IDEAS Research Institute · Warsaw, Poland
Dr. Hubert Plisiecki
Computational psycholinguist
I study meaning in human language
and the behaviour of language models.
Research
How language gives us access to psychological phenomena.
My work approaches this question from two directions: the meanings people give to shared concepts, and the ways artificial systems describe themselves.
Analysis of Meaning
How do people differ in what they mean?
The same word can carry different meanings for different people. I develop computational methods for studying how these differences relate to attitudes, traits, and experiences.
Supervised Semantic Differential (SSD) connects individual differences with interpretable patterns in language. It estimates how meaning shifts semantically and explains it in plain words.
More about this programme
This research project directly follows from my PhD program where I advanced the cross-section of sentiment analysis and emotion psychology by creating numerical representations of emotions to support theory, as well as created emotion detection algorithms that are less prone to human bias.
By expanding this approach to meaning I want to realize the promise that natural language processing holds for the psychological science – providing it with a numerical substrate for how people describe their first-person experience.
Related papers:
Post-Anthropocentric Psychology
How to study AI models using psychological tools?
I study psychological constructs and measures that emerge from the behaviour of language models, rather than assuming that human categories transfer unchanged.
Initial research shows that models reliably differ in the extent to which they consider themselves to possess consciousness, and these behaviors are systematically modulated by supervised finetuning.
Selected publications
A selection of papers on meaning, measurement, and models.
The Two-Process Theory of Machine Self-Report
A theory and instrument for understanding model self-description.
Interpretable Semantic Gradients in SSD: A PCA Sweep Approach and a Case Study on AI Discourse
Choosing dimensionality with interpretability and stability in view.
Reducing social biases in text-based emotion prediction using semantic blinding and semantic propagation graph neural networks
Restricting information to reduce bias in emotion prediction.
Extrapolation of affective norms using transformer-based neural networks and its application to experimental stimuli selection
Extending affective and semantic norms across six languages.
About
I’m a computational psycholinguist at the IDEAS Research Institute in Warsaw, with a PhD in psychology from the Institute of Psychology, Polish Academy of Sciences.
My research brings psychological measurement and interpretable natural language processing together. My doctoral work concerned numerical representations of emotions and NLP methods suited to psychological research.
I also co-founded the Society for Open Science, where together with colleagues we organize scientific events and analyze the robustness of polish psychological science.
Contact me:
I welcome research collaborations and consulting enquiries involving language, psychological measurement, and interpretable NLP.
hplisiecki@gmail.com