KR101501610B1 - System and method for recognition quote/speaker - Google Patents

Abstract

A quotation / speaker recognition system and method are disclosed. A citation / speaker recognition system includes a citation / speaker recognition unit for analyzing a document and recognizing a citation and a speaker, a speaker tagging unit for tagging the recognized speaker with respect to the citation, and a speaker And a search result providing unit for searching for a citation and providing the search result as a search result.

Quotation, speaker, document, sentence structure, morpheme analyzer, sentence pattern

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a system and method for recognizing a speaker,

The present invention relates to a system and method for recognizing a citation / speaker from a sentence, and more particularly, to a system and method for recognizing a citation from a sentence through morphological analysis and pattern processing, and recognizing a speaker who uttered the citation.

As Internet technology advances, there are many services that extract specific information from a document and provide it to users. Especially, it is required to discover services that can provide more useful information to users by analyzing documents.

In this regard, documents such as news are a technology that extracts issues from WHO / WHERE / WHERE attributes and news among news frameworks of 5W 1H that are breaking news, timeliness, and sociality Is required.

Specifically, a document such as news is composed of a form in which a person has uttered a certain content, so that a quote and a speaker that utter a quotation are often described in the document. However, since the document is described in various words, it is not easy to recognize a speaker who uttered a quotation and a quotation from a news source such as a computer apparatus, not a person who can easily grasp the document structure.

In particular, since a document is not always created in a fixed pattern, a method of efficiently and accurately recognizing a quotation and a speaker from various documents is required.

The present invention provides a citation / speaker recognition system and method capable of improving accuracy by recognizing a speaker who has uttered a citation and a quotation through a sentence analysis approach and a pattern recognition processing method.

The present invention provides a citation / speaker recognition system and method for building a pattern processing dictionary to extract text tokens and improving the accuracy of recognizing citation and speaker by recognizing the citation and the speaker of the sentence based on the text token.

The present invention provides a citation / speaker recognition system and method that improves the accuracy of recognizing quotations and speakers by recognizing quotations and speakers using morpheme analyzers.

The present invention recognizes a quotation and a speaker by using a morpheme analyzer, and when a sentence includes a plurality of quotations and a speaker by tagging a speaker with respect to a quotation based on a predetermined sentence pattern indicating a relationship between the speaker and a quotation A citation / speaker recognition system and method capable of accurately matching a citation and a speaker are provided.

The present invention recognizes a quotation and a speaker by using a morpheme analyzer, tags a speaker on a quotation on the basis of a question keyword included in the sentence, a preset sentence pattern indicating a relationship between the speaker and the quotation, And a speaker / speaker recognition system and method capable of accurately matching a speaker with a speaker even in exceptional situations.

According to the present invention, it is possible to combine a plurality of recognition methods for recognizing a citation and a speaker, or to recognize a citation and a speaker from each sentence by selecting one of the recognition methods according to the degree of the error, A speaker / speaker recognition system and a speaker recognition method capable of improving a speaker / speaker recognition system.

A citation / speaker recognition system according to an embodiment of the present invention includes a citation / speaker recognition unit for analyzing a document and recognizing a citation and a speaker, a speaker tagging unit for tagging the recognized speaker with respect to the citation, And a search result providing unit for searching the cited talk sentence of the speaker upon receiving the request and providing the search result as a search result.

According to an embodiment of the present invention, the citation / speaker recognition unit includes a citation recognition unit for extracting a sentence from the document and recognizing a citation in the sentence, a speaker candidate determination unit for analyzing the sentence to determine a speaker candidate for the citation, A speaker selection unit for analyzing the sentence to select a speaker who has uttered the quotation from among the speaker candidates, and a speaker recognition unit for recognizing the unique name of the selected speaker.

A method of recognizing a citation / speaker according to an embodiment of the present invention includes analyzing a document to recognize a citation and a speaker, tagging the recognized speaker with respect to the citation, and receiving a search request for the speaker, And providing the retrieved result as a search result.

According to an embodiment of the present invention, a citation / speaker recognition system and method capable of improving the accuracy by recognizing a speaker who uttered a citation and a quotation through a sentence analysis approach and a pattern recognition processing method are provided.

According to an embodiment of the present invention, various cognitive methods for recognizing a speaker and a speaker can be combined, or a citation and a speaker can be recognized according to respective recognition methods, and a citation can be selected from a sentence And a citation / speaker recognition system and method capable of improving the accuracy of recognizing the speaker.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the present invention is not limited to or limited by the embodiments. Like reference symbols in the drawings denote like elements.

FIG. 1 is a block diagram illustrating an entire configuration of a citation / speaker recognition system according to an embodiment of the present invention. Referring to FIG.

Referring to FIG. 1, the citation / speaker recognition system 100 may include a citation / speaker recognition unit 101, a speaker tagging unit 102, and a search result providing unit 103.

The quotation / speaker recognition unit 101 can analyze the document 108 and recognize the quotation and the speaker. At this time, the document 108 may include at least one Quote and Speaker. In one example, document 108 may be a news article that includes quotes and speakers.

For example, the quotation / speaker recognition unit 101 can recognize a citation and a speaker by applying any one of the first recognition method to the fourth recognition method or a combination thereof to a sentence.

At this time, the first recognition method may mean constructing a pattern processing dictionary to extract a text token, and recognize a citation and a speaker of the sentence based on the text token. The second recognition method may be to extract the parts of speech information from the sentence using the morpheme analyzer and recognize the citation and the speaker of the document 108 based on the part of speech information.

If the sentence contains a plurality of quotations, the third recognition method and the fourth recognition method extract speech-of-speech information from the sentence using the morpheme analyzer, and recognize the citation and the speaker of the document 108 based on the speech- can do. At this time, the third recognition method refers to tagging the speaker with respect to the quotation based on a predetermined sentence pattern indicating the relationship between the speaker and the quotation. The fourth recognition method includes the question keyword included in the sentence, It may mean tagging a speaker for a quotation based on a predetermined sentence pattern indicating a relationship with the quotation.

For example, the quotation / speaker recognition unit 101 can recognize the citation and the speaker from the sentence by combining the above-mentioned four recognition methods. As another example, the quotation / speaker recognition unit 101 recognizes a quotation and a speaker from a sentence by applying a preset citation and a recognition method for recognizing the speaker, and recognizes the citation and the speaker from the sentence, You can select a citation and a speaker according to the method.

In FIGS. 2 to 10, the first to fourth recognition methods can be specifically described.

For example, the citation / speaker recognition unit 101 may include a citation recognition unit 104, a speaker candidate determination unit 105, a speaker selection unit 106, and a speaker recognition unit 107.

The citation recognizing unit 104 can extract a sentence from the document 108 and recognize the citation in the sentence. The document 108 may be composed of a plurality of sentences, and each of the plurality of sentences may be composed of at least one speaker and a quotation. For example, the citation recognizing unit 104 can recognize a citation by extracting a citation from a sentence by tokenizing the sentence.

For example, when the above-mentioned first to fourth recognition methods are applied, the citation recognizing unit 104 can recognize a phrase from a point at which a quotation mark starts to a point at which a quotation mark ends, as a quotation have.

The speaker candidate determination unit 105 can analyze the sentence and determine the speaker candidate of the quotation. In other words, since there can be a plurality of quotations and a plurality of speakers in one sentence, it is possible to determine a speaker candidate that has uttered a quotation from among a plurality of speakers.

For example, when the above-mentioned first to fourth recognition methods are applied, the speaker candidate determining unit 105 can determine a phrase as a speaker candidate in the sentence.

The speaker selection unit 106 can analyze a sentence and select a speaker who has uttered a quotation from among speaker candidates. Specifically, the speaker selection unit 106 can determine the relationship between the quotation and the speaker through the context analysis of the quotation.

The speaker recognition unit 107 can recognize a unique name of the speaker. For example, when the speaker is the A chairman, the speaker recognition unit 107 can recognize the original name of the A chairman.

The speaker identification unit 107 can recognize a unique name of the speaker even when the speaker is composed of a part indicating the gender and a part indicating the title. Specifically, in the case of news articles, "Kim Dae Jung President" or "Kim Young Sang President", which is a proper noun called by the subject, is first presented before using the subject of "President Kim".

The citation / speaker recognition unit 101 according to an embodiment of the present invention may analyze the entire document and store the first name of the first speaker. Then, the speaker recognition unit 107 compares and matches the distinguished name of the speaker recognized with respect to the specific quotation with the name of the speaker previously stored, so that if the speaker recognized for the specific quotation is composed of "sex + position (title) , The speaker's unique name can be recognized.

If the unique name of the speaker is not described in the preceding sentence or a plurality of similar names appear, the speaker recognition unit 107 may analyze the context of the preceding and following sentences and perform exceptional processing.

Through this process, the quotation / speaker recognition unit 101 can recognize the quotation and the speaker from the sentences included in the document.

The speaker tagging unit 102 can tag the speaker for the recognized quotation. That is, the speaker tagging unit 102 can index the citation as a speaker by constructing one set 109 by connecting the speaker and the quotation. If the speaker is input through the person search, the citation / speaker recognition system 100 may extract at least one citation tagged in the speaker and provide the extracted result as a search result.

For example, the process of tagging a speaker with respect to a citation recognized from a sentence may be performed in real time, or may be stored in a specific DB before the user searches for a speaker.

When the search result providing unit 103 receives the search request for the speaker, the search result providing unit 103 may search the quoted speech made by the speaker and provide the search result. For example, when the name of the speaker is input from the user, the search result providing unit 103 may search at least one citation tagged in the corresponding speaker and provide the search result to the user. Quotations can be provided to the user in various forms depending on the service.

For example, the search result providing unit 103 may search for a quotation sent by a specific speaker and provide it to the user. Then, the search result providing unit 103 receives the specific search query and the name of the speaker together, and can search for and provide a quotation containing the search query among the quotations that the speaker has uttered. In addition, the search result providing unit 103 may extract and provide the words most frequently referred to in the quotation for a predetermined period of time by a specific speaker.

As another example, if the speaker is an expert, the search result providing unit 103 may provide data related to the quotation sent by the expert together with the quotation. For example, if the speaker is a securities analyst, the speaker may provide a citation analyzing the stock quotes and a securities index at the time the quotation is uttered.

As another example, when the search result providing unit 103 receives a specific speaker from the user, the search result providing unit 103 may search for and provide a quotation that the person associated with the speaker has uttered.

2 is a diagram illustrating a first recognition method for recognizing a quotation / speaker according to an embodiment of the present invention.

Fig. 2 shows a process of recognizing a citation and a speaker from a sentence through an automata. Specifically, FIG. 2 shows a process of extracting a text token by constructing a pattern processing dictionary before recognizing a citation and a speaker from the sentence.

If the inputted sentence passes through each of the states 202 to 207 shown in FIG. 2, a plurality of text tokens can be extracted from the sentence. The table in Fig. 2 may mean a pattern processing dictionary. As can be seen in the table, the pattern processing dictionary can mean a "pharase" unit consisting of words constituting a sentence. As shown in FIG. 2, after the AS 202 is given a quotation, the BQ 203 is a quote, Q 204 is a quotation, AQ 205 is a quote, BV 206 is a verb phrase, ) Represents the end of the verb phrase. Then, S (201) may indicate the start and F (208) may indicate the end. That is, the citation / speaker recognition system extracts the text token from the sentence through the pattern processing dictionary, and recognizes the citation and the speaker using the extracted text token.

The citation and the speaker can be recognized from the sentence while the sentence enters each state included in the process from S (201) to F (208). The quotation / speaker recognition system may determine that the sentence has entered the AS 202 after the sentence examination (/ / / / / /). Then, if quotation marks ("or") begin after AS 202, the statement may enter Q 204. Also, if quotation marks ("or") do not start after AS 202 , The sentence can enter the BQ 203. Referring to FIG. 2, the BG 203 state can be repeated continuously. Then, if the quotation marks start after the BQ 203 state, the sentence can enter Q 204.

When Q (204) is followed by a quote, the sentence can enter the AQ (205). Then, if the quotation check (say / / and / /) is displayed after the AQ 205, the sentence can enter the BV 206. On the other hand, the sentence may enter the BQ 205 again after the AQ 204.

If a new quote begins after BV 206, the sentence can enter Q 204. If the verb (~) appears after the BV 206, the sentence can enter the AV 207. If a period appears in the sentence, the sentence may enter F (208) and the process may end.

FIG. 3 is a diagram illustrating an example in which a first recognition method is applied to a sentence according to an embodiment of the present invention. In FIG. 3, the first recognition method can be applied to the following sentence.

"Hong has appeared on BBS Radio 'morning journal of Kim Jae Won' on this day, taking as an example the cooperation of Bismarck and La Salle in Germany in the nineteenth century." Park has patriotism and there is a field to share. "

If the above-mentioned sentence is applied to the first recognition method, the sentence can enter the AS 202, which is in a state after the start of the sentence '201', after the ' In other words, 'Hong-won' can mean subject. Since the quotation mark does not start immediately after the AS 202, the sentence can enter the BQ 203, which is a state before the quotation. Thereafter, when the quotation mark (") is started, the sentence can enter Q (204), which is a quote state. Then, at the end of the quotation, the sentence can enter the quoted state AQ 205 . At this time, the quotation may mean "there is a field to share because patriotism is patriotic", which is the phrase from the point where the quotation marks start to the point where it ends.

Then, since a quotation check "high" appears after the quotation, the sentence can enter the BV 206, which is a state before the verb phrase. After that, if the verb phrase 'say' is passed, the sentence can enter AV (207), which is the state after the verb phrase. Since a period appears after the verb phrase, the sentence can be terminated (208).

At this time, the quotation / speaker recognition system can recognize the phrase from the point where the quotation mark is started to the point where the quotation mark ends, as a quotation. Taking the above sentence as an example, a quotation such as "Park has patriotism and there is a field to share" can be recognized as a quotation.

In addition, the quotation / speaker recognition system can determine a noun phrase located at the beginning of the phrase investigation as a speaker candidate in a sentence. Taking the above sentence as an example, the "Mr. Hong" can be determined as the speaker candidate of the quotation. And, the quotation / speaker recognition system recognizes the speaker for the quotation by recognizing the 'Hong doctor' determined by the speaker candidate as the unique name of Hong xx.

4 is a diagram illustrating a second recognition method for recognizing a quotation / speaker according to an embodiment of the present invention.

FIG. 4 shows a process of recognizing a citation and a speaker from a sentence through an automata. Specifically, FIG. 4 shows a process of extracting part of speech information from a sentence using a morpheme analyzer as a partial parser. The quotation / speaker recognition system can recognize the quotation and the speaker of the document based on the extracted part-of-speech information.

If the inputted sentence passes through each of the states 402 to 407 shown in FIG. 4, parts of speech information can be extracted from the sentence. The table in FIG. 4 may refer to a morpheme analyzer. Referring to the morpheme analyzer, it can be seen that noun phrases, quotation marks, endings, and verbs are stored in the sentence and the rest are passed.

Referring to FIG. 4, N (402) is a noun phrase, R (403) is a guitar, Q (404) is a quotation, C (405) is a quote, E (406) is an end and V (407) is a verb. Then, S (401) may indicate the start and F (408) may indicate the end. In this case, the noun phrase may include a suffix, a prefix, a noun, a Hangul numeral, an adjective, a syllable noun, a pronoun, a proper noun, an alphabet or a number.

The sentence may enter N 402 via start 401. Then, when the nomination survey (E / / / / /) is displayed, the sentence can enter R (403). After that, if the quotation mark (") is started, the sentence enters Q (404), and if the quotation mark is ended, the sentence can enter C (405) ), The sentence can enter R (403). After that, if another quote begins, the sentence can again enter Q (404).

If NOUN appears after R (403), the sentence enters E (406), and if VERB appears after R (403), the sentence can enter V (407). If the noun appears after entering V (407), the sentence can again enter R (403). If V (407) appears followed by a period, the sentence can enter F (408). As a result, a plurality of parts of speech information can be extracted from the sentence through the process shown in FIG.

Then, the quotation / speaker recognition system can recognize the phrase from the point where the quotation mark starts to the point where the quotation mark ends, as a quotation. In addition, the quotation / speaker recognition system can determine a noun phrase located in the front part of the sentence in the sentence as a speaker candidate.

5 is a diagram illustrating an example in which a second recognition method is applied to a sentence according to an embodiment of the present invention.

5 is a diagram illustrating an example of applying a second recognition method to a sentence according to an embodiment of the present invention. In FIG. 5, the second recognition method can be applied to the following sentence.

"In the meantime, Kwon said," Now, I can see an inter-Korean relationship that I can not even see before. "" Peace on the Korean peninsula should be solved. "

If the above sentence is applied to the second recognition method, the sentence is recognized as a noun phrase through the start (401), and when the 'noun member' appears, it can enter N (402). Then, after passing through the 'N', the sentence can enter R (403). Thereafter, when the quotation marks begin, the sentence can enter Q (404).

Then, when the quotation mark ends, the sentence enters C (405), and after that, if the quotation check is performed, the sentence can enter R (403). Thereafter, once again a new quote begins, the sentence enters Q 404, and when the quote ends, the sentence can enter C (405). Finally, the sentence in Fig. 4 can consist of a quotation such as "now seeing now" and "a thing to be solved on the Korean peninsula".

(406) and V (407), the sentence becomes "R" (403), and the sentence becomes " . ≪ / RTI > Finally, if a period appears, the sentence can be terminated via F (408). As a result, parts of speech information of a sentence can be extracted through this process.

Then, the quotation / speaker recognition system can recognize the quotation and the speaker in the sentence based on the part-of-speech information. For example, the quotation / speaker recognition system can recognize a phrase from the point where the quotation mark starts to the point where the quotation mark ends, as a quotation. According to the sentence above, quotations such as "now looking at" and "what should be solved on the Korean peninsula" in the document can be recognized as quotes.

And, in the quotation / speaker recognition system, a noun phrase located at the beginning of the narrative search in the sentence can be determined as the speaker candidate. According to the above sentence, the 'lawmaker' in the document can be determined as the speaker candidate. Then, the quotation / speaker recognition system can recognize the unique name xx, which is the name of the 'member'. According to the example of FIG. 5, it can be seen that the volume xx is the speech of two quotations, and the two quotations can be tagged as the volume xx.

6 is a diagram illustrating a third recognition method for recognizing a quotation / speaker according to an embodiment of the present invention.

Figure 6 shows the process of recognizing a citation and a speaker from a sentence through a logic diagram. In particular, FIG. 6 can have meaning when a plurality of quotations and a plurality of speakers exist in a sentence. Specifically, the quotation / speaker recognition system extracts parts-of-speech information from a sentence of the morpheme analyzer, and recognizes a quotation and a speaker of the document based on parts-of-speech information. Then, the quotation / speaker recognition system can tag the speaker for the quotation based on a predetermined sentence pattern indicating the relationship between the speaker and the quotation.

If the inputted sentence passes through each of the states 602 to 604 shown in Fig. 6, parts of speech information can be extracted from the sentence. The table of FIG. 6 may refer to a morpheme analyzer. Referring to the morpheme analyzer, it can be seen that noun phrases and quotations are stored in the sentence and the rest are passed.

If the sentence starts at the beginning (601), the sentence can enter N (602) after the noun phrase is extracted from the sentence and the parts of speech information of the noun phrase is determined. Also, if the quotation marks begin immediately after the beginning of the sentence 601, the sentence can enter Q (604).

Then, if the N (602) state is maintained and the noun examination ((/, / / / /)) appears, the sentence can enter R (603). At this time, if a period is displayed in R (603), the sentence can be terminated by entering F (605). Then, when the quotation mark (") is started after R (603), the sentence can be extracted by entering Q (604).

At this time, the quotation / speaker recognition system can recognize the phrase from the point where the quotation mark starts to the point where the quotation mark ends, as a quotation. In addition, the quotation / speaker recognition system can determine the form in which the nominal phrase set in the sentence is combined with the nominative examination as the speaker candidate.

As a result, according to FIG. 6, a plurality of subjects and quotations included in a sentence can be extracted through an iterative process. The extracted subject and quotation can be recognized as a speaker and a quotation, respectively. If multiple speakers and quotations are recognized, it is important to know which citation the speaker has uttered. In this process, the relationship between the speaker and the quotation can be grasped through a predetermined sentence pattern. An example of a sentence pattern can be described in Fig.

7 is a diagram illustrating an example in which a third recognition method is applied to a sentence according to an embodiment of the present invention. FIG. 7 shows a process of recognizing a speaker and a quotation by applying a third recognition method to the following sentence, and tagging a speaker with respect to the quotation.

"Democratic Party chairman Chung Sye-kyun said on January 8 that Gohung-gil, the chairman of the National Assembly Culture, Sports, Tourism and Broadcasting Communications Commission, said that the Grand National Party lawmaker" will introduce the media relations law to the extraordinary session of the National Assembly in February. "" "

If we apply the above sentence to the third recognition method, the sentence will be recognized as a noun phrase after the start (601) and if the 'democrat representative' appears, it can enter N (602). Then, after passing the noun clause '', the sentence can enter R (603). Again, if the subject noun phrase appears, the sentence can enter N (602). Then, if the sentence exceeds' ',' the sentence can enter R (603). At this time, if R (603) is maintained and the quotation mark is started, the sentence can keep Q (604) until Q (604) is reached and the quote ends. Then, when the quotation marks are terminated, the sentence can enter R (603). And, once the quote begins again, the sentence can enter Q (604). When the quotation marks are terminated, the sentence can enter R (603). If a period appears after that, the sentence can be terminated via F (605).

In this way, the part of speech information of the sentence is extracted, and the quotation / speaker recognition system can determine the phrase as a speaker candidate located at the beginning of the frequency search and recognize the quotation between quotation marks as a quotation. According to the above sentence, 'Jeong Seok-gyun's Democratic Representative' and 'Gohung-gil's GN Grand Assembly' can be determined as speaker candidates, and 'I will assume the media relations law' and 'what can be assumed ~' can be recognized as quotes. At this time, the speaker can be recognized as Jeongseekun and Goheunggil, which are unique names of the speaker candidate.

However, in the case of the sentence, since a plurality of quotations and speakers are recognized, it is important that the speaker tagging process determines which citation the utterer has uttered. Although humans can easily tag a speaker on a citation through a contextual structure, the system is not able to grasp this makeup tagging process. According to an embodiment of the present invention, the citation / speaker recognition system can preliminarily define a frequently used sentence pattern, thereby grasping the relationship between the speaker and the citation according to the sentence pattern. The sentence pattern can be explained in detail in Fig.

8 is a table showing an example of tagging a speaker according to a sentence pattern of a speaker and a quotation according to an embodiment of the present invention.

Referring to FIG. 8, when there are a plurality of recognized speakers and quotations from the sentence, the relationship between the speaker and the quotation can be grasped according to the sentence patterns. 8, N is a speaker and Q is a quotation.

If each of the recognized speakers and the quotation is N1-Q1, then the quotation Q1 can be tagged as the speaker N1 uttered. In the case of a sentence pattern such as N1-Q1-Q2, the quotation Q1 and the quotation Q2 can be tagged as a speech of the speaker N1, respectively.

On the other hand, in the case of a sentence pattern such as N1-Q1-N2-Q2, the quotation Q1 is spoken by the speaker N1, and the quotation Q2 is tagged as spoken by the speaker N2. Also, in the case of a sentence pattern such as N1-N2-Q1-Q2, the quotation Q1 may be tagged as a speaker N2, and the quotation Q2 may be tagged as a speaker N1. Also, in the case where there is no speaker such as Q1 and Q1-Q2, the speaker may not be tagged with respect to the quotation.

8, the above sentence represents a sentence pattern such as N1-N2-Q1-Q2, so it is the utterance of the phrase "I will assume the press relations law" , And the quotation "of course, what can be assumed" can be tagged as the utterance of the speaker, Jung Sye-kyun.

The example shown in FIG. 8 is merely an example of a relationship between a speaker and a citation according to a sentence pattern, and the citation / speaker recognition system according to an embodiment of the present invention is not limited to the example of FIG.

9 is a diagram illustrating a fourth recognition method for recognizing a quotation / speaker according to an embodiment of the present invention.

Figure 9 shows the process of recognizing a citation and a speaker from a sentence through a logic diagram. In particular, FIG. 9 can have a meaning when a plurality of quotations and a plurality of speakers exist in the sentence and the question keyword is included in the sentence. That is, the fourth recognition method relates to an exceptional situation (when a query keyword such as a question or a question is included in the sentence) among the third recognition method.

Specifically, the quotation / speaker recognition system extracts parts-of-speech information from a sentence of the morpheme analyzer, and recognizes a quotation and a speaker of the document based on parts-of-speech information. Then, the quotation / speaker recognition system can tag the speaker with the query keyword based on the query keyword included in the sentence, the preset sentence pattern indicating the relationship between the speaker and the quotation, and the like. At this time, a description of the sentence pattern can be referred to Fig.

If the input sentence passes through each of the states 902 to 905 shown in Fig. 9, parts of speech information can be extracted from the sentence. The table in FIG. 9 may refer to a morpheme analyzer. Referring to the morpheme analyzer, it can be seen that noun phrases, quotations, and question keywords are stored in the sentence, and the rest are passed.

If the sentence passes the start (901), the sentence can enter N (902) after the part of speech information of the phrase is determined from the sentence. Also, if a quote begins immediately after the sentence has gone through the start (901), the sentence can enter Q (905).

Then, if the N (902) state is maintained and the noun examination ((/, / / / /)) appears, the sentence can enter R (903). At this time, if a period appears in R (903), the sentence can enter F (905) and terminate. Then, if a question keyword such as a question or a question appears after R (903), the sentence can enter EX (904). Then, the query keyword can be extracted from the sentence and stored. Question keywords can be extracted repeatedly in sentences. Further, if the quotation mark (") starts after R (903), the sentence can be extracted by entering Q (904).

At this time, the quotation / speaker recognition system can recognize the phrase from the point where the quotation mark starts to the point where the quotation mark ends, as a quotation. In addition, the quotation / speaker recognition system can determine a noun phrase located at the beginning of the phrase investigation as a speaker candidate in the sentence.

As a result, according to FIG. 9, a plurality of subjects and quotations included in a sentence can be extracted through an iterative process. In particular, if the sentence includes a question keyword, the question keyword can be extracted. The extracted subject and quotation can be recognized as a speaker and a quotation, respectively. If multiple speakers and quotations are recognized, it is important to know which citation the speaker has uttered. In this process, the relationship between the speaker and the quotation can be grasped through a predetermined sentence pattern. An example of a sentence pattern can be described in Fig. However, the citation and the speaker recognized according to the sentence pattern can be changed according to the question keyword.

10 is a diagram illustrating an example in which a fourth recognition method is applied to a sentence according to an embodiment of the present invention.

10 is a diagram illustrating an example in which a fourth recognition method is applied to a sentence according to an embodiment of the present invention. FIG. 10 shows a process of recognizing a speaker and a quotation by applying a fourth recognition method to the following sentence, and tagging a speaker with respect to the quotation.

"Democratic Party leader Won Hye-kyung will appear on the news show of Kim Hyun-jung, the CBS Liao, saying," Do you think North Korea's hard-line attitude toward South Korea is a temporary phenomenon during Kim's death? "" It should be judged carefully. "

When the above sentence is applied to the fourth recognition method, the sentence is recognized as a noun phrase when the 'Democratic Party Won Hye-kyung Representative' appears in the beginning (901), and can enter N (902). Then, after passing through the 'N', the sentence can enter R (903). Thereafter, when the quotation mark is started while maintaining R (903), the sentence can enter Q 904 and maintain Q (904) until the quote ends. Then, when the quotation marks are terminated, the sentence can enter R (903).

At this time, if the question keyword is displayed while maintaining R (903), the sentence can return to R (903) after entering EX (904). Thereafter, when the quotation mark is started while maintaining R 903, the sentence enters Q 905 and keeps Q 905 until the quotation is terminated. Then, when the quotation marks are terminated, the sentence can enter R (903). If a period appears after that, the sentence can be terminated via F (905).

In this way, the part of speech information of the sentence is extracted, and the quotation / speaker recognition system can determine the phrase as a speaker candidate located at the beginning of the frequency search and recognize the quotation between quotation marks as a quotation. According to the above sentence, 'Democratic Party Hye-kyung Won-kyu's representative' can be decided as speaker candidate. In addition, quotes such as "whether North Korea's hard-line attitude toward North Korea is a temporary phenomenon" and "should be judged with caution" should be recognized as a quotation. At this time, the speaker can be recognized as a unique name of the speaker candidate, Won Hye-kyung.

However, in the case of the sentence, since a plurality of quotations and speakers are recognized, it is important that the speaker tagging process determines which citation the utterer has uttered. Although humans can easily tag a speaker on a citation through a contextual structure, the system is not able to grasp this makeup tagging process. According to an embodiment of the present invention, the citation / speaker recognition system can preliminarily define a frequently used sentence pattern, thereby grasping the relationship between the speaker and the citation according to the sentence pattern. The sentence pattern can be explained in detail in Fig. However, when the sentence includes a question keyword such as 'question' as shown in FIG. 10, the sentence pattern of FIG. 8 can not be applied as it is. Therefore, it is required to accurately match the recognized citation and the speaker with each other in consideration of the question keyword.

If the sentence of FIG. 10 is applied to FIG. 8, since N1-Q1-Q2, the quotation Q1 and the quotation Q2 can be tagged as having been spoken by the speaker N1, respectively. However, this conclusion is actually a contextual error. In other words, the quotation Q1 can not be tagged as the speaker N1 because the quotation Q1 is not a result of N1 being uttered but a third party distinguishing it from N1.

Therefore, even if the sentence in FIG. 10 is N1-Q1-Q2, the quotation / speaker recognition system can tag the speaker N1 for the quotation Q2. In other words, the quotation that "should be judged as temporary" should be tagged as the speaker Won Hye-kyung's utterance, and the phrase "North Korea's hard-line attitude toward the North as temporary phenomenon" can be tagged as having no speaker.

11 is a diagram illustrating an example of determining a combination of a citation and a speaker by combining a citation / speaker recognition method according to an embodiment of the present invention.

According to Fig. 11, four recognition methods for recognizing a quotation and a speaker from a document are shown. At this time, the recognition methods 1, 2, 3, and 4 correspond to the first, second, third, and fourth recognition methods described above.

According to an embodiment of the present invention, the citation / speaker recognition system can recognize a citation and a speaker from a sentence included in the document by combining preset citation and at least one recognition method for recognizing the speaker.

That is, the citation / speaker recognition system can recognize the citation and the speaker from the sentence by mixing the first, second, third, and fourth recognition methods.

According to another embodiment of the present invention, the citation / speaker recognition system recognizes a citation and a speaker from a sentence included in the document by applying preset citation and at least one recognition method for recognizing the speaker, And the speaker and the speaker according to the recognition method having the lowest error degree can be selected.

That is, the citation / speaker recognition system recognizes the citation and the speaker from the sentences by applying the first, second, third, and fourth recognition methods, and compares the errors of the results according to each recognition method to select the lowest recognition method voting to select the final citation and speaker.

11 is only an example of recognizing a quotation and a speaker, and the citation / speaker recognition system according to an embodiment of the present invention is not limited to the configuration of FIG.

FIG. 12 is a flowchart illustrating an entire process of a citation / speaker recognition method according to an embodiment of the present invention.

In step S1201, the quotation / speaker recognition system may analyze the document to recognize the quotation and the speaker. Referring to FIG. 12, step S1201 of recognizing a quotation and a speaker may include steps S1204, S1205, S1206, and S1207.

In step S1204, the quotation / speaker recognition system extracts a sentence from the document and recognizes the quotation in the sentence.

In step S1205, the quotation / speaker recognition system may analyze the sentence to determine the speaker candidate of the quotation. According to an embodiment of the present invention, the citation and the speaker candidate can be recognized through the first through fourth recognition methods according to the sentence analysis approach and the pattern recognition processing method.

In step S1206, the quotation / speaker recognition system analyzes the sentence and selects a speaker who has uttered the quotation from among the speaker candidates. Specifically, the speaker selection unit 106 can determine the relationship between the quotation and the speaker through the context analysis of the quotation. In particular, when a sentence includes a plurality of quotations, it is important to select a speaker who has uttered the quotation.

In step S1207, the quotation / speaker recognition system may recognize a unique name of the selected speaker. If the speaker is composed of "last name + title (title) ", the quotation / speaker recognition system analyzes the document as a whole and recognizes the speaker's unique name by pre-storing the original name of the speaker.

In step S1202, the quotation / speaker recognition system may tag the recognized speaker for the quotation. That is, when a sentence contains a plurality of quotations and a plurality of speakers, the quotation / speaker recognition system can index the quotations as a speaker by tagging the speakers for each quotation to form one set. The indexed quotation can be detected later by a speaker search.

In step S1203, upon reception of the search request for the speaker, the quotation / speaker recognition system may search for a quotation sent by the speaker and provide the search result. Quotations can be provided to the user in various forms depending on the service.

The step of analyzing the document and recognizing the quotation and the speaker (S1201) may be performed according to the following first to fourth recognition methods.

For example, the citation / speaker recognition system may construct a pattern processing dictionary to extract a text token and recognize a citation and a speaker of the sentence based on the text token (first recognition method). At this time, the quotation / speaker recognition system can recognize the phrase from the point where the quotation mark starts in the sentence to the point where the quotation mark ends, as a quotation (S1204). Then, the quotation / speaker recognition system can determine a candidate noun phrase located at the beginning of the nominative search as a speaker candidate in the sentence (S1205).

For example, the quotation / speaker recognition system can extract the parts of speech information from the sentence using the morpheme analyzer, and recognize the citation and the speaker of the document based on the part of speech information (second recognition method). At this time, the quotation / speaker recognition system can recognize the phrase from the point where the quotation mark starts to the point where the quotation mark ends, as a quotation in the sentence (S1204). Then, the quotation / speaker recognition system can determine the candidate word positioned in the front part of the sentence in the sentence as the speaker candidate (S1205).

For example, when a plurality of quotations are included in a sentence, the quotation / speaker recognition system can extract the part of speech information from the sentence using the morpheme analyzer, and recognize the quotation and the speaker of the document based on the part of speech information Recognition method). Then, the quotation / speaker recognition system can tag the speaker on the quotation based on a predetermined sentence pattern indicating the relationship between the speaker and the quotation.

At this time, the quotation / speaker recognition system can recognize the phrase from the point where the quotation mark starts to the point where the quotation mark ends, as a quotation in the sentence (S1204). Then, the quotation / speaker recognition system can determine a candidate noun phrase located at the beginning of the sentence examination as a speaker candidate (S1205).

Then, the quotation / speaker recognition system can recognize the relationship between the speaker and the quotation according to the sentence pattern shown in FIG. 8, and match the speakers that have uttered the quotation among the speaker candidates.

For example, when a plurality of quotations are included in a sentence and a question keyword is included in the sentence / speaker recognition system, speech information is extracted from the sentence using a morpheme analyzer, and a quotation and a speaker of the document are recognized (Fourth recognition method). Particularly, the fourth recognition method can be applied to an exceptional situation (when a query keyword such as a question or a question is included in a sentence) when a plurality of quotations exist in a sentence.

Then, the quotation / speaker recognition system can tag the speaker on the quotation based on the question keyword included in the sentence, the preset sentence pattern indicating the relationship between the speaker and the quotation. However, when the sentence includes the question keyword, the relationship between the speaker and the quotation according to the sentence pattern shown in Fig. 8 can not be applied as it is. Specifically, the quotation / speaker recognition system can tag the speaker for the quotation according to the sentence pattern shown in FIG. 8 in consideration of the relationship between the quotation indicated by the question keyword and the remaining quotations and the speaker candidate.

At this time, the quotation / speaker recognition system can recognize the phrase from the point where the quotation mark starts in the sentence to the point where the quotation mark ends, as a quotation (S1204). Then, the quotation / speaker recognition system can determine a candidate noun phrase located at the beginning of the sentence examination as a speaker candidate (S1205).

If the speaker candidate is determined according to the first to fourth recognition methods described above, the quotation / speaker recognition system may analyze the sentence and select a speaker who has uttered the quotation among the speaker candidates (S1206).

In particular, when the sentence includes a plurality of quotations, the quotation / speaker recognition system may select a speaker who has uttered the quotation by matching the relationships between the plurality of quotations and the speaker according to a predetermined sentence pattern through the third recognition method have. If the sentence contains a plurality of quotations and the question keyword (question or question) is included, the quotation / speaker recognition system searches the plurality of quotations and the speaker Can be selected by matching the relationships between the words.

The parts not described in FIG. 12 can be referred to the description of FIG. 1 to FIG.

In addition, the citation / speaker recognition method according to an embodiment of the present invention may include a computer-readable medium including program instructions for performing various computer-implemented operations. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The media may be program instructions that are specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and perform program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions may include machine language code such as those generated by a compiler, as well as high-level language code that may be executed by a computer using an interpreter or the like.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, Modification is possible. Accordingly, the spirit of the present invention should be understood only in accordance with the following claims, and all equivalents or equivalent variations thereof are included in the scope of the present invention.

FIG. 1 is a block diagram illustrating an entire configuration of a citation / speaker recognition system according to an embodiment of the present invention. Referring to FIG.

2 is a diagram illustrating a first recognition method for recognizing a quotation / speaker according to an embodiment of the present invention.

FIG. 3 is a diagram illustrating an example in which a first recognition method is applied to a sentence according to an embodiment of the present invention.

4 is a diagram illustrating a second recognition method for recognizing a quotation / speaker according to an embodiment of the present invention.

5 is a diagram illustrating an example in which a second recognition method is applied to a sentence according to an embodiment of the present invention.

6 is a diagram illustrating a third recognition method for recognizing a quotation / speaker according to an embodiment of the present invention.

7 is a diagram illustrating an example in which a third recognition method is applied to a sentence according to an embodiment of the present invention.

8 is a table showing an example of tagging a speaker according to a sentence pattern of a speaker and a quotation according to an embodiment of the present invention.

9 is a diagram illustrating a fourth recognition method for recognizing a quotation / speaker according to an embodiment of the present invention.

10 is a diagram illustrating an example in which a fourth recognition method is applied to a sentence according to an embodiment of the present invention.

11 is a diagram illustrating an example of determining a combination of a citation and a speaker by combining a citation / speaker recognition method according to an embodiment of the present invention.

FIG. 12 is a flowchart illustrating an entire process of a citation / speaker recognition method according to an embodiment of the present invention.

Description of the Related Art

100: Quotations / Speaker Recognition System

101: Quote / Speaker Recognition

102: Speaker tagging part

103: Search result providing service

Claims (19)

A citation / speaker recognition unit for analyzing the document and recognizing the citation and the speaker; And A speaker tagging unit for tagging the recognized speaker with respect to the quotation; Lt; / RTI > Wherein the quotation / A citation recognition unit for extracting a sentence from the document and recognizing a citation in the sentence; A speaker candidate determining unit for analyzing the sentence to determine a speaker candidate of the quotation; A speaker selection unit for analyzing the sentence and selecting a speaker from among the speaker candidates that uttered the quotation; And A speaker recognition unit for recognizing a unique name of the selected speaker, Quot; < / RTI > The method according to claim 1, Wherein the quotation / Constructing a pattern processing dictionary to extract a text token, and recognizing the citation and the speaker of the sentence based on the text token. The method according to claim 1, Wherein the quotation / Extracts the part of speech information from the sentence using the morpheme analyzer, and recognizes the quotation and the speaker of the document based on the part of speech information. The method according to claim 1, Wherein the quotation / Extracting part of speech information from the sentence using a morpheme analyzer when the sentence includes a plurality of quotations and a speaker, recognizing a quotation and a speaker of the document based on the part of speech information, The speaker tagging unit, And a speaker is tagged with respect to the quotation based on a predetermined sentence pattern indicating a relationship between the speaker and the quotation. The method according to claim 1, Wherein the quotation / Extracting part of speech information from the sentence using a morpheme analyzer when the sentence contains a plurality of quotations and a speaker, recognizing a quotation and a speaker of the document based on the part of speech information, The speaker tagging unit, And a speaker is tagged with respect to the quotation based on a question keyword included in the sentence and a predetermined sentence pattern indicating a relationship between the speaker and the quotation. 6. The method according to any one of claims 2 to 5, The citation- Recognizes a phrase from a point at which a quotation mark starts to a point at which the quotation mark ends, as a quotation, Wherein the speaker candidate determining unit determines, Wherein the phrase in the sentence is determined as a candidate speaker. The method according to claim 1, Wherein the quotation / And a citation and a speaker are recognized from the sentence by combining preset citation and at least one recognition method for recognizing the speaker. The method according to claim 1, Wherein the quotation / Recognizing a citation and a speaker from the sentence by applying a preset citation and at least one recognition method for recognizing the speaker, And selecting a citation and a speaker according to the recognition method having the lowest error degree in consideration of the degree of error of the recognition method. The method according to claim 1, And a search result providing unit for searching a cited talk sentence of the speaker upon receiving a search request for the talker, / RTI > < / RTI > Analyzing the document to recognize a quotation and a speaker; And Wherein the citation / speaker recognition system tags the recognized speaker for the citation Lt; / RTI > Analyzing the document to recognize the citation and the speaker, Extracting a sentence from the document by the citation / speaker recognition system and recognizing a citation in the sentence; Analyzing the sentence to determine a speaker candidate of the quotation; Analyzing the sentence by the quotation / speaker recognition system to select a speaker who has uttered the quotation from among the speaker candidates; And Recognizing the unique name of the selected speaker by the quotation / speaker recognition system; / RTI > < / RTI > 11. The method of claim 10, Analyzing the document to recognize the citation and the speaker, Wherein the citation / speaker recognition system constructs a pattern processing dictionary to extract a text token, and recognizes the citation and the speaker of the sentence based on the text token. 11. The method of claim 10, Analyzing the document to recognize the citation and the speaker, The citation / speaker recognition system extracts parts of speech information from the sentence using a morphological analyzer, and recognizes the citation and the speaker of the document based on the part of speech information. 11. The method of claim 10, Analyzing the document to recognize the citation and the speaker, Wherein when the sentence includes a plurality of citation and a speaker, the citation / speaker recognition system extracts parts of speech information from the sentence using a morpheme analyzer, recognizes the citation and the speaker of the document based on the parts of speech information, Wherein the tagging of the recognized speaker for the quotation comprises: And the citation / speaker recognition system tags the citation on the citation based on a predetermined sentence pattern indicating the relationship between the speaker and the citation. 11. The method of claim 10, Analyzing the document to recognize the citation and the speaker, Wherein when the sentence includes a plurality of citation and a speaker, the citation / speaker recognition system extracts parts of speech information from the sentence using a morpheme analyzer, recognizes the citation and the speaker of the document based on the parts of speech information, Wherein the tagging of the recognized speaker comprises: Wherein the citation / speaker recognition system tags a speaker on the citation based on a query keyword included in the sentence and a preset sentence pattern indicating a relationship between the speaker and the citation. 15. The method according to any one of claims 11 to 14, The step of recognizing the quotation in the sentence comprises: Recognizing the phrase from the point at which the quotation mark begins at the quote mark / speaker recognition system to the point at which the quotation mark ends, as a quotation, The step of determining a speaker candidate of the quotation includes: Wherein the quotation / speaker recognition system determines, as a speaker candidate, a noun phrase located at the beginning of the nominal search in the sentence. 11. The method of claim 10, Wherein the recognizing of the quotation and the speaker comprises: Wherein the citation / speaker recognition system recognizes the citation and the speaker from the sentence by combining the preset citation and at least one recognition method for recognizing the speaker. 11. The method of claim 10, Wherein the recognizing of the quotation and the speaker comprises: Wherein the citation / speaker recognition system recognizes a citation and a speaker from the sentence by applying a preset citation and at least one recognition method recognizing the speaker, and considering the degree of error of the recognition method, the citation / And selecting a citation and a speaker according to a recognition method having the lowest degree of error. 11. The method of claim 10, Searching the citation / speaker recognition system for a search request for the speaker, and providing the search result as a search result; / RTI > < / RTI > A computer-readable recording medium on which a program for executing the method of any one of claims 10 to 14 or 16 to 18 is recorded.

Images