Back to results
US 10,762,302 B2Granted
Translation method and apparatus, and translation system
AssigneeSamsung Electronics Co., Ltd.
InventorsJihyun Lee, Hwidong Na, Hoshik Lee
Life of this patent
Estimating the term…
Abstract
A translation method includes: selecting a source word from a source sentence; generating mapping information including location information of the selected source word mapped to the selected source word in the source sentence; and correcting a target word, which is generated by translating the source sentence, based on location information of a feature value of the target word and the mapping information.
Claims43
A processor-implemented machine translation method, the translation method comprising:
- selecting a source word from a source sentence;
- generating mapping information including location information of the selected source word mapped to the selected source word in the source sentence;
- and
- correcting a target word, which is generated by translating the source sentence, based on location information of determined feature values of the target word and the mapping information,
- wherein the translating of the source sentence further includes determining the determined feature values of the target word at a current decoding time based on a previous target word being generated at a previous decoding time, and
- wherein the correcting of the target word includes:
- determining whether corresponding location information of a feature value, of the determined features values of the target word, is included in the mapping information;
- and
- replacing the target word with the selected source word according to the correcting, in response to a determination that the corresponding location information of the feature value, of the determined features values, is included in the mapping information.
6 more elements
A processor-implemented machine translation method, the translation method comprising:
- selecting a source word from a source sentence;
- generating mapping information including location information of the selected source word mapped to the selected source word in the source sentence;
- and
- correcting a target word, which is generated by translating the source sentence, based on location information of determined feature values of the target word and the mapping information,
- wherein the target word is replaced with the selected source word according to the correcting, and
- wherein the correcting of the target word comprises
- selecting a target word corresponding to a preset type from a target sentence generated by completing translation of the source sentence,
- determining whether location information of a maximum feature value of the selected target word is included in the mapping information, and
- in response to the location information of the maximum feature value being included in the mapping information, replacing the selected target word with a source word mapped to the location information of the maximum feature value.
6 more elements
A processor-implemented machine translation method, the translation method comprising:
- selecting a source word from a source sentence;
- generating mapping information including location information of the selected source word mapped to the selected source word in the source sentence;
- and
- correcting a target word, which is generated by translating the source sentence, based on location information of determined feature values of the target word and the mapping information,
- wherein the translating of the source sentence further includes determining the determined feature values of the target word at a current decoding time based on a previous target word being generated at a previous decoding time,
- wherein the target word is replaced with the selected source word according to the correcting,
- wherein the target word is generated by a neural machine translation of the source sentence, and
- wherein the correcting of the target word comprises
- selecting a target word corresponding to a preset type from a target sentence generated by completing translation of the source sentence,
- determining whether location information of a maximum feature value, from among the determined features values, of the selected target word is included in the mapping information, and
- in response to the location information of the maximum feature value being included in the mapping information, replacing the selected target word with a source word mapped to the location information of the maximum feature value.
8 more elements
A translation apparatus comprising:
- a controller;
- and
- a memory storing an instruction executable by the controller,
- wherein the controller is configured to, in response to executing the instruction, select a source word from a source sentence,
- generate mapping information including location information of the selected source word mapped to the selected source word in the source sentence, and
- correct a target word, which is generated by translating the source sentence, based on location information of determined feature values of the target word and the mapping information,
- wherein the translating of the source sentence further includes determining the determined feature values of the target word at a current decoding time based on a previous target word being generated at a previous decoding time, and
- wherein for the correcting of the target word, the controller is configured to determine whether corresponding location information of a feature value, of the determined features values of the target word, is included in the mapping information, and replace the target word with the selected source word, in response to a determination that the corresponding location information of the feature value, of the determined features values, is included in the mapping information.
5 more elements
A translation system comprising:
- a sentence analyzer configured to select a source word from a source sentence, and to generate mapping information including location information of the selected source word mapped to the selected source word in the source sentence;
- a translator configured to determine target words based on a translation of the source sentence, and
- a corrector configured to, each time a target word, among the target words, is determined, determine whether to correct the determined target word based on location information of determined feature values of the determined target word and the mapping information, or configured to, in response to a target sentence being completed by the translator, determine whether to correct a target word, among the target words, selected from the target sentence based on location information of determined feature values of the selected target word and the mapping information,
- wherein the translation of the source sentence further includes determining the corresponding determined feature values of the determined target word or the selected target word respectively generated at a current decoding time based on a previous target word being generated at a previous decoding time, and
- wherein the corrector is configured to determine whether corresponding location information of a feature value, of the determined features values of the target word, is included in the mapping information, and replace the determined target word with the selected source word according to the correcting, in response to a determination that the corresponding location information of the feature value, of the determined features values, is included in the mapping information.
2 more elements
An apparatus with machine translation, comprising:
- one or more processors configured to:
- implement a machine learning translation model to translate a source sentence, including:
- an implementation of the machine learning translation model, for the translating of the source sentence, to generate a target word of a target sentence;
- and
- another implementation of the machine learning translation model, for the translating of the source sentence, to generate another target word for the target sentence
- based on results of the implementation of the machine learning translation model to generate the target word, and
- based on respectively calculated, based on the implementation of the machine learning translation model to generate the target word,
- features of the plural source words of the source sentence,
- wherein, for the implementation of the machine learning translation model to translate the source sentence, the one or more processors are configured to generate a corresponding target word of the target sentence through a selective correction, based on the respectively calculated features, of the generated other target word for the target sentence.
6 more elements
A processor-implemented machine translation method, the translation method comprising:
- generating mapping information including location information of respective one or more select source words in the source sentence;
- determining respective features with respect to each of plural source words of the source sentence;
- translating the source sentence to generate a target sentence;
- selecting a target word, from the target sentence, corresponding to a preset type;
- determining whether location information, corresponding to a determined threshold meeting feature of the determined respective features, is included in the mapping information;
- and
- in response to the location information of the determined threshold meeting feature, of the determined respective features, being determined to be included in the mapping information, correcting the selected target word based on the location information,
- wherein the correcting of the target word includes replacing the target word based on a source word corresponding to the location information.
5 more elements
Dependencies are inferred from the claim text.