Language, Perspectives, and Generalization: the Language of (Large) Language Models

This ACDH lecture by Pia Sommerauer discuss that (Large) Language Models achieve impressive results on various NLP tasks. However, LLMs are based on purely distributional data; they still rely on learning (possibly complex and sophisticated) associations between words and their contexts. It is difficult to tell to what degree language models can reflect a human-like understanding of semantics and how specific information is reflected.

The lecture investigates in how far models can perform interpretation and abstraction to such a degree that they can solve analogies? Given that models are trained on large amounts of text, what social perspectives do they reflect? Do they engage in stereotyping? And finally, how does the scientific community talk about and think about language models and their abilities? Pia Sommerauer presents findings from several studies investigating the behavior of (large) language models and discusses them with respect to an analysis of the language that people use to talk about AI.

Learning Outcomes

After watching this video, learners will be able to:

• describe the use of systematic language patterns for (large) language models fluency and challenges
• critically assess how well (large) language models detect analogies in stories
• discuss in which way stereotypes and biases influence how language models describe people
• understand how the scientific community uses anthropomorphic language to describe LLMs and AI

ACDH Lecture