Abstract: A task-oriented dialog system determines an endpoint in a user utterance by receiving incremental portions of a user utterance that is provided in real time during a task-oriented communication session between a user and a virtual agent (VA). The task-oriented dialog system recognizes words in the incremental portions using an automated speech recognition (ASR) model and generates semantic information for the incremental portions of the utterance by applying a natural language processing (NLP) model to the recognized words. An acoustic-prosodic signature of the incremental portions of the utterance is generated using an acoustic-prosodic model. The task-oriented dialog system can generate a feature vector that represents the incrementally recognized words, the semantic information, the acoustic-prosodic signature, and corresponding confidence scores of the model outputs. A model is applied to the feature vector to identify a likely endpoint in the user utterance.

Abstract: A natural language understanding (NLU) system generates in-place annotations for natural language utterances or other types of time-based media based on stand-off annotations. The in-place annotations are associated with particular sub-sequences of an annotation, which provides richer information than stand-off annotations, which are associated only with an utterance as a whole. To generate the in-place annotations for an utterance, the NLU system applies an encoder network and a decoder network to obtain attention weights for the various tokens within the utterance. The NLU system disqualifies tokens of the utterance based on their corresponding attention weights, and selects highest-scoring contiguous sequences of tokens between the disqualified tokens. In-place annotations are associated with the selected sequences.

Abstract: Structure of conversations between users and agents and/or systems is discovered and interactively displayed to analysts, thereby better supporting development of automated conversation handling systems for different domains. A corpus of prior dialogs of users with agents (without preexisting semantic labels indicating purposes for different parts of the dialogs) is taken as input, and embeddings are generated for textual units (e.g., rounds) of the dialogs. The embeddings are used to cluster the textual units, and the clusters and their relationships are visualized within a user interface that analysts may use to explore and fine-tune the structure of the conversations.

Abstract: A speech recognition system includes, or has access to, conventional speech recognizer data, including a conventional acoustic model and pronunciation dictionary. The speech recognition system generates restructured speech recognizer data from the conventional speech recognizer data. When used at runtime by a speech recognizer module, the restructured speech recognizer data produces more accurate and efficient results than those produced using the conventional speech recognizer data. The restructuring involves segmenting entries of the conventional pronunciation dictionary and acoustic model according to their constituent phonemes and grouping those entries with the same initial N phonemes, for some integer N (e.g., N=3), and deriving a restructured dictionary with a corresponding semi-word acoustic model for the various grouped entries.

Abstract: A virtual assistant system for communicating with customers uses human intelligence to correct any errors in the system AI, while collecting data for machine learning and future improvements for more automation. The system may use a modular design, with separate components for carrying out different system functions and sub-functions, and with frameworks for selecting the component best able to respond to a given customer conversation.

Abstract: A virtual assistant system for communicating with customers uses human intelligence to correct any errors in the system AI, while collecting data for machine learning and future improvements for more automation. The system may use a modular design, with separate components for carrying out different system functions and sub-functions, and with frameworks for selecting the component best able to respond to a given customer conversation. The system may have agent assistance functionality that uses natural language processing to identity concepts in a user conversation and to illustrate that concepts within a graphical user interface of a human agent so that the human agent can more accurately and more rapidly assist the user in accomplishing the user's conversational objectives.

Abstract: Within an environment in which users converse at least partly with human agents to accomplish a desired task, a server assists the agents by identifying workflows that are most applicable to the current conversation. Workflow selection functionality identifies one or more candidate workflows based on techniques such as user intent inference, conversation state tracking, or search, according to various embodiments. The identified candidate workflows are either automatically selected on behalf of the agent, or are presented to the agent for manual selection.

Abstract: A virtual assistant system for communicating with customers uses human intelligence to correct any errors in the system AI, while collecting data for machine learning and future improvements for more automation. The system may use a modular design, with separate components for carrying out different system functions and sub-functions, and with frameworks for selecting the component best able to respond to a given customer conversation.

Abstract: Systems and methods are disclosed herein for discerning aspects of user speech to determine user intent and/or other acoustic features of a sound input without the use of an ASR engine. To this end, a processor may receive a sound signal comprising raw acoustic data from a client device, and divides the data into acoustic units. The processor feeds the acoustic units through a first machine learning model to obtain a first output and determines a first mapping, using the first output, of each respective acoustic unit to a plurality of candidate representations of the respective acoustic unit. The processor feeds each candidate representation of the plurality through a second machine learning model to obtain a second output, determines a second mapping, using the second output, of each candidate representation to a known condition, and determines a label for the sound signal based on the second mapping.

Abstract: A computer-implemented method for providing agent assisted transcriptions of user utterances. A user utterance is received in response to a prompt provided to the user at a remote client device. An automatic transcription is generated from the utterance using a language model based upon an application or context, and presented to a human agent. The agent reviews the transcription and may replace at least a portion of the transcription with a corrected transcription. As the agent inputs the corrected transcription, accelerants are presented to the user comprising suggested texted to be inputted. The accelerants may be determined based upon an agent input, an application or context of the transcription, the portion of the transcription being replaced, or any combination thereof. In some cases, the user provides textual input, to which the agent transcribes an intent associated with the input with the aid of one or more accelerants.

Abstract: Systems and methods are disclosed for determining a move driven by an interaction. In some embodiments, a processor determines an operational state of an interaction with a user based on parameter values of a data structure. The processor identifies a plurality of candidate moves for changing the operational state by determining a domain in which the interaction is occurring, retrieving a set of candidate moves that correspond to the domain from a knowledge graph, and adding the set to the plurality of candidate moves. The processor encodes input of the user received during the interaction into encoded terms, and determines a move for changing the operational state based on a match of the encoded terms to the set of candidate moves. The processor updates the parameter values of the data structure based on the move to reflect a current operational state led to by the move.

Abstract: A masking system prevents a human agent from receiving sensitive personal information (SPI) provided by a caller during caller-agent communication. The masking system includes components for detecting the SPI, including automated speech recognition and natural language processing systems. When the caller communicates with the agent, e.g., via a phone call, the masking system processes the incoming caller audio. When the masking system detects SPI in the caller audio stream or when the masking system determines a high likelihood that incoming caller audio will include SPI, the caller audio is masked such that it cannot be heard by the agent. The masking system collects the SPI from the caller audio and sends it to the organization associated with the agent for processing the caller's request or transaction without giving the agent access to caller SPI.

Abstract: Machine translation is used to leverage the semantic properties (e.g., intent) already known for one natural language for use in another natural language. In a first embodiment, the corpus of a first language is translated to each other language of interest using machine translation, and the corresponding semantic properties are transferred to the translated corpuses. Semantic models can then be generated from the translated corpuses and the transferred semantic properties. In a second embodiment, given a first language for which there is a semantic model, if a query is received in a second, different language lacking its own semantic model, machine translation is used to translate the query into the first language. Then, the semantic model for the first language is applied to the translated query, thereby obtaining the semantic properties for the query, even though no semantic model existed for the language in which the query was specified.

Abstract: An interactive response system directs input to a software-based router, which is able to intelligently respond to the input by drawing on a combination of human agents, advanced recognition and expert systems. The system utilizes human “intent analysts” for purposes of interpreting customer input. Automated recognition subsystems are trained by coupling customer input with IA-selected intent corresponding to the input, using model-updating subsystems to develop the training information for the automated recognition subsystems.

Abstract: A natural language processing system has a hierarchy of user intents related to a domain of interest, the hierarchy having specific intents corresponding to leaf nodes of the hierarchy, and more general intents corresponding to ancestor nodes of the leaf nodes. The system also has a trained understanding model that can classify natural language utterances according to user intent. When the understanding model cannot determine with sufficient confidence that a natural language utterance corresponds to one of the specific intents, the natural language processing system traverses the hierarchy of intents to find a more general user intent that is related to the most applicable specific intent of the utterance and for which there is sufficient confidence. The general intent can then be used to prompt the user with questions applicable to the general intent to obtain the missing information needed for a specific intent.

Abstract: An interactive response system combines human intelligence (HI) subsystems with artificial intelligence (AI) subsystems to facilitate overall capability of multi-channel user interfaces. The system permits imperfect AI subsystems to nonetheless lessen the burden on HI subsystems. A combined AI and HI proxy is used to implement an interactive omnichannel system, and the proxy dynamically determines how many AI and HI subsystems are to perform recognition for any particular utterance, based on factors such as confidence thresholds of the AI recognition and availability of HI resources. Furthermore the system uses information from prior recognitions to automatically build, test, predict confidence, and maintain AI models and HI models for system recognition improvements.

Abstract: A computer-implemented method for providing agent assisted transcriptions of user utterances. A user utterance is received in response to a prompt provided to the user at a remote client device. An automatic transcription is generated from the utterance using a language model based upon an application or context, and presented to a human agent. The agent reviews the transcription and may replace at least a portion of the transcription with a corrected transcription. As the agent inputs the corrected transcription, accelerants are presented to the user comprising suggested texted to be inputted. The accelerants may be determined based upon an agent input, an application or context of the transcription, the portion of the transcription being replaced, or any combination thereof. In some cases, the user provides textual input, to which the agent transcribes an intent associated with the input with the aid of one or more accelerants.

Abstract: A speech interpretation module interprets the audio of user utterances as sequences of words. To do so, the speech interpretation module parameterizes a literal corpus of expressions by identifying portions of the expressions that correspond to known concepts, and generates a parameterized statistical model from the resulting parameterized corpus. When speech is received the speech interpretation module uses a hierarchical speech recognition decoder that uses both the parameterized statistical model and language sub-models that specify how to recognize a sequence of words. The separation of the language sub-models from the statistical model beneficially reduces the size of the literal corpus needed for training, reduces the size of the resulting model, provides more fine-grained interpretation of concepts, and improves computational efficiency by allowing run-time incorporation of the language sub-models.

Abstract: A natural language processing system has a hierarchy of user intents related to a domain of interest, the hierarchy having specific intents corresponding to leaf nodes of the hierarchy, and more general intents corresponding to ancestor nodes of the leaf nodes. The system also has a trained understanding model that can classify natural language utterances according to user intent. When the understanding model cannot determine with sufficient confidence that a natural language utterance corresponds to one of the specific intents, the natural language processing system traverses the hierarchy of intents to find a more general user intent that is related to the most applicable specific intent of the utterance and for which there is sufficient confidence. The general intent can then be used to prompt the user with questions applicable to the general intent to obtain the missing information needed for a specific intent.

Abstract: An interactive response system directs input to a software-based router, which is able to intelligently respond to the input by drawing on a combination of human agents, advanced recognition and expert systems. The system utilizes human “intent analysts” for purposes of interpreting customer input. Automated recognition subsystems are trained by coupling customer input with IA-selected intent corresponding to the input, using model-updating subsystems to develop the training information for the automated recognition subsystems.