ScienceDirect

ScienceDirect
Available online at www.sciencedirect.com

ScienceDirect
Procedia Computer Science 171 (2020) 2522–2531

Third International Conference on Computing and Network Communications (CoCoNet’19)

Automatic Source Code Documentation using Code Summarization

Technique of NLP
Menaka Pushpa Arthur*
Associate Professor, Department of Information Technology, St.Joseph’s Institute of Technology, Chennai 600119, India

Abstract

Source code documentation is an important process for software project maintenance and management. The documentation process
always consumes a lot of time and effort from human experts. The software project document should be concise and clear without
any ambiguity. Existing source code documentation tools like JavaDoc are very limited in the market. Also, tools can identify only
the predefined methods of corresponding programming languages. In this paper, a novel system is proposed to automate the source
code documentation process for C programming language using the source code summarization technique of NLP. The core
component of this proposed system i.e., Software Word Usage Model (SWUM) build using Context-Free Grammars and NLP
preprocessing techniques. This system can successfully generate the documentation for a C program along with predefined and
user-defined methods using Natural Language Generation technique. This proposed system can document a program in two major
formats; method-based, abstract level and statement-based, detailed level. The proposed system efficiency is evaluated by
comparing system-generated source code documentation with an expert generated documentation. Results obtained from that
comparison shows that the proposed system can give better performance for small and medium-size software projects.
© 2020 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review under responsibility of the scientific committee of the Third International Conference on Computing and Network
Communications (CoCoNet’19).
Keywords: Source code documentation; CFG; NLP; code summarization; SWUM; etc

* Corresponding author. Tel.: +91-9994133025.

E-mail address: [email protected], [email protected]

1877-0509 © 2020 The Authors. Published by Elsevier B.V.

This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review under responsibility of the scientific committee of the Third International Conference on Computing and Network
Communications (CoCoNet’19).
10.1016/j.procs.2020.04.273
 Menaka Pushpa Arthur / Procedia Computer Science 171 (2020) 2522–2531 2523

1. Introduction

Documentation of the software project is mostly written by the human in software industries. This process is one
of the important steps in the software life cycle model. Better documentation will reduce the cost of project
maintenance and update processes. Software project Documentation maintains detail about a requirement, analysis,
coding, testing and installation procedures of the software development cycle. This research work concentrates only
on automatic documentation generation for coding. Coding documentation should be concise and clear without any
ambiguity for developers of the maintenance process. Usually, documentation consumes a lot of effort and time from
a human. The number of documentation pages per KiloCode is one among the metric which is used for estimating
software project cost. Efficient documentation will be a very helpful part of the version update of the software project.
The documentation development time and cost can be reduced by this proposed automatic source code documentation
system using NLP techniques. This proposed system can generate documentation for C programs due to its simple
syntax set among other object-oriented programming languages like Java. This research work develops a prototype
model for an automatic source code documentation process by following the simple syntax-based, C-lite programming
language compiler. Existing source code tools are few and have limited functionalities like JavaDoc, it generates
documentation only for predefined methods of Java by extracting information from the Java programming language’s
official documentation. JavaDoc cannot automatically generate source code documentation for user-defined methods.
Depends on user comments JavaDoc can generate documentation for user-defined methods.
In this paper, the proposed system can automatically generate the documentation for C program by Natural Learning
Processing techniques. Two major documentation types are generated by this system; (i) In method-based
documentation, methods like user-defined and pre-defined are automatically documented using extended method call
graph technique. In this, user-defined methods are documented by the Natural Language Generation (NLG) grammars
whereas pre-defined methods can be documented with the help of documentation of C programming language’s
compiler (ii) In the statement-based documentation, each program lines are documented along with methods if any
exist in the given program. The proposed system uses C program as an input in order to generate documentation as an
output. Program levels of complexity vary with the existence of user or pre-defined methods in this. Context-Free
Grammars are formed for the syntax of C programming language. This set of rules can identify types of program
statements such as expression, condition, declaration and function definitions, etc. Then, each statement type is
converted into line documentation with necessary identifiers using NLG technique. Software Word Usage Model
(SWUM) of this system is developed by a unique technique that acts as the core module for the software code
documentation process.
The organization of this paper is as follows: The related work is discussed in Section II. In Section III, the proposed
automatic source code documentation system is well described. In Section IV, the simulation results are discussed to
demonstrate the proposed system’s efficiency. Finally, conclusions and future enhancements of this work are discussed
in Section V.

2. Related Works

The existing research works in software source code interpretation using NLP techniques are limited. G.Sridhara
et al. [3, 4] and E. Hill et al. [7] have been initiated and contributed a lot of research works in this domain. P. W.
McBurney et al. [1] proposed a system to generate source code summarization of context for Java methods. This
system generates a summarization for method signatures. Method signatures are interpreted by the NLP techniques
and summarization is generated by NLG technique. Hence, this system requests a strict way to define a method
signature with an appropriate name for the method and its parameters and then the return value of this method. The
method signature is parsed and tagged then identified the meaning of a method from its name by considering this as a
verb in POS tagging process. The actual intention of this system is to replace the well-known JavaDoc for method
documentation in Java. The limitation of this P. W. McBurney et al. [1] system is signature-based code summarization.
The quality of the summarization depends only on the signature of the method. If the method signature is general or
irrelevant to functions task then we cannot expect meaningful summarization from this system. JavaDoc converts user
given comments and Java programming language documentation for the corresponding method into documentation
for both user-defined and pre-defined Java methods. Whereas P. W. McBurney et al. [1] system converts programming
2524 Menaka Pushpa Arthur / Procedia Computer Science 171 (2020) 2522–2531

language statements into natural language description for Java methods. This system cannot interpret the syntax of
programming language other than the method signature.
S. Badihi et al. [2] proposed a system for automatic code summarization for Java programs using Eclipse-plug-in
along with web-based crowd summaries. They have designed such a system using NLP and NLG techniques for
interpreting programming sentences and code automatic summarization. Program summary from experts are collected
using crowd sourcing. Crowd sourcing is an additional overhead process in this project also it reduces the quality of
the expert’s code summarization content. An important method is identified based on its frequency in the program.
Method’s linguistics such as action, arguments and themes are identified using NLP techniques that used to generate
a summary for the concern Java method. This method expects the user to strictly follow naming conventions when
they write the program. Irrelevant and meaningless naming for methods and their arguments in general identifier
reduces the quality of this system. G.Sridhara et al. [3] generate code summary for Java method from its method call
and signature using the combination of the techniques NLP and NLG. Lexicalization used to identify variables from
the methods. This system identifies the S_Units from the method that contains starting, ending lines and important
lines such as a loop from the method, in order to know the return type of the method. Software Word Usage Model
has been build for software analysis using NLP by Pollock.L et al. [4]. Parsing the program sentence and analyzing
method signature has been done via SWUM is explained well in this paper. Possible applications of software
analysis using their system were also listed out in this work. They have mainly highlighted identifier extraction,
POS tagging and abbreviation expansion in order to improve the accuracy of the system. This system generates
a code summary for methods mainly from its signature.
J. Fowkes et al. [5] designed an unsupervised, extractive source code summarization system using an auto-
folding method. This system creates Abstract Syntax Tree (AST) for source code written in Java. They have
formulated the summarization problem mathematically and code blocks are identified to generate a program
summary. Finally, the system generated summary is compared with the summary generated by Javadoc and then
find out the best code summary from this analysis. They didn’t use NLP technique and claimed that NLP is the
extension of their research work. Few more systems have been designed with the help of Natural Language
Processing to interpret the source code that is not for generating a code summary. Francu et al. [6] proposed a
system to automatically generate executable Java source code from its system requirement using NLP. The system
requirement is represented in the form of UML. E.Hill et al. [7] designed a system that finds an exact position in the
large size software for maintenance and reuse within minimum search time. Identifier split and POS tagging are used
to identify the important keywords in the software code. Improved NL query can be generated using the set of extracted
important keywords. Lei et al. [8] introduced a reverse engineering process in software engineering. They have
designed a system to automatically generate C++ parser from English language specification about the input files of
the software program.

3. Automatic Source Code Documentation System

This section presents an architecture and working procedure of the proposed automatic source code documentation
system using the code summarization technique of Natural Language Processing. The system accepts C program as
an input with the option of the existence of pre-defined or/and user-defined functions. This proposed system can
automatically generate concise source code documentation in a natural language without any human intervention.
Two major types of source code summarization are generated by this system; (i) program or source code
documentation and (ii) method documentation. The proposed system contains three major modules; (i) Preprocessing
of C program (ii) Building a Software Word Usage Model (SWUM) (iii) Important methods extraction and (iv) Source
code documentation generation using NLG for an entire C program or only for extracted important methods of a given
program as per user choice. The detailed architecture of the proposed system is shown in Figure 1 with sub-
components. The working procedure of this system is given in Figure 3 in the form of an algorithm. This system can
generate documentation for each and every line of C program. Existing systems or tools can generate documentation
only for the pre-defined or user-defined methods of the program and it cannot automatically generate documentation
for the entire program code.
 Menaka Pushpa Arthur / Procedia Computer Science 171 (2020) 2522–2531 2525

C Program Line Tokenization

Line Parsing

Method Term Frequency

Preprocessing

Modified Method Page Rank Algorithm Extract Top Rank

Call Graph on Call Graph Method

Important Method Identification

Extract Method Extract Keyword

Block using NLP
SWUM
Domain DB
Sentence Formation
using NLG
NLG

Fig. 1. The Architecture of Automatic Source Code Documentation System.

3.1. Program Preprocessing

This module generates an output from the given C program and that output will be utilized to build a software word
usage model; extract important methods from the program and generate the source code documentation for the
program lines. Program statements or lines are tokenized using newline delimiter in a regular expression. The total
number of program lines and then the set which contains the program line is obtained from the line-based program
tokenization process. Then, POS tagging is applied to each retrieved program lines from the generated line set. From
this, sentence components like a verb, noun, pronoun, etc. are identified for each line except the program lines that
contain only the keywords of the programming language. Parsing of program lines using space delimiter in the regular
expression can identify the type of program statements/lines such as the declaration, control, expression, assignment,
etc. Program lines are parsed using regular expressions with POS tagging and then identified by CFG rules. The line
parsing process is explained in the next section as it is with the software word usage model building phase. The
frequency of method call is identified by the term-frequency technique.

3.2. Software Word Usage Model

SWUM is the core component of the code summarization technique in NLP. The way to build SWUM is varying
depends on its purpose. This work creates SWUM for the source program by considering individual program lines
with its type, parsing detail, type of the methods (pre-defined or user-defined) and method frequency, etc. Parsing of
2526 Menaka Pushpa Arthur / Procedia Computer Science 171 (2020) 2522–2531

program lines is carried out based on the syntax of programming language that is in the form of Context-Free
Grammars. The syntax of C programming language is defined in the form of CFG rules. The sample CFG rules of this
proposed system are given in Figure 2. Regular expression technique maps the program lines with a set of CFG
programming language syntax rules in order to identify the type of that program statement. In this work, syntax rules
are formed based on the syntax of the C-Lite compiler. The syntax rules of the C-Lite compiler is very light and
compact than the C compiler. Due to this reason, this work adopts the syntax rules of the C-Lite compiler for
developing a prototype model for an automatic source code documentation system. In future work, this prototype
system model will be enhanced and modified with respect to other high-end programming language compilers like
object-oriented programming languages and task-specific programming languages, etc. The information about the
program statement type gives the actual purpose of the corresponding statement in the logic of the problem-solving.
In the next step, POS applied to each line and find the statement components like verb, noun, adverb, etc. Those
components are given as one among the input for NLG process to generate the descriptive program line
documentation. This line POS information acts as a basic building block for sentence formation in the documentation
generation phase. Methods are identified separately along with its frequency and the relationship among methods like
method interaction type, interaction frequency, the importance/rank of the method and the order of method call and
so on. This information is the main input for the Modified Method Call Graph phase of the system.

Statement → Stat | Control | Library | Method | Key

Stat → Expr1 | Declaration1 | Key Integer Key | Key Identifier Key
Expr1 → Expr ‘;’
Declaration1 → Declaration ‘;’
Declaration → DataType Identifier | DataType Identifier Key Identifier | DataType Identifier ‘=’ Integer | DataType
Identifier ‘=’ Identifier | DataType Identifier ‘=’ Expr | Identifier '=' Identifier | Identifier '=' Integer | Identifier '='
Expr | Identifier ‘=’ MethodCall1 | DataType Identifier ‘=’ MethodCall1
Expr → Identifier Op Identifier | Identifier Op Integer | Integer Op Identifier | Integer Op Integer
Library → Key ‘;’
Method → MethodDecl1 | MethodCall1 | MethodDefn
MethodDec1 → MethodDec ‘;’
MethodCall1 → MethodCall ‘;’
MethodDec → DataType Method
MethodCall → Method
MethodDef → DataType Method Key Statement Key
Method → Identifier Key ParamList Key
ParamList → DataType Identifier | DataType Identifier Key ParamList | Identifier | Identifier Key ParamList |
Integer | Integer Key ParamList
Control → Key Condition Key Statement Key | Key Condition Key Statement Key Statement | Key Condition Key
Statement Key Key Statement Key
DataTypes → ‘int’ | ‘float’ | ‘char’ | ‘char[’ Integer ‘]’ | ‘double’
Key → Keywords | Keywords Key
Keywords → 'printf' | 'scanf' | '#include<stdioh.h>' | '#include<math.h>' | '#include<string.h>' |'#include<conio.h>'
| 'return' | 'main' | 'int' | 'void' | ‘String’ | ‘{‘ | ‘}’ | ‘;’ | ’,’ | ’>’ | ‘(‘ | ‘)’ | ‘switch’ | ‘case’ | ‘break’ | ‘do’ | ‘if’ | ‘else’ |
‘while’ | ‘for’
Identifier → Alphabet | Identifier Integer | Alphabet Identifier | Alphabet Identifier Alphabet
Alphabet → 'a' | 'b' | 'c' | 'd' | 'e' | 'f'|….|’z’|’A’|’B’|…..|’Z’
Integer → Digit | Digit Integer
Digit → '0'|'1'|'2'|'3'|'4'|'5'|'6'|'7'|'8'|'9'
Op → '+' | '-' | '*' | '/' | '**' | '%'

Fig 2. Context-Free Grammar for Programming Language Syntax

 Menaka Pushpa Arthur / Procedia Computer Science 171 (2020) 2522–2531 2527

3.3. Modified Method Call Graph

Method call graph explains the way of calling a method from other methods. It gives the relationship between
methods. The order of method calling given by the method call graph uses for proper documentation about function
calling within the program or project. This project modified the method call graph by including the detail about the
nature of methods such as pre-defined or used-defined or from other program files etc. An arc of this modified graph
carries a weight that indicates the calling frequency of the corresponding method. The methods which all have a direct
call from the main method are marked as important methods. And, based on the total number of lines of code of the
method also used to identify the importance of that method with respect to a software project. The methods which
have large LOC are considered as important methods. As per the input program is given in Figure 4(a), method main()
contains 19 lines whereas other methods addition and subtraction has a single line. Form this, we cannot claim the
method main as the important method of the given program based on the maximum LOC. In the modified method call
graph always consider the main () as the highest priority important method of any program regardless of its LOC. The
actual task of the method is also considered to identify the important methods in the program. This phase needs further
research to calculate the exact weight of each method of the program. Important methods are only identified by the
modified method call graph and extracted for the method documentation process.

3.4. Code Documentation using NLG

The source code documentation is carried out in two ways; method-description (short documentation) and line-
wise description (detailed documentation). A set of natural language generation rules are formed to generate sentences
either for the entire program or extracted important methods of the given program. The rules set is majorly classified
as two, based on the input of this phase. The tag value and type of the program sentence play a major role to form its
corresponding natural language sentence. The detailed documentation may help in the reverse engineering process of
software engineering in the future.
Input : C Program Code and CFG, NLG rules
Output : Source Code Documentation
1. procedure preprocessing (program p, CFG rules)
2. split program lines using newline delimiter
3. create an array of program lines ‘l’ by line tokenization
4. for each line / element of ‘l’
5. find line type using CFG rules
6. store line type in ‘type’
7. for each method type of statement
8. find the frequency of method calling and store it in ‘freq’
9. create software word usage model ‘sw’
10. return sw, l, type and freq

11. procedure modified_methodcall_graph (number_of_methods, method, freq)

12. create ‘n’ vertices that is equal to number_of_methods
13. assign weight ‘w’ for each m1 and m2 link; if method ‘m1’ calls a method ‘m2’
14. ‘w’ is assigned as ‘freq’
15. Other link weights ‘w’ set as ‘0’
16. find the page rank method to extract ‘m’ with high ‘freq’
17. return top rank methods ‘m’

18. procedure code_summary(program p, CFG rules, NLG rules1, swum sw)

19. call preprocessing (p, rules)
20. for each line ‘l’ in ‘p’
21. call part_of_speech tagging procedure
22. for each ‘l’ and its ‘tag’
23. find ‘verb’ and ‘noun’ and other components
24. create a sentence using ‘components’ and ‘NLG’ rules
25. return sentence / documentation of each ‘l’
2528 Menaka Pushpa Arthur / Procedia Computer Science 171 (2020) 2522–2531

26. procedure method_summary(program p, CFG rules, NLG rules1, swum sw)

27. tm = call modified_methodcall_graph(p, rules)
28. extract top ranked ‘tm’ method from ‘p’
29. call code_summary(tm, rules, rules1)
30. return important method documentation

31. procedure main (program p, CFG rules, NLG rules1)

32. sw = call preprocessing (p, rules)
33. call code_summary(p, rules, rules1,sw)
34. call method_summary(p, rules, rules1,sw)
35. create documentation and print

Fig 3. Algorithm of the Proposed System

4. Results and Discussion

#include <stdio.h> Library

int my_add(int x, int y) Method → MethodDef
{ BlockStart
return (x + y); Expression → Addition → Return
} BlockEnd
float my_sub(float x, float y) Method → MethodDef
{ BlockStart
return (x-y); Expression → Subtraction → Return
} BlockEnd
int main() Method → MethodDef → Predefined → Main
{ BlockStart
int a, b, c; Identifier → a,b,c → Integer
float a1, b1, c1; Identifier → a1, b1, c1 → Float
printf("Enter a : "); Method → MethodCall → Predefined → Print
scanf("%d", &a); Method → MethodCall → Predefined → Read → a
printf("Enter b : "); Method → MethodCall → Predefined → Print
scanf("%d", &b); Method → MethodCall → Predefined → Read → b
c = my_add(a,b); Method → MethodCall → UserDef → my_add → a,b
Return → c
printf("Result of Addition : %d", c); Method → MethodCall → Predefined → Print → c

printf("Enter a1 : "); Method → MethodCall → Predefined → Print

scanf("%f", &a1); Method → MethodCall → Predefined → Read → a1
printf("Enter b1 : "); Method → MethodCall → Predefined → Print
scanf("%f", &b1); Method → MethodCall → Predefined → Read → b1
c1 = my_sub(a1,b1); Method → MethodCall → UserDef → my_sub → a1,b1
Return → c1
printf("Result of Addition : %f", c1); Method → MethodCall → Predefined → Print → c1

return 0; Keyword → Return → 0

} BlockEnd

Fig 4(a). Source Program Fig. 4(b) Parser – Sentence Type Identification
 Menaka Pushpa Arthur / Procedia Computer Science 171 (2020) 2522–2531 2529

The proposed system is implemented by Python with necessary Natural Language Processing and Natural
Language Generation packages. The sample input C program is given in Figure 4(a) that will help to identify the
output generated by the proposed system along with intermediate results such as output of the sentence parser and
modified method call graph. This program does not have any developers’ comments. Without the developer’s
comments, the proposed system can generate the documentation for any source program. In contrast, Javadoc requests
code comments to generate the method summary. Figure 4(a) shows the C program that performs the addition of two
integers and subtraction of two float numbers. User-defined methods of this program are my_add() and my_sub() with
two arguments in the argument list along with the program’s entry point. The program statement type is identified by
the parser and SWUM modules of the system. The program statement type generated by the Parser and SWUM is
given in Figure 4(b) for each and every line of the source code including header file-inclusion. From this parser, user-
defined and pre-defined method sections are extracted and then analyzed to generate the method call graph of the
given program. This proposed system has a modified method call graph component that lists the detail such as
important methods in the program, LOC, interaction type, interaction direction, arguments of the interaction and
purpose of the method. The output of the method call graph module for the given program is given in Figure 5.

Modified Method Call Graph Information:

int main () → my_add(a,b)

Count: 1
Parameter from int main () → my_add(a,b): 2 integer
Parameter from my_add(a,b) → int main (): 1 integer
Task: addition

int main () → my_sub(a1,b1)

Count: 1
Parameter from int main () → my_sub(a1,b1): 2 float
Parameter from my_sub(a1,b1) → int main (): 1 float
Task: subtraction

Important Methods with LOC:

int main () : 19
my_add(a,b) : 1
my_sub(a1,b1) : 1
Fig 5. The Output of Modified Method Call Graph for the Program given in Fig 4(a)

Library “stdio” included

User Defined Method “my_add” for Addition of “a”
(integer), “b” (integer). Return result (integer).
User Defined Method “my_add” for Addition of “a” User Defined Method “my_sub” for Subtraction of “a1”
(integer), “b” (integer). (float) and “b1” (float). Return result (float).
Return result (integer).
Store in “c” (integer). Main Starts
Declaring Identifiers a(int), b(int), c(int),
a1(float),b1(float),c1(float)
User Defined Method “my_sub” for Subtraction of Read a(int), b(int), a1(float), b1(float)
“a1” (float) and “b1” (float). Call Method my_add with a(int), b(int). Save Result in c
Return result (float). (int)
store in “c1” (float). Call Method my_sub with a1(float), b1(float). Save Result
in c1(float)
Print the Result c and c1

Main Returns 0. Main Ends.

Fig 6(a) Source Code Documentation for Methods Fig 6(b) Source Code Documentation for Entire Program
2530 Menaka Pushpa Arthur / Procedia Computer Science 171 (2020) 2522–2531

System Gen. Summary Vs. Rouge-1 Rouge-L Rouge-W

References
P R F1 P R F1 P R F1
Method – Expert generated summary 87.48 71.25 79.37 81.59 68.84 75.22 61.24 30.34 45.79
Program – Expert generated summary 50.93 82.5 66.72 44.57 64.16 54.37 33.46 32.54 33
Algorithm 74.18 74.79 74.49 77.73 78.27 78 49.63 38.12 43.88
Fig 7. Average ROUGE Results of the Proposes System

Two types of documentation are generated by the system based on user choice; method-based documentation is
given in Figure 6(a) and entire source code documentation is in Figure 6(b). Method based documentation gives the
code summary only for user-defined methods of the program. Whereas, detailed documentation can describe all the
statements of the program that include user-defined and pre-defined methods. Nearly, 20 programs are tested in this
proposed system. All the results are in precise quality. The summary generated by the proposed system is compared
with the documentation developed by the software experts. The ROUGE measurement [9] is used to evaluate the
quality of the system generated source code documentation. Figure 7 shows the ROUGE metrics (P-Precision, R-
Recall, F1-Average) of the automatic source code documentation system under its method-based and program-based
documentation. Also, system generated documentation is compared with an expert’s generated algorithm since its
format looks like an algorithm. Figure 7 shows that ROUGE-1 results are better than ROUGE-L and ROUGE-W.
Recall of detailed (program) documentation is better than its Precision in ROUGE-1. Whereas, method-based
documentation has better Precision than its Recall under the same category. F1 metric solves this confusion. It shows
the proposed system performance is better in method-based documentation than entire program documentation. When
compare the system generated detailed-documentation with expert’s algorithm, Precision and Recall are very similar.
Though the documentation generated by the system is enough to satisfy the end-user, the ROUGE metric results are
failed to justify that point. It shows the importance to identify a new evaluation metric to analyze the performances of
the source code summarization system without any ambiguity.

5. Conclusion

The proposed system of this work can automatically generate source code documentation for C programs. Existing
software documentation tools can generate documentation only for methods of the software code. This restriction is
solved by this proposed system. The automatic source code documentation system uses the code summarization
technique of NLP along with NLG. This system can document each and every line of the C program. Two major
documentation types are offered by this system; method-based documentation and program-wise documentation. The
core component of code summarization system, Software Word Usage Model (SWUM) is successfully built using the
programming language’s syntax-based Context-Free Grammar rules. This automatic source code documentation
system reduces the cost of the documentation process of the software life cycle model. Also, this automation can
reduce the overall cost of software project development. In this research work, we proposed a system that works only
with C-Lite compiler-based C programs. C-Lite is a lightweight compiler and we are unable to check the complete
capability of this proposed system prototype against the complicated syntax-based programming languages. In future
work, this system will be enhanced and modified in order to generate code documentation for other high-end, objected-
oriented programming languages like Java, Python and Web designing programming languages.

References

[1] P. W. McBurney and C. McMillan, "Automatic Source Code Summarization of Context for Java Methods," in IEEE Transactions on Software
Engineering, vol. 42, no. 2, pp. 103-119, 1 Feb. 2016.
[2] S. Badihi and A. Heydarnoori, "CrowdSummarizer: Automated Generation of Code Summaries for Java Programs through Crowdsourcing,"
in IEEE Software, vol. 34, no. 2, pp. 71-80, Mar.-Apr. 2017.
 Menaka Pushpa Arthur / Procedia Computer Science 171 (2020) 2522–2531 2531

[3] Giriprasad Sridhara, Emily Hill, Divya Muppaneni, Lori Pollock, and K. Vijay-Shanker. 2010. Towards automatically generating summary
comments for Java methods. In Proceedings of the IEEE/ACM international conference on Automated software engineering (ASE '10). ACM,
New York, NY, USA, 43-52.
[4] Pollock L., Vijay-Shanker K., Hill E., Sridhara G., Shepherd D. (2013) Natural Language -Based Software Analyses and Tools for
Software Maintenance. In: De Lucia A., Ferrucci F. (eds) Software Engine ering. Lecture Notes in Computer Science, vol 7171.
[5] J. Fowkes, P. Chanthirasegaran, R. Ranca, M. Allamanis, M. Lapata and C. Sutton, "Autofolding for Source Code Summarization," in IEEE
Transactions on Software Engineering, vol. 43, no. 12, pp. 1095-1109, 1 Dec. 2017.
[6] Francu, Jan and Petr Hnetynka. “Automated Code Generation from System Requirements in Natural Language.” e-Informatica 3,2009: 72-88.
[7] E. Hill, L. Pollock and K. Vijay-Shanker, "Automatically capturing source code context of NL-queries for software maintenance and
reuse," 2009 IEEE 31st International Conference on Software Engineering, Vancouver, BC, 2009, pp. 232-242.
[8] Lei, Tao et al. “From Natural Language Specifications to Program Input Parsers.” ACL (2013).
[9] C.-Y. Lin. “ROUGE: A package for automatic evaluation of summaries.” Proceedings of the ACL-04 Workshop, 2004.