QA-Noun: Representing Nominal Semantics via Natural Language Question-Answer Pairs

TL;DR

QA-Noun extends QA-based semantic representation to nominal semantics by introducing nine interpretable templates that elicit noun-centered predicate–argument relations. The authors release a dataset of over 2,000 noun mentions and a trained model integrated with QA-SRL to produce a unified, fine-grained decomposition of sentence meaning. Empirical results show QA-Noun nearly covers AMR noun arguments and, when combined with QA-SRL, yields substantially higher granularity than prior fact-based decomposition methods, improving cross-text alignment. The work also demonstrates the practicality of open-weight model pipelines (ICL and LoRA FT) for scalable noun-focused semantic annotation, and provides guidelines, tools, and a foundation for future cross-domain semantic decomposition research.

Abstract

Decomposing sentences into fine-grained meaning units is increasingly used to model semantic alignment. While QA-based semantic approaches have shown effectiveness for representing predicate-argument relations, they have so far left noun-centered semantics largely unaddressed. We introduce QA-Noun, a QA-based framework for capturing noun-centered semantic relations. QA-Noun defines nine question templates that cover both explicit syntactical and implicit contextual roles for nouns, producing interpretable QA pairs that complement verbal QA-SRL. We release detailed guidelines, a dataset of over 2,000 annotated noun mentions, and a trained model integrated with QA-SRL to yield a unified decomposition of sentence meaning into individual, highly fine-grained, facts. Evaluation shows that QA-Noun achieves near-complete coverage of AMR's noun arguments while surfacing additional contextually implied relations, and that combining QA-Noun with QA-SRL yields over 130\% higher granularity than recent fact-based decomposition methods such as FactScore and DecompScore. QA-Noun thus complements the broader QA-based semantic framework, forming a comprehensive and scalable approach to fine-grained semantic decomposition for cross-text alignment.

Figures (6)

• Figure 1: Example QA-Noun annotations for a single target noun (highlighted), explicating each "atomic" fact involving the noun as an individual QA pair.
• Figure 2: Example questions illustrating QA-Noun question templates. NOUN refers to the target noun.
• Figure 3: Reconciliation phase between two QA-Noun annotators. The annotators adjudicate between two proposed arguments that disagree either by extent or by semantic role. The selected argument and role after adjudication is shown schematically under annotator A (in green), while the discarded argument-role is shown under annotator B (in red). The top example showcases role (question) disagreement between the two annotators, while the bottom example depicts different extents (phrases) of the same argument.
• Figure 4: Comparison between example sentences with AMR and QA-Noun annotations. The noun predicate in each sentence is marked in bold and its argument is highlighted inline. Left Comparison between semantic roles when the argument is mutually annotated. Right Diverse arguments captured by QA-Noun's annotators that were out of scope for AMR. They represent different implied meanings, memberships and other relations.
• Figure 5: Sentence decomposition with QA-Noun and QA-SRL compared to fact-based approaches. QA-Noun captures noun-centered relations (e.g., Whose exhibitions?) and surfaces implicit links such as client–consultant relations, which fall under inferential arguments. Combined with QA-SRL verbal roles, they yield a structured predicate–argument breakdown. FactScore and R-ND generate declarative “atomic facts” but typically do not capture such inferential relations.