JP3470930B2 - Natural language analysis method and device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language analyzing method and apparatus, and more particularly, to realizing a process of extracting a word in a specific semantic field from an input character string and conversely masking it. In particular, the present invention relates to a natural language analysis method and apparatus for processing a character string expressing an address.

More specifically, the present invention relates to a natural language analysis method and apparatus for accurately describing an address in which building names, building numbers, floors, and room numbers are mixed.

[0003]

2. Description of the Related Art Conventionally, morphological analysis is performed on an input natural sentence, a particle is referred to by referring to a word dictionary and kanji, hiragana, katakana, English letters, numbers, etc. are recognized, and the meaning is determined without using a dictionary. There is a way to do it.

There are also character strings in which particles do not appear, such as addresses. An example of analyzing a conventional character string including numbers such as addresses (Japanese Patent Laid-Open No. 4-42354) is shown. FIG. 6 shows the configuration of a conventional address analysis system. The analysis system shown in the figure has an input unit 10 for inputting a character string including numbers such as addresses.
A character acquisition unit 20 that reads the input character string one by one, a character determination unit 30 that determines the character type for each character acquired by the character acquisition unit 20, and a number is included depending on the determined character type. The lot number data storage unit 40 stores lot number data.

[0005]

However, in the above-mentioned conventional method, in the case of an address, a particle does not appear,
The method of recognizing particles cannot be used. JP-A-4-423
The method shown in 54 is effective because the analysis up to the chome and lot number is not ambiguous, but it is effective, but when analyzing the building number, floor, and room number, there are the following problems.

(1) A building name including alphanumeric characters cannot be dealt with. Example a: In a building name such as "XX Building Part 2",
The room number is interpreted as "2". Example b: In the case of “Tsukiji No. 2 warehouse”, “No. 2” is interpreted as a room number, and it cannot be recognized that it is a unique name given to the building.

(2) It cannot handle the ambiguity of alphanumeric names. Example c: "B" in "B1-23" means underground, but "B" in "B1-23" is determined to be part of the room number. Example d: Originally, "123F" represents a room number, and "5F"
In the case of, it indicates the floor, but these are indistinguishable.

As in the example above, the name of the unique building, or
Alternatively, there is a problem that it is impossible to distinguish whether the room number or the number of floors is displayed. The present invention has been made in view of the above points, and provides a natural language analysis method and apparatus capable of solving the above conventional problems and accurately analyzing a character string that does not include a particle such as an address. It is an object of the present invention to provide a natural language analysis method and device capable of performing the same.

A further object of the present invention is to provide a natural language analysis method and apparatus capable of determining the meaning of an ambiguous alphanumeric string such as a building, floor, room number, etc., to an appropriate meaning. .

[0010]

FIG. 1 is a diagram for explaining the principle of the present invention. The present invention , the input address sentence
In the natural language analysis method that acquires the address element from the character string,
The morphological analysis unit shapes the input address character string.
Performs a morphological analysis, word division using a word dictionary, and
The word segmentation process and (step
1), multiple characters are consecutive in the word segmentation process
If it is a letter, consider it as a company name and
The post-processing process of adding and storing in the morphological analysis result storage unit
(Step 2), the unit word analysis unit stores the morphological analysis result
A clause containing the unit words related to the address read from the department,
Classify into meanings of address elements using the unit analysis rule
The unit analysis process (Step 3), the dependency analysis unit,
Sentences containing unit words read from the morphological analysis result storage unit
About the clause, using the dependency rule, is it a part of the building name?
Dependency analysis process that checks and analyzes compound words
(Step 5), the digit number / symbol analysis unit determines the result of the morpheme analysis.
For phrases that do not contain unit words read from the memory
Using the number of digits and symbol analysis rules, use only alphanumeric characters.
Number of digits / symbol analysis error that classifies the character string into the meaning of the address element
And (step 6).

The present invention also relates to the digit number / symbol analysis process.
The digit number / symbol analysis section is based on the result of morphological analysis.
Expressed by the number of alphanumeric digits or symbols including hyphens.
Role of the alphanumeric character by using the appearance position of the delimiter
To judge.

FIG . 2 is a block diagram showing the principle of the present invention. Starting
Ming is the one that gets the address element from the entered address string.
It is a natural language analysis device, and about the input address character string
Morphological analysis is performed and word division is performed using the word dictionary 204.
If multiple letters are consecutive English letters, enter the company name.
Morphological analysis result record
A morphological analysis unit 203 stored in the storage unit 201, and a morphological solution
Select the phrase to be analyzed from the analysis result storage unit 201
The phrase selection unit 205 to be read and the morphological analysis result storage unit 2
01 related to the address read by the phrase selection unit 205.
Use the unit word analysis rule 209 for the phrase containing the unit word
And the unit word analysis unit 206 that classifies the meanings of address elements
From the morphological analysis result storage unit 201 to the phrase selection unit 205
About the clause including the unit word read by
Use the rule 210 to check if it is part of the building name,
Dependency analysis unit 207 for analyzing compound words, and morphological solution
Read from the analysis result storage unit 201 by the phrase selection unit 205
Digit / symbol analysis for clauses that do not include the specified unity
Use rule 211 to save a string consisting of only alphanumeric characters.
The number of digits / symbol analysis unit 208 for classifying the meanings of the place elements
Have.

Further, the digit number / symbol analysis unit is a morpheme.
The number of alphanumeric digits or hyphens based on the analysis result
Using the position of the delimiter represented by a symbol containing
Means for determining the role of alphanumeric characters.

According to the present invention, as described above, in order to correctly extract the information of each meaning from the data in which the building name, the building number, the floor, and the room number are mixed, the morpheme analysis of the entire character string is performed and the input is performed. It is necessary to find the meaning of the words that make up the data. The result is used to determine the meaning of the alphanumeric characters.

First, the meaning of a bunsetsu having a unit word is determined. Next, the dependency relation is analyzed, and it is analyzed whether or not the bunsetsu with the unitary word becomes a part of the compound word. Finally, it can be judged from the meanings of the nouns before and after the bunsetsu without unitary words, the meanings of bunsetsus with unitary words, and the presence of symbols such as hyphens.

Therefore, the above-mentioned problem that the building name including alphanumeric characters cannot be dealt with can be dealt with by determining the meaning of the number from the meaning of the surrounding words. In the case of the above-mentioned example a, it can be determined by the meaning that the meaning of "part" is a unit word preceding a number, and in the case of example b, the word "warehouse" follows the word. In addition, the above-mentioned problem that cannot be dealt with in the ambiguous meaning of the alphanumeric name can be judged by the number of digits of the number given to the alphabet. Usually "floor"
Is within 2 digits, and the "room number" can have 1 to 4 digits, but it can be determined by using the knowledge that there are many 3 to 4 digits.

Thus, according to the present invention, the meaning of a bunsetsu is determined by the meaning given to each word and the unit word, the meanings of the preceding and following character strings are analyzed, and the bunsetsu having the unit word can be a compound word. To determine. Furthermore, for clauses that do not have unit words, the meaning of the string is determined based on the number of alphanumeric digits and the type of symbol, so that the alphanumeric string is part of the building name or "floor", or Can accurately determine whether it is a "room number".

It is possible to accurately determine whether it is a “floor” or a “room number”.

[0019]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 3 shows the configuration of the address analysis system of the present invention. The system shown in the figure includes a morphological analysis result storage unit 201, an analysis control unit 202, and a morphological analysis unit 20.
3, word dictionary 204, phrase selection unit 205, unit word analysis unit 206, dependency analysis unit 207, digit number / symbol analysis unit 20
8, unit word analysis rule 209, dependency rule 210,
And the number of digits / symbol analysis rule 211.

The analysis control unit 202 controls each component of the morpheme analysis result storage unit 201, the morpheme analysis unit 203, and the phrase selection unit 205. The morpheme analysis unit 203 refers to the word dictionary 204, divides the input natural language character string into words, and performs morpheme analysis. In the word dictionary 204, “building” is added as a meaning that constitutes an address in addition to general registered words.
It is assumed that words such as “building”, “floor”, and “room number” are registered. As a result, the morpheme analysis unit 203 gives the meanings of “building”, “floor”, and “room number” to the alphanumeric characters having the meanings of “building”, “floor”, and “room number”, respectively, as the unit words. If there are two or more consecutive alphabetic characters in the word division, the alphabetic character is the “company name”.

The morpheme analysis result storage unit 201 holds the result analyzed by the morpheme analysis unit 203, and the analysis control unit 20.
It is read out to the phrase selection unit 205 via 2. The phrase selecting unit 205 selects and reads a phrase to be analyzed from the morphological analysis result storage unit 201. The read clauses are transferred to the unit word analysis unit 206, the dependency analysis unit 208, and the digit number / symbol analysis unit 208, respectively.

The unit word analysis unit 206 refers to the unit word analysis rule 209 to determine the meaning of the phrase including the alphanumeric character of the phrase including the unit word. At this time, the unit word analysis rule 20
With reference to 9, determine the meaning of a number or a single letter.
Further, when the number and one alphabetic character are consecutive and there are two or more semantic candidates for the character, the number of digits / symbol analysis unit 208 is input to select the candidate.

The dependency analysis unit 207 refers to the dependency rule 210 and analyzes the compound word based on the dependency relationship with the preceding and following words. That is, the existence of a word that has a meaning related to the input word in the word before the input word or in the subsequent word is confirmed, and if the words before and after the word give the input word meaning. , The meaning is given to the input word as the dependency relationship is established.

The digit number / symbol analysis unit 208 determines the digit number / symbol rule 21 for an alphanumeric string to which no unit words are added.
With reference to 1, the meaning is analyzed from the positional relationship of symbols such as the number of digits and hyphens. At this time, when there are a plurality of candidates in the unit word analysis unit 206, one of the candidates is selected.

The unit word analysis rule 209 is "building".
This is a rule for determining the meaning of a phrase including unit words such as “building”, “floor”, and “room number”. Dependency analysis rule 2
Reference numeral 10 is a rule for determining whether a phrase with a unitary word can form a part of a compound word.

The digit number / symbol analysis rule 211 determines the meaning of a noun before and after a phrase without a unitary phrase, the meaning of a phrase having a unitary word, the presence or absence of a symbol such as a hyphen, and the meaning according to the number of digits of a number. Is a rule to add. Regarding the number of digits, the rule is to add the meaning of “floor number” if it is within 2 digits and “room number” if it is 3 to 4 digits.

FIG. 4 is a flow chart showing the operation of the address analysis system of the present invention. Step 101) First, as a word division process, the morpheme analysis unit 203 divides the input address character string into words by referring to the analysis word dictionary 204. Step 102) In the word segmentation process, the morphological analysis unit 203 uses a rule that is registered in advance as “analyzing two or more letters as a corporate name” as an analysis process for a corporate name in addition to normal morphological analysis. When two or more consecutive alphabetic characters are input using, the meaning is given as the company name. As a result, two or more letters can be
They are not regarded as "floor" or "room number".

Step 103) Next, as a unit word analysis process, the phrase selecting unit 205 selects a phrase including a unit word. Here, when there is a bunsetsu containing a unit word by referring to the unit word analysis rule 209, the meaning of a numeral or one letter is determined by referring to the meaning given to the word. For example, since "No. 2 building" has a unit word "gokan", the whole is classified into the meaning of "building". Since "2nd floor" has "floor", it is classified as "floor". However, "2F" here means two kinds of solution candidates, meaning "floor" and alphanumeric string "2F".

Step 104) Dependency analysis unit 207
Refers to the dependency rule 210, analyzes the dependency relationship with the preceding and following words, and analyzes the compound word. For example, there are words that have the meaning of a building such as "warehouse", "complex", and "dormitory" in the latter words, so in the case of "No. 2 warehouse",
“No. 2” does not mean “room number”, but includes the following words as “building”. It also analyzes nouns related to positions such as "underground".

Step 105) Finally, as a digit number / symbol analysis process, the phrase selecting unit 205 determines the meaning of the alphanumeric characters to which no unit words are added. When there is a phrase whose meaning is determined in step 102 in the preceding and following phrases, the meaning is changed to another meaning.

[0031]

EXAMPLES Examples of the present invention will be described. FIG. 5 is a diagram for explaining the operation of the embodiment of the present invention. Example 1 below
~ Each process of Example 4 will be described.

[Example 1] Word division process (step 101) The input character string "A ridge first floor 123" is input, and the morpheme analysis unit 203 divides it into words as follows.

A / building / 1 / floor / 123 "A" is an alphabetic character, "building" is a unit word, "1" is a number, "floor"
Is a unit word, and "123" is a number. Post-Processing Process (Step 102) The morpheme analysis unit 203 does not consider the company name because the letter “A” is one character. Therefore, the meaning of the letter "A" is "building". Also, the meaning of the number “1” is the meaning of “floor”, and the number “123” is simply determined as a number because no unit word is given after that.

Unit-Word Analysis Process (Step 103) Next, the unit-word analysis unit 206 includes “A” including a unit word.
Since "ridge" and "first floor" have "ridge" and "floor" respectively,
Each meaning is given. Dependency Analysis Process (Step 104) Next, the dependency analysis unit 207 determines whether the meaning of the building is a character string. In the case of this example, the words "warehouse", "complex", "dormitory", etc., which represent buildings, are not included, so the process proceeds to the next step.

Digit number / symbol analysis process (step 10
5) The phrase selection unit 205 displays the numerical value “123” without a unit word.
Therefore, the dependency relationship for the number “123” is determined. With reference to the number-of-digits / symbol analysis rule 211, since there is a clause meaning “floor” immediately before “123” and the number of digits is three, it is analyzed as a “room number”.

As described above, in Example 1, since the unit word "building" is placed after the alphabetic character "A" represented by one character, it is not judged as a company name but "A building". Also, since there is a unit word "floor" after the number "1", it is referred to as "1st floor". The last number "123" is determined as the number of digits and is 3 digits, so it is determined as "room number".

[Example 2] Word division process (step 101) The input character string "No. 2 warehouse" is input, and the morphological analysis unit 20
In 3, the words are divided as follows.

2 / No. / Warehouse "2" is a number, "No." is a unit word, and "Warehouse" is a building. Post-Processing Process (Step 102) Since the character string in this example does not include an alphabetic character, the analysis result of is transferred to the next process.

Unit-Word Analysis Process (Step 103) Next, in the unit-word analysis unit 206, since there is a unit-word “go”, it is analyzed into the meaning of “2”, which represents a room number. Dependency analysis process (step 104) Next, since there is a word having a building meaning "warehouse" after "No. 2", the name of the warehouse is called "No. 2 warehouse" instead of "room number". .

Digit number / symbol analysis process (step 10
5) In this example, there are no alphanumeric characters without unit words, so
Skip processing. Regarding Example 2 above, since the unit word “go” is after the number “2”, the analysis result is determined to be the room number “2”, but further, since there is the “warehouse” that represents the building later, It is judged as "No. 2 Warehouse" that represents the building.

[Example 3] Word division process (step 101) The input character string "ABC building basement 1st floor" is input, and the morphological analysis unit 203 divides it into words as follows.

ABC / Building / Underground / First floor "ABC" is an alphabet, "Building" is a building, "Underground" is a position, "1" is a number, and "Floor" is a unit word. Post-Processing Process (Step 102) Since the character string in this example includes the alphabetic character "ABC" and this alphabetic character is 3 characters, it is regarded as a company name.

Unit Particle Analysis Process (Step 103) Next, in the unit word analysis unit 206, since there is a unit word “floor”, it is analyzed into a meaning representing the floor number “first floor”. Dependency Analysis Process (Step 104) Next, the dependency analysis unit 207 changes the company name “ABC” to the name of the building because there is a word “building” meaning “building” after “ABC”. ABC building ",
Since there is a word "underground" that indicates the location, "ABC building"
It will be the "1st basement floor".

Digit number / symbol analysis process (Step 10
5) In this example, there are no alphanumeric characters without unit words, so
Skip processing. It is analyzed as an issue. Example 3 above
In this case, since there are two or more alphabetic characters "ABC", it is judged as the company name, and since "ABC" is followed by "Building" which indicates a building, it is judged as "ABC Building", and then " Since there is “underground”, and the unit number “floor” follows the subsequent number “1”, the dependency relation analysis unit 207 determines that it is “underground first floor”.

[Example 4] Word division process (step 101) The input character string "ABC building B1F" is input, and the morpheme analysis unit 203 divides it into words as follows.

ABC / Building / B / 1 / F "ABC" is a letter, "Building" is a building, "B" is a letter, "1" is a number, and "F" is a letter and a unit word representing a floor. is there. Post-Processing Process (Step 102) Since the character string in this example includes the alphabetic character "ABC" and this alphabetic character is 3 characters, it is regarded as a company name.

Unit word analysis process (step 103) Next, in the unit word analysis unit 206, since there is an English unit word “F”, in this example, the meaning of “1st floor” and “1st floor”
Two solution candidates of "F" are left. Dependency Analysis Process (Step 104) Next, since the dependency analysis unit 207 has a word “building” meaning “building” after “ABC”, “AB
Change "C" from the company name to the building name, and call it "ABC Building".

Digit number / symbol analysis process (Step 10
5) In this example 4, there are two solution candidates in the above processing. Therefore, since there is an alphabetic character "B" in front of the one-digit number "1", it is assumed that "B" has the meaning of "underground".
It is analyzed that the subsequent "1F" represents the floor number.

As described above, the meanings of the words forming the input data are registered in the word dictionary 204, and the address is analyzed by the unit word analysis rule 209, the dependency analysis rule 210, and the digit number / symbol analysis rule 211, respectively. By pre-registering the knowledge to do so, each analysis unit determines the meaning of alphanumeric characters after word division, and further determines the meaning of the bunsetsu for the bunsetsu with a unit, and the dependency between the preceding and following words By analyzing the relationship, the meaning of bunsetsu with unit words,
It can be judged by the presence or absence of symbols such as hyphens and the positional relationship.

In the above embodiment, the unit word analysis unit 2
06, dependency analysis unit 207 and digit / symbol analysis unit 20
In FIG. 8, the analysis is performed by referring to each rule, but the present invention is not limited to this example, and it is also possible to store a rule in each analysis unit and perform various analyzes by referring to the rule. is there.

Further, in the above embodiment, the processing when the character string relating to the address is input as the natural language has been described, but the present invention is not limited to this example, and a character having a special form in the analysis of the natural language is used. Various input can be applied to input of strings, and classification and rules corresponding to character strings having a special form can be set in advance and analysis can be performed based on the correspondence.

Further, the analysis result according to the above-described embodiment is stored in the storage means, and at the time of work such as address editing, a certain editing standard is set because the meaning by analysis has already been completed, and unnecessary description is made. It is also possible to process data such as omission. The present invention is not limited to the above embodiments, and various modifications and applications are possible within the scope of the claims.

[0053]

As described above, according to the natural language analysis method and apparatus of the present invention, it is possible to determine the meaning of the ambiguous alphanumeric string of the ridge / floor / room number to a more appropriate meaning. is there. Therefore, according to the present invention, only a building name is extracted from an address in which a building name, a building number, a floor, and a room number are mixed, a database is created, and it is used for creating data of a business such as converting a building name into a formal address. , In order to protect privacy while keeping the data required to distinguish the same surname of individual names, the building name is omitted and it is used to create data such as displaying only the floor / room number, or a telephone notebook. Since the address is stored in a limited space like the above, it is possible to apply to the business of extracting only the floor / room number by omitting the building name.

[Brief description of drawings]

FIG. 1 is a diagram for explaining the principle of the present invention.

FIG. 2 is a principle configuration diagram of the present invention.

FIG. 3 is a block diagram of an address analysis system of the present invention.

FIG. 4 is a flowchart showing the operation of the address analysis system of the present invention.

FIG. 5 is a diagram for explaining the operation of the embodiment of the present invention.

FIG. 6 is a block diagram of a conventional address analysis system.

[Explanation of symbols]

201 Morphological analysis result storage unit 202 Analysis control unit 203 Morphological analysis unit 204 word dictionary 205 clause selection section 206 Unit-word analysis part 207 Dependency analysis unit 208 Digits / Symbol Analysis Department209 Unit analysis rules 210 Dependency Analysis Rule 211 Digits / Symbol analysis rules

Front Page Continuation (56) Bibliography The Institute of Electronics, Information and Communication Engineers, April 25, 1987, J70-D, Vol. 4, p. 832-835 49th Annual Meeting Lecture Proceedings (3), September 30, 1994, p. 3-261, 3-262 (58) Fields investigated (Int.Cl. ⁷ , DB name) G06F 17/27 G06F 17/21 550 JISST file ( JOIS)

Claims (4)

(57) [Claims] 1. A natural language analysis method for obtaining an address element from an input address character string , wherein a morpheme analysis unit forms a shape for the input address character string.
Performs a morphological analysis, word division using a word dictionary, and
The word segmentation process stored in the morphological analysis result storage unit and the morphological analysis unit are alphabetic characters in which a plurality of characters are continuous.
If it is regarded as a company name and given meaning as a company name,
The post-processing process stored in the morpheme analysis result storage unit and the unity word analysis unit read from the morpheme analysis result storage unit.
The phrase that contains the unit words related to the
Unit analysis parsing that classifies the meanings of address elements using
And the dependency analysis unit reads it from the morphological analysis result storage unit.
For the clause that includes the issued unit phrase, set the dependency rule
Check if part of the building name, and analyze the compound word.
The dependency analysis process and the digit number / symbol analysis unit read from the morphological analysis result storage unit.
Number of digits / symbols for bunsetsu that does not include the detected unit
Using the parsing rules, enter a string of alphanumeric characters
It consists of the number of digits and the symbol analysis process, which are classified into the meanings of elements.
A natural language analysis method characterized by the above. 2. In the digit number / symbol analysis process,
The number of digits / symbol analysis unit is based on the result of the morphological analysis.
The number of alphanumeric digits or a symbol including a hyphen.
Using the appearance position of the represented delimiter,
The natural language analysis method according to claim 1, wherein a role is determined . 3. An address element is extracted from the input address character string.
A natural language analyzer for obtaining, performing morphological analysis on the input address character string,
Morphological analysis results are stored by dividing words using a word dictionary.
It is an alphabetic character that stores multiple characters and has multiple consecutive characters.
In the case, it is regarded as a company name, a meaning is given as a company name, and
The morpheme analysis unit to be stored in the morpheme analysis result storage unit and the phrase to be analyzed are selected from the morpheme analysis result storage unit.
The phrase selection unit that selects and reads, and the phrase selection unit that reads from the morphological analysis result storage unit.
The phrase containing the unit word related to the found address is analyzed as a unit word.
A unit word solution that classifies address element meanings using analysis rules
Read by the analysis unit and the phrase selection unit from the morphological analysis result storage unit.
Dependency rules for bunsetsu that contains the unit word found
Check if a part of the building name using
The dependency analysis unit that performs the reading and the phrase selection unit read from the morphological analysis result storage unit.
Number of digits / symbols for bunsetsu that does not include the detected unit
Using the parsing rules, enter a string of alphanumeric characters
It has a digit number / symbol analysis section that classifies elements into meanings.
Natural language analyzer characterized by. 4. The digit number / symbol analysis unit , based on the result of the morphological analysis, the digit number of the alphanumeric characters,
Or, for a delimiter represented by a symbol that includes a hyphen
It includes means for determining the role of the alphanumeric characters by using the appearance position.
4. The natural language analysis device according to claim 3 .