公开(公告)号：US10185713B1

公开(公告)日：2019-01-22

申请号：US14867932

申请日：2015-09-28

Applicant: Amazon Technologies, Inc.

Inventor： Michael Denkowski ,  Alon Lavie ,  Gregory Alan Hanneman ,  Austin Matthews ,  Matthew Ryan Fiorillo ,  Robert Thomas Olszewski ,  Christopher James Dyer ,  William Joseph Kaper ,  Alexandre Alexandrovich Klementiev ,  Gavin R. Jewell

IPC: G06F17/28 ,  G06F17/27

Abstract: Technologies are disclosed herein for statistical machine translation. In particular, the disclosed technologies include extensions to conventional machine translation pipelines: the use of multiple domain-specific and non-domain-specific dynamic language translation models and language models; cluster-based language models; and large-scale discriminative training. Incremental update technologies are also disclosed for use in updating a machine translation system in four areas: word alignment; translation modeling; language modeling; and parameter estimation. A mechanism is also disclosed for training and utilizing a runtime machine translation quality classifier for estimating the quality of machine translations without the benefit of reference translations. The runtime machine translation quality classifier is generated in a manner to offset imbalances in the number of training instances in various classes, and to assign a greater penalty to the misclassification of lower-quality translations as higher-quality translations than to misclassification of higher-quality translations as lower-quality translations.

公开(公告)号：US10268684B1

公开(公告)日：2019-04-23

申请号：US14868166

申请日：2015-09-28

Applicant: Amazon Technologies, Inc.

Inventor： Michael Denkowski ,  Alon Lavie ,  Gregory Alan Hanneman ,  Matthew Ryan Fiorillo ,  Laura Josephine Kieras ,  Robert Thomas Olszewski ,  William Joseph Kaper ,  Alexandre Alexandrovich Klementiev ,  Gavin Richard Jewell

IPC: G06F17/28 ,  G06F17/24

Abstract: Technologies are disclosed herein for statistical machine translation. In particular, the disclosed technologies include extensions to conventional machine translation pipelines: the use of multiple domain-specific and non-domain-specific dynamic language translation models and language models; cluster-based language models; and large-scale discriminative training. Incremental update technologies are also disclosed for use in updating a machine translation system in four areas: word alignment; translation modeling; language modeling; and parameter estimation. A mechanism is also disclosed for training and utilizing a runtime machine translation quality classifier for estimating the quality of machine translations without the benefit of reference translations. The runtime machine translation quality classifier is generated in a manner to offset imbalances in the number of training instances in various classes, and to assign a greater penalty to the misclassification of lower-quality translations as higher-quality translations than to misclassification of higher-quality translations as lower-quality translations.

公开(公告)号：US09959271B1

公开(公告)日：2018-05-01

申请号：US14868083

申请日：2015-09-28

Applicant: Amazon Technologies, Inc.

Inventor： Kartik Goyal ,  Alon Lavie ,  Michael Denkowski ,  Gregory Alan Hanneman ,  Matthew Ryan Fiorillo ,  Robert Thomas Olszewski ,  Ehud Hershkovich ,  William Joseph Kaper ,  Alexandre Alexandrovich Klementiev ,  Gavin R. Jewell

IPC: G06F17/30 ,  G06F17/28 ,  G06F17/27 ,  G06F17/24

CPC classification number: G06F17/2818 ,  G06F17/2854

Abstract: Technologies are disclosed herein for statistical machine translation. In particular, the disclosed technologies include extensions to conventional machine translation pipelines: the use of multiple domain-specific and non-domain-specific dynamic language translation models and language models; cluster-based language models; and large-scale discriminative training. Incremental update technologies are also disclosed for use in updating a machine translation system in four areas: word alignment; translation modeling; language modeling; and parameter estimation. A mechanism is also disclosed for training and utilizing a runtime machine translation quality classifier for estimating the quality of machine translations without the benefit of reference translations. The runtime machine translation quality classifier is generated in a manner to offset imbalances in the number of training instances in various classes, and to assign a greater penalty to the misclassification of lower-quality translations as higher-quality translations than to misclassification of higher-quality translations as lower-quality translations.