NS-RAD: A Neuro-Symbolic Framework for Conflict and Ambiguity Detection in Natural-Language Requirements

Abstract

High-quality software requirements are critical for delivering dependable systems on time and within budget, yet naturallanguage Software Requirements Specifications (SRS) frequently introduce defects, rework, and project delays. We introduce NS-RAD, a multi-layer neuro-symbolic feedback-driven framework for automated detection of two common requirement anomalies: (i) conflicts, where statements cannot be simultaneously satisfied, and (ii) ambiguities, where vague or underspecified language allows for divergent interpretation. It consists of four parts: requirement ingestion and normalization, multi-view semantic and linguistic encoding, dual analysis for conflict and ambiguity detection, and a feedback-driven adaptation layer. We show that NS-RAD achieves high accuracy, cross-dataset generalization, and practical utility on several public SRS datasets. The framework provides detection performance on par with representative recent approaches, calibrated risk assessment, explicit evidence traces, and actionable guidance for analysts. The results show that combining transformer-based learning with linguistically informed rules in a feedback-driven architecture is a powerful and scalable approach to enhance clarity, consistency, and reliability of industrial requirements engineering. 

Received: 1 May 2026 | Revised: 29 July 2026 | Accepted: 30 August 2026

Conflicts of Interest 

The authors declare that they have no conflicts of interest to this work. 

Data Availability Statement 

The data that support the findings of this study are openly available in UAV at https://www.kaggle.com/datasets/ziya07/uav-autonomous-navigation-dataset, in NLP4RE at https://data.mendeley.com/datasets/zs9vkk7sb7/1, in OpenCOSS at https://github.com/open-compass/opencompass, and in WordlVista at https://worldvista.org/.

Author Contribution Statement 

Anis Zarrad: Conceptualization, Methodology, Formal analysis, Data curation, Supervision, Project administration. Jaber Jemai: Conceptualization, Validation, Investigation, Resources, Writing – review & editing, Visualization, Supervision. Danyal Abbasi: Software, Data curation, Writing – original draft, Visualization. Yassine Daadaa: Validation, Resources.