Natural Language Processing in Requirements Engineering

Natural Language Processing (NLP) has emerged as a transformative tool within Requirements Engineering (RE), enabling automated parsing, classification, validation and transformation of textual specifications into machine-interpretable representations. Requirements—often expressed in natural language—pose challenges including ambiguity, incompleteness and inconsistency. NLP techniques span rule-based parsing, statistical learning and deep learning to extract entities, relationships and semantic patterns. In classification tasks, NLP models distinguish functional from non-functional requirements and further identify subcategories such as performance or security constraints. Advances in transformer-based architectures have raised classification accuracy and generalisation across domains. Beyond classification, NLP supports automated traceability link recovery, defect detection in user stories, and the generation of systems modelling artefacts such as SysML diagrams. Recent work also explores low-resource approaches including zero-shot learning to overcome labelled data scarcity. Integration of NLP with graph-based methods leverages syntactic structures and dependency parsing to enhance understanding of complex requirement statements. These developments address growing demands for scalability and consistency in agile, model-based and safety-critical system engineering contexts worldwide, offering more efficient, reliable and standardised requirement handling.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent efforts have fine-tuned transformer models for domain-specific requirement classification. For example, a bespoke language model was developed to categorise aerospace requirements into design, functional and performance types, demonstrating superior accuracy over generic classifiers while using a modest labelled corpus. Another study introduced a zero-shot learning framework that exploits contextual embeddings and transformer language models to classify functional, non-functional and security-related requirements without any task-specific training data, achieving competitive F1-scores and alleviating the dependency on expensive annotations. Complementary research has combined pre-trained transformers with graph attention networks to capture both semantic and syntactic features of requirement statements. By constructing dependency parse trees and applying attention mechanisms, this approach significantly improved generalisation in both seen and unseen projects, underscoring the value of integrating graph-based relational information with deep contextual embeddings.

Natural Language Processing in Requirements Engineering publication trend

The graph below shows the total number of articles in natural language processing in requirements engineering across all publications each year (not limited to Nature Index journals).

Technical terms

Natural Language Processing (NLP): Computational techniques for analysing and generating human language.

Requirements Engineering (RE): Discipline concerned with the elicitation, specification, validation and management of system requirements.

Functional Requirement: A statement of a service or function that a system must perform.

Non-Functional Requirement: A constraint on system services or functions, often relating to performance, security or usability.

Transformer: A deep learning architecture using self-attention mechanisms to capture contextual relationships in sequential data.

Zero-Shot Learning: An approach that applies models to identify classes not seen during training, by leveraging semantic descriptions or embeddings.

Graph Attention Network (GAT): A neural network that processes graph-structured data by assigning attention weights to nodes and edges.

References

1. Natural Language Processing for systems engineering: Automatic generation of Systems Modelling Language diagrams. Knowledge-Based Systems (2023).
2. aeroBERT-Classifier: Classification of Aerospace Requirements Using BERT. Aerospace (2023).
3. Zero-shot learning for requirements classification: An exploratory study. Information and Software Technology (2023).
4. Automatic Requirements Classification Based on Graph Attention Network. IEEE Access (2022).