Abstract: Aspects of the technology provide systems and methods for implementing an asymmetric dual encoder architecture. The architecture includes a token embedder layer section having a first token embedding section associated with a first input and a second token embedding section associated with a second input, and an encoder layer section having a first encoder section receiving token embeddings from the first token embedding section and a second encoder section receiving token embeddings from the second token embedding section. A shared projection layer receives encodings from both the first and second encoder sections and generates a set of projections. An embedding space is configured, based on the set of projections, to generate a question embedding and an answer embedding, in which the question and answer embeddings are used in identifying a set of candidate answers to an input answer.

Abstract: Methods, systems, apparatus, including computer programs encoded on computer storage medium, for generating a sentence summary. In one aspect, the method includes actions of tokenizing the sentence into a plurality of tokens, processing data representative of each token in a first order using an LSTM neural network to initialize an internal state of a second LSTM neural network, processing data representative of each token in a second order using the second LSTM neural network, comprising, for each token in the sentence: processing the data representative of the token using the second LSTM neural network in accordance with a current internal state of the second LSTM neural network to (i) generate an LSTM output for the token, and (ii) to update the current internal state of the second LSTM neural network, and generating the summarized version of the sentence using the outputs of the second LSTM neural network for the tokens.

Abstract: News documents from one or more sources are aggregated. The news documents are grouped into a plurality of news collections. Each of the news collections includes a sub-set of the news documents having related content. Objects described by the news collections are determined. The objects collectively form a set of objects. A relevance of each of the news collections is measured with respect to the objects respectively described by the news collections and one or more news collections are determined from the plurality of news collections to be associated with a first object included in the set of objects based on the relevance of the one or more news collections to the first object.

Abstract: Sets of equivalent syntactic patterns are learned from a corpus of documents. A set of one or more input documents is received. The set of one or more input documents is processed for one or more expressions that match a set of equivalent syntactic patterns from among the sets of equivalent syntactic patterns. A syntactic pattern from among the set of equivalent syntactic patterns is selected for a headline. The syntactic pattern reflects a main event described by the set of one or more input documents. The headline is generated using the syntactic pattern.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for extracting instance attributes from text are described. In one aspect, a method exploits weakly-supervised and unsupervised instance relatedness data, available in the form of labeled classes of instances and distributionally similar instances. The method organizes the data into a graph containing instances, class labels, and attributes. The method propagates attributes among related instances, through random walks over the graph.

Abstract: Sets of equivalent syntactic patterns are learned from a corpus of documents. A set of one or more input documents is received. The set of one or more input documents is processed for one or more expressions that match a set of equivalent syntactic patterns from among the sets of equivalent syntactic patterns. A syntactic pattern from among the set of equivalent syntactic patterns is selected for a headline. The syntactic pattern reflects a main event described by the set of one or more input documents. The headline is generated using the syntactic pattern.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for determining query suggestions from multiple correlation measures. In one aspect, a method includes receiving a first query and second queries, each of the first and second queries including one or more terms; for each second query and a linear model, receiving correlation scores measuring the correlation between the first query and the respective second query, each correlation score received from a respective correlation process, and each respective correlation process being different from the other respective correlation processes, and applying the linear model to the plurality of correlation scores to determine a combined correlation score that quantifies a combined correlation between the first query and the respective second query based on the plurality of correlation scores. The second queries are ranked in an order according to their respective combined correlations scores.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for extracting instance attributes from text are described. In one aspect, a method exploits weakly-supervised and unsupervised instance relatedness data, available in the form of labeled classes of instances and distributionally similar instances. The method organizes the data into a graph containing instances, class labels, and attributes. The method propagates attributes among related instances, through random walks over the graph.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for determining query suggestions from multiple correlation measures. In one aspect, a method includes receiving a first query and second queries, each of the first and second queries including one or more terms; for each second query and a linear model, receiving correlation scores measuring the correlation between the first query and the respective second query, each correlation score received from a respective correlation process, and each respective correlation process being different from the other respective correlation processes, and applying the linear model to the plurality of correlation scores to determine a combined correlation score that quantifies a combined correlation between the first query and the respective second query based on the plurality of correlation scores. The second queries are ranked in an order according to their respective combined correlations scores.

Abstract: A word is split into one or more first substrings. A morpheme is applied to the one or more first substrings to create one or more second substrings. The one or more first and second substrings are selected as one or more splittings, and a score is generated for each of the one or more splittings. One of the one or more splittings with a score higher than the other one or more splittings is selected as a keyword.

Abstract: A word is split into one or more first substrings. A morpheme is applied to the one or more first substrings to create one or more second substrings. The one or more first and second substrings are selected as one or more splittings, and a score is generated for each of the one or more splittings. One of the one or more splittings with a score higher than the other one or more splittings is selected as a keyword.

Abstract: A word is split into one or more first substrings. A morpheme is applied to the one or more first substrings to create one or more second substrings. The one or more first and second substrings are selected as one or more splittings, and a score is generated for each of the one or more splittings. One of the one or more splittings with a score higher than the other one or more splittings is selected as a keyword.

Abstract: A word is split into one or more first substrings. A morpheme is applied to the one or more first substrings to create one or more second substrings. The one or more first and second substrings are selected as one or more splittings, and a score is generated for each of the one or more splittings. One of the one or more splittings with a score higher than the other one or more splittings is selected as a keyword.