requirements traceability

This study addresses the challenges of automatically generating test cases from natural language requirements—namely semantic ambiguity, weak traceability, and inconsistent evaluation—which hinder existing approaches from achieving a balance among automation, accuracy, and reliability. Through a systematic literature review conducted in accordance with the Kitchenham and Charters guidelines, the authors analyze AI- and NLP-driven research from 2000 to 2025, proposing a three-phase evolutionary framework and introducing a novel six-dimensional quality gap analysis encompassing automation, ambiguity handling, domain applicability, traceability, evaluation adequacy, and hallucination control. An examination of 21 core studies reveals that current methods fail to simultaneously satisfy all six dimensions. Based on these findings, the paper outlines four actionable research directions: hallucination mitigation, enhanced traceability, complexity-aware modeling, and compliance assurance.

Natural-language software requirements frequently suffer from high ambiguity and significant variability in stakeholders’ expressive capabilities. Method: This study proposes the first systematic, large language model (LLM)-integrated framework for enhancing requirements quality. It innovatively embeds domain-specific guidelines and agile practices into an LLM-driven semantic constraint mechanism, enabling collaborative and traceable requirements generation and refactoring. Contribution/Results: Through architectural design and a proof-of-concept experiment, the framework demonstrates substantial improvements in requirement clarity, consistency, and traceability. Its applicability boundaries and optimization pathways are empirically identified. The work fills a critical research gap in the structured application of LLMs to requirements engineering and provides a scalable, methodology-driven foundation for intelligent requirements specification.

Informal natural language requirements in safety-critical systems impede direct application to formal verification. Method: This paper proposes a semi-automated specification generation approach integrating large language models (LLMs) with domain ontologies, comprising ontology-driven semantic parsing of requirements, LLM-guided instantiation of formal specification templates, and structured reuse of existing specification assets—thereby enhancing verifiability and domain consistency. Contribution/Results: We establish a challenge analysis framework addressing requirement ambiguity, logical incompleteness, and formal mapping deviation. Preliminary validation in aviation and rail transit domains demonstrates a 32% improvement in specification generation accuracy and a 45% reduction in manual correction effort. The work provides a scalable, empirically grounded methodology for trustworthy natural-language-to-formal-specification translation.

Ambiguity in natural-language requirements frequently leads large language models (LLMs) to generate incorrect code. To address this, we propose SpecFix—a fully automated ambiguity resolution method that requires no LLM metacognitive capabilities. SpecFix first models the distribution of program interpretations induced by an LLM’s responses to the original requirement, using program testing and automated repair. It then infers a revised, minimal, and verifiable specification by reversely analyzing distributional shifts and iteratively contracting the requirement space under logical constraints. Crucially, SpecFix decouples ambiguity resolution into two novel phases: (1) modeling the program-interpretation distribution and (2) constraint-driven, contraction-based specification inference—thereby eliminating reliance on self-reflection. Evaluated on HumanEval+ and MBPP+, SpecFix improves Pass@1 by 4.3% on average across GPT-4o, DeepSeek-V3, and Qwen2.5-Coder, and boosts majority-vote solution rates by 3.4%.

Natural language requirements in software development suffer from poor formalization and weak cross-phase traceability. Method: This paper proposes VERIFAI, a novel framework that integrates large language models (LLMs), domain ontologies, and similarity-based retrieval of reusable software components to enable end-to-end automated generation of formal specifications from natural language requirements, with full traceability across phases. VERIFAI leverages LLM-driven semantic understanding, ontology-guided structural mapping, and component-level similarity search to support bidirectional traceability between design, implementation, and verification—and AI-assisted verification. Contribution/Results: Experiments demonstrate that VERIFAI improves formalization efficiency by 3.2× over baseline approaches and achieves an F1-score of 0.89 for specification accuracy. It provides automated, auditable requirement verification tailored for safety-critical systems.

Current evaluations of natural language requirements lack a unified, evolvable knowledge foundation, hindering both requirements engineering practice and research. Method: This paper introduces QF-Ontology—the first quality-factor ontology for natural language requirements—systematically synthesizing 105 empirically validated quality factors (e.g., passive voice, sentence length, referential clarity) across syntactic, semantic, and structural dimensions. Developed via rigorous ontology engineering—including domain analysis, conceptual modeling, OWL formalization, and web-based implementation—it establishes a “living ontology” paradigm enabling standardized organization, dynamic evolution, and community-driven maintenance. Contribution/Results: We release an open-source, web-accessible knowledge repository supporting quality-factor search, reuse, and community contribution. QF-Ontology fills a critical structural gap in the requirements quality knowledge base, providing a scalable, interoperable foundation for automated quality assessment, requirement optimization, and cross-project experience reuse.

This work addresses the challenge of automatically generating semantic-aligned and verifiable formal properties from unstructured natural language requirements. The authors propose a novel large language model (LLM)-based agent architecture that, for the first time, explicitly integrates modeling and verification constraints into the requirement formalization pipeline. Through a modular design, the approach unifies requirement extraction, formalization compatibility filtering, and property translation into a cohesive workflow. Evaluated across three real-world scenarios, the method achieves an accuracy of 77.8%, substantially improving the syntactic correctness, semantic alignment, and verifiability of the generated formal properties.

High-quality paired requirements-code datasets are scarce, significantly hindering empirical research in requirements engineering. This work proposes an agent-based approach that integrates large language models (LLMs), retrieval-augmented generation (RAG), and human-in-the-loop validation to systematically evaluate, for the first time, the feasibility of LLMs in automatically generating realistic requirements from source code, synthesizing requirement smells, and identifying unimplemented requirements. The study reveals that current LLMs struggle to reliably generate or detect such requirement-related issues, and that neither fully automated nor purely manual review alone is sufficient. These findings underscore substantial challenges in automatically constructing trustworthy requirements-code datasets and highlight the critical need for enhanced human oversight in the process.

This study addresses the inefficiency and error-proneness of manual identification of dependencies among natural language requirements in complex software systems, particularly in the face of high coupling, ambiguity, and frequent changes. To overcome these challenges, we propose LEREDD, a novel approach that integrates Retrieval-Augmented Generation (RAG) with In-Context Learning (ICL) to leverage large language models for automatically detecting multiple fine-grained dependency types. We construct a labeled dataset comprising 813 requirement pairs and evaluate our method experimentally, achieving an accuracy of 0.93 and an F1 score of 0.84. Notably, the non-dependency class attains an average F1 score of 0.96, and the “Requires” dependency type shows an improvement of over 94% in average F1 compared to baseline methods, significantly enhancing both precision and generalization in requirement dependency detection.

This work addresses inconsistencies arising from structural mismatches between natural language and formal languages during requirements formalization. It proposes a “consistency through formalization” principle, mandating strict logical alignment among natural language, the structured language FRETish, and the formal temporal logic MTL. Guided by this principle, the authors refine the FRETish-to-MTL translation pipeline in NASA’s FRET tool. Their approach uniquely integrates cross-layer consistency constraints into a collaborative framework combining large language models and formal verification tools. This integration not only uncovers and corrects multiple inconsistencies in the original translation but also demonstrates superior correctness and reliability, as substantiated by formal equivalence proofs and empirical evaluation.

This work addresses the error-prone and labor-intensive process of manually translating regulatory texts such as the GDPR and the EU AI Act into actionable software requirements. The authors propose Reg2Req, the first end-to-end automated pipeline that leverages natural language processing to identify regulatory provisions, generate system-agnostic software requirements accompanied by plain-language explanations, and establish traceability links. The approach supports requirement classification, use case seed generation, and cross-reference analysis, achieving macro-averaged F1 scores of 0.82 on the GDPR and 0.78 on the EU AI Act. A user study demonstrates that the generated plain-language explanations significantly enhance users’ comprehension and confidence in taking compliance actions (p < 0.001), with all participants expressing willingness to adopt the output as a starting point for compliance efforts.