CN107301173A - A kind of automatic post-editing system and method for multi-source neutral net that mode is remixed based on splicing - Google Patents
A kind of automatic post-editing system and method for multi-source neutral net that mode is remixed based on splicing Download PDFInfo
- Publication number
- CN107301173A CN107301173A CN201710491848.2A CN201710491848A CN107301173A CN 107301173 A CN107301173 A CN 107301173A CN 201710491848 A CN201710491848 A CN 201710491848A CN 107301173 A CN107301173 A CN 107301173A
- Authority
- CN
- China
- Prior art keywords
- language material
- translation
- source
- training
- post
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/40—Processing or translation of natural language
- G06F40/58—Use of machine translation, e.g. for multi-lingual retrieval, for server-side translation for client devices or for real-time translation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a kind of automatic post-editing system and method for multi-source neutral net that mode is remixed based on splicing, belong to Computer Natural Language Processing and machine translation mothod field.Including the system, training module and decoder module are included again；This method is divided into training process and decoding process.Training system process is set up on traditional neural network machine translation model basis, source language material is replaced with by translation original text with the new language material that preliminary translation result is generated after simple sentence splices and remixes, target language material is replaced with by double reference translation, preliminary translation result is set to be aided in mutually in the training process with translation original text, cross validation.The system trained and obtained can be used directly in translation decoding process, translation original text is decoded to preliminary translation result by the corresponding source language material spliced, translation is obtained in fluency, the degree of accuracy and quality on the whole in the preliminary translation result acted on without this post-editing method.
Description
Technical field
The present invention relates to a kind of multi-source neutral net post-editing system and method remixed based on splicing, belong to calculating
Machine application, natural language processing and machine translation mothod field.
Technical background
In recent years, with the propulsion of globalization wave, international exchange is increasingly frequent, demand of all trades and professions to translation service
It is all more urgent.Although machine translation has more efficient easily advantage, however, its translation still has not relative to artificial translation
Small gap.Therefore, the post-editing automated to machine translation result has important practical valency to improve translation quality
Value.
The automatic post-editing system of neutral net is the improvement to traditional automatic post-editing, it be good at generation fluency compared with
High sentence, can improve the word order problem of machine translation translation.The existing automatic post-editing system of neutral net is mostly single
It is pure using preliminary translation result as the original language of input, the raising in terms of language fluency degree is realized substantially, but can not be complete
, often there is more serious leakage and translates problem, reduce overall translation quality in the information content of reduction translation original text.
The content of the invention
Problem is translated the invention aims to a large amount of leakages for solving to occur during existing neutral net post-editing, is carried
Go out a kind of multi-source neutral net post-editing system and method remixed based on splicing.
A kind of multi-source neutral net post-editing system and method remixed based on splicing is based on splicing again including a kind of
The multi-source neutral net post-editing system of mixing, referred to as the system, and a kind of multi-source neutral net remixed based on splicing
Post-editing method, referred to as this method；
Wherein, multi-source refers to the input of post-editing and may come from a variety of different original language, including machine translation translation
With translation original text；
The system enables to translation original text to be influenced each other with preliminary translation result during post-editing, and intersection is tested
Card, improves translation informativeness, and then improve the total quality of post-editing result；
The system includes training module and decoder module；
The function of each module is as follows：
The function of training module is the automatic post-editing system of multi-source neutral net that training remixes mode based on splicing,
Export training pattern；This training pattern be also known as post-editing system model；
The function of decoder module is that the post-editing system model exported using training module is decoded；
Annexation between each module is as follows：
Training module is connected with decoder module, the training pattern exported especially by training module, i.e. post-editing system
Model of uniting is connected.
To achieve the above object, the technical solution adopted in the present invention is as follows：
Define 1：One preliminary machine translation system, referred to as Machine Translation, i.e. MT systems are set；
Define 2：One constant N is set, training original text and the language of reference translation for training module in this method is represented
Material has assumed that N；
Define 3：One constant M is set, the translation original text in this method for decoder module is represented and assumes there are M；
On basis defined above, this method includes the training process of training module and the decoding process of decoder module
Two parts, wherein training module complete the instruction of the automatic post-editing system of multi-source neutral net to remixing mode based on splicing
Practice, export training pattern；The training pattern that decoding process is exported using training process is decoded；
The training process of training module, be specially：
Step 1: each language material required for the system training process is collected, and to training original text language material therein through MT systems
System is tentatively translated, and draws preliminary translation result language material；
Wherein, each language material mainly includes training original text language material and reference translation language material；Wherein, training original text language material and reference
Translation language material is bilingual parallel corporas；
Original text language material is trained, is designated as：{source1,source2,…,sourceN,
Translation language material is trained, { ref is designated as1,ref2,…,refN,
The preliminary translation result language material of original text language material is trained, is designated as：
{mt-outs1,mt-outs2,…,mt-outsN}；
Step 2: being spliced and combined to the language material of step one, rear language material is translated before output source；
Rear language material is translated before source, is designated as：
{sourcemt-outs1,sourcemt-outs2,…,sourcemt-outsN, per in short according to training original text
Preceding, its corresponding preliminary posterior order of translation result is spliced successively；
Step 3: being spliced and combined to the language material of step one, preceding language material is translated after output source；
Language material before being translated behind source, is designated as：
{mt-outsssource1,mt-outssource2,…,mt-outssourceN, per former in short according to training
Wen Hou, and its corresponding preceding order of preliminary translation result are spliced successively；
Integrally mixed again Step 4: being translated before the source exported to step 2 and step 3 and preceding language material being translated behind rear language material and source
Close, draw mixing language material, be used as the source language material of training process；
Wherein, Step 2: splicing and mixed process in step 3 and step 4 are construction multi-source translation language material
Process, that is, refer to translate the source language material that original text together form post-editing system with preliminary translation result；
Language material is mixed, is designated as：{sourcemt-outs1,sourcemt-outs2,…, sourcemt-outsN,mt-
outsssource1,mt-outssource2,…,mt- outssourceN, it is used as the source language material of training module；
Step 5: it is integrally double to the training translation language material of step one, generate the target language material of training process；
Wherein, overall double to training translation language material progress, its output is designated as：
{ref1,ref2,…,refN,ref1,ref2,…,refN, it is used as the target language material of training module；
Step 6: the target language material that the source language material obtained with step 4 is obtained with step 5 is based on neutral net translation model
The system is trained, post-editing system model is exported；
So far, from step one to step 6, the training process of training module in this method is completed；
Step 7: each language material needed in the decoding step of setting the system；
Wherein, each language material needed in decoding step mainly includes translation original text language material and preliminary translation result language material, just
Step translation result language material is obtained by translation original text language material through the translation of MT systems；
Original text language material is translated, is designated as：{src1,src2,…,srcM,
Preliminary translation result language material, is designated as：{mt1,mt2,…,mtM}；
Step 8: being spliced and combined to the language material of step 7, rear language material is translated before the source that output decoding process needs；
Rear language material is translated before source, is designated as：{srcmt1,srcmt2,…,srcmtM}；
Per in short according to translation original text, preceding, the corresponding preliminary posterior order of translation result is spliced successively；
Step 9: being spliced and combined to the language material of step 7, preceding language material is translated behind the source that output decoding process needs；
Preceding language material is translated behind source, is designated as：{mtsrc1,mtsrc2,…,mtsrcM}；
Per in short according to translation original text, rear, the corresponding preliminary preceding order of translation result is spliced successively；
Step 10: will translate before the source of step 8 and step 9 output translated behind rear language material and source before both language materials optionally first,
The post-editing system model of step 6 output is input to, post-editing translation is exported；
So far, from step 8 to step 10, the decoding process of decoder module in this method is completed.
Beneficial effect
The present invention is a kind of automatic post-editing system and method for neutral net based on multi-source mode, contrasts existing skill
Art, has the advantages that：
It is neutral net post-editing system 1. of the invention directly add neutral net post-editing process by translation original text
Training provide complete semantic support,, can be with compared with not adding the method for translation original text and by way of splicing
Extremely low cost improves the fluency of machine translation；
2. the present invention is used to translation original text and preliminary translation result on the basis of sentence splicing, progress entirety is mixed again
Method, compared with the multi-source post-editing method only spliced, neutral net can learn simultaneously to translation original text with
Preliminary two kinds of original language of translation result are to the translation process of translation, two kinds of original language effectively mutual shadow during post-editing
Ring, cross validation, while the informativeness and fluency of translation are improved, so as to improve overall translation quality.
Brief description of the drawings
Fig. 1 is a kind of training of the multi-source neutral net post-editing system and method remixed based on splicing of the present invention
Journey and decoding process.
Embodiment
Model proposed by the invention and method are based on neural network machine translation model, below in conjunction with the accompanying drawings and embodiment
The present invention will be further described.
Embodiment 1
The present embodiment combination accompanying drawing 1, describes after a kind of multi-source neutral net remixed based on splicing of the present invention is translated and compiles
Collect the detailed composition and training and decoding process of system and method.
Training module is connected with decoder module as can be seen from Figure 1.
The training process of training module is comprised the steps of：
Step A：Collect each language material required for the system training process；
Wherein, each language material mainly includes training original text language material and reference translation language material；Wherein, training original text language material and reference
Translation language material is parallel corpora；It is assumed that N=600000, that is, train original text to have 60000；
Original text language material is trained, is designated as：{source1,source2,…,source600000,
Translation language material is trained, { ref is designated as1,ref2,…,ref600000,
The preliminary translation result language material of original text language material is trained, is designated as：
{mt-outs1,mt-outs2,…,mt-outs600000}；
Wherein, preliminary translation result is obtained by training original text by the translation of Moses translation systems；
Step B：Splicing and combining for different order is carried out to step A language material, can according to every a word training original text
Preceding, its corresponding preliminary posterior order of translation result is spliced successively, and rear language material is translated before output source, can also be according to every
The training original text of a word is rear, and its corresponding preceding order of preliminary translation result is spliced successively, is translated after output source
Preceding language material；
Wherein, rear language material is translated before source, is designated as：
{sourcemt-outs1,sourcemt-outs2,…,sourcemt-outs600000,
Language material before being translated behind source, is designated as：
{mt-outsssource1,mt-outssource2,…,mt-outssource600000}；
Step C：Language material before translating behind rear language material and source is translated before the source for exporting step B integrally to mix again, is built mixed
Language material is closed, the source language material of training process is used as；
Wherein, language material is mixed, is designated as：
{sourcemt-outs1,sourcemt-outs2,…,sourcemt-outsN,
mt-outsssource1,mt-outssource2,…,mt-outssource600000}；
Step D：It is integrally double to step A reference translation language material, generate the target language material of training process；
Wherein, overall double to training translation language material progress, its output is designated as：
{ref1,ref2,…,ref600000,ref1,ref2,…,ref600000}；
Step E：The system is trained based on neutral net translation model using source language material and target language material, post-editing is exported
System model；
With based on the multi-source neutral net post-editing system for splicing the mode that remixes between decoder module and training module
It is connected, decoding process is comprised the steps of：
Step F：Each language material needed in the decoding step that the system is set, it is assumed that turning in M=1597, i.e. decoding process
Translating original text has 1597；
Wherein, each language material needed in decoding step mainly includes translation original text language material and preliminary translation result language material, just
Step translation result language material is obtained by translation original text language material through the translation of Moses translation systems；
Original text language material is translated, is designated as：{src1,src2,…,src1597,
Preliminary translation result language material, is designated as：{mt1,mt2,…,mt1597}；
Step G：Splicing and combining for different order is carried out to step F language material, can according to every a word translation original text
Preceding, its corresponding preliminary posterior order of translation result is spliced successively, and rear language material is translated before output source, can also be according to every
The translation original text of a word is rear, and its corresponding preceding order of preliminary translation result is spliced successively, is translated after output source
Preceding language material；
Wherein, rear language material is translated before source, is designated as：
{srcmt1,srcmt2,…,srcmt1597}；
Wherein, preceding language material is translated behind source, is designated as：
{mtsrc1,mtsrc2,…,mtsrc1597}；
Step H：Language material input before being translated behind rear language material or source is translated before the source that any one connecting method is generated in selection step G
The post-editing system of step E outputs, output is the translation handled by post-editing.
Embodiment 2
The present embodiment elaborates the effect of the system and method by taking specific sentence as an example.
In instantiation, translation quality is intuitively being embodied with informativeness and fluency, wherein, the raising of informativeness is thin
Change onto the raising for selecting word accuracy.
It is assumed that translation original text is " but, past challenge, in terms of not terminating in subsidy public building, private house is also filled with
Major test." one.
Preliminary machine translation system uses Moses statictic machine translation systems, and translation result is " however, the past
challenge,not in the funding of public housing, private housing is full of
Challenge. ", in this sentence, the keyword " subsidy " of translation original text has been translated into " funding ", looks like " to be ...
Provide with funds ", lack the implication of help aspect, it is not accurate enough, meanwhile, the clause " not terminating in " of translation original text is translated into
" not ... ", overall language fluency is not good enough.
After the present invention remixes the automatic post-editing system compensation of multi-source neutral net of mode based on splicing, translate
Text is " however, the challenges in the past were not limited to subsidizing
public housing,and private houses were also a major challenge.”。
Either on selecting in word accuracy in keyword " subsidy ", or overall sentence fluency, all closer to correct
Reference translation " nevertheless, past challenges are not limited to subsidized public
Housing.Private housing is also full of serious ordeals. ", quality is far above preliminary translation
As a result, reached that translation original text interacts with preliminary translation result, cross validation so that post-editing translation quality is higher.
Embodiment 3
The present embodiment elaborates the system and method relative to translation original text is not added in statistical significance, directly using just
The automatic post-editing system of single source neutral net and only done splicing and unmixed side that step translation result is trained as original language
Advantage of the automatic post-editing system of formula multi-source neutral net in overall translation quality.
It is assumed that the training original text for training module has 600000 with reference translation data set, for turning over for test module
Translating plaintext data collection has 1597, and preliminary machine translation system uses Moses statictic machine translation systems, and scoring uses multi-
Bleu scripts, BLEU values represent overall translation quality, and it is the quantizating index of informativeness and fluency respectively that unitary to quaternary, which is given a mark,
Specific score is described in table 1 below：
Table 1：Preliminary translation system, single source post-editing system, the multi-source post-editing system based on connecting method and base
Contrast of the multi-source post-editing system of mode to translation original text treatment effect in statistical significance is remixed in splicing
From table 1 it follows that in terms of overall translation quality (BLEU), after the multi-source for remixing mode based on splicing is translated
Editing system translates being translated before source first two joining method formation language material translation quality behind rear or source is all significantly larger than it
His system, and unitary marking and quaternary marking are all highests in all systems, and this explanation translation is in informativeness and fluency
Aspect is obtained for raising.
Described above is presently preferred embodiments of the present invention, and the present invention should not be limited to the embodiment and accompanying drawing institute is public
The content opened.It is every not depart from the lower equivalent or modification completed of spirit disclosed in this invention, both fall within the model that the present invention is protected
Enclose.
Claims (10)
1. a kind of automatic post-editing system and method for multi-source neutral net that mode is remixed based on splicing, it is characterised in that：
Including a kind of automatic post-editing system of multi-source neutral net that mode is remixed based on splicing, referred to as a kind of the system, and base
The automatic post-editing method of multi-source neutral net of mode, referred to as this method are remixed in splicing；
Wherein, multi-source refers to the input of post-editing and may come from a variety of different original language, including machine translation translation with turning over
Translate original text；
The system enables to translation original text to be influenced each other with preliminary translation result during post-editing, and cross validation is carried
Height translation informativeness, and then improve the total quality of post-editing result；
The system includes training module and decoder module；
The function of each module is as follows：
The function of training module is the automatic post-editing system of multi-source neutral net that training remixes mode based on splicing, output
Training pattern；This training pattern be also known as post-editing system model；
The function of decoder module is that the post-editing system model exported using training module is decoded；
Annexation between each module is as follows：
Training module is connected with decoder module, the training pattern exported especially by training module, i.e. post-editing system mould
Type is connected；
To achieve the above object, the technical solution adopted in the present invention is as follows：
Define 1：One preliminary machine translation system, referred to as Machine Translation, i.e. MT systems are set；
Define 2：Set a constant N, represent in this method for training module training original text and reference translation language material all
Assuming that there is N；
Define 3：One constant M is set, the translation original text in this method for decoder module is represented and assumes there are M.
2. a kind of automatic post-editing system of multi-source neutral net that mode is remixed based on splicing according to claim 1
And method, it is characterised in that：This method includes the training process of training module and decoding process two parts of decoder module, wherein
Training module completes the training of the automatic post-editing system of multi-source neutral net to remixing mode based on splicing, and output has been instructed
Practice model；The training pattern that decoding process is exported using training process is decoded；
The training process of training module, be specially：
Step 1: collecting each language material required for the system training process, and training original text language material therein is entered through MT systems
The preliminary translation of row, draws preliminary translation result language material；
Step 2: being spliced and combined to the language material of step one, rear language material is translated before output source；
Step 3: being spliced and combined to the language material of step one, preceding language material is translated after output source；
Integrally mixed again Step 4: being translated before the source exported to step 2 and step 3 and preceding language material being translated behind rear language material and source,
Mixing language material is drawn, the source language material of training process is used as；
Wherein, Step 2: splicing and mixed process in step 3 and step 4 are to construct the process that multi-source translates language material,
Refer to translate the source language material that original text together form post-editing system with preliminary translation result；
Step 5: it is integrally double to the training translation language material of step one, generate the target language material of training process；
Step 6: the target language material that the source language material obtained with step 4 is obtained with step 5 is trained based on neutral net translation model
The system, exports post-editing system model；
So far, from step one to step 6, the training process of training module in this method is completed；
Step 7: each language material needed in the decoding step of setting the system；
Step 8: being spliced and combined to the language material of step 7, rear language material is translated before the source that output decoding process needs；
Step 9: being spliced and combined to the language material of step 7, preceding language material is translated behind the source that output decoding process needs；
Step 10: both language materials are optionally first, input before rear language material is translated before the source that step 8 and step 9 are exported and is translated behind source
The post-editing system model exported to step 6, exports post-editing translation；
So far, from step 8 to step 10, the decoding process of decoder module in this method is completed.
3. a kind of automatic post-editing system of multi-source neutral net that mode is remixed based on splicing according to claim 2
And method, it is characterised in that：In step one, each language material mainly includes training original text language material and reference translation language material；Wherein, train
Original text language material and reference translation language material are bilingual parallel corporas；
Original text language material is trained, is designated as：{source1,source2,…,sourceN,
Translation language material is trained, { ref is designated as1,ref2,…,refN,
The preliminary translation result language material of original text language material is trained, is designated as：
{mt-outs1,mt-outs2,…,mt-outsN}。
4. a kind of automatic post-editing system of multi-source neutral net that mode is remixed based on splicing according to claim 2
And method, it is characterised in that：Rear language material is translated before source in step 2, is designated as：
{sourcemt-outs1,sourcemt-outs2,…,sourcemt-outsN, exist per a word according to training original text
Before, its corresponding preliminary posterior order of translation result is spliced successively.
5. a kind of automatic post-editing system of multi-source neutral net that mode is remixed based on splicing according to claim 2
And method, it is characterised in that：Language material before being translated behind source in step 3, is designated as：
{mt-outsssource1,mt-outssource2,…,mt-outssourceN, exist per a word according to training original text
Afterwards, and its corresponding preceding order of preliminary translation result is spliced successively.
6. a kind of automatic post-editing system of multi-source neutral net that mode is remixed based on splicing according to claim 2
And method, it is characterised in that：Mixing language material in step 4, is designated as：{sourcemt-outs1,sourcemt-outs2,…,
sourcemt-outsN,mt-outsssource1,mt-outssource2,…,mt-outssourceN, it is used as training module
Source language material.
7. a kind of automatic post-editing system of multi-source neutral net that mode is remixed based on splicing according to claim 2
And method, it is characterised in that：Overall double to training translation language material progress in step 5, its output is designated as：
{ref1,ref2,…,refN,ref1,ref2,…,refN, it is used as the target language material of training module.
8. a kind of automatic post-editing system of multi-source neutral net that mode is remixed based on splicing according to claim 2
And method, it is characterised in that：In step 6, each language material needed in decoding step mainly includes translation original text language material and tentatively turned over
Result language material is translated, preliminary translation result language material is obtained by translation original text language material through the translation of MT systems；
Original text language material is translated, is designated as：{src1,src2,…,srcM,
Preliminary translation result language material, is designated as：{mt1,mt2,…,mtM}。
9. a kind of automatic post-editing system of multi-source neutral net that mode is remixed based on splicing according to claim 2
And method, it is characterised in that：Rear language material is translated before source in step 7, is designated as：{srcmt1,srcmt2,…,srcmtM}；
Per in short according to translation original text, preceding, the corresponding preliminary posterior order of translation result is spliced successively.
10. a kind of automatic post-editing system of multi-source neutral net that mode is remixed based on splicing according to claim 2
System and method, it is characterised in that：Preceding language material is translated behind source in step 8, is designated as：{mtsrc1,mtsrc2,…,mtsrcM}；
Per in short according to translation original text, rear, the corresponding preliminary preceding order of translation result is spliced successively.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710477974 | 2017-06-22 | ||
| CN2017104779742 | 2017-06-22 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107301173A true CN107301173A (en) | 2017-10-27 |
| CN107301173B CN107301173B (en) | 2019-10-25 |
Family
ID=60134998
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710491848.2A Active CN107301173B (en) | 2017-06-22 | 2017-06-26 | A kind of automatic post-editing system and method for multi-source neural network remixing mode based on splicing |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107301173B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635269A (en) * | 2019-01-31 | 2019-04-16 | 苏州大学 | A kind of post-editing method and device of machine translation text |
| CN110765791A (en) * | 2019-11-01 | 2020-02-07 | 清华大学 | Automatic post-editing method and device for machine translation |
| US11537660B2 (en) | 2020-06-18 | 2022-12-27 | International Business Machines Corporation | Targeted partial re-enrichment of a corpus based on NLP model enhancements |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101957815A (en) * | 2009-07-13 | 2011-01-26 | 白劲实 | Automatic translation method and system based on correct translation result and corresponding relation |
| CN104199813A (en) * | 2014-09-24 | 2014-12-10 | 哈尔滨工业大学 | Pseudo-feedback-based personalized machine translation system and method |
-
2017
- 2017-06-26 CN CN201710491848.2A patent/CN107301173B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101957815A (en) * | 2009-07-13 | 2011-01-26 | 白劲实 | Automatic translation method and system based on correct translation result and corresponding relation |
| CN104199813A (en) * | 2014-09-24 | 2014-12-10 | 哈尔滨工业大学 | Pseudo-feedback-based personalized machine translation system and method |
Non-Patent Citations (1)
| Title |
|---|
| LUCAS NUNES VIERIA: "Indices of cognitive effort in machine translation post-editing", 《MACHINE TRANSLATION》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635269A (en) * | 2019-01-31 | 2019-04-16 | 苏州大学 | A kind of post-editing method and device of machine translation text |
| CN109635269B (en) * | 2019-01-31 | 2023-06-16 | 苏州大学 | Post-translation editing method and device for machine translation text |
| CN110765791A (en) * | 2019-11-01 | 2020-02-07 | 清华大学 | Automatic post-editing method and device for machine translation |
| US11537660B2 (en) | 2020-06-18 | 2022-12-27 | International Business Machines Corporation | Targeted partial re-enrichment of a corpus based on NLP model enhancements |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107301173B (en) | 2019-10-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102236645B (en) | Based on the natural language man-machine conversation device of semantic logic | |
| CN100543727C (en) | A kind of interpretation method that has merged sentence pattern template and statistical machine translation technology | |
| CN107301173B (en) | A kind of automatic post-editing system and method for multi-source neural network remixing mode based on splicing | |
| CN108363704A (en) | A kind of neural network machine translation corpus expansion method based on statistics phrase table | |
| MXPA04006075A (en) | Adaptive machine translation service. | |
| CN109726293A (en) | A kind of causal event map construction method, system, device and storage medium | |
| CN105723405A (en) | Guided article authorship | |
| CN105573994B (en) | Statictic machine translation system based on syntax skeleton | |
| CN110046261A (en) | A kind of construction method of the multi-modal bilingual teaching mode of architectural engineering | |
| Preidel et al. | BIM-based code compliance checking | |
| CN108491459B (en) | Optimization method for software code abstract automatic generation model | |
| Zuo | China’s policy towards minority languages in a globalising age | |
| CN108804407A (en) | A kind of newer method and apparatus of quick response e-bidding document template | |
| CN107301174A (en) | A kind of automatic post-editing system and method for integrated form based on splicing | |
| CN104391984B (en) | A kind of recommendation degree methods of marking of network user's comment of Chinese and English mixing | |
| van der Stijl et al. | Recommendations for a Dutch sustainable biobanking environment | |
| CN110399936A (en) | It is a kind of for training English Grammar to correct mistakes the text data generation method of model | |
| CN114564933A (en) | Personalized machine translation training method and system | |
| CN104572629B (en) | Monolingual analytical model generation method, monolingual analytical model generation device and language transformation device | |
| Yousef | Ugarit: translation alignment visualization | |
| Ustun | A critical review of IEC 61850 testing tools | |
| Zhou et al. | [Retracted] Innovation and Exploration of Computer‐Aided New Media Translation Course Teaching Mode under the Ecological Environment | |
| CN107122337A (en) | One kind translation official documents and correspondence generation method and device | |
| Svoboda | Computing and Translation: An Overview for Technical Communicators | |
| Charles et al. | Mapping cross-domain metadata to the Europeana data model (EDM) |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |