Language Technology: Computational, linguistic and cognitive perspectives

In our approach to computational cognitive modelling, we focus on models of language and speech processing, and the extent to which these models make use of cognitive signals and data from the visual gestural modality, and combine them with linguistic knowledge. An application of this area we investigate is the development of tools supporting detection and monitoring of mental health conditions including psychosis, depression and cognitive decline.

We develop methods for compiling NLP resources and language models that encompass cultural and societal diversity – with a particular focus on Danish and other low and medium-resourced languages. We perform culture-aware evaluation of language models and provide benchmark data for Danish that span linguistic and lexical variety. We work with annotated datasets, including in particular lexical-semantic lexicons and corpora, also across languages.

We contribute to explainable artificial intelligence by exploring the internal representations and learning mechanisms in computational models of natural languages. In particular, through a multilingual approach, we investigate how language models cope with linguistic diversity at various architectural and linguistic levels. Our research seeks to improve the transparency and performance of multilingual language models and ensure more robust and accurate multilingual language processing.

This area includes computational NLP models for the analysis and generation of textual data in its widest forms including poems, novels, letters, manuscripts, news articles, scientific articles and lyrics. We perform a continuous upgrade of our NLP pipelines and corpus tools, in particular for Danish, and work to ensure that appropriate methods and gold standards are compiled for evaluating them. This research opens many collaboration opportunities with researchers from the Humanities at large, and NorS in particular.