Abstract: A method and system are provided for training word embedding of domain-specific words. The method includes training, by a processor, a first word embedding, using a general domain corpus, on one or more terms inputted by a user. The method further includes retraining, by the processor, the first word embedding, using a specific domain corpus, for a Neuro-Linguistic Programming task, to create a tuned word embedding. The method also includes training, by the processor, a Neural Network for the Neuro-Linguistic Programming task, using the specific domain corpus. The method additionally includes incorporating, by the processor, the trained Neural Network and tuned word embedding into a Neural Network-based Neuro-Linguistic Programming task. The retraining of the first word embedding and the training of the Neural Network are performed together, and the tuned word embedding is accelerated due to a change in a hyper parameter for domain-specific words.

Abstract: A computer-implemented method, computer program product, and system are provided for learning mapping information between different modalities of data. The method includes mapping, by a processor, high-dimensional modalities of data into a low-dimensional manifold to obtain therefor respective low-dimensional embeddings through at least a part of a first network. The method further includes projecting, by the processor, each of the respective low-dimensional embeddings to a common latent space to obtain therefor a respective one of separate latent space distributions in the common latent space through at least a part of a second network. The method also includes optimizing, by the processor, parameters of each of the networks by minimizing a distance between the separate latent space distributions in the common latent space using a variational lower bound. The method additionally includes outputting, by the processor, the parameters as the mapping information.

Abstract: A computer-implemented method, computer program product, and system are provided for deep reinforcement learning to control a subject device. The method includes training, by a processor, a neural network to receive state information of a target of the subject device as an input and provide action information for the target as an output. The method further includes inputting, by the processor, current state information of the target into the neural network to obtain current action information for the target. The method also includes correcting, by the processor, the current action information minimally to obtain corrected action information that meets a set of constraints. The method additionally includes performing an action by the subject device based on the corrected action information for the target to obtain a reward from the target.

Abstract: A computer-implemented method, computer program product, and system are provided for estimating a reward in reinforcement learning. The method includes preparing a state prediction model trained to predict a state for an input using visited states in expert demonstrations performed by an expert. The method further includes inputting an actual state observed by an agent in reinforcement learning into the state prediction model to calculate a predicted state. The method also includes estimating a reward in the reinforcement learning based, at least in part, on similarity between the predicted state and an actual state observed by the agent.

Abstract: A computer-implemented method executed by a robotic system for performing a positional search process in an assembly task is presented. The method includes decomposing, by the robotic system, a perturbation motion into a plurality of actions, the perturbation motion being a motion for an assembly position searched by the robotic system, each action of the plurality of actions related to a specific direction. The method further includes performing reinforcement learning by selecting an action among decomposed actions and assembly movement actions at each step of the positional search process based on corresponding force-torque data received from at least one sensor associated with the robotic system. The method also includes outputting a best action at each step for completion of the assembly task as a result of the reinforcement learning.

Abstract: Symbol sequences are estimated using a computer-implemented method including detecting one or more candidates of a target symbol sequence from a speech-to-text data, extracting a related portion of each candidate from the speech-to-text data, detecting repetition of at least a partial sequence of each candidate within the related portion of the corresponding candidate, labeling the detected repetition with a repetition indication, and estimating whether each candidate is the target symbol sequence, using the corresponding related portion including the repetition indication of each of the candidates.

Abstract: Symbol sequences are estimated using a computer-implemented method including detecting one or more candidates of a target symbol sequence from a speech-to-text data, extracting a related portion of each candidate from the speech-to-text data, detecting repetition of at least a partial sequence of each candidate within the related portion of the corresponding candidate, labeling the detected repetition with a repetition indication, and estimating whether each candidate is the target symbol sequence, using the corresponding related portion including the repetition indication of each of the candidates.

Abstract: Methods and systems are provided for finding a target document in spoken language processing. One of the methods includes calculating a score of each document in a document set, in response to a receipt of first n words of output of an automatic speech recognition (ASR) system, n being equal or greater than zero. The method further includes reading a prior distribution of each document in the document set from a memory device, and updating, for each document in the document set, the score, using the prior distribution, and a weight for interpolation, the weight for interpolation being set based on a confidence score of output of the ASR system. The method additionally includes finding a target document among the document set, based on the updated score of each document.

Abstract: A method, a system, and a computer program product for building an n-gram language model for an automatic speech recognition. The method includes reading training text data and additional text data both for the n-gram language model from a storage, and building the n-gram language model by a smoothing algorithm having discount parameters for n-gram counts. The additional text data includes plural sentences having at least one target keyword. Each discount parameter for each target keyword is tuned using development data which are different from the additional text data so that a predetermined balance between precision and recall is achieved.

Abstract: A computer-implemented method for generating an input for a classifier. The method includes obtaining n-best hypotheses which is an output of an automatic speech recognition (ASR) for an utterance, combining the n-best hypotheses horizontally in a predetermined order with a separator between each pair of hypotheses, and outputting the combined n-best hypotheses as a single text input to a classifier.

Abstract: Symbol sequences are estimated using a computer-implemented method including detecting one or more candidates of a target symbol sequence from a speech-to-text data, extracting a related portion of each candidate from the speech-to-text data, detecting repetition of at least a partial sequence of each candidate within the related portion of the corresponding candidate, labeling the detected repetition with a repetition indication, and estimating whether each candidate is the target symbol sequence, using the corresponding related portion including the repetition indication of each of the candidates.

Abstract: A reading accuracy-improving system includes: a reading conversion unit for retrieving a plurality of candidate word strings from speech recognition results to determine the reading of each candidate word string; a reading score calculating unit for determining the speech recognition score for each of one or more candidate word strings with the same reading to determine a reading score; and a candidate word string selection unit for selecting a candidate to output from the plurality of candidate word strings on the basis of the reading score and speech recognition score corresponding to each candidate word string.

Abstract: Methods, a system, and a classifier are provided. A method includes preparing, by a processor, pairs for an information retrieval task. Each pair includes (i) a training-stage speech recognition result for a respective sequence of training words and (ii) an answer label corresponding to the training-stage speech recognition result. The method further includes obtaining, by the processor, a respective rank for the answer label included in each pair to obtain a set of ranks. The method also includes determining, by the processor, for each pair, an end of question part in the training-stage speech recognition result based on the set of ranks. The method additionally includes building, by the processor, the classifier such that the classifier receives a recognition-stage speech recognition result and returns a corresponding end of question part for the recognition-stage speech recognition result, based on the end of question part determined for the pairs.

Abstract: Methods, a system, and a classifier are provided. A method includes preparing, by a processor, pairs for an information retrieval task. Each pair includes (i) a training-stage speech recognition result for a respective sequence of training words and (ii) an answer label corresponding to the training-stage speech recognition result. The method further includes obtaining, by the processor, a respective rank for the answer label included in each pair to obtain a set of ranks. The method also includes determining, by the processor, for each pair, an end of question part in the training-stage speech recognition result based on the set of ranks. The method additionally includes building, by the processor, the classifier such that the classifier receives a recognition-stage speech recognition result and returns a corresponding end of question part for the recognition-stage speech recognition result, based on the end of question part determined for the pairs.

Abstract: A method for transferring acoustic knowledge of a trained acoustic model (AM) to a neural network (NN) includes reading, into memory, the NN and the AM, the AM being trained with target domain data, and a set of training data including a set of phoneme data, the set of training data being data obtained from a domain different from a target domain for the target domain data, inputting training data from the set of training data into the AM, calculating one or more posterior probabilities of context-dependent states corresponding to phonemes in a phoneme class of a phoneme to which each frame in the training data belongs, and generating a posterior probability vector from the one or more posterior probabilities, as a soft label for the NN, and inputting the training data into the NN and updating the NN, using the soft label.

Abstract: A method and system are provided for training word embedding of domain-specific words. The method includes training, by a processor, a first word embedding, using a general domain corpus, on one or more terms inputted by a user. The method further includes retraining, by the processor, the first word embedding, using a specific domain corpus, for a Neuro-Linguistic Programming task, to create a tuned word embedding. The method also includes training, by the processor, a Neural Network for the Neuro-Linguistic Programming task, using the specific domain corpus. The method additionally includes incorporating, by the processor, the trained Neural Network and tuned word embedding into a Neural Network-based Neuro-Linguistic Programming task. The retraining of the first word embedding and the training of the Neural Network are performed together, and the tuned word embedding is accelerated due to a change in a hyper parameter for domain-specific words.

Abstract: A construction method for a speech recognition model, in which a computer system includes; a step of acquiring alignment between speech of each of a plurality of speakers and a transcript of the speaker; a step of joining transcripts of the respective ones of the plurality of speakers along a time axis, creating a transcript of speech of mixed speakers obtained from synthesized speech of the speakers, and replacing predetermined transcribed portions of the plurality of speakers overlapping on the time axis with a unit which represents a simultaneous speech segment; and a step of constructing at least one of an acoustic model and a language model which make up a speech recognition model, based on the transcript of the speech of the mixed speakers.

Abstract: Methods and systems are provided for finding a target document in spoken language processing. One of the methods includes calculating a score of each document in a document set, in response to a receipt of first n words of output of an automatic speech recognition (ASR) system, n being equal or greater than zero. The method further includes reading a prior distribution of each document in the document set from a memory device, and updating, for each document in the document set, the score, using the prior distribution, and a weight for interpolation, the weight for interpolation being set based on a confidence score of output of the ASR system. The method additionally includes finding a target document among the document set, based on the updated score of each document.

Abstract: Methods and systems for information retrieval include analyzing audio data to produce one or more word hypotheses, each word hypothesis having an associated confidence value. The one or more word hypotheses are displayed in motion across a graphical display. Information retrieval is performed in accordance with one or more selected terms from the displayed word hypotheses.

Abstract: A method, a system, and a computer program product for building an n-gram language model for an automatic speech recognition. The method includes reading training text data and additional text data both for the n-gram language model from a storage, and building the n-gram language model by a smoothing algorithm having discount parameters for n-gram counts. The additional text data includes plural sentences having at least one target keyword. Each discount parameter for each target keyword is tuned using development data which are different from the additional text data so that a predetermined balance between precision and recall is achieved.