An Approach to Circumstantial Knowledge Management for Human-like

Interaction

Alejandro Baldominos, Javier Calle and Dolores Cuadra

Computer Science Dept., Universidad Carlos III de Madrid, Madrid, Spain

Keywords:

Circumstantial Knowledge, Situation Model, Context, Context-awareness, Ubiquitous Computing, Human-

like Interaction, Natural Interaction.

Abstract:

This paper proposes the design of a general-purpose domain-independent knowledge model formalizing and

managing the circumstantial knowledge involved in the human interaction process, i.e., a Situation Model.

Its design is aimed to be embodied into a human-like interaction system, thus enriching the quality of the

interaction by providing context-aware features to the interaction system. The proposal differs from similar

work in that it is supported by the spatio-temporal databases technology. Additionally, since the proposed

model requires to be fed with real knowledge obtained from each speciﬁc interaction domain, this paper also

proposes an edition tool for acquiring and managing that circumstantial knowledge. The tool also supports the

simulation over the model to check the correctness and completeness of the acquired knowledge. Finally, some

scenario examples are provided in order to illustrate how the Situation Model works, and to gain perspective on

its future possibilities of application in different systems where context-aware services can make a difference.

1 INTRODUCTION

The appearance of many handheld computer devices

in the last years, such as smartphones and tables,

is leading to a progressive achievement of ubiqui-

tous computing. However, the current situation still

falls short of the original vision introduced by Weiser

(Weiser, 1991), where computers are invisible for hu-

mans. As increasingly people are gaining access to

these computer devices, the development of new in-

terfaces is required in order to enable a simple human-

computer interaction not requiring from the user spe-

ciﬁc skills or technical capabilities; and for this rea-

son ubiquitous computing introduces new challenges

regarding the interaction with devices, which must be

as simple as possible, so that the user can focus on his

goals. Moreover, this interaction must attend the issue

of location, meaning that it must adapt its behavior to

circumstances in signiﬁcant ways.

Human-like interaction (also referred as natu-

ral interaction) is a way of interaction that aims to

achieve a human-computer interaction that process

what comes naturally (Oviatt and Cohen, 2000) and

that imitates human interactive behavior (Bernsen,

2000). This interaction style seeks to provide inter-

faces which attend and express human messages (ut-

tered as if aimed at other humans) and which imitate

the human interactive reasoning in order to provide

access to the system services. Therefore, it turns the

human-computer interaction process into a conversa-

tion between the user and the system, which should be

as natural as possible (i.e. it should resemble a con-

versation between two humans as much as possible).

Because it resembles the interaction between hu-

mans, human-like interaction presents some attractive

features that turns it into an adequate way of interac-

tion for achieving ubiquitous computing. Particularly,

it provides access to the technology to certain dis-

abled users and to people without speciﬁc technical

skills. Moreover, human-like interaction may result

in an easy and comfortable interaction with the sys-

tem in some scenarios, even when the user is already

familiarized with the technology.

In order to achieve human-like interaction the

knowledge regarding the interaction between humans

must be studied and formalized so that a computer

system can process it. It is generally expected that

the better the knowledge is formalized, more natu-

ral the interaction would be. Due to this, research

in human-like interaction involves some challenges,

as it requires the study and the formalization of all

the knowledge regarding the communication between

humans. This includes the knowledge required for the

concepts and terms regarding the interaction and the

Baldominos A., Calle J. and Cuadra D..

An Approach to Circumstantial Knowledge Management for Human-Like Interaction.

DOI: 10.5220/0004861600710078

In Proceedings of the 16th International Conference on Enterprise Information Systems (ICEIS-2014), pages 71-78

ISBN: 978-989-758-029-1

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

way they are expressed, the dialogue strategies, what

is known about the interlocutors, their own beliefs,

goals and emotions, the context of the interaction, etc.

This paper focuses in the circumstantial knowl-

edge, a subset of the knowledge regarding the in-

teraction between humans that deals with the non-

linguistical context of the interaction. In particular,

the purpose of this paper is to describe the design

of a general-purpose domain-independent Situation

Model, a computer model storing a formalization of

the circumstantial knowledge. Moreover, an intuitive

edition tool is presented, which aims to handle the

knowledge base of the Situation Model.

2 THEORETICAL BACKGROUND

Regarding an interaction process, the circumstantial

knowledge deals with the context of the interaction,

i.e., the set of circumstances applying in an environ-

ment. This paper focuses in this kind of knowledge,

and speciﬁcally will attend the taxonomy of the con-

text provided by Gee (Gee, 1999), which observes

ﬁve different aspects of the context: (i) a semiotic as-

pect regarding the language used during the commu-

nication, (ii) an activity aspect concerning the trans-

actions (or major tasks) underlying the interaction,

(iii) a material aspect dealing with the physical con-

text, such as the spatio-temporal circumstances, in-

cluding people and objects taking part in the interac-

tion, and also other environmental issues, (iv) a politi-

cal aspect regarding the roles of each interlocutor and

(v) a sociocultural aspect observing the inﬂuence of

social and cultural conditions on the interaction.

These ﬁve aspects are not independent, and each

of them inﬂuences the other. For instance, the place

where the interaction is taking place may restrict the

roles that each of the interlocutors can play.

Human-like interaction often represents knowl-

edge by means of two different approaches: con-

versational analysis and discourse analysis (Levin-

son, 1983). While the ﬁrst one aims at generating

the interaction based on the analogy of the context

with its knowledge base (e.g., with case-based rea-

soning), the second one focuses on the design of for-

mal methods to produce coherent and ﬂexible interac-

tions. The human-like interaction system in which the

proposed Situation Model will be integrated follows

the discoursive approach, as it would require much

less corpora than the conversational one and aims

at achieving a knowledge-based (rather than sample-

based) reasoning for generating the interaction.

The human-like interaction system formalizes the

interaction knowledge by means of a cognitive archi-

Presentation Model

Interface

Voice R+S

N.L.P.

Prosodic

G.U.I.

Gesture Acq.

3D Character

component

. . .

Dialogue Model

Discourse

Interpretation

Linguistic

Structures

Intentional

Mgmt.

Generator

Task Model

Ontology

Emotional

Model

Situation

Model

Self-Model

Session

Model

User Model

External

Agents

Interface Agents

Interaction Agent

Intelligent / Ext.

Agents

Figure 1: Cognitive architecture for human-like interaction.

tecture, i.e., a set of independent models each of them

handling a certain subset of the interaction knowledge

and which collaborate to produce the interaction. The

next section describes the framework where the pro-

posed Situation Model will be integrated.

2.1 Framework

The model-based cognitive architecture for a human-

like interaction system which will integrate the pro-

posed Situation Model (Calle, 2004) is shown in ﬁg-

ure 1, where each of the models deals with a subset of

the knowledge regarding the interaction.

The Situation Model, dealing with the circumstan-

tial knowledge, is highlighted in the ﬁgure. The inclu-

sion of the Situation Model in a human-like interac-

tion system would provide some advantages (Rivero

et al., 2007), such as: (i) allowing the system to

adapt itself to the current context and consequently

enhancing the naturality of the produced interaction,

(ii) enabling the use of circumstantial information

for ﬁltering the knowledge of other models, thus re-

ducing the ambiguity and increasing the efﬁciency

and efﬁcacy of the system, (iii) providing mecha-

nisms for situation triggering so that an action is per-

formed when some circumstantial conditions are met,

and (iv) providing new context-related functionalities

such as door-to-door navigation or predictions on an

object situation in the future.

Besides the inclusion of the Situation Model in

the human-like interaction architecture, the proposed

management tool will be incorporated to Cognos

Toolkit (Calle et al., 2009; Calle et al., 2011), aimed at

easing the acquisition and management of the knowl-

edge underlying a human-like interaction system.

ICEIS2014-16thInternationalConferenceonEnterpriseInformationSystems

2.2 Related Systems

This section brieﬂy describes some of the most rel-

evant research and commercial applications where a

context model was successfully integrated within a

human-like interaction system.

One of the earlier approaches is TRIPS (Ferguson

and Allen, 1998) (1998), a system integrating sev-

eral AI techniques in order to provide an interactive

problem-solving assistant for a logistic domain. Par-

ticularly, it aims to assist the user to build plans in cri-

sis situations which may have to meet some speciﬁc

circumstantial constraints. In order to manage the cir-

cumstantial knowledge, TRIPS architecture includes

a Problem-Solving Manager dealing with a database

storing the problem-solving context and with special-

ized reasoners, whose purpose is to keep a represen-

tation of the task to be performed, the generation of a

feasible plan coherent with the current circumstances

and maintain the state of the proposed solution.

Secondly, introduced in 2000, Deep Map (Malaka

and Zipf, 2000) is a research framework of a tourist

information system which aimed to build a mobile

solution offering a trip planner and a city naviga-

tor, and which would consider the user preferences

and needs, and some contextual information such as

weather conditions or trafﬁc, providing an accessible

interface for untrained users. Deep Map core compo-

nent is the Geographical Information System (GIS),

which is in charge of managing the location infor-

mation, relating it to the user needs. As the user

may require some historical information about a given

location, GIS deals with spatial information, and a

database agent extends information on speciﬁc loca-

tions by retrieving related multi-media (subject to a

given moment in time, it is also able of providing his-

toric information of past buildings or monuments in

the speciﬁc location). While a third dimension stores

knowledge about the visibility of objects from a given

location, the fourth dimension is required to store in-

formation about the historical evolution of such ob-

jects. As the system is intended to be used in scenar-

ios where the user may be walking or driving, Deep

Map provides several interfaces, including some re-

garding the interaction by means of natural language

for both the input and output, which turn to be unin-

trusive and therefore adequate in such scenarios.

In the third place, another system closed in the

early 2000s is SmartKom (Wahlster, 2006), which

provides symmetric multimodality to enable users ac-

cessing its functionalities. SmartKom architecture is

based on models, each of them representing and rea-

soning about a certain type of knowledge. SmartKom

is motivated for non-desktop scenarios, and for that

reason the circumstantial knowledge plays a key role

of the interaction process. To support the reasoning

over this knowledge, SmartKom includes a context

model (Porzel et al., 2006) supporting four aspects

of context: dialogical (what have been said), onto-

logical (regarding the concepts on the domain), user

(knowledge about the interlocutors) and situational

(the spatio-temporal context).

Finally, examples of interaction systems incorpo-

rating a context model can also be found in recent

commercial applications, such as Siri, a personal as-

sistant introduced by Apple in the iPhone 4S in 2011.

This assistant enables the user asking for some infor-

mation or for an action to be performed naturally. Siri

takes advantage of the smartphone localization mech-

anisms to retrieve the user location and the Internet

connection to obtain more information about the con-

text. This enables Siri to provide context-aware ser-

vices such as looking for places of interest close to

the user, creating new reminders at a speciﬁed time or

asking for the weather at a certain place.

While all of these models observe the material as-

pect of the context, they lack from a complete support

for the remaining aspects of the context. Moreover,

they are all designed ad-hoc for speciﬁc domains or

systems, and their knowledge is introduced manually

or through speciﬁc tools. While these features are not

meant to be actual handicaps, they prevent the model

from being reusable in other systems.

3 PROPOSAL

This paper proposes the design of a general-purpose

domain-independent Situation Model to be embodied

within the cognitive architecture for a human-like in-

teraction system. This model will support all aspects

of the context, according to the taxonomy described

in section 2. Moreover, this paper describes a sys-

tem for editing the knowledge stored in the Situa-

tion Model through an easy-to-use desktop applica-

tion, which also serves for performing a simulation

over the Situation Model in order to check the cor-

rectness of the knowledge and the model itself.

3.1 Situation Model

The proposed Situation Model architecture is shown

in ﬁgure 2 (top). The Situation Model will be im-

plemented over a spatio-temporal relational database

(Cuadra et al., 2009), as it enables efﬁcient storage

for the location and form description of objects as

well as their evolution through time. Moreover, it

formalizes the circumstantial knowledge by means of

AnApproachtoCircumstantialKnowledgeManagementforHuman-LikeInteraction

Model Knowledge

Edition Knowledge

-id : int

-description : string

Network

-id : int

-description : string

-geometry

-basic_cost : double

NetworkItem

Node

Link

-name : string

-description : string

Situation

-name : string

Feature

-multiplier : double

CostFactor

VisualNetwork

-name : string

-bgImage

Plane

-x : double

-y : double

VisualNode

VisualLink

Figure 2: Architecture for the Situation Model supporting

the edition.

a N-dimensional graph, a comprehensive yet power-

ful representation which can be easily processed by

a computer system. In particular, each aspect of the

context is represented by means of one or more di-

mensions. For instance, the spatial aspect will be

represented by three dimensions, the time will add

a fourth dimension and the remaining aspects will

include additional dimensions. For this reason, the

Situation Model is composed of networks (graphs),

containing both nodes and links. The spatio-temporal

database stores the geometry of each of the network

items, i.e., their N-dimensional location and shape.

Furthermore, network items may have a cost, i.e.,

how expensive it is to traverse through them. Besides

the basic cost of a network item (the one that always

apply), it may also have some features modifying its

cost depending on the current situations. For instance,

a node placed in the stairs of a building may have a

cost; however, it may implement a feature that multi-

plies the cost when the situation indicates that the user

uses a wheelchair. Thus, the use of user-deﬁned situa-

tions may observe any factor conditioning the context

and alter the cost of going through a path depending

on the existence of that factor.

Finally, ﬁgure 3 describes the way all the elements

are physically interconnected by showing the physical

architecture of the human-like interaction system, in-

Human-Like Interaction SystemPhysical Layer

End User

LAN

Wi-Fi

Multi-Agent

Blackboard

NLP Server

Dialog Model

Server

Situation Model

Server

GPS Wireless RFID

...

Figure 3: Physical architecture for the system.

cluding the server handling the Situation Model. In

this architecture, the user carries a mobile device run-

ning the client application for the interaction system.

This device is assisted by several physical layer tech-

nologies, such as GPS, RFID or Wireless; enabling

the client to obtain information about the context.

The client remotely connects to the human-like in-

teraction system, supported by a multi-agent architec-

ture, which processes the interaction and provides the

required services, including context-aware ones.

3.2 Edition Tool

Figure 2 shows the complete architecture of the Sit-

uation Model, including the knowledge required to

edit the circumstantial knowledge. It can be seen that

the Situation Model knowledge base is replicated and

some additional entities supporting the edition func-

tionalities are added. Replication provides some ad-

vantages, as the translation from the edition database

to the model knowledge base is trivial. Moreover, it

also enables the simulation directly over the edition

model, and the proposed management tool will take

advantage of this feature.

Regarding the edition model, the main entity is

the visual network, serving as a wrapper for the net-

work class of the knowledge model. A visual network

is composed of planes, an entity aimed at provid-

ing a bidimensional representation from the original

N-dimensional graph, in order to show the network

graphically in the edition tool. The plane may contain

a background image, which would ease the placement

of nodes. Finally, planes contain visual nodes, an en-

tity serving as a wrapper for nodes and providing a

virtual location for the node over the plane (this vir-

tual location is expressed in pixels, whereas the node

location may refer to actual coordinates). The visual

link also serves as a wrapper for the link class, though

visual links do not belong to a plane, given that a link

may connect nodes in different planes.

Based on the described architecture, a manage-

ment tool is proposed in order to enable the user to

ICEIS2014-16thInternationalConferenceonEnterpriseInformationSystems

Figure 4: Screen for the network management.

manage the circumstantial knowledge. Particularly,

the tool provides functionality for (i) managing situ-

ations, (ii) managing features setting cost factors to

network items based on current situations, (iii) man-

aging networks, (iv) designing of a network by plac-

ing nodes and links over a plane with a background

image and (v) simulating over an existing network.

Figure 4 shows the screen which allows the man-

agement of a network, with a plane already loaded.

The application allows to create, delete and move

nodes and links across the plane. Moreover, a node

or link can be selected in order to edit its description,

location (only for nodes), basic cost or features. To

ease the assignment of coordinates to the nodes, the

application includes a reference feature enabling the

deﬁnition of a spatial reference system by means of

the deﬁnition of the coordinates of two nodes, thus

inferring the location of all the other nodes once the

reference system is deﬁned. Additionally, the possi-

bility of changing the size of the plane and of the net-

work items independently serves for increasing the

tool’s accessibility. Finally, as the edition model is

built over a transactional relational database, the ap-

plication allows to commit or rollback the changes.

Besides the functionality for the edition of the cir-

cumstantial knowledge, the tool includes a feature

to allow the execution of a simulation over the Sit-

uation Model in order to check the correctness of

the acquired and formalized knowledge. The plane,

which is loaded in the application, shows the network

graph already edited. The user can place an agent

(shown as a Pac-Man) over the plane, which can be

moved through it, simulating the actual motion of a

real agent. In order to perform this motion, the system

emulates different technologies of the physical layer,

such as RFID or GPS. The simulation over the model

allows so far the execution of two different services.

The description service describes the agent’s current

location, while the navigation service provides the

user with a route between his current location and a

desired target. The left panel allows the user to choose

situations applying during the simulation. When situ-

ations are enabled or disabled the network items’ cost

change dynamically, altering the route proposedto the

user.

3.3 Preliminary Experimentation

After the development of the edition tool, a prelimi-

nary evaluation was carried out over six subjects with

different academic proﬁles. It is worth noting that the

aim of this evaluation is to check whether this tool

provides actual advantages over manual edition of the

circumstantial knowledge. A complete validation of

the interaction system as a whole, including the cir-

cumstantial knowledge introduced with the proposed

tool, has been left for future work.

The experiment asked the subjects to feed the

Situation Model with an actual network, which was

printed and distributed among them. The real coor-

dinates for each of the nodes from the network was

hidden to the users, yet they were given two refer-

ence points over the network with their actual loca-

tion. The experiment subjects were assigned the tasks

of (i) modeling the network manually by estimating

the position of each node and introducing this infor-

mation into the Situation Model by means of some

provided predeﬁned SQL sentences and (ii) modeling

the network through the edition tool.

For each of these tasks, the execution time was

measured for each subject. Additionally, the quality

of the resulting network was also measured by check-

ing the accuracy of the modeled networks in terms of

the deviation of the nodes placed by the users with

respect to their original locations. The results for the

objective evaluation are shown in table 1 and in ﬁg-

ures 5(a) and 5(b). Regarding the execution time (in

seconds) for both the manual edition and the edition

tool, and it can be concluded that the time required

by the users for modeling the network is signiﬁcantly

less when the edition tool is used, as the manual edi-

tion of the network is between three and four times

slower. Concerning the quality of the modeled knowl-

edge, the results show that the absolute error (the sum

of the misplacements of the nodes in meters) is much

lower when the network is modeled by means of the

AnApproachtoCircumstantialKnowledgeManagementforHuman-LikeInteraction

Table 1: Results for the evaluation.

Time Error Comfortable Intuitive Reliable Agile

Manual

Avg. 1639.5 10.9 2.33 2.67 2.83 2.17

St. Dev. 283.19 4.62 1.03 1.63 1.33 1.6

Tool

Avg. 441.17 3.39 4.83 4.33 4 4.83

St. Dev. 93.36 1.29 0.41 0.82 0.89 0.41

edition tool, thus resulting in a higher accuracy and

quality.

Regarding the subjective experiment, a question-

naire was distributed among the subjects, which were

asked about the degree in which they considered each

of the two different edition methods to be comfort-

able, intuitive, reliable and agile using a 5-point Likert

scale; and also for the advantages and disadvantages

of each of these methods, being able to express other

considerations as well. The users’ average degree of

agreement with those subjectives aspects are shown

in table 1 and in ﬁgures 6(a), 6(b), 6(c) and 6(d), and

it can be drawn that the developed tool improves the

user experience and satisfaction when performing the

edition of the network.

4 SCENARIOS

This section describes real scenarios where the Situ-

ation Model is applied, either by itself or integrated

with the human-like interaction system, which illus-

trate how the model can be used to assist the user or

to improve the user experience with other application.

To carry out these scenarios, a smartphone ap-

plication has been designed and developed, provid-

ing a navigation service which takes advantage of

the Situation Model. This smartphone application al-

lows the user to choose a target (either from a list

or by speech) and to enable or disable certain situa-

tions; and it will provide the users on-screen (and also

voice-synthesised) navigation instructions to reach

their chosen targets. The ﬁrst scenario illustrates how

this mobile application would be used (further de-

scription of the tool usage is shown in http://youtu.be/

d-OsRm59z64).

Manual Edition Edition Tool

Absolute Error (meters)

(a) Absolute error.

Manual Edition Edition Tool

400

800

1200

1600

2000

Time (seconds)

(b) Edition time.

Figure 5: Box plots for the objective evaluation.

Manual Edition Edition Tool

Score (1-5)

(a) Perceived comfort.

Manual Edition Edition Tool

Score (1-5)

(b) erceived intuitiveness.

Manual Edition Edition Tool

Score (1-5)

Manual Edition Edition Tool

Score (1-5)

(d) Perceived agility.

Figure 6: Box plots for the subjective evaluation.

4.1 Improving a Navigation System

This scenario, illustrated in ﬁgure 7 shows how a

pedestrian navigation system, which usually only

considers the spatial aspect of the situation, can be en-

hanced where a broader view of the context is taken

into account. For this case, we will consider that the

system also observes the temporal and ambiental con-

ditions of the material aspect of the situation. Martha

(M) wants to go from the Torres Quevedo building

to the Sabatini building, in the campus of Legan´es of

Universidad Carlos III. It is 9.00 PM and there is a

heavy rain. Usually, Martha would take the shortest

route, which passes by the entry of the library and en-

ters the Sabatini park with the fountain, just before

going into the building by gate A.

This route is shown in the ﬁgure as M-A. How-

ever, there are some issues for which this route is not

Figure 7: Map of the campus illustrating the ﬁrst scenario.

ICEIS2014-16thInternationalConferenceonEnterpriseInformationSystems

9:41 AM

100%

Cognos.S Physical

Targets SituationsNavigation

Circumstantial Knowledge Management

for Human-Like Interaction

Edicio Sabatini

Turn Right

or choose a new target

and then go straight until the fountain

(a) Navigation screen.

9:41 AM

100%

Cognos.S Physical

Navigation Targets

Rainy

Ambiental

Night (over 9 PM)

Temporal

Situations

(b) Situations screen.

Figure 8: Smartphone application providing the navigation

service for the Situation Model.

convenient. In the ﬁrst place, entry A to Sabatini is

closed after 5.00 PM, so Martha will not be able to

access the building by this gate. Secondly, Martha

did forget her umbrella and she really hates walking

under the rain, something that wants to avoid as long

as possible.

Martha, who is a very clever girl, starts her navi-

gation system empowered by the Situation Model and

asks for the best route, as shown in ﬁgure 8(a). The

system is perfectly aware of the context: it knows that

entry A is closed after 5.00 PM, and it also knows that

the weather is really annoying for Martha (these situ-

ations are shown in ﬁgure 8(b)). Finally, the system

proposes an alternative route, M-B, which goes from

Martha location to entry B, which still opens at 9.00

PM. Moreover, this route is covered by the buildings,

so Martha will get to her goal mostly dry. However,

given that there is a small part of the route which is

not covered, the system can evaluate additional op-

tions; for instance, it could start a conversation with

Martha to ask whether she prefer the proposed path or

waiting unless the rain is over.

4.2 Enhancing a Natural Int. System

This scenario shows a sample dialog between a user

and a natural interaction system which is enriched by

a Situation Model. While the Situation Model itself

does not allow such a complex interaction, it provides

an added value by incorporating context-awareness

into the system, e.g., in this scenario the material and

political aspects of the context are considered.

Arthur is in the building of CafreSoft Corporation,

a young and promising software company. He has just

ﬁnished a job interview and is really satisﬁed. It is

11.30 AM and it takes a few hours for him to go home,

so he has decided to have his lunch in the building. He

starts a dialogue with his personal assistant, Hal:

ARTHUR: HAL, I’d like to eat something. What do you say?

[The system realizes it is 11.30 in the morning. It knows

that ARTHUR usually have breakfast at 9.00 AM and lunch

at 1.00 PM. HAL asks ARTHUR for more information.]

HAL: Nice! It’s 11:30, would you have breakfast, or lunch?

ARTHUR: Today I’ll have something for lunch.

[The system looks for places near ARTHUR.]

HAL: There’s a snack bar in this ﬂoor, you may have a sand-

wich there. Or you can go to the restaurant and eat from

the menu until 3 PM.

ARTHUR: Fine, let’s go there.

[HAL disambiguates the anaphora by asking ARTHUR.]

HAL: To the restaurant, right?

ARTHUR: Right.

HAL: Go ahead until the end of the corridor. Then go to the

ﬁrst ﬂoor. You can take the elevator.

ARTHUR: Nice. By the way, where can I ﬁnd a bathroom?

[HAL looks for bathrooms, and ﬁnds one for staff in the

ﬂoor. He looks for alternatives and assists ARTHUR.]

HAL: There’s a bathroom in this ﬂoor, only for personnel.

But you can still use the restaurant restroom.

ARTHUR: Ok, I’ll wait.

[ARTHUR calls the elevator.]

HAL: Remember, go to the ﬁrst ﬂoor.

ARTHUR: Thanks, HAL.

[ARTHUR steps out of the elevator in the ﬁrst ﬂoor.]

HAL: Ok, turn right and you’ll ﬁnd the restaurant.

ARTHUR: Ok, thanks a lot!

4.3 Assisting Disabled People

This scenario shows how the Situation Model could

take into account information about the user in order

to improve its experience.

Alan is working on his University entrance exams

these days. He has to do these exams in the campus of

Universidad Carlos III, and he does not know the col-

lege. Unfortunately he is using a wheelchair since he

was ﬁve and, while he is really used to his wheelchair,

he is still having some mobility issues.

Alan has downloaded in his smartphone a mobile

application developedby the Universitywich contains

the map of the campus and includes a simple naviga-

tion feature to guide the user to a searched location.

Furthermore, from the last update a simple Situation

Model was implemented which takes into account,

among others, some information about the user. For

instance, he could select in a proﬁle that he is a dis-

abled person, and the system will provide alternative

routes. In his case, the system will ﬁnd a route that

goes through ramps and elevators, rather than doing

through stairs. This route will also prefer the auto-

matic doors. Even if the route is longer and requires

more time to get to the goal, the system will prefer it

as long as it has less cost for Alan.

AnApproachtoCircumstantialKnowledgeManagementforHuman-LikeInteraction

5 CONCLUSIONS

This paper has proposed a design for a general-

purpose domain-independent Situation Model for-

malizing the circumstantial knowledge taking place

in an interaction. By incorporating this knowledge to

a human-like interaction system, it will be able to pro-

vide a more realistic and natural interaction, as well as

speciﬁc context-aware functionalities increasingly re-

quired in the highly dynamic environments in which

computer devices are used to work.

The Situation Model is implemented over a spatio-

temporal relational database, which aims to provide a

native support of the material aspect of the context,

while the inclusion of user-deﬁned situations also en-

ables a partial support for the other aspects of the con-

text. The circumstantial knowledge is formalized as a

N-dimensional graph, as it turns out to be a compre-

hensive yet powerful computable formalization.

Besides the Situation Model itself, a management

tool is proposed to edit the circumstantial knowl-

edge, providing means to manage networks and user-

deﬁned situations, as well as for starting a simulation

over the Situation Model, which eases the validation

of the edited knowledge. To support the network edi-

tion, a speciﬁc edition database was designed as a su-

perset of the original knowledge base adding speciﬁc

classes to display the network in the edition tool.

After both the Situation Model and the applica-

tion were implemented, an evaluation was carried out

to check whether the knowledge modeling by means

of the edition tool provides actual advantages to the

user. The evaluation was designed to compare the ef-

ﬁciency and quality of both manual modeling and the

edition using the tool. In all cases, the developed ap-

plication provided better results both in terms of time

required by the users to complete the task and accu-

racy of the resulting model. Finally, a subjective sur-

vey also proved that all the users preferred the edition

by means of the application regarding its reliability,

comfort, agility and intuitiveness.

Future work involves designing and running an

evaluation to check if the integration of the Situation

Model within a human-like interaction system would

actually provide the advantages described in section

2.1. To do so, the system should be evaluated before

and after the Situation Model is integrated, conclud-

ing whether these advantages take place.

Moreover, a mobile application has been devel-

oped supporting real-time edition of the Situation

Model. Evaluating the performance of this applica-

tion and comparing it with the desktop edition tool

remains as a future work.

ACKNOWLEDGEMENTS

This work has been be applied in a research

project funded by the Spanish Ministry of Industry

(CADOOH, TSI-020302-2011-21).

REFERENCES

Bernsen, N. O. (2000). What is Natural Interactivity? In

Proceedings of the 2nd International Conference on

Language Resources and Evaluation, pages 34–37.

Calle, F. J. (2004). Interacci´on Natural mediante Proce-

samiento Intencional: Modelo de Hilos en Di´alogos.

PhD thesis, Universidad Polit´ecnica de Madrid.

Calle, F. J., Albacete, E., S´anchez, E., del Valle, D.,

Rivero, J., and Cuadra, D. (2009). Cognos: A Nat-

ural Interaction Knowledge Management Toolkit. In

NLDB’2009, pages 303–304.

Calle, F. J., Albacete, E., S´anchez, E., Olaziregi, G., del

Valle, D., Rivero, J., and Cuadra, D. (2011). Cognos:

A Pragmatic Annotation Toolkit for the Acquisition

of Natural Interaction Knowledge. Procesamiento del

Lenguaje Natural, pages 269–276.

Cuadra, D., Calle, F. J., Rivero, J., and del Valle, D. (2009).

Applying Spatio-Temporal Databases to Interaction

Agents. In DCAI’2009, pages 536–540.

Ferguson, G. and Allen, J. F. (1998). TRIPS: An Integrated

Intelligent Problem-Solving Assistant. In AAAI-98,

pages 567–572.

Gee, J. P. (1999). Introduction to Discourse Analysis. Rout-

ledge.

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

Press.

Malaka, R. and Zipf, A. (2000). Deep Map: Challenging IT

Research in the Framework of a Tourist Information

System. In Information and Communication Tech-

nologies in Tourism 2000, pages 15–27. Springer.

Oviatt, S. and Cohen, P. (2000). Multimodal Interfaces that

Process what Comes Naturally. Communications of

the ACM, 43:45–53.

Porzel, R., Gurevych, I., and Malaka, R. (2006). In Context:

Integrating Domain- and Situation-Speciﬁc Knowl-

edge. In SmartKom: Foundations of Multimodal Dia-

logue Systems, chapter 18, pages 269–284. Springer.

Rivero, J., Cuadra, D., del Valle, D., and Calle, F. J.

(2007). Incorporating Circumstantial Knowledge In-

ﬂuence over Natural Interaction. In ICPS’2007, pages

415–420.

Wahlster, W. (2006). Dialogue Systems Go Multimodal:

The SmartKom Experience. In SmartKom: Foun-

dations of Multimodal Dialogue Systems, chapter 1,

pages 3–27. Springer.

Weiser, M. (1991). The Computer for the 21st Century.

Scientiﬁc American, 265:94–104.

ICEIS2014-16thInternationalConferenceonEnterpriseInformationSystems