Natural Language Understanding Lab

Prof. Dr. Michael Roth

As humans, we use language to communicate effectively and we understand each other intuitively, most of the time even if statements are ambiguous or lack essential information. But how can we also enable computational models to infer implicit information from context and background knowledge, to resolve ambiguities or, in case of doubt, to simply ask targeted clarification questions? These are the kinds of questions we address in the Natural Language Understanding Lab.

Our overarching goal is to research reliable and intelligible models that are capable of automatically understanding language. At the center of this research lies the interplay of linguistic meaning and background knowledge as well as inherent challenges of language, such as implicitness and ambiguity. Our research also has a particular focus on diversity and inclusion: differences in knowledge and diverse communication needs must be explicitly taken into account so that future models can represent diverse parts of society and be equally usable for different user groups.

As examples of socially relevant tasks, our research investigates the linguistic causes of potential misunderstandings, the adaptation of texts to specific target groups and the influence of sociocultural factors and individual preferences on language use and comprehension. In order to do justice to the diversity of natural language in such tasks, we combine innovative approaches to the development of customized data sets with modern machine learning methods and theoretically grounded models. Through joint research with other disciplines, we aim to establish necessary foundations to break down communication barriers and to improve linguistic interaction in an increasingly digital world.