In an era among where globalization is mostly driven by digital material, the importance
of efficient interaction Indian languages (ILs) is growing. Effective machine translation
(MT) systems are critical to facilitate seamless communication and remove obstacles
between diverse linguistic communities. Machine Translation (MT) is the process of
translating text from one language to another without the need for human intervention.
In consideration with the Indian environment, the development of a quality machine
translation system (MTS) for the Indian languages (ILs) is in huge demand. Still, it
remains a challenging task, since many ILs have low in terms of resources, resulting in an
adverse impact on their translation quality. The vast linguistic diversity and socioeconomic
significance of Indian languages both within the Indian Union and internationally serve
as the driving forces for the attention on these languages. The several linguistic groups
that make up Indian languages, such as Dravidian, Indo-European, Indo-Aryan, and others,
are distinguished by their distinct syntactic patterns and cultural subtleties. The variability
in morphology, syntax, and semantic expression poses significant obstacles to machine
translation. Hence, this thesis presents five important developments aimed at improving
the quality and efficiency of machine translation (MT) systems for ILs.
To start, the first contribution presents an effective Statistical Machine Translation (SMT)
system designed especially to translate text from English into 11 Indian languages and
viceversa. To enhance translation quality and clean up the data, many data filtration
techniques are studied. The effects of distance-based reordering and Morpho-syntactic
Descriptor Bidirectional Finite-State Encoder (msd-bidirectional-fe) reordering on ILs are
analyzed. Following this work, the second contribution describes a Neural Machine
Translation (NMT) system that can translate between English and 11 Indian languages in
both directions. Backtranslation (BT) is utilized for data augmentation to enlarge the dataset.
BT has advantages to many languages. However, whether it has a significant impact on ILs
remains debatable. Hence, the influence of data augmentation on neural machine translation
(NMT) for Indian languages is investigated to evaluate its efficacy in enhancing translation
robustness and quality.
However, NMT systems are restricted in translating low-resource languages as a huge
quantity of data is required to learn useful mappings across languages. Hence, the third
contribution investigates methods for translating from English into eleven Indian languages using Multilingual Neural Machine Translation (MNMT) and vice versa. Multilingual
neural machine translation (MNMT) is a technique for MT that builds a single model
for multiple languages. It is preferred over other approaches, since it decreases training
time and improves translation in low-resource contexts, i.e., for languages that have
insufficient corpus. The integration of techniques like pivot-based machine translation
(MT), backtranslation, and language-relatedness improves translation accuracy and fluency
in linguistically varied environments.
The fourth contribution expands MNMT&rsquos capability to include intra-Indian language pairs.
The effect of the grouping of related languages, namely, East Indo-Aryan (EI), Dravidian
(DR), and West Indo-Aryan (WI) on the MNMT model are examined. The role of
pivot-based MNMT models in enhancing translation quality is investigated. Owing to the
presence of large good-quality corpora from English (EN) to ILs, MNMT IL-IL models,
using EN as a pivot are built and examined. Furthermore, the effect of transliteration
on ILs is also analyzed in this contribution. To explore transliteration, the best MNMT
models from the previous approaches (in most of cases pivot model using related groups)
are determined and built on corpus transliterated from the corresponding scripts to a modified
Indian language Transliteration script (ITRANS).
The final contribution assesses how SMT, NMT, and MNMT systems for Indian
languages are affected by subword tokenization techniques like SentencePiece, WordPiece,
and Byte-Pair Encoding (BPE). These methods are essential for improving translation
performance and managing morphologically complex languages. All the approaches and
each contribution are evaluated using standard evaluation metrics.