Computer-assisted transformation of design documents from a natural language description to structured modeling languages

Lei ChenSpectrum 2008

Abstract

This research addresses the challenge of translating ambiguous natural language (NL) design requirements into structured modeling languages like UML and FBS. By introducing the Recursive Object Model (ROM) as an intermediate semantic representation, the proposed framework automates the extraction of technical specifications from customer descriptions. Software prototypes R2U and R2FBS demonstrate the feasibility of this approach across various engineering domains, including software and mechanical engineering.

Introduction

Engineering design typically begins with the "voice of the customer," captured in unrestricted natural language. However, the inherent ambiguity and lack of structure in NL pose significant challenges for downstream technical modeling. Traditionally, engineers manually translate these requirements into formal diagrams—a process that is subjective, time-consuming, and prone to error. This paper presents a systematic, computer-assisted approach to bridge this gap using a two-step transformation process that reduces human bias and increases efficiency.

The Recursive Object Model (ROM) Bridge

The core of the methodology is the Recursive Object Model (ROM). Unlike standard grammatical parsers that focus on syntax, ROM treats every word in a text as an "object" and maps the semantic relations between them. These relations are categorized into three types:

  1. Constraint: Defining properties or limitations.
  2. Connection: Establishing links between objects.
  3. Predicate: Describing actions or states.

By converting NL text into a ROM diagram—a mathematical, directed graph—the system captures the underlying semantic information necessary for technical modeling. This intermediate step acts as a semantic "bridge," allowing for a more rigorous mapping to structured languages.

Automated Transformation: R2U and R2FBS

Two software prototypes were developed to operationalize the transformation from ROM to specific modeling languages:

R2U (ROM to UML)

R2U focuses on generating Unified Modeling Language (UML) diagrams, primarily Use Case and Class diagrams. The system employs semantic mining to identify the "Product" (the central node with the highest connectivity) and uses heuristics to distinguish between external actors and internal system classes. Verbs in the ROM diagram are mapped to methods, while nouns and their constraints form the basis of the class hierarchy.

R2FBS (ROM to FBS)

R2FBS transforms ROM diagrams into Function-Behavior-State (FBS) models, commonly used in mechanical and systems engineering. The tool simulates human analytic logic to decompose high-level functions into specific behaviors and physical states. Prepositions are particularly critical here, often serving as indicators of state transitions or environmental interactions.

Case Study Validation

The proposed approach was validated through several engineering case studies:

  • Point-of-Sale (POS) System: Successfully generated UML diagrams identifying actors like "Customer" and "Cashier" and their associated actions.
  • Mechanical Riveting Tool: Demonstrated the ability to handle technical descriptions of physical tools for brake linings.
  • Auto-Coating System: Decomposed complex machine requirements into a hierarchical FBS model involving components like conveyor belts and extruders.
  • Design Patents: The system proved exceptionally effective at processing patent language (e.g., US Patent 7,377,052 B2), which is naturally more restricted and precise than general requirements documents.

Conclusion

The research demonstrates that using an intermediate semantic representation like ROM can significantly reduce the subjectivity and effort involved in requirements modeling. While the current system is dependent on the accuracy of the initial ROM analysis and focuses on static diagrams, it provides a robust foundation for future work in generating dynamic models like Sequence or Activity diagrams. Ultimately, these tools serve to assist engineers by providing a consistent, automated baseline for technical design.

Summary updated

Abstract

In the present thesis, a novel approach is proposed to transform design documents described by a natural language into a structured modeling languages, particularly UML diagrams and FBS models. The transformation consists of two steps: (i) From natural language to an intermediate graphic language called Recursive Object Model (ROM). (ii) From a ROM diagram to a modeling language. The ROM diagram corresponding to a text includes the main semantic information implied in the text by modeling the relations between the words in a text. Based on the semantics implied in the ROM diagram, a set of criteria is proposed to mine the semantic meaning of the original text corresponding to the ROM diagram. Once the semantic meaning of the design documents through their corresponding ROM diagram is captured, a set of mapping rules from the ROM diagram criteria to the modeling language elements is proposed. After that, a set of generation rules to explore the relationship between these elements is proposed to generate UML diagrams and FBS models based on a ROM diagram. A software prototype R2U is presented as a proof of concept for transforming ROM diagrams to UML diagrams. Another software prototype R2FBS is also presented as a proof of concept for transforming ROM diagrams to FBS models. Several case studies show that the proposed approach is feasible. The proposed approach can be applied to requirements modeling in various engineering fields such as software engineering, automotive engineering, and aerospace engineering. Future work is indicated at the end of the present thesis
BibTeX
@inproceedings{chen2008computer,
  author = {Lei Chen},
  title = {{Computer-assisted transformation of design documents from a natural language description to structured modeling languages}},
  booktitle = {Spectrum},
  year = {2008},
}
Figure