=Paper=
{{Paper
|id=Vol-2096/paper1
|storemode=property
|title=Similarity Detection among Academic Contents through Semantic Technologies and Text Mining
|pdfUrl=https://ceur-ws.org/Vol-2096/paper1.pdf
|volume=Vol-2096
|authors=Victor Saquicela,Fernando Baculima,Gerardo Orellana,Nelson Piedra,Marcos Orellana,Mauricio Espinoza
|dblpUrl=https://dblp.org/rec/conf/iwsw/SaquicelaBOPOE18
}}
==Similarity Detection among Academic Contents through Semantic Technologies and Text Mining==
https://ceur-ws.org/Vol-2096/paper1.pdf
Similarity Detection among Academic Contents
through Semantic Technologies and Text Mining
Victor Saquicela1 , Fernando Baculima1 , Gerardo Orellana2 , Nelson Piedra3 ,
Marcos Orellana2 , and Mauricio Espinoza1
1
Departamento de Ciencias de la Computación, Universidad de Cuenca, Ecuador
2
Escuela de Ingenierı́a de Sistemas y Telemática, Universidad del Azuay, Ecuador
3
Departamento de Ciencias de la Computación, Universidad Técnica Particular de
Loja, Ecuador
{victor.saquicela, fernando.baculima, mauricio.espinoza}@ucuenca.edu.ec
{gorellana, marore}@uazuay.edu.ec
nopiedra@utpl.edu.ec
Abstract. Nowadays, the information of university courses is managed
by means of syllabus based systems, this is the case of Ecuadorian Higher
Education Institutions (IES for its Spanish acronym). However, the syl-
labus structure is not normalized among all universities, since there is a
wide variety of formats and data models used for each IES which natu-
rally, affects academic processes such as the students mobility or credits
validation between IES. We have addressed these issues by presenting a
proposal based on semantic technologies and text mining methods whose
goal is to identify similarities among academic contents.
Keywords: Higher Education, Semantic Web, Ontologies, Text Mining
1 Introduction
In recent years, aiming to improve the common learning procedures in Higher
Education Institutions, several researchers have proposed technological initia-
tives in the education field [2,19,44,17]. Specifically, Ecuadorian IES have been
exposed to intensive changes through strict evaluation processes in the academic,
administrative and structural points of view [8,26].
The 2011 Organic Law of Higher Education, disposed all IES undergo to an
evaluation, acreditation and categorization process [7], and in consequence the
tools to improve quality in IES gained special interest.
Regarding academic evaluation and with the aim to ensure quality in higher
education, exhaustive analysis have been performed to every career in every IES.
Nowadays, in Ecuador, IES are ruled by the Regulation of Academic Regime
(RRA for its Spanish acronym), which specifies that each IES is composed by
academic units, each academic unit by careers, and each career by subjects. A
subject is defined as a set of contents about a specific area [9].
Although this hierarchical structure, each IES maintains the autonomy over
its study plans, defining its own contents for each career, thus, this creates a large
2
heterogeneity over subjects contents. Also, there is a large variety of formats and
careers program models in which syllabus are built and kept across the different
IES.
Such heterogeneity isolates each IES from others, thus, the creation of a
common national repository to store career related data and subject contents
(syllabus), represents a very challenging task. A syllabus is a document in which
contents and learning methods for students are defined by the members of aca-
demic units [21,44]. In addition, the lack of methods, processes and procedures
to identify similarities among career contents in different IES makes students
mobility through credits validation a difficult task [17]. With this background,
students mobility among IES, at a national or international level, becomes a chal-
lenge, mainly due to the fact of not owning a systematized format for subject
contents, hence, there is a large dependency of manual processes for homologa-
tions or equivalences matching between subjects.
Currently, efforts like [11,16,30] have introduced the benefits of introducing
new information technologies into the education field. Knowledge representa-
tion has been applied to the learning context, specifically to represent learning
resources, mainly implementing on-line learning modalities (e-learning) [10].
In this paper, we propose a new similarity detection approach for academic
contents among Ecuadorian IES. This approach is aligned with the application of
semantic and text mining technologies altogether, which will allow in the future,
the construction of a computer system which will be able to provide solutions
for the students mobility issue.
The remain of this paper is organized as follows: first, we describe the back-
ground and related works about semantic web and text mining, it pays special
attention to the higher education field; next, we present the process for similarity
detection among syllabus; finally, we present some conclusions and future works.
2 Background and Related Works
The need to improve processes, techniques and methodologies around activities
such as learning and teaching has led knowledge representation researchers to
improve how curricula are modeled and managed semantically [10,11,16,17,30].
The semantic web is an emerging technology to describe resources in the web
aiming to make information understandable not only to humans but also to
machines [6].
In the academic context, institutions produce information and store it in
digital repositories. However, the lack of use of open standards and a semantic
approach have caused difficulties when integrating and re-using contents through
the Web [36]. Yet, important advances in the application of the Semantic Web in
the educational context have been made. In [1] and [37] integration architectures
and distributed interoperability of digital repositories are proposed, based on a
Semantic Web approach, Linked Data technologies and federated queries. Those
architectures have been successfully tested to integrate a group of institutional
3
repositories belonging to universities [35], and they have enabled a federated
query environment within this context [40].
An Ontology is an important technology which allows data to be represented
in the Semantic Web [6]. Ontologies are models that can be used to structure
knowledge, they may create a better interaction between teachers and students
and improve the learning outcomes and teaching methods of the academic con-
tents [11]. In [11], an ontology based system is presented to allow the integration
and classification of heterogeneous systems, learning objects and curriculums,
nevertheless, this work is not focused in solving similarities among syllabus or
solving the students mobility problems. On the other hand, Demartini et al.,
in [17] describes the development of the Bowlogna ontology, this work reflects
the interests of the Bologna project for the renewal and standardization of high
education in Europe, authors mention the problem of student mobility regarding
credits recognition, situation that motivated the development of such ontology
and its applications.
Despite the mentioned efforts, similarity detection among syllabus is still
an unsolved problem. To address this problem, some related works have been
reviewed, [42] presents an approach to identify common research areas using
semantic and data mining technologies. In [22], authors developed a software
to help students of the Agder University to discover existing relations between
computer science courses by using ontologies and semantic web. However, this
approach is oriented to an unique field of science within the mentioned university.
Finally, [32] presents an ontology model for the computer science area, using an
extended version of the Wu & Palmer algorithm described in [46] to calculate
semantic similarity between computer science courses, nevertheless, as the work
presented in [22], this methodology is limited to one specific knowledge area and
the use of an algorithm to measure conceptual distances based with the use of
Wordnet service [18] as a knowledge representation approach for English words.
3 Similarity Detection among Academic Contents
With the aim to solve the students mobility issue among higher education insti-
tutions, this paper presents a process for academic contents similarity detection
among IES. This proposal is inspired by the Linked Data Life Cycle described
by Auer in [4] and the approach for the identification of common research ar-
eas proposed by Sumba et al. in [42]. The implementation of this process aims
to develop a similarity detection software among IES syllabus. Figure 1, shows
the proposed process which consists of the following components: i) Data Ex-
traction, module designed to retrieving syllabus data from different IES through
data extraction and cleaning processes, ii) Ontological Modeling Module, this
is where the model for syllabus description is developed, adapted or extended
based on existing methodologies, standards and recommendations for ontology
development, iii) RDF-Ization Module, based on the ontological model, it gen-
erates syllabus data in RDF format, feeding the new semantic repository. RDF
(Resource Description Framework) is a standard model for data interchange on
4
the Web [20], iv) Patterns Detection Module, through the use of text mining
techniques, it discovers similarities and patterns among syllabus and v) Visu-
alization, module designed to exploiting and displaying the semantic data in a
comprehensible format to the final users.
Fig. 1. Similarity detection process among syllabus.
3.1 Data Extraction
Syllabus data will be collected either manually or automatically. The access
method will be defined according to the data format, since these can be presented
in different formats such as PDF, HTML, WORD, Relational Database, among
others. This module execute data pre-processing and cleaning techniques, which
are necessary tasks that need special methods for their treatment for its later
semantic annotation process [3].
Fig. 2. Data Extraction Process.
Figure 2, shows data extraction process and its tasks: 1) Data sources, refers
to documents or databases where syllabus of the IES reside; 2) Connectors, the
5
software components that must be developed to connect with data sources; 3)
ETL process, pre-process and cleaning the data that comes from data sources
using specialized software; and 4) Processed data, a new temporary repository
containing clean data for the subsequent semantic annotation.
3.2 Ontological Model
One of the main goals of this paper is the use of semantic technologies in order
to make syllabus data understandable by both, humans and computers. It is
important to reuse existing ontological models in the best possible way to facil-
itate the inclusion and interoperability of the new data in the web of data [23].
In the background and related works section, several ontological models were
presented, which were developed with the aim of solving education and learning
related issues. As part of the proposed process, this work will reuse, integrate and
extend some ontologies such as BOWLOGNA, FOAF, BIBO, among others, fol-
lowing the NEON ontology development methodology [43], which is widely used
in the ontological engineering field. In addition, to apply NEON in the ontology
development, this work will use the software PROTÉGÉ, which, due to its ease
of use and community support, is widely used in many projects in the field of
ontology development [31]. Finally, with the aim of identifying inconsistencies or
problems during the ontology development, this work will use the OOPS! Plat-
form, an OntOlogy Pitfall Scanner, which allows an easy ontology evaluation
on-line [38].
Fig. 3. Ecuadorian Syllabus Ontology Definition Process.
6
Figure 3 shows the process to reuse, adapt and extend an ontology for
Ecuadorian IES syllabus, such process is composed by: 1) Requirements defi-
nition, it identifies the requirements of the problem and assigns a team of people
for the specific tasks, 2) Application of the NEON methodology, which through
the selection of a set of scenarios allows planning the extension/creation of on-
tology networks, 3) Ontology creation, it uses the PROTÉGÉ software to reuse,
unify, extend and generate the required ontology and 4) Ontology Evaluation,
it verifies the effectiveness and functionality of the ontology through the data
instantiation and query execution.
3.3 RDF-Ization
Once the ontological model has been defined and created, the RDF triplets com-
posed of syllabus data will be created and stored in a repository (TripleStore).
This process depends on the data extraction and ontological modelling modules.
The Linked Data principles proposed by Berners-Lee in [5] will be followed, also
specialized frameworks will be used to generate the data in RDF format.
The RDF-Ization process is shown in Figure 4, where the resulting data from
the extraction module will be mapped with the ontology model by using the
LOD-GF framework, which consists of a friendly graphical interface with drag
& drop functionalities allowing the RDF generation in an easy and intuitive way
[13]. Finally, the new RDF data will be stored in a TripleStore for the subsequent
exploitation through SPARQL queries.
Fig. 4. RDF Generation Process.
3.4 Similarity Detection Techniques
As described in the background and related works section, there are many
projects implementing data mining techniques to discover behavioral patterns
from raw data. In this paper, we propose the implementation of Text Mining
Techniques to analyze unstructured texts, aiming to discover similarities among
syllabuses of different Ecuadorians IES careers. Sailaja et al., presented in [39] a
”Text Mining Framework”, shown in figure 5, which is described by the following
3 stages:
7
• Stage I: Text Pre-processing, including the data cleaning process such
as text lemmatization, stopwords removal, dimensionality reduction, among
others.
• Stage II: Text Mining Techniques, indicates the criteria to select the proper
algorithms to process documents.
• Stage III: Text Analysis, indicates the use of several tools for information
discovery; it means unstructured texts will be converted into meaningful
information that helps decision-making.
Fig. 5. Text Mining Framework [39].
Based on the Text Mining Framework described above, this paper presents
the process shown in Figure 6 to confront the application of text mining on
syllabus data. This process will allows patterns discovering, similarities or dif-
ferences among syllabus, helping to solve the students mobility problem and
address the problems discovered in academic content.
Fig. 6. Text Mining Implementation Process.
8
Text Pre-processing: The mining process of unstructured texts, consists of
several stages; first, there is a data pre-processing [3,33], this stage depends on
the specific domain to be analyzed, and involves the elimination of non-relevant
words or terms (Stopwords), transformation of text to binary matrices, etc. The
platform that will be developed, will extract RDF data from the TripleStore then
use NLTK tool to delete insignificant terms. Second, Text Mining algorithms
need information in a format they can process, thus, the text will be transformed
to numerical matrices using the TF-IDF method recommended in [3].
Text Mining Techniques: There are several techniques and algorithms in the
Artificial Intelligence field which perform text mining. As described in [3], clus-
tering is one of the most popular text mining technique and is widely used in
classification, visualization and document organization applications. Clustering
allows the determination of groups of documents that share common features
or have some similarity within the documents collection [14]. Amid the differ-
ent techniques of clustering, this approach proposes the implementation of the
K-means algorithm which is widely used in the field of Data-Mining and Text-
Mining, this algorithm divides n documents into k different clusters. In addition,
we propose the use of semantic similarity techniques to measure the relatedness
of the documents within each group (cluster) [28]. Among some techniques for
analyzing the semantic content of a text, we can refer to Genetic Algorithms
described in [15], Latent Semantic Analysis (LSA) described in [45], Machine
Learning algorithms described in [3], and Word Embeddings techniques that
have gained popularity in the text mining community [25]. The algorithms based
on WordNET and distance measurement between terms have presented good re-
sults [32,29] as well as those based on the information contents [24]. In the
process proposed by this paper, we intend to use a combination of the methods,
techniques and algorithms described above which will allows the development of
a similarity detection platform to solve the students mobility issue.
Analysis of Text: First, with the aim of reducing search spaces, clustering
algorithms will be used to identify relatedness among syllabus. Subsequently,
on each identified cluster, different text mining algorithms or methods will be
executed to compare resulting texts among the different documents (syllabus).
Different methods such as cosine similarity will be used to determine the similar-
ity degree between two syllabuses. In addition, new triplets will be created and
load on the TripleStore that will allow to identify and query the syllabus clusters
as well as the existing similarity among them. In general, this stage indicates the
knowledge extraction over the data set and finally, with the help of visualization
methods, the detected knowledge and similarities will be exploited.
3.5 Visualization
Recently, visualization techniques have gained interest in the data management
field. Visualization of information is an important tool to tell the hidden stories
about the data [41]. For example, in [27] and [12] the authors propose techniques
related with visualization of the uncertainty in a data set, while in [34] the author
9
propose an algorithm for visualization of word clouds and semantic relationships
existing between these.
The generation of RDF data has several advantages: standardization, inter-
operability, structured data, etc., however, knowledge of the RDF query language
(SPARQL) is needed when accessing this information. Therefore, as part of the
similarity detection process, visualization models modules will be developed in
order to exploit the RDF data. Text mining algorithms allow extracting valuable
information (knowledge) from raw data, so, we propose to implement models or
techniques to visualize this information with the purpose of help to final users
to envision the students mobility issue and credits recognition in a improved
manner.
4 Conclusions
In this work, we analyzed several proposals that want to improve related aspects
of syllabus data management for the subjects taught in different careers of several
Ecuadorian IES. In all these works the implementation of new technologies are
proposed to improve teaching and learning methods and techniques.
The aim of this new approach is to propose a methodological solution for the
creation of a common semantic syllabus repository and to discover behavioral
patterns and similarities among them.
In the future, with the implementation of the proposed methodology, we
intend to create a syllabus platform based on semantic technologies, as well
as to implement a series of text mining techniques that will help solving the
students mobility issue among IES.
5 Acknowledgment
This work is part of the ”Detección de similitudes entre contenidos académicos
de carrera a través de la aplicación de tecnologı́as semánticas y minerı́a de datos”
project, supported by the Ecuadorian Corporation for Research and Academia
(CEDIA for its Spanish acronym).
References
1. Abad, K., Carvallo, J.P., Espinoza, M., Saquicela, V.: Towards the creation of a
semantic repository of istar-based context models. In: Mejia, J., Munoz, M., Rocha,
Á., Calvo-Manzano, J. (eds.) Trends and Applications in Software Engineering. pp.
125–137. Springer International Publishing, Cham (2016)
2. Afros, E., Schryer, C.F.: The genre of syllabus in higher education. Journal of
English for Academic Purposes 8, 224233 (2009)
3. Allahyari, M., Pouriyeh, S.A., Assefi, M., Safaei, S., Trippe, E.D., Gutierrez, J.B.,
Kochut, K.: A brief survey of text mining: Classification, clustering and extraction
techniques. CoRR abs/1707.02919 (2017), http://arxiv.org/abs/1707.02919
10
4. Auer, S., Bhmann, L., Dirschl, C., Erling, O., Hausenblas, M., Isele, R., Williams,
H.: Managing the life-cycle of linked data with the lod2 stack. The Semantic Web
– ISWC 2012: 11th International Semantic Web Conference (2012)
5. Berners-Lee, T.: Linked data (2009), https://www.w3.org/DesignIssues/
LinkedData.html
6. Berners-Lee, T., Hendler, J., Lassila, O.: The semantic web. Scientific American
(2001)
7. CEAACES: Reforma al reglamento transitorio para la tipologa
de universidades y escuelas politcnicas y de los tipos de car-
reras o programas que podrn ofertar cada una de estas institu-
ciones (2012), http://www.ceaaces.gob.ec/sitio/wp-content/uploads/
2013/10/REFORMA-AL-REGLAMENTO-TRANSITORIO-PARA-LA-TIPOLOGI%CC%
81A-DE-UNIVERSIDADES-Y-ESCUELAS-POLITECNICAS.pdf
8. CEAACES: Suspendida por falta de calidad. el cierre de catorce universidades en
ecuador (2013), http://www.ceaaces.gob.ec/sitio/wp-content/uploads/2013/
10/CIERRE-DE-UNIVERSIDADES-placas-ok.pdf.
9. CES: Reglamento de rgimen acadmico (2013), http://www.snna.gob.ec/
wp-content/themes/institucion/dw-pages/Descargas/regimen_academico.
pdf
10. Chung, H., Kim, J.: Semantic model of syllabus and learning ontology for intel-
ligent learning system. Computational Collective Intelligence. Technologies and
Applications: 6th International Conference, ICCCI 2014 p. 175183 (2014)
11. Chung, H., Kim, J.: An ontological approach for semantic modeling of curricu-
lum and syllabus in higher education. International Journal of Information and
Education Technology 6, 365369 (2016)
12. Collins, C., Carpendale, S., Penn, G.: Visualization of uncertainty in lattices to
support decision-making. IEEE VGTC Conference on Visualization p. 5158 (2007)
13. de Cuenca, U.: Lod-gf: An integral open data generation framework (2017), https:
//ucuenca.github.io/lodplatform/
14. Cutting, D.R., Karger, D.R., Pedersen, J.O., Tukey, J.W.: Scatter/gather: A
cluster-based approach to browsing large document collections. In Proceedings of
the 15th Annual International ACM SIGIR Conference on Research and Develop-
ment in Information Retrieval p. 318329 (1992)
15. Dahiya, R., Singh, A.: A survey on text mining using genetic algorithm. Interna-
tional Journal Of Innovative Research And Development 3(5) (2014)
16. Deliyska, B., Manoilov, P.: Ontologies in intelligent learning systems. Intelligent
Learning Systems and Advancements in Computer-Aided Instruction: Emerging
Studies pp. 31–48 (2012)
17. Demartini, G., Enchev, I., Gapany, J., Cudr-Mauroux, P.: The bowlogna ontology:
Fostering open curricula and agile knowledge bases for europes higher education
landscape. Semantic Web 4(1), 5363 (2013)
18. Fellbaum, C.: Wordnet and wordnets. Encyclopedia of Language and Linguistics
p. 2665 (2005)
19. Garrido, A., Morales, L., Serina, I.: On the use of case-based planning for e-learning
personalization. Expert Systems with Applications 60, 115 (2016)
20. Group, R.W.: Resource description framework (rdf) (2004), https://www.w3.org/
RDF/
21. Habanek, D.: An examination of the integrity of the syllabus. College Teaching
53(2), 62–64 (2005), http://www.jstor.org/stable/27559222
11
22. Hokstad, T.: Ontology based study planning and classification of uni-
versity subjects (2015), https://brage.bibsys.no/xmlui/bitstream/handle/
11250/299458/Tor-Erik%20Hokstad.pdf
23. Hyland, B., Atemezing, G., Villazon-Terrazas, B.: Best practices for publishing
linked data (2014), http://www.w3.org/TR/ld-bp/
24. Jeong, Y., Song, M.: Applying content-based similarity measure to author co-
citation analysis. iConference 2016 Proceedings p. 17 (2016)
25. Jiang, Z., Li, L., Huang, D., Jin, L.: Training word embeddings for deep learning
in biomedical text mining tasks. IEEE International Conference on Bioinformatics
and Biomedicine (BIBM) pp. 625–628 (2015)
26. Joffre, C.P., Delgado, B., Kosik, R.O., Huang, L., Zhao, X., Su, T.P., ... Fan, A.P.:
Medical education in ecuador. Medical Teacher 35(12), 979–984 (2013)
27. Johnson, C.R., Sanderson, A.R.: A next step: Visualizing errors and uncertainty.
IEEE Computer Graphics and Applications 23(5), 6–10 (2003)
28. Lee, M., Chang, J., Hsieh, T.: A grammar-based semantic similarity algorithm for
natural language sentences. The Scientific World Journal (2014)
29. Meng, L., Huang, R., Gu, J.: A review of semantic similarity measures in wordnet.
International Journal of Hybrid Information Technology 6(1), 1–12 (2013)
30. Miltenoff, P., Keengwe, J., Schnellert, G.: Technological strategic planning and
globalization in higher education. Learning Tools and Teaching Approaches
through ICT Advancements pp. 348–358 (2013)
31. Musen, M.A., the Protg Team: The protg project: A look back and a look forward.
AI Matters 1(4), 4–12 (2015)
32. Nuntawong, C., Namahoot, C.S., Brckner, M.: A semantic similarity assessment
tool for computer science subjects using extended wu & palmers algorithm and
ontology. Information Science and Applications pp. 989–996 (2015)
33. Patel, R., Sharma, G.: A survey on text mining techniques. International Journal
Of Engineering And Computer Science 3(1), 5621–5625 (2014)
34. Paulovich, F.V., Toledo, F.M.B., Telles, G.P., Minghim, R., Nonato, L.G.: Semantic
wordification of document collections. Computer Graphics Forum 31(3pt3), 1145–
1153 (2012)
35. Piedra, N., Chicaiza, J., Lopez-Vargas, J., Caro, E.T.: Guidelines to producing
structured interoperable data from open access repositories pp. 1–9 (2016)
36. Piedra, N., Chicaiza, J., Lpez, J., Caro, E.T.: A rating system that open-data
repositories must satisfy to be considered oer: Reusing open data resources in
teaching pp. 1768–1777 (2017)
37. Piedra, N., Chicaiza, J., Quichimbo, P., Saquicela, V., Cadme, E., Lpez, J., ... To-
var, E.: Framework for the integration of digital resources based-on a semantic web
approach. RISTI-Revista Ibrica de Sistemas e Tecnologias de Informao pp. 55–70
(2015)
38. Poveda-Villaln, M., Surez-Figueroa, M., Garca-Delgado, M., Gmez-Prez, A.: Oops!
(ontology pitfall scanner!): supporting ontology evaluation on-line. International
Journal on Semantic Web & Information Systems pp. 7–34 (2014)
39. Sailaja, N.V., Padmasree, L., Mangathayaru, N.: Survey of text mining techniques,
challenges and their applications. International Journal of Computer Applications
146(11), 3035 (2016)
40. Segarra, J., Ortiz, J., Espinoza, M., Saquicela, V.: Integration of digital repositories
through federated queries using semantic technologies. In Computing Conference
(CLEI), 2016 XLII Latin American pp. 1–9 (2016)
41. Segel, E., Heer, J.: Narrative visualization: Telling stories with data. IEEE Trans-
actions on Visualization and Computer Graphics 16(6), 11391148 (2010)
12
42. Sumba, X., Sumba, F., Tello, A., Baculima, F., Espinoza, M., Saquicela, V.: De-
tecting similar areas of knowledge using semantic and data mining technologies.
Electronic Notes in Theoretical Computer Science 329, 149–167 (2016)
43. Surez-Figueroa, M.: Neon methodology for building ontology networks: Spec-
ification, scheduling and reuse (2010), http://oa.upm.es/3879/2/MARIA_DEL-_
CARMEN_SUAREZ_DE_FIGUEROA_BAONZA.pdf
44. Tokatl, A.M., Keli, Y.: Syllabus:how much does it contribute to the effective com-
munication with the students? Social and Behavioral Sciences 1(1), 1491–1494
(2009)
45. Tu, H.T., Phan, T.T., Nguyen, K.P.: An adaptive latent semantic analysis for
text mining. In 2017 International Conference on System Science and Engineering
(ICSSE) pp. 588–593 (2017)
46. Wu, Z., Palmer, M.: Verbs semantics and lexical selection. 32Nd Annual Meeting
on Association for Computational Linguistics pp. 133–138 (1994)