|
|
|
|
 |
Address: Abteilung für Computerlinguistik Institut für Linguistik Heinrich-Heine-Universität Düsseldorf Universitätsstr. 1 40225 Düsseldorf Germany |
Office: Phone: Fax: Email: Office hours: |
23.21.04.43 +49 (0)211 81-11705 +49 (0)211 81-03170 zinova@uni-duesseldorf.de by appointment |
Research Interests
- Deep Learning
- Explainable AI
- Machine Learning in Low-Resource Setting
- Distributional Semantics
- Frame Semantics, Semantic Composition with Frames, Decomposition of Event Structure
- Morphology-Semantics Interface
- Pragmatics, Optimality Theory, Game Theory, Rational Speech Act Theory
- Syntax-Semantics Interface
- Tree Adjoining Grammars
- Aspect, Semantics of Aspect
- Verbal Morphology, Semantics of Verbal Affixes
- Slavic Linguistics
Research
I have defended my PhD thesis on aspect and multiple prefixation in Russian (supervised by Laura Kallmeyer and Hana Filip).
In my dissertation, I have addressed both theoretical and computational problems of verbal prefixation and aspect.
The main goal was to provide semantics for verbal affixes and construct morphology-semantics interface
in such way that both the existence and the aspect of a given combination of affixes with the stem
(in the non-lexicalized domain) could be predicted. I have developed (and partially implemented) a system that predicts
the existence, semantics, and properties of complex verbs using basic morphological, syntactic, and semantic principles.
In COVID time I've taken and successfully completed a series of machine learning classes, followed by the relevant summer schools.
My current biggest passion is the combination of machine learning and symbolic approaches.
I believe that both linguistic theory and machine learning can greatly profit by information exchange between them.
I am especially interested in looking at morphology, semantics and pragmatics through the lens of a language model
in order to understand better what kind of information is represented by the model and what can we learn about the data through discovering certain patterns in it.