NSF Award
2438420
In today's global world, cultural competence is essential in communication. Translators adapt their translations to match the culture of the target audience. Companies localize their product ads to boost sales. In line with this, the project aims to study how well a machine grasps cultural nuances for communication. In particular, it will focus on language tools' potential to seek common ground across cultures. The new idea is to make these tools not just about words but also about the culture behind them. This paradigm shift will open the door to endow language tools (e.g., a chatbot) with cultural competence. This, in return, will benefit the users of these tools to enhance mutual understanding and break the cultural gaps.<br/><br/>The technical aims of the project are divided into two thrusts. The first part is to design a new knowledge base to bridge cultures by learning from human experiences. The project will collect raw data from crowdsourcing platforms (e.g., Wikipedia). It will then bridge cultures via human-in-the-loop design. Building on top of this knowledge base, the second part will develop new tools to diagnose large language models (LLMs). The project will focus on capturing the rationale behind LLMs' success and failures in connecting to cultures. Next, the project will design and develop novel interpretation techniques. Based on the findings from this project, the project team will provide proposed guidelines for LLM improvements. The established new resources and tools will be shared with researchers and developers.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
