There has been, in recent years, an enormous growth  in  research  on  and  use  of  artificial  intelligence. Currently, most  attention  goes  to machine learning and big data,  but society increasingly also demands transparencyaccountability and reliability,  and  the  co-existence  of  very  many  AI  systems  necessitates  thinking  about  cooperation, competition  and  communication  between  those  systems. In the light of those developments, ILLC has defined a number of key research themes, which cross its existing research programmes.


(1) Explainable and Ethical AI

Whereas machine learning methods excel in mimicking human behavior, logic and other model-based methods are much more amenable to normative use and to user intervention.  The combination of logic,  probabilistic methods for AI and machine  learning,  are expected to offer new insights that can help us build explanations for acts and decisions of artificial agents while taking into account their legal and ethical consequences.  With its long and rich tradition in pure and applied logic and research on the relation between symbolic and subsymbolic frameworks,  the ILLC is in a unique position to contribute to research that combines the toolsused in the main paradigms (both in science and the humanities) that have shaped the field of AI until today


(2) Interpretable Machine Learning for NLP

ILLC’s computational linguists have successfully participated in the paradigm shift towards deep learning methods that the field has gone through in recent years, while building up a strong, distinctive profile in interpretablemachine learning methods in parsing, machine translation, semantics and dialogue modelling.  Via collaborations with logicians, semanticists, and linguists, they are ideally placed to continue investigating the formal relations between (classic and modern) modelling frameworks and to contribute significantly to make NLP more transparent, accountable and reliable.


(3) Cognitive  Modelling

Work that combines data-driven, learning methods and high-level symbolic descriptions, is also highly relevant for cognitive science and neuroscience.  Strengthened by our expertise in designing computational tools and working out quantitative models to analyse different cognitive processes (from music cognition to the psychology of reasoning) as well as by our participation in the RPA ‘Amsterdam Brain and Cognition’, our participation in the Gravitation programme ‘Language in Interaction’ and themes (1) and (2) above, we will continue with increasedintensity our research on interpretablemodels of higher-order cognition (reasoning, language, music).


(4) Logic,  Games  and  Social  agency

In  the  study  of  information,  both  the  concept,  development and exchange of information are major topics.  In this view, game theory and ‘social’ aspects of information naturally come into play and have been one of the unifying themes across the differentILLC programmes for many years.  This theme now re-emerges with increased urgency in the light of  the  cooperative  and  competitive  interactions  between  actors  –  whether  computer  systems  orhuman users – in modern society.  Our work on epistemics, rational behaviour in strategic games, the mechanisms for collective decision making in social choice theory, logics for social networks andgame theoretic analyses of the evolution of stable communicative conventions, is directly relevantto current trends in AI, and to our collaborations with researchers in the social sciences, law and economics. 




Natural Language Processing


The research of ILLC's Language & Computation group is situated in the interdisciplinary territory between humanities, cognitive science and artificial intelligence. In most of our work in computational linguistics, this translates into a focus on models that incorporate many more insights from cognitive science and linguistics than is common in the Natural Language Processing field. In the past, this has led to pioneering contributions to statistical parsing, syntax based machine translation and semantic role labelling. In our recent work, this focus is reflected in our work on graph convolution, recursive neural networks, interpretability and accomodation in dialogue.

Raquel Fernández
Khalil Sima`an
Ekaterina Shutova
Jelle Zuidema
  Wilker Aziz


Digital & Computational Humanities


We explore how statistical and neural models can retrieve information from text that help answer questions in humanities disciplines ranging from history to philosophy.



Rens Bod
Jaap Kamps
  Ashley Burgoyne
Arianna Betti
Colavizza_G Giovanni Colavizza


 Language & Cognition



Jakub Szymanik
Tom Lentz
Floris Roelofsen


Multi-agent systems


Modern AI systems will rarely operate in isolation, but rather should be expected to interact with other AI systems as well as humans. The field of multiagent systems is dedicated to the design and analysis of mechanisms that enable such interaction. As any AI system interacting with others is ultimately an economic agent, multiagent systems research can greatly benefit from insights gained in fields such as game theory, mechanism design, and social choice theory. This perspective leads to a number of important challenges. (1) How can we incentivize AI systems interacting with each other and human users to behave in line with relevant societal norms? (2) How can we facilitate mutually beneficial cooperation between AI systems with diverse objectives? (3) How can we ensure that resources (including, in particular, computational resources) are shared fairly among competing AI systems? (4) What are the best methods for aggregating the information generated by different AI systems? (5) How can we control and explain emergent phenomena when different AI systems interact with each other?


Ulle Endriss
Alexandru Baltag
Sonja Smets


Causal modelling


Robert van Rooij
  Katrin Schulz