Requirements engineering digest, week 32 of 2026

1. Automatic Translation of Unstructured Requirements into Linear Temporal Logic through Large Language Models

   Alexandra Newcomb, Omar Ochoa

   arXiv (cs.SE) · preprint

   Benchmarks six off-the-shelf large language models on few-shot translation of unstructured natural language requirements into Linear Temporal Logic, scoring 450 generated formulas by manual semantic evaluation, pass@k and a self-consistency measure, and pairing each formula with a model-written explanation for readers who are not formal methods experts.

   Abstract

   Automatically translating unstructured natural language requirements into formal specifications remains a challenge in requirements engineering and formal methods, particularly for safety- and mission-critical systems whose verification depends on mathematically precise specifications. This paper evaluates whether contemporary off-the-shelf Large Language Models (LLMs) can help bridge this gap by generating Linear Temporal Logic (LTL) formulas directly from unstructured requirements. The study examines six modern LLMs using a few-shot prompting strategy on a heterogeneous benchmark of 15 structurally varied requirements. Five independent generations were collected for each requirement-model pair, yielding 450 candidate LTL formulas in total. Performance was assessed through manual semantic evaluation, pass@k for k in {1, 3, 5}, and a self-consistency measure capturing syntactic reproducibility across stochastic trials. The results indicate that current general-purpose LLMs can achieve practically significant performance on the unstructured NL-to-LTL task without task-specific fine-tuning. The study also considers understandability for non-experts by pairing generated formulas with model-produced natural language explanations and discussing the complementary use of timeline-based LTL visualization. The findings suggest that modern LLMs are becoming viable front-end assistants for semi-automated formalization workflows.

   Source: arxiv.org/abs/2608.06287

2. In Terms of Explainability: Refining Requirements for Self-Explainable Systems

   Arno Leue, Akhila Bairy, Maike Schwammberger

   arXiv (cs.SE) · preprint

   Merges the competing definitions of explainability into one set, derives from it the structured explainability requirements a self-explainable system has to meet, and folds explanation goodness, and with it the correctness of an explanation, into the definitions, as groundwork for the standards the EU AI Act and IEEE 7001-2021 call for.

   Abstract

   Autonomous and software-intensive systems have been growing in occurrence, complexity, and assumed responsibility. Due to the high complexity of these systems, properties like transparency and explainability must be a focus of investigation. To date, no universally applicable definition and guide for the development of (self-)explainable systems exists. A need for explainability standards has already been recognized in the EU AI Act and the IEEE Transparency Standard 7001-2021. To address this need, we propose unified definitions in explainability based on an analysis and combination of existing definitions. Additionally, we present structured explainability requirements that are necessary to build (self-)explainable systems. By analysing the resulting taxonomy, we propose the incorporation of explanation goodness and thus correctness of explanations into the unified definitions. With our approach, we support the development of formal standards for (self-)explainable systems. Establishing such a uniform taxonomy also is a beneficial step towards certifying or auditing explainable systems.

   Source: arxiv.org/abs/2608.06049

3. CTRAG: An In-Context Retrieval-based Framework for Automated Compliance Checking using LLMs

   Muhammad Roman, Karen Rafferty, Barry Devereux

   arXiv (cs.CL) · preprint

   A retrieval-augmented pipeline that extracts control questions from regulatory texts and cross-references them against unstructured company documentation, reaching an F1 of 78% and a recall of 85% in a proof of concept deployed at a Big Four firm, including cases where compliance runs indirectly through a third-party provider.

   Abstract

   Trust is fundamental in modern regulatory ecosystems, and compliance checking plays a critical role in fostering that trust. Regulatory compliance verification is essential for businesses operating in highly controlled environments, as it ensures alignment with sector-specific guidelines across domains such as financial reporting, data privacy, and cybersecurity. Manual compliance testing, however, is often time-intensive and prone to inconsistencies, particularly when compliance depends indirectly on third-party services such as cloud providers, where vendors rely on external providers to meet regulatory standards. In this paper, we present CTRAG, a novel Retrieval-Augmented Generation (RAG) pipeline designed for automated compliance checking. CTRAG employs advanced strategies, including adaptive chunking, dynamic retrieval configurations, and in-context learning, to improve the precision and relevance of compliance assessments. By extracting control questions from regulatory texts and cross-referencing them with unstructured company documentation, CTRAG achieves highly accurate, document-informed compliance verification, even in cases of indirect compliance through third-party services. Empirical evaluations demonstrate significant improvements, with CTRAG achieving an F1-score of 78% and a recall of 85% in the final deployed configuration, ensuring minimal missed non-compliance cases while reducing manual reviewer effort in a real-world deployment. To validate CTRAG value, we developed and deployed a POC within a Big Four professional services firm, applying it to real-world cases and cross-checking results against manual compliance reports. These findings highlight CTRAG potential to streamline compliance workflows, mitigate risks, and enhance regulatory trust in complex, high-stakes environments.

   Source: arxiv.org/abs/2608.02472

4. AI-assisted Script Management for Requirements Elicitation Interviews

   Anmol Singhal, Paulo Carvalho, Travis Breaux

   arXiv (cs.SE) · preprint

   An elicitation workflow that writes a theory-guided interview script from business goals and then, during the interview, tracks which topics have been covered and offers follow-up questions on demand. A between-subjects study against trained but unassisted interviewers finds it changes the shape of the interview: fewer topics, more of the script covered, three times as many follow-ups per topic, and more refined goal models.

   Abstract

   Requirements elicitation interviews require interviewers to balance topic coverage, active listening, and adaptive probing while responding to stakeholders in real time. Although prior work has explored AI support for isolated interviewing tasks, such as script generation and follow-up question generation, little is known about how integrated support affects the interview and what requirements artifacts emerge. Furthermore, script management---which helps the interviewer track topic coverage in real time and decide when to probe further---remains underexplored. This paper presents an AI-assisted elicitation workflow that combines theory-guided script generation grounded in business goals with live support for topic coverage tracking and on-demand follow-up question generation. We evaluate the workflow in a between-subjects quasi-experimental study comparing a no-training, AI-assisted condition with a training, AI-unassisted condition. Based on a rubric derived from elicitation best practices, the AI-generated scripts score higher than training-only scripts (92.8 vs. 74.8 out of 100). AI-assisted interviews cover fewer topics (9.6 vs. 14.5), cover more scripted questions (86% vs. 69%), ask more follow-ups per topic (3.43 vs. 1.15), and produce more refined goal models (lowest-level goal fraction 0.653 vs. 0.598). Participants find script management useful, rating topic tracking as the most useful workflow feature (86% agreement). Collectively, these results show that the AI-assisted condition is associated with a different interview trajectory and different elicited requirements than a training-only condition, positioning AI-assisted workflows as elicitation scaffolds for future studies.

   Source: arxiv.org/abs/2608.01640

5. A Triple-Robustness Analysis of Retrieval-Augmented Generation for Multi-Hop Requirements Traceability

   Meftun Akarsu, Burak Ozdemir, Dogancan Buyukcolak, Recep Kaan Karaman

   arXiv (cs.IR) · preprint

   Holds a five-pipeline GraphRAG against vector-RAG matrix fixed while varying the embedder, the corpus, DO-178C typed-edge requirements against Wikipedia paragraph chains, and the LLM judge, over thousands of runs. Where citation quality is measured turns out to flip the ranking between architectures, and single-judge faithfulness scoring proves unstable across retrieval states.

   Abstract

   Reported verdicts on GraphRAG versus vector RAG disagree, and the evidence is typically tied to a single corpus, embedder, and judge -- and, we show, to where citation quality is measured. We present a triple-robustness analysis that holds a five-pipeline architecture matrix fixed and varies embedder (local e5-small vs. Azure text-embedding-3-small), corpus (DO-178C typed-edge requirements vs. Wikipedia paragraph chains via MuSiQue), and judge (paired GPT-5.4 x GPT-4.1 on both corpora), over 2x4,440 main-matrix runs, 600 cross-corpus runs, and over 5,000 faithfulness judgments. (C2a) GraphRAG's graph walk floods the context window at precision 0.12-0.23, but the synthesizer cites selectively at precision 0.48-0.65; scoring the retrieved set as the attribution set inverts the architecture ranking, which reconciles part of the disagreement in prior reports. (C1) Answer-level citation winners are corpus- and stratum-conditional but embedder-robust: GraphRAG ties vanilla on short-hop DO-178C queries and wins every MuSiQue stratum, while agentic pipelines lead only on 3+-hop requirements queries. (C2b) Faithfulness is corpus-conditional: on DO-178C it declines with hop distance (trend p<0.05 in three of four judge x embedder combinations); on Wikipedia chains neither judge shows a collapse. (C3) Single-judge LLM faithfulness is fragile to retrieval state: GPT-5.4's self-kappa across embedders is 0.137 (41% verdict change) against a same-day test-retest floor of 0.76, and re-judging frozen inputs eleven weeks later gives kappa <= 0.14 for both judges. A learned router on dense embeddings alone reaches macro-F1 0.86 on hop classification (C4). We argue that RAG architecture claims should be tested at this level of robustness -- including robustness to the citation-measurement point -- before they are trusted.

   Source: arxiv.org/abs/2608.00705

6. Requirements-driven database evolution: A systematic literature review

   Hana Litavska, Marek Suchanek, Petr Kroha

   Information and Software Technology 199 · in press

   Classifies 77 studies published between 2020 and 2025 on database design and evolution under a six-dimensional taxonomy covering problem scope, target environment, solution input, workload dependency, transformation strategy and reversibility. Two gaps stand out: approaches driven by textual requirements are scarce next to model-driven ones, and reversibility of an evolution step is largely unsupported.

   Abstract

   Context: Database evolution and management are often complex and error-prone processes, largely driven by the need to adapt to changing requirements. Objective: This study aimed to collect information on current research in database design and evolution, examine the approaches used, categorize the environments that existing solutions target, along with their inputs, and evaluate whether the existing methods are reversible. In addition, the study aimed to develop a taxonomy for systematically classifying and comparing the identified approaches. Method: The systematic literature review was conducted following established guidelines, employing both backward and forward snowballing techniques. The literature search was performed across four databases. After filtering and snowballing, 76 relevant papers published between 2020 and 2025 were identified, with one additional key study included manually, resulting in a final set of 77 papers. The studies were grouped into classification units and classified using an inductively developed taxonomy covering problem scope, target environment, solution input, workload dependency, transformation strategy, and reversibility support. Results: We categorized existing research using a six-dimensional taxonomy, enabling a structured comparison of the existing approaches. Most solutions primarily focused on NoSQL databases. Conceptual models were predominantly used as input. We compared various approaches identified in the literature and highlighted their lack of reversibility. The approaches differed significantly between solutions for relational databases. In NoSQL databases, model-driven and workload-driven strategies were prevalent. In multi-model database solutions, concepts from category theory were widely employed, and several approaches also addressed polystore systems. Conclusion: Our findings indicate a significant gap in the recent literature regarding the reversibility of evolution as a means to enhance testability and robustness. In terms of input specifications, although the use of textual instructions is increasing, there is a notable lack of solutions based on textual requirements. In contrast, solutions that employ conceptual models are much more common.

   Source: doi.org/10.1016/j.infsof.2026.108276 · doi:10.1016/j.infsof.2026.108276

7. Embrace ad-hoc requirements - A characterization based on a multi-method industry study

   Andrea Wohlgemuth, Fabiano Dalpiaz, Erik Kamsties

   Journal of Systems and Software 243 · in press

   An in-depth multi-method case study at a large-scale software and mechatronic systems provider, followed by an external validation beyond it, that names and defines the ad-hoc requirement: one written, stored or communicated in a suitable-for-the-moment manner, in meeting notes, e-mails and slides rather than in the specification or the requirements tool. The paper argues these are common, gives evidence of their existence inside and outside the case company, and puts them forward as a class of requirements worth studying.

   Abstract

   Motivation: Requirements are not only represented in specification documents, stored in requirements management tools, and communicated via official channels. Neither are they always phrased according to academic recommendations. Practitioners use quick-and-dirty representations that, while not compliant with guidelines, can be more suitable and efficient for the moment. These include informal representations of requirements in meeting notes, e-mails, and presentation slides. Problem: Although we presume such less disciplined ways of representing and communicating requirements are common, and various authors have hinted at their existence, the scientific community lacks concrete evidence and a precise characterization. Studying these requirements is important to address the trade-off between immediate suitability and long-term impact. Principal ideas: We conduct an in-depth case study at a large-scale software and mechatronic systems provider – a context where systems and humans cooperate –, followed by an initial external validation to find evidence for generality beyond the case company. We use multi-method research, including surveys, interviews, and an analysis of processes and artifacts. Exploring instances and the context of these requirements, we conceptualize the observations and derive a definition of an ad-hoc requirement (AhR) as one that is written, stored, or communicated in a suitable-for-the-moment manner. We then validate the existence of AhRs and our findings concerning AhRs outside the case company. Contribution: This paper positions ad-hoc requirements in the requirements engineering landscape, provides evidence of their existence both within and beyond a case company, and puts forward a new class of requirements for researchers to study.

   Source: doi.org/10.1016/j.jss.2026.113062 · doi:10.1016/j.jss.2026.113062

8. AI Transparency: Governance Compliance or Stakeholder Requirements?

   Muneera Bano, Didar Zowghi

   arXiv (cs.CY), to appear at IEEE RE 2026 · accepted

   Scores 92 AI transparency statements published by Australian Government agencies against the disclosure criteria they must meet, and against a Risk-Control-Involvement-Need framework that separates stakeholders by their structural position. Compliance is widespread, calibration is not: the criteria that matter most to high-risk, low-control stakeholders are the ones least substantively addressed, which the authors name the Transparency Illusion.

   Abstract

   Transparency is increasingly mandated for public-sector AI systems, with organisations required to publish statements describing their AI use and oversight arrangements. However, the existence of such artefacts is often treated as equivalent to transparency itself, despite limited evidence that they proportionately serve relevant stakeholder groups. From a requirements engineering perspective, this raises a validation concern: compliance with mandated disclosure criteria does not necessarily ensure transparency adequacy for stakeholders with different levels of risk exposure, decision control, and involvement. This paper presents an empirical analysis of 92 publicly available AI transparency statements published by Australian Government agencies under the national AI governance mandate. We introduce the stakeholder Risk--Control--Involvement--Need (RCIN) framework to differentiate stakeholder classes according to their structural position and transparency needs. Using a structured rubric derived from the mandated criteria, we evaluate how both the mandate and published statements are calibrated to each stakeholder class. The findings show that while structural compliance is widespread, transparency calibration is uneven. Criteria serving high-control stakeholders are consistently realised, whereas criteria most critical for high-risk, low-control stakeholders are fewer and less substantively addressed. We conceptualise this as the Transparency Illusion: a condition in which transparency appears satisfied through compliant artefacts yet remains unevenly calibrated to stakeholders bearing the greatest exposure to AI-supported decisions. The study frames transparency as a stakeholder-calibrated validation problem, demonstrating that artefact-level compliance does not constitute requirements validation in this context.

   Source: arxiv.org/abs/2606.30652

9. A Taxonomy of Human-Robot Teamwork Requirements

   Anastasia Mavridou, Hazel Taylor, Sandy Lozito, Louise Dennis, Michael Fisher, Marie Farrell

   arXiv (cs.SE), to appear at IEEE RE 2026 · accepted

   Builds a two-level taxonomy of human-robot teamwork requirements, 6 categories and 21 subcategories, from 361 requirements extracted out of 14 cross-domain sources of literature, standards and regulatory guidance, then validates it with five domain experts and against a separate corpus of 448 requirements from six domains.

   Abstract

   Autonomous systems are increasingly deployed in safety- and mission-critical domains where humans and robots must operate as a team to complete complex tasks. Existing requirements for Human-Robot teamwork remain fragmented across disparate sources, with no unified framework that addresses complexities of collaborative Human-Robot tasks. We address this gap by presenting a taxonomy of Human-Robot Teamwork (HRT) requirements derived from analysis of (academic and industrial) literature, standards and regulatory guidance. We extracted a construction corpus of 361 requirements from 14 cross-domain sources. Through iterative classification and refinement, we develop a two-level hierarchical taxonomy comprising 6 high-level categories and 21 low-level subcategories that distinguish information provision, relational control, decision support, safety mechanisms, performance monitoring, and foundational system capabilities. We validate the taxonomy through expert evaluation with 5 domain specialists and a utility demonstration on an independently assembled corpus of 448 requirements drawn from 19 sources spanning six HRT domains.

   Source: arxiv.org/abs/2607.27302

10. From Business Problems to AI Solutions: Where Does Transformation Support Fail?

    Abir Trabelsi, Imen Benzarti, Hafedh Mili, Darine Ameyed

    arXiv (cs.SE), to appear at IEEE RE 2026 · accepted

    Compares 18 approaches drawn from requirements engineering, machine learning project management and automation against a seven-stage transformation framework. Most name the machine learning task or algorithm among their outputs, only four give partial guidance for deriving it and none gives systematic guidance, a hole the authors name the Analytics Translation Problem.

    Abstract

    Translating business problems into well-specified machine learning solutions is a prerequisite for successful AI systems, yet this upstream translation is still one of the least supported steps in existing methodologies. We conduct a structured narrative literature review of 18 approaches spanning requirements engineering (RE), machine learning (ML) project management, and automation. We organize these approaches into a taxonomy of four families and compare them across six input artifact categories, six output artifact categories, and a transformation framework of seven stages, grounded in RE refinement theory and ML lifecycle process. Our study shows that most approaches list ML task or algorithm specification among their expected outputs, yet only four provide partial guidance for deriving it, and none provides systematic guidance. We characterize this gap as the Analytics Translation Problem (ATP) and derive five research recommendations addressing multi-formulation exploration, task derivation guidance, constraint-algorithm filtering, probabilistic traceability, and data-triggered revision.

    Source: arxiv.org/abs/2604.18770

11. From Community Forums to Issue Trackers: A Moodle Case Study

    Pragyan K C, Mitra Bokaei Hosseini

    arXiv (cs.SE), to appear at IEEE RE 2026 · accepted

    Traces how feature requests travel from Moodle's community forum to its Jira tracker, combining link analysis with interviews of developers and forum participants. Cross-channel traceability turns out to be rare, about 3.5% of feature request issues link back to a forum thread, and the hand-off is largely ad hoc, with unclear role ownership and little tool support.

    Abstract

    Sustaining open-source software (OSS) requires effective practices for evolution and change management. In OSS projects, evolution is largely driven by feature requests and enhancements proposed by diverse stakeholders. These requests are often discussed across multiple communication channels, particularly community forums and issue trackers, where stake-holders negotiate intent, clarify requirements, and coordinate development. Despite prior research on OSS forums and issue trackers, we lack an empirical understanding of who creates and maintains links between forum posts and tracker issues, and how these links support clarification, feedback, and coordination throughout feature request lifecycles. To address these questions, we conduct an in-depth case study of Moodle, a widely used open-source learning management system. Our study combines (1) an empirical analysis of cross-channel trace links between Moodle's community forum and its Jira issue tracker, (2) semi-structured interviews with developers, and (3) semi-structured interviews with forum participants. Our results show that cross-channel traceability is rare: only 818 of 23,169 (~3.5%) feature request issues in Moodle's Jira link back to a community forum, and authorship differs by channel, with developers authoring 52.8% of tracker issues, while forum feature requests are predominantly authored by users, and only 230 linked pairs share the same author. The qualitative findings further reveal that the transition from forum posts to issues is largely ad hoc, with limited tool support and unclear role ownership, and that users often experience the process as opaque or weakly responsive.

    Source: arxiv.org/abs/2606.20876

12. From Failure to Alignment: A Requirements Engineering Framework for Machine Learning Systems

    Amel Bennaceur, Gopi Rajbahadur, Prince Mercy, Bashar Nuseibeh, Faeq Alrimawi

    arXiv (cs.SE), to appear at IEEE RE 2026 · accepted

    REAL, a model-based framework that weaves requirements for data, models and the whole system together, uses failure to drive the search for alternative requirements, and refines them iteratively and traceably. Demonstrated on an autonomous driving example, with a replication package.

    Abstract

    Organisations designing, developing, and deploying machine learning systems (MLS) need to be able to check that these systems are trustworthy, and communicate this clearly to their stakeholders, be they different categories of users, engineers, or wider society. By focusing on stakeholders, Requirements Engineering is well positioned to drive the design and engineering of MLS that align with the needs of their stakeholders. Yet, we still need a systematic process for modelling and reasoning about requirements for MLS that is driven both by stakeholders' needs and constraints for MLS development. This paper proposes a framework entitled REAL (Requirements Engineering for mAchines that Learn - and Fail) to help develop MLS that align with stakeholders' needs by adopting a requirements engineering approach. This model-based framework is based on three principles. First, weaving together requirements for data, models, and the system as a whole. Second, using failure to drive the exploration of alternative requirements. Third, iterative and traceable refinement of MLS requirements. We demonstrate the proposed framework using an example from autonomous driving and show that REAL supports the development of MLS that better align with stakeholders' requirements. A replication package is available online.

    Source: arxiv.org/abs/2606.31589

13. From Quality Properties to Practice: A Guideline and Workflow for Explainability Requirements

    Martin Obaidi, Jakob Droste, Hannah Deters, Marc Herrmann, Michel Krahl, Kurt Schneider

    arXiv (cs.SE), to appear at IEEE RE 2026 · accepted

    Derives ten core quality properties for explainability requirements from a literature review, developer interviews and a practitioner survey, turns them into a guideline with formulation instructions, and operationalises the guideline in a drafting, checking and revision tool. Formulation time fell 23.5% with tool support, while perceived quality did not differ from manually written requirements.

    Abstract

    Explainability is increasingly required in AI-enabled software systems to support transparency, user trust, and compliance. Yet, explainability requirements are often written ad hoc, and unguided large language model support can yield vague, inconsistent, or incomplete statements. This paper presents a sequential, guideline-driven workflow for formulating explainability requirements and evaluates its tool-based operationalization. We first elicited candidate quality properties through a structured literature review and developer interviews. We then prioritized these properties in an online survey with practitioners (n = 20) and derived a concise guideline of ten core properties with actionable formulation instructions. Next, we operationalized the guideline in a web-based tool that supports an iterative workflow of drafting, property-based checks, and revision. We evaluated the workflow in two complementary studies. In a task-based study with requirements engineers (n = 6), formulation time was 23.5% lower with tool support (mixed-effects model p = 0.049, Wilcoxon sensitivity analysis p = 0.021). In an independent online study with software developers (n = 18), tool-supported and manually written requirements did not differ significantly in implementability or formulation quality, with a descriptive slight preference tendency toward the tool-supported versions. Overall, our results suggest that combining a prioritized quality guideline with lightweight LLM support can reduce formulation effort without significant differences in perceived quality from manually written requirements.

    Source: arxiv.org/abs/2606.10882

14. HM-Req: A Framework for Embedding Values within CPS Human Monitoring Requirements

    Zoe Pfister, Ruth Breu, Michael Vierhauser

    arXiv (cs.SE), to appear at IEEE RE 2026 · accepted

    An elicitation framework with a controlled natural language for human monitoring requirements in cyber-physical systems, which augments each requirement with the human values of the stakeholders it touches and surfaces conflicts between those values in a dashboard, so that implicit and competing values become something a team can discuss rather than something buried in vague prose.

    Abstract

    Monitoring humans, for example, their movement or location, is essential for safe and efficient human-machine collaboration in Cyber-Physical Systems (CPS). This information allows CPS to ensure safety properties, adapt their behaviour dynamically, and coordinate with humans. To ensure that the design of a CPS respects ethical principles and the privacy of its stakeholders, system requirements, particularly those related to human monitoring, must reflect the human values of all involved stakeholders. However, human values are often underrepresented in Software Engineering -- particularly during requirements elicitation and system design, crucial phases when introducing ethically critical functionality. Stakeholder values are often implicit and conflicting, yet rarely systematically captured. Furthermore, unstructured natural language requirements introduce ambiguity and vagueness, complicating conflict resolution. To address these problems, we propose HM-Req, a requirements elicitation framework including a Controlled Natural Language (CNL) for defining human monitoring requirements. These requirements are then augmented with human values from relevant stakeholders and integrated into a Value Dashboard to detect potential conflicts that require further discussion and resolution. Validation results, applying the CNL to different datasets and conducting a survey and expert interview, provide evidence of the CNL's ability to capture diverse human monitoring requirements and demonstrate HM-Req's usefulness for requirements elicitation activities.

    Source: arxiv.org/abs/2605.12100

15. How Requirements Quality Makes (or Breaks) Traceability Link Recovery

    Tobias Hey, Julian Frattini

    arXiv (cs.SE), to appear at IEEE RE 2026 · accepted

    Annotates 28 types of quality defect across 189 use case descriptions and runs five traceability link recovery approaches over them to measure what each defect does to recovery performance. Some defects hurt, some help, and different approaches respond differently, so the best approach depends on the quality of the dataset it is pointed at.

    Abstract

    Traceability information between requirements and source code greatly benefits the maintenance of a software system. Since manually establishing trace links is cumbersome and error-prone, previous research explored automated traceability link recovery (TLR) approaches to support this task. However, quality defects in requirements impact subsequent activities such as TLR, yet evidence about this remains scarce. Our objective is to contribute empirical evidence on this impact. At the same time, we aim to understand how the performance of TLR approaches varies given these quality defects. To this end, we annotated 28 types of quality defect in 189 use case descriptions from two datasets. Then, we executed five distinct TLR approaches on the dataset and measured their performance in recovering trace links. Finally, we performed statistical tests to quantify the defects' effect strength on this performance. Our results show that some quality defects harm TLR performance, e.g., sentences that do not start with noun phrases, while others actually benefit performance, e.g., use cases that include implementation details. Moreover, different types of approaches respond differently to these defects. As a consequence, the performance-optimizing choice of a TLR approach depends on the quality of the dataset.

    Source: arxiv.org/abs/2606.11834

16. Writing Better Software Explanations: A Guideline-Based Approach

    Martin Obaidi, Jean-Carl Kremser, Hannah Deters, Jakob Droste, Marc Herrmann, Kurt Schneider

    arXiv (cs.SE), to appear at IEEE RE 2026 · accepted

    A tool that pairs LLM-assisted drafting of user-facing software explanations with an empirically derived quality guideline, structuring the work into generation, quality checking and revision while leaving domain control with the developer. Tool-supported writing was 24.4% faster in a six-practitioner experiment, and its explanations were preferred in 204 paired comparisons.

    Abstract

    As software systems increasingly rely on natural-language explanations to address user-reported explanation needs in requirements communication and support, ensuring that such explanations are consistent, relevant, and well formulated remains a major challenge. Purely automatic large language model (LLM) generation often lacks reliable grounding and controllable output quality. In this paper, we present a guideline-based formulation support tool for software explanations that combines LLM-assisted text generation with an empirically derived quality guideline. The tool structures the writing process into generation, quality checking, and iterative revision, while keeping domain control with developers. We evaluated the approach in a two-phase study consisting of an interview-based developer experiment and a controlled user survey. Six industry practitioners with software development or DevOps experience formulated explanations for real explanation needs in a human-only manual condition and in a human-with-LLM-support condition. In this small-scale evaluation, tool-supported formulation was on average 24.4% faster, although inferential analyses indicated only a trend for efficiency. In a subsequent user study with 17 participants and 204 paired comparisons, tool-supported explanations were rated significantly higher in overall satisfaction than manual explanations (p=0.003, rank-biserial correlation=0.86). Our findings suggest potential efficiency gains and higher perceived formulation quality through guideline-driven LLM assistance. Future work should examine long-term industrial use and integration into existing development workflows.

    Source: arxiv.org/abs/2606.10880

17. Agentic Persona Generation with Critique-Refinement: An Industrial Evaluation

    Mohammad Hossein Amini, David Dewar, Shiva Nejati, Mehrdad Sabetzadeh

    arXiv (cs.SE), to appear at IEEE RE 2026, industrial innovation track · accepted

    PerGent replaces single-shot persona generation with a generator agent, a critic agent and an orchestrator running a bounded critique-refinement loop over interviews, surveys and job postings. Deployed at Kinaxis, it reached the highest expert approval rate of the four methods compared, and reproduced more of the hand-written pre-LLM personas while adding substantial new content.

    Abstract

    Personas are widely used in software engineering to support requirements elicitation, design, and validation, but their manual creation is costly, time-consuming, and hard to scale. Recent LLM-based approaches automate persona generation from textual data; however, they typically rely on single-shot generation and subjective evaluations, limiting practical reliability. We present PerGent, an industry-grade method for persona generation built around an iterative critique-refinement loop. Specifically, PerGent uses a generator and a critic LLM agent, coordinated by an orchestrator, to iteratively refine personas using external resources such as interviews, surveys, and job postings through a critique-refinement loop with a user-defined maximum number of rounds. We deploy and evaluate PerGent in an industrial setting at Kinaxis, comparing it with three baselines, including one-shot methods. In an expert in-situ evaluation, PerGent achieved the highest expert approval rate (96.9%), exceeding all baselines. We further compare PerGent-generated personas with best-practice personas manually created by domain experts prior to the adoption of LLMs. Compared to baselines, PerGent reproduces a larger proportion of expert content while also contributing substantial new content beyond the pre-LLM personas. We conclude with lessons learned from deploying and evaluating PerGent at Kinaxis.

    Source: arxiv.org/abs/2606.09637

18. From Regulation to Requirements: An Automated Requirement Derivation and Explanation Pipeline

    Pavithra Nair, Preethu Rose Anish

    arXiv (cs.SE), to appear at IEEE RE 2026, industrial innovation track · accepted

    A pipeline that finds requirement-bearing clauses in regulatory text and derives system-agnostic requirements from them, each with a plain-language explanation traceable back to its legal source, evaluated over the full GDPR and EU AI Act clause sets and released as the Reg2Req tool. In a study with 25 practitioners the explanations significantly improved both comprehension and confidence to act.

    Abstract

    Ensuring software compliance with regulations such as the General Data Protection Regulation (GDPR) and the Artificial Intelligence Act (EU AI Act) poses a significant challenge, as requirements engineers must translate complex legal text into actionable software requirements - a process that remains largely manual and error-prone in practice. We present an automated regulation-to-requirements pipeline that identifies requirement-bearing clauses in regulatory documents and derives system-agnostic software requirements, accompanied by plain-language explanations, traceable to their legal sources. We evaluate the pipeline on the full clause sets of the GDPR (398 clauses) and the EU AI Act (574 clauses). For requirement-bearing clause identification, the approach achieves macro-averaged F1 scores of 0.82 and 0.78, respectively, outperforming a SetFit-based baseline. Human evaluation shows high completeness (4.60 and 4.45) and correctness (3.74 and 3.54) of derived requirements, while explanation clarity scores are near-ceiling (4.92 and 4.94) on a 1-5 scale. We implement the approach in Reg2Req, a publicly released tool that further supports requirement classification, use case seeding, cross-reference analysis, definition indexing, and a traceability matrix to operationalize regulatory compliance in practice. A user study with 25 practitioners shows that the plain-language explanations significantly improve comprehension of derived requirements and confidence in acting on them (p < 0.001), and that all participants would use Reg2Req as a starting point for deriving software requirements from a regulation.

    Source: arxiv.org/abs/2607.04448

19. Supporting the Adoption of Privacy-Enhancing Technologies through Requirements Engineering

    Oleksandr Kosenkov, Vadym Honcharenko, Abhinava Pratap Singh, Volodymyr Spirin, Danica Vranjanin

    arXiv (cs.SE), to appear at IEEE RE 2026, industrial innovation track · accepted

    An industrial challenge paper arguing that privacy-enhancing technologies stall in practice because the engineering, legal and business viewpoints on them are handled separately. It examines the adoption barriers across PET developers, integrators and adopters, and makes the case that specifying those three viewpoints explicitly in requirements is what lets the stakeholders coordinate.

    Abstract

    In recent decades, privacy-enhancing technologies (PETs) have been recognized as a means of meeting regulatory and user privacy requirements in software systems that process personal data. Despite substantial research efforts, support from regulators, contributions by large technology companies such as Google and Microsoft, and growing interest among software practitioners, the practical adoption of PETs remains limited. Existing research consistently identifies recurring challenges to PETs adoption in SE, such as technical complexity and insufficient training. Despite ongoing research efforts, these challenges largely remain unresolved in practice. In this industrial challenge paper, we apply a practical, requirements engineering (RE)-driven perspective to examine challenges to PET adoption across multiple stakeholder groups (PET developers, integrators, and adopters) as well as across different disciplinary perspectives (engineering, law, and business). We argue that RE can facilitate the adoption of PETs by systematically addressing each of the complementary engineering, business, and legal viewpoints on privacy. Neglecting challenges in any of these viewpoints (e.g., the impact of PETs on software architecture, their business implications, and their contribution to regulatory compliance) can increase the impediments or even lead to implementation failure. In practice, explicit specification of these viewpoints within RE can enable meaningful coordination among stakeholders to more effectively realize the benefits of PETs in software engineering.

    Source: arxiv.org/abs/2606.17387

20. Automated Alignment between Elicitation Interviews and Requirements

    Francesco Dente, Fabiano Dalpiaz, Paolo Papotti

    arXiv (cs.CL), to appear at RE@Next! 2026 · accepted

    Formalises the alignment of an interview transcript with a set of user stories through two metrics, requirements faithfulness and interview coverage, and tests whether language and embedding models can compute them. An LLM-based solution reaches 0.86 macro-F1 on manually labelled chunk-story pairs, with embedding models used as blockers to keep it scalable.

    Abstract

    Software requirements are derived from a variety of elicitation techniques, many of which have a conversational nature, like interviews. However, evaluating whether those derived requirements faithfully reflect the stakeholders' needs remains a challenging manual task. In this paper, we formalize the task of aligning the transcript of an interview with a collection of requirements represented as user stories. We propose two heuristic metrics for alignment, called (i) requirements faithfulness: the proportion of stories supported by the transcript, and (ii) interview coverage: the proportion of transcript supported by at least one story. Then, we run experiments with large language models and embedding models that assess the ability of evaluating these metrics automatically. Experiments over four datasets show that an LLM-based solution achieves 0.86 macro-F1 on manually labeled chunk-story pairs. We also show how embedding models can be used as blockers to make the approach more scalable. This work paves the way for more research on linking conversational artifacts with requirements. The formal framework and the automated matching techniques are basic components that can be used for emerging tasks such as tracing requirements to interviews and generating requirements from conversations.

    Source: arxiv.org/abs/2510.08622

21. Information is all you need: Requirements Engineering Quality Reframed

    Henning Femmer, Julian Frattini

    arXiv (cs.SE), to appear at RE@Next! 2026 · accepted

    A vision paper that models requirements engineering as the transfer of information particles, discrete pieces of domain knowledge, between roles and artefacts, and makes quality a property of that flow rather than of an artefact or a process alone. A simulation shows why a high-quality specification gets bypassed in an agile setting.

    Abstract

    To move beyond this vague appeal to context, this vision proposes a novel holistic theory of requirements engineering (RE) quality. This theory models how information particles, i.e., discrete pieces of domain knowledge, are transferred between roles and artifacts. Since the RE process is ultimately an information transfer, holistic RE quality depends on the properties of information flow, i.e., how effectively and efficiently information is transferred from sources (like stakeholders) to targets (like developers and testers), uniting both artifact- and process-based perspectives on RE quality. In an exemplary simulation of the theory we illustrate why a high-quality specification gets bypassed in an agile context, thereby demonstrating that a simulation can provide actionable insights into calibrating the RE process to optimize the information flow. Beyond organizational applications, we envision that the theory can serve as a coherent theoretical framework for understanding the success or failure of RE processes and artifacts.

    Source: arxiv.org/abs/2607.21319

22. Transforming Privacy Artifacts into Accessible Reports for Non-Technical Stakeholders

    Zoe Pfister, Clemens Sauerwein, Benedikt Dornauer, Tina Mersch, Christian Wolf, Ruth Breu, Michael Vierhauser

    arXiv (cs.SE), to appear at RE@Next! 2026 · accepted

    A framework that carries human monitoring use cases and requirements through to privacy reports written for the workers and unions affected, using language models to turn the technical artefacts into something a non-technical reader can act on. Initial insights come from two industrial use cases, with the generated reports evaluated for quality.

    Abstract

    The transition toward Industry 5.0 is reshaping industrial work environments with an emphasis on human-centricity, enabling close collaboration between humans and machines to enhance productivity and flexibility. However, such systems typically require monitoring of human workers and operators, often involving sensitive data, raising significant privacy concerns. As a result, affected workers and unions frequently reject human-machine collaboration features due to a lack of transparency regarding privacy threats and implemented mitigation strategies. To enable early stakeholder involvement, establish trust, and support informed decision-making, privacy implications must be communicated in a way understandable to non-technical stakeholders. Yet, current Requirements Engineering (RE) practices provide limited methodological support for making privacy threats and mitigations accessible to non-technical stakeholders (e.g., individual workers or their representative unions). In this paper, we propose a conceptual framework that guides software design from human monitoring-related use cases and requirements to informed decision-making guidance focusing on non-technical stakeholders. Building on principles such as Privacy by Design, the framework leverages Large Language Models (LLMs) to transform technical artifacts into accessible privacy reports. We share initial insights from two industry use cases, evaluate the quality of the generated reports, and outline future research directions toward integrating privacy transparency into RE processes for human-centric industrial systems.

    Source: arxiv.org/abs/2605.21269

23. Classification of quality characteristics in online user feedback using linguistic analysis, crowdsourcing and LLMs

    Eduard C. Groen, Fabiano Dalpiaz, Martijn van Vliet, Boris Winter, Joerg Doerr, Sjaak Brinkkemper

    arXiv (cs.SE), Journal of Systems and Software, journal-first at IEEE RE 2026 · accepted

    Compares three ways of classifying software quality characteristics in online user feedback when there is no training corpus to learn from: keyword-driven language patterns, crowdsourced micro-tasks, and LLM prompts. Crowdsourcing and the LLMs land in the same accuracy range and both beat the language patterns, which suggests either could be used to build the missing training corpora.

    Abstract

    Software qualities such as usability or reliability are among the strongest determinants of mobile app user satisfaction and constitute a significant portion of online user feedback on software products, making it a valuable source of quality-related feedback to guide the development process. The abundance of online user feedback warrants the automated identification of quality characteristics, but the online user feedback's heterogeneity and the lack of appropriate training corpora limit the applicability of supervised machine learning. We therefore investigate the viability of three approaches that could be effective in low-data settings: language patterns (LPs) based on quality-related keywords, instructions for crowdsourced micro-tasks, and large language model (LLM) prompts. We determined the feasibility of each approach and then compared their accuracy. For the complex multiclass classification of quality characteristics, the LP-based approach achieved a varied precision (0.38-0.92) depending on the quality characteristic, and low recall; crowdsourcing achieved the best average accuracy in two consecutive phases (0.63, 0.72), which could be matched by the best-performing LLM condition (0.66) and a prediction based on the LLMs' majority vote (0.68). Our findings show that in this low-data setting, the two approaches that use crowdsourcing or LLMs instead of involving experts achieve accurate classifications, while the LP-based approach has only limited potential. The promise of crowdsourcing and LLMs in this context might even extend to building training corpora.

    Source: arxiv.org/abs/2506.11722

24. Exploring Ethical Concerns of Mobile Applications from App Reviews: A Literature Survey

    Aakash Sorathiya, Gouri Ginde

    arXiv (cs.SE), journal-first at IEEE RE 2026 · accepted

    Surveys 37 studies published since 2012 on mining app reviews for users' ethical concerns, and reports the objectives, methods and extra sources such as privacy policies that researchers bring to the task. Users keep reporting the same barriers around privacy, security, accessibility, transparency, fairness, accountability and safety, and the survey closes with a four-part research agenda.

    Abstract

    Privacy, security, and accessibility, like ethical concerns in mobile applications (a.k.a. apps), commonly subsumed under non-functional requirements, are generally reported by users through app reviews available in app stores. However, these remain unidentified among other types of reviews, such as user experiences, problem reports, and new feature discussions. Over the past decade, extensive research has focused on extracting valuable information from app reviews, including feature requests and bug reports. However, there remains a lack of a synthesis of research related to app review analysis for exploring users' ethical concerns. This paper presents a comprehensive survey of this research area, covering 37 relevant studies published since 2012, identified from the initial 553 studies using specific inclusion and exclusion criteria. The studies examined vary in review counts, ranging from 500 to 626 million, and include between a single and 1.3 million apps. Our detailed analysis highlights diverse objectives, methodologies, and strategies, along with additional resources such as app privacy policies, which researchers generally utilize to analyze ethical concerns. Our findings also identify persistent barriers to privacy, security, accessibility, transparency, fairness, accountability, and safety, as reported by users in app reviews. Furthermore, we propose a research agenda that focuses on four key areas, including automated extraction and classification of ethical concerns-related app reviews. Our survey outcomes can assist developers and system architects in recognizing and prioritizing non-functional requirements at the initial stages of the development lifecycle, whereas researchers can expand upon this synthesis to create tools for the automated detection of ethical concerns.

    Source: arxiv.org/abs/2604.16770

25. From Law to Gherkin: A Human-Centred Quasi-Experiment on the Quality of LLM-Generated Behavioural Specifications from Food-Safety Regulations

    Shabnam Hassani, Mehrdad Sabetzadeh, Daniel Amyot

    arXiv (cs.SE), Information and Software Technology, journal-first at IEEE RE 2026 · accepted

    A quasi-experiment in which ten participants rated 60 Gherkin specifications generated from food-safety regulations by two language models, across relevance, clarity, completeness, singularity and time saved. Ratings were uniformly high, with no reliable difference between the models, but the qualitative feedback found omissions, hallucinations and mixed intents that make human review non-optional.

    Abstract

    Context: Laws and regulations increasingly shape software design, development, and quality assurance in regulated domains. Because legal provisions are written in technology-neutral language, deriving concrete specifications, requirements, and acceptance criteria to verify software compliance is difficult and error-prone. Recent advances in generative AI, especially large language models (LLMs), may help automate this process. Objective: We present the first systematic human-subject evaluation of LLMs' ability to derive Gherkin behavioural specifications from legal texts using a quasi-experimental design. Gherkin is a domain-specific language for scenario-based system behaviour descriptions in Given-When-Then form and is well suited to automation in software development. Methods: Ten participants evaluated 60 Gherkin specifications generated from food-safety regulations by Claude and Llama. Each participant assessed 12 specifications across five criteria: relevance, clarity, completeness, singularity, and time savings. Each specification was evaluated by two participants, yielding 120 assessments with quantitative ratings and qualitative feedback. Results: Ratings were uniformly high in the top two categories: relevance 95%, clarity 100%, completeness 94.2%, singularity 93.4%, and time savings 91.7%. No statistically reliable differences were found across participants or between LLMs. Qualitative feedback noted occasional omissions, hallucinations, and mixed intents, underscoring the need for human oversight, especially in safety-critical domains. Conclusion: In food safety, LLMs can assist in deriving Gherkin specifications from legal texts, but omissions and hallucinations require systematic human review.

    Source: arxiv.org/abs/2508.20744

26. Generative AI for Requirements Engineering: A Systematic Literature Review

    Haowei Cheng, Jati H. Husen, Yijun Lu, Teeradaj Racharak, Nobukazu Yoshioka, Naoyasu Ubayashi, Hironori Washizaki

    arXiv (cs.SE), journal-first at IEEE RE 2026 · accepted

    Reviews 238 articles published between 2019 and 2025 on generative AI in requirements engineering. Attention is uneven across the phases, analysis and elicitation take most of it while management is left largely alone, and reproducibility, hallucination and interpretability recur together often enough that the authors argue they have to be tackled as one problem rather than three.

    Abstract

    Introduction: Requirements engineering faces challenges due to the handling of increasingly complex software systems. These challenges can be addressed using generative AI. Given that GenAI based RE has not been systematically analyzed in detail, this review examines related research, focusing on trends, methodologies, challenges, and future directions. Methods: A systematic methodology for paper selection, data extraction, and feature analysis is used to comprehensively review 238 articles published from 2019 to 2025 and available from major academic databases. Results: Generative pretrained transformer models dominate current applications (67.3%), but research remains unevenly distributed across RE phases, with analysis (30.0%) and elicitation (22.1%) receiving the most attention, and management (6.8%) underexplored. Three core challenges: reproducibility (66.8%), hallucinations (63.4%), and interpretability (57.1%) form a tightly interlinked triad affecting trust and consistency. Strong correlations (35% cooccurrence) indicate these challenges must be addressed holistically. Industrial adoption remains nascent, with over 90% of studies corresponding to early stage development and only 1.3% reaching production level integration. Conclusions: Evaluation practices show maturity gaps, limited tool and dataset availability, and fragmented benchmarking approaches. Despite the transformative potential of GenAI based RE, several barriers hinder practical adoption. The strong correlations among core challenges demand specialized architectures targeting interdependencies rather than isolated solutions. The limited deployment reflects systemic bottlenecks in generalizability, data quality, and scalable evaluation methods. Successful adoption requires coordinated development across technical robustness, methodological maturity, and governance integration.

    Source: arxiv.org/abs/2409.06741

27. Decoupling in AI Ethics: Learning how to Walk the Talk

    Ravit Dotan, Tomer Gershoni, Irit Hadar, Gil Luria

    Empirical Software Engineering 31(5), journal-first at IEEE RE 2026 · published

    Interviews with 32 practitioners, from early-stage startups to large corporations, on the gap between what companies declare about AI ethics and what they actually operate, read through the organizational psychology notion of decoupling. The paper characterises the common declaration types, operation types and rationales, derives distinct decoupling profiles, and gives adoption recommendations tailored to each profile.

    Abstract

    In recent years, AI ethics declarations, commitments, and frameworks for AI systems development have proliferated. Yet, implementation remains persistently low. This phenomenon, often termed “AI ethics washing,” has been widely criticized but lacks empirical investigation. Our paper examines these gaps between declarations and operations in AI ethics through the organizational psychology concept of “decoupling”—the disconnect between what organizations say and what they do. Using data collected through in-depth interviews with 32 practitioners across diverse companies, from early-stage startups to large corporations, we present a systematic analysis of decoupling between declarations and operations in AI ethics, producing the first analysis of decoupling not only in AI ethics but in any technology development field. Our findings identify and characterize (i) common types of AI ethics declarations, such as policies and internal communications, (ii) common types of AI ethics operations, such as reviews and testing, (iii) common rationales behind companies’ approaches to AI ethics, and (iv) distinct decoupling profiles, i.e., common ways in which AI ethics declarations come apart from operations. Our discussion includes recommendations for increasing AI ethics adoption tailored to each profile. These recommendations differ from traditional AI ethics frameworks. While traditional frameworks prescribe ideal practices based on regulatory or industry expectations, this paper offers recommendations grounded in an empirical analysis of how AI ethics efforts succeed or fail in practice. Our decoupling-informed perspective fundamentally reshapes how practitioners and scholars can approach the challenge of AI ethics implementation.

    Source: doi.org/10.1007/s10664-026-10861-z · doi:10.1007/s10664-026-10861-z