International Business Machines Corporation

Nuance Communications, Inc.

Pryon Incorporated

Steven John Rennie

Vaibhava Goel

David Nahamoo

Etienne Marcheret

Ellen Eide Kislal

Igor Roditis Jablokov

Marie Wenzel Meteer

Soonthorn Ativanichayaphong

Peder Andreas Olsen

Chul Sung

John Pruitt

Jing Chang Huang

Osamu Ichikawa

Samuel Jonathan Thomas

Ewout Van Den Berg

Pierre L Dognin

Petr Fousek

Bryan Dempsey

Carmi Rothberg

Ajinkya Jitendra Zadbuke

Joseph Allen Pruitt

Takashi Fukuda

Gregory Pelton

Ramesh Ambat Gopinath

John Randall Hershey

Lucas Siler

Neil Rohit Mallinar

Chui Sung

Trausti Thor Kristjansson

Dominique O'Donnell

Dominique O'donnell

A question-answering system that receive a natural-language question includes a database to provide a basis for that answer and a structured-query generator that constructs a structured query from the question and uses it to obtain an answer to the question from the database. The method further comprises generating a first answer to said first question, wherein generating said first answer comprises causing said question to traverse a first path through said question-answering system, wherein the first path comprises a knowledge base constructed from content of the document collection using a machine learning BERT model; and causing said question to traverse a second path through said question-answering system to generate a second answer to said question, said second answer differing from said first answer, wherein the second path comprises a structured query generator and the database, the database being distinct from the knowledge base.

Question-answering system for answering relational questions by utilizing two paths where at least one path uses BERT model

Disclosed are methods, systems, devices, apparatus, media, and other implementations that include a method for document processing (particularly for training of a machine learning question answering platform, and for ingestion of documents). The method includes obtaining a question dataset (e.g., either from public or private repositories of questions) comprising one or more source questions for document processing by a machine learning question-and-answer system that provides answer data in response to question data submitted by a user, modifying a source question from the question dataset to generate one or more augmented questions with equivalent semantic meanings as that of the source question, and processing a document with the one or more augmented questions.

Methods and systems for improved document processing and information retrieval

Disclosed are methods, systems, devices, apparatus, media, and other implementations, including a method that includes obtaining a query set (e.g., a universal set of questions), performing a question-and-answer (Q-A) search on one or more documents using the query set to produce answer data responsive to one or more questions included in the query set, with the answer data characterizing concepts associated with the one or more documents, and deriving structured output information for the one or more documents based on the answer data produced in response to performing the Q-A search.

Supervised summarization and structuring of unstructured documents

A question-answering system that receive a natural-language question includes a database to provide a basis for that answer and a structured-query generator that constructs a structured query from the question and uses it to obtain an answer to the question from the database.

Visual structure of documents in question answering

Disclosed are methods, systems, devices, apparatus, media, design structures, and other implementations, including a method that includes receiving a source document, applying one or more pre-processes to the source document to produce contextual information representative of the structure and content of the source document, and transforming the source document, based on the contextual information, to generate a question-and-answer searchable document.

Document pre-processing for question-and-answer searching

Disclosed are methods, systems, devices, apparatus, media, design structures, and other implementations, including a method that includes receiving, at a local device from a remote device, query data representative of a question relating to source content of a source document, and determining whether one or more pre-determined questions stored in a question-answer cache maintained at the local device matches the query data according to one or more matching criteria. The method further includes obtaining from the question-answer cache, in response to a determination that at least one of the pre-determined questions matches the query data received from the remote device, at least one answer data item, associated with at least one pre-determined question, corresponding to an answer to the question relating to the source content.

Systems and methods for question-and-answer searching using a cache

Systems and methods for training a neural network to optimize network performance, including sampling an applied dropout rate for one or more nodes of the network to evaluate a current generalization performance of one or more training models. An optimized annealing schedule may be generated based on the sampling, wherein the optimized annealing schedule includes an altered dropout rate configured to improve a generalization performance of the network. A number of nodes of the network may be adjusted in accordance with a dropout rate specified in the optimized annealing schedule. The steps may then be iterated until the generalization performance of the network is maximized.

Annealed dropout training of neural networks

A method and system for generating training data for a target domain using speech data of a source domain. The training data generation method including: reading out a Gaussian mixture model (GMM) of a target domain trained with a clean speech data set of the target domain; mapping, by referring to the GMM of the target domain, a set of source domain speech data received as an input to the set of target domain speech data on a basis of a channel characteristic of the target domain speech data; and adding a noise of the target domain to the mapped set of source domain speech data to output a set of pseudo target domain speech data.

Method, apparatus, and program for generating training speech data for target domain

According to some aspects, a method of classifying speech recognition results is provided, using a neural network comprising a plurality of interconnected network units, each network unit having one or more weight values, the method comprising using at least one computer, performing acts of providing a first vector as input to a first network layer comprising one or more network units of the neural network, transforming, by a first network unit of the one or more network units, the input vector to produce a plurality of values, the transformation being based at least in part on a plurality of weight values of the first network unit, sorting the plurality of values to produce a sorted plurality of values, and providing the sorted plurality of values as input to a second network layer of the neural network.

Order statistic techniques for neural networks

A speech processing method and arrangement are described. A dynamic noise adaptation (DNA) model characterizes a speech input reflecting effects of background noise. A null noise DNA model characterizes the speech input based on reflecting a null noise mismatch condition. A DNA interaction model performs Bayesian model selection and re-weighting of the DNA model and the null noise DNA model to realize a modified DNA model characterizing the speech input for automatic speech recognition and compensating for noise to a varying degree depending on relative probabilities of the DNA model and the null noise DNA model.

System and method for dynamic noise adaptation for robust automatic speech recognition

A method and apparatus are provided for training a transformation matrix of a feature vector for an acoustic model. The method includes training the transformation matrix of the feature vector. The transformation matrix maximizes an objective function having a regularization term. The method further includes transforming the feature vector using the transformation matrix of the feature vector, and updating the acoustic model stored in a memory device using the transformed feature vector.

Regularized feature space discrimination adaptation

A probability model represented as the product of the probability distribution of a mismatch vector g (or clean speech x) with an observed value y as a factor and the probability distribution of a mismatch vector g (or clean speech x) with a confidence index &#x3b2; for each band as a factor, executes MMSE estimation on the probability model, and estimates a clean speech estimated value x^. As a result, each band influences the result of MMSE estimation, with a degree of contribution in accordance with the level of its confidence. Further, the higher the S/N ratio of observation speech, the more the output value becomes shifted to the observed value. As a result, the output of a front-end is optimized.

Noise reduction method, program product, and apparatus

Techniques disclosed herein include using a Maximum A Posteriori (MAP) adaptation process that imposes sparseness constraints to generate acoustic parameter adaptation data for specific users based on a relatively small set of training data. The resulting acoustic parameter adaptation data identifies changes for a relatively small fraction of acoustic parameters from a baseline acoustic speech model instead of changes to all acoustic parameters. This results in user-specific acoustic parameter adaptation data that is several orders of magnitude smaller than storage amounts otherwise required for a complete acoustic model. This provides customized acoustic speech models that increase recognition accuracy at a fraction of expected data storage requirements.

Growing community of inventors

Yorktown Heights, NY, United States of America

Steven John Rennie