MULTILINGUAL TRANSLATION

MINI PROJECT REPORT

for
21CSE356T - NATURAL LANGUAGE PROCESSING

Submitted by

SHRESHTHA SRIVASTAVA [RA2211003010872]

KALP AGARWAL [RA2211003010879]
GORANTLA GAYATRI [RA2211003010880]
P.VIVEKANANDA REDDY[RA2211003010882]
KESARLA ABHIRAM[RA2211003010908]

Under the Guidance of

Mr.S.Prabu
(Assistant Professor, Department of Computing Technologies)

In partial fulfillment of the requirements for the degree of

BACHELOR OF TECHNOLOGY
in
COMPUTER SCIENCE AND ENGINEERING

DEPARTMENT OF COMPUTING TECHNOLOGIES

SRM INSTITUTE OF SCIENCE AND TECHNOLOGY
KATTANKULATHUR- 603 203
MAY 2025
 SRM INSTITUTE OF SCIENCE AND TECHNOLOGY
KATTANKULATHUR – 603 203

BONAFIDE CERTIFICATE

Certified that Natural Language Processing Mini Project report titled

“MULTILINGUAL TRANSLATION” is the bonafide work of
“SHRESHTH SRIVASTAVA”[RA2211003010872], “KALP AGRAWAL”
[RA2211003010872], “GORANTLA GAYATRI” [RA2211003010880],
“PANNURU VIVEKANANDA REDDY”[RA2211003010882],
“KESARLA ABHIRAM”[RA2211003010908] carried out the project work
under my supervision. Certified further, that to the best of my knowledge the
work reported herein does not form any other work

Mr.S.PRABU Dr. NIRANJANA G

Guide
Professor & Head
Assistant Professor
Dept. of Computing Technologies Dept. of Computing Technologies
 INDEX

CHAPTER NO TITLE PageNo

1 OBJECTIVE 1

2 ABSTRACT 2

3 INTRODUCTION 3

HARDWARE & SOFTWARE

4 5
REQUIREMENTS

5 CONCEPTS WORKING PRINCIPLE 6

6 PROGRAM 8

7 OUTPUT 15

8 CONCLUSIONS 19

9 REFERENCES 20
 MULTILINGUAL TRANSLATION

OBJECTIVE

The primary objective of this project is to develop an automated system that leverages
Natural Language Processing (NLP) techniques to streamline text tokenization and
multilingual translation. The growing demand for efficient communication across language
barriers presents a significant challenge, particularly in industries where time-sensitive and
accurate translation is essential. Traditional translation methods, especially manual
approaches, often face limitations such as slow turnaround times, high error rates, and
difficulty handling large volumes of data. These challenges become even more complex
when dealing with multiple languages that require constant manual expertise.

This project aims to address these challenges by automating key language processing tasks.
Automating text tokenization and translation will reduce the time and effort involved in
manual translation while improving the accuracy and scalability of the process. Tokenization
is the first step in text processing, where text is broken down into smaller components like
words or phrases. This is essential for understanding the structure of the text and preparing
it for the translation phase. In this project, the system uses spaCy, a powerful NLP library,
to handle tokenization, segmenting the text into manageable units efficiently.

For multilingual translation, the system integrates the MarianMT model from HuggingFace,
a state-of-the-art machine translation model that supports numerous language pairs. By
leveraging deep learning algorithms, this model can translate text accurately and efficiently
between different languages, providing a reliable alternative to traditional methods that are
often slow and prone to errors. The system's ability to automatically translate across multiple
languages makes it scalable and well-suited for a range of applications, from business
communications to personal use.
 ABSTRACT

In today’s interconnected and globalized world, language barriers continue to impede

effective communication, limiting interactions between people from different linguistic
backgrounds. The demand for efficient and reliable translation systems is greater than ever,
driven by the growth of international business, scientific collaborations, online education,
and multicultural interactions. Traditional translation methods often rely heavily on manual
effort, making them time-consuming and prone to human error. These challenges are further
exacerbated when large volumes of data need to be processed in a timely manner.

This project addresses these issues by developing an automated Natural Language

Processing (NLP) system that integrates two critical components: text tokenization and
multilingual translation. By utilizing advanced NLP techniques, the system automates key
stages of language processing, enabling faster and more accurate text handling and
translation. The system uses spaCy for text tokenization, breaking down the input text into
smaller units such as words or phrases, which is essential for understanding the structure of
the text and preparing it for the translation phase.

For translation, the system integrates the MarianMT model from HuggingFace, an AI-
powered machine translation model that uses deep learning to perform highly accurate
multilingual translations. The use of MarianMT reduces the time required for translation and
minimizes errors that often occur in human-driven processes. Unlike traditional manual
methods, this approach is more scalable and can easily handle large datasets, making it an
ideal solution for businesses, academic institutions, and multinational organizations. This
system demonstrates how automated translation through NLP can significantly improve
communication across different languages, reducing the barriers to global interaction.
 INTRODUCTION:

Language serves as a foundational medium for communication, but when people speak
different languages, effective interaction can become a significant barrier. Manual
translation methods, while useful in limited contexts, are often time-consuming,
inconsistent, and impractical for large-scale or real-time applications. This challenge has led
to the adoption of Natural Language Processing (NLP) as a powerful approach to automate
language-related tasks.

This project, “Multilingual Translation using NLP,” addresses these challenges by

building an end-to-end pipeline for translating text across multiple languages with
automation and accuracy. The solution integrates several modern NLP techniques to deliver
a seamless translation experience. The core components of the system include:

1. Language Detection: Before translation, the input language is automatically detected

using the langdetect library to ensure the appropriate translation model is applied.

2. Text Tokenization: The input text is preprocessed and tokenized using spaCy, a robust
NLP library. Tokenization divides text into meaningful units, such as words or phrases,
preparing it for accurate analysis and translation.

3. Multilingual Translation: Translation is performed using the MarianMT models from

HuggingFace Transformers. These pretrained models support multiple language pairs
and use English as an intermediate language for indirect translation paths.

4. Sentiment Analysis (Optional): The system can also assess the sentiment of English
text using TextBlob, providing additional contextual understanding beyond literal
translation.

By automating each of these stages, the project demonstrates how NLP technologies can be
used to efficiently bridge language gaps and enhance communication in multilingual
environments. It offers a scalable and real-time solution suitable for personal, academic, or
business applications.
 HARDWARE/SOFTWARE REQUIREMENTS:

Hardware Requirements:

• A computer or cloud-based platform capable of running Python and accessing

required external libraries and models.
• Minimum of 4GB RAM to run the NLP models effectively.
• A stable internet connection to interact with cloud-based models.

Software Requirements:

• Python: The primary programming language for implementing NLP processing and
integrating the various tools and models.
• spaCy: An industrial-strength NLP library used for tokenization, part-of-speech
tagging, and entity recognition.
• TextBlob: A lightweight library for performing sentiment analysis on English text.
• langdetect: Used to automatically detect the language of the input text before
initiating translation.
• HuggingFace Transformers: Specifically, MarianMT pretrained models are used for
translating text between multiple languages with English as an intermediate.
• Gradio: Used to build a web-based graphical user interface, enabling real-time user
interaction and showcasing the system's capabilities in a simple and accessible format.
• Development Environment: Google Colab (for cloud-based development) or any
local IDE such as VS Code or PyCharm for writing and testing code.
 CONCEPTS/WORKING PRINCIPLE

The system operates through a multi-stage pipeline involving language detection,

translation via intermediate English, and natural language processing (NLP) analysis. The
following outlines the detailed working principles:

1. Language Detection

The input text is first analyzed using the langdetect library to identify the source language.
This enables the system to handle diverse input languages dynamically, without requiring
the user to specify the language manually.

Example:

Input: "Bonjour, comment ça va?"

Detected Language: fr (French)

2. Translation via English Intermediate

• The system uses MarianMT models from Helsinki-NLP (hosted on HuggingFace)

to perform translations.
• If the input language is not English, the text is first translated to English using a
source-to-English MarianMT model.
• Then, the English text is translated to the selected target language using an English-
to-target MarianMT model.
• This intermediate English step ensures better coverage and translation accuracy,
even for language pairs that do not have direct translation models.

Example:
Input: "Wie heißt du?" (German)
Intermediate English: "What is your name?"
Final Output (French): "Comment vous appelez-vous ?"
3. Natural Language Processing (NLP) Analysis

Once the English version of the text is obtained (either directly or via translation), it is
processed using the spaCy and TextBlob libraries to extract detailed linguistic insights.

Tokenization:

The text is split into individual words or tokens for further processing.

Example: "Hello world" → ['Hello', 'world']

Part-of-Speech (POS) Tagging:

Each token is labeled with its grammatical role such as noun, verb, adjective, etc.

Dependency Parsing:

The syntactic structure of the sentence is analyzed by identifying dependencies between

tokens, such as subject–verb or object–verb relationships.

Named Entity Recognition (NER):

Entities such as names of people, places, organizations, and dates are identified from the
text.

Sentiment Analysis:

The polarity of the sentence is computed using TextBlob. Based on the polarity score, the
sentence is classified as Positive, Neutral, or Negative.

4. End-to-End Automation

The system integrates all the above modules in a single interface using Gradio. It provides
a user-friendly interface where users can input any sentence, select the target language, and
receive both the translated output and comprehensive NLP insights automatically
Fig. 5.1: NLP Translation Pipeline - Stages from Data Preprocessing to Evaluation
APPROACH/METHODOLOGY/PROGRAMS:
OUTPUT:
 CONCLUSIONS:

This project demonstrates the practical application of Natural Language Processing (NLP)
techniques to build an automated multilingual translation system. By leveraging a modular
pipeline consisting of language detection, intermediate translation, and sentiment analysis, the
system addresses the limitations of manual translation workflows and improves efficiency,
scalability, and accuracy in multilingual communication.

The system utilizes the langdetect library to automatically identify the source language,
enabling seamless support for various input languages. Translation is performed in two stages
using pretrained MarianMT models from HuggingFace, with English serving as the
intermediate language. This ensures greater consistency and compatibility between diverse
language pairs. For analyzing the sentiment of English text, the project integrates the TextBlob
library, which allows classification of the emotional tone before producing the final output in
the target language. Additionally, spaCy is used for linguistic tokenization, named entity
recognition, and part-of-speech tagging to facilitate deeper language understanding.

The entire process is executed within a Gradio-based web interface, enabling real-time
interaction and accessibility without requiring local installations. This end-to-end pipeline—
from language detection and translation to sentiment analysis—provides a robust and
automated approach for cross-lingual communication. The system not only reduces manual
effort but also enhances reliability when dealing with large or dynamic datasets.

Looking forward, the system could be extended to incorporate more complex NLP tasks such
as contextual emotion analysis, multilingual summarization, and real-time chatbot interactions.
Such enhancements would broaden its applicability across domains like customer service,
healthcare, education, and global collaboration. Overall, the project lays a strong foundation
for future advancements in intelligent, language-aware applications.
 REFERENCES:

• NLTK – Natural Language Toolkit. Available at: https://www.nltk.org/

• spaCy – Industrial-Strength Natural Language Processing in Python. Available

at:https://spacy.io/

• TextBlob – Simplified Text Processing. Available at:

https://textblob.readthedocs.io/

• HuggingFace Transformers – MarianMT Models by Helsinki-NLP. Available

at: https://huggingface.co/Helsinki-NLP

• Langdetect – Port of Google's Language Detection Library. Available at:

https://pypi.org/project/langdetect/