An Educational Talking Toy based on an Enhanced Chatbot

Eberhard Grötsch

, Alfredo Pina

, Roman Balandin

, Andre Barthel

and Maximilian Hegwein

Fakultät Informatik und Wirtschaftsinformatik, Hochschule für Angewandte Wissenschaften,

Sanderheinrichsleitenweg 20, Würzburg, Germany

Departamento de Informática, Universidad Pública de Navarra, Campus Arrosadia, Pamplona, Spain

Keywords: Chatbot, Autonomous Toy, Children, Memory, Educational Patterns, Educational Talk.

Abstract: Children are often motivated in their communication behaviour by pets or toys. Our aim is to investigate,

how communication with “intelligent“ systems affects interaction of children. Enhanced chatbot technology

– hidden in toys - is used to talk to children. In the Háblame-project (started as part of the EU-funded

Gaviota project) a first prototype talking German is available. We outline the technical solution, and discuss

further steps.

1 INTRODUCTION

Within the Gaviota project (Bossavit et al., 2014),

we worked on a system to investigate how oral

communication with “intelligent“ systems affects the

oral interaction of children with typical or untypical

development.

To answer that question, we started to develop

those intelligent systems first. While talking and

understanding systems are widely in use (e.g. Siri,

provided on some Apple devices), we did not find a

system, which can be configured to our special

needs.

So we decided to develop our own system.

2 REQUIREMENTS

Our goal is to develop a multi-client educational

system, which talks to different children having

individual knowledge about them (e.g. homework,

friends, parents, favourite games). We assume that it

also can talk to people suffering from dementia

about their personal daily routine. The system

should talk like an adult human.

A special tool has to be provided to enable non-

programmers to administrate basic knowledge about

individual persons. Client data have to be protected

against unauthorized access.

Different from standard chatbots, the system

does not just answer questions of the clients, but it is

able to start a conversation, or able to begin with a

new topic.

3 LEVELS OF TALKING

Figure 1: Levels of talking (Bossavit et al., 2014)

There are different levels of talking (fig. 1): from

just checking facts to deep and open dialogs. Only

answering questions about facts does not meet the

requirements, so small talk should also be covered in

the project. Usually chatbots are used to cover small

talk, but that is not sufficient for educational

purposes. The purpose of the system requires

meaningful and target oriented talk.

Our system does not try to cover deep and open

dialogs.

So the first attempt to solve the problem given is

to use somehow enhanced chatbot technology.

ques%ons(

about(facts((

“when&will&the&

next&plane&leave&

to&Porto&Alegre”&

small(talk(

“I&like&the&

conference&today&

very&much“&

(

dialog(with(

fix

d(target(

“I&think&you&should&

clean&your&shoes,&

before&you&leave&

the&house“&

(

deep(and(

open(dialog(

“Do&you&s@ll&love&

me?“(

(

360

Groetsch E., Pina A., Balandin R., Barthel A. and Hegwein M..

An Educational Talking Toy based on an Enhanced Chatbot.

DOI: 10.5220/0005452403600363

In Proceedings of the 7th International Conference on Computer Supported Education (CSEDU-2015), pages 360-363

ISBN: 978-989-758-107-6

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

4 PREVIOUS WORK

4.1 The Beginning: Eliza

Already in 1966 Weizenbaum (Weizenbaum, 1966)

implemented Eliza, a talking system (based on text

strings, without speech). The Eliza implementation

used to react to a limited number of key words

(family, mother, ...) to continue a dialog. Eliza had

no (deep) knowledge about domains - not even

shallow reasoning, rather a smart substitution for

strings. Modern versions of Eliza can be tested on

several websites, e.g. (ELIZA, 2013).

4.2 Traditional Dialog Systems

Most dialog systems (e.g. the original Deutsche

Bahn system giving information about train time

tables, or the extended system by Philips) are able to

guide people who call a hotline and execute

standardized business processes (delivering account

data, changing address data, etc.). The dialogs in

such systems are predefined and follow strict rules.

They work well, but within a limited domain.

Chatbots are more flexible. Depending on the

size of their database, they can talk without fixed

sequences about a wide area of topics (cf. 4.4).

4.3 Natural Language Processing

(NLP)

A spectacular demonstration of natural language

processing was given by IBM’s artificial intelligence

computer system Watson in 2011, when it competed

on the quiz show Jeopardy! former human winners

of that popular US television show (JEOPARDY,

2011).

IBM used the Apache UIMA framework, a stan-

dard widely used in artificial intelligence (UIMA,

2013). UIMA means “Unstructured Information

Management Architecture“.

The source code for a reference implementation

of this framework is available on the website of the

Apache Software Foundation.

Systems that are used in medical environments

to analyse clinical notes serve as examples.

4.4 Chatbots

Chatbots like Siri or Alice are popular among users

of mobile phones to ease interaction with the system.

Their domains are small talk, access to apps like

calendars, or access to services like weather or

traffic information.

Today chatbot technology is accepted in those

areas mentioned, but it is not widely used.

However, in general chatbots do not have special

knowledge about their users, and they do not initiate

interaction with the user.

This is a severe limitation to educational systems

– the system has to have at least basic knowledge

about its clients, and it has to have a concept how to

start communication and how to overcome breaks in

the interaction. Therefore we concluded, that

chatbots are a suitable tool to begin with, but we had

to enhance them to meet the additional requirements

of educational systems.

5 THE ”HÁBLAME“ PROJECT

5.1 Concept of a Dialog System

In the beginning, we intended to achieve our goals

by real natural language processing (NLP), i.e., we

studied basic concepts of NLP, and started with

syntax and semantic parsing (Grötsch et.al., 2013).

Figure 2: Concept of a dialog system based on (Schmitt,

2014).

But we soon recognized, that it is rather tedious

to build such an NLP system based on deep

language understanding – although there are some

powerful tools available.

So we decided to use chatbot technology as

described above. Fig. 2 shows the basic overall

architecture of our system: we use a client

(smartphone) for speech-to-text and text-to-speech

conversion, the input strings recognized are sent to a

server, which implements the enhanced chatbot

functions and generates an output string, which is

returned to the client.

5.2 Chatbot Architecture

None of the chatbots available fulfill the

requirements of section 2. Therefore we looked for

an open system, which we could adapt to our needs.

Client'

(

speech'to'text'

(

Android&

(

text'to'speech'

(

Server'

(

analysis'

(

database,&

AIML,&

A.L.I.C.E.&

(

evalua4on'

(

input(string(

output(string(

speech(in(

speech(out(

AnEducationalTalkingToybasedonanEnhancedChatbot

361

We selected the “Artificial Linguistic Internet

Computer Entity” (A.L.I.C.E.)

“A.L.I.C.E. is an artificial intelligence natural

language chat robot based on an experiment

specified by Alan M. Turing in 1950. The A.L.I.C.E.

software utilizes AIML, an XML language we

designed for creating stimulus-response chat

robots.” (Wallace, 2014). It is released by the Alice

A.I. Foundation under the GNU General Public

License.

So A.L.I.C.E. uses the Artificial Intelligence

Markup Language (AIML) to store information. It

can generate new phrases, but they are stored under

supervision of a human to prevent wrong input.

5.3 Client Functions

The clients are Android based systems. Therefore

the built-in speech to text system of Android is used,

which produces text strings. Those strings can be

forwarded to an app which also handles the

communication to and from the server.

Text to speech is covered by an app provided by

Google.

5.4 Server Functions

The server runs A.L.I.C.E. and a file which contains

AIML, which describes the patterns and templates of

the dialog. (Our work is based on AIML 2.0, and its

reference implementation ProgramAB)

A.L.I.C.E. is enhanced by the individual

personal memory (IPM, cf. 5.5). To act as web

server it is based on Apache Tomcat.

A short example shows the structure of AIML

(table 1).

Each category (fundamental unit of knowledge

in AIML) consists at least of a pattern (human input)

and a template (answer of the bot). Wildcards (“*”)

grant more flexibility in describing the input. In

table 1, line 2 the wildcard “*” covers any arbitrary

input, and whatever the human says in the

beginning, the answer of the bot will be: “What is

your name?” The content of the wildcard in the

pattern of line 6 is used in the answer of template

line 8. Line 14 and line 20 are references to a

already defined template: to the template of line 6,

and the answer to all three patterns (line 6, 14 and

20) will be template line 8: “* is a beautiful name”.

AIML offers more elements than the ones

shown: e.g. several prioritized wildcards, topics,

which cover several categories, or random, which

selects by accident, or conditional branching.

Table 1: Example of simple AIML (Balandin, Hegwein,

Barthel, 2014)

1 <category>

2 <pattern>*</pattern>

3 <template> What is your name? </template>

4 </category>

5 <category>

6 <pattern>my name is * </pattern>

7 <template>

8 <star index="1" /> is a beautiful name.

9 </template>

10 </category>

11 <category>

12 <pattern>i am called * </pattern>

13 <template>

14 <srai>my name is <star/></srai>

15 </template>

16 </category>

17 <category>

18 <pattern>one calls me * </pattern>

19 <template>

20 <srai>my name is <star/></srai>

21 </template>

22 </category>

5.5 Enhancing AIML

It is necessary to implement four new features to add

the functionality of the project requirements:

1. data structures to store individual data of users

(IPM – individual personal memory),

(Balandin et.al., 2014)

2. multi-client properties,

3. timeout functions to recover from breaks / start

a dialog, (Balandin et.al., 2014),

4. an editor tool for adding new phrases and new

personal data by educators.

Below those enhancements are explained in

more detail:

1. Individual Memory - IPM

When talking to people with dementia or to

children, it is essential to “know” details about

the life and the environment of the person to

whom one is talking. We added such a basic

memory, but still have to refine it (the data

structure and the bot program processing it).

AIML 2.0 allows defining new tags. To use

IPM, we added 4 new tags to write new

CSEDU2015-7thInternationalConferenceonComputerSupportedEducation

362

individual information into IPM, or to read

from IPM. We introduced one tag to write

information, and two tags, which read

information from the IPM: one tag leads to

reading randomly, the other selectively.

Another tag checks, whether a information

selected is available.

2. Multi-client properties

We have decided to implement in a first step a

central architecture based on a server. So we

are able to learn about the system’s deficits,

and the needs to improve the system. To gain

flexibility, we will implement multi-client

properties, to serve more than one client

simultaneously.

3. Timeout

Bot and database are modified, to be able to

start a dialog, or to continue talking, even if the

human does not answer.

4. Editor tools

There are tools available to edit AIML, e.g.

GaitoBot AIML Editor. (GaitoBot, 2015). We

have to investigate those tools and select one,

which is suitable to be used by educators. In

addition, we need editor functions to edit the

individual personal memory.

If none of the available editors is appropriate,

we will write an Habláme specific editor.

6 RESULTS

We have implemented a prototype which is able to

talk within a limited domain to an adult person. So

the text to speech component, speech to text, the

server, enhanced AIML, the timeout function, and

the proper function of the database can be

demonstrated.

Three major tasks have yet to be accomplished:

1. Complete the system and add

- multi-client properties, and an

- editor for educators.

2. Next step is to complete the AIML patterns so

that children are motivated to talk to the system

(current status: project already started).

3. Test the system together with educators with

healthy children first, and then with children

with atypical development with respect to its

educational benefits (current status: project will

start in summer)

ACKNOWLEDGEMENTS

This work had been partially funded by the EU

Project GAVIOTA (DCI-ALA/19.09.01/10/21526/

245-654/ALFA 111(2010)149).

REFERENCES

Balandin, R., Hegwein, M., Barthel, A., 2014.

Erweiterung und Analyse eines Chatbot. Project

Report, HAW Würzburg-Schweinfurt (FHWS).

Bossavit, B., Ostiz, M., Lizaso, L., Pina, A., Grötsch, E.,

2014. Software with Augmented Reality, Virtual

Reality, and Natural Interaction for Atypical

Development Individuals / Software con realidad

avanzada, realidad virtual e interacción natural para

personas de desarrollo no típico. In: Fernández, M.I.

(Ed.) Augmented Virtual Realities for Social

Development – Experiences between Europe and Latin

America / Realidades Virtuales aumentadas para el

desarrollo social – Experiencias entre Europa y

Latinoamérica, pages 103 – 129, Universidad de

Belgrano, Buenos Aires, 2014, ISBN 978-950-757-

047-6.

ELIZA, 2013.

www.med-ai.com/models/eliza.html (Jan 7, 2015).

GaitoBot, 2015.

www.gaitobot.de/gaitobot/ (March 25, 2015).

Grötsch E., Pina A., Schneider M., Willoweit B., 2013.

“Artificial Communication“ - Can Computer

Generated Speech Improve Communication of

Autistic Children?. In Proceedings of the 5th

International Conference on Computer Supported

Education, Aachen, pages 517-521

DOI: 10.5220/0004412805170521.

JEOPARDY, 2011.

http://www.nytimes.com/2011/02/17/science/17jeopar

dy-watson.html?_r=0, (Jan 6, 2015).

Schmitt, S., 2014. Ein plattformunabhängiges

Dialogsystem für gesprochene Sprache, Bachelor

thesis, HAW Würzburg-Schweinfurt (FHWS).

UIMA, 2013.

http://uima.apache.org/ (Jan 6, 15).

Wallace, R., 2014. The Anatomy of A.L.I.C.E.

http://www.alicebot.org/documents/Anatomy.doc.

(Jan. 6, 15).

Weizenbaum, J., 1966. ELIZA - A Computer Program for

the Study of Natural Language, Communication

between Man and Machine. Communications of the

ACM. New York 9.1966,1. ISSN 0001-0782.

AnEducationalTalkingToybasedonanEnhancedChatbot

363