Abstract: Embodiments of the present disclosure include systems and methods for packing tokens to train sequence models. In some embodiments, a plurality of datasets for training a sequence model is received. Each dataset in the plurality of datasets includes a sequence of correlated tokens. A set of training data is generated that includes a subset of a sequence of tokens from a first dataset in the plurality of datasets and a subset of a sequence of tokens from a second, different dataset in the plurality of datasets. The sequence model is trained using the set of training data.

Abstract: Disclosed herein are system, apparatus, article of manufacture, method and/or computer program product embodiments, and/or combinations and sub-combinations thereof, for generating a scene emotion value for a scene based on a sequence of frame emotion values for a sequence of frames within the scene of a content. The content can include multiple scenes, and a scene can include multiple frames, where a frame emotion value can be generated for each frame. A frame emotion value can be generated based on scene metadata related to the scene, content metadata related to the content, and a frame metadata related to the frame.

Abstract: A technique is described herein for assisting a user in editing a file. The technique involves producing current context information that includes an input message and selected file content. The input message describes a user's editing objective, while the selected file content describes a portion of the file to which the editing objective is to be applied. The technique then requests a pattern-completion engine to generate edit information based on the current context information. The edit information describes one or more changes to the selected file content that satisfy the objective of the user. The pattern-completion engine uses a machine-trained autoregressive text-completion model that is trained on revision history information. The model can be trained in a process that incorporates various tests to ensure that the edit information that is generated works as expected, satisfies various performance metrics, and fulfills the editing objectives of the user.

Abstract: An information processing apparatus according to the present disclosure includes a reception unit that receives pre-training data that is data used for pre-training in machine learning, and a search condition for similar pre-training data that is data similar to the pre-training data, a search unit that searches for similar pre-training data in accordance with the search condition, and a generation unit that performs pre-training based on the retrieved similar pre-training data, and generates a trained model by using a result obtained through the pre-training.

Abstract: A categorization system can include a computing device that is configured to obtain a plurality of data items over a threshold analysis period from an incoming data database in response to a threshold analysis interval elapsing. The computing device can also be configured to select a categorization model from a model database. The computing device can also be configured to, for each data item of the plurality of data items, apply the categorization model to the data item to identify at least one topic associated with the corresponding data item. The computing device can also be configured to generate a categorization visualization indicating a frequency of data items corresponding to each topic. The computing device can also be configured to transmit the categorization visualization to at least one of: (i) a user interface of an analyst device and (ii) a categorized database.

Abstract: Computer systems and processes configured to collect empirical data from a plurality of observations of a person, and to analyze the data to identify a particular state of the person characterized by at least a particular property selected from the group consisting of types of behaviors, types of actions, types of activities, and/or types of emotions. The computer system facilitates transmission of a digital message, the content of which may be determined in response to the instance of the one particular state identified. The content of some digital messages may include experiments performed by the computer system on the person, to test the validity of the state-identification-process. The state-identification-process can then be updated with the observed responses of the person to the experiments, and with the results of the experiments. These experiments and the updating of the state-identification-process might be performed by the computer system to autonomously refine the state-identification-process.

Abstract: An apparatus for generating an emoji includes an analyzer that is configured to, in response to an utterance of a user being input, acquire at least one of context information or information about a sentiment of the user. AN emoji generator generates a new emoji based on emoji generation information including information related to at least two among the context information, the information about the sentiment of the user, and information about an intent of the user corresponding to the utterance. The emoji generator is configured to select a plurality of emojis that match the emoji generation information from among a plurality of stored emojis and combine at least some of the plurality of emojis to generate the new emoji.

Abstract: A system determines a baseline cyberthreat-risk score for a user, and displays the baseline cyberthreat-risk score via a user interface. The system presents at least one cyberthreat-education activity via the user interface, and receives, via the user interface, at least one user input associated with the presented at least one cyberthreat-education activity. The system generates an updated cyberthreat-risk score at least in part by updating the baseline cyberthreat-risk score based at least in part on the user input, and displays the updated cyberthreat-risk score via the user interface.

Abstract: Disclosed systems relate to generating questions from text. In an example, a method includes forming a first semantic tree from a first reference text and second semantic tree from a second reference text. The method includes identifying a set of semantic nodes that are in the first semantic tree but not in the second semantic tree. The method includes forming a first syntactic tree for the first reference text and a second syntactic tree for the second reference text. The method includes identifying a set of syntactic nodes that are in the first syntactic tree but not in the second syntactic tree. The method includes mapping the set of semantic nodes to the set of syntactic nodes by identifying a correspondence between a semantic node and a syntactic node, forming a question fragment from a normalized word, and providing the question fragment to a user device.

Abstract: A method for autonomously identifying symptom terms in free running text data includes the acts of defining a plurality of symptom terms associated with a particular pathology or therapeutic substance or procedure, labeling in a text data set any defined symptom terms and associating a tag indicating any of a positive, negative, or other status with relation to the labeled symptom term, and processing with a natural language processing algorithm multiple different subsets of the text data containing labeled symptom terms to identify a frequency of occurrence of a symptom term and to improve identification accuracy.

Abstract: Systems and methods are provided that train a machine-learned language encoding model through the use of a contrastive learning task. In particular, the present disclosure describes a contrastive learning task where the encoder learns to distinguish input tokens from plausible alternatives. In some implementations, on each training example the proposed method masks out some subset (e.g., 15%) of the original input tokens, replaces the masked tokens with samples from a “generator” (e.g., which may be a small masked language model), and then trains the encoder to predict whether each token comes from the original data or is a replacement produced by the generator.

Abstract: Techniques described herein relate to facilitating end-to-end multilingual communications with automated assistants. In various implementations, speech recognition output may be generated based on voice input in a first language. A first language intent may be identified based on the speech recognition output and fulfilled in order to generate a first natural language output candidate in the first language. At least part of the speech recognition output may be translated to a second language to generate an at least partial translation, which may then be used to identify a second language intent that is fulfilled to generate a second natural language output candidate in the second language. Scores may be determined for the first and second natural language output candidates, and based on the scores, a natural language output may be selected for presentation.

Abstract: Sustainability measurement is critical to determine whether industry performance is heading in intended direction. State of the art systems in the field of sustainability measurement fail to consider many parameters which are indicative of the sustainability of industries. The disclosure herein generally relates to industry monitoring, and, more particularly, to a method and system for sustainability measurement in an industrial environment. The system calculates similarity score which indicates similarity between different sentences and indicators, and used the calculated similarity scores and extracted features to classify the sentences as belonging to specific classes. This information is in turn used for measuring sustainability of organization from which input data have been collected.

Abstract: Techniques are disclosed for generating anomaly scores for a neuro-linguistic model of input data obtained from one or more sources. According to one embodiment, generating anomaly scores includes receiving a stream of symbols generated from an ordered stream of normalized vectors generated from input data received from one or more sensor devices during a first time period. Upon receiving the stream of symbols, generating a set of words based on an occurrence of groups of symbols from the stream of symbols, determining a number of previous occurrences of a first word of the set of words, determining a number of previous occurrences of words of a same length as the first word, and determining a first anomaly score based on the number of previous occurrences of the first word and the number of previous occurrences of words of the same length as the first word.

Abstract: Systems and methods for detecting and using semantic relatedness to classify segments of digital text are disclosed. More particularly, embodiments determine the semantic relatedness of segments of text to abstract categories where the abstract categories are not defined by a single word or semantic concept. Detecting semantic relatedness includes analyzing text, embedding the text, and determining semantic relatedness to a set of concepts for a category where each concept may include a set of words/phrases embedded in a similar fashion. The text embedding can be projected onto each concept embedding and reduced to a score representing semantic relatedness. The text is classified based on the semantic relatedness.

Abstract: A computer system for processing unstructured data, the computing system comprising a computer processor, a computer memory operatively coupled to the computer processor and the computer memory having disposed within it computer program instructions that, when executed by the processor, cause the computing system to carry out the steps of receiving unstructured data input from a client device, analyzing the unstructured data for features that satisfy logical segment criteria by using natural language processing (NLP), partitioning the unstructured data into logical segments based on satisfaction of the logical segment criteria, and linking data from a repository to the unstructured data based on the logical segments.

Abstract: Improved intelligent personal assistant (IPA) software agents are disclosed that are configured to interact with various people, service providers, files, and/or smart devices. More particularly, this disclosure relates to an improved Natural Language Processing (NLP) Intent Determination Service (IDS) that is able to determine the likely best action to take in response to generic user commands and queries. The improved NLP IDS disclosed is said to be ‘search-based’ because, rather than attempt to parse incoming user commands and queries up front, the incoming user commands and queries are searched against a pre-generated database of exemplary user commands (e.g., having associated action or parsing identifiers) to determine the most relevant search result(s). The associated system actions and known grammar/parsing rules of the most relevant search result(s) may then be used to process the incoming user command or query—without having to actually parse the incoming user command or query from scratch.

Abstract: An estimation device (100), which is an estimation device that estimates a duration of a speech section, includes: a representation conversion unit (11) that performs representation conversion of a plurality of words included in learning utterance information to a plurality of pieces of numeric representation data; an estimation data generation unit (12) that generates estimation data by using a plurality of pieces of the learning utterance information and the plurality of pieces of numeric representation data; an estimation model learning unit (13) that learns an estimation model by using the estimation data and the durations of the plurality of words; and an estimation unit (20) that estimates the duration of a predetermined speech section based on utterance information of a user by using the estimation model.

Abstract: The information processing apparatus analyzes the results of modification performed by user for a recommended character string and determines whether there is an excess or a lack in units of predetermined character string areas in a recommended area corresponding to the recommended character string. In a case where it is determined that there is an excess or a lack as a result of the determination, a recommended character string that eliminates the excess or the lack is specified. Then, area information for specifying the recommended area, which is registered in a database, is updated so that the character string corresponding to the specified recommended area becomes the recommended character string.

Abstract: [Problem] Provided is a system that can objectively evaluate a person who makes a presentation (presenter) [Solution] A presentation evaluation system 1 includes: a voice analysis unit 3 that analyzes a content of a conversation, a presentation material related information storage unit 5 that stores information related to a presentation material, a keyword storage unit 7 that stores information related to a keyword in each page of the presentation material, a related term storage unit 9 that stores a related term of each keyword, and an evaluation unit 11 that evaluates the content of the conversation analyzed by the voice analysis unit 3 or a person who had the conversation.

Abstract: An operation method of a storage device includes receiving a first write request; adding the first write request to a first fragment; selecting at least “n” (e.g., at least two) streams among a plurality of pre-allocated streams when a size of the first fragment, when a size of the first fragment is >=a reference value, based on a cosine similarity between the first fragment and each of the pre-allocated streams; applying input information to a machine learning model to detect a first sequential stream associated with the first fragment from among the at least “n” streams; allocating a stream identifier of the first sequential stream to the first fragment; and storing write data included in the first fragment based on the stream identifier of the first sequential stream. The input information includes statistical information of at least one of the “n” streams and the first fragment.

Abstract: A body language system for determining a body language message of a living being in a context comprising an artificial intelligence system, said AI system running a computer program that: retrieves an image of said living being showing body language; labels said living being in said image, resulting in a labeled living being; determines said context from said image using a trained machine learning model; determines a baseline body language of said labeled living being from said image using a trained machine learning model; adapts a trained machine learning model of said AI system using said baseline body language and said context; applies the adapted trained machine learning model of said AI system to the one image for categorizing said body language resulting in a category, and applying said category for determining said body language message.

Abstract: Described is a system for producing a causal map from a body of text. The system receives multiple textual documents as input. Pairs of cause-effect phrases are extracted from the textual documents and embedded into a vector space. The embedded data is clustered into clusters using a probabilistic technique. A causal map having nodes and edges is generated from the clusters. Using the causal map, causal connections between clusters are obtained, where each node represents an event and each edge represents a causal relationship between events. The causal map is provided as an interactive graph.

Abstract: One embodiment provides a system for processing natural-language entries. The system obtains a plurality of historical natural-language entries associated with a first domain and pre-processes the historical natural-language entries to obtain a set of generic terms and a set of domain-specific terms. The system trains a machine learning model in the first domain using the plurality of historical natural-language entries associated with the first domain. The training comprises learning weight values of one or more generic terms, a weight value of a respective generic term indicating likelihood that the generic term is related to a trigger event. The system generalizes the machine learning model trained in the first domain, thereby allowing the model to be applied to a second domain.

Abstract: An item recall method includes: behavior data is acquired, where the behavior data includes items and item information of each item; target behavior data containing a retrieval category word is extracted from the behavior data; retrieval words of each item and a retrieval frequency of each retrieval word are acquired in a reverse correlation manner; word segmentation is performed on the item information to obtain multiple item segmented words; a similarity between all retrieval words and the multiple item segmented words is calculated; whether the similarity is greater than a first preset threshold or not is determined, and if yes, then a retrieval word is extracted as an expansion word of the retrieval category word; and item recall is performed according to the retrieval category word and the expansion word.

Abstract: This disclosure provides systems, methods, and media for creating a data graph database from various unstructured and unstructured data items for use by various services. The method comprises the operations of identifying unstructured data items in data subjects; recognizing regions of interest (ROIs) in the unstructured data items; and extracting the ROIs from the unstructured data items. The method further comprises encoding the extracted ROIs into ROI vectors; creating a data graph to represent the data subjects, the data items, and the ROI vectors; and storing the data graph into a graph database. The various embodiments can manage data items of different data formats together rather than separately, thus creating a data management system for managing data across data formats. The data management system can also store structured data items into the graph database, thus complementing the existing ETL procedure for structured data items.

Abstract: Apparatus for generating a putative ontology from a data structure associated with a data store, the apparatus including an electronic processing device that generates a putative ontology by determining at least one concept table in the data structure, determining at least one validated attribute within the at least one concept table, determining at least one selected attribute value from the at least one validated attribute and generating at least one ontology class using the at least one attribute value.

Abstract: A computerized method for extracting domain specific insights from a corpus of files containing large documents comprising: breaking down large chunks of text into smaller sentences/short paragraphs in a domain specific way, identifying and removing domain noise; identifying the sentence intents of the non-noise sentences; tagging the sentences with other domain specific attributes; defining a semantic ontology using a graph database based on the sentence intents, a multitude of mini-dictionaries and domain attributes; applying a pre-defined ontology to tag documents with domain specific hashtags; and combining the hashtags using machine learning techniques into insights.

Abstract: A computer-implemented method according to one embodiment includes receiving, at a deep neural network (DNN), a plurality of sentences each having an associated label; training the DNN, utilizing the plurality of sentences and associated labels; and producing a linguistic expression (LE) utilizing the trained DNN.

Abstract: Methods and systems for speech recognition in an aircraft are disclosed. Methods and systems include executing an air traffic control transcription application using first acoustic and language models and executing a command and control speech recognition application using second acoustic and language models. Flight context data is processed to identify additional words not included in training of the second acoustic and language model but included in training of the first acoustic and language models. Acoustic and language model parameters are extracted corresponding to the additional words from the first acoustic and language models. The extracted acoustic and language model parameters are added to the second acoustic and language models. An aircraft control command is generated that encapsulates at least one of the additional words using the command and control speech recognition application.

Abstract: Described are methods and systems are for generating dynamic conversational queries. For example, as opposed to being a simply reactive system, the methods and systems herein provide means for actively determining a user's intent and generating a dynamic query based on the determined user intent. Moreover, these methods and systems generate these queries in a conversational environment.

Abstract: A text input system is described for inputting text to a computing device. The text input system has a memory storing a composing region comprising a plurality of text items selected by a user for potential input into the computing device. The text input system has a composing region updater which detects one of the plurality of text items as being a designated symbol. The composing region updater is configured to detect a corrective action acting to correct associated text associated with the selected text items and, when the corrective action is detected, to return the plurality of text items including the designated symbol to the composing region.

Abstract: A translation model training method for a computer device includes obtaining a training sample set, the training sample set including a plurality of training samples. Each training sample is a training sample pair having a training input sample in a first language and a training output sample in a second language. The method also includes determining a disturbance sample set corresponding to each training sample in the training sample set, the disturbance sample set comprising at least one disturbance sample, and a semantic similarity between the disturbance sample and the corresponding training sample being greater than a first preset value; and training an initial translation model by using the plurality of training samples and the disturbance sample set corresponding to each training sample to obtain a target translation model, such that the training output sample remains same for the disturbance sample and the corresponding training sample.

Abstract: Systems and methods for e-commerce systems using natural language understanding are described. A computing device is configured receive a user utterance including at least one identified semantic component and at least one missing semantic component and generate a context stack including a set of context entries. Each of the context entries includes a root intent element, an entity list element, and a dialogue stack and each context entry in the set of context entries is associated with one of a user utterance or a system utterance. The computing device is further configured to review at least one context entry in the set of context entries to locate the at least one missing semantic element within the dialogue stack and generate an intent flow execution request including the at least one semantic element from the first speech data and the missing semantic element.

Abstract: A digital human development platform can enable a user to generate a digital human. The digital human development platform can receive user input specifying a dialogue for the digital human and one or more behaviors for the digital human, the one or more specified behaviors corresponding with one or more portions of the dialog on a common timeline. Scene data can be generated with the digital human development platform by merging the one or more behaviors with one or more portions of the dialogue based on times of the one or more behaviors and the one or more portions of the dialog on the common timeline.

Abstract: Devices and techniques are generally described for process optimization using reinforcement learning. In various examples, first input data is received and a first process for processing the first input data is determined. In some examples, a second process for processing the first input data is determined. A first machine learning model is used to generate a first prediction for processing the first input data by the first process. The first process and/or the second process are controlled based at least in part on the first prediction.

Abstract: The present disclosure provides systems and methods for relationship extraction. Embodiments of the present disclosure provide a relationship extraction network trained to identify relationships among entities in an input text. The relationship extraction network is used to generate a dependency path between entities in an input phrase. The dependency path includes a set of words that connect the entities, and is used to predict a relationship between the entities. In some cases, the dependency path is related to a syntax tree, but it may include additional words, and omit some words from a path extracted based on a syntax tree.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for detecting a continued conversation are disclosed. In one aspect, a method includes the actions of receiving first audio data of a first utterance. The actions further include obtaining a first transcription of the first utterance. The actions further include receiving second audio data of a second utterance. The actions further include obtaining a second transcription of the second utterance. The actions further include determining whether the second utterance includes a query directed to a query processing system based on analysis of the second transcription and the first transcription or a response to the first query. The actions further include configuring the data routing component to provide the second transcription of the second utterance to the query processing system as a second query or bypass routing the second transcription.

Abstract: In an embodiment, the disclosed technologies are capable of receiving, by a digital model, data representing a first text sequence in a first language; using the digital model, modifying the first text sequence to result in creating and digitally storing a second text sequence in the first language; and outputting, by the digital model, the second text sequence in the first language. The modifying may include any one or more of: deleting text from the first text sequence, adding text to the first text sequence, modifying text of the first text sequence, reordering text of the first text sequence, adding a digital markup to the first text sequence. The digital model may have been fine-tuned, after having been machine-learned, using a subset of values of model parameters associated with an encoding layer or an embedding layer or both the encoding layer and the embedding layer.

Abstract: A computer-implemented method for generating quote-based search summaries from a plurality of documents includes receiving information identifying a meaning taxonomy, the meaning taxonomy including a normalized term and at least one syntactic structure that identifies an entity; locating, within at least one document of the plurality of documents, a statement attributable to the entity; receiving a search query comprising the normalized term; and displaying a summary of the at least one document, the summary including the statement.

Abstract: An electronic apparatus is disclosed.

Abstract: Methods, apparatus, and computer readable media are described related to automated assistants that proactively incorporate, into human-to-computer dialog sessions, unsolicited content of potential interest to a user. In various implementations, in an existing human-to-computer dialog session between a user and an automated assistant, it may be determined that the automated assistant has responded to all natural language input received from the user. Based on characteristic(s) of the user, information of potential interest to the user or action(s) of potential interest to the user may be identified. Unsolicited content indicative of the information of potential interest to the user or the action(s) may be generated and incorporated by the automated assistant into the existing human-to-computer dialog session.

Abstract: A method for providing candidate recommendations for an open position based on problem solving proficiency and culture matching includes: receiving job posting data from a first computing system, the job posting data including employer culture markers and job criteria values; receiving candidate information for a candidate, the candidate information including candidate culture markers, text submissions, and candidate values; determining a culture match score by comparing the employer culture markers against the candidate culture markers; determining a problem solving proficiency level for the candidate by applying a problem solving algorithm to the text submissions; and transmitting a hiring recommendation for the candidate to the first computing system based on the determined culture match score and problem solving proficiency level for the candidate and the criteria values.

Abstract: Techniques described herein relate to facilitating end-to-end multilingual communications with automated assistants. In various implementations, speech recognition output may be generated based on voice input in a first language. A first language intent may be identified based on the speech recognition output and fulfilled in order to generate a first natural language output candidate in the first language. At least part of the speech recognition output may be translated to a second language to generate an at least partial translation, which may then be used to identify a second language intent that is fulfilled to generate a second natural language output candidate in the second language. Scores may be determined for the first and second natural language output candidates, and based on the scores, a natural language output may be selected for presentation.

Abstract: Generating automated conversation responses by receiving a conversation input message, determining an intent associated with the conversation input message, detecting content associated with the intent in a data stream in response to determining the intent, and generating a conversation output according to the content and the intent.

Abstract: Methods and systems provide for dynamically generated topic segments for a communication session. In one embodiment, the system connects to a communication session with a number of participants; receives a list of topics; receives a transcript of a conversation between the participants produced during the communication session, the transcript including timestamps for a number of utterances associated with speaking participants; for each topic in the list of topics, segments the utterances into one or more topic segments based on the topic; for each of the segments, classifies whether the topic segment is related to the topic, and transmits, to one or more client devices, a list of the topic segments for the communication session.

Abstract: A method for redacting sensitive information from an audio stream, such as a voice signal in a telephone call, in real time is provided. The method includes: receiving an audio stream; conveying the audio stream through a channel that includes a valve; detecting, from within the audio stream, a first event that indicates an onset of sensitive information; closing the valve so that the conveying of the audio stream through the channel is temporarily stopped; detecting, from within the audio stream, a second event that indicates an ending of the sensitive information; and reopening the valve so that the conveying of the audio stream through the channel is resumed. The sensitive information may include payment card industry (PCI) information, such as a card number and/or a card verification value (CVV).

Abstract: This document relates to machine learning. One example includes a method or technique that can be performed on a computing device. The method or technique can include obtaining a task-semantically-conditioned generative model that has been pretrained based at least on a first training data set having unlabeled training examples and semantically conditioned based at least on a second training data set having dialog act-labeled utterances. The method or technique can also include inputting dialog acts into the semantically-conditioned generative model and obtaining synthetic utterances that are output by the semantically-conditioned generative model. The method or technique can also include outputting the synthetic utterances.

Abstract: A system and method are disclosed that enable rapid and cost-effective human-in-the-loop synthesis of domain-specific textual training data for a deep learning model. The data augmentation process incorporates a sentence generator, a sentence classifier, and weak-supervision by a domain expert that is ‘in the loop.’ Generally, both the sentence generator and the sentence classifier are implemented as machine learning models. The sentence generator generates new sentences based on manually labeled sentences and the sentence classifier generates labels for the newly generated sentences. The new sentences are corrected or verified by a domain expert and then used to retrain one or both of the sentence generator and the sentence classifier.

Abstract: Methods, systems, and apparatuses for improved speech recognition and transcription of user utterances are described herein. A user utterance may be processed by a speech recognition computing device. One or more acoustic features associated with the user utterance may be used to determine whether one or more actions are to be performed based on a transcription of the user utterance.