Method for generating paraphrases for use in machine translation system

• US 10,318,642 B2
• Filed: 12/12/2016
• Issued: 06/11/2019
• Est. Priority Date: 02/01/2016
• Status: Active Grant

• Current Assignee: Panasonic Intellectual Property Management Co., Ltd. (Panasonic Holdings Corporation)
• Original Assignee: Panasonic Intellectual Property Management Co., Ltd. (Panasonic Holdings Corporation)
• Inventors: Fujiwara, Nanami, Yamauchi, Masaki
• Primary Examiner(s): He, Jialong
• Application Number:
  US15/375,267
• Publication Number:
  US 20170220559A1
• Time in Patent Office: 911 Days
• CPC Class Codes: G06F 40/45 Example-based machine translation; Alignment
• CPC Class Codes: G06F 40/51 Translation evaluation
• CPC Class Codes: G06F 40/58 Use of machine translation, e.g. for multi-lingual retrieval, for server-side translation for client devices or for real-time translation

Abstract

A paraphrase generation method according to the present disclosure generates one or more paraphrases of an original text by paraphrasing, within an acceptable limit for accepting paraphrasing, one or more of a plurality of fragments included in the original text into another expression in the language of the original text, the plurality of fragments being obtained by dividing the original text in accordance with a predetermined rule.

Claims

1. A paraphrase generation method, comprising:

receiving an original text;

generating, using a processor, one or more paraphrase candidate sentences by paraphrasing one or more of a plurality of fragments included in the original text into another expression in a language of the original text, the plurality of fragments being obtained by dividing the original text in accordance with a predetermined rule;

determining whether the paraphrasing of one or more of the plurality of fragments in each of the one or more paraphrase candidate sentences is acceptable by comparing each of a cumulative total of paraphrase acceptability scores with a first acceptable limit, each of the cumulative total of paraphrase acceptability scores being calculated by adding each of paraphrase acceptability scores in each of the one or more paraphrase candidate sentences, each of the paraphrase acceptability scores being assigned to a paraphrase pair including a first fragment and a second fragment that represents another expression of the first fragment, each of the paraphrase acceptability scores indicating a degree to which paraphrasing from the first fragment into the second fragment is accepted;

determining whether each of the one or more paraphrase candidate sentences is acceptable by comparing each of linguistic acceptability scores with a second acceptable limit, each of the linguistic acceptability scores indicating a degree to which the paraphrase candidate sentence is accepted as having a linguistically correct meaning, and each of the linguistic acceptability scores being obtained from a language model; and

outputting each of the one or more paraphrase candidate sentences when (i) each of the cumulative total of paraphrase acceptability scores is within the first acceptable limit, and (ii) each of the linguistic acceptability scores is within the second acceptable limit.

2. The paraphrase generation method according to claim 1,

wherein the paraphrasing in generating each of the one or more paraphrase candidate sentences is executed until (i) the determining determines that the paraphrasing is not within the first acceptable limit, or (ii) the determining determines that the paraphrase candidate sentence is not within the second acceptable limit.

3. The paraphrase generation method according to claim 1, further comprising:

creating a translation corpus that includes a collection of a plurality of text pairs, the plurality of text pairs each including a first text in a first language paired with a second text in a second language,

wherein the receiving includes further receiving the second text, with the original text being the first text, and

wherein the creating of the translation corpus includes creating one or more new text pairs, the one or more new text pairs each including the received second text and the one or more paraphrase candidate sentences, the created one or more new text pairs forming a new part of the translation corpus.

7. The paraphrase generation method according to claim 1, further comprising:

presenting the one or more paraphrase candidates to user;

receiving a value of evaluation of the user; and

changing one of a value of the first acceptable limit and a value of the second acceptable limit using the value of evaluation.

4. An apparatus, comprising:

a processor; and

a medium having a computer program stored thereon, the computer program causing the processor to execute operations including

receiving an original text;

generating, using the processor, one or more paraphrase candidate sentences by paraphrasing one or more of a plurality of fragments included in the original text into another expression in a language of the original text, the plurality of fragments being obtained by dividing the original text in accordance with a predetermined rule;

determining whether the paraphrasing of one or more of the plurality of fragments in each of the one or more paraphrase candidate sentences is acceptable by comparing each of a cumulative total of paraphrase acceptability scores with a first acceptable limit, each of the cumulative total of paraphrase acceptability scores being calculated by adding each of paraphrase acceptability scores in each of the one or more paraphrase candidate sentences, each of the paraphrase acceptability scores being assigned to a paraphrase pair including a first fragment and a second fragment that represents another expression of the first fragment, each of the paraphrase acceptability scores indicating a degree to which paraphrasing from the first fragment into the second fragment is accepted;

determining whether each of the one or more paraphrase candidate sentences is acceptable by comparing each of linguistic acceptability scores with a second acceptable limit, each of the linguistic acceptability scores indicating a degree to which the paraphrase candidate sentence is accepted as having a linguistically correct meaning, and each of the linguistic acceptability scores being obtained from a language model; and

outputting each of the one or more paraphrase candidate sentences when (i) each of the cumulative total of paraphrase acceptability scores is within the first acceptable limit, and (ii) each of the linguistic acceptability scores is within the second acceptable limit.

5. A non-transitory recording medium having a computer program stored thereon, the computer program causing a processor to execute operations comprising:

receiving an original text;

generating, using the processor, one or more paraphrase candidate sentences by paraphrasing one or more of a plurality of fragments included in the original text into another expression in a language of the original text, the plurality of fragments being obtained by dividing the original text in accordance with a predetermined rule;

determining whether the paraphrasing of one or more of the plurality of fragments in each of the one or more paraphrase candidate sentences is acceptable by comparing each of a cumulative total of paraphrase acceptability scores with a first acceptable limit, each of the cumulative total paraphrase acceptability scores being calculated by adding each of paraphrase acceptability scores in each of the one or more paraphrase candidate sentences, each of the paraphrase acceptability scores being assigned to a paraphrase pair including a first fragment and a second fragment that represents another expression of the first fragment, each of the paraphrase acceptability scores indicating a degree to which paraphrasing from the first fragment into the second fragment is accepted;

determining whether each of the one or more paraphrase candidate sentences is acceptable by comparing each of linguistic acceptability scores with a second acceptable limit, each of the linguistic acceptability scores indicating a degree to which the paraphrase candidate sentence is accepted as having a linguistically correct meaning, and each of the linguistic acceptability score being obtained from a language model; and

outputting each of the one or more paraphrase candidate sentences when (i) each of the cumulative total of paraphrase acceptability scores is within the first acceptable limit, and (ii) each of the linguistic acceptability scores is within the second acceptable limit.

6. A machine translation system, comprising:

a processor; and

a medium having a computer program stored therein, the computer programs causing the processor to execute operations, including

receiving an original text;

generating, using the processor, one or more paraphrase candidate sentences by paraphrasing one or more of a plurality of fragments included in the original text into another expression in a language of the original text, the plurality of fragments being obtained by dividing the original text in accordance with a predetermined rule;

determining whether the paraphrasing of one or more of the plurality of fragments in each of the one or more paraphrase candidate sentences is acceptable by comparing each of a cumulative total of paraphrase acceptability scores with a first acceptable limit, each of the cumulative total of paraphrase acceptability scores being calculated by adding each of paraphrase acceptability scores in each of the one or more paraphrase candidate sentences, each of the paraphrase acceptability scores being assigned to a paraphrase pair including a first fragment and a second fragment that represents another expression of the first fragment, each of the paraphrase acceptability scores indicating a degree to which paraphrasing from the first fragment into the second fragment is accepted;

determining whether each of the one or more paraphrase candidate sentences is acceptable by comparing each of linguistic acceptability scores with a second acceptable limit, each of the linguistic acceptability scores indicating a degree to which the paraphrase candidate sentence is accepted as having a linguistically correct meaning and each of the linguistic acceptability scores being obtained from a language model;

outputting each of the one or more paraphrase candidate sentences when (i) each of the cumulative total of paraphrase acceptability scores is within the first acceptable limit, and (ii) each of the linguistic acceptability scores is within the second acceptable limit;

creating a translation corpus including a collection of a plurality of text pairs, the plurality of text pairs each including a first text in a first language paired with a second text in a second language different from the first language; and

translating a source text between the first language and the second language using the translation corpus, the source text representing a text to be translated,

wherein the creating of the translation corpus creates one or more new text pairs, the one or more new text pairs each including the second text and the one or more paraphrase candidate sentences, the created one or more new text pairs forming a new part of the translation corpus.