In narrative AIED, we employ a library of narrative templates derived from classic literature, including Aesop’s fables, the Panchatantra, and Grimm's fairy tales.
These templates serve as FILL-IN-THE-SLOT FRAMEWORKS for story generation, allowing the LLM to customize narratives based on:
Pupil's Profile:: Personal interests, preferences, and learning needs.
Knowledge State:: Current mastery level in various subjects.
Learning Objectives:: Specific curricular goals targeted during the session.
In
narrative
AIED
''
we
employ
a
library
of
narrative
templates
derived
from
classic
literature
''
including
Aesop
’
s
fables
''
the
Panchatantra
''
and
Grimm
'
s
fairy
tales
. <
div
><
br
>
div
><
div
>
These
templates
serve
as
FILL
-
IN
-
THE
-
SLOT
FRAMEWORKS
for
story
generation
''
allowing
the
LLM
to
customize
narratives
based
on
:<
p
class
=
„
fragment
“
>
Pupil
'
s
Profile
::
Personal
interests
''
preferences
''
and
learning
needs
.
p
><
p
class
=
„
fragment
“
>
Knowledge
State
::
Current
mastery
level
in
various
subjects
.
p
><
p
class
=
„
fragment
“
>
Learning
Objectives
::
Specific
curricular
goals
targeted
during
the
session
.
p
>
div
>
A human-machine peer learning curriculum (i.e., a HMPL-C) is a planned sequence of educational instructions–i.e., a curriculum–which involves:
1. At least one human learner G, H, I, ... which gradually develops her/his/their skill Γ.
2. At least one artificial learner a, b, c, ... which gradually develops its/her/his/their skill σ.
3. Activities by means of which G (resp. H, I, etc.) develops her/his/their skill Γ, which directly involve knowledge and competence exhibited by a (resp. b, c, etc.).
4. Activities by means of which a (resp. b, c, etc.) develops her/his/their skill σ, which directly involve knowledge and competence exhibited by G (resp. H, I, etc.).
1162
A
human
-
machine
peer
learning
curriculum
(
i
.
e
.''
a
HMPL
-
C
)
is
a
planned
sequence
of
educational
instructions
–
i
.
e
.''
a
curriculum
–
which
involves
:<
br
>
1
.
At
least
one
human
learner
G
''
H
''
I
''
...
which
gradually
develops
her
/
his
/
their
skill
Γ
.<
br
>
2
.
At
least
one
artificial
learner
a
''
b
''
c
''
...
which
gradually
develops
its
/
her
/
his
/
their
skill
σ
.<
br
>
3
.
Activities
by
means
of
which
G
(
resp
.
H
''
I
''
etc
.)
develops
her
/
his
/
their
skill
Γ
''
which
directly
involve
knowledge
and
competence
exhibited
by
a
(
resp
.
b
''
c
''
etc
.).<
br
>
4
.
Activities
by
means
of
which
a
(
resp
.
b
''
c
''
etc
.)
develops
her
/
his
/
their
skill
σ
''
which
directly
involve
knowledge
and
competence
exhibited
by
G
(
resp
.
H
''
I
''
etc
.).<
br
>
The concept of Non-Player Educators (NPEs) extends the idea of AI agents serving as interactive educational companions. In the NP system:
Role Modeling: NPEs can embody various characters or personas within stories, modeling behaviors and conveying lessons.
Diversity of Perspectives: Multiple NPEs can provide different viewpoints, enriching the educational content.
Consistency and Patience: As AI-driven entities, NPEs offer consistent interactions and infinite patience, accommodating the child’s pace.
In technical terms, NPEs are realized by means of individual Low-Rank Adaptors (LoRA) which are loaded and unloaded atop the underlying language model.
1605
<
p
>
The
concept
of
<
em
>
Non
-
Player
Educators
em
>
(
NPEs
)
extends
the
idea
of
AI
agents
serving
as
interactive
educational
companions
.
In
the
NP
system
:
p
><
p
class
=
„
fragment
“
><
strong
>
Role
Modeling
:
strong
>
NPEs
can
embody
various
characters
or
personas
within
stories
''
modeling
behaviors
and
conveying
lessons
.
p
><
p
class
=
„
fragment
“
><
strong
>
Diversity
of
Perspectives
:
strong
>
Multiple
NPEs
can
provide
different
viewpoints
''
enriching
the
educational
content
.
p
><
p
class
=
„
fragment
“
><
strong
>
Consistency
and
Patience
:
strong
>
As
AI
-
driven
entities
''
NPEs
offer
consistent
interactions
and
infinite
patience
''
accommodating
the
child
’
s
pace
.
p
><
p
class
=
„
fragment
“
>
In
technical
terms
''
NPEs
are
realized
by
means
of
individual
Low
-
Rank
Adaptors
(
LoRA
)
which
are
loaded
and
unloaded
atop
the
underlying
language
model
.
p
>
By integrating speech-to-text (STT), text-to-speech (TTS), mid-sized large-language models (MLMs) and sufficient and necessary knowledge base stored in the vector database, the NP provides a personalized and interactive learning experience.
It emphasizes the traditional educational practice of storytelling, enhanced by modern AI capabilities, to promote basic literacy, arithmetic, musical skills and uncorruptable personality.
1593
<
div
>
By
integrating
speech
-
to
-
text
(
STT
)''
text
-
to
-
speech
(
TTS
)''
mid
-
sized
large
-
language
models
(
MLMs
)
and
sufficient
and
necessary
knowledge
base
stored
in
the
vector
database
''
the
NP
provides
a
personalized
and
interactive
learning
experience
.
div
><
div
><
br
>
div
><
div
>
It
emphasizes
the
traditional
educational
practice
of
storytelling
''
enhanced
by
modern
AI
capabilities
''
to
promote
basic
literacy
''
arithmetic
''
musical
skills
and
uncorruptable
personality
.<
br
>
div
>
The first property of the Primer is that it is either book-like—that is, it looks like a classic analog book—or "embooked," meaning it is physically embedded into a classic analog book. This first characteristic implies that the Primer:
uses reflected, non-emitted light (i.e., E-Ink rather than LED or OLED) to display content
ideally consists of multiple pages that can be read on both sides and are bound together using special binding techniques,
is equipped with both internal and external content encoding on the front and back covers and can include an additional layer of protection (e.g., a protective cover),
has a standardized format (e.g., A5) that allows it to be easily carried in existing wearables (e.g., school bags) and stored or archived in existing furniture (e.g., bookshelves)
1600
The
first
property
of
the
Primer
is
that
it
is
either
book
-
like
—
that
is
''
it
looks
like
a
classic
analog
book
—
or
„
embooked
''
“
meaning
it
is
physically
embedded
into
a
classic
analog
book
.
This
first
characteristic
implies
that
the
Primer
:<
p
class
=
„
fragment
“
>
uses
reflected
''
non
-
emitted
light
(
i
.
e
.''
E
-
Ink
rather
than
LED
or
OLED
)
to
display
content
p
><
p
class
=
„
fragment
“
>
ideally
consists
of
multiple
pages
that
can
be
read
on
both
sides
and
are
bound
together
using
special
binding
techniques
,
p
><
p
class
=
„
fragment
“
>
is
equipped
with
both
internal
and
external
content
encoding
on
the
front
and
back
covers
and
can
include
an
additional
layer
of
protection
(
e
.
g
.''
a
protective
cover
),
p
><
p
class
=
„
fragment
“
>
has
a
standardized
format
(
e
.
g
.''
A5
)
that
allows
it
to
be
easily
carried
in
existing
wearables
(
e
.
g
.''
school
bags
)
and
stored
or
archived
in
existing
furniture
(
e
.
g
.''
bookshelves
)
p
>
Linux
open-source voice identification (Speechbrain), speech-to-text (DeepSpeech 2, whisper), speech-to-text (XTTS2, bark) and text-to-image (Stable Diffusion)
small language models (Llama-3-SauerkrautLM-8b-Instruct, udkai/Turdus, Ministral...)
ChromaDB vector database (embeddings through NV-Embed-v1)
on-demand loading of LoRA adapter through LoRAx (loraexchange.ai)
Kastalia Knowledge Management System for lesson & learning graph management
https://github.com/hromi/personal_Primer https://github.com/hromi/primer-backend
1599
<
p
class
=
„
fragment
“
>
Linux
p
><
p
class
=
„
fragment
“
>
open
-
source
voice
identification
(
Speechbrain
)''
speech
-
to
-
text
(
DeepSpeech
2
''
whisper
)''
speech
-
to
-
text
(
XTTS2
''
bark
)
and
text
-
to
-
image
(
Stable
Diffusion
)
p
><
p
class
=
„
fragment
“
>
small
language
models
(
Llama
-
3
-
SauerkrautLM
-
8b
-
Instruct
''
udkai
/
Turdus
''
Ministral
...)
p
><
p
class
=
„
fragment
“
>
ChromaDB
vector
database
(
embeddings
through
NV
-
Embed
-
v1
)
p
><
p
class
=
„
fragment
“
>
on
-
demand
loading
of
LoRA
adapter
through
LoRAx
(
loraexchange
.
ai
)
p
><
p
class
=
„
fragment
“
>
Kastalia
Knowledge
Management
System
for
lesson
&
learning
graph
management
p
><
p
class
=
„
fragment
“
>
https
://
github
.
com
/
hromi
/
personal
Primer
https
://
github
.
com
/
hromi
/
primer
-
backend
p
>
book-like; voluminous; modular; unique; adaptable; robust; bilateral; environmentally aware; circadian; moody; cooperative; behavior-oriented; habit-disrupting; playful and funny; mnemonic; multimodal; speech-based; narrative; cybertextual and encyclopedic; online-offline; protected; eye-to-eye; avatarized
(c.f. Hromada, 2018; Hromada, Seidler & Kapanadze, 2020; Hromada & Kim, 2023 etc.)
1611
<
div
>
book
-
like
;
voluminous
;
modular
;
unique
;
adaptable
;
robust
;
bilateral
;
environmentally
aware
;
circadian
;
moody
;
cooperative
;
behavior
-
oriented
;
habit
-
disrupting
;
playful
and
funny
;
mnemonic
;
multimodal
;
speech
-
based
;
narrative
;
cybertextual
and
encyclopedic
;
online
-
offline
;
protected
;
eye
-
to
-
eye
;
avatarized
div
><
div
><
br
>
div
><
div
>(
c
.
f
.
Hromada
''
2018
;
Hromada
''
Seidler
&
Kapanadze
''
2020
;
Hromada
&
Kim
''
2023
etc
.)
<
br
>
div
>
form-factor: A5 size
computational core: NVIDIA Orin
main input modalities: microphone, gesture sensors,
main output modalities: speakers, electrophoretic (e-ink) display(s)
other peripherals attachable through standard USB-C and 40-pin GPIO connectors
NVMe disk, WLAN cards (both fully operational)
power-related: solar-panel positive
our current main challenge: thermal regulation
1598
<
p
class
=
„
fragment
“
>
form
-
factor
:
A5
size
p
><
p
class
=
„
fragment
“
>
computational
core
:
NVIDIA
Orin
p
><
p
class
=
„
fragment
“
>
main
input
modalities
:
microphone
''
gesture
sensors
,
p
><
p
class
=
„
fragment
“
>
main
output
modalities
:
speakers
''
electrophoretic
(
e
-
ink
)
display
(
s
)
p
><
p
class
=
„
fragment
“
>
other
peripherals
attachable
through
standard
USB
-
C
and
40
-
pin
GPIO
connectors
p
><
p
class
=
„
fragment
“
>
NVMe
disk
''
WLAN
cards
(
both
fully
operational
)
p
><
p
class
=
„
fragment
“
>
power
-
related
:
solar
-
panel
positive
p
><
p
class
=
„
fragment
“
>
our
current
main
challenge
:
thermal
regulation
p
>
spectacular things are already happening in "open source" branch of AIED
with USA gradually becoming the prey of the dark side, immediate deployment of "walled garden" approaches is of utmost importance
all "bricks " to build Your educational "cathedral" are available out there (GitHub, Huggingface) and ready to serve
the future will be more weird than a dream and the key to that dream is ...
... education
1612
<
p
class
=
„
fragment
“
>
spectacular
things
are
already
happening
in
„
open
source
“
branch
of
AIED
p
><
p
class
=
„
fragment
“
>
with
USA
gradually
becoming
the
prey
of
the
dark
side
''
immediate
deployment
of
„
walled
garden
“
approaches
is
of
utmost
importance
p
><
p
class
=
„
fragment
“
>
all
„
bricks
“
to
build
Your
educational
„
cathedral
“
are
available
out
there
(
GitHub
''
Huggingface
)
and
ready
to
serve
p
><
p
class
=
„
fragment
“
>
the
future
will
be
more
weird
than
a
dream
and
the
key
to
that
dream
is
...
p
><
p
class
=
„
fragment
“
>...
education
p
>
1613
Mail
:
daniel
@
udk
-
berlin
.
de
/
daniel
@
wizzion
.
com
<
br
/>
Matrix
:
@
d
.
hromada
:
medienhaus
.
udk
-
berlin
.
de
<
br
/>
LinkedIn
(<
a
href
=
„
https
://
www
.
linkedin
.
com
/
in
/
dhromada
“
target
=
„
blank
“
rel
=
„
noopener
“
>@
dhromada
a
>)
/
ResearchGate
(<
a
href
=
„
https
://
www
.
researchgate
.
net
/
profile
/
Daniel
-
Hromada
“
target
=
„
blank
“
rel
=
„
noopener
“
>
Daniel
-
Hromada
a
>)
/
Instagram
(<
a
href
=
„
https
://
www
.
instagram
.
com
/
DigiEduBerlin
/
“
target
=
„
blank
“
rel
=
„
noopener
“
>@
DigiEduBerlin
a
>)
×
- You start the recording with a start button (You can click on it, touch it, or simply move Your finger/cursor over it).
- Subsequently, You move the finger/cursor over the first syllable/word. You pronounce the segment only when its background has green color.
- You gradually progress from syllable to syllable and segment to segment. You always start pronouncing certain segment only when it is green.
- Note that accentuated & long syllables are marked with blue color, while short/non-accentuated syllables are marked with green color.
- After You are done with the last syllable You move Your finger towards the stop button. Then you can playback the whole recording, when you are satisfied you click on "Upload" and the next text appears.
×
In order to be fully compliant with European data-protection Law, we need Your explicit consent regarding use of Your voice data. Please choose one among following consent types:
- Do not upload :: You do not give us Your consent. Thus, no data will be uploaded from Your browser to our server. But You can still use the interface for testing purposes.
- Only speech-to-text-models :: Your recordings will become part of the corpus from which automatic speech recognition (ASR) models will be trained. Corpus itself will not be published but the final model will be published.
- Only text-to-speech :: Similar to previous option but this time, the resulting model will not be used for ASR but for synthesis of artificial voices.
- STT & TTS :: Both ASR and voice synthesis models could be trained from datasets containing Your recordings. Again, the recordings themselves will not be published.
- Public Dataset :: Your recordings will become part of a publicly available dataset. This is the most permissive option.
Note that in all cases, Your recordings will be anonymized and asides voluntary gender / age / zodiac sign / mother language information no metadata is collected.