[Paper Review] Annotating Predicate-Argument Structure for a Parallel Treebank

Lea Cyrus, Hendrik Feddes+1 more|ArXiv.org|Jul 1, 2004

Natural Language Processing Techniques4 references3 citations

TL;DR

This paper presents a parallel treebank project (FuSe) that annotates predicate-argument structures in aligned English and German sentences to enable cross-linguistic alignment. By treating predicate-argument structure as a dedicated layer, the project captures translational equivalence beyond syntactic or functional categories, enabling robust alignment of monolingual treebanks through a relational database architecture with specialized alignment and binding layers.

ABSTRACT

We report on a recently initiated project which aims at building a multi-layered parallel treebank of English and German. Particular attention is devoted to a dedicated predicate-argument layer which is used for aligning translationally equivalent sentences of the two languages. We describe both our conceptual decisions and aspects of their technical realisation. We discuss some selected problems and conclude with a few remarks on how this project relates to similar projects in the field.

Motivation & Objective

• To develop a multilingual, multi-layered parallel treebank with syntactic, functional, and predicate-argument annotations for English and German.
• To enable precise alignment of translationally equivalent sentences by modeling predicate-argument structures as a dedicated linguistic layer.
• To support both monolingual and contrastive linguistic research by providing a resource that captures cross-linguistic structural correspondences.
• To overcome limitations of existing annotation schemes by focusing on predicate-argument structure as the core alignment mechanism, rather than relying on constituent or functional structure alone.
• To establish a scalable database architecture that integrates monolingual annotation with a dedicated alignment layer for language pairs.

Proposed method

• Adopt and adapt the TIGER annotation scheme for syntactic and functional structure in both English and German, with language-specific adjustments.
• Introduce a dedicated predicate-argument structure layer that annotates predicates and their arguments independently of constituent or functional structure.
• Use a relational database system (Annotate) for monolingual annotations, extended with a custom alignment database to fuse English and German predicate-argument structures.
• Implement a binding layer to link predicate-argument structures across languages, enabling cross-linguistic comparison and alignment.
• Develop a text-based import pipeline for predicate-argument and alignment data, with a graphical annotation tool in development to improve consistency and efficiency.
• Leverage the parallel structure to explore data-driven clustering of predicates and arguments via translation equivalence, without relying on predefined semantic frames.

Experimental results

Research questions

• RQ1 How can predicate-argument structure serve as a reliable basis for aligning translationally equivalent sentences in a parallel treebank?
• RQ2 To what extent can predicate-argument structure capture translational equivalence when constituent or functional structure fails to do so?
• RQ3 What are the limitations of a minimal predicate-argument annotation scheme in capturing cross-linguistic semantic correspondences?
• RQ4 Can data-driven clustering of predicates and arguments be achieved through alignment in a parallel treebank without relying on predefined semantic frames?
• RQ5 How can a scalable, extensible database architecture support the integration of monolingual annotation with a dedicated alignment layer for multilingual treebanks?

Key findings

• Predicate-argument structure enables alignment of translationally equivalent constituents even when constituent or functional structure differs, as shown in examples involving nominalized verbs in German and their verbal counterparts in English.
• The project successfully demonstrates that a simple, non-frames-based predicate-argument annotation scheme can support reliable cross-linguistic alignment, despite limitations in capturing deeper semantic distinctions.
• The use of a dedicated alignment layer in a relational database architecture allows for the fusion of two monolingual treebanks into a single, coherent parallel resource.
• The project identifies that while deeper schemes like FrameNet or Levin classes could resolve some inconsistencies, they introduce complexity and language-specific challenges, especially in cross-linguistic transfer.
• The alignment mechanism enables the discovery of semantic clusters—such as equivalent predicates (e.g., 'buy' and 'purchase') and arguments—through translation equivalence, suggesting feasibility for data-driven frame semantics.
• The current workflow, involving text-based import of predicate-argument and alignment data, is functional but shows promise for improvement through a dedicated graphical annotation tool to enhance consistency and reduce annotation burden.