A SERIOUS GAME FOR SECOND LANGUAGE ACQUISITION

Marilisa Amoia, Claire Gardent

INRIA Nancy Grand Est, 54500 Vandoeuvre-les-Nancy, France

Laura Perez-Beltrachini

INRIA Nancy Grand Est, 54500 Vandoeuvre-les-Nancy, France

Keywords:

Virtual learning environments, Computer assisted learning, Intelligent tutoring systems, Immersive learning.

Abstract:

This paper describes an interactive learning system speciﬁcally designed for second language acquisition. In

order to render the learning experience more fun, to engage the learner and to help him maintaining long-term

motivation, the system was implemented as a 3D video game. It brings together the ability of virtual reality

environments such as Second Life to reproduce immersive experiences and NLP language technology, thereby

providing both situated learning and automatic authoring of training activities in context.

1 INTRODUCTION

Interactive or situated learning has been acknowl-

edged as one of the main pedagogical principle in

second language acquisition. Indeed, many observa-

tions substantiate the role of immersive environments

in facilitating learning. It has been noticed (Krashen,

1997), for instance, that adult learners acquire a sec-

ond language more easily and their knowledge is

more anchored, if they are exposed during learning

to situations similar to real life, like those that chil-

dren experience by acquiring the ﬁrst language and

further, that (Rutherford, 1987) raising the awareness

of the learner on the phenomena of the target language

in context, i.e. by noticing or highlighting them in a

particular situation, fosters learning.

In the last few years, there has been an increas-

ing interest in the e-learning research community for

gaming and simulation technology as they allow close

reproductions of immersive experiences. The ﬁrst ex-

periments in language teaching go back to the 1990’s

when the ﬁrst computer-aided learning software was

produced on CD-ROM. Escape From Planet Arizona

and Who is Oscar Lake

for instance, are examples of

those early language games created for learning En-

Escape from Planet Arizona: An EF Multimedia Lan-

guage Game. [Software]. (1995). Stockholm, Sweden: EF

Education.

Who is Oscar Lake: A Language Learning Adventure

Game. (1995). Jersey COW Software.

glish as a second language. These language games

represent ﬁrst attempts to integrate traditional learn-

ing content, i.e. vocabulary, grammar exercises, etc.,

in a situational context (an adventure story), thus pro-

ducing in the learner an impression of immersive ex-

perience.

In the last decade, some research on computer

aided second language acquisition has focused on us-

ing online 3D virtual reality environments and video

game technology for teaching languages. This type

of environment further promotes learning as a social

experience allowing learners to practice active com-

munication over the web by means of chatting, emails

etc. Thethis (Segond et al., 2005) , for instance, im-

plements a web application providing a learning soft-

ware for the language training of hotel receptionists.

The learner is exposed to similar situations as if he

were in the reception of an hotel. He can interact both

with virtual agents simulating telephone calls, hotel

guests arriving, etc. by means of preset dialogues or

with a human tutor or other learners by means of chat-

ting. The main innovative aspect of Thethis is the so-

cial, communicative aspect of the learning platform

which allows learners to share the learning experience

with fellow students and tutors.

More recently, the 3D video game technology has

been used in so called culturally-aware tutorial sys-

tems such as ATL (Raybourn et al., 2005), TLCTS

(Johnson and Valente, 2009) and BiLAT (Kim et al.,

2009) to train social cultural skills in a military sce-

394

Amoia M., Gardent C. and Perez-Beltrachini L..

A SERIOUS GAME FOR SECOND LANGUAGE ACQUISITION.

DOI: 10.5220/0003338503940397

In Proceedings of the 3rd International Conference on Computer Supported Education (CSEDU-2011), pages 394-397

ISBN: 978-989-8425-49-2

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

nario. The learner acts as an avatar in the simulated

environment and must provide natural language ut-

terances and in some systems even gestures thereby

learning not only grammar but also cultural skills of

the target language/society.

In these systems however, the virtual world is used

mainly as a mean to immerse the learner in a simula-

tion of the societal and cultural world of the L2 lan-

guage. The linguistic sophistication of the exercises

remains limited covering, e.g. preset dialogs and sim-

ple language exercises such as vocabulary training or

drills. Moreover, these exercises are hard coded and

all the systems described above rely on human author-

ing for the learning content. Recently, some work

has aimed at automatizing the generation of learn-

ing content and learning activities. Indeed, teachers

very often lament the high expense on time to pro-

duce different learning activities. Examples of works

in this direction are for instance, VISL (Bick, 2005),

a visual interactive syntax learning tool accessible

though the internet for learning the syntax of different

languages and TAGARELLA (Amaral and Meurers,

2007), an intelligent web-based workbook for learn-

ing Portuguese and more recently WERTI

, a proto-

type of a system for the automatic generating of exer-

cises (Metcalf and Meurers, 2006) based on arbitrary

web content selected by the learner.

I-FLEG, the language game presented in this pa-

per, integrates both these research approaches and

provides a situated language learning environment to-

gether with automatic generation of learning activities

in context.

This paper is structured as follows. Section 2 de-

scribes the architecture of the language game I-FLEG

and illustrates how learning activities are automati-

cally generated in the context of the game. Section

3 concludes describing the preliminary evaluation of

the system and discussing pointers for future research.

2 I-FLEG

I-FLEG is a prototype interactive 3D game for sec-

ond language learning. To ensure portability to differ-

ent platforms, the system is implemented in Java and

uses Second Life

as a graphical interface. The sys-

tem consists of a reasoning module that implements

the game logic linked to a natural language genera-

tion module and to a database.

I-FLEG is a sort of an adventure game. Its goal

is to teach vocabulary (on some speciﬁc topics such

http://prospero.ling.ohio-state.edu/WERTi/

http://secondlife.com/

as house, food, etc.) and some grammar features

(e.g. prepositions, adjective morphology) to learners

of a second language. The current implementation is

addressed to a target audience including learners of

French at A1-A2 levels, i.e. beginners and interme-

diate level learners. However, the modular architec-

ture of the system allows to easily set it to another

language. At present, the system also includes an En-

glish grammar and lexicon that can be used for train-

ing.

Following pedagogical approaches (Uhl-Chamot

and O’Malley, 2009) that emphasized the importance

of learner awareness of the acquired language skills

during the learning process and the role of indepen-

dent learning, in the game framework we adopt a free

learning ﬂow strategy and let the learner free to ex-

plore learning contents and to organize the learning

process meeting his own individual needs.

2.1 Game Scenario

The game scenario is a house containing different

rooms such as a living room, a kitchen, a library, etc.

A ﬁrst person perspective is used. The learner is an

avatar that can freely move in the game world. He

can interact with the physical objects of the virtual

world by touching, moving or taking them. By doing

so he learns their names and characteristics or triggers

learning activities. The player communicates with the

system by typing text in a chat box.

The game aspect is represented by the challenge

the player has to master. Somewhere in the house,

there are strange objects hidden, that are not part of

the furniture. These objects when touched trigger

learning activities, i.e. exercises. The player has a

limited amount of time to ﬁnd out where they are,

touch them and solve the learning activities linked to

them. By solving exercises the player earns credits

points (the score). The game consists of different lev-

els. Each level represents a language proﬁciency level

in the second language (e.g. A1) and includes a set

of training activities covering several teaching goals

ranging over different grammar topics, e.i. vocabu-

lary, syntax, morphology, etc. A level is complete

when the learner has totalized the score that deﬁne

the accomplishment of that language level.

2.2 Learner Model

The system monitors the user during the whole game

session and maintains a learner model. After each

game session, the interactions of the user with the

system are stored in a database. The content of the

database is retrieved each time the same user logs in

A SERIOUS GAME FOR SECOND LANGUAGE ACQUISITION

395

Figure 1: Game scores examples from the demo session.

Player Vocabulary Score Syntax Score Total Score Start Time End Time

AvatarT2 9 4 13 11:59:22 12:12:11

AvatarT4 3 2 5 11:40:35 11:53:45

AvatarT3 6 1 7 10:49:48 11:12:45

AvatarT1 14 8 22 11:12:01 11:36:01

the game. Before a new game starts, the results of

the previous game are evaluated so that the level and

type of new interactions depend on the evaluation of

previous results.

In the prototype, a simpliﬁed model of learning

performance is used based on scoring. The score is

a structured datatype mapping learning performance

to the credit points the learner has totalized in each

training activity and to the time needed for the game

session. Scoring is used both to evaluate the learning

progress of the learner and to give him feedback. The

total score is shown to the learner during the game as

an immediate simpliﬁed form of feedback. Whereas

at the end of the game, a more structured output is

shown including the results achieved in each grammar

ﬁeld that has been trained.

2.3 Generation of Test Activities

I-FLEG supports the automatic authoring of test ac-

tivities. As mentioned in Section 2.1, exercises are

triggered by the learner interactions with the world.

Whenever the learner touches a teaching object, an

exercise is produced whose precise content depends

on the learner proﬁle (his level, his game score for

each activity type) and on the teaching goals being

pursued (e.g., teaching the passive form, the use of cl-

itics or adjectival morphology). Importantly, the sys-

tem is non-deterministic thereby supporting the pro-

duction of varied output. That is, the same touching

event might trigger different exercises.

This automation of learning material relies on a

generation module which combines constraint driven

content selection and surface realization. Starting

from an ontology describing the content of the 3D

world, the generation module ﬁrst selects a set of facts

(content selection) and then turns this set of facts into

a sentence (surface realization). From this output sen-

tence, the I-FLEG system will then derive a test item

consisting of the exercise itself (e.g., a ﬁll in the blank

utterance) and its solution. More speciﬁcally, the gen-

eration of a test item is parametrized for

• Language Content: the syntactic form the output

should have, e.g. a sentence, a noun phrase, etc.

• Semantic Content: the set of individuals the

test item refers to, generally corresponds to the

touched object.

• Teaching Goal: the linguistic topic which the test

item should illustrate, e.g. adjective morphology.

• Type of Activity: the type of test item that should

be generated, e.g. vocabulary, ﬁll in the blank,

scramble sentence, etc.

The generator then uses these parameters together

with a grammar of the speciﬁed language to build all

possible syntactic realizations (i.e. paraphrases) that

satisfy the constraints above. At this point, only one

of these realizations is selected by the system and

used to instantiate the expected answer. This selec-

tion strategy ensures that repeated interactions with

the same object can generate different test items.

Next, the test query is build from the expected an-

swer on the basis of both the teaching goal and the re-

quired activity type. The test query is used as a train-

ing item to test the language knowledge of the learner.

The generated expected answer is used to evaluate the

answer of the learner and is sent as feedback by the

system in case his answer is wrong.

Figure 2 for instance, shows the parameters sent

to the generation module after the learner touched a

table (table

) together with the test item generated by

the system for an exercise of type ﬁll-in-the-blank fo-

cusing on adjective morphology.

Generation Parameters

Semantic Content = [table

]

Language Content = Sentence

Teaching Goal = [ADJ

morph

Type of Activity = FILL IN T HE BLANK

Generates Test Item

Test Query = “C’est une table (blanc) ”

Expected Answer = “C’est une table blanche”

Game Session

Please, complete the following sentence.

System: C’est une table ... (blanc).

User: C’est une table blanc.

System: Non, c’est une table blanche.

Figure 2: An example of automatic generated test item.

In the different levels of the game, language con-

tent and teaching goal are chosen to meet the lan-

guage proﬁciency level of the learner. Therefore, the

complexity of the generated test items depends on the

CSEDU 2011 - 3rd International Conference on Computer Supported Education

396

game level. For instance, after the user has touched

the same table, the system outputs (1a) if the user is a

beginner. However, if the learner is more advanced

and the teaching goal is for instance, to teach pro-

nouns the system outputs (1b).

(1) a. C’est une table.

This is a table.

b. C’est une petite table. Elle est blanche.

This is a small table. It is white.

Figure 3: An example of tutorial interaction.

3 CONCLUSIONS

This paper presented I-FLEG, the prototype of an in-

teractive language game speciﬁcally designed to fos-

ter second language acquisition. The game is inte-

grated in Second Life and is accessible for everyone

through the web. It brings together 3D graphics, vir-

tual reality and NLP technologies thereby providing

immersive, situated language learning and context-

driven automatic generation of learning activities.

We conducted a preliminary evaluation of the

game prototype during a demo session at our insti-

tution open house day. The game was set to train En-

glish and was presented to a general audience mostly

composed by French speakers. The test participants

played short game sessions including a tutorial, a vo-

cabulary and a syntax test unit. We asked them to

express their subjective opinion about the game. They

found the learning platform very fun and natural to in-

teract with. Further, most of them judged playing the

game a very engaging and entertaining experience.

Despite the limited number of test participants

(about 10 people) we believe the feedback we re-

ceived is very encouraging.

In future work, we plan to improve the language

learning system so to allow the automatic generation

of learning activities for more complex language phe-

nomena. Further, we want to formalize the process

of evaluation of the learner output so to deﬁne more

efﬁcient ways to present feedback to learners.

Finally, we envisage a large-scale web-based eval-

uation of the system as well as a comparison of the

efﬁciency of different teaching methodologies, such

as for instance free vs. prescriptive learning ﬂow.

ACKNOWLEDGEMENTS

Many thanks to the TALARIS Team at LORIA and

especially to Alexandre Denis for his collaboration in

the implementation.

The research described here was conducted as part

of ALLEGRO, an on-going project funded within the

EU Interreg IV program focusing on the development

of new technologies for foreign language learning.

REFERENCES

Amaral, L. and Meurers, D. (2007). Designing Learner

Models for Intelligent Language Tutors. CALICO

2007.

Bick, E. (2005). Grammar for Fun: IT-based Grammar

Learning with VISL. In CALL for the Nordic Lan-

guages, Copenhagen Studies in Language, pages 49–

64. Henriksen, P. Juel (ed.).

Johnson, W. L. and Valente, A. (2009). Tactical Language

and Culture Training Systems: Using AI to Teach For-

eign Languages and Cultures. AI Magazine, pages 72–

83.

Kim, J., Hill, R., Durlach, P., Lane, H. C., Forbell, E., Core,

M., Marsella, S., Pynadath, D., and Hart, J. (2009).

BiLAT: A Game-Based Environment for Practicing

Negotiation in a Cultural Context. Journal of Artiﬁ-

cial Intelligence in Education, 19(2):289–308.

Krashen, S. (1997). The monitor model for adult second

language performance. In Viewpoints on English as a

Second Language, New York: Regents. Burt, Dulay,

Finocchiaro(eds).

Metcalf, V. and Meurers, D. (2006). Generating Web-

based English Preposition Exercises from Real-World

Texts. EUROCALL 2006.

Raybourn, E., Deagle, E., Mendini, K., and Heneghan, J.

(2005). Adaptive Thinking & Leadership simulation

game training for Special Forces Ofﬁcers. In Proceed-

ings of I/ITSEC (Interservice/Industry Training, Sim-

ulation and Education Conference).

Rutherford, W. (1987). Second Language Grammar:

Learning and Teaching. Longman.

Segond, F., Parmentier, T., Stock, R., Rosner, R., and

Muela, M. U. (2005). Situational language training for

hotel receptionists. In Proceedings of the 2nd ACL-

Workshop on Building Educational Applications Us-

ing NLP, pages 85–92, Ann Arbor, Michigan.

Uhl-Chamot, A. and O’Malley, J. M. (2009). The Calla

Handbook: Implementing the Cognitive Academic

Language Learning Approach. Addison Wesley.

A SERIOUS GAME FOR SECOND LANGUAGE ACQUISITION

397