Requirement Formalisation Using Natural Language Processing and

Machine Learning: A Systematic Review

Shekoufeh Kolahdouz-Rahimi

1 a

, Kevin Lano

2 b

and Chenghua Lin

3 c

School of Arts, University of Roehampton, London, U.K.

Department of Informatics, King’s College London, London, U.K.

Department of Computer Science, University of Shefﬁeld, U.K.

ﬁ

Keywords:

Requirements Engineering, Requirement Formalisation, Natural Language Processing, Machine Learning,

Deep Learning, Systematic Mapping Study.

Abstract:

Improvement of software development methodologies attracts developers to automatic Requirement Formal-

isation (RF) in the Requirement Engineering (RE) ﬁeld. The potential advantages of applying Natural Lan-

guage Processing (NLP) and Machine Learning (ML) in reducing the ambiguity and incompleteness of re-

quirements written in natural languages are reported in different studies. The goal of this paper is to survey

and classify existing works on NLP and ML for RF, identifying the challenges in this domain and providing

promising future research directions. To achieve this, we conducted a systematic literature review to outline

the current state-of-the-art of NLP and ML techniques in RF by selecting 257 papers from commonly used

libraries. The search result is ﬁltered by deﬁning inclusion and exclusion criteria and 47 relevant studies

between 2012 and 2022 are selected. We found that heuristic NLP approaches are the most common NLP

techniques used for automatic RF, primarily operating on structured and semi-structured data. This study also

revealed that Deep Learning (DL) techniques are not widely used, instead, classical ML techniques are pre-

dominant in the surveyed studies. More importantly, we identiﬁed the difﬁculty of comparing the performance

of different approaches due to the lack of standard benchmark cases for RF.

1 INTRODUCTION

Productive management of Requirement Engineering

(RE) accelerates the process of software development.

Requirement Formalisation (RF) relates to the pro-

cess of transforming requirements in natural language

to speciﬁc formal notations by removing ambiguities.

Formal speciﬁcation of requirements is applicable in

different stages of software development especially in

the validation phase. Manual formalisation of natu-

ral language requirements is an error-prone and time-

consuming task and infeasible for complex systems

(Zaki-Ismail et al., 2021). To this end many auto-

matic and semi-automatic approaches have been in-

troduced to formalise requirements by applying Nat-

ural Language Processing (NLP) techniques (Rolland

and Proix, 1992; Ryan, 1993). Additionally, leverag-

ing Machine Learning (ML) and Deep Learning (DL)

techniques in this domain pushes the research ﬁeld

forward.

https://orcid.org/0000-0002-0566-5429

https://orcid.org/0000-0002-9706-1410

https://orcid.org/0000-0003-3454-2468

In the last decade, there has been a noticeable in-

crease in the number of papers using NLP and ML

techniques for RF. Each research applied a partic-

ular technique and mostly there are no comprehen-

sive guidelines for the reason of applying those tech-

niques. A large number of research reviews have been

carried out to survey this domain (Alzayed and Al-

Hunaiyyan, 2021). To overcome the limitations of

existing studies, in this paper we conducted a system-

atic mapping study of NLP and ML approaches for

RF considering the guidelines presented by Kitchen-

ham and Charters (Brereton et al., 2007), (Keele et al.,

2007), and Petersen et al. (Petersen et al., 2015).

We investigated 47 studies from an initial set of 257.

The papers are selected from commonly used libraries

including ACM Digital Library, IEEE Xplore, Sci-

enceDirect, Springer Link, and Scopus. The search

results are ﬁltered by deﬁning inclusion and exclu-

sion criteria to decide whether a publication found

in the search should be included in the study or ex-

cluded. Three research questions were formulated for

our Systematic Literature Review (SLR). By answer-

ing these research questions, the state of the art of RF

Kolahdouz-Rahimi, S., Lano, K. and Lin, C.

Requirement Formalisation Using Natural Language Processing and Machine Learning: A Systematic Review.

DOI: 10.5220/0011789700003402

In Proceedings of the 11th International Conference on Model-Based Software and Systems Engineering (MODELSWARD 2023), pages 237-244

ISBN: 978-989-758-633-0; ISSN: 2184-4348

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

237

is evaluated. We identify current challenges in the

community and provide guidelines to address those

challenges. The actions that were taken by the author

of this paper as part of MDENet project

are also dis-

cussed. Finally, for further maturity of the research

ﬁeld, future research directions in this area are pro-

vided.

The remainder of this paper is organised as fol-

lows: In Section 2, related review papers in RF using

NLP and ML are summarised. Section 3, presents

background information related to this domain. The

applied research method in this study is described in

Section 4. Section 5 outlines the key ﬁndings of our

study. A discussion of action taken by the authors to

address the challenges is provided in Section 6. Fi-

nally, the conclusion and future direction of research

are provided in Sections 7 and 8.

2 RELATED WORK

A comprehensive survey in the application of NLP to

RE is provided in (Zhao et al., 2020), by considering

404 works. This paper emphasise is on the insufﬁ-

cient application of NLP techniques for RE studies in

industrial cases. Importantly, the lack of expertise in

selecting appropriate NLP techniques in RE domain

is discussed. Although challenges are introduced in

this work, practical solutions are not provided. Addi-

tionally, investigating ML techniques in this domain

is not the main focus of research.

A survey in the application of NLP techniques to

requirements in the form of user stories is provided

in (Raharjana et al., 2021) and the potential advan-

tages of those techniques in RF domain are discussed.

A comprehensive classiﬁcation based on the uses of

NLP techniques for user stories is provided and a

model is recognised as a common target for RF in the

form of user stories. However, ML techniques for RF

are not considered in this research, and in general, the

main challenges of the domain are not sufﬁciently dis-

cussed. Additionally, input datasets for user stories in

investigated papers are not provided.

Yalla and Sharma (Yalla and Sharma, 2015) sur-

vey the current literature that leverages RE and NLP

for different phases of software development. How-

ever, only limited future research directions and

guidelines are provided in this work. Selected arti-

cles that generate UML diagrams by applying NLP

techniques are investigated in (Dawood et al., 2017;

Abdelnabi et al., 2021). These works emphasize the

immaturity of the research area as most of the cur-

https://mde-network.com/

rent processes are not automated. The advantages and

disadvantages of different studies that generate UML

diagrams by applying heuristic rules are identiﬁed in

(Ahmed et al., 2022). This work highlighted the no-

ticeable application of ML in this domain.

The are many related literature studies in the RF

domain. However, the challenges of the domain are

not deeply recognized and guidelines for addressing

those challenges and clear research direction are lim-

ited in most studies. This proves the immaturity of

the research area and its potential for further improve-

ment. Therefore, the main aim of this research is

to identify current challenges in the community by

classifying applied techniques and providing practi-

cal guidelines for addressing those issues.

3 BACKGROUND

In this section, the related concepts to this research

including RE, NLP and DL are explained.

3.0.1 Requirement Formalisation

RE is an important process in software development

for discovering stakeholder’s needs and classifying

them for other phases of software development (Pohl,

2010). Formalising requirement is one of the key

tasks in RE that is automated by applying different

NLP and ML techniques and tools. It enables the

translation of the requirement in natural language into

a structured formal form (e.g., UML modeling dia-

grams). This transformation reduces the ambiguities

of natural language and provides a convenient way for

validation and veriﬁcation (Sch

on et al., 2017; Tukur

et al., 2021).

3.1 Natural Language Processing

NLP is an area of research in Artiﬁcial Intelligence

(AI) that enables a computer to process a large

amount of structured/unstructured data in natural lan-

guage that exist in today’s world. Different NLP

methods, approaches, processes, and procedures are

introduced to process data in different phases of RE.

Tokenization, POS tagging, and dependency parsing

are the commonly used techniques in this domain

(Kulkarni and Shivananda, 2019).

3.2 Machine Learning

ML is one of the core technical terms in Artiﬁcial In-

telligence (AI), which refers to the learning and iden-

tiﬁcation of patterns from examples and existing data

MODELSWARD 2023 - 11th International Conference on Model-Based Software and Systems Engineering

238

Table 1: Terms for Selecting Relevant Research Studies.

Group Term

A Natural Language Processing

Natural Language, NLP

B Machine Learning

Deep Learning

C Requirement Formalisation

Model Generation

UML Generation

OCL Generation

Usecase or Use case Diagram Generation

Class Diagram Generation

Sequence Diagram Generation

ER Diagram Generation

Activity Diagram Generation

(Samuel, 1967). Different algorithms and process-

ing techniques are introduced in this domain. DL is

an ML technique, which is based on Artiﬁcial Neu-

ral Network (ANN) (Goodfellow et al., 2016) and is

applicable in a variety of domains including RF.

4 RESEARCH METHOD

The provided guidelines by Kitchenham and Charters

(Brereton et al., 2007), (Keele et al., 2007), and Pe-

tersen et al. (Petersen et al., 2015) are applied for

systematic mapping study in this research. The fol-

lowing research questions are the main target of this

research.

• Q1: What are the most commonly used NLP/ML

approaches for automatic/semi-automatic RF?

• Q2: What are the input and output of RF ap-

proaches?

• Q3: What are the gaps and deﬁciencies in existing

RF work?

The three phases of the study protocol including

planning, conducting, and reporting are explained in

the following sections.

4.1 Review Planning

The review process and search strategy are explained

in this part. To provide comprehensive coverage of

existing publications most major publishers in Soft-

ware Engineering are investigated.

According to the objectives of this study and re-

search questions, three terms were selected in this pa-

per. Each term includes different keywords and at

least one of the keywords has to be presented in a pa-

per. Table 1 presents the list of selected terms in this

research.

To identify the largest number of studies in the do-

main of RF, the following search string is followed:

Search String= (A ∨ B ∨ (A ∧ B)) ∧ C

Additionally, inclusion and exclusion criteria to

decide which of the selected articles should be con-

sidered as primary studies and which ones should be

excluded are deﬁned as follows:

4.1.1 Inclusion Criteria

• Published between January 2012 and March 2022

• Publications that generate model or any formali-

sation from requirement

• Publications in peer-reviewed journals, confer-

ences, and workshops

• Publication in English

4.1.2 Exclusion Criteria

• Publications not written in English

• Publications before 2012

• Summary, survey, or review publications

• Non peer-reviewed publications

• Publications not focusing on RF

• Books, web sites, technical reports, pamphlets, tu-

torials, duplicate papers, and white papers.

In this research abstracts, titles, and keywords of

papers are evaluated according to the inclusion and

exclusion criteria. Furthermore, in some cases the

whole text of the paper is also investigated.

4.2 Review Conducting

The review conduction stage presents the selection

process for LR in this research.

4.2.1 Article Selection

This phase is divided into three sub tasks including,

pilot study, article selection and quality assessment of

selected primary studies.

Pilot Study. A pilot study is carried out to investigate

the reliability of provided selection criteria as sug-

gested by Kitchenham and Charters (Brereton et al.,

2007), (Keele et al., 2007), and Petersen et al. (Pe-

tersen et al., 2015) before the selection of primary ar-

ticles. In this stage, ﬁve papers are selected by the

ﬁrst and second authors. These articles are investi-

gated by a third author, who is an expert in the NLP

domain and was not involved in the search process

by considering the inclusion and exclusion criteria. A

satisfactory result is presented from the pilot study,

which proves the suitability of the deﬁned criteria in

this research.

Requirement Formalisation Using Natural Language Processing and Machine Learning: A Systematic Review

239

Initial Search

First Filtering

(Abstract, Title,

keywords, Intro,

Conclusion)

Applying

Inclusion/Exclusion

Criteria

Detailed Filtering

257 studies fund

115 studies removed

142 studies remained

71 studies removed

71 studies remained

24 studies removed

47 studies remained

Figure 1: Primary studies selection process.

Primary Study Selection. In this part, the relevant

articles are searched using the provided search string.

The result of this selection is presented in Figure 1.

For the initial search process, 257 results are recog-

nised. To reﬁne the selected papers in the next it-

eration, titles, abstracts, keywords, introduction, and

conclusion sections are reviewed. As a result, 115

papers are removed from the list of selections, and

then the rest are kept for the next iteration. Following

that by applying inclusion and exclusion criteria 71

papers are rejected. Next for this iteration, the con-

tent of the paper is investigated deeply and 24 papers

are rejected. Finally, 47 studies are remaining. Figure

3 presents the distribution of the resulting papers in

each year. As can be seen in this ﬁgure, in 2021 the

domain gained more interest and the highest number

of publications were published. Additionally, Figure

2 indicates the publication type of result papers. Con-

ferences are the target of publication in most studies.

Quality Assessment. The quality of the selected

study is also assessed in this research. Therefore, a

checklist with four quality assessment questions is

presented in Table 2. The questions are answered

by the ﬁrst author by selecting from ’yes’, ’no’, and

’partly’ options.

Figure 2: Primary studies per publication type.

Figure 3: Distribution of NLP papers in each year.

4.2.2 Data Extraction and Synthesis

To answer each research question, the data extraction

process is performed by developing a predeﬁned data

extraction form in Table 3. The form enables us to

record essential information about primary studies to

answer each research question. The form is ﬁlled out

by the ﬁrst author manually and then the second and

third authors reviewed the results and ﬁnally, the is-

sues are ﬁxed.

4.2.3 Reporting the Review

Based on the results of data extraction phase, the re-

view result is presented and each research question is

answered and discussed.

Table 2: Quality assessment Questions.

QID Topic Question

A1 Objective Did the study clearly deﬁne

the research objectives?

A2 Related Did the study provide a review

work of previous work?

A3 Research Are the research methodology

methodology clearly established?

A4 Validity Did the study include a

discussion on the validity and

reliability of the procedure used?

A5 Future work Did the study point out potential

further research?

MODELSWARD 2023 - 11th International Conference on Model-Based Software and Systems Engineering

240

Table 3: Data Extraction Form.

Study data Description Relevant RQ

Title Study

Overview

Author Study

Overview

Year Study

Overview

Article Study

Source Overview

Type of Journal,

Article Conference

Workshop

Research What is the main RQ1

goal goal of study?

Research Model extraction/generation, RQ1

goal Requirement formalisation,

category UML, Usecase, Class

Activity, ER diagram

extraction/generation

Research What research methods RQ1

method did the study employ?

Data What are the datasets RQ2

for evaluation of the study

Evaluation what are the evaluation RQ1

criteria in the study?

NLP What are the applied RQ1

techniques NLP techniques?

ML What are the applied RQ1

techniques ML techniques?

NLP What NLP tools RQ1

tools did the study use?

Challenges What are the challenges RQ3

of the study?

Future Work What are the suggested RQ3

future work?

5 RESULTS

This section presents the result of the review of this

research. We selected 47 primary studies for SLR.

5.1 Summary of Studies

Investigated studies for classifying NLP and ML tech-

niques to formalise requirements are available in

(RFr, 2022). The studies are summarised according to

the applied NLP/ML techniques, input and output ar-

tifacts, datasets, approach, applied tools and libraries,

input structure, and evaluation criteria.

This section discusses the classiﬁcation results

of the investigated approaches according to each re-

search question.

Figure 4: Frequency of NLP Technologies in Different

Studies.

5.2 Q1: What Are the Most Commonly

Used NLP/ML Approaches for

Automatic/Semi-Automatic RF?

To answer this question, NLP and ML approaches ap-

plied in selected studies are deeply investigated.

5.2.1 NLP Techniques

Figure 4 presents the applied NLP techniques and fre-

quency of using those techniques through out investi-

gated papers.

• Applied Techniques. Tokenization, POS tag-

ging and Type dependencies are the most common

used NLP techniques in those research.

• Heuristic Rules. Majority of research applied

heuristic rules for formalising requirements.

• Frequent Tools and Libraries. Stanford core

NLP is the most common-used NLP tools in in-

vestigated studies.

• Evaluation Criteria. Accuracy in terms of preci-

sion, recall and F-measure are the most common

criteria for evaluation in selected studies.

5.2.2 ML Techniques

There are not many studies that apply ML techniques

for RF and mostly classical ML techniques such as

decision trees and Support Vector Machine (SVM) are

used in those studies. The applied ML techniques and

frequency of application of these techniques are pre-

sented in Figure 5. Around 20% of selected studies in

this research applied both NLP and DL techniques.

5.3 Q2: What Are the Input and

Output of RF Approaches?

This question is answered according to the applied

type and structure of input elements and generated

output elements.

Requirement Formalisation Using Natural Language Processing and Machine Learning: A Systematic Review

241

Figure 5: Frequency of DL Technologies in Different Stud-

ies.

Figure 6: Frequency of Input for RF.

• Input Types. The frequency of input types in se-

lected studies is presented in Figure 6. English

text and user story are the most common-used

type in most studies.

• Input Structure. The unstructured English text is

the most frequent input structure for most of the

studies as presented in Figure 7. The inputs in the

format of user story are semi-structured.

• Output Types. In most studies, the formalisation

is in the form of UML diagrams. Figure 8 indi-

cates that class and use case diagrams are the most

common input types in these studies.

Figure 7: Structure of Input Elements.

Figure 8: Frequency of Generated Output for RF.

5.4 Q3: What Are the Gaps and

Deﬁciencies in Existing RF Work?

The RF ﬁeld remains at an experimental stage, in par-

ticular evaluation of approaches is not performed sys-

tematically and it is difﬁcult to compare different ap-

proaches. The published results of studies were often

not reproducible due to the unavailability of tools or

data.

Investigating different works, we identiﬁed three

main deﬁciencies for formalising requirements. A list

of deﬁciencies is provided below:

5.4.1 Lack of Completeness in Heuristic Rules

The performance and completeness of heuristic ap-

proaches are typically not evaluated on a broad range

of input cases, thus it is not possible to determine

which are the best to use in different situations. A

lot of papers applied heuristic approaches and deﬁned

rules manually. It is not possible to come up with any

speciﬁc number of rules for formalising requirements

and generating relevant artifacts. This issue is not

deeply investigated in the community as there is not

an adequate comparative evaluation in this domain.

5.4.2 Lack of Application of DL

There is under-use of DL techniques, which seem to

be relevant and applicable to RF tasks and could help

to avoid the limitations of heuristic approaches, espe-

cially for unstructured source data. It is assumed that

the limited number of training data in the community

is the main reason that developers do not use DL mod-

els in different tasks. Most of the learning methods

used in the community are standard methods such as

decision tree or regression model. It can be concluded

that most of the studies use learning in the wrong way

and do not exploit the full potential of deep learning

techniques in this domain. Therefore, in theory, ap-

plication of ML provides a potential solution for the

MODELSWARD 2023 - 11th International Conference on Model-Based Software and Systems Engineering

242

different task of RF. Community can beneﬁt from the

DL models and tools by applying some modern learn-

ing architecture such as OpenAI Codex (Chen et al.,

2021) as in these models it is not essential to have lots

of labels data for performing particular task.

5.4.3 Lack of Evaluation Benchmark

Framework

To systematically compare different RF cases, stan-

dard benchmarks and evaluation criteria need to be

established e.g., there are well-established bench-

marks in Natural Language Generation (GEM, ) and

Speech Processing (SUP, ). Many of the evaluation

datasets cited in selected papers are no longer avail-

able. Therefore, a repository of standard cases, pro-

posed approaches, and evaluation procedures are nec-

essary. This is an important issue in the community

and it is essential to ﬁll this gap.

6 DISCUSSION ON ACTION

TAKEN

Our systematic review results show open issues and

research challenges for formalising requirements.

This research is part of MDENet project and some

actions are taken by the authors of this paper to solve

part of the issues. These actions are summerised be-

low:

• In order to strengthen the area of RF research,

we developed a DSL for NLP pipelines, based on

the SQLite grammar of GitHub - antlr/grammars-

v4/SQLite. This enables the high-level deﬁnition

of NLP pipelines for RF, independent of any par-

ticular NLP platform such as NLTK or Apache

OpenNLP. Common RF processing such as POS-

tagging, segmentation, chunking, and parsing can

be speciﬁed. A transformation from the DSL to

Python was deﬁned to support implementation in

NLTK.

• To provide a central point of reference for RF re-

search, we established a GitHub repository (RFr,

2022), which will hold links to state-of-the-art re-

search in the area, evaluation cases, evaluation

tools, and the results of evaluations. The repos-

itory will be a resource for the RF community and

aims to improve the practical application of RF

research to real-world software problems.

• To compare the effectiveness of RF approaches,

there needs to be an established set of require-

ments statements that can be applied. We selected

25 cases of real-world requirements statements in

Figure 9: Three Kinds of Evaluation Strategy for RF Ap-

proaches.

the format of user stories, which reﬂect a diversity

of linguistic styles and scales, and added these to

the RF repository.

• To evaluate the results of applying RF approaches

to the evaluation cases, we provide tools to (i)

compare the formalized models produced by an

approach to manually constructed reference mod-

els for the cases, to identify a measure of similar-

ity of these models; (ii) compare the formalized

models to the source document, to check the com-

pleteness of the formalization; (iii) to evaluate the

internal quality of the formalized model. Three

kinds of evaluation strategy are presented in Fig-

ure 9. Example evaluations have been provided

for three RF approaches, evaluated on two user

story case studies.

7 FUTURE DIRECTIONS

In the following we discuss directions to complete the

current actions and further future work on RF.

• Generate a platform to guide the user in selecting

appropriate NLP techniques occurring to their re-

quirements. This will occur by considering more

case studies and evaluating them by applying the

evaluating tools speciﬁed in (RFg, 2022).

• We will develop the RF repository with more

example case studies, including examples of

unstructured requirements statements/background

documentation, evaluation tools and evaluations,

and publicise this in MDE forums and invite con-

tributions from RF researchers.

Requirement Formalisation Using Natural Language Processing and Machine Learning: A Systematic Review

243

8 CONCLUSION

This research carried out a systematic survey of exist-

ing approaches for RF, including NLP and ML ap-

proaches across a wide range of applications. 250

publications were examined, and 47 speciﬁc publica-

tions were selected for deeper analysis. We identiﬁed

that:

• Heuristic NLP approaches are the most common

RF technique in the research, primarily operating

on structured and semi-structured data.

• Deep learning techniques are not widely-used, in-

stead classical ML techniques such as decision

trees and Support Vector Machine (SVM) are used

in the surveyed studies.

• There is a lack of standard benchmark cases for

RF and therefore it is difﬁcult to compare the per-

formance of different approaches.

REFERENCES

Gembenchmark. https://gem-benchmark.com/.

Superbbenchmark. https://superbbenchmark.org/.

(2022). Repository of state-of-the art requirements formal-

isation approaches.

(2022). Requirement formalisation using natural language

processing and machine learning repository. https://

doi.org/10.5281/zenodo.7337229/.

Abdelnabi, E. A., Maatuk, A. M., and Hagal, M. (2021).

Generating uml class diagram from natural language

requirements: A survey of approaches and techniques.

In 2021 IEEE 1st International Maghreb Meeting of

the Conference on Sciences and Techniques of Au-

tomatic Control and Computer Engineering MI-STA,

pages 288–293. IEEE.

Ahmed, S., Ahmed, A., and Eisty, N. U. (2022). Au-

tomatic transformation of natural to uniﬁed model-

ing language: A systematic review. arXiv preprint

arXiv:2204.00932.

Alzayed, A. and Al-Hunaiyyan, A. (2021). A bird’s eye

view of natural language processing and requirements

engineering. Int. J. Adv. Comput. Sci. Appl., 12(5):81–

90.

Brereton, P., Kitchenham, B. A., Budgen, D., Turner, M.,

and Khalil, M. (2007). Lessons from applying the sys-

tematic literature review process within the software

engineering domain. Journal of systems and software,

80(4):571–583.

Chen, M., Tworek, J., Jun, H., Yuan, Q., Pinto, H. P. d. O.,

Kaplan, J., Edwards, H., Burda, Y., Joseph, N., Brock-

man, G., et al. (2021). Evaluating large language mod-

els trained on code. arXiv preprint arXiv:2107.03374.

Dawood, O. S. et al. (2017). From requirements engineer-

ing to uml using natural language processing–survey

study. European Journal of Industrial Engineering,

2(1):pp–44.

Goodfellow, I., Bengio, Y., and Courville, A. (2016). Deep

learning. MIT press.

Keele, S. et al. (2007). Guidelines for performing system-

atic literature reviews in software engineering. Tech-

nical report, Technical report, ver. 2.3 ebse technical

report. ebse.

Kulkarni, A. and Shivananda, A. (2019). Natural language

processing recipes. Springer.

Petersen, K., Vakkalanka, S., and Kuzniarz, L. (2015).

Guidelines for conducting systematic mapping stud-

ies in software engineering: An update. Information

and software technology, 64:1–18.

Pohl, K. (2010). Requirements engineering: fundamentals,

principles, and techniques. Springer Publishing Com-

pany, Incorporated.

Raharjana, I. K., Siahaan, D., and Fatichah, C. (2021). User

stories and natural language processing: A systematic

literature review. IEEE Access, 9:53811–53826.

Rolland, C. and Proix, C. (1992). A natural language

approach for requirements engineering. In Interna-

tional conference on advanced information systems

engineering, pages 257–277. Springer.

Ryan, K. (1993). The role of natural language in require-

ments engineering. In [1993] Proceedings of the IEEE

International Symposium on Requirements Engineer-

ing, pages 240–242. IEEE.

Samuel, A. L. (1967). Some studies in machine learning

using the game of checkers. ii—recent progress. IBM

Journal of research and development, 11(6):601–617.

Sch

on, E.-M., Winter, D., Escalona, M. J., and

Thomaschewski, J. (2017). Key challenges in agile

requirements engineering. In International Confer-

ence on Agile Software Development, pages 37–51.

Springer, Cham.

Tukur, M., Umar, S., and Hassine, J. (2021). Require-

ment engineering challenges: A systematic mapping

study on the academic and the industrial perspec-

tive. Arabian Journal for Science and Engineering,

46(4):3723–3748.

Yalla, P. and Sharma, N. (2015). Integrating natural lan-

guage processing and software engineering. Interna-

tional Journal of Software Engineering and Its Appli-

cations, 9(11):127–136.

Zaki-Ismail, A., Osama, M., Abdelrazek, M., Grundy, J.,

and Ibrahim, A. (2021). Arf: Automatic require-

ments formalisation tool. In 2021 IEEE 29th Inter-

national Requirements Engineering Conference (RE),

pages 440–441. IEEE.

Zhao, L., Alhoshan, W., Ferrari, A., Letsholo, K. J., Ajagbe,

M. A., Chioasca, E.-V., and Batista-Navarro, R. T.

(2020). Natural language processing (nlp) for re-

quirements engineering: A systematic mapping study.

arXiv preprint arXiv:2004.01099.

MODELSWARD 2023 - 11th International Conference on Model-Based Software and Systems Engineering

244