Jader Martins of the D4H explains his new article the Proceedings of the 1st Worskhop on Towards Ethical and Inclusive Conversational AI and invites us to recognize the continued limits of AI chatbots in historical translation.
Author: Jader Martins
Jader Martins is a doctoral researcher at the Luxembourg Institute of Science and Technology LIST.
Historical events profoundly shape the evolution of language, imbuing many words with nuances that depend on the specific time and culture. Grasping that socio-cultural context is crucial for accurately translating historical documents. Despite advancements in language models, chatbots often struggle to fully comprehend the socio-cultural contexts inherent in human interactions. Effective and respectful communication necessitates translating not only the words but also their cultural implications. Misunderstandings due to cultural differences can result in miscommunication or even offend. Achieving a balance between linguistic accuracy and cultural sensitivity remains a complex challenge.
While conversational AI systems, such as chatbots, are not primarily designed for translation, they are sometimes utilized for this purpose. However, chatbots frequently face difficulties in recognizing socio-cultural contexts during user interactions. The figure below illustrates how a state-of-the-art chatbot (ChatGPT 3.5) struggles with translating region-specific terms.
Our experiments above reveal how ChatGPT 3.5 struggles with regional word variations. While it accurately translated “potatoes and tomatoes” into Standard German, it faced challenges with Swiss German in prompt 7 and failed to find “Erdäpfel und Paradeiser” the correct translation for Austrian German in prompt 8.
Researchers are investigating methods to enhance chatbots’ comprehension of socio-cultural contexts, including diverse cultural factors and communication styles, to achieve more accurate and culturally aware translations. Our study proposes that Knowledge Graphs can improve chatbots’ sensitivity to users’ socio-cultural dimensions, enhancing text understanding and translation. This approach would enable chatbots to access advanced lexical resources similar to professional translators using specialized dictionaries, improving translation quality. For more details on our proposal click here.