MODEL FOR PEDAGOGICAL INDEXATION OF TEXTS FOR

LANGUAGE TEACHING

Mathieu Loiseau, Georges Antoniadis and Claude Ponton

LIDILEM, Universit

e Stendhal Grenoble 3, Grenoble, France

Keywords:

Computer Assisted Language Learning (CALL), Natural Language Processing (NLP), Educational Metadata.

Abstract:

In this communication we propose to expose the main pedagogical ressource description standards limitations

for the description of raw ressources, through the scope of pedagogical indexation of texts for language teach-

ing. To do so we will resort to the testimony of language teachers reagarding their practices. We will then

propose a model supposed to exceed these limitations. This model is articulated around the notion of text

facet, which we introduce here.

1 CORPORA FOR LANGUAGE

TEACHING

Thanks to the communicative approach’s widespread

use (cf. (Levy, 1997)) authentic text is at the heart of

the teachers set of problems.

However, corpora, despite numerous, are not dedi-

cated to text search for language teaching. Querying

mecanisms mostly rely on traditional keywords

queries. Teachers display an ability to adapt com-

puter tools of which they were not meant to be

the end user, such as in Tim Johns’ Data Driven

Learning (DDL)(Higgins and Johns, 1984). All the

same, some of the ﬂaws of CALL systems mentioned

in (Antoniadis et al., 2004) accurately describe the

situation of language corpora for language teaching:

when a teacher seeks to ﬁnd a text in a corpus , there

is no system that allows him/her to express his/her

request in terms of his/her set of problems, using

pedagogical concepts.

1.1 Pedagogical Indexation for

Language Teaching

The project to create a pedagogically indexed text

base for language teaching directly stems from these

considerations. This project will lead to the imple-

mentation of a prototype (under development). It

should fulﬁll the following use cases: text query and

text addition.

Lef

evre’s deﬁnition of “documentary lan-

guage”

(Lef

evre, 2000) explicitely puts the users at

the center of the indexation process. Consequently:

Deﬁnition (Pedagogical Indexation). Pedagogical

indexation is performed following a documentary lan-

guage, which describes objects according to peda-

gogical criteria (relevant to didactics).

In our project, the considered objects are texts and

we want the users (language teachers) to be able to

ﬁnd those objects by formulating questions that are

relevant to their set of problems, i.e. language didac-

tics. The scope of this article is that of pedagogical

indexation of texts for language teaching.

2 USERS’ NEEDS

In order to deﬁne pertinent criteria for text pedagog-

ical indexation for language teaching, rather than fa-

voring a given teaching approach, we have adopted an

empirical approach: a preliminary qualitative study

based on eight interviews with language teachers; a

short questionnaire destined to grasp how teachers

handle authentic texts, the classiﬁcation and research

of texts and to validate the apparently self-evident -

“Artiﬁcial language, which provides a formalised and

univocal representation of the documents of a corpus and of

the questions interesting a group of users, so as to allow the

simple spotting of the documents of the corpus which an-

swer the questions of those users”, translated by the authors

212

Loiseau M., Antoniadis G. and Ponton C. (2008).

MODEL FOR PEDAGOGICAL INDEXATION OF TEXTS FOR LANGUAGE TEACHING.

In Proceedings of the Third International Conference on Software and Data Technologies - ISDM/ABF, pages 212-217

DOI: 10.5220/0001893202120217

 SciTePress

yet fraught with consequences - hypothesis that a va-

riety of pedagogical contexts

can correspond to one

given text; and a long questionnaire aiming at ﬁne-

tuning the information gathered in the ﬁrst question-

naire and isolating research criteria.

2.1 Questionnaires Results

First Questionnaire. 112 out of 115

language

teachers declare “being able to use a same text in

various different contexts”. It is not only prospec-

tive thinking since 106 of these 112 teachers declared

having actually done it. Besides the conﬁrmation of

our hypothesis, we concluded that: teachers favor the

use of authentic texts

; they resort to specially con-

structed texts

when they want to control their linguis-

tic content (grammatical structures, vocabulary), par-

ticularly for beginners groups; the type of activity and

the audience seem to be the most frequent and com-

mon research criteria; ﬁnally, we could not to draw

conclusions concerning the teachers’ own text collec-

tions organization.

Second Questionnaire. We have been able,

through the description of the teachers’ own text

collection classiﬁcation, to isolate some research

criteria, the most widely used of which were: theme,

“linguistic content

or objective

” and level.

Aside from assessing the teachers’ expectations

towards a pedagogically indexed text base, the rest

of the questionnaire was dedicated to confront the

teachers’ practices with the hypothesis that some

criteria inﬂuenced one another. We have been able to

demonstrate that the activity type had an effect on:

• text length, [F(5,143)=3,362; p<,01]

;

• the number of “representative elements” of the

notion the activity is about, [F(4,127)=4,739;

p<,005]

;

by “pedagogical context” we mean the didactical goals

and all the characteristics of the audience (level, age, in-

terests, etc.) and of the institution (track/diploma, material

constraints, number of learners, etc.)

The corresponding question was not on the ﬁrst paper

version of the questionnaire, which 18 persons answered

in (Taylor, 1994) Taylor quotes various consistant def-

initions of “authentic text”, among which Nunan’s: “A rule

of thumb for authentic here is any material which has not

been speciﬁcally produced for the purposes of language

teaching.”

Unlike authentic texts, specially constructed text are

written for the purpose of being used as teaching material.

of the text

of the activity the text is to be used in

Anova test results, for information only

• the amount of unknown structures (other than the

object of the class), [χ

=32,177; dl=10; p<,001]

;

• the amount of unknown vocabulary (other than the

object of the class), [χ

=28,949; dl=10; p<,005]

Moreover, teachers declare that the level of the stu-

dents inﬂuences the quantity of unknown structures

tolerated in a text. They also state that the students’

native tongue, when taken into account, has an inﬂu-

ence on the quantity of unknown structures tolerated

in a text.

2.2 Conclusions

A given text can be used in various different peda-

gogical contexts. For instance, the quantity of un-

known structures and vocabulary sought (or tolerated)

depends on the activity type. This “un-knowledge” is

part of the learner’s level. Therefore, we can say that,

depending on the type of activity, a given text can cor-

respond to more than one audience.

In our opinion, this example illustrates the fact that

some “pedagogical” characteristics of a text are not

ﬁxed characteristics. They depend on the combina-

tion of some inherent characteristics of the text (such

as its linguistical content) and on the pedagogical con-

text in which the teacher plans to use it.

Our prototype will not cover all the teachers’ needs

and those which will be covered will only be partially

so, hence the need for evolutivity.

The system will not be able to provide only the

one most relevant text for every query. Whereas, it

would not be a problem to fetch a text, based on its

author and title; for other types of request, some cri-

teria will be too hard to model.

Let us consider the example of the ”theme” of the text.

There is no reliable NLP tool that would extract the

various themes of the text. Manual annotation would

raise the issue of consistency and exhaustiveness.

Even if the annotation in itself was consistent and

exhaustive, the issue of linking the annotation with

teachers queries remains. This type of criteria thus

requires a certain number of approximations both for

the indexation and for the expression/interpretation of

the queries. Other criteria such as “how interesting

the text will be to the students” are, to this moment,

almost impossible to model. They inﬂuence the teach-

ers’ choices all the same. Our system cannot substi-

tute for this choice process, it can facilitate it: act as

decision-assistance by providing a subset of candidate

texts highlighting certain elements of the text depend-

ing on the query that was performed.

test results, for information only

MODEL FOR PEDAGOGICAL INDEXATION OF TEXTS FOR LANGUAGE TEACHING

213

3 “FLAWS” OF PEDAGOGICAL

RESOURCE DESCRIPTION

STANDARDS

Pedagogical resource description standards do not

seem to bring an answer to the needs mentioned in the

previous section. In this article, we will only explain

our remarks concerning Learning Object Metadata

(LOM) but they can be extended to most educational

metadata standards.

3.1 LOM

LOM speciﬁcations propose more than seventy data

elements. For each data element, the LOM speciﬁca-

tions(IEEE, 2002) deﬁne the meaning of the data ele-

ment, a cardinality, a datatype - sometimes accompa-

nied by a vocabulary, which despite examples can be

vague and bound to construct its meaning through the

experience of the community of users. For instance,

data element 5.3, “Interactivity Level” will take its

value among very low, low, medium, high, very high.

Each element is non-mandatory and can be re-

peated to describe a “pedagogical object”: “any entity

- digital or non-digital - that may be used for learn-

ing, education or training”.

LOM descriptor 1.8 (“Agregation Level”) and its four

levels, spanning from raw resource to fully integrated

curriculum, illustrates this deﬁnition of pedagogical

object. Pernin justly remarks that the ambiguities and

imprecisions of the model stem from their “will to in-

tegrate within the same model entities of conceptually

very different levels”(Pernin, 2004). (Translation by

the authors)

3.2 Inherent Pedagogical Properties

All these metadata standards rely on the same prin-

ciple: making an inventory of the properties of an

object. The essence of this metadata introduces a

notion of ﬁxedness of the properties, including ped-

agogical properties. If ﬁxed pedagogical properties

can adequately describe objects that are already ped-

agogically exploited

, it does not seem ﬁt for raw

resources, a fortiori texts.

LOM category 5 (Educational) descriptors des-

ignate either properties we have presented as peda-

gogical context, or properties, which are linked to it:

5.10, “Description”, which contains “Comments on

By “pedagogically exploited”, we mean that they ben-

eﬁtted from some pedagogical added value, which can be

through the creation of an activity or other type of peda-

gogy oriented commentaries

how this learning object is to be used”; 5.7, “Typical

Age Range”, the “age of the typical intended user”;

and the self explanatory 5.8, “Difﬁculty”. We have

shown that a text could have various uses. Each dif-

ferent use (5.10 Description), will also affect both the

object’s difﬁculty and the audience to which it is des-

tined. LOM allows various “5. Educational” groups.

In our case an exhaustive description of the resource

would require all possible uses of every text to be

given in their LOM description. For each text, the

“Educational” description would be an n-tuple, each

component of which would be a set of “Educational”

data elements.

Under these conditions, a manual pedagogical de-

scription of texts is too tedious to imagine thus index-

ing them, might it be with LOM or any other peda-

gogical resource description standard.

4 THE NOTION OF “FACET”

Given the deﬁnition we provided, we have proved

that the pedagogical indexation of texts for language

teaching can be difﬁcult to achieve if one considers

all pedagogical properties as inherent to the object.

To consider the various parameters inﬂuencing the

properties of the text, we suggest the notion of “text

facet” :

Deﬁnition (Facet). A text facet is a property deﬁned

with a view to the latter’s pedagogical exploitation in

laguage teaching, accompanied by at least one mech-

anism to compute (automatically or not) the value of

this property for any text depending on a given peda-

gogical context.

The creation of a facet stems from the texts’ sub-

sequent pedagogical exploitation. Any characteristic

that can be useful to a language teacher for the search

or the choice of a text can be a facet of a text, pro-

vided that its value can be assigned or computed. The

computation of a facet value might involve any text

charasteristic. Those characteristics do not need to be

pedagogically relevant, as such.

4.1 Facets and Parameters

The difference between facet and element, as de-

ﬁned in metadata standards, resides less in its seman-

tics than in the means to obtain and use them. A

parametrized facet could be one that numbers the rep-

resentative elements of a language didactics relevant

notion present in a text. From now on we will desig-

nate this facet by F

RepE

. The value of this facet, is

contextual and depends on the pedagogical context:

ICSOFT 2008 - International Conference on Software and Data Technologies

214

the notion on which to work. A teacher of Span-

ish might ﬁnd it interesting to work on the expres-

sion of duty, for instance with haber* que + Inf

tener* que + Inf.

. Obviously a text will not neces-

sarily have the same number of occurrences of both

structures, the value of F

RepE

will thus be different

according to which structure (pedagogical context in

this case) is sought. We will discuss in 5.2 how to

make these facets available to the user.

To implement F

RepE

, a morphological parser (NLP)

can be combined with a pattern matching program (to

ﬁnd the sought structure) and a counter.

Constant Facet. In the same way the constant func-

tion f (x) → 0 is a function of x nontheless, a facet as-

signing the same value to a text for any pedagogical

context is all the same a facet.The length of the text

length

) or its author are thus facets. These facets, like

pedagogical resource description standards metadata,

consider properties that are intrinsic to the text.

...

1~2

Text

Figure 1: Text facets.

Compound Facets. All the previous examples ei-

ther rely on user submitted information (cf. Author)

or on underlying text properties revealed by computer

programs (including NLP programs). It is also possi-

ble to combine various existing facets to create others

with higher pedagogical added value (Figure 1).

Let us combine two facets we have already discussed:

length

and F

RepE

. We can perform an approximation

of a facet evaluating the adequation of a text for a ped-

agogical context depending on two parameters: the

activity type and the notion at the center of the activity

(2.1). Since we have shown that the activity type both

had an effect on text lengths and the number of rep-

resentative elements, we can reuse the data from our

questionnaire to establish threshold values depending

on the activity type.

Figure 1 emphasizes the fact, that combining two

facets is different than just giving both a value. In this

case the use of the compound facet infers from the

Hay que comer / One must eat. (Impersonal)

Tienes que comer / you must eat.

activity type the desired number of words of the text

as well as the number of representative elements of

the notion (along with the tolerance). Of course, this

data comes from the teachers’ declared practices, it

is therefore likely that the facet will not be very pow-

erful. Though, studying the actual practices with the

prototype should allow a reﬁnement of those values

and the iterative creation of more powerful facets. For

instance, this particular compound facet could then be

improved, by taking into account the students level.

4.2 Facet vs Descriptor

These examples evoke the major difference between

metadata descriptors and facets. The descriptors con-

tain a set of a priori deﬁned ﬁxed values. Whereas the

facets’ can also be dynamically computed according

to a pedagogical context entered by the user.

For some simple facets, one could imagine a mech-

anism that would create the Educational elements of

an application standard of LOM. For a “level” facet

that would take into account the activity type, it would

be possible to create an “Educational” tuple for each

possible pedagogical context. But other facets, such

as the F

RepE

, depending on the formalism used to de-

scribe the notions, the number of possible pedagogi-

cal contexts (hence the number of tuples) would make

exhaustiveness an issue.

5 THE MODEL

Facets allow to integrate processes to the description

of resources, without actually containing them.

5.1 Prism

The revelation of a facet requires certain processes

which can be performed by a human (e.g. “author”

facet) or by a program. Hence the notion of prism:

Deﬁnition (Prism). The prism is a set of computer

function(s) created or combined in order to reveal the

facets of a text.

The functions in Figure 2 (Fn

→ Fn

) are in-

spired from the MIRTO model(Antoniadis et al.,

2005): in both cases the functions in themselves do

not have a didactic interest. Their combination at the

level of the prism (respectively scripts) is what cre-

ates the pedagogical added value. Tokenization is not

a functionality of the prism, but the succession of to-

kenizer and counter is: it gives access to F

length

In Figure 2, one can see two facets to each of

which a treatment in two parts is associated. The ﬁrst

MODEL FOR PEDAGOGICAL INDEXATION OF TEXTS FOR LANGUAGE TEACHING

215

Prism

Vi1

Vi2

Figure 2: The prism.

part of the treatment is to be performed once and for

all when the text is added to the system. It is meant

to add the underyling data from which the values of

the facets are computed. The second part is the actual

process which computes the values of the facets using

the pedagogical context. The processes associated to

two different facets can both use the same function

(Fn

, in Figure 2, e.g. a counter).

The creation and the modiﬁcation of the prism re-

quire computer science expertise which cannot be ex-

pected from a language teacher. This task ought to be

performed by developpers, but cannot be so without

language didactics knowledge considering the pivotal

position of the prism in our model. We thus advo-

cate the inter-disciplinary dialogue with teachers and

language didactics experts for this phase.

5.2 Access to the Facets of a Text

To make the prism revealed facets available to teacher

users, we propose two non exclusive means: views

and visualizations.

5.2.1 View and Empty View

Facets are text properties, revealed thanks to the con-

junction of computer programs combined in the sys-

tem prism. The functions of the prism not only allow

the necessary processes for revealing facets, they are

also used to perform queries on these values.

Deﬁnition (View). The view of a text t according to

the facet F under the pedagogical context C, is the

value of F computed for t using the characteristics of

In other words, given a facet and a text, the prism

computes a different view of this text for each peda-

gogical context speciﬁed by the teacher through the

system interface.

By extension, the view of a text set is a set of the

respective views of these texts. By constraining the

values of the facets, thus extending the pedagogical

context to the expected values, the view of a set of

texts acts as a ﬁlter: the view of a text is “empty” if it

does not satisfy the constraints.

In F

RepE

, there is a view for each language fact of

which we might want to count the occurences. Let us

use again the pedagogical context haber* que + Inf. A

view of a text for the facet and pedagogical context is

the number of occurences of haber* que + Inf. A con-

strained view of the same text, using the same facet

and extending the pedagogical context, could impose

that the corresponding view is greater than a given

number: e.g. the view of F

RepE

for the context count

of “haber* que + Inf” structures ≥ 4. The view of a

text, that does not satisfy this constraint will be con-

sidered an empty view. By extension the view of a text

set will contain all the non-empty views of the texts of

the set. I.e. the views for all the texts having at least 4

occurences of haber* que + Inf structures in this case.

5.2.2 Homogeneous Text Collection

The notion of “homogeneous text collection” directly

stems from that of view of a group of text.

A text collection, will be considered homogeneous if

all the texts it contains explicitly share one or more

properties.

Deﬁnition (Homogeneous Text Collection). An ho-

mogeneous text collection is a constrained view of a

text collection.

The homogeneity of the collection therefore de-

pends on the satisfaction of the research criteria. This

notion allows to perform a research on a subset of

texts: the user will be able to reﬁne his research step

by step. Each step being the creation of a view of the

previous step (Figure 3).

These intermediary collections will allow better

performances than adding parameters to the query and

performing it on the whole collection all over again.

5.2.3 Visualizations

To provide the user with text choice assistance (cf.

Figure 3), the system provides the user with graphical

representations of the underlying information added

by the prism in relation with a given facet. These rep-

resentations are called “visualizations”. The informa-

tion of the visualizations is qualitative, whereas the

view is quantitative. On Figure 2, F

is associated

to two visualizations: Vi

, uses the underlying infor-

mation associated to the facet and Vi

uses the view

itself.

To help the user choose between the texts satisfy-

ing a new view of F

RepE

- texts with 5 occurences of

English Preterite - the system could provide a visu-

alization highlighting each occurence or another one,

ICSOFT 2008 - International Conference on Software and Data Technologies

216

which would only show the list of the occurences of

the structure (Figure 4).

System Interface

Teacher

User

View

Visualization change

View

for one or more of V

's facets

on one ore more texts

visualization of

view of C

Collection

visualization of

view of C

Collection

= V

)

texts the V

view

of which is not

empty

= V

)

Figure 3: Interaction diagram.

Preterite

Pipepline (8)Pixies (5)

3 bears (13)

Visualizations

The Story of Goldilocks and the Three Bears

Once upon a time, there was a little girl named

Goldilocks. She went for a walk in the forest.

Pretty soon, she came upon a house. She

knocked and, when no one answered, she

walked right in. At the table in the kitchen, there

were three bowls of porridge. Goldilocks was

hungry. She tasted the porridge from the ﬁrst

bowl. "This porridge is too hot!" she explained.

So, she tasted the porridge from the second

bowl. "This porridge is too cold," she said. So,

she tasted the last bowl of porridge.

Text

List

Preterite

Visualizations

Pipepline (8)Pixies (5) 3 bears (13)

The Story of Goldilocks and the Three Bears

tasted (3)

was (2)

answered

came

explained

knocked

said

walked

went

were

Text

List

Figure 4: Visualization examples.

6 CONCLUSIONS

A study of language teachers’ needs and practices

proved pedagogical resource description standards in-

adequate to describe raw resources, in particular texts,

in the context of pedagogical indexation for language

teaching. We have presented a model, which attemps

to take into account the inﬂuence of the pedagogical

context on the value of the texts’ properties. To do so

we introduced the notions of facet, prism, view, visu-

alization and homogeneous text collection.

The implementation of such a system must be as

generic as can be: its quality will strongly depend on

its ability to integrate different kinds of tools and on

the quality of the latter. While writing these lines, we

are implementing the architecture we have described

using php/mySQL. The prism of the prototype will in-

tegrate a few simple tools to grant access to a few ba-

sic facets. The prototype is meant to be an implemen-

tation example of the architecture, showing its viabil-

ity. But any urge to make a product out of it would re-

quire several distinct tasks. The analysis of the teach-

ers needs: the study of the teachers’ actual practices

would reﬁne our declared practices based asumptions

of the teachers’ needs; such a study could use our pro-

totype. Isolating text properties from a language di-

dactics point of view that would prove useful to the

teachers. And ﬁnally, developing reliable NLP tools,

which would add information to the texts that can be

used for facet creation. Of course, each these three

leads must feed off the conclusions of the other two.

REFERENCES

Antoniadis, G.,

Echinard, S., Kraif, O., Lebarb

e, T.,

Loiseau, M., and Ponton, C. (2004). Nlp-based script-

ing for call activities. In Coling Workshop on eLearn-

ing for Computational Linguistics and Computational

Linguistics for eLearning Proceedings, Gen

eve.

Antoniadis, G.,

Echinard, S., Kraif, O., Lebarb

e, T., and

Ponton, C. (2005). Mod

elisation de l’int

egration de

ressources tal pour l’apprentissage des langues : la

plateforme mirto. ALSIC, 8:65–79.

Higgins, J. and Johns, T. (1984). Computers in language

learning. Collins ELT / Addison-Wesley, World Lan-

guage Division, London.

IEEE (2002). Final 1484.12.1 lom draft standard document.

Lef

evre, P. (2000). La recherche d’informations : du texte

int

egral au th

esaurus. Herm

es Science, Paris.

Levy, M. (1997). Computer-Assisted Language Learn-

ing, context and conceptualization. Oxford University

Press.

Pernin, J.-P. (2004).

A propos des objets p

edagogiques. In

Entre technique et p

edagogie : la cr

eation de con-

tenus multim

edia pour l’enseignement et la formation,

Neuch

atel.

Taylor, D. (1994). Inauthentic authenticity or authentic in-

authenticity? Teaching English as a Second or For-

eign Language, 1(2):A–1.

MODEL FOR PEDAGOGICAL INDEXATION OF TEXTS FOR LANGUAGE TEACHING

217