Word sense disambiguation using wsd specific wordnet of polysemy words

International Journal on Natural Language Computing (IJNLC) Vol. 3, No.4, August 2014
WORD SENSE DISAMBIGUATION USING WSD
SPECIFIC WORDNET OF POLYSEMY WORDS
Udaya Raj Dhungana1, Subarna Shakya2, Kabita Baral3 and Bharat Sharma4
1, 2, 4Department of Electronics and Computer Engineering, Central Campus, IOE,
Tribhuvan University, Lalitpur, Nepal
3Department of Computer Science, GBS, Lamachaur, Kaski, Nepal
ABSTRACT
This paper presents a new model of WordNet that is used to disambiguate the correct sense of polysemy
word based on the clue words. The related words for each sense of a polysemy word as well as single sense
word are referred to as the clue words. The conventional WordNet organizes nouns, verbs, adjectives and
adverbs together into sets of synonyms called synsets each expressing a different concept. In contrast to the
structure of WordNet, we developed a new model of WordNet that organizes the different senses of
polysemy words as well as the single sense words based on the clue words. These clue words for each sense
of a polysemy word as well as for single sense word are used to disambiguate the correct meaning of the
polysemy word in the given context using knowledge based Word Sense Disambiguation (WSD) algorithms.
The clue word can be a noun, verb, adjective or adverb.
KEYWORDS
Word Sense Disambiguation, WordNet, Polysemy Words, Synset, Hypernymy, Context word, Clue Words
1. INTRODUCTION
Words that express two or more different meanings when used in different contexts are referred to
as polysemy or multi-sense words. Every natural language contains such polysemy words in its
vocabulary. Such polysemy words create a big problem during the translation of one natural
language to another. To translate the correct meaning of the polysemy word, the machine must
first know the context in which the polysemy word has been used. Only after this, the machine
can find out the correct meaning of the word in the particular context and can translate the
meaning of that word into the correct word in another language. The process of finding the
correct meaning of the polysemy word using machine by analyzing the context in which the
polysemy word has been used is referred as Word Sense Disambiguation (WSD).
Although different methods have been tested to find the correct sense of the polysemy word,
accuracy at satisfactory level has not been obtained yet. Among the different methods used for
WSD, this research focused its study on the knowledge-based approach. The knowledge-based
approaches use the resources such as dictionaries thesauri, ontology, collocation etc to
disambiguate a word in a given context [1].
These days, the WordNet is becoming popular as a resource to be used in knowledge-based
approach to disambiguate the meanings of polysemy words. WordNet is a lexical database
developed at Princeton University for English language [2]. The WordNet organizes nouns, verbs,
adjectives and adverbs into the groups of synonyms and describes the relationships between these
DOI : 10.5121/ijnlc.2014.3405 73

synonym groups forming a semantic network among the words. After the development of English
WordNet, many other WordNet on other languages such as Spanish WordNet, Italian WordNet,
Hindi WordNet etc were built. These WordNet are used as one important resource to
disambiguate the different meanings of a polysemy words in respective languages. To
disambiguate the meaning of a polysemy word using the WordNet, the related words from synset,
gloss and different levels of hypernym are collected from the WordNet database and these related
words are compared to find the overlaps using different WSD algorithm. However, the collection
of related words from synset, gloss and different level of hypernym that are taken from the
WordNet contains only very few words that can be used to disambiguate the correct sense of a
polysemy word in the given context [3]. This increases only the computational overhead for the
WSD algorithm and for the system. Moreover, according to [3] as we go increasing the levels of
hypernym to collect the related words, the sense of the words becomes more general and the two
different senses of a polysemy word are found to have the same hypernym. This therefore creates
another ambiguity in ambiguity. To overcome this problem, we have developed a new model to
organize both the single sense and multi-sense words.
74
2. RELATED TASKS
In 1986, Lesk Michael [4] developed an algorithm called Lesk algorithm to identify senses of
polysemy words. He used the overlap of word definition from the Oxford Advanced Learner’s
Dictionary of Current English (OALD) to disambiguate the word senses.
Banerjee and Pedersen [5] adapted the original Lesk algorithm to use the lexical database
WordNet. They used Senseval-2 word sense disambiguation exercise to evaluate their system and
the overall accuracy was found to be 32%. To compare two glosses, they used the longest
sequence of one or more consecutive words that occurs in both glosses. For each overlap, a square
of the number of words in the overlap is calculated and the final score is the sum of all overlaps.
Sinha M. et. al. [6] developed automatic WSD for Hindi language using Hindi WordNet at IIT
Bombay. They used statistical method for determining the senses. The system could disambiguate
the nouns only. They compared the context of the polysemy word in a sentence with the contexts
constructed from the WordNet. They evaluated the system using the Hindi corpora provided by
the Central Institute of Indian Languages (CIIL). The accuracy of their algorithm was found in the
range from 40% to 70%. They used simple overlap method to determine the winner sense.
In [7], Shrestha N. et. al. used the Lesk algorithm to disambiguate the Nepali ambiguous words.
They modified the Lesk algorithm in such a way that only the words in the sentence without their
synset, gloss, examples and hypernym are taken as context words. Each word in the context
words is compared with each word in the collection of words formed by the synset, gloss,
examples and hypernym of each sense of the target word. They did not include the synset, gloss,
example and hypernym of the context words in the collection of context words. Moreover, the
number of example for each sense of the target word was only one.
In [3], Dhungana and Shakya used the adapted Lesk algorithm to disambiguate the polysemy
word in Nepali language. The experiments performed on 348 words (including the different
senses of 59 polysemy words and context words) with the test data containing 201 Nepali
sentences shows the accuracy of their system to be 88.05%. This accuracy is found to be
increased by 16.41%, in compared to the accuracy of the system developed by [7].
Context word refers to the word that is used in context with the target word. The target word is
the word which has multiple senses at different contexts and its meaning needs to be
disambiguated in the given context. Hypernymy is a semantic relation between two synsets to

show super-set hood (is a kind of) relation. For example, (Transliteration: Belpatra) is a
kind of (Transliteration: Paat, Transliteration in English: Leaf). Therefore, is a
hypernymy and is hyponymy [8].
75
3. STATEMENT OF RESEARCH PROBLEM
3.1. Problem Statement
From analyzing many research works on WSD using WordNet such as [4], [5], [6], [7], [9] and
[3], it was noticed that the WordNet is very useful resource to use for word sense disambiguation.
However, it is not exactly suitable for knowledge-based, overlap selection WSD approaches. The
reason is that the WordNet is built for general purpose in NLP tasks but is not focused for WSD.
WordNet contains huge amount of information for words that are arranged with semantic relation.
In all WSD methods, the WordNet is used to take a large number of words to disambiguate the
meaning of multi-sense word. However, only very few words taken from the WordNet are used to
disambiguate the different senses of a multi-sense word. In this sense, all the efforts such as
processing time for CPU and memory to store large number of unused words are wasted.
Furthermore, it is noticed that the words taken from the WordNet to disambiguate the multiple
meaning of the multi-sense word itself creates the ambiguity resulting in decrease in accuracy.
Another important point noticed from [3] is that when deeper levels of the hypernymy from the
WordNet are used, the correctly disambiguated polysemy words are also incorrectly
disambiguated. The reason is that the deeper (probably from second and/or third) levels of
Hypernymy of all the different sense of a polysemy word are same as shown in Fig.1. That’s why,
there is no any basis to disambiguate the different sense of the polysemy word.
Example: WSD and Hypernymy in English WordNet: Let us consider a polysemy word Pen in
English language. In WordNet, the word has five different meanings as a noun. The five
meanings of the word Pen with the different levels of hypernymy are shown in Fig. 1. First look
at the Fig.1b and Fig.1c. Here, we can see that the only the meaning of two different words are
different. The hypernymy of these two senses are the same. In such case, there is no meaning of
using the hypernymy of the two different senses of the polysemy word to disambiguate their
meanings. Inclusion of these hypernymy is just waste of computational effort and waste of
memory for the system.
3.2. Solution Approach
What we need is that the organization of the polysemy words must be organized in such a way
that it should contain only those words that are sufficient to disambiguate the different senses of a
polysemy word and it should not again introduce ambiguity as it is in the case of using WordNet.
Although WordNet is very useful lexical resource that can be used to disambiguate the different
meanings of a polysemy word, it has still some limitations to use in WSD algorithms as discussed
in previous section. Therefore, to improve the accuracy of the knowledge-based overlap selection
WSD algorithms dramatically, we need to develop a new logical model that organizes the words
in such a way that it disambiguates the meanings of a polysemy word correctly and efficiently
resulting in higher accuracy than that can be obtained from the use of WordNet.

76
3.2.1. Organization of Clue Words
Unlike in WordNet, we organized the different senses of polysemy word as well as single sense
words. We grouped each sense of a polysemy word based on the verb, noun, adverb and adjective
with which the sense of the polysemy word can be used in a sentence. The organization of the
single sense is also done in the same way. This organization of words results in a new model of
WordNet which is focused on WSD. This new model of WordNet contains all the necessary and
sufficient words/information that can be used to disambiguate the senses of a polysemy word. It
restricts the unnecessary words/information and therefore does not create ambiguity in ambiguity
like WordNet. For instance, a sample organization of different meanings of a polysemy word Pen
in English language as a noun is shown in Fig. 2.
Figure 2. Organization of different senses of noun Pen in new model of WordNet based on clue words
3.2.1. Algorithm
The algorithm is very simple. First, we find collection of clue words for each sense of a polysemy
word. The number of collection of words for a polysemy word is equal to the number of different
senses of the polysemy. Let say these collection of words be T1, T2, T3,...,Tn where n is the
number of different senses that the polysemy word has. After this, we find the collection of words
(say C) from the context window.
4. INTENDS
The main objectives of this research work include:
1. To organize the senses of polysemy words based on clue words.
2. To develop a new model of WordNet based on clue words for both polysemy and
single sense words.
3. To test whether using the conventional or using the new model WordNet, the WSD
algorithm gives the higher accuracy.

77
5. HYPOTHESIS
The conventional WordNet is very useful resource for language learning and language processing
tasks. However, it is not suitable for knowledge-based WSD algorithms to get higher accuracy.
We belief that organizing the different senses of polysemy words and context words based on
clue words best suits for knowledge-based WSD algorithms resulting in higher accuracy.
Moreover this reduces the computational effort for the system and saves the system memory
while processing.
6. RESEARCH METHODOLOGY
Depending upon the nature of our research, we used experiment type of research strategy. We
used the algorithm and test data already developed by [3] to test our new model of WordNet for
WSD. Based on the research question, we set up two experiments with different settings. In first
experiment, we used the system developed by [3] using the conventional sample Nepali WordNet.
In second experiment, we only replaced the conventional sample Nepali WordNet by our new
model of WordNet organized with clue words keeping all the other settings constant as in first
experiment.
6.1. Experiment 1
Lexical Resource: In first experiment, we used the conventional sample Nepali WordNet
developed by [3]. This conventional sample Nepali WordNet contains 348 words including 59
polysemy words. The words in this WordNet are organized as in English WordNet.
Experiment Setting: In this setting, we included the synset, gloss, example and hypernym of
each word of in the collection of context words to form the final collection of context words and
compared these words with the collection words of each sense of target word to determine
overlaps. The number of examples for each word provided in this case is more than four in
average. Moreover, we did not include the target word in the collection of context word.
6.2. Experiment 2
Lexical Resource: In this Second experiment, we used our new model of WordNet that is
focused for word sense disambiguation.
Experiment Setting: The experiment setting is same as the setting in Experiment 1 except that
we replaced the conventional sample Nepali WordNet by our new model of WordNet organized
with clue words.
7. RESULT AND DISCUSSIONS
After setting the two experiments, we executed the experiments using the test data developed by
[3]. The test data consists of 201 different Nepali sentences each containing polysemy word. In
experiment 1, we found that the polysemy words in 117 test sentences out of 201 are correctly
disambiguated. This shows the accuracy of the system with conventional sample Nepali WordNet
found to be 88.059%. In experiment 2, we found that the polysemy words in 184 test sentences
out of 201 are correctly disambiguated. At this time, the accuracy of the system is found to be
91.543%. Therefore, the accuracy of the system using our new model of WordNet that is specific

to WSD is found to be increased by 3.484%. The accuracies obtained in the two experiments are
shown in Table 1.
78
Table 1. Heading and text fonts.
Experiment
No
No. Of polysemy words
that are correctly
disambiguated
No. Of polysemy words
that are correctly
disambiguated
Accuracy
in
percentage
1 177 24 88.059
2 184 17 91.543
8. CONCLUSIONS
The WordNet organizes the words in the lexical database based on their meanings of the words
instead of their forms as in dictionaries. It groups the nouns, verbs, adjectives and adverbs
together into synonym sets, each expressing a distinct concept [2]. The words in a synonym set
can be interchangeably used in many contexts. The main relationship among the words in
WordNet is the synonym. From analyzing many research works on WSD using WordNet such as
[4], [5], [6], [7], [9] and [3], we have noticed that the WordNet is not exactly suitable to use with
knowledge-based, overlap selection WSD approaches. The reason is that the WordNet is built for
general purpose in NLP tasks but not focused for WSD. In all WSD methods that used WordNet,
the WordNet is used to take a large number of words to disambiguate the meaning of multi-sense
word. But it is noticed that only very few words taken from the WordNet are used to
disambiguate the different senses of a multi-sense word. In this sense, all the efforts such as
processing time for CPU and memory to store large number of unused words are wasted. To
defeat with these problems of WordNet, we developed a new model of WordNet that organizes
the different senses of polysemy words as well as single sense word using the clue words. The
experiment shows that the accuracy of the system using our new model of WordNet is found to be
increased by 3.484% in compare with the accuracy of the system in [3]. This evidence leads us to
conclude that our new model of WordNet gives the higher accuracy for the knowledge-based
WSD algorithms rather than using the conventional WordNet.
ACKNOWLEDGEMENTS
We would like to kindly express our sincere thanks to Madan Puraskar Pustakalaya (MPP), Patan
Dhoka, Lalitpur for providing Nepali corpus and Nepali online dictionary which formed the main
resources of this research. We are equally thankful to Mr. Balram Prasain, Lecturer, Tribhuwan
University for his valuable help in this research.
REFERENCES
[1] N. Ide and J. Véronis, “Word sense disambiguation: The state of the art,” Computational
Linguistics, pp. 1–40, 1998.
[2] G. A. Miller, R. Beckwith, C. Fellbaum, D. Gross, and K. J. Miller, “Introduction to wordnet: An
on-line lexical database,” International Journal of Lexicography, 1998.
[3] U. R. Dhungana and S. Shakya, “Word sense disambiguation in nepali language,” in The Fourth
International Conference on Digital Information and Communication Technology and Its
Application (DICTAP2014), Bangkok, Thailand, 2014, pp. 46–50.
[4] M. E. Lesk, “Automatic sense disambiguation using machine readable dictionaries: How to tell a
pine cone from a ice cream cone,” in SIGDOC Conference, Toronto, Ontario, Canada, 1986.

[5] S. Banerjee and T. Pedersen, “An adapted lesk algorithm for word sense disambiguation using
wordnet,” in Third International Conference on Intelligent Text Processing and Computational
Linguistics, Gelbukh, 2002.
[6] M. Sinha, M. K. Reddy, P. Bhattacharyya, P. Pandey, and L. Kashyap, “Hindi word sense
79
disambiguation,” Master’s thesis, Indian Institute of Technology Bombay, Mumbai, India, 2004.
[7] N. Shrestha, A. V. H. Patrick, and S. K. Bista, “Resources for nepali word sense disambiguation,”
in IEEE International conference on Natural Language Processing and Knowledge Engineering
(IEEE NLP-KE’08), Beijing, China, 2008.
[8] P. Bhattacharyya, P. Pande, and L. Lupu, “Hindi wordnet,” Indian Institute of Technology
Bombay, Mumbai, India, Tech. Rep., 2008.
[9] N. Shrestha, A. V. H. Patrick, and S. K. Bista, “Nepali word sense disambiguation using lesk
algorithm,” Master’s thesis, Kathmandu University, Dhulikhel, Kavre, Nepal, 2004.

Word sense disambiguation using wsd specific wordnet of polysemy words

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to Word sense disambiguation using wsd specific wordnet of polysemy words (20)

Recently uploaded (20)

Word sense disambiguation using wsd specific wordnet of polysemy words