Available via license: CC BY-NC 4.0
Content may be subject to copyright.
On Neural Machine Translation Based on Cloud
Platform in the Context of Artificial Intelligence
Renna Gui(B)
School of Foreign Languages, Nanchang Institute of Technology, Nanchang, Jiangxi, China
86050651@qq.com
Abstract. The research of machine translation has gone through the era of
machine translation based on grammar rules, case data and statistical methods.
Up to now, it has become a neural machine translation designed with coding and
decoding as the basic framework and using neural network to model the trans-
lation process. It is illustrated how the human-computer co-translation can be
applied based on translation technology cloud platforms so as to greatly enhance
the translation productivity.
Keywords: Machine Translation ·Cloud Platform ·Artificial Intelligence ·
Human-Computer Translation
1 Introduction
After decades of development, machine translation has completed the evolution and pro-
motion from “mechanical brain” to rule-based machine translation, case-based machine
translation, statistics-based machine translation and neural network machine translation
based on deep learning [2]. People have actively explored in model research and devel-
opment, pre-translation processing of original language, post-translation editing which
have been widely used. The rapid development of machine translation strongly empow-
ers the trend of human-computer co-translation communication in this era of artificial
intelligence.
2 Transition from CAT to NMT
Since the birth of computer in 1947, Machine aided translation (CAT) began to be
used in the academic circles. The development of CAT translation method reflects the
different stages of automatic translation by computer. The first stage is based on language
rules, which holds that human translation can be simulated as long as the theories and
methods of human translation are “injected” into the computer, but language rules cannot
fully cover the complex and changeable language phenomena. The second stage is an
empirical approach to learn from the actual language. Obtaining translation templates
from translation examples or statistical model translation (SMT) fall into this category.
© The Author(s) 2023
K. Subramanian et al. (Eds.): CTMCD 2022, ACSR 99, pp. 436–443, 2023.
https://doi.org/10.2991/978-94-6463-046-6_52
On Neural Machine Translation Based on Cloud Platform 437
The third stage is neural network translation (NMT), which is an end-to-end holistic
translation model [1].
With the deepening of machine translation research and the progress of machine
learning and other related disciplines, people gradually find that there are many unavoid-
able problems in statistical machine translation. For example, the translation process
depends on the assumption of hidden structure, the definition of translation features
requires manual design, feature engineering is time-consuming and laborious, and often
does not have universal significance. For these problems, people have tried a new path
- neural machine translation. The so-called neural machine translation is to use neural
networks to directly model translation problems. This process does not assume that trans-
lation has hidden structure and does not rely on artificially-defined features. The whole
translation model can be completed in an end-to-end mode, and translation decoding
becomes a process of forward calculation or inference of neural networks.
Although neural networks have been applied in many tasks, substantial progress in
machine translation has not achieved until 2013. The main reasons are as follows: 1)
there is no very effective framework to deal with the transformation from text sequence
to text sequence; 2) The learning of deep neural network is not very effective. [4]The
problem of deep network learning has made continuous progress in recent years, and
the problem of frame selection in neural machine translation has been alleviated by
the “coding and decoding” structure. The so-called “encoding and decoding” structure
defines the transformation from sequence to sequence as a two-stage modeling problem.
Firstly, the input word sequence X =x1x2…xncontaining N words is encoded. The
encoding result is a real vector hn, which represents the information of the whole input
sequence up to the Nth word; In the second stage, the encoded vector is used for decoding
to generate the output sequence Y =y1y2…ym, which can be described as follows:
ˆ
Y=argmaxY
m
i=1
Pr(yi|{y0,...,yi−1},X)
Pr(yi|{y0,…,yi-1},X) describes the generation probability of the ith word of the
target language. Since the encoder has expressed x as hn, the conditional part of
Pr(yi|{y0,…,yi-1},X) is only related to hn and {y0,…,yi-1 }. Since Pr(yi|{y0,…,yi-1},X)
can be calculated by neural network, a network is used to complete the transformation
from input sequence to output sequence, and the network can be trained by relatively
mature back-propagation method. More importantly, this model completely uses the
representation of continuous space when calculating Pr(yi|{y0,…,yi-1},X). Compared
with the discrete space representation of traditional statistical machine translation, the
representation ability of the model is greatly enhanced. Figure 1shows an example of
an encoding and decoding structure.
Since the encoding and decoding framework completely transforms the machine
translation problem into a network computing problem from input sequence to output
sequence, it does not rely on the characteristics of manual design, so it can better capture
the complex correspondence between different languages. It can be said that neural
machine translation based on coding and decoding framework has become the standard
configuration of relevant research institutions and enterprises. Based on this framework,
researchers have also made a lot of improvements and upgrades. Compared with the
438 R. Gui
Fig. 1. Encoding and decoding structure of neural machine translation.
level three to five years ago, the quality of machine translation has been improved by
leaps and bounds.
3 Application of Human-Computer Co-translation
3.1 Present Situation of Human-Computer Co-translation
With the application of deep learning in translation machine, the translation quality of
translation machine based on artificial neural network has been significantly improved.
The machine translation finds its way in political, economic, cultural and other commu-
nication activities, especially is widely used in the language service industry. According
to China Language Service Industry Development Report 2022, with the innovation of
artificial intelligence, machine translation is more and more widely used in the indus-
try, and there are 252 enterprises with machine translation and artificial intelligence
business. The service mode of “machine translation +Post editing” has been generally
recognized by the market. According to the survey, more than 90% of enterprises said
that adopting this mode can improve translation efficiency as well as translation quality
and reduce translation costs. In the field of low-end text translation, machine translators
have gradually replaced manual translators. In the translation and communication activ-
ities of other scenes, the participation of machine translators is becoming more and more
common, such as using translation machines for cross language communication at sym-
posiums, using machine translation systems for preliminary text translation in translation
projects, and using their own translation systems to communicate with foreign friends on
Wechat. Machine translator has become an important subject in translation practice and
an unavoidable research topic in the study of translation and translation communications
[6].
“Human-Computer Co-translation” is an intelligent translation mode based on big
data, neural machine translation technology, artificial intelligence and mobile Internet,
which combines machine intelligence and artificial intelligence, balances the high effi-
ciency of machine translation and the high quality of human translation, and forms the
output chain of “manuscript-machine translation-human translation”. Neural network
machine translation technology has greatly improved the speed and accuracy of the
translation system.
On Neural Machine Translation Based on Cloud Platform 439
3.2 Cloud Translation Platform Employed at Home and Abroad
In recent years, driven by cloud computing and big data technology, speech recogni-
tion, translation technology and translation platform technology have been continuously
developed, and translation tools have shifted from local to network [5]. Some language
service organizations and localization service companies began to apply cloud comput-
ing technology to the translation industry to build a service translation cloud platform
that can meet the growing needs of users. Integrating the cloud platform with specific
translation technology is the translation technology cloud platform. Its biggest advantage
is that it will be distributed in different places. The translation service human resources
are integrated to form a large-scale and standardized translation project management,
so as to provide comprehensive services for the development of local business and trade
and cultural exchanges [3].
Translation technology cloud platforms at home and abroad can be divided into
four categories: translation trading platform, translation production platform, translation
corpus data platform and artificial intelligence machine translation platform.
3.2.1 Translation Trading Platform
Translation trading platform is an online third-party trading security platform connecting
language service talents and customers to ensure the safety and integrity of both parties
in the transaction. The client finds the required translator through the translation trading
platform and delivers the task to the translator who will complete the translation task
with the agreement of both parties.
The most famous translation trading platform in the world is ProZ, which is the largest
translation community in the world. It provides translation resources and employment
opportunities for translators, translation companies and other personnel in the language
industry, and reduces the cost of obtaining resources between language service tal-
ents and customers. ProZ also has a TM town platform, which can intelligently match
customers with professional translators according to customer needs, translator qualifi-
cations, translation subjects and other screening conditions, select the best results, and
help customers find translators faster.
3.2.2 Translation Production Platform
Under the traditional translation operation mode, translation companies often encounter
problems such as difficult corpus reuse, tight cycle, complex software operation and
inconsistent translation style, while translators will also encounter problems such as
cumbersome term query, data loss and long typesetting time. Therefore, the online aided
translation production platform came into being. The Internet+ language service trans-
lation production platform effectively combines online translation, corpus management,
team management, project management, collaborative translation and other functions
to help translation enterprises and translation teams improve translation efficiency and
reduce translation costs.
Famous translation production platforms abroad include Memsource and Matecat.
Take Memsource as an example. It is a Czech technology company that provides tech-
nical solutions such as cloud translation management system and CAT tools. It has won
440 R. Gui
international recognition for providing easy-to-use and fully functional translation tools
and management system to customers, translation companies and language service per-
sonnel. Now it has become a leader in cloud translation technology. Memsource cloud
is an online translation management system of MEM source company, with embedded
CAT function, which supports project managers to view work progress in real time
and multi-person cooperative translation. Famous translation production platforms in
China include Tmxmall, Yi-cat, jeemaa.com and Twinslator. Taking Tmxmall, Yi-cat as
an example, it is an online translation management platform, which establishes a fully
automated, scientific and process-based translation management model for freelancers,
translation companies and enterprise translation departments. YiCat embedded CAT tool
supports 46 languages, 27 file formats and 27 QA inspection rules. It has the functions
of real-time project supervision, team management, synchronization of translating and
reviewing and intelligent QA inspection. YiCat integrates hundreds of millions of sen-
tence pairs of corpus data from Tmxmall platform. Users can retrieve the total memory
of their public cloud corpus sharing platform in real time during translation, and store
them in the personal memory of private cloud corpus management platform, which can
help translators save time and improve translation quality.
In the translation stage, human translators can use online and offline computer-aided
translation software such as Trados, Wordfast and Twinslator to preprocess documents,
that is, match the repeated contents of the original text and the translation memory and
carry out pre-translation, so as to reduce the repeated workload and improve the transla-
tion efficiency. At the same time, Twinslator, jeemaa.com, zhimaky.utranshub.com and
other platforms integrate translation memory and translation management functions,
which can realize the complex division of translation, translation reviewing and quality
inspection, and greatly improve the efficiency of translation management.
As is shown in Table 1, Human and machine interact and cooperate to complete
project, knowledge base construction and machine learning. After obtaining information
through image recognition, character recognition and audio and video recognition, the
term corpus and online translation complete the basic work of the machine part, including
term extraction, term unification, term management and corpus recovery after translation.
Online translation completes tasks through automatic text analysis, intelligent settings
Tabl e 1 . Human-computer cooperation in translation.
User Projects and Tasks Information Tools
Team/individual terminology management Terminology database management tool
individual Memory-based translation Translation memory management tool
Spelling check Spelling Check tool
Syntax check Syntax check tool
Data inquiry Search engine
Tea m Quality
monitoring
Translation quality
inspection tool
Schedule management Translate project management tools
On Neural Machine Translation Based on Cloud Platform 441
and a variety of machine assistance. The intelligent error correction system checks the
low-level errors in translation, such as spelling errors, digital errors and omissions, unit
errors and omissions, terminology omissions, sentence and paragraph omissions and
punctuation errors and omissions, which saves the time and cost of proofreading.
3.2.3 Corpus Data Platform
With the development of machine translation and the wide application of deep learn-
ing technology, more and more neural network structures are introduced into machine
translation, such as machine translation based on convolution neural network structure,
and machine translation based on cyclic neural network, which improve the quality of
machine translation. Machine translation based on neural network usually needs to use a
large number of high-quality corpus for training in order to get good translation results.
Corpus refers to a large-scale electronic library with considerable capacity, which is
based on the guidance of certain linguistic principles and the method of random sampling
to collect the text or speech fragments of continuous language without any processing.
At present, corpus has been widely used in language teaching, language research and
language engineering.
Machine translation methods are divided into rule-based translation methods and
corpus-based translation methods. Corpus based translation methods can be divided into
case-based translation and statistics-based translation. The difference between them is
that in the former case, corpus will participate in translation as a kind of translation
knowledge for the translation subject to query while in the latter one corpus is used to
find the sentences that are most likely to become the target language without specific
translation practice.
The statistics-based machine translation method starts with language phenomena
and obtains the translation through a relatively rational model. The case-based machine
translation method obtains the translation from the perspective of machine learning
through the analysis and reasoning of cases. However, the two methods are not mutu-
ally exclusive. The prospect of corpus-based machine translation is to organically com-
bine the advantages of various methods to further improve the performance of machine
translation system.
3.2.4 Machine Translation Platform
Today, with the rapid development of big data +Internet, the application scenarios of
machine translation continue to expand. Google translation is a famous foreign artificial
intelligence machine translation platform. Google translation is a multilingual translation
platform developed by Google, which supports 103 languages.
In other words, users only need to input the content to be translated and select the
language pair to generate the target language translation. The interface is friendly and
the operation is simple. Corporate giants such as Facebook, Amazon and eBay are also
developing machine translation engines. Domestic machine translation platforms include
Baidu translation, Youdao translation, Sogou translation and “iFLYTEK voice cloud”
platform, while technology giants such as Tencent and Alibaba are also strengthening
442 R. Gui
the research and development of machine translation products and emerging in the field
of machine translation.
4 Prospect of Human-Computer Co-translation
Machine translation has experienced rule-based system and statistics-based system. At
present, the system based on deep learning has reached a more mature stage. In 2015,
baidu released the neural network translation system, becoming the world’s first Internet
neural network translation system. Google and Youdao followed suit and successively
released neural network translation systems. Compared with the previous translation
system, the translation quality is greatly improved. Some experts believe that at present,
the translation quality of Google translation system has reached the level of medium-
sized manual translators. Machine translation is widely used in various fields, such as
business translation, conference and so on. According to the Research Report on the
Market Prospect and Investment of China’s Translator Industry in 2020 issued by askci
Corporation, it is expected that the market scale of translation machines will reach 30 to
40 million in three to five years, and the sales volume of translation machines in China
will exceed 100 billion yuan in 2025.
Although machine translation has many advantages such as high speed, wide cov-
erage and high economic benefits, its technology is technical, formal and computa-
tional. Although machine translation continues to explore in the interpretation of human
brain, cognition and language, it faces a series of problems and challenges because it
involves the most advanced and deepest core problems of human intelligence. The cur-
rent machine translation technology has not possessed the strength of human brain cell
computing and ability of cell neural connection and operation. Therefore, although the
mainstream machine translation technology has been neural machine translation based
on deep learning, it has not revealed the mystery of human brain. Even if the deep learn-
ing model can be used to deal with translation problems, machine translation cannot
completely replace human translation in a short time, The mode of human-computer
co-translation is still the direction of development in the future. The human-computer
co-translation platform will gather more machine capacity and the wisdom of manual
translators, which will be more closely combined in the operation of actual translation
projects. The machine capacity will make more use of the platform to learn manual
capacity and output translations close to manual capacity and quality.
5 Conclusions
Human-Computer Co-translation will gather more machine capacity and artificial intel-
ligence, which will be more closely combined in the operation of actual translation
projects. With the development of neural network technology, machine translation
improves the accuracy of translation, and the role of translator mainly turns to post-
translational editing. The translation platform integrating machine translation, computer-
aided translation and translation management systems provides a more efficient transla-
tion environment for human translation. The translation platform can be integrated, iter-
ated and upgraded continuously, and effectively meet the needs of human personalized
translation.
On Neural Machine Translation Based on Cloud Platform 443
Acknowledgements. This work was financially supported by Nanchang Institute of Technology
Teaching Reform Project: An Empirical Study of Blended Interpretation Teaching Mode under
the Background of “Internet Plus” Project No. 2020JG043.
References
1. Bundgaard, K. , Christensen, T. P. , 2016, Schjoldager A. “Translator-computer interaction
in action - an observational process study of computer-aided translation”, The Journal of
Specialised Translation
2. Christensen, T. P. , Schjoldager, A .2016 “Computer-aided translation tools - the uptake and
use by Danish translation service providers, The Journal of Specialised Translation
3. Excell, D. 2019, Some Challenges in Using Computer-Aided Translation Tools to Facilitate
Second Language Fluency in Education. Annals of Emerging Technologies in Computing
4. Xiao, T., Li, Y.Q., Chen, Q., Zhu,J.B., 2018, Machine Translation in the Era of Deep Learning.
Artificial Intelligence
5. Yan, X., Chen, R.Z., Zhang, J. 2019, The development Status andTrend of Translation
Technology Cloud Platform. Chinese Science & Technology Translator Journal
6. Yin, F.Z., Yu, C.F., Deng, Y.L., 2020, Ten Chapters about Translation Communications, Hunan
Normal University Press, Changsha
Open Access This chapter is licensed under the terms of the Creative Commons Attribution-
NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/),
which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any
medium or format, as long as you give appropriate credit to the original author(s) and the source,
provide a link to the Creative Commons license and indicate if changes were made.
The images or other third party material in this chapter are included in the chapter’s Creative
Commons license, unless indicated otherwise in a credit line to the material. If material is not
included in the chapter’s Creative Commons license and your intended use is not permitted by
statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder.
