Recent research areas and publications
Data-driven computational linguistics
Can machines learn the meaning of words and expressions from data? Contrary to what Chomsky claimed, much is possible and publications here discuss models of learning semantic relations.
- T. Lindh-Knuutila and T. Honkela (2015): Exploratory analysis of semantic categories: comparing data-driven and human similarity judgments.
- T. Honkela and K. Förger (2013): Modeling Action Verb Semantics Using Motion Tracking
Modeling subjectivity of meaning
Contrary to what early Wittgenstein claimed, each of us has a private language. In other words, each expression in any language is understood by each person in a different manner at least to a some degree. My “beautiful yellow house” does not look the same as yours. Logicians ran into trouble with their limited representations but continuous high-dimensional vectorial representations can be used to deal with intricacies of linguistic meaning.
- T. Honkela et al. (2015): GICA: Grounded Intersubjective Concept Analysis. A Method for Improved Research, Communication and Participation
- Sintonen et al. (2014): Quantifying the Effect of Meaning Variation in Survey Analysis
Neural networks and statistical machine learning
Neural networks and machine learning methods based on biological inspiration or rigorous theoretical basis derived from probability theory, statistics and information theory are the cornerstones of modern data-driven information processing. In digital humanities, these methods are used for a wide range of tasks ranging from topic analysis to the study of historical processes. In humanities, an interesting challenge is the inherent presence of human interpretation. A critical view is necessary as no such thing as objective data exists.
- T. Vatanen et al. (2015): Self-organization and missing values in SOM and GTM.
- J. Strahl et al. (2014): A Gaussian Process Reinforcement Learning Algorithm with Adaptability and Minimal Tuning Requirements