Histoire contemporaine du Luxembourg

History's New Chatty Companion: Putting the Chat Eluxemburgensia to the Test

History's New Chatty Companion: Putting the Chat Eluxemburgensia to the Test
Members of the research area Contemporary History of Luxembourg and participants of the Strand Colloquium for Digital and Transnational History had a workshop to assess the capabilities of the Chat Eluxemburgensia, an AI-powered chatbot designed to provide historical information in response to natural language queries. The workshop identified the chatbot's strengths in answering specific, structured questions about historical figures and events.

Members of the research area Contemporary History of Luxembourg, alongside participants of the Strand Colloquium for Digital and Transnational History, were introduced to the Chat Eluxemburgensia by Yves Maurer, the head of IT at the National Library of Luxembourg (BNL). This workshop was designed to try out and critically assess the capabilities of the Chat Eluxemburgensia, an AI-powered chatbot crafted to deliver historical information in response to user queries, formulated in natural language. Users can simply formulate a question in their preferred language and the chat will provide a summary of the relevant content it has identified and about a dozen suggested articles.

The session was structured around engaging with the chatbot through a series of predefined and user-generated prompts, allowing for a thorough evaluation of its ability to furnish relevant historical data.

Throughout the workshop, participants unearthed several lessons and proffered advice for enhancing user interaction with the Chat Eluxemburgensia. It was observed that the chatbot did not fare well with queries containing humoristic or broad language, often yielding overly general responses that spanned multiple time periods. Conversely, the chatbot offered relevant content when presented with specific, structured questions, especially those concerning historical figures and events, demonstrating its potential for accurate and relevant information retrieval.

Challenges were identified in the chatbot's handling of homonyms and its limited semantic search capabilities, which sometimes hindered performance for certain types of queries. Additionally, the influence of capitalization and special characters on the chatbot's response accuracy was noted, underscoring the need for further refinement of its training data to minimize errors and inaccuracies.

For instance, the query for “Gustave Simon” offered answers on two persons, the nazi Gauleiter of Luxembourg during the Occupation and the singer and music teacher at the Luxembourg conservatoire. Another example illustrated the limits linked to the historical corpus on which the chat relies: looking for information on the Bommeleeër affair will yield only limited answers as the digitised newspaper collection is predominantly from the prewar period, post-war titles include the Luxemburger Wort until 1980, the d’Lëtzebuerger Land until 2007 (and again from 2017-2022) and the Revue (until 2020).

Overall, the chat eluxemburgensia’s strenghts are in the transparency of the communication: the summary of the relevant materials may state the absence of result or take a cautious tone in presenting the information.