Abstract: A method includes performing, using at least one processor, feature extraction of input audio data to identify extracted features associated with the input audio data. The method also includes detecting, using the at least one processor, a language associated with each of multiple portions of the input audio data by processing the extracted features using a plurality of language models, where each language model is associated with a different language. In addition, the method includes directing, using the at least one processor, each portion of the input audio data to one of a plurality of automatic speech recognition (ASR) models based on the language associated with the portion of the input audio data.

Abstract: An intelligent question and answer (Q&A) system and method for interactively guiding users through a procedure is disclosed. The intelligent Q&A system can dynamically generate process trees (or procedural trees) from the content or procedures presented in a raw document, such as a reference manual. The intelligent Q&A system can include a virtual agent that uses the dynamically generated process trees for interactive conversation with a user. Using the system, the virtual agent can interactively guide users through completing tasks such as updating software or connecting an IoT device to an existing system.

Abstract: A method for performing hierarchical entity classification of an entity mention within a context, wherein ontological classes are computed for the entity mention levelwise using a contextual representation of the context and a state representation obtained by running an end-to-end trained decoding recurrent neural network on a mention representation of the entity mention.

Abstract: Computer-implemented method includes developing, via a processor, a words model from a plurality of natural language text based articles relating to a subject and generating, via the processor, a static vector based upon the words model. The computer-implemented method further includes developing, via the processor, an actual articles model from actual articles, generating, via the processor, a bootstrapped vector using the actual articles model, generating, via the processor, a n-dimensional depth item using the static vector and the bootstrapped vector, and determining, via the processor, evidence based on the n-dimensional depth item. The computer-implemented method still further includes presenting, via the processor and a display, the evidence base upon an input query from a user.

Abstract: A method of configuring a conversational computing interface. The method comprises maintaining a branching dialogue representing a plurality of multi-turn conversations each including a plurality of turns, wherein a turn indicates one or both of a conversational computing interface action and a conversation event, the branching dialogue including a shared prefix of turns common to the plurality of multi-turn conversations and at least two different counterfactual branches descending from the shared prefix of turns, each different counterfactual branch including one or more descendent turns corresponding to a different multi-turn conversation of the plurality of multi-turn conversations.

Abstract: Embodiments relate to a system, program product, and method directed at natural language (NL) and a virtual dialog platform. An NL statement is detected and analyzed to identify one or more entities expressed in the statement. The identified entities are leveraged to parse the statement into keywords. The intent of the received statement is represented as a relationship between two or more of the keywords. A knowledge representation is identified to represent the statement with respect to a formatted module having two or more components and a component relationship structure. Each statement keyword is assigned to a designated module component based on an alignment of the component relationship with the keyword relationship. The statement intent is expressed based on the relationship between the keywords, and a statement response is inferred. The inferred statement is communicated to the virtual dialog platform.

Abstract: This disclosure describes, in part, context-based device arbitration techniques to select a voice-enabled device from multiple voice-enabled devices to provide a response to a command included in a speech utterance of a user. In some examples, the context-driven arbitration techniques may include determining a ranked list of voice-enabled devices that are ranked based on audio signal metric values for audio signals generated by each voice-enabled device, and iteratively moving through the list to determine, based on device states of the voice-enabled devices, whether one of the voice-enabled devices can perform an action responsive to the command. If the voice-enabled devices that detected the speech utterance are unable to perform the action responsive to the command, all other voice-enabled devices associated with an account may be analyzed to determine whether one of the other voice-enabled devices can perform the action responsive to the command in the speech utterance.

Abstract: An artificial intelligence apparatus for recognizing speech of a user includes a microphone and a processor configured to acquire, via the microphone, first speech data including speech of a user, generate a first speech recognition result corresponding to the first speech data, perform control corresponding to the generated first speech recognition result, generate an alternative speech recognition result corresponding to the first speech data if negative feedback is acquired from the user, and perform control corresponding to the generated alternative speech recognition result.

Abstract: A significant correlation framework is provided herein for translating input commands to intents. The input commands may be natural language commands, received from a variety of input channels, which may be translated to intents or other runtime-bindable execution objects. The significant correlation framework may use interpreter nodes for translating the input commands by calculating the strength of correlation between an input command and an intent. The significant correlation framework may analyze the sequence of intents or the timing of translated intents to enhance the accuracy of the translation. The significant correlation framework may maintain a history of command translations, and may compare current translations against the history to improve accuracy of the translations. The significant correlation framework may switch between a depth-first mapping method and a breadth-first mapping method. Depth-first mapping may translate commands through a single interpreter node.

Abstract: A method for supplementing context in dialog flows for chatbot systems includes responsive to receiving identification information associated with a user and a topic for a chatbot conversation, initializing a chatbot conversation between a chatbot and the user. The method identifies a set of entities and a set of one or more relationships, wherein each relationship from the set one or more relationships is between two entities from the set of entities. The method determines an initial set of relevancy scores for each entity in the set of entities and the set of one or more relationships, wherein the initial set of relevancy scores are based at least on a domain for the topic of the chatbot conversation. The method generates a first response to the user based on the initial set of relevancy scores, wherein the first response includes at least one entity and at least one relationship.

Abstract: A computer-implemented method for answering natural language medical information questions posed by a user of a medical conversational interface of a cognitive artificial intelligence system, the method comprising: receiving from a medical conversational user interface a user-generated natural language medical information query at an artificial intelligence-based medical conversation cognitive agent; extracting a medical question from the query; compiling a medical conversation language sample; extracting internal medical concepts and medical data from the sample, where the internal medical concepts include descriptions of medical attributes of the medical data entities; inferring a therapeutic intent of the user; generating a therapeutic paradigm logical framework for interpreting of the medical question, wherein logical framework comprises a catalog of medical logical progression paths from the medical question to respective therapeutic answers, each of the logical progression paths includes one or more med

Abstract: A system includes a processor and a memory storing instructions that, when executed by the processor, cause the system to receive an utterance, transmit the utterance to a cloud to generate an intent and an entity, receive the intent and the entity, and perform an action with respect to a graphical user interface. A method includes receiving an utterance, transmitting the utterance to a cloud to generate an intent and an entity, receiving the intent and the entity, and performing an action with respect to a graphical user interface. A non-transitory computer readable medium includes program instructions that when executed, cause a computer to receive an utterance, transmit the utterance to a cloud to generate an intent and an entity, receive the intent and the entity, and perform an action with respect to a graphical user interface.

Abstract: Methods, apparatus, and computer readable media related to receiving textual input of a user during a dialog between the user and an automated assistant (and optionally one or more additional users), and generating responsive reply content based on the textual input and based on user state information. The reply content is provided for inclusion in the dialog. In some implementations, the reply content is provided as a reply, by the automated assistant, to the user's textual input and may optionally be automatically incorporated in the dialog between the user and the automated assistant. In some implementations, the reply content is suggested by the automated assistant for inclusion in the dialog and is only included in the dialog in response to further user interface input.

Abstract: According to some embodiments, a multi-layer speech recognition transcript post processing system may include a data-driven, statistical layer associated with a trained automatic speech recognition model that selects an initial transcript. A rule-based layer may receive the initial transcript from the data-driven, statistical layer and execute at least one pre-determined rule to generate a first modified transcript. A machine learning approach layer may receive the first modified transcript from the rule-based layer and perform a neural model inference to create a second modified transcript. A human editor layer may receive the second modified transcript from the machine learning approach layer along with an adjustment from at least one human editor. The adjustment may create, in some embodiments, a final transcript that may be used to fine-tune the data-driven, statistical layer.

Abstract: A quantization device includes: a trellis-structured vector quantizer which quantizes a first error vector between an N-dimensional (here, “N” is two or more) subvector and a first predictive vector; and an inter-frame predictor which generates a first predictive vector from the quantized N-dimensional subvector, wherein the inter-frame predictor uses a predictive coefficient comprising an N×N matrix and performs an inter-frame prediction using the quantized N-dimensional subvector of a previous stage.

Abstract: A system and method for federated context-sensitive language models comprising a federated language model server and a plurality of edge devices. The federated language model server may comprise one or more machine learning models trained and developed centrally on the server, and distribute these one or more machine learning models to edge devices wherein they may be operated locally on the edge devices. The edge devices may gather or generate context data that can be used by a speech recognition engine, and the local language models contained therein, to develop adaptive, context-sensitive, user-specific language models. Periodically, the federated language model server may select a subset of edge devices from which to receive uploaded local model parameters, that may be aggregated to perform central model updates wherein the updated model parameters may then be sent back to edge devices in order to update the local model parameters.

Abstract: A computer-implemented process includes obtaining a natural-language-text document comprising a first and second clause and determining first and second embedding sequences based on n-grams of the first and second clauses. The process includes generating data model objects based on the embedding sequences and determining an association between the first data model object and the second data model object based on a shared parameter of the first and second clauses. The process includes receiving a query including the first category and the first n-gram and causing a presentation of a visualization of data model objects that includes shapes based on the data model objects and a third shape based on the association between the first data model object and the second data model object.

Abstract: Methods for classification of microblogs using semi-supervised open domain targeted sentiment classification. A hidden Markov model support vector machine (SVM HMM) is trained with a training dataset combined with discrete features. A portion of the training dataset is clustered by k-means clustering to generate cluster IDs which are normalized and combined with the discrete features. After formatting, the combined dataset is applied to the SVM HMM and the C parameter, which is optimized by calculating a zero-one error at each iteration. The open domain targeted sentiment classification methods uses less labelled data than previous sentiment analysis techniques, thus decreasing processing costs. Additionally, a supervised learning model for improving the accuracy of open domain targeted sentiment classification is presented using an SVM HMM.

Abstract: Embodiments of the present disclosure provide a query processing method and an apparatus, a server and a storage medium. The method includes: determining a word vector representation of a query sequence and an entity vector representation of the query sequence respectively based on respective words and respective entities included in the query sequence; determining a word vector representation of a paragraph and an entity vector representation of the paragraph respectively based on respective words and respective entities included in the paragraph; and determining a similarity between the query sequence and the paragraph according to the word vector representation of the query sequence, the entity vector representation of the query sequence, the word vector representation of the paragraph, and the entity vector representation of the paragraph.

Abstract: The present disclosure provides a method, device and system for detecting a working state of a tower controller, the method includes: collecting voice data of a tower controller, and extracting a keyword from the voice data; acquiring a video image of the tower controller, and acquiring a gaze area of the tower controller from the video image; analyzing and detecting whether the tower controller has correctly accomplished an observation action according to the gaze area of the tower controller and the keyword. The present disclosure implements more efficient and accurate detection on the working state of the tower controller, and at the same time ensures the safety of an aircraft in an airport area and reduces a risk of colliding with other obstacles.