CROWD SOURCING INFORMATION TO FULFILL USER REQUESTS

[0001] This application claims priority of U.S. Provisional Application Ser. No. 61/646,831, filed May 14, 2012, which is incorporated herein by reference in its entirety.

[0002] The disclosed embodiments relate generally to digital assistants, and more specifically, digital assistants that provide crowd-sourced responses to users' speech-based requests.

[0003] Just like human personal assistants, digital assistants or virtual assistants can perform requested tasks and provide requested advice, information, or services. An assistant's ability to fulfill a user's request is dependent on the assistant's correct comprehension of the request or instructions. Recent advances in natural language processing have enabled users to interact with digital assistants using natural language, in spoken or textual forms, rather than employing a conventional user interface (e.g., menus or programmed commands). Such digital assistants can interpret the user's input to infer the user's intent; translate the inferred intent into actionable tasks and parameters; execute operations or deploy services to perform the tasks; and produce output that is intelligible to the user. Ideally, the output produced by a digital assistant should fulfill the user's intent expressed during the natural language interaction between the user and the digital assistant.

[0004] The ability of a digital assistant system to produce satisfactory responses to user requests depends on the natural language processing, knowledge base, and artificial intelligence implemented by the system. At any time, a digital assistant may be limited by its particular implementation, however sophisticated that implementation may be, and fail to produce a satisfactory response to a user's request. A well-designed response procedure in such a situation can improve a user's experience in interacting with the system and prevent the user's loss of confidence in the system's service.

[0005] The embodiments disclosed herein provide methods, systems, computer readable storage medium and user interfaces for a digital assistant to crowd source assistance or information from one or more external information sources (so-called “crowd sourcing information sources” or “CS information sources”), and generate a response to a user request based on the crowd sourced information or assistance. These external information sources, for example, include expert information services, general information sources, and forums where answers to questions are provided in structured, semistructured, and unstructured forms by members of the public. In addition, crowd sourced information and answers can be stored, e.g., in a crowd-sourced knowledge base, in a manner that facilitates searching based on natural language queries or structured queries derived from subsequent user requests.

[0006] Accordingly, some embodiments provide a method for providing a response to a user request, the method including, at a computer system including one or more processors and memory storing one or more programs: receiving a user request from a mobile client device, the user request including at least a speech input and seeks an informational answer or performance of a task: detecting a failure to provide a satisfactory response to the user request; in response to detecting the failure, crowd-sourcing information relevant to the user request by querying one or more crowd sourcing information sources; receiving one or more answers from the crowd sourcing information sources; and generating a response to the user request based on at least one of the one or more answers received from the one or more crowd sourcing information sources.

[0007] In some embodiments, crowd-sourcing the information relevant to the user request further includes: generating one or more queries based on the user request; and sending the one or more queries to the one or more crowd sourcing information sources.

[0008] In some embodiments, at least one of the queries includes an audio recording of the speech input.

[0009] In some embodiments, the crowd-sourcing further comprises identifying, from a set of crowd sourcing information sources, the one or more crowd sourcing information sources to be queried.

[0010] In some embodiments, detecting the failure to provide a satisfactory response to the user request further includes determining that a web-search based on information contained in the user request is unsatisfactory to the user.

[0011] In some embodiments, detecting the failure to provide a satisfactory response to the user request comprises receiving feedback from the user that a previous response provided to the user for the user request was unsatisfactory.

[0012] In some embodiments, detecting the failure to provide a satisfactory response to the user request comprises analyzing logs of the usage of the digital assistant.

[0013] In some embodiments, the method further includes: prior to the crowd-sourcing: requesting permission from the user to send the information contained in the user request to the one or more crowd sourcing information sources; and receiving permission from the user to send the information contained in the user request to the one or more crowd sourcing information sources.

[0014] In some embodiments, the method further includes: sending a list of the one or more crowd sourcing information sources to the mobile client device.

[0015] In some embodiments, the one or more crowd sourcing information sources represented in the list are separately selectable by the user.

[0016] In some embodiments, the information contained in the user request is sent to a real-time answer-lookup database.

[0017] In some embodiments, the information contained in the user request is sent to one or more non-real-time expert services.

[0018] In some embodiments, the method further includes: receiving at least one real-time answer from a real-time answer-lookup database; upon receipt of the at least one real-time answer, sending to the mobile client device the at least one real-time answer; receiving at least one non-real-time answer from a non-real-time expert service after receiving the at least one real-time answer; and upon receipt of the at least one non-real-time answer, sending to the mobile client device the at least one non-real-time answer.

[0019] In some embodiments, the at least one real-time answer and the at least one non-real-time answer are presented to the user at different times.

[0020] In some embodiments, the method further includes: not receiving any answer from at least one of the one or more crowd sourcing information sources before generating the remedial response.

[0021] In some embodiments, the method further includes: when more than one answer is received from the one or more crowd sourcing information sources, ranking the answers in accordance with predetermined criteria.

[0022] In some embodiments, the method further includes: selecting a subset of answers from the one or more answers in accordance with the ranking.

[0023] In some embodiments, the method further includes: providing the response in speech form to the user.

[0024] In some embodiments, receiving the one or more answers from the crowd sourcing information sources further includes: receiving at least one of the one or more answers from individual members of the public in non-real-time.

[0034] Like reference numerals refer to corresponding parts throughout the drawings.

[0035] FIG. 1 is a block diagram of an operating environment 100 of a digital assistant according to some embodiments. The terms “digital assistant.” “virtual assistant,” “intelligent automated assistant,” or “automatic digital assistant,” refer to any information processing system that interprets natural language input in spoken and/or textual form to infer user intent, and performs actions based on the inferred user intent. For example, to act on a inferred user intent, the system can perform one or more of the following: identifying a task flow with steps and parameters designed to accomplish the inferred user intent, inputting specific requirements from the inferred user intent into the task flow; executing the task flow by invoking programs, methods, services, APIs, or the like; and generating output responses to the user in an audible (e.g. speech), textual, and/or visual form.

[0036] Specifically, a digital assistant is capable of accepting a user request at least partially in the form of a natural language command, request, statement, narrative, and/or inquiry. Typically, the user request seeks either an informational answer or performance of a task by the digital assistant. A satisfactory response to the user request is either provision of the requested informational answer, performance of the requested task, or a combination of the two. For example, a user may ask the digital assistant a question, such as “Where am I right now?” Based on the user's current location, the digital assistant may answer, “You are in Central Park near the west gate.” The user may also request the performance of a task, for example, “Please invite my friends to my girlfriend's birthday party next week.” In response, the digital assistant may acknowledge the request by saying “Yes, right away,” and then send a suitable calendar invite on behalf of the user to each of the user friends listed in the user's electronic address book. There are numerous other ways of interacting with a digital assistant to request information or performance of various tasks. In addition to providing verbal responses and taking programmed actions, the digital assistant can also provide responses in other visual or audio forms, e.g., as text, alerts, music, videos, animations, etc.

[0037] An example of a digital assistant is described in Applicant's U.S. Utility application Ser. No. 12/987,982 for “Intelligent Automated Assistant,” filed Jan. 10, 2011, the entire disclosure of which is incorporated herein by reference.

[0038] As shown in FIG. 1, in some embodiments, a digital assistant is implemented according to a client-server model. The digital assistant includes a client-side portion 102a, 102b (hereafter “DA client 102”) executed on a user device 104a, 104b, and a server-side portion 106 (hereafter “DA server 106”) executed on a server system 108. The DA client 102 communicates with the DA server 106 through one or more networks 110. The DA client 102 provides client-side functionalities such as user-facing input and output processing and communications with the DA-server 106. The DA server 106 provides server-side functionalities for any number of DA-clients 102 each residing on a respective user device 104.

[0039] In some embodiments, the DA server 106 includes a client-facing I/O interface 112, one or more processing modules 114, data and models 116, and an I/O interface to external services 118. The client-facing I/O interface facilitates the client-facing input and output processing for the digital assistant server 106. The one or more processing modules 114 utilize the data and models 116 to determine the user's intent based on natural language input and perform task execution based on inferred user intent. In some embodiments, the DA-server 106 communicates with external services 120 through the network(s) 110 for task completion or information acquisition. The I/O interface to external services 118 facilitates such communications.

[0040] Examples of the user device 104 include, but are not limited to, a handheld computer, a personal digital assistant (PDA), a tablet computer, a laptop computer, a desktop computer, a cellular telephone, a smart phone, an enhanced general packet radio service (EGPRS) mobile phone, a media player, a navigation device, a game console, a television, a remote control, or a combination of any two or more of these data processing devices or other data processing devices. More details on the user device 104 are provided in reference to an exemplary user device 104 shown in FIG. 2.

[0041] Examples of the communication network(s) 110 include local area networks (“LAN”) and wide area networks (“WAN”), e.g., the Internet. The communication network(s) 110 may be implemented using any known network protocol, including various wired or wireless protocols, such as e.g., Ethernet, Universal Serial Bus (USB), FIREWIRE, Global System for Mobile Communications (GSM), Enhanced Data GSM Environment (EDGE), code division multiple access (CDMA), time division multiple access (TDMA), Bluetooth, Wi-Fi, voice over Internet Protocol (VoIP), Wi-MAX, or any other suitable communication protocol.

[0042] The server system 108 can be implemented on one or more standalone data processing apparatus and/or a distributed network of computers. In some embodiments, the server system 108 also employs various virtual devices and/or services of third party service providers (e.g., third-party cloud service providers) to provide the underlying computing resources and/or infrastructure resources of the server system 108.

[0043] Although the digital assistant shown in FIG. 1 includes both a client-side portion (e.g., the DA-client 102) and a server-side portion (e.g., the DA-server 106), in some embodiments, the functions of a digital assistant can be implemented as a standalone application installed on a user device. In addition, the divisions of functionalities between the client and server portions of the digital assistant can vary in different embodiments. For example, in one example embodiment, the DA client can be a thin-client that provides only user-facing input and output processing functions, and delegates all other functionalities of the digital assistant to a backend server.

[0044] As described later in this specification, the digital assistant can implement a crowd sourcing functionality. The crowd sourcing functionality allows the digital assistant to gather information from third party information sources (i.e., so-called “crowd-sourcing information sources” or “CS information sources”), and use the crowd sourced information to facilitate request fulfillment, and in some cases, intent inference, in an extended time frame. In some embodiments, the information crowd sourcing is only invoked when other real-time response mechanisms of the digital assistant have failed to produce a satisfactory response to a user request. In some embodiments, the information crowd sourcing is available to produce a response to a user request without the presence of a prior failure by another response mechanism of the digital assistant. In some embodiments, the information crowd sourcing is performed offline when failures are detected in a user interaction log, and the crowd sourced information is subsequently used to improve the response mechanisms of the digital assistant for future user requests and interactions.

[0045] FIG. 2 is a block diagram of a user-device 104 in accordance with some embodiments. The user device 104 includes a memory interface 202, one or more processors 204, and a peripherals interface 206. The various components in the user device 104 are coupled by one or more communication buses or signal lines. The user device 104 includes various sensors, subsystems, and peripheral devices that are coupled to the peripherals interface 206. The sensors, subsystems, and peripheral devices gather information and/or facilitate various functionalities of the user device 104.

[0046] For example, a motion sensor 210, a light sensor 212, and a proximity sensor 214 are coupled to the peripherals interface 206 to facilitate orientation, light, and proximity sensing functions. Other sensors 216, such as a positioning system (e.g., GPS receiver), a temperature sensor, a biometric sensor, and the like, can also be connected to the peripherals interface 206, to facilitate related functionalities.

[0047] A camera subsystem 220 and an optical sensor 222 are utilized to facilitate camera functions, such as taking photographs and recording video clips. Communication functions are facilitated through one or more wired and/or wireless communication subsystems 224, which can include various communication ports, radio frequency receivers and transmitters, and/or optical (e.g., infrared) receivers and transmitters. An audio subsystem 226 is coupled to speakers 228 and a microphone 230 to facilitate voice-enabled functions, such as voice recognition, voice replication, digital recording, and telephony functions.

[0048] An I/O subsystem 240 is also coupled to the peripheral interface 206. The L/O subsystem 240 includes a touch screen controller 242 and/or other input controller(s) 244. The touch-screen controller 242 is coupled to a touch screen 246. The touch screen 246 and the touch screen controller 242 can, for example, detect contact and movement or break thereof using any of a plurality of touch sensitivity technologies, such as capacitive, resistive, infrared, surface acoustic wave technologies, proximity sensor arrays, and the like. The other input controller(s) 244 can be coupled to other input/control devices 248, such as one or more buttons, rocker switches, thumb-wheel, infrared port, USB port, and/or a pointer device such as a stylus.

[0049] The memory interface 202 is coupled to memory 250. The memory 250 can include high-speed random access memory and/or non-volatile memory, such as one or more magnetic disk storage devices, one or more optical storage devices, and/or flash memory (e.g., NAND, NOR).

[0050] The memory 250 stores an operating system 252, a communication module 254, a graphical user interface module 256, a sensor processing module 258, a phone module 260, and applications 262. The operating system 252 includes instructions for handling basic system services and for performing hardware dependent tasks. The communication module 254 facilitates communicating with one or more additional devices, one or more computers and/or one or more servers. The graphical user interface module 256 facilitates graphic user interface processing. The sensor processing module 258 facilitates sensor-related processing and functions. The phone module 260 facilitates phone-related processes and functions. The application module 262 facilitates various functionalities of user applications, such as electronic-messaging, web browsing, media processing, Navigation, imaging and/or other processes and functions.

[0051] As described in this specification, the memory 250 also stores client-side digital assistant instructions (e.g., in a digital assistant client module 264) and various user data 266 (e.g., user-specific vocabulary data, preference data, and/or other data such as the user's electronic address book, to-do lists, shopping lists, etc.) to provide the client-side functionalities of the digital assistant.

[0052] In various embodiments, the digital assistant client module 264 is capable of accepting voice input, text input, touch input, and/or gestural input through various user interfaces (e.g., the I/O subsystem 244) of the user device 104. The digital assistant client module 264 is also capable of providing output in audio, visual, and/or tactile forms. For example, output can be provided as voice, sound, alerts, text messages, menus, graphics, videos, animations, vibrations, and/or combinations of two or more of the above. During operation, the digital assistant client module 264 communicates with the digital assistant server using the communication subsystems 224.

[0053] In some embodiments, the digital assistant client module 264 utilizes the various sensors, subsystems and peripheral devices to gather additional information from the surrounding environment of the user device 104 to establish a context associated with a user input. In some embodiments, the digital assistant client module 264 optionally provides the context information or a subset thereof with the user input to the digital assistant server to help infer the user's intent.

[0054] In some embodiments, the context information that can accompany the user input includes sensor information, e.g., lighting, ambient noise, ambient temperature, images or videos of the surrounding environment, etc. In some embodiments, the context information also includes the physical state of the device, e.g., device orientation, device location, device temperature, power level, speed, acceleration, motion patterns, cellular signals strength, etc. In some embodiments, information related to the software state of the user device 106, e.g., running processes, installed programs, past and present network activities, background services, error logs, resources usage, etc., of the user device 104 can also be provided to the digital assistant server as context information associated with a user input.

[0055] In some embodiments, the DA client module 264 selectively provides information (e.g., user data 266) stored on the user device 104 in response to requests from the digital assistant server. In some embodiments, the digital assistant client module 264 also elicits additional input from the user via a natural language dialogue or other user interfaces upon request by the digital assistant server 106. The digital assistant client module 264 passes the additional input to the digital assistant server 106 to help the digital assistant server 106 in intent inference and/or fulfillment of the user's intent expressed in the user request.

[0056] In various embodiments, the memory 250 can include additional instructions or fewer instructions. Furthermore, various functions of the user device 104 may be implemented in hardware and/or in firmware, including in one or more signal processing and/or application specific integrated circuits.

[0057] FIG. 3A is a block diagram of an example digital assistant system 300 in accordance with some embodiments. In some embodiments, the digital assistant system 300 is implemented on a standalone computer system. In some embodiments, the digital assistant system 300 is distributed across multiple computers. In some embodiments, some of the modules and functions of the digital assistant are divided into a server portion and a client portion, where the client portion resides on a user device (e.g., the user device 104) and communicates with the server portion (e.g., the server system 108) through one or more networks, e.g., as shown in FIG. 1. In some embodiments, the digital assistant system 300 is an embodiment of the server system 108 (and/or the digital assistant server 106) shown in FIG. 1. It should be noted that the digital assistant system 300 is only one example of a digital assistant system, and that the digital assistant system 300 may have more or fewer components than shown, may combine two or more components, or may have a different configuration or arrangement of the components. The various components shown in FIG. 3A may be implemented in hardware, software, firmware, including one or more signal processing and/or application specific integrated circuits, or a combination of thereof.

[0058] The digital assistant system 300 includes memory 302, one or more processors 304, an input/output (I/O) interface 306, and a network communications interface 308. These components communicate with one another over one or more communication buses or signal lines 310.

[0059] In some embodiments, the memory 302 includes a non-transitory computer readable medium, such as high-speed random access memory and/or a non-volatile computer readable storage medium (e.g., one or more magnetic disk storage devices, flash memory devices, or other non-volatile solid-state memory devices).

[0060] The I/O interface 306 couples input/output devices 316 of the digital assistant system 300, such as displays, a keyboards, touch screens, and microphones, to the user interface module 322. The I/O interface 306, in conjunction with the user interface module 322, receive user inputs (e.g., voice input, keyboard inputs, touch inputs, etc.) and process them accordingly. In some embodiments, e.g., when the digital assistant is implemented on a standalone user device, the digital assistant system 300 includes any of the components and I/O and communication interfaces described with respect to the user device 104 in FIG. 2. In some embodiments, the digital assistant system 300 represents the server portion of a digital assistant implementation, and interacts with the user through a client-side portion residing on a user device (e.g., the user device 104 shown in FIG. 2).

[0061] In some embodiments, the network communications interface 308 includes wired communication port(s) 312 and/or wireless transmission and reception circuitry 314. The wired communication port(s) receive and send communication signals via one or more wired interfaces, e.g., Ethernet, Universal Serial Bus (USB), FIREWIRE, etc. The wireless circuitry 314 receives and sends RF signals and/or optical signals from/to communications networks and other communications devices. The wireless communications may use any of a plurality of communications standards, protocols and technologies, such as GSM, EDGE, CDMA, TDMA, Bluetooth, Wi-Fi, VoIP, Wi-MAX, or any other suitable communication protocol. The network communications interface 308 enables communication between the digital assistant system 300 with networks, such as the Internet, an intranet and/or a wireless network, such as a cellular telephone network, a wireless local area network (LAN) and/or a metropolitan area network (MAN), and other devices.

[0062] In some embodiments, memory 302, or the non-volatile and/or non-transitory computer readable storage media of memory 302, stores programs, modules, instructions, and data structures including all of a subset of: an operating system 318, a communications module 320, a user interface module 322, one or more applications 324, and a digital assistant module 326. The one or more processors 304 execute these programs, modules, and instructions, and reads/writes from/to the data structures.

[0063] The operating system 318 (e.g., Darwin, RTXC, LINUX, UNIX, OS X, WINDOWS, or an embedded operating system such as VxWorks) includes various software components and/or drivers for controlling and managing general system tasks (e.g., memory management, storage device control, power management, etc.) and facilitates communications between various hardware, firmware, and software components.

[0064] The communications module 320 facilitates communications between the digital assistant system 300 with other devices over the network communications interface 308. For example, the communication module 320 may communicate with the communication interface 254 of the device 104 shown in FIG. 2. The communications module 320 also includes various software components for handling data received by the wireless circuitry 314 and/or wired communications port 312.

[0065] The user interface module 322 receives commands and/or inputs from a user via the I/O interface 306 (e.g., from a keyboard, touch screen, and/or microphone), and generates user interface objects on a display.

[0066] The applications 324 include programs and/or modules that are configured to be executed by the one or more processors 304. For example, if the digital assistant system is implemented on a standalone user device, the applications 324 may include user applications, such as games, a calendar application, a navigation application, or an email application. If the digital assistant system 300 is implemented on a server farm, the applications 324 may include resource management applications, diagnostic applications, or scheduling applications, for example.

[0067] The memory 302 also stores the digital assistant module (or the server portion of a digital assistant) 326. In some embodiments, the digital assistant module 326 includes the following sub-modules, or a subset or superset thereof: an input/output processing module 328, a speech-to-text (STIT) processing module 330, a natural language processing module 332, a dialogue flow processing module 334, a task flow processing module 336, a service processing module 338, a failure management module 340, and a crowd sourcing module 342. Each of these processing modules has access to one or more of the following data and models of the digital assistant 326, or a subset or superset thereof: ontology 360, vocabulary index 344, user data 348, task flow models 354, service models 356, and crowd-sourced knowledge base 358.

[0068] In some embodiments, using the processing modules, data, and models implemented in the digital assistant module 326, the digital assistant performs at least some of the following: identifying a user's intent expressed in a natural language input received from the user; actively eliciting and obtaining information needed to fully infer the user's intent (e.g., by disambiguating words, names, intentions, etc.); determining the task flow for fulfilling the inferred intent; and executing the task flow to fulfill the inferred intent. In some embodiments, the digital assistant also takes appropriate actions when a satisfactory response was not or could not be provided to the user for various reasons.

[0069] As shown in FIG. 3B, in some embodiments, the I/O processing module 328 interacts with the user through the I/O devices 316 in FIG. 3A or with a user device (e.g., a user device 104 in FIG. 1) through the network communications interface 308 in FIG. 3A to obtain user input (e.g., a speech input) and to provide responses to the user input. The I/O processing module 328 optionally obtains context information associated with the user input from the user device, along with or shortly after the receipt of the user input. The context information includes user-specific data, vocabulary, and/or preferences relevant to the user input. In some embodiments, the context information also includes software and hardware states of the device (e.g., the user device 104 in FIG. 1) at the time the user request is received, and/or information related to the surrounding environment of the user at the time that the user request was received. In some embodiments, the I/O processing module 328 also sends follow-up questions to, and receives answers from, the user regarding the user request. When a user request is received by the I/O processing module 328 and the user request contains a speech input, the I/O processing module 328 forwards the speech input to the speech-to-text (STT) processing module 330 for speech-to-text conversions.

[0070] The speech-to-text processing module 330 receives speech input (e.g., a user utterance captured in a voice recording) through the I/O processing module 328. In some embodiments, the speech-to-text processing module 330 uses various acoustic and language models to recognize the speech input as a sequence of phonemes, and ultimately, a sequence of words or tokens written in one or more languages. The speech-to-text processing module 330 can be implemented using any suitable speech recognition techniques, acoustic models, and language models, such as Hidden Markov Models, Dynamic Time Warping (DTW)-based speech recognition, and other statistical and/or analytical techniques. In some embodiments, the speech-to-text processing can be performed at least partially by a third party service or on the user's device. Once the speech-to-text processing module 330 obtains the result of the speech-to-text processing, e.g., a sequence of words or tokens, it passes the result to the natural language processing module 332 for intent inference.

[0071] More details on the speech-to-text processing are described in Applicant's U.S. Utility application Ser. No. 13/236,942 for “Consolidating Speech Recognition Results,” filed on Sep. 20, 2011, the entire disclosure of which is incorporated herein by reference.

[0072] The natural language processing module 332 (“natural language processor”) of the digital assistant takes the sequence of words or tokens (“token sequence”) generated by the speech-to-text processing module 330, and attempts to associate the token sequence with one or more “actionable intents” recognized by the digital assistant. An “actionable intent” represents a task that can be performed by the digital assistant, and has an associated task flow implemented in the task flow models 354. The associated task flow is a series of programmed actions and steps that the digital assistant takes in order to perform the task. The scope of a digital assistant's capabilities is dependent on the number and variety of task flows that have been implemented and stored in the task flow models 354, or in other words, on the number and variety of “actionable intents” that the digital assistant recognizes. The effectiveness of the digital assistant, however, is also dependent on the assistant's ability to infer the correct “actionable intent(s)” from the user request expressed in natural language.

[0073] In some embodiments, in addition to the sequence of words or tokens obtained from the speech-to-text processing module 330, the natural language processor 332 also receives context information associated with the user request, e.g., from the I/O processing module 328. The natural language processor 332 optionally uses the context information to clarify, supplement, and/or further define the information contained in the token sequence received from the speech-to-text processing module 330. The context information includes, for example, user preferences, hardware and/or software states of the user device, sensor information collected before, during, or shortly after the user request, prior interactions (e.g., dialogue) between the digital assistant and the user, and the like.

[0074] In some embodiments, the natural language processing is based on an ontology 360. The ontology 360 is a hierarchical structure containing many nodes, each node representing either an “actionable intent” or a “property” relevant to one or more of the “actionable intents” or other “properties”. As noted above, an “actionable intent” represents a task that the digital assistant is capable of performing, i.e., it is “actionable” or can be acted on. A “property” represents a parameter associated with an actionable intent, a domain concept or entity, or a sub-aspect of another property. A linkage between an actionable intent node and a property node in the ontology 360 defines how a parameter represented by the property node pertains to the task represented by the actionable intent node.

[0075] In some embodiments, the ontology 360 is made up of actionable intent nodes and property nodes. Within the ontology 360, each actionable intent node is linked to one or more property nodes either directly or through one or more intermediate property nodes. Similarly, each property node is linked to one or more actionable intent nodes either directly or through one or more intermediate property nodes. For example, as shown in FIG. 3C, the ontology 360 may include a “restaurant reservation” node (i.e., an actionable intent node). Property node “restaurant,” (a domain entity represented by a property node) and property nodes “date/time” (for the reservation) and “party size” are each directly linked to the actionable intent node (i.e., the “restaurant reservation” node). In addition, property nodes “cuisine,” “price range,” “phone number,” and “location” are sub-nodes of the property node “restaurant,” and are each linked to the “restaurant reservation” node (i.e., the actionable intent node) through the intermediate property node “restaurant.” For another example, as shown in FIG. 3C, the ontology 360 may also include a “set reminder” node (i.e., another actionable intent node). Property nodes “date/time” (for the setting the reminder) and “subject” (for the reminder) are each linked to the “set reminder” node. Since the property “date/time” is relevant to both the task of making a restaurant reservation and the task of setting a reminder, the property node “date/time” is linked to both the “restaurant reservation” node and the “set reminder” node in the ontology 360.

[0076] An actionable intent node, along with its linked concept nodes, may be described as a “domain.” In the present discussion, each domain is associated with a respective actionable intent, and refers to the group of nodes (and the relationships therebetween) associated with the particular actionable intent. For example, the ontology 360 shown in FIG. 3C includes an example of a restaurant reservation domain 362 and an example of a reminder domain 364 within the ontology 360. The restaurant reservation domain includes the actionable intent node “restaurant reservation,” property nodes “restaurant,” “date/time,” and “party size,” and sub-property nodes “cuisine,” “price range,” “phone number,” and “location.” The reminder domain 364 includes the actionable intent node “set reminder,” and property nodes “subject” and “date/time.” In some embodiments, the ontology 360 is made up of many domains. Each domain may share one or more property nodes with one or more other domains. For example, the “date/time” property node may be associated with many different domains (e.g., a scheduling domain, a travel reservation domain, a movie ticket domain, etc.), in addition to the restaurant reservation domain 362 and the reminder domain 364.

[0077] While FIG. 3C illustrates two example domains within the ontology 360, other domains (or actionable intents) include, for example. “initiate a phone call,” “find directions,” “schedule a meeting,” “send a message,” and “provide an answer to a question,” and so on. A “send a message” domain is associated with a “send a message” actionable intent node, and may further include property nodes such as “recipient(s)”, “message type”, and “message body.” The property node “recipient” may be further defined, for example, by the sub-property nodes such as “recipient name” and “message address.”

[0078] In some embodiments, the ontology 360 includes all the domains (and hence actionable intents) that the digital assistant is capable of understanding and acting upon. In some embodiments, the ontology 360 may be modified, such as by adding or removing entire domains or nodes, or by modifying relationships between the nodes within the ontology 360.

[0079] In some embodiments, nodes associated with multiple related actionable intents may be clustered under a “super domain” in the ontology 360. For example, a “travel” super-domain may include a cluster of property nodes and actionable intent nodes related to travels. The actionable intent nodes related to travels may include “airline reservation,” “hotel reservation,” “car rental,” “get directions,” “find points of interest,” and so on. The actionable intent nodes under the same super domain (e.g., the “travels” super domain) may have many property nodes in common. For example, the actionable intent nodes for “airline reservation,” “hotel reservation,” “car rental,” “get directions,” “find points of interest” may share one or more of the property nodes “start location,” “destination,” “departure date/time,” “arrival date/time,” and “party size.”

[0080] In some embodiments, each node in the ontology 360 is associated with a set of words and/or phrases that are relevant to the property or actionable intent represented by the node. The respective set of words and/or phrases associated with each node is the so-called “vocabulary” associated with the node. The respective set of words and/or phrases associated with each node can be stored in the vocabulary index 344 in association with the property or actionable intent represented by the node. For example, returning to FIG. 3B, the vocabulary associated with the node for the property of “restaurant” may include words such as “food,” “drinks,” “cuisine,” “hungry,” “eat,” “pizza,” “fast food,” “meal,” and so on. For another example, the vocabulary associated with the node for the actionable intent of “initiate a phone call” may include words and phrases such as “call,” “phone,” “dial,” “ring,” “call this number,” “make a call to,” and so on. The vocabulary index 344 optionally includes words and phrases in different languages.

[0081] The natural language processor 332 receives the token sequence (e.g., a text string) from the speech-to-text processing module 330, and determines what nodes are implicated by the words in the token sequence. In some embodiments, if a word or phrase in the token sequence is found to be associated with one or more nodes in the ontology 360 (via the vocabulary index 344), the word or phrase will “trigger” or “activate” those nodes. Based on the quantity and/or relative importance of the activated nodes, the natural language processor 332 will select one of the actionable intents as the task that the user intended the digital assistant to perform. In some embodiments, the domain that has the most “triggered” nodes is selected. In some embodiments, the domain having the highest confidence value (e.g., based on the relative importance of its various triggered nodes) is selected. In some embodiments, the domain is selected based on a combination of the number and the importance of the triggered nodes. In some embodiments, additional factors are considered in selecting the node as well, such as whether the digital assistant has previously correctly interpreted a similar request from a user.

[0082] In some embodiments, the digital assistant also stores names of specific entities in the vocabulary index 344, so that when one of these names is detected in the user request, the natural language processor 332 will be able to recognize that the name refers to a specific instance of a property or sub-property in the ontology. In some embodiments, the names of specific entities are names of businesses, restaurants, people, movies, and the like. In some embodiments, the digital assistant can search and identify specific entity names from other data sources, such as the user's address book, a movies database, a musicians database, and/or a restaurant database. In some embodiments, when the natural language processor 332 identifies that a word in the token sequence is a name of a specific entity (such as a name in the user's address book), that word is given additional significance in selecting the actionable intent within the ontology for the user request.

[0083] For example, when the words “Mr. Santo” are recognized from the user request, and the last name “Santo” is found in the vocabulary index 344 as one of the contacts in the user's contact list, then it is likely that the user request corresponds to a “send a message” or “initiate a phone call” domain. For another example, when the words “ABC Café” are found in the user request, and the term “ABC Café” is found in the vocabulary index 344 as the name of a particular restaurant in the user's city, then it is likely that the user request corresponds to a “restaurant reservation” domain.

[0084] User data 348 includes user-specific information, such as user-specific vocabulary, user preferences, user address, user's default and secondary languages, user's contact list, and other short-term or long-term information for each user. The natural language processor 332 can use the user-specific information to supplement the information contained in the user input to further define the user intent. For example, for a user request “invite my friends to my birthday party,” the natural language processor 332 is able to access user data 348 to determine who the “friends” are and when and where the “birthday party” would be held, rather than requiring the user to provide such information explicitly in his/her request.

[0085] Other details of searching an ontology based on a token string is described in Applicant's U.S. Utility application Ser. No. 12/341,743 for “Method and Apparatus for Searching Using An Active Ontology,” filed Dec. 22, 2008, the entire disclosure of which is incorporated herein by reference.

[0086] Once the natural language processor 332 identifies an actionable intent (or domain) based on the user request, the natural language processor 332 generates a structured query to represent the identified actionable intent. In some embodiments, the structured query includes parameters for one or more nodes within the domain for the actionable intent, and at least some of the parameters are populated with the specific information and requirements specified in the user request. For example, the user may say “Make me a dinner reservation at a sushi place at 7.” In this case, the natural language processor 332 may be able to correctly identify the actionable intent to be “restaurant reservation” based on the user input. According to the ontology, a structured query for a “restaurant reservation” domain may include parameters such as {Cuisine}, {Time}, {Date}, {Party Size}, and the like. Based on the information contained in the user's utterance, the natural language processor 332 may generate a partial structured query for the restaurant reservation domain, where the partial structured query includes the parameters {Cuisine=“Sushi”} and {Time=“7 pm”}. However, in this example, the user's utterance contains insufficient information to complete the structured query associated with the domain. Therefore, other necessary parameters such as {Party Size} and {Date} are not specified in the structured query based on the information currently available. In some embodiments, the natural language processor 332 populates some parameters of the structured query with received context information. For example, if the user requested a sushi restaurant “near me,” the natural language processor 332 may populate a (location) parameter in the structured query with GPS coordinates from the user device 104.

[0087] In some embodiments, the natural language processor 332 passes the structured query (including any completed parameters) to the task flow processing module 336 (“task flow processor”). The task flow processor 336 is configured to receive the structured query from the natural language processor 332, complete the structured query, if necessary, and perform the actions required to “complete” the user's ultimate request. In some embodiments, the various procedures necessary to complete these tasks are provided in task flow models 354. In some embodiments, the task flow models include procedures for obtaining additional information from the user, and task flows for performing actions associated with the actionable intent.

[0088] As described above, in order to complete a structured query, the task flow processor 336 may need to initiate additional dialogue with the user in order to obtain additional information, and/or disambiguate potentially ambiguous utterances. When such interactions are necessary, the task flow processor 336 invokes the dialogue processing module 334 (“dialogue processor 334”) to engage in a dialogue with the user. In some embodiments, the dialogue processor 334 determines how (and/or when) to ask the user for the additional information, and receives and processes the user responses. (The questions are provided to and answers are received from the users through the I/O processing module 328.) In some embodiments, the dialogue processor 334 presents dialogue output to the user via audio and/or visual output, and receives input from the user via spoken or physical (e.g., clicking) responses. Continuing with the example above, when the task flow processor 336 invokes the dialogue flow processor 334 to determine the “party size” and “date” information for the structured query associated with the domain “restaurant reservation,” the dialogue flow processor 335 generates questions such as “For how many people?” and “On which day?” to pass to the user. Once answers are received from the user, the dialogue flow processor 334 can then populate the structured query with the missing information, or pass the information to the task flow processor 336 to complete the missing information from the structured query.

[0089] In some cases, the task flow processor 336 may receive a structured query that has one or more ambiguous properties. For example, a structured query for the “send a message” domain may indicate that the intended recipient is “Bob,” and the user may have multiple contacts named “Bob.” The task flow processor 336 will request that the dialogue processor 334 disambiguate this property of the structured query. In turn, the dialogue processor 334 may ask the user “Which Bob?”, and display (or read) a list of contacts named “Bob” from which the user may choose.

[0090] Once the task flow processor 336 has completed the structured query for an actionable intent, the task flow processor 336 proceeds to perform the ultimate task associated with the actionable intent. Accordingly, the task flow processor 336 executes the steps and instructions in the task flow model according to the specific parameters contained in the structured query. For example, the task flow model for the actionable intent of “restaurant reservation” may include steps and instructions for contacting a restaurant and actually requesting a reservation for a particular party size at a particular time. For example, using a structured query such as: {restaurant reservation, restaurant=ABC Café, date=3/12/2012, time=7 pm, party size=5}, the task flow processor 336 may perform the steps of: (1) logging onto a server of the ABC Café or a restaurant reservation system such as OPENTABLE®, (2) entering the date, time, and party size information in a form on the website, (3) submitting the form, and (4) making a calendar entry for the reservation in the user's calendar.

[0091] In some embodiments, the task flow processor 336 employs the assistance of a service processing module 338 (“service processor”) to complete a task requested in the user input or to provide an informational answer requested in the user input. For example, the service processor 338 can act on behalf of the task flow processor 336 to make a phone call, set a calendar entry, invoke a map search, invoke or interact with other user applications installed on the user device, and invoke or interact with third party services (e.g. a restaurant reservation portal, a social networking website, a banking portal, etc.). In some embodiments, the protocols and application programming interfaces (API) required by each service can be specified by a respective service model among the services models 356. The service processor 338 accesses the appropriate service model for a service and generates requests for the service in accordance with the protocols and APIs required by the service according to the service model.

[0092] For example, if a restaurant has enabled an online reservation service, the restaurant can submit a service model specifying the necessary parameters for making a reservation and the APIs for communicating the values of the necessary parameter to the online reservation service. When requested by the task flow processor 336, the service processor 338 can establish a network connection with the online reservation service using the web address stored in the service model, and send the necessary parameters of the reservation (e.g., time, date, party size) to the online reservation interface in a format according to the API of the online reservation service.

[0093] In some embodiments, the natural language processor 332, dialogue processor 334, and task flow processor 336 are used collectively and iteratively to infer and define the user's intent, obtain information to further clarify and refine the user intent, and finally generate a response (i.e., an output to the user, or the completion of a task) to fulfill the user's intent.

[0094] In some embodiments, after all of the tasks needed to fulfill the user's request have been performed, the digital assistant 326 formulates a confirmation response, and sends the response back to the user through the I/O processing module 328. If the user request seeks an informational answer, the confirmation response presents the requested information to the user. In some embodiments, the digital assistant also requests the user to indicate whether the user is satisfied with the response produced by the digital assistant 326.

[0095] More details on the digital assistant can be found in the U.S. Utility application Ser. No. 12/987,982, entitled “Intelligent Automated Assistant”, filed Jan. 18, 2010, U.S. Utility Application No. 61/493,201, entitled “Generating and Processing Data Items That Represent Tasks to Perform”, filed Jun. 3, 2011, the entire disclosures of which are incorporated herein by reference.

[0096] In many instances, a digital assistant is able to infer a user's intent based on a natural language request provided by the user and fulfill the user's request either by providing information sought by the user's request or by performing tasks according to the user's request. However, sometimes, the digital assistant will fail to provide a satisfactory response to the user's request for information or action. The reasons for the failures can be many, such as imperfect speech recognition, unrecognized terms and concepts in the user request, incorrect or incomplete information and inadequate capability in the digital assistant's services, and so on. Regardless of the reason for the digital assistant's failure to provide a satisfactory response to a user request, it is desirable to implement a suitable failure management procedure for the digital assistant.

[0097] As shown in FIG. 3B, in some embodiments, the digital assistant 326 also implements a failure management module 340 to provide appropriate remedies when a failure is detected. In some embodiments, the failure management module 340 invokes the crowd sourcing module 342 or searches for answers in the crowd-source knowledge base 358 to generate an appropriate remedial or corrective response for a failed user request. The crowd sourcing module 342 issues queries and collects answers from one or more external CS information sources over an extended period of time, and uses the answers to supplement the digital assistant's abilities in speech processing, natural language processing (for intent inference), and/or task flow processing. For example, the crowd sourced answers may help to recognize speech in particular accents in different regions, expand the vocabulary associated with different domains, and/or identify additional domain entities. In some embodiments, the crowd sourced answers are used (either by the digital assistant itself, or by a provider of the digital assistant) to create additional domains and task flows to further expand the capabilities of the digital assistant.

[0098] In some embodiments, the crowd sourcing module 342 establishes and maintains the crowd sourced knowledge base 358. The crowd sourced knowledge base 358 stores crowd sourced information that addresses informational or task requests that the digital assistant might provide to its users. In some embodiments, the contents of the crowd sourced knowledge base are organized by records of previous user requests to which the digital assistant had initially failed to successfully respond, but subsequently fulfilled using crowd-sourced information. The crowd-sourced knowledge base provides references and information to the digital assistant to provide correct and satisfactory responses to the same or similar user requests received in the future. In some embodiments, the crowd sourced knowledge base is organized to facilitate searching by the natural language processor. For example, the information and answers in the crowd sourced knowledge base may be indexed by nodes in an ontology (e.g., the ontology 360 in FIG. 3B) as well, so that the infrastructure of the natural language processor can be leveraged to quickly find past questions and answers in one or more relevant domains.

[0099] In some embodiments, the failure management module 340, the crowd sourcing module 340 and the CS knowledge base 358 are part of the digital assistant system 326, and can various other components through various internal application interfaces. In some embodiments, one or more of the failure management module 340, the crowd sourcing module 340, and the CS knowledge base 358, of one or more sub-components thereof are optionally provided separately from the digital assistant system 326, and the digital assistant system 326 accesses each of these modules and sub-components thereof through one or more external application programming interfaces (APIs). In some embodiments, various aspects of the digital assistant system 326, such as the speech to text processing (e.g., the speech modules, acoustic models, vocabulary used in speech recognition), the natural language processing aspect (e.g., language models, grammar, ontology, etc.), the task-flow, dialogue flow, and service processing, are modified based on the information stored in the CS knowledge base 358 to improve future performance of the digital assistant system 326.

[0100] In some embodiments, the digital assistant maintains usage logs 370 on user requests and interactions between the digital assistant and the users. The usage logs optionally store information such as the user requests received, the context information surrounding the user requests, the responses provided to the users, and feedback provided by the users, the parameters, models, third-party services, and procedures used by the digital assistant to generate and provide the responses, etc. In some embodiments, the usage logs are searchable by various search parameters, such as time, location, user, demographic, response type, feedback type, task type, duration, failure type, etc. More details are provided with respect to the usage log 370 in FIG. 7 and accompanying descriptions.

[0101] Although FIG. 3B does not explicit show the communication interfaces between all components of the digital assistant 326, it is to be understood that the components shown are capable of communicate with any other components of the digital assistant 326 either directly or through one or more other interfaces, such as application programming interfaces, database query interfaces, and/or other interfaces, protocols, and/or communication channels.

[0102] FIG. 4 is a flow diagram illustrating an example process 400 undertaken by a failure management module of a digital assistant (e.g., the failure management module 340 in FIGS. 3A-3B). The example process 400 is merely an illustration of the decision process regarding which remedy option(s) should be provided to the user after a failure to provide a satisfactory response is recognized by the digital assistant. Other methods and processes are possible, and more or fewer remedy options may be implemented by the failure management module of the digital assistant in various embodiments.

[0103] In the example process 400, first, the digital assistant registers a failure to provide a satisfactory response to a user request, and optionally determines the failure type for the failure (402). A failure type is determined based on the reason for which the digital assistant has failed to produce a satisfactory response to the user request. The failure may be discovered by the digital assistant during the natural language processing or task flow execution process, or may be indicated by the user after the unsatisfactory response was provided to the user.

[0104] After the failure has been registered and its failure type determined, the digital assistant selects one or more real-time remedy options suitable for addressing the type of failure, and presents the options to the user (404). Examples of real-time remedy options include doing a web search based on the user request, calling a technical support hotline, calling an emergency number, searching the crowd-sourced knowledge base, and the like. In this specification, a real-time response refers to a response to a user request provided to the user by the digital assistant within a time-frame associated with the same user session or continuous dialogue between the user and the digital assistant in which the user request was received by the digital assistant. Therefore, each of the real-time remedy options should produce a real-time remedial or corrective response to the user request, e.g., within a few minutes of the user request.

[0105] In some embodiments, if the digital assistant had been able to partially infer one or more candidate actionable intents during the natural language processing stage, the candidate intents and their associated properties and domains can be used by the digital assistant in selecting the real-time remedy options to be presented to the user. In some embodiments, the domains (or actionable intents) recognized by the digital assistant may each be associated with one or more real-time remedy options. In some embodiments, the domains (or actionable intents) are clustered into different groups, and each group is associated with one or more real-time remedy options.

[0106] After the selected real-time remedy option(s) are presented to the user, the user may reject or accept the real-time remedy options(s) by providing a user input (e.g., a verbal reply, a gestural input, or the like). Based on the user input, the digital assistant determines whether the user has accepted any of the real-time remedy option(s) (406). If the digital assistant determines that the user has accepted one or more of the real-time remedy option(s) presented to the user, the digital assistant proceeds to execute the accepted real-time remedy option(s) (e.g., performing the web search, or making the call to the technical support hotline, and so on) (408).

[0107] Once the accepted real-time remedy options have been executed (e.g., when search results from the web search have been presented, or when the technical support call is completed), the digital assistant inquires whether the user is satisfied with the results of the real-time remedies just provided (410). If the digital assistant determines that the user is satisfied with the real-time remedial response based on the user's response, the digital assistant considers that the failure has been addressed and ceases further action regarding the failure and associated user request (412).

[0108] In some embodiments, if the digital assistant determines that the user has rejected all of the real-time remedy options presented to the user (e.g., shown as the “No” branch of the decision 406), or if the user is unsatisfied with the real-time remedial response(s) provided to the user (e.g., shown as the “No” branch of the decision 410), the digital assistant proceeds to invoke the delayed remedy procedures (414). The delayed remedy procedures include consulting with external CS information sources and/or crowd sourced knowledge base to generate a delayed remedial or corrective response in an extended time frame.

[0109] In this specification, a delayed response refers to a response to a user request that is provided outside of the time frame of the current user session or continuing dialogue with the user in which the user request was first received. The time-frame in which a delayed response to a user request is provided ranges from several minutes to several hours, days, or weeks, depending on the nature of the user request and the typical time frame that useful information may be crowd sourced from external CS information sources.

[0110] As shown in FIG. 4, the delayed remedy procedures include first obtaining user's authorization to proceed with information crowd sourcing for the user request (416), performing the information crowd sourcing process (418), and generating a delayed response based on the crowd-sourced information, as well as any information that the digital assistant already possessed before the crowd sourcing was started (420). More details on information crowd sourcing for user requests are provided with reference to FIGS. 5, 6A-6C, and 7.

[0111] As set forth above, in some embodiments, the digital assistant may attempt to obtain additional information from external sources and formulate a response based on the additional information in an extended time-frame. In some embodiments, the information crowd sourcing procedures may also be invoked as one of the means to generate a response without first detecting a failure. For example, the digital assistant can allow the user to explicitly request a response to be generated based on crowd sourced information at the outset. In response to the user's explicit request, the natural language processor of the digital assistant can invoke the information crowd sourcing module directly without first detecting a failure.

[0112] FIG. 5 is a block diagram of an example information crowd sourcing module (e.g., the crowd sourcing module 342 shown in FIGS. 3A-3B and later in FIG. 7) in accordance with some embodiments. As shown in FIG. 5, the crowd sourcing module 342 includes an information source selection module 502, a query generation module 504, an answer monitoring module 506, an answer integration module 508, and a response generation module 510. In some embodiments, the crowd sourcing module 342 also includes knowledge building module 512 to build and maintain the crowd-sourced knowledge base based on the crowd sourced information.

[0113] Also shown in FIG. 5, the crowd sourcing module 342 stores data in various data structures and databases to keep track of the CS information sources, user requests, queries, and answers involved in the crowd sourcing process. For example, the information sources database 516 stores the CS information sources available to provide crowd sourced information. The user requests database 518 stores the user requests for which information crowd sourcing is currently being performed. The queries database 520 stores the queries that have been sent to the external CS information sources for each user request. The answers database 522 stores the answers that have been received from the CS information sources for each query.

[0114] In some embodiments, the information source selection module 502 selects, from among multiple CS information sources in the information source database 516, one or more CS information sources suitable to provide useful information for the comprehension and fulfillment of a user request. In some embodiments, the information source selection module is optional, and a fixed set of CS information sources are used for all user requests.

[0115] In some embodiments, the query generation module 504 generates one or more queries for each user request for which information crowd sourcing is to be performed. The queries are generated based on the user request and its context information. The query generation module designs the queries such that they are likely to bring back answers helpful in the comprehension and fulfillment of the user request. In some embodiments, the query generation module 504 also serves to send the queries to the appropriate CS information sources.

[0116] In some embodiments, the answer monitoring module 506 monitors the CS information sources to retrieve answers to queries from the CS information sources. In some embodiments, the answer monitoring module 506 can also receive answers sent to the answer monitoring module 506 by the CS information sources. For different CS information sources and/or queries, the time frame in which monitoring for answers is performed can range in minutes, hours, days, weeks, or even longer. The answer monitoring module 506 stores the answers received for each query in the answers database 522, and keeps track of the answer statuses of the queries.

[0117] Once the answer monitoring module determines that sufficient answers have been collected for the queries issued for a particular user request, the answer integration module filters, ranks, reconciles, and integrates the answers to provide consolidated crowd sourced information relevant to the particular user request to the response generation module. The response generation uses the consolidated crowd sourced information and any information the digital assistant already possesses to generate a response to the particular user request.

[0118] In some embodiments, if the response generated based on the crowd sourced information is satisfactory to the user, the knowledge-base building module 512 stores the consolidated crowd sourced information and/or the queries and answers that contributed to the consolidated crowd sourced information in the crowd-sourced knowledge base 358.

[0119] FIG. 5 is merely an illustration of how a crowd sourcing module may be implemented. In various embodiments, more or fewer components may be used to implement information crowd sourcing for the digital assistant. More or fewer functions may be provided by the digital assistant. More details regarding the crowd sourcing module 342 and the information crowd sourcing process are provided with reference to FIGS. 6A-6C, and 7 below.

[0120] FIGS. 6A-6C illustrate an example process 600 for crowd sourcing information to provide a response to a user request. In some embodiments, the process 600 can be performed as part of a delayed remedy procedure used when one or more real-time response mechanisms have failed to produce a satisfactory response to the user. In some embodiments, the process 600 is a standalone process that is provided independently of a detected failure to fulfill a user request. In some embodiments, the process 600 may be used to provide a response either when a prior failure was detected or without the presence of a prior failure. FIGS. 6A-6C each describes one stage of the information crowd sourcing process. Not all steps shown in FIGS. 6A-6C are necessary in all embodiments. In some embodiments, the process 600 is performed by the information crowd sourcing module 358 shown in FIGS. 3A-3B and 5.

[0121] FIG. 6A illustrates the first stage of the information crowd sourcing process. In the first stage, queries are generated based at least in part on a user request, and CS information sources are selected for information crowd sourcing for the user request. This stage of the information crowd sourcing process can happen quickly and within the same user session in which the user request was first received.

[0122] As shown in FIG. 6A, during the first stage, the digital assistant first seeks express permission from the user that information crowd sourcing is to be used to aid in the generation of a satisfactory response to the user request (602). In some embodiments, the digital assistant notifies the user that the information crowd sourcing does not guarantee to produce a satisfactory response, and that even if the information crowd sourcing does bring back useful information for producing a satisfactory result, it would take some extended time outside of the current user session before the response can be generated. In some embodiments, the digital assistant also notifies the user that the digital assistant may act on behalf of the user to answer questions and provide additional information to external CS information sources during the crowd sourcing process. The digital assistant allows the user to reject the option to perform crowd sourcing for the user request. The digital assistant can also help the user to establish some privacy rules for the digital assistant for interacting with different CS information sources, such that the user's privacy is not inadvertently compromised during the crowd sourcing.

[0123] Based on user's input, the digital assistant determines whether the user has accepted the option to crowd source information for the user request (604). If the user does not accept the information crowd sourcing option for any reason (e.g., for privacy or timing concerns), the digital assistant notifies the user that a satisfactory response cannot be generated based on the current capabilities of the digital assistant. After the notification, the digital assistant can cease further actions or dialogues regarding the user request (606).

[0124] If the digital assistant determines that the user would like to proceed with the information crowd sourcing, the digital assistant proceeds to identify one or more CS information sources suitable for providing information regarding the user request (608). In some embodiments, the CS information source selection is performed by the information source selection module 502 shown in FIG. 5. In some implementations, the information source selection is optional, and a default set of information sources are used for all user requests.

[0125] In some embodiments, the digital assistant optionally selects the suitable CS information sources based on the properties and domains that were “activated” by the words in the user request during the earlier natural language processing of the user request. For example, if the properties of “restaurant” and “birthday party” were activated during the natural language processing of a user request, the digital assistant can select CS information sources such as a life style information portal, rather than a technical support information portal. In some embodiments, other criteria for selecting the CS information sources can be used.

[0126] In some embodiments, the CS information sources that the crowd sourcing module may query for information and answers include public forums. In a public forum, questions can be posted to a wide audience, and answers can be solicited and received from the general public who visit the public forum. Examples of public forums include online chat rooms, online message boards, discussion groups, and the like. In general, a large public forum can have sub-forums focused on different topics and subject matters. In some embodiments, the digital assistant can treat each sub-forum as a separate CS information source.

[0127] In general, public forums are suitable for collecting answers for queries that are difficult to categorize or comprehend for machines but may be easily handled by real people. For example, the digital assistant may fail to comprehend a question or answer such as “How to get rid of ants in my kitchen?” based on the domains and properties it has implemented so far, however, individual members of the public will easily understand the question and may have straight-forward answers right away.

[0128] In some embodiments, the CS information sources may include specialized sources providing more specialized and focused information, such as expert forums, technical support forums, fan-sites for particular subject matter, and the like. The more specialized and focused CS information sources may be more suitable for queries that require specialized knowledge. For example, a user may ask, “Why can't I print this?” after failing to print a webpage opened on her handheld device. The answer may require a diagnostic procedure that only specially-trained technical support staff or other technically savvy individuals can provide. The query regarding the error in printing a webpage from a user device may be more suited for a technical support expert forum than a general purpose public forum or a product review expert forum, for example.

[0129] In some embodiments, the CS information sources include a group of self-identified contributors. Each contributor can be an individual or a group of individuals who have identified themselves as experts in particular fields and agreed to answer questions from the crowd sourcing module in the particular fields.

[0130] For example, a female programmer, who is also an excellent chef, may identify herself as an expert in the fields of computers and cooking. Whenever a query in one of these two fields is issued by the information crowd sourcing module, the programmer can be alerted of the query (e.g., through an automatic notification system of the digital assistant).

[0131] If the programmer is able to provide an answer to the query, she can submit the answer to the information crowd sourcing module (e.g., to the answer monitoring module 506 shown in FIG. 5).

[0132] In some embodiments, queries and answers in a particular field may be posted in a public area for all self-identified experts of the field to see. The answers provided for the queries can be peer reviewed and rated by other self-identified experts in the field. The information crowd sourcing module can utilize the rating of the answers to select the best answers to a query, for example.

[0133] In some embodiments, the interaction between the information crowd sourcing module and the self-identified experts can be through a third-party service, where the third-party service handles the dispatching of queries and collection of answers, as well as screening the self-identified experts, evaluating answers, and/or rating the self-identified experts. In some embodiments, the information crowd sourcing module implements the interfaces and processing components for managing the information crowd sourced from the self-identified experts.

[0134] In some embodiments, the CS information sources include an answer arena where users participate in a game in which participants compete to see who can provide the best answers and the most number of answers to queries posted to the game arena. In some embodiments, the game arena includes many smaller arenas each for a different question domain or subject matter. The game arena can provide rewards for gamers, such as points, credits, and the like. Sometimes, the game arena can be used to collect answers for the more challenging questions, and the participants can utilize various resources they personally have access to as an individual (e.g., either online or in the real world) to figure out an answer. The answers for a particular query can be reviewed and/or voted on by a group panel, or by other disinterested/non-participating users for the particular query.

[0135] In some embodiments, each user can specify a list of preferred CS information sources or the digital assistant of the user may have learned over time that a particular group of CS information sources have worked well for the user. In such embodiments, the information source selection module can choose the suitable CS information sources based on the user's preference or the recommendation by the digital assistant based on past successes.

[0136] In some embodiments, some CS information sources are only available to a specific group of users (e.g., subscribers to a premium CS information source), and the information source selection can be based on the status and identity of the user and whether they have authorization to use particular CS information sources. Other ways of selecting the CS information sources for different users and user requests are possible.

[0137] Once the digital assistant has selected the suitable CS information sources for the user request, the digital assistant can proceed to generate one or more queries based on the user request and any available context information (610). In some embodiments, the query generation is performed by the query generation module 504 shown in FIG. 5. In some embodiments, the queries may be the raw voice input of the user contained in the user request. In some embodiments, the query may be a portion of the voice input of the user, and/or other processed form of the raw voice input. In some embodiments, the query includes some or all of the context information currently associated with the user request. In some embodiments, the query may also include partially instantiated domains and/or concepts related to the user request. In various embodiments, the crowd sourcing module generates different queries for different CS information sources, and according to the respective formats required by the different CS information sources.

[0138] In some embodiments, the CS information sources identified in the information source database 516 can be organized according subject matter and types. Different APIs and/or protocols needed to communicate with each CS information source can be stored in the information source database 516 as well. The query generation module 504 can refer to the CS information source database 516 when formatting the queries for a particular CS information source.

[0139] In some embodiments, the query generation module can generate a query that includes the user request and associated context information their original data form for some CS information sources. The query generation module can also generate a query that includes the user request and context information in a processed form according to the requirements of a CS information source, e.g., according to particular APIs or formatting requirements of the CS information source. In some embodiments, query generation module generates a natural language query that paraphrases the user request with some useful context information, but has all personally identifiable information removed therefrom.

[0140] In some embodiments, the query generation module generates queries that are related to only one or more sub-aspects of the user request. For example, suppose that the user said, “Please make an e-card for me with the words ‘You are the best dad in the world.’” The query generation module may generate a query “How do I make a custom e-card?” The query generation module may also generate other queries such as “What is an e-card?” or “How to make an e-card?” if the word “e-card” and task-flow associated with making an c-card are yet not part of the vocabulary and task-flow models of the digital assistant at the time. In some embodiments, the query generation module uses the natural language processing capabilities of the digital assistant to generate the queries in natural language.

[0141] In some embodiments, the query generation module can generate queries that are natural language variants of the user request. For example, suppose the user said. “Teach me how to cook lobsters.” The query generation module may generate natural language queries such as “How to cook lobsters?” “Lobster cooking tips” “Got an easy lobster recipe?” The crowd sourcing module can use the natural language variants to identify similar questions that have already been asked and answered in the past in various occasions. For example, instead of or in addition to issuing a fresh query generated from the user request to the various CS information sources, the information crowd sourcing module can search on FAQ bulletins, message boards, public forums, and the crowd-source knowledge base for similar or equivalent questions, and use the answers to those similar or equivalent questions to help with formulating a response to the user request.

[0142] In some embodiments, the query generation module identifies user requests that are very similar to one another and can benefit from the answers to the same queries. For example, suppose that the crowd sourcing module has dispatched a first query “Printing error after system upgrade to OS version 7.1.” to a CS information source for a user request “Why can't I print?” accompanied by context information indicating a system upgrade to OS version 7.1, and the first query is now in the answer gathering stage. Further suppose that the information crowd sourcing module now needs to do information crowd sourcing for a second user request, “I upgraded to OS 7.1 and now I can no longer print, what's wrong?” The query generation module will generate a second query based on the second user request, and then recognize that there is already a similar or equivalent query (i.e., the first query) dispatched to one or more CS information sources. Therefore, the crowd sourcing module does not dispatch the newly generated query to the CS information sources that have already received a similar or equivalent query before. Instead, the crowd sourcing module waits for the answers to the first query to be collected, and uses the answers to the first query for generating responses to both user requests.

[0143] The ability to recognize that similar or equivalent queries have been dispatched or answered is important because the information crowd sourcing module handles information crowd sourcing for many user requests received from many users. Many of these user requests may be very similar and the same queries and answers may provide the necessary information to resolve all of these user requests. Thus, by recognizing similarities between user requests, and detecting and filtering out duplicate or nearly duplicate queries generated from the user requests, the information crowd sourcing module can operate more efficiently. The filtering of duplicate and near duplicate queries also helps prevent the need to have contributors answer the same questions over and over again. In some embodiments, the query generation module uses the natural language processing capabilities of the digital assistant to determine if two or more queries are duplicates or near duplicates of each another.

[0144] In some embodiments, the information crowd sourcing module avoids issuing duplicate queries by recognizing the commonalities among different user requests. In some embodiments, the commonality between two user requests can be found based on a large overlap between the domains and properties activated by the two user requests. If two user requests can be fulfilled using the same root solutions or answers (e.g., a common task flow), then the information crowd sourcing only need to be performed for one of the two user requests or a generic combination of the two user requests.

[0145] For example, suppose that, after a new version of a device operating system is released, many users who have upgraded to the new version of operating system may experience similar technical issues due to compatibility with existing applications or due to bugs in the operating system. Therefore, multiple users may issue user requests indicating a problem with their respective devices around the same period of time. The information crowd sourcing module will recognize that the different user requests “Why did my web browser crash?” “What happened to my web browser?” “Why can't I open this webpage?” and “Why can't I open this link from my e-mail?” all relate to the same issue because all of these different user requests map to the same domain of technical questions related to the domain of “web browser application technical issues” in the ontology implemented by the digital assistant. In addition, since devices of the users may collect context information (e.g., the operation that the user was performing right before the user request was received, the current version of the operating system, the device type, and so on), the commonality of the user requests is further reflected in the context information provided along with the user requests. In this example, the commonality of the user requests will be reflected in the context such as the version of the operating system, the time that the user requests were received, the error log that has been recorded, and so on. When these user requests are all directed to the crowd sourcing module around the same time period, the crowd sourcing module will be able to determine and recognize that the information crowd sourcing of these similar user requests can be addressed together using one or more generic queries, such “How to fix the browser problem after operating system upgrade to version x.x, in device XX?” The version information and device information in the generic query is completed in by the context information accompanying the user requests. In some embodiments, some user-specific context information (anonymized to protect user privacy) may be utilized by the experts in the CS information sources to provide answers that may be generally applicable to other users.

[0146] In some embodiments, the information crowd sourcing component establishes a special query pool for handling queries that are each relevant to the fulfillment of multiple similar or equivalent user requests. The crowd sourcing module optionally sends the queries in the special query pool to a team of experts who can provide solutions, answers, or information in a more speedily manner. Once the answer to a query in the special query pool is received, the answer can help resolve a large number of user requests. In some embodiments, the crowd sourcing module establishes certain criteria for determining when a query may enter the special query pool to receive expedited answers.

[0147] The ability to recognize the similarity between user requests also allows the information crowd sourcing module to determine if a current user request is similar or identical to another user request that has already been successfully fulfilled as the result of a previous information crowd sourcing process. If the crowd sourcing module determines that the current user request and the earlier request are identical or sufficiently similar, the crowd sourcing module attempts to fulfill the current user request in a similar manner. The similarity between the two user requests can be determined based on the similarity between the domains and properties that were activated during the natural language processing (intent inference) process for the two user requests.

[0148] In some embodiments, the query generation module uses the results from the natural language processing (intent inference) process to determine what questions will likely bring forth answers that will aid the digital assistant in producing a satisfactory response to the user request. For example, suppose that the user had said, “Find me a restaurant that serves Caipirinha.” The digital assistant would be able to infer that the user wishes to find a restaurant, but would not know what “Caipirinha” stands for. The query generation module will then use the words that were not found in the vocabulary for the property of “restaurant” (e.g., “Caipirinha” in this case) as the subject of the query. Therefore, the query generation module would generate queries such as “What is Caipirinha?” or “What kind of food or drink is called “Caipirinha”? in addition to the user's original question “Which restaurant serves Caipirinha.” In some embodiments, for unknown vocabulary, the query generation module may also provide a few variations of the spelling in the different queries.

[0149] In addition, in some embodiments, the query generation components may also use the ontology (e.g., the hierarchy of nodes and sub-nodes) of the “restaurant” property to formulate a query such as “What kind of cuisine is Caipirinha?” When a reply is received for this query and the digital assistant finds out that “Caipirinha” is a type of alcoholic drink found in South America, then, the digital assistant will be able to focus its search for restaurants that serve South American cuisine and alcoholic beverages, and finally find an answer to the user's original question.

[0150] In some embodiments, the query generation component 504 may pass at least some of the domains and properties that were activated during natural language processing process of the user request to a CS information source that is capable of extracting information embedded in one or more partially completed structured queries representing those domains and properties. The CS information source can use the information extracted from the partially completed structured queries to decide which contributor should receive the natural language or structured crowd sourcing queries generated by the query generation module 604. In some embodiments, the CS information source can also use the extracted information to provide answers to follow-up questions that a contributor may ask regarding the crowd sourcing queries.

[0151] There are many ways that queries can be generated for a user request. Different query generation methods may have different implications in the effectivencss in bringing forth information helpful in responding to user requests, and in the efficiency of query dispatching and answer collection processes. The methods for generating queries provided above are just some of the examples. In some embodiments, the query generation module 504 stores all the queries that have been handled, and the queries that are in the answer collection stage in a queries database (e.g., the queries database 520 in FIG. 5). The queries in the queries database may be clustered based on their commonalities in terms of subject matter. In some embodiments, equivalent or similar queries may be identified as such in the queries database.

[0152] In some embodiments, the crowd sourcing module selects the suitable CS information sources for a user request before the queries for the user request are generated. In some embodiments, however, the queries can be generated from the user request first, and then CS information sources can be selected for each query based on various criteria, e.g., based on the subject matter of the query. In some embodiments, a group of CS information sources can be selected before the queries are generated, and the selected CS information sources can be further refined and assigned to handle different queries after the queries are generated.

[0153] For example, the queries that are generated for a user request may be related to several different domains of knowledge. Thus, different queries may be dispatched to different CS information sources each suitable for answering some of the queries. Suppose that a user asked, “How do I program my thermostat to conserve energy?” This user request may be partially matched to the domain of technical support due to the terms “program” and “thermostat,” and partially matched to the domain of energy conservation due to the terms “conserve energy” in the user request. Based on these two domains, the CS information source selection module may decide to select a technical support forum for a first query “How to program a thermostat?” generated from the user request, while selecting a conservation focused panel for a second query “What room temperatures conserve energy?” generated from the user request.

[0154] Referring back to FIG. 6A, after the queries are generated for each selected CS information source, the queries and selected CS information sources are optionally presented to the user for review and approval (612). In some embodiments, the user is allowed to modify the queries and/or the selection of CS information sources at this time or accept the queries and CS information sources as suggested by the crowd sourcing module. In some embodiments, the digital assistant does not present the queries and selected CS information sources to the user for review or approval, and simply proceeds to send the queries to the CS information sources.

[0155] In some embodiments, natural language representations of the queries are presented to the user for review and approval, but equivalent structured machine-readable queries are sent to the CS information sources. In some embodiments, the crowd sourcing module will present a list of all the different CS information sources available to the user upon request by the user. The user is allowed to choose one or more of the CS information sources to use for the crowd sourcing. In some embodiments, the user is also allowed to modify some or all of the queries that are to be sent to the different CS information sources.

[0156] In some embodiments, if the information crowd sourcing module is already handling a similar user request, and the queries generated for the current user request are already being answered at various CS information sources, the information crowd sourcing module notifies the user that many users have been experiencing the same issue or have the same question, and assures the user that a solution or answer should be arriving soon.

[0157] In some embodiments, if the crowd sourcing module is invoked as part of the failure management process of a digital assistant, the crowd sourcing module stores the user request in a failure status database. Each user request is assigned a unique identifier in the failure status database, and information associated with the user request, such as the original user input, the context information associated with the user request, the reason of the failure, the partial or complete intent inferred from user input, the current status of the crowd sourcing for the user request, and ultimately, the result of the crowd sourcing and output generation, can be stored in the failure status database. Various modules of the crowd sourcing module can updated the failure status database with appropriate information.

[0158] In some embodiments, when a user's voice input or a portion thereof is included a query, the query generation module implements an algorithm to clean the voice input of the user to remove or obfuscate any offensive or sensitive phrases or words from the voice input before releasing the voice input to any public forums. In some embodiments, the query generation module establishes a screening process that prevents queries involving questionable content from being posted onto the public forums. For example, if the query generation module detects sensitive words related to illegal activities (e.g., copyright violations or drug-related activities) in a user request, the information crowd sourcing component can reject the user request, or avoid sending queries related to the user request to the CS information sources.

[0159] Referring back to FIG. 6A, if suitable CS information sources and/or queries have been presented to the user for approval, the crowd sourcing module determines whether the user has approved the CS information sources and/or queries presented to the user (614). If the presented CS information sources and/or queries are approved, the crowd sourcing module can prepare to enter the second stage of the information crowd sourcing described with reference to FIG. 6B. If the user did not approve the CS information sources and queries, but provided suggestions for modifications, the digital assistant modifies the queries and/or selection of CS information sources according to user's suggestions (616). After the queries and/or selection of CS information sources are modified according to the user's suggestions, the crowd sourcing module can prepare to enter the second stage of the information crowd sourcing described with reference to FIG. 6B. If the request for approval of the CS information sources and/or queries is not implemented by the crowd sourcing module, the crowd sourcing module proceeds to the second stage once the queries are generated.

[0160] In some embodiments, if a query generated for a current user request is a duplicate of another query already provided to an CS information source, the information crowd sourcing module does not sent that query to the CS information source again, but simply monitors the answers for the earlier query for both some earlier user request(s) and the current user request.

[0161] FIG. 6B illustrates the second stage of the information crowd sourcing process for the user request. The second stage can continue from the step 616 in FIG. 6A or the “No” branch of the decision 614 in FIG. 6A. In the second stage, the information crowd sourcing module optionally searches the crowd-sourced knowledge base (e.g., the crowd-sourced knowledge base 358 in FIGS. 3A-3B and 5 and 7) to see if any of the approved queries for the current user request already exists in the crowd-sourced knowledge base (622). If one or more of the approved queries already exist or have equivalents in the crowd-sourced knowledge base 358, the information crowd sourcing module uses the answers to those queries as the answers for the one or more approved queries.

[0162] If one or more of the approved queries are not found in the crowd-sourced knowledge base 358, the information crowd sourcing module proceeds to send those one or more approved queries to the approved CS information sources (624). In some embodiments, sending the queries to a CS information source includes posting a message to a public discussion forum or bulletin board, adding a challenge to a game arena, sending a structure query to a database, and/or other communications according to the respective APIs and protocols of the CS information sources.

[0163] After the queries have been dispatched to the CS information sources, the information crowd sourcing module (or an answer monitoring module thereof) monitors the replies received from the CS information sources for the queries dispatched to the CS information sources (626). In some embodiments, the monitoring includes responding to follow-up questions about the queries, and determining when to close the answer gathering period for each of the queries. In some embodiments, the monitoring also involves requesting the user to provide additional information or participating in a live dialogue with one of the CS information sources. In some embodiments, the monitoring also includes determining whether it is appropriate to request the user to provide information or participate in a live dialogue based on the current state of the user. For example, the answer monitoring module may access the user's calendars and determine the user's current location or speed or current engagement, and decide whether it is appropriate to interrupt the user at the present time.

[0164] In some embodiments, after the queries are sent to different CS information sources, the answer monitoring module proactively (e.g., periodically) check to see if any replies or answers have been provided by the CS information sources. Sometimes, the answer monitoring module receives notifications from the CS information sources when one or more replies or answers have been received for a particular query. Sometimes, if the number of answers are abundant for certain CS information sources, e.g., expert forums having many self-identified experts or popular game arenas, the answer monitoring module optionally sets a limited answer gathering period and stops taking more replies or answers after the limited answer gathering period expires. For different types of CS information sources, the manner by which answer to queries are monitored may be different.

[0165] In some embodiments, the answer monitoring module 506 provides additional information to the CS information sources in response to follow-up questions received from the CS information sources. For example, when a reply is received from a CS information source about a query, the answer monitoring module may utilize the natural language processing capabilities and intent inference process of the digital assistant to determine whether the reply seeks additional clarification information or provides an answer. If the reply seeks additional information, the answer monitoring module determines what information is being sought and whether the digital assistant possesses that information. If the information is available, and sharing of the information with the CS information source is not prohibited by the user's privacy policy or preference, the answer monitoring module provides the information to the CS information source. In some embodiments, the answer monitoring module formulates a natural language response that includes the requested additional information and provides the natural language response back to the CS information source. For example, the natural language response can be posted as a follow-up to the original query in a public forum, and all users reading the original query can now see the follow-up information as well.

[0166] In some embodiments, when the answer monitoring module processes the follow-up questions to determine their meanings and intent, the answer monitoring module also utilizes the domains and properties associated with the original user request to help provide context to the follow-up questions.

[0167] In some embodiments, when a follow-up question is received from a CS information source for a particular query, the answer monitoring module initiates a dialogue with the information source (e.g., an expert or a customer support staff). In some embodiments, the answer monitoring system utilizes the natural language processing and dialogue processing capabilities of the digital assistant to facilitate the dialogue with the CS information source. In some embodiments, the dialogue can be carried out continuously in real-time, or intermittently over an extended period of time. For example, the answer monitoring module may engage in a diagnostic process with a live technical support staff, and can answer questions posed by the technical support staff. In some embodiments, the digital assistant handles the follow-up questions without the active participation of the user in some situations, while in other situations, the answer monitoring module may decide that it is appropriate to bring the user into the follow-up dialogue, so that the user can provide information that the digital assistant did not currently possess.

[0168] Referring back to FIG. 6B, once the information crowd sourcing module (or the answer monitoring module thereof) determines that the enough answers have been gathered for the queries associated with a user request, the information crowd sourcing module proceeds to compile and integrate the answers received for the queries, and formulate a response to the user request based on the integrated answers (628). In some embodiments, the answer integration module 508 compiles and integrates the answers received from the queries for the user request into consolidated crowd sourced information, and the response generation module generates the response to the user request based on the consolidated crowd sourced information and any information the digital assistant already possessed before the information crowd sourcing for the user request.

[0169] In some embodiments, when integrating the answers received from different CS information sources, the answer integration module merges and reconciles information received from all of the answers. In some embodiments, outlier answers may be filtered out. In some embodiments, the answer integration module ranks the answers according to factors such as the votes they received, the quality-level or credibility of their respective contributors, the number of duplicates or supporter replies each answer has, and so on.

[0170] In some embodiments, one or more top-ranked answers are selected and a response is formulated to include all of the one or more top-ranked answers (e.g., when the user request seeks an informational answer). In some embodiments, one or more top-ranked answers are merged into a single answer and a response is formulated to include the single answer. In some embodiments, the answers received for a user request may be related to different aspects of the user request, and the response generation module creates a response based on the consolidated knowledge inferred from answers received for the different aspects of the user request.

[0171] In some embodiments, if the user request is for the performance of a task, the information obtained from the answers can be used by the response generation module to generate a task flow for the user request.

[0172] At the second stage shown in FIG. 6B, the response is merely formulated. For a user request seeking an informational answer, the crowd sourced informational answer has not been presented to the user. For a user request seeking performance of a task, the actions for performing the task have not been carried out by the digital assistant.

[0173] Referring back to FIG. 6B, after the digital assistant has made the attempt to formulate a response to the user request using the crowd sourced information (e.g., information from the crowd-sourced knowledge base and/or from the answers to the queries), the crowd sourcing module determines whether the response has been successfully formulated (630). If the information crowd sourcing module determines that, despite the additional information obtained from the answers received from the CS information sources, it is not able to successfully formulate a response to the user request, the information crowd sourcing notifies the digital assistant of the failure, and ceases further action on this user request (632). In some embodiments, the information crowd sourcing module stores a log of the failures, and retries the crowd sourcing at a later time (e.g., when a new CS information source is added).

[0174] In some embodiments, the answer integration module may determine that, at the end of the allowed answer gathering time period, no satisfactory answer or an insufficient number of answers have been received from the different CS information sources. If so, the crowd sourcing module also determines that a response cannot be successfully formulated at this time. For example, sometimes, the answer integration module may determine that the answers received are not of sufficient quality (e.g., based on a threshold quality criterion such as votes, or based on peer review) for generating a satisfactory response to the user. Sometimes, the response generation module may determine that the answers received do not provide sufficient information or the information is not of sufficient specificity and clarity to help the response generation module to generate a task flow to accomplish the requested task.

[0175] Referring back to FIG. 6B, if the information crowd sourcing module determines that it is able to formulate a response to the user request with the help of the additional information obtained from the crowd sourcing process, the digital assistant can prepare to enter the final stage of providing the crowd sourced response to the user. FIG. 6C illustrates the final stage of the information crowd sourcing, i.e., presenting the crowd sourced response to the user.

[0176] As shown in FIG. 6C, before the digital assistant provides the crowd sourced response to the user, the digital assistant first determines whether it is an appropriate time to reengage the user regarding the earlier user request (642). For example, the digital assistant may determine the current status of the user based on various signals such as the speed by which the user is moving, the current location of the user, the current time, whether there is any meeting scheduled on the user's calendar for the present time, whether the user device is currently idle or being used for other tasks, and so on. Based on the status of the user, the digital assistant determines whether it is appropriate to interrupt the user or wait for a better opportunity later.

[0177] If the digital assistant determines that is it not appropriate to actively reengage the user at the present time, the digital assistant optionally presents a silent notification to the user (e.g., a silent status indicator on the user device) about the availability of the crowd sourced response, such that the user can initiate a dialogue with the digital assistant at a time he or she deems fit or the digital assistant can just wait until a suitable opportunity to reengage with the user is presented (644). For example, at the end of a conversation with the user regarding other matters, the digital assistant can ask the user whether the user wishes to see the crowd sourced response for an earlier user request. In some embodiments, the digital assistant checks periodically to determine if it is appropriate the reengage the user regarding the crowd sourced response for the earlier user request.

[0178] If the digital assistant determines that it is appropriate to reengage the user regarding the earlier user request, the digital assistant proceeds to provide the crowd sourced response to the user. Depending on whether the user request was seeking an informational answer or the performance of a task, different steps are taken by the digital assistant. For example, the digital assistant first determines whether the user request was for an informational answer (648). If the user request seeks an informational answer, the digital assistant proceeds to present the informational answer formulated based on the crowd sourced information to the user (650).

[0179] Alternatively, if the user request seeks the performance of a task, the digital assistant optionally presents the task flow formulated for the user request, and asks the user to confirm that the user wishes to proceed with the execution of the task flow (652). The digital assistant determines whether the user approves the task flow (654). If the user does not approve the task flow, the digital assistant considers the crowd sourcing answer as unsatisfactory, and ceases further actions regarding the user request (662). If the digital assistant determines that the task flow is approved, the digital assistant proceeds to execute the task flow, e.g., using the facilities of the task flow processing and execution infrastructure of the digital assistant (656).

[0180] In some implementations, before a crowd sourced task flow is executed for a user request, the crowed sourcing module forwards the crowd sourced task flow to a board of trusted experts for review and verification. In some embodiments, the digital assistant optionally simulates an operating environment of the user's device and associated services, and executes the crowd sourced task flow in a simulation before the task flow is executed on the user's device.

[0181] Once either the informational answer is presented to the user, or the execution of the proposed task flow is successfully completed, the digital assistant asks the user whether the response provided for the user request was satisfactory (658). If, based on the user's feedback, the digital assistant determines that the crowd sourced response is still not satisfactory to the user, the digital assistant considers the crowd sourced response an unsatisfactory response, and ceases further actions regarding the user request, or escalates the user request to other remedial processes, such as sending the user request to a human personal assistant of the user, or the like.

[0182] Alternatively, if the digital assistant determines that the user is satisfied with the crowd sourced response (e.g., based on the user's feedback), the digital assistant (e.g., the knowledge-base building module 512 shown in FIG. 5) proceeds to record the crowd sourced response, the user request, and/or the queries and answers that contributed to the successful fulfillment of the user request to the crowd-sourced knowledge base (660). In some embodiments, the digital assistant also records other information, such as the CS information sources that provided the best answers to the queries, and the follow-up questions and answers exchanged during the answer gathering stage.

[0183] In some embodiments, the user requests, queries, and answers can be organized by one or more ontologies. The ontologies optionally model the user requests, queries and answers as nodes of domains and properties having various associated vocabulary, attributes, parameters, and task flows. The nodes are organized in one or more hierarchies and may be interrelated by their super-properties, sub-properties, vocabulary, attributes, parameters, task flows, and so on. The organization of the crowd-sourced knowledge base allows the digital assistant to determine whether satisfactory answers to a query or information request already exist in the crowd-sourced knowledge base, and whether a satisfactory response for a user request for performance of a task can be found in the crowd-sourced knowledge base. In some embodiments, the searching for a user request or answer in the crowd-sourced knowledge ontology based on words in the user request or answer is analogous to the identification of an actionable intent based on the words in the user's utterance. For example, a user request node in the crowd-sourced knowledge ontology is optionally associated with one or more property nodes that define different aspects of the user request. Based on the respective vocabulary associated with each node in the crowd-sourced knowledge ontology, a new user request will trigger or activate many of the property nodes associated with an existing user request in the crowd-sourced knowledge base, and be identified as similar to the existing user request.

[0184] In some embodiments, the failure to provide a satisfactory response to a user request is identified by an offline analysis of the data logs of the operation and usage of the digital assistant, rather than in real time while responding to the user. In some embodiments, the failure is identified by automated or manual analysis of log data. For example, in some embodiments, failures are identified in the usage logs by evidence for a task not being completed, or answers not being available from information sources, or from users' repeated requests with similar intents, or from users indicating frustration (e.g., turning off or otherwise interrupts the digital assistant's response or action), and/or other data analysis techniques. In some embodiments, data on failures identified offline are sent to the crowd-sourcing module in a similar manner as failures identified in real-time. In these embodiments, the results of crowd-sourced answers are available for use in future real-time sessions, but generally are not delivered back to the users who experienced the failures that were identified offline.

[0185] FIG. 7 illustrates that, in some embodiments, the digital assistant 326 or the components responsible for providing immediate response to the user are provided separately (e.g., by third-parties, or on different infrastructures) from the failure management module 340, the crowd sourcing module 342, and the crowd-sourced knowledge base 358. However, the functionalities provided by the failure management module 340, the crowd sourcing module 342, and the crowd sourced knowledge base 358 herein, are also applicable to embodiments in which the failure management module 340, the crowd sourcing module 342, and/or the crowd sourced knowledge base 358 are part of the digital assistant system 326.

[0186] Referring now to FIG. 7, content of the usage logs 370 of the digital assistant 326 can be analyzed for failures 720. In some embodiments, a failure detection component 720 periodically, or upon request, scans the usage logs 370 for signals, and/or patterns indicative of failures. For example, if the digital assistant implements a catch-all response (e.g., sending the user to a web search interface whenever the digital assistant has failed to find or provide a satisfactory response after two attempts), each issuance of the catch-all response is logged in the usage log 370. When the failure detection module 720 makes a query against the usage logs 370 for all instances in which a catch-all response has been provided for a particular type of user request, the failure detection module 720 would receive event logs of user requests that had failed to produce a satisfactory response. The failure detection module 720 optionally sends to the failure management module 340 the event logs of these user requests. In some embodiments, the failure management module 340 processes these failures in a similar manner as failures that are detected in real-time. The difference between real-time and offline processing of the failures lies in the lack of direct interactions with the user (e.g., requesting permissions, or confirmations, etc.) during the processing. In addition, in most scenarios, if and when these failures are resolved by crowd-sourcing, the correct responses are not presented back to users who initially experienced those failures. Instead, the offline processing is for the benefit of users that make similar requests in the future.

[0187] In some embodiments, the information obtained by failure-driven crowd-sourcing and stored in crowd-sourced knowledge base 358 is integrated back into the real-time response mechanisms (e.g., the STT processing module 330, the natural language processing module 332, dialogue processing module 334, the task flow processing module 336, the service processing module 356, and the models used by these processing modules) used by the digital assistant 326. For example, if a question posed by a user was not answered by the digital assistant, and the failure was identified and sent to crowd sourcing as described herein, and successfully answered, then the answer is optionally added to the primary information sources used by the digital assistant. The next time that the same question is asked, the assistant optionally uses the updated information sources to answer the question directly without invoking the crowd-sourcing processes.

[0188] In some embodiments, a crowd-sourcing incorporation module 730 is implemented, either as a standalone module or part of the failure management module 340. The crowd-sourcing incorporation module 730 operates in a batch mode to analyze successful crowd-sourced information in the crowd-sourced knowledge base 358, and add that information into the appropriate information stores (e.g., acoustic models, speech models, vocabulary, ontology, language models, task flow models, service models, etc.) of the digital assistant 326.

[0189] In some embodiments, the crowd-sourced information is used to update the vocabulary database 344 used in natural language processing. For example, when the failure involved a failure to recognize certain words which were then later associated by the crowd-sourcing process with nodes in the ontology, as described in earlier parts of the specification, the new vocabulary is optionally added to vocabulary database 344 and indexed by nodes in ontology 360 so that on subsequent requests the assistant can correctly infer the intent based on the new vocabulary.

[0190] In some embodiments the crowd-sourced information is used to update the task flow models 354. For example, the crowd-sourced response to a diagnostic problem might suggest that the next task flow step is to ask the user a certain question to obtain specific diagnostic data. In some embodiments, such task flows are added to the task flow models 354 after an expert panel review, as described in earlier part of the specification.

[0191] In some embodiments the crowd-sourced information is used to update the service models 356 used by service processing module 338. For example, the crowd-source response to a certain question might be a referral to use a particular service. For instance, if the user asked “where can I get blue suede shoes” the crowd-sourced response might be a referral to use a service representing an online shoe sales information site. This is different from storing a specific answer to where to buy the shoes. It is the information used by an assistant to call the shoe sales information service with a query about blue suede shoes, which in turn might respond with stores and inventory results for blue suede shoes.

[0192] In some embodiments, the crowd-sourced knowledge base 358 is used directly by external services which are employed by the digital assistant to respond to user requests. For example, a digital assistant sometimes calls on the services of several question answering services (e.g., “external services” in FIG. 3B). One of these services is optionally a question answering service that uses questions and answers stored in the crowd-sourced knowledge base 358 to provide answers back to the digital assistant 358. In these embodiments, some of these questions are directly answered by the assistant in real-time without requiring the information to be incorporated into the response mechanisms of the digital assistant 326 (e.g., through the operation of the crowd-sourcing incorporation module 730). For example, in some embodiments, the digital assistant optionally uses the information in the crowd-sourced knowledge base 358, either directly or through a third-party service, for a period of time to respond to real-time requests, and further evaluate the user feedback before allowing the crowd-sourcing incorporation module 730 to modify the response mechanism of the digital assistant 326 using such information.

[0193] In some embodiments, when preparing a response to the user, the digital assistant, in addition to calling on a web search engine on unknown queries, also calls a service that is powered by a crowd-sourced knowledge base (e.g., the crowd-sourced knowledge base 358 or other third-party crowd-sourced knowledge bases). In some embodiments, if the crowd-sourced knowledge base powered service returns an answer, the digital assistant provides that answer to the user instead of the web search (e.g., the catch-all response). If the crowd-sourced knowledge base powered service does not return an answer, the digital assistant provides the answer received from the web search (e.g., the catch-all response). In this way, the competence of the digital assistant can be automatically grown as it feeds failures to crowd-sourced services and they provide suitable responses on subsequent requests.

[0194] It should be understood that the particular order in which the operations in the flow charts have been described is merely exemplary and is not intended to indicate that the described order is the only order in which the operations could be performed. One of ordinary skill in the art would recognize various ways to reorder the operations described herein. In addition, unless explicitly specified, features described in various embodiments may be combined and used together. Additionally, it should be noted that details of other processes described herein may be applied in addition to, instead of, or in conjunction with the operations described with reference to the figures.

[0195] The foregoing description, for purpose of explanation, has been described with reference to specific embodiments. However, the illustrative discussions above are not intended to be exhaustive or to limit the invention to the precise forms disclosed. Many modifications and variations are possible in view of the above teachings. The embodiments were chosen and described in order to best explain the principles of the invention and its practical applications, to thereby enable others skilled in the art to best utilize the invention and various embodiments with various modifications as are suited to the particular use contemplated.