Speaker
Description
Poznań Supercomputing and Networking Cernter has been involved in supporting cultural heritage and digital humanities sector for more than 20 years now. One of the recent developments, in collaboration with the Poznań City Hall and Adam Mickiewicz University (AMU), is an online platform for supporting activities of various institutions in the plain language context. The main goal of the platform is to simplify official communication (e.g. letters) sent out by local governments or companies to their audience (citizens, clients). The platform, called Proste pismo (Simple writing) was created as a database of official texts to be simplified in the language layer using semi-automated mechanism based on dictionary of simplified phrases. With time new features has been added, such as templates of documents, a dictionary of complex phrases, and an AI (artificial intelligence) module to automate the whole process. The AI module provides simplification for a given text fragment using relevant LLM model. Initially the model used in the AI module was based on BART, however, in subsequent developments and based on executed experiments, the Polish LLM Bielik model has been added. In BART case the basic LLM is trained based on relevant set of training documents prepared by the language experts from AMU. It is worth mentioning that the data processed by the models is not shared outside the service, i.e. the processing part is integrated in the platform. In addition, the results of simplification are processed using the Pisarek index (readability score for Polish language) and the Fog index, both being related to each other. Using other language model in the context of Proste Pismo is possible after appropriate training and parameterization. An important note is also to remember that the task of the AI model is not to replace the human (user) actions, but only to support them. Because of possible errors and hallucinations, any result obtained using AI should be verified by a human, especially in such a sensitive task as communication.
