Selecting Ontology Entailments for Presentation to Users

Artemis Parvizi, Chris Mellish, Kees van Deemter, Yuan Ren and Jeff Z. Pan

Department of Computing Science, University of Aberdeen, Aberdeen, U.K.

Keywords:

Semantic Web, Entailment Selection, System Feedback, Human Computer Interaction.

Abstract:

Presenting entailments of axioms in a formal ontology is a non-trivial task. This position paper argues that the

problem of selecting entailments for presentation to users is not adequately acknowledged or addressed in the

literature or in implemented systems. We analyse the problem and consider some different approaches that

can help to address the problem.

1 CONTENT SELECTION WITH

OWL

Many tools aim to assist users in creating or modi-

fying ontologies. Most of these tools will at some

point show users certain entailments of the axioms

they have deﬁned; for instance, the new entailments

that follow given the addition of a new axiom. On the

face of it, this might seem a simple matter: a reasoner

is applied to the deﬁned axioms and generates a set

of entailments (perhaps ﬁltering out those that were

already entailed before the last axiom was added) and

these are then presented to the user. This simple situ-

ation could work if the user was developing an ontol-

ogy using only RDFS. But it is well known that, with

any sufﬁciently expressive logic and when there can

be arbitrary background knowledge, the number of en-

tailments of a set of axioms is inﬁnite. Unfortunately

most families of OWL fall into this class of expressive

logics.

For instance, in a variant of the famous pizza

ontology, an ontology author might be interested

in those concepts that subsume PineappleHamPizza

(here abbreviated to PH) after addition of an axiom

deﬁning this concept. If the user is unlucky, the list of

entailments might be presented to her as a list whose

ﬁrst few items are as follows (mercifully limited to

entailments of the form PH ⊑ X):

PH ⊑(¬VegetarianPizza⊓ NamedPizza) (1)

PH ⊑(∀hasBase.ThinAndCrispyBase) (2)

PH ⊑(∀hasBase.(∃isIngredientO f.Cheesey

Pizza⊓ SpicyPizza)) (3)

PH ⊑(∃hasBase.(∃isIngredientO f.SpicyPizza

Equiv) ⊓ ThinAndCrispyBase) (4)

PH ⊑(∃hasIngredient.(∀isToppingOf.Spicy

Pizza) ⊓CheeseTopping)) (5)

PH ⊑(∃hasIngredient.(∀hasSpiciness.Hot)) (6)

PH ⊑(∃hasTopping.(∃isToppingOf.Named

Pizza⊓ SpicyPizzaEquiv)) (7)

PH ⊑(≤ 4.hasTopping) (8)

PH ⊑(Pizza⊓ NamedPizza) (9)

PH ⊑(Pizza⊓ Pizza⊓ NamedPizza) (10)

PH ⊑(¬CheeseTopping) (11)

PH ⊑(≤ 245.hasTopping) (12)

The set of all the entailments of an ontology can be

inﬁnite for two different reasons, giving rise to two

different selection problems:

1. There are inﬁnitely many logically equivalent ver-

sions of any given axiom, especially if one takes

into account the background knowledge of the

ontology at hand. For example, in the above,

Pizza⊓ NamedPizza (axiom 9) is equivalent with

Pizza⊓Pizza⊓NamedPizza (axiom 10), therefore

any axiom that uses one can use the other instead.

Which one(s) to present?

2. In the worst case, there are inﬁnitely many log-

ically distinct entailments, which state inﬁnitely

many different things about the world. For in-

stance, if PH ⊑ (∃hasBase.X) then also PH ⊑

(∃HasBase.(X ⊔ Y)) for every possible class Y.

Which of these will interest the user?

The ﬁrst of these problems is known in the Computa-

tional Linguistics literature as the “problem of logical

equivalence” (Appelt, 1987).

The fact that the number of entailments is inﬁnite

means that an interface supporting ontology develop-

ment in OWL has to make a selection from the inﬁ-

382

Parvizi A., Mellish C., van Deemter K., Ren Y. and Z. Pan J..

Selecting Ontology Entailments for Presentation to Users.

DOI: 10.5220/0005136203820387

In Proceedings of the International Conference on Knowledge Engineering and Ontology Development (KEOD-2014), pages 382-387

ISBN: 978-989-758-049-9

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

nite set

More precisely, given that no reasoner will

deliver the inﬁnite set, the interface needs to select

by making focussed requests to the reasoner and then

perhaps selecting from the results. The challenge, in

our view, is to select those axioms that the user is

most interested in, and which give the user the best

understanding of the total set of entailments; addition-

ally, the selected set should be presented in a way that

helps the user understand the structure of the set.

How do existing interfaces address this problem?

The most widely used ontology authoring environ-

ment, Prot´eg´e

provides a box where the user can

check “Show Inferences”. The interface then makes

certain entailments visible (or ﬁndable), and others of

course not, without further comment. As another ex-

ample, the ROO environment (Denaux et al., 2012)

provides valuable feedback to a user when they add a

new axiom. One form of its output can be something

like “This input implies 6 new relevant facts ... ”, but

without an explanation of what notion of “relevance”

is at work. With both of these systems, a technically

sophisticated person can ﬁnd out and understand the

criteria for selection, but the criteria are not justiﬁed

and it is unclear whether they match the needs of the

user. Worse still, users may end up misunderstanding

the nature of logic by getting the impression that en-

tailments are necessarily ﬁnite and limited in scope.

2 POSSIBLE APPROACHES TO

ENTAILMENT SELECTION

A number of principles can be used to select entail-

ments, but it is surprisingly difﬁcult to ﬁnd a clear-

cut way of categorising them. In the following, we

divide these principles into four categories, based on

the type of information they use. These categories

can of course be combined. The most frequently used

methods use just syntax.

2.1 Approaches using Syntax

Here axioms are selected based on the explicit struc-

ture used to state them. For instance, a syntactic ap-

proach might select axioms of the form A ⊑ B where

A is a named concept. The example in section 1 is of

this kind, with PH in the position of A.

Traditional ontology authoring environments,

such as Prot´eg´e , ClickOnA (Wolters and Nolle, 2013)

and Swoop (Kalyanpur et al., 2006), allow the user to

A similar problem arises (e.g. with a less expressive

logic) when the number of entailments is large but ﬁnite.

http://protege.stanford.edu/

select different aspects of the ontology to be viewed

by offering different options in the user interface. In

terms of axiom presentation, the user is selecting the

material based on the syntactic form of the relevant

axioms. A common facility is to allow the user to

browse around the axioms of the form X ⊑ Y, with

X and Y being named classes, by displaying a hierar-

chy of classes. These facilities rely on standard rea-

soner functionalities, provided for instance through

the OWL API (Horridge and Bechhofer, 2011), such

as to get named superclasses and subclasses of named

classes, named domains and ranges of properties,

named equivalent classes, the object property values

of a named instance and indeed to test an arbitrary

speciﬁc axiom for being entailed.

Surprisingly Controlled Natural Language-based

(CNL) systems, whose main objective is to simplify

the process of exploration and authoring OWL ontolo-

gies, do not seem to address entailment selection. For

example, ACE View (Fuchs et al., 2008) (a Prot´eg´e

extension) presents an “entailments view” showing

class assertions, property assertions and subclass ax-

ioms where the involved individuals, properties, and

classes are always named. OntoVerbal (Liang et al.,

2011) (a Prot´eg´e plugin) concentrates on axiom re-

alisation, grouping and aggregation, and does not

present any approach to tackle axiom selection. The

RoundTrip Ontology Authoring (ROA) (Davis et al.,

2008) environment only presents axioms that have

been explicitly deﬁned, selecting in turn top level

classes, subclasses, instances, class properties and

their respective domain and ranges, and instance prop-

erties.

In the Rabbit to OWL Ontology authoring (ROO)

editor (Denaux et al., 2012), the authors acknowledge

the issue of inﬁnitely many entailments and decide

to use a subset that contains the ﬁnite set of relevant

implications. Relevant axioms are those of the form

A ⊑ B, ⊤ ⊑ B, A ⊑ ⊥ and A(a) such that A and B

are concept expressions that appear in some axiom in

O∪ {a} and a is a named individual in O∪ {a}. This

is probably the widest set of entailments presented by

an ontology authoring environment.

For reducing the set of generated subsumers, Mel-

lish and Pan (Mellish and Pan, 2008) restrict the en-

tailments to be in the sublanguage ALEN where dis-

junctions are not allowed and negations are restricted

to only atomic concepts. To make this subsumers set

ﬁnite, the authors restrict the generated axioms to be

in a speciﬁc normal form and limit the syntactic com-

plexity of this.

SelectingOntologyEntailmentsforPresentationtoUsers

383

2.2 Approaches using Logic

A logic-based approach takes into account how ax-

ioms behave logically, i.e. makes appeal to notions

like entailment and logical derivation in selecting ax-

ioms. In its purest form, such an approach can only

make choices between axioms which are logically dis-

tinct. For instance, one principle might be not to

select any axiom that is a tautology. As another ex-

ample, a logical approach might select axioms of the

form X ⊑ Y which are most speciﬁc forY in the ontol-

ogy O, in which there is no Z such that O |= X ⊑ Z,

O |= Z ⊑ Y and O 6|= Y ⊑ Z; for instance, if the on-

tology has Boy ⊑ Man ⊑ Person then Boy ⊑ Person

should not be presented but Boy ⊑ Man may be.

Mellish and Pan (Mellish and Pan, 2008) adopt

something like the above principle of speciﬁcity.

They also state various logical principles based on the

Gricean maxims (Grice, 1970) of cooperative conver-

sation for further reducing the set of subsumers being

generated for a class.

A further possibility is to use logic in a different

way, by letting the derivation of entailments play a

role. For example, the user may be permitted to ask

to only see entailments equivalent to axioms whose

derivation hinges on the transitivity of a particular

role; entailments whose derivation does not use ax-

ioms containing the universal quantiﬁer; or entail-

ments whose derivation requires more than n reason-

ing steps. Further, it might make sense to prefer en-

tailments that have longer derivations (because they

are less trivial) or entailments that have shorter deriva-

tions (because they are more understandable)

2.3 Approaches using Discourse

Structure

When entailments are being selected for presentation

as part of a conversation, e.g. in the context of an

ontology authoring environment, there may be struc-

ture in the discourse that can be exploited to help de-

termine what entailments might be most relevant. A

discourse has a minimal structure coming from the se-

quence of utterances from the participants, but there

are more sophisticated notions of discourse structure

that might be used (Lascarides and Asher, 2007).

When a user presents a set of axioms in ontology au-

thoring, these axioms should not be considered as in-

dividually independent sentences; there will be a co-

herence and ﬂuency that is helpful for the ontology

authors. For instance, when the user is in the middle

of describing subclasses of a given class, entailments

Richard Power, personal communication

that mention those subclasses may be especially rele-

vant and interesting.

Although we are not aware of any existing ap-

proaches using discourse structure for entailment se-

lection, there are hints of useful approaches in other

work. In particular, work on ordering and grouping of

textual material generated from ontologies (see Sec-

tion 3) is relevant. This is because any principle that

indicates that axioms x and y should be grouped to-

gether (e.g. axioms “about” the same concept should

be grouped together) can be converted into a princi-

ple for axiom selection - given that x is in the recent

discourse context, axiom y should be selected (e.g. se-

lect axioms “about” the concepts currently being dis-

cussed).

2.4 Approaches using Pragmatics

A pragmatics-based approach uses explicit principles

about what will be communicatively successful. In

other words, it takes serious account of the fact that

users will be human beings and will have goals and

intentions.

One approach to modelling user’s goals in ontol-

ogy authoring is the use of Competency Questions

(QCs). A CQ is a natural language question that the

ontology author expects the ontology to be able to an-

swer (Uschold et al., 1996). In (Ren et al., 2014),

the authors have introduced the notion of Competency

Question-driven Ontology Authoring. By collecting

and analysing various CQs from two different do-

mains, the authors proposed a set of patterns for mod-

elling natural language CQs as OWL patterns. The

authors categorised these patterns and extracted a set

of presuppositions for each pattern. For example, a

CQ “Which animals eat grass?” displays the pattern

[WHICH [CE1] [OPE] [CE2]?], where CE1 and CE2

are concepts and OPE is a 2-place relation. Questions

of this pattern suggest that the user has certain expec-

tations, (known as presuppositions in the linguistics

literature (Levinson, 1983)), which may or may not

be entailed by the ontology. For example, the ques-

tion suggests that animals should be able to eat grass.

If an the ontology entails that animal ⊑ ∀eat.¬Grass

then one of the presuppositions of the CQ is violated.

We believe a user is likely to be especially inter-

ested in the presuppositions of his or her CQs, be-

cause if one or more of these are violated then the

CQ is unanswerable. This insight can be exploited in

the selection of entailments.

Some of the suggestions that we have made in the

above for guiding the selection of entailments have

their roots – or at least bear strong similarities – to ex-

KEOD2014-InternationalConferenceonKnowledgeEngineeringandOntologyDevelopment

384

isting ideas and methods in theoretical and computa-

tional linguistics. The same will be true for our (more

tentative) ideas about ordering and grouping entail-

ments.

3 ORDERING AND GROUPING

If we take the shape of the axioms as given (e.g., we

do not translate logic axioms into a more readable for-

mat, as in e.g. (Power, 2014)) then, apart from selec-

tion, the other issues to be addressed are ordering and

groupingof entailments. These affectthe presentation

of ﬁnite as well as inﬁnite sets of axioms

and are

relevant regardless of whether axioms are presented

graphically or in a list.

Interestingly, the task of entailment selection in-

teracts with the other tasks of ordering and grouping.

One of the reasons why the list of axioms presented in

Section 1 is difﬁcult to digest is that the axioms were

listed in a somewhat arbitrary order and without any

further structuring, that is, as a simple enumeration.

There is no point in selecting a set of axioms that can-

not be coherently grouped and ordered. Similarly, if

we are expecting to order or group axioms in a partic-

ular way then this may help us to select axioms in the

ﬁrst place. The following work addresses the prob-

lems of ordering and grouping and presents solutions

that may have relevance for selection.

In OntoVerbal, a distinction between direct and in-

direct axioms is made. This basically means produc-

ing a syntactic characterisation of what concept an ax-

iom is mainly “about” (the “topic” of the axiom). So

A ⊑ B is directly about A but only indirectly about B.

Directness and also syntactic complexity are used as

the basis for grouping together related axioms. The

hypothesis is that by grouping and presenting axioms

in a rhetorically coherent manner (following Rhetori-

cal Structure Theory) with a single topic, not only is

the coherence of the text maintained, but also the user

is aided in recognising topic change.

NaturalOWL (Androutsopoulos et al., 2013) is an-

other OWL verbaliser. When describing an OWL

class or individual(target), it ﬁrst presents all the state-

ments that directly describe the target (primary tar-

gets); second, the system describes the desirable state-

ments that are indirectly related to the target (second-

level targets). Second-level targets are only gener-

ated when the target is an individual. Consistent with

the linguistic literature (Halliday and Hasan, 1976),

See e.g. http://www.w3.org/TR/2002/WD-rdfsyntax-

grammar-20020325/, where stripping and abbreviation are

employed to avoid repeated subjects or objects in adjacent

RDF triples.

an attempt was made to do this in a “coherent” fash-

ion, using text planning; a local coherence measure,

based on Centering Theory (CT) (Grosz et al., 1995))

is applied; an additional mechanism prevents infor-

mation repetition in the generated text. NaturalOWL

also uses the complexity of the generated sentences to

decide how triples are mapped to English sentences.

A text might be overly complicated, which requires

breaking down, or the sentences might be too simple

and in need of aggregating. Aggregation relies on a

set of thresholds for the number of sentences that can

be aggregated.

4 OUR EXAMPLE REVISITED

The example of 1 might serve to show some of the

consequences that our ideas about selection, order-

ing and grouping might have. These remarks will be

highly tentative and incomplete.

Logical Approaches. The rule suggested in Section

2.2 for choosing the most speciﬁc subsumer would re-

move an axiom such as PH ⊑ ¬CheeseTopping (ax-

iom 11) from the entailments listed in Section 1, be-

cause the more general statement PH ⊑ ¬Topping

holds. This principle would also cause replacing the

axiom PH ⊑ (≤ 245.hasTopping) (axiom 12) by a

stronger axiom such as PH ⊑ (≤ 4.hasTopping) (ax-

iom 8). Many other entailments can be removedusing

similar principles.

Furthermore, consider the axiom PH ⊑

(∃hasBase.(∃isIngredientOf.SpicyPizzaEquiv)

⊓ ThinAndCrispyBase) (axiom 4), which implies

PH ⊑ (∃hasBase.ThinAndCrispyBase). Given

that hasBase is functional, this makes the axiom

PH ⊑ (∀hasBase.ThinAndCrispyBase) (axiom 2)

logically redundant, so if the functionality of roles is

presented elsewhere then the latter can be removed.

Pragmatic Approaches. Suppose, for simplicity,

the user has expressed only one Competency Ques-

tion, namely “What base does a pizza have?”. Then

it is presupposed that some pizzas are allowed to

have bases. If the ontology contains an axiom

∃hasBase.⊤ ⊑ Pizza⊓Topping, then this implies that

nothing can have a base (because Pizza and Topping

are disjoint in the ontology). This makes it impossi-

ble for the ontology to meaningfully answer the com-

petency question. If the user deletes the above faulty

axiom, the user is likely to be interested in the conse-

quence that now the ontology supports the idea that

pizzas can have bases. Since this CQ presupposes the

satisﬁability of Pizza ⊓ ∃hasBase.Base, the fact that

this concept is now satisﬁable is also worth pointing

out.

SelectingOntologyEntailmentsforPresentationtoUsers

385

Discourse based Approaches. Suppose the sys-

tem’s analysis of Competency Questions has sug-

gested focussing on entailments that make use of the

hasBase role, but the user has recently performed

a long series of authoring actions centring on the

hasIngredient role. This suggests broadening the se-

lection of axioms, focussing on entailments that make

use of either the hasBase role or the hasIngredient

role; For example, axioms 3, 5, and 6 may be of in-

terest; although we can argue against the inclusion

of axiom 3, if we have already mentioned PH ⊑

(∃hasBase.ThinAndCrispyBase). In the (unlikely)

absence of other information this might mean omit-

ting all other entailments.

If and when genuine (i.e., non-redundant) infor-

mation is omitted, the system should indicate what

principles were used to guide the selection process,

so as to avoid misunderstandings. For example, the

system could say “this presentation focusses on the

hasBase and hasIngredient roles. To view other en-

tailed axioms, do ...”. Explanations of this kind would

ﬁt neatly within the ideas suggested in the following

section.

Syntactic Approaches. Ultimately, syntactic ap-

proaches might be more suitable for ordering and

grouping than for selection, which is better han-

dled on more principled grounds. Axioms could

be grouped by dividing them into broad syntactic

classes, for example, (1) any and all axioms of the

form PH ⊑ (∃hasIngredient.ϕ), (2) axioms of the

form PH ⊑ (∀hasIngredient.ϕ), (3) axioms of the

form PH ⊑ (∃hasBase.ϕ), (4) axioms of the form

PH ⊑ (∀hasBase.ϕ), (5) any and all other axioms.

This is just one example of a possible grouping; users

might be offered a choice between different types of

grouping. The original example might result after

the user has asked to see subsumers of the concept

PineappleHamPizza.

5 INTEGRATION INTO AN

AUTHORING ENVIRONMENT

(Power, 2014) argued that an ontology authoring sys-

tem should 1. only generate informative statements,

2. detect redundancies (deciding whether to ﬁx or re-

port them), 3. identify and report contradictions, and

ﬁnally 4. report implications.

Perhaps the most important aspect of this policy,

in our view, is informativeness. Reducing the set of

entailments and increasing the informativenessof this

set can be achieved by a combination of syntactic, log-

ical, and discourse-oriented approaches. While syn-

tactic structures provide a starting point for limiting

the set of entailments, only a logical framework can

detect and address redundancy. Finally, the user’s fo-

cus of attention may be tracked through a sequence

of the user’s authoring operations. If the system is

able to grasp what the user’s focus of attention is, this

information can inform the selection task as well.

We have suggested extending the content selection

policy by introducing and modelling users’ require-

ments through competency questions. In an ontol-

ogy authoring interface that allows the users to submit

and monitor competency questions and their associ-

ated authoring tests (Ren et al., 2014), the system will

havea better understanding of the users’ requirements.

Thus, in such an interface, when the system is faced

with a large set of entailments, a natural way forward

would be to select entailments and non-entailments

based on such requirements. These requirements can

also affect the grouping and ordering process.

Figure 1 presents a way of displaying CQs and

their associated authoring tests in an ontology author-

ing interface. In this ﬁgure, the CQ “What cake has

which cake ﬁlling?” is shown on top and is expanded

with 5 different authoring tests. The CQ can be an-

swered meaningfully only when all the authoring tests

are passed. As shown in the ﬁgure, a green/red dot is

used to indicate the pass/fail status of the authoring

tests so that the users know what is preventing the CQ

from being answered. In the example shown in Fig-

ure 1, users can ﬁnd out that the hasFilling property

has not been introduced into the ontology so thatCake

and CakeFilling cannot be related.

Figure 1: An example of the Competency Question tree

with its associated authoring tests.

Suppose someone is using the interface to author

an ontology and, at some stage, she is faced with two

CQs that are as yet unfulﬁlled (i.e., marked red in the

interface), and she is currently working on the ﬁrst of

these two. Suppose this is the CQ “What base does

a pizza have?” that was discussed in Section 4, un-

der Pragmatic approaches. Suppose, at some point,

she manages to fulﬁl this CQ. At this point, partic-

ular axioms become worth presenting (as explained

in Section 4). If this set of axioms is of a manage-

able size then axioms that are relevant in connection

with the the last remaining axiom might also be pre-

sented (because they are relevant to the next authoring

task). Presentation of axioms should also be subject

to the logical, syntactical, and discourse-based consid-

KEOD2014-InternationalConferenceonKnowledgeEngineeringandOntologyDevelopment

386

erations that we discussed elsewhere.

It seems to us that content selection, grouping

and ordering should be developed in a way that more

decisions are made by the adaptable authoring envi-

ronment rather than reasoner side. Discourse-based

approaches would have the best chance of coming

to fruition if ontology editing used an intelligent

dialogue-oriented authoring environment. Unlike pre-

vious dialogue-oriented interfaces, such as ROO, we

envisage an ongoing dialogue with the user that can

provide valuable insight into user’s objectives, and in

which occasional disambiguation dialogues could be

initiated by the system.

Predicting all the logical consequences of an au-

thoring action is a difﬁcult task, even for the most ex-

perienced user. It might therefore be useful to offer

users the option of asking hypothetical questions. We

envisage that this could work as follows: If in doubt,

the user would be able to ask what the consequences

of a given authoring action would be. Upon this ques-

tion, the system should present the entailments of this

action in a transparent way (as discussed in this po-

sition paper), to allow him or her to understand the

consequences of the action as clearly as possible. Fi-

nally, the user should be given the choice between

committing or reverting the changes. We hypothesise

that these new features would help to prevent errors

in ontology editing and increase users’ conﬁdence in

the quality of the ontology that they are creating.

6 SUMMARY

This paper has argued that the problem of entailment

selection is largely neglected in software tools that

(perhaps implicitly) claim to present entailments from

OWL ontologies. On the other hand, there are a num-

ber of good ideas in current work for how the problem

might be addressed in a more substantial way. It is in-

teresting that many of the ideas about how to select

entailments come from Linguistics or the Philosophy

of Language, rather than from formal logic (Gricean

maxims, presuppositions, discourse structure). That

is, a solution to the problem needs to look at the na-

ture of the communication within which the selection

takes place. Finally, we discussed the added beneﬁts

of other restructuring mechanism, such as grouping

and ordering.

ACKNOWLEDGEMENTS

This research has been funded by the EPSRC project:

WhatIf: Answering “What if...” questions for Ontol-

ogy Authoring. EPSRC reference EP/J014176/1.

REFERENCES

Androutsopoulos, I., Lampouras, G., and Galanis, D.

(2013). Generating natural language descriptions from

owl ontologies: the naturalowl system. Journal of AI

Research, 48:671–715.

Appelt, D. E. (1987). Bidirectional grammars and the de-

sign of natural language generation systems. In work-

shop on Theoretical issues in natural language pro-

cessing, pages 206–212.

Davis, B., Iqbal, A. A., Funk, A., Tablan, V., Bontcheva,

K., Cunningham, H., and Handschuh, S. (2008).

Roundtrip ontology authoring. Springer.

Denaux, R., Thakker, D., Dimitrova, V., and Cohn, A. G.

(2012). Interactive semantic feedback for intuitive on-

tology authoring. In FOIS, pages 160–173.

Fuchs, N. E., Kaljurand, K., and Kuhn, T. (2008). Attempto

controlled english for knowledge representation. In

Reasoning Web, pages 104–124. Springer.

Grice, H. P. (1970). Logic and conversation. Harvard Univ.

Grosz, B. J., Weinstein, S., and Joshi, A. K. (1995). Center-

ing: A framework for modeling the local coherence of

discourse. Computational linguistics, 21(2):203–225.

Halliday, M. A. and Hasan, R. (1976). Cohesion in English.

London: Longman.

Horridge, M. and Bechhofer, S. (2011). The owl api: A java

api for owl ontologies. Semantic Web, 2(1):11–21.

Kalyanpur, A., Parsia, B., Sirin, E., Grau, B. C., and

Hendler, J. (2006). Swoop: A web ontology editing

browser. Journal of Web Semantics, 4(2):144–153.

Lascarides, A. and Asher, N. (2007). Segmented discourse

representation theory: Dynamic semantics with dis-

course structure. In Computing meaning, pages 87–

124. Springer.

Levinson, S. C. (1983). Pragmatics. Cambridge University

Press.

Liang, S. F., Stevens, R., Scott, D., and Rector, A. (2011).

Automatic verbalisation of snomed classes using on-

toverbal. In Artiﬁcial Intelligence in Medicine, pages

338–342. Springer.

Mellish, C. and Pan, J. Z. (2008). Natural language di-

rected inference from ontologies. Artiﬁcial Intelli-

gence, 172(10):1285–1315.

Power, R. (2014). A formal dialogue model for ontology

authoring. In the 50th Anniversary Convention of the

AISB.

Ren, Y., Parvizi, A., Mellish, C., Pan, J. Z., van Deemter,

K., and Stevens, R. (2014). Towards competency

question-driven ontology authoring. In The Seman-

tic Web: Trends and Challenges, pages 752–767.

Springer.

Uschold, M., Gruninger, M., et al. (1996). Ontologies: Prin-

ciples, methods and applications. Knowledge engi-

neering review, 11(2):93–136.

Wolters, M. and Nolle, A. (2013). Clickona: An editor for

dl-litea based ontology design. In 26th International

Workshop on Description Logics. Ulm, Germany.

SelectingOntologyEntailmentsforPresentationtoUsers

387