Semantic Rules Representation in Controlled Natural Language in FluentEditor

HSI 2013 Sopot, Poland, June 06-08, 2013

Abstract. The purpose of this paper is to present a way of
representation of semantic rules (SWRL) in controlled natural
language (English) in order to facilitate understanding the
rules by humans interacting with a machine. The rule
representation is implemented in FluentEditor – ontology
editor with controlled natural language (CNL). The
representation can be used in a lot of domains where people
interact with machines and use specialized interfaces to define
knowledge in a system (semantic knowledge base), e.g.
representing medical knowledge and guidelines, procedures in
crisis management or in management of any coordination
processes. Such knowledge bases are able to support decision
making in any discipline provided there is a knowledge stored
in a proper semantic way.
Keywords: controlled natural language, decision support,
ontologies, semantic web rule language.
I. INTRODUCTION
ONTROLLED Natural Language (CNL) aims to support a
human – computer communication and thus improve
easiness and efficiency of these interactions.
The top-level goal of this research is to support
decision-making in any domain assuming that we have a
proper knowledge representation and suitable algorithms to
enable reasoning.
Such a decision support system comprises of: a semantic
knowledge base, a program for integrating domain-specific
information with the knowledge base and user interfaces to
allow interacting with the system and get the right
information needed to make the right decision at the right
time [1].
Our first assumption is that we have a knowledge base,
e.g. Ontorion Knowledge Server [2], where the knowledge
is stored/ modeled using semantic technologies (that is
ontologies, semantic rules, reasoning engines).
We state that using semantic technologies applied in our
knowledge-based decision support system can improve
interoperability and effectiveness of decision-making
processes. Generally, ontologies are used for sharing
knowledge and common understanding of a particular
The paper presents controlled natural language and interfaces
implemented and being developed in FluentEditor by Cognitum
(http://www.cognitum.eu/semantics/). The software is developed by P.
Kapłański (a main architect), P. Zarzycki and A. Wróblewska.
domain of interest, which makes communication between
various beings possible and unambiguous. The various
actors may be human users with different levels of expertise
or computer programs (agents). The semantic technologies
allow for easy adaptation to changes in the modeled
knowledge and procedures without the need for intervention
in the system architecture and its internal structure. In this
case, we need to exchange only the knowledge of the system
described using ontologies and semantic rules [3].
Additionally to the whole spectrum of semantic
technologies, suitable communication patterns and
dedicated interfaces can improve human-computer
interactions easiness and efficiency and be more
understandable for human users with different levels of
expertise. We assume that we need an interface for inputting
/ modeling the knowledge into the system and an interface
using the knowledge and the inferred conclusions (similarly
as in [4][5]). Summing up, end-users require less effort to
learn how to interact with the system, since the dedicated
user interfaces are built upon well-known schema (i.e.
natural language).
In the following sections the principal contributions of
this paper are: presentation of reasons to use CNL to
represent semantic rules (section II), a method to transform
SWRL into CNL and vice-versa (section III), simple use
cases dedicated to medical diagnostics and crisis
management showing the usefulness of the representation
and preliminary version of interfaces to input the rules in
FluentEditor [6] (section IV). In the last section we discuss
and conclude the paper.
II.SEMANTIC KNOWLEDGE AND CONTROLLED NATURAL
LANGUAGE
A. Semantic Ways of Defining the Knowledge
Modeling domain ontologies and SWRL rules we are able to
define a knowledge scheme in any semantic knowledge
base. The store for the knowledge base can be implemented
in NoSQL technology (e.g. Cassandra [7], Azure Tables
[8]) or in RDF data stores (e.g. AllegroGraph [9], Virtuoso
[10]).
Relatively simple interface to model ontologies is
supported by Protégé [11] or NeOn editors [12]. However,
these interfaces are rather simple for experienced
knowledge engineers, not for common users that do not
know nuances of ontology engineering. On the other hand,
Semantic Rules Representation
in Controlled Natural Language in FluentEditor
Anna Wróblewska1,2
, Paweł Kapłański2
, Paweł Zarzycki2
, Iwona Ługowska3
1
Institute of Computer Science, Warsaw University of Technology, Warsaw, Poland,
2
Cognitum, Warsaw, Poland,
3
Department of Soft Tissue and Bone Sarcoma and Melanoma, M. Sklodowska-Curie
Cancer Centre/Institute of Oncology, Warsaw, Poland,
awroblew@ii.pw.edu.pl, p.kaplanski@cognitum.eu, p.zarzycki@cognitum.eu
C

the simplest way to represent the knowledge is to use natural
language. Nevertheless, using natural language is
unreachable for the current technology and thus for the
machines that should understand this knowledge. The most
appropriate solution seems to use controlled natural
languages.
As far as CNL is concerned, Controlled English was
successfully used by large corporations to standardize the
language used for internal communication, e.g.: Caterpillar
Technical English [13], IBM Easy English [14], Boeing
Simplified English [15] etc. The novel approach to CE
supported by knowledge representation and reasoning
requires that CE has restricted grammar and vocabulary, in
order to reduce the ambiguity and complexity inherent in
natural language. In the last years, this branch of CE
established itself in various application fields (mostly as an
interface of knowledge bases) as a powerful knowledge
representation language that is readable for humans and
processable by computers.
Attempto Controlled English (ACE) [16][17][18] is
very expressive CE and it is the one that is mostly used.
ACE can be translated into a non-decidable subset of
first-order logic. It also provides its subset called ACEOWL
[19] that can be translated into SROIQ Description Logic
(formal foundation of OWL2). In Protégé ontology editor
there is a plugin implemented to convert modelled ontology
to ACE. However, it has some drawbacks. Among other it
has no implementation of SWRL rules.
In Ontorion Knowledge Server we use CNL with formal
semantic (e.g. description logic, OWL standards, SWRL
rules) that will allow us to provide a domain ontology with
rules. Then we apply automatic reasoning services and we
generate explanations (written in CNL) of the automatically
provided implications [5].
B. Examples of Rules Representation
In the following, to make the issue more visible, we present
an example of simple rule representation that can be applied
in temporary diagnostics that a patient treated with any
therapy should have signed a proper consent before.
In Controlled English (CE) in FluentEditor, the rule can
be written as in Figure 1. The same rule edited in Protégé
editor 4.1 (which does not allow defining the rules properly,
only to edit them) is shown as in Figure 2. The rule defined
in common format of OWL and SWRL (owl/xml) is shown
in Figure 3. As it can be easily seen, the rule written in
Controlled English has a potential to be more
understandable for an average user without any additional
knowledge.
Figure 1 An example of a semantic rule written in
Controlled English in FluentEditor
Figure 2 An example of a semantic rule shown in Protégé
ontology editor
Nevertheless, representation of semantic rules in
Controlled English has some limits in respect to unrestricted
natural language (similarly as other controlled natural
languages). Generally, the sentences should be quite simple
without adjectives. Noun phrases should be written together
with dashes between words. It requires a little understanding
of ontology modelling, i.e. verbs that define semantic
relations or concepts and instances should be written as one
word (with dashes between words in natural language).
Figure 3 An example of a semantic rule shown in
OWL/XML format
III. SEMANTIC RULES IN CNL
A. Semantic Web Rule Language
The OWL 2 Web Ontology Language is an ontology
language for the Semantic Web with formally defined
meaning [20]. OWL 2 ontologies in the abstract syntax
contains a sequence of axioms and facts, i.e. provide
classes, properties, individuals, and data values and can be
stored as Semantic Web documents. Ontology axioms may
be of various kinds, e.g. subClass axioms, equivalentClass
axioms. Semantic Web Rule Language (SWRL) is proposed
to extend the OWL 2 ontologies with rule axioms.
SWRL is based on a combination of the OWL DL and
OWL Lite sublanguages of the OWL Web Ontology
Language with the Unary/Binary Datalog RuleML
sublanguages of the Rule Markup Language [21]. The
SWRL standard extends the set of OWL axioms to include
Horn-like rules and thus enables Horn-like rules to be
combined with an OWL knowledge base.
The SWRL rules (semantic rules) are of the form of an
implication between an antecedent (body) and consequent
(head). The intended meaning can be read as: whenever the
conditions specified in the antecedent hold, then the
conditions specified in the consequent must also hold.

Both the antecedent (body) and consequent (head) consist
of atoms. Atoms in these rules (Figure 4) can be of the form
C(x), P(x,y), sameAs(x,y), differentFrom(x,y) or builtIn(r, x,
…) where C is an OWL description or data range, P is an
OWL property, r is a built-in relation, x and y are either
variables, OWL individuals or OWL data values, as
appropriate. In the context of OWL Lite, descriptions in
atoms of the form C(x) may be restricted to class names. The
main safety condition of SWRL is that only variables that
occur in the antecedent of a rule may occur in the
consequent.
Figure 4 A grammar of SWRL
B. Controlled English for SWRL
We propose own specially designed methods for
representation of SWRL in Controlled English (SR-CE) in
predicative editor FluentEditor [6] (that is an interface to
the Ontorion knowledge base). The editor prohibits one to
enter any sentence that is not grammatically or
morphologically correct and actively helps the user during
sentence writing (Figure 5).
Figure 5 FluentEditor interface helps the user during
sentence writing
The very first version of CE was introduced in [22]. Up to
now, the CE language was extended, specified to be more
logical and natural [23]. Additionally, we added
representation of semantic rules. The EBNF grammar1
of
the CE is designed to be as close as possible to natural
English and can be translated to OWL2 and back easily.
Grammar of CE was implemented using context-free
grammar parser. All the mechanisms were implemented in
FluentEditor.
CE language deals with concepts, roles and instances that
can be represented by symbols in form of extended words2
.
Concept and role identifiers should start with small letter in
opposite to instances and cannot be any of keywords. The
morphology is applied if needed by separated dictionary
based module. Additionally it is required that each sentence
starts with upper-case letter and ends with „full stop‟ sign.
There are five general groups of sentences that are allowed
in FluentEditor:
1) regarding concept specifications, e.g. subsumptions,
complement, equivalence, disjointness, value partitition,
more complex concept definitions,
2) instance specifications,
3) role (object properties) specifications, e.g. inclusion,
equivalence, disjointness and other axioms (symmetry,
transitivity),
4) data attributes (properties) specifications and their data
ranges,
5) semantic rules specifications.
In the following we consider only the semantic rules.
However, in the next section we introduce some examples
considering the whole spectrum of the designed Controlled
English.
The main part of the grammar regarding semantic rules
specification in CE (SR-CE) is shown in Figure 6 and
auxiliary structures are shown in Figure 7. It is more
complicated than the SWRL grammar (Figure 4) because of
specific characteristics of a natural language.
Correspondingly, some constructions in SR-CE cannot be
translated directly into SWRL (and vice-versa).
Figure 6 The main part of EBNF grammar of SR-CE
In SR-CE rules are presented as if-clause sentences.
They consist of antecedent and consequent clauses. The
clauses comprises of atoms, likewise in SWRL.
1
Extended Backus–Naur Form (EBNF) is a family of metasyntax
notations used for expressing context-free grammars.
2
Extended words are a group of words connected with a dash (‘-‘) sign.

Nevertheless, the atoms in antecedent and consequent
(<condition> and <condition-result> respectively) are
slightly different. In <condition> atom one can use all
available objects, however in <condition-result> - only
identified objects. That is because of the safety condition of
SWRL that only variables that occur in the antecedent of a
rule may occur in the consequent. SWRL rules cannot create
new instances, there can be only new associations (new
knowledge/axioms) for the existing instances.
There is only one <condition-result> atom, in which
one can use not identified object. That is
“<identified-object> ‘is’ <not-identified-object>”. It is
used to express that an instance, represented with the
variable in sentence, is assigned to a new concept, e.g. “If a
person is-year-old greater-or-equal-to 18 then the person is
an adult.”
Atoms in antecedent can express:
- OWL property between objects: “<object><name>
<object>” and “<object> is <name> by <object>” for
an inversion of the property,
- OWL description (axioms for concepts): “<object>
exists”,
- sameAs axiom to express that two individuals are the same
and opposed differentFrom axiom: “<object>is
<object>”, “<object>is not<object>”, respectively,
- OWL data range description and data property
assignments: “<object> <name> equal-to <object>”,
“<object><name><abstract-bound>”,
- OWL complex description axioms for concepts and data
properties: “<object>is-and-is-only<complex-object>”.
Atoms in consequence are similar, but without definitional
clauses for OWL descriptions. Additionally, they use almost
everywhere identified objects, as it was mentioned before.
Figure 7 Auxiliary structures in SR-CE grammar
As far as the auxiliary grammar structures are concerned,
one can use not identified and identified object (Figure 7).
They represent variables in the semantic rule sentences. Not
identified objects are those variables that are used for the
first time in a sentence. Identified objects correspond to the
variables defined/introduced with not identified object
names. The variables of kind of “a/the <name>” denote
variables assigned to a class represented with its <name>.
The variables of kind of ”a/the thing” represent an instance
of the top concept “owl:Thing” and thus any variable in the
SR-CE sentence. Additionally, one can use multiple
variables representing instances of the same class in any
sentence. It can be achieved using an identifier denoting
subsequent variable number, e.g. “a thing (1)”,”a thing
(2)”.
There is also a possibility to define complex objects -
class and data descriptions, e.g. “If a person is-and-is-only
something that has-child at-least one person then the
person is a parent”.
C.General Implementation Tips
SR-CE grammar is translated to SWRL grammar and thus
made available to inferring in a reasoning engine (e.g.
HermiT [24]). To implement it we used OWL API [25],
which is a Java API and reference implementation for
creating, manipulating and serializing OWL ontologies. A
backward implementation of SWRL to SR-CE, e.g. to show
the result of reasoning, is also available in FluentEditor.
IV. USE CASES
A. Crisis Management Use Case
Crisis management is an activity of public administration
which is a part of national security management. Crisis
management deals with: crisis prevention, preparation,
response and recovery.
In this use case we consider and construct an ontology of
crisis management to characterize crises (crisis events,
effects, risks and dangers), the studied world (crisis
environment: people, natural sites, goods) and treatment
system (communication and coordination actors,
procedures and tasks and the treatment infrastructure, e.g.
resources).
This use case deals with response to a crisis. It concerns a
tanker accident and its effects. A tanker truck (containing
unknown toxic substance) had an accident. Several children
of the near kindergarten (outside when the accident
happened) feel sick.
We use the ontology and reasoning engine to find
answers to questions: what the dangers can occur and what
emergency services are needed. Some semantic rules in the
use case are presented in Figure 8 and Figure 9. A fragment
of the ontology (also instances modelling the use case),
some dedicated semantic rules and provided inferred new
knowledge that are written in CE are shown in Figure 10.
The developed system infrastructure (FluentEditor and
Ontorion Knowledge Server) can deduce that, e.g. in our use
case we need such rescue teams as police, fire brigade and
emergency medical services and that the crisis (tanker
accident) causes a risk of contamination.

If a crisis has-type a crisis-type and an actor reacts-to-type
the crisis-type then the crisis needs the actor.
If an effect causes a potential-danger and a crisis induces
the effect and an actor assumes a procedure and the procedure
reduces the potential-danger then the crisis needs the actor.
If a crisis induces an effect and the effect concerns a
people-group and the people-group has-victims
greater-or-equal-to 1 and an actor is
Emergency-Medical-Service then the crisis needs the actor.
If a crisis induces an event and the event causes a risk then
the risk appears-during the crisis.
Figure 8 Semantic rules written in SR-CE and used in crisis
management use case
Figure 9 Semantic rules used in crisis management use case
and presented in Protégé ontology editor
Figure 10 A fragment of crisis management ontology and
semantic rules written in CE in FluentEditor. At the bottom
of the figure/interface inferred new knowledge is shown
B. Oncology Guidelines Use Case
Application of clinical practice guidelines is mandatory in
current evidence-base medicine, including oncology [26].
The knowledge contained in the guidelines is dynamic and
improved regularly. The SR-CE allows writing the
knowledge in a relatively easy way for doctors and read
inferred recommendations.
In this use case we consider and build a very simplified
ontology dealing with a few rules from gastrointestinal
stromal tumour (GIST) guidelines [27]. A sarcoma may
have characteristics. A patient may have favourable or
unfavourable prognostic factors. The recommended therapy
is dependent on the factors and a patient overall state.
Semantic rules in the use case are presented in Figure 11
and Figure 12. A fragment of the ontology (also instances
modelling the use case), dedicated semantic rules and
provided inferred new knowledge that are written in CE are
shown in Figure 13. The developed system infrastructure
can deduce that the considered patient has a
recommendation for the adjuvant therapy with a drug -
imatinib.
If a gist has-size greater-or-equal-to 5 then the gist
has-characteristic Large-Gist.
If a patient has a gist and the gist has-characteristic
a characteristic and the characteristic constitutes a
prognostic-factor and the prognostic-factor is
Unfavorable-Prognostic-Factor then the patient
has-prognostic Unfavorable-Prognostic-Factor.
If a patient has a gist and the patient is-in-state
After-Surgery-State and the patient has-prognostic
Unfavorable-Prognostic-Factor then
Adjuvant-Therapy-With-Imatinib is-recommended-to the
patient.
Figure 11 Semantic rules written in SR-CE and used in
oncology guidelines use case
Figure 12 Semantic rules used in oncology guidelines use
case and presented in Protégé ontology editor
Figure 13 A fragment of simplified oncology guidelines
ontology and semantic rules written in CE in FluentEditor.
At the bottom of the figure/interface inferred new
knowledge is shown
V.CONCLUSION
In this paper we investigated a top-level problem of
decision-making support with the use of semantic
technology and knowledge bases. We described a concept
of using controlled natural language to support

human-machine interactions. We presented a grammar of
SWRL rules written in controlled English and the results of
translation from CE to SWRL and vice-versa. We provided
simple use cases showing the usability of the methods
implemented in FluentEditor.
In the future we plan to elaborate CNL mechanisms for
Polish language and improve more usable interfaces for a
non-experienced user.
ACKNOWLEDGMENT
We would like to gratefully thank Dr. Aleksandra Zięba,
PhD in political science and an expert in crisis management,
for her valuable remarks regarding the crisis management
use case.
REFERENCES
[1] Eberhardt J, Bilchik A, Stojadinovic A (2012) Clinical decision
support systems: potential with pitfalls. J. Surgical Oncolology 105
(5): 502-510
[2] Ontorion Knowledge Server,
http://www.cognitum.eu/semantics/Ontorion/
[3] Sheth A, Ramakrishnan C (2003) Semantic (Web) Technology In
Action: Ontology Driven Information Systems for Search, In:
“Integration and Analysis, Bulletin of the Technical Committee on
Data Engineering”, 26 (4): 40-48
[4] R.N. Shiffman, G. Michel, M. Krauthammer, N.E. Fuchs, K.
Kaljurand, T. Kuhn: Writing Clinical Practice Guidelines in
Controlled Natural Language. Springer LNCS 5972, 2010
[5] A. Wróblewska, A. Zięba, R. Mieńkowska –Norkiene, P. Kapłański,
P. Zarzycki: Application of Semantic Knowledge Management
System in Specific Areas of Polish Public Administration. Roczniki
Kolegium Analiz Ekonomicznych SGH (Annals of the College of
Economic Analysis SGH), 2013, in print
[6] FluentEditor, http://www.cognitum.eu/semantics/FluentEditor2
[7] http://cassandra.apache.org
[8] http://www.windowsazure.com/en-us/develop/net/how-to-guides/ta
ble-services
[9] http://www.franz.com/agraph/allegrograph
[10] http://virtuoso.openlinksw.com/rdf-quad-store
[11] Protégé ontology editor, http://protege.stanford.edu/
[12] NeON Toolkit, http://neon-toolkit.org
[13] C. Kamprath, E. Adolphson, T. Mitamura, and E. Nyberg.
“Controlled Language for Multilingual Document Production:
Experience with Caterpillar Technical English”. 1998.
[14] A. Bernth. “EasyEnglish: a tool for improving document quality”. In:
Proceedings of the fifth conference on Applied natural language
processing, pp. 159–165, Association for Computational
Linguistics, Stroudsburg, PA, USA, 1997.
[15] R. H. Wojcik, J. E. Hoard, and K. Holzhauser. “The Boeing
Simplified English Checker”. In: Proceedings of the International
Conference Human Machine Interaction and Artificial Intelligence
in Aeronautics and Space, pp. 43–57, Toulouse: Centre d’Etudes et
de Recherches de Toulouse, 1990.
[16] N. E. Fuchs, U. Schwertel, and R. Schwitter. “Attempto Controlled
English - Not Just Another Logic Specification Language”. In:
LOPSTR ’98: Proceedings of the 8th International Workshop on
Logic Programming Synthesis and Transformation, pp. 1–20,
Springer-Verlag, London, UK, 1990.
[17] J.L. De Coi, N.E. Fuchs, K. Kaljurand, T. Kuhn. Controlled English
for reasoning on the Semantic Web. In F. Bryand, J. Maluszynski,
editors, Semantic Techniques for the Web - The REWERSE
Perspective, LNCS 5500: 276-308, Springer, 2009.
[18] Attempto project, http://attempto.ifi.uzh.ch/site/description/
[19] K. Kaljurand. Attempto Controlled English as a Semantic Web
Language. PhD thesis, Faculty of Mathematics and Computer
Science, University of Tartu, 2007.
[20] OWL 2 Web Ontology Language Document Overview (Second
Edition). W3C Recommendation 11 December 2012.
http://www.w3.org/TR/owl2-overview/
[21] I. Horrocks, P.F. Patel-Schneider, H. Boley, S. Tabet, B. Grosof, M.
Dean: SWRL: A Semantic Web Rule Language. Combining OWL
and RuleML. W3C Member Submission 21 May 2004,
http://www.w3.org/Submission/SWRL/
[22] P. Kapłański, Controlled English interface for knowledge bases,
“Studia Informatica” 2011, vol. 32, no. 2A (96), pp. 486-494
[23] P.Kapłański: Ontology-aided software engineering. PhD thesis.
Gdańsk University of Technology, 2012
[24] B. Glimm, I. Horrocks, B. Motik, G. Stoilos:. Optimising Ontology
Classification. In: P.F. Patel-Schneider, Y. Pan, P. Hitzler, P. Mika,
L. Zhang, J.Z. Pan, I. Horrocks, B. Glimm, editors, Proc. of the 9th
Int. Semantic Web Conf. (ISWC 2010), volume 6496 of LNCS,
pages 225--240, Shanghai, China, November 7-11 2010. Springer.
http://www.hermit-reasoner.com/
[25] M. Horridge, S. Bechhofer. The OWL API: A Java API for OWL
Ontologies. Semantic Web Journal 2(1), Special Issue on Semantic
Web Tools and Systems, pp. 11-21, 2011.
http://owlapi.sourceforge.net/
[26] ESMO guidelines,
http://www.esmo.org/education-research/esmo-clinical-practice-gui
delines.html
[27] NCCN Clinical Practice Guidelines in Oncology. Soft Tissue
Sarcoma. Version 2.2012

Semantic Rules Representation in Controlled Natural Language in FluentEditor

More Related Content

What's hot (19)

Similar to Semantic Rules Representation in Controlled Natural Language in FluentEditor (20)

More from Cognitum (10)

Recently uploaded (20)

Semantic Rules Representation in Controlled Natural Language in FluentEditor