Methods, systems, and apparatus, including computer programs encoded on computer storage media, for information extraction using natural language processing. One of the methods includes determining, for one or more tokens from a plurality of tokens that represent an unstructured sentence, a token type from a plurality of predetermined token types that indicates an element type for a phrase that corresponds to the token and has one or more properties using dependency data and a part-of-speech label for the token; assigning, for a token whose associated dependency data indicates that the token has a child, data for the child token to one of the one or more properties for the token type of the token; and providing, for use by a downstream semantic system and for the token, a textual representation of the phrase for the token and the phrases for one or more of the child tokens.

A machine reading comprehension (MRC) question and answer providing method includes receiving a user question; analyzing the user question; selecting at least one document from at least one domain corresponding to an analyzed user question and searching for a passage, which is a candidate answer determined as being suitable for the user question, in the selected at least one document; obtaining at least one correct answer candidate value by inputting the user question and a corresponding passage into each of at least one MRC question and answer unit; and determining whether the at least one correct answer candidate value is a best answer.

A machine reading comprehension (MRC) question and answer providing method includes receiving a user question; analyzing the user question; selecting at least one document from at least one domain corresponding to an analyzed user question and searching for a passage, which is a candidate answer determined as being suitable for the user question, in the selected at least one document; obtaining at least one correct answer candidate value by inputting the user question and a corresponding passage into each of at least one MRC question and answer unit; and determining whether the at least one correct answer candidate value is a best answer.

A device with a microphone acquires audio data of a user's speech. A neural network accepts audio data as input and provides sentiment data as output. The neural network is trained using training data based on input from raters who provide votes as to which sentiment descriptors they think are associated with a sample of speech. A vote by a rater assessing the sample for a particular semantic descriptor is distributed to a plurality of semantically similar semantic descriptors. Semantic descriptor similarity data indicates relative similarity between possible semantic descriptors in the semantic space. The distributed partial votes may be aggregated to produce training data comprising samples of speech and weights of corresponding semantic descriptors. The training data is then used to train the neural network. For example, the neural network may be trained with the training data using per-instance cosine similarity loss or correlational loss.

Systems and methods are provided to pre-train projection networks for use as transferable natural language representation generators. In particular, example pre-training schemes described herein enable learning of transferable deep neural projection representations over randomized locality sensitive hashing (LSH) projections, thereby surmounting the need to store any embedding matrices because the projections can be dynamically computed at inference time.

A crime type inference system based on text data, may include: a keywords dictionary construction unit configured to receive crime source data, and generate a crime type keywords dictionary by extracting crime keywords; a data set construction unit configured to generate a dataset for crime type learning by using the crime source data and the keywords dictionary; a crime type prediction model training unit configured to generate a crime type prediction model by using the dataset, and train the crime type prediction model; and a crime type inference unit configured to infer a crime type by using new crime data.

A crime type inference system based on text data, may include: a keywords dictionary construction unit configured to receive crime source data, and generate a crime type keywords dictionary by extracting crime keywords; a data set construction unit configured to generate a dataset for crime type learning by using the crime source data and the keywords dictionary; a crime type prediction model training unit configured to generate a crime type prediction model by using the dataset, and train the crime type prediction model; and a crime type inference unit configured to infer a crime type by using new crime data.

An electronic device and method are disclosed. The electronic device includes input circuitry, a display, and a processor. The processor implements the method, including extracting at least one piece of context information based at least in part on an application screen displayed on the display, analyzing the extracted at least one piece of context information to generate a language model based on the extracted at least one piece of context information, receiving a voice input of a user through the input circuitry and convert the voice input into a text string using the generated language model, and resetting the generated language model.

An electronic device and method are disclosed. The electronic device includes input circuitry, a display, and a processor. The processor implements the method, including extracting at least one piece of context information based at least in part on an application screen displayed on the display, analyzing the extracted at least one piece of context information to generate a language model based on the extracted at least one piece of context information, receiving a voice input of a user through the input circuitry and convert the voice input into a text string using the generated language model, and resetting the generated language model.

A method performed by a computing device includes determining a set of identigens for each word of words of a subjective query to produce sets of identigens. The method further includes interpreting, using identigen pairing rules of a knowledge database, the sets of identigens to determine a most likely meaning interpretation of the subjective query and produce a query entigen group that includes query entigens. The method further includes identifying one or more characteristic entigen categories for a subjective category entigen of the query entigen group. The method further includes recovering a set of response entigens for the subjective query from the knowledge database utilizing the query entigen group and based on the one or more characteristic entigen categories. The set of response entigens provides an answer for the subjective query.