HUM data lab > Workflows > Tunes of the World map: Exploring Estonian and Ukrainian Folk Song Heritage

Tunes of the World map: Exploring Estonian and Ukrainian Folk Song Heritage

The project aims to create an interactive digital map for exploring Estonian and Ukrainian folk-song traditions through geographical, thematic, emotional, and structural perspectives. It was eveloped during the HUM Hackathon 2026. The current prototype focuses primarily on Ukrainian folk-song materials, while future versions will integrate Estonian datasets and enable cross-cultural comparison.

The workflow combines digital humanities methods, natural language processing, cultural heritage data management, and interactive visualisation. Two large folk-song corpora 85,550 Estonian songs and 56,726 Ukrainian songs were prepared, harmonised, analysed, and transformed into a searchable map interface. Users can navigate from ethnographic regions to individual singers and complete song texts, explore emotional and thematic patterns, and investigate the geographical distribution of oral traditions.

Workflow steps

Keywords: collecting, sentiment analysis, text categorization

The primary purpose of this stage is to gather and analyze the vast collections of Estonian and Ukrainian folk songs to create a unified foundation for our project. The corpora were balanced by excluding the Estonian runic song corpus, which contains substantial Finnish material. After filtering the dataset to focus on Estonian folk songs only, the Estonian and Ukrainian corpora became roughly comparable in size, allowing for more meaningful cross-corpus comparisons.

We start by processing two massive datasets: 85,550 Estonian songs and 56,726 Ukrainian songs. Because these databases represent different cultural, linguistic, and song traditions, this comparative phase is critical for identifying both shared and unique characteristics before any visual mapping can begin. Key analytical dimensions include: genre groupings, line length distribution (measured in syllables or vowels as a proxy for metrical organization), degree of formulaicity versus narrativity, and structural stability within each tradition. Together, these metrics inform how songs are classified, filtered, and ultimately made navigable in the final interface.

Before analysis, both corpora underwent data preparation and harmonisation. For the Estonian corpus, multiple CSV files were processed, category labels were cleaned and standardised, Finnish materials were filtered out, and genre categories were translated into English. For the Ukrainian corpus, metadata were normalised by extracting collection names from song identifiers, removing technical labels, standardising collection metadata, and validating genre classifications. These procedures transformed heterogeneous archival materials into structured and comparable datasets suitable for computational analysis and visualisation.

The current hackathon prototype includes the analysis of both Ukrainian and Estonian folk-song corpora, including thematic, emotional, genre, and metrical comparisons. The publicly accessible map interface currently focuses on the Ukrainian corpus. Integration of Estonian songs into the interactive map and the development of a full cross-corpus visual exploration interface are planned for future development.

Keywords: content analysis, sentiment analysis, tagging, stylistic analysis, natural language processing, machine learning

This stage transforms raw, unstructured folk song databases into structured, machine-readable datasets through several analytical procedures.

Step 2.1. We use a Large Language Model (Claude) to assign songs to thematic categories. The model identifies recurring motifs and topics in song texts and classifies songs into broad thematic groups such as love, family, wedding, Cossack life, captivity, and death.

Step 2.2. Songs are analysed using sentiment-analysis techniques to identify their emotional characteristics. Emotional scores include valence (how positive or negative the emotional tone of a song is) and arousal (the intensity or strength of emotional expression). Sentiment analysis provides emotional scores describing the overall tone and intensity of each song, while thematic classification assigns songs to broader thematic categories.

Step 2.3. The corpus is analysed according to existing genre classifications in the metadata. This step allows us to examine the distribution of ritual, non-ritual, narrative, and other song categories across regions and collections. Genre metadata were used to compare the distribution of song types across the corpora. While both traditions share genres such as wedding, family, and work songs, the Ukrainian corpus includes distinctive categories such as Cossack songs and Chumak songs, whereas the Estonian corpus is characterised by songs about youth life, singing, entertainment, and social relations.

Step 2.4. We analyse the structural characteristics of songs using vowel-per-line measurements as a proxy for verse length and metrical organisation. This enables comparison of poetic structures across genres and traditions. Songs were analysed through metrical profiling (analysis of verse structure based on the number of vowels per line, used as a proxy for metrical organisation). This allowed us to compare rhythmic patterns across genres and between the Estonian and Ukrainian corpora.

Step 2.5. The outputs of thematic, emotional, genre, and metrical analyses are integrated into a single searchable dataset. Each song is enriched with metadata tags describing its theme, emotional profile, genre affiliation, and metrical characteristics.

The output is a consistent, queryable dataset - songs tagged by theme, emotional dimension, genre group, and metrical signature - that serves as the foundation for visualization in the next stage.

Keywords: data visualization, design, mapping

We chose a map-based design because folk traditions are deeply tied to the land; seeing Estonian and Ukrainian songs together on a single interface helps reveal patterns of cultural migration and shared Baltic-Slavic heritage. This stage bridges the gap between the "Discover" phase (raw data) and the "Deliver" phase (the final tool), ensuring the information is not just accurate but also intuitive and engaging.

This stage translates the results of computational analysis into an accessible and interactive visual environment. The goal is to enable both researchers and the wider public to explore song traditions.

Step 3.1. The first stage focuses on developing the visual framework of the Tunes of the World Map . A map-based interface was chosen because song traditions are closely connected to specific regions and local communities. The interface provides a geographical overview.

Step 3.2. Users can explore songs through regional filters and navigate from broader ethnographic regions to smaller geographical units. This functionality allows users to investigate the spatial distribution of genres, themes, and emotional patterns within the corpus.

Step 3.3. Comparative thematic, emotional, genre, and metrical analyses were conducted for both Estonian and Ukrainian song corpora. By designing a clear visual interface, we make complex analytical data – such as genre distributions, emotional patterns, and metrical characteristics (e.g., the number of vowels per line) – accessible to users without a technical background. While the current map interface focuses on the Ukrainian corpus, future versions of the platform will visualise comparative results through an interactive cross-corpus comparison interface.

Step 3.4. This feature allows users to move from a geographical region to an individual performer and then to specific song records. Each record includes the full song text, metadata, and analytical annotations (theme, emotion, genre). This functionality provides direct access to the original cultural material underlying the visualisations.

Keywords: preserving, dissemination, searching

The deliver stage transforms the prototype into a publicly accessible cultural heritage resource. It represents the culmination of the workflow, where the structured and annotated datasets developed in Steps 1–2 and the visual environment designed in Step 3 are brought together in a functional platform for researchers, musicians, educators, and the wider public. By providing intuitive access to archival folk-song collections, the platform supports both cultural heritage preservation and new forms of comparative research.

Step 4.1. Public-Access Prototype Launch
The interactive map is made available online for researchers, musicians, educators, and the general public.

Step 4.2. Advanced Search and Query Interface
Users can search and filter songs according to region, genre, theme, emotional characteristics, and structural features.

Step 4.3. Heritage Preservation and Dissemination
The platform contributes to the preservation, accessibility, and dissemination of folk-song heritage by making archival materials easier to discover and explore.

Keywords: dissemination, publishing

This workflow produces several research and cultural heritage outputs:

Spep 5.1 Dataset and Tagged Corpus – a structured dataset enriched with thematic, emotional, genre, and metrical annotations.

Step 5.2 Interactive Map – a publicly accessible visualisation platform currently supporting exploration of the Ukrainian folk-song corpus, with Estonian corpus integration planned for future development.

Step 5.3 Research and Heritage Tool – a digital resource supporting textual comparative folklore research, education, and cultural heritage preservation.