International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
The Role of Machine Learning in Enhancing Language
translation
S.sreeja1 , [Link] raj yadav2 , [Link] 3,J. Rochirmai4,D. Nihal5 ,[Link] reddy 6
1
Associate Professor, AI&ML Department, Sreyas Institute of Engineering and Technology ,Hyderabad,
India,Sreeja.s@[Link]
2
AI&ML Department, Sreyas Institute of Engineering and Technology, Hyderabad, India, puneethraaaj@[Link]
3
AI&ML Department, Sreyas Institute of Engineering and Technology, Hyderabad, India,vigneshutkuri@[Link]
4
AI&ML Department, Sreyas Institute of Engineering and Technology, Hyderabad, India,rochirmaijeedula@[Link]
5
AI&ML Department, Sreyas Institute of Engineering and Technology, Hyderabad, India,nihaldonkena11@[Link]
6
AI&ML Department, Sreyas Institute of Engineering and Technology, Hyderabad, India,srideepreddy74@[Link]
ABSTRACT
Machine learning has changed dialect interpretation by making worldwide communication less
demanding and more successful. Conventional rule-based frameworks frequently battled to get it
the setting and meaning behind words, driving to destitute interpretations. In differentiate, machine
learning particularly Neural Machine Interpretation (NMT) learns from huge sums of bilingual
content to supply more precise and common interpretations . This demonstrate supplanted more
seasoned strategies like repetitive and convolutional systems by utilizing self-attention. Coming
about in superior and more significant [Link] profound learning models, counting
transformers and repetitive neural systems, can get it dialect stream and recognize common
expressions or expressions that do not decipher straightforwardly. This makes interpretations of
text more normal and realistic whether they’re used in a daily coversations or for formal purpose
. However the machine interpretation still faces some challenges , it can have problems regarding
social implications and complex informal expressions , which may lead to faulty or incorrect
translations at times . To overcome this analysts are building smarter models and taking feedback
and criticism to work more efficiently to offer a good and real time interpretation with no
challenges.
Keywords - Large language model , Neural network ,Transformer architecture ,Deep Learning ,
Accuracy.
[Link]
1
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
The capability to seamlessly communicate across languages has been a driving force behind mortal
progress, fostering artistic exchange, profitable growth, and scientific collaboration. For centuries,
language restatement reckoned primarily on mortal moxie, a process innately constrained by time,
cost, and the vacuity of professed translators. This limitation has disproportionately affected
communication involving less- resourced languages, where the failure of trained linguists and
resemblant corpora creates significant hurdles. The emergence of machine literacy( ML), still, has
dramatically altered this geography, offering the eventuality to overcome these limitations and
homogenize access to high- quality restatement services [Link] attempts at machine
restatement, frequently producing inaccurate and unnatural restatements. While statistical machine
restatement( SMT) represented a significant step forward, it still faced challenges in handling the
complications of syntax and semantics. The arrival of neural machine restatement( NMT),
particularly with the development of motor infrastructures, NMT models, powered by deep literacy
ways, demonstrated an unknown capability to learn complex patterns in language data, performing
in substantial advancements in restatement delicacy, ignorance, and overall quality. The success
of motor- grounded models is apparent in their wide relinquishment and superior performance
across a wide and different range of language [Link] after these advancements, challenges
still remain. The performance of the neural machine translation systems heavily depend on the
high- quality resemblant dataset, a resource frequently lacking for low- resource languages.
Experimenters are laboriously exploring colorful strategies to address this data failure, including
transfer literacy, data addition, and the application of multilingual models. likewise, the evaluation
of machine restatement systems continues to be a complex undertaking, with traditional criteria
frequently failing to completely capture the craft of mortal judgment. The development of further
sophisticated evaluation criteria that better align with mortal perception is an ongoing area of
[Link] exploration will give us an insight disquisition about the transformative part of the
machine literacy in increasing language restatement. We'll examine the elaboration of ML ways
in this field, from early rule- grounded systems to the current dominance of motor- grounded
models. We'll dissect the colorful strategies employed to address the challenges faced by using
low- resource languages , critically assess strengths and limitations of being evaluation criteria .
Eventually, we will bandy the unborn directions of ML- driven language restatement, considering
the eventuality of arising technologies and their counteraccusations for global communication and
understanding.
2. LITERATURE SURVEY
Surge in use of translators Kolhar et all [1] shares an insight on the advancements of a translation
machine system aimed at enhancing classroom learning experiences in real world scenarios . This
system focused on assisting non-native Arabic-speaking teachers in effectively communicate with
native Arabic-speaking students by providing real-time translations during lectures where teachers
2
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
are interacting with students . Integrating with digital learning and devices, this machine translation
system allows for effective translation of PowerPoint presentations and learning modules ,
improving advanced student engagement and better understanding of students. An demo
experiment was conducted involving 25 students demonstrated that the use of this technology
improved their overall comprehension of technical subjects taught in English and enhanced overall
learning outcomes for efficient teaching. Students reportedly stated benefits in vocabulary
improvement and better subject understanding, particularly helped them when working on
assignments and homework. The study highlights the potential outcomes of merging AI-based
translation models in the educational ecosystem to bridge language barriers and foster better
academic performance between the students .
Continuing with this Genovese et all [2] gives a systematic review of its role in the process of
language translation and interpretation" explores how effective can be AI-based translation models
can be used in medical field . The study analyzed over nine reports focusing on the most used
platforms . These findings indicate and give a clear conclusion that AI translation tools can thereby
provide more detail translations resulting in easy, simple and effective communication, particularly
we translate from English to European languages. However, accuracy decreases when translating
into English or involving non-European languages. The accuracy scores ranged from 76.7% to
96.7%, with patients generally more satisfied and happy with the results (84–96.6%) than
clinicians (53.8–86.7%). Clinicians expressed their dissatisfaction about the authenticity and
standard of AI translations over the long term of the usage of this technology, leading to use it
primarily as a last-resort option or for the brief interactions when human translators are unavailable
in that situations. The study concludes that while AI shows goood promise in clinical translation,
but it should complement, and not replace, human translation services still should be considered
to ensure high-quality patient care.
Additionally , went further exploring David B. Sawyer et all [3] this explores the growing
influence and possibilties of machine learning (ML) on translation practices in real world scenarios
, where the main objective to bridge the gap between human expertise and technological
devolopment and advancement with time. This book highlights on various things where the first
part focuses on how humans and machines approach translation, discussing various techniques
like word embeddings and context-based predictions. The second part dives into key ML tasks
relevant to translation aspect, such as automated translation, quality assessment, and NLP
applications and many more. The final part lays emphasis on the importance of data, highlighting
how translators contribute by creating and managing linguistic datasets that power and power up
ML tools. Throughout this book, the authors build a human-centered perspective, claiming that
ML can enhance translation workflows, human translators remain essential service in daily
applications . They help in providing better communication between translators, developers, and
users of ML systems, aiming to create collaborative environments where the technology
3
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
complements in improving human skill rather than replacing it completely . This book ultimately
encourages translators to use and embrace ML tools as a valuable integration.
In another interesting devlopment Akbar Khanan et all [4] explores the possibilities of how
artificial intelligence has transformed and impacted the field of language translation. It discusses
the evolution from traditional rule-based methods to modern advancments in the domain of AI-
driven approaches, highlighting the benefits as well as the challenges associated with these
advancements. This paper also explores the limitations and implications of AI in translation for
various fileds and sectors .
In this article Aishwarya R et all [5] discusses the improvement of translation systems that
converts English text to Hindi and detect the language of the given input text. The authors have
implemented two different Long Short-Term Memory (LSTM) models using various distinct
training approaches for encoding-decoding techniques on the samilar dataset. The first LSTM has
achieved an accuracy score of 84% on translating English to Hindi, while the second model
attained 71% accuracy.
Recent reviews from Sefara et all [6] emphasizes on translation system designed used for low-
requirement languages. This system integrates language identification for enhancing and improve
the translation accuracy by a lot, particularly for made for languages which have limited linguistic
resources. By integrating transformer-based approach, this technique helps us to improve
translation quality and address and solve challenges related with low-resource language
translation.
Samuel et all [7] .This conference lays focus on a range of research papers presented at SACAIR
2021, focusing on advancements and applications in the field of artificial intelligence. The specific
section starting on page 390 discusses topics related to language translation using machine
learning techniques and its applications in real world.
Kiran Gaykar et all [8] gives us a perspective on building a mobile application which extracts
text from images and translate it into various languages depending on user need. Utilizing Google's
Machine Learning Kit, this application performs text detection and recognition directly on the
device itself , ensuring quick and easy processing and enhanced privacy and no security breach
.This tool aims to assist users, particularly mainly focussed on tourists, in overcoming language
barriers by enabling them to understand textual information from images, such as traffic signs or
menus, in their preferred language.
Pierre Marquis et all [9] explores the potential outcomes for translating lanuages between different
types of machine learning (ML) models with the objective of refining the explainability of
translator. As machine learning models grow in complexity, interpreting their decisions becomes
increasingly difficult with [Link] central contribution is a translation map that outlines which
4
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
models can be converted into others efficiently. This has some serious implications for XAI
because more interpretable models can be derived from the complex ones, like neural networks,
without sacrificing the overall [Link], the paper stresses on understanding the
inter-relationships between model structures is very important for reducing the gap between
performance and interpreting ability, thereby creating trust and transparency on AI systems.
Finally, Saikat Chakraborty et all [10], gives us a detailed research of how deep learning has
revolutionized this field of machine translation domain . It outlines this transition from the
traditional and statistical approaches to modern deep learning models, which have somehow
significantly improved the translation quality,accuracy and [Link] paper delves deep into the
use of various neural network architectures .Furthermore, this paper reviews evaluation metrics
and emphasizes the need for more of robust, human-centric assessment methods. It concludes by
discussing the limtiations like unsupervised translation, zero-shot learning, and the integration of
translation in real-time applications.
Literature survey table
Author et Year of Algorithm Implementatio Evaluation Comments
al.[Ref Publicatio Used n Details Parameters
sNo.] n
Johnny et 2022 k-Nearest Used 5DT Glove Accuracy k-NN
al. [1] Neighbors data performed
(k-NN) best (97.04%
accuracy)
Deng et al. 2022 Systematic PRISMA-P User Mixed
[2] literature methodology attitudes perceptions
review employed analyzed observed
Mohamma 2024 Artificial Literature Translation Advancement
d et al. [3] Intelligence review accuracy s and
methodologie conducted assessed challenges
s
Natarajan 2022 Deep learning Sign language Recognition Challenges
et al. [4] framework processing accuracy and
measured advancements
Kahlon et 2023 Systematic Sign language Performance Advancement
al. [5] literature translation metrics s and
review analyzed challenges
Klimova et 2023 Neural Foreign Language Advanced
al. [6] Machine language skills learners'
Translation education development suitability
5
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
Sitender et 2023 Neural English-Hindi- Corpus size High
al. [7] Machine Sanskrit requirement computationa
Translation translation s l demands
Dhanjal et 2022 Speech English, Hindi, Recognition Future
al. [8] recognition Punjabi accuracy enhancement
models achieved s suggested
Lalrempuii 2023 Multilingual Indic languages Translation Data scarcity
et al. [9] neural integration quality addressed
translation assessed
Kandimalla 2022 Transformer English-Hindi, BLEU scores, Back-
et al. [10] architecture, English-Bengali manual translation
back- evaluation improves
translation performance
3 . METHODS AND MATERIAL
Our suggested technique focuses more on enhancing the accuracy of the language translation using
machine learning techniques such as neural machine translation and large language models aiming
to improve the overall refinement of our translator . By using these two bulk datasets to train our
model, we ensure more improved results compared to previous existing models . Leveraging the
power of machine learning techniques , we will enable users to get a better results to their inputs.
3.1 SYSTEM ARCHITECTURE
Input processing
The input processing is the first step in the system architecture of a language translator which
forms the basis of accurate and efficient translation. This means converting the raw input text into
a format suitable for the translation engine. It all starts with cleaning and normalizing the input
6
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
with text normalization—getting the characters into a consistent form including dealing with
punctuation and an dealing with variances in representations like contractions. After text
normalization, Tokenization is done where the input is split into smaller parts that can still provide
context and allows the systems to recognize the grammatical structure of the input. For some
systems (particularly multilingual ones) the input may have automatic language identification to
determine the language in which the text has been originally written if not mentioned otherwise.
More complex models may also perform morphological scanning to identify roles, structures of
words in terms of grammar. Syntactic parsing is used to identify relationships between words in
a sentence which was especially useful in both rule based and hybrid translation models.
Text processing
One of the critical elements in a language translator's system architecture is the translation engine
responsible for translating the processed input text of the source language in contrast to target
language. Upon receiving and modifying the input data, the translation engine uses a set of
algorithms and linguistic rules to create a corresponding output in another language. In particular,
if NMT systems, the translation engine might employ attention mechanisms or feedback loops to
refine the quality of the given output. At the end, this is the point at which the structured linguistic
representation that was created in the input phase becomes a cogent, contextually appropriate and
grammatically translucently rendering of the target language and indeed the most pivotal step in
the translation process as it occurs.
Translation model
The post-processing also refers to the output generators stepping of the Language translator system
architecture that is the final phase in the process that are responsible for making the output
polished, grammatically correct, and presentable to the user. Given that the translation engine
produces raw machine text in the target language, this phase works on polishing and formatting,
typically to make it more readable. The finalization might correct punctuations, capitalization,
order of words, and compatible construction of a sentence according to the grammar of the target
language you are translating to. The number can also be used to concatenate output values, which
were separated during input or translation, such as compound words or phrases. Post-processing:
In some systems, especially those that employ neural machine translation, post-processing is used
to impose further language models or rules that improve the fluency and coherence of the output.
This stage also checks how named entities, dates and technical words are treated to ensure correct
7
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
translation or preservation. Filtering may also be applied at the post-generation stage, when
offensive or inappropriate content needs to be removed according to the system’s target
application. In conclusion, this step provides a "sparkling" insight, to finally translate the original
version and provide the end user an usual, natural and contextually failed translation.
Translation process
The output interface is the last piece of the puzzle in the architecture of a language translator,
presenting the translated text to the user in an understandable and easy to retrieve format. This
should concern itself with how the content (now in its translated form) is delivered to the user,
and in a way that meets their expectations and presents correctly based on the platform it’s being
delivered on. The output interface depends on the type of translator: a web application or mobile
app will show the text on a screen, while embedded software will use text-to-speech to read the
translated text aloud or export it to another type of application. The initial product used to have a
simple output interface which allowed users to view and copy or paste the translated content easily.
It might also include more advanced features, depending on the specific system, such as side-by-
side comparisons of the translated text with the original, pronunciation guides, or alternative
translations for specific words. Output interface aims to not only display the final translated
sentence but also aims to improve user experience by presenting the output in a very simple, easy
to interpret and user friendly outputs. It is the point at which the translation system connects with
the user, bringing the communication cycle begun with processing input to its conclusion.
Post processing
The concept is crucial to successful language translators and, in particular, modern language
translators that utilize machine learning techniques. Based on this information, the translator can
learn and adapt to improve its results, leveraging user feedback to become more accurate over
time. For example, when users indicate that they choose a better alternative translation, correct a
mistake, or rate the quality of a translation, that feedback is fed into the system in order to update
the system’s models. Such feedback can also be employed in the neural machine translation
systems and helps their neural networks to optimize eventually, allowing a translator to learn for
specific languages, dialects, or even certain usage patterns.
Output delivery
The auxiliary frameworks within the system architecture of a language translator are crucial in
integrating the different functionalities of the translator, making it more efficient and user friendly.
Although they do not participate directly in the text translation process, these modules offer
services which make it possible for the core components to function without any problems.
8
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
Supporting modules include databases that house the bilingual dictionaries and corpora that are
necessary for translation creation. User authentication, session control, and error logging in
addition to error-control modules make basic but vital functions for any system. Modules of this
type are crucial for robust and transparent operation in web and enterprise translation systems.
Language model management modules are relevant in spell and grammar checking, domain
adaptation, and provide inter-subject area relevance that make translations accurate and
appropriate according to the intended use. These supporting modules may also consist of
integration tools that allow for the use of the translator with other document editing applications,
website browsers, or speech to text programs. Supporting modules are therefore the components
of a system that serve the entire scope of the operation by building an infrastructure, controlling
resources, and interfacing with the translation engine and interface components of the system.
3.2 PROPOSED METHODOLOGY
3.2.1 Neural machine translation Algorithm (NMT)
Neural Machine Translation (NMT) is a recent technological advancement in AI-based translation
systems and translation services driven by the deep learning paradigm having subfields such as
neural networks, with attention mechanisms or transformers. In contrast to other approaches,
NMT uses end-to-end archictectures that consider whole sentences as input rather than splitting
sentences into smaller [Link] the flip side , Statistical Machine Translation algorithm
(SMT) is works on probabilistic frameworks derived from aligned bilingual text corpora. It is
translation-focused and interprets text by splitting it into smaller units, such as words and phrases,
which is done separately – each component is translated independently according to the highest
probability of occurrence. SMT was once described as cutting-edge technology, but almost
exclusively relied upon solving coherence in longer sentences and failing to consider context,
which leads to translations that are ungraceful or stilted. In contrast to SMT, NMT provides
seamless, more precise translations, especially in complex linguistic.
3.2.2 Large language models algorithm (LLM’s)
Large Language Models (LLMs) are advanced AI approach systems which are made to process,
generate, and manipulate human languages. They are designed using deep learning techniques,
primarily transformer methodology, and are extensively trained on vast amounts of data sets
from the internet and other sources. Extensive training enables LLMs to understand grammar,
9
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
context, facts, and even reasoning patterns, allowing implement a wide range of tasks such as
translation, summary generation, answering the question, and content creation.
What sets LLMs apart is their scale—both in terms of the number of parameters (often running
into billions) and the volume of data they are trained on. This scale allows them to generate
human-like responses with a high rate of fluency and relevance. Models such as GPT, BERT,
and PaLM have shown impressive capabilities in understanding nuanced language and following
instructions. However, LLMs also come with challenges, such as extreme computational costs,
probable biases in outputs due to biased training datasets, and the possible risk of giving
incorrect and misleading information. Despite these limitations, LLMs represent a major leap
forward in AI and are mostly used in applications ranging from chatbots and virtual assistants to
making content creation and research support.
3.3 FORMULAE -
1) Translation probability - P(Z∣ X)=t=1∏mP(zt∣ z1,...,zt−1,X)
X – input sentence
Y – output sentence
2) Transformer architecture -
Where Q – query , K – key , V – value
10
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
3) Loss function - L=−t=1∑mlogP(xt∗ ∣ x1∗ ,...,xt−1∗ ,Y)
4) Decoding - Score(X)=∣ Y∣ α logP(X∣ Y)
4. RESULT AND DISCUSSION
From our research, we developed a translator using machine learning approaches like neural
machine translations algorithm (NMT) and large language models algorithm (LLM), unlike
traditional translators which are bounded by drawbacks such as performance limitations ,
problems in handling large data sets and acuracy concerns . This integration of machine learning
algorithms helps us to overcome this drawbacks and infact solve these problems with ease. With
further devolopments it can be more efficient and more performance oriented with regular
advancemnets promising a bright future in terms of usage and application.
11
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
The above flowchart demonstrates the system architecture used in the working of this language
translator .There are primarily six phases involved in the working of this language translator . Each
and every phase is very important in the understanding the working of this translator . These steps
involve processes such as lexical analysis , syntax analysis , semantic analysis ,intermediate code
generation , code optimization , code generation and last but not the least code linking and
assembly thus making it very important to understand structure and working of this language
translator . By our testing we have witnessed and recorded better quality and accuracy compared
to the traditional translators . The results were tested on multiple environments with multiple data
sets and the end results were very improved and efficient.
After testing we thought of comparing traditional translators with our translator , we did an in
detail analysis of both and made a comparision table with different parameters such as
adaptability,context handling,learning approach,accuracy,algorithm type .The above table shows
differences between the traditional existing system and executed proposed system .
12
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
Parameters Traditional system proposed System
Hard to adapt to new Easily fine-tuned with
languages/domains more data or prompts
Adaptability
Word-by-word or phrase Full sentence and
Context Handling based, poor long-range document level
context understanding
Manual rules or phrase End-to-end deep learning
tables on large datasets
Learning
Approach
50- 70-85%
Accuracy 70%
Algorithm Type Rule-based or statistical Neural networks (NMT),
(e.g., SMT) transformers (LLMs)
The working interface of how our language translator actually looks like
13
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
This graph showcases the difference in accuracy of our model compared to the traditional model
in different parameters such as adequency,fluency,BLEU,METEOR score,f-measure and atlast
overall rating of both these translators.
14
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
These are outputs of our implemented code where we trained with our own datasets . We have
tested this in multiple languages and the results are as [Link] results are performed on text
to text translation on various environments and conditions.
15
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
5. CONCLUSION
In the course of this project , we have seen the enhancement of machine based translation with the
addition of machine learning techniques such as large language models (LLM’s) and neural
machine translation (NMT) helps in increasing overall accuracy of the translator. Unlike traditional
translation systems such as statistical machine translation ( SMT) which often falls short in
delivering fluent and context – aware translations,especially for complex and idiomatic
expressions . By using NMT we can consider sentences and long-range dependencies enables more
natural and accurate language translation.
Further , the integration of large language models (LLM’s) have increased the performance even
more to a higher level as they are trained on large datasets , multilingual datasets , posses a deep
understanding of grammar and cross - semantic patterns . The transformer based architecture of
allows for exceptional fluency, contextual understanding and adaptability over a large set of
languages . These additions not only just enhance the overall quality of translation but also improve
capabilities such as paraphrasing , summarization and context aware suggestions which enhance
the overall experience of a user.
Overall the integration of both NMT and LLM’s marks a transformative shift in language
translation ecosystem . It bridges gap between human and machine level fluency , enabling more
effective communication across linguistic boundaries . The main objective of this project is to shed
light on the potential of modern AI driven approaches to create smarter , more efficient and more
human like translation systems , setting foundations for multilingual applications and global
connectivity.
The future scope of these machine learning techniques is very promising and with time it gets
better ensuring sustainable and promising future ahead . Further advancements ensure that high
quality translation is delivered with using less and minimal resources thereby not completely being
a toll on computer resources thereby promoting linguistic inclusivity . These translation has
multiple various applications such as voice assistant , international customer support , cross border
communication and many more possible real world applications . Incorporating feedback loops
through reinforcement learning could further penalize and refine quality over the time .
Additionally combining large language models (LLM’s) with speech recognition and synthesis
can lead to more robust speech to speech translation systems . Overall the integration of LLM’s
and NMT will play an important role in fostering global collaboration , ensuring seamless
communication and breaking language barriers among citizens of different nations.
16
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
6. REFERENCES
[1] A Survey on Evaluation of Large Language Models - Yupeng Chang, Xu Wang, Jindong
Wang, Yuan Wu, Linyi Yang, Kaijie Zhu, Hao Chen, Xiaoyuan Yi, Cunxiang Wang, Yidong
Wang, Wei Ye, Yue Zhang, Yi Chang, Philip S. Yu, Qiang Yang, and Xing Xie, no. 3 (2024): 1-
45.
[2] The Prospects of Large Language Models - Timm Teubner, Christoph M. Flath, Christof
Weinhardt, Wil van der Aalst & Oliver Hinz ,Bus. Inf. Syst. Eng 65 (2023): 95-101.
[3] Evaluating Large Language Models Trained on Code - Mark Chen, Jerry Tworek, Heewoo
Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan , arXiv preprint
arXiv:2107.03374 (2021).
[4] Emergent Abilities of Large Language Models - Jason Wei, Yi Tay, Rishi Bommasani, Colin
Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald
Metzler, Ed H. Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, William Fedus
, arXiv preprint arXiv:2206.07682 (2022).
[5] Large language models encode clinical knowledge - Karan Singhal, Shekoofeh Azizi, Tao
Tu, S. Sara Mahdavi, Jason Wei, Hyung Won Chung, Nathan Scales, Ajay Tanwani, Heather
Cole-Lewis, Stephen Pfohl, Perry Payne, Martin Seneviratne, Paul Gamble, Chris
Kelly, Abubakr Babiker, Nathanael Schärli, Aakanksha Chowdhery, Philip Mansfield, Dina
Demner-Fushman, Blaise Agüera y Arcas, Dale Webster, Greg S. Corrado, Yossi
Matias, Katherine Chou, Vivek Natarajan , Nature 620, no. 7972 (2023): 172-180.
[6] Risks and Benefits of Large Language Models for the Environment - Matthias C.
Rillig,Marlene Ågerstrand,Mohan Bi,Kenneth A. Gould,Uli Sauerland ,no. 9 (2023): 3464-3466.
[7] Neural Machine Translation: A Review – Felix stalhberg ,Journal of Artificial Intelligence
Research 69 (2020): 343-418.
[8] Challenges and Applications of Large Language Models - Jean Kaddour, Joshua
Harris, Maximilian Mozes, Herbie Bradley, Roberta Raileanu, Robert McHardy , arXiv preprint
arXiv:2307.10169 (2023).
[9] Neural Machine Translation of Rare Words with Subword Units - Rico Sennrich, Barry
Haddow, Alexandra Birch,arXiv preprint arXiv:1508.07909 (2015).
[10] Neural machine translation : past , present and future - Shereen A. Mohamed, Ashraf A.
Elsayed, Y. F. Hassan , Mohamed A. Abdou,Neural Computing and Applications 33 (2021):
15919-15931(2021).
17
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
[11] Artificial Intelligence Based Language Translation Platform - Kolhar, Manjur, and Abdalla
Alameen,Intelligent Automation & Soft Computing 28, no. 1 (2021).
[12] "Artificial intelligence in clinical settings: a systematic review of its role in language
translation and interpretation." - Genovese, Ariana, Sahar Borna, Cesar A. Gomez-Cabello, Syed
Ali Haider, Srinivasagam Prabha, Antonio [Link],and Benjamin [Link]. Annals of
Translational Medicine 12, no. 6 (2024): 117.
[13] "An android based language translator application." - Ogundokun, Roseline Oluwaseun,
Joseph Bamidele Awotunde, Sanjay Misra, Tobe Segun-Owolabi, Emmanuel Abidemi Adeniyi,
and Vivek Jaglan,In Journal of Physics: Conference Series, vol. 1767, no. 1, p. 012032. IOP
Publishing, 2021.
[14] "The impact of artificial intelligence on language translation: a review." - Mohamed, Yasir
Abdelgadir, Akbar Khanan, Mohamed Bashir, Abdul Hakim HM Mohamed, Mousab AE Adiel,
and Muawia A. Elsadig , Ieee Access 12 (2024): 25553-25579.
[15] "Comparative analysis of language translation and Detection System using Machine
Learning." - Verma, Aishwarya R., and R. R. Sedamkar , Int. J. Res. Appl. Sci. Eng. Technol. 9,
no. 8 (2021): 1200-1211.
[16] "Transformer-based machine translation for low-resourced languages embedded with
language identification.-Sefara, Tshephisho J., Skhumbuzo G. Zwane, Nelisiwe Gama,
Hlawulani Sibisi, Phillemon N. Senoamadi, and Vukosi Marivate",pp. 127-132. IEEE, (2021).
[17] "The Role of AI in Modern Language Translation and Its Societal Applications: A
Systematic." - Ssemugabi, Samuel (2024)Artificial Intelligence Research (2024): 390..
[18] "Image based Text Translation using Firebase ML Kit." - Bagal, Vipul, Kiran Gaykar, and
Purnima Ahirao ,Grenze International Journal of Engineering and Technology, June Issue,
Grenze ID: 01. GIJET. 8.2. 201 (2022): 400-404.
[19] "On translations between ML models for XAI purposes." - De Colnet, Alexis, and Pierre
Marquis. pp. 3158-3166. International Joint Conferences on Artificial Intelligence, (2023).
[20] "On ML-based program translation: perils and promises." -Malyala, Aniketh, Katelyn Zhou,
Baishakhi Ray, and Saikat Chakraborty.(2023). New Ideas and Emerging Results (ICSE-NIER)
(pp. 60-65). IEEE.
[21] "Artificial intelligence and human translation: A contrastive study based on legal texts." -
Moneus, Ahmed Mohammed, and Yousef Sahari,Heliyon 10, no. 6 (2024).
18
International Journal of Science and Technology (IJST)
Volume - 9 issue – 2 ISSN No: 1489 – 8956
_______________________________________________________
19