Abstract: Systems and techniques for validation and generation of localized content for audio and video are described herein. The systems and techniques provide for training of twin neural networks to evaluate performance characteristics, sometimes referred to as content-auxiliary characteristics, of a localized performance. The localized performance may be validated or improved by identifying misalignment in the performance characteristics to ensure that localized content preserves content as well as creative intent and performance ability in the final product. The machine learning models trained using the techniques described herein may be used in connection with auto-localization processes to automatically generate high quality localized audio and video content.

Abstract: Certain aspects of the present disclosure include systems and techniques for generating content that indicates a sensation associated with audio. One example method generally includes monitoring audio to be played during display of an associated portion of an interactive content stream provided over a communication network to at least one viewing device during an interactive session, and analyzing, via a machine learning component, the audio to determine a sensation associated with at least a portion of the audio. The method may also include determining an effect indicating the sensation, wherein the effect is associated with one or more output devices associated with the at least one viewing device, and outputting an indication of the effect to the associated output devices, wherein the effect is configured to be output along with the audio in real-time with the display of the associated portion of the interactive content stream.

Abstract: A generation device is a device that generates a translated sentence in a second language from an input sentence in a first language to be translated. The generation device includes: an acquisition unit that acquires the input sentence; a normalization unit that converts the input sentence into a normalized sentence that is a grammatically correct sentence in the first language; and a first translation unit that generates the translated sentence by translating the normalized sentence into the second language using first parallel translation data that is parallel translation data between the first language and the second language. The normalization unit generates the normalized sentence by using second parallel translation data that is parallel translation data between a third language and the first language. A data amount of the second parallel translation data is larger than a data amount of the first parallel translation data.

Abstract: Provided is a system and method for processing Compliance Data Subject Requests (DSRs) across a network by providing an executable planning model that contains a DSR domain description language, a DSR action description language, DSR problem templates, and predetermined logical, temporal, and dependency constraints. The DSR domain description language includes domains, with each domain including a domain name, and an object hierarchy representing various objects and their inter-relationship within each domain. The DSR action description language includes at least DELETE, REMOVE, OBFUSCATE, and ENCRYPT actions. The DSR problem templates each correspond to an operation that operates upon a compliance object within one of the domains, using at least one of the DSR actions. An instance of DSR processing may comprise receiving from a user DSR that includes a user object on which DSR processing is required and automatically processing the user DSR by executing the executable planning model.

Abstract: A method, computer system, and a computer program product for text detection is provided. The present invention may include training a text detection model. The present invention may include performing text detection on an inputted image using the trained text detection model. The present invention may include determining whether at least one of a plurality of bounding boxes generated using the inputted image has an aspect ratio above a threshold. The present invention may include based upon determining that at least one of the plurality of bounding boxes generated using the inputted image has the aspect ratio above the threshold, upscaling any text within the at least one bounding box and performing text detection on a new image using the trained text detection model. The present invention may include outputting an output image.

Abstract: Techniques are disclosed for augmenting data sets used for training machine learning models and for generating predictions by trained machine learning models. These techniques may increase a number and diversity of examples within an initial training dataset of sentences by extracting a subset of words from the existing training dataset of sentences. The techniques may conserve scarce sample data in few-shot situations by training a data generation model using general data obtained from a general data source.

Abstract: Methods and apparatuses for performing context completion to messages in a session are provided in the present disclosure. A message may be obtained. It may be detected that there exists context ellipsis in the message. It may be determined whether the message is retained in the current domain of the session. In response to determining that the message is retained in the current domain, a complementary text for recovering the context ellipsis may be selected in the current domain. A completed message may be generated based on the message and the complementary text.

Abstract: System and method are provided for identification and classification of multilingual messages that would be considered inappropriate in an online interactive portal. The system may include processors to generate a set of data of intended inappropriate multilingual messages to train classification model. The set of data with labels is classified by assigning unique identifiers. The system includes pre-processing module to eliminate unwanted characters from set of data to train classification model. The classification model may be trained by multilingual representation module based at least in part on set of data with labels. The classification model determines whether set of data with one or more labels includes intended inappropriate multilingual messages. Furthermore, feedback loop module is utilised to retrain classification model recurrently to update set of data.

Abstract: A smart translation system that translates the input content received from an application based on translation metadata and the application is disclosed. It is initially determined if a translation of the input content exists in a user cache. If it is determined that the translation of the input content exists in the user cache, the translation is retrieved from the user cache. Else, if it is determined that the translation of the input content does not exist in the user cache, the domain and language contexts of the input content are determined and an automatic translation engine is selected based on the contexts and the translation metadata. The translated content is presented to the user via the application while maintaining the look and feel of the application.

Abstract: The possibility of a desired program being automatically generated is increased by a program generation device including: a storage unit that stores a set of pairs each constituted by a program and natural language that describes a specification of the program; a search unit configured to search for a first program having a structure that is the closest to a structure of a program to be generated, from the set, by inputting natural language that describes a specification of the program to be generated and natural language included in the pairs to a model that has been caused to learn a relationship between natural language describing a specification of a program and similarity of a structure of the program; and a change unit configured to change the first program to generate a second program that satisfies at least one pair of an input value and an output value.

Abstract: Methods and apparatus are described for evaluating dubbing of media content. Emotions are identified based on combinations of attributes determined for segments of a source language audio and a dubbed audio. The emotions may be compared to determine emotional prosody transfer between the source audio and dubbed audio. Based on the comparison, a notification is generated indicating whether an emotion classification associated with the source audio matches an emotion classification associated with the dubbed audio.

Abstract: Systems, methods, and non-transitory computer readable media including instructions for interpreting facial skin micromovements are disclosed. An example head mountable system includes a housing configured to be worn on a head of a wearer. The head mountable system also includes at least one detector integrated with the housing and configured to receive light reflections from a facial region of the head and to output associated reflection signals. The head mountable system also includes at least one microphone associated with the housing and configured to capture sounds produced by the wearer and to output associated audio signals. The head mountable system also includes and at least one processor in the housing, configured to use both the reflection signals and the audio signals to generate output that corresponds with words articulated by the wearer.

Abstract: A computer-implemented method for multi-model representation learning is provided. The method includes encoding, by a trained time series (TS) encoder, an input TS segment into a TS-shared latent representation and a TS-private latent representation. The method further includes generating, by a trained text generator, a natural language text that explains the input TS segment, responsive to the TS-shared latent representation, the TS-private latent representation, and a text-private latent representation.

Abstract: Provided is a method for augmented reality-based image translation performed by one or more processors, which includes storing a plurality of frames representing a video captured by a camera, extracting a first frame that satisfies a predetermined criterion from the stored plurality of frames, translating a first language sentence (or group of words) included in the first frame into a second language sentence (or group of words), determining a translation region including the second language sentence (or group of words) included in the first frame, and rendering the translation region in a second frame.

Abstract: The present application provides a database generation method and apparatus, an electronic device and a medium. The method includes: acquiring an image of a Chinese character, determining, based on the image of the Chinese character, elementary components contained in the Chinese character and positions of the elementary components in the image of the Chinese character; determining position codes corresponding to the positions of the elementary components; and storing the elementary components and the corresponding position codes thereof in a corresponding manner to obtain a component library of a database. Thus, the difficulty of generating the database is reduced.

Abstract: Embodiments of a system and a method for classifying blocks of text at varying, possibly simultaneous, and possibly interacting levels of scope (e.g., sentence, paragraph, section, document) are disclosed. A system includes a processor coupled to: a data reader to receive an input comprising of a data stream, and convert the data stream into one or more logical data blocks of varying scopes. The system includes a dependency tree generator to create a dependency tree to define the scope and dependencies of each of a plurality of natural language processors (NL Processors) with respect to each other such that the dependency tree identifies when an input of a NL Processor (depender) depends on an output of another NL Processor (dependee). Each NL Processor is configured based on a type of at least one logical data block of the one or more logical blocks to which it applies.

Abstract: To easily construct a translation model environment in which the sentences of a plurality of second languages corresponding to a sentence of a first language can be generated. An internal state changing device 1 includes a change unit 13 configured to change an internal state that is a final hidden state of an encoder, in a translation model that is a learned model constituted by an encoder-decoder model, as the translation model for translating an original sentence that is a sentence of a first language into a translated sentence that is a sentence of a second language.

Abstract: A source content routing system is described for distributing source content received from clients such as documents, to translators for performing translation services on the source content. The routing system extracts source content features, which may be represented as vectors. The vectors may be assembled into an input matrix, which may be processed using an artificial neural network, classifier, perceptron, CRF model, and/or the like, to select a translator such as a machine translation system and/or human. The translator provides translation services translation from a source language to a target language, post translation editing, proof reading, quality analysis of a machine, quality analysis of human translation, and/or the like and returns the product to the content routing system or clients.

Abstract: There is provided a computer-implemented method for ensuring that a large language model (LLM) generates original text, including (i) providing or accessing a database of previous text that the LLM should not generate, wherein the database includes text used to train the LLM; (ii) checking potential continuations generated by the LLM against the database; (iii) when a potential continuation generated by the LLM matches text in the database, adjusting the potential continuation generated by the LLM to no longer match that text in the database, to produce an adjusted potential continuation, and (iv) storing the adjusted potential continuation.

Abstract: An electronic device is provided. The electronic device includes an input unit configured to receive a voice or text, an output unit, and a processor. The processor is configured to determine context information, translate the received voice or text based on the context information, convert the translated voice or text, and output the converted voice or text using the output unit.

Abstract: Systems and methods for evaluating a pre-trained artificial intelligence (AI) model using layered prompts. The system obtains a set of application domains in which the AI model will be used, and a set of guidelines that define one or more operational boundaries of the AI model. The system determines a set of layers, where each layer is associated with corresponding guidelines and mapped to a set of variables and benchmarks. Each variable represents an attribute within the guidelines and each benchmark indicates the degree of satisfaction of the AI model with the guidelines. The AI model is dynamically evaluated against these benchmarks using a series of assessments. Subsequent assessments are dynamically constructed based on the outcomes of previous assessments. Scores are assigned to the AI model for each layer by comparing the expected and actual responses. The results are then displayed in a graphical user interface (GUI).

Abstract: Communication systems, methods, and non-transitory machine-readable storage media are disclosed herein. A communication system may include a communication device configured to receive a video stream including sign language content and any content indicators associated with the video stream during a real-time communication session within a single communication device or between a plurality of communication devices. The communication system may also include a translation engine configured to automatically translate the sign language content into word content during the real-time communication session without assistance of a human sign language interpreter. Further, the communication system may be configured to output the word content translation to a communication device during the real-time communication session.

Abstract: A device, system and method for processing images that include amounts is provided. A receipt image is received at a controller of a computing device. The receipt image is rendered, at a display screen, the receipt image comprising a total amount region including a total amount of expenses in the receipt image. A total amount field, including a submission total amount associated with the total amount, is rendered at the display screen adjacent the receipt image. A link between the total amount region and the total amount field is rendered at the display screen, at the total amount region of the receipt image. Input is received confirming the submission total amount, via an actuatable option rendered at the display screen in association with the total amount field. The submission total amount is transmitted, via a communication interface, to an expense management system.

Abstract: An electronic device includes a memory storing at least one instruction, and a processor configured to execute the at least one instruction to: input a first text into a first neural network model and acquire a first translation text; input the first translation text into a second neural network model and acquire a first liberal translation text; identify whether the first liberal translation text is to be outputted based on the first text and the first liberal translation text; output the first liberal translation text according to the identification result; based on a second text being received subsequently to the first text, input the first and second texts into the first neural network model and acquire a second translation text; and input the identification result and the second translation text into the second neural network model and acquire a second liberal translation text corresponding to the second translation text.

Abstract: A server accesses a natural language query. The server facilitates a mapping of the natural language query to a vector using a query-to-vector engine. The server matches the vector to an intent representing a prediction associated with the natural language query. The server provides a response to the natural language query based on the intent.

Abstract: A method for providing sentiment analysis of a natural language service request to automatically identify and map a user state of mind by utilizing artificial intelligence is disclosed. The method includes receiving, via a graphical user interface, a raw input, the raw input including a computer file corresponding to a natural language request; parsing the raw input into component parts; annotating, by using a model, each of the component parts with a predetermined indicator, the predetermined indicator corresponding to the user state of mind; mapping, by using the model, the component parts based on the predetermined indicator; compiling the mapped component parts into a structured input, and determining, by using the model, a quality that corresponds to the structured input.

Abstract: Implementations of the disclosure provide a method, system and computer program product for model localization. In an implementation of the disclosure, a method for model localization includes parsing a model to identify translatable terms, generating a seed file associating each of the translatable terms with a corresponding tag and replacing each translatable term in the model with a corresponding tag and submitting each of the translatable terms to machine translation for a target language to produce a different translation file mapping each tag from the seed file with a translated term in the target language of a corresponding one of the translatable terms. Then, the model may be deployed in a data analytics application using the different translation file to dynamically translate each translatable term into a corresponding translated term within a user interface to the data analytics application.

Abstract: Systems and methods for automatic evaluation of the quality of NLG outputs. In some aspects of the technology, a learned evaluation model may be pretrained first using NLG model pretraining tasks, and then with further pretraining tasks using automatically generated synthetic sentence pairs. In some cases, following pretraining, the evaluation model may be further fine-tuned using a set of human-graded sentence pairs, so that it learns to approximate the grades allocated by the human evaluators. In some cases, following fine-tuning, the learned evaluation model may be distilled into a student model.

Abstract: In some aspects, a computing device may access audio information comprising an audio stream from a client device. The computing device may provide an audio segment from the audio stream to a language identification process of the computing device comprising a machine learning model that is trained to identify a language of a plurality of languages within recorded speech. The computing device may identify an identified-language of the plurality of languages for the speech based at least in part on the audio segment. The computing device may provide the identified-language to the client device. Numerous other aspects are described.

Abstract: Methods are provided, such as a method of interacting with a large language model (LLM), including the step of a processing system using a structured, machine-readable representation of data that conforms to a machine-readable language, such as a universal language, to provide new context data for the LLM, in order to improve the output, such as continuation text output, generated by the LLM in response to a prompt; and such as a method of interacting with a LLM, including the step of providing continuation data generated by the LLM to a processing system that uses a structured, machine-readable representation of data that conforms to a machine-readable language, such as a universal language, in which the processing system is configured to analyse the continuation output generated by the LLM in response to a prompt to enable an improved version of that continuation output to be provided to a user. Related computer systems are provided.

Abstract: Provided is a method of training a real-time simultaneous interpretation model based on external alignment information, the method including: receiving a bilingual corpus having a source language sentence as an input text and a target language sentence as an output text; generating alignment information corresponding to words or tokens (hereinafter, words) in the bilingual corpus; determining a second action following a first action in the simultaneous interpretation model on the basis of the alignment information to generate action sequence information; and training the simultaneous interpretation model on the basis of the bilingual corpus and the action sequence information, wherein the first action and the second action represent a read action of reading the word in the input text or a write action of outputting an intermediate interpretation result corresponding to the read action performed up to a present.

Abstract: In some embodiments, a method can include generating a resume document image having a standardized format, based on a resume document having a set of paragraphs. The method can further include executing a statistical model to generate an annotated resume document image from the resume document image. The annotated resume document image can indicate a bounding box and a paragraph type, for a paragraph from a set of paragraphs of the annotated resume document image. The method can further include identifying a block of text in the resume document corresponding to the paragraph of the annotated resume document image. The method can further include extracting the block of text from the resume document and associating the paragraph type to the block of text.

Abstract: Exemplary embodiments relate to the automatic generation of captions for visual media, including photos, photo albums, non-live video, and live video. The visual media may be analyzed to determine contextual information (such as location information, people and objects in the video, time, etc.). A system may integrate this information with information from the user's social network and a personalized language model built using public-facing language from the user. The personalized language model captures the user's way of speaking to make the generated captions more detailed and personalized. The language model may account for the context in which the video was generated. The captions maybe used to simplify and encourage content generation, and may also be used to index visual media, rank the media, and recommend the media to users likely to engage with the media.

Abstract: Methods and systems for generating and using a semantic index are provided. In some examples, content data is received. The content data includes a plurality of subsets of content data. Each of the plurality of subsets of content data are labelled, based on a semantic context corresponding to the content data. The plurality of subsets of content data and their corresponding labels are stored. The plurality of subsets of content data are grouped, based on their labels, thereby generating one or more groups of subsets of content data. Further, a computing device is adapted to perform an action, based on the one or more groups of subsets of content data.

Abstract: A machine translation method includes: receiving a sentence, the sentence including a plurality of words; calling a machine translation model obtained through training, the machine translation model including a partitioning model and a translation model; partitioning the sentence based on the partitioning model and according to word vectors of the words, to obtain to-be-translated blocks, each to-be-translated block including at least one of the words; and translating the sentence based on the translation model and the to-be-translated blocks, to obtain a translation result.

Abstract: A translation device comprises: a memory that stores first dictionary information including a replacement target term in a first language and a translated word in a second language for the replacement target term in association with each other, and second dictionary information including an alternative term instead of the replacement target term; a controller that replaces the replacement target term included in an input sentence with the alternative term to generate a replaced sentence; and a communication interface that outputs the replaced sentence to an external device and acquires a translation result of the replaced sentence in the second language from the external device. The controller detects a specific character string in the translation result of the replaced sentence acquired from the communication interface, and replaces the detected character string with a translated word associated with the replacement target term in the first dictionary information, to generate a translated sentence.

Abstract: Provided is a method of performing hybrid translation using a general-purpose neural network machine translator by a device, including: (a) receiving an original document configured of a first language from a user terminal; (b) determining a field of the original document by comparing the original document with a standard document for each field stored in a database; (c) generating a primary translation configured of a second language for the original document by using the general-purpose neural network machine translator; and (d) converting some expressions included in the primary translation into expressions in another second language with reference to a professional translation dictionary for the determined field to generate a secondary translation, wherein the expression includes any one of a word, a phrase, and a sentence.

Abstract: Method, system and product for performing private and non-private tasks in Augmented Reality (AR) systems. A privacy policy of a user using an AR device is obtained. A frame sensor is utilized to obtain frames for processing. An overlay display for the frame is generated. The overlay display is generated based on an execution of a non-private task on a non-trusted device and based on an execution of a private task on a trusted device. The private task and the non-private task are determined based on the privacy policy. The overlay display is displayed by an AR device.

Abstract: Method and server for training an Attention-based Neural Network (ANN) to generate a textual output sequence being a content summary to be used as a response to a query. The ANN has an encoder and a decoder sub-network. The method includes, (i) inputting a query and input sequence into the encoder sub-network where the input sequence is a sequence of input groups associated with corresponding content snippets, (ii) generating an encoded representation of the input sequence, which includes generating attention-type outputs for respective words from the input sequence by applying an attention-limiting mask configured to attend only to words from the given input group, (iii) generating a decoded representation being a predicted textual output sequence, (iv) generating a penalty and (iii) adjusting the ANN based on the penalty score.

Abstract: An automated system and method of converting legacy decision logic to a target format. The legacy files are received by the decision logic translation system, which outputs the business rule content in a standard rule structure, according to the selected target format. The process involves decision logic-based rule extraction. In general, methods or processes for extracting business rules have been difficult to reproduce and do not present clearly the extracted rules regarding the concepts of business rules, their composition and categorization. These drawbacks lead to incomplete extraction of rules and massive manual effort to achieve a complete extraction and verification. In contrast, the proposed system overcomes these drawbacks, and outputs files that can be easily used to migrate the business rules to a new platform.

Abstract: An example system and method for selectively conveying content via User Interface (UI) display screen sections of Business Intelligence (BI) software and for facilitating user interaction therewith, e.g., so as to impart usefulness or preference ratings thereto, which can then affect subsequently displayed content. An example method includes providing one or more UI controls enabling a user to provide an input query and to interact with resulting content (i.e., query results, also called search results), where the user interaction with the content provides information (e.g., context information) indicative of a usefulness or preference rating pertaining to the content; thereby enabling modifications to subsequently retrieved content in accordance with the usefulness or preference rating(s). Such rating(s) may be aggregated from plural users of an associated software application.

Abstract: Systems and methods for using a generative artificial intelligence (AI) model using a multi-turn process to generate a suggested draft reply to a selected message. A first turn of the multi-turn process uses a shorter prompt including at least a portion of the body of the selected message and that requests multiple draft replies from the AI model. The resulting AI-generated draft replies are shortened, summarized, and/or otherwise converted into a plurality of shortened summaries that are presented as reply options to a user. Upon selecting a shortened summary, a more robust prompt is generated in a second turn with the AI model with the selected reply option to generate a more complex suggested draft reply to the selected message. Additionally, various customization options are provided, which when selected, reframe a query presented to the AI model to generate a more relevant and personalized response.

Abstract: The systems and methods described herein can generate a voice-based interface to increase the accuracy of translations. The voice-based interface can result in fewer input audio signals being transmitted between devices of a network. A method includes: receiving a first input audio signal; generating, based on the first input audio signal, a first translation string in a second language and a second translation string in a first language; determining a first translation score based on a likelihood that the first input audio signal includes an utterance in the first language and a second translation score based on a likelihood that the first input audio signal includes an utterance in the second language; selecting the first translation string based on the first translation score and the second translation score; generating an output signal from the first translation string; and transmitting the output signal to the client device.

Abstract: A computer-implemented method of processing language, includes: calculating a value indicating a degree of easiness of association between words with regard to one or more combinations of words in a first word group included in first text data and a second word group included in second text data; associating a first word and a second word based on the value indicating the degree of easiness of association; calculating a value indicating a degree of easiness of association between word groups with regard to a word group in the first text data including the first word and a word group in the second text data including the second word, using a result of the associating and the value indicating the degree of easiness of association between the words; and updating the value indicating the degree of easiness of association between the words by using the value indicating the degree of easiness of association between the word groups.

Abstract: Certain aspects of the present disclosure provide techniques for orchestrating a user experience using natural language input. A user experience is orchestrated within an ecosystem of features for which a plurality of corresponding tokens is defined. Natural language describing a desired user experience result is received by a user experience orchestrator. A sequence of tokens corresponding to operations belonging to an ecosystem of features which produce a correct result for the natural language input can be identified by a trained large language model and executed by the user experience orchestrator using a token operator. The output operations determined by the model to produce or be likely to produce the correct result based on the natural language input can be disambiguated, confirmed, and/or executed.

Abstract: Systems and methods for handling multilingual queries are provided. One example method includes receiving, at a computing device, an input, wherein the input comprises a multi-lingual query comprising at least a first source language and a second source language. The multi-lingual query is translated, word for word, into a destination language to produce a monolingual query, with the word order of the multilingual query and the word order of the monolingual query being the same. The monolingual query is processed using natural language processing to map the mono-lingual query to a natural language query in the destination language.

Abstract: Computer-readable media, systems and methods may improve automatic summarization of transcripts of financial earnings calls. For example, a system may generate segments, such as by disambiguating sentences, from a transcript to be summarized. The system may use an estimator that assesses whether or not the segment should be included in the summary. Different types of estimators may be used. For example, the estimator may be rule-based, trained based on machine-learning techniques, or trained on based on machine-learning with language modeling using natural language processing to fine-tune language models specific to financial earnings calls.

Abstract: A pharmacy management system includes a processor and a memory storing instructions that, when executed by processor, cause the pharmacy management system to train a machine learning model; receive a sig code utterance; analyze the sig code utterance using the trained machine learning model; and generate an output. A computer-implemented method includes training a machine learning model; receiving a sig code utterance; analyzing the sig code utterance using the trained machine learning model; and generating an output. A non-transitory computer readable medium includes program instructions that when executed, cause a computer to train a machine learning model; receive a sig code utterance; analyze the sig code utterance using the trained machine learning model; and generate an output.

Abstract: Aspects of the invention include a computer-implemented method including extracting a domain-specific object feature from a first image data, wherein the feature describes an object in the first image data. A domain-specific semantic meaning of text data is determined. The object feature is mapped to a portion of the text data, wherein the portion of the text data describes the object. A joint representation of the object and the portion of the text data is created. A second image data and a query directed towards an object in the second image data is received. An answer to the query is generated based on the joint representation.

Abstract: Described herein are systems, methods, and computer readable media for dynamically determining a language variant to use for vehicle output to a vehicle occupant based on the vehicle's location. A geographic region may include multiple sub-regions, each of which may be associated with a respective one or more language variants. As an example, a geographic region may be a state or province, and each sub-region may have one or more dialects that are spoken by individuals in that sub-region. In some cases, a particular dialect may be predominant in a given sub-region. As a vehicle traverses a travel path, it may determine its current location, which geographic sub-region includes that location, and which language variant (e.g., dialect) is predominant there. That language variant may then be selected for in-vehicle communication with a vehicle occupant. The vehicle location determination may be made at or near where the occupant entered the vehicle.