IRJET- An Analysis of Recent Advancements on the Dependency Parser

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 05 | May 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 2139
An Analysis of Recent Advancements on the Dependency Parser
Patel Hemalkumar Ratilal1
1Student,School of Computer Science and Engineering, VIT University, Vellore, Tamil Nadu, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Dependency Grammar (DG) is a class of current syntactic hypotheses that are all in light of the dependency
connection as opposed to the constituency relation. Dependency is the thought that linguistic units, e.g. words, areassociatedwith
each other by coordinated connections. The finite verb is taken to be the structural focal point of condition structure. All other
syntactic units are either straightforwardly or in a roundabout way associatedwiththeverbasfarasthecoordinatedconnections,
which are called dependencies. Dependency grammar has been a great help in NLP (Natural Language Processing).Dependency
parser and parsing techniques have an upper hand when it comes to the understanding of the NLP. This paper reviews the recent
advancement in the dependency parser and its applications.
Key Words: Dependency Grammar , Dependency Parser, Parse Tree, NLP, Semantic Dependencies, Morphological
Dependencies, Prosodic Dependencies
1. INTRODUCTION
Dependency Grammar is different fromtheconventional grammarintermsofsyntacticstructure. WhileContextFreeGrammar
(CFG) is a generative grammar, Dependency Grammar is completely descriptive. It only depends on the dependency relation
that exists between lexical items present in the sentence. This is a great virtue of the DG which makes it a suitablegrammar for
the free order languages. Even though it is used for the free order languages, some frameworksaretherewhichfollowstheDG,
such as Link Grammar, Operator Grammar, Lexicase, Word Grammar, Meaning Text Theory etc. There are various ways to
represent the dependencies which are purely based upon the type of the dependency, mostly semantic, morphological,
prosodic and syntactic. Semantic Dependencies are based upon the predicates and arguments, or triples. Morphological
Dependencies are constructed on the words and sometimes parts of words. When a word oritspartaffectsanother word,then
it becomes the morphological. Nature of the clitics isrecognised bytheProsodicDependencies.Cliticgeneratesitsownword by
merging with its host. DG has its own style of approaching the sentences in linear order. DGs force to focus on the content by
being minimal than the conventional Constituency-based Grammars. The Syntactic Function is considered primitive in DGs.
These functions are annotated as labels in the Parse Tree.
2. Different Directions of Development
M. Kumarand M. Dua [1] used the ability of the Paninian Grammar to providethe greatsolution forthelanguageswiththeword
free order and applied that to Stanford parser which in fact provides the better dependencies for the languages having fixed
order. English is one such language which supports StanfordParser.PreviousresultsshowedthatPaninianGrammariscouldbe
applied to the other Indian Languages with much higher efficiency.
Syntaxalong with the semantics wascaught and handled by VerbNet. They have exhibited the issues thatexperiencedwhile
doing an adjustment and proposed the answer for beating these issues. They utilizedHindiDependencyparserforconfirmation
of comes about. With this adjustment of Stanford Parser hand in hand syntax tree of English and Hindi can be made.
M. Savic´, G. Rakic´,Z. Budimac and M. Ivanovic [2] displayed SNEIPL,a novelwaytodealwiththeextractionofprogramming
systems that depends on a dialect autonomous, enhanced solid linguistic structure tree representation of the source code. The
appropriateness of the methodology is exhibited by the extraction of programming systems speaking to genuine, medium to
extensive programming frameworks written in various languages which have a place to various programming paradigms. To
examine the fulfilmentandrightnessofthemethodology,classcoordinatedeffortsystems(CCNs)extricatedfromcertifiableJava
programming frameworks are contrasted with CCNs acquired by different devices. Dependency finder was used to fetch the
dependencies having the entity levels from the Java bytecode. In addition to that, they used the Doxygen, which is efficient in
extracting the dependency using the fuzzy parsing, which in turn is complete to independent to the language being used. They
compared these results with that of SNEIPL. They demonstrated that the SNEIPL can form highly precise networks better than
the Dependency Finder and Doxygen.
S. N, S. M. Jasmine and S. Joseph [3] proposedan machinelearning approacheslike ADTree and Naïve Bayes Classifier along
with the dependency parsingto extract the hypernym and meronym relations in the given sentences. The approachutilizedthe
path between nouns in the syntax tree generated.

W. Z. YUAN, Z. MIAO, W. ZHU and W. LIU [4] developed a cross breed strategy for identifying ASR blunder in talked turns is
produced. The mistaken content is locally dissected first by neighboringco-eventrelationsutilizingngrammodel.Theyusedthe
ability of Dependency Parser to find the long distance dependency relationstoanalyzethetextglobally.Theirtestsdemonstrate
that we can utilize data from a Dependency parsing stage together with n-gram model not just to identify incorrect ASR
speculations thatcan cause understanding blunders, additionallydependablyfindmistakesandhereandthererightthemasthe
speculations are being handled.TheyalsoshowedthatbuildingChineselanguageparserhasmuchmorecomplexitythannormal
languages.
M. Shen, D. Kawahara and S. Kurohashi [5] they demonstrated another approach for Dependency Parsing that can use
complex subtree representations by applying productive subtree choice methods. Chinese Treebank along with the Penn
Treebank were analyzed using thisapproach to show the viability of the novel method.Their frameworkaccomplishesthebest
execution among known administered frameworks assessed on these datasets, progressing the benchmark precision from
91.88% to 93.42% forEnglish, what's more, from 87.39% to 89.25% forChinese. Other semi-supervised parsing methodssuch
as word-clustering along with the subtree feature integration show no overlap with their novel approach which is a clear
advantage of this method.
C. Lee, G. G. Lee and M. Jang [6] brought the novel model for dependency structure called Dependency Structure Language
Model tocounter the weakness of modelslike unigramandbigram in TDT (Topic Detection and Tracking).Modelslikeunigram
and bigram lack the ability to capture the semantics in records. In addition to that, dependencies having the long distance
between them are not supported by those models. The novel model they proposed utilized the Chow Expansion Theory along
with the dependency parse tree produced by a dependency parser.Long-distance dependencies can be easilytakencarebythe
dependency structure language model. They concluded that Dependency Structure Language Model beat the conventional
language model. In addition to that, they showed that the proposed model has advantages overthebigraminTDT.However,the
proposed model has high computational cost, which needs to be taken care in future.
W. Chen, M. Zhang and Y. Zhang [7] demonstrated an approach to tackle the problem of feature sparseness in dependency
parsing. Their methodleads to learning of featureembeddingsautomatically.Intheirmethodology,thecomponentembeddings
are gathered from a lot of auto-parsed information. To begin with, the sentences in crude information are parsed by a gauge
framework, what's more, they acquired Dependency Trees. They proposed twolearningmethodstoinducefeatureembeddings
using the representation of every model feature by utilizing neighboring features on Dependency Trees. They extracted the
bunch of features forDependencies that are based onGraphbyutilizingfeatureembeddings.Thenewparsersarehighlycapable
of utilizing settled hand-composed features along with thehidden class representationoffeatures.Thenewparsersaccomplish
critical execution changes over a solid pattern.
S. Pyysalo, F. Ginter, T. Pahikkala, J. Boberg, J. J¨arvinen and T. Salakoski [8] demonstrated an assessment of Link Grammar
along with the Connexor Machinese Syntax (CMS), two noteworthy and highly researched Dependency Parsers, on a custom
hand-explained corpus comprising of sentences with respect to the interactions between protein and protein. To express the
protein-protein correspondence, interaction graph was used. They did the performance benchmark of the both the parsers in
with measurements like a number of full parses along with the individual dependencies. They concluded that both the parsers
lack the ability to parse the biomedical English. While CMSsignificantly beat theLinkGrammar,theiranalysisimpliedthatthere
are some ways in which the Link Grammar could perform better in the specific domain. They did the deep evaluation of Link
Grammar, addressing issues including panic mode and sampling.
W. Chen, M. Zhang, Y. Zhang and X. Duan [9] proposedthehighlyaccurateframeworktoaddressthesparsenessproblem.The
framework was used to train the discriminative models.
It works on the concept of frequency bucketing on big data processed through automation. In addition to that, the feature
clusters are utilized to define the meta features. The proposed approach is highly efficient when it comes to dealing with
dependency parsing along with the part of speech tagging. The meta-parser is proved highlycompetitivewiththebestavailable
parsers and most accurate in Chinese data for the Dependency Parsing Task. The tagger showed the reduced error rate of 5%
than the state-of-the-art baseline systems when processed on the Chinese and English language data. In addition to that, meta
features are proven efficient while dealing with the unknown features.
J. Yu and W. Chen [10] proposed an automatic identification method to address the recognition problem for co-ordinate
structures in Chinese Dependency. They introduced the new features related to the word pairs collected into the Dependency
Parser. In the first step, they influenced two hand-made standards to remove exceptionally precise direction word sets as seed
words. The second step is touse seed words to extricatecoordinatestructures in the corpus forfurtherutilizationofcoordinate
word pair extraction. Exploratory results demonstrated that the extricated coordinate word sets can essentially enhance the
precision on coordinate structure Dependency Analysis.

R. Goutam [11] utilized the Partial Parsers to investigate the impact of applying bootstrapping strategies such as self-
preparing along with the co-preparing on Hindi Dependency Parsers. They went through distinctive criteria for selecting very
sure Partial Parses that are going to be useful for bootstrapping. Our outcomes appeared that co-preparing between Partial
Parser, Malt and MST utilising understanding check as determination criteria played out the best and gave huge change in
execution improvement at baseline.Utilisingco-training,wecouldenhancetheexactnessofthelow-performingPartialParserto
that tantamount to the cutting edge precision for programmed Hindi Dependency Parsing.
W. Wang and M. P.Harper [12] showed the assessment of a language model based on Statistical Constraint Dependency
Grammar (CDG) parser. Grammatical structureof a sentence is represented bydependencyrelationsbetweenwordsutilizedin
the sentences. CDG applies some restrictions to decide that structure. This Language Model outperforms the conventional CDG
based Language Model by showing reduced WER(Word Error Rate) compared to the latter. In addition to that, other Language
Model based on stateof the art parser were lacking in the ability of reduced the WER(Word Error Rate).Thereasonof the being
powerful is its ability to tightly integrate with various sources.
S. Majidi andG. Crane [13] did the detailed study on the error rateofthedifferentannotators fortheancientGreekLanguage.
They compared the error rate of the twotypeof annotators,StudentAnnotatorsandDependencyParserAnnotators.Theyfound
out the error ratealmostsimilar for both type. Both students and parserfindthesamedifficulty.Theerrorsofautomaticparsers
and that of the student overlap at some stage. We can consolidate these different tasks. Teaching students an Ancient Greek
Language with increasing the accuracy of the linguistic information.
M. Wang, H. Xia, D. Sun, Z. Chen, M. Wang and A.L [14] proposed a novel content digging technique foreffectivelyrecovering
and extricating protein phosphorylation data from writing. They utilized the Natural Language Processing (NLP) techniquesto
change everysentence into Dependency Parse Tree. This provides the edge to the Dependency Parse Trees in reflecting inborn
relationships of phosphorylation-related watchwords. This helped to extract the detailed insight about substrates,kinasesand
phosphorylation locales. Contrasted and other existing methodologies, the proposed technique exhibits essentially enhanced
execution, recommending it is a capable bioinformatics way to deal with recovering phosphorylation insight from a vast
literature. Their method has the ability to simplify the complex sentences which give that the ability to outperform any existing
method. Even information extracted from the Dependency Trees are precise and highly accurate.
Weichselbraun and N. S¨usstrunk [15] proposed a way to optimizetheexistingmethodsbyfocusingonthefeatureextraction
along with the parsingconstraint optimization.Inadditiontothat,thisapproachcouldbeappliedtotheexistingmethodssuchas
MDParser (Volokh, 2013) and Stanford parser (Chen and Manning, 2014). They compared them by applying them for the
English Universal Dependencies Corpus. They measured the performance by two factors - Unlabeled Attachment Score-UAS
(correctly assigned head) and Labeled Attachment Score-LAS (fraction of heads and types which have been correctly
recognized).
The results show that Syntactic Parser outperforms rest. After applying the optimization throughput was increased four
times of the original throughput of MDParser. In addition to that, this performance gain was not acquired at the cost of the
accuracy. Not only that, Syntactic Parser provided better accuracy in some cases. Syntactic Parser also showed the very good
performance in other languages other than English like French, German in LAS outperforming MaltParser and MST.
Rahul.C, Dinunath.K, R. Ravindran and K.P.Soman [16] proposed their work for English to Malayalam Statistical Machine
Translation by specifying rule based reordering along withthemorphologicalinformation.Therearetwowayswhichhavebeen
demonstrated extremely successful – rearranging the English Based Sentences to Malayalam Syntax – Word separation by
utilizing the root suffix method on both the languages. The phrase-based system is beaten by the proposed method. Their
method leads to the conclusion that the Indian Languages -whicharenotrichinmorphologyandparsingtoolsshouldbetreated
with this method. Their method reduced the need of such toolsfortheIndianLanguagesfollowingSOV(Subject–Object–Verb)
format.
L. D. Caro and M. Grella [17] proposed a novel algorithm for Sentiment Analysis (SA) along with the model which is context-
based. Their algorithm depends on upon the rules of propagation for the Dependency Tree at the syntactic level. Their model
tunes the user’s opinions on the basis of the context. They implemented their ideas through SentiVisutilizingDataVisualization
ability of the SentiVis for the Sentiment Analysis. Separated sentiments were arranged in the 2D format for further processing.
2D format foruseful for calculating the strength with variability. Not only that this2Dtablecouldbeusedaccordingtotheuser’s
wish such as finding fruitful facts along with the removing the false positives and negatives.
B. V. S. Kumari and R. R. Rao [18] gained the insight about some of the popular parsers including MaltParser, TurboParser,
MSTParser, Easy First Parser and ZPar by different settings of features. They showed how different parser gets affected by
different settings. Theyalso did the benchmark on the performance of the different parsers.TeluguDependencyTreebankfrom

the ICON 2010 tools contest was used to test the reports. The state-of-the-art performancewas91.8%inUASand70.0%inLAS.
They pointed out that while building the datasets weneed to keep in mind that a good mixtureoflongandshortsentencesalong
with the short and long distance dependencies should be there in the test language. Telugu is one such language. They also
concluded that a parser should be given the ability to add/remove some of the features so that parser could learn from the
datasets and develop on its own.
H. Mohamed, N. Omar, M. J. A. Aziz and S. A. Rahman [19] examined the Dependency Grammar needed to develop the Malay
Language Grammar Parser. They proposed the possibilities for generating Probabilistic Dependency Malay Parser utilizing
annotated relation based on the English Language. They defined the rules of the inter-relation of the Malay and English. They
projected the Tree Structurewhich was utilized to train the stochastic analyzer. That training produced the tree lattices having
Malay Dependency Trees. Those lattices also had the probabilities so that information could be gained and tested were
performed using the decoders.
P. N. Pelja,R. P, S. G. Mand R.Binu [20] proposed and automatic AMR tool to address the issue of time-consuming tasks such
asabstract meaning representation. Their experimentsprovesthattheirmethodperformsuptothepointwith80%accuracyon
simple sentences. Although they did not perform the tests on the complex sentences, they claim that their method could be
developed to support and handle the complex sentences of the English Language.
Table 1: Comparative Analysis
Ref.
NO.
Problem Addressed Methodology Merits Demerits
1 Using Paninian Grammar for
annotation scheme for English
Use of Karaka
relations mapping
English-Hindi
parallel treebank
with Paninian
Grammar
Max accuracy
achieved so far is
still less than 75%
2 Extraction of Software
Networks
SNEIPL – a novel
approach based on
language-
independent and
concrete syntax
tree (eCST) of
source code
Highly precise
networks better
than Dependency
Finder and
Doxygen,
Language-
independent, more
precise than
available tools
It can only fetch
software networks
and cannot do
further reverse
engineering
3 Extraction the relationships
between entities exist in the
text
Use of the
Dependency
parsing techniques
along with the
machine learning
techniques
Better than
WordNet, Uses
probability,F-score
with ADTree
classifier (75%) for
hypernym relation,
Naïve Bayes
classifier (70%) for
meronym relation.
F-Score for
hypernym is less
than expected
4 Detection the Automatic Speech
Recognition (ASR) errors in
Chinese Language
Using Dependency
Parsing and n-gram
model to detect the
error
More efficient than
only n-gram model
Max efficiency
achieved is still not
considerable and
needs some
improvement, it
can’t correct the
error
5 Highly accurate syntactic
analysis
Novel feature set,
feature back-off
strategy
No overlap with
semi-supervised
parsing methods,
efficient than
previous methods
6 Weakness of unigram and Novel approach Better handling of Highcomputational

bigram models for topic
detection and tracking
called Dependency
StructureLanguage
Model using Chow
expansion theory
and linguistic
parser
long-distance
dependencies,
more efficient and
effective than
previous methods
cost O(n3)
7 Solving feature sparseness in
Dependency Parsing
Parsing of
sentences with
baseline system,
feature embedding,
inferring feature
embedding
Improved
performance when
tested on English
and Chinese
Languages
Results are
unknown for other
languages like
Japanese
8 Information Extraction in
biomedical protein- protein
interactions
Tests were done on
a custom hand
annotated corpus
for Link Grammar
and Connexor
Machinese Syntax
Link Grammar and
Connexor
Machinese Syntax
evaluated, failure
sources identified
and improvements
proposed
Results are
unknown for
machine annotated
corpus
9 Higher sparseness in
Discriminative methods used in
NLP
Frequency
bucketing
High accuracy on
English and
Chinese data
Smaller data was
used for tests
10 Identification of coordinate
word pairs in Chinese Language
Utilizing seeds to
find new rules for
unlabelled data,
generating
candidate
coordinate word
pairs and use them
to Dependency
Parser model
Simple and
effective approach
No re-use of final
word pairs as seed
words
11 Parsing languages having small
treebank
Using Partial
Parser joined with
DependencyParser
for self-training,
using Malt Parser
and MST Parser
with Partial Parser
for co-training
Performance
improvement,
accurate for Hindi
Parsing
Results are
unknown for
Telugu and Bangla.
12 Evaluation of Statistical
Constraint Based Dependency
Parser
The 20k open
vocabulary DARPA
Hub1 WSJ CSR is
used
to evaluate SCDG
parser-based LM
Reduction in Word
Error Rate and
more accuracythan
other Constraint
based parsers
Accuracy is still not
considerable
13 Comparing human andmachine
annotators for Greek Language
Comparisons of
errors using
Spearman’s Rank
correlation
Overlapofmistakes
show that we can
combine two
different methods
Test done only on
the Greek Language
14 Text mining for protein
phosphorylation
Simplification of
complex sentences
using Dependency
Parsing, Pattern
derivation from
Dependency Tree
Efficient, accurate,
can be applied to
large texts in
biology literature
No test done for
entire text of
literature
15 OptimizingDependencyParsing
Throughput
Replacing
infrequent word
Four fold
improvement with
Not good UAS score
for other languages

with ‘unknown’
label, Feature
Encoding in 64 bits
no cost of accuracy,
Best UAS for
English
for
16 Poor Statistical Machine
Translation from Malayalam to
English
Rearrangement of
English sentences
to Malayalam
Good forLanguages
which follow SOV
order (Indian
Languages) ,
reduces the use of
parsing tools
Only for languages
which follow SOV
order
17 Performing Sentiment Analysis
using Dependency Parsing
Using Dependency
Parsing for
sentiment
propagation rules,
Context based
model to tune
sentiments, Using
SentVis for
implementation
Detection of
outliers, Efficient
on the simple
sentences of
English
Conditional
statements cannot
be analysed,no
results availablefor
other languages
18 Exploration of effect of
statistical parser for Telugu
parsing
Using MSTParser,
Malt Parser, ZPar
,TurboParser and
Easy-First Parser
with different
feature settings
Test language was
Telugu –
morphologically
rich language
Small size ofTelugu
Treebank
19 Generating Parser for
knowledge acquisition in Malay
Language
Using existing
parsers for English
along with English-
Malay parallel
corpus
New possibilities
are developed for
Malay Parsers
Model is only
proposed , not
created in practice
20 Automation of the AMR
(Abstract Meaning
Representation)forinformation
retrieval
Developed an
editor for AMR, Use
of Dependency
Parser
80% accuracy on
simple sentencesof
the English
Tests on the
complex sentences
are unknown,
dependent on the
English
3. SCOPE OF RESEARCH
AMR automation is yet not made for the complex sentences [20]. Construction of the Chinese Language Parser having high
accuracy is still a poorly addressed question. Chinese Language is very complex and different from the other languages.
Dependency Re-ranking System Proposed above [5] is still not combined with the other methods. The effect of the methods
proposed [11] on the Bangla and Telugu languages which have scarce resources has not beenstudied.Inadditionto that,more
advanced strategies need to be used as a selection criteria. The alternate clustering algorithm couldbecombined withmethod
proposed [9]. Enhancing of the parser [9] by utilising semi-supervised methods is yet not done. Parsers for Malayalamarenot
still much efficient [16]. Morphologically rich languages are difficult to parse and very interesting work is going on them.
Different parsers having complementary features are combined to increase the accuracy.
REFERENCES
[1] M. Kumar and M. Dua, "Adapting Stanford Parser’s Dependencies to Paninian Grammar’sKaraka relationsusingVerbNet,"
Procedia Computer Science, no. 58, pp. 363-370, 2015.
[2] M. Savic´, G. Rakic´, Z. Budimac and M. Ivanovic, "A language-independent approach to the extraction of dependencies
between source code entities," Information and Software Technology, no. 56, pp. 1268-1288, 2014.
[3] S. N, S. M. Jasminea and S. Joseph, "Automatic Extraction of Hypernym & Meronym Relations in English Sentences Using
Dependency Parser," Procedia Computer Science, no. 93, pp. 539-546, 2016.

[4] W. Z. YUAN, Z. MIAO, W. ZHU and W. LIU, "Detecting errors in Chinese spoken dialog system using ngram anddependency
parsing," ICWMMN Proceedings, 2008.
[5] M. Shen, D. Kawahara and S. Kurohashi, "Dependency Parse Reranking with Rich Subtree Features," IEEE/ACM
TRANSACTIONS ON AUDIO, SPEECH, AND LANGUAGE PROCESSING, vol. 22, no. 7, 2014.
[6] C. Lee, G. G. Lee and M. Jang, "Dependency structure language model for topic detection and tracking," Information
Processing and Management, no. 43, pp. 1249-1259, 2007.
[7] W. Chen, M. Zhang and Y. Zhang, "Distributed Feature Representations for Dependency Parsing," IEEE/ACM
TRANSACTIONS ON AUDIO, SPEECH, AND LANGUAGE PROCESSING, vol. 23, no. 3, 2015.
[8] S. Pyysalo, F. Ginter, T. Pahikkala, J. Boberg, J. J¨arvinen and T. Salakoski, "Evaluation of two dependency parsers on
biomedical corpus targeted at protein—protein interactions,"International Journal ofMedical Informatics,no.75, pp.430-
442, 2006.
[9] W. Chen, M. Zhang, Y. Zhang and X. Duan, "Exploiting meta features for dependency parsing and part-of-speech tagging,"
Artificial Intelligence, no. 230, p. 173–191, 2016.
[10] J. Yu and W. Chen, "Extracting Coordinate Word Pairs for Dependency Parsing," Collaborative Innovation Center of Novel
Software Technology and Industrialization, Suzhou, China.
[11] R. Goutam, "Exploring self-training and co-training for Hindi Dependency Parsing using partial parses," in International
Conference on Asian Language Processing, Hyderabad, India, 2012.
[12] W. Wang and M. P.Harper, LANGUAGE MODELING USING A STATISTICAL DEPENDENCY GRAMMAR PARSER,IEEE,2003.
[13] S. Majidi and G. Crane, "Human and Machine Error Analysis on Dependency Parsingof AncientGreek Texts,"inIEEE,2014.
[14] M. Wang, H. Xia, D. Sun, Z. Chen, M. Wang and A. Li, "Literature mining of protein phosphorylationusingdependencyparse
trees," Methods, no. 67, pp. 386-393, 2014.
[15] A. Weichselbraun and N. S¨usstrunk, Optimizing Dependency Parsing Throughput, Chur, Switzerland: Department of
Information, University of Applied Sciences Chur.
[16] Rahul.C, Dinunath.K, R. Ravindran and K.P.Soman, "Rule Based Reordering and Morphological Processing For English-
Malayalam Statistical Machine Translation," in IEEE, Coimbatore, 2009.
[17] L. D. Caro and M. Grella, "Sentiment analysis via dependency parsing," Computer Standards & Interfaces , no. 35, p. 442–
453, 2013.
[18] B. V. S. Kumari and R. R. Rao, "Telugu dependency parsing using different statistical parsers," Journal of King Saud
University – Computer and Information Sciences, 2015.
[19] H. Mohamed, N. Omar, M. J. A. Aziz and S. A. Rahman, "Statistical Malay Dependency Parser for Knowledge Acquisition
Based on Word Dependency Relation," Procedia - Social and Behavioral Sciences , no. 27, p. 188 – 193, 2011.
[20] P. N. Pelja, R. P, S. G. M and R. Binu, "Automatic AMR Generation for Simple Sentences Using Dependency Parser,"Procedia
Technology, no. 24, pp. 1528-1533, 2016.

IRJET- An Analysis of Recent Advancements on the Dependency Parser

More Related Content

What's hot (18)

Similar to IRJET- An Analysis of Recent Advancements on the Dependency Parser (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- An Analysis of Recent Advancements on the Dependency Parser