Microsoft Technology Licensing, LLC

Paul Anthony Crook

Ruhi Sarikaya

Marius Alexandru Marin

Omar Zia Khan

Jean-Philippe Robichaud

Vipul Agarwal

Alexandre Rochette

Nikhil Holenarsipur Ramesh

Li Deng

Jianfeng Gao

Imed Zitouni

Hisami Suzuki

Daniel Boies

Ravi Kumar Bikkula

Xiujun Li

Vasiliy Radostev

Yun-Nung Vivian Chen

Xuesong Yang

Danko Panic

Conversational understanding systems allow users to conversationally interface with a computing device. In examples, a query may be received that includes a request for execution of a task. A data exchange task definition may be accessed. The data exchange task definition assists a conversational understanding system in managing task state tracking for information needed for task execution. Using the data exchange task definition, a per-turn policy for interacting with the user computing device is generated based on the state of a dialogue with a computing device and an evaluation of a process flow chart provided by a task owner resource. The task owner resource may be independent from the conversational understanding system. A response to the query may be generated and output based on the per-turn policy. In examples, the per-turn policy is used to generate one or more responses during a dialogue with a user via a computing device.

Policy authoring for task state tracking during dialogue

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.

Knowledge graph for conversational semantic search

Aspects herein provide third party application authors with a user interface authoring platform that automates and simplifies a task definition process while also providing the ability to leverage pre-existing language understanding models and canonicalization and resolution modules that are provided by the operating system on which the CU system resides or as provided by other third parties. In particular, the present disclosure provides a method and system for authoring a task using a user interface authoring platform.

Natural language task completion platform authoring for third party experiences

Systems and methods for building conversational understanding systems are provided. More specifically, the systems and methods utilize prebuilt conversational items that can form a CU system upon selection by a builder without requiring any further input from the builder. Accordingly, the systems and methods for building a conversational understanding system reduce the expertise, time, and resources necessary to build a conversational understanding system for an application when compared to systems and methods that utilize conversational items that require further input from the builder.

Systems and methods for building conversational understanding systems

A computer-implemented technique is described herein for training a personal digital assistant (PDA) component and a simulated user (SU) component via a self-learning strategy. The technique involves conducting interactions between the PDA component and the SU component over the course of plural dialogs, and with respect to plural tasks. These interactions yield training data. A training system uses the training data to generate and update analysis components used by both the PDA component and the SU component. According to one illustrative aspect, the SU component is configured to mimic the behavior of actual users, across a range of different user types.

Self-learning technique for training a PDA component and a simulated user component

Non-limiting examples of the present disclosure describe generation and application of a universal hypothesis ranking model to rank/re-re-rank dialog hypotheses. An input is received through a user interface of an application for dialog processing. A plurality of dialog hypotheses are generated based on input understanding processing of the received input. The plurality of dialog hypotheses are ranked using a universal hypothesis ranking model that is applicable to a plurality of languages and locales. The ranking of the plurality of dialog hypotheses comprises using the universal hypothesis ranking model to analyze language independent features of the plurality of dialog hypotheses for policy determination. Other examples are also described including examples directed to generation of the universal hypothesis ranking model.

Generation and application of universal hypothesis ranking model

Non-limiting examples of the present disclosure describe decoupling task state tracking that is managed by a shared task completion platform from execution of tasks by a task resource owner. Task registration data is received at a shared task state platform for a task that is executable by a task owner resource. Task registration data comprises parameters to be collected for execution of the task and ancillary information, such as the name of the task and whether to confirm the values of the parameters after collection. During interaction with a user, the shared task completion platform receives an input and determines the task is associated with the received input. During the interaction, parameters of the received task registration data are utilized to collect data for execution of the task. The collected data is transmitted to the task owner resource for execution of the task.

Task state tracking in systems and services

A processing unit can operate an end-to-end recurrent neural network (RNN) with limited contextual dialog memory that can be jointly trained by supervised signals-user slot tagging, intent prediction and/or system action prediction. The end-to-end RNN, or joint model has shown advantages over separate models for natural language understanding (NLU) and dialog management and can capture expressive feature representations beyond conventional aggregation of slot tags and intents, to mitigate effects of noisy output from NLU. The joint model can apply a supervised signal from system actions to refine the NLU model. By back-propagating errors associated with system action prediction to the NLU model, the joint model can use machine learning to predict user intent by a binary classification obtained by both forward and backward output, and perform slot tagging, and make system action predictions based on user input, e.g., utterances across a number of domains.

Joint language understanding and dialogue management using binary classification based on forward and backward recurrent neural network

In language evaluation systems, user expressions are often evaluated by speech recognizers and language parsers, and among several possible translations, a highest-probability translation is selected and added to a dialog sequence. However, such systems may exhibit inadequacies by discarding alternative translations that may initially exhibit a lower probability, but that may have a higher probability when evaluated in the full context of the dialog, including subsequent expressions. Presented herein are techniques for communicating with a user by formulating a dialog hypothesis set identifying hypothesis probabilities for a set of dialog hypotheses, using generative and/or discriminative models, and repeatedly re-ranks the dialog hypotheses based on subsequent expressions. Additionally, knowledge sources may inform a model-based with a pre-knowledge fetch that facilitates pruning of the hypothesis search space at an early stage, thereby enhancing the accuracy of language parsing while also reducing the latency of the expression evaluation and economizing computing resources.

Dialogue evaluation via multiple hypothesis ranking

Methods and systems are provided for collaborative completion of tasks using task frames. Upon receiving a request to perform a task, a system utilizes task frames in completing the requested task. A task frame is a data structure that contains the parameters and status signals that represent a particular task and captures the combined system's understanding of a current state of the task. Input is received at a client device and sent to a server, where the input is processed. Based on the processed input, a task frame is retrieved and filled. The filled task frame is sent to the client device, where the client device performs actions based on the task frame and updates the task frame parameters and the state of the task. The updated task frame is returned to the server. The shared task frame provides improvements to the overall task completion process.

Building multimodal collaborative dialogs with task frames

Systems and methods for responding to spoken language input or multi-modal input are described herein. More specifically, one or more user intents are determined or inferred from the spoken language input or multi-modal input to determine one or more user goals via a dialogue belief tracking system. The systems and methods disclosed herein utilize the dialogue belief tracking system to perform actions based on the determined one or more user goals and allow a device to engage in human like conversation with a user over multiple turns of a conversation. Preventing the user from having to explicitly state each intent and desired goal while still receiving the desired goal from the device, improves a user's ability to accomplish tasks, perform commands, and get desired products and/or services. Additionally, the improved response to spoken language inputs from a user improves user interactions with the device.

Growing community of inventors

Bellevue, WA, United States of America

Paul Anthony Crook