CN110276082A - Translation processing method and device based on dynamic window - Google Patents
Translation processing method and device based on dynamic window Download PDFInfo
- Publication number
- CN110276082A CN110276082A CN201910490402.7A CN201910490402A CN110276082A CN 110276082 A CN110276082 A CN 110276082A CN 201910490402 A CN201910490402 A CN 201910490402A CN 110276082 A CN110276082 A CN 110276082A
- Authority
- CN
- China
- Prior art keywords
- word
- window
- target
- translation
- source
- 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
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
- G06F3/048—Interaction techniques based on graphical user interfaces [GUI]
- G06F3/0484—Interaction techniques based on graphical user interfaces [GUI] for the control of specific functions or operations, e.g. selecting or manipulating an object, an image or a displayed text element, setting a parameter value or selecting a range
- G06F3/0485—Scrolling or panning
-
- 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)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Human Computer Interaction (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Machine Translation (AREA)
Abstract
The invention proposes a kind of translation processing method and device based on dynamic window, wherein method includes: to slide in the source word of input according to preset window sliding state modulator target window；Target word after translation and the target word in target window present scope are subjected to similarity calculation；Speech synthesis, which is carried out, according to similarity calculation result exports target translation.The translation delay for reducing simultaneous interpretation as a result, improves translation efficiency.
Description
Technical field
The present invention relates to voice processing technology field more particularly to a kind of translation processing methods and dress based on dynamic window
It sets.
Background technique
In general, first identifying to voice signal to be translated during carrying out simultaneous interpretation, sentence side is then carried out
Boundary's identification, identifies a sentence.Sentence is handled by punctuation mark marking model, forming one can completely be turned over
The sentence translated carries out the generation of target translation by MT engine.
However, above-mentioned translation process, voice output generates the translation of target side to the end since speaker, link compared with
It is more, it is delayed larger.Such as in English-Chinese simultaneous interpretation scene, generally requires that the pause of speaker is waited to make pauses in reading unpunctuated ancient writings, receiving
It just can recognize that a complete sentence after content even after more than ten seconds within several seconds, carry out translation generation.
Summary of the invention
For this purpose, the first purpose of this invention is to propose a kind of translation processing method based on dynamic window, reduce
The translation of simultaneous interpretation is delayed, and improves translation efficiency.
Second object of the present invention is to propose a kind of translation processing unit based on dynamic window.
Third object of the present invention is to propose a kind of computer equipment.
Fourth object of the present invention is to propose a kind of computer readable storage medium.
First aspect present invention embodiment proposes a kind of translation processing method based on dynamic window, including following step
It is rapid: to be slided in the source word of input according to preset window sliding state modulator target window；By the target word after translation
Similarity calculation is carried out with the target source terminal word language in the target window present scope；Language is carried out according to similarity calculation result
Sound synthesis output target translation.
In addition, the translation processing method based on dynamic window of the embodiment of the present invention, also has following additional technology special
Sign:
Optionally, it is slided in the source word of input described according to preset window sliding state modulator target window
Before, further includes: judge whether the length of source word currently entered meets the initial length of the target window；If knowing
The length of the source word currently entered meets the initial length of the target window, then states source currently entered to described
Terminal word language carries out translation and generates target word.
Optionally, further includes: calculate the alignment relation of each word according to alignment schemes, obtain the corresponding sample of sample object word
Origin terminal word sentence；According to the initial length of the sample object word corresponding sample source terminal word sentence training target window.
It is optionally, described to be slided in the source word of input according to preset window sliding state modulator target window,
It include: to obtain the target window current initial position and end position；According to preset function and preset threshold calculating
The state value of initial position and the end position；According to the control of the state value of the initial position and the end position
Target window slides in the source word of input.
Optionally, it is slided in the source word of input described according to preset window sliding state modulator target window
Before, further includes: obtained according to the current initial position of the target window and the current location of the source word of input presetting
Whole sequence；If according to sequencing function trained in advance determine the source word of the input current location and the target window
Corresponding semantic similarity meets preset condition, then the adjustment of word position is carried out to the pre-adjustment sequence.
Second aspect of the present invention embodiment proposes a kind of translation processing unit based on dynamic window, comprising: sliding die
Block, for being slided in the source word of input according to preset window sliding state modulator target window；Computing module is used for
Target word after translation and the target source terminal word language in the target window present scope are subjected to similarity calculation；Synthesize mould
Block exports target translation for carrying out speech synthesis according to similarity calculation result.
In addition, the translation processing unit based on dynamic window of the embodiment of the present invention, also has following additional technology special
Sign:
Optionally, further includes: judgment module, for judging whether the length of source word currently entered meets the mesh
Mark the initial length of window；Generation module, for knowing that the length of the source word currently entered meets the target
When the initial length of window, to it is described state source word currently entered carry out translation generate target word.
Optionally, the sliding block, comprising: acquiring unit, for obtaining the current initial position of the target window
And end position；Computing unit, for calculating the initial position and the end position according to preset function and preset threshold
State value；Control unit, for controlling the target window according to the state value of the initial position and the end position
It is slided in the source word of input.
Optionally, further includes: module is obtained, for the source according to the target window current initial position and input
The current location of word obtains pre-adjustment sequence；Module is adjusted, it is described defeated for being determined in basis sequencing function trained in advance
When the current location of the source word entered semantic similarity corresponding with the target window meets preset condition, to described
The adjustment of pre-adjustment sequence progress word position.
Third aspect present invention embodiment proposes a kind of computer equipment, including processor and memory；Wherein, described
Processor is corresponding with the executable program code to run by reading the executable program code stored in the memory
Program, for realizing the translation processing method based on dynamic window as described in first aspect embodiment.
Fourth aspect present invention embodiment proposes a kind of computer readable storage medium, is stored thereon with computer journey
Sequence realizes the translation processing method based on dynamic window as described in first aspect embodiment when the program is executed by processor.
Technical solution provided in an embodiment of the present invention at least has following additional technical characteristic:
Can be according to speaker's content, dynamic adjusts the window size of attention, generates translation in real time, reduces simultaneous interpretation
Time delay.
Detailed description of the invention
Above-mentioned and/or additional aspect and advantage of the invention will become from the following description of the accompanying drawings of embodiments
Obviously and it is readily appreciated that, in which:
Fig. 1 is the flow diagram of the simultaneous interpretation of the prior art；
Fig. 2 is the flow chart of the translation processing method according to an embodiment of the invention based on dynamic window；
Fig. 3 is that the attention of attention mechanism according to an embodiment of the invention calculates schematic diagram；
Fig. 4 is that the attention of attention mechanism in accordance with another embodiment of the present invention calculates schematic diagram；
Fig. 5 is the dynamic change schematic diagram of dynamic window according to an embodiment of the invention；
Fig. 6 is the structural schematic diagram of the translation processing unit according to an embodiment of the invention based on dynamic window；
Fig. 7 is the structural schematic diagram of the translation processing unit in accordance with another embodiment of the present invention based on dynamic window；
Fig. 8 is the structural schematic diagram of the translation processing unit based on dynamic window of another embodiment according to the present invention；
And
Fig. 9 is the structural schematic diagram of the translation processing unit based on dynamic window of a still further embodiment according to the present invention.
Specific embodiment
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to attached
The embodiment of figure description is exemplary, it is intended to is used to explain the present invention, and is not considered as limiting the invention.
For what is mentioned in above-mentioned background technique, during traditional simultaneous interpretation, as shown in Figure 1, needing first to identify
Voice is made pauses in reading unpunctuated ancient writings automatically based on the pause in the voice recognized, in turn, where identifying the punctuates such as fullstop after punctuate
Position, the position based on punctuate identification carry out the addition of punctuate, and the voice messaging after addition punctuate is carried out machine translation, into
And speech synthesis output is carried out according to the text after translation, in the method, it can all cause delay very big to a certain extent,
Such as the possible word speed of speaker is very fast, just pauses after finishing more than ten seconds.Or speech recognition errors cause punctuate module accuracy rate
Decline, just judges whether be a complete sentence, the delay of translation is larger after more than ten seconds.
In order to solve the above-mentioned technical problem, the present invention proposes that a kind of simultaneous interpretation using dynamic window attention mechanism fills
It sets, finishes a complete sentence without waiting for speaker, can dynamically slide attention window, generate target in real time and translate
Text.
Specifically, Fig. 2 is the process of the translation processing method according to an embodiment of the invention based on dynamic window
Figure, as shown in Fig. 2, this method comprises:
Step 101, it is slided in the source word of input according to preset window sliding state modulator target window.
Wherein, source word is the word to be translated received.
It is appreciated that in the embodiment of the present application, using neural network translation model end to end, being translated in neural network
In model, notice that power module plays very important effect for the promotion of translation quality.In practical applications, as shown in figure 3,
The identification of the distal end words and phrases of input and the similarity of the sentence after translation may be implemented in attention mechanism based on attention power module,
Determine that the word that needs important in target word are paid close attention to, reference Fig. 3 are (similar with gray value mark in figure based on similarity
Degree, gray value is higher, and similarity is higher), it may be exactly the word for more needing to pay close attention to the more similar word of target word after translation, need
Higher weight etc. is assigned, carrys out the adjustment that target word carries out the attention of different words as a result, so that being translated in the subsequent target that obtains
The quality of translation can be improved when literary.
Certainly, it is contemplated that in attention mechanism as shown in Figure 3, be still using obtained entire sentence as pay attention to
The basis of power processing, it is desirable that after model can obtain all words of source words and phrases (complete sentence), just can be carried out attention meter
It calculates.And obtain that complete sentence cost is higher, it is easy to produce the translation of high delay.Therefore, as shown in figure 4, introducing slidable
Target window only calculates the attention of the source words and phrases within the scope of target window every time.It is last in generation every time referring to Fig. 4
Before translation, it is thus only necessary to the attention force information of the source word within the scope of calculation window, without identifying the sentence of source sentence
Sub- boundary carries out attention calculating as a long sentence without using all identified contents.We only need to count
Calculate the attention force information within the scope of current window.
It should be noted that the target window of the embodiment of the present invention can slide, wherein the initial length of target window
Degree can be obtained according to experimental data training.
As a kind of possible implementation, low volume data collection can be manually marked, indicating needs when target word generates
Which words and phrases of source are investigated, when training, the alignment relation of each word is calculated based on alignment thereof, it is corresponding to obtain sample object word
Sample source terminal word sentence, according to the initial length of the corresponding sample source terminal word sentence training objective window of sample object word, the training
Some training methods in traditional technology can be used in process, determine that the relationship of the source words and phrases of target word and dependence is restrained
Training, determines initial length.
In this example, judge whether the length of source word currently entered meets the initial length of target window, if
Know that the length of source word currently entered meets the initial length of target window, then triggering starts the condition of translation, to stating
Source word currently entered carries out translation and generates target word.At this time due to the source words and phrases that initial length includes obviously be less than it is whole
The words and phrases quantity of a sentence improves translation efficiency therefore, it is possible to quickly trigger translation process.
In this example, after initial length determines, with the big of the variation dynamic adjustment window for translating obtained target word
It is small, to guarantee that few as far as possible source words and phrases and the target word translated are consistent as far as possible.
Specifically, control target window sliding mechanism are as follows: obtain current goal window current initial position and
End position, the state value of initial position and end position is calculated according to preset function and preset threshold, which is used for
Determine whether to slide the starting position of window and end position, be controlled according to the state value of initial position and end position
Target window slides in the source word of input.In this example, the direction of sliding is the sliding of window right direction, the state value
It can be the information of prespecified multiple format, for example can be 0 and 1, wherein 0 representative is not slided to the right, and 1 representative is slided to the right.
Based on this, a target window is as shown in figure 5, may only end position slide to the right, it is also possible to only start
It slides to the right position, it is also possible to which starting position and end position slide to the right, this is needed when generation dependent on target word
Investigate which words and phrases of source, the considering of this dependence can restrain training by RL method and obtain, based on RL method convergence instruction
White silk can determine the state value of initial position and end position according to the comparison of preset function and preset threshold, as a kind of possibility
Implementation, which is Sigmoid function or is Bernoulli function etc., both functions are based primarily upon out
Beginning position and end position between the source words and phrases covered whether can translate accurate target word and calculate state value.
In the method, dynamic window initial position can be such as defined as s, the value that end position is defined as e, s and e has
0 or 1 two kind as a result, 0 indicate do not slide to the right, 1 indicate to the right slide a position.The value of s and e can use Bernoulli
Profile samples obtain, or judge whether state value judges more than 0.5 using Sigmoid function.Wherein, the training of RL has two
Kind of mode obtains carrying out after decision back-propagation gradient after reward calculating according to sampling such as policy gradient.Or
Imitation learning designs a teacher agent, using teacher agent generate word alignment as a result,
Corresponding action sequence is generated, then using there is supervision to be trained, the above-mentioned method by the convergence training of RL method can be by existing
There is technology to obtain, will herein be described in detail.
Step 102, the target word after translation and the target source terminal word language in target window present scope are subjected to similarity
It calculates.
Wherein, the target word after translation is corresponding to the corresponding translation word of source sentence in current goal window, currently
Target source terminal word language in range is source sentence included in current goal window.
Specifically, the target word after translation and the target source terminal word language in target window present scope are subjected to similarity meter
It calculates, similarity is higher, it is clear that the target word after generating translation more relies on, and therefore, is existed based on the degree of dependence, that is, similarity degree
Increase when being subsequently generated translation its translation weight, for example, when target source terminal word language be Gonna make it right,
Target word is " wanting to make it correct ", since the translation result of standard of comparison is " it is wanted to carry out ", it is, therefore, apparent that source word
The similarity of " right " is not very high, it is clear that the higher source word of similarity is " make ", " it ", " Gonna " at this time, is made
For a kind of possible implementation, similarity can be determined based on contextual information and semantic similarity.
Step 103, speech synthesis is carried out according to similarity calculation result and exports target translation.
Specifically, after obtaining similarity, speech synthesis is carried out according to similarity calculation result and exports target translation, such as
The weight of the higher target word of similarity is improved and generates corresponding translation, and obtains speech synthesis corresponding with translation.
Certainly, because of the particularity of machine translation problem: the word string of translation often depends on the sequencing of long range, that is, works as
Previous existence at word may need to see a far source words and phrases, when especially for translator of English scene, possible hero's
Name appears in most starting for long sentence, it would therefore be desirable to carry out certain pre- sequencing to source sentence.
In one embodiment of the invention, working as according to the source word of the current initial position of target window and input
Front position obtains pre-adjustment sequence, for example, as shown in fig. 6, the initial position of current window is the 16th source word, and it is current defeated
The current location of the source word entered is the 18th source word, at this point, based on the target word result and source words and phrases currently translated
Between semantic corresponding relationship, and the 17-18 word newly obtaining determine that the 18th word is added to the progress of current goal window
Translation, specifically, can determine current location and the target window of the source word of input based on sequencing function trained in advance
Corresponding semantic similarity meets preset condition, i.e., the source words and phrases newly inputted have with the source words and phrases in target window
Stronger semantic relation, this pre-adjustment sequence carry out the adjustment of word position, for example, current goal window is added in the 18th word
In, select the little word of semantic contribution degree to be put into except window.
Wherein, as a kind of possible example, trained sequencing function can be the corresponding formula of following function in advance
(1), wherein in formula (1), htFor the semantic expressiveness of current t moment；Tanh is tangent non-linear transform function；τFor sequencing
The size of window；T is model size；σ is Sigmoid function, and value is between 0~1；I is integer variable, and value is in 0~2 τ
Between；E is the expression of embedded term vector, and w is the parameter for needing to learn.
To sum up, the translation processing method based on dynamic window of the embodiment of the present invention, can be according to speaker's content, dynamic
The window size for adjusting attention generates translation in real time, reduces simultaneous interpretation time delay.
In order to realize above-described embodiment, the present invention also proposes a kind of translation processing unit based on dynamic window.
Fig. 6 is the structural schematic diagram of the translation processing unit according to an embodiment of the invention based on dynamic window, such as
Shown in Fig. 6, should translation processing unit based on dynamic window include: sliding block 10, computing module 20, synthesis module 30,
In,
Sliding block 10, for sliding in the source word of input according to preset window sliding state modulator target window
It is dynamic.
Computing module 20 is carried out for the target word after translating and the target source terminal word language in target window present scope
Similarity calculation.
Synthesis module 30 exports target translation for carrying out speech synthesis according to similarity calculation result.
In one embodiment of the invention, as shown in fig. 7, on the basis of as shown in Figure 6, the device further include: sentence
Disconnected module 40, generation module 50, wherein
Judgment module 40, for judging whether the length of source word currently entered meets the initial length of target window
Degree.
Generation module 50, for knowing that the length of source word currently entered meets the initial length of target window
When, translation generation target word is carried out to source word currently entered is stated.
In one embodiment of the invention, as shown in figure 8, sliding block 10 includes: on the basis of as shown in Figure 6
Acquiring unit 11, computing unit 12, control unit 13, wherein
Acquiring unit 11, for obtaining the current initial position of target window and end position.
Computing unit 12, for calculating the state value of initial position and end position according to preset function and preset threshold.
Control unit 13, for controlling target window in the source of input according to the state value of initial position and end position
It is slided in word.
In one embodiment of the invention, as shown in figure 9, on the basis of as shown in Figure 6, the device further include: obtain
Modulus block 60 and adjustment module 70, wherein
Module 60 is obtained, for obtaining according to the current location of the current initial position of target window and the source word of input
Take pre-adjustment sequence.
Adjust module 70, for the current location that the source word of input is determined according to sequencing function trained in advance with
When the corresponding semantic similarity of target window meets preset condition, the adjustment of word position is carried out to pre-adjustment sequence.
It should be noted that previous embodiment is equally applicable to the explanation of the translation processing method based on dynamic window
In the translation processing unit based on dynamic window of the present embodiment, details are not described herein again.
To sum up, the translation processing unit based on dynamic window of the embodiment of the present invention, can be according to speaker's content, dynamic
The window size for adjusting attention generates translation in real time, reduces simultaneous interpretation time delay.
In order to realize above-described embodiment, the present invention also proposes a kind of computer equipment, including processor and memory；Its
In, processor runs journey corresponding with executable program code by reading the executable program code stored in memory
Sequence, for realizing the translation processing method based on dynamic window as described in aforementioned any embodiment.
In order to realize above-described embodiment, the present invention also proposes a kind of computer readable storage medium, is stored thereon with calculating
Machine program realizes the translation processing side based on dynamic window as described in aforementioned any embodiment when the program is executed by processor
Method.
In the description of the present invention, it is to be understood that, term " center ", " longitudinal direction ", " transverse direction ", " length ", " width ",
" thickness ", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom" "inner", "outside", " up time
The orientation or positional relationship of the instructions such as needle ", " counterclockwise ", " axial direction ", " radial direction ", " circumferential direction " be orientation based on the figure or
Positional relationship is merely for convenience of description of the present invention and simplification of the description, rather than the device or element of indication or suggestion meaning must
There must be specific orientation, be constructed and operated in a specific orientation, therefore be not considered as limiting the invention.
In addition, term " first ", " second " are used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance
Or implicitly indicate the quantity of indicated technical characteristic.Define " first " as a result, the feature of " second " can be expressed or
Implicitly include at least one this feature.In the description of the present invention, the meaning of " plurality " is at least two, such as two, three
It is a etc., unless otherwise specifically defined.
In the present invention unless specifically defined or limited otherwise, term " installation ", " connected ", " connection ", " fixation " etc.
Term shall be understood in a broad sense, for example, it may be being fixedly connected, may be a detachable connection, or integral；It can be mechanical connect
It connects, is also possible to be electrically connected；It can be directly connected, can also can be in two elements indirectly connected through an intermediary
The interaction relationship of the connection in portion or two elements, unless otherwise restricted clearly.For those of ordinary skill in the art
For, the specific meanings of the above terms in the present invention can be understood according to specific conditions.
In the present invention unless specifically defined or limited otherwise, fisrt feature in the second feature " on " or " down " can be with
It is that the first and second features directly contact or the first and second features pass through intermediary mediate contact.Moreover, fisrt feature exists
Second feature " on ", " top " and " above " but fisrt feature be directly above or diagonally above the second feature, or be merely representative of
First feature horizontal height is higher than second feature.Fisrt feature can be under the second feature " below ", " below " and " below "
One feature is directly under or diagonally below the second feature, or is merely representative of first feature horizontal height less than second feature.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.In the present specification, schematic expression of the above terms are not
It must be directed to identical embodiment or example.Moreover, particular features, structures, materials, or characteristics described can be in office
It can be combined in any suitable manner in one or more embodiment or examples.In addition, without conflicting with each other, the skill of this field
Art personnel can tie the feature of different embodiments or examples described in this specification and different embodiments or examples
It closes and combines.
Although the embodiments of the present invention has been shown and described above, it is to be understood that above-described embodiment is example
Property, it is not considered as limiting the invention, those skilled in the art within the scope of the invention can be to above-mentioned
Embodiment is changed, modifies, replacement and variant.
Claims (11)
1. a kind of translation processing method based on dynamic window, which comprises the following steps:
It is slided in the source word of input according to preset window sliding state modulator target window；
Target word after translation and the target source terminal word language in the target window present scope are subjected to similarity calculation；
Speech synthesis, which is carried out, according to similarity calculation result exports target translation.
2. the method as described in claim 1, which is characterized in that described according to preset window sliding state modulator target window
Before mouth slides in the source word of input, further includes:
Judge whether the length of source word currently entered meets the initial length of the target window；
If knowing, the length of the source word currently entered meets the initial length of the target window, works as to described state
The source word of preceding input carries out translation and generates target word.
3. method according to claim 2, which is characterized in that further include:
The alignment relation of each word is calculated according to alignment schemes, obtains the corresponding sample source terminal word sentence of sample object word；
According to the initial length of the sample object word corresponding sample source terminal word sentence training target window.
4. the method as described in claim 1, which is characterized in that described according to preset window sliding state modulator target window
It is slided in the source word of input, comprising:
Obtain the current initial position of the target window and end position；
The state value of the initial position and the end position is calculated according to preset function and preset threshold；
The target window is controlled in the source word of input according to the state value of the initial position and the end position
Sliding.
5. the method as described in claim 1, which is characterized in that described according to preset window sliding state modulator target window
Before mouth slides in the source word of input, further includes:
Pre-adjustment sequence is obtained according to the current initial position of the target window and the current location of the source word of input；
If according to sequencing function trained in advance determine the source word of the input current location and the target window pair
The semantic similarity answered meets preset condition, then the adjustment of word position is carried out to the pre-adjustment sequence.
6. a kind of translation processing unit based on dynamic window characterized by comprising
Sliding block, for being slided in the source word of input according to preset window sliding state modulator target window；
Computing module carries out phase with the target source terminal word language in the target window present scope for the target word after translating
It is calculated like degree；
Synthesis module exports target translation for carrying out speech synthesis according to similarity calculation result.
7. device as claimed in claim 6, which is characterized in that further include:
Judgment module, for judging whether the length of source word currently entered meets the initial length of the target window；
Generation module, for knowing that the length of the source word currently entered meets the initial length of the target window
When, to it is described state source word currently entered carry out translation generate target word.
8. device as claimed in claim 6, which is characterized in that the sliding block, comprising:
Acquiring unit, for obtaining the current initial position of the target window and end position；
Computing unit, for calculating the state of the initial position and the end position according to preset function and preset threshold
Value；
Control unit is being inputted for controlling the target window according to the state value of the initial position and the end position
Source word in slide.
9. device as claimed in claim 6, which is characterized in that further include:
Module is obtained, for according to the current initial position of the target window and the acquisition of the current location of the source word of input
Pre-adjustment sequence；
Module is adjusted, in current location and the institute for determining the source word of the input according to sequencing function trained in advance
When stating the corresponding semantic similarity of target window and meeting preset condition, the tune of word position is carried out to the pre-adjustment sequence
It is whole.
10. a kind of computer equipment, which is characterized in that including processor and memory；
Wherein, the processor is run by reading the executable program code stored in the memory can be performed with described
The corresponding program of program code, at for realizing the translation according to any one of claims 1 to 5 based on dynamic window
Reason method.
11. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
The translation processing method according to any one of claims 1 to 5 based on dynamic window is realized when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910490402.7A CN110276082B (en) | 2019-06-06 | 2019-06-06 | Translation processing method and device based on dynamic window |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910490402.7A CN110276082B (en) | 2019-06-06 | 2019-06-06 | Translation processing method and device based on dynamic window |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110276082A true CN110276082A (en) | 2019-09-24 |
| CN110276082B CN110276082B (en) | 2023-06-30 |
Family
ID=67962041
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910490402.7A Active CN110276082B (en) | 2019-06-06 | 2019-06-06 | Translation processing method and device based on dynamic window |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110276082B (en) |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100220364A1 (en) * | 2007-06-01 | 2010-09-02 | Advanced Track And Trace | Method and device for securing documents |
| CN103257798A (en) * | 2012-02-17 | 2013-08-21 | 阿里巴巴集团控股有限公司 | Window sliding method and window sliding device |
| CN104243970A (en) * | 2013-11-14 | 2014-12-24 | 同济大学 | 3D drawn image objective quality evaluation method based on stereoscopic vision attention mechanism and structural similarity |
| CN107704453A (en) * | 2017-10-23 | 2018-02-16 | 深圳市前海众兴电子商务有限公司 | A kind of word semantic analysis, word semantic analysis terminal and storage medium |
| CN108132931A (en) * | 2018-01-12 | 2018-06-08 | 北京神州泰岳软件股份有限公司 | A kind of matched method and device of text semantic |
| CN108304390A (en) * | 2017-12-15 | 2018-07-20 | 腾讯科技（深圳）有限公司 | Training method, interpretation method, device based on translation model and storage medium |
| CN108647214A (en) * | 2018-03-29 | 2018-10-12 | 中国科学院自动化研究所 | Coding/decoding method based on deep-neural-network translation model |
| CN108664632A (en) * | 2018-05-15 | 2018-10-16 | 华南理工大学 | A kind of text emotion sorting algorithm based on convolutional neural networks and attention mechanism |
| CN109034378A (en) * | 2018-09-04 | 2018-12-18 | 腾讯科技（深圳）有限公司 | Network representation generation method, device, storage medium and the equipment of neural network |
| CN109086869A (en) * | 2018-07-16 | 2018-12-25 | 北京理工大学 | A kind of human action prediction technique based on attention mechanism |
| CN109145190A (en) * | 2018-08-27 | 2019-01-04 | 安徽大学 | A kind of local quotation recommended method and system based on neural machine translation mothod |
| CN109145315A (en) * | 2018-09-05 | 2019-01-04 | 腾讯科技（深圳）有限公司 | Text interpretation method, device, storage medium and computer equipment |
| CN109344413A (en) * | 2018-10-16 | 2019-02-15 | 北京百度网讯科技有限公司 | Translation processing method and device |
-
2019
- 2019-06-06 CN CN201910490402.7A patent/CN110276082B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100220364A1 (en) * | 2007-06-01 | 2010-09-02 | Advanced Track And Trace | Method and device for securing documents |
| CN103257798A (en) * | 2012-02-17 | 2013-08-21 | 阿里巴巴集团控股有限公司 | Window sliding method and window sliding device |
| CN104243970A (en) * | 2013-11-14 | 2014-12-24 | 同济大学 | 3D drawn image objective quality evaluation method based on stereoscopic vision attention mechanism and structural similarity |
| CN107704453A (en) * | 2017-10-23 | 2018-02-16 | 深圳市前海众兴电子商务有限公司 | A kind of word semantic analysis, word semantic analysis terminal and storage medium |
| CN108304390A (en) * | 2017-12-15 | 2018-07-20 | 腾讯科技（深圳）有限公司 | Training method, interpretation method, device based on translation model and storage medium |
| CN108132931A (en) * | 2018-01-12 | 2018-06-08 | 北京神州泰岳软件股份有限公司 | A kind of matched method and device of text semantic |
| CN108647214A (en) * | 2018-03-29 | 2018-10-12 | 中国科学院自动化研究所 | Coding/decoding method based on deep-neural-network translation model |
| CN108664632A (en) * | 2018-05-15 | 2018-10-16 | 华南理工大学 | A kind of text emotion sorting algorithm based on convolutional neural networks and attention mechanism |
| CN109086869A (en) * | 2018-07-16 | 2018-12-25 | 北京理工大学 | A kind of human action prediction technique based on attention mechanism |
| CN109145190A (en) * | 2018-08-27 | 2019-01-04 | 安徽大学 | A kind of local quotation recommended method and system based on neural machine translation mothod |
| CN109034378A (en) * | 2018-09-04 | 2018-12-18 | 腾讯科技（深圳）有限公司 | Network representation generation method, device, storage medium and the equipment of neural network |
| CN109145315A (en) * | 2018-09-05 | 2019-01-04 | 腾讯科技（深圳）有限公司 | Text interpretation method, device, storage medium and computer equipment |
| CN109344413A (en) * | 2018-10-16 | 2019-02-15 | 北京百度网讯科技有限公司 | Translation processing method and device |
Non-Patent Citations (12)
| Title |
|---|
| ABHISHEK SHARMA: "Attention-based Neural Machine Translation", 《TOWARDS DATA SCIENCE》 * |
| ABHISHEK SHARMA: "Attention-based Neural Machine Translation", 《TOWARDS DATA SCIENCE》, 10 March 2019 (2019-03-10), pages 1 * |
| 张传强: "基于显著性的目标检测与识别算法", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 张传强: "基于显著性的目标检测与识别算法", 《中国优秀硕士学位论文全文数据库信息科技辑》, no. 6, 15 June 2017 (2017-06-15), pages 138 - 1082 * |
| 张金超等: "基于多编码器多解码器的大规模维汉神经网络机器翻译模型", 《中文信息学报》 * |
| 张金超等: "基于多编码器多解码器的大规模维汉神经网络机器翻译模型", 《中文信息学报》, vol. 32, no. 9, 15 September 2018 (2018-09-15), pages 20 * |
| 梅桐: "利用深度学习进行人机对话", 《通讯世界》 * |
| 梅桐: "利用深度学习进行人机对话", 《通讯世界》, 25 August 2018 (2018-08-25), pages 220 * |
| 王炜: "基于深度学习的中文单文档自动文摘方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 王炜: "基于深度学习的中文单文档自动文摘方法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》, no. 1, 15 January 2019 (2019-01-15), pages 138 - 5486 * |
| 王超超等: "基于双语合成语义的翻译相似度模型", 《北京大学学报(自然科学版)》 * |
| 王超超等: "基于双语合成语义的翻译相似度模型", 《北京大学学报(自然科学版)》, vol. 51, no. 02, 1 December 2014 (2014-12-01), pages 335 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110276082B (en) | 2023-06-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| KR102401942B1 (en) | Method and apparatus for evaluating translation quality | |
| US10991366B2 (en) | Method of processing dialogue query priority based on dialog act information dependent on number of empty slots of the query | |
| US11081102B2 (en) | Systems and methods for comprehensive Chinese speech scoring and diagnosis | |
| JP5901001B1 (en) | Method and device for acoustic language model training | |
| CN110838289A (en) | Awakening word detection method, device, equipment and medium based on artificial intelligence | |
| CN105404621B (en) | A kind of method and system that Chinese character is read for blind person | |
| US11081109B2 (en) | Speech processing method using artificial intelligence device | |
| US20200035216A1 (en) | Speech synthesis method based on emotion information and apparatus therefor | |
| WO2020098269A1 (en) | Speech synthesis method and speech synthesis device | |
| US11189282B2 (en) | Intelligent voice recognizing method, apparatus, and intelligent computing device | |
| CN111833845A (en) | Multi-language speech recognition model training method, device, equipment and storage medium | |
| WO2023108994A1 (en) | Sentence generation method, electronic device and storage medium | |
| CN110010136B (en) | Training and text analysis method, device, medium and equipment for prosody prediction model | |
| CN110852040B (en) | Punctuation prediction model training method and text punctuation determination method | |
| CN112133287B (en) | Speech recognition model training method, speech recognition method and related device | |
| CN111859964A (en) | Method and device for identifying named entities in sentences | |
| CN112016271A (en) | Language style conversion model training method, text processing method and device | |
| CN111161726B (en) | Intelligent voice interaction method, device, medium and system | |
| US10614170B2 (en) | Method of translating speech signal and electronic device employing the same | |
| CN112560475B (en) | Triplet extraction method and system | |
| CN108446278A (en) | A kind of semantic understanding system and method based on natural language | |
| CN115827854B (en) | Speech abstract generation model training method, speech abstract generation method and device | |
| CN110544470A (en) | voice recognition method and device, readable storage medium and electronic equipment | |
| US20230205994A1 (en) | Performing machine learning tasks using instruction-tuned neural networks | |
| CN113362815A (en) | Voice interaction method, system, electronic equipment and storage medium |
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 |