Not all public-private partnerships get off the ground, but in Estonia, the tendency has been for success and, mostly, widespread adoption in recent years.
Such collaborations have yielded such platforms as Texta, an open-source toolkit for text analytics; Hans, a speech recognition-based system for producing verbatim parliamentary records; Hoia, a mobile app for preventing the spread of COVID-19; and Neurotõlge, a translation engine that can translate from Estonian into six other languages.
“The way we are doing these kinds of partnerships is the right way,” says Texta CEO Silver Traat. “We do smaller pilots to see if something is viable, and then if the data is there to support such projects, we go ahead with them,” he says.
What is best about Estonia´s approach is our country´s constant need for innovation. Even in the field of public-private partnerships where we aim to reach another dimension by opening the country to act as a global hackathon. If you have an idea for a better public service, or maybe you have already developed one, then we want to hear from you, let you pilot it here, and if all goes well – act as your global marketers. How so? Meet the Digital Testbed Framework – your chance to pilot a service or a product in Estonia by using our governmental tech stack, for free. Send your ideas by October 10, 2021 👉 https://e-estonia.com/testbed/
From licensed product to open source
In the case of Texta, the project coalesced from the company of the same name, which Traat founded in Tallinn in 2016. The company initially planned to make its text analytics toolkit available as a licensed product but found little traction, as companies tended to be unaware of opportunities around analytics. “Then we decided, let’s make it open source,” says Traat. Once Texta embarked in that direction, it ended up in discussion with public institutions. The effort grew from a pilot project to having its toolkit available in the Estonian state code repository.
These tools include Searcher, an application for creating searches for the toolkit’s other applications and browsing or summarizing the data; Tagger, an application that allows categorization of documents into classes; and Lexicon Miner, which enables users to train language models to extract terminology from datasets. Since its debut, the toolkit has gained some well-known users, including the Estonian Ministry of Education and Research, media company Ekspress Meedia, and Õhtuleht, one of the country’s most widely read newspapers.
Traat said that users like being able to use the tools to browse their data and classify documents such as emails so that they can make business decisions more efficiently. Texta also continues to innovate and assists the private sector, such as media companies, to automate the monitoring of online comments.
According to Ott Velsberg, the government chief data officer of Estonia, the partnership with Texta was driven by the state’s AI strategy, part of which was to provide open source tools to stakeholders. He characterizes the relationship as a success as organizations have adopted the tools. “The use cases differ, but having the same underlying technology has made it easier for them to use them, rather than requiring full-blown coding experience,” says Velsberg. He adds that the collaboration has no doubt helped Texta too, which has gained references from work.
Currently, 10 AI projects in the government are under discussion that relies on Texta, he adds.
Hans is another example of a public-private partnership. Finestmedia, a Tallinn-based digital solutions firm, worked last year with Tallinn University of Technology to develop and deploy speech recognition technology to produce verbatim records of parliamentary sessions. Previously, the Estonian parliament had relied on stenographers. The tool was trained using 1,500 hours of recordings and accompanying verbatim records.
Finestmedia’s job was to develop Hans and integrate this technology to other parliamentary information systems – the agendas of sittings, the MPs and their contact details, the bills under discussion, identification of those present, voting results, statistics, the sound and video files of the sittings and other relevant information. This was no easy task, as each new test function had to be tested live in Parliament.
“Creating an innovative solution for a conservative institution such as the Parliament of Estonia with a team of different organizations was definitely a challenge for us,” underscores Finestmedia COO Jan Urva. “In the end, the solution met the expectations of the Chancellery,” he says. “It’s operational and fully integrated into the everyday work of the parliament.”
The resulting system breaks recordings into 10-minute fragments, transcribes those segments, and then forwards the transcripts to editors, as Hans only provides an accuracy of about 95 percent. However, the use of Hans has allowed the Parliament to reduce the work of stenographers. The solution is currently in use only within the Estonian parliament, but Urva says that similar system could be adopted by municipalities, courts, or other public and private services in the future.
Tracing infected contacts
Hoia is another example of an app created in partnership between the state and Estonia’s deep bench of IT firms. More than a dozen companies worked together with the government on Hoia, which is used for tracing the contacts of people infected with SARS-CoV-2, the virus that causes COVID-19. The app was rolled out in August 2020 and has since been downloaded more than 285,217 times. It was developed for free and on the fly, something that could only have been achieved via a public-private partnership, according to its developers.
“Hoia is a system that has high requirements for security and privacy,” says Veiko Raime, CEO of MobiLab, one of the firms that contributed to its development. “Only by building together with government agencies and standing on top of the existing trust of Estonian e-governance were we able to gain the market penetration and usage of Hoia we needed to succeed,” he says.
New features in development
The Hoia app relies on Bluetooth low energy technology or BLE, and phones with the app installed can pick up Bluetooth signals from nearby phones. If the signal is close and frequent enough, anonymous codes are collected and stored on the phone that refers to that individual. Should a person with the app be infected, they can alert the app, and those who are considered to have been in close contact with that person will be notified. The identity of the infected person remains anonymous. The app can be downloaded from Google Play and the Apple App Store.
According to Karilin Engelbrecht, a spokesperson for the Health and Welfare Information Systems Center (TEHIK), a data and communication center overseen by the Estonian Ministry of Social Affairs, the team behind Hoia continues to innovate. The app now works across borders and exchanges information with nearly 20 European countries. Engelbrecht added that a new feature would be added to the app shortly.
“Hoia does not have the opportunity for minors and people under guardianship to mark themselves as positive and thus warn their close contacts,” she says. “We are developing an opportunity where a parent or guardian can do it for them.”‘ She declined to provide a timeline.
Neurotõlge, which means Neurotranslation in Estonian, is a somewhat different kind of example of state partnership. In this case, the core collaboration was between state actors and the University of Tartu’s Natural Language Processing group, headed up by investigator Mark Fišel. Fišel was engaged in machine learning translation, and in 2017, he received state backing through a national effort to support Estonian-language technology. Two tranches of funding were set aside for Neurotõlge, translating between Estonian and Finnish, Latvian and Lithuanian, and English, Russian, and German.
“English and German are the most common input languages for translation,” noted Fišel, adding that Neurotõlge aims to add French, Spanish, Swedish, and perhaps others to the tool in the future. The tool can also fix grammatical errors within languages so that an Estonian text can be checked for accuracy using Neurotõlge. Fišel’s team has also worked with the private sector, such as Luisa’s translation agency, to develop specialized search engines.
“It’s nice to have collaborators,” says Fišel of such partnerships. “Doing these things theoretically is boring, and seeing them put into practice fuels research and makes things more interesting.”
💡 Want to help us make e-Estonia better, faster, stronger? Apply to the Digital Testbed Framework by October 10, 2021!