Design and Testing of Automatic Machine Translation System ...

Research ArticleDesign and Testing of Automatic Machine Translation SystemBased on Chinese-English Phrase Translation

Jing Ning and Haidong Ban

College of Foreign Language Xijing University Xirsquoan 710123 Shaanxi China

Correspondence should be addressed to Jing Ning 20120080xijingeducn

Received 29 June 2021 Revised 2 September 2021 Accepted 9 September 2021 Published 30 September 2021

Academic Editor Sang-Bing Tsai

Copyright copy 2021 Jing Ning and Haidong Ban +is is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

With the development of linguistics and the improvement of computer performance the effect of machine translation is gettingbetter and better and it is widely used+e automatic expression translation method based on the Chinese-English machine takesshort sentences as the basic translation unit and makes full use of the order of short sentences Compared with word-basedstatistical machine translation methods the effect is greatly improved +e performance of machine translation is constantlyimproving +is article aims to study the design of phrase-based automatic machine translation systems by introducing machinetranslation methods and Chinese-English phrase translation explore the design and testing of machine automatic translationsystems based on the combination of Chinese-English phrase translation and explain the role of machine automatic translation inpromoting the development of translation In this article through the combination of machine translation experiments andmachine automatic translation system design methods the design and testing of machine automatic translation systems based onChinese-English phrase translation combinations are studied to cultivate peoplersquos understanding of language knowledge andintelligence and then help solve other problems Language processing issues promote the development of corpus linguistics +eexperimental results in this article show that when the Chinese-English phrase translation probability table is changed from 82to 51 the BLEU translation evaluation system for the combination of Chinese-English phrases is improved Automatic machinetranslation saves time and energy of translation work which shows that machine translation shows its advantages due to its shortdevelopment cycle and easy processing of large-scale corpora

1 Introduction

People express their emotions through language which is animportant tool for communication between people+erefore it is more and more important to overcomecommunication barriers between languages in the 21stcentury Machine automatic translation is a meaningful andcomplicated research and full of challenges and difficultiesContinuous and high-quality automatic translation machineresearch is one of the ultimate goals of computing andlanguage research which is the main trend of futuredevelopment

Machine automatic translation is becoming more andmore important in todayrsquos society and its potential is hugewith the rapid economic development Everyday peoplefrom all walks of life deal with a large number of documents

and people of different languages communicate with eachother +erefore machine automatic translation has a greatmarket demand and only a very large amount of infor-mation can meet the needs of translation With the com-bination of Chinese and English automatic machinetranslation has become the most common method atpresent which has the benefit of greatly facilitating peoplersquoslives and it is also a simple data warehouse

Sangeetha and Jothilakshmi proposed a speech-to-speech translation system which mainly focuses on trans-lation from English to Dravidian +e three main technol-ogies involved in the SST system are automatic continuousspeech recognition machine translation and text-to-speechsynthesis systems Based on automatic associative neuralnetwork vector support mechanism and hidden Markovmodel automatic continuous speech recognition has been

HindawiMobile Information SystemsVolume 2021 Article ID 3539155 8 pageshttpsdoiorg10115520213539155

developed Compared with SVM and AANN HMM pro-duces better results but it currently lacks specific data toprove [1] Shereen and Mohamed believes that deaf-mutepeople are an important part of the growing community andthey use sign language However communication betweennormal people and hearing-impaired people becomes dif-ficult because most normal people cannot understand themeaning of sign language gestures while deaf-mute peoplecannot understand natural spoken language +ere are ap-proximately 70 million deaf and hearing-impaired people inthe world as well as people who use sign language as theirmother tongue or mother tongue+e analysis of the existingsystem provides us with the necessary information about itswork process success rate shortcomings and limitationsand its development is relatively vague [2] In order toimprove the accuracy of automatic machine translationSangeetha and Jothilakshmi proposed a study to improve theefficiency of machine translation when necessary For thisreason based on the adjustment of English context and themutual information between words in English words theyproposed an automatic translation system based on semanticrelations [1]

+e innovation of this article lies in the investigation andstudy of the method probability of Chinese-English phrasetranslation and the combined machine automatic translationsystem of Chinese-English phrase translation +e systematicresearch and experimentation of automatic translators are ofgreat significance To a certain extent it can promote the rapidand in-depth dissemination of international information

2 Machine Translation Method of Chinese-English Phrases

21 Process of Machine Translation of Phrases Machinetranslation experiments found that the translationmodel in thebasic IBM machine translation equation was replaced by thereverse translation model but the accuracy of translation wasnot reduced by the automatic translationmachine which couldnot be passed through the channel theory [3 4] +erefore themaximum entropy based on machine translation is proposed+is more general method is a statistical method of machinetranslation based on the source channel [5 6] Characterizingthe maximum entropy language format and translation modeand adding them to the model framework the main advantageis that it can easily integrate knowledge sources and auto-matically weight between knowledge sources Most currentstatistical machine translation methods use the highest entropymodeling framework [7 8]+e automatic machine translationmodeling of phrases is shown in Figure 1

22 Method and Process of Machine Translation Based onPhrase Structure

221 Corpus Preprocessing +e processing level of thecorpus directly affects the translation results Statisticalmachine translation usually uses a bilingual corpus andprepares Chinese and English corpora separately [9 10] +eresults of corpus preprocessing are shown in Table 1

222 Implementing the Title Translation System in theAviation Field On the basis of researching related machinetranslation theory using some existing resources and toolswe complete the phrase translation model module realizethe phrase-based statistical machine translation system andintroduce the basic working principle of the system systemimplementation and system operating environment settingsand parameters [11 12]

223 Automatic Evaluation Technology of MachineTranslation Based on the research of machine automatictranslation technology the results of Chinese-English ma-chine automatic translation are automatically evaluated Inthe field of statistical machine learning there are alreadysome methods to solve domain adaptation problems[13 14] But most of them are only used to solve simplelearning problems (such as classification or regression) Inthe face of structured learning problems such as machinetranslation different domain adaptive methods are used tosolve them separately under the machine learning frame-work [15] +e application of machine translation automaticevaluation technology is shown in Figure 2

23 Stack Search Translation Method Stack search utilizes aresearch and exploratory method Before strengthening thesearch of n heaps the number n is the number of words inthe source language sentence and each state data hypothesisis stored in the stack extension [16] ldquoIrdquo is translated as ldquosherdquoand ldquoflowerrdquo is derived from the word ldquoflowersrdquo Both hy-potheses are in the first stack of the stack search translationmethod [7] Also in the second stack the molecule adds

Machinetranslation of

phrases

Figure 1 Automatic machine translation modeling of phrases

Table 1 Experimental data

Step Complaint text1 Introduce custom dictionary2 Segment words and mark part of speech3 Remove stop words4 Keep important part of speech words5 Keep processing results

2 Mobile Information Systems

information about the source language translation of the twoterms For the source language words that have beentranslated the stack cost is low so they are determined as thebest translation [8] +e stack search conversion is shown inFigure 3

3 Chinese-English Machine TranslationProbability Experiment

31 Phrase-Based Statistical Machine Translation +e basicidea is to use phrases as the basic unit of translation In theprocess of transfer everyonersquos translation of phrases is not thesame everywhere and there are various opinions and inter-pretations at the same time In grammatical sense if only thephrase lines are not continuous we still need to solve theproblem of the overall coherence of the full text In order toexpand the transmission of these contents we can easily solvethe local problem in the same way Context-dependent issuesand explanations of phrases in all languages using this methodcan maintain the original state of the language to the greatestextent Generally speaking the so-called free grammar methodcan be a continuous line subnavigation +erefore Chinese-English translation of words must be carried out to extract theviewpoint of double-body protection and the process of rule-based machine translation is shown in Table 2

32 Defining the Format of Phrase Translation ProbabilityTable In the output file of the phrase output module eachline contains some Chinese phrases English phrases andtranslation probability values

P( brarr

ararr

) N( b

rarr ararr

)

1113936tN( brarr

ararr

) (1)

Lexicalized translation probability

lex aj b

j c1113872 1113873 1113945

j

jm

1

| i(j i) isin c| 1113936j

p aj bj1113872 1113873⎧⎨

⎩

(2)

+e BLEU evaluation tool is currently the most widelyused indicator in international machine translation

evaluation It compares the system translation with thereference translation calculates the accuracy of each systemtranslation and finally records the entire translation It iscalculated as follows

sore BPlowast exp 1113944N

n1fnlog hn

⎛⎝ ⎞⎠Wi (3)

33 Vector Machine Algorithm Where P is the penaltylength factor B is the shortest length of the referencetranslation of the tested sentence and R is the translationlength of the tested sentence that is the number of wordscontained in the entire output translation

Figure 2 Application of machine translation automatic evaluation technology

2754

15

14

12

23

3

2140

30

36

30 60

Edge

Leafnodes

Subtree

Siblings

Figure 3 Stack search conversion

Table 2 Rule-based machine translation process

Step Translation Knowledge base1 Source language text Target2 Lexical and morphological analysis Rule base3 Syntax analysis Dictionary4 Semantic analysis Target language text5 Structure transformation Language generation

Mobile Information Systems 3

BP min 1 exp 1 minusRref

Rsys1113888 11138891113896 (4)

In the current statistical method the shared mo-dernity of indecent words indicates the fidelity oftranslation It means that a word has been translated inthe original text and a dictionary with more than twoyuan appears at the same time to indicate the fluency ofthe target language

Hn 1113936

uu1 δn

u 11113936uan

(5)

+e way and form of this formula equal to half is cal-culated as follows

score 2lowastplowast r

p + r (6)

+is is the minimum error rate during editing +e scoreranges from 0 to 1 +e scores are different for editing +eso-called edit distance is the minimum cost of insertiondeletion and replacement operations performed by con-verting the system output into a reference translation

P M

L (7)

34 Automatic Evaluation Model of Machine Translation+e logarithmic linear model is introduced into statis-tical translation which can add any number of features tothe translation process and determine the contribution ofeach feature to the translation result by weighting thesefeatures +erefore the effect of phrase-based translationsystem developed by them is far better than that of word-based translation system For formal syntax model rulesthe formula is as follows

F Fa

y1113888 1113889

count(a y)

1113936acount ai y1113872 1113873

F Fa

r a1113874 1113875 1113945

n

i1

1j(i j) isin a1113864 1113865

F Fr

a1113874 1113875

count(a y)

1113936rcount a yi

1113872 1113873

(8)

According to CKY algorithm we can construct hyper-graph from sentences of source language When we calculatethe k-best derivation of a node the ranking of the dimensionof rules no longer only depends on the score of syntactic rulefeatures We use heuristic function H (R) to sort the rules

h(Y) u 1113944r

i0InP wi( 1113857 minus u(r + 1)

f a middot middot middot al( 1113857 1113945 flm

ai

a middot middot middot aiminus11113888 1113889

(9)

4 Chinese-English Phrase TranslationCombined Machine AutomaticTranslation System

41 Phrase-Based Statistical Machine Translation +e basicidea is to use machine translation as the basic unit of phrasetranslation In the translation process each translated wordmust be combined with context and constrained translationduring the translation process But generally speaking nogrammar is performed in the same way In this way the two-body alignment should be removed from the bilingual ex-cerpt Given a source language sentence the sentences usedfor the translation process model are as follows the source isdivided into phrase sentences and language word view-points and the order is adjusted according to the inter-pretation target model of each sentence Phrases are used asthe basic unit of translation +e Chinese sentence inter-pretation system is used to divide many sentences into so-called ldquophrasesrdquo and then translate them into English +egenerated phrases and output are shown in Table 3

42 Translation Process It mainly includes the followingparts model phrase translation translation model traininglanguage training and trial transmission of decoding results+ese parts are scattered in the form of a flowchart From theperspective of the translation science model each table isbest to learn Chinese phrases from the English interpretationof English sentences and arrange them in a row as shown inthe flowchart in Figure 4

Traditional word alignment-based heuristic phrase ex-traction methods will have word alignment errors and word-to-space problems which leads to the loss of many bisyn-tactic phrases On the other hand the bilingual phrasesextracted from bilingual phrases in this paper are bilingualphrases with better quality +erefore we consider addingthe extracted bisyntactic phrases to the phrase table to makeup for the bisyntactic phrases lost by the heuristic phraseextraction method +e experiment uses the providedtraining set development set and test set +e source lan-guage of the corpus is Chinese and the target language isEnglish +e scale of the experimental data is shown inTable 4

It can be seen from the table that English sentences areon average longer than Chinese sentences and both Chineseand English sentences are longer especially when the av-erage length of English sentences reaches one word whichbrings difficulty to syntactic analysis We analyze the syntaxof the source language and the target language and extractbilingual phrases using an iterative phrase extraction algo-rithm According to the Chinese-English phrase translationtraining set there are 120000 Chinese-English bilinguallyaligned sentences and the test corpus contains 141 sen-tences In the experiment this paper uses the C value and thedegree of adhesion to reduce the source language It is addedto the translation model as a function and the translationresults are compared with the reference frame First nomatter how long the sentence is the possibility of translationis lower than the C value of the source code and it can be


seen that the BLEU evaluation can be improved by 002 atmost compared with the benchmark system while thephrase translation probability table is only 78 of theoriginal When the phrase translation probability table isreduced to 51 of the original the BLEU evaluation is stillslightly higher than the benchmark system +e experi-mental results are shown in Figure 5

+e input of the bilingual phrase extraction algorithm isan aligned Chinese-English bilingual tree so it is necessaryto perform syntactic analysis on the source language end andthe target language end of the training corpus separately+ebilingual phrase extraction algorithm extracts bilingualphrases based on word alignment +e training corpus is thetraining corpus that has been word aligned so it is no longernecessary to apply word alignment to the training corpusWe run the bisyntactic phrase extraction algorithm and

temporarily store the extracted bisyntactic phrases +isexperiment needs to run four different machine translationsystems +ese systems are statistical machines based onbisyntactic phrases that are generated after the extractedbisyntactic phrases are applied to the system

43 Chinese-English Translation Corpus +e Chinese-Englishtranslation corpus is used +is corpus contains more than10000 words and 10000 pairs of sentences +is article findsthat the best translation effect can be achieved by using it as ameans of word alignment extraction in Chinese-Englishtranslation Used as the evaluation standard the calculationscript adopts the standard script +is article uses tenthousand sentence pairs in ten thousand pairs of sentencesas the training corpus and the number of short sentence

Table 3 Example of phrase-based statistical machine translation process

Original She will attend the party on May 1stPhrase division She will On May 1st Attend the partyTranslation She will On May 1st Attend the partyAdjusting the order She will Attend the party On May 1st

Training corpus

Chinese segmentation tool

Corpus format conversion

Standard training library

Training corpus pre-processing

Word segmentation training corpsus

Words are right

Harmony corpus of grammar

Short phrase extract

chinesesegmentati

on tool Engl

ishse

gmen

tatio

n to

ol

Figure 4 Phrase translation model training process


Chinese EnglishSentences in the training collection 99933 99777Number of words in training set 245699 348766+e training setrsquos average sentence length 25 30Number of sets 400 400Development setrsquos number of single words 1012 1122Test setrsquos average sentence length 25 33


pairs extracted by themethod is regarded as the parameter inthe linear rearrangement model as shown in Figure 6

As shown in Figure 6 the fragment probability phrasesare used in the Chinese-English translation thereby im-proving performance +rough the data test of an exampleafter completing the translation process machine transla-tion is introduced into the system and a partition system isestablished +e model and module are given and theexisting local resources and document resources are used

87 8577 75

60

69

58

0

65

047 045 044052

045

066 065

041037

0

10

20

30

40

50

60

70

80

90

100

Prob

abili

ty (

)

phrase table sizeBLEU C value threshhold

C va

lue

07

05

09

15

18

24

25

26

Figure 5 Phrase translation probability table size and BLEU value under different C value thresholds

2555 25372245

26872567 2833

0

5

10

15

20

25

30

dl=0 dl=3

Perio

d

Athletes

Figure 6 +e result of the improved system under Chinese-English translation

Table 5 Dataset characteristics

DatasetChinese English

Sentence number Word number Sentence number Word numberTraining set 3000 230 4000 4455Test set 4000 456 7000 788Development set 490 2788 3891 342

Table 6 +e characteristics of phrase model and formal syntaxmodel

Model Phrase modelMaximum rule length 20Model number 334KModel size 12KBefore filtering 34MBAfter filtering 76MB


including some open source translation tools and publiclyauthorized translation tools +ese tools are based on theresearch and development of the comprehensive decision-making mechanism of the statistical system

44ModelTrainingandParameter Setting +e evaluation ofmachine translation mainly includes manual evaluationand automatic evaluation +e advantage of manual eval-uation is high accuracy but the disadvantage is that thelabor cost and time cost are too high +e advantages ofautomatic evaluation are low cost fast speed and the abilityto be used repeatedly +e disadvantage is low accuracy Atpresent the focus of machine translation evaluation re-search is how to improve the rate of automatic evaluation+e test set of CSTAR 2003 is the development set of theexperiment Some features of the corpus are shown inTable 5

+e phrase is extracted from the training set and theEnglish part of the training set is trained by language modeltool +e feature model of phrase model and formal syntaxmodel is reduced by a 3-element language model in order tospeed up the training of minimum error rate and savememory space the development set and test set are used tofilter these models +e characteristics of the model areshown in Table 6

+e evaluation of machine translation plays an importantrole in the research of machine translation technology and thepromotion of market Manual evaluation refers to the evalu-ation of candidate translations given by machine translationsystem according to certain standards and norms Automaticevaluation is the use of machines to complete the scoringprocess but it requires that the results of scoring are consistentas much as possible with the personrsquos score the training ofmachine translation is shown in Figure 7

Machine translation evaluation in short is the evaluation ofall aspects of machine translation in order to correctly andobjectively reflect the achievements and functions of machinetranslation+e significance ofmachine translation evaluation isto find out the problems existing in the research and devel-opment of machine translation system by evaluating the per-formance and development level of machine translation definethe goal find solutions provide direction for the improvementof the existing machine translation system and constantlyimprove the translation quality of machine translation systemthe paradigm of machine translation is shown in Table 7

Machine translation is a reliable way to evaluate theperformance of a translation system However it usuallytakes time and effort to organize a manual evaluation +euse of automatic evaluation tools can greatly reduce the costof evaluation analyze the system performance in time

Initialparameter model

decodebatch

newparameter

calculateloss

finalparameter

automatic translation to reference translation

Figure 7 Training of machine translation

Table 7 +e paradigm of machine translation

Intermediate language Transformation method Direct translation method1 Source language deep representation Target language deep representation2 Source language text Target language text3 Chinese English

Figure 8 Neural machine translation system (httpalturlcomacrvo)


improve the system targeted and shorten the product de-velopment cycle the neural machine translation system isshown in Figure 8

5 Conclusions

+is article extends the discussion from the perspective ofautomatic translation systems for mechanical design +emachine translation system is a large-scale system composedof several modules which can complete the translationwork+is article makes full use of the existing resources andtools in the literature briefly describes the phrase andprobability of phrase translation and integrates these toolsand modules and we believe that building a machinetranslation system based on statistical results means anattempt that cannot be done by learning translators Au-tomatic machine translation is a complete process that in-tegrates the development of concepts opens up the use ofexisting resources and adds modules such as repositoriesdictionaries and so on +e decision is based on the resultsof statistical machine translation methods that can achievebetter translation results

Data Availability

No data were used to support this study

Conflicts of Interest

+e authors declare that they have no conflicts of interest

Acknowledgments

+is study was supported by the Scientific Research Programfunded by Shaanxi Provincial Education Department (grantno 18JK1188) and the Scientific Research Foundation ofXijing University (grant nos XJ180113 and XJ130134)

References

[1] J Sangeetha and S Jothilakshmi ldquoSpeech translation systemfor English to dravidian languagesrdquo Applied Intelligencevol 46 no 3 pp 534ndash550 2017

[2] A Shereen and A Mohamed ldquoA cascaded speech to Arabicsign language machine translator using adaptationrdquo Inter-national Journal of Computer Application vol 133 no 5pp 5ndash9 2016

[3] J Zhang Y Zhou and C Zong ldquoAbstractive cross-languagesummarization via translation model enhanced predicateargument structure fusingrdquo IEEEACM Transactions on Au-dio Speech and Language Processing vol 24 no 10pp 1842ndash1853 2016

[4] Z Qin P Wang J Sun J Lu and H Qiao ldquoPrecise roboticassembly for large-scale objects based on automatic guidanceand alignmentrdquo IEEE Transactions on Instrumentation andMeasurement vol 65 no 6 pp 1398ndash1411 2016

[5] H A Bouarara R M Hamou and A Rahmani ldquoBHA2 bio-inspired algorithm and automatic summarisation fordetecting different types of plagiarismrdquo International Journalof Swarm Intelligence Research vol 8 no 1 pp 30ndash53 2017

[6] D Tolic and S Hirche ldquoStabilizing transmission intervals fornonlinear delayed networked control systemsrdquo IEEE Trans-actions on Automatic Control vol 62 no 1 pp 488ndash494 2017

[7] D Kuehn M Schilling T Stark M Zenzes and F KirchnerldquoSystem design and testing of the hominid robot charlierdquoJournal of Field Robotics vol 34 no 4 pp 666ndash703 2017

[8] S F Rafique J Zhang M Hanan W Aslam A U Rehmanand Z W Khan ldquoEnergy management system design andtesting for smart buildings under uncertain generation (windphotovoltaic) and demandrdquo Journal of Tsinghua UniversityEnglish Edition vol 23 no 3 pp 254ndash265 2018

[9] C Wang ldquoDesign and research of ultrasonic nondestructivetesting system for conveyor beltrdquo Machinery ManagementDevelopment vol 33 no 1 pp 98ndash100 2018

[10] J Li S Yang H Zhang G Liu and T Sun ldquoDesign and fieldtesting of a nitrogen circulation drilling systemrdquo Chemistryand Technology of Fuels and Oils vol 53 no 3 pp 428ndash4352017

[11] H Totoki Y Ochi M Sato and K Muraoka ldquoDesign andtesting of a low-order flight control system for quad-tilt-wingUAVrdquo Journal of Guidance Control and Dynamics vol 39no 10 pp 2423ndash2431 2016

[12] L Yang and W Li ldquoDesign and implementation of indoorenvironment testing system based on android platformrdquoEnvironmental Science and Management vol 42 no 5pp 26ndash29 2017

[13] S Lei and Z Liping ldquoDesign and implementation of auto-matic testing system for LTE-M based TAUrdquo ElectronicsWorld no 14 pp 40-41 2017

[14] S Zhou D Zou and T Xiao ldquoDesign and experiment of thevelocity-pressure characteristic testing system for seafloorsedimentsrdquo Ocean Technology vol 36 no 5 pp 55ndash61 2017

[15] Y Cheng X Chen and H Wang ldquoDesign and precisionanalysis for PLC-based energy efficiency testing system ofelectric fansrdquo Journal of Testing Technology vol 30 no 1pp 1ndash5 2016

[16] Y Xu W Haikun and S Fang ldquo+e design of the testingsystem of the diesel generator under the low temperature andlow pressurerdquo Electrical Automation vol 38 no 3 pp 85ndash872016


developed Compared with SVM and AANN HMM pro-duces better results but it currently lacks specific data toprove [1] Shereen and Mohamed believes that deaf-mutepeople are an important part of the growing community andthey use sign language However communication betweennormal people and hearing-impaired people becomes dif-ficult because most normal people cannot understand themeaning of sign language gestures while deaf-mute peoplecannot understand natural spoken language +ere are ap-proximately 70 million deaf and hearing-impaired people inthe world as well as people who use sign language as theirmother tongue or mother tongue+e analysis of the existingsystem provides us with the necessary information about itswork process success rate shortcomings and limitationsand its development is relatively vague [2] In order toimprove the accuracy of automatic machine translationSangeetha and Jothilakshmi proposed a study to improve theefficiency of machine translation when necessary For thisreason based on the adjustment of English context and themutual information between words in English words theyproposed an automatic translation system based on semanticrelations [1]

+e innovation of this article lies in the investigation andstudy of the method probability of Chinese-English phrasetranslation and the combined machine automatic translationsystem of Chinese-English phrase translation +e systematicresearch and experimentation of automatic translators are ofgreat significance To a certain extent it can promote the rapidand in-depth dissemination of international information

2 Machine Translation Method of Chinese-English Phrases

21 Process of Machine Translation of Phrases Machinetranslation experiments found that the translationmodel in thebasic IBM machine translation equation was replaced by thereverse translation model but the accuracy of translation wasnot reduced by the automatic translationmachine which couldnot be passed through the channel theory [3 4] +erefore themaximum entropy based on machine translation is proposed+is more general method is a statistical method of machinetranslation based on the source channel [5 6] Characterizingthe maximum entropy language format and translation modeand adding them to the model framework the main advantageis that it can easily integrate knowledge sources and auto-matically weight between knowledge sources Most currentstatistical machine translation methods use the highest entropymodeling framework [7 8]+e automatic machine translationmodeling of phrases is shown in Figure 1

22 Method and Process of Machine Translation Based onPhrase Structure

221 Corpus Preprocessing +e processing level of thecorpus directly affects the translation results Statisticalmachine translation usually uses a bilingual corpus andprepares Chinese and English corpora separately [9 10] +eresults of corpus preprocessing are shown in Table 1

222 Implementing the Title Translation System in theAviation Field On the basis of researching related machinetranslation theory using some existing resources and toolswe complete the phrase translation model module realizethe phrase-based statistical machine translation system andintroduce the basic working principle of the system systemimplementation and system operating environment settingsand parameters [11 12]

223 Automatic Evaluation Technology of MachineTranslation Based on the research of machine automatictranslation technology the results of Chinese-English ma-chine automatic translation are automatically evaluated Inthe field of statistical machine learning there are alreadysome methods to solve domain adaptation problems[13 14] But most of them are only used to solve simplelearning problems (such as classification or regression) Inthe face of structured learning problems such as machinetranslation different domain adaptive methods are used tosolve them separately under the machine learning frame-work [15] +e application of machine translation automaticevaluation technology is shown in Figure 2

23 Stack Search Translation Method Stack search utilizes aresearch and exploratory method Before strengthening thesearch of n heaps the number n is the number of words inthe source language sentence and each state data hypothesisis stored in the stack extension [16] ldquoIrdquo is translated as ldquosherdquoand ldquoflowerrdquo is derived from the word ldquoflowersrdquo Both hy-potheses are in the first stack of the stack search translationmethod [7] Also in the second stack the molecule adds

Machinetranslation of

phrases

Figure 1 Automatic machine translation modeling of phrases


Step Complaint text1 Introduce custom dictionary2 Segment words and mark part of speech3 Remove stop words4 Keep important part of speech words5 Keep processing results






P( brarr

ararr

) N( b

rarr ararr

)

1113936tN( brarr

ararr

) (1)


lex aj b

j c1113872 1113873 1113945

j

jm

1

| i(j i) isin c| 1113936j

p aj bj1113872 1113873⎧⎨

⎩

(2)




n1fnlog hn

⎛⎝ ⎞⎠Wi (3)



2754

15

14

12

23

3

2140

30

36

30 60

Edge

Leafnodes

Subtree

Siblings






Rsys1113888 11138891113896 (4)


Hn 1113936

uu1 δn

u 11113936uan

(5)



p + r (6)


P M

L (7)


F Fa

y1113888 1113889

count(a y)

1113936acount ai y1113872 1113873

F Fa

r a1113874 1113875 1113945

n

i1

1j(i j) isin a1113864 1113865

F Fr

a1113874 1113875

count(a y)

1113936rcount a yi

1113872 1113873

(8)


h(Y) u 1113944r



ai


(9)













Training corpus






Words are right



chinesesegmentati

on tool Engl

ishse

gmen

tatio

n to

ol







87 8577 75

60

69

58

0

65

047 045 044052

045

066 065

041037

0

10

20

30

40

50

60

70

80

90

100

Prob

abili

ty (

)


C va

lue

07

05

09

15

18

24

25

26


2555 25372245

26872567 2833

0

5

10

15

20

25

30

dl=0 dl=3

Perio

d

Athletes















decodebatch

newparameter

calculateloss

finalparameter








5 Conclusions


Data Availability




Acknowledgments


References






















P( brarr

ararr

) N( b

rarr ararr

)

1113936tN( brarr

ararr

) (1)


lex aj b

j c1113872 1113873 1113945

j

jm

1

| i(j i) isin c| 1113936j

p aj bj1113872 1113873⎧⎨

⎩

(2)




n1fnlog hn

⎛⎝ ⎞⎠Wi (3)



2754

15

14

12

23

3

2140

30

36

30 60

Edge

Leafnodes

Subtree

Siblings






Rsys1113888 11138891113896 (4)


Hn 1113936

uu1 δn

u 11113936uan

(5)



p + r (6)


P M

L (7)


F Fa

y1113888 1113889

count(a y)

1113936acount ai y1113872 1113873

F Fa

r a1113874 1113875 1113945

n

i1

1j(i j) isin a1113864 1113865

F Fr

a1113874 1113875

count(a y)

1113936rcount a yi

1113872 1113873

(8)


h(Y) u 1113944r



ai


(9)













Training corpus






Words are right



chinesesegmentati

on tool Engl

ishse

gmen

tatio

n to

ol







87 8577 75

60

69

58

0

65

047 045 044052

045

066 065

041037

0

10

20

30

40

50

60

70

80

90

100

Prob

abili

ty (

)


C va

lue

07

05

09

15

18

24

25

26


2555 25372245

26872567 2833

0

5

10

15

20

25

30

dl=0 dl=3

Perio

d

Athletes















decodebatch

newparameter

calculateloss

finalparameter








5 Conclusions


Data Availability




Acknowledgments


References



















Rsys1113888 11138891113896 (4)


Hn 1113936

uu1 δn

u 11113936uan

(5)



p + r (6)


P M

L (7)


F Fa

y1113888 1113889

count(a y)

1113936acount ai y1113872 1113873

F Fa

r a1113874 1113875 1113945

n

i1

1j(i j) isin a1113864 1113865

F Fr

a1113874 1113875

count(a y)

1113936rcount a yi

1113872 1113873

(8)


h(Y) u 1113944r



ai


(9)













Training corpus






Words are right



chinesesegmentati

on tool Engl

ishse

gmen

tatio

n to

ol







87 8577 75

60

69

58

0

65

047 045 044052

045

066 065

041037

0

10

20

30

40

50

60

70

80

90

100

Prob

abili

ty (

)


C va

lue

07

05

09

15

18

24

25

26


2555 25372245

26872567 2833

0

5

10

15

20

25

30

dl=0 dl=3

Perio

d

Athletes















decodebatch

newparameter

calculateloss

finalparameter








5 Conclusions


Data Availability




Acknowledgments


References
























Training corpus






Words are right



chinesesegmentati

on tool Engl

ishse

gmen

tatio

n to

ol







87 8577 75

60

69

58

0

65

047 045 044052

045

066 065

041037

0

10

20

30

40

50

60

70

80

90

100

Prob

abili

ty (

)


C va

lue

07

05

09

15

18

24

25

26


2555 25372245

26872567 2833

0

5

10

15

20

25

30

dl=0 dl=3

Perio

d

Athletes















decodebatch

newparameter

calculateloss

finalparameter








5 Conclusions


Data Availability




Acknowledgments


References




















87 8577 75

60

69

58

0

65

047 045 044052

045

066 065

041037

0

10

20

30

40

50

60

70

80

90

100

Prob

abili

ty (

)


C va

lue

07

05

09

15

18

24

25

26


2555 25372245

26872567 2833

0

5

10

15

20

25

30

dl=0 dl=3

Perio

d

Athletes















decodebatch

newparameter

calculateloss

finalparameter








5 Conclusions


Data Availability




Acknowledgments


References

























decodebatch

newparameter

calculateloss

finalparameter








5 Conclusions


Data Availability




Acknowledgments


References



















5 Conclusions


Data Availability




Acknowledgments


References


















Date post:	23-Oct-2021
Category:	Documents
Upload:	others
View:	4 times
Download:	0 times

Design and Testing of Automatic Machine Translation System ...

Documents