In 2004, I met Larry Lessig at a conference in Taipei. I was deep in the Perl community, maintaining over 100 open-source modules on the Comprehensive Perl Archive Network, or CPAN. Lessig’s Creative Commons, or CC, framework crystallized something I had only felt intuitively: that the same spirit of openness behind free software could apply to culture, education and knowledge itself.¹

It would take nearly a decade, and a dictionary website with more than a few cracks, for that insight to change everything.

This moment arrived in October 2012, when a government advertisement admitting the incomprehensibility of a national budget went viral in Taiwan.² A group of civic hackers, frustrated by how little the government trusted citizens with public information, decided to stop complaining and start building. They registered g0v.tw — a zero replacing the ‘o’ in gov.tw — and held their inaugural Hackathon at Academia Sinica Dec. 1, 2012.³

My connection to this world stretched back years. I was the original coder for Open Foundry, the free software forge hosted by Academia Sinica’s Institute of Information Science, the same institute where Tyng-Ruey Chuang — a key figure in Creative Commons Taiwan — worked.⁴ These communities overlapped inevitably.

At the next g0v Hackathon, Jan. 27, 2013, my old friend Ping Yeh — a quantum physicist turned Google engineer who had moved to California — joined remotely. He published a blog post titled Return Writing to the People 〈還文於民〉.⁵ His children were trying to learn Mandarin in the United States, and the Ministry of Education’s online dictionary was unusable. Built in 1996 on technology that predated modern web standards, it logged you out after 30 minutes — despite having no login button. It displayed rare characters as tiny bitmap images you could not copy or paste. It said “best viewed in IE 5 or Netscape 4.7+,” the plus being redundant since Netscape had long been discontinued. 20 years of procurement cycles had faithfully replicated these “cracks.”

Ping proposed a vision: download the dictionary data, clean it, structure it and rebuild it as a modern web application. Within 24 hours, dozens of hackers had done exactly that — scraping the HTML, converting it to structured data, building a website, an input method extension, offline and mobile apps.

With hundreds of people trying simultaneously to identify Unicode code points from those tiny bitmap characters, we overwhelmed Google Spreadsheets. We quickly shifted to EtherCalc, a real-time collaborative spreadsheet I built with Dan Bricklin’s help.⁶ On Feb. 1, 2013, I pushed the “First commit of prototype webkit client” — timestamped 2:02 a.m.

The domain Ping registered was 3du.tw — a play on edu.tw. We called the project 萌典 (MoeDict): 萌 for the Ministry of Education, or MOE, but also in the sense of “the first sprouting” — and yes, in the Japanese internet sense of something you find endearing. The name was a hack, just like the project.

Then came the legal problem. The MOE’s website declared “All copyrights reserved” and explicitly prohibited linking to individual dictionary entries. We were not linking — we had taken the entire dataset. In Taiwan at the time, this kind of reproduction could be criminal.

This is where Creative Commons engineered a sea change.

We released all our code and converted data under CC0. By claiming zero copyright ourselves, we made the government’s own data the only copyrighted material in play.

The strategy was precise: if we claimed no copyright over our work, we were only performing data conversion for the government. The proportion of original creative expression we claimed was exactly zero, matching the 100 percent of government data we had used. Proportional fair use.

CC0 also meant the MOE could merge our work back into their own systems without crediting us — without any legal friction. We were not building a rival; we were doing their work better and handing it back, as an extension of the public service the ministry was supposed to provide.

We had learned from what happened to the “Real Price Map” — a civic project by Google Taiwan engineers that used government real estate data and was closed down over copyright.⁷ Aaron Swartz’s prosecution was fresh on everyone’s mind.⁸ Everyone was well aware we could face criminal charges. MoeDict was hosted on servers in Tokyo, partly to complicate jurisdictional questions. We organized the work as distributed civil disobedience: dozens of people each downloading a portion, each asserting individual fair use, each relinquishing their own copyright. If one of us were prosecuted, everyone else could pick up where they left off.

While the MOE’s lawyers spent a year debating our case, we kept building. MoeDict grew to 7 million visits per month. When we added an Open Graph image generator — type any phrase and receive a banner in beautiful calligraphy — the site became a tool for civic mobilization. During the marriage equality movement, organizers used MoeDict to create shareable banners from dictionary definitions. We had redirected the Facebook algorithm with lexicography.

Then we launched a crowdsourced error-correction campaign. Our software identified dictionary entries that cited the same classical source but differed by one character — likely typos accumulated over decades. We asked readers to research which was correct. Within 18 days, we collected over 5,000 corrections.⁹ The MOE’s own parallel campaign had yielded about 6,000 total — the majority from MoeDict users.

On the day we delivered those corrections, the MOE gave us an award and declared our work fair use.¹⁰ The ministry could no longer take action against us — our contributors were their core constituency: teachers, students, linguists and language lovers. To go after us would mean going after them.

What happened next surprised even us. The project kept forking.

When I added Taigi (Taiwanese Hokkien) to MoeDict alongside Mandarin, the Google Play ratings dropped. Some users complained about “bloating their phones with a dialect they don’t use.” Within the Taigi community itself, factions disagreed over romanization — Pe̍h-ōe-jī versus Tâi-lô — and whether to use Chinese characters at all.

We held monthly MoeDict Hackathons and invited all factions. Because the code was open and the data was CC0, anyone who disagreed could fork the project. I wrote converters between romanization systems so each community could have the version they preferred. One fork, iTaigi, became popular by letting users add contemporary vocabulary — all the Pokemon names the MOE’s editorial board would take decades to approve.¹¹

We digitized the Amis-Mandarin-English trilingual dictionary — an indigenous language treasure existing only as low-quality 1970s scans — in 2.5 days through crowdsourced OCR correction.¹² Then came Hakka, followed by more Austronesian languages. Each community adapted the shared infrastructure to their own needs.

This was CC not as a static licence but as a living practice. The freedom to fork dissolved conflicts that no committee could resolve.

At the same time, I was working with Apple and Oxford University Press, helping prepare the MOE’s dictionary data for other platforms. I was, quite literally, taking all the sides. When there is no artificial scarcity, you can work with everyone at once.

After about two years, the MOE announced that all its dictionaries — past and future — would be released under Creative Commons. They chose CC BY-ND 3.0 Taiwan, restricting derivative works.¹³ But the shift was profound: from “all rights reserved, do not even link to us” to Creative Commons, because a few dozen people relinquished their own rights first.

Through g0v’s advocacy and MoeDict’s example, Taiwan’s ranking on the Global Open Data Index rose from 36th to 11th to 1st.¹⁴ A national regulation followed in 2015: any government system under US$1.5 million must now release its data under Creative Commons by default.¹⁵

In 2016, I joined the Cabinet as minister without portfolio responsible for Taiwan’s digital affairs; the news conference for my first day was held just before a MoeDict Hackathon. The same principles — radical transparency, rough consensus and running code — became part of national policy. Over the decade that followed, while democratic scores declined across much of the world, Taiwan’s rose on indexes from Freedom House, the Economist Intelligence Unit and V-Dem.¹⁶

Despite being the primary target of the largest foreign influence operations on earth,¹⁷ we remained among the least affectively polarized developed societies,¹⁸ and 91 percent of citizens continued to consider our democracy at least “fairly good.”¹⁹ The crack that let the light in had become a window onto an entire plural, caring society.

I look back on that period and see CC as a grammar for collaboration between people who may never meet. CC0 allowed us to make an irrefutable argument to a sceptical bureaucracy. CC BY-ND allowed the MOE to share while retaining the integrity it cared about. The freedom to fork allowed communities with irreconcilable differences to collaborate anyway.

Larry Wall, Perl’s creator, would say: instead of singularity, focus on plurality. That is what CC made possible for us. Not a single correct answer, but a shared infrastructure for many answers.

Today, the questions CC posed about sharing extend to the frontiers of machine learning. We call the answer Civic AI (仁工智慧): Replacing the 人 in artificial intelligence (人工智慧) with 仁, the character for civic care. Our 6-Pack of Care framework is CC0.²⁰ The forks keep forking.

There is a Leonard Cohen verse I often return to: “There is a crack in everything, and that is how the light gets in.” A 20-year-old dictionary website, cracked in every technical sense, let the light of a movement through. Creative Commons was the framework that ensured nobody monopolised that light, and that everyone could help it shine on.