If you’ve ever received an official-looking piece of mail from the government and opened it to realise (perhaps with a sense of panic) that you have no idea what it’s about, you are not alone.
As early as the 1950s, a “plain language” movement to simplify bureaucratic text started in the United States. It gradually worked its way across the globe, spreading into law, medicine, and academia. According to a Harvard University study, plain language benefits all users, including people with cognitive disabilities, low reading literacy, and people who are encountering an unknown topic or language.
At government level, specialised translators generally have the job of turning convoluted official documents into easy-to-read text. But it’s an expensive and time-consuming task, and sometimes —as during the COVID-19 outbreak— society cannot wait to receive important information.
In 2022, three graduates from the Technical University of Munich started a company, SUMM AI, that uses large language models to simplify any kind of text quickly and inexpensively. One of the co-founders, Flora Geske, says, “We are on a mission to make the world understandable.”
SUMM AI was a finalist in the European Investment Bank Institute’s 2023 Social Innovation Tournament, which recognises startups that are making a difference socially, ethically, or environmentally.
Material out of reach for millions
At first, the SUMM AI project’s modest goal was to help Flora’s aunt, who has a cognitive disability. While studying information systems and AI as part of an interdisciplinary master's degree in finance and information management, Flora met Vanessa Theel, a mathematician, and Nicholas Wolf, a computer scientist.
“This was pre-Chat GPT. Really at the beginning of seeing computers being able to understand language and generate language,” Flora recalls. “So we said, let’s build a tool that lets my aunt break down complicated stuff into easy language.”
In Germany, nearly 18% of adults have extremely limited literacy skills, mostly due to socio-demographic factors, or because they are foreign-language immigrants. According to the OECD, Germany’s numbers are comparable to most of the countries surveyed, where “relatively large proportions of the adult population” have trouble understanding all but the most basic written material.
After building the prototype, the trio realised there was a market for it, since many public institutions are required to offer accessible information by law. Flora says, “We decided we wanted to use this tool for a social cause, for something that is relevant for society, rather than just a commercial use case.”
Plain text at the push of a button
Flora compares the SUMM AI software to Google Translate: you enter any complicated text, push a button, and it automatically generates a plain text translation. But rather than translating text word for word, the AI distils the meaning of the original, rewriting it with shorter sentences, simpler syntax, and explanations of difficult terms.
Is the tool just another example of AI replacing jobs? “Of course, AI is shifting the field, and institutions no longer need to fully rely on manual translators,” says Flora. “However, the number of translators working on easy language is very small compared to the huge volume of texts. Therefore, we are not taking anything away from them. A large number of texts would otherwise not be translated at all, due to effort and costs. An AI-translated text is still a lot better than no understandable text at all.”
She is adamant that AI tools should be used with a knowledge of their limits. “We teach our users [mostly employees of public institutions] what to look out for when working with an AI-generated text. We ask them to review the text, to check that everything from the original source is reflected correctly, that the facts are all true, and then we publish it. It’s important that it’s not fully AI-based, to prevent misinformation, and that we have humans in the loop.”
No more fear of the small print
SUMM AI’s customers are governments and companies located in Germany. One is the city of Hamburg, where Senate Spokesman Marcel Schweitzer says, “This efficient tool enables us to inform our target groups quickly and clearly, especially in situations of crisis. This would not be possible with conventional methods.” The startup has also begun translating French and English into plain language, conducting a pilot project with the French government and working with an institution in Canada.
A recent EU directive, the EU Accessibility Act, requires all companies, from financial institutions to e-commerce, to communicate in a more user-friendly way by 2025. The law creates a whole new market for SUMM AI, translating documents such as corporate fine print and bank contracts into plain language.
The new directive has attracted funding, too. Flora says, “Social impact is not always the first topic for investors. If there are big companies that need to comply with this, who can become our customers, it is a really powerful thing that allows us to scale our impact.”
So far, SUMM AI has translated around 70 000 texts, and is working constantly to fine tune its technology and improve the quality of its translations.
The company has grown to a staff of 18 people, more than half of them women, which is notable in the technology field. Many are non-native German speakers. “It’s really interesting to see them say: I used the tool myself today because I got a letter from the government about my immigration status,” Flora says.
But, she admits, they are not the only ones: “Even me, when I get legal documents, sometimes I use the tool.”
