Simply start by asking a question using one or more full sentences. In a first step, the chatbot identifies the language of your text. Next, it searches for potential answers among the eight million articles from the digitised newspapers on eluxemburgensia.lu. During this process, it tries to find articles in German and French, irrespective of the language in which you typed your question. Finally, the chatbot connects to the ChatGPT application. The latter is a text generator based on artificial intelligence. The chatbot asks it to analyse these articles and to suggest a coherent answer to your question. You may ask follow-up questions, which use the current chat history as context.

The chatbot is not meant to replace the classic keyword search; it is rather a complementary and experimental service, which answers your query based on a predefined number of articles. The chatbot is therefore not as exhaustive as a classic search. It chooses articles according to their subject, not according to specific keywords or spellings. It generates a structured response based on a sample of articles, so you don’t have to make an initial selection and provides a starting point to further deepen your research. In contrast, the search bar allows you to find desired keywords within the whole collection of articles. This can be important, for example, if you are looking for all references to certain entities, such as people or places. It also allows you to search for a specific spelling, an exact combination of words or dates.

Any question, which cannot be answered by reading a few articles as a human, can also not be answered by the chatbot. For example, it would not be possible to list all opening dates of the annual Schueberfouer funfair, unless there is a single article that contains the exact answer. You should also keep in mind that a statistical language model is generating the replies. This means that, even if the underlying articles correctly address the question, there is no guarantee that the chatbot correctly identifies the relevant information.

You should use full sentences and describe your question and any relevant context in detail. Your instructions to the chatbot should be as clear as possible. It is also important to use correct spelling. For example, in order to get a list of house prices, you could ask: 'Please generate a price list of houses sold on Boulevard Royal. Add the exact address and the selling price to the list.' The chatbot will then produce a list based on the information found in the articles.

With every reply, the chatbot lists all the articles that it used to gather information in order to answer your question.

The chatbot is still experimental and should only be utilised in a trial context. If you are looking for a higher degree of confidence in the results or if answering your question requires analysing a larger volume of articles, using the classic search bar is recommended. However, you can always ask follow-up questions if you are not happy with the chatbot's reply. A good strategy is to narrow down your initial question by being more specific or detailed.

Newspaper articles vary heavily in their length. Therefore, the chatbot needs to split longer articles into smaller sections in order to improve the overall search. In some cases, it can happen that the less relevant part of a split article is used to formulate the reply.