BERT-based models are currently used for solving nearly all Natural Language Processing (NLP) tasks and most often achieve state-of-the-art results. Therefore, the NLP community conducts extensive research on understanding these models, but above all on designing effective and efficient training procedures. Several ablation studies investigating how to train BERT-like models have been carried out, but the vast majority of them concerned only the English language. A training procedure designed for English does not have to be universal and applicable to other especially typologically different languages. Therefore, this paper presents the first ablation study focused on Polish, which, unlike the isolating English language, is a fusional language. We design and thoroughly evaluate a pretraining procedure of transferring knowledge from multilingual to monolingual BERT-based models. In addition to multilingual model initialization, other factors that possibly influence pretraining are also explored, i.e. training objective, corpus size, BPE-Dropout, and pretraining length. Based on the proposed procedure, a Polish BERT-based language model – HerBERT – is trained. This model achieves state-of-the-art results on multiple downstream tasks.

This paper studies the generation methods for paraphrasing in the Russian language. There are several transformer-based models (Russian and multilingual) trained on a collected corpus of paraphrases. We compare different models, contrast the quality of paraphrases using different ranking methods and apply paraphrasing methods in the context of augmentation procedure for different tasks. The contributions of the work are the combined paraphrasing dataset, fine-tuned generated models for Russian paraphrasing task and additionally the open source tool for simple usage of the paraphrasers.

Abusive phenomena are commonplace in language on the web. The scope of recognizing abusive language is broad, covering many behaviors and forms of expression. This work addresses automatic detection of abusive language in Russian. The lexical, grammatical and morphological diversity of Russian language present potential difficulties for this task, which is addressed using a variety of machine learning approaches. Finally, competitive performance is reached over multiple domains for this investigation into automatic detection of abusive language in Russian.

Not all topics are equally “flammable” in terms of toxicity: a calm discussion of turtles or fishing less often fuels inappropriate toxic dialogues than a discussion of politics or sexual minorities. We define a set of sensitive topics that can yield inappropriate and toxic messages and describe the methodology of collecting and labelling a dataset for appropriateness. While toxicity in user-generated data is well-studied, we aim at defining a more fine-grained notion of inappropriateness. The core of inappropriateness is that it can harm the reputation of a speaker. This is different from toxicity in two respects: (i) inappropriateness is topic-related, and (ii) inappropriate message is not toxic but still unacceptable. We collect and release two datasets for Russian: a topic-labelled dataset and an appropriateness-labelled dataset. We also release pre-trained classification models trained on this data.

In this paper we describe a transformer model pre-trained on 8 billion tokens of crawled text from the Croatian, Bosnian, Serbian and Montenegrin web domains. We evaluate the transformer model on the tasks of part-of-speech tagging, named-entity-recognition, geo-location prediction and commonsense causal reasoning, showing improvements on all tasks over state-of-the-art models. For commonsense reasoning evaluation we introduce COPA-HR - a translation of the Choice of Plausible Alternatives (COPA) dataset into Croatian. The BERTić model is made available for free usage and further task-specific fine-tuning through HuggingFace.

The success of pre-trained transformer language models has brought a great deal of interest on how these models work, and what they learn about language. However, prior research in the field is mainly devoted to English, and little is known regarding other languages. To this end, we introduce RuSentEval, an enhanced set of 14 probing tasks for Russian, including ones that have not been explored yet. We apply a combination of complementary probing methods to explore the distribution of various linguistic properties in five multilingual transformers for two typologically contrasting languages – Russian and English. Our results provide intriguing findings that contradict the common understanding of how linguistic knowledge is represented, and demonstrate that some properties are learned in a similar manner despite the language differences.

In this study, we present an exploratory analysis of a Slovenian news corpus, in which we investigate the association between named entities and sentiment in the news. We propose a methodology that combines Named Entity Recognition and Subgroup Discovery - a descriptive rule learning technique for identifying groups of examples that share the same class label (sentiment) and pattern (features - Named Entities). The approach is used to induce the positive and negative sentiment class rules that reveal interesting patterns related to different Slovenian and international politicians, organizations, and locations.

Parallel language corpora where regular texts are aligned with their simplified versions can be used in both natural language processing and theoretical linguistic studies. They are essential for the task of automatic text simplification, but can also provide valuable insights into the characteristics that make texts more accessible and reveal strategies that human experts use to simplify texts. Today, there exist a few parallel datasets for English and Simple English, but many other languages lack such data. In this paper we describe our work on creating an aligned Russian-Simple Russian dataset composed of Russian literature texts adapted for learners of Russian as a foreign language. This will be the first parallel dataset in this domain, and one of the first Simple Russian datasets in general.

This paper describes the University of Ljubljana (UL FRI) Group’s submissions to the shared task at the Balto-Slavic Natural Language Processing (BSNLP) 2021 Workshop. We experiment with multiple BERT-based models, pre-trained in multi-lingual, Croatian-Slovene-English and Slovene-only data. We perform training iteratively and on the concatenated data of previously available NER datasets. For the normalization task we use Stanza lemmatizer, while for entity matching we implemented a baseline using the Dedupe library. The performance of evaluations suggests that multi-source settings outperform less-resourced approaches. The best NER models achieve 0.91 F-score on Slovene training data splits while the best official submission achieved F-scores of 0.84 and 0.78 for relaxed partial matching and strict settings, respectively. In multi-lingual NER setting we achieve F-scores of 0.82 and 0.74.

This document describes our participation at the 3rd Shared Task on SlavNER, part of the 8th Balto-Slavic Natural Language Processing Workshop, where we focused exclusively in the Named Entity Recognition (NER) task. We addressed this task by combining multi-lingual contextual embedding models, such as XLM-R (Conneau et al., 2020), with character- level embeddings and a biaffine classifier (Yu et al., 2020). This allowed us to train downstream models for NER using all the available training data. We are able to show that this approach results in good performance when replicating the scenario of the 2nd Shared Task.

Named entity recognition, in particular for morphological rich languages, is challenging task due to the richness of inflected forms and ambiguity. This challenge is being addressed by SlavNER Shared Task. In this paper we describe system submitted to this task. Our system uses pre-trained multilingual BERT Language Model and is fine-tuned for six Slavic languages of this task on texts distributed by organizers. In our experiments this multilingual NER model achieved 96 F1 score on in-domain data and an F1 score of 83 on out of domain data. Entity coreference module achieved F1 score of 47.6 as evaluated by bsnlp2021 organizers.

We present a collection of Named Entity Recognition (NER) systems for six Slavic languages: Bulgarian, Czech, Polish, Slovenian, Russian and Ukrainian. These NER systems have been trained using different BERT models and a Frustratingly Easy Domain Adaptation (FEDA). FEDA allow us creating NER systems using multiple datasets without having to worry about whether the tagset (e.g. Location, Event, Miscellaneous, Time) in the source and target domains match, while increasing the amount of data available for training. Moreover, we boosted the prediction on named entities by marking uppercase words and predicting masked words. Participating in the 3rd Shared Task on SlavNER, our NER systems reached a strict match micro F-score of up to 0.908. The results demonstrate good generalization, even in named entities with weak regularity, such as book titles, or entities that were never seen during the training.

In this paper we describe TraSpaS, a submission to the third shared task on named entity recognition hosted as part of the Balto-Slavic Natural Language Processing (BSNLP) Workshop. In it we evaluate various pre-trained language models on the NER task using three open-source NLP toolkits: character level language model with Stanza, language-specific BERT-style models with SpaCy and Adapter-enabled XLM-R with Trankit. Our results show that the Trankit-based models outperformed those based on the other two toolkits, even when trained on smaller amounts of data. Our code is available at https://github.com/NaiveNeuron/slavner-2021.

In this paper we describe our submissions to the 2nd and 3rd SlavNER Shared Tasks held at BSNLP 2019 and BSNLP 2021, respectively. The tasks focused on the analysis of Named Entities in multilingual Web documents in Slavic languages with rich inflection. Our solution takes advantage of large collections of both unstructured and structured documents. The former serve as data for unsupervised training of language models and embeddings of lexical units. The latter refers to Wikipedia and its structured counterpart - Wikidata, our source of lemmatization rules, and real-world entities. With the aid of those resources, our system could recognize, normalize and link entities, while being trained with only small amounts of labeled data.

This paper describes Slav-NER: the 3rd Multilingual Named Entity Challenge in Slavic languages. The tasks involve recognizing mentions of named entities in Web documents, normalization of the names, and cross-lingual linking. The Challenge covers six languages and five entity types, and is organized as part of the 8th Balto-Slavic Natural Language Processing Workshop, co-located with the EACL 2021 Conference. Ten teams participated in the competition. Performance for the named entity recognition task reached 90% F-measure, much higher than reported in the first edition of the Challenge. Seven teams covered all six languages, and five teams participated in the cross-lingual entity linking task. Detailed valuation information is available on the shared task web page.