Hierarchical Bracketing Encodings for Dependency Parsing as Tagging

2025-05-16

AI Summary

This work addresses the challenges of structural redundancy in sequence-labeling dependency parsing and the difficulty of jointly accommodating both projective and non-projective structures. We propose a novel unified encoding family based on hierarchical bracket nesting, modeling dependency trees as bracket sequences enriched with hierarchical information. We formally prove that the existing 4-bit projective encoding belongs to this family and derive an optimal hierarchical scheme, reducing the required label set size for projective trees from 16 to 12. Moreover, we present the first compact encoding capable of representing arbitrary non-projective structures. Theoretically, we establish a formal framework for tree encoding; empirically, we validate the approach across multiple multilingual treebanks. Results show competitive parsing accuracy, significantly reduced label-space complexity, and simultaneous improvements in model generalization and training efficiency.

Technology Category

• Natural Language Processing: Syntax — Tagging, Chunking & Parsing
• Reasoning under Uncertainty: Relational Probabilistic Models
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

We present a family of encodings for sequence labeling dependency parsing, based on the concept of hierarchical bracketing. We prove that the existing 4-bit projective encoding belongs to this family, but it is suboptimal in the number of labels used to encode a tree. We derive an optimal hierarchical bracketing, which minimizes the number of symbols used and encodes projective trees using only 12 distinct labels (vs. 16 for the 4-bit encoding). We also extend optimal hierarchical bracketing to support arbitrary non-projectivity in a more compact way than previous encodings. Our new encodings yield competitive accuracy on a diverse set of treebanks.

Problem

• Develops hierarchical bracketing encodings for dependency parsing
• Optimizes label count for projective tree encoding
• Extends encoding to handle non-projective structures compactly

Innovation

• Hierarchical bracketing for dependency parsing
• Optimal encoding with 12 distinct labels
• Supports non-projectivity more compactly