Open domain question answering system using semantic role labeling

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 03 Special Issue: 15 | Dec-2014 | IWCPS-2014, Available @ http://www.ijret.org 104
OPEN DOMAIN QUESTION ANSWERING SYSTEM USING
SEMANTIC ROLE LABELING
Sreelakshmi V1
, Sangeetha Jamal2
1
Department of Computer Science, Rajagiri School of Engineering and Technology, Kochi, India
2
Department of Computer Science, Rajagiri School of Engineering and Technology, Kochi, India
Abstract
The World Wide Web is an attractive source of information retrieval and can be used for seeking simple factual answers for user
questions. Although usual information retrieval systems like search engines helps in finding the relevant documents for a set of
keywords, there are situations where we have more specific information need. Search engines only provide the actual facts as a
ranked list of documents and links that contain the keywords. However, what a user really wants is often a precise answer to a
question. A Question Answering system gives us a precise solution to this scenario. A Question Answering system retrieves a
precise answer to any natural language question posed by the user. A typical QA system uses several language processing
techniques. In this paper, an Open Domain Question Answering that answers simple Wh-questions using online search has been
proposed.
Keywords: Information Retrieval, Question Answering, Semantic Role Labeling
--------------------------------------------------------------------***------------------------------------------------------------------
1. INTRODUCTION
The Question Answering task is one of the major research
area under Information Retrieval (IR) with applications from
Natural Language Processing (NLP). The goal of a Question
Answering system is to deliver an exact answer to the
arbitrary questions formulated by users. In the process of
normal Information Retrieval using search engines, the user
is provided with a ranked list of documents and links, but
not the exact answer. A Question Answering system
addresses this problem. These systems allow a user to ask
questions in everyday language and provide the user with a
quick and precise answer.
Question Answering systems can be classified into two
groups namely closed domain and open domain. Closed
domain Question Answering Systems is restricted to a
specific domain (like medical, music etc.). They also contain
a domain specific ontology. Whereas, an open domain
Question Answering system deals with questions about
nearly everything and can on universal ontology and
information such as World Wide Web.
A typical Question Answering system can be divided into 3
modules namely: Question Processing module, Document
Processing or Information Retrieval module and Answer
Processing module. Each module contains several sub
modules and these modules use several Natural Language
Processing Techniques inorder to extract the proper answer.
The usual Question Answering system is designed to answer
simple wh-questions like “who”, “what”, “when”, “where”,
etc. But the recent QA research focuses on extending the
system to answer complex questions, summary questions,
opinion questions etc. The paper proposes a Question
Answering system that answers simple factoid,
wh-questions by using a technique called Semantic Role
Labeling.
The rest of the paper is organized as follows. The next
section describes the general architecture of a Question
Answering System. Section 3 discusses some of the related
works in this area. The proposed system architecture is
described in section 4. The paper concludes with the
experimental setup and results.
2. QA SYSTEM ARCHITECTURE
A general Question Answering system is composed of three
main modules. They are:
 Question Processing
 Information Retrieval or Document Processing
 Answer Processing module
In the Question Processing module, the user query is
processed. In this phase, the user question is classified based
on the stem word like ‘who’, ‘what’, ‘when’, ‘where’ etc.
and hence the answer type information can be easily
obtained from this. For example, the question “Where is the
headquarters of WHO” looks for a location as answer. In
this phase, the keywords from the questions are also
extracted. This phase is crucial because the answer
extraction strategy completely depends on the information
collected from this module. An optional phase of query
reformulation can also be included in this module.
In the Document Processing module, the question is
question is passed for Information retrieval. This module
uses a search engine for IR and the documents related to the
user questions are saved. Then they are ranked based on
certain heuristics.

_______________________________________________________________________________________
Answer Processing module is the distinguishing feature
between usual IR systems and QA systems. In this module,
the exact answer for the user question is identified and
extracted from the retrieved documents. This phase uses the
query information collected from the first phase. Then the
answer is validated for correctness and returned to the user.
3. RELATED WORKS
Before the emergence of web, Question Answering systems
had a limited audience and were mostly domain restricted.
The first QA system developed was BASEBALL (Green et
al. 1961). It answered domain–specific questions about the
baseball games played in American league. Another early
work was LUNAR system (Woods et al. 1972) used to
access, compare and evaluate the chemical analysis data on
lunar rock and soil composition that was accumulating as a
result of the Apollo moon mission.
When the web became popular, QA systems became open
domain and millions were able to access these systems.
START (Katz, 1997) Information Server built at the MIT
Artificial intelligence Lab is one of the first web-based QA
systems. A new method called Knowledge annotation
approach was introduced in the system.
The work in QA system, for the past few years, has been
flourished by the inclusion of QA task in the yearly Text
Retrieval Conference (TREC), sponsored annually by the
U.S. National Institute of Standards and Technology (NIST)
and the Cross Language Evaluation Forum (CLEF). These
competitions helped to develop and improve the research in
this field.
A semantics based QA system accessible via the web was
developed bt Hovy et al. The system was named
Webclopedia. In the system, the parsed question is expanded
and converted to multi unit words and related documents are
retrieved. Answers are selected using machine learning
based grammar parser.
Radev et al. Proposed a method called Probabilistic Phrase
Reranking (PPR) in their web-based question answering
system (NLQA). The method was fully implemented at the
University of Michigan as a Web-accessible system, called
NSIR which relies on an existing search engine to return
documents that are likely to contain the answer to a user
question.
A general-purpose, fully automated question-answering
system available on the web was MULDER developed by
Kwok et al. MULDERs architecture relies on multiple
search engine queries, natural-language parsing, and a novel
voting procedure to yield reliable answers coupled with high
recall.
Harabagiu et al proposed a system called FALCON that was
presented in TREC-9. The system used a method called
boosting, which integrates different forms of syntactic,
semantic and pragmatic knowledge for the goal of achieving
better performance. The answer engine handles question
reformulations, finds the expected answer type from a large
hierarchy using WordNet semantics and extracts the answers
after performing unifications on the semantic forms of the
question and its candidate answers. FALCON also used
Named Entity recognition technique and the answers were
extracted from semantically rich passages that match the
question type.
Dan Moldovan et al. proposed LASSO, A Tool for Surfing
the Answer Net in TREC 8. In order to find answers in
large collections of documents, novel language processing
techniques were included in the system. The question was
processed by combining syntactic information with semantic
information that characterized the question in which eight
heuristic rules were defined to extract the keywords used for
identifying the answer.
Recent research in QA systems use the ontology based
methods to improve the efficiency. Current works in QA are
based on semantic web. Bollegala et al. proposed an
automatic method to estimate the semantic similarity
between words or entities using web.
Semantic similarity is computed by mapping terms to an
ontology and by examining their relationships in that
ontology. For that, ontologies like Wordnet are used. The
similarity is measured by page counts and by analyzing the
retrieved snippets. Coefficients like WebJaccard ,Web PMI,
WebOverlap etc. can be used to find the semantic similarity
between question answer pair, and hence improve the
performance.
4. PROPOSED SYSTEM
A simple Question Answering system was implemented
using the technique called Semantic Role Labeling. The
system consisted of three phases called Query Processing,
Document Processing, Answer Processing. In each of these
phases, several language processing components were used.
The system is web based and hence uses the search engine
to extract information from the web.

_______________________________________________________________________________________
Fig -1: Proposed QA System Working Example
The first phase is to identify the keywords from the user
question. In this phase, the user question is analyzed and
classified based on the stem word. From the question stem
word, The expected answer type information can be
obtained.
The next phase is Information Retrieval using search engine.
The user question is passed on to an online search engine
and the documents related to query are retrieved. These
documents undergoes some linguistic techniques like POS
tagging, Parsing etc.
The third and final phase is the Answer Processing phase,
which distingushes question answering systems from the
usual text retrieval systems. The proper answer for the user
question is identified and extracted from the retrieved
documents. The query information collected from the first
phase can be used for validating the answer. For identifying
the exact answer, a technique called semantic role labeling
is been used.
4.1 Semantic Role Labeling
The technique called SRL or Semantic Role Labeling is
used to label the semantic roles in a sentence. Also known
as Shallow Semantic Parsing, Case Role Analysis, Thematic
Analysis etc., this technique has applications in the research
areas of Question Answering systems, Text summarization,
Machine Translation etc. The goal is to identify all the
constituents that fill a semantic role, i.e. to determine the
roles like Agent, Patient, Instrument, Location, Time etc. in
a sentence.
SRL can be used for identifying the exact answer for a user
question. For example, ‘Who’ question uses Agents or
Actors i.e. Persons as the answer. Similarly, ‘Where’
question requires a Location as answer and ‘When’
question uses Time or Temporal tags as answer. By
combining these information, a simple factoid Question
Answering system can be implemented.
5. RESULT
A simple QA system that answers simple factoid questions
was implemented in Python. For the purpose of Semantic
Role Labelling, SENNA [20] tool was used. The
information retrieval was done using Google search engine.
The proposed system performance was compared with a QA
system that used simple pattern matching technique for
answer extraction. Results shows that the proposed system
that uses SRL has got more accuracy than a normal pattern
matching QA system. The results can be seen in the
following table.
Chart -1: Result Comparison
The accuracy was calculated as:
Accuracy = (Correct Answer / Total Questions)
The system was tested for 4 types of questions namely
‘who’, ‘what’, ‘when’ and ‘where’. The result was classified
into two types. If the system provides the necessary
information, then it was classified as a correct answer. And
if the retrieved result does not contain the necessary
information, or if the answer was not produced, then it is
classified as a wrong answer. The answer size varied from a
word , a few words, to a sentence based on the question
type. It can be observed that the QA system accuracy has
improved by using the semantic role labeling technique.
6. CONCLUSION
The wealth of information on the web makes it an attractive
source for seeking information. Retrieval of accurate
information like quick answers to simple factual questions
from web is done with the help of a QA system. A Question
Answering system provides the user a precise answer for the
wh-question. The question Answering systems helps the

_______________________________________________________________________________________
user to get a quick and precise answer for simple wh-
questions. For this purpose, QA systems heavily use
language processing components like Parser, Tagger etc. In
this paper, a simple QA system was implemented using a
technique called Semantic Role Labeling and the
performance of the system is compared with a QA system
that uses pattern matching. As a part of future enhancement,
the system can be extended for answering complex
questions.
REFERENCES
[1]. M Ramprasath, S Hariharan Improved Question
Answering System by semantic reformulation, IEEE- Fourth
International Conference on Advanced Computing, 2012.
[2]. Ali Mohamed Nabil Allam, and Mohamed Hassan
Haggag, The Question Answering Systems: A Survey,
International Journal of Research and Reviews in
Information Sciences (IJRRIS), September 2012 Science
Academy Publisher, United Kingdom.
[3]. B.F. Green , A.K. Wolf, C. Chomsky, and K. Laughery,
Baseball: An automatic question answerer, In Proceedings
Western Computing Conference, 1961.
[4]. Woods W., Progress in natural language understanding
an application to lunar geology, American Federation of
Information Processing Societies (AFIPS) Conference
Proceedings 1973.
[5]. Hovy E., Gerber L., Hermjakob U., Junk M., and Lin
C.Y., Question answering in webclopedia ; The Ninth Text
REtrieval Conference (TREC 9), 2002.
[6]. Moldovan D., Lasso: A Tool for Surfing the Answer
Net, In Proceedings of the Eighth Text Retrieval Conference
(TREC-8).
[7]. Text Retrieval Conference (TREC), http://trec.nist.gov
[8]. Cross Language Evaluation Forum (CLEF)
(http://www.clefcampaign.org/).
[9]. Deepak Ravichandran, Eduard Hovy, Learning Surface
Text Patterns for a Question Answering System , In
Proceedings of the ACL Conference, 2002.
[10]. Soubbotin M. M., Patterns of potential answer
expressions as clues to the right answers, In NIST Special
Publication, The Tenth Text Retrieval Conference TREC-
10, Gaithersburg 2002.
[11]. L. Hirschman, R. Gaizauskas, Natural language
question answering: the view from here, Natural Language
Engineering 7, 2001 Cambridge University Press.
[12]. Marco De Boni, Suresh Manandhar, Implementing
clarification dialogues in open domain question answering,
Natural Language Engineering, Pages 343-361, 01/2005.
[13]. Danushka Bollegala, Yutaka Matsuo, and Mitsuru
Ishizuka , A Web Search Engine-Based Approach to
Measure Semantic Similarity between Word , IEEE
Transactions On Knowledge And Data Engineering, July
2011.
[14]. Abney S., Collins M., and Singhal A., Answer
extraction , In the Proceedings of ANLP 2000.
[15]. A Harabagiu, Dan Moldovan, Marius Paaca, Rada
Mihalcea, Mihai Surdeanu, Rzvan Bunescu, Roxana Girju,
Vasile Rus, Paul Morarescu, FALCON: Boosting
Knowledge for Answer Engines , In NIST Special
Publication 500-249:The Ninth Text REtrieval Conference
(TREC 9)
[16]. Kwok C., Etzioni O., and Weld D. S., "Scaling
question answering to the web", In the Proceedings of the
10th World Wide Web Conference (WWW 2001), Hong
Kong.
[17]. Kangavari M., Ghandchi S. & Golpour M, Information
Retrieval: Improving Question Answering Systems by
Query Reformulation and Answer Validation, World
Academy of Science, Engineering and Technology, 2008.
[18]. Boris Katz,From Sentence Processing to Information
Access on the World Wide Web , AAAI Technical Report,
1997
[19]. Sreelakshmi V, Sangeetha Jamal, Web Based Question
Answering System using Pattern Matching , in International
Conference on Information Science - ICIS, July 2014,
College of Engineering Cherthala.
[20]. Ronan Collobert, Weston , L. Bottou, M. Karlen, K.
Kavukcuoglu and P. Kuksa. , Natural Language Processing
(Almost) from Scratch, Journal of Machine Learning
Research (JMLR), 2011.
[21]. R. Collobert, J. R. Collobert. Deep Learning for
Efficient Discriminative Parsing, in International
Conference on Artificial Intelligence and Statistics
(AISTATS), 2011.
[22]. D. Gildea, D. Jurafsky, Automatic labeling of semantic
roles , Computational Linguistics.
[23]. S. Narayanan, Harabagiu, Question Answering on
Semantic Structures, In Proceedings of the 19th
COLING.
[24]. Sreelakshmi V, Sangeetha Jamal, Survey Paper :
Question Answering Systems, in National Conference on
Computing and Communication - (NCCC ), March 2014,
GEC Idukki.
[25]. Poonam Gupta, Vishal Gupta, “A Survey of Text
Question Answering Techniques”, International Journal of
Computer Applications, Volume 53, September 2012.
[26]. SENNA SRL Tool[Online], Available : http://ml.nec-
labs.com/senna/
[27]. Li X. , Roth D. , “Learning question classifiers” , In
Proceedings of the 19th International Conference on
Computational Linguistics (COLING 2002), p.556-562.
[28]. Dragomir Radev, Weiguo Fan, Hong Qi, Harris Wu,
Amardeep Grewal, “Probabilistic Question Answering on
the Web”, Proceedings of the 11th international conference
on World Wide Web, ACM 2002
BIOGRAPHIES
Sreelakshmi V is a postgraduate
engineering student in Computer Science
and Engineering-Information Systems at
Rajagiri School of Engineering and
Technology, Kochi, India. She completed
her graduation in Computer Science and
Engineering from the University of Calicut.
Ms. Sangeetha Jamal is currently working
as Assistant Professor in the Department
of Computer Science and Engg., RSET,
Cochin, Kerala. She has completed her
M.Tech Degree in Software Engineering
from Cochin University of Science &
Technology, and her B.Tech Degree in Computer Science &
Engg from the University of Calicut. She is currently doing
her Ph.D. in Natural Language Processing.

Open domain question answering system using semantic role labeling

More Related Content

What's hot (17)

Viewers also liked (20)

Similar to Open domain question answering system using semantic role labeling (20)

More from eSAT Publishing House (20)

Recently uploaded (20)

Open domain question answering system using semantic role labeling