Navigation auf uzh.ch
Navigation auf uzh.ch
The research project InvestigaDiff focuses on the synchronization of text documents across languages. The project is funded by the Swiss National Science Foundation and is scheduled to run for three years (2024–2027).
The term “Generative Artificial Intelligence” refers to technology that, among other things, facilitates the editing of text documents. For example, large language models like GPT can rephrase texts in a desired style or generate drafts for new texts. Additionally, neural machine translation can produce translations in many languages.
However, it is unclear how well this technology is suited for the maintenance of existing texts—such as updating texts or aligning content in different language versions.
The research project “InvestigaDiff” aims to fill this gap and lay the technological foundation for new applications of this kind.
We distinguish between two complementary applications:
(1) Automatically identifying differences between texts. Here, we see it as important to distinguish well between content and stylistic differences and to correctly assess the severity of a detected difference.
(2) Generating suggestions for how differences between texts can be carefully resolved (“synchronization”). Special challenges are expected when different languages or text types need to be processed simultaneously.
With the increasing use of generative AI in society, the number of texts and documents that need to be maintained and updated over time is expected to increase as well.
At the same time, the research project offers a new perspective on language models and related systems: What are their limits, and are there simple algorithms or technical adjustments that can further enhance their potential?
