SYSTEM AND METHOD FOR AUTOMATIC, UNSUPERVISED PARAPHRASE GENERATION USING A NOVEL FRAMEWORK THAT LEARNS SYNTACTIC CONSTRUCT WHILE RETAINING SEMANTIC MEANING

• US 20180075016A1
• Filed: 09/15/2016
• Published: 03/15/2018
• Est. Priority Date: 09/15/2016
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: BENNETT, Laura J., KRISHNAMURTHY, Lakshminarayanan, PARAMESWARAN, Niyati, SUDARSAN, Sridhar
• Granted Patent: US 9,953,027 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 40/289 Phrasal analysis, e.g. finite state techniques or chunking
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/01 Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
• CPC Class Codes: G06N 5/04 Inference or reasoning models

Abstract

A system includes a question answering system executed by a computer, a processor, and a memory coupled to the processor. The memory is encoded with instructions that when executed cause the processor to provide training for training the question answering system. The training system is configured to receive a plurality of bidirectional disjunctive logical forms which include two directional disjunctions of differences between a first logical form of a first sentence and second logical form of a second sentence, realize the plurality of bidirectional disjunctive logical forms to generate a first plurality of paraphrases of the first and second sentence, score each of the first plurality of paraphrases based on textual similarity between the first plurality of paraphrases and the first and second sentences, and prune the first plurality of paraphrases to generate a second plurality of paraphrases based on the scores of each of the first plurality of paraphrases.

Claims

1. A method, comprising:

receiving, by a surface realizer of a paraphrase generation system, a plurality of bidirectional disjunctive logical forms, wherein the plurality of bidirectional disjunctive logical forms includes two directional disjunctions of differences between a first logical form of a first sentence and a second logical form of a second sentence;

realizing, by the surface realizer, the plurality of bidirectional disjunctive logical forms to generate first paraphrases of the first and second sentences;

determining a first score for a third paraphrase of the first paraphrases;

determining a second score for a fourth paraphrase of the first paraphrases, wherein the first score is higher than the second score based in part on a first syntactic variation between the third paraphrase and the first sentence and the second sentence being greater than a second syntactic variation between the fourth paraphrase and the first sentence and the second sentence; and

pruning, by the paraphrase generation system, the first paraphrases to generate second paraphrases, wherein the first paraphrases are pruned based on the first score and the second score such that the third paraphrase is included in the second paraphrases and the fourth paraphrase is not included in the second paraphrases based on the first score being higher than the second score.

2. The method of claim 1, wherein the differences include a first word that is an alternative to a second word and has a similar meaning to the second word, and wherein the two directional disjunctions include the first word combined with the second word while maintaining the same semantic relations with surrounding elements of a third sentence.

3. The method of claim 1, wherein realizing the plurality of bidirectional disjunctive logical forms includes generating natural language text from syntactic representations of text from the plurality of bidirectional disjunctive logical forms.

4. The method of claim 1, further comprising:

converting, by the paraphrase generation system, the first sentence into the first logical form by dependency parsing the first sentence to identify parts of speech of the first sentence thereby generating a first parsed representation of the first sentence; and

converting, by the paraphrase generation system, the second sentence into the second logical form by dependency parsing the second sentence to identify parts of speech of the second sentence thereby generating a second parsed representation of the second sentence.

5. The method of claim 4, further comprising generating, by the paraphrase generation system, at least one of the plurality of bidirectional disjunctive logical forms by comparing the first parsed representation of the first sentence with the second parsed representation of the second sentence to determine a word from the first sentence that corresponds with a word from the second sentence, and wherein realizing the plurality of bidirectional disjunctive logical forms includes generating natural language text from syntactic representations of text from the plurality of bidirectional disjunctive logical forms.

6. The method of claim 1, further comprising, providing the second of paraphrases to a training system as ground truth for training.

7. The method of claim 6, further comprising, training a question answering system utilizing the ground truth.

8. A system, comprising:

a processor; and

a memory coupled to the processor, the memory encoded with instructions that when executed by the processor cause the processor to;

receive a plurality of bidirectional disjunctive logical forms, wherein the plurality of bidirectional disjunctive logical forms includes two directional disjunctions of differences between a first logical form of a first sentence and a second logical form of a second sentence;

realize the plurality of bidirectional disjunctive logical forms to generate first paraphrases of the first and second sentences;

determine a first score for a third paraphrase of the first paraphrases;

determine a second score for a fourth paraphrase of the first paraphrases, wherein the first score is higher than the second score based in part on a first syntactic variation between the third paraphrase and the first sentence and the second sentence being greater than a second syntactic variation between the fourth paraphrase and the first sentence and the second sentence; and

prune the first paraphrases to generate second paraphrases, wherein the first paraphrases are pruned based on the first score and the second score such that the third paraphrase is included in the second paraphrases and the fourth paraphrase is not included in the second paraphrases based on the first score being higher than the second score.

9. The system of claim 8, wherein the differences include a first word that is an alternative to a second word and has a similar meaning to the second word, wherein the two directional disjunctions include the first word combined with the second word.

10. The system of claim 8, wherein realizing the plurality of bidirectional disjunctive logical forms includes generating natural language text from syntactic representations of text from the plurality of bidirectional disjunctive logical forms.

11. The system of claim 8, wherein the instructions, when executed by the processor, further cause the processor to:

convert the first sentence into the first logical form by dependency parsing the first sentence to identify parts of speech of the first sentence thereby generating a first parsed representation of the first sentence; and

convert the second sentence into the second logical form by dependency parsing the second sentence to identify parts of speech of the second sentence thereby generating a second parsed representation of the second sentence.

12. The system of claim 11, wherein the instructions, when executed by the processor, further cause the processor to generate at least one of the plurality of bidirectional disjunctive logical forms by comparing the first parsed representation of the first sentence with the second parsed representation of the second sentence to determine a word from the first sentence that corresponds with a word from the second sentence.

13. The system of claim 8, wherein the processor is configured to cause the memory to store the second paraphrases as ground truth for training.

14. The system of claim 13, further comprising a training system coupled to the processor, wherein the training system is configured to train a question answering system utilizing the ground truth.

15. A computer program product for generating paraphrases, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to:

receive a plurality of bidirectional disjunctive logical forms, wherein the plurality of bidirectional disjunctive logical forms includes two directional disjunctions of differences between a first logical form of a first sentence and a second logical form of a second sentence;

realize the plurality of bidirectional disjunctive logical forms to generate a first plurality of paraphrases of the first and second sentences;

determine a first score for a third paraphrase of the first paraphrases;

determine a second score for a fourth paraphrase of the first paraphrases, wherein the first score is higher than the second score based in part on a first syntactic variation between the third paraphrase and the first sentence and the second sentence being greater than a second syntactic variation between the fourth paraphrase and the first sentence and the second sentence; and

prune the first paraphrases to generate second paraphrases, wherein the first paraphrases are pruned based on the first score and the second score such that the third paraphrase is included in the second paraphrases and the fourth paraphrase is not included in the second paraphrases based on the first score being higher than the second score.

16. The computer program product of claim 15, wherein the differences include a first word that is an alternative to a second word and has a similar meaning to the second word, and wherein the two directional disjunctions include the first word combined with the second word.

17. The computer program product of claim 15, wherein realizing the plurality of bidirectional disjunctive logical forms includes generating natural language text from syntactic representations of text from the plurality of bidirectional disjunctive logical forms.

18. The computer program product of claim 15, wherein the program instructions are further executable by the computer to cause the computer to:

convert the first sentence into the first logical form by dependency parsing the first sentence to identify parts of speech of the first sentence thereby generating a first parsed representation of the first sentence; and

convert the second sentence into the second logical form by dependency parsing the second sentence to identify parts of speech of the second sentence thereby generating a second parsed representation of the second sentence.

19. The computer program product of claim 18, wherein the program instructions are further executable by the computer to cause the computer to generate at least one of the plurality of bidirectional disjunctive logical forms by comparing the first parsed representation of the first sentence with the second parsed representation of the second sentence to determine a word from the first sentence that corresponds with a word from the second sentence.

20. The computer program product of claim 15, wherein the program instructions are further executable by the computer to cause the computer to store the second plurality of paraphrases as ground truth for training.