Next Generation Learner Modeling by Theory of Mind Model Induction

Klaus P. Jantke

, Bernd Schmidt

and Rosalie Schnappauf

Fraunhofer Institute for Digital Media Technology, Erfurt, Germany

Fachhochschule Erfurt, Erfurt, Germany

University of Rostock, Rostock, Germany

Keywords:

Technology-enhanced Learning, User Modeling, Learner Modeling, Player Modeling, Theory of Mind,

Learner Model Induction, Inductive Inference, ToMMI Technology.

Abstract:

Learning is a spectrum of involved processes requiring the learner’s engagement and building upon the

learner’s prior knowledge and other prerequisites. Educators know how to adapt to their learners’ needs

and desires. User modeling is a key technology to enable digital systems such as e-learning environments and

serious games to adapt to their users’s peculiarities. There is a huge corpus of scientiﬁc research on user mod-

eling, on implementation of user modeling and related system adaptivity, and on the impact on teaching and

learning. The aim of the present contribution is to go even further. The concept of theories of mind is adopted

and adapted from animal behavioral research. Theory of mind user models allow for the identiﬁcation and

representation of user/learner/player peculiarities beyond the limits of all other preceding approaches to user

modeling. Theory of mind learner models allow for the representation of higher quality proﬁles describing,

for instances, intention, misconceptions, or even fear. The acquisition of suchlike expressive proﬁles is an

inductive learning process of the digital system. The inductive inference of learner proﬁles requires particular

concepts and algorithms. An implementation serves as proof of concept.

1 MOTIVATION

This is a technological paper. Although the authors

have some running implementation (Schmidt, 2014),

application and evaluation are considered secondary.

Emphasis is put on an introduction to the innovative

technology. The authors’ intention is to coin the term

theory of mind model induction (nickname: ToMMI)

and to discuss how to utilize this technology for the

purpose of learner modeling.

The implementation does not only serve as proof

of concept, one may go even further. One of the key

results in section 6 demonstrates the system develop-

ers’ ability to proof mathematically that the modeling

algorithms succeed in practice.

The present paper is the authors’ ﬁrst publication

in the ﬁeld and, thus, the ﬁrst publication on this novel

technology at all. Therefore, technology is in focus.

The developed technology relates in an intriguing

way to other areas of research such as, prominently,

theories of mind (Carruthers and Smith, 1996) and

inductive inference (Jain et al., 1999).

The key idea is to adopt and adapt theory of mind

concepts for user modeling. In doing so, user model-

ing becomes inductive learning. This is investigated

in cases where users are learners and/or players.

Besides theory and technology development, the

authors work in the area of technology enhanced-

learning, in general, and on game-based learning, in

particular ranging from earlier publications such as,

e.g., (Jantke et al., 2003), (Jantke et al., 2004), and

(Jantke and Knauf, 2005) to recent contributions like,

e.g., (Knauf et al., 2010), (Jantke and Schulz, 2011),

(Fujima and Jantke, 2012), (Arnold et al., 2013),

(Krebs and Jantke, 2014), (Jantke and Hume, 2015).

When studying digital games, there arises a really

enormous manifold of exciting questions. Playing a

game, usually, is fun (Koster, 2005). Among the great

excitements of game play, there is the anticipation

of an adversary player’s intentions. Knowing or, at

least, hypothesizing what a human player wants, may

form the basis for advanced ideas of game play such

as, e.g., setting a snare for the adversary. Potentially,

this does apply to serious games as well (Egenfeldt-

Nielsen, 2007). To go even further, in serious games,

this knowledge may be used to assist the learner.

Theories of theories of mind (Carruthers and

Smith, 1996) allow for underpinning those studies.

To warm up, so to speak, the authors brieﬂy sketch

their preliminary case study in which a prototypical

theory of mind induction has been implemented for

the ﬁrst time ever (Schmidt, 2014).

Jantke, K., Schmidt, B. and Schnappauf, R.

Next Generation Learner Modeling by Theor y of Mind Model Induction.

In Proceedings of the 8th International Conference on Computer Supported Education (CSEDU 2016) - Volume 1, pages 499-506

ISBN: 978-989-758-179-3

499

2 THE GORGE CASE STUDY

In this work, the theory of mind induction means

player modeling.

This application is based on a digital game named

GORGE (Jantke et al., 2010) (see also (Gaudl et al.,

2009) and (Jantke, 2010)). What has been done for

game playing is now ready for a transfer to computer

supported education. Seen from this perspective, this

contribution aims at the transfer of technologies from

player modeling to learner modeling.

In GORGE there are different teams of robots (see

ﬁgure 1) operated by different players who may be

humans or computer programs. Originally, GORGE

has been designed and implemented as a research tool

for studies of the perception of Artiﬁcial Intelligence.

Figure 1: Screenshot of an Earlier Version of GORGE

GORGE is turn-based. A dice is rolled and players

may select one of their robots to move it accordingly.

When a robot reaches a cell on the game path where

another robot is sitting, this one is jostled backwards

to the next free cell. Due to this game mechanics,

robots tend to form clusters on the path. This leads to

opportunities for taking revenge.

Human players may have largely varying inten-

tions about jostling others such as, e.g., some grand-

parent’s intention not to frustrate the own grandchild.

Another aim may be to take immediate revenge when-

ever possible. Next, consider reciprocal altruism:

“If X jostled me never before and if I have the choice

to jostle X or to do something else, I will not jostle X.”

Intentions like this may be easily represented by

logical formulas. A computer program can monitor

human behavior when playing GORGE. Based on the

observations, the computer program hypothesizes the

player’s goals. The computer learns a theory of mind.

3 CONVENTIONAL MODELING

There is more than 30 years of work on user modeling

and adaptation. Consequently, it is not easy to relate

the authors’ present efforts.

For the sake of comparison, the authors of the

present paper have analyzed (Houben et al., 2009),

(De Bra et al., 2010), (Konstan et al., 2011), (Mas-

thoff et al., 2012), (Carberry et al., 2013), (Dim-

itrova et al., 2014), (Ricci et al., 2015), and the pa-

pers therein, as well as some of the impressive intro-

ductory and survey papers such as (Brusilovsky et al.,

1995), (Specht and Weber, 1997) (Brusilovsky, 2001),

(Brusilovsky and Mill

an, 2007).

There is a principle of conventional user model-

ing. A certain ﬁnite number n of features are selected

and human beings are characterized accordingly by

points in an n-dimensional space (Jung, 1921).

The Myers-Briggs Type Indicator (MBTI, for

short) (Briggs Myers and Briggs, 1980) relies explic-

itly on Carl G. Jung’s theory of psychological types.

Expressed in formal terms, according to MBTI, pro-

ﬁles of humans are points in a 4-dimensional space.

The Felder-Silverman approach (Felder and Sil-

verman, 1988) is very similar to the MBTI, but puts

explicit emphasis on learning such that one may un-

derstand it as an attempt to compromise between the

MBTI and the David Kolb Learning Style Inventory

(LSI, for short).

The LSI (see (Kolb, 1984), (Kolb and Fry, 1975)),

analogously, builds a 4-dimensional space to host

learner proﬁles. In contrast to the before-mentioned

approaches, the LSI is enriched by a cyclic learning

process model that underpins the four dimensions.

The history of learner modeling relies on model

spaces of varying dimension (from (Brusilovsky et al.,

1995) to (Brusilovsky and Mill

an, 2007)).

Occasionally, approaches are coming up in which

the authors attempt to abandon the conventional limi-

tations and aim at something like learner preferences

(Kassak et al., 2015), (Schewe et al., 2007), (Smith

et al., 2015). But saying that “preferences are ex-

pressed on the level of items” (Kassak et al., 2015)

means to fall back to conventional approaches.

However impressive and practically successful,

contemporary user modeling has apparent limitations.

For illustration, educators who strive hard to treat

their students with empathy know about the critical

impact of misconceptions and about the importance

of conceptual change and want to know their learner’s

peculiarities ((Carey, 1985; Carey, 2000), (Thagard,

2012), (Vosniadou, 2013b), and the chapters therein).

CSEDU 2016 - 8th International Conference on Computer Supported Education

500

4 THEORIES OF MIND

The present short section deals with an inspiration for

understanding other individuals’ motivations, goals,

desires, preferences, fears, and the like.

The aim of the authors’ present work is to take the

inspiration from the present section to proceed from

conventional learner modeling (see section 3) to in-

novative learner modeling of a higher expressiveness

(see section 6) and, therefore, of a higher utility.

The source of inspiration is theories of mind as

surveyed in (Carruthers and Smith, 1996).

The gist of the concept is well illustrated in

sources such as (Emery, 2004), (Emery et al., 2004),

(Goldman, 2006), (Clayton et al., 2006), (Call and

Tomasello, 2008), (Emery and Clayton, 2009). Al-

though the treatment in recent publications such as

(Call and Tomasello, 2008) and (Emery and Clay-

ton, 2009) is slightly more sceptical than in earlier

work ((Carruthers and Smith, 1996) and the chapters

therein), the application of the theories of mind per-

spective to human beings is scientiﬁcally justiﬁed and

practically useful (Mauer, 2012).

Loosely speaking, theories of mind refer to two

different individuals, say A and B.

In some of the above-cited sources, both agents

A and B are birds, e.g., animals of the food-caching

species western scrub jay (Aphelocoma californica).

In the authors’ work, A is a computer program and

B is a human learner.

The scenario under consideration is as follows.

While agent B is acting–chaching food, interacting

with an e-learning system, playing a digital game, or

whatsoever–the agent A is monitoring B’s behavior.

A is pondering the observations made and tries to ﬁnd

explanation for B’s behavior in terms of B’s thoughts.

What A constructs is, in some sense, a model of B.

In cases where B is a human interacting with a digital

system, the result is a user model or, more speciﬁcally,

a learner model or a player model or both at once.

There are particular aspects in theories of mind in-

vestigations such as time travel (Suddendorf, 2007)

that are relevant to game play.

When game-based learning moves into focus,

those particular investigations open completely new

opportunities of application.

Notice that the present approach goes beyond the

reach of all the above-cited sources. The present work

aims at computer programs that are able to monitor

human-computer interaction and learn about humans

by algorithmic induction of user proﬁles which have

the particular form of a computer’s theory of mind

about a human user/learner/player.

5 EDUCATIONAL RELEVANCE

OF THEORIES OF MIND

Educators think about the thoughts of their students.

This covers a wide spectrum of aspects and enables

them to adapt to their students’ peculiarities, needs,

and desires (Davis et al., 2000). When computers

in educational settings shall become adaptive as well,

there arises the need to equip them with digitally, i.e.,

formally represented knowledge accordingly.

By way of illustration, let us look at preconcep-

tions and misconceptions (Vosniadou, 2013b).

There are wide-spread misconceptions which

cause difﬁculties to learners. A typical one is the be-

lief that motion is caused by a force (Hammer, 1996).

In chemistry, many misconceptions are due to the

misinterpretation of molecular equations. This leads

learners to the belief that water is just a large amount

of H

O molecules, a theory of mind, so to speak, that

makes the electrical conductivity of water completely

incomprehensible.

In biology, a quite prominent misconception is the

confusion of osmosis and diffusion accompanied by

a large number of different misbelieves such as water

molecules cease movement at osmotic equilibrium.

Comprehensive publications such as the hand-

book (Vosniadou, 2013b) or, e.g., (Chi et al., 1994),

(diSessa and Sherin, 1998), (MacBeth, 2000), and

(Vosniadou, 2013a) illustrate the omnipresence of

learners’ thoughts that are likely to hinder learning.

There is abundant evidence for the need of knowing a

learner’s preoccupation.

Misconceptions are wide-spread in science as

demonstrated in the case of biology (Kayoko and

Hatano, 2013), in chemistry (Barke et al., 2009), in

physics (Brown and Hammer, 2013), and in so-called

earth science (Phillips, 1991).

The situation is not different in the humanities;

see, e.g., (Leinhardt and Ravi, 2013) for history and

(Arabatzis and Kindi, 2013) and (Thagard, 2013) for

the history of science.

For curiosity, the rise and fall of phlogiston–the

concept and the theory–may be of interest (Wisniak,

2004).

The theories of mind are collections of thoughts

ascribed to other individuals. If computers shall be

able to construct theories of mind that are intended to

characterize human beings, the hypothesized thoughts

must be represented formally–inside the computer,

constructed and written by the computer and readable

by the computer. Theories of mind are sets of logical

formulas.

This logical point of view is taken subsequently.

Next Generation Learner Modeling by Theory of Mind Model Induction

501

6 INNOVATIVE TECHNOLOGIES

OF LEARNER MODELING

Recall a human player’s reciprocal altruism when

playing GORGE (section 2): “If X jostled me never

before and if I have the choice to jostle X or to do

something else, I will not jostle X.”

If a learner is driven by an intention like this one,

how can a computer program learn about the human’s

thoughts to represent this particular learner’s highly

individual intentions, desires, goals, ideas, fears, and

so on in a proﬁle as expressive as a theory of mind ?

This section is intended to survey an answer that,

hopefully, will inform the readers about the essentials.

6.1 Spaces of ToM Hypotheses (I)

The theories of mind that hypothetically characterize

human learners are represented as ﬁnite sets of logical

formulas.

Modal operators like  and ♦ are appropriate to

make complex expressions readable. The operators,

as usual, express the two modalities of necessity and

possibility, respectively (Blackburn et al., 2001).

The intention of reciprocal altruism, e.g., may be

represented as follows. Let us assume, that predi-

cates such as jostle are ternary. The ﬁrst of the argu-

ments contains time information. The other two argu-

ments contain player names, where the distinguished

constant ∗ names the human player who is currently

modeled. The symbols π

...

, perhaps, with upper and

lower decorations denote strings of events describing

human-computer interactions. The binary relation 

holds for any two strings, exactly if the ﬁrst one is

an initial segment of the second one. All remaining

syntax is conventional ﬁrst order predicate calculus

(see, e.g., (Richter, 1978)).

The following formula is a statement about some

human-computer interaction encoded by π and some

recent player’s action µ. It formalizes a variant of the

player’s aim at reciprocal altruism (see section 1).

( 6 ∃π

: π

 π ∧ jostle(π

, X, ∗) → ¬jostle(πµ, ∗, X) )

It may be difﬁcult to realise “¬jostle(πµ, ∗, X)”,

if the action µ is enforced. The following ﬁts better.

( 6 ∃π

: π

 π ∧ jostle(π

, X, ∗) ∧ ¬jostle(π, ∗, X) →

¬jostle(πµ, ∗, X ) )

Logical formulas like the two above form the

space of hypotheses. A user proﬁle is a ﬁnite set of

formulas expressing player intentions. In other words,

those formulas express “the computer’s thoughts”

about the human user currently modeled.

The question is how to learn those formulas from

observations.

6.2 Inductive Inference of

Human Learner Proﬁles

. . . it is not really difﬁcult to construct a series of

inferences, each dependent upon its predecessor and each

simple in itself. If, after doing so, one simply knocks out

all the central inferences and presents one’s audience with

the starting-point and the conclusion, one may produce

a startling, though possibly a meretricious, effect.

Sherlock Holmes to Dr. Watson in

‘The Adventure of the Dancing Men’

by Arthur Conan Doyle, 1915

Proﬁles of human learners are inductively inferred

from observations of human-computer interactions.

Basic steps of inference are simple, but the overall

result of learner modeling may appear meretriciously

as Arthur Conan Doyle put it (Doyle, 1915).

Logic programming

is the authors’ software

technology of choice ((Clocksin and Mellish, 1981),

(Sterling and Shapiro, 1986)) to implement the steps

of logical inference.

Observations are formally represented in the form

πµ where π describes the history of interaction and µ

denotes the human learner’s current action.

As long as the current user proﬁle is sufﬁcient to

explain the human user’s behavior, a property named

consistency, there is no need to change the proﬁle.

More formally, assume the necessary background

knowledge (domain knowledge, system behavior, . . . )

BK and a current learner proﬁle ϕ. The consistency is

expressed in formal terms

as BK ∪ {πµ} |= ϕ.

Note that there is no method, in general, to prove

consistency (Jain et al., 1999). Usually, this prop-

erty is only co-enumerable (Rogers jr., 1967; Sipser,

1997). Therefore, the logical reasoning strategy is

refutation. Fortunately, Prolog is refutation-complete.

Every hypothesis ϕ of the learner proﬁle is kept

until its refutation succeeds. Our system in form-

ing learner proﬁles tries to validate BK ∪ {πµ} 6|= ϕ.

Only if this succeeds, the hypothesis is given up and

replaced by a reﬁnement. But how to do this . . . ?

Logic programming is an appropriate approach to the

authors’ ambitious task of modeling learners by theories of

mind. For comparison, look at IBM’s Watson which is pow-

ered by 10 racks of IBM Power 750 servers running Linux,

and uses 15 terabytes of RAM, 2,880 processor cores and is

capable of operating at 80 teraﬂops. Watson was written in

mostly Java but also signiﬁcant chunks of code are in C++.

A core part to perform reasoning is implemented in Prolog.

This, in fact, requires some more precision. To be a

user proﬁle, ϕ must contain two free variables π about the

current (history of) interaction (Note that this includes time

information.) and the learner’s current action µ. The present

notation is simpliﬁed.

CSEDU 2016 - 8th International Conference on Computer Supported Education

502

6.3 Spaces of ToM Hypotheses (II)

To learn learner proﬁles, one needs to agree about

“what to say about a human learner”. After such a

decision has been made, one has a potentially inﬁnite

set of formulas which may describe characteristics of

human learners, i.e. theories of mind.

As described before, the refutation-completeness

of Prolog allows for a fully computerized refutation

of logical expressed hypotheses. But how to reﬁne

refuted hypotheses within a theory of mind? How to

step forward from one hypothetical learner model to

the next one, if necessary?

To arrive at a completely algorithmic approach to

learning of theories of mind forming learner models,

we adopt and adapt a fundamental concept introduced

by Dana Angluin (Angluin, 1980). Her ingenious

concept is called indexed family of formal languages

in which every particular language has a decidable

word problem

. In contrast to Dana Angluin’s ap-

proach, we can not assume decidability. This means

that our approach is more expressive, but somehow

less comfortable. We have to invest more algorithmic

effort. These thoughts lead to the concept below that

refers to some underlying basic knowledge BK and to

a set Π describing sequences of possible interactions.

A so-called indexed family of logical formulas is a

sequence of formulas Ψ = {ψ

}

n=0,1,2,...

as follows.

(i) Ψ = {ψ

}

n=0,1,2,...

is recursively enumerable.

(ii) For any π ∈ Π and for any index n, BK ∪{π} 6|= ψ

is recursively enumerable.

(iii) For any two indices i and j with i < j, ψ

does not

logically imply ψ

Given an indexed family of logical formulas Ψ,

one can easily implement a computer program able

to learn whatever formula characterizes a human

learner’s intentions.

According to fundamental results of inductive

learning based on usually incomplete information

(Jain et al., 1999), there is a computer program able

to learn whatever formula in Ψ = {ψ

}

n=0,1,2,...

might

describe a human learner’s peculiarities. The infer-

ence principle is called identiﬁcation by enumeration.

For every set of observed learner behavior, it returns

the ﬁrst formula which is not (or not yet) refuted.

Note that there may be any ﬁnite number of spaces

of hypotheses in use, i.e., any collection of indexed

families of logical formulas Ψ

, Ψ

000

, . . . , Ψ

(k)

Similar inference procedures may run on the different

enumerations in parallel returning one statement from

each of the (sub-)spaces of hypotheses.

The word problem is the question whether or not any

word belongs to a language (Hopcroft and Ullman, 1979).

6.4 Features of Learner Modeling

by Theory of Mind Induction

To sum up sections 6.1, 6.2, and 6.3, if we are able

(i) to describe potential human learner peculiarities of

interest by means of logical formulas and if we can

(ii) enumerate these formulas appropriately, there is a

universal computational method that is able to learn

whatever human peculiarity might be on hand.

Recall that a formula ψ occurs as a component of

a proﬁle of some human learner, if it is listed within

an indexed family of logical formulas, which is in use

for theory of mind induction, where this formula has

some index n, i.e., ψ = ψ

. Furthermore, for every

preceding index i, the formula ψ

has been refuted

Finally, there is no refutation of ψ

, at least, not yet.

The key insight is that a component such as ψ

within a learner proﬁle is unavoidably hypothetical.

This has to be stressed, though it is not a big surprise.

When humans think about other human beings and

about their thoughts, they usually are aware of uncer-

tainty. This applies to computer “thoughts” as well.

Consequently, it is highly advisable to deal with

hypothetical learner proﬁles carefully. A particularly

interesting option is to reﬁne adaptivity by reﬂection

(Jantke et al., 2013). This approach has deep roots

in research about reﬂective inductive inference (see

(Grieser, 2008) which expands upon (Jantke, 1994),

(Grieser and Jantke, 1995), and (Jantke, 1995)).

The crux is that there is no way to deﬁnitely say

whether or not a hypothesis is consistent with given

observations. The only way of automated reasoning

is to try refutation. If a hypothesis is not consistent,

Prolog will ﬁnd this out after some time.

A closer look at the methodology described in

the previous subsections reveals a further peculiarity.

First, naturally, if a formula has been found that truly

describes a learner’s intention, goal, motivation, or so,

this is consistent and, thus, will never be abandoned.

But, second, what might be the reason not to arrive at

such a correct hypothesis? The only reason may be

another formula, say ψ

, which occurs earlier in the

enumeration, but which has not yet been refuted.

In such a case, this earlier hypothesis the learn-

ing algorithm gets stuck with is equivalent–within the

limitated information available–to the “true” one.

By means of logic programming and due to the known

refutation completeness of Prolog, this means that taking

(a) the basic knowledge, (b) the observation of the learner’s

behavior, and (c) the candidate formula ψ

, there has been

derived the empty clause. This means the discovery of a

contradiction and, thus, it establishes a refutation of ψ

See (Bl

asius and B

urckert, 1978) and (Richter, 1978)

for background and (Popper, 1934) for the bigger picture.

Next Generation Learner Modeling by Theory of Mind Model Induction

503

7 SUMMARY & CONCLUSIONS

To the authors’ very best knowledge, the present

approach of modeling users, especially learners, by

theories of mind–more precisely by the induction of

user proﬁles that form theories of mind–is a novelty.

By theories of mind one may express peculiarities

of human users far beyond the limits of conventional

user modeling (see the rich corpus of work (Houben

et al., 2009), (De Bra et al., 2010), (Konstan et al.,

2011), (Masthoff et al., 2012), (Carberry et al., 2013),

(Dimitrova et al., 2014), (Ricci et al., 2015), and the

papers therein). This is of a particular relevance to

technology-enhanced learning. Teachers and digital

systems to assist them for purposes of teaching and

learning need to “know” about their human learners.

There may be individual goals, misconceptions,

and even fears of human learners. Theories of mind

are appropriate to express such individual conditions

and the so-called theory of mind model induction is

appropriate for the computerized learning of human

peculiarities. This is a bio-inspired technology that

has a ﬁrm foundation in behavioral research.

But do we really need so much technology on a

CSEDU conference?

Yes, we do. First of all, educators need to be in-

formed about the next opportunities available to them.

Second and even more importantly, they are needed

to carve out the future of education. Only educators

can name the human learner conditions that are not

yet covered sufﬁciently well by conventional learner

modeling. Only educators can help the technologists

to determine the formalism needed to express learner

peculiarities of relevance.

Seen from this perspective, the present paper is

intended to be understood as a call for co-operation.

Let us go together for the design, implementation,

application, and evaluation of the next generation

of e-learning systems that are able to “understand”

their human learners by the induction of learner needs

from observed behavior.

The observation of learners leads to hypothetical

learner proﬁles (see section 6 above). Based on these

learner proﬁles, personalization takes place. But how

to treat learners appropriately? This is not a question

for technologies. Think of a trainee who strives hard

to avoid repeated conversation with the same staff

member. How to react appropriately?

It needs psychologists, educational psychologists,

educators, and domain experts to determine suitable

adaptive behavior in response to hypothesized learner

proﬁles. As a ﬁrm basis, the specialists need to under-

stand learner proﬁles and the way they are created.

REFERENCES

Angluin, D. (1980). Inductive inference of formal lan-

guages from positive data. Information and Control,

45:117–135.

Arabatzis, T. and Kindi, V. (2013). The problem of concep-

tual change in the philosophy and history of science.

In (Vosniadou, 2013b), pages 343–359.

Arnold, S., Fujima, J., Jantke, K. P., Karsten, A., and Simeit,

H. (2013). Game-based training for executive staff

of professional disaster management: Storyboarding

adaptivity of game play. In Tan, D., editor, Proceed-

ings of the International Conference on Advanced In-

formation and Communication Technology for Edu-

cation (ICAICTE 2013), Sept. 20-22, 2013, Hainan,

China, pages 68–73. Atlantis Press.

Barke, H.-D., Hazari, A., and Yitbarek, S. (2009). Mis-

conceptions in Chemistry. Addressing Perceptions in

Chemical Education. Springer.

Blackburn, P., de Rijke, M., and Venema, Y. (2001). Modal

Logic. Cambridge, U.K.: Cambridge University

Press.

asius, K. H. and B

urckert, H.-J., editors (1978).

Deduktionssysteme. Automatisierung des logischen

Denkens. M

unchen, Wien: Oldenbourg.

Briggs Myers, I. and Briggs, K. (1980). Gifts Differing: Un-

derstanding Personality Types. Mountain View, CA:

Davies-Black.

Brown, D. E. and Hammer, D. (2013). Conceptual change

in physics. In (Vosniadou, 2013b), pages 121–137.

Brusilovsky, P. (2001). Adaptive hypermedia. User Model-

ing and User-Adapted Interaction, 11(1-2):87–110.

Brusilovsky, P. and Mill

an, E. (2007). User models for

adaptive hypermedia and adaptive educational sys-

tems. In Brusilovsky, P., Kobsa, A., and Neijdl, W.,

editors, The Adaptive Web. Methods and Strategies of

Web Personalization, volume 4321 of LNCS, chap-

ter 1, pages 3–53. Berlin & Heidelberg: Springer-

Verlag.

Brusilovsky, P., Specht, M., and Weber, G. (1995). Towards

adaptive learning environments. In GI Jahrestagung

1995, pages 322–329.

Call, J. and Tomasello, M. (2008). Does the chimpanzee

have a theory of mind? 30 years later. Trends in Cog-

nitive Sciences, 12(5):187–192.

Carberry, S., Weibelzahl, S., Micarelli, A., and Semer-

aro, G., editors (2013). User Modeling, Adaptation,

and Personalization. Proc. 21th International Conf.,

UMAP 2013, Rome, Italy, June 2013. Number 7899

in LNCS. Springer.

Carey, S. (1985). Conceptual Change in Childhood. Cam-

bridge, MA, USA: The MIT Press.

Carey, S. (2000). Science education as conceptual change.

J. Applied Developmental Psychology, 21:13–19.

Carruthers, P. and Smith, Peter, K., editors (1996). Theories

of Theories of Mind. Cambridge, U.K.: Cambridge

University Press.

Chi, M., Slotta, J. D., and de Leeuw, N. (1994). From things

to processes: A theory of conceptual change. Learn-

ing and Instruction, 4:27–43.

CSEDU 2016 - 8th International Conference on Computer Supported Education

504

Clayton, N. S., Emery, N. J., and Dickinson, A. (2006).

The rationality of animal memory: Complex caching

strategies of western scrub jays. In Hurley, S. and

Nudds, M., editors, Rational Animals?, pages 197–

216. Oxford, UK: Oxford University Press.

Clocksin, W. F. and Mellish, C. S. (1981). Programming in

Prolog. Berlin, Heidelberg, New York etc.: Springer.

Davis, B., Sumara, D., and Luce-Kapler, R. (2000). En-

gaging Minds: Learning and Teaching in a Complex

World. Mahwah, NJ, USA: Lawrence Erlbaum.

De Bra, P., Kobsa, A., and Chin, D., editors (2010). User

Modeling, Adaptation, and Personalization. Proc.

18th International Conf., UMAP 2010, Big Island, HI,

USA, June 2010. Number 6075 in LNCS. Springer.

Dimitrova, V., Kuﬂik, T., Chin, D., Ricci, F., Dolog, P., and

Houben, G.-J., editors (2014). User Modeling, Adap-

tation, and Personalization. Proc. 22nd International

Conf., UMAP 2014, Aalborg, Denmark, July 2014.

Number 8538 in LNCS. Springer.

diSessa, A. A. and Sherin, B. L. (1998). What changes in

conceptual change? International Journal of Science

Education, 20(10):1155–1191.

Doyle, A. C. (1915). The adventure of the dancing men. In

The Return of Sherlock Holmes. London: Smith, Elder

& Co., 3.1 edition.

Egenfeldt-Nielsen, S. (2007). Educational Potential of

Computer Games. Continuum Studies in Education.

London, New Delhi, New York, Sydney: Bloomsbury

Publishing, formerly Continuum Intl. Publ. Group.

Emery, N. J. (2004). Are corvids ‘feathered apes’? Cogni-

tive evolution in crows, jays, rooks and jackdaws. In

Watanabe, S., editor, Comparative Analysis of Minds,

pages 181–213. Tokyo: Keio University Press.

Emery, N. J. and Clayton, N. S. (2009). Comparative social

cognition. Annual Review of Psychology, 60:87–113.

Emery, N. J., Dally, J. M., and Clayton, N. S. (2004). West-

ern scrub-jays (Aphelocoma californica) use cognitive

strategies to protect their caches from thieving con-

speciﬁcs. Animal Cognition, 7:37–43.

Felder, R. M. and Silverman, L. K. (1988). Learning and

teaching styles in engineering education. Engineering

Education, 78(7):674–681.

Fujima, J. and Jantke, K. P. (2012). The potential of the

direct execution paradigm: Toward the exploitation

of media technologies for exploratory learning of ab-

stract content. In Urban, B. and M

usebeck, P., edi-

tors, eLearning Baltics 2012: Proceedings of the 5th

International eLBa Science Conference, pages 33–42.

Fraunhofer Verlag.

Gaudl, S., Jantke, K. P., and Woelfert, C. (2009). The

good, the bad and the ugly: Short stories in short

game play. In Iurgel, I., Zagalo, N., and Petta, P.,

editors, Proceedings of the 2nd International Confer-

ence on Digital Storytelling, Dec. 9-11, 2009, Erfurt,

Germany, number 5915 in LNCS, pages 127–133.

Springer-Verlag Berlin Heidelberg 2009.

Goldman, A. I. (2006). Simulating Minds: The Philosophy,

Psychology, and Neuroscience of Mindreading. New

York, NY: Oxford University Press.

Grieser, G. (2008). Reﬂective inductive inference of recur-

sive ffnctions. Theoretical Computer Science, 397(1-

3):57–69.

Grieser, G. and Jantke, K. P. (1995). Ans

atze zur

Reﬂexion in der Induktiven Inferenz. Studie der

Forschungsgruppe Algorithmisches Lernen, HTWK

Leipzig (FH), FB Informatik, Mathematik & Natur-

wissenschaften. Studie #02/95, Version 1.0.

Hammer, D. (1996). Misconceptions or P-prims: How may

alternative perspectives of cognitive structure inﬂu-

ence instructional perceptions and intentions? The

Journal of the Learning Sciences, 5(2):97–127.

Hopcroft, J. E. and Ullman, J. D. (1979). Introduction

to Automata Theory, Languages, and Computation.

Boston: Addison-Wesley.

Houben, Geert-Jan ANDMcCalla, G., Pianesi, F., and Zan-

canaro, M., editors (2009). User Modeling, Adap-

tation, and Personalization. Proc. 17th International

Conf., UMAP 2009, Trento, Italy, June 2009. Number

5535 in LNCS. Springer.

Jain, S., Osherson, D., Royer, J. S., and Sharma, A. (1999).

Systems That Learn. Cambridge, MA, USA: The MIT

Press.

Jantke, K. P. (1994). Towards reﬂecting inductive in-

ference machines. GOSLER Report 24/93, HTWK

Leipzig (FH), FB Informatik, Mathematik & Natur-

wissenschaften.

Jantke, K. P. (1995). Reﬂecting and self-conﬁdent induc-

tive inference machines. In Jantke, K. P., Shinohara,

T., and Zeugmann, T., editors, Proc. 6th International

Workshop on Algorithmic Learning Theory (ALT’95),

October 18–20, 1995, Fukuoka, Japan, volume 997 of

LNAI, pages 282–297. Springer-Verlag.

Jantke, K. P. (2010). The Gorge approach. Digital game-

control and play for playfully developing technology

competence. In Cordeiro, J., Shishkov, B., Verbraeck,

A., and Helfert, M., editors, CSEDU 2010. 2nd Inter-

national Conference on computer Supported Educa-

tion, Proc., Vol. 1, Valencia, Spain, April 7-10, 2010,

pages 411–414. INSTICC.

Jantke, K. P., Beick, H.-R., Brovko, Y., and Drefahl, S.

(2013). Reﬁnement of adaptivity by reﬂection. In

Yetongnon, K., Dipanda, A., and Chbeir, R., editors,

9th International Conference on Signal Image Tech-

nology & Internet-based Systems, Dec. 2-5, 2013, Ky-

oto, Japan, pages 309–316. IEEE.

Jantke, K. P., Grieser, G., Lange, S., and Memmel,

M. (2004). DaMiT: Data Mining lernen und

lehren. In Lernen, Wissensentdeckung und Adap-

tivit

at (LWA-2004), Fachgruppentreffen Maschinelles

Lernen (FGML), 4.-6. Oktober 2004, Berlin.

Jantke, K. P., Hoppe, I., Lengyel, D., and Neumann, A.

(2010). Time to play Gorge – Time to learn AI: A

qualitative study. In Hambach, S., Martens, A., Tavan-

garian, D., and Urban, U., editors, eLearning Baltics

2010, Proc. 3rd Intl. eLBa Science Conference, pages

99–110. Fraunhofer Verlag.

Jantke, K. P. and Hume, T. (2015). Effective learning

through meaning construction in digital role playing

games. In IEEE International Conference on Con-

Next Generation Learner Modeling by Theory of Mind Model Induction

505

sumer Electronics (ICCE), Las Vegas, January 2015,

pages 686–689. IEEE Consumer Electronics Society.

Jantke, K. P. and Knauf, R. (2005). Didactic design through

storyboarding: Standard concepts for standard tools.

In Proc. 4th Intl. Symp. on Information and Com-

munication Technologies, Cape Town, South Africa,

January 3–6, 2005, pages 20–25. Computer Science

Press, Trinity College Dublin, Ireland.

Jantke, K. P., Memmel, M., Rostanin, O., Thalheim, B., and

Tschiedel, B. (2003). Decision support by learning-

on-demand. In CAiSE Workshop 2003, Klagenfurt,

Osterreich.

Jantke, K. P. and Schulz, A. (2011). Adaptivity in moo-

dle beyond the limits of adaptivity in moodle. In Ver-

braeck, A., Helfert, M., Cordeiro, J., and Shishkov,

B., editors, 3rd International Conference on Com-

puter Supported Education, CSEDU 2011, May 6-8,

2011, Noordwijkerhout, The Netherlands, pages 418–

421. INSTICC.

Jung, C. G. (1921). Psychologische Typen. Z

urich: Rascher

Verlag.

Kassak, O., Kompan, M., and Bielikova, M. (2015). User

preference modeling by global and individual weights

for personalized recommendation. Acta Polytechnica

Hungarica, 12(8):27–41.

Kayoko, I. and Hatano, G. (2013). Conceptual change in

ıve biology. In (Vosniadou, 2013b), pages 195–219.

Knauf, R., Sakurai, Y., Tsuruta, S., and Jantke, K. P. (2010).

Modeling didactic knowledge by storyboarding. Jour-

nal of Educational Computing Research, 42(4):355–

383.

Kolb, D. and Fry, R. E. (1975). Towards an applied the-

ory of experiential learning. In Cooper, C. L., editor,

Theories of Group Processes, pages 33–58.

Kolb, D. A. (1984). Experiential Learning: Experience as

the Source of Learning and Development. Upper Sad-

dle River, New Jersey: Prentice Hall.

Konstan, J. A., Conejo, R., Marzo, J. L., and Oliver, N.,

editors (2011). User Modeling, Adaptation, and Per-

sonalization. Proc. 19th International Conf., UMAP

2011, Girona, Spain, July 2011. Number 6787 in

LNCS. Springer.

Koster, R. (2005). A Theory of Fun for Game Design.

Scottsdale, AZ, USA: Paraglyph Press.

Krebs, J. and Jantke, K. P. (2014). Methods and tech-

nologies for wrapping educational theory into serious

games. In Zvacek, S., Restivo, M. T., Uhomoibhi, J.,

and Helfert, M., editors, Proceedings of the 6th Inter-

national Conference on Computer Supported Educa-

tion, CSEDU 2014, Barcelona, Spain, May 1-3, 2014,

pages 497–502. SCITEPRESS.

Leinhardt, G. and Ravi, A. (2013). Changing historical con-

ception of history. In (Vosniadou, 2013b), pages 343–

359.

MacBeth, D. (2000). On an apparatus for conceptual

change. Science Education, 84(2):228–264.

Masthoff, J., Mobasher, B., Desmarais, M. C., and Nkam-

bou, R., editors (2012). User Modeling, Adaptation,

and Personalization. Proc. 20th International Conf.,

UMAP 2012, Montreal, Canada, July 2012. Number

7379 in LNCS. Springer.

Mauer, M. C. (2012). Das Konstrukt der Theory of Mind

bei Erwachsenen. PhD thesis, LMU M

unchen, Fak.

Psychologie und P

adagogik.

Phillips, W. (1991). Earth science misconceptions. The

Science Teacher, 58(2):21.

Popper, K. (1934). Logik der Forschung. T

ubingen.

Ricci, F., Bontcheva, K., Conlan, O., and Lawless, S., ed-

itors (2015). User Modeling, Adaptation, and Per-

sonalization. Proc. 23rd International Conf., UMAP

2015, Dublin, Ireland, June/July 2015. Number 9146

in LNCS. Springer.

Richter, M. M. (1978). Logikkalk

ule. Stuttgart: Teubner.

Rogers jr., H. (1967). Theory of Recursive Functions

and Effective Computability. Hoboken, NJ, USA:

McGraw-Hill.

Schewe, K.-D., Thalheim, B., and Tretiakov, A. (2007).

Formalisation of user preferences, obligations and

rights. In Kaschek, R. H., editor, Intelligent Assis-

tent Systems.Concepts, Techniques and Technologies,

chapter VI, pages 114–143. Hershey, London, Mel-

bourne, Singapore: Idea Group Publ.

Schmidt, B. (2014). Theory of Mind Player Model-

ing. Konzeptentwicklung, Implementierung und Er-

probung mit logischer Programmierung. Bachelor

thesis, FH Erfurt – University of Applied Sciences,

Angewandte Informatik.

Sipser, M. (1997). Introduction to the Theory of Computa-

tion. Boston, MA, USA: PWS Publ. Co.

Smith, A., Min, W., Mott, B. W., and Lester, J. C.

(2015). Diagrammatic student models: Modeling stu-

dent drawing performance with deep learning. In

Ricci, F., Bontcheva, K., Conlan, O., and Lawless,

S., editors, User Modeling, Adaptation, and Personal-

ization. Proc. 23rd International Conf., UMAP 2015,

Dublin, Ireland, June/July 2015, pages 216–227.

Specht, M. and Weber, G. (1997). Kognitive Lernermodel-

lierung. Kognitionswissenschaft, 6(4):165–176.

Sterling, L. and Shapiro, E. (1986). The Art of Prolog. Cam-

bridge, MA, USA: The MIT Press.

Suddendorf, T. (2007). The evolution of foresight: What is

mental time travel and is it unique to humans? Behav-

ioral and Brain Sciences, 30:299–313.

Thagard, P. (2012). The Cognitive Science of Science: Ex-

planation, Discovery, and Cognitive Change. Cam-

bridge, MA, USA: The MIT Press.

Thagard, P. (2013). Conceptual change in the history of sci-

ence: Life, mind, and disease. In (Vosniadou, 2013b),

pages 360–374.

Vosniadou, S. (2013a). Conceptual change in learning and

instruction: The framework theory approach. In (Vos-

niadou, 2013b), pages 1–30.

Vosniadou, S., editor (2013b). International Handbook

of Research on Conceptual Change. Second Edition.

New York, Milton Park: Routledge.

Wisniak, J. (2004). Phlogistion: The rise and fall of a the-

ory. Indian Journal of Chemical Technology, 11:732–

743.

CSEDU 2016 - 8th International Conference on Computer Supported Education

506