Numerous
languages
exhibit
shared
characteristics,
especially
in
morphological
features.
For
instance,
Arabic
and
Russian
both
belong
to
the
fusional
language
category.
The
question
arises:
Do
such
common
traits
influence
comprehension
across
diverse
linguistic
backgrounds?
This
study
explores
possibility
of
transferring
skills
a
zero-shot
scenario.
Specifically,
we
demonstrate
that
training
models
on
other
can
enhance
Arabic,
as
evidenced
by
our
evaluations
three
key
tasks:
natural
inference,
answering,
named
entity
recognition.
Our
experiments
reveal
certain
morphologically
rich
(MRLs),
Russian,
display
similarities
when
assessed
context,
particularly
tasks
like
answering
inference.
However,
this
similarity
is
less
pronounced
We
describe
the
findings
of
fourth
Nuanced
Arabic
Dialect
Identification
Shared
Task
(NADI
2023).
The
objective
NADI
is
to
help
advance
state-of-the-art
NLP
by
creating
opportunities
for
teams
researchers
collaboratively
compete
under
standardized
conditions.
It
does
so
with
a
focus
on
dialects,
offering
novel
datasets
and
defining
subtasks
that
allow
meaningful
comparisons
between
different
approaches.
2023
targeted
both
dialect
identification
(Subtask1)
dialect-to-MSA
machine
translation
(Subtask
2
Subtask
3).
A
total
58
unique
registered
shared
task,
whom
18
have
participated
(with
76
valid
submissions
during
test
phase).
Among
these,
16
in
1,
5
2,
3
3.
winning
achieved
87.27
F1
14.76
Bleu
21.10
3,
respectively.
Results
show
all
three
remain
challenging,
thereby
motivating
future
work
this
area.
methods
employed
participating
briefly
offer
an
outlook
NADI.
Engineering Technology & Applied Science Research,
Journal Year:
2025,
Volume and Issue:
15(2), P. 20737 - 20742
Published: April 3, 2025
Large
Language
Models
(LLMs)
have
recently
demonstrated
outstanding
performance
in
a
variety
of
Natural
Processing
(NLP)
tasks.
Although
many
LLMs
been
developed,
only
few
models
evaluated
the
context
Arabic
language,
with
significant
focus
on
ChatGPT
model.
This
study
assessed
three
two
NLP
tasks:
sentiment
analysis
and
machine
translation.
The
capabilities
LLaMA,
Mixtral,
Gemma
under
zero-
few-shot
learning
were
investigated,
their
was
compared
against
State-Of-The-Art
(SOTA)
models.
experimental
results
showed
that,
among
models,
LLaMA
tends
to
better
comprehension
abilities
for
outperforming
Mixtral
both
However,
except
Arabic-to-English
translation,
where
outperforms
transformer
model
by
4
BLEU
points,
all
cases,
fell
behind
that
SOTA
Disinformation
involves
the
dissemination
of
incomplete,
inaccurate,
or
misleading
information;
it
has
objective,
goal,
purpose
deliberately
intentionally
lying
to
others
aboutthe
truth.
The
spread
disinformative
information
on
social
media
serious
implications,
and
causes
concern
among
internet
users
in
different
aspects.
Automatic
classification
models
are
required
detect
posts
media,
especially
Twitter.
In
this
article,
DistilBERT
multilingual
model
was
fine-tuned
classify
tweets
either
as
dis-informative
not
Subtask
2A
ArAIEval
shared
task.
system
outperformed
baseline
achieved
F1
micro
87%
macro
80%.
Our
ranked
11
compared
with
all
participants.
