Abstract: Various embodiments of the disclosure provide an electronic device and a method of operating the same, the electronic device including: a speech recognition module; a memory configured to store information corresponding to a plurality of domains related to a collaborative task; and a processor operatively connected to the speech recognition module or the memory, wherein the processor is configured to receive a user voice from a user, analyze the received user voice using the speech recognition module to determine whether or not to perform a collaborative task, if the collaborative task is determined to be performed, select at least one participant related to the collaborative task, collect information related to the collaborative task from the user or an electronic device of the selected participant, and perform the collaborative task, based on the collected information. Various embodiments may be provided.

Abstract: An embodiment for identifying and replacing logically neutral phrases in natural language queries may include receiving a natural language query. The embodiment may also identify one or more logically neutral or non-logically neutral anchors in the natural language query. The embodiment may also identify boundaries containing one or more logically neutral phrases. The embodiment may further include detecting semantic and logical relations between verbal phrases and functional language between and adjacent to the one or more logically neutral and non-logically neutral anchors to reintroduce non-logically neutral language back into a non-logically neutral portion of the natural language query. The embodiment may also include generating a modified natural language query by automatically removing the boundaries and optionally replacing the one or more logically neutral phrases in the natural language query.

Abstract: A system, method and computer-readable storage device provides an improved speech processing approach in which hyper parameters used for speech recognition are modified dynamically or in batch mode rather than fixed statically. The method includes estimating, via a model trained on audio data and/or metadata, a set of parameters useful for performing automatic speech recognition, receiving speech at an automatic speech recognition system, applying, by the automatic speech recognition system, the set of parameters to processing the speech to yield text and outputting the text from the automatic speech recognition system.

Abstract: A clustered metasearch system receives a search query from a user. The system uses Natural Language Processing to identify an object of the search query and descriptors of the search query. The system sorts the search into an applicable realm based on the object of the search query. The system then conducts the search across a variety of search engines and collects root domains from the search results. Root domains within the same realm as the search query are prioritized and additional factors such as the presence of descriptors in the result, the recency of the result, the search engine rank of the result, and the distance from the center of the realm are used to determine the final ranking of the results. The results are then displayed to a user.

Abstract: Methods, systems, and computer program products for active learning for natural language question answering are provided herein. A computer-implemented method includes generating a semantic signature of a natural language query; generating a SQL signature for a SQL query corresponding to the natural language query; determining whether a set of mappings includes a semantic signature matching the generated semantic signature, wherein each mapping in the set is between (i) a given semantic signature and (ii) a SQL signature representing a class of SQL queries corresponding to the given semantic signature; and in response to determining that the set of mappings does not include the generated semantic signature, adding a mapping between the generated semantic signature and the generated SQL signature to the set of mappings.

Abstract: Techniques for using enhanced logit values for classifying utterances and messages input to chatbot systems in natural language processing. A method can include a chatbot system receiving an utterance generated by a user interacting with the chatbot system and inputting the utterance into a machine-learning model including a series of network layers. A final network layer of the series of network layers can include a logit function. The machine-learning model can map a first probability for a resolvable class to a first logit value using the logit function. The machine-learning model can map a second probability for a unresolvable class to an enhanced logit value. The method can also include the chatbot system classifying the utterance as the resolvable class or the unresolvable class based on the first logit value and the enhanced logit value.

Abstract: A system may determine relevance prompts based on input documents and a relevance prompt template and may transmit the plurality of relevance prompts to a large language model for completion. The system may receive response messages including chunk relevance scores. The system may select a subset of the input documents based on the chunk relevance scores. The system may determine query response prompts including text from the selected input documents the natural language query, and a second set of natural language instructions to address the natural language query. The system may determine a response to the natural language query based on answers determined in response to the query response prompts.

Abstract: Certain aspects of the present disclosure provide techniques for rendering visual artifacts in virtual worlds using machine learning models. An example method generally includes identifying, based on a machine learning model and a streaming natural language input, an intent associated with the streaming natural language input; generating, based on the identified intent associated with the streaming natural language input, one or more virtual objects for rendering in a virtual environment displayed on one or more displays of an electronic device; and rendering the generated one or more virtual objects in the virtual environment.

Abstract: Embodiments of the present disclosure relate to a solution for translation verification and correction. According to the solution, a neural network is trained to determine an association degree among a group of words in a source or target language. The neural network can be used for translation verification and correction. According to the solution, a group of words in a source language and translations of the group of words in a target language are obtained. An association degree among the group of words and an association degree among the translations can be determined by using the trained neural network. Then, whether there is a wrong translation can be determined based on the association degrees. In some embodiments, corresponding methods, systems and computer program products are provided.

Abstract: Various embodiments are provided for intelligent application of operational rules to operational data in a computing environment by a processor. One or more operational rules may be extracted and formalized from a knowledge graph, a domain knowledge, or a combination thereof describing one or more operational policies and conditions. The one or more operational rules may be applied to operational data to identify and filter non-compliant operational data.

Abstract: Systems and methods for using a generative artificial intelligence (AI) model to generate a suggested draft reply to a selected message. A message generation system and method are described that use inferred context to improve the suggested draft reply message for the user. Various message data and additional context are obtained and included in a prompt provided to the AI model to improve suggested content. In some examples, the message data and additional context include a message thread history and previously sent messages, profile information of the sender and recipient(s) of the selected message, known relationship information between the sender and the user, etc. For instance, the user's preferred communication style and talking points can be inferred based on the profile data, relationship data, and the user's past communications with similar participants and used to tailor the suggested draft reply to the user.

Abstract: The present disclosure relates to systems, non-transitory computer-readable media, and methods that generate a dynamic cross-platform ask interface and utilize a cross-platform language processing model to provide platform-specific, contextually based responses to natural language digital text queries. In particular, in one or more embodiments, the disclosed systems utilize machine learning models to extract registered intents from digital text queries to identify platform-specific configurations associated with the registered intents. Utilizing the platform-specific configurations, the disclosed systems can generate tailored platform-specific requests for information, as well as customized end-user search results that cause client devices to efficiently, accurately, and flexibly render platform-specific search results.

Abstract: Computing units provided at local sites or edge locations are programmed to execute conversational tools that generate pertinent, domain-specific responses to queries received from workers at such sites or locations. The conversational tools are large language models that are trained with domain-specific knowledge documents. Data representing queries are received from workers at such sites or locations and provided as inputs to the conversational tools along with text representing nearest knowledge documents from a knowledge base associated with the domain, as well as contextual data. Responses identified based on outputs received from the conversational tools in response to the inputs are provided to the workers that generated the queries.

Abstract: Systems and methods for pre-fetching results from large language models (LLMs) are provided. The method includes acquiring a context of an interaction between a user and an Artificial Intelligence (AI) character; predicting, based on the context, one or more anticipated words to be uttered by the user; generating, based on the one or more anticipated words, at least one query to an LLM; providing the at least one query to the LLM; generating, based on at least one response obtained from the LLM, an anticipated reply of the AI character model to the one or more anticipated words to be pronounced by the user; receiving one or more words uttered by the user; determining that a level of a discrepancy between the one or more words and the one or more anticipated words is below a predetermined threshold; and providing the anticipated reply to the user.

Abstract: Systems and methods are presented for inferring an embedding vector of an item of a first type into the embedding space. Upon receiving a first time for which there is no embedding vector, documents of a document corpus that include (co-occurrence) both the received item and other items of the same type are identified. Of those other items that have embedding vectors, those embedding vectors are retrieved and averaged. The resulting averaged embedding vector is established as an inferred embedding vector for the received item.

Abstract: Aspects discussed herein may relate to using machine learning models as part of methods and techniques for ingesting, creating, storing, editing, and managing a document. The document may be a legal contract that includes one or more clauses. Among other things, one or more machine learning models may be configured to recognize clauses and/or classifications, or types, of clauses. For example, the one or more generative language models may be used to generate one or more recommended edits to a clause, generate one or more suggested clauses that are missing from the contract, and/or generate one or more suggested locations where a clause may be inserted into or moved within the contract.

Abstract: A method, apparatus and computer program product for establishing a logical connection between an indirect utterance and a transaction is described. An indirect utterance is received from a user as an input to a conversational system. The indirect utterance is parsed to a first logical form. A first set of predicates and terms is mapped from the first logical form to a first subgraph in a knowledge graph. A second set of predicates and terms is mapped from a second logical form belonging to a transaction to a second subgraph of the knowledge graph. A best path in the knowledge graph between the first subgraph and the second subgraph is searched for while transforming the first logical form using the node and edge labels along the best path to generate an intermediate logical form. A system action is performed for a transaction if a graph structure of the intermediate logical form matches the graph structure of the logical form of the transaction above a threshold.

Abstract: Systems and methods for observation-based training of an Artificial Intelligence (AI) character model are provided. An example method includes receiving log data including interactions of a user and a first AI character model, receiving internal parameters of a second AI character model including a first plurality of heuristic machine learning models and a second plurality of primary machine learning models, pre-processing the log data to obtain one or more data streams including behavioral characteristics of the user, running the one or more data streams through the first plurality of heuristic machine learning models to produce intermediate outputs, composing the intermediate outputs into templated formats, and providing the templated formats to the second plurality of primary machine learning models. The internal parameters of the second AI character model are adjusted based on the templated formats such that the second AI character model mimics the behavioral characteristics of the user.

Abstract: A method for intelligent editing of legal documents is described. The method may include receiving a rule, accessing a plurality of documents, and generating a list of tokens as a function of the rule and the plurality of documents. The method may also include ranking each token of the list of tokens, wherein a ranking of each token is a function of a frequency of occurrence in the plurality of documents, accessing a user inputted legal text, and suggesting one or more alterations to the user inputted legal text as a function of the ranked list of tokens. An apparatus for intelligent editing of legal documents is also described.

Abstract: To improve prediction accuracy of utterance types in a dialog. A learning data generation device (10) according to the present invention comprises: a sort unit (11) configured to perform, based on information appended to utterances in a dialog amongst more than one speaker and that is indicative of a dialogue scene that is a scene in which the utterances in the dialog were made, sorting regarding whether the utterances are to be targets for generation of the learning data, wherein the sorter (11) is configured to exclude utterances of a dialogue scene that includes utterances similar to utterance of the particular type from the targets for generation of learning data.

Abstract: By formulizing a specific company's internal knowledge and terminology, the ontology programming accounts for linguistic meaning to surface relevant and important content for analysis. The ontology is built on the premise that meaningful terms are detected in the corpus and then classified according to specific semantic concepts, or entities. Once the main terms are defined, direct relations or linkages can be formed between these terms and their associated entities. Then, the relations are grouped into themes, which are groups or abstracts that contain synonymous relations. The disclosed ontology programming adapts to the language used in a specific domain, including linguistic patterns and properties, such as word order, relationships between terms, and syntactical variations. The ontology programming automatically trains itself to understand the domain or environment of the communication data by processing and analyzing a defined corpus of communication data.

Abstract: The invention concerns linguistic analysis. In particular the invention involves a method of operating a computer to perform linguistic analysis. In another aspect the invention is a computer system which implements the method, and in a further aspect the invention is software for programming a computer to perform the method.

Abstract: Certain aspects of the present disclosure provide techniques for executing structured query language queries having a schema associated therewith against an application programming interface using natural language. The schema can be chunked such that embeddings of the resulting chunks are stored in a vector store. Schemas (or subschemas) generated using on or more chunks of the vector store may be provided to a large language model along with a NL query to generate a structured query language query which may be executed against the application programming interface. This allows large language models to produce structured query language queries, such as GraphQL queries even if a GraphQL schema is too large to be provided to the model as context. Aspects disclosed herein also provide techniques for client code generation and client software development kit generation.

Abstract: Artificial intelligence (AI) technology can be used in combination with composable communication goal statements to facilitate a user's ability to quickly structure story outlines using “explanation” communication goals in a manner usable by an NLG narrative generation system without any need for the user to directly author computer code. This AI technology permits NLG systems to determine the appropriate content for inclusion in a narrative story about a data set in a manner that will satisfy a desired explanation communication goal such that the narratives will express various ideas that are deemed relevant to a given explanation communication goal.

Abstract: A mechanism is provided in a data processing system for presentation delivery. The mechanism delivering a presentation content to a group of users and receives a plurality of questions concerning the presentation content from the group of users. The mechanism stores the plurality of questions in a question history database and clusters the plurality of questions in the question history database into one or more question clusters. The mechanism determines a topic for each of the one or more question clusters to form one or more question topics and generates feedback for updating the presentation content based on the one or more question topics.

Abstract: Automatic license plate recognition occurs when a light sensor that continually captures video detects motion as a vehicle is driven through a gate. The light sensor detects the vehicle and license plate in the video stream captured by the light sensor. An algorithm associated with the video stream of the light sensor is trained to detect license plates. The light sensor starts executing the recognition algorithm when it detects motion. Recognition of characters in the license plate is based upon an aggregation of several captured video frames in which a license plate is detected.

Abstract: The present disclosure relates to systems, non-transitory computer-readable media, and methods that utilize recurrent neural networks to determine the existence of one or more open intents in a text input, and then extract the one or more open intents from the text input. In particular, in one or more embodiments, the disclosed systems utilize a trained intent existence neural network to determine the existence of an actionable intent within a text input. In response to verifying the existence of an actionable intent, the disclosed systems can apply a trained intent extraction neural network to extract the actionable intent from the text input. Furthermore, in one or more embodiments, the disclosed systems can generate a digital response based on the intent identified from the text input.

Abstract: A device and method for determining at least one part of a knowledge graph. A body of text is made available; for one sentence from the body of text, a first, second, and third input respectively for a first, second, and third classifier is determined. Each of the first, second, and third inputs includes a numerical representation of at least one part of the sentence. A numerical representation of a first probability is determined by the first classifier as a function of the first input, which indicates whether or not the sentence relates to the knowledge graph. If the numerical representation of the first probability satisfies a first condition, a numerical representation of a second probability is determined by the second classifier as a function of the second input, which defines a first type for the word from the sentence.

Abstract: Implementations relate to managing multimedia content that is obtained by large language model(s) (LLM(s)) and/or generated by other generative model(s). Processor(s) of a system can: receive natural language (NL) based input that requests multimedia content, generate a response that is responsive to the NL based input, and cause the response to be rendered. In some implementations, and in generating the response, the processor(s) can process, using a LLM, LLM input to generate LLM output, and determine, based on the LLM output, at least multimedia content to be included in the response. Further, the processor(s) can evaluate the multimedia content to determine whether it should be included in the response. In response to determining that the multimedia content should not be included in the response, the processor(s) can cause the response, including alternative multimedia content or other textual content, to be rendered.

Abstract: An analysis device includes processing circuitry configured to perform parsing on a first character string based on a grammar described in a PEG in which a variable is associated with a predetermined terminal symbol, add, to the variable, an element in which a predetermined attribute is imparted to a part of the first character string, the part being a second character string analyzed as corresponding to the terminal symbol, extract an element that is latest from elements of each predetermined attribute from the variable, and determine whether the element extracted satisfies a predetermined condition regarding a context.

Abstract: Systems and methods for video processing are described. Embodiments of the present disclosure generate a plurality of image feature vectors corresponding to a plurality of frames of a video; generate a plurality of low-level event representation vectors based on the plurality of image feature vectors, wherein a number of the low-level event representation vectors is less than a number of the image feature vectors; generate a plurality of high-level event representation vectors based on the plurality of low-level event representation vectors, wherein a number of the high-level event representation vectors is less than the number of the low-level event representation vectors; and identify a plurality of high-level events occurring in the video based on the plurality of high-level event representation vectors.

Abstract: Systems, methods and non-transitory computer readable media for prompt-based attribution of generated media contents to training examples are provided. In some examples, a first media content generated using a generative model in response to a first textual input may be received. The generative model may be a result of training a machine learning model using a plurality of training examples. Each training example of the plurality of training examples may include a respective textual content and a respective media content. Properties of the first textual input and properties of the textual contents included in the plurality of training examples may be used to attribute the first media content to a first subgroup of the plurality of training examples. The training examples of the first subgroup may be associated with a source. Further, a data-record associated with the source may be updated based on the attribution.

Abstract: A conversation augmentation system can automatically augment a conversation with content items based on natural language from the conversation. The conversation augmentation system can select content items to add to the conversation based on determined user “intents” generated using machine learning models. The conversation augmentation system can generate intents for natural language from various sources, such as video chats, audio conversations, textual conversations, virtual reality environments, etc. The conversation augmentation system can identify constraints for mapping the intents to content items or context signals for selecting appropriate content items. In various implementations, the conversation augmentation system can add selected content items to a storyline the conversation describes or can augment a platform in which an unstructured conversation is occurring.

Abstract: Method and system to control a conversational bot uses a directed acyclic graph to specify a desired conversation flow. A graph node has synthetic conversation transcripts annotated with events, wherein an event in a synthetic conversation transcript has preconfigured event expressions that represent ways in which dialogue at the node can unfold. During an on-going conversation with an actor, the system provides a data model uniquely associated with the conversation and that specifies a linear sequence of observations. The data model includes events representing semantically-related conversation fragments located in annotated historical conversation transcripts. In response to receipt of an input in association with a current graph node, the system determines whether the input extends an event in the synthetic conversation transcript associated with the node. If so, a response that continues a current conversation flow in the graph is provided.

Abstract: A method includes receiving audio data encoding an utterance spoken by a native speaker of a first language, and receiving a biasing term list including one or more terms in a second language different than the first language. The method also includes processing, using a speech recognition model, acoustic features derived from the audio data to generate speech recognition scores for both wordpieces and corresponding phoneme sequences in the first language. The method also includes rescoring the speech recognition scores for the phoneme sequences based on the one or more terms in the biasing term list, and executing, using the speech recognition scores for the wordpieces and the rescored speech recognition scores for the phoneme sequences, a decoding graph to generate a transcription for the utterance.

Abstract: Systems and methods for natural language processing are described. Embodiments of the inventive concept are configured to receive an input sequence and a plurality of candidate long forms for a short form contained in the input sequence, encode the input sequence to produce an input sequence representation, encode each of the plurality of candidate long forms to produce a plurality of candidate long form representations, wherein each of the candidate long form representations is based on a plurality of sample expressions and each of the sample expressions includes a candidate long form and contextual information, compute a plurality of similarity scores based on the candidate long form representations and the input sequence representation, and select a long form for the short form based on the plurality of similarity scores.

Abstract: Aspects described herein may allow for the application of stochastic gradient boosting techniques to the training of deep neural networks by disallowing gradient back propagation from examples that are correctly classified by the neural network model while still keeping correctly classified examples in the gradient averaging. Removing the gradient contribution from correctly classified examples may regularize the deep neural network and prevent the model from overfitting. Further aspects described herein may provide for scheduled boosting during the training of the deep neural network model conditioned on a mini-batch accuracy and/or a number of training iterations. The model training process may start un-boosted, using maximum likelihood objectives or another first loss function.

Abstract: Systems and methods for improved Spoken Language Understanding (“SLU”) are provided. The methods may comprise receiving an utterance from a user, contextualizing a plurality of words in the utterance, providing the contextualized words to the slot detector to determine the probability of a word forming the beginning or end of a slot to determine slots and nested slots, an intent classifier to determine the probability of a word conveying a user intent, and a slot classifier that applies specific labels to each slot and nest slot. The SLU method may employ a model and jointly trains the model for each task (determining beginning and end of slots, intents, and slot classifications) using a combined loss function.

Abstract: Embodiments described herein provide methods and systems for effectively and efficiently summarizing long documents. A transformer is provided with bottom-up and top-down inference combined to effectively capture long-range dependency. In the bottom-up inference, each token only attends to nearby tokens within a window of a specified size. In the top-down inference, full self-attention is given using units with coarser granularity. The bottom-up-inferred token representations are then updated with the top-down representations, which is achieved with cross-attention between the top and token levels. Multiple levels of top-down representations with increasingly coarser granularity can be used if documents are extremely long.

Abstract: Systems and methods for conversational user experiences and conversational database analysis disclosed herein improve the efficiency and accessibility of low-latency database analytics.

Abstract: Methods and systems for a multimodal conversational system are described. A method for interactive multimodal conversation includes parsing multimodal conversation from a physical human for content, recognizing and sensing one or more multimodal content from the parsed content, identifying verbal and non-verbal behavior of the physical human from the one or more multimodal content, generating learned patterns from the identified verbal and non-verbal behavior of the physical human, training a multimodal dialog manager with and using the learned patterns to provide responses to end-user multimodal conversations and queries, and training a virtual human clone of the physical human with interactive verbal and non-verbal behaviors of the physical human, wherein appropriate interactive verbal and non-verbal behaviors are provided by the virtual human clone when providing the responses to the end-user multimodal conversations and queries.

Abstract: Methods, systems, and computer programs are presented for providing access to a cloud data platform including a machine learning model for performing a plurality of iterations, by at least one hardware processor, to generate a Natural Language Processing (NLP) model. The cloud data platform performs each iteration by receiving real-world documents and enabling information retrieval from the real-world documents without annotated training data. Each iteration includes receiving data comprising text data, layout data, and image data and analyzing the text data, the layout data, and the image data. The cloud data platform generates one or more outputs from the machine learning model by applying the iterative training on new data, based at least in part on the analyzing of the text data, the layout data, and the image data.

Abstract: Disclosed embodiments relate to natural language processing. Techniques can include receiving input text, extracting, from the input text, at least one modifier and aspect pair, receiving data from a knowledgebase, based on the at least one modifier and aspect pair and commonsense data, generate one or more premise embeddings, convert the input text into tokens, generating at least one vector for one or more of the tokens based on an analysis of the tokens, combine the at least one vector with the one or more premise embeddings to create at least one combined vector, and analyze the at least one combined vector wherein the analysis generates an output indicative of a feature of the input text.

Abstract: A data query method supporting a natural language, an open platform, and a user terminal are provided, where the method includes: receiving, by the open platform, a natural language query statement sent by the user terminal, and transforming the natural language query statement into a query statement that is recognizable by a third-party content provider; sending, by the open platform, the query statement that is recognizable to one or more third-party content providers; receiving, by the open platform, one or more query response messages returned by the one or more third-party content providers according to the query statement that is recognizable; and sending, by the open platform, the one or more query response messages to the user terminal.

Abstract: A target set of texts, for training and/or evaluating a text classification model, is augmented using insertions into a base text within the original target set. In an embodiment, an expanded text, including the base text and an insertion word, must satisfy one or more inclusion criteria in order to be added to the target set. The inclusion criteria may require that the expanded text constitutes a successful attack on the classification model, the expanded text has a satisfactory perplexity score, and/or the expanded text is verified as being valid. In an embodiment, if a number of expanded texts added into the target set is below a threshold number, insertions are made into an expanded text (which was generated based on the base text). Inclusion criteria are evaluated against the doubly-expanded text to determine whether to add the doubly-expanded text to the target set.

Abstract: Provided are a semantic classification method and apparatus, a neural network training method and apparatus and a storage medium.

Abstract: Embodiments of the present disclosure use natural language processing, machine learning and relevant corpora to detect social engineering attacks with a high degree of accuracy. In various embodiments, lexical features, spelling features and topical features are automatically analyzed from a source text and a model is employed to assess the likelihood that the source message is a social engineering attack.

Abstract: A hypothesis stitcher for speech recognition of long-form audio provides superior performance, such as higher accuracy and reduced computational cost. An example disclosed operation includes: segmenting the audio stream into a plurality of audio segments; identifying a plurality of speakers within each of the plurality of audio segments; performing automatic speech recognition (ASR) on each of the plurality of audio segments to generate a plurality of short-segment hypotheses; merging at least a portion of the short-segment hypotheses into a first merged hypothesis set; inserting stitching symbols into the first merged hypothesis set, the stitching symbols including a window change (WC) symbol; and consolidating, with a network-based hypothesis stitcher, the first merged hypothesis set into a first consolidated hypothesis.

Abstract: The present disclosure relates to systems, non-transitory computer-readable media, and methods for utilizing machine learning models to determine predicted client intent classifications and/or client-agent escalation classes to generate personalized digital text reply options within an automated interactive digital text thread. For example, disclosed systems utilize the machine learning model to generate predicted client-agent escalation classes and corresponding probabilities. The disclosed systems utilize the predicted client-agent escalation classifications and the escalation class probabilities to generate personalized digital text reply options. Moreover, the disclosed systems can provide personalized digital text reply options to a client device within an automated interactive digital text thread, bypassing the inefficiency of menu options or protocols utilized to guide clients to terminal information.

Abstract: A method for text generation, relates to a field of natural language processing, including: obtaining corpus data; labeling the corpus data to obtain a first constraint element; obtaining a first generation target; and generating a first text matching the first generation target by inputting the corpus data and the first constraint element into a generation model.