Overcoming Language Barriers: Multilingual Analysis of the 2023 Swiss Privacy Law's Impact
This paper presents an LLM-based pipeline for analyzing multilingual privacy policies without translation to demonstrate that Switzerland's 2023 privacy law revision significantly increased data subject rights disclosures, a trend further amplified by the widespread adoption of automated policy generators.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine the internet as a giant, bustling marketplace where every shopkeeper has a sign in their window called a "Privacy Policy." This sign is supposed to tell you exactly what the shopkeeper does with your personal secrets—like your name, your location, or your shopping habits. For a long time, these signs were written in a confusing, legal-sounding language that most people couldn't understand, and they were often different depending on which country you were visiting. In the world of computer science and law, researchers have been trying to figure out if new rules actually make these signs clearer and more honest. They've looked at big rules like the ones in Europe (GDPR) and California, but they've struggled to check what's happening in places with many different languages, like Switzerland, where people speak German, French, Italian, and English all at once. The big question is: When a country changes its privacy laws, do websites actually update their signs to be more transparent, or do they just ignore the new rules?
This paper takes a deep dive into that question by looking at Switzerland's major privacy law update in 2023. The researchers faced a tricky problem: how do you read and compare thousands of privacy policies written in four different languages without getting lost in translation? To solve this, they built a super-smart AI assistant (a Large Language Model) that can read German, French, Italian, and English all at once, acting like a universal translator that understands the meaning of the law without needing to translate the words first. They taught this AI to look for specific promises, like "we will delete your data if you ask" or "we will tell you who we share your data with."
The team tested their AI on 120 real privacy policies written by human experts to make sure it was accurate, and it got the answers right more than 90% of the time. Then, they used this AI to scan over 35,000 websites before and after the new Swiss law kicked in. They found that after the law changed, Swiss websites suddenly started adding many more details to their privacy signs, especially about your rights to see or delete your data. It looks like the new law worked! But here's the twist: the researchers discovered that a huge chunk of these new, detailed signs weren't written by lawyers. Instead, about 18% of the websites used automated "policy generators"—basically, online tools that fill in the blanks for you. The paper suggests that these tools might actually be doing a good job; websites that used them were up to 15 percentage points more likely to include all the required details than those that didn't. So, while the law pushed for change, it was these automated drafting tools that helped many small websites actually keep up, turning a complex legal requirement into a simple, filled-out form.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.