LEARNING CHINESE HANDWRITING WITH A HAPTIC

INTERFACE IN DIFFERENT VELOCITY MODE

Min Xiong

1,2,3

, Isabelle Milleville-Pennel

, Cédric Dumas

and Richard Palluel-Germain

Robotique, L'Institut de Recherche en Communications et Cybernétique de Nantes (IRCCyN), France

Psychologie, Cognition, Technologie, L'Institut de Recherche en Communications

et Cybernétique de Nantes (IRCCyN), France

School of Software Engineering, Chongqing University, Chongqing, P.R.China

Laboratoire de Psychologie et Neurocognition CNRS UMR 5105, Université Pierre Mendes France, Grenoble, France

Keywords: Haptic Interface, Human Computer Interaction, Learning Method.

Abstract: This paper presents an experiment by using haptic interface for Chinese handwriting learning. Based on the

strategy of “record-and-play”, this haptic interface records the teacher’s information and transfers the

writing skill to users. There are two kinds of transfer methods. One is using the real speed that recorded

from the teacher, called variable velocity mode. The other is using a constant speed which is re-programmed,

called constant velocity mode. The objective of this experiment is to determine what kind of velocity mode

benefits the handwriting learning most. Also, another purpose is to see if use of haptic device in learning

a given Chinese character could influence learning of other characters with common strokes. The result

shows that haptic device does benefit handwriting learning. In order to improve shape or decrease inair

time, c-v mode (constant velocity mode first and variable velocity mode second) shows statistical

significance and increases performance; separately, constant velocity mode gets better improvement than

variable velocity mode with haptic learning. For writing velocity or size, no significant effect can be made.

Using haptic device to learn a given Chinese character nearly cannot influence learning of other characters

with common strokes.

1 INTRODUCTION

Generally, haptic interface guidance has been widely

used for handwriting learning for a long time.

Numerous studies have been made in order to

evaluate the advantage and find a good way via

virtual environment for this skill training. No matter

in hieroglyphic writing or phonetic writing,

researchers has proved that using haptic device

benefited handwriting learning.

In 1996, Y. Yokokohji et al. (1996) investigated

a possibility of skill mapping from human to human

via a visual/haptic display system, based on a

strategy name “record-and-play”. Although the

chosen task was too easy, no remarkable result was

obtained, this strategy gave a good idea. Later in

1998, Kazuyuki Henmi and Tsushikawa (1998) used

this strategy on training calligraphy. After using this

system, student's trajectories resembled more and

more to those of the teacher's. So they judged that

there was some positive effect of Japanese

handwriting learning by using the haptic system.

However, on this experiment, they did not give a full

evaluation and statistical analysis on the training

result. Solis et al. (2002) built a similar skill transfer

system based on the same strategy, besides they

increased the flexibility ratio of users, added a real-

time capability of understanding the user movements,

changed their behaviors as dynamic response to user

inputs. This system is so called a bi-directional skill

transfer system which can guide users dynamically.

Meanwhile, Teo et al. (2002) used a 6-DOF haptic to

develop a Chinese handwriting teaching system.

They detailed the learning process into motion

guidance and path guidance, and quantified the

performance assessment included shape, motion,

force and smoothness. Bluteau et al. (2008) made a

further research on the effects of two types of haptic

guidance-control in position (HGP) and in force

(HGF), also on the basis of “play-and-record”. A

statistical analysis was made to evaluate the

difference of number of velocity peaks, mean

Xiong M., Milleville-Pennel I., Dumas C. and Palluel-Germain R. (2010).

LEARNING CHINESE HANDWRITING WITH A HAPTIC INTERFACE IN DIFFERENT VELOCITY MODE.

In Proceedings of the 2nd International Conference on Computer Supported Education, pages 39-45

DOI: 10.5220/0002778900390045

 SciTePress

velocity, and shape matching score before and after

haptic training. Result showed that HGF improved

performances whereas HGP and NHG (non haptic

guidance) showed no significant improvement.

From all these researches above, the strategy of

“record-and-play” has been applied extensively. All

of them lay particular emphasis on either technique,

or the comparison between difference learning

processes. These learning processes were divided by

different haptic-guidance mode, based on motion,

path, position, and force. The quantitative indexes of

performance assessment such as shape, motion,

force and time were just used as evaluation criteria.

Actually, all these quantitative indexes can be

embodied in the learning process, which can vary

numerically with a high level of precision. If we take

these indexes as variables in haptic training process,

we can study different effects on learning by

changing them. Therefore, we can find a good

learning method specifically with the use of haptic

device.

In the training ways of using haptic guidance,

according to the interaction between the user and

haptic device, two general training methods can be

divided: (Wu et al, 2007)

Passive mode: users use haptic device passively,

the haptic device guides users to move under a pre-

designed velocity, force, and path.

Active mode: users use haptic device on their

own initiative and practive some movement, only

when they deviate from the usual route, the haptic

device will output a corrective force and compel

them to go back.

In the two modes, the first one passive mode is

normally used for beginners, and the second one,

active mode, which gives major autonomy to users,

would be better applied for intermediate or higher

level.

In our experiment, our participants were all

beginners to Chinese handwriting. Therefore, in the

main, we used the passive mode. Specifically, we

chose the velocity as variable, based on a haptic

guidance control in position mode (Bluteau et al,

2008), and studied the influence on learning Chinese

handwriting from changing different velocity mode.

Two different velocities were designed in this haptic

guidance, variable and constant velocity modes.

After using this haptic device in different velocity,

the results are evaluated to see which one benefit the

handwriting learning better. Also, another purpose is

to see if using haptic device to learn a Chinese

character will influence the other characters that

have the common strokes.

2 METHOD

This experiment has three main parts: pre-test,

haptic-training, and post-test. During the pre-test and

post-test, a tablet (wacom) is used, all the writing

data on the tablet are recorded. During the haptic-

training, a haptic device (phantom omni) is used to

help teach participants how to write Chinese

characters. This haptic device moves in two different

speed mode based on predefined program. The first

mode is called c-v mode. In this mode, a constant

velocity is first used, and then a variable velocity is

secondly used. Another mode is named v-c mode. In

this mode, a revise order of velocity mode is used,

that is, a variable velocity at first and a constant

velocity second. By analyzing and evaluating the

different data collected from pre-test and post-test,

we can study the influence factors of using haptic

device for handwriting learning and achieve our

purpose.

2.1 Participants

Seventeen adults between the ages of 20 and 44

years old participated in this study. All of them came

from Ecole des Mines de Nantes, France. They were

divided into two groups (8 participants and 9

participants in each group) in this experiment. One

group was constant-variable-tested group (c-v

group), and the other group was variable-constant--

tested group (v-c group). All these Participants had

never learnt Chinese writing before.

2.2 Experimental Setup

The experimental setup included a tablet (wacom) to

collect the writing data from participants, a

computer screen for showing traces, a haptic arm (a

phantom omni with six degrees of freedom which

can move in different speed based on predefined

program.) to teach the writing movement of

participants, and another computer screen for

simulating the paper sheet. (See figure. 1) 3 basic

Chinese characters were used in this experiment: 歹

(dai), 反(fan), 瓦(wa).

2.3 Procedure

Mainly, this experiment contained three parts: pre-

test, haptic-training, and post-test (post-test 1 and

post test 2). The schematic view of this experiment

can be shown in figure 1.

The first step was pre-test. During the pre-test,

participants were asked to write freely on a digital

tablet (Wacom) and to try their best to write. The

order of strokes in each character was not given to

CSEDU 2010 - 2nd International Conference on Computer Supported Education

participants. Every time when the participant was

writing, a model of character was shown on the side.

By observing the standard model, participant tried to

write the same character on the tablet freely. Totally,

there were three Chinese characters to write, and

each character should be written for 3 times. These

three characters were tested and verified to be proper

for beginners. They were neither too hard nor too

easy to learn. The criterions included the character

shape, the speed and the time of writing. During the

process of pre-test, the positions of the pen and the

time of writing each character were recorded.

Figure 1: Schematic view of this experiment.

The second step was Haptic-training for the two

groups. Participants were asked to write passively

along with a haptic arm on a horizontal screen. It is

note that the pen (haptic arm) nearly touched the

video screen. One group was asked to choose the c-v

mode and the other was asked to choose v-c mode.

The haptic arm moved under a programmed

trajectory and moved either in constant speed or in

variable speed according to the chosen mode.

Therefore, the participants’ hand moved along with

the haptic arm in two ways. During this part, only

the first character (歹(dai)) was used. Compare to

the other two ones, the character (歹(dai)) has the

most common strokes. It is can be used to assay

whether the haptic interface can help to learn writing

different Chinese characters when the character

patterns are similar. After writing this character for

20 times, the participant could go on to do the post-

test part.

In the third part of post-test, the whole procedure

was the same as in the pre-test. The participants

were asked to write three Chinese characters freely

on the same digital tablet. Then, they were asked to

do haptic-training again but chose the reverse

velocity mode. Therefore, the participants did the

haptic-training and post-test again with different

velocity mode.

After one week, the same group did the same post-

test again for memory checking. Finally, we can

receive all the writing data from both before and

after haptic-training. Hence, the data can be used for

analysis and evaluation.

3 RESULT

In both two groups of c-v and v-c, for each

parameter, a paired samples T-Test was performed

on the periods of pre-test and after-first-test (1vs2);

pre-test and after-second-test (1vs3); pre-test and

after-one-week-test (1vs4); after-first-test and after-

second-test (2vs3); after-second-test and after-one-

week-test (3vs4)); after-first-test and after-one-

week-test (2vs4). For each parameter and period of

test, an independent samples T-Test was performed

on the c-v group and v-c group. For each analysis, a

significance level of 0.05 was chosen.

By using the formula of paired samples t-test, the

result of t and S

can be calculated, and then p can be

observed. If p>0.05, we can say that there are no

differences statistically significant between the two

samples.

In this test, in c-v group, we have 9 participants;

in v-c group, we have 8 participants. Therefore, by

using formula of independent samples T-Test, n

=9,

=8. If │t┃< t

0.05, 15

=2.131, there is no significant

difference between the two groups.

3.1 Shape

We evaluated the shapes by analyzing all strokes in

each character. The perfect score of each character is

defined as 1. The closer to 1, the better score of

writing shape is.

By using independent samples T-Test, compare

the writing of “dai” between the c-v group and v-c

group, we get the result that, at the time of pre-test

(condition 1), │t┃=0.276328545< t

0.05, 15

=2.131,

there is no significant difference between c-v group

and v-c group. Therefore, we can consider that the

two groups are the same in condition 1; they are the

samples from the same population.

Then by using paired t-test function in Excel,

tables can be shown

Table 1: Paired t-test result of “dai” in c-v group.

test

vs2

vs3

vs4

vs3

vs4

dai 0

.312

.012

374

.873

495

.059

838

.076

089

.059

LEARNING CHINESE HANDWRITING WITH A HAPTIC INTERFACE IN DIFFERENT VELOCITY MODE

Table 2: Paired t-test result of “dai” in v-c group.

test

vs2

vs3

vs4

vs3

vs4

dai 0

.364

505

.278

837

.291

005

.801

831

.884

975

.895

In the two tables, all the values below 0.05 are

coloured in red, which indicate the significant

difference between two paired samples.

From the two tables above, comparing condition

1 and condition 3, the c-v group has a distinct

increase, which has statistical significances in the

conditions (p=0.012374<0.05). While at the same

situation, there is no significant difference between

condition 1 and condition 3 in v-c group. Therefore,

after second-test, we can say that the c-v group

increases more performance than v-c group. In pre-

test, the average score of shape was 0.9 and

0.915625 for the c-v group and v-c group. After

second-test, the average score was 0.947222 for the

c-v group and 0.93125 for the v-c group.

In the same way, by using independent samples

T-Test, compare the writing of “fan” between the c-

v group and v-c group, we get the result that, at the

time of pre-test (condition 1), │t┃= 2.32474> t

0.05,

=2.131, there is a significant difference between c-

v group and v-c group. That is to say, the starting

values of two groups are quite different. Therefore,

we do not compare these conditions anymore;

instead, we compare the difference between every

two conditions. Specifically, we compare

condition2-condition1 (c2-c1), condition3-

condition1 (c3-c1), condition4-condition1 (c4-c1),

condition3-condition2 (c3-c2), condition4-

condition3 (c4-c3) in the two groups. The two

groups have been compared on the base of the

difference between every two conditions.

Table 3: Paired t-test result of “fan” in c-v group.

c2-

c3-

c4-

c3-

c4-

fan 0

.529

841

.611

441

.586

308

.967

912

1.72

276

Because from all the t values in the table, │t┃< t

0.05,

=2.131. Therefore, we can say that there are no

significant differences of between-conditions in c-v

group and v-c group. In another word, no matter

what kind of training is made, there is no significant

effect on shape of writing “fan”.

Then, by comparing the two groups of writing “wa”,

same result is gotten. No matter what kind of

training is made, there is no significant effect on

shape of writing “wa”.

To sum up, when the participant writes “dai”,

considering the shape of characters, the first constant

then variable velocity training mode is better than

the reverse mode. For the other two characters, by

using the haptic device, no significant effect has

been made.

3.2 Velocity

For all the three characters: “dai”, “fan”, “wa” (歹，

反，瓦), at pre-test, by using independent samples

T-Test to compare the two groups, the t= -0.90729, -

1.83878, -1.9627. All the |t| are smaller than t

0.05,

=2.131. There are no significant differences

between the group c-v and group v-c at the

beginning, no matter which character is written.

Therefore, we can consider that the writing velocity

of two groups is the same in condition 1; they are the

samples from the same population. Then in the test

conditions, all these conditions were tested paired.

We can find out that all the p value are bigger than

0.05. That is to say, there is no significant difference

between any two conditions.

To sum up, by testing velocity, no statistical

significant result can be found. In another word, the

two velocity modes haptic training methods have no

significant effect on changing velocity of

handwriting.

3.3 Size

For all the three characters: “dai”, “fan”, “wa” (歹，

反，瓦), at pre-test, by using independent samples

T-Test to compare the two groups, the t= 0.704133, -

0.28657, 0.599112; all |t|< t

0.05, 15

=2.131. There are

no significant differences between the group c-v and

group v-c at the beginning, no matter which

character is written. Therefore, we can consider that

the writing size of two groups is the same in

condition 1; they are the samples from the same

population. Then in the test conditions, all these

conditions were tested paired. All these p values are

bigger than 0.05. That is to say, there is no

significant difference between any two conditions.

To sum up, by testing size, no statistical

significant result can be found. In another word, the

two velocity modes haptic training methods have no

significant effect on changing size of handwriting.

3.4 Order

Every Chinese character has many strokes, only

when all the strokes are written in right order, we

can say that the order of the character is good.

CSEDU 2010 - 2nd International Conference on Computer Supported Education

Therefore, we define the good order as 100%, the

score of order is equal to n

right strokes

all

*100% (n

right

strokes

=number of all strokes in one character, n

all

=the

whole number of strokes in one character).

For the training of writing “dai”:

Table 4: Paired t-test result of “dai” in c-v group.

test

1vs

3vs

2vs

dai 0

.103

- - -

Table 5: Paired t-test result of “dai” in v-c group.

test

1vs

3vs

2vs

.103

- - -

All the p>0.05, it seems that there is no significant

difference between every two conditions. However,

if we have a look at the original data, it is clear that,

at the very beginning, there were just a little

participants wrote “dai” in wrong order, but no

matter what order was at pre-test, after the first

training, it turns to 100%, which is the good order.

After, it stays the same. We can tell that, the first-test

is good. However, we cannot find if the following

tests are even better because all the data stays the

same as 100%.

For the training of “fan”:

At pre-test, by using independent samples T-Test

to compare the two groups, the t= -0.66323, |t|=

0.66323< t

0.05, 15

=2.131. There is no significant

difference between the group c-v and group v-c at

the beginning. Therefore, we can consider that the

writing order of two groups is the same in condition

1; they are the samples from the same population.

Then in the test conditions, all these conditions were

tested paired. We can find out that all the p value are

bigger than 0.05. That is to say, there is no

significant difference between any two conditions.

In another word, the two velocity modes haptic

training methods have no effect on improving the

writing order of “fan”.

For the training of “wa”, the exactly same test

was done, the result is the same: the two velocity

modes haptic training methods have no effect on

improving the writing order of “wa”.

To sum up, the training has good effect on

improving the order of “dai”, but no effect on the

other two characters.

3.5 Inair Time

By using independent samples T-Test, compare the

writing of “dai” between the c-v group and v-c

group, we get the result that, at the time of pre-test

(condition 1), │t┃=1.063509< t

0.05, 15

=2.131, there

is no significant difference between c-v group and v-

c group. Therefore, we can consider that the two

groups are the same in condition 1; they are the

samples from the same population.

Then by using paired t-test function in Excel,

tables can be shown:

able 6: Paired t-test result of “dai” in c-v group.

test

1vs

2vs

3vs

2vs

dai 0

.052

.037

.088

.709

.872

.982

Table 7: Paired t-test result of “dai” in v-c group.

test

1vs

2vs

3vs

2vs

dai 0

.410

.164

.093

.036

.643

.098

By using paired t-test in each group, compare

condition 1 and condition 2, the c-v group almost

has a significant increase, with p=0.052916

approach to 0.05. While at the same situation, there

is no significant difference between condition 1 and

condition 3 in v-c group. Therefore, after first-test,

we can say that constant velocity mode gets

significant improvement. In pre-test, the average

inair time was 3.35 and 2.41 for the c-v group and v-

c group. After second-test, the average score was

1.95 for the c-v group and 1.95 for the v-c group.

Compare condition 1 and condition 3, the c-v group

has a distinct increase, which has statistical

significances in the conditions (p=0.037507<0.05).

While at the same situation, there is no significant

difference between condition 1 and condition 3 in v-

c group. Therefore, after-second-test, we can say

that the c-v group gets significant improvement.. In

pre-test, the average inair time was 3.35037 and

2.414167 for the c-v group and v-c group. After

second-test, the average score was 1.905 for the c-v

LEARNING CHINESE HANDWRITING WITH A HAPTIC INTERFACE IN DIFFERENT VELOCITY MODE

group and 1.635208333 for the v-c group. Again, by

using paired t-test in each group, compare condition

2 and condition 3, the v-c group has a distinct

increase, which has statistical significances in the

conditions (p=0.036594<0.05). While at the same

situation, there is no significant difference between

condition 2 and condition 3 in c-v group.

Considering the condition 2, │t┃=0.00089< t

0.05,

=2.131, there is no significant difference between

c-v group and v-c group at that time. In condition 2,

the inair times of two groups are almost the same.

Therefore, from condition 2 to condition 3, we can

say that the v-c group gets better results, compared

to c-v group. After second-test, the average inair

time was 1.949074074 and 1.949375 for the c-v

group and v-c group. After second-test, the average

score was 1.905 for the c-v group and 1.635208333

for the v-c group. Specifically, since in condition 2,

the c-v group and v-c group are almost the same,

from condition 2 to condition 3, the c-v group chose

variable velocity mode, while at the same time, the

v-c group chose constant velocity mode, therefore,

the difference shows that constant velocity mode

increased more performance than variable velocity

mode in this situation.

In the same way, by using independent samples

T-Test, compare the writing of “fan” between the c-

v group and v-c group, we get the result that, at the

time of pre-test (condition 1), │t┃=1.412317<t

0.05,

=2.131, there is no significant difference between

c-v group and v-c group. Therefore, we can consider

that the two groups are the same in condition 1; they

are the samples from the same population.

By using paired t-test function in Excel, tables

can be shown

Table 8: Paired t-test result of “fan” in c-v group.

test

c-v 1vs

1vs

2vs

3vs

2vs

fan 0

.734

949

.123

727

.117

448

.260

535

.429

.286

Table 9: Paired t-test result of “fan” in v-c group.

test

v-c 1vs

1vs

2vs

3vs

2vs

fan 0

.013

806

.152

893

.194

225

.475

.624

437

.352

And then, by using paired t-test in each group,

compare condition 1 and condition 2, the v-c group

has a distinct increase, which has statistical

significances in the conditions (p=0.013806<0.05).

While at the same situation, there is no significant

difference between condition 1 and condition 2 in c-

v group. Therefore, after first-test, we can say that

the v-c group gets significant improvement,

compared to c-v group. That is to say, the variable

velocity mode training increases performance in this

situation for reducing inair time. In pre-test, the

average inair time was 3.33 and 1.66 for the c-v

group and v-c group. After first-test, the average

score was 2.77 for the c-v group and 1.12 for the v-c

group.

For the three character of “wa”, at pre-test, by

using independent samples T-Test, t= 1.462937,

|t|1.462937< t

0.05, 15

=2.131. There is no significant

difference between the group c-v and group v-c at

the beginning. Therefore, we can consider that the

inair time of writing in the two groups is the same in

condition 1; they are the samples from the same

population. Then in the test conditions, all these

conditions were tested paired. All these p values can

be seen in the following tables.

By using paired t-test function in Excel, tables

can be shown

Table 10: Paired t-test result of “wa” in c-v group.

test

1vs

2vs

3vs

2vs

wa 0

.708

.175

.298

.373

.760

.420

Table 11: Paired t-test result of “wa” in c-v group.

test

1vs

2vs

3vs

2vs

wa 0

.497

.146

.623

.216

.860

.738

From the two tables above, we can find out that all

the p value are bigger than 0.05. That is to say, there

is no significant difference between any two

conditions. In another word, the two velocity modes

haptic training methods have no significant effect on

reducing inair time of handwriting “wa”.

CSEDU 2010 - 2nd International Conference on Computer Supported Education

4 CONCLUSIONS

Our purpose of this study is to evaluate the

advantage of using haptic device on learning, to

determine what kind of velocity mode benefits the

handwriting learning, and to see if use of haptic

device in learning a given Chinese character could

influence learning of other characters with common

strokes. Therefore, a good Chinese handwriting

learning method with haptic interface may be found.

In order to answer these questions, a haptic interface

including a haptic device (phantom omni) and tablet

was used to teach beginners to write. By comparing

all the five parameters in section 3 and three Chinese

characters, we can get the result that, when the

haptic device is used to learning a specific Chinese

character, in order to improve shape or decrease

inair time, c-v mode shows statistical significance

and increases performance; separately, constant

velocity mode gets better improvement than variable

velocity mode with haptic learning. For writing

velocity or size, no significant effect can be made.

Using haptic device to learn a Chinese character

writing nearly cannot influence the other characters

that have the common strokes. Only after variable

velocity mode training, the inair time of writing

similar characters may reduce.

This experiment is using visuo-haptic interface

under different velocity mode. The test of non-visual

will be evaluated in future experiment.

ACKNOWLEDGEMENTS

We would like to thank Jeremy Bluteau and Richard

Palluel-Germain (Laboratoire de Psychologie et

Neurocognition, CNRS UMR 5105, Grenoble) for

providing the original program.

REFERENCES

Bluteau, J., Coquillart, S., Payan, Y. & Gentaz, E., 2008.

Haptic Guidance Improves the Visuo-Manual

Tracking of Trajectories.,.

CL Teo, E.B.L., 2002. A Robotic Teacher of Chinese

Handwriting. In Haptic Interfaces for Virtual

Environment and Teleoperator Systems.

FLANDERS Martha, M.L.A.G.C., 2006. Planning and

Drawing complex shapes.: Springer-verlag. pp.116-28.

Henmi, K. & Yoshikawa, T., 1998. Virtual lesson and its

application to virtual calligraphy system. In Robotics

and Automation. IEEE.

Palluel-Germain, R. et al., 2007. A Visuo-Haptic Device -

Telemaque - Increases Kindergarten Children's

Handwriting Acquisition.,.

Sara Kiesler, P.H., 2004. Introduction to the Special Issue

on Human–Robot Interaction. Special Issue of Human-

Computer Interaction,.

Solis, J., Avizzano, C.A. & Bergamasco, M., 2002.

Teaching to write Japanese characters using a haptic

interface. Haptic Interfaces for Virtual Environment

and Teleoperator Systems,. pp.255-62.

Srimathveeravalli, G. & Thenkurussi, K., 2005. Motor

Skill Training Assistance Using Haptic Atributes.

Haptic Interfacees for Virtual Environment and

Teleoperator Systems pp.452-57.

Wang, D.-x., Zhang, Y.-r. & Yao, C., 2006. Experimental

evaluation of haptic enabled interactive motor skill

training system. Journal of System Simulation.

Wu, J., Wang, D.-x. & Zhang, Y.-r., 2007. A calligraphy

motor skill training system based on record-play

method. Journal of System Simulation, 19.

Yokokohji, Y. et al., 1996. Toward Machine Mediated

Training of Motor Skills-Skill Transfer from Human

to Human via Virtual Environment. In IEEE

International Workshop on Robot and Human

Communication. IEEE.

LEARNING CHINESE HANDWRITING WITH A HAPTIC INTERFACE IN DIFFERENT VELOCITY MODE