Speakers
Description
In the era of data-driven linguistics, the transition from traditional, human-oriented lexicography to machine-readable and interoperable language resources is paramount. The Digital Dictionary Database of Slovene (DDDS), developed by the Centre for Language Resources and Technologies (CJVT) at the University of Ljubljana, represents a paradigm shift in how national lexicographical data is curated, stored, and disseminated. Rather than treating dictionaries as isolated digital products, the DDDS functions as a centralized, relational infrastructure designed to support a multi-layered ecosystem of language resources—ranging from the Thesaurus of Modern Slovene to the Collocations Dictionary and morphological database Sloleks. This paper will describe the structural design, the underlying data model, and the multifaceted API-based application of the DDDS, highlighting its role in the "responsive dictionary" framework.
At its core, the DDDS is built on a PostgreSQL database managed via the Django Object-Relational Mapping (ORM) framework. Unlike traditional lexicographical databases that often rely on XML hierarchies, the DDDS employs a highly normalized relational model that prioritizes the atomization of linguistic data. The model is organized into several clusters, with the primary entities being Lexical Units and Senses. A key innovation in the DDDS is its approach to "Lexemes," which correspond to tokens in a corpus but are treated as distinct structural components within the database. This allows for the precise representation of complex multi-word expressions, compounds, and syntactically varied forms. By utilizing a single primary key system and Django’s "content types" and "generic relations," the database maintains a flexible architecture where additional metadata—such as dictionary-specific labels or administrative information—can be associated with any table without necessitating disruptive schema changes. This modularity ensures that the database remains a "living" resource capable of accommodating various lexicographical workflows.
The DDDS serves as the "backbone" for a suite of modern Slovene resources. Most notably, it powers the Thesaurus of Modern Slovene, which introduced the concept of the responsive dictionary. In this model, the dictionary is not a static text but an evolving dataset where automatically generated candidates are continuously refined through expert validation and crowdsourcing. The database manages these dynamic relationships, tracking synonymy, sense disambiguation, and collocation frequency extracted from the Gigafida reference corpus.
A critical component of the DDDS project is its commitment to the FAIR (Findable, Accessible, Interoperable, and Reusable) principles. This is achieved through a comprehensive REST API that allows external developers, researchers, and NLP systems to interact with the database in real-time. The API design distinguishes between read-only public routes and restricted read-write routes.
A significant modern application of the DDDS is its role in the PoVeJMo project, which aims to optimize Large Language Models for the Slovene language. By providing high-quality, structured grammatical and lexical data, the DDDS enables models to learn the nuances of Slovene grammar and semantics more effectively than they would from unstructured web-crawl data alone. The extraction of sense-specific examples and sense-relation mappings from the database provides a "gold standard" dataset for training models in tasks such as Natural Language Inference (NLI) and Word Sense Disambiguation (WSD).
The development of the DDDS highlights a shift toward a "digital-first" lexicographic workflow. Traditional dictionary writing involved drafting entries that were later converted to digital formats. In the DDDS framework, the database is the dictionary. The interface used by lexicographers—a new dedicated online editor—is integrated with the database, ensuring that updates are immediately reflected in the underlying data structure. The Digital Dictionary Database of Slovene demonstrates that a central, well-structured database is essential for the survival and growth of a national language in the digital age. By decoupling the linguistic data from the final presentation layer (the web interface), CJVT aims to create a robust infrastructure that supports both human users and automated systems. Future developments involve the expansion of the data model to include more granular semantic relations and the continued integration of crowdsourced feedback, ensuring that the DDDS remains the authoritative, yet dynamic, reference for the modern Slovene language.
In the final paper we will describe the specialized data model with REST API, the online editor, and various online visualizations of DDDS data.