Abstract: A computer-implemented method is provided. The method includes acquiring a seed word; calculating a similarity score of each of a plurality of words relative to the seed word for each of a plurality of models to calculate a weighted sum of similarity scores for each of the plurality of words; outputting a plurality of candidate words among the plurality of words; acquiring annotations indicating at least one of preferred words and non-preferred words among the plurality of the candidate words; updating weights of the plurality of models in a manner to cause weighted sums of similarity scores for the preferred words to be relatively larger than the weighted sums of the similarity scores for the non-preferred words, based on the annotations; and grouping the plurality of candidate words output based on the weighted sum of similarity scores calculated with updated weights of the plurality of models.

Abstract: Text data including at least named entities can be received. From the named entities, continuous entities, overlapping entities and disjoint entities can be identified. The overlapping entities can be transformed into continuous entities. The continuous entities, the transformed entities and the disjoint entities can be encoded. The encoded entities can be input to a machine learning language model to train the machine learning model to predict candidate entities. The predicted entities can be decoded to reconstruct the predicted entities.

Abstract: In order to solve a problem that the level of precision in inferring a crosslingual topic of documents or words in a non-parallel corpus associated in the document level is not high, a topic inferring apparatus includes: a word distribution information storage unit in which word distribution information is stored in association with each of two or more languages; a document topic distribution generating unit that acquires document topic distribution information of a multilingual document set; a segment topic distribution generating unit that acquires segment topic distribution information of each segment, using the document topic distribution information; and a word topic determining unit that determines, for each word contained in two or more documents contained in the multilingual document set, a topic of each word using the segment topic distribution information. Accordingly, it is possible to improve the level of precision in inferring a topic.

Abstract: A method includes maintaining a question repository in which each question corresponds to a set of decision trees. A distance matrix encodes a distance between each pair of questions. In response to a request for a new question, the method converts the new question into a set of tokens. For each question of the existing questions, the method determines a minimum distance between each token of the new question and the tokens of the question and sums the minimum distances to calculate a distance between the question and the new question. The method includes performing cluster analysis on the distance matrix. Performing cluster analysis includes normalizing the distance matrix and applying a hierarchical clustering process to the normalized distance matrix. Based on the cluster analysis, the method transmits an alternative question proposal or adds the new question to the question repository.

Abstract: Provided is a method, system, and computer program product for context-dependent spellchecking. The method comprises receiving context data to be used in spell checking. The method further comprises receiving a user input. The method further comprises identifying an out-of-vocabulary (OOV) word in the user input. An initial suggestion pool of candidate words is identified based, at least in part, on the context data. The method then comprises using a noisy channel approach to evaluate a probability that one or more of the candidate words of the initial suggestion pool is an intended word and should be used as a candidate for replacement of the OOV word. The method further comprises selecting one or more candidate words for replacement of the OOV word. The method further comprises outputting the one or more candidates.

Abstract: Techniques and solutions are described for translating strings in a human language that are associated with code, such as code that declares user interface elements in association with which the strings can be displayed. The strings can be embedded in code, or can be included in files that are accessed by the code or used in creating executable code. Strings can be extracted from data, such as files, received in a translation request. After translation, such as by a human translator or using automatic translation, the translation results can be provided in response to the translation request. The translation process can be automated, such that changes to files in a code repository can trigger a translation request. Disclosed technologies can facilitate the production of programs that are localized using different human languages.

Abstract: Techniques are disclosed for training and/or utilizing an automatic post-editing model in correcting translation error(s) introduced by a neural machine translation model. The automatic post-editing model can be trained using automatically generated training instances. A training instance is automatically generated by processing text in a first language using a neural machine translation model to generate text in a second language. The text in the second language is processed using a neural machine translation model to generate training text in the first language. A training instance can include the text in the first language as well as the training text in the first language.

Abstract: Described herein is a system for adapting an output to a user input over a period of time based on how often the user interacts with the system. The system may determine a user's level of familiarity of the system, and may determine to personalize the output to a user request based on his level of familiarity. The user's level of familiarity may be determined by analyzing historical interactions between the user and the system. The level of personalization applied to the output may be determined based on the user's level of familiarity. As user becomes more familiar with the system, the output may be more personalized.

Abstract: A first neural network model of a user device processes audio data to extract audio embeddings that represent vocal characteristics of a user of an utterance represented in the audio data. The audio embeddings may then be hashed to remove characteristics specific to the user while still maintaining a unique set of characteristics. The hashed embeddings may be sent to a remote system, which may use them to identify the user.

Abstract: Embodiments are disclosed for a method for dynamic text reading. The method includes performing pre-processing for a text document. Pre-processing includes determining the text document comprises an emotional statement based on an indicator of an emotion associated with the emotional statement. Pre-processing also includes identifying a speaker of the emotional statement. Further, pre-processing includes generating a role-to-voice map that associates the speaker with a digital representation of a voice for the speaker. The method additionally includes generating, based on the pre-processing, the voice for the speaker reading aloud a text of the text document using the digital representation of the voice with a tonal modulation that conveys the emotion.

Abstract: One embodiment provides a method, including: receiving, at an information handling device, user input comprising a potential wake word; determining, using a processor, whether the potential wake word is associated with a stored wake word; and responsive to determining that the potential wake word is associated with the stored wake word, activating, based on the potential wake word, a digital assistant associated with the information handling device. Other aspects are described and claimed.

Abstract: Various embodiments include methods and devices for implementing automatic grammar augmentation for improving voice command recognition accuracy in systems with a small footprint acoustic model. Alternative expressions that may capture acoustic model decoding variations may be added to a grammar set. An acoustic model-specific statistical pronunciation dictionary may be derived by running the acoustic model through a large general speech dataset and constructing a command-specific candidate set containing potential grammar expressions. Greedy based and cross-entropy-method (CEM) based algorithms may be utilized to search the candidate set for augmentations with improved recognition accuracy.

Abstract: A method for audio processing includes receiving, in a computer, a recording of a teleconference among multiple participants over a network including an audio stream containing speech uttered by the participants and conference metadata for controlling a display on video screens viewed by the participants during the teleconference. The audio stream is processed by the computer to identify speech segments, in which one or more of the participants were speaking, interspersed with intervals of silence in the audio stream. The conference metadata are parsed so as to extract speaker identifications, which are indicative of the participants who spoke during successive periods of the teleconference. The teleconference is diarized by labeling the identified speech segments from the audio stream with the speaker identifications extracted from corresponding periods of the teleconference.

Abstract: Disclosed embodiments relate to systems and methods for automatically processing diversely structured operational policies. Techniques include identifying a policy associated with an application or computer code; applying a language processing protocol to the policy to interpret the policy and extract attribute(s) of the policy, where the policy is defined using a vocabulary and syntax; normalizing the policy to define the policy using a standardized vocabulary and syntax agnostic to an infrastructure or service associated with the application or computer code, where one or more of the vocabulary and syntax are respectively different from the standardized vocabulary and syntax and where normalizing the policy comprises translating the attribute(s) of the policy; and evaluating the policy based on the normalizing to determine whether a potentially malicious activity is associated with the application or computer code.

Abstract: A method includes identifying a set of identigens for each word of a first phrase of a phrase group to produce a first plurality of sets of identigens and determining whether first and second identigen rules are applicable to the first plurality of sets of identigens. When the first and second identigen rules are applicable to the first plurality of sets of identigens, the method further includes identifying a set of identigens for each word of a second phrase of the phrase group to produce a second plurality of sets of identigens and determining that the first identigen rules are applicable to the second plurality of sets of identigens. The method further includes identifying one valid identigen of each set of identigens of the first plurality of sets of identigens by applying the first identigen rules to the first plurality of sets of identigens to produce a first entigen group.

Abstract: Techniques and a framework are described herein for constructing and/or updating, e.g., on top of a general-purpose knowledge graph, an “event-specific provisional knowledge graph.” In various implementations, live data stream(s) may be analyzed to identify entity(s) associated with a developing event. The entity(s) may form part of a general-purpose knowledge graph that includes entity nodes and edges between the entity nodes. Based on the identified one or more entities, an event-specific provisional knowledge graph may be constructed or updated in association with the developing event. In some implementations, the event-specific provisional knowledge graph may be queried for new information about the developing event. Computing devices may be caused to render, as output, the new information.

Abstract: This speech processing device is provided with: a contribution degree estimation means which calculates a contribution degree representing a quality of a segment of the speech signal; and a speaker feature calculation means which calculates a feature from the speech signal, for recognizing attribute information of the speech signal, using the contribution degree as a weight of the segment of the speech signal.

Abstract: A speech synthesis system includes an operating interface, a storage unit and a processor. The operating interface provides a plurality of language options for a user to select one output language option therefrom. The storage unit stores a plurality of acoustic models. Each acoustic model corresponds to one of the language options and includes a plurality of phoneme labels corresponding to a specific vocal. The processor receives a text file and generates output speech data corresponding to the specific vocal according to the text file, a speech synthesizer, and one of the acoustic models which corresponds to the output language option.

Abstract: Streaming machine learning unidirectional models is facilitated by the use of embedding vectors. Processing blocks in the models apply embedding vectors as input. The embedding vectors utilize context of future data (e.g., data that is temporally offset into the future within a data stream) to improve the accuracy of the outputs generated by the processing blocks. The embedding vectors cause a temporal shift between the outputs of the processing blocks and the inputs to which the outputs correspond. This temporal shift enables the processing blocks to apply the embedding vector inputs from processing blocks that are associated with future data.

Abstract: A system, method and non-transitory computer readable medium for providing call quality driven communication management wherein an audio data stream of a communication session having one or more utterances is processed to generate a transcript of the communication session. The generated transcript is analyzed to determine whether a quality of the audio data stream, and one or more quality improvement measures when one or more audio artifacts are determined to be present in the audio data stream.