SPEECH UNDERSTANDING SYSTEM USING AN EXAMPLE-BASED SEMANTIC REPRESENTATION PATTERN

• US 20110054883A1
• Filed: 11/19/2009
• Published: 03/03/2011
• Est. Priority Date: 09/01/2009
• Status: Active Grant

• Current Assignee: Electronics and Telecommunications Research Institute
• Original Assignee: Electronics and Telecommunications Research Institute
• Inventors: Kim, Seung Hi, Lee, Ilbin, Park, Jun, Yun, Seung, Lee, Soo Jong, Kim, Sanghun, Kim, Jeong Se, Park, Sang Kyu
• Granted Patent: US 8,370,130 B2
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G06F 40/268 Morphological analysis
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning

Abstract

A speech understanding apparatus includes: a speech recognition unit for recognizing an input speech to produce a speech recognition result; a sentence analysis unit for performing morpheme analysis on a sentence corresponding to the speech recognition result, extracting additional information, and performing syntax analysis; a hierarchy describing unit for describing hierarchy of the sentence; a class transformation unit for performing class transformation on the sentence; a semantic representation determination unit for marking optional expressions for the sentence, deleting meaningless expressions and the additional information, converting the sentence into its base form, and deleting morphemic tags or symbols to determine a semantic representation; a semantic representation retrieval unit for retrieving the determined semantic representation from an example-based semantic representation pattern database; and a retrieval result processing unit for selectively producing a retrieved semantic representation.

Claims

1. A pattern database building apparatus, comprising:

a sentence analysis unit for segmenting an input corpus into example sentences with reference to a phrase dictionary and performing morpheme analysis and syntax analysis on the segmented example sentences;

a hierarchy describing unit for describing hierarchy of the example sentences based on results of the morpheme analysis and the syntax analysis;

a class transformation unit for performing class transformation on the example sentences, whose hierarchy has been described, with reference to a thesaurus/ontology and a classification rule dictionary; and

a semantic representation pattern determination unit for marking optional expressions for the example sentences that have been undergone the class transformation, deleting meaningless expressions and additional information, converting the example sentences into their base form, deleting morphemic tags or symbols to determine a semantic representation pattern of the example sentences, and storing the determined semantic representation pattern in a database.

2. The pattern database building apparatus of claim 1, wherein the sentence analysis unit assigns the morphemic tags or symbols to the example sentences through the morpheme analysis, and indicates dependencies of the example sentences by text-indent based on a dependency grammar through the syntax analysis.

3. The pattern database building apparatus of claim 1, wherein the hierarchy of each of the example sentences is described by bracketing.

4. The pattern database building apparatus of claim 1, wherein the class transformation unit converts a noun into a hypernym class, converts synonyms from declinable words into a single word, assigns particles subject to phonological change due to a final consonant to the same class, and classifies numeric representations of time and amount by using automata.

5. The pattern database building apparatus of claim 1, wherein the additional information includes at least one of aspect, modality, tense, and sentence pattern.

6. A speech understanding apparatus, comprising:

a speech recognition unit for recognizing an input speech to produce a speech recognition result;

a sentence analysis unit for performing morpheme analysis on a sentence corresponding to the speech recognition result, recognizing and extracting additional information, and performing syntax analysis;

a hierarchy describing unit for describing hierarchy of the sentence based on the additional information and results of the morpheme analysis and the syntax analysis;

a class transformation unit for performing class transformation on the sentence, whose hierarchy has been described, with reference to a thesaurus/ontology and a classification rule dictionary;

a semantic representation determination unit for marking optional expressions for the sentence that has been undergone the class transformation, deleting meaningless expressions and the additional information, converting the sentence into its base form, and deleting morphemic tags or symbols to determine a semantic representation;

a semantic representation retrieval unit for retrieving the determined semantic representation from an example-based semantic representation pattern database; and

a retrieval result processing unit for selectively producing a retrieved semantic representation depending on whether or not the determined semantic representation exists in the example-based semantic representation pattern database.

7. The speech understanding apparatus of claim 6, wherein the sentence analysis unit assigns the morphemic tags or symbols to the sentence through the morpheme analysis, and indicates dependencies of the sentence by text-indent based on a dependency grammar through the syntax analysis.

8. The speech understanding apparatus of claim 6, wherein the hierarchy describing unit describes the hierarchy of the sentence by bracketing.

9. The speech understanding apparatus of claim 6, wherein the class transformation unit converts a noun into a hypernym class, converts synonyms from declinable words into a single word, assigns particles subject to phonological change due to a final consonant to the same class, and classifies numeric representations of time and amount by using automata.

10. The speech understanding apparatus of claim 6, wherein the retrieval result processing unit produces the speech recognition result if the determined semantic representation does not exist in the example-based semantic representation pattern database.

11. The speech understanding apparatus of claim 10, wherein the retrieval result processing unit creates a sentence corresponding to the retrieved semantic representation, or produces the speech recognition result, the retrieved semantic representation and the additional information, if the determined semantic representation exists in the example-based semantic representation pattern database.

12. A speech understanding method comprising:

recognizing an input speech to produce a speech recognition result;

performing morpheme analysis on a sentence corresponding to the speech recognition result, recognizing and extracting additional information, and performing syntax analysis;

describing the hierarchy of the sentence based on the additional information and results of the morpheme analysis and the syntax analysis;

performing class transformation on the sentence whose hierarchy has been described;

marking optional expressions for the sentence that has been undergone the class transformation, deleting meaningless expressions and the additional information, converting the sentence into its base form, deleting morphemic tags or symbols to determine a semantic representation;

retrieving the determined semantic representation from an example-based semantic representation pattern database; and

selectively producing a retrieved semantic representation depending on whether or not the determined semantic representation exists in the example-based semantic representation pattern database.

13. The speech understanding method of claim 12, wherein the morpheme analysis is performed by assigning the morphemic tags or symbols to the sentence, and the syntax analysis is performed by indicating dependencies of the sentences by text-indent based on a dependency grammar.

14. The speech understanding method of claim 12, wherein the hierarchy of the sentence is described by bracketing.

15. The speech understanding method of claim 12, wherein said performing class transformation includes:

converting a noun into a hypernym class;

converting synonyms from declinable words into a single word;

assigning particles subject to phonological change due to a final consonant to the same class; and

classifying numeric representations of time and amount by using automata.

16. The speech understanding method of claim 12, wherein said selectively producing the retrieved semantic representation produces the speech recognition result if the determined semantic representation does not exist in the example-based semantic representation pattern database.

17. The speech understanding method of claim 16, wherein said selectively producing the retrieved semantic representation creates a sentence corresponding to the retrieved semantic representation, or produces the speech recognition result, the retrieved semantic representation and the additional information, if the determined semantic representation exists in the example-based semantic representation pattern database.