Task 1: Spotting Machine-Generated Text from Language Models for Polish (ŚMIGIEL)
Task description TBA
Task 2: Gender-inclusive LLMs for Polish
Polish is a grammatical gender language in which all nouns inherently encode gender markers as an integral part of the grammatical system. For example, śliwka [a plum] is feminine, jabłko [an apple] is neutral, whereas pomidor [a tomato] is masculine. All adjectives and verbs associated with a noun must match the noun’s grammatical gender. Additionally, personal nouns have distinct feminine, e.g., nauczycielka [a teacherfem] and masculine forms, e.g., nauczyciel [a teachermasc]. While feminine personal nouns typically denote female individuals or groups of females, masculine personal nouns refer not only to male individuals or male groups but also to mixed-gender groups and even females, a phenomenon known as the generic masculine, e.g., niemiecka polityk Ursula von der Leyen [Germanfem politicianmasc Ursulafem von der Leyen].
Although the grammatical system of Polish allows for naming individuals according to their natural gender (i.e., female or male), standard Polish remains heavily masculine-centric. This is reflected in a strong dominance of masculine expressions over feminine ones, which may be interpreted as reinforcing gender bias and exclusion.
One social consequence of this linguistic system is that current large language models (LLMs) trained on Polish texts inherit and reinforce masculine bias, generating gender-imbalanced outputs. As LLMs become increasingly integrated into communication, translation, and content generation systems, ensuring their outputs reflect gender inclusivity is crucial, particularly in gender-rich languages like Polish.
The dominance of masculine expressions over feminine ones in a language is a form of gender discrimination (GEC, GNL-EU). Acknowledging the harmful effects of sexist language, the Council of Europe encourages its member states to eliminate sexism from language and to adopt practices that support gender equality. In line with this recommendation, we introduce a task focused on developing gender-inclusive LLMs for Polish.
Task 3: Polish Language Document Layout Detection
The aim of the task is to advance the research concerning document layout detection in Polish. Document layout recognition involves identifying such elements as the title, section header, main text, table, footnote, image, list, etc. The input to the system consists of page images in graphic file format, while the output is the set of identified areas representing individual structural elements (rectangle coordinates) and the corresponding element category label.
Task 4: Polish Speech Emotion Recognition Challenge
Speech emotion recognition (SER) represents a critical area of research due to its extensive potential applications. Recent advancements in automatic speech recognition (ASR) and large language models (LLMs) have led to new possibilities for the development of SER systems. However, given that SER combines both audio processing and natural language understanding, the field still faces challenges. The subtle emotional cues are conveyed not only by what is said, but also by how it is said.
The difficulty of this task is caused by the subjective nature of emotions, both in their expression and perception. Each person may experience and interpret emotions differently, depending on factors such as language, cultural background, and situational context. Additionally, even subtle variations in speech can complicate generalization and reduce robustness, especially in the case of low-resource languages or challenging acoustic conditions.
In order to promote research in this area, we introduce the Polish Speech Emotion Recognition Challenge. The goal of this task is to evaluate how well current systems can identify emotional states from speech across diverse conditions, languages and speakers.