ABSTRACT
Abstract
Techniques are disclosed for training and/or utilizing an automatic post-editing model in correcting translation error(s) introduced by a neural machine translation model. The automatic post-editing model can be trained using automatically generated training instances. A training instance is automatically generated by processing text in a first language using a neural machine translation model to generate text in a second language. The text in the second language is processed using a neural machine translation model to generate training text in the first language. A training instance can include the text in the first language as well as the training text in the first language.
Description
BACKGROUND
Machine translation can use one or more computing devices to translate text, speech, etc. from a first language to a second language. For example, a machine translation process can substitute words in a first language with corresponding words in a second language. However, a simple substitution of words can generate inaccurate translations because a phrase can have a specific meaning that is different than a literal translation of each word within a phrase. For example, the English phrase âleap yearâ has a specific meaning (i.e., a year that includes February 28 th ) but a word for word substitution can translate âleap yearâ into the second language as words that indicate a year that leaps.
Neural machine translation (âNMTâ) models can use artificial neural networks to predict the likelihood of a sequence of words. In many implementations, NMT models can translate an entire phrase and/or sentence at a time. Translating an entire sentence and/or phrase, rather than individual tokens (i.e., individual characters, individual words, etc.) can leverage the overall context of the textual and/or spoken input. NMT models can utilize a variety of neural network learning techniques to perform translation including feature learning techniques, deep learning techniques, etc.
SUMMARY
Implementations described herein are directed towards generating edited text, that corrects translation error(s) introduced by a NMT model, by processing text containing translation error(s) using an automatic post-editing (âAPEâ) model. In many implementations, translation error(s) can include one or more words incorrectly translated using the NMT model from a source language into a target language. For example, the word club meaning a cudgel in the source language can incorrectly be translated into the word club meaning an organization in the target language by a NMT model.
An APE model can be trained, for use in correcting translation error(s), by using automatically generated training instances. For example, training instances can be automatically generated in an unsupervised manner. As described herein, automatic generation of training instances can conserve network and/or client device resources. In many implementations, a training instance can include training text and ground truth text. The training text can be automatically generated by processing the ground truth text using a NMT model to translate the ground truth text from a first language into a second language. The text generated in the second language is translated back into the first language using a NMT model to generate the training text in the first language. The training text and ground truth text can be stored as a training instance. For example, German ground truth text can be translated into Spanish text. The Spanish text can be translated back into German to generate the training text using an NMT model.
In many implementations, an APE model can be trained to correct translation errors in text translated into a target language from a source language. For example, a first APE model can be trained to correct translation errors in German text translated from English text; a second APE model can be trained to correct translation errors in German text translated from Spanish text; a third APE model can be trained to correct translation errors in French text translated from Spanish text; etc. Additionally or alternatively, a multilingual APE model can be trained to correct translation errors in a target language translated from many source languages. For example, the multilingual APE model can be trained to correct Italian text translated from English text, to correct Italian text translated from French text, to correct Italian text translated from Spanish text, and to correct Italian text translated from additional and/or alternative source language(s). Such a multilingual APE model can occupy less electronic storage space than multiple monolingual APE models. This savings in storage space can be beneficial, for example, when the multilingual APE model is utilized on a client device with limited storage capacity. Further, and as described herein, training of a multilingual model for N languages can occur with less training iterations than training of N separate monolingual models for each language.
Additionally or alternatively, translation error(s) can include grammatical gender translation error(s). Nouns in some (but not all) languages include an inherent gender. For example, French nouns can either be masculine or feminine, German nouns can be either masculine, feminine, or neuter, etc. Some nouns can have different gender depending on the context. For example, the English word âaideâ can be translated into French as âun aideâ meaning a male assistant as well as âune aideâ meaning a female assistant. Furthermore, some nouns can have different meanings depending on the gender. As another example, the French word rose has a masculine form and a feminine form. The masculine form âle roseâ can be translated into English as pink meaning the color while the feminine form âla roseâ can be translated into English as a rose meaning the flower. In contrast, nouns in some languages, such as English, lack a gender. Gender translation error(s) can occur when translating from a source language with gendered nouns into a target language without gendered nouns, from a source language without gendered nouns into a target language with gendered nouns, and/or from a source language with gendered nouns into a target language with gendered nouns.
An APE model in accordance with many implementations can be used to correct the gender of translated word(s). In many implementations, training instances to train an APE model to correct gender errors can additionally include preceding text to add additional context to the translated text. For example, English phrase B can be immediately preceded by English phrase A. English phrase B can be translated into German, and the German translation can be translated back into English training text using NMT model(s). The English phrase B (i.e., the ground truth text), the English training text, and the preceding English phrase A can be saved as a training instance. The APE model can be trained by processing the English training text and the preceding English phrase A to generate edited text. In many implementations, the preceding English phrase A can provide additional and/or alternative context for editing gender error(s).
APE models can additionally or alternatively be used to edit text generated using a template engine. In many implementations, a template engine can determine a text template and a set of structured data responsive to user interface input. For example, a user can provide input of âhow many lights are on in the kitchenâ. Template engine can determine the text template of âThere are [#] of lights on in the [room]â as well as the set of structured data of â#=3; room=kitchenâ. Furthermore, template engine can generate natural language output of âThere are 3 lights on in the kitchenâ based on the determined text template and the set of structured data. However, a template engine can introduce errors(s) into the natural language output. For example, when only one light is turned on in the kitchen, the template engine can introduce error(s) into the natural language output of âThere are 1 lights on in the kitchenâ based on the text template of âThere are [#] of lights on in the [room]â and the set of structured data â#=1; room=kitchenâ. An APE model can process natural language output of âThere are 1 lights on in the kitchenâ to generate an edited output of âThere is 1 light on in the kitchenâ. Additionally or alternatively, the APE model can be trained to generate a variety of edited natural language outputs. For example, the text of âIt is 75 degrees and sunny in Mountain Viewâ can be processed using the APE model to generate varying outputs of: âIn Mountain View it is 75 degrees and sunnyâ; âIt is sunny in Mountain View. The temperature is 75 degreesâ; âIn Mountain View it is currently 75 degrees and sunnyâ; etc.
A training instance for training an APE model to edit natural language generated using the template engine can be automatically generated by processing a ground truth instance of natural language text using random noise to generate training text. In many implementations, random noise can include adding one or more words to the text, removing one or more words from the text, reordering one or more words in the text, and/or additional automatic alternation(s) of the text. The ground truth text and the training text can be stored as a training instance. For example, the text of âThe thermostat is currently set to 72 degreesâ can be processed using random noise to generate training text of âThe thermostat is lights currently set to 72 degreesâ (i.e. adding the word lights); âThe thermostat is currently to 72 degreesâ (i.e., removing the word set); âThe thermostat is currently set 72 to degreesâ (i.e., reordering the words to and 72); etc. Additional and/or alternative noise may added to ground truth text to generate training text.
Accordingly, various implementations set forth techniques for training automatic post-editing models using automatically generated training instances. Generating training instances in a fully supervised manner can be resource and/or labor intensive. For example, in generating a single training instance, relevant data must be transmitted to a client device of a human reviewer, resources of the client device utilized to enable the reviewer to provide an annotation, and the annotation transmitted back for training instance generation. Implementations disclosed herein utilize automatically generated training instances, enabling systems to quickly generate training data corresponding to a wide number of languagesâwithout transmission of data to client devices of reviewers and without utilization of resources of those client devices in obtaining reviewer provided annotations. Additionally or alternatively, a large number of training instances can be automatically generated from a variety of diverse resources, which can improve the accuracy and/or robustness of edited translated text generated using an APE model trained on such training instances. In many implementations, a multilingual APE model can be trained using automatically generated training data. Multilingual models can frequently be trained with fewer training instances from each language. Computational resources (e.g., processor cycles, memory, battery power, and/or additional resources of a computing system) can be conserved by this reduced number of training instances required to train the multilingual APE model for many languages.
As described herein, edited translated text generated using the APE model can be utilized to cause performance of one or more computer actions. For example, the edited translated text can be visually and/or audibly rendered. Since the edited translated text is generated to remove translational and/or other errors, it is more quickly ascertainable by a user to which it is rendered. This can prevent the user from needing to request the text again, wasting computational resources, and/or can enable the user to more quickly provide further user interface input that is responsive to the edited translated text, reducing computer/user interaction duration and conserving resources. As another example, the edited translated text can be utilized in generating and/or transmitting command(s) (e.g., a command to control a smart device in accordance with the edited translated text). Since the edited translated text is generated to remove translational and/or other errors, it can prevent the wasteful transmission of incorrect commands and/or the wasteful implementation of incorrect commands.
The above description is provided only as an overview of some implementations disclosed herein. These and other implementations of the technology are disclosed in additional detail below.
It should be appreciated that all combinations of the foregoing concepts and additional concepts described in greater detail herein are contemplated as being part of the subject matter disclosed herein. For example, all combinations of claimed subject matter appearing at the end of this disclosure are contemplated as being part of the subject matter disclosed herein.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 illustrates an example of generating edited text using an automatic post-editing model in accordance with various implementations disclosed herein.
FIG. 2 illustrates an example environment in which various implementations disclosed herein can be implemented.
FIG. 3 illustrates automatically generating a training instance in accordance with implementations disclosed herein.
FIG. 4 is a flowchart illustrating an example process in accordance with implementations disclosed herein.
FIG. 5 is a flowchart illustrating another example process in accordance with implementations disclosed herein.
FIG. 6 is a flowchart illustrating another example process in accordance with implementations disclosed herein.
FIG. 7 illustrates another example environment in which implementations disclosed herein can be implemented.
FIG. 8 is a flowchart illustrating another example process in accordance with implementations disclosed herein.
FIG. 9 is a flowchart illustrating another example process in accordance with implementations disclosed herein.
FIG. 10 illustrates an example architecture of a computing device.
DETAILED DESCRIPTION
FIG. 1 illustrates an example of generating edited text using an APE model in accordance with various implementations. In the illustrated example 100 , text 102 is processed using NMT model 104 to generate translated text 106 . In many implementations, text 102 is in a source language and translated text 106 is in a target language. For example, source language English text can be translated to target language French text, source language German text can be translated to target language Chinese text, source language French text can be translated to target language Spanish text, and/or additional source language(s) text can be translated to additional target language(s) text. NMT model 104 can be trained for translating a specific source language into a specific target language (e.g., a distinct model translates French to English, another distinct model translates English to French, another distinct model translates French to German, etc.). Additionally or alternatively, NMT model 104 can be a multilingual model trained for translating many source languages into many target languages. For example, multilingual NMT model can be trained to translate English to French, French to English, German to English, English to German, etc.
NMT model 104 can introduce translation errors into translated text 106 . For example, the word club meaning a cudgel can incorrectly be translated as the word club meaning an organization. APE model 108 can process translated text 106 to generate edited <figure-callout id="110" label="text" filename
BACKGROUND
Machine translation can use one or more computing devices to translate text, speech, etc. from a first language to a second language. For example, a machine translation process can substitute words in a first language with corresponding words in a second language. However, a simple substitution of words can generate inaccurate translations because a phrase can have a specific meaning that is different than a literal translation of each word within a phrase. For example, the English phrase âleap yearâ has a specific meaning (i.e., a year that includes February 28 th ) but a word for word substitution can translate âleap yearâ into the second language as words that indicate a year that leaps.
Neural machine translation (âNMTâ) models can use artificial neural networks to predict the likelihood of a sequence of words. In many implementations, NMT models can translate an entire phrase and/or sentence at a time. Translating an entire sentence and/or phrase, rather than individual tokens (i.e., individual characters, individual words, etc.) can leverage the overall context of the textual and/or spoken input. NMT models can utilize a variety of neural network learning techniques to perform translation including feature learning techniques, deep learning techniques, etc.
SUMMARY
Implementations described herein are directed towards generating edited text, that corrects translation error(s) introduced by a NMT model, by processing text containing translation error(s) using an automatic post-editing (âAPEâ) model. In many implementations, translation error(s) can include one or more words incorrectly translated using the NMT model from a source language into a target language. For example, the word club meaning a cudgel in the source language can incorrectly be translated into the word club meaning an organization in the target language by a NMT model.
An APE model can be trained, for use in correcting translation error(s), by using automatically generated training instances. For example, training instances can be automatically generated in an unsupervised manner. As described herein, automatic generation of training instances can conserve network and/or client device resources. In many implementations, a training instance can include training text and ground truth text. The training text can be automatically generated by processing the ground truth text using a NMT model to translate the ground truth text from a first language into a second language. The text generated in the second language is translated back into the first language using a NMT model to generate the training text in the first language. The training text and ground truth text can be stored as a training instance. For example, German ground truth text can be translated into Spanish text. The Spanish text can be translated back into German to generate the training text using an NMT model.
In many implementations, an APE model can be trained to correct translation errors in text translated into a target language from a source language. For example, a first APE model can be trained to correct translation errors in German text translated from English text; a second APE model can be trained to correct translation errors in German text translated from Spanish text; a third APE model can be trained to correct translation errors in French text translated from Spanish text; etc. Additionally or alternatively, a multilingual APE model can be trained to correct translation errors in a target language translated from many source languages. For example, the multilingual APE model can be trained to correct Italian text translated from English text, to correct Italian text translated from French text, to correct Italian text translated from Spanish text, and to correct Italian text translated from additional and/or alternative source language(s). Such a multilingual APE model can occupy less electronic storage space than multiple monolingual APE models. This savings in storage space can be beneficial, for example, when the multilingual APE model is utilized on a client device with limited storage capacity. Further, and as described herein, training of a multilingual model for N languages can occur with less training iterations than training of N separate monolingual models for each language.
Additionally or alternatively, translation error(s) can include grammatical gender translation error(s). Nouns in some (but not all) languages include an inherent gender. For example, French nouns can either be masculine or feminine, German nouns can be either masculine, feminine, or neuter, etc. Some nouns can have different gender depending on the context. For example, the English word âaideâ can be translated into French as âun aideâ meaning a male assistant as well as âune aideâ meaning a female assistant. Furthermore, some nouns can have different meanings depending on the gender. As another example, the French word rose has a masculine form and a feminine form. The masculine form âle roseâ can be translated into English as pink meaning the color while the feminine form âla roseâ can be translated into English as a rose meaning the flower. In contrast, nouns in some languages, such as English, lack a gender. Gender translation error(s) can occur when translating from a source language with gendered nouns into a target language without gendered nouns, from a source language without gendered nouns into a target language with gendered nouns, and/or from a source language with gendered nouns into a target language with gendered nouns.
An APE model in accordance with many implementations can be used to correct the gender of translated word(s). In many implementations, training instances to train an APE model to correct gender errors can additionally include preceding text to add additional context to the translated text. For example, English phrase B can be immediately preceded by English phrase A. English phrase B can be translated into German, and the German translation can be translated back into English training text using NMT model(s). The English phrase B (i.e., the ground truth text), the English training text, and the preceding English phrase A can be saved as a training instance. The APE model can be trained by processing the English training text and the preceding English phrase A to generate edited text. In many implementations, the preceding English phrase A can provide additional and/or alternative context for editing gender error(s).
APE models can additionally or alternatively be used to edit text generated using a template engine. In many implementations, a template engine can determine a text template and a set of structured data responsive to user interface input. For example, a user can provide input of âhow many lights are on in the kitchenâ. Template engine can determine the text template of âThere are [#] of lights on in the [room]â as well as the set of structured data of â#=3; room=kitchenâ. Furthermore, template engine can generate natural language output of âThere are 3 lights on in the kitchenâ based on the determined text template and the set of structured data. However, a template engine can introduce errors(s) into the natural language output. For example, when only one light is turned on in the kitchen, the template engine can introduce error(s) into the natural language output of âThere are 1 lights on in the kitchenâ based on the text template of âThere are [#] of lights on in the [room]â and the set of structured data â#=1; room=kitchenâ. An APE model can process natural language output of âThere are 1 lights on in the kitchenâ to generate an edited output of âThere is 1 light on in the kitchenâ. Additionally or alternatively, the APE model can be trained to generate a variety of edited natural language outputs. For example, the text of âIt is 75 degrees and sunny in Mountain Viewâ can be processed using the APE model to generate varying outputs of: âIn Mountain View it is 75 degrees and sunnyâ; âIt is sunny in Mountain View. The temperature is 75 degreesâ; âIn Mountain View it is currently 75 degrees and sunnyâ; etc.
A training instance for training an APE model to edit natural language generated using the template engine can be automatically generated by processing a ground truth instance of natural language text using random noise to generate training text. In many implementations, random noise can include adding one or more words to the text, removing one or more words from the text, reordering one or more words in the text, and/or additional automatic alternation(s) of the text. The ground truth text and the training text can be stored as a training instance. For example, the text of âThe thermostat is currently set to 72 degreesâ can be processed using random noise to generate training text of âThe thermostat is lights currently set to 72 degreesâ (i.e. adding the word lights); âThe thermostat is currently to 72 degreesâ (i.e., removing the word set); âThe thermostat is currently set 72 to degreesâ (i.e., reordering the words to and 72); etc. Additional and/or alternative noise may added to ground truth text to generate training text.
Accordingly, various implementations set forth techniques for training automatic post-editing models using automatically generated training instances. Generating training instances in a fully supervised manner can be resource and/or labor intensive. For example, in generating a single training instance, relevant data must be transmitted to a client device of a human reviewer, resources of the client device utilized to enable the reviewer to provide an annotation, and the annotation transmitted back for training instance generation. Implementations disclosed herein utilize automatically generated training instances, enabling systems to quickly generate training data corresponding to a wide number of languagesâwithout transmission of data to client devices of reviewers and without utilization of resources of those client devices in obtaining reviewer provided annotations. Additionally or alternatively, a large number of training instances can be automatically generated from a variety of diverse resources, which can improve the accuracy and/or robustness of edited translated text generated using an APE model trained on such training instances. In many implementations, a multilingual APE model can be trained using automatically generated training data. Multilingual models can frequently be trained with fewer training instances from each language. Computational resources (e.g., processor cycles, memory, battery power, and/or additional resources of a computing system) can be conserved by this reduced number of training instances required to train the multilingual APE model for many languages.
As described herein, edited translated text generated using the APE model can be utilized to cause performance of one or more computer actions. For example, the edited translated text can be visually and/or audibly rendered. Since the edited translated text is generated to remove translational and/or other errors, it is more quickly ascertainable by a user to which it is rendered. This can prevent the user from needing to request the text again, wasting computational resources, and/or can enable the user to more quickly provide further user interface input that is responsive to the edited translated text, reducing computer/user interaction duration and conserving resources. As another example, the edited translated text can be utilized in generating and/or transmitting command(s) (e.g., a command to control a smart device in accordance with the edited translated text). Since the edited translated text is generated to remove translational and/or other errors, it can prevent the wasteful transmission of incorrect commands and/or the wasteful implementation of incorrect commands.
The above description is provided only as an overview of some implementations disclosed herein. These and other implementations of the technology are disclosed in additional detail below.
It should be appreciated that all combinations of the foregoing concepts and additional concepts described in greater detail herein are contemplated as being part of the subject matter disclosed herein. For example, all combinations of claimed subject matter appearing at the end of this disclosure are contemplated as being part of the subject matter disclosed herein.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 illustrates an example of generating edited text using an automatic post-editing model in accordance with various implementations disclosed herein.
FIG. 2 illustrates an example environment in which various implementations disclosed herein can be implemented.
FIG. 3 illustrates automatically generating a training instance in accordance with implementations disclosed herein.
FIG. 4 is a flowchart illustrating an example process in accordance with implementations disclosed herein.
FIG. 5 is a flowchart illustrating another example process in accordance with implementations disclosed herein.
FIG. 6 is a flowchart illustrating another example process in accordance with implementations disclosed herein.
FIG. 7 illustrates another example environment in which implementations disclosed herein can be implemented.
FIG. 8 is a flowchart illustrating another example process in accordance with implementations disclosed herein.
FIG. 9 is a flowchart illustrating another example process in accordance with implementations disclosed herein.
FIG. 10 illustrates an example architecture of a computing device.
DETAILED DESCRIPTION
FIG. 1 illustrates an example of generating edited text using an APE model in accordance with various implementations. In the illustrated example 100 , text 102 is processed using NMT model 104 to generate translated text 106 . In many implementations, text 102 is in a source language and translated text 106 is in a target language. For example, source language English text can be translated to target language French text, source language German text can be translated to target language Chinese text, source language French text can be translated to target language Spanish text, and/or additional source language(s) text can be translated to additional target language(s) text. NMT model 104 can be trained for translating a specific source language into a specific target language (e.g., a distinct model translates French to English, another distinct model translates English to French, another distinct model translates French to German, etc.). Additionally or alternatively, NMT model 104 can be a multilingual model trained for translating many source languages into many target languages. For example, multilingual NMT model can be trained to translate English to French, French to English, German to English, English to German, etc.
NMT model 104 can introduce translation errors into translated text 106 . For example, the word club meaning a cudgel can incorrectly be translated as the word club meaning an organization. APE model 108 can process translated text 106 to generate edited text 110 , where translation error(s) introduced by NMT model 104 are corrected in edited text 110 . In a variety of implementations, APE model 108 is an encoder-decoder model (e.g., a transformer model, a sequence to sequence model, and/or or additional encoder-decoder model(s)). Additionally or alternatively, APE model 108 can include a variety of types of machine learning models such as a feed forward neural network model, a recurrent neural network model, a convolutional neural network model, and/or additional types of neural network model(s). APE model 108 can be utilized in correcting a variety of translation errors including incorrectly translated word error(s), gender error(s), and/or additional translation error(s). In many implementations, a transformer model can include a transformer encoder as well as a transformer decoder. The transformer decoder can be utilized in processing text (such as translated text 106 ) to generate an encoded representation of the text. Predicted output (such as edited text 110 ) can be generated, token-by-token, using the transformer decoder attended to the encoded representation of the text. In many implementations, generating the predicted output, token-by-token, using the transformer decoder attended to the encoded representation of the text includes generating a probability distribution over a vocabulary of tokens at each iteration of processing using the transformer decoder. Each token in the vocabulary of tokens can be a distinct word in the vocabulary of the system. Additionally or alternatively, a corresponding token can be selected for an iteration from the vocabulary of tokens based on the probability distribution for the iteration. Furthermore, the selected token can be used as part of the input to the transformer decoder in a subsequent iteration.
In many implementations, APE model 108 is trained to correct errors introduced when translating a source language into a target language. In additional or alternative implementations, APE model 108 is a multilingual model and can correct errors introduced when translating a first language into a target language, a second language into a target language, a third language into a target language, etc. For example, a NMT model can translate Spanish text into English text. Similarly, a NMT model can translate French text into English text. A multilingual APE model can be utilized to edit both the English text translated from the Spanish text as well as the English text translated from the French text.
FIG. 2 illustrates an example environment in which implementations disclosed herein may be implemented. The example environment 200 in FIG. 2 includes client device 202 , APE engine 204 , APE model 206 , APE training engine 208 , APE training instances 210 , APE training instance engine 212 , and/or additional engine(s) and/or model(s) (not depicted). The client device 102 may be, for example, a standalone voice-activated speaker device, a desktop computing device, a laptop computing device, a tablet computing device a mobile phone computing device, a computing device of a vehicle of the user (e.g., an in-vehicle communications system, an in-vehicle entertainment system, an in-vehicle navigation system), and/or a wearable apparatus of the user that includes a computing device (e.g., a watch having a computing device, glasses having a computing device, a virtual or augmented reality computing device, etc.). Additional and/or alternative client devices may be provided.
APE engine 204 , APE training engine 208 , and APE training instance engine 212 are example components in which techniques described herein may interface. The operations performed by one or
more engines
204 , 208 , 212 of FIG. 2 may be distributed across multiple computing systems. In some implementations, one or more aspects of
engines
204 , 208 , 212 may be combined in a single system and/or one or more aspects may be implemented on the client device 202 . For example, in some of those implementations, aspects of APE engine 204 may be combined with aspects of APE training engine 208 . Engines in accordance with many implementations may each be implemented in one or more computing devices that communicate, for example, through a communication network. A communication network may include a wide area network such as the Internet, one or more local area networks (âLANâs) such as Wi-Fi LANs, mesh networks, etc., and/or one or more bus subsystems. A communication network may optionally utilize one or more standard communication technologies, protocols, and/or inter-process communication techniques.
APE model 206 can be trained to process natural language text to generate edited text correcting word translation error(s), gender error(s), etc. In a variety of implementations, APE model 206 can be trained for editing text translated from a specific source language to a specific target language. For example, a first APE model can be trained to generate edited text based on text translated from French to English; a second APE model can be trained to generate edited text based on text translated from English to French; a third APE model can be trained to generate edited text based on text translated from French to German; etc. Additionally or alternatively, a multilingual APE model 206 can be trained to generate edited text based on multiple source languages translated into a target language. For example, the multilingual APE model can be trained to generate edited text based on text translated from German to English, based on text translated from Spanish to English, based on text translated from French to English, and/or based on text translated from additional and/or alternative language(s) to English. Multilingual APE model(s) in accordance with many implementations can be trained to generate text based on translations into additional and/or alternative target languages.
In a variety of implementations, client device 202 , utilizing APE training engine 208 , can train APE model 206 using APE training instances 210 . Utilizing training instances to train an APE model is described with respect to process 500 of FIG. 5 . APE training instance engine 212 can be utilized to generate APE training instances 210 . Generating training instances is described with respect to process 400 of FIG. 4 .
FIG. 3 illustrates an example of automatically generating an APE training instance in accordance with many implementations. Example 300 includes processing ground truth text 302 using NMT model 304 to generate intermediate text 306 . In many implementations, ground truth text 302 is in first language and intermediate text is in a second language. Intermediate text 306 can be processed by NMT model 308 to generate training text 310 . In many implementations, training text 310 is in the first language. Training text 310 in the first language and ground truth text 302 in the first language can be stored as a training instance. In a variety of implementations, NMT model 304 and NMT model 308 can be distinct models. In other implementations, NMT model 304 and NMT model 308 can be included as portions of a multilingual NMT model which is trained to translate multiple language pairs. In a variety of implementations, translating text from a first language into a second language, and translating the resulting text in the second language back into the first language can introduce one or more translation errors. These translation error(s) can be corrected by processing the translated text using an APE model in accordance with many implementations.
FIG. 4 is a flowchart illustrating a process 400 of automatically generating a training instance according to implementations disclosed herein. For convenience, the operations of the flowchart are described with reference to a system that performs operations. This system may include various components of various computer systems, such as one or more components of client device 202 of FIG. 2 . Moreover, while operations of process 400 are shown in a particular order, this is not meant to be limiting. One or more operations may be reordered, omitted, and/or added.
At block 402 , the system processes text in a first language using a NMT model to generate text in a second language. In many implementations, the text in the first language is utilized as the ground truth portion of the training instance. For example, the system can process French text using the NMT model to generate Spanish text.
At block 404 , the system processes the text in the second language generated at block 402 using a NMT model to generate training text in the first language. For example, the system can process the Spanish text generated at block 402 using a NMT model to generate additional French text.
At block 406 , the system stores the training text in the first language and the ground truth text in the first language as a training instance. For example, the system can store the training text generated at block 404 in French as well as the French text processed at block 402 as a training instance.
At block 408 , the system determines whether to generate additional training instances. If so, the system proceeds back to block 402 , and processes additional text in a first language before proceeding to blocks 404 and 406 . If the system determines not to generate additional training instances, the process ends.
In some implementations, preceding text (i.e., a preceding sentence and/or phrase) can be stored as an additional portion of the training instance. For example, the French phrase immediately preceding the French text processed in block 402 can be stored as the preceding text portion of the training instance. The preceding text portion can be utilized to provide additional and/or alternative context to the APE model.
FIG. 5 is a flowchart illustrating a process 500 of training an APE model according to implementations disclosed herein. For convenience, the operations of the flowchart are described with reference to a system that performs operations. This system may include various components of various computer systems, such as one or more components of client device 202 of FIG. 2 . Moreover, while operations of process 500 are shown in a particular order, this is not meant to be limiting. One or more operations may be reordered, omitted, and/or added.
At block 502 , the system automatically generates training instances. In many implementations, the system can automatically generate training instances in accordance with process 400 of FIG. 4 .
At block 504 , the system selects a training instance including a ground truth text portion and a training text portion.
At block 506 , the system applies the training text as input to an APE model to generate predicted output.
At block 508 , the system determines a difference between the ground truth text and the predicted output.
At block 510 , the system updates one or more weights in the APE model based on the difference determined at block 508 . For example, the system can backpropagate the determined difference over the neural network model to update weights of the APE model.
At block 512 , the system determines whether there are one or more additional unprocessed APE training instances. If so, the system proceeds back to block 504 , selects an additional APE training instance, then performs
blocks
506 , 508 , and 510 based on the additional APE training instance. In some implementations, at block 512 , the system may determine not to process any additional APE training instances if one or more training criteria has been satisfied (e.g., a threshold number of epochs have occurred, a threshold duration of training has occurred, and/or all APE training instances have been processed). Although process 500 is described with respect to a non-batch learning technique, batch learning may additionally and/or alternatively be utilized.
Furthermore, although process 500 is described with respect to a training instance including a training text portion and a ground truth text portion, a training instance may additionally or alternatively include a preceding text portion. In some such implementations, the training text as well as the previous text can be processed using the APE model to generate predicted output at block 506 . A difference between the predicted output and the ground truth text can be determined at block 508 .
FIG. 6 is a flowchart illustrating a process 600 of using an APE model to generate edited text according to implementations disclosed herein. For convenience, the operations of the flowchart are described with reference to a system that performs operations. This system may include various components of various computer systems, such as one or more components of client device 202 of FIG. 2 . Moreover, while operations of process 600 are shown in a particular order, this is not meant to be limiting. One or more operations may be reordered, omitted, and/or added.
At block 602 , the system receives text in a target language translated using a NMT model from a source language. For example, the system can receive French text translated from German text using the NMT model. In many implementations, the NMT model introduces error(s) into the text in the target language when translating from the source language to the target language.
At block 604 , the system processes the text in the target language using an APE model to generate edited text. In many implementations, the translation error(s) introduced by the NMT model at block 402 are corrected in the edited text by the APE model.
At block 606 , the system causes a client device to perform one or more actions based on the edited text. For example, the system can process the edited text using a text to speech engine to generate an audio waveform corresponding to the edited text. The audio waveform can be rendered using one or more speakers of the client device. Additionally or alternatively, the system can process the edited text to determine one or more actions to control a device, and the device can perform the one or more actions. Networked hardware devices can include: a smart thermostat, a smart light, networked cameras, smart light switches, and/or other device(s). For example, the edited text of âturn off all the lightsâ can be used to turn off all networked smart lights.
Turning to FIG. 7 , an example environment in which implementations disclosed herein can be implemented. FIG. 7 includes a client computing device 702 , which execute an instance of an automated assistant client 704 . One or more cloud-based automated assistant components 710 can be implemented on one or more computing systems (collectively referred to as a âcloudâ computing system) that are communicatively coupled to client device 702 via one or more local and/or wide area networks (e.g., the Internet) indicated generally at 708 .
An instance of an automated assistant client 704 , by way of its interactions with one or more cloud-based automated assistant components 710 , may form what appears to be, from the user's perspective, a logical instance of an automated assistant 700 with which the user may engage in a human-to-computer dialog. It thus should be understood that in some implementations, a user that engages with an automated assistant client 704 executing on client device 702 may, in effect, engage with his or her own logical instance of an automated assistant 700 . For the sakes of brevity and simplicity, the term âautomated assistantâ as used herein as âservingâ a particular user will often refer to the combination of an automated assistant client 704 executing on a client device 702 operated by the user and one or more cloud-based automated assistant components 710 (which may be shared amongst multiple automated assistant clients of multiple client computing devices). It should also be understood that in some implementations, automated assistant 700 may respond to a request from any user regardless of whether the user is actually âservedâ by that particular instance of automated assistant 700 .
The client computing device 702 may be, for example: a desktop computing device, a laptop computing device, a tablet computing device, a mobile smartphone computing device, a standalone interactive speaker, a smart appliance, and/or a wearable apparatus of the user that includes a computing device (e.g., a watch of the user having a computing device, glasses of the user having a computing device, a virtual or augmented reality computing device). Additional and/or alternative client computing devices may be provided. Additionally or alternatively, operations of client computing device 702 may be distributed between multiple computing devices. For example, one or more operations of client computing device 702 may be distributed between a mobile smartphone and a vehicle computing device. Furthermore, operations of client computing device 702 may be repeated between multiple computing devices (which in some cases may be communicatively coupled). As a further example, a mobile smartphone as well as a vehicle interface device may each implement operations of automated assistant 700 , such as a mobile smartphone and a vehicle interface device both including an invocation engine (described below). In various implementations, the client computing device 702 may optionally operate one or more other applications that are in additional to automated assistant client 704 , such as a message exchange client (e.g., SMS, MMS, online chat), a browser, and so forth. In some of those various implementations, one or more of the other applications can optionally interface (e.g. via an application programming interface) with the automated assistant 704 , or include their own instance of an automated assistant application (that may also interface with the cloud-based automated assistant component(s) 710 ).
Automated assistant 700 engages in human-to-computer dialog sessions with a user via user interface input and output devices of the client device (not pictured). To preserve user privacy and/or to conserve resources, in many situations a user must often explicitly invoke the automated assistant 700 before the automated assistant will fully process a spoken utterance. The explicit invocation of the automated assistant 700 can occur in response to certain user interface input received at the client device 702 . For example, user interface inputs that can invoke the automated assistant 700 via the client device 702 can optionally include actuations of a hardware and/or virtual button of the client device 702 . Moreover, the automated assistant client can include one or more local engines 706 , such as an invocation engine that is operable to detect the presence of one or more spoken invocation phrases. The invocation engine can invoke the automated assistant 700 in response to detection of one or more of the spoken invocation phrases. For example, the invocation engine can invoke the automated assistant 700 in response to detecting a spoken invocation phrase such as âHey Assistantâ, âOK Assistantâ, and/or âAssistantâ. The invocation engine can continuously process (e.g., if not in an âinactiveâ mode) a stream of audio data frames that are based on output from one or more microphones of the client device 702 , to monitor for an occurrence of a spoken invocation phrase. While monitoring for the occurrence of the spoken invocation phrase, the invocation engine discards (e.g., after temporary storage in a buffer) any audio data frames that do not include the spoken invocation phrase. However, when the invocation engine detects an occurrence of a spoken invocation phrase in processed audio data frames, the invocation engine can invoke the automated assistant 700 . As used herein, âinvokingâ the automated assistant 700 can include causing one or more previously inactive functions of the automated assistant 700 to be activated. For example, invoking the automated assistant 700 can include causing one or more local engines 706 and/or cloud-based automated assistant components 710 to further process audio data frames based on which the invocation phrase was detected, and/or one or more following audio data frames (whereas prior to invoking no further processing of audio data frames was occurring).
The one or more local engine(s) 706 of automated assistant 704 can include, for example, the invocation engine described above, a local speech-to-text (âSTTâ) engine (that converts captured audio to text), a local text-to-speech (âTTSâ) engine (that converts text to speech), a local natural language processor (that determines semantic meaning of audio and/or text converted from audio), an APE engine (such as APE engine 204 in FIG. 2 ), a template engine and/or other local components. Because the client device 702 is relatively constrained in terms of computing resources (e.g., processor cycles, memory, battery, etc.), the local engines 706 may have limited functionality relative to any counterparts that are included in cloud-based automated assistant components 710 .
In many implementations, the template engine can identify a text template and a set of structured data response to user interface input. For example, a user can provide input of âturn on the lights in hereâ. The template engine can identify a text template of âOK, turning on [#] of lights in the [room]â and a set of structured data of â#â4, roomâliving roomâ. In many implementations, the template engine can generate natural language text responsive to the user interface input, such as âOK, turning on 4 lights in the living roomâ based on the identified text template and the set of structured data. The template engine can introduce one or more errors into
CLAIMS
Claims ( 19 )
What is claimed is:
1. A method implemented by one or more processors, the method comprising:
detecting a spoken utterance at a client device, wherein the spoken utterance includes an indication of one or more device actions of a device associated with the client device, and wherein the spoken utterance is directed to an automated assistant executing at least in part on the client device;
in response to the spoken utterance being detected at the client device:
generating, by the automated assistant, natural language text, wherein generating the natural language text comprises applying, to a text template, structured data that is responsive to the spoken utterance;
processing the natural language text generated based on applying the structured data that is responsive to the spoken utterance to the text template, using an automatic post-editing model to generate edited text, wherein the edited text, generated based on processing the text using the automatic post-editing model, corrects one or more errors in the natural language text; and
causing the client device to perform one or more actions based on the edited text, wherein causing the client device to perform the one or more actions based on the edited text comprises:
processing the edited text to determine the one or more device actions of the device associated with the client device; and
causing the device to perform the one or more device actions.
2. The method of claim 1 , wherein the one or more errors in the natural language text include a subject-verb agreement error in the natural language text.
3. The method of claim 1 , wherein the text template comprises one or more fixed terms and one or more variables, and wherein generating the natural language text comprises using the set of structured data in populating the one or more variables.
4. The method of claim 1 , wherein the automatic post-editing model is a transformer model that includes a transformer encoder and a transformer decoder, and wherein processing the natural language text using the automatic post-editing model to generate the edited text comprises:
processing the natural language text using the transformer encoder to generate an encoded representation of the natural language text; and
generating the edited text, token-by-token, using the transformer decoder attended to the encoded representation of the natural language text.
5. The method of claim 4 , wherein generating the predicted output, token-by-token, using the transformer decoder attended to the encoded representation of the natural language text comprises:
generating, at each of a plurality of iterations of processing using the transformer decoder, a probability distribution over a vocabulary of tokens;
selecting, from the vocabulary of tokens and based on the probability distribution for the iteration, a corresponding token for the iteration; and
using the selected token as part of input to the transformer decoder in a subsequent iteration of the iterations of processing.
6. The method of claim 1 , wherein the automatic post-editing model is a sequence to sequence model.
7. The method of claim 1 , wherein causing the client device to perform one or more actions based on the edited text comprises:
processing the edited text using a text to speech engine to generate an audio waveform corresponding to the edited text; and
causing the client device to render the audio waveform via one or more speakers of the client device.
8. The method of claim 1 , wherein the device associated with the client device is a light, a thermostat, or a camera.
9. The method of claim 1 , wherein the automatic post-editing model is trained using a training instance including training text and ground truth text, and wherein the training text is generated by:
processing the ground truth text using random noise to generate the training text wherein processing the ground truth text using random noise to generate the training text comprises inserting one or more words into the ground truth text, deleting one or more words from the ground truth text, and/or reordering one or more words in the ground truth text.
10. The method of claim 9 , wherein processing the ground truth text using random noise to generate the training text comprises:
inserting one or more words into the ground truth text.
11. The method of claim 10 , wherein processing the ground truth text using random noise to generate the training text comprises:
reordering one or more words in the ground truth training text.
12. The method of claim 9 , wherein processing the ground truth text using random noise to generate the training text comprises:
deleting one or more words from the ground truth text.
13. The method of claim 1 , wherein the automatic post-editing model is trained for use in correcting one or more translation errors introduced by a neural machine translation model translating text from a source language into a target language.
14. A system, comprising:
memory storing instructions; and
one or more processors executing the instructions to:
detect a spoken utterance at a client device, wherein the spoken utterance includes an indication of one or more device actions of a device associated with the client device, and wherein the spoken utterance is directed to an automated assistant executing at least in part on the client device;
in response to the spoken utterance being detected at the client device:
generate natural language text, wherein in generating the natural language text one or more of the processors are to apply, to a text template, structured data that is responsive to the spoken utterance;
process the natural language text generated based on applying the structured data that is responsive to the spoken utterance to the text template, using an automatic post-editing model to generate edited text, wherein the edited text, generated based on processing the text using the automatic post-editing model, corrects one or more errors in the natural language text; and
cause the client device to perform one or more actions based on the edited text, wherein causing the client device to perform the one or more actions based on the edited text comprises:
process the edited text to determine the one or more device actions of the device associated with the client device; and
cause the device to perform the one or more device actions.
15. The system of claim 14 , wherein the text template comprises one or more fixed terms and one or more variables, and wherein generating the natural language text comprises using the set of structured data in populating the one or more variables.
16. The system of claim 14 , wherein the automatic post-editing model is a transformer model that includes a transformer encoder and a transformer decoder, and wherein in processing the natural language text using the automatic post-editing model to generate the edited text one or more of the processors are to:
process the natural language text using the transformer encoder to generate an encoded representation of the natural language text; and
generate the edited text, token-by-token, using the transformer decoder attended to the encoded representation of the natural language text.
17. The system of claim 14 , wherein in generating the predicted output, token-by-token, using the transformer decoder attended to the encoded representation of the natural language text one or more of the processors are to:
generate, at each of a plurality of iterations of processing using the transformer decoder, a probability distribution over a vocabulary of tokens;
select, from the vocabulary of tokens and based on the probability distribution for the iteration, a corresponding token for the iteration; and
use the selected token as part of input to the transformer decoder in a subsequent iteration of the iterations of processing.
18. The system of claim 14 , wherein in causing the client device to perform one or more actions based on the edited text one or more of the processors are to:
process the edited text using a text to speech engine to generate an audio waveform corresponding to the edited text; and
cause the client device to render the audio waveform via one or more speakers of the client device.
19. The system of claim 14 , wherein the automatic post-editing model is trained using a training instance including training text and ground truth text, and wherein the training text is generated by:
processing the ground truth text using random noise to generate the training text, wherein processing the ground truth text using random noise to generate the training text comprises inserting one or more words into the ground truth text, deleting one or more words from the ground truth text, and/or reordering one or more words in the ground truth text.
US17/700,123
2019-07-15
2022-03-21
Automatic post-editing model for generated natural language text
Active
2039-07-15
US12039286B2
( en )
Priority Applications (2)
Application Number
Priority Date
Filing Date
Title
US17/700,123
US12039286B2
( en )
2019-07-15
2022-03-21
Automatic post-editing model for generated natural language text
US18/773,129
US20240370666A1
( en )
2019-07-15
2024-07-15
Automatic post-editing model for generated natural language text
Applications Claiming Priority (2)
Application Number
Priority Date
Filing Date
Title
US16/511,806
US11295092B2
( en )
2019-07-15
2019-07-15
Automatic post-editing model for neural machine translation
US17/700,123
US12039286B2
( en )
2019-07-15
2022-03-21
Automatic post-editing model for generated natural language text
Related Parent Applications (1)
Application Number
Title
Priority Date
Filing Date
US16/511,806
Continuation
US11295092B2
( en )
2019-07-15
2019-07-15
Automatic post-editing model for neural machine translation
Related Child Applications (1)
Application Number
Title
Priority Date
Filing Date
US18/773,129
Continuation
US20240370666A1
( en )
2019-07-15
2024-07-15
Automatic post-editing model for generated natural language text
Publications (2)
Publication Number
Publication Date
US20220215183A1
US20220215183A1 ( en )
2022-07-07
US12039286B2
true
US12039286B2 ( en )
2024-07-16
Family
ID=74343962
Family Applications (3)
Application Number
Title
Priority Date
Filing Date
US16/511,806
Active
2040-03-19
US11295092B2
( en )
2019-07-15
2019-07-15
Automatic post-editing model for neural machine translation
US17/700,123
Active
2039-07-15
US12039286B2
( en )
2019-07-15
2022-03-21
Automatic post-editing model for generated natural language text
US18/773,129
Pending
US20240370666A1
( en )
2019-07-15
2024-07-15
Automatic post-editing model for generated natural language text
Family Applications Before (1)
Application Number
Title
Priority Date
Filing Date
US16/511,806
Active
2040-03-19
US11295092B2
( en )
2019-07-15
2019-07-15
Automatic post-editing model for neural machine translation
Family Applications After (1)
Application Number
Title
Priority Date
Filing Date
US18/773,129
Pending
US20240370666A1
( en )
2019-07-15
2024-07-15
Automatic post-editing model for generated natural language text
Country Status (1)
Country
Link
US
( 3 )
US11295092B2
( en )
Families Citing this family (24)
* Cited by examiner, â Cited by third party
Publication number
Priority date
Publication date
Assignee
Title
US11159597B2
( en )
2019-02-01
2021-10-26
Vidubly Ltd
Systems and methods for artificial dubbing
US11202131B2
( en )
2019-03-10
2021-12-14
Vidubly Ltd
Maintaining original volume changes of a character in revoiced media stream
US11295092B2
( en )
*
2019-07-15
2022-04-05
Google Llc
Automatic post-editing model for neural machine translation
CN112364658B
( en )
*
2019-07-24
2024-07-26
é¿éå·´å·´éå¢æ§è¡æéå ¬å¸
Translation and speech recognition method, device and equipment
US20210224319A1
( en )
*
2019-12-28
2021-07-22
Ben Avi Ingel
Artificially generating audio data from textual information and rhythm information
US11475220B2
( en )
*
2020-02-21
2022-10-18
Adobe Inc.
Predicting joint intent-slot structure
US11508360B2
( en )
*
2020-09-15
2022-11-22
Microsoft Technology Licensing, Llc
Synthetic data generation for training of natural language understanding models
US20220231873A1
( en )
*
2021-01-19
2022-07-21
Ogoul Technology Co., W.L.L.
System for facilitating comprehensive multilingual virtual or real-time meeting with real-time translation
EP4272109A1
( en )
*
2021-01-29
2023-11-08
Huawei Technologies Co., Ltd.
Cross-lingual apparatus and method
US11966711B2
( en )
*
2021-05-18
2024-04-23
International Business Machines Corporation
Translation verification and correction
US20230057835A1
( en )
2021-10-31
2023-02-23
Ron Zass
Analyzing Image Data to Report Events
US12361230B2
( en )
2021-12-02
2025-07-15
Y.E. Hub Armenia LLC
Method and server for performing domain-specific translation
US12412048B2
( en )
*
2021-12-21
2025-09-09
International Business Machines Corporation
Automatically integrating user translation feedback
US12254874B2
( en )
*
2022-02-20
2025-03-18
Google Llc
False suggestion detection for user-provided content
US12153879B2
( en )
*
2022-04-19
2024-11-26
International Business Machines Corporation
Syntactic and semantic autocorrect learning
US12282755B2
( en )
2022-09-10
2025-04-22
Nikolas Louis Ciminelli
Generation of user interfaces from free text
US12493838B2
( en )
2022-10-12
2025-12-09
Sdl Limited
Translation decision assistant
US11704507B1
( en )
2022-10-31
2023-07-18
Kudo, Inc.
Systems and methods for automatic speech translation
US12346666B2
( en )
*
2022-11-09
2025-07-01
Sdl Limited
Translation review suitability assessment
CN115618891B
( en )
*
2022-12-19
2023-04-07
æ¹å大å¦
A method and system for multimodal machine translation based on contrastive learning
EP4459511A1
( en )
*
2023-04-28
2024-11-06
Fossick LLC
Reduced data machine learning customization and outcome refinement
US11995414B1
( en )
*
2023-08-28
2024-05-28
Sdl Inc.
Automatic post-editing systems and methods
US12380736B2
( en )
2023-08-29
2025-08-05
Ben Avi Ingel
Generating and operating personalized artificial entities
CN117034968B
( en )
*
2023-10-10
2024-02-02
ä¸å½ç§å¦é¢èªå¨åç 究æ
Neural machine translation method, device, electronic equipment and medium
Citations (18)
* Cited by examiner, â Cited by third party
Publication number
Priority date
Publication date
Assignee
Title
US20060178868A1
( en )
*
2005-01-14
2006-08-10
Classified Ventures
Methods and systems for generating natural language descriptions from data
US20090076792A1
( en )
2005-12-16
2009-03-19
Emil Ltd
Text editing apparatus and method
US20090326913A1
( en )
2007-01-10
2009-12-31
Michel Simard
Means and method for automatic post-editing of translations
US20100138210A1
( en )
2008-12-02
2010-06-03
Electronics And Telecommunications Research Institute
Post-editing apparatus and method for correcting translation errors
EP2299369A1
( en )
2009-09-22
2011-03-23
Celer Soluciones S.L.
Management, automatic translation and post-editing method
US8065151B1
( en )
*
2002-12-18
2011-11-22
At&T Intellectual Property Ii, L.P.
System and method of automatically building dialog services by exploiting the content and structure of websites
US20140163963A2
( en )
*
2010-09-24
2014-06-12
National University Of Singapore
Methods and Systems for Automated Text Correction
US20150220513A1
( en )
2014-01-31
2015-08-06
Vivint, Inc.
Systems and methods for personifying communications
US20150279366A1
( en )
*
2014-03-28
2015-10-01
Cubic Robotics, Inc.
Voice driven operating system for interfacing with electronic devices: system, method, and architecture
US9164982B1
( en )
*
2008-11-25
2015-10-20
Yseop Sa
Methods and apparatus for automatically generating text
US9176952B2
( en )
2008-09-25
2015-11-03
Microsoft Technology Licensing, Llc
Computerized statistical machine translation with phrasal decoder
US20170091177A1
( en )
2015-09-30
2017-03-30
Kabushiki Kaisha Toshiba
Machine translation apparatus, machine translation method and computer program product
EP3447655A1
( en )
2017-08-21
2019-02-27
Televic Education NV
A revision system and method for revising translated texts with reduction of false positives
US10248651B1
( en )
*
2016-11-23
2019-04-02
Amazon Technologies, Inc.
Separating translation correction post-edits from content improvement post-edits in machine translated content
US20200104371A1
( en )
2018-09-28
2020-04-02
Baidu Usa Llc
Systems and methods for simultaneous translation with integrated anticipation and controllable latency (stacl)
US20200152186A1
( en )
*
2018-11-13
2020-05-14
Motorola Solutions, Inc.
Methods and systems for providing a corrected voice command
US10896295B1
( en )
*
2018-08-21
2021-01-19
Facebook, Inc.
Providing additional information for identified named-entities for assistant systems
US11295092B2
( en )
*
2019-07-15
2022-04-05
Google Llc
Automatic post-editing model for neural machine translation
Family Cites Families (5)
* Cited by examiner, â Cited by third party
Publication number
Priority date
Publication date
Assignee
Title
US8972268B2
( en )
*
2008-04-15
2015-03-03
Facebook, Inc.
Enhanced speech-to-speech translation system and methods for adding a new word
US9606988B2
( en )
*
2014-11-04
2017-03-28
Xerox Corporation
Predicting the quality of automatic translation of an entire document
US20170286376A1
( en )
*
2016-03-31
2017-10-05
Jonathan Mugan
Checking Grammar Using an Encoder and Decoder
KR102463567B1
( en )
*
2016-05-06
2022-11-07
ì´ë² ì´ ì¸í¬.
Using meta-information in neural machine translation
US10235362B1
( en )
*
2016-09-28
2019-03-19
Amazon Technologies, Inc.
Continuous translation refinement with automated delivery of re-translated content
2019
2019-07-15
US
US16/511,806
patent/US11295092B2/en
active
Active
2022
2022-03-21
US
US17/700,123
patent/US12039286B2/en
active
Active
2024
2024-07-15
US
US18/773,129
patent/US20240370666A1/en
active
Pending
Patent Citations (18)
* Cited by examiner, â Cited by third party
Publication number
Priority date
Publication date
Assignee
Title
US8065151B1
( en )
*
2002-12-18
2011-11-22
At&T Intellectual Property Ii, L.P.
System and method of automatically building dialog services by exploiting the content and structure of websites
US20060178868A1
( en )
*
2005-01-14
2006-08-10
Classified Ventures
Methods and systems for generating natural language descriptions from data
US20090076792A1
( en )
2005-12-16
2009-03-19
Emil Ltd
Text editing apparatus and method
US20090326913A1
( en )
2007-01-10
2009-12-31
Michel Simard
Means and method for automatic post-editing of translations
US9176952B2
( en )
2008-09-25
2015-11-03
Microsoft Technology Licensing, Llc
Computerized statistical machine translation with phrasal decoder
US9164982B1
( en )
*
2008-11-25
2015-10-20
Yseop Sa
Methods and apparatus for automatically generating text
US20100138210A1
( en )
2008-12-02
2010-06-03
Electronics And Telecommunications Research Institute
Post-editing apparatus and method for correcting translation errors
EP2299369A1
( en )
2009-09-22
2011-03-23
Celer Soluciones S.L.
Management, automatic translation and post-editing method
US20140163963A2
( en )
*
2010-09-24
2014-06-12
National University Of Singapore
Methods and Systems for Automated Text Correction
US20150220513A1
( en )
2014-01-31
2015-08-06
Vivint, Inc.
Systems and methods for personifying communications
US20150279366A1
( en )
*
2014-03-28
2015-10-01
Cubic Robotics, Inc.
Voice driven operating system for interfacing with electronic devices: system, method, and architecture
US20170091177A1
( en )
2015-09-30
2017-03-30
Kabushiki Kaisha Toshiba
Machine translation apparatus, machine translation method and computer program product
US10248651B1
( en )
*
2016-11-23
2019-04-02
Amazon Technologies, Inc.
Separating translation correction post-edits from content improvement post-edits in machine translated content
EP3447655A1
( en )
2017-08-21
2019-02-27
Televic Education NV
A revision system and method for revising translated texts with reduction of false positives
US10896295B1
( en )
*
2018-08-21
2021-01-19
Facebook, Inc.
Providing additional information for identified named-entities for assistant systems
US20200104371A1
( en )
2018-09-28
2020-04-02
Baidu Usa Llc
Systems and methods for simultaneous translation with integrated anticipation and controllable latency (stacl)
US20200152186A1
( en )
*
2018-11-13
2020-05-14
Motorola Solutions, Inc.
Methods and systems for providing a corrected voice command
US11295092B2
( en )
*
2019-07-15
2022-04-05
Google Llc
Automatic post-editing model for neural machine translation
Non-Patent Citations (41)
* Cited by examiner, â Cited by third party
Title
Artetxe, M. et al., " Unsupervised Neural Machine Translation "; in International Conference on Learning Representations; 2018; 12 pages.
Artetxe, M. et al., " Unsupervised Statistical Machine Translation "; in Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing; 2018; pp. 3632-3642.
Bahdanau, D. et al., " Neural Machine Translation by Jointly Learning to Align and Translate "; in 3rd International Conference on Learning Representations; 15 pages; 2015.
Bannard, C. et al., " Paraphrasing with Bilingual Parallel Corpora "; in Proceedings of the 43rd Annual Meeting on Association for Computational Linguistics; 2005; pp. 597-604.
Bojar, O. et al., " Findings of the 2017 Conference on Machine Translation (WMT17) "; in Proceedings of the Second Conference on Machine Translation; 2017; 68 pages.
Brants, T. et al., " Large Language Models in Machine Translation "; in Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning; 2007; 10 pages.
Cheng, Y. et al., " Semi-Supervised Learning for Neural Machine Translation "; in Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics; 2016; vol. 1, pp. 1965-1974.
Cotterell, R. et al., " Explaining and Generalizing Back-Translation through Wake-Sleep "; arXiv preprint arXiv:1806.04402; 2018; 7 pages.
Domhan, T. et al., " Using Target-Side Monolingual Data for Neural Machine Translation through Multi-task Learning "; in Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing; 2017; pp. 1500-1505.
Edunov, S. et al., " Understanding Back-Translation at Scale "; in Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing; 2018; pp. 489-500.
Gehring, J. et al., " Convolutional Sequence to Sequence Learning "; in Proceedings of the 34th International Conference on Machine Learningâvol. 70; 2017; pp. 1243-1252.
Gellerstam, M. " Translationese in Swedish Novels Translated from English "; Translation Studies in Scandinavia; 1986; pp. 88-95.
Gulcehre, C. et al., " On Integrating a Language Model into Neural Machine Translation "; Comput. Speech Lang.; 2017; pp. 137-148.
Gulcehre, C. et al., " On Using Monolingual Corpora in Neural Machine Translation "; arXiv preprint arXiv:1503.03535; 2015; 9 pages.
He, D. et al., " Dual Learning for Machine Translation "; in Conference on Advances in Neural Information Processing Systems (NeurIPS); 2016; 9 pages.
Hoang, C. et al., " Iterative Backtranslation for Neural Machine Translation "; in Proceedings of the 2nd Workshop on Neural Machine Translation and Generation, vol. 1; 2018; pp. 18-24.
Junczys-Dowmunt, M., " Microsoft's Submission to the WMT2018 News Translation Task: How I Learned to Stop Worrying and Love the Data "; in Proceedings of the Third Conference on Machine Translation, vol. 2: Shared Task Papers; Association for Computational Linguistics; 2018; pp. 425-430.
Kiesel et al. " Clarifying False Memories in Voice-based Search ", CHIIR '19, Mar. 10-14, 2019, Glasgow, United Kingdom (Year: 2019).
*
Koppel, M. et al., " Translationese and Its Dialects "; in Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologiesâvol. 1; 2011; pp. 1318-1326.
Lample, G. et al., " Phrase-Based & Neural Unsupervised Machine Translation "; in Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP); 2018; 11 pages.
Lample, G. et al., Unsupervised Machine Translation Using Monolingual Corpora Only; in International Conference on Learning Representations; 2018; 14 pages.
Liyanapathirana et al. " Using the TED Talks to Evaluate Spoken Post-editing of Machine Translation ". In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16), May 2016, pp. 2232-2239 (Year: 2016).
*
Mallinson, J. et al., " Paraphrasing Revisited with Neural Machine Translation "; in Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: vol. 1, Long Papers, vol. 1; 2017; pp. 881-893.
Niu, X. et al., " Bi-Directional Neural Machine Translation with Synthetic Parallel Data "; ACL 2018; 8 pages.
Papineni, K. et al., " Bleu: A Method for Automatic Evaluation of Machine Translation "; in Proceedings of the 40th Annual Meeting on Association for Computational Linguistics, . Association for Computational Linguistics; 2002; 8 pages.
Peter, J. et al., " The QT21/HimL Combined Machine Translation System "; in Proceedings of the First Conference on Machine Translation; 2016; pp. 344-355.
Poncelas, A. et al., " Investigating Backtranslation in Neural Machine Translation "; in Proceedings of the 21st Annual Conference of the European Association for Machine Translation; 2018; pp. 249-258.
Post, M. " A Call for Clarity in Reporting Bleu Scores "; arXiv preprint arXiv:1804.08771; 2018; 6 pages.
Selinker, L. " Interlanguage "; International Review of Applied Linguistics; 1972; pp. 209-231.
Sennrich, R. et al., " Edinburgh Neural Machine Translation Systems for WMT 16 "; in Proceedings of the First Conference on Machine Translation; 2016; pp. 371-376.
Sennrich, R. et al., " Improving Neural Machine Translation Models with Monolingual Data "; in Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (vol. 1: Long Papers); 2016; pp. 86-96.
Shaw, P. et al., " Self-Attention with Relative Position Representations "; in Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies; 2018; pp. 464-468.
Shen, J. et al., " Lingvo: A Modular and Scalable Framework for Sequence-to-Sequence Modeling "; CoRR, arXiv:1902.08295; 2019; 17 pages.
Stahlberg, F. et al., " The University of Cambridge's Machine Translation Systems for WMT18 "; in Proceedings of the Third Conference on Machine Translation, vol. 2: Shared Task Papers; 2018; pp. 504-512.
Tebbifakhr et al. " Multi-source Transformer for Automatic Post-Editing of Machine Translation Output ". Emerging Topics from the Fifth Italian Conference on Computational Linguistics, vol. 5, No. 1, Jun. 2019 (Year: 2019).
*
Vaswani, A. et al. " Attention Is All You Need "; in Advances in Neural Information Processing Systems; 2017; 11 pages.
Warner, M., " The Politics of Translation "; London Review of Books; 40 No. 19:22; 2018.
Wu, W. et al. " Improving Neural Machine Translation with Neural Sentence Rewriting "; in 2018 International Conference on Asian Language Processing (IALP).
Xia, Y. et al., " Dual Supervised Learning "; in International Conference on Machine Learning; 2017; 11 pages.
Zhang, Z. et al., " Joint Training for Neural Machine Translation Models with Monolingual Data "; in Thirty-Second AAAI Conference on Artificial Intelligence; 2018; 8 pages.
Zoph et al.; Multi-Source Neural Translation; arXiv perprint arXiv:1601.00710v1 [cs.CL]; dated Jan. 5, 2016.
Also Published As
Publication number
Publication date
US20220215183A1
( en )
2022-07-07
US11295092B2
( en )
2022-04-05
US20210019373A1
( en )
2021-01-21
US20240370666A1
( en )
2024-11-07
Similar Documents
Publication
Publication Date
Title
US20240370666A1
( en )
2024-11-07
Automatic post-editing model for generated natural language text
JP7063932B2
( en )
2022-05-09
Appropriate agent automation assistant call
US11797772B2
( en )
2023-10-24
Word lattice augmentation for automatic speech recognition
US11545157B2
( en )
2023-01-03
Speaker diartzation using an end-to-end model
US9805718B2
( en )
2017-10-31
Clarifying natural language input using targeted questions
US12073187B2
( en )
2024-08-27
Automatic evaluation of natural language text generated based on structured data
US12204569B2
( en )
2025-01-21
Natural language understanding clarifications
US12080271B2
( en )
2024-09-03
Speech generation using crosslingual phoneme mapping
US11942091B2
( en )
2024-03-26
Alphanumeric sequence biasing for automatic speech recognition using a grammar and a speller finite state transducer
EP4086904A1
( en )
2022-11-09
Speaker awareness using speaker dependent speech model(s)
WO2022191892A1
( en )
2022-09-15
Device arbitration for local execution of automatic speech recognition
EP4254256A1
( en )
2023-10-04
Spoken language