Facebook, Inc.

Juan Miguel Pino

Stanislav Funiak

Gaurav Lochan

Mridul Malpani

Matthias Gerhard Eck

Rui Andre Augusto Ferreira

Ying Zhang

Shaomei Wu

Madhu Ramanathan

Irina-Elena Veliche

Debnil Sur

Amit Bahl

Xian Li

Embodiments are disclosed for transliterating text entries across different script systems. A method according to some embodiments includes steps of: receiving an input string in a first script system input using a keyboard; segmenting, using a probabilistic model, the input string into phonemes that correspond to characters or sets of characters in a second script system; converting the phonemes in the first script system into the characters or sets of characters in the second script system, the characters or sets of characters forming a word or a word prefix in the second script system; and outputting the word or the word prefix in the second script system.

Transliteration of text entry across scripts

Embodiments are disclosed for transliteration based on a machine translation model training pipeline. A method according to some embodiments includes steps of: training a probabilistic model for transliteration from a first script system to a second script system using a machine translation model training pipeline; segmenting, using the probabilistic model, an input string in the first script system into phonemes that correspond to characters in the second script system; converting the phonemes in the first script system into the characters in the second script system, the characters forming a word or a word prefix in the second script system; and outputting the word or the word prefix in the second script system.

Transliteration using machine translation pipeline

In one embodiment, a method includes identifying a plurality of dyslexic users on an online social network. The plurality of dyslexic users may be identified based on content objects posted by these users over a particular time period, where the content objects may include one or more of word-level errors or sentence-level errors. A machine-learning model may be trained for text correction using a corpus of social network data, which may include at least the content objects with one or more of word-level errors or sentence-level errors, and a corresponding set of corrected content objects. A text string including one or more errors may be received from a client system associated with a first user. The text string may be transformed into a vector representation using an encoder of the machine-learning model. A corrected text string may be generated from the vector representation using a decoder of the machine-learning model.

Text correction for dyslexic users on an online social network

Technology is disclosed for building correction models that correct natural language snippets. Correction models can include rules comprising pairs of word sequences identified from viable correction snippet pairs, where a first sequence of words in the pair should be replaced with a second sequence of words in the pair. Viable correction snippet pairs can be identified from among pairs of language snippets, such as a post to a social media website and a subsequent update to that post. Viable corrections can be the snippet pairs that both have no more unaligned words than a word alignment threshold and have no aligned word pair with a character edit difference above an edit distance threshold. In some implementations, word alignments can be found by aligning all the characters between a pair of language snippets, and identifying aligned words as those that have at least one aligned letter in common.

Machine-translation based corrections

Embodiments are disclosed for transliteration decoding using a tree structure. A method according to some embodiments includes steps of: generating a tree structure for an input string in a first script system, the tree structure including nodes representing segments of the input string; identifying segmentation candidates for the input string based on paths of the tree structure, the segmentation candidates segmenting the input string into character groups; selecting a segmentation candidate based on probabilities of the segmentation candidates predicted by a probabilistic model; segmenting the input string into character groups that correspond to characters in a second script system; decoding the character groups in the first script system into the characters in the second script system, the characters forming a word or a word prefix in the second script system; and outputting the word or the word prefix in the second script system.

Transliteration decoding using a tree structure

Specialized language processing engines can use author-specific or reader-specific language models to improve language processing results by selecting phrases most likely to be used by an author or by tailoring output to language with which the reader is familiar. Language models that are author-specific can be generated by identifying characteristics of an author or author type such as age, gender, and location. An author-specific language model can be built using, as training data, language items written by users with the identified characteristics. Language models that are reader-specific can be generated using, as training data, language items written by or viewed by that reader. When implementing a specialized machine translation engine, multiple possible translations can be generated. An author-specific language model or a reader-specific language model can provide scores for possible translations, which can be used to select the best translation.

Growing community of inventors

Cambridge, United Kingdom

Juan Miguel Pino