JP6672344B2 - System and method for multi-user multi-language communication - Google Patents

Description

CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority and benefit of US patent application Ser. No. 13 / 908,979, filed Jun. 3, 2013, which is hereby incorporated by reference in its entirety. I do.

FIELD OF THE INVENTION The present invention relates generally to language translation, and more particularly to language translation with multiple users and multiple languages.

Background of the Invention Prior to the advent of machine-based language translation (hereinafter "machine translation"), translation between two languages was only possible by human-trained human intervention or interpretation in both languages. In contrast, typical machine translators generally operate based on statistical / probabilistic analysis of context and grammar, usually without the need for human intervention / interpretation.

  Typical machine translation is often error prone, especially if the text to be translated has minimal context. Text with minimal context is often found in conversations that employ a concise sentence structure. In addition, machine translation often involves abbreviations, acronyms, contractions, colloquial words / phrases, proper nouns, and common nouns. These are also commonly found in conversational texts.

SUMMARY OF THE INVENTION Various embodiments described herein provide systems and methods related to multilingual communication between multiple users, possibly at two or more client systems. Modes of communication facilitated by embodiments include Internet-based chat (eg, Apple® iMessage®, Windows® Live Messenger, etc.), e-mail (eg, embedded forum messaging, Yahoo (registered) Trademark) email, RFC 5322, etc.), text-based cellular communications (eg, SMS or MMS messages), online forum postings (eg, web-based hobby forum postings), and online social media services (eg, Twitter (registered) Trademark), Facebook, etc.). For example, the systems and methods may implement a multilingual multi-user chat system.

For some embodiments, the provided method comprises: identifying a first language and a second language; and providing, from a first person in a first chat client system communicating in the first language, Receiving an initial message in one language and querying a data store for a first corresponding message in a second language based on the initial message in the first language. If the data store includes a first corresponding message, the method assists in transmitting the corresponding message to a second person in a second chat client system communicating in a second language. Depending on the embodiment, the initial message may include text, emoticons, ASCII base art, or other content suitable or customary for human-readable messages transmitted over the network. In addition, the initial message may be part of a larger message being communicated between chat client systems (eg, the initial message is a sentence in a multi-sentence message).

  If the data store does not include the first corresponding message, the method may attempt to utilize a conversion engine to convert at least a portion of the initial message into a converted message in the first language. With the transformed message, the method may query the data store for a second corresponding message in a second language based on the transformed message.

  For certain embodiments, the system or method may attempt to transform the initial message using a series of transformation operations before querying the data store for a second corresponding message based on the transformed message. Alternatively, in some embodiments, the system or method may perform the conversion and query iteratively, such that the initial message is converted using a subset of the available conversion operations, resulting in The data store is queried for a second corresponding message based on the resulting transformed message, and if the second corresponding message is not identified, another iteration of the transformation and query is performed (eg, the resulting result). The transformed message is further transformed with another subset of the available transformation operations, and the data store is queried for a second corresponding message based on the resulting transformed message). In some such embodiments, a subset of the transform operations applied at each iteration may be applied to the initial message, or may be applied to the resulting latest transformed message.

  Ultimately, the method may assist in translating the initial or transformed message into a corresponding message in a second language. In some embodiments, when the first corresponding message for the initial message is not in the data store and the conversion engine does not convert at least a portion of the initial message, the initial message may be translated into a corresponding message. In addition, in various embodiments, when the first corresponding message for the initial message is not in the data store, when the conversion engine provides a converted message that includes a conversion of at least a portion of the initial message, and when converted. When the data store does not include a second corresponding message for the message, the transformed message may be translated into a corresponding message.

Depending on the embodiment, transforming a portion of the initial message identifies words or phrases of the chat term in the initial message (eg, “lol”, “gr8t”) and converts the words or phrases of the chat term to non-chat terms. Replacing with a word or phrase, performing a spell check on a portion of the initial message, or identifying an abbreviation in a portion of the initial message, and replacing the abbreviation with a word or phrase corresponding to the abbreviation (eg, represented by an abbreviation). (Eg, replacing “CA” with “California” or “brb” with “be right back”).

Moreover, translating a portion of the initial message identifies the acronym in the portion of the initial message, and replaces the acronym with a word or phrase (eg, represented by an acronym) corresponding to an acronym (eg, “USA”). Or identifying colloquial words or phrases in a portion of the initial message and replacing the colloquial words or phrases with words or phrases that represent colloquial words or phrases. Further, translating a portion of the initial message may include identifying abusive words or phrases in the portion of the initial message, and converting the abusive words or phrases (eg, representing the abusive words or phrases) to abuse. It may include replacing with non-target words or phrases, or removing abusive words or phrases from the initial message.

  For some embodiments, transforming the portion of the initial message includes flagging the portion of the initial message from being translated. For example, some parts of the initial message may include proper nouns, common nouns, contractions, abbreviations, or acronyms, and the method may flag the parts so that they are not translated in subsequent operations.

  Certain embodiments provide a system that includes various components configured to perform the various operations described herein. Similarly, certain embodiments provide a computer program product that includes computer instruction codes configured to cause a computer system to perform various operations described herein.

  In one aspect, the invention relates to a computer-implemented method. The method includes selecting a word or phrase from a data store that is associated with a translation failure from a first language to a second language of a message containing the word or phrase, and providing a user seeking user feedback about the translation failure. Selecting, determining a value of the incentive to provide to the user in exchange for the user feedback, and sending a request for the feedback to the user's computing device, the request comprising the incentive Wherein the method further comprises receiving the user feedback from the computing device, wherein the user feedback comprises a word or phrase in each of the first or second languages, the method comprising: Furthermore, the above user feedback Includes determining to be approved, based on the approval, according to the value of the incentive and providing credit to the account of the user.

  In some embodiments, the failure is due to an actual failure to translate the message. The failure may be identified by the user flagging the message as potentially incorrect or may be obtained by flagging. Selecting the user may be based on the user's certainty measure, the quota associated with the user, the user's previous credit for the account, the user's preferences, or the user's linguistic ability. In some embodiments, the word or phrase includes the first language chat term. The response may include chat terms in the second language. In some implementations, the query includes a field configured to receive a text-based value.

  In one embodiment, the request includes a pre-selected set of definitions that allows the user to select a definition for the word or phrase. The pre-selected set of definitions may include, for example, at least one definition provided by another user in response to another request, wherein the other request has previously received the word or phrase from the other user. Has been previously generated to get user feedback. The other request may include another set of the preselected definitions from which the other user has selected the definition. The method may further include evaluating the user feedback to determine a most common response.

In various embodiments, the method further includes determining the suitability of the user based on the user feedback. The method may further include updating a translation or translation of the word or phrase from the first language to the second language based on the user feedback. In some embodiments, determining that the user feedback is approved may include determining that the user feedback is not incorrect and / or determining that the user feedback is accurate. . Determining that the user feedback is approved may be based on a comparison of the user feedback to at least one previous user feedback provided by another user in response to the other request, wherein the other request comprises: It has been previously generated to get feedback on the word or phrase from the other user.

  In some embodiments, the incentive includes (or is an offer for) in-game currency or in-game items. The value of the incentive may be determined, for example, based on the complexity of the word or phrase or the importance of the word or phrase. In some examples, determining the value of the incentive includes: (i) the complexity of the word or phrase; (ii) the importance of the word or phrase; (iii) the response method used by the user; (Iv) considering the type of word or phrase, and / or (v) the language involved in the translation failure.

  In another aspect, the invention is directed to a system that includes one or more computers that are programmed to perform operations. The operation includes selecting a word or phrase from a data store that is associated with a translation failure from a first language to a second language of a message containing the word or phrase, and providing a user seeking user feedback about the translation failure. Selecting, determining a value of the incentive to provide to the user in exchange for the user feedback, and sending a request for the feedback to the user's computing device, the request comprising the incentive Wherein the operation further comprises receiving the user feedback from the computing device, wherein the user feedback includes a respective word or phrase of the first or second language, and wherein the operation Furthermore, the above user feedback Includes determining to be approved, and providing credit to the account of the user based on the approval.

  In some embodiments, the failure is due to an actual failure to translate the message. The failure may be identified by the user as flagging the message as potentially incorrect, or may be obtained by flagging. Selecting the user may be based on the user's certainty measure, the quota associated with the user, the user's previous credit for the account, the user's preferences, or the user's linguistic ability. In some embodiments, the word or phrase includes the first language chat term. The response may include a second language chat term. In some implementations, the query includes a field configured to receive a text-based value.

  In one embodiment, the request includes a pre-selected set of definitions that allows the user to select a definition for the word or phrase. The pre-selected set of definitions may include, for example, at least one definition provided by another user in response to another request, wherein the other request has previously received the word or phrase from the other user. Has been previously generated to get user feedback. The other request may include another set of the preselected definitions from which the other user has selected the definition. The operations may further include evaluating the user feedback to determine a most common response.

In various embodiments, the act further includes determining the suitability of the user based on the user feedback. The operations may further include updating a translation or translation of the word or phrase from the first language to the second language based on the user feedback. In some embodiments, determining that the user feedback is approved may include determining that the user feedback is not incorrect and / or determining that the user feedback is accurate. . Determining that the user feedback is approved may be based on a comparison of the user feedback to at least one previous user feedback provided by another user in response to the other request, wherein the other request comprises: It has been previously generated to get feedback on the word or phrase from the other user.

  In some embodiments, the incentive includes (or is an offer for) in-game currency or in-game items. The value of the incentive may be determined, for example, based on the complexity of the word or phrase or the importance of the word or phrase. In some examples, determining the value of the incentive includes: (i) the complexity of the word or phrase; (ii) the importance of the word or phrase; (iii) the response method used by the user; (Iv) considering the type of word or phrase, and / or (v) the language involved in the translation failure.

  In another aspect, the present invention relates to a computer program product stored on one or more storage media for improving language translation through incentive feedback. The computer program product, wherein the data processing device selects a word or phrase from a data store that is associated with a failure to translate a message containing the word or phrase from a first language to a second language. Selecting a user seeking user feedback on the failure; determining a value of an incentive to provide to the user in exchange for the user feedback; and sending a request for the feedback to the user's computing device. Wherein the request includes the incentive, and wherein the computer program product further comprises: the data processing device further comprising: the user feedback from the computing device. Being executable by the data processing device to perform an operation including receiving, the user feedback comprising a respective word or phrase in the first or second language, and wherein the computer program product further comprises: The data processing device is executable by the data processing device to perform operations including determining that the user feedback is approved and crediting the user's account based on the approval.

  In some embodiments, the failure is due to an actual failure to translate the message. The failure may be identified by the user as flagging the message as potentially incorrect, or may be obtained by flagging. Selecting the user may be based on the user's certainty measure, the quota associated with the user, the user's previous credit for the account, the user's preferences, or the user's linguistic ability. In some embodiments, the word or phrase includes the first language chat term. The response may include a second language chat term. In some implementations, the query includes a field configured to receive a text-based value.

  In one embodiment, the request includes a pre-selected set of definitions that allows the user to select a definition for the word or phrase. The pre-selected set of definitions may include, for example, at least one definition provided by another user in response to another request, wherein the other request has previously received the word or phrase from the other user. Has been previously generated to get user feedback. The other request may include another set of the preselected definitions from which the other user has selected the definition. The operations may further include evaluating the user feedback to determine a most common response.

In various embodiments, the act further includes determining the suitability of the user based on the user feedback. The operations may further include updating a translation or translation of the word or phrase from the first language to the second language based on the user feedback. In some embodiments, determining that the user feedback is approved may include determining that the user feedback is not incorrect and / or determining that the user feedback is accurate. . Determining that the user feedback is approved may be based on a comparison of the user feedback to at least one previous user feedback provided by another user in response to the other request, wherein the other request comprises: It has been previously generated to get feedback on the word or phrase from the other user.

  In some embodiments, the incentive includes (or is an offer for) in-game currency or in-game items. The value of the incentive may be determined, for example, based on the complexity of the word or phrase or the importance of the word or phrase. In some examples, determining the value of the incentive includes: (i) the complexity of the word or phrase; (ii) the importance of the word or phrase; (iii) the response method used by the user; (Iv) considering the type of word or phrase, and / or (v) the language involved in the translation failure.

  In one aspect, the invention relates to a method implemented by a data processing device. The method includes providing a text message chat system to a plurality of users, receiving an original text message in a first language from a first user, and first translating the original text message in a second language. Generating; providing the second user with the original text message and the first translation; receiving a translation correction from the second user to respond to an error in the first translation; (a Identifying the most accurate translation correction from the plurality of translation corrections, including the translation correction from the second user; and (b) selecting a word-based feature, a language-based feature, and a word alignment feature. Assessing the accuracy of said translation correction from said second user using at least one of: Chi comprising at least one.

  In certain embodiments, the method includes providing an incentive (eg, virtual goods and / or virtual currency used in an online game) to encourage the second user to submit the translation correction. . Determining the most accurate translation correction may include receiving at least one additional translation correction from at least one additional user to respond to the error in the initial translation. The one additional translation correction and the translation correction from the second user define the plurality of translation corrections, and determining the most accurate translation correction further comprises determining the plurality of translation corrections. This may include receiving feedback from a user regarding accuracy and identifying the most accurate translation correction from the plurality of translation corrections based on the feedback.

  In some implementations, the method further includes providing a reward (eg, virtual goods and / or virtual currency used in online games) to the user who submitted the most accurate translation correction. The method may further include providing a reward (eg, virtual goods and / or virtual currency used in online games) to the user who provided the feedback used to identify the most accurate translation. Word-based features may include, for example, word counts, character counts, emoticons, numbers and / or punctuation. Using the language-based features may include identifying the part of speech in the original text message and the translation correction from the second user.

In some embodiments, the method further comprises: identifying a number of verbs present in each of the original text message and the translation correction from the second user; With the number of verbs in the translation correction from the second user. Lack of part of speech in the original text message and / or the translation correction from the second user may indicate language detection failure. The method may further include rejecting the translation correction from the second user if the translation correction from the second user is the same as the first translation.

  In another aspect, the invention relates to a system that includes a computer readable medium having stored instructions, and a data processing device. The data processing apparatus includes: providing a text message chat system to a plurality of users; receiving an original text message in a first language from a first user; Generating a translation; providing the original text message and the first translation to a second user; receiving a translation correction from the second user to respond to an error in the first translation; (A) identifying the most accurate translation correction from the plurality of translation corrections including the translation correction from the second user; and (b) word-based features, language-based features, and / or words. Evaluating the accuracy of the translation correction from the second user using the alignment features. Chino configured to execute instructions to perform the operations including the performing at least one.

  In some embodiments, the act includes providing an incentive (eg, virtual goods and / or virtual currency used in an online game) to encourage the second user to submit the translation correction. . Determining the most accurate translation correction may include receiving at least one additional translation correction from at least one additional user to respond to the error in the initial translation. The one additional translation correction and the translation correction from the second user define the plurality of translation corrections, and determining the most accurate translation correction further comprises determining the plurality of translation corrections. This may include receiving feedback from a user regarding accuracy and identifying the most accurate translation correction from the plurality of translation corrections based on the feedback.

  In some implementations, the act further includes providing a reward (eg, virtual goods and / or virtual currency used in online games) to the user who submitted the most accurate translation correction. The operations may further include providing a reward (eg, virtual goods and / or virtual currency used in online games) to the user who provided the feedback used to identify the most accurate translation. Word-based features may include, for example, word counts, character counts, emoticons, numbers and / or punctuation. Using the language-based features may include identifying the part of speech in the original text message and the translation correction from the second user.

  In some embodiments, the operation further comprises: identifying a number of verbs present in each of the original text message and the translation correction from the second user; and determining a number of verbs in the original text message. Comparing with the number of the verbs in the translation correction from the second user. Lack of part of speech in the original text message and / or the translation correction from the second user may indicate language detection failure. The operation may further include rejecting the translation correction from the second user if the translation correction from the second user is the same as the first translation.

In another aspect, the invention relates to a computer program product stored on one or more storage media for controlling a processing mode of a data processing device. The computer program product includes: the data processing device providing a text message chat system to a plurality of users; receiving an original text message in a first language from a first user; Generating a first translation of a second language, providing the second user with the original text message and the first translation, and translating from the second user to respond to errors in the first translation. Receiving a correction; (a) identifying the most accurate translation correction from the plurality of translation corrections including the translation correction from the second user; and (b) word-based features, language-based features. And / or using at least one of the word alignment features Serial executable by said data processing apparatus to perform operations including at least one of evaluating the accuracy of the translation modifications from the second user.

  In some embodiments, the act includes providing an incentive (eg, virtual goods and / or virtual currency used in an online game) to encourage the second user to submit the translation correction. . Determining the most accurate translation correction may include receiving at least one additional translation correction from at least one additional user to respond to the error in the initial translation. The one additional translation correction and the translation correction from the second user define the plurality of translation corrections, and determining the most accurate translation correction further comprises determining the plurality of translation corrections. This may include receiving feedback from a user regarding accuracy and identifying the most accurate translation correction from the plurality of translation corrections based on the feedback.

  In some implementations, the act further includes providing a reward (eg, virtual goods and / or virtual currency used in online games) to the user who submitted the most accurate translation correction. The operations may further include providing a reward (eg, virtual goods and / or virtual currency used in online games) to the user who provided the feedback used to identify the most accurate translation. Word-based features may include, for example, word counts, character counts, emoticons, numbers and / or punctuation. Using the language-based features may include identifying the part of speech in the original text message and the translation correction from the second user.

  In some embodiments, the act of identifying the number of verbs present in each of the original text message and the translation correction from the second user, and determining the number of verbs in the original text message Comparing with the number of the verbs in the translation correction from the second user. Lack of part of speech in the original text message and / or the translation correction from the second user may indicate language detection failure. The action may include rejecting the translation correction from the second user if the translation correction from the second user is the same as the first translation.

  In one aspect, the invention relates to a method implemented by a data processing device. The method includes identifying a first language and a second language, and audible chat terms in the first language from a first person in a first chat client system communicating in the first language. Receiving a message, converting the chat term audible message to the first language chat term text message, and converting the chat term text message to a plain language text message of the first language. Translating the plain language text message into a corresponding plain language text message in the second language; and translating the corresponding plain language text message into a corresponding chat term in the second language. Converting to a text message and the corresponding chat term text message above Converting the corresponding chat term audible message into the second language and sending the corresponding chat term audible message to a second person in a second chat client system communicating in the second language. Including.

In some embodiments, converting the chat term audible message to the first language chat term text message includes providing the chat term audible message to a speech recognition system. Converting the chat term text message may include identifying a chat term word or phrase in the chat term text message and replacing the chat term word or phrase with a non-chat word or phrase. . In some examples, converting the corresponding chat term text message to a corresponding chat term audible message in the second language comprises providing the corresponding chat term text message to a text-to-speech system. including.

  In one implementation, the speech recognition system and / or the text-to-speech system uses data including a plurality of accents and dialects for each of the first and second languages. The data may include a chat term and plain language format for each of the first and second languages. In various embodiments, the method receives feedback from a user regarding the accuracy of at least one of the corresponding plain text message, the corresponding chat term text message, and / or the corresponding chat term audible message. Including. The method may further include offering the user an incentive (eg, virtual goods and / or virtual currency used in an online game) for providing feedback. In some cases, the method includes processing the feedback to improve the accuracy of the speech recognition system and / or the text-to-speech system.

  In another aspect, the invention is directed to a system that includes a computer readable medium storing instructions and a data processing device. The data processing device is configured to identify a first language and a second language, and to provide a chat in the first language from a first person in a first chat client system communicating in the first language. Receiving a term audible message; converting the chat term audible message to the first language chat term text message; and converting the chat term text message to a plain language text message of the first language. Translating the plain language text message into a corresponding plain language text message in the second language, and translating the corresponding plain language text message in the second language. Convert to chat term text message and the corresponding chat term text above Converting the message into a corresponding chat term audible message in the second language, and sending the corresponding chat term audible message to a second person in a second chat client system communicating in the second language. And an instruction to perform an operation including:

  In some embodiments, converting the chat term audible message to the first language chat term text message includes providing the chat term audible message to a speech recognition system. Converting the chat term text message includes identifying a chat term word or phrase in the chat term text message and replacing the chat term word or phrase with a non-chat word or phrase. May be included. In some examples, converting the corresponding chat term text message to a corresponding chat term audible message in the second language comprises providing the corresponding chat term text message to a text-to-speech system. including.

In one implementation, the speech recognition system and / or the text-to-speech system uses data including a plurality of accents and dialects for each of the first and second languages. The data may include a chat term and plain language format for each of the first and second languages. In various embodiments, the response includes receiving feedback from a user regarding the accuracy of the corresponding plain text message, the corresponding chat term text message, and / or the corresponding chat term audible message. . The operations may further include providing the user with incentives (eg, virtual goods and / or virtual currency used in online games) for providing feedback. In some cases, the operations include processing feedback to improve the accuracy of the speech recognition system and / or the text-to-speech system.

  In another aspect, the invention relates to a computer program product stored on one or more storage media for controlling a processing mode of a data processing device. The computer program product may further comprise: the data processing device identifying a first language and a second language; and a first person in a first chat client system communicating in the first language. Receiving a chat term audible message in a first language; converting the chat term audible message into a chat term text message in the first language; and converting the chat term text message to a plain language in the first language. Converting the plain text message into a corresponding plain text message in the second language; and converting the corresponding plain text message into the second plain text message. Convert to the corresponding chat term text message in two languages Converting the corresponding chat term text message into a corresponding chat term audible message in the second language; and providing the second person in a second chat client system communicating in the second language with the second person. Executable by the data processing device to perform operations including sending a corresponding chat term audible message.

  In some embodiments, converting the chat term audible message to the first language chat term text message includes providing the chat term audible message to a speech recognition system. Converting the chat term text message includes identifying a chat term word or phrase in the chat term text message and replacing the chat term word or phrase with a non-chat word or phrase. May be included. In some examples, converting the corresponding chat term text message to a corresponding chat term audible message in the second language comprises providing the corresponding chat term text message to a text-to-speech system. including.

  In some implementations, the speech recognition system and / or the text-to-speech system uses data including a plurality of accents and dialects for each of the first and second languages. The data may include a chat term and plain language format for each of the first and second languages. In various embodiments, the act of receiving receiving feedback from a user regarding the accuracy of the corresponding plain text message, the corresponding chat term text message, and / or the corresponding chat term audible message. Including. The operations may further include providing the user with incentives (eg, virtual goods and / or virtual currency used in online games) for providing feedback. In some cases, the operations include processing feedback to improve the accuracy of the speech recognition system and / or the text-to-speech system.

In one aspect, the invention relates to a method implemented by a data processing device. The method may be implemented by combining old training data (e.g., including one or more old text messages for which a correct translation into a different language is known) and one or more new training data (e.g., the correct translation to a different language is not known). Selecting a mixture of new training data (including a text message); and (i) a respective request that the user translate at least one of the old training data and the new training data, and / or (ii) Sending a plurality of respective requests including respective incentives to the translation to a user's client device (or a plurality of user's multiple client devices) at different times; and, after sending the specific request, the specific request. About the old training data of Receiving the translation from the computer; comparing the received translation with the correct translation for the old training data; determining the accuracy of the received translation based on the comparison; Updating the certainty score for the user based on The certainty score then indicates the likelihood that the user will provide an accurate translation of the text message into the different language.

  In some embodiments, the user is a participant in an online game. Each of the above incentives may include, for example, virtual goods and / or virtual currency for online games. Determining the accuracy of the translation received from the user may include (i) calculating a word error rate (WER) and / or (ii) determining a bilingual evaluation understudy (BLEU). May include using. In some cases, updating the certainty score for the user includes using item response theory to identify deviations from standards in the accuracy of the user's translation. The method may further include revoking the user's translation rights if the certainty score falls below a threshold.

  In various implementations, the method includes providing the respective incentive to the user if the translation of the user is determined to be correct. The method may further include detecting collusion between the user and the second user by identifying a pre-existing relationship between the user and the second user. In one example, identifying the pre-existing relationship includes analyzing a social network of the user and / or the second user.

  In another aspect, the invention relates to a system that includes a computer readable medium having stored instructions, and a data processing device. The data processing device may include old training data (e.g., including one or more old text messages whose correct translation into a different language is known) and one or more old training data (e.g., the correct translation into the different language is not known). (I) a respective request for the user to translate at least one of the old training data and the new training data, and / or (Ii) sending a plurality of respective requests, including respective incentives for the translation, to a user's client device (or a plurality of user's client devices) at different times; Of the old training data Receiving a translation from an ant device; comparing the received translation with the correct translation for the old training data; determining, based on the comparison, the accuracy of the received translation; And executing instructions to perform an action including updating a certainty score for the user based on the translation. The certainty score then indicates the likelihood that the user will provide an accurate translation of the text message into the different language.

In one embodiment, the user is a participant in an online game. Each of the above incentives may include, for example, virtual goods and / or virtual currency for online games. Determining the accuracy of the translation received from the user may include (i) calculating a word error rate (WER) and / or (ii) calculating a bilingual evaluation under study (BLEU). May include using. In some cases, updating the certainty score for the user includes using item response theory to identify deviations from standards in the accuracy of the user's translation. The operation may further include revoking the translation authority of the user when the certainty score falls below a threshold.

  In various implementations, the act includes providing the respective incentive to the user if the translation of the user is determined to be correct. The operation may further include detecting a collusion between the user and the second user by identifying a pre-existing relationship between the user and the second user. In one example, identifying the pre-existing relationship includes analyzing a social network of the user and / or the second user.

  In another aspect, the invention relates to a computer program product stored on one or more storage media for controlling a processing mode of a data processing device. The computer program product may be configured such that the data processing device includes old training data (e.g., including one or more old text messages for which a correct translation into a different language is known) and a correct translation (e.g., into the different language) Selecting one or more new training data mixes (including one or more new text messages for which unknown is not known) and (i) the user translating the old training data and at least one of the new training data. Requesting and / or (ii) sending a plurality of respective requests including respective incentives to the translation to a user's client device (or multiple user devices) at different times, and After sending, on the above specific request Receiving a translation from the client device for the old training data, comparing the received translation with the correct translation for the old training data, and determining the accuracy of the received translation based on the comparison. The data processing device is operable to perform operations including determining and updating a certainty score for the user based on the translation. The certainty score then indicates the likelihood that the user will provide an accurate translation of the text message into the different language.

  In some embodiments, the user is a participant in an online game. Each of the above incentives may include, for example, virtual goods and / or virtual currency for online games. Determining the accuracy of the translation received from the user may include (i) calculating a word error rate (WER) and / or (ii) determining a bilingual evaluation understudy (BLEU). May include using. In some cases, updating the certainty score for the user includes using item response theory to identify deviations from standards in the accuracy of the user's translation. The operation may further include revoking the translation authority of the user when the certainty score falls below a threshold.

  In various implementations, the operation includes providing a respective incentive to the user if the translation of the user is determined to be correct. The operation may further include detecting a collusion between the user and the second user by identifying a pre-existing relationship between the user and the second user. In one example, identifying the pre-existing relationship includes analyzing a social network of the user and / or the second user.

In one aspect, the invention relates to a method implemented by a data processing device. The method comprises: obtaining a text message in a first language that includes at least one word; providing the text message to a machine translation system; and obtaining a translation of the text message from the machine translation system. Determining that the text message and the translation both include the at least one word in the first language, and that the at least one word is correctly spelled; )
Determining the frequency of occurrence of the at least one word in a previous text message; (b) determining Bayesian probabilities for adjacent words that occur before and after the at least one word; and (c) determining synonyms. Performing one or more of performing k-means clustering to identify clusters of contained words. If the frequency is above a first threshold, if the Bayesian probability is above a second threshold, and / or if the cluster includes the at least one word, the method includes: Adding the at least one word to the vocabulary.

  In one embodiment, the at least one word includes a word that is not in the vocabulary or is a word that is not in the vocabulary. The at least one word may be a new chat term word or may include the new chat term word. The method may include determining whether the vocabulary in the data store includes the at least one word. The text message may be received from a player's client device in an online game. In various examples, the vocabulary includes or consists of words in a vocabulary of the first language.

  Determining Bayesian probabilities includes: (i) considering a previous use of the at least one word in a previous text message; and (ii) appearing before and after the at least one word in the previous text message. Identifying the word, if any, may be included. The Bayesian probabilities may provide an indication of the likelihood that the adjacent word will appear before and after the at least one word in the text message.

  In various implementations, identifying the cluster includes reviewing a previous text message and, in a similar context, identifying a word that is used as the at least one word in the text message. . The method may further include analyzing the syntax and semantics of the text message to determine the parts of speech present in the text message.

  In another aspect, the invention relates to a system that includes a computer readable medium having stored instructions, and a data processing device. The data processing device obtains a text message in a first language including at least one word, provides the text message to a machine translation system, and obtains a translation of the text message from the machine translation system. Determining that the text message and the translation both include the at least one word in the first language, and that the at least one word is correctly spelled; (A) determining a frequency with which the at least one word appears in a previous text message; (b) determining a Bayesian probability for adjacent words that occur before and after the at least one word; and (c). K-means cluster to identify clusters of words containing synonyms Configured to execute the instructions to perform operations including and performing one or more of performing a ring. If the frequency is above a first threshold, if the Bayesian probability is above a second threshold, and / or if the cluster includes the at least one word, the method comprises: Adding the at least one word to the vocabulary.

In one embodiment, the at least one word includes a word that is not in the vocabulary or is a word that is not in the vocabulary. The at least one word may be a new chat term word or may include the new chat term word. The operations may include determining whether the vocabulary in the data store includes the at least one word. The text message may be received from a player's client device in an online game. In various examples, the vocabulary includes or consists of words in a vocabulary of the first language.

  Determining Bayesian probabilities includes: (i) considering a previous use of the at least one word in a previous text message; and (ii) appearing before and after the at least one word in the previous text message. Identifying the word, if any, may be included. The Bayesian probabilities may provide an indication of the likelihood that the adjacent word will appear before and after the at least one word in the text message.

  In various implementations, identifying the cluster includes reviewing a previous text message and, in a similar context, identifying a word that is used as the at least one word in the text message. . The operations may further include analyzing the syntax and semantics of the text message to determine the parts of speech present in the text message.

  In another aspect, the invention relates to a computer program product stored on one or more storage media for controlling a processing mode of a data processing device. The computer program product may be configured such that the data processing device obtains a text message in a first language including at least one word; provides the text message to a machine translation system; Obtaining a translation of the text message, that the text message and the translation both include the at least one word of the first language, and that the at least one word is correctly spelled; (A) determining the frequency with which the at least one word appears in the previous text message; and (b) determining the Bayesian probabilities for adjacent words that occur before and after the at least one word. And (c) words containing synonyms Executable by said data processing apparatus to perform operations comprising and performing one or more of the performing the identifying so k-means clustering clusters. If the frequency is above a first threshold, if the Bayesian probability is above a second threshold, and / or if the cluster includes the at least one word, the method comprises: Adding the at least one word to the vocabulary.

  In one embodiment, the at least one word includes a word that is not in the vocabulary or is a word that is not in the vocabulary. The at least one word may be a new chat term word or may include the new chat term word. The operations may include determining whether the vocabulary in the data store includes the at least one word. The text message may be received from a player's client device in an online game. In various examples, the vocabulary includes or consists of words in a vocabulary of the first language.

  Determining Bayesian probabilities includes: (i) considering a previous use of the at least one word in a previous text message; and (ii) appearing before and after the at least one word in the previous text message. Identifying the word, if any, may be included. The Bayesian probabilities may provide an indication of the likelihood that the adjacent word will appear before and after the at least one word in the text message.

In various implementations, identifying the cluster includes reviewing a previous text message and, in a similar context, identifying a word that is used as the at least one word in the text message. . The operations may further include analyzing the syntax and semantics of the text message to determine the parts of speech present in the text message.

  In one aspect, the invention relates to a method implemented by a data processing device. The method includes: (a) receiving a request from a multi-user chat session to review a portion of a history of text messages, wherein the history includes a plurality of text messages, each text message being in a respective language. And originating from each chat session participant, the method further comprises: (b) (i) selecting a different respective text message from the portion of the text message history; and (ii) selecting Performing a plurality of parallel processes each including translating the translated text message into a target language; and (c) providing a translated text message from the plurality of parallel processes to a client device of a user. , (D) different parts of the history of the text message The includes receiving a request to consider, and repeating the step (e) (b) and for the different parts of the history of the text message (c).

  In an embodiment, selecting the different respective text messages comprises querying a storage device for the portion of the history of the text messages. Translating the selected text message may include converting at least a portion of the text message from chat terms to plain language. In some implementations, the method includes receiving a request from the user to stop browsing the history of the text message. The plurality of parallel processes may include one process for each respective chat session participant. Alternatively or additionally, the plurality of parallel processes may include one process for each language used in the respective chat session.

  In another aspect, the invention relates to a system that includes a computer readable medium having stored instructions, and a data processing device. The data processing device is configured to execute instructions to perform an operation including: (a) receiving a request from a multi-user chat session to review a portion of a history of a text message, wherein the history comprises a plurality of history messages. A text message, wherein each text message is in a respective language and originated from a respective chat session participant, and wherein the data processing device further comprises: (b) (i) a history of the text message. Performing a plurality of parallel processes, each including selecting a different respective text message from the portion; (ii) translating the selected text message into a target language; and (c) translating the translated text message into a target language. Text messages from the above multiple parallel processes Providing to the device; (d) receiving a request to review different portions of the text message history; and (e) performing steps (b) and (c) for the different portions of the text message history. It is configured to execute the instruction to perform an operation including repeating.

  In an embodiment, selecting the different respective text messages comprises querying a storage device for the portion of the history of the text messages. Translating the selected text message may include converting at least a portion of the text message from chat terms to plain language. In some implementations, the act includes receiving a request from the user to stop browsing the history of the text message. The plurality of parallel processes may include one process for each respective chat session participant. Alternatively or additionally, the plurality of parallel processes may include one process for each language used in the respective chat session.

  In another aspect, the invention relates to a computer program product stored on one or more storage media for controlling a processing mode of a data processing device. The computer program product is executable by the data processing device to perform operations including: (a) receiving a request from a multi-user chat session to review a portion of a history of text messages; The history includes a plurality of text messages, each text message being in a respective language, and originating from each chat session participant, wherein the computer program product further comprises: b) a plurality of parallel processes each including: (i) selecting a different respective text message from the portion of the text message history; and (ii) translating the selected text message into a target language. Do this (C) providing a translated text message from the plurality of parallel processes to a user's client device; (d) receiving a request to review different portions of the text message history; ) The data processing device may be operable to perform operations including repeating steps (b) and (c) for the different portions of the text message history.

  In an embodiment, selecting the different respective text messages comprises querying a storage device for the portion of the history of the text messages. Translating the selected text message may include converting at least a portion of the text message from chat terms to plain language. In some implementations, the act includes receiving a request from the user to stop browsing the history of the text message. The plurality of parallel processes may include one process for each respective chat session participant. Alternatively or additionally, the plurality of parallel processes may include one process for each language used in the respective chat session.

  In one aspect, the invention relates to a method implemented by a data processing device. The method includes providing a text message chat system to a plurality of users (eg, of an online game), and requesting a first user of the text message chat system to block a second user of the text message chat system. Receiving the request and preventing a text message from the second user from being displayed to the first user after receiving the request.

  In one embodiment, the method blocks, after receiving the request, a future invitation from the second user to the first user to join a chat session using the text message chat system. It further includes. The method may include receiving a second request from the first user to unblock the second user. In some cases, the method includes, after receiving the second request, allowing a text message from the second user to be displayed to the first user. The method allows, after receiving the second request, a future invitation to be sent from the second user to the first user to join a chat session using the text message chat system. Can include:

In some embodiments, the plurality of users includes or defines an alliance in the online game. The method may include translating at least a portion of a text message in the text message chat system from a first language to a second language. The method may further include converting at least a portion of the text message from chat terms to plain language. In some implementations, translating and / or transforming may include or utilize a parallel process. For example, the parallel processes may include or utilize one process for each of the plurality of users of the chat system (or one process for each language used by the users).

  In another aspect, the invention relates to a system that includes a computer readable medium having stored instructions, and a data processing device. The data processing apparatus includes: providing a text message chat system to a plurality of users (eg, of an online game); and a first user of the text message chat system to block a second user of the text message chat system. Executing instructions to receive a request from the first user, and after receiving the request, preventing a text message from the second user from being displayed to the first user. It is configured to

  In one embodiment, the operation blocks, after receiving the request, a future invitation from the second user to the first user to participate in a chat session using the text message chat system. It further includes. The action may include receiving a second request from the first user to unblock the second user. In some cases, the act includes, after receiving the second request, allowing a text message from the second user to be displayed to the first user. The operation permits, after receipt of the second request, a future invitation to be sent from the second user to the first user to join a chat session using the text message chat system. May further include:

  In some embodiments, the plurality of users includes or defines an alliance in the online game. The operation may include translating at least a portion of a text message in the text message chat system from a first language to a second language. The operations may further include converting at least a portion of the text message from chat terms to plain language. In some implementations, translating and / or transforming may include or utilize a parallel process. For example, the parallel processes may include or utilize one process for each of the plurality of users of the chat system (or one process for each language used by the users).

  In another aspect, the invention relates to a computer program product stored on one or more storage media for controlling a processing mode of a data processing device. The computer program product, wherein the data processing device provides a text message chat system to a plurality of users (eg, of an online game) and the text message chat system blocks a second user of the text message chat system. Act of receiving a request from a first user of the system and preventing, after receiving the request, a text message from the second user from being displayed to the first user. Can be executed by the data processing device.

  In one embodiment, the operation blocks, after receiving the request, a future invitation from the second user to the first user to participate in a chat session using the text message chat system. Including. The action may include receiving a second request from the first user to unblock the second user. In some cases, after receiving the second request, the operation includes allowing a text message from the second user to be displayed to the first user. After receiving the second request, the operation permits a future invitation to be sent from the second user to the first user to participate in a chat session using the text message chat system. Can include:

  In some embodiments, the plurality of users includes or defines an alliance in the online game. The operation may include translating at least a portion of a text message in the text message chat system from a first language to a second language. The operations may further include converting at least a portion of the text message from chat terms to plain language. In some implementations, translating and / or transforming may include or utilize a parallel process. For example, the parallel processes may include or utilize one process for each of the plurality of users of the chat system (or one process for each language used by the users).

  Elements of the embodiments described for a given aspect of the invention may be used in various embodiments of another aspect of the invention. For example, it is conceivable. Features of the dependent claims dependent on one independent claim may be used in a device and / or method of any of the other independent claims.

  Other features and aspects of the various embodiments will become apparent from the following detailed description when read in connection with the accompanying drawings. The accompanying drawings illustrate, by way of example, features of such embodiments.

  Various embodiments are described in detail with reference to the following figures. The drawings are provided for illustrative purposes only and depict only some embodiments. These drawings should not be deemed to limit the width, scope, or applicability of the embodiments.

FIG. 1 is a block diagram illustrating a typical environment utilizing a multilingual communication system according to various embodiments. FIG. 1 is a block diagram illustrating an exemplary communication conversion and translation system according to various embodiments. FIG. 3 is a block diagram illustrating an exemplary conversion module according to various embodiments. FIG. 1 is a block diagram illustrating an exemplary chat client system according to various embodiments. 4 is a flowchart illustrating an exemplary method of multilingual communication according to various embodiments. 5 is a flowchart illustrating an exemplary method of converting communication according to various embodiments. FIG. 2 illustrates a typical multilingual chat session between chat client systems according to various embodiments. 4 is a flowchart illustrating the operation of an exemplary multilingual communication method according to various embodiments. 4 is a flowchart illustrating the operation of an exemplary multilingual communication method according to various embodiments. 4 is a flowchart illustrating the operation of an exemplary multilingual communication method according to various embodiments. 4 is a flowchart illustrating the operation of an exemplary multilingual communication method according to various embodiments. FIG. 2 is a block diagram illustrating an exemplary digital device that may be utilized in accordance with implementations of various embodiments. FIG. 2 is a block diagram illustrating an exemplary user feedback system according to various embodiments. FIG. 3 is a block diagram illustrating an exemplary user feedback client system according to various embodiments. 5 is a flowchart illustrating an exemplary method for user feedback, according to various embodiments. FIG. 2 is a block diagram illustrating an exemplary data flow for a user feedback system, according to various embodiments. FIG. 9 illustrates a screen shot illustrating an example of receiving user feedback on a word, according to various embodiments. FIG. 7 illustrates a screen shot illustrating an example of skipping user feedback, according to various embodiments. FIG. 4 illustrates a screen shot illustrating an example of receiving user feedback on a phrase, according to various embodiments. FIG. 9 illustrates a screen shot illustrating an example of receiving user feedback through a list of select form responses, according to various embodiments. FIG. 9 illustrates a screen shot illustrating an example of creating a list of select form responses, according to various embodiments. FIG. 9 illustrates a screen shot illustrating an exemplary incentive notification according to various embodiments. FIG. 7 is a screen shot illustrating an example where translation fails between client chat systems, according to various embodiments. FIG. 4 illustrates a screen shot illustrating a typical list of words or phrases available for user feedback, according to various embodiments. FIG. 4 illustrates a screen shot illustrating a typical list of words or phrases available for user feedback, according to various embodiments. FIG. 4 illustrates a screen shot illustrating an example of defining a word, according to various embodiments. FIG. 9 illustrates a screen shot illustrating an exemplary list of select form responses, according to various embodiments. FIG. 9 illustrates a screen shot illustrating an exemplary list of status for a submitted response, according to various embodiments. FIG. 9 illustrates a screen shot illustrating a typical incentive notification, according to various embodiments. 5 is a flowchart of a method for detecting unreadable phrases in a language corpus, according to various embodiments. FIG. 3 is a schematic diagram of a fraud detection module, according to various embodiments. 4 is a flowchart of a method for detecting fraud in incentived translations, according to various embodiments. 1 is a schematic diagram of a system for performing audio-to-text transcription and translation in a group chat environment, according to various embodiments. FIG. 4 is a schematic diagram of a chat history module for converting and / or translating a chat history, according to various embodiments. 5 is a flowchart of a method for converting and / or translating a chat history, according to various embodiments. FIG. 9 includes a screenshot of a user interface for blocking one or more users of a chat session, according to various embodiments. FIG. 8 includes a screenshot of a user interface for unblocking one or more users of a chat session, according to various embodiments. FIG. 4 includes a flowchart of a method for blocking one or more users in a chat session, according to various embodiments. FIG. 7 includes a screenshot of a user interface that allows a user to modify a language translation of an original message, according to various embodiments. FIG. 7 includes a screenshot of a user interface that allows a user to modify a language translation of an original message, according to various embodiments. FIG. 7 includes a screenshot of a user interface that allows a user to modify a language translation of an original message, according to various embodiments. FIG. 7 includes a screenshot of a user interface that allows a user to modify a language translation of an original message, according to various embodiments. FIG. 7 includes a screenshot of a user interface that allows a user to modify a language translation of an original message, according to various embodiments. FIG. 7 includes a screenshot of a user interface that allows a user to modify a language translation of an original message, according to various embodiments. FIG. 9 includes a screenshot of a user interface that allows a user to determine translation corrections submitted by other users in exchange for possible rewards, according to various embodiments. FIG. 9 includes a screenshot of a user interface that allows a user to determine translation corrections submitted by other users in exchange for possible rewards, according to various embodiments. FIG. 4 is a schematic diagram of a translation accuracy module for assessing translation accuracy, according to various embodiments. 5 is a flowchart of a method for assessing translation accuracy according to various embodiments.

DETAILED DESCRIPTION The various embodiments described herein relate to and facilitate multilingual communication. The systems and methods of some embodiments include, for example, Internet-based chat (eg, Apple® iMessage®, Windows® Live Messenger, etc.), e-mail (eg, embedded forum messaging, Yahoo® Trademark) mail, RFC 5322, etc.), text-based cellular communications (eg, SMS or MMS messages), online forum postings (eg, web-based hobby forum postings), online social media service postings (eg, Twitter) (Registered trademark), Facebook (registered trademark), etc.). Certain embodiments may also be used to translate past transcripts of communications or conversations (eg, stack transcripts or chat histories). Various embodiments include specialized / domain-related terminology (eg, chat terms), abbreviations, acronyms, proper nouns, common nouns, contractions, colloquial words or phrases, and abusive text in text. Communication systems and methods for translating text between two or more languages (e.g., in spoken language) while addressing / corresponding to one or more of the typical words or phrases. For example, some systems and methods described herein are utilized in connection with chat systems, such as those used in massively multiplayer online (MMO) games, which tend to have users chatting in different foreign languages. Can be done. In some embodiments, chat dialogs between two or more users can be translated in a manner that is transparent to intervention and presented to each user in their respective native or selected languages. In addition, due to the use of a multi-layer / multi-module conversion process, certain embodiments are faster between two or more users (eg, in their respective native languages) than are otherwise possible with conventional translation systems alone. 2.) It can facilitate translation of communications (eg, translation in a few microseconds).

  According to some embodiments, the system or method may perform a translation from a chat term in a first language, such as English, to a chat term in a second language, such as French. In another example, a system or method may include converting a chat term in a first language (eg, English) into a formal language in a first language (eg, English) before attempting to translate it into a second language (eg, French). Can be converted to a proper language. Some embodiments include data stores (eg, translations) that may include translations entered manually by a human operator, or translations based on those previously performed by a translation system (eg, historical translations performed by one embodiment). Such a text translation may be achieved by first querying the cache. Embodiments convert one or more portions of the text (eg, chat terms, acronyms, abbreviations, acronyms, proper nouns, common nouns, spoken words, and base words in the text) to be more suitable for accurate text translation. Processing one or more of the abusive words). For example, an embodiment may translate a given text to account for (current or past) idiom usage across different languages. Embodiments may again attempt to query the data store after conversion of a portion of the text. If this translation lookup retry fails, the embodiment uses a machine translation service (eg, a third-party cloud-based translation service such as Google® translation) to provide the text (which may have been translated). May try to translate

  Embodiments may convert a single translated text into chat terms in a new language (e.g., converting a formal French phrase into a French chat term) to translate the final generated text. Further improvements can be attempted. Accordingly, certain embodiments provide a chat between chat terms in a first language (eg, English) and chat terms in a second language (eg, Russian, French, Spanish, Chinese, Hindi, etc.). Facilitate translation.

  Some embodiments may help to reduce or avoid the need to use machine translation (thus reducing the time, costs, and other overhead associated with machine translation) and have minimal context Or may facilitate accurate translation of text including short sentence structures. If machine translation is facilitated by a third-party service or by a secure network connection (eg, a secure socket layer [SSL] connection), the costs or overheads avoided by certain embodiments may be significant.

  As understood herein, "converting" means manipulating a first text segment in a first language to form a second text segment in a first language. . The resulting second text segment may also be referred to herein as "converted text." "Translation" will be understood to mean converting a text segment in a first language into a corresponding text segment in a second language.

  Also, as will be understood herein, a "translated translation" has been translated in accordance with the embodiments described herein (e.g., from a text of a chat term in a first language to a first translation). Means the translation (from a first language to a second language) of a text segment (which has been converted to formal text in another language). “Untranslated translation” shall mean a translation (from a first language to a second language) of a text segment before the text segment is translated according to the embodiments described herein. Will be appreciated.

  Various embodiments may implement different conversion / translation procedures, with certain procedures well suited for a particular translation application. For example, for a particular chat system application, the conversion procedure implemented is a set of conversion related modules: chat term module, acronym module, proper noun module, common noun module, spoken language module, spell check module, abbreviation module, And applying the abusive modules in the order listed. Generally, the translation / translation procedure employed will determine which translation operations are performed, when the translation operations will be performed throughout the translation process (eg, before or after machine translation), or in any order. (For example, the priority or the priority of the conversion operation) is determined. The translation / translation procedure determines what translations are pre-populated in the data store (eg, translations can be stored in translation "caches" to speed up the overall process) and when the translation cache is used for the entire translation process You can also decide what will be done. For certain embodiments, the conversion / translation procedure employed may be determined dynamically based on the state of the environment in which the embodiment is used. For example, if the burden on the user of the chat system is greater than normal, the conversion / translation procedure may switch to one that reduces the processing burden on the chat system (eg, relies more on machine translation rather than on data stores).

  FIG. 1 is a block diagram illustrating an exemplary environment 100 utilizing a multilingual system according to various embodiments. As shown in FIG. 1, a typical environment 100 includes clients 102-1 through 102-N (collectively, "clients 102"), a chat server 108, and a translation server 110, each of which is a computer. Networks 106 may be communicatively coupled to one another. According to some embodiments, the computer network 106 is implemented or facilitated using one or more local or wide area communication networks, such as the Internet, a WiFi network, a WiMax network, a private network, a public network, etc. Can be Depending on the embodiment, some or all of the communication connections with the computer network 106 may be transmitted between the various entities depicted in the exemplary environment 100 utilizing encryption (eg, Secure Sockets Layer [SSL]). Information can be secure.

Each of client 102, chat server 108 and translation server 110 may be implemented with one or more digital devices, which may be similar to the digital devices described below with reference to FIG. For example, client 102-1 may receive user input (eg, configured for user interaction) and may include one or more other clients (eg, one of clients 102-2 through 102-N). And any form of computing device that can provide a client user interface that facilitates communication with the chat server 108 via the computer network 106. Such computing devices are known to mobile phones, tablet computing devices, laptops, desktop computers, personal digital assistants, portable gaming units, wired gaming units, thin clients, set-top boxes, portable multimedia players, or others skilled in the art. Any other type of network accessible user device. Further, one or more of chat server 108 and translation server 110 may include one or more cloud-based services (eg, System-as-a-Service [SaaS], Platform-as-a-Service [PaaS] or Infrastructure-as- a-Service [IaaS]), or may consist of one or more servers that can be implemented using one or more cloud-based services.

The clients 102 may be configured to communicatively connect to a chat server 108 that provides or otherwise facilitates a chat session between the clients 102. Each of the clients 102-1 through 102-N includes a chat client system (104-1 through 104-N, respectively) that allows a user at each of the clients 102 to access a chat session via the chat server 108. obtain. In addition, depending on the embodiment, each of the chat client systems 104-1 through 104-N (hereinafter collectively referred to as "chat client systems 104") may be non-chat applications (e.g., video games) as stand-alone chat applications. , Or by a chat service accessible at the client via a web browser. Those skilled in the art will recognize that for some embodiments, the chat client systems 104 may not be heterogeneous with respect to one another and still be able to establish a chat session between them. The chat client system 104 receives chat input (eg, chat messages) from each user in a language (and corresponding character set) selected by the user (eg, based on the user's settings or preferences) and receives another input. It may be possible to send chat input to chat server 108 to be relayed to a user (eg, another user in another chat client system). Chat client system 104 receives chat output (eg, from another user in another chat client system) from chat server 108 (eg, a chat session dialog) and receives the chat output by the user (eg, user settings or It may also be possible to display in a selected language (and corresponding character set) (based on preferences).

  With the use of some embodiments, when a dialog is passed between chat client systems 104, the translation of the chat dialog may be transparent to the user. Thus, for some embodiments, all chat dialogs presented in a given chat client system 104 are independent of the user's language in other chat client systems 104 contributing to the same chat dialog, It may be a language specific to (or selected by) the user in that given chat client system 104. For example, if a user in chat client system 104-1 and a user in chat client system 104-2 are contributing to the same chat dialog (ie, involved in the same chat session), The user may have chosen to enter and receive the chat dialog in English, while the user at the chat client system 104-2 may have chosen to enter and receive the chat dialog in Russian. . Users at client systems 104-1 and 104-2 will see the same chat content, but the chat dialog will be presented in their respective selected language.

  As shown, chat server 108 includes a chat host system 112 configured to establish and / or facilitate a chat session between chat client systems 104, and various systems and methods described herein. And a communication translation and translation (CTT) system 114 configured to perform the translation and / or translation operations in accordance with For some embodiments, chat client systems 104 may establish chat sessions with each other via chat host system 112. Chat host system 104 may utilize the capabilities of CTT system 114 in facilitating transparent translation of chat dialogs between chat client systems 104. One skilled in the art will recognize that for some embodiments, the chat host system 112 and the CTT system 114 may be part of separate servers, and the entity operating the chat host system 112 may be different from the entity operating the CTT system 114. It will be appreciated that they may be different. For example, chat host system 112 may be a third party chat host system that uses the services of CTT system 114.

  As also shown, translation server 110 may include a translation module 116 configured to receive a machine text translation request and respond to the request. According to some embodiments, CTT system 114 may utilize the operations / services of translation module 116 in performing machine translation of text. CTT system 114 may obtain access to services provided by translation module 116 using one or more translation application programming interfaces (APIs). Depending on the embodiment, the translation module 116 (and the server 110 to which it belongs) may be operated by a third party, such as Google, which may provide machine translation services for free or for a fee. Although the translation module 116 is shown as a component running on a separate server from the CTT system 114, those skilled in the art will appreciate that for some embodiments, the translation module 116 runs on the same server as the CTT system 114. And / or may be an integrated component of the CTT system 114.

  FIG. 2 is a block diagram illustrating an exemplary communication conversion and translation system 114 according to various embodiments. As shown, CTT system 114 includes communication conversion and translation (CTT) control module 202, communication conversion and translation (CTT) communication module 204, language module 206, conversion module 208, translation data store 210, and translation application programming. An interface (API) module 212 may be included. CTT control module 202 controls and / or adjusts the performance of various operations within CTT system 114 as CTT system 114 performs conversion or translation operations in accordance with certain systems and methods described herein. It can be configured as follows. For some embodiments, CTT control module 202 controls the operation of other components of CTT system 114, such as CTT communication module 204, language module 206, conversion module 208, translation data store 210, and translation API module 212. obtain.

  CTT communication module 204 may be configured to facilitate communication between CTT system 114 and systems and components outside CTT system 114, such as chat server 108 and / or translation server 110. Thus, via the CTT communication module 204, the CTT system 114 may receive a chat dialog (including one or more chat messages) to be translated or translated by the CTT system 114, and may be translated by the CTT system 114. Output a chat dialog.

  The language module 206 may be configured to identify one or more languages used in connection with the chat dialog received by the CTT system 114. For some embodiments, the language module 206 identifies the language by parsing the content of the received chat dialog and / or each chat client system (eg, chat client system) involved in the received chat dialog. 104), language preference / setting information can be obtained.

  The conversion module 208 is configured to perform a conversion operation on a chat dialog (including one or more chat messages) received by the CTT system 114 according to some systems and methods described herein. obtain. According to some embodiments, the conversion operations performed by the conversion module 208 may include, but are not limited to, those related to chat terms, acronyms, abbreviations, proper nouns, common nouns, spoken words, and abusive words. Additional details of the transform module 208 are described in FIG.

The translation data store 210 stores the pre-translated text, and then, when the CTT system 114 performs the translated and untranslated translations in accordance with certain systems and methods described herein. CT pre-translated text
It may be configured to provide to the T system 114. As described herein, the translation data store 210 may operate as a cache for translations previously performed by the CTT system 114 and / or by a human operator (eg, via a translation training system). ) The translations entered and stored manually may be stored. For some embodiments, translations that will speed up the performance of CTT system 114 for certain chat contexts may be populated in translation data store 210. For example, if the CTT system 114 is utilized with a chat system associated with an MMO game, translations (translated and unconverted) specifically associated with the MMO game are stored in the translation data store 210 (eg, the CTT system). (By the operator at 114). For some embodiments, the multi-layer / multi-module approach to transforming text used by the transform module 208 is particularly well-suited for dealing with chat text in MMO games, which can be complex in nature.

Depending on the embodiment, the data store 210 stores the untranslated translation (eg, <official English> “you” → <official French> “vous”) and the translated translation (eg, <English chat terms>). u "→ <official French>" vous ") or both.
For some embodiments, translation data store 210 may store translations such that corresponding chat messages can be identified using a hash value / tag. For example, to store the Spanish translation for the original message in English, the Spanish translation may be stored based on the hash value of the English message, which translates the Spanish translation into the hash value of the English message. To retrieve and retrieve it later. One skilled in the art will recognize that for some embodiments, the translation data store 210 may include a separate data store for translation between two specific languages. Thus, when a chat message is being translated / translated between English and French, the corresponding data English-French data store may be utilized for operations associated with the translation data store 210.

  Translation API module 212 may be configured to provide CTT system 114 with access to machine translation services provided outside of CTT system 114 (eg, by translation module 116 of translation server 110). As described herein, when there is no translation in translation data store 210, translation API module 212 may be utilized by CTT system 114.

  FIG. 3 is a block diagram illustrating an exemplary conversion module 208 according to various embodiments. As shown, conversion module 208 includes chat term module 302, acronym module 304, proper noun module 306, common noun module 308, spoken language module 310, spell check module 312, abbreviation module 314, and / or abusive language A module 316 may be included. According to some embodiments, during operation, the conversion module 208 may process the chat message in whole or in part (eg, split the message into tokens or logical parts, and then process those tokens / parts). Do). In some embodiments, various modules of conversion module 208 may be called in parallel.

Chat term module 302 may be configured to identify one or more words or phrases in a chat message associated with a chat terminology (ie, chat term), and for the identified words or phrases, It may be configured to suggest a replacement (eg, not a corresponding formal / ie chat term) word or phrase. In some embodiments, chat term module 302 may identify identified chat term words or phrases such that they are skipped or otherwise ignored during subsequent machine translation (eg, by translation module 116). May be flagged. In addition, in some embodiments, the identified chat term words or phrases may be flagged for later review and action by a human operator (eg, an administrator of the CTT system 114). In order to identify chat term words or phrases and / or their corresponding (formal) words or phrases, some embodiments include chat term words or phrases and / or chat term words or phrases and their (E.g., stored on a data store) that includes mappings to the corresponding words and phrases. The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. For example, chat term module 302 may employ statistical machine translation in its functionality. For some embodiments, the statistical machine translation employed employs a bilingual text extracted from a transformation that protects contextual information and / or adds grammar to otherwise non-grammatical sentences, and And / or may be trained using phrase-level pairs. The results from the chat term module 302 may be the words or phrases of the chat term flagged by the chat term module 302 to be ignored, suggested substitutions, or by the chat term module 302 (eg, the identified chat term It may include words or phrases that are not chat terms inserted into the message (instead of words or phrases). Depending on the embodiment, the message resulting from chat term module 302 may be provided to another transform module (in transform module 208) for further processing, or the message transformed by chat term module 302 The suggested replacement may be provided to CTT control module 202 to determine if is in data store 210.

  The acronym module 304 may be configured to identify one or more acronyms in the chat message, and further suggest a replacement word or phrase (eg, represented by the acronym) that corresponds to the acronym. It may be configured as follows. In some embodiments, the acronym module 304 flags the identified acronyms so that they are skipped or otherwise ignored during subsequent machine translation (eg, by the translation module 116). obtain. Additionally, in some embodiments, the identified acronym may be flagged for later review and action by a human operator (eg, the administrator of the CTT system 114). To identify acronyms and / or their corresponding words or phrases, some embodiments include acronyms and / or a mapping of acronyms to their corresponding words and phrases (eg, on a data store). The data set (stored in the data set) is available. The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. The result from the acronym module 304 is an acronym flagged by the acronym module 304 to be ignored, a suggested replacement, or a message (eg, instead of the identified acronym) by the acronym module 304. May include words or phrases inserted into the. Depending on the embodiment, the message resulting from the acronym module 304 may be provided to another translation module (in the translation module 208) for further processing, or the message translated by the acronym module 304 The suggested replacement may be provided to the CTT control module 202 to determine if is in the data store 210.

Proper noun module 306 may be configured to identify one or more proper nouns in the chat message, and further propose a replacement word or phrase (e.g., represented by a proper noun) that corresponds to the proper noun. It may be configured as follows. In some embodiments, proper noun module 306 flags the identified proper noun so that it is skipped or otherwise ignored during subsequent machine translation (eg, by translation module 116). obtain. Additionally, in some embodiments, a human operator (eg, CTT
The identified proper noun may be flagged for later review and action by the system 114 administrator). In order to identify proper nouns and / or their corresponding words or phrases, some embodiments use proper nouns (eg, well-known proper nouns or common names of individuals, such as Disneyland®), and / or proper nouns. A data set (e.g., stored on a data store) that includes a mapping between the data and their corresponding words and phrases may be utilized. The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. The result from the proper noun module 306 may be a proper noun flagged by the proper noun module 306 to be ignored, a suggested substitution, or a message (eg, instead of an identified proper noun) by the proper noun module 306. May include words or phrases inserted into the. Depending on the embodiment, the message resulting from proper noun module 306 may be provided to another conversion module (in conversion module 208) for further processing, or the message converted by proper noun module 306 The suggested replacement may be provided to CTT control module 202 to determine if is in data store 210.

  The common noun module 308 may be configured to identify one or more common nouns in the chat message and further suggest a replacement word or phrase (e.g., represented by a common noun) that corresponds to the common noun. It may be configured as follows. In some embodiments, the common noun module 308 flags identified common nouns to be skipped or otherwise ignored during subsequent machine translation (eg, by the translation module 116). obtain. Additionally, in some embodiments, identified common nouns may be flagged for later review and action by a human operator (eg, the administrator of CTT system 114). To identify common nouns and / or their corresponding words or phrases, some embodiments include common nouns and / or a mapping of common nouns to their corresponding words and phrases (eg, on a data store). The data set (stored in the data set) is available. The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. The result from the common noun module 308 may be a common noun flagged by the common noun module 308 to be ignored, a suggested substitution, or a message (eg, instead of the identified common noun) by the common noun module 308. May include words or phrases inserted into the. Depending on the embodiment, the message resulting from the common noun module 308 may be provided to another conversion module (in the conversion module 208) for further processing, or the message converted by the common noun module 308 The suggested replacement may be provided to the CTT control module 202 to determine if is in the data store 210.

The colloquial module 310 may be configured to identify one or more colloquial words or phrases in the chat message, and may further substitute for the identified words or phrases (eg, corresponding formal / ie It may be configured to suggest (non-colloquial) words or phrases. In some embodiments, the spoken language module 310 may add the identified spoken words or phrases to be skipped or otherwise ignored during subsequent machine translation (eg, by the translation module 116). Can flag. Additionally, in some embodiments, the identified spoken words or phrases may be flagged for later review and action by a human operator (eg, an administrator of CTT system 114). To identify colloquial words or phrases and / or their corresponding (formal) words or phrases, some embodiments include colloquial words or phrases and / or colloquial words or phrases and (E.g., stored on a data store) that includes mappings to the corresponding words and phrases. The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. The results from the colloquial module 310 may be colloquial words or phrases flagged by the colloquial module 310 to be ignored, suggested substitutions, or by the colloquial module 310 (eg, the identified colloquial words or phrases). (Instead of a non-spoken word or phrase) inserted into the message. Depending on the embodiment, the message resulting from the spoken module 310 may be provided to another conversion module (in the conversion module 208) for further processing, or the message converted by the spoken module 310 may be stored in a data store. The proposed replacement may be provided to CTT control module 202 to determine if it is within 210.

  The spell check module 312 may be configured to identify one or more misspelled words or phrases in the chat message, and further replace (eg, corrected) words for the identified words or phrases. Or it may be configured to suggest phrases. For example, the spell check module 312 may be configured to automatically correct a word or phrase to a suggested replacement word or phrase. In some embodiments, the spell check module 312 may include the identified misspelled word or phrase so that it is skipped or otherwise ignored during subsequent machine translation (eg, by the translation module 116). May be flagged. Additionally, in some embodiments, identified misspelled words or phrases may be flagged for later review and action by a human operator (eg, an administrator of CTT system 114). In order to identify a misspelled word or phrase and / or its corresponding (corrected) word or phrase, some embodiments may include a misspelled word or phrase and / or a misspelled word or phrase. A data set (eg, stored on a data store) that includes mappings to their corresponding words and phrases may be utilized. The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. The results from the spell check module 312 may be used by a misspelled word or phrase flagged by the spell check module 312 to be ignored, a suggested replacement, or by the spell check module 312 (eg, a misspelled word or phrase). (Instead of) may include the corrected word or phrase inserted in the message. Depending on the embodiment, the message resulting from the spell checking module 312 may be provided to another conversion module (in the conversion module 208) for further processing, or the message converted by the spell checking module 312 may be The suggested replacement may be provided to CTT control module 202 to determine if it is in data store 210.

The abbreviation module 314 may be configured to identify one or more abbreviations in the chat message and is further configured to suggest a replacement word or phrase (eg, represented by an abbreviation) that corresponds to the abbreviation. You may. In some embodiments, the abbreviation module 314 may flag identified abbreviations to be skipped or otherwise ignored during subsequent machine translation (eg, by the translation module 116). In addition, in some embodiments, the identified abbreviations may be flagged for later review and action by a human operator (eg, an administrator of the CTT system 114). To identify abbreviations and / or their corresponding words or phrases, some embodiments include abbreviations and / or mappings of abbreviations to their corresponding words and phrases (e.g., stored on a data store. D) Data sets may be used. The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. The result from the abbreviation module 314 may be an abbreviation flagged by the abbreviation module 314 to be ignored, a suggested substitution, or a word inserted into the message by the abbreviation module 314 (eg, instead of the identified abbreviation). Or may contain phrases. Depending on the embodiment, the message resulting from the abbreviation module 314 may be provided to another translation module (in the translation module 208) for further processing, or the message translated by the abbreviation module 314 may be stored in a data store. The proposed replacement may be provided to CTT control module 202 to determine if it is within 210.

  The abusive module 316 may be configured to identify one or more abusive words or phrases (hereinafter referred to as “abuses”) in the chat message, and may further correspond to the abusive words. May be configured to suggest a replacement word or phrase (eg, a suitable alternative) (eg, a toned down euphemism). In some embodiments, the abusive language module 316 flags the identified abusive language to be skipped or otherwise ignored during subsequent machine translation (eg, by the translation module 116). Can be established. In addition, in some embodiments, the identified abusive words may be flagged for later review and treatment by a human operator (eg, the administrator of the CTT system 114). To identify abusive words and / or their corresponding words or phrases, some embodiments include mapping abusive words and / or abbreviations to their corresponding words and phrases (eg, data store). (Stored above). The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. The results from the abusive module 316 may be output by the abusive, suggested replacement, or abusive modules flagged by the abusive module 316 to be ignored (eg, the identified abusive (Instead of a word or phrase) inserted into the message. Depending on the embodiment, the message resulting from the abusive module 316 may be provided to another conversion module (in the conversion module 208) for further processing, or may be converted by the abusive module 316. A suggested replacement may be provided to CTT control module 202 to determine if the message is in data store 210.

For some embodiments, one or more of the various conversion modules 208 may be configured to insert one or more characters before and / or after the flagged portion to convert one or more characters in the chat message. Flags can be set for these parts. For example, if the chat term module 302 flags the word “LOL” in a portion of a chat message, the chat term module 302 inserts certain characters (“_”) before and / or after the word (eg, "_LOL_"), which may indicate that the flagged portion should be ignored by translation module 116.

For some embodiments, the conversion module 208 may perform two or more conversion operations on the initial message in parallel, and in response, each of the two or more conversion operations provides a separate response. From there, conversion module 208 may then select one converted message for further processing (eg, for use in operation 514). Depending on the embodiment, each response may include a flagged text portion, a suggested replacement, or a word or phrase inserted into the initial message. Thereafter, the selected transformed message may follow the selection priority. The selection priority can determine which transformed messages are selected for further processing according to which precedent. In some embodiments, the priority selection also depends on which translation operation is most likely to generate a translated message suitable for subsequent lookup or subsequent machine translation in the translation data store 210). Good. In addition, in some embodiments, the priority of selection may be according to which conversion operation produces the most formal converted message. The selection priority may depend on the conversion / translation procedure selected by the embodiment.

Table 1 below shows an example of how the conversion module 208 may process a portion of a chat message according to various embodiments. As shown, conversion module 208 may process the chat message based on tokens or proximal tokens, and may stop processing for a particular token once the conversion has been performed.

Table 1. Example Chat Message Processing FIG. 4 is a block diagram illustrating an exemplary chat client system 104 according to various embodiments. As shown, chat client system 104 may include a chat client controller 402, a chat client communication module 404, and a chat client graphical user interface (GUI) module 406. The chat client control module 402 controls and / or adjusts the performance of various operations within the chat client system 104 when the chat client system 104 performs a chat-related operation (eg, a communication chat dialog with the chat server 108). May be configured. For some embodiments, chat client control module 402 may control the operation of other components of chat client system 104, including, for example, chat client communication module 404 and chat client GUI module 406.

  Chat client communication module 404 may be configured to facilitate communication between chat client system 104 and systems and components outside chat client system 104, such as chat server 108. Accordingly, the chat client module 404 may cause the chat client system 104 to receive a chat dialog to be presented in the chat client system 104 from the chat server 108 (eg, via the chat client GUI module 406). At 104 (eg, via the chat client GUI module 406), the chat dialog received from the user may be sent to the chat server.

  Chat client GUI module 406 may be configured to provide graphical input / output access to a chat session with another chat client system to a user at chat client system 104. Thus, for some embodiments, the chat client GUI module 406 may present a client GUI adapted to receive user interactions by the client 102 to a user at the client 102. For some embodiments, the chat client GUI module 406 may provide chat dialogs (eg, as they are received from the chat server 108) to the user in their selected language (eg, according to the user's language preferences / settings). It can be configured to present. Additionally, the chat client GUI module 406 may be configured to receive chat input from the user in a language of the user's selection (eg, according to the user's language preferences / settings). As described herein, the language used to present and receive chat dialogs in chat client system 104 is different from the language used to present and receive chat dialogs in another chat client system. May be. The chat client GUI module 406 will be further described with reference to FIG.

  FIG. 5 is a flowchart illustrating an exemplary method 500 for multilingual communication according to various embodiments. As described below, for some embodiments, the method illustrated by the method 500 includes the chat client system 104-1, the chat client system 104-2, the CTT system 114 (eg, of the chat server 108), and (eg, translation). Operations related to translation module 116 (of server 110) may be performed.

The method 500 may begin at operation 502, where the language module 204 (of the CTT system 114) communicates with a first language used by a user in a first chat client system (eg, 104-1) and a second chat. A second language being used by the user at the client system (e.g., 104-2) may be identified. According to some embodiments, language module 204 may identify a first language and a second language by obtaining language preferences / settings from respective chat client systems 104.

  At operation 504, the CTT communication module 204 (of the CTT system 114) may receive an initial message in a first language. In some embodiments, CTT communication module 204 may receive an initial message from chat host system 112, and chat host system 112 may have received the initial message from a chat client system (eg, 104-1). Is also good.

  At operation 506, the CTT control module 202 (of the CTT system 114) may query the translation data store 210 for a corresponding message in the second language corresponding to the initial message. At operation 508, CTT control module 202 may determine whether a corresponding message is found in translation data store 210. If one exists, at operation 510, the CTT communication module 204 may assist in sending a corresponding message to a second chat client system (eg, chat client system 104-2). In some embodiments, a corresponding message may be sent to chat host system 112. Chat host system 112 may relay the corresponding message to a second chat client system (eg, 104-2). The method 500 may then end.

  If the corresponding message does not exist in the translation data store 210, at operation 512, the conversion module 208 may attempt to convert at least a portion of the initial message into a converted message in the first language. As described herein, messages resulting from the conversion module 208 may be converted or remain intact (e.g., when the conversion operation of the conversion module 208 is not applied to the initial message). Good. For some embodiments, the translation module 208 may perform more than one translation operation on the initial message in parallel, and in response, each of the two or more translation operations provides a separate response. From there, conversion module 208 may then select one converted message for further processing (eg, for use in operation 514). Depending on the embodiment, each response may include a flagged text portion, a suggested replacement, or a word or phrase inserted into the initial message. Thereafter, the selected transformed message may follow the selection priority. The selection priority can determine which transformed messages are selected for further processing according to which precedent. In some embodiments, the priority selection is based on which conversion operation is most likely to generate a translated message suitable for subsequent lookup or subsequent machine translation in the translation data store 210). Good. In addition, in some embodiments, the priority of selection may be according to which conversion operation produces the most formal converted message. The selection priority may depend on the conversion / translation procedure selected by the embodiment.

At operation 514, assuming that the transformation module 208 has transformed the message, the CTT control module 202 (of the CTT system 114) translates the translated data store 210 for the corresponding message in the second language corresponding to the transformed message. Can be queried. At operation 516, CTT control module 202 may determine whether a corresponding message is found in translation data store 210. If one exists, at operation 518, the CTT communication module 204 may assist in sending a corresponding message to a second chat client system (eg, chat client system 104-2). In some embodiments, a corresponding message may be sent to chat host system 112. Chat host system 112 may then relay the corresponding message to a second chat client system (eg, 104-2). Then method 50
0 may end.

  For some embodiments, if the corresponding message is not already present in translation data store 210, then in operation 520, CTT control module 202 causes any of conversion modules 208 to be performed on chat messages that have not yet been performed. It may be determined whether there is an additional conversion operation.

  If there are additional conversion operations, method 500 returns to operation 512 to perform additional conversion operations. Depending on the embodiment, the additional conversion operation may include applying a different conversion operation to that already performed by the conversion module 208 on the initial message, or the same conversion operation performed may be performed in English. May apply to different parts of a chat message, or some combination thereof. For example, during the first execution of operation 512, if conversion module 208 applies an action associated with the chat term to the initial message (to create a first converted message), then a second execution of operation 512 During, the conversion module 208 may apply an action associated with the abbreviation to the second converted message. Following the subsequent execution of operation 512, method 500 may proceed to operations 514 and 516, where CTT control module 202 causes the corresponding translated message in the second language to correspond to the latest updated message. The translation data store 210 may be queried again for the message to be sent, and the CTT control module 202 may determine whether the corresponding message is found in the translation data store 210. By iteratively performing the conversion and query operations in this manner, an embodiment may be able to find a corresponding message before performing all available conversion operations. One skilled in the art will appreciate that for certain embodiments, the conversion and query operations may be performed sequentially, such that the query operation (eg, operation 514) includes all available conversion operations provided by the conversion module 208 in the chat message. It will be appreciated that it will only be done after it has been done for the

  If the corresponding message is not present in the translation data store 210 and there are no additional translation operations (of the translation module 208), at operation 522 (via the translation API module 212), the translation module 116 It may assist in translating the converted message into a corresponding message in a second language. Thereafter, at operation 524, the CTT communication module 204 may assist in transmitting a corresponding message to a second chat client system (eg, chat client system 104-2). According to some embodiments, a corresponding message may be sent to the chat host system 112. Chat host system 112 may then relay the corresponding message to a second chat client system (eg, 104-2). The method 500 may then end.

  For certain embodiments, the conversion module 208 may be utilized to convert the corresponding message to a second language before the corresponding message is sent to the chat host system 112. As described herein, a further conversion process may be performed on the corresponding message to further improve the translation for the user at the second chat client system (eg, 104-2). For example, if the initial message includes chat terms in a first language (eg, English), the additional conversion process may add chat terms in a second language to the extent possible.

Although the steps of the above method may be represented and described in some order, those skilled in the art will recognize that the order in which the steps are performed may vary between embodiments. In addition, those skilled in the art will recognize that the components described above with respect to method 500 are merely examples of components used in the method, and for some embodiments, other components may be utilized in some embodiments. Will do.

  FIG. 6 is a flowchart illustrating an exemplary method 600 for converting communications according to various embodiments. As described below, for some embodiments, the method 600 may perform operations associated with the transform module 208 (eg, of the CTT system 114).

  The method may begin at operation 602, where an initial message is received by the conversion module 208 for a conversion process. According to some embodiments, the conversion module 208 may fail to identify the message in the translation data store 210, and possibly after a third party or proprietary translation process (eg, as a cloud-based service). The initial message may be received for translation before the initial message is machine translated by translation module 116), which may be provided. As described herein, especially text is minimal context, concise sentence structure, specialized / domain related terminology (eg, chat terms for Internet-based chat), abbreviations, acronyms, The conversion module 208 is used in various embodiments to facilitate or otherwise enhance text translation when including colloquial, proper noun, common noun, abusive, or some combination thereof. obtain. Text translations that may take advantage of the operation of the conversion module 208 include, but are not limited to, conversations (eg, transcripts), offline or online Internet-based chats (eg, instant messaging), and cell phone messaging (eg, SMS or MMS). It may include translation of the resulting text.

  At operation 604, chat term module 302 may identify one or more words or phrases in the initial message associated with the chat terminology (ie, chat terms), and may further identify the words or phrases for the identified words or phrases. May be suggested (eg, not the corresponding formal / ie chat term). In some embodiments, chat term module 302 may identify identified chat term words or phrases such that they are skipped or otherwise ignored during subsequent machine translation (eg, by translation module 116). May be flagged. Additionally, in some embodiments, identified chat term words or phrases may be flagged for later review and action by a human operator (eg, an administrator of CTT system 114). In order to identify chat term words or phrases and / or their corresponding (formal) words or phrases, some embodiments may use chat term words or phrases and / or chat term words or phrases and their (E.g., stored on a data store) that includes mappings to the corresponding words and phrases. The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. Messages resulting from act 604 (hereinafter “first intermediate messages”) may be words or phrases of a chat term flagged by chat term module 302 to be ignored, suggested substitutions, or chat term module 302 (Eg, instead of the identified chat term words or phrases) may include words or phrases that are not chat terms inserted into the initial message. In some examples, the first intermediate message may be the same as the initial message (eg, when no replacement is performed by chat term module 302). Depending on the embodiment, the first intermediate message resulting from the chat term module 302 may be provided to another transform module (in the transform module 208) for further processing, or may be transformed by the chat term module 302. A suggested replacement may be provided to CTT control module 202 to determine whether the provided message is in data store 210. Following operation 604, a first intermediate message may be provided to a next operation of conversion module 208 for processing (eg, operation 606).

  At operation 606, the acronym module 304 may identify one or more acronyms in the chat message and further suggest a replacement word or phrase (e.g., represented by an acronym) corresponding to the acronym. You may. In some embodiments, the acronym module 304 flags the identified acronyms so that they are skipped or otherwise ignored during subsequent machine translation (eg, by the translation module 116). obtain. Additionally, in some embodiments, the identified acronym may be flagged for later review and action by a human operator (eg, the administrator of the CTT system 114). To identify acronyms and / or their corresponding words or phrases, some embodiments include acronyms and / or a mapping of acronyms to their corresponding words and phrases (eg, on a data store). The data set (stored in the data set) is available. The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. The message resulting from act 606 (hereinafter the “second intermediate message”) may be identified (eg, identified) by an acronym, suggested replacement, or acronym module 304 flagged by the acronym module 304 to be ignored. (Instead of quoted acronyms) may include words or phrases inserted into the message. In some examples, the second intermediate message may be the same as the first intermediate message (eg, when no replacement is performed by the acronym module 304). Depending on the embodiment, the second intermediate message resulting from the acronym module 304 may be provided to another conversion module (in the conversion module 208) for further processing, or may be converted by the acronym module 304 A suggested replacement may be provided to CTT control module 202 to determine whether the provided message is in data store 210. Following operation 606, a second intermediate message may be provided to a next operation of conversion module 208 for processing (eg, operation 608).

At operation 608, the proper noun module 306 may identify one or more proper nouns in the chat message and further propose a replacement word or phrase (e.g., represented by a proper noun) corresponding to the proper noun. You may. In some embodiments, proper noun module 306 flags the identified proper noun so that it is skipped or otherwise ignored during subsequent machine translation (eg, by translation module 116). obtain. Additionally, in some embodiments, identified proper nouns may be flagged for later review and action by a human operator (eg, the administrator of CTT system 114). In order to identify proper nouns and / or their corresponding words or phrases, some embodiments use proper nouns (eg, well-known proper nouns or common names of individuals, such as Disneyland®), and / or proper nouns. A data set (e.g., stored on a data store) that includes a mapping between the data and their corresponding words and phrases may be utilized. The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. The message resulting from act 608 (hereinafter the “third intermediate message”) may be identified by proper noun, suggested replacement, or proper noun module 306 flagged by proper noun module 306 to be ignored (eg, identifying (Instead of written proper nouns) may include words or phrases inserted into the message. In some examples, the third intermediate message may be the same as the second intermediate message (eg, when no replacement is performed by proper noun module 306). Depending on the embodiment, the third intermediate message resulting from proper noun module 306 may be provided to another transform module (in transform module 208) for further processing, or by proper noun module 306. A suggested replacement may be provided to the CTT control module 202 to determine whether the transformed message is in the data store 210. Following operation 608, a third intermediate message may be provided to a next operation of conversion module 208 for processing (eg, operation 610).

  At operation 610, the common noun module 308 may identify one or more common nouns in the chat message and further suggest a replacement word or phrase corresponding to the common noun (eg, represented by a common noun). You may. In some embodiments, the common noun module 308 flags identified common nouns to be skipped or otherwise ignored during subsequent machine translation (eg, by the translation module 116). obtain. Additionally, in some embodiments, identified common nouns may be flagged for later review and action by a human operator (eg, the administrator of CTT system 114). To identify common nouns and / or their corresponding words or phrases, some embodiments include common nouns and / or a mapping of common nouns to their corresponding words and phrases (eg, on a data store). The data set (stored in the data set) is available. The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. The message resulting from act 610 (hereinafter the “fourth intermediate message”) may be a common noun flagged by common noun module 308 to be ignored, a suggested replacement, or by common noun module 308 (eg, identifying (Instead of written common nouns). In some examples, the fourth intermediate message may be the same as the third intermediate message (eg, when no replacement is performed by common noun module 308). Depending on the embodiment, the fourth intermediate message resulting from the common noun module 308 may be provided to another conversion module (within the conversion module 208) for further processing, or by the common noun module 308. A suggested replacement may be provided to the CTT control module 202 to determine whether the transformed message is in the data store 210. Following operation 610, a fourth intermediate message may be provided to a next operation of conversion module 208 for processing (eg, operation 612).

At operation 612, the spoken language module 310 may identify one or more spoken words or phrases in the chat message, and further substitute for the identified words or phrases (eg, corresponding formal / ie A (non-colloquial) word or phrase may be suggested. In some embodiments, the spoken language module 310 may add the identified spoken words or phrases to be skipped or otherwise ignored during subsequent machine translation (eg, by the translation module 116). Can flag. Additionally, in some embodiments, the identified spoken words or phrases may be flagged for later review and action by a human operator (eg, an administrator of CTT system 114). To identify colloquial words or phrases and / or their corresponding (formal) words or phrases, some embodiments include colloquial words or phrases and / or colloquial words or phrases and (E.g., stored on a data store) that includes mappings to the corresponding words and phrases. The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. The message resulting from act 612 (hereinafter the “fifth intermediate message”) may be colloquial words or phrases flagged by colloquial module 310 to be ignored, suggested substitutions, or colloquial module 310 ( Non-spoken words or phrases inserted into the message (eg, instead of identified spoken words or phrases) may be included. In some examples, the fifth intermediate message may be the same as the fourth intermediate message (eg, when no replacement is performed by colloquial noun module 310). Depending on the embodiment, the fifth intermediate message resulting from colloquial module 310 may be provided to another translator module (in translator module 208) for further processing, or may be translated by colloquial module 310 A suggested replacement may be provided to CTT control module 202 to determine if the message is in data store 210. Following operation 612, a fifth intermediate message may be provided to a next operation of conversion module 208 for processing (eg, operation 614).

  At operation 614, the spell check module 312 may identify one or more misspelled words or phrases in the chat message, and further substitute (eg, corrected) words for the identified words or phrases. Or you may suggest a phrase. For example, the spell check module 312 may automatically correct a word or phrase with a suggested replacement word or phrase. In some embodiments, the spell check module 312 may include the identified misspelled word or phrase so that it is skipped or otherwise ignored during subsequent machine translation (eg, by the translation module 116). May be flagged. Additionally, in some embodiments, identified misspelled words or phrases may be flagged for later review and action by a human operator (eg, an administrator of CTT system 114). In order to identify a misspelled word or phrase and / or its corresponding (corrected) word or phrase, some embodiments may include a misspelled word or phrase and / or a misspelled word or phrase. A data set (eg, stored on a data store) that includes mappings to their corresponding words and phrases may be utilized. The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. Messages resulting from act 614 (hereinafter “sixth intermediate message”) are misspelled words or phrases flagged by spell check module 312 to be ignored, suggested replacements, or spell check module 312. (E.g., in place of a misspelled word or phrase). In some examples, the sixth intermediate message may be the same as the fifth intermediate message (eg, when no replacement is performed by spell check module 312). Depending on the embodiment, the sixth intermediate message resulting from spell checking module 312 may be provided to another conversion module (in conversion module 208) for further processing, or by spell checking module 312. A suggested replacement may be provided to the CTT control module 202 to determine whether the transformed message is in the data store 210. Following operation 614, a sixth intermediate message may be provided to a next operation of conversion module 208 for processing (eg, operation 616).

At operation 616, the abbreviation module 314 may identify one or more abbreviations in the chat message and may suggest a replacement word or phrase corresponding to the abbreviation (eg, represented by an abbreviation). In some embodiments, the abbreviation module 314 may flag identified abbreviations to be skipped or otherwise ignored during subsequent machine translation (eg, by the translation module 116). In addition, in some embodiments, the identified abbreviations may be flagged for later review and action by a human operator (eg, an administrator of the CTT system 114). To identify abbreviations and / or their corresponding words or phrases, some embodiments include abbreviations and / or mappings of abbreviations to their corresponding words and phrases (e.g., stored on a data store. D) Data sets may be used. The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. The message resulting from act 616 (hereinafter the “seventh intermediate message”) may be an abbreviation flagged by the abbreviation module 314 to be ignored, a suggested replacement, or by the abbreviation module 314 (eg, the identified abbreviation). (Instead of a word or phrase) inserted into the message. In some examples, the seventh intermediate message may be the same as the sixth intermediate message (eg, when no replacement is performed by the abbreviation module 314). Depending on the embodiment, the seventh intermediate message resulting from the abbreviation module 314 may be provided to another translation module (in the translation module 208) for further processing, or may have been translated by the abbreviation module 314. A suggested replacement may be provided to CTT control module 202 to determine if the message is in data store 210. Following operation 616, a seventh intermediate message may be provided to a next operation of conversion module 208 for processing (eg, operation 618).

  At act 618, the abusive module 316 may identify one or more abusive words or phrases in the chat message (hereinafter, referred to as “abuses”) and further respond to the abusive words. A replacement word or phrase (e.g., a suitable alternative) (e.g., a toned euphemism) may be suggested. In some embodiments, the abusive language module 316 flags the identified abusive language to be skipped or otherwise ignored during subsequent machine translation (eg, by the translation module 116). Can be established. In addition, in some embodiments, the identified abusive words may be flagged for later review and treatment by a human operator (eg, the administrator of the CTT system 114). To identify abusive words and / or their corresponding words or phrases, some embodiments include mapping abusive words and / or abbreviations to their corresponding words and phrases (eg, data store). (Stored above). The dataset may be configured as a training or learning system, may be proprietary (eg, manually collected “in an organization” by the administrator of the CTT system 114), or may be commercially available. Or may be obtained from a publicly available Internet knowledge base. The message resulting from act 618 (hereinafter the “eighth intermediate message”) is sent by the abusive, suggested replacement, or abusive module 316 flagged by the abusive module 316 to be ignored. It may include words or phrases inserted into the message (eg, instead of identified abusive words). In some examples, the eighth intermediate message may be the same as the seventh intermediate message (eg, when no replacement is performed by the abusive language module 316). Depending on the embodiment, the eighth intermediate message resulting from the abusive module 316 may be provided to another conversion module (in the conversion module 208) for further processing, or the abusive module 316 The suggested replacement may be provided to the CTT control module 202 to determine if the message transformed by is in the data store 210. Following operation 618, an eighth intermediate message may be provided to the next operation of the conversion module 208 for processing. The method 600 may then end.

  According to some embodiments, the message ultimately resulting from the conversion module 208 (eg, an eighth intermediate message resulting from operation 618) may also be subsequently used to query the translation data store 210 for a corresponding message. Good. The corresponding message can serve as a translation for the resulting message. One skilled in the art will appreciate that in some instances, messages resulting from the conversion module 208 (eg, messages subsequently used to query the translation data store 210) have no translation applied to the initial message (e.g., If the initial message passed operations 604-618 without any transformations applied, it will be appreciated that the initial message may be the same as the initial message received (eg, in operation 602).

Those skilled in the art will also recognize that various embodiments may perform more or fewer operations than shown, may perform operations that differ from those illustrated, and may perform operations in a different order. Will. In general, the types of conversion operations performed, and the order in which they are performed, may depend on the conversion procedure employed by the embodiment. As noted herein, various embodiments may implement different conversion / translation procedures in implementing each conversion. Certain procedures are well suited for a particular translation application or translation context. The conversion / translation procedure employed may determine which conversion operations are performed, when the conversion operations are performed, and in what order the conversion operations are performed. The translation / translation procedure may determine what translations are populated in the translation data store and when the translation data store is utilized during the entire translation / translation process.

  For some embodiments, an intermediate message resulting from an operation in method 600 may have an influential and / or cascading effect on a message resulting from a subsequent operation in method 600. In addition, for some embodiments, when a chat message is processed by the method 600, each operation of the flowchart 600 may be performed on the chat message before the method ends. Alternatively, for some embodiments, the method of flowchart 600 terminates prematurely by performing only a subset of the illustrated operations (eg, after at least one operation has resulted in the conversion of a chat message). obtain. According to some embodiments, after execution of each operation of flowchart 500, a query to translation data store 210 for a corresponding message in a desired language based on the resulting updated translated message follows. obtain. If a corresponding message is identified, the method of flowchart 500 may terminate early.

  For various embodiments, method 600 may perform acts 604-612 in parallel. For example, CTT control module 202 may perform two or more operations 604-612 on the initial message in parallel and receive a separate response from each of the two or more operations. Each response may include a flagged text portion, a suggested replacement, or a word or phrase inserted into the initial message. Thereafter, the CTT control module 202 may optionally follow the priority of the selection (e.g., for further processing and according to any precedent, to determine which transformed message is selected) One of the received responses may be selected for processing (eg, querying the translation data store 210 or translating by the translation module 116).

  For example, during the method 600, the CTT control module 202 may perform an operation 604 for identifying chat term processing, an operation 610 for common noun processing, and an operation 616 for abbreviation processing for an initial message. . In response, operation 604 may return an initial message transformed for the chat term. Act 610 can return the initial message as is. Act 616 may return an initial message translated for the abbreviation. Thereafter, based on the selection priority, CTT control module 202 may select the transformed message returned from operation 616 for further processing.

  For certain embodiments, a time limit is enforced for performing various operations of method 600. The time limit may cause the conversion operation of method 600 to stop executing if no response / result is received before the time limit expires. In so doing, various embodiments may ensure that certain translation operations do not unnecessarily obstruct the entire translation / translation process.

  Although the steps of the above method may be represented and described in some order, those skilled in the art will recognize that the order in which the steps are performed may vary between embodiments. In addition, those skilled in the art will recognize that the components described above with respect to the method of flowchart 600 are merely examples of components that may be used in the method, and for some embodiments, other components may be utilized in some embodiments. You will recognize that.

  FIG. 7 is a diagram 700 illustrating a typical multilingual chat session between chat client systems 104 (eg, 104-1 and 104-2), according to various embodiments. As shown, chat client system 104-1 may include chat client GUI module 406-1. Chat client system 104-2 may include chat client GUI module 406-2. As described herein, each of the chat client GUI modules 406-1 and 406-2 provides graphical input / output access to the chat session shared between them, to the chat client system 104-1. And 104-2, respectively. For some embodiments, the chat client GUI modules 406-1 and 406-2 may present to each user a client GUI adapted to receive user interactions regarding sent and received chat dialogs.

  When a chat dialog 712 (represented by a double arrow in FIG. 7) is passed between the chat client systems 104-1 and 104-2, the chat client GUI modules 406-1 and 406-2 cause the respective chat client The chat dialog 712 may be presented in a language selected (implicitly or explicitly) by a user in the system 104-1 or 104-2. As shown, the chat client GUI module 406-1 presents a chat dialog 712 in an output area 708 in a first language (eg, English) and sends a chat input in the first language to a second area 710. May include a chat dialog box 702 configured to receive at. Chat client GUI module 406-2 is configured to present chat dialog 712 in a second language (eg, French) in output area 720 and receive chat input in the second language in second area 722. A chat dialog box 714 may be included. For some embodiments, when the chat dialog 712 is presented in the dialog boxes 702 and 714, it may include the presentation of a username (eg, a user online identifier) associated with the user entering the chat message in the chat dialog 712.

In the illustrated embodiment of FIG. 7, the language selected for chat client system 104-1 is English and the language selected for chat client system 104-2 is French. Thus, while chat messages 704 (“LOL”) and 706 (“Who laughin at?”) Are presented in English in dialog box 702 of chat client GUI module 406-1, respective chat messages 716 (“MDR”) are displayed. ") And 718 (" Quite fair rire? ") Are presented in French in a dialog box 714 of the chat client GUI module 406-2. Translation of chat messages 704, 706, 716, and 718 may be facilitated by the various systems and methods described herein. The translation of messages similar to chat messages 704, 706, 716, and 718 will be further described with reference to FIGS.

FIG. 8 is a flowchart illustrating the operation of an exemplary multilingual communication method 800 according to various embodiments. As described below, for some embodiments, the method 800 may perform operations associated with the chat client system 104-1, the chat client system 104-2, and the CTT system 114 (eg, of the chart server 108). In particular, FIG. 8 illustrates the translation of an English chat message containing the text “LOL” into a French chat message, according to some embodiments. Such a situation occurs when the language used by the user in the first chat client system 104-1 is English and the language used by the user in the second chat client system 104-2 is French. Can occur. According to some embodiments, CTT system 114 may automatically detect these language selections / preferences for chat client systems 104-1 and 104-2.

  As shown, at operation 802, the first chat client system 104-1 sends an English message for transmission (eg, via the chat host system 112) to the second chat client system 104-2. Can send. The English message may be sent to the CTT control module 202 of the CTT system 114 for translation processing.

At operation 804, CTT control module 202 may query translation data store 210 for chat messages that correspond to English chat messages ("LOL") and have been previously translated into French. In response, at operation 806, translation data store 210 may return a corresponding French message ("MDR") corresponding to the English chat message ("LOL") to CTT control module 202. afterwards,
At operation 808, CTT control module 202 may assist in transmitting a corresponding French chat message ("MDR") to second chat client system 104-2 (e.g., CTT system 114 may transmit Send the corresponding French chat message to the chat host system 112).

FIG. 9 is a flowchart illustrating the operation of an exemplary multilingual communication method 900 according to various embodiments. As described below, for some embodiments, the method illustrated by flowchart 900 includes the chat client system 104-1, the chat client system 104-2, the CTT system 114 (eg, of the chart server 108), and (eg, translation). Operations related to translation module 116 (of server 110) may be performed. In particular, FIG. 9 illustrates the translation of an English chat message containing the text “LOL” into a French equivalent chat message, according to some embodiments. Unlike the illustrated embodiment of FIG. 8, FIG. 9 illustrates the use of a translation module 208 and a translation module 116 (eg, of the CTT system 114).

  As shown, at operation 902, the first chat client system 104-1 transmits a transmission (eg, via the chat host system 112) to a second chat client system 104-2 having a French-speaking user. In order to be able to send English chat messages. The English chat message may be sent to the CTT control module 202 of the CTT system 114 for translation processing.

At operation 904, CTT control module 202 may query translation data store 210 for a French equivalent of the chat message corresponding to the English chat message ("LOL"). In response, at operation 906, translation data store 210 returns a query failure to CTT control module 202, and translation data store 210 has a corresponding French chat message for the English chat message ("LOL"). You may indicate that you have not. In such a case, at operation 908, the CTT control module 202 may send an English chat message to the conversion module 208 for the conversion process according to an embodiment. As described herein, conversion module 208 may include a plurality of conversion-related modules 932 configured to convert the chat message into a suitable message by a further translation process.

In operation 910, the chat term module 302 of the conversion module 208 converts the English chat message (“LOL”) into a converted English chat message (“laugh out”).
loud "), and the converted English chat message can be returned to the CTT control module 202 for further processing. Those skilled in the art will recognize that, for some embodiments, the English chat message may be processed by an additional module of the conversion module 208 before the converted English chat message is returned to the CTT control module 202. Would.

  At operation 912, CTT control module 202 may query translation data store 210 for a French equivalent chat message corresponding to the converted English chat message ("laugh out loud"). In response, at operation 914, translation data store 210 returns the query failure to CTT control module 202 and translates the corresponding French chat message for the translated English chat message ("laugh out loud"). It may indicate that the data store 210 does not have it. In such a case, at operation 916, the CTT control module 202 may send the translated English chat message to the translation module 116 for a machine translation process according to an embodiment.

At operation 918, translation module 116 may return a machine translated French chat message ("mort de rire") corresponding to the translated English chat message. The resulting machine translated French chat message ("mort de rire") is an example of a translated translation of the English chat term chat message ("LOL").

In operation 920, the CTT control module 202 converts the machine-translated French chat message (“mort de rire”) into the conversion module 208 for further conversion processing of the machine-translated French chat message according to an embodiment. Can be sent to As noted herein, machine translated text may be sent for further conversion processing to further improve the French translation. For example, if the original English chat message included English chat terms, additional conversion processing could add French chat terms to the extent possible. Accordingly, in operation 922, the chat term module 302 of the conversion module 208 may convert the machine-translated French chat message ("mort de rire") into the converted French chat message ("MDR"). The converted French chat message may be returned to the CTT control module 202 for further processing.

Finally, at operation 924, CTT control module 202 may assist in transmitting a corresponding French chat message ("MDR") to second chat client system 104-2 (e.g., CTT system 114 may include: Send the corresponding French chat message to chat host system 112 for transmission). In addition, at operation 926, the CTT control module 202 translates the translation mapping of the translated translation between the original English chat message (“LOL”) and the translated French chat message (“MDR”). It may be stored in data store 210. Once the mapping is stored in translation data store 210, it can be used to store translation entries to speed up future translations, for example, as illustrated in FIG. As noted herein, translation data store 210 may store mappings of translated and untranslated translations.

For some embodiments, CTT control module 202 may also store the corresponding (transformed and untranslated) translation mapping determined during operation of method 900. For some embodiments, the translation mapping includes chat messages that were not initially in the translation data store 210 (eg, the chat message shown for operation 904 and the chat message shown for operation 912), and a lookup of the translation data store 210. May be between a corresponding message determined during subsequent operations (eg, a query to the translation data store 210 that returns no results, via the translation module 208 and / or the translation module 116, and a decision after the query. Mapping between corresponding chat messages).

For example, as shown in FIG. 9, the CTT control module 202 determines whether the original English chat message (“LOL” in operation 904 and the translated English chat message in operation 912 (“laugh out loud”) (both) The CTT control module 202 will then query the translation data store 210 from the translation data store 210 for the result (in operations 906 and 914, respectively). However, at operation 916, CTT control module 202 finally sends the translated English message ("laugh out loud") to machine translation module 116 for machine translation, and in response, at operation 918 Receive a machine translated French chat message ("mort de rire"). Accordingly, in operation 928, CTT control module 202 translates the translated translation between the original English chat message ("LOL") and the machine translated French chat message ("mort de rire"). May be stored in the translation data store 210. Similarly, in operation 930, CTT control module 202 converts between the translated English chat message ("laugh out loud") and the machine translated French chat message ("mort de rire"). A translation mapping for the translation may be stored in translation data store 210. In doing so, the method 900 then proceeds with the original English chat message (“LOL”) or the converted English chat message (“laugh out”
loud ") to the translation data store 210, the translation data store 210
A corresponding translated translation will be provided.

FIG. 10 is a flowchart illustrating the operation of an exemplary multilingual communication method 1000 according to various embodiments. As described below, for some embodiments, the method 1000 includes a chat client system 104-1, a chat client system 104-2, a CTT system 114 (eg, of the chart server 108), and a translation (eg, of the translation server 110). Operations related to module 116 may be performed. In particular, FIG. 10 illustrates the translation of an English chat message containing the text “Who laughin at?” Into a French chat message, according to some embodiments.

  As shown, in operation 1002, the first chat client system 104-1 transmits an English chat message to a second chat client system 104-2 for transmission (eg, via the chat host system 112). May be sent. The English chat message may be sent to the CTT control module 202 of the CTT system 114 for translation processing.

At operation 1004, the CTT control module 202 may query the translation data store 210 for a French equivalent chat message corresponding to the English chat message (Who laughin at?). In response, at operation 1006, the translation data store 210 returns a query failure to the CTT control module 202 and translates the corresponding French chat message for the English chat message (Who u laughin at?) Into the translation data store 210. May not have any. In such a case, at operation 1008, the CTT control module 202 may send an English chat message to the conversion module 208 for the conversion process according to an embodiment. As described herein, conversion module 208 may include a plurality of conversion-related modules 1036 configured to convert the chat message into a suitable message through a further translation process.

In operation 1010, the chat term module 302 of the conversion module 208 converts the English chat message (Who laughin at?) Into a converted English chat message (Who you laughin at?), And converts the converted English chat message. The message may be passed to an additional module of the conversion module 208 for further processing, such as the spell check module 312.

As described herein, various modules of the conversion module 208, including the chat term module 302, identify one or more words or phrases in a chat message and replace for the identified words or phrases. It can be configured to suggest a word or phrase. Thus, those skilled in the art will recognize that for some embodiments, the conversion performed / suggested by the modules of the conversion module 208 may involve word-to-phrase or phrase-to-phrase conversion of the chat message. Will do. For example, in operation 1010, chat term module 302 proposes / replaces the phrase "who u" with "who are you", possibly during conversion (then the word "laughin" is replaced with "laughing"). ) To replace the English chat message ("Who u laughin at?") With the converted English chat message ("Who are you laughing at?") I can do it. In doing so, the various modules of the transformation module 208, such as the chat term module 302, may be used to improve the grammar for each transformation, possibly eliminating the need for a separate module in the transformation module 208 to implement the grammar improvement. Can be brought.

For some embodiments, before the converted English chat message is passed to an additional module of conversion module 208, chat term module 302 passes the converted English chat message to CTT control module 202 in operation 1010. obtain. Next, the CTT control module 202 checks the corresponding chat message in French corresponding to the converted English chat message (“Who you laughin at?”) (
The translation data store 210 may be queried (at operation 1012). In response, at operation 1014, the translation data store 210 returns the query failure to the CTT control module 202 and responds to the translated English chat message ("Who you laughin at?") With the corresponding French chat message. May not be included in the translation data store 210.

At operation 1016, the spell check module 312 may perform a spell check process at operation 1018 on the converted English chat message ("Who you laughin at?"). During the spell check process, the spell check module 312 may correct the converted English chat message into a corrected English chat message ("Who you laughing at?"), And the corrected English chat message May be returned to the CTT control module 202. One skilled in the art will appreciate that for some embodiments, the corrected English chat message may be processed by an additional module of the conversion module 208 before the converted English chat message is returned to the CTT control module 202. Will recognize.

At operation 1020, CTT control module 202 may query translation data store 210 for a French equivalent of the chat message corresponding to the corrected English chat message ("Who you laughing at?"). In response, at operation 1022, the translation data store 210 returns a query failure to the CTT control module 202 and returns a corresponding French chat message for the corrected English chat message ("Who you laughing at?"). May not be included in the translation data store 210. In such a case, at operation 1024, the CTT control module 202 may send the corrected English chat message to the translation module 116 for a machine translation process according to an embodiment.

At operation 1026, the translation module 116 determines whether the translated English chat message corresponds to a machine translated French chat message ("Quite teit rire?").
) Can be returned. At operation 1028, the CTT control module 202 converts the machine-translated French chat message ("Quite teit rire?") For further conversion processing of the machine-translated French chat message according to an embodiment. It can be sent to module 208.

As noted herein, the machine translated text may be sent for further conversion processing to further refine the translation of the text. For example, if the original English chat message included English chat terms, additional conversion processing could add French chat terms to the extent possible. At operation 1030, conversion module 208 may return the unmodified, machine-translated French chat message ("Quite teit rire?") To CTT control module 202 for further processing (eg, module of conversion module 208). Does not apply any changes to machine-translated French chat messages).

In operation 1032, CTT control module 202 may assist in transmitting a machine-translated French chat message (“Quite fait rire?”) To second chat client system 104-2 (eg, CTT system 114). Sends the corresponding French chat message to chat host system 112 for transmission). In addition, at operation 1034, the CTT control module 202 translates between the original English chat message ("Who laughin at?") And the translated French chat message ("Quite fait rire?"). The mapping may be stored in the translation data store 210. As described herein, in an additional operation (not shown), the CTT control module 202 converts the previously failed queries to the translation data store 210 and the corresponding messages determined after those queries. The corresponding translation mapping that is based may also be stored in translation data store 210 (eg, similar to operations 928 and 930 of FIG. 9).

  According to some embodiments, the conversion operations performed by conversion module 208 include performing certain conversion operations in parallel, and may perform certain conversion operations in succession. The order in which the conversion operations are performed in parallel and sequentially can vary between the various embodiments. If the conversion operations are performed in parallel, as described herein, some embodiments may have a selection option that determines which converted message is selected according to which precedent for further processing. Priorities may be adopted.

FIG. 11 is a flowchart illustrating the operation of an exemplary multilingual communication method 1100 according to various embodiments. As described below, for some embodiments, the method 1100 includes a chat client system 104-1, a chat client system 104-2, a CTT system 114 (eg, for the chart server 108), and a translation (eg, for the translation server 110). Operations related to module 116 may be performed. In particular, FIG. 11 illustrates the translation of an English chat message containing the text “Who laughin at?” Into a French chat message with a parallel conversion operation, according to some embodiments.

  As shown, at operation 1102, the first chat client system 104-1 transmits an English chat message to a second chat client system 104-2 for transmission (e.g., via the chat host system 112). May be sent. The English chat message may be sent to the CTT control module 202 of the CTT system 114 for translation processing.

In operation 1104, the CTT control module 202 may query the translation data store 210 for a French equivalent chat message corresponding to the English chat message (Who laughin at?). In response, in operation 1106, the translation data store 210 returns the query failure to the CTT control module 202 and translates the corresponding French chat message for the English chat message (Who u laughin at?) Into the translation data store 210. May not be provided.

In such a case, the CTT control module 202 may send an English chat message according to an embodiment to the conversion module 208 for the conversion process. As described herein, the conversion module 208 may include a plurality of conversion-related modules 1130 configured to convert the chat message into a suitable message by a further translation process. As shown in FIG. 11, during operation 1108, CTT control module 202 sends an English chat message (“Who laughin at?”) In parallel to two or more conversion related modules 1130 of conversion module 208. I can do it. In addition, during operation 1108, the CTT control module 202 may receive results from the transformation related module 1130 in parallel and send queries to the translation data store 210 based on the transformation results in parallel.

Thus, in operation 1110a, the CTT control module 202 converts the English chat message (“Who laughin at?”) Into a chat term module 3 for conversion processing.
02. In parallel, at operation 1110b, the CTT control module 202 may send an English chat message ("Who laughin at?") To the spell check module 312 for conversion processing. Thereafter, the CTT control module 202 determines in operation 1112a the first converted English chat message ("Who you laughin at?"
) May be received from chat term module 302, while in operation 1112 b, CTT control module 202 may receive a second converted English chat message (“Who laughin at?”) From spell check module 312. . Depending on the respective conversion processing time, chat term module 302, spell check module 312 and other conversion related modules 1130 may respond to CTT control module 202 sequentially or in parallel with each other.

Thereafter, in operation 1114a, the CTT control module 202 may query the translation data store 210 for a French equivalent chat message corresponding to the first translated English chat message ("Who you laughin at?"). In operation 1114b, the CTT control module 202 may query the translation data store 210 for a French equivalent chat message corresponding to the second converted English chat message ("Who laughing at?"). For some embodiments, during operations 1114a and 1114b, the CTT control module 202 may query the translation data store 210 sequentially or in parallel. In some embodiments, the timing of the query may depend on when the translation related module 1130 of the translation module 208 returns a respective response. As shown in FIG. 11, translation data store 210 may return a query failure (eg, <failure>) for the query at acts 1116a and 1116b.

Finally, the CTT control module 202 may select one transformed message from the two or more messages resulting from the parallel operation 1108 for further processing. If only one of the transformation related modules 1130 results in a transformed message, CTT control module 202 may select that particular transformed message for further processing. As noted herein, CTT control module 202 may select the transformed message based on the priority of the selection. The priority of the selection may be determined according to the conversion / translation procedure selected by the embodiment. For some embodiments, the priority of selection may be that the converted message has the most formal content, that the converted message has the most conversions, or that the converted message is suitable for machine translation. May be based on whether it is due to a conversion-related module that is known to be more likely to generate the generated message.

  Once the translated message is selected, at operation 1118, the CTT control module 202 may send the translated English chat message to the translation module 116 for a machine translation process according to an embodiment. For example, as shown in FIG. 11, the CTT control module 202 sends the first translated English chat message generated by the chat term module 302 (“Who you laughin at?”) To the translation module 116 for transmission. ").

In operation 1120, the translation module 116 determines that the misspelled word "laughin
"), But may return a machine translated French chat message (" Quite teit rire? ") Corresponding to the first converted English chat message. In operation 1122, the CTT control module 202 converts the machine-translated French chat message ("Quite teit rire?") For further conversion processing of the machine-translated French chat message according to an embodiment. It can be sent to module 208.

As noted herein, the machine translated text may be sent for further conversion processing to further refine the translation of the text. For example, if the original English chat message included English chat terms, additional conversion processing could add French chat terms to the extent possible. At operation 1124, conversion module 208 may return the unmodified, machine-translated French chat message (“Quite teit rire?”) To CTT control module 202 for further processing (eg, conversion module 208's If the module does not apply any changes to machine translated French chat messages).

At operation 1126, CTT control module 202 may assist in transmitting a machine-translated French chat message ("Quite teit rire?") To second chat client system 104-2 (e.g., CTT system 114). Sends the corresponding French chat message to the chat host system 112 for transmission). In addition, in operation 1128, CTT control module 202 translates between the original English chat message ("Who laughin at?") And the translated French chat message ("Quite fait rire?"). The mapping may be stored in the translation data store 210. As described herein, in an additional operation (not shown), the CTT control module 202 converts the previously failed queries to the translation data store 210 and the corresponding messages determined after those queries. The corresponding translation mapping that is based may also be stored in translation data store 210 (eg, similar to operations 928 and 930 of FIG. 9).

For some embodiments, the conversion operations may be performed in a hybrid series / parallel arrangement, whereby some conversion operations are performed in parallel and other conversion operations are performed sequentially. For example, as shown in FIG. 11, an English chat message (“Who laughin at?”) Is sent in parallel to the chat term module 302 and the spell check module 312 in acts 1110a and 1110b. Thereafter, when one of the resulting converted messages is selected (eg, based on a priority of selection), other conversion related modules 1130 of conversion module 208 (eg, acronym module 304, proper noun module, etc.) 306 and common noun module 308) may operate in parallel on the selected transformed message.

  FIG. 12 is a block diagram of an exemplary digital device 1200. Digital device 1200 includes a processor 1202 communicatively coupled to a bus 1214, a memory system 1204, a storage system 1206, a communication network interface 1208, an I / O interface 1210, and a display interface 1212. Processor 1202 is configured to execute executable instructions (eg, programs). In some embodiments, processor 1202 includes circuitry or any processor capable of processing executable instructions.

  Memory system 1204 is any memory configured to store data. Some examples of the memory system 1204 are storage devices such as RAM or ROM. The memory system 1204 can include a ram cache. In various embodiments, data is stored in memory system 1204. The data in memory system 1204 may be cleared or eventually moved to storage system 1206.

  Storage system 1206 is any storage device configured to retrieve and store data. Some examples of storage system 1206 are flash drives, hard drives, optical drives, and / or magnetic tapes. In some embodiments, digital device 1200 includes a memory system 1204 in the form of RAM and a storage system 1206 in the form of flash data. Both memory system 1204 and storage system 1206 include a computer-readable medium that can store instructions or programs executable by a computer processor, including processor 1202.

  A communication network interface (com. Network interface) 1208 can be coupled to a network (eg, computer network 106) via link 1216. Communication network interface 1208 may support communication over, for example, an Ethernet connection, a serial connection, a parallel connection, or an ATA connection. Communication network interface 1208 may also support wireless communication (eg, 802.11a / b / g / n, WiMax®). It will be apparent to those skilled in the art that communication network interface 1208 can support many wired and wireless standards.

  Optional input / output (I / O) interface 1210 is any device that receives input from a user and outputs data. Optional display interface 1212 is any device configured to output graphics and data to a display. In one example, display interface 1212 is a graphics adapter.

  It will be appreciated by those skilled in the art that the hardware components of the digital device 1200 are not limited to those depicted in FIG. Digital device 1200 may include more or fewer hardware elements than those shown. Further, hardware elements share functionality and may nevertheless be within the various embodiments described herein. In one example, encoding and / or decoding may be performed by processor 1202 and / or a coprocessor located on a GPU (ie, Nvidia®).

The above functions and components may be configured with instructions stored on a recording medium such as a computer-readable medium. Instructions can be retrieved and executed by a processor. Some examples of instructions are software, program code, and firmware. Some examples of storage media are memory devices, tapes, disks, integrated circuits, and servers. The instructions are operable, when executed by the processor, to direct the processor to operate consistent with some embodiments. Those skilled in the art are familiar with instructions, processors, and storage media.

  Various embodiments are described herein by way of example. It will be apparent to those skilled in the art that various modifications may be made and other embodiments may be used without departing from the broader scope of the invention presented herein. These and other variations on the exemplary embodiments are intended to be encompassed by the present invention.

  FIG. 13 is a block diagram illustrating an exemplary user feedback system according to various embodiments. For illustrative purposes, in FIG. 13, an exemplary system for user feedback is shown as part of an exemplary communication conversion and translation (CTT) system 1300 according to various embodiments. For example, a typical user feedback system is part of a chat translation system employed by various multilingual chat systems, including game chat systems available in connection with MMO games (eg, in-game chat systems). obtain. Through the use of a typical user feedback system, the various players of an MMO game can use a flagged word or word, possibly in exchange for in-game currency / credit / items, as an incentive for feedback / approved feedback. For phrases, one may choose to provide user feedback through, for example, an in-game chat system. For some embodiments, those skilled in the art will appreciate that some or all of a typical user feedback system and its associated components may be separate and / or independent of a typical communication conversion and translation system 1300. Will recognize.

  It is understood that, as used herein, "feedback" refers to "user feedback" or "user feedback response", possibly in response to a query requesting feedback on a translation or translation. Should be. Further, it should be understood that the user feedback may include user feedback about the translation or user feedback about the translation. User feedback may include a definition for a given word or phrase. The definition allows (i) a given word or phrase to be converted from a given word or phrase in the first language to a corresponding word or phrase in the (same) first language, ii) allows a given word or phrase to be converted from a given word or phrase in a first language to a corresponding word or phrase in a (different) second language, thereby Performing a translated translation and / or (iii) allowing a given word or phrase to be translated from a first language to a second language, thereby untranslated translation of the word or phrase Execute As described herein, a user feedback response may be received in response to a query requesting user feedback in connection with a translation failure.

  As used herein, "translation failure" or "translation failure" refers to the actual failure to translate the text (e.g., the translated text is identical to the original text) or the potential Text translations that are flagged as containing translation failures (e.g., reading a translated text, and if the translated text is questionable or the translated text is an incorrect translation) It is understood to include text translations flagged by the user who believes to include them.

As shown in FIG. 13, the CTT system 1300 includes a communication conversion and translation (CTT) control module 1302, a communication conversion and translation (CTT) communication module 1304, a language module 1306, a conversion module 1308, and a translation data store. 1310 and a translation application programming interface (API) module 1312. In some embodiments, the CTT control module 1302, the CTT communication module 1304, the language module 1306, the conversion module 1308, the translation data store 1310, and the translation API module 1312 are included in the CTT system 114, as described herein. It can be similar to a component.

  As further shown in FIG. 13, the CTT system 1300 includes a translation failure management module 1314, a user feedback (UF) query generation module 1316, a user feedback (UF) query audience selection module 1318, and a query / response value module. 1320, a query application program interface (API) module 1322, a response evaluation module 1324, a translation data store update module 1326, an audience suitability evaluation module 1328, and an incentive reward module 1330.

  Translation failure management module 1314 may be configured to facilitate managing translation failures (eg, failure to translate text from a first language to a second language). For some embodiments, the translation failure management module 1314 may cause the system 1300 to use the second language during the chat session from the first user using the first language during the chat session. CTT system 1300 may be adapted to handle one or more translation failures that it may experience in attempting to translate a chat message sent to a second user. As described herein, the text translated by CTT system 1300 may be a chat message, which may include chat terms, abbreviations, spoken words, idioms, and the like. During the conversion or translation of a chat message, some or all of the chat messages may fail to translate because the CTT system 1300 translates and / or translates the portions of the chat message that cause those failures. It will be appreciated that it may not be possible.

  The functions provided by the translation failure management module 1314 include automatically detecting when a text translation failure has occurred, and automatically detecting when a word or phrase in the text cannot be translated. Allowing the user to flag some or all of the "translated" text as containing actual or potential translation errors after the translation process, and various audience members Managing which translation failures are selected for user feedback from (eg, chat members, such as players using an in-game chat system), and which words or phrases associated with translation failures are associated with user feedback (eg, How often in the text to be translated Managing whether the word is selected (based on whether it appears / used) and when the translation failure is no longer a candidate for selection for user feedback (eg, if the word or phrase associated with the translation failure is Management from the selected pool).

In various embodiments, the translation failure management module 1314 may manage the overall availability of a word or phrase for user feedback by audience members. As used herein, "user feedback opportunities" may include words or phrases available for user feedback by audience members. The translation failure management module 1314 may determine one or more based on various conditions, such as availability of computing resources for user feedback, incentives provided in connection with user feedback, quality of feedback responses received in the past, and the like. Of the user feedback opportunities may be suppressed or suspended. For example, continuing to provide incentives (eg, in-game currency or in-game items) for an approved user feedback response may have a negative impact on the economy (eg, in-game economy) associated with the incentive. It may be determined that. Thus, some embodiments may (temporarily) stop the availability of user feedback opportunities that provide incentives, or the current incentive strategy is detrimental to the target economy (eg, the in-game economy) And those embodiments detect, the incentive (eg, amount or type of incentive) provided for the user feedback opportunity may be adjusted.

  As described herein, the audience member for whom user feedback is sought is, in some cases, a member who has volunteered to provide such feedback as a participant in a user feedback program and / or has submitted. A member may volunteer to provide such feedback in return for an incentive if the feedback is approved (eg, as being correct, acceptable, or useful in defining a transformation or translation). For some embodiments, the user feedback systems described herein provide an opportunity for audience members to provide user feedback opportunities for incentives (eg, in-game currency or in-game items) that are potentially useful in the gaming system. It may be part of a gaming system, such as an online MMO game, which is a game player who chooses to participate in another manner. In this way, the user feedback system may be presented as a game feature that game audience members consider to be a "game" that is "played" for reward, thereby leveraging the competition inherent in the gaming environment. . Much more about the selection of audience members is described with respect to the UF query audience selection module 1318.

  As described herein, words or phrases that may prevent translation of the communicated text or cause inaccurate translation may be flagged by the communications system user. For example, in a multilingual multi-user chat system associated with an online game, a game player may flag all chat messages received as potentially presenting translation problems or failures, or some portion of the chat messages. (E.g., the word or phrase). Words or phrases associated with translation failures include, for example, specialized / domain-related terminology, abbreviations, acronyms, proper nouns, common nouns, contractions, colloquial words or phrases, and abusive words Or may contain phrases. In addition, a word or phrase may be flagged by a system or method that automatically detects that the word or phrase is untranslatable and in some cases prevents translation of a larger phrase or sentence.

  Translation management module 1314 may be responsible for selecting a word or phrase as being the subject of user feedback, where the selected word or phrase includes the actual translation of the text from the first language to the second language. Or, it may be flagged as being associated with a potential failure. Translation management module 1314 may select a word or phrase based on a number of factors. Some of these factors include the current importance of the word or phrase in translation (eg, importance based on the overall usage of the word or phrase) and the complexity of the word or phrase (eg, the word or phrase Difficulties or how long the word or phrase has been a problem), the aptitude of the user who has chosen or volunteered to provide user feedback (eg, the user has a second language aptitude), and provides user feedback And preferences of the user who has selected / willed to do so. One of skill in the art will appreciate that other factors for selecting a word or phrase for user feedback may be utilized by various embodiments.

For various embodiments, a user selected to provide user feedback may be a user who volunteers to provide such feedback. For example, setting a user preference to be associated, selecting a system ad in a chat for which user feedback is desired, and / or providing a section listing one or more user feedback opportunities available for the selection. By browsing, a user may choose to provide user feedback.

  As shown herein, translation failure management 1314 manages when a particular word or phrase associated with an actual or potential translation failure is no longer a candidate for selection by an audience member for user feedback. obtain. Various conditions may cause the translation failure manager 1314 to determine when a particular number of feedback responses have been received, for example, in connection with a word or phrase, or in connection with an associated translation failure, Determining when a certain number of consistent feedback responses have been received in connection with the phrase or associated translation failure, and determining whether a given feedback response is a valid response for the requested user feedback And determining when it has been approved (eg, a particular response from the set of unique feedback responses has been manually approved by the administrator as the correct response).

  The UF query generation module 1316 may be configured to generate, for a given word or phrase selected for user feedback, a query for obtaining user feedback from a selected audience member. As described herein, the selected audience members may, in some cases, improve the translation of the problem when the submitted user feedback is approved as correct, accepted as an accepted feedback response, or May be a user who has volunteered to provide user feedback in return for an incentive when approved to be useful. Much more about the selection of audience members is described with respect to the UF query audience selection module 1318.

  For some embodiments, the generated query may include one or more ways to receive a query response from the selected audience member. For example, the generated query is a predefined response (also referred to herein as a "select-form response") that an audience member can select as a response to the generated query. A list may be included as a response method. In another example, the generated query is a response to the generated query, wherein one or more fields configured to receive the text value entered into the field by the audience member (as referred to herein as a "freeform response"). (Also referred to as "free-form response"). Other response methods may include graphical user interface (GUI) elements, text values, or some combination thereof.

  For example, one or more response methods included in the generated query may follow many factors. Many such factors include, for example, audience member preferences, the importance and / or complexity of the word or phrase for which user feedback is sought, and the number of feedback responses ever received for the word or phrase for which user feedback is sought. And the number of consistent feedback responses received so far for the word or phrase for which user feedback is sought, and whether there are enough freeform feedback responses to create a selection form response. For some embodiments, the audience members responding to the generated query may be selected from more than one response method for the generated query.

Further, the language in which the audience members are provided with user feedback opportunities may follow a number of factors. The factors include, for example, whether the audience member is a multilingual speaker by monitoring the audience member's chat history, whether the audience member's linguistic ability meets or exceeds a certain level of linguistic certainty, and whether the audience member Language settings on the user device (eg, default language settings for the audience member's device) that are being used to participate in user feedback opportunities. For example, if the default device language of the audience member's device is German and the audience member chooses to participate in user feedback opportunities, the queries generated and sent to the audience member will be based on German. (Eg, a query to define an English word or phrase to a German word or phrase, or vice versa). For some embodiments, the generated query may provide the audience member with an option to select the language of the submitted response (eg, if the generated query does not specify or require a particular language). And / or may provide the audience member with the option of providing two or more responses, each response being in a possibly different language. For some such embodiments, the audience members are given a variety of language factors as discussed herein, such as whether the audience members are multilingual speakers and the language settings on the audience members' user devices. May be provided with a language response option.

  The UF query audience selection module 1318 may be configured to select audience members for which user feedback is desired. In some embodiments, the UF referral audience selection module 1318 may optionally include a pool of audience members who volunteered to provide user feedback in return for incentives (which may be received when user feedback is approved / accepted). Audience members can be selected. As described herein, an audience member may volunteer to participate in a translation, at which time the audience member may be selected by the UF inquiry audience selection module 1318 to request user feedback. Pool. In some embodiments, when the UF query audience selection module 1318 selects an audience member, a query generated by the UF query generation module 1316 may be provided to the audience member to obtain user feedback. The query generated by the UF query generation module 1316 may be provided to the audience member selected by the UF query audience selection module 1318 as part of a list of user feedback available for participation by the selected audience member. Once the generated query is provided to the selected audience member, the selected audience member can provide the generated query to the audience member (eg, via one or more response methods associated with the generated query). The selected audience member may initiate a user feedback session that can provide one or more responses (eg, via one or more response methods included with the generated query).

Depending on the embodiment, if a word or phrase is selected for user feedback before the audience member is selected, the selected word or phrase (e.g., if the selected word or phrase is an audience member's eligibility or Audience members may be selected from a set of candidate audience members based on whether they match a reference). Alternatively, if an audience member is selected before a word or selection is selected for user feedback, the selected word or phrase may be based on the selected audience member. The eligibility of a given audience member is determined by one or more feedback responses previously provided by that given audience member (e.g., in connection with previous queries generated and provided to the given audience member). And / or may be determined based on one or more linguistic abilities of the audience members. Audience members may be assessed, for example, as having the requisite fitness to provide user feedback on one or more words or phrases associated with the translation failure if they achieve a certain number of approved feedback responses. In another example, the audience member may include, at least, feedback responses previously submitted for the particular language, audience member preferences, and / or information associated with the audience member indicating fluency in the particular language. Based on that, it can be assessed as having an aptitude in a particular language. In yet another example, an audience member who has submitted a feedback response previously evaluated as incorrect or incorrect (eg, a meaningless or forged response) may be determined to have low aptitude.

  In some embodiments, once a given audience member has provided a response for a word or phrase (and the response is optionally approved), the given audience member will be able to respond to the user for the same word or phrase. It need not be reselected to provide feedback. Certain embodiments may make an exception to this restriction if a given audience member provides a response in a language different from the language of the response previously submitted for the same word or phrase.

  Where incentives are provided in return for user feedback (eg, upon approval of a feedback response), various embodiments include a time-based quota (eg, hourly, daily, monthly for submitting a feedback response). Audience members may be selected based on a quota, such as a per-year, per-year limit) or an incentive acquisition quota (eg, an in-game currency or item limit given to incentive-based user feedback). The quota may be associated with individual audience members, groups of audience members, or some combination thereof. For some embodiments, the UF query audience selection module 1318 may include computing resource availability for user feedback, incentives provided in connection with user feedback, quality of feedback responses received in the past, and the like. May reduce or stop the availability of user feedback opportunities to one or more audience members.

  The query / response (QR) bidding module 1320 is configured to determine the value or significance of the query or query response based on the words or phrases for which the query or query response seeks user feedback. obtain. Examples of factors that may be considered by various embodiments in determining the value of a query or query response include word or phrase complexity (e.g., higher complexity has a higher value), word or phrase to the translation / translation process. The importance of the phrase (e.g., the more important the higher the value), the response method used by the query or query response (e.g., a higher value for the free form response method than the selection form response method), word or phrase (Eg, a chat term, abbreviation, or colloquial) or one or more languages included (eg, receive a French response, a query that includes an English word or phrase, receives an English response, Has a higher value than queries containing English words or phrases). In some embodiments, the incentive provided for a given user feedback may be based on a value associated with the query or query response. In some embodiments, the QR bidding module 1320 determines a value for a query or query response based on the validity of a previous query or query response when seeking useful, useful, or accurate user feedback. It can be configured to Thus, for some embodiments, the value may be dynamically adjusted based on the latest validity of that achieved by a previous query or query response. In some embodiments, the QR quoting module 1320 is configured to allow an administrator (eg, of the CTT system 1300) or another authorized user to manually assign a value to a query and / or query response in connection with one or more user feedback. May be assigned or adjusted.

The query API module 1322 may be configured to facilitate sending a user feedback query to a selected audience member and / or receiving a query response from the selected audience member. In some embodiments,
The query API 1322 provides a list of user feedback opportunities available to the selected audience member, provides a query regarding one or more user feedback opportunities selected by the selected audience member, and provides a query by the selected audience member. Receiving a response for the selected one or more user feedback opportunities and / or the current status of the inquiry response previously submitted for approval (eg, approved, rejected, or pending / waiting approval) Status) to selected audience members. For some embodiments, the query API 1322 may include one or more associated with a given audience member from a chat client system, possibly used by a given audience member, to interact with the CTT system 1300. Can be obtained. As set forth herein, the preferences associated with an audience member may determine the selection of a word or phrase for which user feedback is sought from the audience member and / or for user feedback for a given word or phrase. It may be determined whether the audience member is selected to receive the inquiry.

  The response evaluation module 1324 may be configured to evaluate a query response submitted in connection with the query generated to obtain user feedback. For some embodiments, response evaluation module 1324 may evaluate the query response in one or more phases.

  During the activation phase, response evaluation module 1324 treats one or more unique responses submitted for a given generated query as being approved or rejected for a given generated query. obtain. In some embodiments, the unique response submitted in connection with the query generated to obtain user feedback confirms that the unique response is valid for the word or phrase associated with the generated query. Until it is done, or until it is rejected as invalid for the word or phrase associated with the generated query. As used herein, a given “unique response” may include a set of responses that are similar in syntax but not necessarily identical (eg, different punctuation or spacing). A given unique response may be received by more than one response method. The unique response is also referred to herein as a "unique query response" and a "unique user feedback response." Depending on the embodiment, more than one unique query response may be approved for a given generated query. For example, more than one unique response may be accepted as defining a given word or phrase with respect to the user feedback being sought. The response can be manually validated, for example, by an administrator reviewing and treating the response (eg, only a unique response in some cases). Additionally or alternatively, the responses may be validated by an automated process that may approve and / or reject the submitted responses based on their counts and / or thresholds associated with the response counts. In some embodiments, the auto-validation process may filter the top submitted responses for a given word or phrase and provide those top submitted responses to an administrator or the like for review and treatment I can do it. The activation phase may be based on a response evaluation module 1324 periodically (eg, based on a schedule) or based on conditions (eg, when the number of submitted responses pending approval meets or exceeds a threshold). Can be performed by

In some embodiments, if multiple unique user feedback responses (e.g., definitions) are provided for a given word or phrase, the response evaluation module 1324 may automatically (and / or manually administer the input). ), The top-ranked unique user feedback response of the plurality of unique user feedback responses (eg, the most common response) may be accepted, but the top-ranked unique user feedback response is statistically significant. This can only be done if For example, if a first unique user feedback response was received 26 times for a given word and a second unique user feedback response was received 24 times for the same given word, then a first unique user feedback response would be received. Even though the user feedback response is a top ranked unique user feedback response for a given word, it cannot be acknowledged as a response for the given word. This is because 26 times are not statistically significant over 24 times. For example, the first unique user feedback may not be selected until the first unique user feedback response remains the top ranked unique user feedback response and the response count reaches 40.

Given that phrases are not identical but are similar in nature and can convey the same intent, in some embodiments, the word error rate (WER: word) is used to group unique user feedback responses that include the phrase. error rate) can be used. WER for two phrases
May measure word substitutions, deletions and insertions to convey similarities between phrases.

  In various embodiments, if multiple unique user feedback responses (eg, definitions) are provided for a given phrase, response evaluation module 1324 may automatically (and / or through manual administrative input) rank the top rank. May approve the unique user feedback response.

  During the check phase, response evaluation module 1324 may determine whether the response submitted by the audience member has been treated as approved, rejected, or pending verification (eg, pending approval). For some embodiments, the data store may indicate whether a unique response submitted by an audience member associated with a given word or phrase has been approved or rejected as a valid definition for the given word or phrase. Can maintain status. Thus, the check phase determines the disposition of the submitted response for a given word or phrase by querying a data store that maintains the disposition status of the previously submitted unique response for the given word or phrase. The submitted response may share the disposition status of the unique response corresponding to the submitted response. Depending on the embodiment, the check phase for a submitted response may be performed immediately or immediately after the response has been submitted. If the submitted response is still determined to be pending validation during the check phase, the check phase may be re-executed later, and possibly after the enablement phase, thereby providing the same or similar The status of the submitted response will be affected. The status of the submitted response may be updated according to the current action of the submitted response as determined during the check phase. As described herein, the current status of one or more responses submitted by a given audience member may be provided as a list reflecting the current status for those responses. Much more about the response status is discussed below in connection with FIG.

  Evaluating the response of response evaluation module 1324 may include determining whether the response is approved. The response may be approved based on at least one previous response provided by another person in response to another query previously generated to obtain feedback on the word or phrase from the another person. Once the response is determined to define exactly the word or phrase, the response can be approved.

The translation data store update module 1326 updates the translation or translation (eg, stored in the translation data store 210), possibly based on the evaluation of the response submitted by the selected audience member for the user feedback query. Can be configured. For example, if during the response evaluation it is determined that the submitted response in the first language is an accepted definition for a given word in the (same) first language, then the first A translation that maps a given word in the first language to a response in the first language may be added or updated. In another example, during response evaluation, if the submitted response in the second language is determined to be an accepted definition for a given word in the first language, then the first
A translation that maps a given word in one language to a response in a second language may be added or updated. Updating the translation or translation may be performed by the translation data store update module 1326 during or after the check phase, with the action taken to ensure that the submitted response is approved.

  Audience suitability assessment module 1328 may be configured to determine the suitability of an audience member that may indicate a level of certainty associated with the ability of the audience member to provide accurate and / or useful user feedback responses. As described herein, the eligibility of a given audience member may be previously provided by a given audience member (eg, in connection with a previous query generated and provided to the given audience member). A determination may be made based on one or more feedback responses made and / or one or more linguistic abilities of the audience member. For example, an audience member may be assessed as having the necessary aptitude to provide user feedback on one or more words or phrases associated with a translation failure if they achieve a certain number of approved feedback responses. In another example, the audience member may include, at least, feedback responses previously submitted for the particular language, audience member preferences, and / or information associated with the audience member indicating fluency in the particular language. Based on that, it can be assessed as having an aptitude in a particular language. In yet another example, an audience member who has submitted a feedback response (eg, a junk response) that has been previously evaluated as incorrect or incorrect may be determined to have low aptitude.

  The incentive reward module 1330 may be configured to provide an incentive to the audience member based on the evaluation of the response submitted by the audience member in connection with the query for user feedback. As described herein, incentives may be given to audience members upon approval of a submitted response. The amount or type of incentive given may be determined based on a number of factors. The many factors include the value of the query or query response assigned by the QR quoting module 1320, the response method used by the audience member in responding to the query (eg, in connection with a particular word or phrase or The amount of incentives already given to the audience member or all audience members through incentive-based user feedback, the language of the inquiry or response provided, and the type of word or phrase for which the response was submitted (eg, chat terms , Abbreviations or special domain words or phrases). The incentive given is a real world currency, or an in-game currency that may or may not have a value (eg, a monetary value in the real world economy) outside its associated virtual economy. Or it may include virtual currency, such as in-game items. For some embodiments, incentives may include real world goods or services or virtual goods or services that may have an associated monetary value. One skilled in the art will recognize that other forms of incentive may be provided in different embodiments.

  For some embodiments, the incentive reward module 1330 may be configured to determine whether one or more of the submitted responses have been approved and / or if an incentive has been given to an audience member for the approved submitted responses. May be responsible for notifying audience members. In various embodiments, the incentive reward module 1330 may notify audience members of incentive rewards through notification messages (eg, in-chat messages such as pop-up messages) and / or updates to a list of statuses for submitted responses. .

One skilled in the art will appreciate that for various embodiments, a system for user feedback is shown in FIG.
It will be appreciated that each of the components shown in FIG. 13 may perform more or less than the operations described for each component. .

  FIG. 14 is a block diagram illustrating an exemplary user feedback client system according to various embodiments. For illustrative purposes, in FIG. 14, the user feedback client system is shown as part of a chat client system 1400 according to various embodiments. For example, a typical user feedback client system may be part of a game chat client system (eg, an in-game chat client system) that is available in connection with an MMO game. Various players of the MMO game may choose to provide user feedback on the flagged word or phrase, possibly in exchange for in-game currency / credits / items as an incentive for feedback. Those skilled in the art will recognize that, in some embodiments, some or all of the exemplary system user feedback, its associated components, may be separate from the exemplary communication conversion and translation system 1300. Will.

  As shown in FIG. 14, a chat client system 1400 may include a chat client controller 1402, a chat client communication module 1404, and a chat client graphical user interface (GUI) module 1406. In some embodiments, as described herein, chat client controller 1402, chat client communication module 1404, and chat client GUI module 1406 are similar to components of chat client system 104.

  As further shown in FIG. 14, the chat client system 1400 includes a conversion / user feedback (UF) query preference module 1408 and a conversion / user feedback (UF) query graphical user interface (GUI) module 1410. For some embodiments, the UF query preference module 1408 and / or the UF query GUI module 1410 facilitate user feedback interaction with the CTT system 1300. In the context of chat client system 1400, a chat user of chat client system 1400 may be an audience member for a user feedback system (eg, CTT system 1300) in various embodiments.

The UF query preference module 1408 may preview, define, and / or adjust preferences with respect to a user feedback function provided in connection with a user feedback system (eg, CTT system 1300) in some embodiments. It can be configured to manage and otherwise allow chat users. Examples of preferences managed by the UF query preference module 1408 include, for example, language preferences for user feedback (eg, the language of the word or phrase for which user feedback is sought and / or the language of user feedback sought), user feedback query. Or a preferred word or phrase type (eg, abbreviation, chat term, physics related, or idiom) and the like.

As used herein, a select form response is a response that is predetermined and is selectable from a list of two or more select form responses. Depending on the embodiment, the list of select form responses may, if applicable, allow the audience member to select more than one response. A freeform response is a response (eg, a character or string value) that includes a text-based value entered into a field by an audience member.

  The UF query GUI module 1410 may present a query generated for user feedback and provided to the chat user (eg, by the CTT system 1300), present one or more response methods associated with the query, and / or be presented. Graphically facilitating receiving a response from the chat user through the response method. The UF query GUI module 1410 may further facilitate the management of preferences through the UF query preference module 1408. Much more about the graphical user interface that can be presented in the chat client system is described below with respect to FIGS. 17-23 and 24-31.

  One skilled in the art will appreciate that for various embodiments, the client system for user feedback may include more or less components than those shown in FIG. 14, and each component shown in FIG. 14 will be described for each component. It will be appreciated that more or less operations may be performed than those performed.

  FIG. 15 is a flowchart illustrating an exemplary method 1500 for user feedback according to various embodiments. At step 1502, the translation failure management module 1314 may identify potential translation or translation failures of the text from the first language to the second language. At step 1504, translation failure management module 1314 may further select a word or phrase from the identified potential failure for user feedback. At step 1506, the UF query audience selection module 1318 may select an audience member to seek user feedback. At step 1508, the UF query generation module 1316 may generate a query to possibly obtain user feedback from the selected audience members. At step 1510, response evaluation module 1324 may receive a response to the generated query. The query API module 1322 may be responsible for providing the generated query to the selected audience members and receiving a response to the generated query. At step 1512, response evaluation module 1324 may evaluate the received response. At step 1514, the audience qualification module 1328 evaluates the qualification of the selected audience member, possibly based on the evaluation of the response provided at step 1510 and / or the response received as performed in step 1512. I can do it. At step 1516, the incentive reward module 1330 may provide an incentive to the selected audience member based on the response evaluation. As shown herein, in evaluating the response and determining that it is approved, the incentive reward module 1330 may determine whether the query response as determined by the value of the query and / or the QR quoting module 1320 Some may therefore reward audience members. At step 1518, translation data store update module 1326 may update the translation or translation based on the response evaluation. As shown herein, upon evaluating the response and determining that it is approved, the translation data store update module 1326 updates the translation or translation corresponding to the query and the words or phrases of the submitted query response. obtain.

FIG. 16 is a block diagram illustrating an exemplary data flow 1600 for a user feedback system, according to various embodiments. As shown, data flow 1600 includes a chat client system 1400, a translation failure management module 1314, a UF query generation module 1316, a UF query audience selection module 1318, a response evaluation module 1324, a translation data store update module 1326, and an incentive reward module. 1330. Data flow 1600 further includes an undefined word / phrase data store 1602, a recorded response data store 1604, a feedback audience data store 1606, a response approval data store 1608, a chat data store 1610, and a dictionary data store 1612.

  The undefined word / phrase data store 1602 may include words or phrases associated with translation failures and for which user feedback is sought. The undefined word / phrase data store 1602 includes a word or phrase, a sample sentence in which the word or phrase is used (eg, the context of the word or phrase), and a confidence measure indicating how important the word or phrase is (eg, Word importance), the source language for the word or phrase, the target language for the required user feedback, and the like. In some embodiments, the significance of a word of a word or phrase in the undefined word / phrase data store 1602 may be initially equal for all words, but the word or phrase appears and becomes problematic and / or Or increase if you can't translate.

  Recorded response data store 1604 may include user feedback responses received from audience members and recorded for words or phrases contained in undefined word / phrase data store 1602. In some embodiments, the user feedback response includes a response received for a query generated to obtain user feedback on a word or phrase. The undefined word / phrase data store 1602 includes a recorded user feedback response, an identifier for the audience member submitting the user feedback response, and a timestamp when the user feedback response was received and / or recorded. , Indicating whether the recorded user feedback response is approved, a timestamp of when the recorded user feedback response was approved, and the like.

  The feedback audience data store 1606 may include a set of identifiers for audience members selected to participate in user feedback on words or phrases included in the undefined word / phrase data store 1602. The feedback audience data store 1606 may include each identifier for the audience member and a certainty score that reflects the consistency, suitability and / or certainty of the audience member in providing a user feedback response.

  Response approval data store 1608 may include each unique user feedback response received in connection with a word or phrase included in undefined word / phrase data store 1602. The response approval data store 1608 stores each unique user feedback response and the unique user feedback response as an approved response (eg, a correct response), a rejected response (eg, an incorrect response), a pending review response (eg, an incorrect response). For example, a response that requires review) or an indication of whether the response has any other status. In some embodiments, the response approval data store 1608 may be used in determining when a user feedback response received from an audience member and recorded in the recorded response data store 1604 has been approved.

According to some embodiments, the translation failure management module 1314 reviews the chat logs, possibly provided by the chat data store 1610, and may include one or more words or words associated with actual or potential translation failures. It can be configured to identify phrases. In various embodiments, translation failure management module 1314 may include a standard dictionary (eg, an Oxford dictionary) and / or words or phrases that embodiments described herein may parse, recognize and / or handle (eg, (Words or phrases of chat terms) may be configured to exclude those words or phrases defined in dictionary data store 1612, which may include a dictionary. The words or phrases identified by the translation failure management module 1314 may be added to the undefined words / phrases data store 1602 so that those added words and phrases may be added to user feedback from selected audience members. Allows you to be selected.

  Translation failure management module 1314 may be configured to select one or more words or phrases from possibly undefined word / phrase data store 1602 for user feedback. For some embodiments, translation failure management module 1314 may select from a set of words or phrases that are designated as having the highest importance in undefined word / phrase data store 1602 (eg, undefined word / phrase). Selected from the top 10 important words or phrases in the phrase data store 1602). In some embodiments, the translation failure management module 1314 may select more than one user feedback to respond to (eg, allow audience members to choose the user feedback they feel most confident in responding). More than one word or phrase may be selected so that a set of can be provided to the audience members. The selection process by the translation failure management module 1314 from the undefined word / phrase data store 1602 includes the importance of the word or phrase, the time of the word or phrase in the undefined word / phrase data store 1602, and the UF query audience selection module 1318. And whether the audience member selected by the UF query audience selection module 1318 has already responded to the selected word or phrase (eg, recorded for the selected word or phrase). (Based on checking response data store 1604), and so on.

  The UF query audience selection module 1318 may be configured to select one or more audience members from which a feedback audience data store 1606 may potentially be sought for user feedback. As described herein, user feedback may be determined for a word or phrase selected by the translation failure management module 1314, possibly from the chat data store 1610. Selection of an audience member from the feedback audience data store 1606 may depend on the eligibility level associated with the audience member.

  The UF query generation module 1316 is configured to generate one or more queries for a word or phrase selected by the translation failure management module 1314, possibly from the undefined word / phrase data store 1602, for user feedback. obtain. As shown, translation failure management module 1314 may provide UF query generation module 1316 with selected words or phrases for which one or more queries are to be generated. As described herein, the UF query generation module 1316 includes, for example, the audience member preferences selected by the UF query audience selection module 1318 and the words or phrases selected for user feedback by the translation failure management module 1314. Many different factors may be considered when generating a query that includes As a result, the UF query generation module 1316 may provide one or more queries generated by the UF query generation module 1316 to the chat client system 1400, with a different query for each word selected and provided by the translation failure management module 1314. Could be produced.

As a result, one or more queries generated by the UF query generation module 1316 may be provided to the chat client system 1400, thereby presenting the provided queries for selection by a user at the chat client system 1400. I do. Depending on the embodiment, the UF query generation module 1316 may provide the generated query to the chat client system, or, alternatively, may provide another way to provide the generated query to the chat client system. Components can play a role. Once the generated query is presented for selection at the client chat system 1400, a user of the client chat system 1400 may select to respond to one or more of the presented queries and the chat client system Those responses provided by 1400 may be added (eg, recorded) to recorded response data store 1604.

  When a response is added to the recorded response data store 1604, some embodiments check that the added response can be evaluated by the response evaluation module 1324. As described herein, response evaluation module 1324 may evaluate a response by checking and treating the status of the response and the response.

  As shown in FIG. 16, the response evaluation module 1324 includes a response check module 1614, which is configured to perform a status check of the user feedback response during the evaluation of the user feedback response. Can be done. Response check module 1614 extracts the user feedback response from recorded response data store 1604 and checks the status of the unique response in response acknowledgment data store 1608 corresponding to the extracted user feedback response. The status of the user feedback response from chat system 1400 may be checked. In so doing, response check module 1614 may determine whether a given user feedback response is approved or rejected. The approval status of the extracted user feedback response in the recorded response data store 1604 may be updated according to the latest status check performed by the response check module 1614. If the response check module 1614 determines that the extracted user feedback response has been approved, the approval status of the extracted user feedback response in the recorded response data store 1604 reflects the approval and updates the approval status. It can be updated to include a timestamp for when it was done. As a result, with the approval reflected in the recorded response data store 1604 for the extracted user feedback response, the incentive reward module 1330 may provide an incentive to the audience member who submitted the approved user feedback response.

  If the status of the extracted translation response is still pending review, response checking module 1614 may recheck the status of the extracted user feedback response at a later time (eg, according to a predetermined schedule). If the status of the extracted translation response is rejection, the approval status of the extracted user feedback response in the recorded response data store 1604 may be updated to reflect the rejection.

  If no unique response corresponding to the extracted user feedback response is found, the extracted user feedback response is unique (eg, by the user of the chat client system 1400) for the word or phrase for which the user feedback response was provided. A response may be added to the response approval data store 1608. Further, if the extracted user feedback response is added as a unique response to response approval data store 1608, the unique response may have an initial approval pending status, and the initial approval pending status may be unique. The status of the response remains until manually or automatically (e.g., through the evaluation response module 1324).

As shown in FIG. 16, the response evaluation module 1324 further includes a response validator 1616, which responds to the status of the unique user feedback response with an approval, rejection, or pending approval. It can be configured to treat any. As described herein, a unique user feedback response submitted in connection with a query generated to obtain user feedback is a unique user feedback response where the unique user feedback response is the word or word associated with the generated query. Approval may be considered pending until it is approved as valid for the phrase or rejected as invalid for the word or phrase associated with the generated query.

  For some embodiments, response evaluation module 1324 may be configured to determine when a given word or phrase no longer requires further user feedback. Response evaluation module 1324 may determine how many unique user feedback responses have been approved for a given word or phrase, and the count of unique and approved user feedback responses has met or exceeded a particular threshold. Such a determination may be made based on an example of a factor such as whether or not the event has occurred. Upon determining that a given word or phrase no longer requires further user feedback, response evaluation module 1324 may be configured to remove the given word or phrase from undefined word / phrase data store 1602. , Thereby removing words or phrases from future selections for user feedback (eg, by translation failure management module 1314).

  As described herein, once the provided user feedback response is approved (eg, by an audience member through the chat client system 1400) as a valid response for the word or phrase for which the user feedback response was provided, Incentive reward module 1330 may provide incentives to the audience member. The incentive reward module 1330 may have been recently approved in the recorded response data store 1604 (eg, the approval status has been updated recently to reflect the approval) and / or the recorded response having the incentive reward module 1330 approved status. One or more user feedback responses that have been approved since the last time they attempted to identify one or more user feedback responses in data store 1604 may be identified. The incentive reward module 1330 may determine when a given user feedback response was last approved based on the approval timestamp included in the recorded response data store 1604 for the user feedback response. For some embodiments, once an incentive is provided for the translated feedback response in recorded response data store 1604, the translated feedback response may be removed from recorded response data store 1604. Alternatively, once an incentive is given to the translated feedback response in the recorded response data store 1604, the translated feedback response may be, for example, when the incentive was given, the amount of the incentive given. Updated in the recorded response data store 1604 to indicate the type of incentive given, when the audience member was notified about this reward, and / or how the audience member was notified about the given incentive. Can be done.

FIG. 17 illustrates an exemplary screenshot for receiving user feedback for a word, according to various embodiments. In particular, FIG. 17 shows screen shots 1702, 1704, and 1706 that illustrate examples of a GUI that may be presented to an audience member (eg, through a chat client system 1400) to facilitate a user feedback process. Screenshot 1702 presents an example of a banner 1708 that asks one or more audience members to participate in user feedback on a word or phrase associated with a translation failure. Audience members may select to participate in user feedback by selecting banner 1708, which may lead to the initiation of a user feedback session and / or may select audience members to participate. Audience members may be directed to a list of available user feedback opportunities. As described herein, user feedback opportunities may allow audience members to provide definitions for words or phrases associated with actual or potential translation failures. According to some embodiments, the audience member may select one of the available user feedback opportunities associated with the word or phrase, and then, when prompted, provide a definition for the associated word or phrase. obtain.

Screenshot 1704 presents a list 1710 of available user feedback opportunities for various words (eg, “Skrilla”, “Booty”, “Cray”, “Hecka” and “Freshness”). Screenshot 1706 provides an example of a query 1712 presented to an audience member to obtain user feedback on the word “Skrilla”. As shown, query 1712 provides a typical context 1714 in which the word “Skrilla” is used, and further provides a field 1716 that is configured to receive a freeform response for query 1712. If the audience member selects user feedback for the word "Skrilla" from the list 1710 in screenshot 1704, the audience member may be directed to screenshot 1706.

  FIG. 18 shows an exemplary screenshot for skipping user feedback according to various embodiments. In particular, FIG. 18 shows screen shots 1802 and 1804 that illustrate examples of a GUI that may be presented to an audience member (eg, through a chat client system 1400) to facilitate a user feedback process. Screenshot 1802 presents a list 1806 of user feedback opportunities available for selection by audience members. As shown, list 1806 provides audience members with the option to skip one or more of the listed user feedback opportunities.

  Screenshot 1804 provides an example of a query 1808 presented to an audience member to obtain user feedback on various words. As shown, the query 1808 provides the audience member with an option to skip the process of providing a response to the query 1808. Some embodiments avoid inaccurate and / or forged responses to various user feedback queries by providing audience members with the option to skip certain user feedback opportunities and / or various user feedback queries. obtain.

FIG. 19 illustrates an exemplary screenshot for receiving user feedback for a phrase, according to various embodiments. In particular, FIG. 19 shows screenshots 1902 and 1904 that illustrate an example of a GUI that may be presented to an audience member (eg, via chat client system 1400) to facilitate a user feedback process. Screenshot 1902 presents a list 1906 of user feedback opportunities available for selection by audience members. As shown, there is a list 1906 of user feedback opportunities available for various words and phrases (eg, “Skrilla” and “Pardon my french”).

  Screenshot 1904 provides an example of a query 1908 presented to an audience member to obtain user feedback on the phrase “Pardon my french”. As shown, query 1908 provides a typical context 1910 in which the phrase "Pardon my french" is used, and further provides a field 1912 configured to receive a freeform response for query 1910. When the audience member selects user feedback for the phrase “Pardon my french” from the list 1906 of screenshot 1902, the audience member can be directed to screenshot 1904.

FIG. 20 illustrates an exemplary screenshot for receiving user feedback through a list of select form responses, according to various embodiments. In particular, FIG. 20 shows screenshots 2002, 2004, and 2006 that illustrate examples of a GUI that may be presented to an audience member (eg, through a chat client system 1400) to facilitate a user feedback process. Screenshot 2002 provides an example of a banner 2008 that prompts one or more audience members to participate in user feedback on a word or phrase associated with a translation failure. Audience members may select to participate in user feedback by selecting banner 2008, which may lead to the initiation of a user feedback session and / or may select audience members to participate. Audience members may be directed to a list of available user feedback opportunities.

Screenshot 2004 provides an example of a query 2010 presented to an audience member to obtain user feedback on the word “Skrilla”. The inquiry 2010 includes a select form response 2012, and the select form response 2012 lists responses that can be selected by the audience member. Screenshot 2006 presents the incentive (eg, five gold coins) provided by notifications 2014 and 2016 once the correct response “money” has been selected for the word “Skrilla”.

FIG. 21 illustrates an exemplary screenshot for creating a list of choices, according to various embodiments. In FIG. 21, a screenshot 2102 shows the word “Skrilla”
Provides an example of a query 2106 presented to an audience member to obtain user feedback on. As shown, query 2106 provides a typical context 2108 in which the word “Skrilla” is used, and further provides a field 2110 that is configured to receive a freeform response for query 2106.

According to some embodiments, the select form response method used to obtain user feedback for a given word or phrase is a predefined form selected from free form responses collected for a given word or phrase. May include a list of responses that have been made. Thus, as shown in screenshot 2104, various audience members provide freeform responses (eg, "A lot of money", "Cash", "Money", and "Really Rich") for the word "Skrilla" through field 2110. As collected, the collected responses may be useful in creating a list of selected form responses 2112.

FIG. 22 shows screen shots illustrating exemplary incentive notifications according to various embodiments. In FIG. 22, a screenshot 2200 shows an audience member notifying of the approval of the response of “money” for the word “Skrilla” and notifying the incentive (eg, XXXX Gold) given to the approved response. Provide an example notification. Screenshot 2202 presents an example notification to an audience member notifying the rejection of a "money" response for the word "Skrilla". Screenshot 2204 provides an example of a push notification to an audience member notifying of approval of the response.

FIG. 23 shows screen shots illustrating an example where translation failed between client chat systems, according to various embodiments. In FIG. 23, screenshot 2300 depicts a typical interface of a first chat client system, and screenshot 2302 depicts a typical interface of a second chat client system. Double arrow 2304 represents chat communication between the first chat client system and the second chat client system. As shown, when the chat user "Aramis" enters a chat communication in English into the interface of the first chat client system, the input chat communication is translated into French and the second chat of the chat user "tapir" is performed. Presented on the interface of the client system. Similarly, when the chat user “tapir” inputs a chat communication in French into the interface of the second chat client system, the input chat communication is translated into English and the first chat client system of the chat user “Aramis” is input. Presented on the interface.

As shown in FIG. 23, the chat communication 2306 (ie, “Tru dat bro”) entered by the chat user “Aramis” into the interface of the first chat client system is the second chat client of the chat user “tapir”. Translation failed when sent to the system interface. The chat communication 2308 (ie, “Tru dat bro?”) Presented to the chat user “tapir” presents the original chat communication input by the chat user “Aramis”, and the chat communication is performed by the chat user “Aramis”. This translation failure is reflected by indicating to the chat user "tapir" that the original chat message was input by "".

The translation failure shown by FIG. 23 may be one that can benefit from user feedback according to some embodiments. According to some embodiments, the translation failure shown in FIG. 23 may be identified by the translation failure management module 1314, wherein one or more words from the original chat communication 2306 (ie, “Tru dat bro?”) User feedback from participating audience members may be added to the undefined word / phrase data store 1602 for future selection. For example, each word “Tru”, “dat” and “bro” may be added to the undefined word / phrase data store 1602 for each future user feedback.

FIGS. 24 and 25 show screen shots illustrating exemplary lists of words or phrases available for user feedback according to various embodiments. In FIG. 24, screenshot 2400 presents a list 2402 of user feedback opportunities available for audience member selection for words and phrases (including “Tru”) that are available for audience member selection. In FIG. 25, screenshot 2500 presents another list 2502 of user feedback opportunities available for audience member selection for phrases available for audience member selection. 24 and 25, screenshots 2400 and 2500 may be part of an in-game chat system, whereby a game player may provide user feedback for a word or phrase, and may approve a feedback response. , In-game credits (eg, in-game gold).

FIG. 26 shows screen shots illustrating examples of defining words according to various embodiments. In FIG. 26, a screenshot 2600 shows a query 2602 that includes a typical context 2604 in which the word “Tru” is used, and further provides a field 2606 configured to receive a freeform response to the query 2602. .

  FIG. 27 shows a screen shot illustrating an exemplary list of select form responses, according to various embodiments. In FIG. 27, screenshot 2700 presents a list 2702 of responses that an audience member can select to define the word “nemore”.

FIG. 28 shows a screenshot showing an exemplary list of status for responses submitted according to various embodiments. As shown in FIG. 28, the list of submitted response statuses 2802 includes the pending status for the first response 2804 and the approval status for the second and third responses 2806 and 2808. For some embodiments, the list 2802 may provide further information about response status including, for example, why a particular response is approved, rejected, or still pending review.

FIG. 29 shows a screenshot illustrating an exemplary incentive notification, according to various embodiments. In particular, FIG. 29 shows a screen presenting members with a typical notification 2902 to accurately define the phrase “U still thr” as “You still there?” During a user feedback process according to an embodiment. A shot 2900 is provided. The notification indicates that the user will be given 10 gold pieces that may be valuable or useful as in-game currency as an incentive for the user feedback provided. As shown, the notification further provides a summary of the user feedback (ie, the word or phrase in question and the provided user feedback response).

In one embodiment, a learning system for data selection is provided wherein feedback obtained from a user is automated by a machine learning system having a check and balance for player consistency. The system adds parallel sentence received from the player to a parallel corpora that can be used to retrain the statistical machine translation (SMT) system from time to time.

The chat conversion system may be, or may include, a system that converts chat terms into plain speak. For example, the chat conversion may convert "Ur da king" (a chat term message) to "You are the king" (a plain language message). In certain embodiments, "plain language" refers to a normal language typically spoken and / or written by a normal individual outside of an electronic chat environment where chat terms may dominate. Plain language tends to be more grammatical than chat terminology.

  The learning system may further utilize or include a language translation system that translates one language into another. For example, a language translation would replace "How are you doing kind sir" (an English message) with "

(Spanish message).
In some embodiments, the "translation corpus" is two texts, one for each language, arranged in parallel so that line n in one text corresponds to line n in the second translated text. It is understood to mean text. A bilingual corpus is also referred to as a "training corpus" in such a context.

  In various embodiments, “machine learning” refers to a supervised, semi-supervised, or unsupervised system that can learn from patterns in input data and develop mechanisms to detect, transform, or predict behavior. Is understood.

In general, the construction of a chat conversion system and a language translation system requires a large amount of bilingual corpora for proper syntax rules or statistical learning. The systems and methods described generally herein utilize a reliable statistical chat translation and language translation system using a bilingual corpus. However, in some situations, this initial training data set may be limited in its content and scope. For example, new chat words are created daily and added to chat rooms. To maintain an accurate and reliable conversion and translation system, these new chat words should be supplemented to the chat conversion training corpus.

In various embodiments, “Out of Vocabulary (OOV)” (
Systems and methods are provided for identifying words that are, for example, words that are not present in a given vocabulary. In some embodiments, with reference to FIG. 30, a method 3000 is provided for detecting and processing OOV words. In step 3002, an OVV word is first detected by transmitting the OVV word through a translator system, such as CTT system 114 or CTT system 1300 and / or one or more modules thereof. If the output from the translator system is the same as the input for a given word, the translator system indicates a lack of translatability that suggests that the word may be OOV. To further evaluate the word as a potential OOV word, the systems and methods may determine whether the word is a new word, not just a misspelled word (step 3004). Both misspelled words and new words will appear as OOV. Thus, the words that can be modified with the spell checker can be considered to be misspelled words rather than OOV words.

  Further, OOV words that appear frequently in chat are generally more likely to be OOV words (eg, new chat utterances). For example, if a word has been used by a user in a previous text message, such previous use indicates that the word may be an OOV word. In some embodiments, an ensemble of machine learning and language processing methods is used in parallel to detect whether a word is an OOV word (step 3006).

  Additionally or alternatively, a Bayesian probability may be calculated to provide a statistical probability that the OOV is a new word rather than a misspelled word (step 3008). Pure chat terms generally tend to follow a word used before or after the chat term. In comparison, spelling errors have a less consistent distribution of adjacent words. Calculating prior and posterior Bayesian probabilities helps to distinguish useful OOV words that can be added to the vocabulary from spelling errors that should not be added to the vocabulary. You. For example, consider the phrase "Wassup, how's it going". Since "Wassup" does not exist in the standard vocabulary, it is considered as an OOV word. However, "Wassup" is almost always followed by the word "How's it going", or "Wassup" is often used at the beginning of a sentence. This pattern or consistent behavior is captured by Bayesian probabilities. The system is trained on text that has misspelled words but no chat words.

Alternatively or additionally, a machine learning method such as k-means clustering may be used to distinguish different types of OOV words, such as new chat words, misspelled words or junk (step 3010). . K-means clustering tends to result in hidden similarities between classes of words. Words belonging to similar topics tend to be clustered together, thereby indicating hidden similar semantic relationships between them. Again, consider the example of "Wassup, how's it going". By clustering groups of sentences using the k-means algorithm, "Wassup" appears in clusters of greeting words such as "Hi", "What's up", "Hello", "Hi!" It becomes clear that is included. In contrast, the spelling error will be located at the edge of the cluster or not at all in any defined cluster. These hidden relationships help distinguish useful OOV words from errors. The sentence syntax and semantics can be analyzed to determine what kind of OOV word (eg, verb, noun or adjective) the sentence contains.

  When the system and method detect a new chat word or other OOV word, the new chat word is presented to a human translator to define a translated or language translated version of the new chat word. . Thereafter, translated or translated versions of the new chat words may be added to the translated vocabulary and used by the systems and methods described herein.

  As described herein, where incentives are provided for manual translation of chat between languages, there is a potential for a user to manipulate the system fraudulently to exploit incentives (eg, in-game currency). is there. The systems and methods described herein can detect fraudulent submissions but are generally tolerant of the ability of human translators.

  When a user of the system acts as a translator, the user translates one or more words or sentences into a specified target language. However, if the user gives an incorrect, incomplete or improper translation to deceive the system or to gain incentives without fulfilling the purpose of the system, the user commits wrongdoing. I have.

  In certain embodiments, with reference to FIG. 31A, the systems and methods described herein utilize a fraud detection module 3100. The fraud detection module 3100 detects fraud in incentived translations by presenting both new and old training data (eg, a bilingual corpus) to the user. Old training data corresponds to translations for which the correct answer is known, while new training data corresponds to translations for which the correct answer is unknown. The percentage of new data over old data can be varied for the user over time. For example, more older data may be shown initially, then the percentage may be gradually reduced.

  In some embodiments, fraud detection is performed by checking the translation accuracy of old data received from the user. Based on this accuracy, a certainty score is assigned to each user. Large or sudden changes in translation accuracy, or consistently low accuracy, indicate cheating or poor translation ability in the user. Even after establishing the certainty of the translator's capabilities, preferably, at least 10-20% of the time for periodic fraud checks is randomly populated with old data.

  Using this basic structure, the fraud detection module 3100 may include and / or utilize a supervised fraud detection module 3102 and / or an unsupervised fraud detection module 3104. According to the supervised fraud detection module 3102, for example, the presented input sentence, the resulting translation, the existing known known true translation, the user's current certainty score, and the translator over time The reporting tool may show the output from each user in a user interface having a field such as a graph showing the variation in the certainty score of the user. When reviewing a translation, a human teacher may accept or reject the translation, thereby adjusting the user's certainty score. If the cumulative report shows incorrect behavior, the teacher may remove the user (can revoke the user's translation authority). Removing a user or revoking a user's translation authority may be performed using the translation authority module 3106.

Alternatively or additionally, fraud detection module 3100 may utilize unsupervised fraud detection module 3104. According to the unsupervised fraud detection module 3104, translation accuracy can be measured by various metrics such as WER (Word Error Rate) and BLEU (Machine Translation Accuracy Metric comparing machine translation with good quality reference translation). Can be calculated using: Certainty in user translation ability is changed or fluctuated (eg up or down)
Can be checked for Similar sentences presented to one translator may be presented to other independent translators using the system. Reliability between translators can also be calculated. For example, through random sampling, social network analysis (e.g., ascertaining whether two translators are not socially connected or previously involved), and chat sessions and / or By detecting repeated interactions between users in online games, collusion between translators can be avoided. Two users that regularly interact together online (eg, in an online game or chat session) may be more likely to engage in collusion. In some embodiments, item response theory (ie, the theory and test theory used in linguistic psychology) is used to augment the translator's ability to measure translator certainty. Fraud detection can be performed using item response theory to perform unsupervised fraud detection in translation enhancement systems with incentive mechanisms. Item response theory defines how translators' accuracy can be measured against peers and against themselves to measure consistency over time. In this way, deviations from the standard can be identified. Inter-translator credibility may also be calculated by re-presenting the same sentence to a translator after a set period of time. Various thresholds in reliability and translator certainty can be set, and if the translator certainty falls below such threshold, the translator can use the translation authority module 3106 to configure the system. Can be removed or blocked (eg, the user's translation authority can be revoked). In some implementations, translations from the high certainty system are added to the translation versus vocabulary.

  FIG. 31B includes a flowchart of a method 3110 for detecting fraud in incentived translations, according to an embodiment of the present invention. The method includes selecting a mixture of old training data and new training data (step 3112). Old training data includes one or more old text messages whose correct translation into a different language is known. The new training data includes one or more new text messages for which the correct translation into a different language is unknown. A plurality of respective requests are sent to the user's client device at different times (step 3114). The request includes (i) a request to the user to translate the old training data and / or the new training data, and (ii) an incentive for the translation. After sending the particular request, a translation is received from the client device for the old training data for the particular request (step 3116). The received translation is compared to the correct translation for the old training data (step 3118). The accuracy of the received translation is determined based on the comparison (step 3120). Next, the certainty score is updated for the user based on the translation (step 3122). The certainty score indicates the likelihood that the user will later provide an accurate translation of the text message into a different language.

  In various embodiments, the systems and methods described herein utilize speech translation or speech recognition technology to translate audible speech in one language into another for users of a group voice chat system. translate. For chat terms, a speech-to-text transcription system transcribes a user's chat terms into text, which is then translated into plain language (eg, non-chat terms) and translated into a foreign language. Can be realized. Thereafter, a final conversion is performed to create a foreign language chat term, which is then output to the end user through the foreign language text-to-speech system. The system and method preferably use state of the art speech recognition and statistical machine translation techniques with very fast decoders.

FIG. 32 is a schematic illustration of a group chat system 3200 that enables groups 3202 of people who speak different languages to interact with one another using chat terms. As described herein, system 3200 may identify a language spoken by people participating in group chat system 3200. If a first user 3204 wishes to send an audible chat term message to a second user 3206, the first user 3204 may send an audible chat term message 3208 in a first language (eg, English) to a user input device (eg, a chat client system). Microphone). The voice recognition module 3210 converts the audible chat term message into a chat term text message 3212 in the first language. The conversion module 3214 is used to convert the chat term text message 3212 into a plain language (eg, non-chat term) first language text message 3216. Next, translation module 3218 is used to translate the plaintext text message 3216 into a corresponding plaintext text message 3220 in a second language (eg, French). Thereafter, the conversion module 3222 is used to convert the corresponding plain text message 3220 into a corresponding chat term text message 3224 in the second language. As one skilled in the art will appreciate, the conversion module 3222 may be the same as or form part of the conversion module 3214. Thereafter, text-to-speech module 3226 is used to convert the corresponding chat term text message 3224 in the second language into a corresponding chat term audible message 3228. Finally, the corresponding chat term audible message 3228 is delivered to the second user 3206 using an output device (eg, a speaker on the second chat client system).

In various embodiments, the speech recognition module 3210 includes a hidden Markov model, a dynamic time warping (DTW) based speech recognition, to convert the audible chat term text message 3208 into a chat term text message 3212. And / or a neural network may be utilized. Similarly, text-to-speech module 3226 may use speech synthesis to convert a corresponding chat term message into a corresponding chat term audible message. As will be appreciated by those skilled in the art, speech synthesis may involve concatenative synthesis (eg, unit selection synthesis, diphone synthesis, and / or domain-specific synthesis). , Formant synthesis, articulatory synthesis, HMM-based synthesis, and / or sine wave synthesis.

An important aspect of creating such a speech processing system involves collecting speech samples from multiple accents and dialects for the language that can be processed. The nature of the audio data may include the format of chat terms and plain language in each language so as to maintain the relevance of the system to the domain to which the system corresponds. The incentive feedback mechanism described herein can be used to transcribe these speech samples, which in turn can be used by the speech recognition module 3210 and / or the text-to-speech module 3226. Can be used to train. If sparse, field adaptation techniques can be used to replace the data points. This may be necessary in the case of chat term speech samples where data tends to be sparse. For example, audio data collected in the gaming arena (eg, for an online game) may be replaced with plain language data that is fully available data. The domain adaptation is preferably a rule that governs chat terms in a given language—small speech fluctuations from plain language (eg, conversion of chat terms into plain language or chat terms in plain language from plain language). Rules that govern the conversion to Plain language sentences that do not have audio samples in the corresponding sentences of the chat terms may be converted to chat terms using these domain-level rules. The user feedback loop may tune acoustic model parameters (eg, for speech recognition module 3210 and / or text-to-speech module 3226) to a level that makes the acoustic model field more restrictive, and thus more accurate. Can be used. For example, if the speech recognition module 3210 consistently has difficulty with a particular accent, additional audible samples of various words at that accent may be provided to the system (eg, by the user). This will help the speech recognition module 3210 learn how to better recognize words spoken with the accent.

  As described above, embodiments of the systems and methods described herein are used to translate text or chat messages from a group chat environment into different languages. Archiving such translated chats can lead to a great deal of text in different languages remaining in the repository.

  Referring to FIG. 33A, in one embodiment, to reduce storage requirements and facilitate user review of a chat history, the chat history module 3300 stores the chat history as it is browsed by the user. Used to translate in real time. Chat history module 3300 includes a chat storage module 3302 (eg, a register or other storage device) for storing chat histories from various users. Chat history module 3300 further includes a chat history conversion module 3304 that converts the text message before and / or after the text message is translated into a different language. For example, chat history conversion module 3304 may perform a real-time conversion of chat history text messages from chat terms to formal or plain language. In some embodiments, the chat history conversion module 3304 is the same or similar to the conversion module 208. Chat history module 3300 further includes a chat history translation module 3306 that may be used to perform real-time translation of chat history text messages (eg, formal or plain language) into different languages (eg, French to English). Chat history translation module 3306 may be or include other modules or components described herein, such as language module 206 and / or conversion data store 210.

  Once the user has reviewed the chat history, the translated and / or translated text generated by the chat history module 3300 may be deleted or removed from memory. This reduces storage requirements for systems and methods. If the user wishes to review the chat history at a later time, the chat history module 3300 can be used again to translate and translate the text in the chat history as needed.

  In some embodiments, chat history module 3300 translates chat history for a user in real time. Chat history module 3300 receives a request from a user to review the history of a text message from a chat session. The chat history module 3300 receives from the chat history storage module 3302 a history of text messages including text messages in a plurality of languages. The chat history conversion module 3304 and the chat history translation module 3306 are then used to translate and / or translate the initial portion of the chat history into the language used by the user as needed. After viewing the translated first portion of the chat history, the user may want to view a different portion of the chat history. Thereafter, chat history module 3300 may receive a request from the user to view different portions of the text message history. Thereafter, the chat history conversion module 3304 and the chat history translation module 3306 are used to translate and / or translate different portions of the chat history into the language used by the user as needed. Chat history module 3300 preferably performs the conversion and / or translation in real time as the user scrolls through the chat history.

  In some cases, scrolling through the chat history presents scale and data storage issues. Providing infinite scrolling of chat history presents the problem of fast real-time access of data across multiple databases and multiple users. This can be done by performing multiple processes in parallel that fetch historical messages from all users present in the chat room. The translation on these messages and the associated chat conversion may be performed in parallel as the messages are fetched from data storage. The resulting output implemented by the end user is the output of a seamless transition from one screen of the chat to the next screen where data lookup from the database has already been performed. This is because the systems and methods described herein may have no limitations on data storage and parallel computing can be reused between previously performed processes, resulting in an infinite number of processes. You can continue about the screen.

  FIG. 33B is a flowchart of a method 3310 for translating chat history in real time, according to an embodiment of the present invention. The method 3310 includes receiving a request from a person to review a history of text messages from a chat session (step 3312). The history preferably includes text messages in multiple languages and from multiple users. At least two parallel processes are executed (step 3314). Each parallel process receives (i) receives or selects a text message (ie, a text message that forms at least part of a history of text messages) generated by a respective user of the chat session; and (ii) receives the text message. Translating into a target language. Translated text messages from the plurality of parallel processes are provided to one's client device (step 3316). A request is received from the person to review different parts of the text message history (step 3318). Steps 3314 and 3316 are repeated for different parts of the history of the text message.

  In some cases, users of the systems and methods described herein may want to avoid interacting with some other users in a group chat or gaming environment. In previous chat systems, barring and silencing chat users is typically handled by the administrator or moderator of the chat server. However, embodiments of the systems and methods described herein allow a user to have direct control over who can send chat messages and / or chat contact invitations to the user. I do. For example, user A may prevent user A from seeing communications from user B in any chat room anymore and / or if user A no longer invites personal chat contacts from user B (ie, one-on-one chat). User B may be allowed to block so as not to receive it.

  In various implementations, an alliance is a group of players in a game (eg, a multiplayer online game) that can be grouped together as a unit to enhance game play. Each alliance preferably has its own chat room, and members of the alliance can talk to each other or send text messages. This presents a need to sometimes block certain users from the Alliance chat room.

FIG. 34A includes a screenshot of a user interface 3400 that allows a first user of a gaming system to block communication from a second user of the gaming system, according to an embodiment. As shown, the first user selects the administration settings icon 3402 (eg, by tapping on a touch screen), which opens a settings window 3404. Thereafter, the first user selects the “Block from Alliance” button 3406 on the setting window 3404. A message window 3408 appears to notify the first user that future communications from the second user will be blocked. The second user may be added to a list of other users blocked by the first user. The first user may have the option to edit the list to add or remove users from the list. For example, referring to FIG. 34B, the next time the user selects the management settings icon 3402, the settings window 3404 may include unblocking from the alliance button 3410. If the first user selects unblock from the alliance button 3410, future communications from the second user may be unblocked and the first user may be notified that such communications have been unblocked. A message window 3412 may appear.

  In some cases, the complexity of the system is brought about or reduced by the scale at which blocking and unblocking are performed. Parallel computing may provide the flexibility to perform blocking and deblocking in real time, without the lags seen in conventional systems. For example, parallel processing may be used to translate and / or convert text messages in a text message chat system. A separate parallel process may be assigned to each user of the chat session and / or each language used for the chat session. Such parallel processing may simplify the task of blocking and unblocking users. For example, separate parallel processes may be removed or added from the chat system when a user is blocked or unblocked, respectively.

  FIG. 35 is a flowchart of a method 3500 of blocking a user from a chat session, the method 3500 including providing a text message chat system to multiple users of an online game (step 3502). A request to block a second user of the text message chat system is received from a first user of the text message chat system (step 3504). After receiving the request, prevent a text message from the second user from being displayed to the first user (step 3506). In some cases, text messages in a chat session are translated and / or converted using the systems and methods described herein. Parallel processes may be used to perform translation and / or conversion of text messages. For example, a separate parallel process may be assigned to handle translation and / or translation of text messages for each particular user of the chat session and / or for each language included in the chat session.

  Automatic translation services are not always accurate and can sometimes benefit from human intervention to correct certain errors. In some implementations, the translation systems and methods described herein allow a user to identify translation errors and provide corrections to correct these errors. For example, a bilingual or foreign language user (eg, a French player of an online game) may view the chat window and see a translation that is incorrect (eg, a translation into or from French). The user may submit a suggestion for a correction for the incorrect translation and may be rewarded (eg, with in-game currency or virtual items) for submitting the correction.

In some implementations, the original text message and the corresponding translation are displayed on a single screen, thereby providing an opportunity for those skilled in the language to provide immediate feedback on the translation. For example, a user may recognize a translation error and select an option to submit a corrected translation. Thereafter, the user may enter and submit the modified translation and receive a reward when the modified translation is approved. In submitting a modified translation, a user may be prevented from submitting additional modified translations for the original message. Thus, the user may not be able to get multiple rewards from a single incorrect translation.

In some cases, the system and method cannot translate the original message because the original message was not correctly entered by the user. For example, FIG. 36A shows an original Spanish message 3602 representing “Eres el peor!”. An automatic English translation 3604 of this message is shown in FIG. 36B and represents "You are the best!" Referring to FIGS. 36C and 36D, the user may recognize that the translation was incorrect because the original message was not entered in the proper Spanish. To respond to this error, the user may select a "correct translation button" 3606 to open a correction window 3608 where the user can enter corrections for the original message. Referring to FIGS. 36E and 36F, in this case, the user inputs “Ustedes son los mejores!” In correction window 3608, and selects submit button 3610. A confirmation window 3612 appears to notify the user that the submission will be processed. Upon closing confirmation window 3612, the user returns to the original chat page.

  Referring to FIGS. 37A and 37B, a user interface 3700 is provided that allows a user to review translation corrections received from other users. Users reviewing translation corrections may be rewarded for their efforts and may be able to select the particular type of reward they receive (eg, virtual goods or currency for online games). In general, after a translation correction has been submitted by a user, other users may determine whether the correction is better than the original translation and any other translation corrections submitted by other users. If the user's translation is determined to be the best translation, the user may receive an award and the user's translation may be added to a translation dictionary (eg, translation data store 210). Users participating in determining various translations may also receive rewards. However, such rewards may only be given to the user who selects the translation correction best selected by all of the judges.

  In general, by allowing a user to submit translation suggestions and to determine the submissions of other users, systems and methods utilize feedback that the user is willing to provide. The data collected in this process, once approved, can be used to modify translation cache entries, thereby improving the overall translation capabilities of the systems and methods described herein. This may ensure that when the original message is resubmitted for translation, the correct translation will be indicated in the future.

  In a typical implementation, there are two types of users, monolingual users and bilingual users, who can submit and / or determine translation corrections. Bilingual users can generally understand sentences in the original language and provide more accurate translations in different languages. In contrast, monolingual users may not understand the original language phrase, but may still consider translations (presented in the language of the user's device) and submit corrections in exchange for rewards. The translations obtained from the two types of users tend to be different, with bilingual users generally providing more accurate translations. The systems and methods can preferably determine or detect whether a user is monolingual or bilingual, and user feedback can be weighted according to the determination. For example, a user may be able to identify themselves to the system and method as being monolingual or bilingual.

  In some situations, most users are monolingual and speak the same language (English). The greater the supply of speakers of a language, the more users generally will submit translation corrections for that language, and generally the less demand for translation corrections to or from that language. Become. To stimulate the provision of translation corrections for other languages, the user may be rewarded according to the translation correction needs. For example, if the majority of users speak one language and there is no shortage of translation corrections provided in that language, such users will receive a lower reward (eg, 75% of nominal amount) for submitting translation corrections. ). At the same time, a small number of users who speak different languages may receive greater rewards (eg, 125% of nominal amount) due to greater demand for translation corrections for the different languages.

  The number of translations that a user modifies during a given time period (eg, one day) may or may not be limited. For example, if no reward is given for submitting a correction, the number of translation corrections may be unlimited. On the other hand, if the user is rewarded for such submission, during that time period the user may be allowed to submit a limited number of translation corrections. Such a restriction can result in bilingual users or users who tend to submit numerous translation corrections gaining an unfair advantage in underlying games (eg, multiplayer online games) by receiving excessive rewards. Can be prevented.

  In some cases, feedback on incorrect translations may be received from only a small number of users (eg, two or three users), thereby determining the accuracy of translation submissions and automatically rewarding them. It can be difficult to generate. For example, chat occurs in a continuous stream, and many users may be more focused on chatting with other users and / or playing the underlying game and less focused on submitting translation corrections. do not do. Furthermore, users can select a chat based on what they see in their window, and few users can select the same chat. Thus, if more than one translation amendment is received, the amendment suggestion may be made available to other users to make a decision to gain consensus on the correct translation in exchange for a reward.

Rewards for submitting translation corrections may be given to users according to a raffle system. In such a system, not all submissions will be rewarded, but users who submit more modifications may be randomly distributed so that they are more likely to be rewarded. Such an approach may allow a player to achieve an unfair advantage over other users due to his / her ability and / or desire to translate messages, rather than his / her own ability or effort in the underlying game. Reduce the likelihood.

  In addition to allowing the user to correct bad translations, the user should also provide feedback regarding misdetected languages, unfiltered abusive words, and specified entity detection Can be done. For example, when viewing the original message and the translated message, the user may recognize that the automatic translation system has improperly detected the original language. The user may then inform the system about this language detection failure in exchange for a possible reward. Similarly, the user may be able to notify the system about any abusive words that appear in the message, so that the system filters or removes such abusive words from future messages. Enable. The user may also be able to notify the system about specified entities that appear in the message, such as company, brand, trademark, etc. This may allow the system and method to recognize when designated entities appear in a message and to ensure that such entities are properly designated and / or identified.

  In general, translation corrections submitted by a user require careful evaluation to ensure that the user is only rewarded for the correct correction. This improves the overall accuracy of the system and prevents users from committing fraud by submitting fraudulent modifications. In some implementations, the accuracy of the translation correction depends on word-based features, language-based features, and other features (eg, word alignment match features) to prevent users from abusing the system. Automatically evaluated using Part of speech (POS) based language models can be used to check sentences for grammatical accuracy. Further, some users may submit translation corrections that are grammatically correct but have nothing to do with the original message. For such cases, the features of word alignment match analysis may be useful and may be implemented as a periodic process to approve and / or reject a user's submission. In the translation systems and methods described herein, a machine learning approach may be used to enable sparse user feedback.

  Table 2 shows an example of a translation correction proposal submitted by a user, according to an embodiment of the present invention. In these examples, the original message in the source language is "aaa bbb ccc" and the correct translation in the target language is "xxx yyy zzz". The column labeled "Indicated Translation" contains an example of the first translation proposed by the automated system described herein.

Table 2. Examples of User Modifications and Preferred Results As shown in Table 2, when a user submits a correct and enhanced translation, the user's submission should be approved and the user may receive an appropriate reward. However, if the user submits a low quality or incorrect translation (eg, a random message), his submissions should be rejected and no reward should be given to the user. The system and method preferably approve or reject such instances as indicated in the "Status" column of this table.

  In some embodiments, the translation of the original message is categorized according to whether the translation is appropriate for the original message. The classification can be treated as a binary classification task where features are extracted from the translation and the original message. Classification techniques may be used to ensure that translation corrections submitted by users are accurate. For example, in some cases, the majority-based validation described herein is not preferred due to the small number of responses (eg, 1-3) that may be received for an incorrect translation. Classification techniques may also be used to identify and / or respond to hash collisions that appear in cached transformed data. For example, about 10% or more of the translation entries in the data table can be corrupted by hash collision.

  Referring to FIG. 38, in various implementations, translation accuracy is evaluated using a translation accuracy module 3800 including a word-based feature module 3802, a language-based feature module 3804, and a word alignment module 3806. . The word-based feature module 3802 is used to evaluate word-based features such as word counts, character counts, pictograms, numbers and / or punctuation. For example, if the translation is correct, the number of words in the original message and the number of words in the translation are generally about the same. Thus, if the number of words in the two messages differs by more than a threshold amount (eg, a factor of 2), the translation may be considered incorrect or likely incorrect. In one example, if the number of words in one of the messages (eg, translations) is （(or less than の) the number of words in another message (eg, original message), the word-based feature module 3802 may conclude that the translation is or is likely to be incorrect.

  Another word-based feature that can be used to assess the accuracy of the translation is the number of letters (eg, letters and numbers) in the original message and the translation. In general, if the number of characters in the original message is approximately the same as the number of characters in the translation, the translation is more likely to be accurate. The threshold amount may be used to determine when character counts in the two messages are significantly different. For example, if the translation has more than 3/2 characters of the original message, the word-based feature module 3802 may conclude that the translation is likely to be incorrect or incorrect.

  Another word-based feature that can be used to assess the accuracy of translation is emoji, which generally remains unchanged between the original message and the translation (eg, ideographs or smileys used in Japanese electronic messages) (Smiley)) count and order. Emoticons tend to fall into certain Unicode text ranges that can be used to detect them in a given sentence. Using this Unicode range, regular expressions can be used to identify or extract emoji from both of the messages. For example, if the input contains three pictographs consecutively and the output contains only one pictograph, a difference is indicated between the input and the output. If the glyph count and / or order is different between the two messages, the word-based feature module 3802 may conclude that the translation is or is likely to be incorrect.

An additional word-based feature that can be used to assess the accuracy of the translation is the count of any digits and punctuation in the two messages. For example, numbers and punctuation, if present, can be identified or extracted in the original message and translation, and the length of the longest common subsequence (LCS) is determined between them after sorting. Can be This length divided by the maximum of the two message lengths gives a real value for this word-based feature. In general, real values provide an indication of the percentage of duplicate numbers and punctuation in the two messages. Experimental results show that for this particular feature, better results are obtained using real values rather than binary values. For example, an English input sentence "I am going to meet you at 4:30 Cya !!" may have an equivalent output of "Je vais vous recontrer a 4:30 Au revoir !!". When extracting punctuation and numbers, "4:30 !!" is obtained for both the English and French versions. In this case, the LCS is 6 (unit is character), and the maximum length of the English version and the French version is 36 (unit is character). The resulting real value for this word-based feature is 6/36 = 0.167.

  Relying solely on word-based features may be insufficient for assessing translation accuracy. For example, a user may be able to deceive at least some of the word-based features by submitting a translation correction in which each word of the original message is replaced with a dummy word (eg, “xxx”). , Thereby creating a fraudulent modification with the same number of words and letters present in the original message.

To avoid this problem, translation accuracy module 3800 may use language-based feature module 3804 to evaluate language-based features in addition to or instead of word-based features. For example, in one embodiment, the words present in the original message and the translation (e.g., using an open source POS tagger) to identify the part of speech (POS) (e.g., verb, noun, adjective, etc.) in the two messages. Using). Each word in the message may be tagged according to the part of speech with a different number of tags, using different tags set for each language. For example, the sample sentence "The device is easy to use" can be tagged with a POS tagging machine as "The_DT device_NP is_VBZ easy_JJ to_TO use_VB", indicating the part of speech of each word in the sentence. In this case, the tags are a determiner (DT), a noun phrase (NP), a singular present adjective (VBZ), an adjective (JJ), To (TO), and a single verb (VB). The main subject tag for this purpose is typically a verb, followed by adjectives and adverbs.

  In some cases, the original message and the translation are tagged separately (e.g., using a POS tagging machine) and the resulting tags for each message are counted to identify the number of verbs, adjectives, adverbs, etc. in each message. Is done. Due to the different types of verbs used in each language (e.g., legal verbs, infinitive verbs, past verbs, future verbs, etc.), a simplified verb tag VB maps all types of verbs in each language. Can be obtained using For example, an English verb part-of-speech tag may be mapped to a singular verb tag VB as follows. That is, 'VBD' (verb, past tense) => 'VB', 'VBG' (verb, gerund) => 'VB', 'VBN' (verb, past participle) => 'VB', 'VBP' (Verb, non-third-person singular present) => 'VB', and 'VBZ' (verb, third-person singular present) => 'VB'. POS tags in tagged messages can be replaced with this simplified set of POS tags.

After simplifying the POS tags, the number of verb tags VB can be counted in both the original message and the translation. Ideally, the number of verbs in each message should be the same, with some exceptions. For example, translate "was sleeping" in English to "dormais" in French. An English POS tagger may tag "was" and "sleeping" as two different verbs, while a French POS tagger may tag "dormais" as a single verb . "Is", "was", and "can"
Such verbs are known as auxiliary verbs in English. Other languages may not have equivalents of these auxiliary verbs; instead, a single verb may be used as a replacement. Given such differences in verb usage between languages, systems and methods may use a threshold (eg, 2 or 3) for the difference in the number of verbs between the original message and the translation. For example, if the difference between the number of verbs in two messages is greater than two, language-based feature module 3804 may consider that the translation is likely to be incorrect or incorrect. This threshold of 2 has been found to give reasonable results through trial and error. Other parts of speech (eg, adjectives and adverbs) can be counted and compared between the two messages using one or more additional thresholds.

  However, in some cases, the user may deceive this translation accuracy check by simply copying and submitting the existing translation as a modification to the translation. In that case, the submission may be classified as a valid modification, but the user cannot be rewarded for the submission. In some cases, a user may even more simply change the case of some words in an existing translation to create and submit a valid correction, and the user deserves a reward and submits the submission. You can be rewarded for things. Thus, in some embodiments, systems and methods determine whether an existing translation and a user's submission are the same. If the existing translation (including, for example, case and capitalization) and the user submission are the same, no reward can be given to the user.

  In some embodiments, POS tag checking is used to identify cases where the automatic translation system did not correctly identify the language of the original message. For example, the language of the original message may be incorrectly detected when the user's translation correction passes the word count check, but fail in the POS tag check. Further, if the number of verbs is equal to zero, or if all tags are nouns in one message and not in the other, there is a possibility of incorrect language detection. For example, the original Spanish message is "

May be represented. However, if the language is detected to be English, the English POS tagger will most likely not be able to tag the message, and may tag all words as nouns by default. The output of the English POS tagging machine may be, for example, as follows.

As a comparison, the output of a Spanish tagger for the same original message could be:

The tags "NN", "RB", and "PPX" refer to the noun (s), adverb, and possessive pronoun, respectively.

  Thus, in some cases, the original message and the part of speech of the translation are compared to determine whether the language was properly identified in the original message. In general, if one of the messages (eg, translations) is tagged as having a non-zero number of verbs and the other message (eg, original message) is tagged as having no verbs, a language detection failure will occur. More likely to occur. Further, if all words in one message are tagged as nouns, while other messages have some type of POS tags (eg, nouns, verbs and adjectives), the probability of language detection failure is higher. .

  In various embodiments, the accuracy of the translation is assessed by identifying and reviewing the original message and proper nouns in the translation. In general, if the translation is accurate, proper nouns (eg, people's names and city names) are the same in the translation and in the original message. Thus, comparing two messages and filtering for common words that remain untranslated may be useful as a feature for identifying pure translations. In some cases, the presence of such untranslated proper nouns may help improve translation accuracy, but the absence of untranslated proper nouns may provide any information about translation accuracy. Absent. If the proper noun is identified in the original message rather than in the translation, the accuracy of the translation may be considered likely to be inaccurate or incorrect. A penalty may be added to the score of the real value returned for this proper noun feature, thereby identifying any bad translations and helping to improve translation accuracy. For example, if the proper noun is inconsistent between the two messages, the accuracy score for the translation may be reduced by a penalty.

  Alternatively or additionally, the accuracy of the translation can be assessed by analyzing and comparing the grammar in the original message and the translation. Working in multiple languages can make it difficult to parse the tree for all languages to understand the grammar of the sentence. Messages are often also written in chat languages that follow a different grammar compared to plain or formal speech in their native language.

Thus, to recognize patterns between grammars in chat languages, sentences can be tagged with POS tags to build an N-gram language model for the POS tags, thereby approximating the underlying grammatical structure. I will provide a. An n-gram may be defined as a set of n consecutive words. These n-gram models may be typical for a given language and / or may be used to represent n consecutive words in a given language. In one implementation, the word-based n-gram method is extended to part-of-speech-based n-grams. In other words, a shallow method of parsing the sentence, where the words in the sentence are tagged by the POS tagging machine, may be used. In one approach, a BLEU score is calculated for the POS tag rather than the actual text.

A trigram (3 gram) language model can be created on POS-tagged sentences for each language. For example, the sentence "The device is easy to use" becomes "The_DT device_NP is_VBZ
It has a POS tagged output of "easy_JJ to_TO use_VB". The word-based trigrams in this sentence are {The, device, is}, {device, is, easy}, {is, easy, to}, {easy, to, use}. The corresponding POS based trigrams are {DT, NP, VBZ}, {NP, VBZ, JJ}, {VBZ, JJ, TO}, {JJ, TO, VB}.

  Trigrams with higher probabilities may be used to indicate a partial structure of the grammar. For example, a trigram language model built on a POS tag may have a probability associated with each trigram. The probability can be calculated as the ratio of the number of times a given trigram has occurred in the text corpus compared to all trigrams in the same text. Grammar trigrams tend to be repeated often and therefore will have a higher probability (also known as the language model score). Thus, if a message receives a higher score on this language model, the message is more likely to be grammatically correct. This score can be useful in detecting when a user types a spam message to get a reward. The score may further be useful in determining when language detection has failed. For example, sentences in the wrong language may have even lower scores because a separate model may be used for each language. The score may further be useful in detecting when the quality of the translation is good. Separate models for human translation and machine translation can be trained for this purpose.

  In some embodiments, the language model is trained using translations that have been validated to be accurate. A trigram model can be built on a POS tag.

  An inherent problem with n-grams of any size is the lack of all possible hypotheses. In such cases, a backoff method is employed in which the n-1 and n-2 grams are identified. For example, if an unknown POS trigram such as {DT, NP, JJ} is found, the model determines whether the bigrams {DT, NP} and {NP, JJ} have linguistic model scores indicating grammar. Can be checked retroactively. If the bigram has no probability (or if the probability is too low), another backoff can be performed to check the unigram language model score for {DT}, {NP}, {JJ}. All three models are also linearly interpolated with more weight on the trigram model, less weight on the bigram model, and less weight on the unigram model. Can be grouped by

The most frequently occurring trigrams of POS tags for chat English are <IN><VBD><PP> = 1.0, <FW><NN><IN> = 1.0, <DT><WP><VBP> = 1.0 and <RB><//S1> // S2> = 1.0. A score of 1.0 indicates that a given trigram sequence has a 100% probability of being grammatical at all times it occurs. In general, chat languages use more pronouns starting with "Wh" followed by a verb, which is captured by <DT><WP><VBP>. Also, people tend to end chat language sentences with adverbs or adjectives, as in "You are cool" or "This is awesome", which is captured by <RB></S1></S2>. , S1 and S2 are sentence end tags. Thus, these trigrams can be used to recognize at least some structure of the underlying grammar in each language. Since language models are built for all languages of interest, models can be saved in JSON format and loaded immediately at any time.

In some implementations, after a message has been POS-tagged, the systems and methods may count the number of trigrams in the message that match existing trigram language models. Each trigram in the model may have a probability score associated with it, and for each trigram in the message, the corresponding score of the trigram is identified in the model. In some cases, if the score is above a threshold (eg, 0.05), the trigram is counted as a match. Otherwise, the trigram is not counted as a match. The system and method may calculate the number of trigram matches against the total number of trigrams in a sentence, and this ratio may be used as a real value feature to assess translation accuracy. For example, a trigram of {easy, to, use}, which may occur frequently in grammatical text, may have a high probability score (language model score) of about 0.68. In contrast, the nongrammatical trigram of {easy, use, to} may have a smaller probability of occurrence of about 0.04. When compared to a defined threshold of 0.05, the ungrammatical trigram may be filtered, and the number of trigrams above the threshold may provide a quantitative value for textual grammar.

After obtaining the POS-tagged message, the sentence probabilities may be calculated using a trigram language model. In one embodiment, the log probability of each trigram in the sentence is determined and the sum of the log probabilities is calculated. This sum is then divided by the number of words in the sentence to get a score for that sentence. The score may be used as a feature of the real value to assess the accuracy of the translation. The log probability of the trigram is preferably calculated as a linear interpolation of the trigram, bigram and unigram probabilities. For example, in the sentence "The device is easy to use", the POS-tagged output is "The_DT device_NP is_VBZ easy_JJ to_TO use_VB". The POS-based trigrams for this sentence are {DT, NP, VBZ}, {NP, VBZ, JJ}, {VBZ, JJ, TO}, and {JJ, TO, VB}. Each of these trigrams has a probability of occurrence in a given language corpus. Assuming these probabilities are 0.12, 0.44, 0.32 and 0.89 for each trigram, the combined score for the grammar of the sentence can be calculated as a log probability. In this case, the log probability is calculated as log (0.12) + log (0.44) + log (0.32) + log (0.89), which is equal to -1.82. On a relative scale, this log probability provides a numerical indication of the grammar of the sentence.

  In addition to or in place of the word-based and language-based features described above, the translation accuracy module 3800 may evaluate the word alignment between the original message and the translation. , The word alignment module 3806 may be used. To generate a list of potential word alignment pairs, a translation accuracy check algorithm is performed on the database using only word-based features, and a bilingual corpus having translation pairs with probability values> 0.90 is generated. Extracted. This indicates that only good quality translated message pairs are filtered to create a bilingual corpus. 100K sentences were collected for English and French pairs and English and Spanish pairs. The bilingual corpus for these 100K sentences was supplied to the Statistical Machine Translation Toolkit (ie, GIZA ++) to extract word alignments. The toolkit extracted approximately 25-30k word alignment pairs and the associated probability scores for these pairs.

Given that there is typically more than one word alignment per word, it may be beneficial to select only those alignments that have a probability score greater than some threshold (eg, 0.01). However, even with such thresholds, more than one alignment per word can be obtained, most often due to spelling errors or the same word in a different tense (eg, past or future). A sample word alignment between English and French is shown in Table 3. Separate word alignments may be obtained for both possible orders of the two languages (eg, English to French and French to English). Therefore, two word alignment files can be extracted for each language pair. A word alignment for the source language (ie, the language of the original message) with respect to the target language (ie, the language of the translation) may be referred to as a source alignment, while a word alignment for the target language, relative to the source language, may be referred to as a target alignment.

Table 3. Typical word alignment between English and French Source and target alignments can be loaded into two separate files. For each word in the original message that is also present in the source alignment, the system and method determines whether at least one corresponding word is present in the translation and also in the target alignment. Then the same process is applied to the translation. Finally, the percentage of words found in the word alignment file is returned as a feature for assessing translation accuracy.

  In some implementations, word alignments are extracted for language pairs that include English as one of the languages. For example, a language pair is a combination of English and one of the following: Spanish, French, Polish, Portuguese, Dutch, German, Danish, Swedish, Turkish, Italian, and Russian. Possible, there are a total of 11 language pairs and 11 × 2 = 22 word alignment files. For language pairs that do not include English (eg, translated from French to Russian), English may be used as an intermediate language to extract the alignment. For example, if you enable French to Russian translation, French-English may be used to extract word alignment from the original message, and Russian-English may be used to extract word alignment from translation. . The intersection of these two sets gives the percentage of matches between word alignments in both messages. This method can be extended to all language pairs.

  In various embodiments, the user's certainty is calculated to provide an indication of whether the user's translation submission can be trusted. Apart from the other approaches discussed herein, a user may be more or less trusted based on his translation history. The number of modified translations and the number of rewarded translations may be stored in data tables for various users. This allows the percentage of corrections made by the player to be calculated and the translation corrections to be approved based at least in part on the certainty certainty threshold of the particular user. This allows spammers to be separated from pure players who easily modify translations.

FIG. 39 is a flowchart of a method 3900 for correcting translation errors, according to an embodiment. The method 3900 includes providing a text message chat system to multiple users of the online game (step 3902). An original text message is received from a first user of an online game in a first language (step 3904). A first translation of the original text message is generated in a second language (step 3906). The original text message and the first translation are provided to a second user of the online game (step 3908). For example, a second user may view the two translations together or separately on a display of the client device. A translation correction is received from a second user to respond to an error in the first translation (step 3910). The method 3900 may further include identifying the most accurate translation correction from a plurality of translation corrections (ie, including a translation correction from a second user) (step 3912). Alternatively or additionally, method 3900 uses at least one of a word-based feature, a language-based feature, and a word alignment feature to assess the accuracy of translation corrections from a second user. (Step 3914).

  The word-based features and language-based features described above (eg, four word-based features and four language-based features) may be fitted using a linear regression model. After training, the model preferably returns the number of real values for each translation pair, and a threshold can be used to classify each translation pair as good or bad. For example, if the feature returns a numerical value of x1, x2, x3... X8, the regression equation may be y = a1 × x1 + a2 × x2 +. And y is the output value. The preferred value for threshold y was found to be 0.65 after precision and reproducibility experiments using ROC curves.

  After adding the word alignment-based features described above and again performing a linear regression on the word-based features and the language-based features, the preferred threshold was changed to 0.76. In addition, the addition of word alignment-based features increased the AUC value from 0.853 to 0.976.

  Table 4 shows the regression coefficients obtained from training 13k sentences for each of the word-based features, the language-based features, and the word alignment features. The results in the table show that, after normalization, the emoji features and character counts have small regression coefficients, which means that these features were found to contribute little to assessing translation accuracy. .

Table 4. Regression coefficients for translation accuracy features The regression model was evaluated by 10-fold cross-validation on 13k sentences of French-English pairs and Spanish-English pairs. Representative labels for the data were calculated using BING translation with some human supervision. Table 5 shows the accuracy, reproducibility, accuracy, and F-measure value in percentage for evaluation. Accuracy represents the ratio of true positive translation pairs among all translation pairs marked as true by our method. Reproducibility is the ratio of true positive translation pairs among all translation pairs known to be true pairs. Accuracy is the ratio of the sum of the results marked as true positive and true negative to the total test set size. The F scale is a harmonic mean of accuracy and reproducibility. These metrics can be used in classification tasks to measure system performance and reliability. 13k sentences were used as a training set, and 400 manually annotated sentences were used for testing. Because the test set was small, the number for tests was less than for 13k sentences.

Table 5.1 Results of Observing Translation Accuracy Using 13,000 Sentences Table 6 shows the results of 10-fold cross validation for a 13k sentence dataset, where W refers to the use of word-based features, L refers to the use of language-based features and A refers to the use of word alignment features. These results indicate that the word-based features help improve accuracy and that the word alignment features help significantly improve reproducibility. Language-based features give a slight boost to both accuracy and reproducibility. In general, recall indicates how accurately a bad decision is detected from the total dataset. The results in the above table show that adding word alignment-based features improves recall. Accuracy indicates how accurately a good translation was predicted from the total translations marked correct by the system. The results in the table show that adding word-based features improves accuracy.

Table 6.1 Results of 10-fold cross validation for the sentence dataset of 13,000 Table 7 shows that various word-based features, language-based features, and word alignment features were fitted with other machine algorithms, according to one embodiment. The results are shown below. Previous results have been shown using a linear regression technique to connect different features together. There are machine learning algorithms that can be used to combine variables (features in this context) together to produce better ensemble results than the individual parts. Linear regression shows a one-dimensional method for combining variables. In the literature on machine learning algorithms, there are two-dimensional and multi-dimensional methods for combining variables. These algorithms were used to find a better way to combine the features used in the task of predicting good translation pairs.

The results in Table 7 were obtained by combining the features with various machine learning algorithms. The dataset used contained 13k sentences and the parameters were adjusted using a gridSearch algorithm. From the methods listed in the table, the gradient boosting classifier and the random forests method are ensemble based methods, which is why these methods gave better results. explain. Gradient Boosting Machines (GBMs) take longer to train, but GBMs and random forests give very good results. However, training time is generally not important, as the model only needs to be trained once.

Table 7. Results of Fitting Translation Accuracy Features by Machine Algorithm The final translation accuracy check algorithm is tested against a provided test set of 3045 English-French sentence pairs. The results are shown in Table 8. The performance of the WLAplain feature is poor due to the fact that the word alignment was extracted from a plain language database. As the message is modified after a series of transformations, a sharp decrease in these results can be found. Word alignments were extracted from the chat term database and some smoothing was done on smaller sentences to avoid zero scores in features. WLA shows results for the chat term database. Poor accuracy due to WLA features indicates that many true translations have been rejected. Features were fitted by linear regression as random forests overfitted the results. In general, these results can be used as a basis for selecting the final set of features used in the algorithm. While higher accuracy is generally preferred, higher accuracy rewards more accurate translation pairs entered by the user, and higher recall results in fewer false entries being incorrectly classified as correct. Guarantee. According to the results in the table, a WLA feature set is preferred. A threshold of 0.75 may be selected for higher reproducibility (e.g., to get the least malicious entry that is marked correct), and a threshold of 0.68 if high accuracy is desired. A threshold may be selected.

Table 8. Results from the Translation Accuracy Check Algorithm Although the invention has been particularly shown and described with reference to certain preferred embodiments, it departs from the spirit and scope of the invention as defined by the appended claims. It should be understood by those skilled in the art that, if not so, various changes may be made in form and detail.

Claims (18)

A method implemented by a data processing device,
(A) receiving a request from a multi-user chat session to review a portion of the history of text messages, wherein the history includes a plurality of text messages, each text message being in a respective language, and Of the chat session participants, the method further comprises:
(B) (i) and selecting the history each text message that is different from a portion of said text message, a plurality of parallel each including a translating the text message selected (ii) the target language and a carrying out the process,
The plurality of parallel processes includes one process for each respective chat session participant or each language used for a respective chat session, the method further comprising:
(C) providing translated text messages from the plurality of parallel processes to a user's client device;
(D) receiving a request to review different parts of the history of the text message;
(E) repeating steps (b) and (c) for the different portion of the history of the text message. Wherein selecting a different respective text message includes querying the storage device for the portion of the history of the text message, the method according to claim 1.   3. The method of claim 1 or 2, wherein translating the selected text message comprises converting at least a portion of the text message from chat terms to plain language.   The method of claim 1, further comprising receiving a request from the user to stop browsing the history of the text message.   The method according to any of the preceding claims, wherein the plurality of parallel processes comprises one process for each respective chat session participant. Wherein the plurality of parallel processes, including one process with each language used in the respective chat session, the method according to any one of claims 1 to 5. The system
A computer-readable medium for storing instructions;
A data processing device,
The data processing device includes:
(A) is configured to execute the instructions to perform operations including receiving a request to consider part of the history of text messages from a multi-user chat sessions, the history comprises a plurality of text messages, Each text message is in a respective language and originates from a respective chat session participant, and the data processing device further comprises:
(B) (i) and selecting the history each text message that is different from a portion of said text message, a plurality of parallel each including a translating the text message selected (ii) the target language Executing the instruction to perform an operation including performing a process .
The plurality of parallel processes includes one process for each respective chat session participant or each language used for a respective chat session, wherein the data processing device further comprises:
(C) providing translated text messages from the plurality of parallel processes to a user's client device;
(D) receiving a request to review different parts of the history of the text message;
(E) the different portions configured to execute the instructions to perform operations including a repeating steps (b) and (c) the system of the history of the text message. Wherein selecting a different respective text message includes querying the storage device for the portion of the history of the text message, according to claim 7 system.   The system of claim 7 or 8, wherein translating the selected text message comprises converting at least a portion of the text message from chat terms to plain language.   The system of any one of claims 7 to 9, wherein the action further comprises receiving a request from the user to stop browsing the history of the text message.   The system according to any one of claims 7 to 10, wherein the plurality of parallel processes comprises one process for each respective chat session participant. Wherein the plurality of parallel processes, including one process with each language used in the respective chat session, according to any one of claims 7 to 11 systems. A computer program, wherein the computer program, when executed by a data processing device, the data processing device,
(A) causing an operation including receiving a request from a multi-user chat session to review a portion of the history of the text messages, wherein the history includes a plurality of text messages, each text message being in a respective language. and are transmitted from each of the chat session participants, the computer program further, when executed by said data processing apparatus, the data processing device,
(B) (i) and selecting the history each text message that is different from a portion of said text message, a plurality of parallel each including a translating the text message selected (ii) the target language Performing an operation including performing a process,
The plurality of parallel processes includes a process for each respective chat session participant, or each language used for a respective chat session, wherein the computer program is further executed by the data processing device, In the data processing device,
(C) providing translated text messages from the plurality of parallel processes to a user's client device;
(D) receiving a request to review different parts of the history of the text message;
(E) above for the different parts of the text message history to perform operations including a repeating steps (b) and (c), the computer program. Said different selecting the respective text message includes querying the storage device for the portion of the history of the text message, the computer program of claim 13. Translating said selected text message, at least in part comprises converting into plain words from the chat term, computer program according to claim 13 or claim 14 of the text message. The operation further includes the computer program according to any one of claims 13 to 15 to receive a request to stop to view the history of the text message from the user. Wherein the plurality of parallel processes, for each respective chat session participants including one process, the computer program according to any one of claims 13 to 16. The plurality of parallel processes, including one process with each language used in the respective chat session, the computer program according to any one of claims 13 to 17.