Today we’ll tell you about interesting events planned for the upcoming conference: reports, discussion topics and three technological competitions within Dialogue Evaluation – to extract semantic relationships from business texts, automatic search for hyperonyms and morphosyntactic analysis of texts. Go!
This year at the Dialogue Center – competition summary between developers of linguistic text analysis systems. This year competitions in three tasks took place. One way or another, they were all related to the topic of research and development of new language models that take into account the context of the sentence.
- Competition RuREBus (Russian Relation Extraction for Business) on the automatic extraction of semantic relations in a statement close to business tasks. The goal of the track is to compare methods for extracting relationships from business texts in Russian. The participants used the building of the Ministry of Economic Development (about 280 million tokens) as a dataset for the competition. These are various reports of regional bodies, as well as forecasts and development plans.
- Competition Grammval on automatic morphosyntactic analysis of texts. Participants in the competition developed systems that determine the morphological features of a word, a lemma (a dictionary form of a word), and syntactic relationships in a sentence. The winner of the competition, MIPT graduate Daniil Anastasiev, beat the state-of-the-art results (best at the moment) for the Russian language, obtained earlier at major international conferences.
- Competitions for the automatic search for hyperonyms in Russian (a hyperonym is a concept that expresses a more general name of objects in relation to another concept. For example, “cat” is a hyperonym in relation to the term “maine coon”). Competition Terms were close to reality: the participants at the entrance were given only the words themselves without definitions, as well as the contexts in which these words were found.
On June 18 and 19, the winners of each competition will tell you how they solved the problems, what experiments they set and what methods of machine learning they used.
At the opening of the conference, June 17, 11: 00-11: 30, the director of the Physics and Technology School of Applied Mathematics and Informatics (FPMI) MIPT will speak Andrey Raygorodsky. FPMI is interesting in dialogue with the humanities and activities at the intersection of artificial intelligence and language. ABBYY has been cooperating with this PhysTech School for a long time: it is open two basic departments and laboratory ABBYY Lab.
Traditionally, the Dialogue is attended by well-known international computer linguists:
Manfred Stade, professor of applied computer linguistics, University of Potsdam. His scientific area of interest is argumentation mining, based on the theory of rhetorical relations in the text and markup of the logic of text development. At the Dialogue, he will tell how, based on this logic, set up machine learning experiments and extract arguments. June 17, Wednesday, 15: 30-16: 45.
Igor Boguslavsky, professor at the Technological University of Madrid and head of the Laboratory of Computer Linguistics, IPPI RAS, one of the largest specialists in ontological semantics. His report will be devoted to how to train a computer to understand texts. June 17, Wednesday, 12: 00-14: 00.
Olga Inkova, professor at the University of Geneva, will speak in the sections “Analysis of Sentiment and Intentions” June 18, Thursday, 16: 30-18: 30; “Linguistic models of language” June 19, Friday, 15: 30-18: 30.
Sergey Sharov, Professor, Department of Translation Studies, University of Leeds (UK). One of the most respected experts in the world in the field of analysis of Internet texts. He will make a report on the reliability of case studies and how one case differs from another. June 19, Friday, 11 a.m. – 2 p.m.
June 18, at 19:00, we invite to the round table “UPGREAT technological NLP-barrier “PRO // READ”: how to correctly pose (and solve) the problem of understanding the text“. The Up Great “PRO // READING” contest is dedicated to creating an AI system that can automatically detect and annotate semantic errors in real texts. The one who overcomes the technological barrier will win: he will create a solution that has no analogues in the world and can cope with a meaningful analysis of the text at a level no worse than specialists. The participants in the discussion at the Dialogue will discuss the task of understanding the text and ways to solve it.
On June 19, at 19:30, a presentation and discussion of RuSuperGlue, a new resource for conducting open-source testing for the Russian language, will take place. This is a Russian analogue of the English project Superglue (General Language Understanding Evaluation). It will contain test tasks for logic and goal setting with open source and the ability to add your own transformer model for the Russian language to the leaderboard and compare with others.
On June 20, from 15:00 to 18:30, a student session will also be held in the main areas of the Dialogue.
The working languages of the conference are Russian and English. Detailed conference program laid out hereIt will be supplemented. Conference proceedings will be published in the yearbook “Computational Linguistics and Intelligent Technologies“, Which is part of the international citation system Scopus.
You can register for “Dialog” here until June 20. Two days before each performance, the Dialogue Organizing Committee will open to participants access to speaker presentations.