Abstract: A computer-implemented method includes generating an empirically derived acoustic confusability measure by processing example utterances and iterating from an initial estimate of the acoustic confusability measure to improve the measure. The method can further include using the acoustic confusability measure to selectively limit phrases to make recognizable by a speech recognition application.

Abstract: Provided are systems and methods for mapping customer feedback to one or more processes. In one example, the method may include receiving text content that includes feedback, mapping, via execution of a mapping algorithm, the text content to a plurality of processes based on a synonym graph, generating mapping values for the plurality of processes, where a mapping value for a process is generated based on distance within the synonym graph of a mapping between a word in the text content and a word identifying the process, and determining a process among the plurality of processes that is most correlated to the feedback based on the generated mapping values.

Abstract: A system (100) includes a feature extraction engine (130), a finding code comparison engine (140), and a mapping interface (160). The feature extraction engine (130) extracts features of a statement of a finding code in a source vocabulary (110) and features of a second statement of a second finding code in a target vocabulary (112). The finding code comparison engine (140) determines a mapping between the statement of the source vocabulary and the second statement of the target vocabulary by comparing the extracted features based on at least one identified concept that comprises the extracted features. The mapping interface (160) presents the determined mapping on a display device (162).

Abstract: Systems and methods may be used to generate and use intent predictions to enhance user experience. The intent predictions may describe the data required to resolve a user request included in a user input (e.g., question, search query, and the like) submitted by a user. The intent predictions may be generated using a machine learning model that comprises a model framework for extracting features and classifying user inputs into intent classes based on the extracted features. The intent predictions may be integrated into an information service to improve business metrics including contact rate, transfer rate, helpful rate, and net total promoter score.

Abstract: Aspects of the subject disclosure may include, for example, a method in which a processing system collects information associated with trouble tickets each including a problem abstract and a log text. The method includes analyzing the log text to obtain a problem resolution for that ticket; defining ticket clusters according to the problem abstracts, and labeling the clusters. The processing system creates a library of the labeled clusters, each entry including a cluster label, a problem abstract for that cluster, and a resolution summary for that problem abstract, indicating a mapping of the problem abstract to the resolution summary for that cluster. The method includes training, based on the mapping, machine-learning applications for a predicted resolution summary for each cluster and for classifying a new ticket. The method includes assigning the new ticket to a cluster according to the classifying. Other embodiments are disclosed.

Abstract: The disclosure relates to identifying a user intent of a statement. When a user intent of a to-be-analyzed statement needs to be determined, at least one historical statement of the to-be-analyzed statement can be determined from an human-computer interaction session. A historical vector determined with reference to a statement vector of the to-be-analyzed statement and a statement vector of the at least one historical statement may comprise related content in the historical statement, and the content may be associated with the content of the to-be-analyzed statement. If the user intent of the to-be-analyzed statement is determined with reference to the historical vector, then the content related to the to-be-analyzed statement comprised in the historical vector may be used as an additional determination reference, so that the accuracy of determining the user intent of the to-be-analyzed statement can be improved.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for information extraction using natural language processing. One of the methods includes determining, for one or more tokens from a plurality of tokens that represent an unstructured sentence, a token type from a plurality of predetermined token types that indicates an element type for a phrase that corresponds to the token and has one or more properties using dependency data and a part-of-speech label for the token; assigning, for a token whose associated dependency data indicates that the token has a child, data for the child token to one of the one or more properties for the token type of the token; and providing, for use by a downstream semantic system and for the token, a textual representation of the phrase for the token and the phrases for one or more of the child tokens.

Abstract: A computer-executable method for moderating publication of data content with a moderator tool. The data contents are labelled as acceptable or unacceptable. The moderator tool receives the training data and executes a first algorithm that identifies features that exist in the training data and extracts them and ending up with a feature space. The moderator tool executes a second algorithm in the feature space for defining a distribution of data features that differentiate between the acceptable contents and the unacceptable contents in order to create a moderation model. When the moderator tool receives a new data content to be moderated, it executes the moderator tool on the new data content for identifying the data features in the new data content to be moderated in accordance with the moderation model created, and for producing a moderation result for the new data content by indicating whether the new data content is acceptable.

Abstract: A method includes: initializing a list of token embeddings, each of the token embeddings corresponding to a tokenized word from text in a corpus of text; generating a graph for a group of consecutive words s from the text, said graph including binary relations between pairs of tokenized words of the group of consecutive words; selecting the token embeddings representing the words of the group of consecutive words from the list of token embeddings; computing a tensor of binary relations as the product between a matrix of the selected token embeddings and a tensor representing discourse relations, the computed tensor representing the binary relations between the pairs of tokenized words; computing a loss using the computed tensor; optimizing the list of token embeddings using the computed loss. The above may be repeated until the computed loss is within a predetermined range.

Abstract: A role switching method of robots with similar appearance includes: at a robot, the robot being capable of communicating with other robots within a predetermined range: identifying other robots having the same appearance as the robot itself within the predetermined range; and performing role exchange with the other robots having the same appearance as the robot itself within the predetermined range.

Abstract: Disclosed is a system for generating a recommendation associated with an extraction rule for big-data analysis. The system may receive a set of data blocks and an extraction rule. The extraction rule comprises an identifier and a first range associated with the identifier in the set of data blocks. The system identifies a set of words associated with the identifier in one or more data blocks from the set of data blocks based on a provenance knowledge. The provenance knowledge is indicative of a relationship between the word and the identifier. The system identifies a set of locations associated with the set of words in the one or more data blocks. The system generates a second range associated with the identifier based on the set of locations. The system recommends the second range for modifying the extraction rule for big-data analysis based on an accuracy change.

Abstract: An approach to induction of unknown terms into a term taxonomy graph may be provided. The approach may include analyzing a domain specific corpus to generate a term taxonomy graph using a term taxonomy graph generation model with a term knowledge base and determining which terms within the domain specific corpus are out of vocabulary (OOV) terms. The approach may also analyze the terms in the domain specific corpus with a semantic representation model to generate feature vectors of the OOV terms and terms known within the generated term taxonomy graph. The approach may determine if an OOV can be a hyponym of a term within the term taxonomy graph based on the feature vectors and insert the OOV term into the graph at the appropriate location.

Abstract: The embodiments describe a generative-discriminative (GeDi) language modeling for determining a next token in a text sequence. A class conditional language model and a positive control code determine a first class conditional probability for each token candidate. The class conditional language model and a negative control code determine a second class conditional probability for the each token candidate. A logarithmic probability difference between the first class conditional probability and the second class conditional probability is determined for each token candidate. An unconditional language model determines an unconditional probability for each token candidate. A combined probability is determined by combining the unconditional probability and the logarithmic probability difference for each token candidate. The next token is selected from the token candidates based on the combined probabilities of the token candidates.

Abstract: An audio encoder has a first information sink oriented encoding branch such as a spectral domain encoding branch, a second information source or SNR oriented encoding branch such as an LPC-domain encoding branch, and a switch for switching between the first and second encoding branches, the second encoding branch having a converter into a specific domain different from the spectral domain such as an LPC analysis stage generating an excitation signal, and the second encoding branch having a specific domain coding branch such as LPC domain processing branch, and a specific spectral domain coding branch such as LPC spectral domain processing branch, and an additional switch for switching between the specific domain coding branch and the specific spectral domain coding branch. An audio decoder has a first domain decoder, a second domain decoder, and a third domain decoder as well as two cascaded switches for switching between the decoders.

Abstract: An information processing apparatus includes a voice acquisition section, a reliability generation section, and a processing execution section. The voice acquisition section acquires an ambient voice. The reliability generation section generates reliability indicating a degree in which the acquired voice is uttered from the particular position on the basis of a predetermined transfer characteristic. As the predetermined transfer characteristic, a phase difference or acoustic characteristic of the voice can be assumed. The processing execution section executes a process according to the generated reliability. As the process according to the reliability, a notification according to the reliability or a predetermined command can be assumed to be executed.

Abstract: The present system ensures that external noise sources do not taint sensor data recorded about a user's sleep parameters, to reduce false positives and false negatives, and provides a more accurate record of the user's sleep sounds, by cycling on and off noise sources. Additionally, the present system, in one embodiment, may be used to track changes in sound patterns from the user and from the environment. This may lead to early warning of health issues, as well as issues with the devices in the bedroom.

Abstract: Methods and systems for multi-modality soft-agents for an enterprise virtual assistant tool are disclosed. An exemplary method comprises capturing, with a computing device, one or more user requests based on at least one multi-modality interaction, populating, with a computing device, soft-queries to access associated data sources and applications, and mining information retrieved by executing at least one populated soft-query. A soft-query is created from user requests. A multi-modality user interface engine annotates the focus of user requests received via text, speech, touch, image, video, or object scanning. A query engine populates queries by identifying the sequence of multi-modal interactions, executes queries and provides results by mining the query results. The multi-modality interactions identify specific inputs for query building and specific parameters associated with the query. A query is populated and used to generate micro-queries associated with the applications involved.

Abstract: A computer-implemented method identifies one or more concepts from a document, wherein the document comprises information on completing a task. The method computes a relevance value of an utterance with respect to completing the task using the one or more identified concepts. The method removes the utterance from a dialog model to be used for completing the task when the relevance value of the utterance is below a given threshold value.

Abstract: Systems and methods for presence ground truth approximation and utilization are disclosed. For example, a system detects the presence of a predefined subject, such as a person associated with a given user profile, and/or determines that authentication criteria for performing an action in association with the user profile has been satisfied. A period of time to associate data is determined, and data of one or more data types is labeled as being associated with the speaker identification event. That data may be formatted and input into one or more models to train those models to more accurately detect presence and/or determine whether authentication of a user profile should succeed.

Abstract: Techniques for selectively requesting feedback from users are described. A system may receive a user input and perform an action responsive to the user input. The system may determine whether feedback should be requested from the user. Such determination may be based on various signals such as the user's present emotional state, whether the user input included profanity, whether the user input interrupted an output of the system, etc. When the system determines feedback should be requested, the system may select a feedback prompt pre-established by a domain, skill, or the like. After the system outputs the response to the user input, the system may output the feedback prompt.