Bridging Eras: Reviving Ancient

Languages with Modern Technology

Aneesh : AIE24103
Ankur : AIE24105
Avyuktha : AIE24107
Jiya : AIE24123
 Introduction
● Reviving ancient languages for modern
audiences.
● Preserving cultural and philosophical essence.
● Expanding accessibility for diverse audiences.
● Enhancing accessibility through visual and
audio.
 Introduction
● Addressing translation challenges and
limitations.
● Bridging cultural gaps in translations.
● Utilizing rule-based and neural methods.
● Combining machine learning and human
intervention.
 Research Problem Statement

● Neglecting the possibility of words having more

than one meaning
● Translations done without considering context
can be erroneous
● Mistranslations due to lack of words with exact
meanings
 Research Problem Statement
● Visualizing descriptions may become arduous
● Difficulty in translating text with improper
grammar
● Translators are not always accurate
● Thesauruses/Dictionaries are may not be
sufficient
Example from Varadambika Parinaya Campu
 Research Problem Statements
nirantarāndhakārita-digantara-kandaladamanda-
sudhārasa-bindu-sāndratara-ghanāghana-vṛnda-
sandehakara-syandamāna-makaranda-bindu-
bandhuratara-mākanda-taru-kula-talpa-kalpa-mṛdul̥ a-
sikatā-jāla-jaṭila-mūla-tala-maruvaka-miladalaghu-laghu-
laya-kalita-ramaṇīya-pānīya-śālikā-bālikā-karāra-vinda-
galantikā-galadelā-lavaṅga-pāṭala-ghanasāra-
kastūrikātisaurabha-medura-laghutara-madhura-
śītalatara-saliladhārā-nirākariṣṇu-tadīya-vimala-vilocana-
 Objectives of the Research

●To provide captions, audio descriptions and

voiceovers
●To use AI-powered translation techniques to
decipher old texts
● To focus on the use of examples to understand
aspects of ancient languages
 Objectives of the Research

● To develop effective machine translation

methods for ancient literature
● To incorporate the essence and philosophy of
ancient texts in modern translations
● To implement grammar corrector
 Stakeholders
● Linguists, historians, language experts, AI
researchers
● Individuals with impairments, technology
developers
 Literature Review

1) Translating lost languages using Machine Languages

Keywords: Decipherment algorithm, Cognate-based
decipherment, Permitted patterns, Historical Linguistics
Authors: Barzilay, Luo, Adam conner-simons
Research gaps:
1. Unknown Language Families and Isolates
2. Limited Application of Historical Context
 Literature Review
2) Study of various approaches in machine translation for
Sanskrit language
Keywords: Machine translation, Morphology, Rule based
model, Computational linguistics
Authors: Prof. Deepak Mane, Aniket Hirve
Research Gaps:
1) Scalability
2) Context Handling
 Literature Review
3) SentiWordNet for Bangla
Keywords: Lexicons, Labeling, stemming, Sentiment
tag extraction program
Authors: Amitava Das, Sivaji Bandyopadhyay
Research Gaps:
1) Language-specific words
2) Domain Adaptability
 Research Hypotheses

• How can machine translation be optimized to

handle the complexity of ancient languages like
Sanskrit?
○ Gather quality data
○ Use Grammar rules
○ Transfer Learning
○ Train for Context
○ Continuous Expert Feedback
 Research Hypotheses

• Can a hybrid model of machine learning and human

intervention provide better translations than purely
computational models?
○ Error Correction
○ Handling Variations
○ Model Training
○ Specialized Vocabulary
○ Efficiency
 Research Hypotheses

• How can technology ensure equal visual and audio

access?
○ Screen Readers and Text-to-speech
○ Closed Captions and Transcripts
○ Audio Description
○ Speech Recognition
○ Sign Language Interpretation
 Research Hypotheses

• Will the integration of tools like SentiWordNet

improve user experience?
○ Better Sentiment Analysis
○ Personalized Interactions
○ Improved Content Recommandations
○ Product Feedback Insights
○ Sentiment-Based Filtering
 Methodology

● Machine Translation Approaches:

○ RBMT: Uses linguistic rules and bilingual
dictionaries
○ PBMT: breaks the source sentence into phrases
○ SMT: Statistical models based on large datasets
○ EBMT:retrieves similar sentences from a database
○ NMT: uses deep learning and artificial neural
networks ( encoder-decoder architectures )
 Methodology

● Lexical Transfer Techniques

○ To connect languages with different
structures and vocabularies
○ Direct translation :source language to target
language (Eg: “Apple”,“Seb”)
○ Cognate matching: Finds words of common
origin (Eg."apple" (English)"manzana"
(Spanish))
 Methodology

3) Loan word recognition : recognizing words that are

borrowed (Eg: "Pizza" (Italian))
4) False friend Identification: false cognates ("Actual"
(English: real),"actual" (Spanish: current))
5) Contextual disambiguation:Translates words
based on their context (Eg : "Bank")
 Methodology
● MIT Linguistic Research:
○ Studied the language of Ugaritic
○ Ugaritic language is similar to Hebrew
○ Development by MIT’s Computer Science
and Artificial Intelligence Laboratory
(CSAIL):
○ A new system to decipher lost language
without advanced knowledge.
 Methodology
● Audio Visual Documentation
○ Neural Networks: Text-to-image
○ Generator and discriminator
○ Super resolution (ESRGAN)
○ Natural rendering: generates high quality,
photorealistic images
○ Reinforcement learning and fine tuning
 Methodology
● Neural Machine Translation and Sanskrit
*Mishra and Mishra:
In 2008,
Implemented example and rule based approaches
In 2010,
Improved Rule based Machine Translation
approach
 Methodology
● Neural Machine Translation and Sanskrit
Kaul and Manvi(2019):
Proposed encoder-decoder based NMT
Recent Years:
Sequence to Sequence Learning
Transfer Learning
Reinforcement Learning
 Methodology

● Neural Machine translation And Sanskrit.

○ Offers strength in handling sanskrit translation
like morphology, contextual understanding
and character level translation
○ Use of various PLM’s
Provide a set of initial biases that can be fine
tuned for a task
*PLM’s - Pre-trained Language Models
 Methodology

● Combining rule-based and Neural Machine

Translation (NMT) techniques for translating
ancient texts
● Neural Transfer Techniques
○ Aim: To understanding and translating
ancient languages, morphology and complex
grammar.
 Technologies
● SentiWordNet and Subjectivity Word List
● SentiWordNet:
○ Description: A resource that assigns
sentiment scores (positive, negative, neutral)
to words
○ Use Case: Analyzing the emotional tone of
text for sentiment analysis
 Technologies
Polarity Classification:
● Task of classifying subjective text as either:
○ Positive
○ Negative
○ Neutral
● Uses:
○ In business to analyze consumer opinion
○ In social media to check nature of comments.
 Technologies
● Subjectivity Word List:
○ Description: Categorizes words based on
their level of subjectivity or objectivity
○ Use Case: Identifying and classifying text
based on personal opinion versus factual
content
 Technologies

● POS Tagging Technology

○ Machine Translation
Improves translation quality by providing
structural understanding of sentences
 Technologies
Subjectivity Classification:
● Recognizes contextual polarity of Options,
emotions etc
○ Categorizes text as subjective or objective
○ Facts are objective
○ Opinions are subjective
● Has good accuracy on strong sentences
 Technologies
● Speech Recognition.
○ Improves translation quality by providing
structural understanding of sentences.
● Information Retrieval
○ Helps system understand relationship
between words and queries
 Technologies
● Translation tools
● Neural Machine Translation (NMT)
● Multilingual Neural Machine Translation
(Multilingual NMT)
● Contextual Word Embeddings
 Technologies
● Sanskrit Language Tools :
○ Transliteration tools
○ Sanskrit Dictionaries
○ Grammar Tools
○ Optical Character Recognition (OCR)
○ Text Parsing Tools
 Significance of the Study
●Contributions to improving accessibility through
technology
●Impact on enhancing user experience for
diverse audiences
● This study enhances understanding and
accessibility of ancient literature
 Significance of the Study
● It preserves cultural and historical significance
for modern audiences
● The goal is to make texts more understandable
to contemporary readers
 Scope and Limitations
● Scope:
○ Focus on highly inflected languages such as
Sanskrit, applying NMT models to bridge the
gap between ancient and modern language
structures
○ Focus on language processing tools and
accessibility technology
 Scope and Limitations
● Limitations:
○ Not addressing all disabilities or tech gaps
○ Limited availability of digitized training data
for ancient languages. Human intervention
remains necessary to ensure translation
accuracy
 Timeline
Months 1-8 Data collection

Months 9-13 Publishing research paper

Months 14-25 Trial run
Months 26-36 Prototype development
 Expected Results
● Improvements in accessibility and greater
translation accuracy
● Development of a hybrid translation model for
ancient Sanskrit texts
● Balances literal and interpretive translation
techniques
● Combines rule-based and Neural Machine
Translation (NMT) methods
 Expected Results
● Enhances translation accuracy while preserving
cultural and contextual meaning
● Contributes to linguistic research in machine
translation for ancient languages
● Provides insights into challenges and
opportunities in translating historically rich
languages
 Conclusion
● NMT offers significant potential for translating
Sanskrit
● Leveraging modern approaches like transfer
learning with neural systems improves
translation quality
● The importance of creating visual and audio
access
 Conclusion
● This research develops a hybrid translation
model for ancient literature
● It bridges the gap between ancient texts and
modern understanding
● The project makes valuable texts accessible to a
wider audience
● It promotes cultural preservation and
intellectual exploration
 References
● SentiWordNet for Bangla
○ Jadavpur University Research
○ https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=99
61fbfddc5dece785d9016b4d5b94a79bc3d317
● Automatically creating datasets for measures of semantic relatedness
○ ACM Digital Library
○ https://dl.acm.org/doi/abs/10.1145/1963405.1963455
● Machine Learning for Ancient Languages: A Survey
○ MIT Research papers
○ https://direct.mit.edu/coli/article/49/3/703/116160/Machine-Learning
-for-Ancient-LanguagesA-Surv
● Translating Lost Languages Using Machine Learning
○ https://news.mit.edu/2020/translating-lost-languages-using-machine-le
arning-1021
 References
● Machine Translation Systems for Indian Languages:
Review of Modelling Techniques, Challenges, Open
Issues, and Future Research Directions
https://link.springer.com/article/10.1007/s11831-020-09
449-7

● Study of Various Approaches in Machine Translation for

Sanskrit Language
Research Gate
https://www.researchgate.net/profile/Deepak-Mane/public
ation/327287889_Study_of_various_approaches_in_Machi
 Acknowledgement
A heartfelt gratitude to Smitha Sail Ma’am for guiding us and making this
presentation possible.
Thanks to my parents and those of my teammates for supporting us.
I would now like to extend my thanks to this university for giving us an
opportunity to learn a skill so vital in the current world.
Lastly, I thank my team members for their unwavering support without
which, this presentation would have been far from complete.
THANK YOU