Meet UD_Czech-PDTC: A Large and Genre-Rich Treebank in Universal Dependencies

This paper introduces the UD_Czech-PDTC treebank, a new resource for the Universal Dependencies (UD) project. By converting the Prague Dependency Treebank-Consolidated (PDT-C 2.0) into the UD format, the authors have created a massive, manually annotated dataset (3,440K words) that spans diverse genres, including journalism, spoken language, and user-generated content. This resource is significant because it bridges the gap between the highly detailed, multi-layered Prague Dependency Treebank (PDT) and the cross-linguistically consistent UD framework.

Converting PDT to UD is not a simple mapping exercise. While both frameworks are dependency-based and share similar core concepts, they differ in their handling of specific linguistic phenomena. The authors detail how they reconciled these differences, particularly in:

• Part-of-Speech (POS) Tagging: The authors had to map Czech-specific categories to the universal inventory, such as identifying determiners (DET) and auxiliaries (AUX), which are not explicitly distinguished in traditional Czech grammar.
• Syntactic Structure: The frameworks diverge in how they represent coordination and ellipsis. For instance, PDT uses a "technical head" approach for coordination, whereas UD treats the first conjunct as the parent of subsequent conjuncts.
• Function Words: The treatment of copulas and prepositions differs, requiring careful adjustments to ensure the resulting UD trees remain consistent with the project's global guidelines.

This conversion is vital for both computational linguistics and theoretical research. By integrating the rich, multi-layered annotations of the Prague family of treebanks into the UD ecosystem, researchers can now perform large-scale, cross-lingual analyses on a high-quality, genre-diverse Czech dataset. The paper demonstrates that despite the differences in "universality" versus language-specific depth, the two frameworks can be successfully aligned to support both natural language understanding and broader linguistic studies.