Hello, I am Alexander, an assistant professor for Natural Language Processing (NLP). My main research interest is computational lexical semantics, including word sense embeddings, word sense induction, extraction of lexical resources, and other related topics. I am also interested in argument mining. More generally, I am interested in neural and statistical natural language processing, information retrieval, knowledge bases, machine learning and intersections/interactions of these fields. You can find the list of my publications below on this page and also at Google Scholar.
I am with the Skoltech since 2019. My background is almost a decade of exciting research and developments in the field of NLP: I worked on a range of problems and tasks, such as semantic relatedness, word sense disambiguation, and induction, sentiment analysis, gender detection, taxonomy induction, etc . Before Skoltech, I was a Postdoctoral researcher in the group of Chris Biemann at the University of Hamburg, Germany. Prior to the appointment in Hamburg, I had a position of Postdoc at TU Darmstadt. I received my PhD in Computational Linguistics from the Université catholique de Louvain, Belgium. During these years, I (co-)authored more than 40 peer-reviewed research publications, including papers in top-tier conference proceedings, such as ACL, EMNLP, EACL, and ECIR receiving (with co-authors) the best paper awards at the “Representation Learning for NLP” (RepL4NLP) workshop at ACL 2016 and SemEval’2019 competition on “Unsupervised Frame Induction”. I co-organised two shared tasks on semantic relatedness and word sense induction evaluation for the Russian language (RUSSE’15 and RUSSE’18). I served also as a co-editor of a data science conference on Analysis of Social Networks, Images, and Texts (AIST) with the proceedings published in Springer LNCS series.
Information for prospective students willing to do a research project on a topic related to NLP.
Below is a list of some selected research highlights in grouped by fields of interest. The full list of references can be found on the publication tab.
|Idea:||To induce sense inventories for 158 languages and design a word sense disambiguation (WSD) algorithm for them using only word embeddings with no training data. These way we enable WSD for low-resourced languages.|
|Paper:||Word Sense Disambiguation for 158 Languages using Word Embeddings Only (LREC-2020)|
|Ideas:||(1) To perform lexical substitution in context, integration of information about the target work is important; (2) The majority of the lexical substitutes are co-hyponyms and synonyms, but the distribution varies across parts of speeches and models.|
|Paper:||Always Keep your Target in Mind: Studying Semantics and Improving Performance of Neural Lexical Substitution (COLING-2020)|
|Idea:||Induce from text interpretable representations of word senses fitted with images, hypernyms, definitions, and so on in a completely unsupervised fashion and design a system for word sense disambiguation on the basis of this sense inventory.|
|Reference:||Unsupervised, Knowledge-Free, and Interpretable Word Sense Disambiguation (EMNLP-2017)|
|Idea:||Similarly to a social ego-network, where an individual has several circles of close groups that do not overlap, related words cluster in a similar way forming word senses. An ego-network based graph clustering can be used to automatically identify word senses based on any pre-trained word embedding model.|
|Paper:||Making Sense of Word Embeddings (ACL-2016)|
|Idea:||It is possible to mine the Web for comparative statements to help to answer comparative questions (like “Is python better than Matlab for deep learning?”) and design a system that would fulfill the information needs of the users more efficiently than the usual web search.|
|Paper:||Answering Comparative Questions: Better than Ten-Blue-Links? (CHIIR-2019)|
|Idea:||Argument mining can be cast as a tagging task, we make available the recent neural models readily available for the integration into various NLP pipelines as well as for interactive analysis of user texts.|
|Paper:||TARGER: Neural Argument Mining at Your Fingertips (ACL-2019)|
A more complete list of research interests is listed below:
Chekalina, V., Bondarenko, A., Biemann, C., Beloucif, M., Logacheva, V., Panchenko, A. (2021): Which is better for Deep Learning: Python or MATLAB? Answering Comparative Questions in Natural Language. The 2021 Conference of the European Chapter of the Association for Computational Linguistics – System Demonstrations. Kyiv, Ukraine (Online)
I supervised research-oriented Master theses, usually also aiming to publish a conference paper on the basis of the produced materials.
I help to write research proposals to funding organizations which let researchers visit our faculty and do interesting short-term research project together.