Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for processing event data. In one aspect, a method includes assigning events to event bundles based on timestamps associated with the events. Each event bundle contains events having timestamps that are within a pre-specified period of time. Event batches are created, where each event batch includes a pre-specified number of event bundles. A first event batch is provided to a first computing group and a second computing group. The first computing group is configured to perform a first processing stage, and the second computing group is configured to perform a second processing stage. A determination is made that a threshold number of the event bundles in the first event batch have been processed by the first computing group. In response to the determination, a second event batch is provided to each of the computing groups.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for identifying non-compositional compounds. In one aspect, a method includes the actions of receiving a collection of phrases, each phrase including two or more words; for each phrase, determining if the phrase is a non-compositional compound, a non-compositional compound being a phrase of two or more words where the words composing the phrase have different meanings in a compound than their conventional meanings individual, the determining including: identifying a similar term for a term of the phrase, substituting the similar term for the term of the phrase to generate a substitute phrase, calculating a similarity between the phrase and the substitute phrase, and identifying the phrase as a non-compositional compound when the calculated similarity is less than a specified threshold value.

Abstract: Aspects directed to phrase generation are provided. A method is provided that includes identifying a plurality of phrase candidates from a plurality of text string entries in a corpus. For each phrase candidate: identifying a plurality of left contexts and a plurality of right contexts for the phrase candidate, each left context of the plurality of left contexts being a nearest unique feature to the right of the phrase candidate in a text string entry and each right context of the plurality of right contexts being the nearest unique feature to the right of the phrase candidate, and calculating a left context vector including a score for each left context feature and a right context vector including a score for each right context feature of the phrase candidate. A similarity is determined between pairs of phrase candidates using the respective left and right context vectors for each phrase candidate.

Abstract: Methods, systems and apparatus, including computer program products, for identifying properties of an electronic document. In one aspect, a sequence of bytes representing text in a document is received. A plurality of byte-n-grams are identified from the bytes. For multiple encodings, a respective likelihood of each byte-n-gram occurring in each of the respective multiple encodings is identified. A respective encoding score for each of the multiple encodings is determined. A most likely encoding of the document is identified based on a highest encoding score among the encoding scores. In another aspect, a sequence of characters, having an encoding, are identified in a document. The sequence is segmented into features, each corresponding to two or more characters. A respective score for each of multiple languages is determined based on the features and a respective language model. A language of the document is identified based on the scores.

Abstract: Aspects directed to phrase generation are provided. A method is provided that includes identifying a plurality of phrase candidates from a plurality of text string entries in a corpus. For each phrase candidate: identifying a plurality of left contexts and a plurality of right contexts for the phrase candidate, each left context of the plurality of left contexts being a nearest unique feature to the right of the phrase candidate in a text string entry and each right context of the plurality of right contexts being the nearest unique feature to the right of the phrase candidate, and calculating a left context vector including a score for each left context feature and a right context vector including a score for each right context feature of the phrase candidate. A similarity is determined between pairs of phrase candidates using the respective left and right context vectors for each phrase candidate.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for identifying non-compositional compounds. In one aspect, a method includes the actions of receiving a collection of phrases, each phrase including two or more words; for each phrase, determining if the phrase is a non-compositional compound, a non-compositional compound being a phrase of two or more words where the words composing the phrase have different meanings in a compound than their conventional meanings individual, the determining including: identifying a similar term for a term of the phrase, substituting the similar term for the term of the phrase to generate a substitute phrase, calculating a similarity between the phrase and the substitute phrase, and identifying the phrase as a non-compositional compound when the calculated similarity is less than a specified threshold value.