=Paper=
{{Paper
|id=Vol-1523/STIDS_2015_T10_Donohue_etal
|storemode=property
|title=Controlled and Uncontrolled English for Ontology Editing
|pdfUrl=https://ceur-ws.org/Vol-1523/STIDS_2015_T10_Donohue_etal.pdf
|volume=Vol-1523
|dblpUrl=https://dblp.org/rec/conf/stids/DonohueKBBMGPRS15
}}
==Controlled and Uncontrolled English for Ontology Editing==
https://ceur-ws.org/Vol-1523/STIDS_2015_T10_Donohue_etal.pdf
Controlled and Uncontrolled English for Ontology
Editing
Brian Donohue Robert Ganger Tien Pham Amardeep Barry Smith
CUBRC CUBRC Army Research Bhattal University at
Buffalo, NY Buffalo, NY Lab IBM Buffalo
Adelphi, MD Southampton, UK Buffalo, NY
Douglas Kutach Ron Rudnicki
CUBRC CUBRC Geeth de Mel Dave Braines
Buffalo, NY Buffalo, NY IBM IBM
White Plains, NY Portsmouth, UK
Abstract—Ontologies formally represent reality in a way not represented in the underlying ontology, but this can be
that limits ambiguity and facilitates automated reasoning and perceived by users as unfriendly. Second, applications can
data fusion, but is often daunting to the non-technical user. let users enter such information, but this allows the ontology
Thus, many researchers have endeavored to hide the formal to be modified in ways that do not follow best practices.
syntax and semantics of ontologies behind the constructs of
Controlled Natural Languages (CNLs), which retain the In this report, we specifically examine the interplay
formal properties of ontologies while simultaneously between OWL 2 DL [1] and the International Technology
presenting that information in a comprehensible natural Alliance’s Controlled English, ITA-CE [2]-[6] to determine
language format. In this paper, we build upon previous work the feasibility of using ITA-CE as a medium through which
in this field by evaluating prospects of implementing ontologies can be correctly modified by non-technical users.
International Technology Alliance Controlled English (ITA-
CE) as a middleware for ontology editing. We also discuss at There are two main conclusions we have drawn from this
length a prototype of a natural language conversational investigation. First, a conversational interface application
interface application designed to facilitate ontology editing via can assist users in ontology editing tasks. We developed a
the formulation of CNL constructs. prototype software application that can be used either as a
command-line application or as part of the conversational
Keywords—Ontology; Controlled English; Intelligence
Collection panel found in IBM’s ITA-CE processing environment
called ‘CE Store’. Our application allows people to converse
I. INTRODUCTION with a computer in everyday English so that the user’s
intentions regarding ontology can be rendered into a CNL
Ontologies formally represent reality in a way that limits
ambiguity and facilitates automated reasoning and data equivalent. The CNL command can then be passed on to
additional machine agents, which modify the ontology and
fusion. Many technologies are available for building,
store the result in an OWL file. The application can easily
sharing, and using ontologies, including Web Ontology
Language (OWL) and controlled natural languages (CNLs). be extended to ingest other data formats such as relational
On the one hand, OWL provides effective representation databases and ontology formats other than OWL. At the
constructs and enables efficient reasoning procedures but is same time, we discovered that such software does not
daunting to the non-technical user. On the other hand, strictly require the use of CE statements in general or of
ITA-CE statements in particular. However, within the
CNLs, which are restricted versions of natural languages,
current implementation of the conversational interface
provide a human-friendly representation format that is easier
for non-technical users but there is no established standard application, ITA-CE is presented to human users as
for how statements of CNLs should map onto assertions unambiguous confirmation prompts to ensure that the user’s
natural language commands were interpreted correctly.
defining an ontology.
Motivated by the accessibility of CNL, we explored how to The second conclusion was that ITA-CE might serve as a
create software infrastructure that would enable users to convenient communication medium for analysts and
interact with an OWL knowledge base through CNL developers handling information sources in a variety of
constructs. We see the value of such an infrastructure for the formats, and by enabling machine agents to exploit
intelligence, defense and security communities as being additional information sources when attempting to interpret
realized in the use of ontology-driven information collection requests made by users.
applications. Such applications typically have one of two The rest of the document is structured as follows. Section
opposing shortcomings. First, users can be prevented from II discusses existing approaches to utilizing CNL-based
entering information about an entity if that type of entity is approaches to modifying ontologies. Section III provides an
STIDS 2015 Proceedings Page 74
overview of ITA-CE. In Section IV, we provide a detailed 2014 connects users with OWL2 DL’s full expressive
discussion of our implementation to support conversational potential.
OWL ontology editing. In Section V, we introduce a few Second, some of the editors employ insufficient resources
illustrative scenarios to show the applicability of our work. for explaining and correcting user input error. CLOnE
We conclude in Section VI by sketching future directions of Editor users, for example, complained of receiving little
our work. guidance for inputting CNL expressions and no feedback to
II. CONTROLLED NATURAL LANGUAGE ONTOLOGY explain syntactical errors [11]. Similarly, users of ROO
EDITORS Editor complained of receiving no feedback on semantic
errors [13].
The formal underpinnings of semantic technologies are
substantial obstacles for a casual end-user. This usability Third, the documented experiments conducted with CNL
problem has been widely noted already within the Semantic ontology editors suggest that all but the simplest editing
Web. For example, Rector [7] documented numerous errors tasks lie beyond the ordinary capabilities of the CNL
commonly made by non-expert ontology users. These ontology editor user. Of the editors surveyed above, GINO,
include (1) the failure to make all information explicit, (2) CLOnE, and ROO ran experiments with non-expert
ignorance of the effects of range and domain restrictions, (3) ontology users. In each case, however, users were successful
mixing up defined and primitive classes, (4) only in performing basic tasks (e.g., most users could create
misunderstanding common logical constructs (‘and’, ‘or’, a class, property, or instance), and for the most part
‘some not’, ‘not some’), (5) presuming that classes are unsuccessful in executing any more sophisticated task (e.g.,
disjoint by default, and (6) being insensitive to open world correctly adding an axiom). If the usability problem is to be
reasoning. Thus, non-expert ontology users face an immense solved altogether, then end-users will need an interface that
hurdle developing and utilizing ontology-based information relieves much of the burden of expressing the wide array of
sources. OWL constructs.
In response to the usability problem, previous research [8]- Fourth, all of the editors require end-users to master the
[14] has sought to hide the formal syntax and semantics of stringent lexical and syntactic rules governing the
ontologies behind CNL constructs. Several projects in implemented CNL. For instance, users must be sensitive to
particular have sought to exploit existing CNLs, or develop quantifiers (‘every’, ‘some’), disallowed terms (‘or’, ‘not’),
new CNLs, in order to simplify the tasks of creating, and peculiar lexical conventions (e.g., in ACE, dashes
managing, and navigating ontologies.1 Several software between multiple elements of a term, e.g., ‘stretch-of-river’).
applications now allow users to edit ontologies by writing Although the GINO and Fluent editors assist users in
English sentences that are restricted in admissible entering CNL through predictive mechanisms, and although
vocabulary and grammatical constructions, yet relatively CLOnE and ROO employ less stringent lexical and syntactic
easily comprehended. This sets them apart from traditional restrictions, the fact remains that users are expected to write
tree-structured and graph-structured ontology editors such as impeccable CNL sentences. Thus, even if CNL-based
Protégé and TopBraid Composer™. editors are more human friendly than traditional tree-
structured or graph-structured ontology editors, they
To a great extent, these CNL ontology editors have helped nevertheless run the risk of alienating the non-expert user, as
to bridge the gap between casual users and ontological put forward by Smart [15].
formalisms. However, their success is limited by at least one
of four recalcitrant problems. First, most of the editors are Related projects include formulations of alternative OWL
not fully compliant with OWL 2 DL, the most widely used syntaxes, aimed at simplifying OWL for non-expert users,
member of the OWL family of languages. ACE View does and resulting in an ontology language that resembles, in
not currently support sentences that express data properties many of its properties, a CNL. In particular, we note
and their corresponding datatypes (e.g., having a date-time Manchester OWL Syntax [16], which is recognized by
value associated with an event). GINO Editor and CLOnE W3C, and Sydney OWL Syntax [17].
Editor restrict users to only very basic OWL constructs (e.g., The software discussed in this paper aims to overcome the
there appears to be no support for inserting class axioms or usability problem in a different way. If it were fully
cardinality restrictions). The ROO Editor employs a more developed, it would minimize people’s direct interaction
expressive language than ACE View, GINO Editor, and with CNLs and eliminate altogether the requirement for
CLOnE Editor, but it shies away from OWL 2 DL them to write syntactically correct sentences of a CNL.
expressivity for simplicity [12]. Only the Fluent Editor™ Instead, a non-technical user could engage in ontology
creation, editing, and management entirely by means of a
1 natural language human-machine conversation. At most a
There has been little comparative study of the vast array of CNLs actively
in use. Kuhn (2014) surveys 100 recent CNLs, and Schwitter (et al.,
user would need to read a sentence of CNL in order to
2008) offers a detailed comparison of three of the more prominent CNLs in confirm that the conversational agent has correctly
use today: Attempto Controlled English (ACE), Ordnance Survey Rabbit, interpreted the natural language input. Behind the scenes,
and Sydney OWL Syntax (SOS). the conversational agent would translate the user’s
STIDS 2015 Proceedings Page 75
commands into a CNL, which would then be passed along This rule states that the “is married to” relation is
to other agents that have access to particular knowledge symmetric. If a user includes this rule in a model and also
representation documents, e.g., an OWL/XML file. In our includes the sentence “the person Steve is married to the
current preliminary implementation of this idea, all of the person Jane,” then software can easily and correctly infer
interaction can take place within the CNL processing “the person Jane is married to the person Steve.”
environment designed by the International Technology Models can be defined and extended using the ITA-CE
Alliance (ITA) and within the context of previous ITA
processing environment, called ‘CE Store’ [5].2 ITA-CE’s
research on CNL-based tools and their military applications.
CE Store software allows users to define and execute
III. OVERVIEW OF ITA CONTROLLED ENGLISH custom “CE agents,” including conversational agents with
which users can converse in ordinary English. Below, we
In 2010, the International Technology Alliance (ITA) began present a pair of CE agents, one of which assists the user in
developing a CNL known as ITA Controlled English (ITA- formalizing natural language expressions and the other of
CE) for the purpose of supporting tasks within the Data-to- which retrieves and emends information within OWL files.
Decisions (D2D) framework, “specifically to assist coalition
decision makers in distributed information environments IV. CONVERSATIONAL INTERFACE APPLICATION
through automated or semi-automated fusion processes” [4].
The conversational interface application that we coded
Previous ITA-CE research has addressed the problem of the cooperates with the CE Store software, which takes the form
miscommunication between US and UK military personnel of a web application that accepts ITA-CE sentences as input
rooted in lexical and cultural discrepancies and the problem and responds by remembering the ITA-CE sentence and
of enhancing shared understanding and communication for possibly triggering other behavior based on the content of
military decision-making, especially through the the ITA-CE sentence. Our software is an add-on to the CE
exploitation of sensor resources. An example of this is
Store in the sense that it is a WAR file that can be placed in
MOIRA (Mobile Intelligence Reporting App), which aims
the same folder as the CE Store's WAR files and is
to expedite data requisition within ISR missions.
configured by storing a few CE statements in the CE Store.
At base, the syntax and semantics of ITA-CE are adopted
Given the suitability of the CE Store for incorporating
from Sowa’s Common Logic Controlled English [18]-[19],
ontology information, one possible strategy for editing
which in turn aligns itself closely to first-order logic. As ontologies by means of ITA-CE is first to translate the
with most CNLs, the resultant expressions of the language
English into ITA-CE and then ITA-CE into OWL. In
are readily comprehensible to ordinary English speakers. practice, however, such a strategy faces some obstacles.
For example:
there is a person named Steve.
To see why, first observe that in order for the software to
the person Steve is married to the person Jane. guide the user in editing an ontology, it must somehow
the person Steve has the person Jane as spouse. access all the relevant information about the ontology.
By writing ITA-CE sentences, users can gradually construct Because the CE Store software stores its data in the form of
a model, in which all pertinent entities within that model – ITA-CE statements, the ontology would first need to be
including types, properties, relations, and individuals – are loaded into CE Store. This would incur some data
specified. Thus, to construct a model, a user would write redundancy and would require the addition of some rules to
sentences defining the objects, properties, and relationships draw inferences not explicitly asserted by the ontology.
within that domain. Neither of these is a sizable obstacle. However, if we were
to adopt this strategy, and then a user made changes to the
New terms are introduced to the model by means of ontology, then the previous statements might become false.
“conceptualise” statements. For example: For example, an ontology might define a hierarchy in which
conceptualise a ~ Chihuahua ~ C that is a dog. Z is a direct subtype of X. If a user chooses to insert a new
conceptualise the Chihuahua C ~ barks at ~ the type Y so that X is a parent of Y and Y is a parent of Z, the
person P1.
conceptualise the Chihuahua C has the person P2 as
previous statement will need to be deleted. Deletion can
~ owner ~. pose a problem for the CE Store because new statements are
sometimes automatically inferred according to the rules
The term being added or modified is set off by tildes (~), already present in the store.
and if the new term is a noun, it is followed by an uppercase
variable name. Although limited deletion of sentences is no obstacle, some
kinds of changes to an ontology (that ought to be allowed)
Additionally, users can write rules of inference to the model could potentially require identifying and deleting sentences
in the form of “if-then” statements. For example: en masse, at least if there are already statements about
if instances of the classes in the ontology. For example, we
( the person P1 is married to the person P2 )
then
2
( the person P2 is married to the person P1 ) . An alpha version is publically available for download at
http://ibm.co/RDIa53.
STIDS 2015 Proceedings Page 76
might have a relation “receives information from” between sometimes responding with advice or questions. The output
artifacts, with some instances like “the machine KRF343 to the user is often in ordinary English, but when the
receives information from the machine EELR.” and “the OntologyAgent is prompting the user for confirmation, it
machine KRF343 receives information from the machine provides a controlled English statement to coach the user so
KRF343.” Now suppose that someone alters the ontology by that on future occasions, the user can use controlled English
marking the “receives information from” relation as for quicker unambiguous communication.
reflexive, which triggers (for every artifact X) the addition
When the OntologyAgent needs information about the
of a statement, “X receives information from X.” Then, if
current state of the ontology, it sends out requests for
the reflexivity of this relation is removed, the proper information by placing appropriately structured ITA-CE
behavior would not be to have all such statements removed, sentences in the CE Store, which in turn trigger a response
but to remove only those that statements whose existence by another specially designed Java class, OwlAgent, which
were generated by the rule. In that case, we might need to
loads OWL ontology files and operates automated reasoners
keep “the machine KRF343 receives information from the to answer questions about the inferred ontology, sending
machine KRF343,” because, say, it is a special radio that answers back to the OntologyAgent through ITA-CE
broadcasts messages and also records what it broadcasts in statements placed in the CE Store. Information is shuttled
addition to what it receives from other broadcasters. The CE
back and forth between the OwlAgent and OntologyAgent
Store can keep track of which rules its sentences are derived until the OntologyAgent feels confident about the meaning
from (as “rationale graphs”); even then, however, there of the user’s initial request. Once the user has confirmed the
remains a problem of managing which information should accuracy of the OntologyAgent’s interpretation, it passes an
be deleted.
ITA-CE string to an OwlAgent via the CE Store
A second related worry is that the ontology editing software conversational interface, the OwlAgent analyzes the
does not have exclusive control over the CE Store, nor can it requested modification, and, so long as it protects the
verify ahead of time whether all the rules CE Store will be integrity of the OWL file, updates the ontology accordingly.
compatible with changes that a user makes. If a user Given this structure, the program can (and does) log changes
accidentally adds a statement to the ontology which renders to all ontologies, keeps different users' ontologies separate,
it inconsistent, it is possible that rules that trigger on the and saves changes to the ontologies incrementally to allow
statements in the ontology will generate conflicting users to undo changes.
assertions, which can trigger further undesired effects that
In order to enhance its interpretive capabilities, the
might be hard to predict and difficult to undo.
OntologyAgent attempts to leverage information already
A third deficiency of the CE Store for maintaining the present within the OWL file. It does so by posing queries to
ontology information that users are editing is that it does not the OwlAgent (again, via the CE Store), such as, ‘Does the
provide a means for keeping different users’ ontology class C exist in the ontology?’ and ‘Does the user’s request
changes separate. In its current state, the CE Store does not violate any domain or range restrictions on object
allow any name spaces, which could differentiate between properties?’ The OwlAgent, who manages changes to OWL
conflicting definitions of an entity. files, answers the OntologyAgent’s queries, for example,
Consequently, we found it advisable to adopt two design ‘There is no such class in the OWL file’ or ‘That requested
policies. First, it should not be required for the CE Store to change would violate a domain restriction’. As a result, the
ingest the ontology information, which is already in the OntologyAgent can make suggestions based on the
information already contained within the OWL file and thus
OWL file. Second, we should not require the content of the
OWL file match what ontology information (if any) kept in provide guidance to users wishing to modify an ontology.
the CE Store. At the same time, we determined that we Currently, the software allows users to add classes anywhere
could still use the CE Store by inserting ITA-CE statements in the class hierarchy and to add any desired existential
about the user’s intention to make an ontology change or restrictions. With the basic framework having been coded, it
is straightforward to expand the software to allow other
about what information about the ontology the user would
types of ontology changes.
like to obtain.
The structure of the resultant application is such that a user Consider an example. The user, who is logged in as ‘Steve’,
types “I want to add Citizen as a subclass of Person.” into
types ordinary English sentences into either a terminal
window or the conversational interface of the CE Store, the conversational interface. The OntologyAgent is part of
which triggers a response from a specially designed Java the CUBRC software and conducts some basic natural
class, OntologyAgent. This Java class parses English input language processing to identify that the string matches one
in attempt to infer the user's intentions regarding the loaded of the allowed forms for stating that that one class X should
ontology. In the special case where the user's input is in be made a subclass of another class Y. The user could have
controlled English, the OntologyAgent can act on it without just typed “citizen is person” and gotten the same result. The
needing to ask the user for further clarification, but when the interface is designed to be very forgiving about what it
user states a command in ordinary English, the accepts. After getting confirmation from the user, the
OntologyAgent will do its best to interpret the input,
STIDS 2015 Proceedings Page 77
OntologyAgent passes the user’s command into the CE The reason for passing two strings to the OwlAgent that
Store using a controlled English statement express the same command is that (1) it is easy for the
there is a request card named 'msg_1' that is from OntologyAgent to generate a tiny amount of additional text
the service 'OntologyAgent' and is to the to express the same command in different formats, and (2)
individual 'OwlAgent' and has and has 'citizen' as any agents that are coded to receive commands from an
argument one and has 'person' as argument two and
has 'Steve' as human and has 'Command_AddClass' as
OntologyAgent can be coded to respond to whichever
command and has 'conceptualise a ~ citizen ~ C format is easiest to parse. In our conversational interface
that is a person.' as content. software, the Java methods called by the OwlAgent
This string, when it enters the CE Store is recognized as a correspond exactly with the command strings that the
OntologyAgent sends to the OwlAgent. The correspondence
exists because there are only a small number of types of
ontology alterations that it is reasonable to allow. Because
OntologyAgents and OwlAgents understand that adding
class X as a subclass of class Y is something a user probably
wants to do, an OntologyAgent could encode that command
in ITA-CE and have an OwlAgent decode it, but it is a bit
simpler for the CUBRC software (and no harder for the CE
Store) to have the OwlAgent ignore the ITA-CE altogether
and check the command string and its arguments directly.
This raises interesting questions about what role, if any,
controlled English can or should play in a conversational
interface for editing ontologies. In the current version of the
software, ITA-CE strings are used to provide a logically
unambiguous yet human-readable confirmation prompt, but
are not needed for processing the ordinary English input.
Three questions in particular arise. Should some CNL be
used even for the confirmation prompt? Should some CNL
be used in this software for transmitting messages between
the OntologyAgent and OwlAgent? And is ITA-CE the best
form of CE to be used if we do use CE for user
confirmation?
Concerning the first, the simplicity and lack of ambiguity
provides a reason to use it for a command confirmation. But
this alone is not a strong reason because it is often possible
to make CNL statements even clearer in meaning by
legitimate CE statement. Because it is recognized as a card dropping some of the formality. For example, instead of the
to the individual ‘OwlAgent’, the OwlAgent coded as part formula in ITA-CE:
of the CUBRC software is passed the string through a Java conceptualise a ~ citizen ~ C that is a person.
method. The OwlAgent then parses the string to find that it
should check whether ‘citizen’ can be added as a subclass of the computer could offer
‘person’. It does so by making calls to a layer of the conceptualise a citizen that is a person.
CUBRC software that keeps track of which ontologies are without any loss of content. The tildes and the variables just
being edited by which humans and limits the kind of indicate which concept is being defined, which is often clear
ontology changes permitted. This OWLAPI wrapper then from context or can be emphasized with boldface if desired.
makes calls to a publicly available Java API called A stronger reason for using CNL is that it helps to coach the
OWLAPI, which is more complex and permits a much user so that the next time the user wants to enter the same
wider range of operations on ontologies, especially those kind of request, the user can use the sort of phrasing that
stored as OWL files. appeared in previous prompts.
Note, however, that the OntologyAgent sends redundant Concerning the second question of whether CNL should be
information to the OwlAgent. In the content string it sends a used for the transmission of messages, there are some good
statement that obeys the syntax and semantics of ITA-CE: arguments for its use. In our particular software, we
conceptualise a ~ citizen ~ C that is a person. controlled both ends: the natural language processing and
In the command, argument one, and argument two strings, it the ontology processing. But if we think more broadly about
conveys in effect the same information: how the conversational interface could be used, we find that
an OntologyAgent could potentially get clues about how
Command_AddClass 'Citizen' 'Person'
best to advise the human user from a wide range of
STIDS 2015 Proceedings Page 78
additional sources beyond what is kept in the edited The user has other ways to communicate the desire to add a
ontology. Given the existing software framework, one could class. The following statements are equivalent to “An
code agents that respond to an OntologyAgent’s information animal is an object.”
requests by looking at a relational database, dictionaries, or animal is object
even special purpose ontology advisors. Or one could code Animals are objects.
agents that edit ontologies stored in Open Biological and I would like to make animal a subclass of object.
Make animal a kind of object.
Biomedical Ontology (OBO) files rather than OWL files.
The separation of the functions requires some sort of The software also allows users to insert a new class in
common communication format. Because CNL is relatively between two existing classes, one of which is a direct
easy to parse and easy to extend in order to add additional subclass of the other.
functionality, it could serve well in this role. A further Insert the class organism between object and
benefit of allowing CNL to play this role would be that one animal.
could easily set up logging and auditing software to work
B. Adding Existential Restrictions
with the CE Store in order to keep track of which ontology
changes were being made and by whom. Such information The other main capability of the software allows the user to
could be handy for later analysis to ascertain common add a new existential restriction on a class. The user can
ontology editing patterns, which could then be used to type something like
automate revisions to ontologies. Every sensor observation is about some detected
material entity.
Concerning the third question, it is important to keep in
or
mind that ITA-CE is severely restricted in its core lexicon Sensor observations are about detected material
and syntax. For example, the words ‘every’, ‘some’, ‘or’, entities.
and ‘not’ are not considered grammatical. While these same
ideas can be expressed indirectly, the resultant statements When receiving such a statement, the OntologyAgent sends
and rules necessary for that expression tend to be too a test message to the OwlAgent to see if the class “sensor
observation” can have an existential restriction with the
awkward for casual users (e.g., ‘or’ is defined in terms of
complex statements involving numerous cases of ‘and’ and relation “is about” and the class “detected material entity”.
If it can, the OwlAgent lets the OntologyAgent know this,
‘it is false that’). Thus, extending the lexicon and syntax of
ITA-CE would greatly facilitate user understanding if these and then the OntologyAgent asks the user for confirmation.
terms could appear in the confirmation prompts and in any If the user answers, “yes”, the OntologyAgent tells the
communication to reduce complexity. In this way, ITA-CE OwlAgent to add the appropriate axiom and the OwlAgent
could emulate some of the existing CNL-based ontology does so, logging the change in case the user later wants to
editors mentioned above, insofar as these editors utilize undo it. If the user answers “no” or doesn’t respond or
CNL constructs that more closely resemble OWL issues some alternative statement, the OntologyAgent will
constructs. forget about the attempt to add the existential restriction.
C. Adding Existential Restrictions in the Face of
V. USE CASES: MODIFYING THE SENSOR ONTOLOGY Obstacles
A. Adding New Classes In order to demonstrate some of the more sophisticated
It is easy to add a new class to an ontology using the capabilities of our application, we created a use case with
conversational interface. The easiest way to do this is for the the following vignette:
user to use the “is a” form that is standard in Basic Formal Your current version of the CUBRC Sensor Ontology
Ontology. The user types includes a class called ‘Detected Material Entity’ but
An animal is an object. does not include any additional information about it.
At this point, the OntologyAgent sends a test message to the You want to introduce a class called ‘Sensor
OwlAgent to see if the class “animal” can be assigned as a Observation’ and make the ontology understand that a
direct subclass of “object”. If it can, the OwlAgent lets the sensor observation is the kind of thing that is about
OntologyAgent know this, and the OntologyAgent asks the detected material entities. The ontology already has an
user for confirmation. If the user answers, “yes”, the ‘is about’ relationship but does not know the term
OntologyAgent tells the OwlAgent to add the appropriate ‘Sensor Observation’.
axiom and the OwlAgent does so, logging the change in
Your goals, then, are (1) to add a new class for sensor
case the user later wants to undo it. If the user answers “no”
observations, (2), correctly situate that class within the
or doesn’t respond or issues some alternative statement, The
existent hierarchy of classes, and (3) describe its
OntologyAgent will forget about the attempt to add
relationship to detected material entities.
“animal”.
This task is more complicated than simply adding an
existential restriction, but nonetheless quickly doable for the
user who interacts with the OntologyAgent. The process
STIDS 2015 Proceedings Page 79
begins when the user enters a natural language expression 4. An information content entity
If is one of these, just let me know the number,
that informally captures his request to add an existential or you can write something of the form, "a sensor
restriction. The user could type anything like the following: observation is a ___."
I want to make every sensor observation be about As it turns out, the user Steve decides to situate Sensor Ob-
some detected material entity.
I want to make sensor observation be about servation under the parent class of Descriptive ICE, so he
detected material entities. types
I want sensor observations to be about detected
1
material entities.
Sensor observations should be about detected The OntologyAgent responds by adding ‘sensor
material entities.
Sensor observations are about detected material observation’ as a direct subclass of the ‘descriptive
entities. information content entity’ class. Once the OwlAgent
a sensor observation is about a detected material reports that that change has been made, the OntologyAgent
entity. knows that the obstacles have been cleared to allow the
With such a request, the OntologyAgent queries for the class existential restriction to be added. So the OntologyAgent
‘sensor observation’. Discovering – as the vignette stipulates immediately sends the command to the OwlAgent without
– that there is no class by that name in the ontology, the needing further confirmation. When the OwlAgent reports
OntologyAgent begins to search for any clues that will help that this further command has succeeded, the
interpret the user’s intention. The OntologyAgent first OntologyAgent reports to the user the following:
checks the various parts of the user’s phrase. In this case, it OntologyAgent: I successfully added 'sensor
sees that ‘observation’ is a word in ‘sensor observation’ and observation' as a direct subclass of 'descriptive
checks whether there is a class ‘observation’ that could be a information content entity'. I successfully added
the existential restriction you requested.
superclass for ‘sensor observation’. When the OwlAgent
tells the OntologyAgent that ‘observation’ is also not in the In summary, Steve submitted a single sentence to convey his
ontology, the OntologyAgent looks for restrictions upon desired ontology change. After that, Steve only needed to
relations (e.g., domain or range restrictions) in this case, select a multiple choice answer in order to clarify where the
whether the ‘is about’ relation already has any existential previously unknown ‘sensor observation’ class should be
restrictions. (Although we have not coded further placed in the hierarchy, and all the necessary changes were
capabilities into the OntologyAgent yet, it would be easy to made without any further effort.
extend the CUBRC software to also look for synonyms of In practice, all ontology changes would be audited before
class names and annotations like developer comments, being integrated into widely shared OWL files.
definitions, and class labels.) When searching the ‘is about’
relation for restrictions, the OwlAgent reports to the VI. FUTURE WORK
OntologyAgent the following clue: there is an axiom about
Our research on this topic was merely exploratory because
the class Information Content Entity to the effect that
of resource limitation and a number of future efforts are
Information Content Entity is equivalent to ‘is about some
warranted. Greater editing functionality should be added
entity’ (a generic class comprising all other classes). Since
first because it provides clear value and is not an ambitious
the user is trying to express that sensor observations are
task. The harder task will be to improve the software's
about detected material entities, and since being an
ability to help the user with the most appropriate questions
Information Content Entity is equivalent to being about
and advice. Still, even these improvements can be
some Entity, the OntologyAgent formulates a guess that (a)
incremental and piece-meal, as the natural language is
we need to insert a new class called Sensor Observation,
meant only to help guide and coach when the user does not
and (b) the class Sensor Observation needs to be a subclass
know the appropriate controlled English to use.
of Information Content Entity, though not necessarily a
direct subclass. So, the OntologyAgent gets the direct Earlier, we showed that although our application extends the
subclasses of Information Content Entity from the capability of the CE Store software, it does so without
OwlAgent and presents to the user the following text. necessarily using ITA-CE to represent ontological claims.
OntologyAgent: I understand you as wanting to add The role played by ITA-CE might easily be played by a
an existential restriction. You are trying to more expressive CNL or a standard data format like XML or
express that every sensor observation is about JSON. There is potential for ITA-CE to serve as a common
some detected material entity. I think you want me format for capturing the semantic intention of natural
to issue the following command:
conceptualise a ~ sensor observation ~ C that is language user inputs. Its value in this regard will largely
about the detected material entity R. depend on how much other infrastructure uses ITA-CE and
However, the ontology is not aware of what 'sensor how well the information being passed back and forth can
observation' means. Because of the 'is about'
relation, a sensor observation should probably be
be leveraged to automate some ontology development.
assigned to be one of the following: At this stage, our application is limited to assistance with
1. A descriptive information content entity
2. A designative information content entity OWL-based ontology editing. It is our hope that this project
3. A directive information content entity will eventually be extended further, so that ITA-CE
STIDS 2015 Proceedings Page 80
sentences can be exploited to mediate user interaction with Acknowledgments
other ontology languages (e.g., OBO) and other formats
(SQL, relational databases). Thus, in dividing labor between Work on ITA-CE as a middleware for ontology editing was
the OntologyAgent and OwlAgent, we have left open the supported by the US Army Research Laboratory. We also
possibility of mapping ITA-CE to these formats and of thank David Mott for his consultations on ITA-CE.
writing additional back-end agents, whose jobs, like the REFERENCES
OwlAgent’s, would be to exchange information with the
[1] Bock, C., Fokoure, A., Haase, P., Hoekstra, R., Horrocks, I.,
OntologyAgent and edit the appropriate document on behalf Ruttenberg, A., Sattler, U., Smith, M. 2012. “OWL 2 Web Ontology
of the user. We also leave open the possibility of alternative Language Structural Specification and Functional-Style Syntax,” 2nd
OntologyAgents, who pass messages in ITA-CE only. [2] Mott, D. Hendler, J. 2009. “Layered Controlled Natural
Languages,” Proceedings of the Third Annual Conference of the
To illustrate this proposal, consider the following two International Technology Alliance.
scenarios. In Scenario #1, a user interacts with a different [3] Braines, D., Mott, D., Laws, S., de Mel, G., Pham, T. 2013.
OntologyAgent that outputs ITA-CE strings only, not the “Controlled English to Facilitate Human/Machine Analytical
Processing,” in Proceedings of SPIE.
command strings described earlier, and sends the messages [4] Braines, D., Preece, A., de Mel, G., Pham T. 2014. “Enabling
to an OwlAgent. The problem is that the OwlAgent does not CoIST Users: D2D at the Network Edge,” in 2014 Proceedings of the
understand ITA-CE strings; it ingests command strings only. 17th International Conference on Information Fusion (FUSION).
In this scenario, it would be desirable to introduce a further [5] Poteet, S., Xue, P., Kao, A., Mott, D., Braines, D., Giammanco,
C. 2013. “Controlled English for Effective Communication during
intermediary agent to translate ITA-CE strings into Coalition Operations,” Proceedings of ICCRTS.
command strings amenable to the OwlAgent’s work. By [6] Preece, A., Pizzocaro, D., Braines, D., Mott, D., de Mel, G.,
introducing this further layer, we allow the OwlAgent to Pham, T. 2012. “Integrating Hard and Soft Information Sources for
continue interpreting messages in terms of command strings. D2D using Controlled Natural Languages,” Proceedings of the 15th
International Conference on Information Fusion.
At the same time, this makes room for ITA-CE to act as [7] Rector, A., Drummond, N., Horridge, M., Rogers, J.,
middleware for various other agents. Knublauch, H., Stevens, R., Wang H., Wroe, C. 2004. “OWL Pizzas:
Practical Experience of Teaching OWL-DL: Common Errors &
In Scenario #2, a user interacts with the OntologyAgent, Common Patterns,” European Conference on Knowledge Acquisition
who passes messages in turn to an OwlAgent that (EKAW-2004), Whittlebury, UK.
understands ITA-CE strings only, not command strings. [8] Bernstein, A., Kaufmann E. 2006. “GINO – A Guided Input
This new OwlAgent would have access to a relational Natural Language Ontology Editor,” in Cruz, I., Decker, S.,
Allemang, D., Preist, C., Schwabe, D., Mika, P., Uschold, M., Aroyo,
database and would query that database to provide the L. (eds.), ISWC 2006. Lecture Notes on Computer Science, Vol 4273.
OntologyAgent with information about the user’s intended [9] Bernstein, A., Kaufmann, E., Kiefer, C. 2009. “Querying the
request. In this scenario, it becomes indispensable that the Semantic Web with Ginseng – A Guided Input Natural Language
OntologyAgent transmit not just command strings, but ITA- Search Engine,” Searching Answers: Festschrift in Honour of
Michael Hess on the Occasion of His 60th Birthday, Clematide, S.,
CE strings as well. Thus, if ITA-CE is to be the common Klenner, M., Volk, M. (eds.). Munster: MV-Wissenschaft.
format for various agents and information sources, we ought [10] Kaufmann, E., Bernstein, A. 2007. “How Useful Are Natural
to ensure that OntologyAgents are conversant in ITA-CE. Language Interfaces to the Semantic Web for Casual End-Users?”
Lecture Notes in Computer Science, Vol. 4825, 281-294.
These scenarios highlight another potential benefit of [11] Funk, A., Tablan, V., Bontcheva, K., Cunningham, H., Davis,
employing ITA-CE as middleware. If ITA-CE is used as a B., Handschuh, S. 2007. “CLOnE: Controlled Language for Ontology
Editing,” Lecture Notes in Computer Science: The Semantic Web
common format, then additional agents could be
[12] Hart, G., Johnson, M., Dolbear, C. 2008. “Rabbit: Developing a
programmed to provide OntologyAgents with answers to his Controlled Natural Language for Authoring Ontologies,” The
queries which are based on the access they enjoy to various Semantic Web: Research and Applications, Lecture Notes in
information sources. In this case, ITA-CE appears to be Computer Science, Vol. 5021, 348-360.
[13] Kaljurand, K. 2008. “ACE View – An Ontology and Rule Editor
promising as a common format for the exchange of Based on Controlled English,” Proceedings of the Poster and
information among agents. Demonstration Session at the 7th International Semantic Web
Conference (ISWC 2008), CUER Workshop Proceedings.
If ITA-CE were to be harnessed in these ways, then in will [14] Wroblewska, A., Kaplanski, P., Zarzycki, P., Lugowska, I.
be necessary to augment the present ITA-CE core lexicon 2013. “Semantic Rules Representation in Controlled Natural
and syntax. At present, ITA-CE does not allow the terms Language in FluentEditor,” The 6th Annual International Conference
‘every’, ‘some’, ‘or’, and ‘not’, which inhibits users from on Human System Interaction (HSI).
[15] Smart, P.R. 2008. “Controlled Natural Languages and the
facile comprehension. Thus, an extension of this project Semantic Web. Technical Report ITA/P12/SemWebCNL, School of
would be to pursue methods laid out by Mott and Hendler Electronics and Computer Science, University of Southampton.
[2], in which new layers of generic syntax are added to the [16] Horridge, M., & Patel-Schneider, P.F. 2012. “OWL 2 Web
core ITA-CE syntax. Mott and Hendler illustrate this with Ontology Language Manchester Syntax,” 2nd edition.
[17] Cregan, A., Schwitter, R., Meyer T. 2007. “Sydney OWL
the adverb ‘only’; to add ‘only’ to the syntax of ITA-CE, Syntax – toward a Controlled Natural Language Syntax for OWL
they invented a language, which defines ‘only’ in terms of 1.1,” presented at OWLED 2007, OWL: Experiences and Directions,
the unaugmented core syntax. Further transformations could Third International Workshop, Innsbruck, Austria, 6-7th, June 2007.
add common quantifiers and connectives such as ‘every’, [18] Sowa, J.F. 2007. “Common Logic Controlled English,”
http://www.jfsowa.com/clce/clce07.htm.
‘some’, ‘or’, and ‘not’, thus enhancing users’ interaction [19] Mott, D. 2010. “Summary of ITA Controlled English,”
with ITA-CE expressions by rendering them more natural. https://www.usukita.org/papers/5658/details.html.
STIDS 2015 Proceedings Page 81