Abstract: A method of parametric coding of a multichannel digital audio signal including coding a signal arising from a channels reduction processing applied to the multichannel signal and coding spatialization information of the multichannel signal. The method includes the following acts: extraction of a plurality of items of spatialization information of the multichannel signal; obtaining at least one representation model of the extracted spatialization information; determination of at least one angle parameter of a model obtained; coding the at least one determined angle parameter so as to code the spatialization information extracted during the coding of spatialization information. Also provided are a method for decoding such a coded signal and corresponding coding and decoding devices.

Abstract: Provided are a data processing method, an electronic device and a storage medium. The method include: acquiring pending word vectors; acquiring multi-head attention vectors based on the pending word vectors, the multi-head attention vectors including first and second multi-head attention vectors; acquiring, based on the first multi-head attention vector, a first hidden semantic representation corresponding to the sentence; acquiring, based on the second multi-head attention vector, a second hidden semantic representation corresponding to the target word; acquiring a classification feature vector, based on the first and second hidden semantic representations; and acquiring, based on the classification feature vector, a sentiment polarity category of the sentence, and performing a sentiment classification on the sentence.

Abstract: An embodiment for detecting and mitigating bias is provided. The embodiment may include receiving real-world data from a database. The embodiment may also include creating an inverted index from the real-world data. The embodiment may further include analyzing words in the inverted index. The analyzation may identify a plurality of categories in the real-world data. The embodiment may also include generating a structure template containing various entities within each category of the plurality of categories. The embodiment may further include receiving a test record of the structure template. The embodiment may also include providing alternative entities in the test record where bias is likely to occur. The embodiment may further include storing the test record. The embodiment may also include in response to determining bias exists, indicating a corrective action.

Abstract: The present disclosure relates to generating digital responses based on digital dialog states generated by a neural network having a dynamic memory network architecture. For example, in one or more embodiments, the disclosed system provides a digital dialog having one or more segments to a dialog state tracking neural network having a dynamic memory network architecture that includes a set of multiple memory slots. In some embodiments, the dialog state tracking neural network further includes update gates and reset gates used in modifying the values stored in the memory slots. For instance, the disclosed system can utilize cross-slot interaction update/reset gates to accurately generate a digital dialog state for each of the segments of digital dialog. Subsequently, the system generates a digital response for each segment of digital dialog based on the digital dialog state.

Abstract: Provided in the present disclosure are an electronic device and a natural language generation method thereof. The natural language generation method of the electronic device comprises: obtaining input data with information on a plurality of slots included for generating a response; obtaining a natural language corresponding to the input data by inputting the information on the plurality of slots to one of artificial intelligence models trained for obtaining a natural language generation template and a natural language; and outputting the obtained natural language.

Abstract: The present disclosure relates to chatbot systems, and more particularly, to techniques for obtaining data items for input to a chatbot. In certain embodiments, a chatbot system includes a component that can be invoked by a chatbot in the chatbot system to obtain data items needed by the chatbot. The component can be invoked based on a reference to the component in a dialog flow definition configured for the chatbot. The reference to the component can indicate a composite entity that the component will use to determine how the data items are obtained from a user. The composite entity acts as a container for the data items and may be configured separately from the dialog flow definition of the chatbot. The data items can be obtained based on rules specified in a composite entity definition configured for the composite entity.

Abstract: Techniques for determining attributable data in a natural language user input that can be used to identify a specific user are described. A system may use various data signals determined using different components. The system may process the various signals to make a final determination on whether the input includes attributable data. The system may use a first component to detect user-identifiable data in the input. The system may use a second component to determine whether the input is potentially attributable to a particular user.

Abstract: Systems and methods are disclosed for utilizing a comment moderation bot for detecting and normalizing neologisms in social media. One method comprises transmitting, by a neologism normalization system, a comment moderation bot for detecting neologisms on an online platform maintained by one or more publisher systems. The comment moderation bot may aggregate data related to user comments and transmit the aggregated data to the neologism normalization system for further processing. The neologism normalization system implements unsupervised machine learning models for detecting neologisms in the aggregated data through tokenization and filtering; and normalizing the neologisms through similarity analysis and lattice decoding.

Abstract: A system may perform speech enhancement of audio data in real-time by suppressing noise components that are present in the audio data while preserving speech components. The system may include an in-ear module and a separate signal processing module that is wirelessly communicatively coupled to the in-ear module. The system may include non-negative matrix factorization (NMF) dictionaries capable of identifying frequency band components associated with speech and frequency band components associated with noise. The NMF dictionaries may be trained using voice samples and noise samples. The NMF dictionaries may be applied to noisy speech data to produce an NMF representation of the speech data which may then be applied using a dynamic mask to the noisy speech data in order to suppress the noise components of the noisy speech data and produce speech enhanced data.

Abstract: Disclosed herein is a system and method for Natural Language Processing (NLP) of real world documents. the system and method combines various models not previously combined and overcomes the challenges of this combination. Models include an encoder-decoder model, a spatial model, and a multi-modal model. An iterative training process receives documents and generates outputs, wherein the iterative training process comprises enabling information retrieval from documents without training data.

Abstract: A method and an apparatus for structuring data are related to information processing technologies in the field of natural language processing. By acquiring an unstructured text and inputting the unstructured text into an encoder-decoder model, an output sequence is obtained. The encoder-decoder model is trained using a training text marked with the attribute value of each attribute. A structured representation is generated based on the attributes corresponding to the attribute elements included in the output sequence and the attribute values comprised in the attribute elements.

Abstract: A dialogue system includes a processor configured to: generate a meaning representation corresponding to an input sentence by performing Natural Language Understanding on the input sentence, generate an output sentence corresponding to the input meaning representation based on Recurrent Neural network (RNN), and determine whether the input sentence cannot be processed using the natural language generator. The processor calculates a parameter representing a probability of outputting the input sentence when the meaning representation corresponding to the input sentence is input to the natural language generator, and determines whether the input sentence cannot be processed based on the calculated parameter.

Abstract: An analysis system connects to a set of data sources and perform natural language questions based on the data sources. The analysis system connects with the data sources and retrieves metadata describing data assets stored in each data source. The analysis system generates an execution plan for the natural language question. The analysis system finds data assets that match the received question based on the metadata. The analysis system ranks the data assets and presents the ranked data assets to users for allowing users to modify the execution plan. The analysis system may use execution plans of previously stored questions for executing new questions. The analysis system supports selective preprocessing of data to increase the data quality.

Abstract: Systems and processes for operating an intelligent automated assistant are provided. For example, a first message is received, wherein a semantic representation of the first message is obtained. In accordance with a determination that the semantic representation of the first message is within a predefined distance of a stored semantic representation, one or more stored reply message representations within the predefined distance of the semantic representation of the first message are identified. The one or more stored reply message representations may correspond to one or more messages previously sent, by a user of the electronic device, in response to a previously received second message. The second message may be semantically similar to the first message. One or more reply messages are provided based on the one or more identified reply message representations, wherein the one or more reply message suggestions are provided as suggested responses to the first message.

Abstract: Systems and methods for guided text generation in task-based dialogue. In some aspects of the technology, an automated assistant system is configured to receive a user request, call multiple APIs, generate dialogue acts based on data received from each API, replace any slot names in the dialogue acts with natural language descriptions of the slots, concatenate the modified dialogue acts, and pass the concatenated result to an NLG model for generation of a natural language response. In some aspects of the technology, the automated assistant may be configured to generate simple templated responses based on the data received from each API, concatenate the simple templated responses, and pass the concatenated sequence to an NLG model trained as a sequence-to-sequence transformer for generation of a final natural language response.

Abstract: Features are disclosed for training and using named entity recognition models based on gazetteer information. A named entity recognition model can be trained with a gazetteer output at a layer of the model to provide deterministic data in the probabilistic model. The named entity recognition model can recognize named entities based on the word embedding and the gazetteer output. The named entity recognition model can tune the gazetteer output to include false positive name entities such that the gazetteer output is not deterministic of the output of the model. In some embodiments, the named entity recognition model can be tuned so as to adjust the gazetteer output.

Abstract: A system and method for federated, context-sensitive, acoustic model refinement comprising a federated language model server and a plurality of edge devices. The federated language model server may comprise one or more machine learning models trained and developed centrally on the server, and distribute these one or more machine learning models to edge devices wherein they may be operated locally on the edge devices. The edge devices may gather or generate context data that can be used by a speech recognition engine, and the local language models contained therein, to develop adaptive, context-sensitive, user-specific language models. Periodically, the federated language model server may select a subset of edge devices from which to receive uploaded local model parameters, that may be aggregated to perform central model updates wherein the updated model parameters may then be sent back to edge devices in order to update the local model parameters.

Abstract: A combination of propagation operations and learning algorithms is applied, using a selected set of labeled conversational logs retrieved from a subset of a plurality of conversational logs, to a remaining corpus of the plurality of conversational logs to train an automated response system according to an intent associated with each of the conversational logs. The combination of propagation operations and learning algorithms may include defining the labels by a user for the selected set of the subset of the plurality of conversational logs; training a probabilistic classifier using the defined labels of features of the selected set, wherein the probabilistic classifier produces labeling decisions for the subset of conversational logs; weighting the features of the selected set in a model optimization process; and/or training an additional classifier using the weighted features of the selected set and applying the additional classifier to the remaining corpus.

Abstract: A mask estimation apparatus for estimating mask information for specifying a mask used to extract a signal of a specific sound source from an input audio signal includes a converter which converts the input audio signal into embedded vectors of a predetermined dimension using a trained neural network model and a mask calculator which calculates the mask information by fitting the embedded vectors to a mixed Gaussian model.

Abstract: An apparatus generates an index including positions of morphemes included in a target text data and semantic attributes between the morphemes corresponding to the positions. The apparatus gives information including positions of morphemes included in an input query and semantic attributes between the morphemes corresponding to the positions to the query, and executes a retrieval on the target text data, based on the information given to the query and the index.