1.1.

Machine Learning

Machine learning is a branch of artificial intelligence based on the idea that systems can learn from data,
identify patterns and make decisions with minimal human intervention. It can also be defined as a
method of data analysis that automates analytical model building. Machine learning addresses the
question of how to build computers that improve automatically through experience. It is one of today’s
most rapidly growing technical fields, lying at the intersection of computer science and statistics, and the
core of artificial intelligence and data science [9]. Machine learning is used to teach machines how to
handle data more efficiently. Sometimes after viewing the data, we cannot interpret the extracted
information from the data. In that case, we apply machine learning. With the abundance of datasets
available, the demand for machine learning is in rising. Many industries apply machine learning to
extract relevant data. The purpose of machine learning is to learn from the data.

1.2. Machine Translation

Machine translation is a machine that is capable to translate automatically from one language to
another. Machine Translation methods are different, and each has its benefits and drawback. No
translation tools can generate an exact version of the source language but give a list of information that
can be utilized to find the type of information contained in the source text [10]. Machine translation
performs simple replacements with basic translated keywords for non-native speakers to understand
content in an original foreign language that they need to assess [11]. Reviewing the above and more
literature we understand that machine translation is essential for modern word linguistic
communication

Machine translation in foreign languages has been evolving before computers were even created. It is
feasible to hint thoughts approximately mechanizing translation processes lower back to the 17th
century, however sensible opportunities got here best in the 20 th century. In the mid-1930s, a French-
Armenian Georges Artsrouni and a Russian Petr Troyanskii implemented patents for ‘translating
machines’. Of the two, Troyanskii's become the more significant, providing now no longer the best
technique for an automated bilingual dictionary. However additionally a scheme for coding inter-lingua
grammatical roles and a definition of the way evaluation and synthesis would possibly work. However,
royanskii’s thoughts have been now no longer acknowledged till the end of the 1950s. By then, the
computer have been born [27]. Most machine translation is conducted either by rule-based, statistical,
or neural network approaches some examples of these are; “rule-based machine translation systems
adapted to English to Finnish translation” [28] which the feasibility of the rule-based approach to the
present-day challenge is that the greater grammatical the sentences are, the higher the result. Ordinary
information reporting may be translated satisfactorily. However, sports activities information and
different varieties of much less grammatical texts are a massive problem… “MIZAN: A Large Persian-
English Parallel Corpus” is an example of statistical machine translation [29] which is public to be had
and the most important Persian-English parallel corpus with approximately 1 million sentence pairs. As it
has an exceedingly larger size, is greater exactly aligned, and consists of better pleasant textual content
in comparison to different present and available Persian-English corpora, we accept as true with it is
able to be an influential useful resource for Persian SMT and plenty of different bilingual and cross-
lingual corpus-associated studies. We compare the MIZAN corpus in an SMT mission with a baseline
BLUE rating of approximately 25 for in-area and 24 for easy out-of-area take a look at sets. We
additionally look into a few required similar research to improve SMT aid for Persian-English and
perhaps different languages with exceptional phrase orders. Nevertheless, the Persian language is an
under-resourced language with many open spots to dig in.

1.3. Neural network

The neural network is a series of algorithms that endeavors to recognize underlying relationships in a set
of data through a process that mimics the way the human brain operates [12]. Neural networks are the
simple clustering of primitive artificial neurons. This clustering occurs by creating layers that are then
connected [13]. Basically, the neural network has three components namely:

 Embedding layer / Input layer

 Encoder and Decoder method / Hidden layer

 Classification layer / Output layer

1.3.1. Embedding layer / Input layer

In a neural network, the embedding layer is defined as the first hidden layer. The embedding layer is
basically a process of mapping a discrete categorical variable to a vector of continuous numbers which
are low-dimensional and learned continuous vector representations of discrete variables [14]. Word
embedding is a term used for the representation of words for text analysis, typically in the form of a
real-valued vector that encodes the meaning of the word such that the words that are closer in the
vector space are expected to be similar in meaning [15]. Embedding layers have three main purposes
which are:

 Finding nearest neighbors in the embedding space

 As input to machine learning model in the case of supervised task

 And last but not least, for visualization of concept and relations between categories.

1.3.2. Encoder and Decoder method / Hidden layer

The introduction of hidden layers makes neural networks superior to most of machine learning
algorithms. Hidden layers reside in-between input and output layers and this is the primary reason why
they are referred to as hidden [16]. The word “hidden” implies that they are not visible to the external
systems and are “private” to the neural network. Neural network translation use encoder and decoder
methods in order to compress and decompress the input data namely called auto-encoder. Auto-
encoders are a type of neural network used to learn efficient coding of unsupervised data [17]. It is also
a special neural network structure, which has an input layer, output layer, and hidden layer. It adjusts
the weights of the hidden layer by training to enable the input and output values to be as close to each
other as possible [18]. Therefore, the hidden layer has important features of the original signal in order
to realize unsupervised feature extraction. There is also another encoder-decoder architecture with
recurrent neural networks that has become an effective and standard approach for neural network
translation. The key benefit of this approach is the ability to train a single end-to-end model directly on
the source and target sentences and the ability to handle variable length input and output sequences of
text [19]. The encoder and the decoder in neural network translation play a great role in getting more
accurate results. Note that there could be zero or more hidden layers in the neural network.

1.3.3. Classification layer / Output layer

The output layer takes in the inputs which are passed in from the layers before it, performs the
calculations via its neurons, and then the output is computed [16]. This layer is the last layer in the
network & receives input from the last hidden layer.

1.4. Evolution of neural network translation

Evaluation of machine translation is pivotal to knowing the research outcome. So as to know whether
the approach that is selected for the research gives a satisfying result or not. Translations are evaluated
via fidelity, fluency, or a combination of both. The most accurate evaluations use human raters to
evaluate each translation along with fidelity and fluency [20]. There are also some automatic evaluation
metrics to evaluate machine translation systems like WER, BLEU, and METEOR