Processing natural-language documents and queries

• US 10,585,924 B2
• Filed: 08/08/2014
• Issued: 03/10/2020
• Est. Priority Date: 08/08/2014
• Status: Active Grant

• Current Assignee: Cuong Duc Nguyen
• Original Assignee: Cuong Duc Nguyen
• Inventors: Nguyen, Cuong Duc
• Primary Examiner(s): Thai, Hanh B
• Application Number:
  US14/455,447
• Publication Number:
  US 20160042058A1
• Time in Patent Office: 2,041 Days
• Field of Search: 707771, 707778-779
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 40/205 Parsing
• CPC Class Codes: G06F 40/216 using statistical methods
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

The subject disclosure presents a natural language processing engine for analyzing an input sentence comprising one or more clauses, and generating a plurality of semantic structures for the sentence and the component clauses, based on statistically parsing an input sentence to generate a syntactic structure of the input sentence, examining the syntactic structure of phrases and subordinate clauses within the input sentence, and generating tuples representing a subject, verb, object, indirect object, supplement, type, etc. Each part of the tuple is a reference to an entity in an external knowledge base. Disclosed operations include linking a plurality of entities identified in the syntactic structure with corresponding entities found in an external knowledge base, operating a co-reference resolution, filtering the references from mentioned entities to external entities by semantic relations, and exporting the set of output tuples.

Claims

1. A system for parsing a natural-language sentence, the system comprising:

a server; and

a memory coupled to the server, the memory to store logical instructions that are executed by the processor to perform operations comprising;

creating a phrase-level syntactic structure based on a syntactic structure of an input query comprising one or more phrases;

wherein the creation of the phrase-level syntactic structure further comprises recognizing one or more clauses in the input query and parsing one or more verb phrases in each clause;

wherein recognizing the one or more clauses comprises matching the syntactic structure of the input query with one or more sentence templates to determine one or more main clauses and one or more clause-support subordinate clauses;

constructing an entity-level syntactic structure based on the phrase-level syntactic structure;

constructing a sentence-level semantic structure based on the entity-level syntactic structure; and

returning a set of tuples in the sentence-level semantic structure as an output.

2. The system of claim 1, wherein the parsing of the one or more verb phrases comprises parsing a verb tense and a verb part.

3. The system of claim 2, wherein the parsing of the verb tense further comprises matching a syntactic structure of the one or more verb phrases with one or more verb tense templates to determine a tense or a voice of a main verb.

4. The system of claim 2, wherein the parsing of the one or more verb parts in each clause comprises

finding all verb types and corresponding verb usage templates for a stemmed main verb;

matching one or more of the corresponding verb usage templates with each verb part;

extracting a plurality of parts from the syntactic structure of the one or more verb phrases based on the matched verb usage templates;

creating a phrase-level record from the extracted plurality of parts; and

creating a corresponding verb record for the stemmed main verb.

5. The system of claim 1, wherein constructing the entity-level syntactic structure further comprises:

parsing a first noun phrase for each phrase-level record;

parsing a second noun phrase in a direct object part of said each phrase-level record;

parsing a third noun phrase in an indirect object part of said each phrase-level record; and

parsing a supplement part of said each phrase-level record.

6. The system of claim 5, wherein the parsing of each noun phrase further comprises:

searching an entire surface form of the noun;

upon finding the entire surface form, parsing the noun phrase as a single noun phrase;

upon determining that the noun is a pronoun, tagging the pronoun for a subsequent co-reference resolution; and

upon determining that the noun phrase is not a single noun phrase, parsing the noun phrase as a complex noun phrase.

7. The system of claim 6, wherein the parsing of the noun phrase as a single noun phrase comprises:

searching within the noun phrase for a longest head noun that matches an entity name found in a knowledge base;

extracting one or more modifiers of the longest head noun;

parsing any existing adjective phrase in the noun phrase to extract additional modifiers; and

outputting an entity-level record for the noun phrase, the entity-level record including the longest head noun and the one or more modifiers and any additional modifiers.

8. The system of claim 6, wherein the parsing of the noun phrase as a complex noun phrase comprises:

matching one or more complex noun phrase templates to a syntactic structure of the noun phrase;

upon a determination that the complex noun phrase comprises a plurality of simple noun phrases, parsing each simple noun phrase and combining results;

parsing any existing prepositional phrases;

parsing any existing subordinate clauses; and

creating an entity-level record for each noun phrase as a returned result.

9. The system of claim 6, wherein parsing the complex noun phrase further comprises:

parsing a main noun phrase within the complex noun phrase to retrieve an entity-level record for the main noun phrase;

parsing a prepositional phrase within the complex noun phrase to retrieve a preposition and an entity-level record of the prepositional phrase;

creating a phrase-level record to the entity-level syntactic structure, the phrase-level record representing a relation between the entity-level record of the main noun phrase with the entity-level record of the prepositional phrase; and

returning the entity-level record of the main noun phrase as its results.

10. The system of claim 1, wherein the operations further comprise parsing a subordinate clause by:

matching one or more subordinate clause templates with a syntactic structure of a subordinate clause;

extracting a Wh-part and an internal clause;

parsing the Wh-part;

parsing one or more noun phrases and verb parts of the internal clause; and

creating a Phrase-Level Record for the subordinate clause.

11. The system of claim 1, wherein the operations further comprise parsing an adjective phrase by:

parsing an adjective in the adjective phrase;

parsing an adverb in the adjective phrase; and

outputting the parsed adjective and parsed adverb.

12. The system of claim 1, wherein the construction of the sentence-level sematic structure comprises:

executing a co-reference resolution on one or more pronouns based on the entity-level semantic structure;

filtering one or more entities by using one or more of a semantic relation in the entity-level semantic structure or a modifier of a noun phrase; and

constructing a list of tuples from the one or more entity-level records.

13. The system of claim 12, wherein the co-reference resolution further comprises:

searching on a list of previously-parsed entities; and

selecting a most appropriate entity as a reference entity for the pronoun based on one or more semantic guidelines.

14. A non-transitory computer-readable medium for storing computer-executable instructions that are executed by a processor to perform operations comprising:

parsing a first sentence received in a query to create a list of stated entities, wherein the query comprises a paragraph; and

parsing each remaining sentence in the paragraph with the list of stated entities and updating the list with one or more recognized entities in said each remaining sentence; and

creating a phrase-level syntactic structure based on a syntactic structure of the first sentence comprising one or more phrases;

wherein the creation of the phrase-level syntactic structure further comprises recognizing one or more clauses in the first sentence and parsing one or more verb phrases in each clause;

wherein recognizing the one or more clauses comprises matching the syntactic structure of the first sentence with one or more sentence templates to determine one or more main clauses and one or more clause-support subordinate clauses.

15. The computer-readable medium of claim 14, wherein the operations further comprise:

constructing a sentence-level semantic structure based on the phrase-level syntactic structure to enable the creating of the list of stated entities.

16. A non-transitory computer-readable medium for storing computer-executable instructions that are executed by a processor to perform operations comprising:

parsing a header part of an input document to retrieve a sentence-level semantic structure for the header part;

parsing a body part of the input document based on a plurality of entities retrieved from the sentence-level semantic structure for the header part;

calculating a related measurement between each of the plurality of entities in the input document; and

creating a phrase-level syntactic structure based on a syntactic structure of the header comprising one or more phrases;

wherein the creation of the phrase-level syntactic structure further comprises recognizing one or more clauses in the header and parsing one or more verb phrases in each clause;

wherein recognizing the one or more clauses comprises matching the syntactic structure of the header with one or more sentence templates to determine one or more main clauses and one or more clause-support subordinate clauses.

17. The computer-readable medium of claim 16, wherein the operations further comprise matching the header part with one or more predefined sentence templates.

18. The computer-readable medium of claim 16, wherein the operations further comprise automatically detecting and adding a new sentence template to the one or more predefined sentence templates.