Abstract: Described herein are technologies pertaining to determining which search engine results page (SERP), from a plurality of SERPs, is preferable to a user. A query is received, and multiple SERPS are retrieved based upon the query, wherein the multiple SERPs are generated independently from one another. Values of features of the query and the multiple SERPs are obtained, and a determination as to which of the SERPs from the multiple SERPs is preferable to the user is made based upon the values of the features. The SERP determined to be preferable over other SERPs in the multiple SERPs is presented to the user.

Abstract: A computer-implemented technique is described herein for expediting a user's interaction with a digital assistant. In one implementation, the technique involves receiving a system prompt generated by a digital assistant in response to an input command provided by a user via an input device. The technique then generates a predicted response based on linguistic content of the system prompt, together with contextual features pertaining to a circumstance in which the system prompt was issued. The predicted response corresponds to a prediction of how the user will respond to the system prompt. The technique then selects one or more dialogue actions from a plurality of dialogue actions, based on a confidence value associated with the predicted response. The technique expedites the user's interaction with the digital assistant by reducing the number of system prompts that the user is asked to respond to.

Abstract: Automatic speech recognition systems can benefit from cues in user voice such as hyperarticulation. Traditional approaches typically attempt to define and detect an absolute state of hyperarticulation, which is very difficult, especially on short voice queries. This disclosure provides for an approach for hyperarticulation detection using pair-wise comparisons and on a real-world speech recognition system. The disclosed approach uses delta features extracted from a pair of repetitive user utterances. The improvements provided by the disclosed systems and methods include improvements in word error rate by using hyperarticulation information as a feature in a second pass N-best hypotheses rescoring setup.

Abstract: Systems and methods are disclosed for personalized communications using semantic memory. In one implementation, a first communication is received from a user and processed to identify a first content element within the communication. The first content element is associated with a second content element within a content repository. A second communication that includes the first content element is received from the user. Based on an association between the first content element and the second content element within the content repository, a third communication that includes the second content element is generated and provided to the user in response to the second communication.

Abstract: Developer and runtime environments supporting multi-modal input for computing systems are disclosed. The developer environment includes a gesture library of human body gestures (e.g., hand gestures) that a previously-trained, system-level gesture recognition machine is configured to recognize. The developer environment further includes a user interface for linking a gesture of the gesture library with a semantic descriptor that is assigned to a function of the application program. The application program is executable to implement the function responsive to receiving an indication of the gesture recognized by the gesture recognition machine within image data captured by a camera. The semantic descriptor may be additionally linked to a different input modality than the gesture, such as a natural language input.

Abstract: Cascaded models may be applied to extract facts from a medical text. A first model may be applied to at least a portion of the medical text. The first model extracts at least one first medical fact. The at least one first medical fact is linked to at least first text in the at least a portion of the medical text. A second model may be applied to the first text. The second model extracts at least one second fact that is an attribute of the at least one first medical fact.

Abstract: Systems and methods are disclosed for task identification and tracking using shared conversational context. In one implementation, a first communication from a first user is received within a communication session. The first communication is processed to identify a first content element within the first communication. A second communication is received within the communication session. The second communication is processed to identify a second content element within the second communication. The first content element is associated with the second content element. Based on an association between the first content element and the second content element, a task is identified. An action is initiated with respect to the task.

Abstract: Developer and runtime environments supporting multi-modal input for computing systems are disclosed. The developer environment includes a gesture library of human body gestures (e.g., hand gestures) that a previously-trained, system-level gesture recognition machine is configured to recognize. The developer environment further includes a user interface for linking a gesture of the gesture library with a semantic descriptor that is assigned to a function of the application program. The application program is executable to implement the function responsive to receiving an indication of the gesture recognized by the gesture recognition machine within image data captured by a camera. The semantic descriptor may be additionally linked to a different input modality than the gesture, such as a natural language input.

Abstract: A computer-implemented technique is described herein for expediting a user's interaction with a digital assistant. In one implementation, the technique involves receiving a system prompt generated by a digital assistant in response to an input command provided by a user via an input device. The technique then generates a predicted response based on linguistic content of the system prompt, together with contextual features pertaining to a circumstance in which the system prompt was issued. The predicted response corresponds to a prediction of how the user will respond to the system prompt. The technique then selects one or more dialogue actions from a plurality of dialogue actions, based on a confidence value associated with the predicted response. The technique expedites the user's interaction with the digital assistant by reducing the number of system prompts that the user is asked to respond to.

Abstract: Systems and methods are disclosed for task identification and tracking using shared conversational context. In one implementation, a first communication from a first user is received within a communication session. The first communication is processed to identify a first content element within the first communication. A second communication is received within the communication session. The second communication is processed to identify a second content element within the second communication. The first content element is associated with the second content element. Based on an association between the first content element and the second content element, a task is identified. An action is initiated with respect to the task.

Abstract: A method for automatically cross-linking a plurality of APIs in an artificial intelligence (AI) graph structure comprises maintaining an AI graph structure defining a plurality of API-agnostic semantic entities, a plurality of function nodes, a plurality of input-adapter edges, and a plurality of output adapter edges. The method further comprises cross-linking a new function from a new API by computer-analyzing documentation of the new API with a natural language processing (NLP) machine in order to recognize the new function, and updating the AI graph data structure to include a new function node based on the new function.

Abstract: Cascaded models may be applied to extract facts from a medical text. A first model may be applied to at least a portion of the medical text. The first model extracts at least one first medical fact. The at least one first medical fact is linked to at least first text in the at least a portion of the medical text. A second model may be applied to the first text. The second model extracts at least one second fact that is an attribute of the at least one first medical fact.

Abstract: A method for automatically cross-linking a plurality of APIs in an artificial intelligence (AI) graph structure comprises maintaining an AI graph structure defining a plurality of API-agnostic semantic entities, a plurality of function nodes, a plurality of input-adapter edges, and a plurality of output adapter edges. The method further comprises cross-linking a new function from a new API by computer-analyzing documentation of the new API with a natural language processing (NLP) machine in order to recognize the new function, and updating the AI graph data structure to include a new function node based on the new function.

Abstract: Developer and runtime environments supporting multi-modal input for computing systems are disclosed. The developer environment includes a gesture library of human body gestures (e.g., hand gestures) that a previously-trained, system-level gesture recognition machine is configured to recognize. The developer environment further includes a user interface for linking a gesture of the gesture library with a semantic descriptor that is assigned to a function of the application program. The application program is executable to implement the function responsive to receiving an indication of the gesture recognized by the gesture recognition machine within image data captured by a camera. The semantic descriptor may be additionally linked to a different input modality than the gesture, such as a natural language input.

Abstract: Described herein is a mechanism to adapt a machine learning model used in a language understanding model that has been trained using a first set of user input having a first set of features to effectively operate using user input having a second set of features. Losses are defined based on the first set of features, the second set of features or features common to both the first set and second set. The losses comprise one or more of a source side tagging loss, a reconstruction loss, an adversarial domain classification loss, a non-adversarial domain classification loss, an orthogonality loss, and target side tagging loss. The losses are jointly minimized using a gradient descent method and the resulting coefficients are used to retrain the machine learning model.

Abstract: Automatic speech recognition systems can benefit from cues in user voice such as hyperarticulation. Traditional approaches typically attempt to define and detect an absolute state of hyperarticulation, which is very difficult, especially on short voice queries. This disclosure provides for an approach for hyperarticulation detection using pair-wise comparisons and on a real-world speech recognition system. The disclosed approach uses delta features extracted from a pair of repetitive user utterances. The improvements provided by the disclosed systems and methods include improvements in word error rate by using hyperarticulation information as a feature in a second pass N-best hypotheses rescoring setup.

Abstract: Examples of the present disclosure describe systems and methods for using online signals to improve judgment quality in Side-by-Side (SBS) evaluation. In aspects, two or more search result lists may be accessed within a query log. The search result lists may be used to generate and/or determine satisfaction metrics between the search result lists. The satisfaction metrics may be aggregated to automatically generate preference judgments for the search result lists. In some aspects, the preference judgments may be compared to the preference judgments of judges to measure the judgment quality of the judges.

Abstract: Automatic speech recognition systems can benefit from cues in user voice such as hyperarticulation. Traditional approaches typically attempt to define and detect an absolute state of hyperarticulation, which is very difficult, especially on short voice queries. This disclosure provides for an approach for hyperarticulation detection using pair-wise comparisons and on a real-world speech recognition system. The disclosed approach uses delta features extracted from a pair of repetitive user utterances. The improvements provided by the disclosed systems and methods include improvements in word error rate by using hyperarticulation information as a feature in a second pass N-best hypotheses rescoring setup.

Abstract: Examples of the present disclosure describe systems and methods of identifying good and bad abandonment using gesture movement. In aspects, user feedback signals may be received by a client device in response to the viewing and/or navigation of query results. The feedback signals may be provided to a framework for determining and/or analyzing query abandonment. The framework may identify gesture data in the feedback signals and extract feature data from the gesture data. The feature data may be provided to a metrics component to determine metrics and/or satisfaction values for the feature data. The metrics and/or feature data may be used to train a predictive model to identify good abandonment in query results.

Abstract: Described herein is a conversation engine that can be used in a system such as a personal digital assistant or search engine that combines a dynamic knowledge graph built during execution of a request and one or more static knowledge graphs holding long term knowledge. The conversation engine comprises a state tracker that holds the dynamic knowledge graph representing the current state of the conversation, a policy engine that selects entities in the dynamic knowledge graph and executes actions provided by those entities to move the state of the conversation toward completion, and a knowledge graph search engine to search the static knowledge graph(s). The conversation is completed by building the dynamic knowledge graph over multiple rounds and chaining together operations that build toward completion of the conversation. Completion of the conversation results in completion of a request by a user.