DIKU Bits: Mike Zhang and Maria Vlachou

Speaker

Maria Vlachou, Postdoc, NLP (Natual Language Prosessing) Section

Title

Large Language Models for Extracting Relevant Information about Credibility Assessment in Danish Asylum Cases

Abstract

To determine the outcome of an asylum application, legal experts engage in a process called credibility assessment, examining whether the applicant’s claim is believable. Large Language models provide an opportunity to automate this process by directly asking the LLM to judge how the decision was processed. However, credibility can be determined by a wide number of indicators. In this talk, Maria will present an overview of ways to extract and synthesize useful relevant information that provide insights about credibility and risk assessment using LLMs and semantic search.

---

Speaker

Mike Zhang, Postdoc, NLP (Natual Language Prosessing) Section

Title

Evaluating the Cultural Awareness of Language Models for Danish with Native Speakers

Abstract

If we prompt a language model on what it means to be ‘a pineapple in its own juice’, we expect the model to understand cultural quirks. Standardized tests simply don't capture true native nuance. Who better to evaluate a model's grasp of Danish culture than native Danes themselves? Mike will present a study where native Danes test and evaluate the cultural awareness of large language models using specific cultural prompts.