We present GEST – a new manually created dataset designed to measure gender-stereotypical reasoning in language models and machine translation systems. GEST contains samples for 16 gender stereotypes about men and women (e.g., Women are beautiful, Men are leaders) that are compatible with the English language and 9 Slavic languages. The definition of said stereotypes was informed by gender experts. We used GEST to evaluate English and Slavic masked LMs, English generative LMs, and machine translation systems. We discovered significant and consistent amounts of gender-stereotypical reasoning in almost all the evaluated models and languages. Our experiments confirm the previously postulated hypothesis that the larger the model, the more stereotypical it usually is.

Large Language Models (LLMs) are increasingly influential in Computational Social Science, offering new methods for processing and analyzing data, particularly in lower-resource language contexts. This study explores the use of OpenAI’s GPT-3.5 Turbo and GPT-4 for automating annotations for a unique news media dataset in a lower resourced language, focusing on stance classification tasks. Our results reveal that prompting in the native language, explanation generation, and advanced prompting strategies like Retrieval Augmented Generation and Chain of Thought prompting enhance LLM performance, particularly noting GPT-4’s superiority in predicting stance. Further evaluation indicates that LLMs can serve as a useful tool for social science text annotation in lower resourced languages, notably in identifying inconsistencies in annotation guidelines and annotated datasets.

We introduce a new Slovak masked language model called SlovakBERT. This is to our best knowledge the first paper discussing Slovak transformers-based language models. We evaluate our model on several NLP tasks and achieve state-of-the-art results. This evaluation is likewise the first attempt to establish a benchmark for Slovak language models. We publish the masked language model, as well as the fine-tuned models for part-of-speech tagging, sentiment analysis and semantic textual similarity.

This position paper discusses the problem of multilingual evaluation. Using simple statistics, such as average language performance, might inject linguistic biases in favor of dominant language families into evaluation methodology. We argue that a qualitative analysis informed by comparative linguistics is needed for multilingual results to detect this kind of bias. We show in our case study that results in published works can indeed be linguistically biased and we demonstrate that visualization based on URIEL typological database can detect it.

Since propaganda became more common technique in news, it is very important to look for possibilities of its automatic detection. In this paper, we present neural model architecture submitted to the SemEval-2020 Task 11 competition: “Detection of Propaganda Techniques in News Articles”. We participated in both subtasks, propaganda span identification and propaganda technique classification. Our model utilizes recurrent Bi-LSTM layers with pre-trained word representations and also takes advantage of self-attention mechanism. Our model managed to achieve score 0.405 F1 for subtask 1 and 0.553 F1 for subtask 2 on test set resulting in 17th and 16th place in subtask 1 and subtask 2, respectively.

In this paper, we present neural model architecture submitted to the SemEval-2019 Task 9 competition: “Suggestion Mining from Online Reviews and Forums”. We participated in both subtasks for domain specific and also cross-domain suggestion mining. We proposed a recurrent neural network architecture that employs Bi-LSTM layers and also self-attention mechanism. Our architecture tries to encode words via word representation using ELMo and ensembles multiple models to achieve better results. We highlight importance of pre-processing of user-generated samples and its contribution to overall results. We performed experiments with different setups of our proposed model involving weighting of prediction classes for loss function. Our best model achieved in official test evaluation score of 0.6816 for subtask A and 0.6850 for subtask B. In official results, we achieved 12th and 10th place in subtasks A and B, respectively.

Using different neural network architectures is widely spread for many different NLP tasks. Unfortunately, most of the research is performed and evaluated only in English language and minor languages are often omitted. We believe using similar architectures for other languages can show interesting results. In this paper, we present our study on methods for improving sentiment classification in Slovak language. We performed several experiments for two different datasets, one containing customer reviews, the other one general Twitter posts. We show comparison of performance of different neural network architectures and also different word representations. We show that another improvement can be achieved by using a model ensemble. We performed experiments utilizing different methods of model ensemble. Our proposed models achieved better results than previous models for both datasets. Our experiments showed also other potential research areas.

Growing amount of comments make online discussions difficult to moderate by human moderators only. Antisocial behavior is a common occurrence that often discourages other users from participating in discussion. We propose a neural network based method that partially automates the moderation process. It consists of two steps. First, we detect inappropriate comments for moderators to see. Second, we highlight inappropriate parts within these comments to make the moderation faster. We evaluated our method on data from a major Slovak news discussion platform.

In this paper, we present neural models submitted to Shared Task on Implicit Emotion Recognition, organized as part of WASSA 2018. We propose a Bi-LSTM architecture with regularization through dropout and Gaussian noise. Our models use three different embedding layers: GloVe word embeddings trained on Twitter dataset, ELMo embeddings and also sentence embeddings. We see preprocessing as one of the most important parts of the task. We focused on handling emojis, emoticons, hashtags, and also various shortened word forms. In some cases, we proposed to remove some parts of the text, as they do not affect emotion of the original sentence. We also experimented with other modifications like category weights for learning and stacking multiple layers. Our model achieved a macro average F1 score of 65.55%, significantly outperforming the baseline model produced by a simple logistic regression.