International Journal of Innovation and Applied Studies

ISSN 2028-9324 Vol. 3 No. 3 July 2013, pp. 857-865

© 2013 Innovative Space of Scientific Research Journals
http://www.issr-journals.org/ijias/

Literature Review of Automatic Single Document Text Summarization Using NLP

1 1 2
Md. Majharul Haque , Suraiya Pervin , and Zerina Begum
1
Department of Computer Science & Engineering,
University of Dhaka,
Dhaka, Bangladesh
2
Institute of Information Technology,
University of Dhaka,
Dhaka, Bangladesh

Copyright © 2013 ISSR Journals. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

ABSTRACT: In the time of overloaded online information, automatic text summarization is especially demanded for salient
information retrieval from huge amount electronic text. For the blessing of World Wide Web, the mass of data is now
enormous in its volume. Researchers realized this fact from various aspects and tried to generate an automatic abstract of
the gigantic body of data from the commencement of the last half century. Numerous ways are there for characterizing
different approaches to passage recapitulation: extractive and abstractive from single or compound document, objective of
content abridgement, characteristic of text summarization, level of processing from superficial to profound and sort of
article’s content. A significant précis is very much helpful in our day to day life which can save valuable time. The
investigation was at first commenced naively on single document abstraction. In this paper, automatic single document text
summarization task is addressed and different methodologies of various researchers are discussed from the very beginning of
this research to this modern age. This literature review intends to observe the trends of abstraction procedure using natural
language processing. Also some promising approaches are indicated and particular concentration is dedicated for the
categorization of diversified methods from raw level to similar like human professionals, so that in future one can get
precious direction for further analysis.

KEYWORDS: Information retrieval, World Wide Web, electronic text, automatic abstract, human professionals.

1 INTRODUCTION

Outcome of the information retrieval becomes necessary for user to find out concrete information for the abstraction
because of the stridently escalation of data on the web. Internet is widely used by people to come across information using
proficient information retrieval (IR) tools, such as Google, Yahoo, AltaVista, etc., where findings are abundant. In most of the
cases, users feel bore with the very tedious and time consuming job to reveal the main gist of the outcome of the IR.
Academics and researchers are very much benefitted by using automatic text summarization system as a tool to lessen the
amount of time spent manually extracting the chief thoughts from large documents. In addition to the above reason,
automatic text summarization also provides its users with numerous benefits as well as:
(i) Increase efficiency of other researches to choose documents/information from search engines’ output, which usually
contain an excess amount of replicated information.
(ii) Solve the limitation of information presentation on small communication devices such as PDA and mobile phone etc.,
which is able to display abridged version of the full document.
(iii) The running time of machine for translation is significantly reduced if a short version of text is given.

Corresponding Author: Md. Majharul Haque ([email protected]) 857

Literature Review of Automatic Single Document Text Summarization Using NLP

American research libraries spawned the initial interest in automatic text shortening during sixties [1]. A considerable
body of research over the last sixty years has explored different levels of analysis of text, to help determine what information
in the text is salient for a given summarization task [1]. Radev et al [2] in 2002 defined a summary as a text that is
engendered from one or more texts, which conveys essential information of the original text(s), and that is no longer than
half of the original text(s) and usually notably less than that. Simply a summary text is a derivative of a source text condensed
by selection and/or generalization on important content [3]. A large document is entered into the computer and a
recapitulated content is returned, which is a non redundant extract from the original passage. Automatic text summarization
can be classified into single document text summarization and multiple documents text summarization [4] as in figure 1. This
paper focused on single document text summarization as single document was the target from the commencement of such
research on automatic abstraction.

Fig. 1. Classification of document summarization

Various methods that utilize passage categorization, such as neural networks, semantic graphs, decision trees, regression
models, fuzzy logic and swarm intelligence, etc. are involved on the study on finding crucial portion of text. The objective of
this paper is to present a comprehensive literature review on automatic single document text summarization using natural
language processing and investigate the movement of document abridgement.
The rest of the paper is organized as follows. Section 2 briefly describes about Natural Language Processing. Section 3
presents a comprehensive literature review about different procedures of automatic single document text summarization. At
a glance comparison among the various techniques is depicted in section 4. Section 5 turns conclusion with a brief about this
paper.

2 NATURAL LANGUAGE PROCESSING

Natural Language Processing (NLP) is a field of computer science, artificial intelligence and linguistics as all those specified
arena brought it into play. Generally it deals with the interactions between machines and human languages that accomplish
task on analyzing, understanding and generating the language, which human use naturally in order to interact with
computers in both oral and written contexts using natural human languages instead of computer languages [5]. It is an
interdisciplinary area based on versatile arena of study including computer engineering, which provides methods for model
illustration, algorithm devise and accomplishment; linguistics, which categorizes linguistic forms and practices; mathematics,
which provides formal models and methods; psychology, which studies models and theories of human behavior; statistics,
which offers procedures for predicting measures based on sample records; and biology, which travels around the underlying
architecture of linguistic processes in the human brain [6].

3 REVIEW ON AUTOMATIC SINGLE DOCUMENT TEXT SUMMARIZATION

In the very beginning of the research in the arena of launching artificial intelligence to generate abridged version of a
large document, disclosed the paradigms for extracting salient features.
Automatic text summarization process model can be divided into three steps [7] as in figure 2: (1) in the preprocessing
step source text interpretation to source text representation, (2) source representation transform to summary text
representation with an algorithm and (3) in the final step, summary text generation from summary representation.

ISSN : 2028-9324 Vol. 3 No. 3, July 2013 858

 Md. Majharul Haque, Suraiya Pervin, and Zerina Begum

Fig. 2. Process flow of automatic text summarization

Numerous types of research work have been accomplished by various researchers where we can be familiar with multiple
types of way of summary generation from single document text. In the following part of this section, methodologies of
several researchers that incorporated this topic are depicted in brief, from pioneering works to the era of modern science
where the thoughts of similar like human professionals’ abstraction techniques are being explored.

3.1 PIONEERING WORKS

The voice over automatic i.e. computerized abstraction initiated around sixty years ago. As accomplishment of automatic
passage summarizer was often cited in the oldest publication in 1958 by H. P. Luhn [8]. This method is based on the word
frequency and clearly emphasized that the most frequent words represent the most important concepts of the text. In the
next step, frequencies are used to score and decide on sentences to be extracted. In this paper, application of machine is
emphasized and expressed that because of the absence of the variations of human capabilities and orientation, auto-
abstracts have a high degree of reliability, consistency, and stability, as they are the product of statistical analysis of the
author’s own words. This paper is worthy of being appreciated as it is almost the earliest paper in this arena of automatic
text abstraction [9]. Besides this, the proposed method mostly works on scientific article and journal paper.
P. B. Baxendale [10] in 1958 investigated machine techniques for reducing scientific credentials to their important sharp
indices. Human scanning patterns were tried to be simulated here for selecting topic sentences special phrases. It was
declared in this paper that sentence’ position and containing certain cue-words (i.e., words like ‘crucial’ or ‘pertinent’ etc.) or
the word exist in the heading are special indicator for being in the salient parts of the document.
G. J. Rath et al [11] in 1961 in their research scrutinized about four types of lexical indicators of content to determine
which one is the best for detecting relevant document from repository and to answer a set of question. After their
experiment, it was claimed that utilizing complete text is better than only abstract for answering question. But for
distinguishing relevant document only abstract is enough. They also proclaimed that purely statistical method of producing
extracts was suspected of being inadequate, and hence other methods were sought.
H. P. Edmundson [12] in 1969 accomplished a notable progress after ten years of the beginning of the research on text
recapitulation. He described three additional methods with the standard keyword method, disregarding the very high
frequency common words to determine the sentences’ weight. Those are:
(i) Cue Method: The hypothesis of this technique is that the presence or absence of certain cue words will compute the
significance of a sentence.
(ii) Title Method: The weight of a sentence is calculated as a sum of all the content words materializing in the title,
headings and sub-headings of a text.
(iii) Location Method: Here relevance is assumed on the basis of location, expressed that sentences taking place in initial
position of paragraphs have a higher probability of being pertinent.
The result was very fruitful and assumed that by using a combination of these three schemes the best correlation
between the automatic and human-made abstracts can be achieved. This paper emphasized on indicative abstracts rather
than on the production of informative abstracts. So, if a user doesn’t know about the document to be summarized earlier,
can’t get summary through this proposed methodology.

ISSN : 2028-9324 Vol. 3 No. 3, July 2013 859

Literature Review of Automatic Single Document Text Summarization Using NLP

3.2 ALGEBRAIC METHODS

Julian Kupiec et al [13] in 1995 explored an innovative algebraic method. Using naïve-bayes classifier, this classification
procedure sorts out each sentence as worthy of being in the abstract or not. A trained classifier is generated at first, where
the authors used a corpus of 188 couples of complete documents/summaries. The distinguishing features used uppercase
words, length of sentence, structure of phrase, position in paragraph besides word frequency.
ChinYew Lin et al [14] in 1997 in their paper addressed the difficulty of identifying probable topics of texts by their
location in the corpus, and scored sentences by its position in given text. They considered techniques of tailoring the position
method towards optimality over a variety and how it can be estimated for efficiency. The idea that texts generally follow a
predictable discourse structure, and that the sentences of greater topic centrality tend to occur in certain specifiable
locations (e.g. title, abstracts etc), arises the position method for topic recognition. Here also predicted that discourse
structure significantly varies over domains, for which position method is a bit tough. Given a text T and the list of topic
keywords, each sentence of the text is labeled with the paragraph number and the sentence number. With the number of
2097 documents in this method, the result illustrated that the first and the last 50 positions fully cover majority of the text.
Eduard Hovy et al [15] in 1999 attempted to create a robust automated text summarization system and named it
SUMMARIST. Instead of irregular term counting, SUMMARIST combines symbolic world knowledge (embodied in WordNet,
dictionaries, and similar resources) with strong NLP processing (using IR and statistical method) to resolute concept
relevance. Their procedure based on the following equation:
Summarization = topic identification + interpretation + summary generation
S. P. Yong et al [16] in 2005 introduced an automatic text summarization system that incorporates learning ability by
combining a statistical approach, keywords extraction and neural network with unsupervised learning. It was claimed that
their proposed technique can extract 83.03% of significant contents. The procedure is made up of three steps, as follows:
(i) Text pre-processing subsystem: two pre-processing methods are applied, one is stop words removal to remove words
like “the”, “a”, “can” and “will” etc. and another one is stemming to convert each word to its stem by eliminating
suffixes and prefixes.
(ii) Keywords extraction subsystem: from a group of sample heterogeneous text documents, main features of each word
are determined by computing term frequency tf(w, s) as the number of times that the word w appears in the
sentences s, and the inverse sentence frequency isf(w) is calculated as the number of sentences in which the word w
occurs. From the generated tf-isf(w, s) matrix, most frequent terms are listed as keyword in the text to be
summarized.
(iii) Summary production subsystem: in the final step the system chooses sentences, which contain the keywords, as part
of the summary. It is suggested to run through another round of stop words checking procedure before selecting
sentences for being ensured that there is no stop word is working as keyword.

3.3 PROCEDURES BASED ON TEXT CORRELATION

Previously discussed methods used statistical probability for choosing a sentence for generating abridged version of text.
But those panoramas are fully dumb about the cohesion of sentences with each other. So, the relations between concepts in
a text can’t be captured using extractive methods. If a sentence is extracted which contains link to any previous context then
the summary will be difficult to understand [17]. In those circumstances various research work done for exploring the
cohesive properties to comprise relations among expressions of text.
M. A. K. Halliday et al [18] in 1976 performed the first research to explore the degree of subjectivity of two aspects of the
lexical cohesion perceived by readers of text: the word cluster (lexical chains) that are formed and the lexical semantics
relations that are perceived between the words. Building blocks of lexical cohesion, cohesive harmony, and the concept of
patterns of lexical affinity are the lexical semantic relations. The linguistic study was emphasized here and tried to forms inter
sentence groups of related words, and the original outlook of them was very wide and universal; but there had to be a
recognizable relation between two words as the only decisive factor.
Ruqaiya Hasan [19] in 1984 extended the investigation to include the concept of cohesive harmony, which adds lexico-
grammatical structure to word groups by first dividing them into two types: (i) identity-of-reference chains that combine
reference and lexical cohesion, (ii) similarity chains that used only classical relations, and linking these chains with
grammatical intra-sentence relations.

ISSN : 2028-9324 Vol. 3 No. 3, July 2013 860

 Md. Majharul Haque, Suraiya Pervin, and Zerina Begum

Rhetorical Structure Theory (RST) of text summarization can also be included into the groups of methods based on text
correlation as the text coherence is attributed principally to the presence of rhetorical relation. RST was developed during
1980s by researchers in natural language generation, which models the discourse structure of a text by means of a
hierarchical tree diagram [20].
William C. Mann et al [21] in 1988 established a new definitional foundation of RST and also scrutinized three claims: the
predominance of nucleus/satellite structural patterns, the functional basis of hierarchy, and the communicative role of text
structure. Various kinds of consequences of RST have also been reviewed in this paper. A quantity of allegorical relations,
which hook up together text units, forms a hierarchical structure called RST tree. The relations tie collectively a nucleus as
central to the author’s objective and a satellite as subsidiary parts. Texts in terminal nodes of RST tree are supposed to be
encoded and non-terminal nodes represent contiguous text spans, whose sibling spans are joined by discourse relations. At
last the nucleus and some tightly connected terminal nodes are selected as the principal theme of passage.
Jane Morris et al [22] in 1991 utilized cohesion chains and Regina Barzilay et al [23] in 1997 exploited lexical chains to
characterize contents of a document. A perception in the document is symbolized by a sequence of associated words in these
representations. The sequence is a list of words that confines a segment of the consistent structure of the document and is
generated as independent of the grammatical formation of the text. The procedure usually launch from a set of words in the
heading of the document to construct lexical chains, adjoining of words that have similar meaning or are related to the title.
WordNet thesaurus has been used for determining cohesive relations between terms (i.e., repetition, synonymy, antonymy,
hypernymy, and homonymy). In the next step, scores are calculated for sentences considered to be important by the
previous step and those scores will be used to generate the final summary.
Branimir Boguraev et al [24] in 1997 proposed a novel technique for saliency-based content characterization of text
document. A procedure was offered, which uses a salience feature as the basis for a “ranking by importance” of an
unstructured referent set, and ultimately for topic stamp identification. Co-reference resolution system is used here, which is
a process of determining whether two expressions in natural language refer to the identical entity. Then a threshold value is
calculated and where the occurrences of frequently mentioned objects overcome the value is included into the summary.
Li Chengcheng [25] in 2010 presented an effective method using RST for successful automatic text summarization, which
is based on natural language generation. The rhetoric structure of the text is extracted with a compound that relates the
sentences in this theory. Here the summarization is accomplished using the nodes i.e. nucleus, those are given weights based
on script based analysis. Then the entire text is divided into individual sentences based on commas, quotes, semicolons and
punctuation marks exist in the sentences. This is then done into a graph, deletes the unimportant sentences and then
summarizes the entire text. Principal scheme of this procedure is as follows: (i) analyzing the candidate sentence, (ii) discover
the rhetoric relations and (iii) forming the essential part of sentence constructive for ultimate recapitulation.

3.4 METHODS CLOSE TO HUMAN ABSTRACTION CONCEPT

Artificial Intelligence is much matured in this modern age, as the application of this is noticeable in various regions such as
robotics, machine learning and knowledge based system etc. Automated summary is also the grace of Artificial Intelligence
through natural language processing. But, still there is a qualitative dissimilarity between synopsis generated by existing
automated summarizers and the abstracts written by human. Extraction is a common technique utilized by most of the
summarizers, which may be ambiguous or incoherent to the original abstract. In spite of the existence of some shortcomings,
a number of methods have started to emerge lately with either sentence compressing capability or re-producing technique.
Hongyan Jing [26] in 2000 presented a novel sentence lessening method, which eliminate extraneous phrases from the
extracted sentences that are chosen for abstraction purpose. A parse tree is generated at first in this procedure and then
grammar is checked to decide which terms of a sentence must not be removed to keep its’ structure accurate. Then it finds
the sentences that are closely related to the main topic by using corpus evidence, consisting of sentences compressed by
human and original sources, delete unnecessary sub-tree from parse tree for producing the final outline.
Kevin Knight et al [27] in 2000 proposed an innovative method by incorporating the procedure of regenerating sentence
for coherent abstract, rather than just extracting from original source. A training corpus is used here, which is a collection of
newspaper articles paired with human written abstracts. Their first goal is to rewrite a compressed version of given input
parse tree. The rewriting process starts with an empty Stack and Input List that contains the sequence of words subsumed by
the given large parse tree. Four types of operations are incorporated here: SHIFT operations move the first word from the
input list into the stack; REDUCE operations pop the top syntactic trees, combine them and again push them to the stack;
DROP operations are used to remove from the input list that has already been compressed; ASSIGN TYPE operations change
label of trees at the top of the stack. The procedure ends when the input list is vacant and the stack contains only one tree.

ISSN : 2028-9324 Vol. 3 No. 3, July 2013 861

Literature Review of Automatic Single Document Text Summarization Using NLP

After that an in-order traversal of the leaves of this tree generates the compressed form of the sentences those are given as
input.
K. McKeown et al [28] in 2000 stated that there is a significant difference between the summaries produced by current
automatic system and the human professionals. Because automatic summarizer cannot always recognize key topics of an
article and automatic procedure has no robust text generation method. This paper presented a “Cut and Paste” strategy for
text summarization that derived from an analysis of human written abstract. “Cut and Paste” method not just extracts
sentences but smooth the extracted sentences by editing them which mainly cutting phrases and pasting them together in a
novel ways. It was stated that generated summary bears a resemblance to the human summarization process more than
extraction does. Six operations were defined to transform chosen sentences from an article into the corresponding summary
sentences in its human written abstracts: (i) sentence reduction, (ii) sentence combination, (iii) syntactic transformation, (iv)
lexical paraphrasing, (v) generalization and specification, and (vi) reordering. An evaluation was done for this procedure with
50 human-written abstracts, consisting of 305 sentences in total and claimed that 93.8% of sentences were correctly
decomposed.

4 COMPARISON AMONG THE TECHNIQUES

At a glance comparison among the discussed techniques of single document text summarization has been shown in table
1:

ISSN : 2028-9324 Vol. 3 No. 3, July 2013 862

 Md. Majharul Haque, Suraiya Pervin, and Zerina Begum

Table 1. Comparison Among the Techniques of Single Document Text Summarization

# Researcher(s), Category Technique Basis of procedure

Year, Reference
1 H. P. Luhn, 1958, Pioneering works Word frequency Frequent words represent the most important concepts of the
[8] text.
2 P. B. Baxendale, Pioneering works Position in text Sentence position and containing certain cue-words or the word
1958, [10] exist in the heading are special indicator for being the salient parts
of the document.
3 G. J. Rath, 1961, Pioneering works Lexical indicator Lexical indicators of content could be utilized best by subjects to
[11] determine relevant from irrelevant documents.
4 H. P. Edmundson, Pioneering works Cue words and heading The sentences that contain specific cue words, words those are
1969, [12] exist in title and heading or sub-heading, are significant to be
selected for summary.
5 Julian Kupiec, 1995, Algebraic method Naïve-Bayes classifier Uppercase words, length of sentence, structure of phrase, position
[13] in paragraph with word frequency are the distinguishable features
of sentence.
6 ChinYew Lin, 1997, Algebraic method Position in text Texts generally follow a predictable discourse structure, and that
[14] the sentences of greater topic centrality tend to occur in certain
specifiable locations (e.g. title, abstracts, etc).
7 Eduard Hovy, 1999, Algebraic method Symbolic world Instead of irregular term counting, this method combines symbolic
[15] knowledge world knowledge (embodied in WordNet, dictionaries, and similar
resources) with strong NLP processing (using IR and statistical
method) to resolute concept relevance.
8 S. P. Yong, 2005, Algebraic method Neural network Using neural based system this method generate summary using
[16] three sub system such as: Summarization = Text pre-processing
sub-system + Keywords extraction sub-system + Summary
production sub-system.
9 M. A. K. Halliday, Text correlation Lexical cohesion Building blocks of lexical cohesion, cohesive harmony, and the
1976, [18] concept of patterns of lexical affinity are the lexical semantic
relations.
10 Ruqaiya Hasan, Text correlation Lexical cohesion Identity-of-reference chains and similarity chains, that linking with
1984, [19] grammatical intra-sentence relations.
11 William C. Mann, Text correlation Rhetorical Structure Texts in terminal nodes of RST tree are supposed to be encoded
1988, [21] Theory(RST) and non-terminal nodes represent contiguous text spans.
12 Jane Morris, 1991, Text correlation Cohesion chains A perception in the document is symbolized by a sequence of
[22] associated words in the representation.
13 Regina Barzilay, Text correlation Lexical chains The procedure usually launch from a set of words in the heading of
1997, [23] the document to construct lexical chains, adjoining of words that
have similar meaning or are related to the title.
14 Branimir Boguraev, Text correlation Saliency-based content This method uses a salience feature as the basis for a “ranking by
1997, [24] characterization importance” of an unstructured referent set.
15 Li Chengcheng, Text correlation Rhetorical Structure Principal scheme of this procedure is as follows: (i) analyzing the
2010, [25] Theory(RST) candidate sentence, (ii) discover the rhetoric relations and (iii)
forming the essential part of sentence constructive for ultimate
recapitulation.
16 Hongyan Jing, Human abstraction Lessening method Find the sentences that are closely related to the main topic and
2000, [26] concept eliminate extraneous phrases from the extracted sentences that
are chosen for abstraction purpose.
17 Kevin Knight, 2000, Human abstraction Regenerating sentence Rewrite a compressed version of given input parse tree. The
[27] concept rewriting process starts with an empty Stack and Input List that
contains the sequence of words subsumed by the given large parse
tree.
18 Hongyan Jing, Human abstraction Cut and paste Smooth the extracted sentences by editing them which mainly
2000, [28] concept cutting phrases and pasting them together after reduction,
combination, transformation, specification, reordering, etc.

ISSN : 2028-9324 Vol. 3 No. 3, July 2013 863

Literature Review of Automatic Single Document Text Summarization Using NLP

5 CONCLUSION

In this paper the concepts of single document text summarization that categorize different approaches in this arena have
been reviewed. This literature review emphasizes on extractive approaches for text summarization. Deriving the classification
of the automatic text abstraction procedure has also been attempted. Recent trend in summarization system that comes
from novice procedure to resemble with human written summary has been scrutinized here. We believe that the study of
document summarization is a productive region for further research, both by linguists performing text analysis and by
computational linguists trying to create techniques to produce summaries conforming to one or more of the characteristics
listed above. Around 18 papers have been briefly discussed and various key topics from other historical publication relevant
with text abstraction have been analyzed here from 1958 to 2010. There exist some other procedures similar with those
briefed in this paper, the discussion of which has not been included here as it will be a large corpus. Nowadays, in the age of
computer and internet, information is found from various sources about a single topic, so the research of multi-document
text abstraction is a burning issue. After all it is expected that any researchers can get help from this literature review for
better understanding of different types of summary generation techniques even for multi-document text abridgement. It will
also assist for better perception of the diversified sorts of abstraction procedure, which will help to construction of new
formula and systems that significantly serve the various principle of summarization in broad-spectrum.

REFERENCES

[1] Martin Hassel, “Evaluation of Automatic Text Summarization - A practical implementation,” Licentiate thesis,
Department of Numerical Analysis and Computer Science, Royal Institute of Technology, Stockholm, Sweden, 2004.
[Online] Available: http://www.csc.kth.se/~xmartin/papers/licthesis_xmartin_notrims.pdf (May 21, 2013)
[2] Dragomir R. Radev, Eduard Hovy and Kathleen McKeown, “Introduction to the special issue on summarization,” Journal
of Computational Linguistics, vol. 28, no. 4, pp. 399-408, December 2002.
[3] Abdelwahab Hamou-Lhadj and Timothy Lethbridge, “Summarizing the Content of Large Traces to Facilitate the
Understanding of the Behaviour of a Software System,” Proceedings of the 14th IEEE International Conference on
Program Comprehension, pp. 181-190, 2006.
[4] S. Suneetha, “Automatic Text Summarization: The Current State of the art,” International Journal of Science and
Advanced Technology (ISSN 2221-8386), vol. 1, no. 9, pp. 283-293, November 2011.
[5] Eugene Charniak and Drew McDermott, Introduction to artificial intelligence, Addison-Wesley Longman Publishing Co.,
Inc., Boston, MA, 1985.
[6] Bill Z. Manaris, Natural Language Processing: A Human-Computer Interaction Perspective, University of Southwestern
Louisiana, Academic Press, New York, vol. 47, pp. 1-66, 1998.
[7] Karen Sparck Jones, Automatic summarizing: factors and directions, In: Inderjeet Mani and Mark T. Maybury (Eds.),
Advances in Automatic Text Summarization, Cambridge MA: MIT Press, pp. 1-12, 1999.
[8] Hans P. Luhn, “The Automatic Creation of Literature Abstracts,” IBM Journal of Research and Development, vol. 2, no. 2,
pp. 159-165, 1958.
[9] Dipanjan Das and André F. T. Martins, “A survey on Automatic Text Summarization,” Language Technologies Institute,
Carnegie Mellon University, 2007.
[Online] Available: http://www.cs.cmu.edu/~afm/Home_files/Das_Martins_survey_summarization.pdf (May 21, 2013)
[10] P. B. Baxendale, “Machine-made Index for Technical Literature -An Experiment,” IBM Journal of Research and
Development, vol. 2, no. 4, pp. 354-361, October 1958.
[11] G. J. Rath, A. Resnick and T. R. Savage, “Comparisons of four types of lexical indicators of content,” Journal of the
American Society for Information Science and Technology, vol. 12, no. 2, pp. 126-130, April 1961.
[12] H. P. Edmundson, “New Methods in Automatic Extracting,” Journal of the Association for Computing Machinery, vol. 16,
no. 2, pp. 264-285, April 1969.
[13] Julian Kupiec, Jan Pedersen and Francine Chen, “A Trainable Document Summarizer,” Proceedings of the 18th annual
international ACM SIGIR conference on Research and development in information retrieval, pp. 68-73, 1995.
[14] Chin-Yew Lin and Eduard Hovy, “Identifying Topics by Position,” In Proceedings of the Fifth conference on Applied
natural language processing, San Francisco, pp. 283-290, 1997.
[15] Eduard Hovy and Chin-Yew Lin, Automated Text Summarization in SUMMARIST, In: Inderjeet Mani and Mark T. Maybury
(Eds.), Advances in Automatic Text Summarization, MIT Press, chapter 8, pp. 18-24, 1999.
[16] S. P. Yong, A. I. Z. Abidin and Y. Y. Chen, “A Neural Based Text Summarization System,” 6th International Conference of
Data Mining, pp. 45-50, 2005.

ISSN : 2028-9324 Vol. 3 No. 3, July 2013 864

 Md. Majharul Haque, Suraiya Pervin, and Zerina Begum

[17] Karel Jezek and Josef Steinberger, “Automatic Text Summarization (The state of the art 2007 and new challenges),”
Znalosti, pp. 1-12, 2008.
[18] M. A. K. Halliday and Ruqaiya Hasan, Cohesion in English, Longman, London, 1976.
[19] Ruqaiya Hasan, Coherence and Cohesive Harmony, In: Flood James (Ed.), Understanding Reading Comprehension:
Cognition, Language and the Structure of Prose. Newark, Delaware: International Reading Association, pp. 181-219,
1984.
[20] William C. Mann and Sandra A. Thompson, Relational Propositions in Discourse, Defense Technical Information Center,
Information Sciences Institute, Marina del Rey, 1983.
[Online] Available: http://www.tandfonline.com/doi/abs/10.1080/01638538609544632?journalCode=hdsp20#preview
[21] William C. Mann, Sandra A. Thompson, “Rhetorical Structure Theory: Toward a functional theory of text organization,”
Text, vol. 8, pp. 243-281, 1988.
[Online] Available: http://www.cis.upenn.edu/~nenkova/Courses/cis700-2/rst.pdf (May 21, 2013)
[22] Jane Morris and Graeme Hirst, “Lexical Cohesion Computed by Thesaural Relations as an Indicator of the Structure of
Text,” Journal of Computational Linguistics, vol. 17, no. 1, pp. 21-48, March 1991.
[23] Regina Barzilay and Michael Elhadad, “Using Lexical Chains for Text Summarization,” In Proceedings of the ACL/EACL-97
Workshop on Intelligent Scalable Text Summarization, Association of Computational Linguistics, pp. 10-17, 1997.
[24] Branimir Boguraev and Christopher Kennedy, “Salience-based Content Characterization of Text Documents,” In
Proceedings of the ACL'97/EACL'97 Workshop on Intelligent Scalable Text Summarization, 1997.
[25] Li Chengcheng, “Automatic Text Summarization Based On Rhetorical Structure Theory,” International Conference on
Computer Application and System Modeling (ICCASM), vol. 13, pp. 595-598, October 2010.
[26] Hongyan Jing, “Sentence Reduction for Automatic Text Summarization,” In Proceedings of the 6th Applied Natural
Language Processing Conference, Seattle, USA, pp. 310-315, 2000.
[27] Kevin Knight and Daniel Marcu, “Statistics-Based Summarization Step One: Sentence Compression,” In Proceeding of
the 17th National Conference of the American Association for Artificial Intelligence, pp. 703-710, 2000.
[28] Hongyan Jing and Kathleen R. McKeown, “Cut and Paste Based Text Summarization,” In Proceedings of the 1st Meeting
of the North American Chapter of the Association for Computational Linguistics, Seattle, USA, pp. 178-185, 2000.

ISSN : 2028-9324 Vol. 3 No. 3, July 2013 865