domains
Source: Language Connect

Top news stories on Google News
Source: newsmap.jp Click for a live, interactive version

My absence from blogging for a few months has been personal (I got married in July) but also work related: I have a number of great project outputs that have just been released. These include a draft paper on social influence and collective action, a presentation at the Oxford Martin School, and a publication of Internet related maps resulting also in an online visualization gallery.

I’ve put my new mapping skills to work on the latest Wikipedia dumps from 30 September 2011 to uncover some patterns in geotagged articles. My methods are not perfect and not all language editions of the encyclopedia have the same level of geo-tagging; nevertheless, I think the patterns revealed are quite telling:

The map above shows which language edition out of German, Portuguese, and Spanish has the most geotagged articles in each country. There are a few ties, but for the most part a clear pattern emerges: countries in the Spanish-speaking world have more Spanish articles, German-speaking regions more German articles, etc. I will parse more dumps and add these (in particular I’d like to add Arabic, French, and English), but I think this pattern will hold across these and other languages.

I’ve received some challenge on my language-related research about what specific benefits multilingual contributors might bring, and I think one answer lies in breadth of content. Better coverage for a particular language edition of Wikipedia might not lie in energizing those in the home regions of a language, but rather in mobilizing the diaspora and language learners. As Brian Hecht’s 2010 article shows, there is little overlap in content and articles between different additions of Wikipedia and thus the possibility of greater coverage exists for all language editions.