An Intelligent Tutoring System for Procedural Training with Natural

Language Interaction

José Paladines

1,2

and Jaime Ramírez

Facultad de Ciencias Técnicas, Universidad Estatal del Sur de Manabí, Manabí, Ecuador

Computer Science School, Universidad Politécnica de Madrid, Madrid, Spain

Keywords: Context-aware Dialogue, Intelligent Tutoring System, Natural Language Processing, Procedural Training,

Virtual Environment.

Abstract: In this paper we present a proposal of an Intelligent Tutoring Systems equipped with dialogue in natural

language to facilitate student interaction with the learning environment, provide hints and answer students’

questions. This system is designed to be integrated with a 2D/3D virtual environment for procedural training,

where it can maintain a dialogue with students adapted to the context. Our notion of context comprises: the

specific features of the student; his/her progress in the development of the task; and the virtual environment

where it is performed. The dialogue will be controlled by a dialogue manager, built on Watson Assistant,

which has been chosen for its versatility. Additionally, we present an application example that describes the

operation of the modules that constitute the proposed approach. Then, we provide some indications on how it

will be evaluated with students shortly.

1 INTRODUCTION

During the last years, Intelligent Tutoring Systems

(ITS) have been developed that show effective results

in the teaching the concepts of physics, mathematics

and computer science. Some of these ITSs have been

able to capture the attention and interest of most

students through mixed conversational dialogues and

have been able to produce significant learning gains

beyond the classroom environment (Graesser et al.,

2001).

Furthermore, 2D/3D Virtual Environments (VEs)

have demonstrated to be valuable training tools by

simulating real scenarios where students can learn

without risks in a cost-efficient way (Dalgarno, 2002;

Duncan, Miller and Shangyi, 2012). 3D VEs offer a

superior learning experience, greater immersion,

greater fidelity and a high level of active student

participation, therefore, in addition to supporting the

learning tasks, can be intrinsically motivating to the

student to make decisions to achieve individual goals

within the environment (Dalgarno and Lee, 2010).

This kind of virtual environments become especially

useful for procedural training, that is, when students

have to learn to perform a task step by step that

eventually they will have to do in the real world.

However, after carrying out a literature review in

the field of ITSs, we can state that there are some ITSs

intended for procedural training, in which the student

can interact with a 2D/3D virtual environment, such

as Steve (develop physical tasks) (Rickel and

Johnson, 1999), SafeChild (pedestrian safety of

children) (Gu, Sosnovsky and Ullrich, 2015),

Lahystotrain (surgeons in laparoscopic operations

and hysteroscopy) (Los Arcos et al., 2000),

TRANSoM (pilots of remote-operated submarine

vehicles (ROV)) (Pioch, Roberts and Zeltzer, 1997),

but only in a few of these ITSs (Jacob, Paco and

Normit-SE) the user can interact with the system

through a dialogue in natural language.

The goal of this paper is to present a proposal for

an ITS capable of maintaining a dialogue with the

student in natural language while the student is

performing a procedural training in a 2D/3D virtual

environment. To generate the dialogue in natural

language, the specific features of the student, his/her

progress in the development of the task and the

physical environment where it is performed will be

taken into account. Therefore, the tutoring feedback

will include a dialogue adapted to the context.

Moreover, in a virtual environment for procedural

training, the actions related to the development of a

procedure become particularly relevant and must

Paladines, J. and Ramírez, J.

An Intelligent Tutoring System for Procedural Training with Natural Language Interaction.

DOI: 10.5220/0007712203070314

In Proceedings of the 11th International Conference on Computer Supported Education (CSEDU 2019), pages 307-314

ISBN: 978-989-758-367-4

307

serve to determine the most suitable tutoring at each

moment.

To generate the dialogue between the student and

the ITS, we will build a dialogue manager with the

cognitive services of some currently available

platforms (Mallios and Bourbakis, 2016).

This paper is structured as follows: section 2

presents the previous works most related to the

objectives of this paper; Section 3 summarizes the

main features of the platforms for the construction

of dialogue managers; Section 4 details the

architecture of the proposed system; Section 5

describes, through an application example, the

operation of the ITS integrated with the dialogue

manager; and finally, section 6 shows the

conclusions and future work.

2 RELATED WORK

In the literature we find few ITSs that use dialogue in

natural language for procedural training, because

most of this kind of ITSs are oriented to the teaching

of concepts. Therefore, we believe that to show a

reasonable view of the state of the art in this field, it

is convenient to first present some remarkable

systems that simulate patterns of discourse and

include pedagogical strategies of a human tutor

through dialogues for declarative instruction, and

then present the existing ITSs with dialogue for

procedural training.

CircSim (Glass, 2001), considered one of the first

systems to implement dialogues in natural language,

can analyse the user input through a syntactic analysis

and admits short answers of one or two words. It has

a tutoring strategy based on Directed Reasoning Lines

to control the variables of a prediction table and is

oriented to the physiology domain. Atlas-Andes

(Rosé et al., 2001) adds dialogue capabilities to the

Andes system (Gertner and VanLehn, 2000),

allowing students to participate in a typed dialogue. It

has a tutoring strategy based on knowledge

construction dialogues (KCD) by means of which it

involves the students in a dialogue about correcting

conceptual misconceptions of physics. Why2-Atlas

(VanLehn et al., 2002) encourages students to write

essays as answers to a question. The analysis of these

long explanations and the discovery of

misconceptions is possible through CARMEL. Like

Atlas-Andes, it is aimed at the teaching of physics.

ITSpoke (Litman and Silliman, 2004) is a system that

involves students in spoken dialogues of conceptual

physics. To analysis the statements of students, it uses

the front-end of the Why2-Atlas system. Regarding

the tutoring strategy, only Atlas-Andes, Why2-Atlas,

ITSpoke ITSs rely on KCDs. These KCDs are based

on the CircSim Directed Reasoning Lines. Other

outstanding ITS for declarative instruction is

AutoTutor (Graesser et al., 2005), a system that

maintains conversations of mixed initiative with

students to allow them to build explanations of

concepts. It uses a dialogue pattern called expectation

and misconceptions tailored dialogue that consists in

comparing the explanations of the students with a set

of expectations (ideal answers) and misconceptions

(incorrect answers) using a statistical technique called

latent semantic analysis. This system has evolved

over time and has managed to cover different

domains such as computer literacy, biology, physics

and critical thinking. Beetle II (Dzikovska et al.,

2014), implements an approach based on a task-based

dialogue system supported by a simulation that

generates a dynamic learning context. The dialogue

with the student is executed through a cycle

“Predicting, verifying, evaluating” with which we can

analyse the predictions and mistakes that the student

may make. It is aimed at teaching electricity and

electronics and uses an ontology to represent the

knowledge domain. Both AutoTutor and Beetle II

support the constructions of explanations through the

dialogue but differ in the approach to interpret the

student's statement. AutoTutor applies a statistical

approach whereas Beetle II employs a hybrid

approach, that is, adds a statistical classifier to the

semantic analyser for a better interpretation of

statements.

Next, we will mention the only ITSs with dialogue

in natural language for procedural training. Jacob

(Evers and Nijholt, 2000) teaches how to solve the

problem of the Towers of Hanoi and provides

instructions and assistance to execute tasks in a

virtual 3D environment. Paco (Rickel et al., 2002)

supports tutoring actions as part of a collaborative

dialogue system (built on collagen) that uses rules for

the generation of speech acts. This generation of acts

is based on a task model, a student model and the

interaction with the student. Normit-SE (Mitrovic,

2005) teaches the database normalization process.

The dialogue starts from the moment the student

makes an error to which he must provide explanations

by selecting one of the options (solutions) offered in

a menu. These systems provide help when requested

and positive comments, although in the case of Jacob

it is occasional. To analyse the students' statements,

they use a symbolic approach based on superficial

semantic grammars.

CSEDU 2019 - 11th International Conference on Computer Supported Education

308

3 DIALOGUE MANAGER

Nowadays, there are some platforms on the market to

implement conversational interfaces; among the most

popular are Google's Dialog Flow, Facebook

Messenger’s Wit.ai, IBM's Watson, Microsoft's

LUIS, etc. Each of them has its own characteristics,

advantages and disadvantages, but to structure the

flow of conversation or dialogue, these platforms use

some common elements such as utterances, intentions

and entities, which facilitate the Natural Language

Processing (Singh, 2017).

• An utterance is the user input that the

application needs to interpret, which in some

cases are not well formulated.

• An intention represents the purpose expressed

in the user's input. The same intention can be

expressed through different user sentences.

• An entity represents an instance of an object

class that is relevant to a user's intention and

can be identified in an utterance.

Something outstanding of these platforms is that

they support the use of a structure called context that

facilitates the adaptation of the dialogue to different

situations in a given scenario. This structure is used

internally and externally to pass information between

a client application and the dialogue manager.

For this work, we have only considered the IBM,

Microsoft and Google platforms because apart from

the dialogue managers services, they also have a

natural language understanding module necessary to

decompose the student's statement into entities and

relationships. However, after comparing these three

platforms, we have chosen IBM because it enables us,

among other things, to handle a larger number of

intents; to define concepts, dictionaries and

relationships through annotators; and to manage the

context information more easily. In addition a recent

comparative study of conversational platforms

(Koplowitz et al., 2018) positions the IBM platform

as the most complete and robust of the market.

From the IBM Watson platform, the following

cloud services will be used:

• Watson Knowledge Studio (WKS), to create

the automatic learning annotator by

identifying the mentions (entities) and

relationships in unstructured texts;

• The Natural Language Understanding (NLU)

module (IBM, 2016) to apply the machine

learning annotator obtained from the WKS;

and,

• The Watson Assistant (IBM, 2018) to build

the nodes of dialogue based on the intents,

entities and context variables necessary for the

training.

4 PROPOSED APPROACH

The dialogue manager associated with the ITS must

have enough information to be able to provide a

contextualized dialogue as part of the tutoring

feedback. To provide this type of dialogue, the

context must contain information related to the

student's knowledge, the virtual environment (with

static and dynamic information) and the student's

current progress in the practice assignment to be

performed. Some examples of personalized tutoring

feedback that may be provided in this way are the

following ones:

• Answer questions related the ubication and

identification of an object in the virtual

environment or how reach it, even though

this object is a distant position to the

student's avatar one

• Answer questions related the next action to

be done in the practice assignment.

• Recommend learning activities to fill gaps in

knowledge demonstrated by the student.

• Provide hints proactively to guide the

student with the execution of a task, if it is

observed that the student needs help, even if

he/she is not asking for it.

• Encourage an affective dialogue to mitigate

students’ inactivity or moments of

discouragement.

Figure 1 describes the architecture of the ITS with

dialogue in natural language. This architecture

contains four main components: the Procedural

Training Environment (PTE), the Natural Language

Understanding (NLU) System, the ITS and the

Dialogue Manager (DM).

The PTE is the module that simulates the real

environment where the task related to the practice are

carried out. In order to give more realism to the tasks

that the student must perform throughout the practice,

this environment can be a virtual world in 3D. The

interaction of the user with the PTE can be generated

through events such as questions, attempts of actions,

etc., that will be delivered to the ITS during the

development of the practice.

The NLU System will be responsible of receiving

the user's sentences and extracting from them their

An Intelligent Tutoring System for Procedural Training with Natural Language Interaction

309

composing entities and relationships. In this way, it

will preprocess the statement of the user, so that later

the ITS modules can work with the semantics of the

statement. The NLU system will be developed using

the Watson Knowledge Studio and the Watson’s

Natural Language Understanding Service.

The ITS will be integrated by the modules

corresponding to a classic ITS plus a world module.

The World Module (WM) will represent the physical

characteristics of the virtual training environment,

that is, it will contain information about the scenarios

and constituent aspects of the virtual Its content will

be useful so that the system can answer questions, for

example, about the situation of an object or how to go

from one place to another. environment, related to

avatars and 2D/3D objects. The Student Module (SM)

will contain information related to the student. In this

work, we will adopt the student model proposed by

(Clemente, Ramírez and de Antonio, 2011), because

it fits very well to your needs. This SM contains

information on: the student's actions; his/her

movements through the virtual environment; the

questions he/she asked; the hints he/she received from

the tutor, etc. From this information, the same module

will infer, with a certain level of reliability, the

student knowledge that, in turn, will be useful to

decide the best tutoring strategy in each moment.

Both the information of the WM and the SM will be

represented by means of ontologies, because they

support the representation of sufficiently abstract and

properties as well as facilitate their own reuse and

even their own extension to other application

contexts, if necessary. To access the information on

these ontologies, Jena framework will be used. The

Expert Module (EM) will contain information about

the knowledge that the student must learn. In this

case, the EM will contain a complete description of

the procedure to be learned. The Tutor Module (TM),

based on the information of the other modules, will

provide students with adequate feedback at every

moment of their learning. The ITS extracts the

information from the different modules to build the

context. This context will be represented by an

ontology and will be filled with information from the

SM (state of knowledge, progress of the activity,

student's trajectory in the virtual environment), the

EM (the correct plan, the next correct action) and the

WM (structure of the virtual world, position of the

student's avatar, descriptions of the objects). Once the

contextual information is collected, before passing it

to the Dialogue Manager, this information will be

transformed into another representation

understandable by the Dialogue Manager.

The DM will contain the definition of the structure

of the dialogue, i.e., the intentions, the entities and the

dialogue nodes specifically intended to the training

environment. This component will be implemented

through the Watson Assistant and will be responsible

for the dialogue with the user taking into account the

contextual information provided by the ITS.

Figure 1: Architecture of proposed approach.

CSEDU 2019 - 11th International Conference on Computer Supported Education

310

5 EXAMPLE OF APPLICATION

For a better understanding of the tutoring process and

the generation of dialogue in natural language, we

present an example taken from a practice assignment

that is performed in a Virtual Biotechnology

Laboratory (In

http://youtu.be/ mAFREZ5_iak you can

find a video of this virtual laboratory). This laboratory

was developed on the platform of virtual worlds

OpenSimulator, as part of a master thesis (Riofrío-

Luzcando, 2012). To carry out the practice

assignment, the student controls an avatar and has he

help of an automatic tutor, who will give him

indications about the actions to be carried out at each

moment and will show him error messages, when he

makes a mistake. These indications and error

messages consist of messages previously configured

and associated with each of the actions that must be

carried out in the practice assignment.

The current version of the automatic tutor doesn’t

implement the architecture proposed in the previous

section. However, the example that will be explained

below describes the behavior of the ITS as if it were

really equipped with the proposed architecture and

implemented the process of tutoring detailed in the

previous section.

In this example, we are going to assume that the

student asks for a chemical that he needs to prepare

the mix, and after receiving the hint, he looks for it

and ends up adding a different chemical to the one he

is looking for. The tutor will give the hint based on

the student's level of knowledge.

STUDENT (Utterance) Where is Casein? /

Where can I find casein?

TUTOR (General hint) “Casein is in the

showcase”

STUDENT (Action attempt) Look in the

showcase and try to add the bisacrylamide

chemical.

The tutor detects that the student has taken an element

that isn’t casein and decides to block the action, send

an error message and give a more specific hint about

the chemical to choose.

TUTOR Error Message “The action attempt is

incorrect”

(Specific hint) “You must add casein to the mix,

that's the right thing!

As we can see, the dialogue is composed of two

iterations; and to show how the tutoring feedback is

generated, Figure 1 has been used as a base, so that

enumerated circles have been drawn on it to specify

the sequence of steps executed by the ITS modules.

Table 1: Iterations of the Dialogue Example.

First Iteration: The student asks the ITS, because he/he

does not know where Casein is.

1. The student asks where Casein is

2. The PTE retransmits the question to the NLU System.

3. The NLU System interprets the question and breaks it

down into entities and relationships and sends them to

the ITS Communication Module (CM).

4. The CM sends the question broken into entities and

relationships to the SM.

• The SM infers that the student doesn’t know where

the casein is and updates the student’s knowledge

state in the ontology.

• The learning objective to be evaluated is: “The

student knows where the casein is”.

• Therefore, the state of knowledge of the student will

contain “The student has not acquired that objective

with a certain degree of certainty”.

5. The CM sends the question to the TM Where is

Casein?

6. The TM, once the student's question arrives, asks the

WM Where is casein? Where is the student? Then,

WM returns:

“Casein is in the showcase” and “The showcase is in

the main room”

“The student is in the main room”

7. The TM asks the SM: Does the student know where

the showcase is? What was the overall performance of

the student so far in practice assignment?

As the student has previously taken another

chemical from the showcase and this was recorded in

the student’s ontology, the SM will answer the TM:

“Yes, the student already knows where the

showcase is”

“The performance was good (made a few

mistakes)”

8. The TM sends to the CM the information obtained

from the WM and SM with which the context

information will be elaborated. In addition, it indicates

that the student is going to need a general hint.

9. The CM sends the student's question to the DM.

Where is Casein? and the context with the following

information:

“The student is in the main room”.

“Casein is in the showcase”.

“The showcase is in the main room”.

“The student knows where the showcase is”.

Hint Level: General

10. The DM builds the hint according to the level

decided by the TM and send it to the PTE.

General: “Casein is in the showcase”

Second Iteration: After receiving the hint, the student

tries to add bisacrylamide to the mix, so the action

attempt is blocked.

1. The student tries to add the bisacrylamide chemical

element to the mix.

2. The PTE retransmits the action attempt (event) to the

CM.

An Intelligent Tutoring System for Procedural Training with Natural Language Interaction

311

3. The CM sends the event to the SM.

• The SM registers in its ontology that the student

tries to perform an action.

• The learning objective to be evaluated is: “The

student must add casein to the mix”.

4. The CM sends the event to the TM.

5. The TM asks the EM if the event is correct and what

is the next action; and EM returns: “The event is

wrong”, “Add casein to the mix”.

6. The TM tells the SM “The action of adding

bisacrylamide to the mix is incorrect” and asks What

was the recent performance of the student so far in

practice assignment?

• The SM infers that the student has not acquired this

learning objective or reinforces his belief that he has

not acquired it, since he has just obtained new

evidence of it.

• The SM returns that the performance was bad for

the next action.

7. The TM tells the CM:

“The action of adding bisacrylamide to the mix is

incorrect”.

NEXT_ACT_PLAN: “Add casein to the mix”.

In addition, it indicates that the student is going to

need a specific hint.

8. The CM tells the DM the following.

“The attempt of action is incorrect”.

NEXT_ACT_PLAN: “Add casein to the mix”.

9. The DM sends the PTE an error message and hint

according to the level decided by the TM.

“The attempt of action is incorrect”.

Specific: “You must add casein to the mix, that's the right

thing!”

5.1 Application of the Natural

Language Understanding

The Natural Language Understanding system has an

automatic learning annotator that was built from some

entities, dictionaries and an annotated corpus of

dialogue. These entities, dictionaries and the corpus

refer to the terminology associated with the

procedural tasks that students must perform as part of

the practice assignment. If we apply this annotator to

the student’s question, the annotator will generate a

file in JSON format that would have the following

information in abbreviated form:

Utterance: Where is casein?/ Where can I find

casein?

Relation: LocateObj

Entities: Adv_ans (Where),

Action (Is),

Chemical (Casein).

5.2 Application of the Student Module

The Student Module is responsible for controlling the

information related to the learning objectives, the

state of the student knowledge, the trace of the

student, etc. In this sense, this information is

represented by a network of ontologies and updated

by means of diagnosis rules (Clemente, Ramírez and

de Antonio, 2011).

First Iteration:

Learning objective: “The student knows where

the casein is”

State of the student's knowledge: Student does not

know where the casein is.

Rule:

Type_Of_Question(question,Where_Is(obj)) →

Add_SM(¬Know(Where_Is(obj)))

Second Iteration:

Learning objective: “The student must add the

casein to the mixture”

Trace of the student: Student tried to add

bisacrylamide.

Rule:

Try_To_Apply(actx) ˄ Next_Act_Plan(acty) ˄

¬Equal(actx, acty) →

Add_SM(¬Know(Next_Act_Plan(acty)))

5.3 Application of the World Module

Within the World Module there is an ontology,

expressed in OWL language, in which information

about the 3D virtual environment has been

represented, that is, the names and locations of the

contained objects, the location of the student’s avatar,

etc. In this case, this module provides the following

information:

First Iteration:

<owl:ObjectProperty rdf:ID="Is">

<rdfs:domain

rdf:resource="#chemical"/>

<rdfs:range

rdf:resource="#container"/>

</owl:ObjectProperty>

Second Iteration:

<owl:ObjectProperty rdf:ID="Is">

<rdfs:domain

rdf:resource="#container"/>

<rdfs:range rdf:resource="#room"/>

</owl:ObjectProperty>

5.4 Application of the Tutor Module

The Tutor Module decides and applies the tutoring

strategy according to the pace of the student's

CSEDU 2019 - 11th International Conference on Computer Supported Education

312

learning, just as a human tutor would. To do this, it

interprets the requests of the communication module;

collects the information coming from the world,

expert and student modules; and encapsulates the

necessary logic to return the information required by

the Dialogue Manager, and thus enables the dialogue

with the student. To build this information in the

example, TM will have to submit the following

requests:

First Iteration, TM requests information to:

WM: Where is the casein? Where is the student?

SM: Does the student know where the showcase

is? What was the overall performance of the

student so far?

Second Iteration, TM requests information to:

EM: Is this event correct or incorrect?

SM: What was the recent performance of the

student?

5.5 Application of the Dialogue

Manager

The Dialogue Manager receives from the CM the

user's event and the context data. Then, with this

information the dialogue manager generates the

response in natural language. For the generation of

natural language, a dialogue structure has been

designed, integrated by the necessary intents, entities

and dialogue nodes. Next, the dialogue structure and

the contexts employed in the example are detailed in

a simplified form:

First Iteration

Intent #locate = Where is Casein?

Entities @chemical = Casein

Context variables

$levHin: “G”,

$posStu: “main room”,

$objLoc: “casein”,

$contObjLoc: “showcase”,

$ubiSpaObj: “in”,

$posObjLoc: “main room”,

Node information

If #locate and (@element || @document ||

@chemical)

If ($levHin== “G”) && ($posStu == “main

room”)

The answer in natural dialogue would be:

The $objLoc is $ubSpaObj the $contObjLoc

Second Iteration

Intent #trylocate = “ ”

Entities @chemical = “bisacrylamide”

Context variables

$levHin: “C”,

$posStu: “main room”,

$objLoc: “casein”,

$contObjLoc: “showcase”,

$posObjLoc: “main room”,

$nextAccPlan: “Add casein to the mix”

Node information

If #trylocate and ($posStu == “main room”)

and ($objLoc!=@chemical)

If ($levHin== “C”)

The answer in natural dialogue would be:

“The attempt of action is incorrect”

“You must $nextAccPlan, that’s the right thing!”

6 CONCLUSION AND FUTURE

WORK

We have presented a proposal of an Intelligent

Tutoring System for Procedural Training with Natural

Language Interaction. To generate the interaction in

natural language, we implemented a dialogue manager

and a natural language understanding module, using

the IBM Watson platform because it supports all the

required services. To detail the proposed solution, we

have presented an application example that describes

how the dialogue manager would be integrated with

the ITS. In the future, we plan to conduct a pilot study

with 25 students of the first semester of the Forestry

Engineering Degree of the Universidad Politécnica de

Madrid. We will select a part of the practice that is

carried out in the Virtual Biotechnology Laboratory, in

which the dialogue manager will help students and

answer their ques-tions. In this way we will evaluate

the effectiveness and robustness of the proposed

approach. The results of this study will serve to identify

and address any inconvenience, so that we can

successfully conduct a second study in which we will

employ the dialogue system in the entire practice. To

formulate the evaluation metrics, we plan to use the

Goal, Question, Metric (GQM) methodology [25].

This methodology is used to identify which metrics are

necessary to achieve the objectives of the pilot study.

The evaluation of the first study will be conducted

through an experiment to test the following research

questions: a) Is the ITS robust? b) Does the ITS answer

the students’ questions properly?. Dialogues will be

collected from the ITS log file and analysed to evaluate

the quality of the ITS answers. Later, after analysing

the results of the first study, we will define the research

questions for the second study.

An Intelligent Tutoring System for Procedural Training with Natural Language Interaction

313

ACKNOWLEDGEMENTS

Paladines would like to acknowledge financial

support from the Universidad Estatal del Sur de

Manabí.

REFERENCES

Clemente, J., Ramírez, J. and de Antonio, A. (2011) ‘A

proposal for student modeling based on ontologies and

diagnosis rules’, Expert Systems with Applications,

38(7), pp. 8066–8078.

Dalgarno, B. (2002) ‘The potential of 3D virtual learning

Environments: A constructivist analysis’, Electronic

Journal of Instructional Science and Technology, 5(2),

pp. 1–19.

Dalgarno, B. and Lee, M. J. W. (2010) ‘What are the

learning affordances of 3-D virtual environments?’,

British Journal of Educational Technology, 41, pp. 10–

32. doi: 10.1111/j.1467-8535.2009.01038.x.

Duncan, I., Miller, A. and Shangyi, J. (2012) ‘A taxonomy

of virtual worlds usage in education’, British Journal of

Educational Technology, 43(6), pp. 949–964.

Dzikovska, M. et al. (2014) ‘BEETLE II: Deep natural

language understanding and automatic feedback

generation for intelligent tutoring in basic Electricity

and Electronics’, Inter. J. Artif. Intell. Educ., 24(3), pp.

284–332.

Evers, M. and Nijholt, A. (2000) ‘Jacob - An animated

instruction agent in virtual reality’, in Tan, T., Shi, Y.,

and Gao, W. (eds) Advances in Multimodal Interfaces -

ICMI 2000. Berlin, Heidelberg: Springer Berlin

Heidelberg, pp. 526–533.

Gertner, A. S. and VanLehn, K. (2000) ‘Andes: A coached

problem solving environment for Physics’, in Gauthier,

G., Frasson, C., and VanLehn, K. (eds) Proc. 5th Inter.

Conf. Intelligent Tutoring System. Berlin, Heidelberg:

Springer Berlin Heidelberg, pp. 133–142. doi:

10.1007/3-540-45108-0_17.

Glass, M. (2001) ‘Processing language input in the

CIRCSIM-Tutor intelligent tutoring system’, in Al, J.

D. M. et (ed.) Proc. Artif. Intell. in Educ. San Antonio:

IOS Press, pp. 210–221.

Graesser, A. et al. (2005) ‘AutoTutor: An Intelligent

Tutoring System With Mixed-Initiative Dialogue’,

IEEE Transactions on Education, 48(4), pp. 612–618.

doi: 10.1109/TE.2005.856149.

Graesser, A. C. et al. (2001) ‘Intelligent tutoring systems

with conversational dialogue’, AI Mag., pp. 39–51.

Gu, Y., Sosnovsky, S. and Ullrich, C. (2015) ‘SafeChild:

An intelligent virtual reality environment for training

pedestrian safety skills’, in Conole, G. et al. (eds)

Design for Teaching and Learning in a Networked

World. Cham: Springer International Publishing, pp.

141–154.

IBM (2016) Natural Language Understanding. Available

at: https://www.ibm.com/watson/services/natural-

language-understanding/ (Accessed: 15 September

2017).

IBM (2018) Watson Assistant. Available at:

https://console.bluemix.net/docs/services/assistant/gett

ing-started.html#getting-started (Accessed: 3 October

2018).

Koplowitz, R. et al. (2018) The Forrester New Wave

Conversational Computing Platforms, Q2 2018.

Litman, D. J. and Silliman, S. (2004) ‘ITSPOKE: An

Intelligent Tutoring Spoken Dialogue System’, in

Demonstration Papers at HLT-NAACL 2004.

Stroudsburg, PA, USA: Association for Computational

Linguistics (HLT-NAACL--Demonstrations ’04), pp.

5–8.

Los Arcos, J. L. et al. (2000) ‘LAHYSTOTRAIN:

Integration of virtual environments and ITS for surgery

training’, in Gauthier, G., Frasson, C., and VanLehn, K.

(eds) Int. Conf. Intelligent Tutoring System. Berlin,

Heidelberg: Springer Berlin Heidelberg, pp. 43–52.

Mallios, S. and Bourbakis, N. (2016) ‘A survey on human

machine dialogue systems’, in Proc. 7th Inter. Conf.

Inform. Intell. Syst. & Applicat., pp. 1–7.

Mitrovic, A. (2005) ‘The Effect of Explaining on Learning:

A Case Study with a Data Normalization Tutor’, in Int.

Conf. on Artificial Intelligence in Education (AIED).

Amsterdam, The Netherlands, The Netherlands: IOS

Press, pp. 499–506.

Pioch, N., Roberts, B. and Zeltzer, D. (1997) ‘A virtual

environment for learning to pilot remotely operated

vehicles’, in Proc. Int. Conf. Virtual Systems and

MultiMedia, pp. 218–226.

Rickel, J. et al. (2002) ‘Collaborative discourse theory as a

foundation for tutorial dialogue’, in Proc. 6th Int. Conf.

on Intelligent Tutoring Systems. Springer-Verlang, pp.

542–551.

Rickel, J. and Johnson, W. L. (1999) ‘Animated agents for

procedural training in virtual reality: Perception,

cognition, and motor control’, Applied Artificial

Intelligence, 13, pp. 343–382.

Riofrío-Luzcando, D. (2012) Diseño e implementación de

un laboratorio virtual de Biotecnología. M.S. thesis,

ETSII, Univ. Politécnica de Madrid, Madrid, España.

Rosé, C. et al. (2001) ‘Interactive conceptual tutoring in

Atlas-Andes’, in Morre, J. D., Redfield, C. L., and

Johnson, W. L. (eds) Inter. Conf. Artifi. Intell. in Educ.

Amsterdam: IOS Press, pp. 256–266.

Singh, B. (2017) Chat Bots - Designing intents and entities

for your NLP models. Available at:

https://blogs.msdn.microsoft.com/brijrajsingh/2017/01

/29/chat-bots - designing-intents-and-entities-for-your-

nlp-models/ (Accessed: 8 February 2017).

VanLehn, K. et al. (2002) ‘The architecture of Why2-Atlas:

A coach for qualitative Physics essay writing’, in Cerri,

S. A., Gouardères, G., and Paraguaçu, F. (eds) Inter.

Conf. on Intelligent Tutoring Systems. Springer Berlin

Heidelberg, pp. 158–167.

CSEDU 2019 - 11th International Conference on Computer Supported Education

314