What Makes Text Simple?

One of the main goals of the IDEAL project is to make the democratic collaboration more accessible to various communities, including those that may have difficulties using the dominant language of the country they reside in. That is why we rely on language technology of text simplification. It aims to build automatic tools that transform a given text piece -- e.g. a sentence, a paragraph, or a whole document -- into a variant that maintains the meaning of the original, but is simpler and more accessible. But here lies the challenge: different people and different groups need different degrees of simplification. For some readers it suffices to replace the most complex words with simpler variants, while others may require a complete rewrite.

That is why the Universitat Pompeu Fabra (UPF), one of the partners of IDEAL, has been working on building text simplification technology that is adaptable, and it produces different simplification depending on the level of language proficiency the target reader possesses -- e.g. A2 or B1. Our solution, called STARLING (Simplifying Text Across Languages Using Generative Models) achieves this by performing multiple rephrasing of any input sentence and chooses the one that best suits the needs of the reader. The research was just presented at the Fourth Workshop on Text Simplification, Accessibility and Readability, co-located with the EMNLP conference in Suzhou, China, and the article is available in open access.

But this is not the end of the story! The experiments so far were performed in English, the language allowing us to benefit from vast language technology resources. Currently, the UPF team is working on expanding the solution into several more languages, including some with very limited resources, like Dari or Farsi. All of that because we believe the language you speak, or the fluency you speak it with, should not affect your ability to get involved in your community.