Abstract: Techniques for predicting a penal code and modifying an annotation based on the prediction are provided. A sentence describing a scene captured with a video capture device may be received form a video description system. The scene may depict a criminal act. An artificial intelligence bot may identify a penal code that has the highest probability of being associated with the criminal act described by the sentence. The sentence may be modified based on a lexicon associated with the identified penal code. Feedback indicating if the identification of the penal code was correct may be received. The feedback may be used to train the artificial intelligence bot.

Abstract: An enterprise data management system with definition quality assessment capabilities for automatically assessing the quality of definitions for terms stored in the enterprise data management system. The system can include a processor programmed to receive a term and a corresponding definition. The processor assess the quality of the definition, including for each of a plurality of quantifiable definition guidelines: deriving feature inputs based on the definition; feeding the feature inputs into a machine learning model corresponding to the definition guideline; and receiving a quality score for the definition guideline from the corresponding machine learning model. An overall quality score is calculated based on the quality score for each of the definition guidelines. The overall quality score and the quality score for each of the plurality of definition guidelines is displayed and if the overall quality score is less than a selected threshold score, a transformation of the definition is recommended.

Abstract: Embodiments of the present disclosure relate to a solution for translation verification and correction. According to the solution, a neural network is trained to determine an association degree among a group of words in a source or target language. The neural network can be used for translation verification and correction. According to the solution, a group of words in a source language and translations of the group of words in a target language are obtained. An association degree among the group of words and an association degree among the translations can be determined by using the trained neural network. Then, whether there is a wrong translation can be determined based on the association degrees. In some embodiments, corresponding methods, systems and computer program products are provided.

Abstract: Certain aspects of the present disclosure provide techniques for generating a replacement sentence with the same or similar meaning but a different sentiment than an input sentence. The method generally includes receiving a request for a replacement sentence and iteratively determining a next word of the replacement sentence word-by-word based on an input sentence. Iteratively determining the next word generally includes evaluating a set of words of the input sentence using a language model configured to output candidate sentences and evaluating the candidate sentences using a sentiment model configured to output sentiment scores for the candidates sentences. Iteratively determining the next word further includes calculating convex combinations for the candidate sentences and selecting an ending word of one of the candidate sentences as the next word of the replacement sentence. The method further includes transmitting the replacement sentence in response to the request for the replacement sentence.

Abstract: A system and method for detecting and resolving mis-transcriptions in a transcript generated by an automatic speech recognition system when transcribing spoken words. The system and method receive a machine language generated transcript of a speech signal by at least one of a first machine learning system and a second machine learning system, and analyze the machine language generated transcript to find a region of low confidence indicative of a mis-transcription and predict an improvement to the region of low confidence indicative of the mis-transcription. The system and method select a replacement word for the mis-transcription based on the predicted improvement to the region of low confidence and replace the mis-transcription by the replacement word to generate a corrected transcript.

Abstract: Generally discussed herein are devices, systems, and methods for generating an automatic interactive digital notebook completion model. A method can include receiving notebook content of an interactive digital notebook, the notebook content including a markdown cell followed by a code cell. The method can include generating input/output examples by, for each input/output example by masking one of (i) content of the markdown cell or (ii) content of the code cell resulting in a masked cell, identifying the masked cell and content of another cell of the markdown cell or the code that is not masked as an input for an input/output example, and identifying the content of the masked cell as an output for the input/output example. The method can include training, based on the input/output examples, a natural language processing model that generates a prediction of the content of a second masked cell as an output.

Abstract: A method, computer system, and a computer program product for contextual digital content highlighting is provided. Discussion is monitored between a plurality of parties in conjunction with a digital presentation and a context of the monitored discussion is then identified. Then a most relevant portion of displayed digital content associated with the presentation is identified based on the identified context and highlighting then is applied to the identified most relevant portion of the displayed content.

Abstract: Described herein are system and method embodiments to improve word representation learning. Embodiments of a probabilistic prior may seamlessly integrate statistical disentanglement with word embedding. Different from previous deterministic methods, word embedding may be taken as a probabilistic generative model, and it enables imposing a prior that may identify independent factors generating word representation vectors. The probabilistic prior not only enhances the representation of word embedding, but also improves the model's robustness and stability. Furthermore, embodiments of the disclosed method may be flexibly plugged in various word embedding models. Extensive experimental results show that embodiments of the presented method may improve word representation on different tasks.

Abstract: This application provides a keyword extraction method. The method includes: receiving an information entity including a title and a text; performing word segmentation on the text, to obtain a plurality of candidate words; and performing character segmentation on the title corresponding to semantics of the text, to obtain a plurality of characters; sequentially inputting the plurality of candidate words to a keyword extraction model, to obtain an attention weight of each candidate word relative to each character; selecting, from the plurality of candidate words, a candidate word that appears in the title; determining an extraction threshold according to an attention weight of the selected candidate word relative to each character; and determining a keyword of the text of the information entity from the candidate words according to the extraction threshold. This application further provide a method for training a keyword extraction model, a computer device, and a storage medium.

Abstract: The present application provides a summary generation model training method, apparatus, electronic device, and non-transitory computer readable storage medium. The summary generation model training method includes: obtaining a first vector set, where vectors in the first vector set are original encoding vectors which have been trained; generating a second vector set based on the first vector set, where the number of vectors in the second vector set is greater than the number of the vectors in the first vector set, and each vector in the second vector set is determined according to one or more vectors in the first vector set; and taking the vectors included in the first vector set and the vectors included in the second vector set as input encoding vectors to perform model training to obtain a summary generation model.

Abstract: A target set of texts, for training and/or evaluating a text classification model, is augmented using insertions into a base text within the original target set. In an embodiment, an expanded text, including the base text and an insertion word, must satisfy one or more inclusion criteria in order to be added to the target set. The inclusion criteria may require that the expanded text constitutes a successful attack on the classification model, the expanded text has a satisfactory perplexity score, and/or the expanded text is verified as being valid. In an embodiment, if a number of expanded texts added into the target set is below a threshold number, insertions are made into an expanded text (which was generated based on the base text). Inclusion criteria are evaluated against the doubly-expanded text to determine whether to add the doubly-expanded text to the target set.

Abstract: A method, apparatus and system for training an embedding space for content comprehension and response includes, for each layer of a hierarchical taxonomy having at least two layers including respective words resulting in layers of varying complexity, determining a set of words associated with a layer of the hierarchical taxonomy, determining a question answer pair based on a question generated using at least one word of the set of words and at least one content domain, determining a vector representation for the generated question and for content related to the at least one content domain of the question answer pair, and embedding the question vector representation and the content vector representations into a common embedding space where vector representations that are related, are closer in the embedding space than unrelated embedded vector representations. Requests for content can then be fulfilled using the trained, common embedding space.

Abstract: A speech emotion detection system may obtain to-be-detected speech data. The system may generate speech frames based on framing processing and the to-be-detected speech data. The system may extract speech features corresponding to the speech frames to form a speech feature matrix corresponding to the to-be-detected speech data. The system may input the speech feature matrix to an emotion state probability detection model. The system may generate, based on the speech feature matrix and the emotion state probability detection model, an emotion state probability matrix corresponding to the to-be-detected speech data. The system may input the emotion state probability matrix and the speech feature matrix to an emotion state transition model. The system may generate an emotion state sequence based on the emotional state probability matrix, the speech feature matrix, and the emotional state transition model. The system may determine an emotion state based on the emotion state sequence.

Abstract: A computer implemented method is described for conducting text sequence machine learning, the method comprising: receiving an input sequence x=[x1, x2, . . . , xn], to produce a feature vector for a series of hidden states hx=[h1, h2, . . . , hn], wherein the feature vector for the series of hidden states hx is generated by performing pooling over a temporal dimension of all hidden states output by the encoder machine learning data architecture; and extracting from the series of hidden states hx, a mean and a variance parameter, and to encapsulate the mean and the variance parameter as an approximate posterior data structure.

Abstract: A system includes: a natural language processing (NLP) model trained in a training domain and configured to perform natural language processing on an input dataset; an accuracy module configured to: calculate a domain shift metric based on the input dataset; and calculate a predicted decrease in accuracy of the NLP model attributable to domain shift relative to the training domain based on the domain shift metric; and a retraining module configured to selectively trigger a retraining of the NLP model based on the predicted decrease in accuracy of the NLP model.

Abstract: A method of performing speech recognition using a two-pass deliberation architecture includes receiving a first-pass hypothesis and an encoded acoustic frame and encoding the first-pass hypothesis at a hypothesis encoder. The first-pass hypothesis is generated by a recurrent neural network (RNN) decoder model for the encoded acoustic frame. The method also includes generating, using a first attention mechanism attending to the encoded acoustic frame, a first context vector, and generating, using a second attention mechanism attending to the encoded first-pass hypothesis, a second context vector. The method also includes decoding the first context vector and the second context vector at a context vector decoder to form a second-pass hypothesis.

Abstract: A method for operating a neural network includes receiving an input sequence at an encoder. The input sequence is encoded to produce a set of hidden representations. Attention-heads of the neural network calculate attention weights based on the hidden representations. A context vector is calculated for each attention-head based on the attention weights and the hidden representations. Each of the context vectors correspond to a portion of the input sequence. An inference is output based on the context vectors.

Abstract: Techniques for presenting an alternative input representation to a user for testing and collecting processing data are described. A system may determine that a received spoken input triggers an alternative input representation for presenting. The system may output data corresponding to the alternative input representation in response to the received spoken input, and the system may receive user feedback from the user. The system may store the user feedback and processing data corresponding to processing of the alternative input representation, which may be later used to update an alternative input component configured to determine alternative input representations for spoken inputs.

Abstract: Examples of the present disclosure describe systems and methods for dynamically expanding acronyms in audio content. In aspects, a user access of an audio resource may be detected. The audio content of the audio resource may be evaluated to identify acronyms. One or more of the identified acronyms may be evaluated based on a user-specific context of the user and/or a global context associated with the user. Based on the evaluated context(s), expansion candidates and corresponding confidence scores may be determined for each identified acronym. Based on the confidence scores, an expansion candidate may be selected and used to replace the identified acronym when the audio content is consumed by the user.

Abstract: A method and system for predicting an intended time interval for a content segment may include receiving a request for natural language processing (NLP) of the content segment, the content segment including one or more temporal expressions, accessing contextual data associated with each of the one or more temporal expressions, decoding the content segment into a program that describes a temporal logic of the content segment based on the one or more temporal expressions, evaluating the program using the contextual data to predict an intended time interval for the content segment, and providing the intended time interval as an output.