Psycholinguistics is concerned with how language is represented and processed in the mind/brain. Our work is concerned with developing NLP models for predicting the psycholinguistic features of words, such as how easily a word can be visualized by a person (e.g. it is easy to picture what a dog looks like, but it is much more difficult to picture an action like running). We also examine parallels between the cognitive development children undergo and the “development” of a neural language model during training.

Recent Publications:

• Can Transformer Language Models Predict Psychometric Properties?

• Modeling Age of Acquisition Norms Using Transformer Networks

If you have ever seen the comment sections on a political Facebook post, or a twitter debate over something controversial, you might notice one invariant fact: people are bad at reasoning. We use fallacies, we are affected by cognitive biases, we mischaracterize positions, and we often have misplaced overconfidence in poorly-thought-out views. Why is this the case? And most importantly, can we do better? Our work attempts to create AI-based tools that can understand argumentative dialogues, assess the quality of arguments people make, and offer suggestions on how to improve those weaknesses.

• Warrant Game - People tend to be somewhat poor at evaluating whether an analogy is any good. For example: "Both coffee and alcohol change your mood when you drink them; both coffee and alcohol are liquids; both coffeeholics and alcoholics can't function until they've had their drink; alcoholics have a serious problem; THEREFORE, coffeeholics have a serious problem." Is this a good argument? Why or why not? How can we program an algorithm to reason in the way you just did?

We have developed a game / interaction framework called WG-A which allows people to evaluate an analogical argument without having to spend a lot of time understanding the deep theory behind what makes an analogy good or bad. WG-A makes it easier to determine key features typically associated with argument strength and may reveal hidden assumptions or fundamental reasoning incompatibilities. By presenting arguers with an issue and positions which relate to conspiratorial or biased thinking, WG-A may operate as an educational tool for breaking biased belief into core values and building cognitive skills. We are currently performing experimentation to test WG-A's effectiveness in combating belief in conspiracy theories about COVID-19 and in reducing anti-Black racial bias.

Related Publications:
Cooper, M., Fields, L., Badilla, M., & Licato, J. (2020). WG-A: A Framework for Exploring Analogical Generalization and Argumentation. In Proceedings of the 42nd Cognitive Science Society Conference (CogSci 2020).

Licato, J. & Cooper, M. (2020). Assessing Evidence Relevance By Disallowing Direct Assessment. In Proceedings of the 12th Conference of the Ontario Society for the Study of Argumentation.

Fields, L. & Licato, J. (2022). Combatting Conspiratorial Thinking with Controlled Argumentation Dialogue Environments. In Oswald, S. and Lewinski, M. and Greco, S. and Vilata, S., eds. The Pandemic of Argumentation: 23. Springer Nature.

• Aporia is an argumentation dialogue environment designed to create structured datasets of opposing interpretive arguments. Players compete by arguing for or against certain interpretations of open-textured rules. The game is designed to be fun in order to incentivize willing participation and obtain useful datasets. Aporia is played in rounds by any group of three or more people. At the beginning of each round, two players are randomly chosen to argue against each other, and a third player is designated as a judge. The players are provided with an ethical rule for a given professional association and a scenario. For example, the rule ``teachers need to act professionally with students'' could be paired with the scenario ``some teacher exchanges some light-hearted jokes with a student during recess.'' Would the teacher's action be considered ``professional'' in the sense meant by the rule?
  
  Related Publications:
  Marji, Z. & Licato, J. (2021). Aporia: The Argumentation Game. In Proceedings of The Third Workshop on Argument Strength (ArgStrength 2021).
  Licato, J. (2022). Automated Ethical Reasoners Must be Interpretation-Capable. In Proceedings of the AAAI 2022 Spring Workshop on ``Ethical Computing: Metrics for Measuring AI's Proficiency and Competency for Ethical Reasoning".
  

We developed the next version of MATR, a highly customizable natural deduction reasoner. Although this line of research is not currently active, we were previously able to show that MATR could develop concise proofs of both of Gödel's incompleteness theorems.

Related Publications:

Malaby, E., Dragun, B., & Licato, J. (2020). Towards Concise, Machine-discovered Proofs of Gödel's Two Incompleteness Theorems. In Proceedings of The 33rd International Florida Artificial Intelligence Research Society Conference (FLAIRS-33). AAAI Press.