See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/258651905

A Survey of Text Question Answering Techniques

Article in International Journal of Computer Applications · September 2012

DOI: 10.5120/8406-2030

CITATIONS READS

104 3,915

2 authors:

Poonam Gupta Vishal Gupta

Maharishi Markandeshwar University, Mullana Panjab University
8 PUBLICATIONS 125 CITATIONS 41 PUBLICATIONS 2,596 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

text summarization View project

All content following this page was uploaded by Poonam Gupta on 30 January 2016.

The user has requested enhancement of the downloaded file.

 International Journal of Computer Applications (0975 – 8887)
Volume 53– No.4, September 2012

A Survey of Text Question Answering Techniques

Poonam Gupta Vishal Gupta

ME, Computer Science & Engineering Assistant Professor, Computer Science &
University Institute of Engineering & Technology, Engineering Department
Panjab University, Chandigarh University Institute of Engineering & Technology
Panjab University, Chandigarh

ABSTRACT NLP systems. This movement leads to an explosion of

activities like information retrieval, natural language
Question Answering (QA) is a specific type of information understanding, etc. [1][2][3]. Information retrieval is an art
retrieval. Given a set of documents, a Question Answering and science of searching for information in documents,
system attempts to find out the correct answer to the question searching for documents themselves, searching for metadata
pose in natural language. Question answering is which describe documents, or searching within databases,
multidisciplinary. It involves information technology, whether relational standalone databases or hypertext
artificial intelligence, natural language processing, knowledge networked databases such as the Internet, for text, sound,
and database management and cognitive science. From the images or data [4].
technological perspective, question answering uses natural or
statistical language processing, information retrieval, and Question answering is a difficult form of information retrieval
knowledge representation and reasoning as potential building characterised by information needs that are at least somewhat
blocks. It involves text classification, information extraction expressed as natural language statements or questions, and
and summarization technologies. In general, question was used as one of the most natural type of human computer
answering system (QAS) has three components such as communication. In comparison with classical information
question classification, information retrieval, and answer retrieval, where complete documents are considered similar to
extraction. These components play a essential role in QAS. the information request, in question answering, specific pieces
Question classification play primary role in QA system to of information are returned as an answer. The user of a
categorize the question based upon on the type of its entity. question answering system is interested in a concise,
Information retrieval method is get of identify success by comprehensible and correct answer, which may refer to a
extracting out applicable answer post by their intelligent word, sentence, paragraph, image, audio fragment, or an
question answering system. Finally, answer extraction module entire document [13]. The main purpose of a QA system is to
is rising topics in the QAS where these systems are often find out ‘‘WHO did WHAT to WHOM, WHERE, WHEN,
requiring ranking and validating a candidate’s answer. HOW and WHY?”[11]. QA systems merge information
retrieval with information extraction methods to identify a set
Most of the Question Answering systems consists of three of likely set of candidates and then to produce the final
main modules: question processing, document processing and answers using some ranking scheme [12].
answer processing. Question processing module plays an
important part in QA systems. If this module doesn't work In current years, there has been a marked increase in the
correctly, it will make problems for other sections. Moreover amount of information available on the Internet. Users often
answer processing module is an emerging topic in Question have specific questions in their mind, for which they expect to
Answering, in which these systems are often required to rank discovering out the answers. They would like to find out the
and validate candidate answers. These techniques aiming at answers to be short and precise, and they always prefer to
discovering the short and precise answers are often based on express the questions in their native language without being
the semantic classification. QA systems give the ability to restricted to a particular query language, query formation
answer questions posed in natural language by extracting, rules, or even a particular knowledge domain. The latest
from a repository of documents, fragments of documents that approach taken to matching the user needs is to carry out
contain material relevant to the answer. actual investigation of the question from a linguistic point of
view and to attempt to understand what the user really means.
General Terms A typical pipeline Question Answering System consists of
Types, Architecture, Applications, Information retrieval. three distinct phases: Question classification, information
retrieval or document processing and answer extraction.
Keywords Question classification is the first phase which classifies user
questions, derives expected answer types, extracts keywords,
Natural language processing, Question answering System, and reformulates a question into semantically equivalent
Information retrieval. multiple questions. Reformulation of a query into similar
meaning queries is also known as query expansion and it
1. INTRODUCTION boosts up the recall of the information retrieval system.
NLP focuses on communications between computers and Information retrieval (IR) system recall is very important for
natural languages in terms of theoretical results and practical question answering. If no correct answers are present in a
applications, and on information sharing now that information document, no further processing could be carried out to find
is exchange as it never has been before and sharing an answer. Precision and ranking of candidate passages can
information becoming the leading theme in the domain of also affect question answering performance in the IR phase.

1
 International Journal of Computer Applications (0975 – 8887)
Volume 53– No.4, September 2012

Answer extraction is a final component in question answering deals with and so on. The table 1 provides the detail of the
system, which is the tag of discrimination[5]. comparisons of these QA systems.
2. GENERAL ARCHITECTURE Table 1, Characterization of QA systems
The user writes a questioni by means of the user query
interface. After that this query is used to extract all the QA system based on vQA systems
possible answers for the input question. The architecture of DIMENSIONS NLP and IR Reasoning
Question-Answering system is as shown in Figure 1. with NLP
The architecture which is given in Figure 1 works in 5 stages.
The function of each stage is as follows [6]: Technique Syntax processing, Semantic
Named Entity tagging Analysis or
and verify com and Information high reasoning
Retrieval
Data Resource Free text documents Knowledge Base
Domain Domain Independent Domain Oriented
Responses Extracted Snippets Synthesized
Responses
Questions Mostly wh- type of Beyond of wh-
Deals with Questions type of questions
Evaluations Uses existing N/ A
Information Retrieval
3.1 Web Based Question Answering
Systemiser
With the wide spread usage of internet a tremendous use of
data is available, web is one of the best source to obtain the
aFigure 1. Architecture of Question-Answering Systemp d information. Web based question answering systems is using
the search engines (Like Google, Yahoo, Alto Vista etc.,) to
2.1 Query Pre-processing get back webpage’s that potentially containing answers to the
Given a natural language question as input, the overall questions. The majority of these Web based QA systems
function of the question preprocessing module is to process works for open domain while some of them works for domain
and analyze the input question. This leads to the classification oriented also. The wealth of information on the web making it
of question as belonging to any of the types supported by the an attractive store for getting quick answers to simple, factual
system. questions[16]. The data that is available on web has the
characteristics of semi structure, heterogeneity and
distributivity.
2.2 Query Generation The Web Based QA systems mostly handles wh-type of
In query generation we will use Query Logic Language (QLL)
questions such as “who killed Indira Gandhi”?
which is used to express the input question.
Or “Which of the following is correct”. This QA system
2.3 Database Search provides answers in various forms like text documents, Xml
Here the search of the possible results is done in the stored documents or Wikipedia. The common levels that are used by
database, the related results that satisfy the given query with different web based Question Answering systems
selected keyword and rules are sent to the next stage. architectures are as follows [10]:

2.4 Related Document Question Classification: This level gives correct answers
The result which was generated by the previous stage is stored by classify the user query into one of the question type to
as a document. which it belongs to. The question classification is made to
provide better accuracy in the results.
2.5 Answer Display
The result is stored as a document which is in wx format . Answer Extraction : This level extracts the correct
Then the result is converted into required text which is possible answers for different classification of questions.
required by the user and displayed to the user.
Answer Selection: Among the possible answers obtained,
ranking approaches are used to find out the best accurate
3. TYPES OF QA SYSTEMS answers based on its weightage factor.
Different types of QA systems which are divided into two Answer classes generally is of factoid and non - factoid types.
major groups based on the methods used by them. First group The factoid is getting short fact based answers like names,
of QA system belongs to simple natural language processing dates, and non-factoid is getting descriptions or
and information retrieval methods, while another group of QA definitions[27].
systems are dependent upon the reasoning with natural Given a user's natural language question, the system will
language. submit the question to a search engine, then extract all
The two QA systems are compared with characteristics of possible answers from the search results according to the
different dimension such as techniques used, question that question type identified by the question classification module,
finally select the most similar answers to return. The

2
 International Journal of Computer Applications (0975 – 8887)
Volume 53– No.4, September 2012

architecture of web based question answering system is shown Question processor which is taking the question as input
in figure 2[18]. and generates asking point for the question which in turn
helps to match for the answer in the text.

Text Processor retrieve named entities keywords from the

text to produce accurate results. Some of the IR systems like
AskJeeves, LaSiE system performs text analysis which use
some of the basic modules like Tokenizer, Sentence splitter,
Parse process, Name matcher, Discourse Interpreter.

The IR/IE based QA systems depends upon the knowledge

base which requires an extension to CE and GE components
to handle yes/no types of questions in the text. This systems
can answer only wh-type of questions but other than wh-type
of questions such as “How can I assemble a computer?” are
not answered. The Architecture of IR/IE based question
answering system is given in figure 3[19].

Figure 2. Architecture of Web based question answering

system

The design of our question answering system was encouraged

by the goal of exploiting the large amounts of text data that is
available on the Web and elsewhere as a useful resource [21].
Huge amounts of data provide several sources of redundancy
that our system capitalizes on. Answer redundancy such that Figure 3. Architecture of IR/IE based question answering
multiple, differently phrased, answer occurrences enable us to
use only simple query rewrites for matching, and facilitates
the extraction of candidate answers. 3.3 Restricted Domain Question Answering
systems
3.2 IR / IE Based Question Answering This type of Question answering system requiring a linguistic
Systems support to understand the natural language text in order to
Most of the IR based QA systems is returning a set of top answer the questions accurately. An efficient approach for
ranked documents or passages as responses to the query. improving the accuracy of QA system was done by restricting
Information Extraction (IE) system is using the natural the domain of questions and the size of knowledge base which
language processing (NLP) systems to parse the question or resulted in the development of restricted domain question
documents returned by IR systems, yielding the “meaning of answering system (RDQA). This system have particular
each word”. IE systems need several resources like Named characteristics like “System must be Accurate” and “Reducing
Entity Tagging (NE), Template Element (TE), Template the level of Redundancy”. RDQA overcomes the difficulties
relation (TR), Correlated Element (CE), and General Element incurred in open domain by achieving better accuracy. Early
(GE). IE systems architecture is build into distinct levels: RDQA systems like LUNAR allows to ask geologist
questions about rocks. BASEBALL is another restricted
Level 1 NE tagger is use to handle named entity elements domain QA system, which can only answer about one
in the text (who, when, where, what etc..,). season’s Baseball data. These early systems has encoded large
amount of domain knowledge in data bases.
Level 2 handles NE tagging + adj like (how far, how long Question answering on restricted domains requiring the
,how often etc..,), processing of complex questions and offering the opportunity
to carry out complex analysis of the text sources and the
Level 3 builds the correlated entities by using the most questions. The main difference between open-domain
important entity in the question and prepares General question answering and restricted-domain question answering
Element(GE) which consists of asking point of view. For Eg: is the existence of domain-dependent information that can be
“Who won the first Nobel Prize in Literature?” The used to improve the accuracy of the system[28].
ASKING POINT is clearly defined i.e. Person (Noun) if we
by passing this question into the separate levels which was A question is linguistically analysed by the Heart of Gold
mentioned above. (HoG) NLP architecture, which flexibly integrates deep and
shallow NLP components, for instance, PoS tagger, named
KEYWORDS such as won, noble, prize etc.., are retrieved. entity recognition and HPSG parser. The semantic
The architecture of IE systems consisting of two common representations which was generated by the Heart of Gold are
modules, they are then interpreted and a question object is generated that
contains a proto query. This proto query can be viewed as an
implementation-independent, ‘higher level’ representation of
a database or ontology query. From this, an instance of a

3
 International Journal of Computer Applications (0975 – 8887)
Volume 53– No.4, September 2012

specific database or ontology query is constructed. From the normal context. The perspectives of these types of questions
result(s) returned by the queried information source, an may fluctuate but the common goal is to obtain accurate
answer object is generated which forms the basis for answer from the system. This section presents a classification
subsequent natural language answer generation. This is shown of different levels of Questioners.
in figure 4 [15].
CASUAL QUESTIONERS: In this type of questioners
normal questions are pose to the system. Majorly it focus in
normal “perspective” to handle the questions like Eg:“ When
he was born?” and “who invented telephone?” . All these
type of questions are having normal context.

TEMPLATE QUESTIONERS: In this type of questioners,

templates are generated for the given question, which focuses
on the “linguistic” knowledge of the question. For Eg: “How
Akshay manage to complete a task?” and “Does any
specific reason to invent bulb?”

CUBE REPORTER: In this type of questioners the complex

questions are broken down into small set of questions. It
majorly consists of context and specific relations to answer
the questions of this type. The QA system needs to search
answers from multiple sources which lies beyond the database
search. It can answer the questions like Eg: “Does any
specific actions performed by US government after
Figure 4. Architecture of Domain Restricted question Lincoln’s death?”. Cube reporter generates small set of
answering system questions which are associated to the chief question, that are
Eg:
“When did Ram died?” “What was the reason behind his
3.4 Rule Based Question Answering death?” and
Systems “What was released by Indian government after Gandhi’s
death?”.
The rule based QA system is an extended form for IR based
QA system. Rule Based QA doesn’t use deep language
PROFESSIONAL INFORMATION ANALYST: These
understanding or specific sophisticated approaches. A broad
questions are having future perspectives. It is used to
coverage of NLP techniques are used in order to achieve
identifies different taxonomies and multiple facts which are
accuracy of the answers retrieved. Some popular rule based
involved in the questions, but it requires much reasoning
QA systems such as Quarc and Noisy channel generates
techniques for answering, the questions like E.g.: “What are
heuristic rules with the help of lexical and semantic features in
the actions done by Indian government to honour
the questions. For each type of questions it generates rules for
Mahatma Gandhi?”-panel menu-selection schemes or
the semantic classes like who, when, what, where and Why
Use Natural Language?
type questions. “Who” rules looks for Names that are mostly
Nouns of persons or things. “What” rules focuses on generic 4. Question answering system based on
word matching function shared by all question types it information retrieval
consists of DATE expression or nouns. “When” rules mainly Currently, the accessible information, predominantly obtained
consists of time expressions only.“Where” rules are mostly through the Internet is gradually increasing. The most
consisting of matching locations such as “in”, “at’, “near” and significant way to access the information is through
inside. “Why” rules are based upon observations, that are information retrieval (IR) systems. IR system takes a user’s
nearly matching to the question. These Rule Based QA query as input and returns a set of documents sorted by their
systems first establish parse notations and generate training relevance to the query. Some standard technologies are used
cases and test cases through the semantic model. This system to perform the IR task such as existing web search engine like
consists of some common modules like IR module and (Google, Askme, Alta vista etc...).
Answer identifier or Ranker Module. Question answering is an information retrieval task
constrained by an expression of all or a part of the information
IR module: It gives the set of documents or sentences that need as a set of natural language questions or statements. IR
includes the answers to the given question and returns the systems are usually based on the segmentation of documents
results back to the ranker module. and queries into index terms, and their relevance is computed
according to the index terms they have in common, as well as
Ranker Module: Assigning ranks or scores to the according to other information such as the characteristics of
sentences which are retrieved from IR module. the documents, for instance number of words, hyperlink
between papers.
Answer Identifier: It identifies the answer substrings from The number of document returned by the IR system huge
the sentences based upon their score or rank. means paragraph filtering concept has used to reduce the no of
candidate document and to reduce the amount of candidate
3.5 Classification of Questioners Levels text from each document[5]. The steps involved for QA
In Question Answering system the questions are classified system based on information retrieval is given below:
into different levels based upon its context. The questions may
be assertive, informative, interrogative or interactive in

4
 International Journal of Computer Applications (0975 – 8887)
Volume 53– No.4, September 2012

4.1 Filtering candidate document QA system take a natural language question as input, convert
The idea of paragraph filtering is based on the principle that the question into a query and forwards it to an IR module.
the most relevant documents should contain the question When a set of appropriate documents is retrieved, the QA
keywords in a few neighboring paragraphs, rather than system extracts an answer for this question. There are
dispersed over all documents. To exploit that idea, the different methods of identifying answers. One of them make
position of the set of question keywords in each document use of a predefined set of entity classes. Given a selected
was examined. If the keywords are all found in some set of N question, the QA system classifies it into those classes based
successive paragraphs, then that set of paragraphs will be on the type of entity it is looking for, identifies entity
returned, otherwise, the document is rejected from further instances in the documents, and selects the most expected one
processing. ‘N’ is again a configurable number that could be from all the entities with the same class as the question. There
tuned based on an evaluation of system performance under are different types of methods available for classify the
changed tolerances of keyword distance in documents. question. In the following section we are going to discuss
important technique for question classification. Such as
identification of question pattern, semantic approach for
4.2 Identifying quality of the document question classification, sub tree kernel using support vector
To estimate the quality of the selected paragraph quality machine to improve the performance of the question
component has used. If the quality of paragraphs is deemed to classification.
be inadequate, then the system returns to the question
keyword extraction module, and alters the heuristics for oFunctional Word Questions: All Non-Wh questions
extracting keywords from the question. Then the IR can
(except how) fall under the category of Functional Word
performed by using new set of key word retrieved from
Questions. These questions generally start with non-
scratch. The reason of re-determining question keywords
significant verb phrases.
stems from including either too many or too few candidate
Example: Name the Ranger who was always after Yogi Bear.
paragraphs after paragraph filtering. In either case, new
queries for the information retrieval system are produced by
When Questions: When Questions starts with ‘‘When”
revisiting the question keywords component, and either
keyword and are temporal in nature. The general pattern for
adding or dropping keywords. This feedback loop offers some
When Questions is When (do|does|did|AUX) NP VP X”,
form of retrieval context that ensures that only a ‘reasonable’
where AUX, NP, and VP auxiliary verbs, noun phrases, and
number of paragraphs are passed onto the Answer Processing
Verb phrases. ‘|’ indicates Boolean OR operation and ‘X’ can
module. Like several other parameters, exactly how many
be any combination of words playing insignificant role in
paragraphs constitute a ‘reasonable’ number should be
answer type determination.
configured, based on performance testing. Next paragraph
Example: When did Israel begin turning the Gaza Strip and
ordering is to rank the paragraphs according to a plausibility
Jericho over to the PLO?
degree of containing the correct answer.
Where Questions: ‘‘Where Questions” starts with Where
4.3 Standard radix sort algorithm for keyword and are related to the location. These may represent
natural entities such as mountains, geographical boundaries,
paragraph ordering manmade locations such as temple, or some virtual location
This algorithm uses different scores to order the paragraph.
such as Internet or fictional place. The general pattern for
The number of words from the question that are recognized in
Where Questions is Where (do|does|did| AUX) NP VP X?”
the identical sequence within the recent paragraph window,
Example: Where is Italy?
the number of words that separate the majority of distant
keywords in the current paragraph window and the number of
unmatched keywords in the recent paragraph window. Which Questions: The general pattern for Which Questions
Paragraph window is defined as the smallest span of text is Which NP X”? The expected answer type of such questions
required to capture each maximally inclusive set of question is decided by the entity type of the NP.
keywords within each paragraph. Radix sorting is performed Example: Which company manufactures sports kit?
for each paragraph window among all the paragraphs. s using
special purpose data languages. Most important, Who/Whose/Whom Questions: Questions falling under this
category have general pattern(Who|Whose|Whom)
4.4 Lexical and Syntactic Knowledge for IR [do|does|did|AUX] [VP] [NP] X? Here [word] indicates the
In our suggestion we adopt the format of parsing the query to optional presence of the term word in the pattern. These
acquire the set of query terms to calculate the TP information, questions usually ask about an individual or an organization.
instead of calculating TP among all possible combinations of Example: Who wrote ‘Hamlet’?
query pairs, but we vary from previous approaches in the
following three points: first we do not carry out a full parsing Why Questions: Why Questions always ask for certain
of the query but chunking the queries into sets of simple reasons or explanations. The general pattern for Why
phrases such as noun, prepositional phrases and sequences of Questions”\ is ‘‘Why [do|does|did|AUX] NP [VP] [NP]” X”.
verbs .In order to reach a more consistent behavior for Example: Why do heavier objects travel downhill rapidly?
different queries, we apply different TP measures depending
on the lexical type of each query term. We apply TP measures How Question: ‘‘How Questions” have two types patterns of
to phrases as well as terms because phrases represent the syntax: ‘‘How [do/does/did/AUX] NP VP X?” or ‘‘How
concepts expressed in a text more accurately than single [big|fast|long|many|much|far] X?” For the first pattern, the
words. answer type is the explanation of some process while second
pattern return some number as a result.
4.5 Question Classification Example: How did the jack gets its name?
Question answering is an alternate of information retrieval,
which retrieves detailed information rather than documents. A

5
 International Journal of Computer Applications (0975 – 8887)
Volume 53– No.4, September 2012

What Questions: What Questions have several types of the process of becoming Bangla), and a fairly large number of
patterns? The most general regular expression for What “loan-words” from Persian, Arabic, Portuguese, English and
Questions can be written as ‘‘What [NP] [do/does/did/AUX] other languages. Also a large number of words are considered
[functional-words] [NP] [VP] X? What Questions can ask for to be of unknown etymology. A translation based on
virtually anything. transliteration and a table look-up method is proposed as an
Example: What is considered the costliest disaster for interface to the actual QA task. The implementation part thus
insurance industry? Many What Questions are disguised in the involves transliterating a Bangla question as an equivalent
form of ‘‘Functional Word Questions”. Latin alphabet (English) version that could be used in an
actual QA task. The Bangla lexicon consists of a good number
of “loan-words” from Arabic, Persian, English and other
5. MULTI-STREAM languages. And most of them are pronounced almost the same
QUESTIONANSWERING way as would be pronounced in the original language. Entire
The selection of the final answer is complicated by the fact work can be divided into two components, the translation
that the final answer has to be selected from various pools of based on transliteration with table look-up and the question
ranked candidates found by different streams[25]. In other answering part[16].
words, the selection of the correct answer from a given set of
replies corresponding to different QA systems. In particular, it
propose a supervised multi-stream approach that decides
about the correctness of answers based upon a set of features
that describe: (i) the compatibility between question and
answer types, (ii) the redundancy of answers across streams,
as well as (iii) the overlap and non-overlap information
between the question–answer pair and the support text[14].

The general scheme of the proposed multi-stream QA

approach. It consists of two chief stages. In the first stage,
called QA stage, several QA systems extract—in parallel—a
candidate answer and its corresponding support text for a
given question. Then, in the second stage, called selection
stage, a classifier evaluates all candidate answers and assigns Figure 5. Components of the System
to each of them a category (correct or incorrect) as well as a
confidence value (ranging from 0 to 1). At the end, the correct An approach to transform the Bangla question could be
answer having the highest confidence value is selected as the
final response. In the case that all answers were classified as • tokenizing the transliterate version of the Bangla question,
an incorrect result, the system returns a nil response. • using translation based upon the transliteration to translate
the named entities (medical terms)
• translating the remaining question by a simple table look-up
6. QUESTION ANSWERING SYSTEM method
FOR INDIAN LANGUAGES
HINDI LANGUAGE Hindi QA system research attempts to 7. APPLICATIONS OF QUESTION
deal with a wide range of question types like when, where,
ANSWERING SYSTEM
what time, how many. The developed Question-Answering Question answering has many applications. We can
system in Hindi is using Hindi Shallow Parser. The shallow subdividing these applications based upon the source of the
parser gives the analysis of the sentence in terms of the answers: structured data (databases), semi-structured data (for
morphological analysis, POS tagging, Chunking etc. Apart example, comment fields in databases) or free text. We can
from the final output, intermediate output of all the individual further distinguish among search over a fixed set of
modules is also available. All outputs are available in Shakti collections, as used in TREC (particularly useful for
evaluation); search over the Web, search over a collection or
Standard Format (SSF).
book, e.g. an encyclopedia or search over a single text, as
done for reading comprehension evaluations. Most of the
TELUGU LANGUAGE Telugu is an important language in companies can use Question Answering techniques internally
India belonging to the Dravidian family. The important for the employees who are searching out the answers for the
component of our QA system is the Dialogue Manager (DM), similar questions. Another application is in education and
to handle the dialogues between user and system. It is medical fields can also find uses for Question Answering in
necessary in generating dialogue for clarifying partially fields where there are frequently asked questions that people
understood questions, resolving Anaphora and Co-reference
wants to search [23].
problems[20].
We can also distinguish between domain-independent
question answering systems and domain specific systems,
BENGALI LANGUAGE The language Bengali is one of
such as help systems. We can even imagine applying question
the Indo-Aryan languages of South Asia with over 200
answering techniques to material in other modalities, such as
million native speakers. Bangla was written in the Brahmi-
annotated images or speech data. Overall, we would expect
derived Bangla script. Bangla underwent a period of vigorous
that as collections become larger and more heterogeneous
Sanskritization that was started in the 12th century and
,finding answers for questions in such collections will become
continued throughout the middle ages. The Bangla lexicon
harder indicates that having multiple answer sources (answer
consists of tatsama (Sanskrit words that have changed
redundancy) increases the likelihood of finding the correct
pronunciation, but have retained the original spelling),
answer for a given question.
tadbhava (Sanskrit words that have changed at least twice in

6
 International Journal of Computer Applications (0975 – 8887)
Volume 53– No.4, September 2012

8. CONCLUSION Answering System”, Journal of World Academy of

Science, Engineering and Technology 42, 2008. 506-513.
The goal of a question answering system is to retrieving
answers to questions rather than full documents or best- [8] Hammo, Bassam., Abu-Salem, Hani. and Lytinen,
matching passages, as most information retrieval systems. In Steven. A Question Answering System to Support the
this paper we discussed some of the approaches used in the Arabic Language.
existing QA system and proposed a new architecture for QA
system retrieve the exact answer. Answering system has [9] Hirachman, L. and Gaizauskas, R. “Natural Language
become an important component of the online education Question Answering: The View From Here”. Journal of
platform. Natural Language Engineering 7 (4). 275{300. c 2001.
Cambridge University Press DOI:
10.1017/S1351324901002807. 275-299.
A survey of different QA techniques have been elaborated.
Question answering system for Indian languages like hindi [10] Guda, Vanitha., Sanampudi, Suresh. Kumar. and
telugu, bengali is discussed. No Punjabi QAS is discovered. Manikyamba, I.Lalkshmi ,”Approaches For Question
Answering Systems” , Vanitha Guda et al. / International
The focus of the system has been mainly on four kind of Journal of Engineering Science and Technology (IJEST)
questions of type What, Where, How many, and what time. ISSN : 0975-5462 Vol. 3 No. 2011. 990-995.
On analysis of the system the overall efficiency of the system
[11] Moreda, Paloma., Llorens Hector., Saquete, Estela. and
was found to be significant.
Palomar, Manuel. “Combining semantic information in
question answering systems” Journal of Information
The next generation of question answering systems will have
Processing and Management 47, 2011.
to take into consideration presently available multimedia data.
870- 885. DOI: 10.1016/j.ipm.2010.03.008. Elsevier.
There exists a mixture of natural language text, images, video,
audio, user added tags, and metadata. On the question side, [12] Ko, Jeongwoo., Si, Luo., and Nyberg Eric. “Combining
users may express their queries using a variety of modalities. evidence with a probabilistic framework for answer
ranking and answer merging in question answering”
Journal : Information Processing and Management 46,
9. ACKNOWLEDGEMENT 2010 541-554. DOI: 10.1016/j.ipm.2009.11.004.
Many thanks to Mr . Vishal Gupta Assistant Professor in
Elsevier.
UIET, Panjab University Chandigarh, for doing this literature
review. [13] Kolomiyets, Oleksander. and Moens, Marie-Francine.
“A survey on question answering technology from an
information retrieval perspective”. Journal of
REFERENCES Information Sciences 181 , 2011.5412-5434. DOI:
[1] Li, DU. Jia. and Fang, YU .Ping. 2010. Towards natural
10.1016/j.ins.2011.07.047. Elsevier.
language processing: A well-formed substring table
approach to understanding garden path sentence. 978-1- [14] Tellez-Valero, Alberto., Montes-y-Gomez, Manuel.,
4244-6977-2/10, IEEE. Villasenor-Pineda, Luis. and Padilla Anselmo Penas.
“Learning to select the correct answer in multi-stream
[2] Suarez, O. S., Riudavets, F. J. C., Figueroa, Z. H., and
question answering”. Journal of Information Processing
Cabrera, A. C. G. “Integration of an XML electronic
and Management,2010. 856 – 869. DOI: 10.1016/j.ipm.
dictionary with linguistic tools for natural language
Elsevier.
processing” Journal of Information Processing &
Management, vol. 43, 2007, 946-957. [15] Frank, Anette., Krieger, Hans-Ulrich., Xu, Feiyu.,
Uszkoreit, Hans., Crysmann, Berthold., Jörg, Brigitte.
[3] Metais, E. “Enhancing information systems management
and Ulrich Schäfer. “Question answering from structured
with natural language processing techniques,” Journal of
knowledge sources”. Journal of Applied Logic 5 , 2007.
Data & Knowledge Engineering, vol. 41, 2002, 247-272.
20 – 48. DOI: 10.1016/j.jal.2005.12.006. Elsevier.
[4] Zhang, Wen., Yoshida,Taketoshi., and Tang, Xijin. 2008.
[16] Haque, Nafid. and Rosner, Mike. A prototype
TFIDF, LSI and Multi-word in Information Retrieval and
framework for a Bangla question answering system using
Text Categorization. International Conference on
translation based on transliteration and table look-up as
Systems, Man and Cybernetics. 1-4244-2384-2/08,
an interface for the medical domain. University of Malta
IEEE.
Gertjan Van Noord, University of Groningen.
[5] Ramprasath, Muthukrishan. And Hariharan,
[17] Zhang Dell. and Lee Sun Wee. A Web-based Question
Shanmugasundram. “A Survey on Question Answering
Answering System.
System”, International Journal of Research and Reviews
in Information Sciences (IJRRIS) Vol. 2, No. 1, 2012, [18] Rodrigo, Alvaro., Perez-Iglesias, joaqum., Penas,
171-178. Anselmo., Garrido, Guillermo. and Araujo,Lourdes. A
Question Answering System based on Information
[6] Sahu, Shriya., Vasnik, Nandkishor., and Roy,Devshri.
Retrieval and Validation.
“Proshanttor : “A Hindi Question Answering
System”International Journal of Computer Science & [19] Reddy, Rami., Reddy, Nandi. and Bandyopadhyay,
Information Technology (IJCSIT) Vol 4, No 2, 2012, Sivaji. Dialogue based Question Answering System in
149-158. Telugu.
[7] Kangavari, Mohammad. Reza., Ghandchi, Samira. and [20] Susan Dumais, Michele Banko, Eric Brill, Jimmy Lin,
Golpour, Manak. “A New Model For Question Andrew Ng “Web Question Answering: Is More Always
Better?”

7
 International Journal of Computer Applications (0975 – 8887)
Volume 53– No.4, September 2012

[21] Zhenqiu, Liang. “Design of Automatic Question [25] Jijkoun, Valentin. and Rijke, Maarten de. “Answer
Answering System Base on CBR”. Journal of Procedia Selection in a Multi-Stream Open Domain Question
Engineering 29, 2011. 981-985. DOI Answering System”.
:10.1016/j.proeng.2012.01.075. Elsevier.
[26] Kwok, Cody.,Etzioni, Oren. and S. Weld, Daniel.
[22] Badia, Antonio. “Question answering and database “Scaling Question Answering to the Web”. ACM
querying: Bridging the gap with generalized Transactions on Information Systems, Vol. 19, No. 3,
quantification”. Journal of Applied Logic 5,2007. 3-19. 2001, 242–262.
DOI:10.1016/j.jal.2005.12.007. Elsevier.
[27] Quarteroni, S. and Manandhar S. “Designing an
[23] Gupta, Vishal. and Lehal, Gurpreet S. “A Survey of Text Interactive Open-Domain Question Answering System”.
Mining Techniques and Applications”. Journal of Journal of Natural Language Engineering 1. 1-23.
Emerging Technologies in web Intelligence, VOL. 1, No.
1. [28] Molla ,Diego. and Vicedo, Jose Luis. “Question
Answering in Restricted Domains: An Overview”.
[24] “Introduction to the special issue on question Association for Computer Linguistics. 41-61.
answering”. Editorial of Information Processing and
Management 47,2011. 805-807. DOI:
10.1016/j.ipm.2011.04.004. Elsevier.

View publication stats