JPH04673A - Method and device for registering compound word - Google Patents

Abstract

PURPOSE:To properly extract a compound word from a text and to register it in a dictionary by possessing a compound word candidate extracting step, a compound word candidate presentation step and an inclusive relation presentation step. CONSTITUTION:A compound word candidate extracting means 6 to extract a compound word candidate based on the dictionary information of a word dictionary 5, a compound candidate file 2 to store an extracted compound word candidate and a list generating means 9 able to grasp the mutual inclusive relation of the compound word candidate of the compound word candidate file 2 are provided. A meaningless word string is removed and the compound word candidate is obtained by executing the deletion of the word string corresponding to necessity from a text 1. Besides, the inclusive relation list of a form able to grasp the inclusive relation between the compound word candidates easily is generated and is displayed. Thus, a compound word can be extracted properly from the text 1 and can be registered in a dictionary.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a collocation registration method and device, and more particularly to a collocation registration method and device for extracting collocations from a text and registering them in a dictionary.

[Prior art] The processing accuracy of natural language processing systems (e.g. natural language question answering systems and machine translation systems) is highly dependent on the accuracy of dictionaries, and it is necessary to register all technical terms used in the target field in the dictionary. There is a need to.

However, technical terms do not remain constant (for example, technical terms in the field of science and technology increase with technological innovation).
, it is necessary to add additional information to the dictionary many times while operating the system.

By the way, since technical terms are often composed of collocations, it is important to efficiently register the collocations in a dictionary.

The conventional technology for registering such collocations is to first mechanically extract a word string from a text, then delete words with a low frequency of occurrence, and use the remaining word strings as collocations.
[Automatic extraction of idiomatic expressions from a text data base/Disclosed in Information Processing Society of Japan 37th National Conference Papers 7B-6P1032-P, 1033J.

In addition, a technology that first mechanically extracts word strings from text, then deletes duplicate word strings based on the frequency of occurrence of the word strings, and converts the remaining words into collocations is disclosed in Japanese Patent Application Laid-Open No. 1-10267.
It is disclosed in No. 9.

Furthermore, the KWIC list for collocations is displayed, and the user is prompted to input information about the collocations using the KWIC list as a guide.
A technique for updating a dictionary is disclosed in Japanese Patent Laid-Open No. 63261467.

[Problems to be Solved by the Invention] Among the above-mentioned conventional techniques, in which a part of a word string is deleted based on the frequency of occurrence of the word string and the remaining words are made into collocations, some of the collocations may not be considered as collocations. There is a problem that the appropriate word string remains.

On the other hand, in the conventional technology that displays a KWIC list for collocations, the user can judge the suitability of the collocation by looking at the display, but sentences containing the word string are displayed for the number of times the word string appears. , there is a problem that the amount of the list becomes enormous and the burden on the user increases.

SUMMARY OF THE INVENTION Therefore, an object of the present invention is to provide a collocation registration method and device that can appropriately extract collocations from text and register them in a dictionary, and that can reduce the burden on the user.

[Means for Solving the Problems] The present invention includes a collocation candidate extraction step of extracting collocation candidates from a given text based on a predetermined collocation candidate extraction rule, and presenting each of the extracted collocation candidates to a user; A collocation candidate presentation step in which one collocation candidate is specified; and an inclusion relationship presentation step in which other collocation candidates that have an inclusive relationship with the collocation candidate specified by the user are selected and these collocation candidates are presented to the user in a format that allows the inclusion relationship to be understood. and has the user input predetermined information about the collocation candidate specified by the user,
The present invention provides a collocation registration method further comprising a dictionary registration step of registering the collocation candidate as a collocation in a dictionary together with the information.

Furthermore, the present invention provides a collocation registration device that suitably implements the collocation registration method described above.

[Operation] In the collocation registration method of the present invention, in the collocation candidate extraction step, a word string is extracted from the text, and if necessary, the word string is deleted based on the frequency of occurrence or by morphological analysis/syntax analysis. By doing this, we remove as many meaningless word strings as possible and obtain collocation candidates. Furthermore, in the inclusion relationship presentation step, an inclusion relationship list is generated and displayed in a format that allows the inclusion relationships between compound word candidates to be easily grasped.

In this way, since the inclusion relationship list is displayed in a format that allows the inclusion relationships between compound word candidates to be easily grasped, the user can easily eliminate meaningless word strings. Therefore, it becomes possible to appropriately extract collocations from the text and register them in the dictionary.

[Embodiments] Hereinafter, embodiments of the present invention will be described in detail based on the drawings. Note that the present invention is not limited thereby.

FIG. 1 is a block diagram of a collocation dictionary maintenance bag W50 according to an embodiment of the present invention.

This collocation dictionary maintenance device 50 stores a text file 1 for storing English text, a text editing means 7 for editing the English text of the text file 1, and rules for extracting collocation candidates from the English text. A collocation candidate extraction rule file, a word dictionary 5 storing word information, and collocation candidates are extracted from the English text of the text file 1 based on the collocation candidate extraction rules of the collocation candidate extraction rule file 3 and dictionary information of the word dictionary 5. A collocation candidate extracting means 6 that extracts collocation candidates, a collocation candidate file 2 that stores the extracted collocation candidates, and a list in a format that allows understanding the mutual inclusion relationship of collocation candidates in the collocation candidate file 2, and collocation candidates appearing. KWI so that you can understand the context
List generation means 9 for generating a C list; dictionary maintenance means 8 for generating dictionary information based on information such as translations set by the user for collocation candidates; and collocation dictionary 4 for storing the dictionary information created in 1. Display means 12 for presenting the collocation candidates, the list, and the dictionary information to the user; input means 11 for the user to input data; and the text editing means 7, the list generation means 9, and the dictionary maintenance. It is provided with a control means 10 which controls the means 8 and also controls data input/output.

FIG. 2 is a flowchart of the operation of the collocation dictionary maintenance device 50.

In step 101, the control means 10 requests the user to input the values of the parameters. The user is
As parameters, the specification information of the English text from which collocation candidates are extracted, the number of words of collocation candidates to be extracted,
Appearance frequency threshold.

Input information such as specifying the processing method for extracting compound word candidates.

In step 102, the collocation candidate extraction means 6 extracts collocation candidates from the English text based on the parameters input by the user, and stores the extracted collocation candidates in the collocation candidate file 2. A flowchart of this compound word candidate extraction process is shown in FIG.

That is, in step 301 in FIG. 3, the control means 1o sets the value of the parameter input by the user in step 101 in FIG. 1 to the collocation extraction means 6.

In step 302, the collocation candidate extraction means 6 extracts a word string that meets the specified conditions from the specified English text using the specified extraction processing method.

Extraction processing methods include (1) a method that extracts all word strings that satisfy a specified word count condition, and (2) a method that extracts only word strings that match a predetermined extraction pattern.

The former is shown in Figure 4 (1) (however, if the specified number of words is
). The latter is shown in FIG. 4 (2) (however, when the extraction pattern is a noun phrase).

Although only one part of speech is assigned to each word in FIG. 4(2), multiple parts of speech may be assigned and all noun phrases formed in that case may be extracted. Also,
In Figure 4 (2), the extraction pattern is fixed to noun phrases, considering the characteristic that technical terms often have noun phrases.
The user may specify an extraction pattern as one of the parameters.

In step 303, meaningless word strings are deleted from the extracted word strings. For example, in Figure 4 (1), word strings that are inappropriate as compound word candidates (for example, It can be removed by setting it to , and applying that rule. For example, if you set a rule that ``articles (or words "a,""the," or "an") cannot be at the right end of a word string,
■ in Figure 4 (1) can be removed. Also, it is checked whether the word string is already registered in the collocation dictionary 4, and if it is already registered, it is deleted.

In step 304, the collocation candidate extraction means 6 sorts the extracted word strings, deletes duplicate word strings, and calculates the frequency of appearance of each word string. At this time, for a word string that includes an inflected form, if it can be generalized as a pattern, it is generalized and the appearance frequency is calculated.

For example, structural ambiguity
j and rstruc-tural ambigui
In the case of two word strings in tiesJ, the latter plural type is generalized to the singular former type, and the frequency of occurrence is summed. When presenting this type of word string to the user, it is presented in a form that allows the user to recognize which word is a generalized conjugation form. On the other hand, for example, rvending machine J
Extract the word string , consider the conjugation form, and generalize it.
As a bend machine J, fvend m
If the word string achineJ does not exist, the original rvending machineJ is used as it is as the word string.

After calculating the appearance frequency, word strings whose appearance frequency is less than the specified appearance frequency threshold are deleted.

In step 305, the collocation candidate extraction means 6 writes the remaining word string as a collocation candidate into the collocation candidate file 2 together with its appearance frequency.

Through the above steps, collocation candidates are extracted from the English text,
The words are stored in the collocation candidate file 2.

In step 103, the control means 10 displays the collocation candidates and other information on the display means 12. FIG. 5 shows an example of the display screen, and 1202 is a display window for compound word candidates. Further, 1201 is a menu window for starting various processes. In FIG. 5, the collocation candidates are displayed in descending order of frequency of occurrence in the English text, but it may also be possible to select to display them in another order (for example, ABC order) in the menu window 1201. It is possible.

In step 104, the user displays the display window 120
2, and check whether there is a collocation candidate of interest.

At this time, by using the scrolling functions "↑" and "↓" of the menu window 1201, it is possible to refer to compound word candidates that cannot be displayed in the window 1202. If there is no target collocation candidate, the collocation extraction process ends. If there is a compound word candidate of interest, the process advances to step 105.

In step 105, the user inputs the input means 1 such as a mouse.
1 to specify the collocation candidate of interest.

In step 106, the user selects menu window 1
201: "Inclusion relationship J rKWIC J r dictionary registration" or "Complete".

If "inclusion relationship" is selected, the process advances to step 107. r
If KWIcJ is selected, the process proceeds to step 108. If "Dictionary Registration" is selected, the process advances to step 109. If "Complete" is selected, the process returns to step 104.

In step 107, the list generating means 9 extracts other compound word candidates that have an inclusive relationship with the compound compound candidate specified in step 105 from the compound word candidate file 2, aligns the positions of common word strings, and creates a list. An inclusion relationship list is generated and displayed on the display device 12. FIG. 6 shows an example of the display screen, and 1203 is a window that displays the inclusion relationship list. Note that FIG. 6 shows a case where the collocation candidate rLINE FEEDJ is specified in step 105, and the collocation candidate rLINE FEEDJ
Other collocation candidates including rLTNE FEED MODE
, J, rLINE FEEDKEYJ, r
PREssING LINE FEEDJ, r
PREssrNGLINE FEED KEYJ etc. are extracted from the collocation candidate file 2 and the common word string [
Align LINE FEEDI and open window 1
It is displayed in 203.

The user can easily find meaningless word strings by referring to collocation candidate series that have an inclusive relationship with the collocation candidate of interest. For example, the collocation candidate r in Figure 5
PREssING LINE FEEDJ is clearly rPREssING LINE from its appearance frequency.
Because it only appears as a part of FEED KEYJ,
It turns out that it has no meaning as an independent combination.

Note that if the appearance frequency of compound word candidates is displayed as a value after subtracting the overlap, meaningless word strings can be found even more easily.

In step 108, the list generating means 9 edits the sentence including the collocation candidate specified in step 105 and creates a KWI.
A C list is generated and displayed on the display means 12. FIG. 7 is an example of the display screen, and 1204 is the KW
This is a window that displays an IC list. In addition, KWI
The C list is a list of sentences in which the positions of the collocation candidates of interest are aligned for easy viewing, and it is possible to easily grasp in what context the collocation candidates of interest appear.

In step 109, the dictionary maintenance means 8 requests the user to input dictionary information necessary for registering the collocation candidate specified in step 105 in the collocation dictionary 4, and when the user inputs the collocation candidate, the collocation candidate is registered in the collocation dictionary 4. . A window 1205 shown in FIG. 8 is a window that requests the user to input dictionary information, and is a window that requests the user to input dictionary information.
EEDJ is displayed in the technical term column. The user is
Headword.

Enter in the part of speech, meaning code, and translation fields. For example, enter rFEEDJ in the headword field, enter rNJ representing a noun in the part of speech field, enter rBJ representing action in the meaning code field, and enter "line break" in the translation field. input.

Note that the dictionary maintenance means 8 analyzes the syntactical and
Determine the central word (head) from the semantic pattern, set the central word as a provisional value in the headword column, and estimate the part of speech, meaning code, and translation value based on the dictionary information of the central word, Automatically set in each column as a provisional value. Therefore, in many cases, the user only needs to make an input to confirm the automatically set value, which reduces the user's input effort.

Examples of the method for determining the central word include a method in which the last word of a collocation candidate or the word before a prepositional phrase is used as the central word.

The user can change the headword while the provisional value is displayed in the window 1205.

When a headword is changed, the dictionary maintenance means 8 changes the part of speech and meaning code based on the dictionary information of the headword.

Automatically set provisional values for translated words.

Further, after the user confirms the value displayed in the window 1205, the user selects the entry word 1 part of speech.

Semantic codes and translations can be changed individually.

As explained above, regarding the collocation candidates extracted by the system, the user can check their frequency of occurrence, inclusion relationships with other collocation candidates, context in the English text, etc. using the inclusion relationship list and KWW.
By checking the IC list, you can determine whether or not it is a collocation that should be registered. If it is determined that the compound word should be registered, the information can be easily set and registered in the compound word dictionary 4.

In another embodiment of the invention, step 302 of FIG.
Alternatively, in step 304, it is checked whether the extracted word string has already been registered in the collocation dictionary 4, and the registered word string is automatically deleted. Alternatively, instead of automatically deleting the registered word string, the word string may be displayed on the window 1202 of the display device 12 by changing the display color so that it can be seen that it has been registered, and the user can delete it. In any case, unnecessary dictionary registration processing can be avoided.

In addition, as other embodiments, the above-mentioned FIGS. 5, 6, and 7
As shown in the figure, a “containment relationship” window 1203,
An example is one in which the rKWI CJ window 1204 and the "Dictionary Registration" window 1205 are not displayed selectively, but these windows 1203, 1204, and 1205 are displayed simultaneously in an overlapping manner.

In addition, as yet another embodiment, step 1 in FIG.
05, the user not only specifies a desired collocation candidate from among the collocation candidates displayed in the collocation candidate display window 1202, but also selects the "inclusion relation" window 1203, r
For example, the word strings displayed in the KWIcJ window 1204 or the ``Dictionary Registration'' window 1205 can also be specified as collocation candidates. In addition, the window (
One example is a system in which word strings displayed in (not shown) can also be specified as compound word candidates.

Still another embodiment is an apparatus to which the present invention is applied as a device for maintaining a translation dictionary for languages other than English and Japanese.

Another example is an apparatus to which the present invention is applied as a device for maintaining a collocation dictionary used in a natural language processing system that performs processing other than translation (for example, database search).

[Effects of the Invention] According to the collocation registration method and device of the present invention, collocations extracted from text are created in a format that allows the user to easily recognize inappropriate collocation candidates that cannot be automatically removed. Since candidates are presented, it becomes possible to register appropriate collocations in the dictionary without placing a large burden on the user.

[Brief explanation of drawings]

FIG. 1 is a block diagram of a collocation dictionary maintenance device according to an embodiment of the present invention, FIG. 2 is a flowchart of processing of the collocation dictionary maintenance device of FIG. 1, FIG. 3 is a flowchart of collocation candidate extraction processing, and FIG. 4 (1) and (2) are explanatory diagrams of the collocation candidate extraction method,
Figure 5 is an example of a screen that presents collocation candidates, Figure 6 is an example of a screen that presents an inclusion relationship list, and Figure 7 is a KWI
FIG. 8 is an example diagram of a screen that presents the C list, and FIG. 8 is an example diagram of a screen that presents dictionary information. (Explanation of symbols) 50... Collocation dictionary maintenance device 1... Text file, 2... Collocation candidate file, 3... Collocation candidate extraction rule file, 4... Collocation dictionary, 5... Word Dictionary, 6... Collocation candidate extraction means, 7... Text editing means, 8... Dictionary maintenance means, 9... List generation means, 10... Control means, 11... Input means, 12 ...Display means. Figure 1

Claims (1)

[Claims] 1. A collocation candidate extraction step of extracting collocation candidates from a given text based on a predetermined collocation candidate extraction rule, and presenting each of the extracted collocation candidates to the user and selecting one collocation candidate. a step of presenting a collocation candidate that the user specifies; a step of presenting a collocation relationship that selects other collocation candidates that have an inclusive relationship with the collocation candidate specified by the user and presenting these collocation candidates to the user in a format that allows the user to specify the inclusive relationship; A collocation registration method comprising the steps of: having a user input predetermined information about the collocation candidate, and registering the collocation candidate as a collocation in a dictionary together with the information. 2. A KW that extracts sentences containing collocation candidates specified by the user from the text and presents them to the user in the form of a KWIC list.
2. The collocation registration method according to claim 1, further comprising the step of presenting an IC. 3. In the dictionary registration step, the central word is estimated according to the pattern of collocation candidates specified by the user, the dictionary information of the central word is extracted from the dictionary, and it is presented to the user as a provisional value of the dictionary information of the collocation candidate. Claim 1 or Claim 2, wherein the user is made to input information as to whether the provisional value is appropriate or not, and if there is an input that the provisional value is appropriate, the provisional value is registered in a dictionary as dictionary information of the compound word candidate.
How to register collocations. 4. In the dictionary registration step, the user selects one of the words constituting the collocation candidate specified by the user as the central word, extracts the dictionary information of the central word from the dictionary, and uses it as the provisional dictionary information of the collocation candidate. 3. The provisional value is presented to the user as a value, the user is made to input information as to whether the provisional value is appropriate, and if the provisional value is appropriate, the provisional value is registered in a dictionary as dictionary information of the compound word candidate. How to register collocations. 5. In the dictionary registration step, the dictionary information of each word constituting the collocation candidate specified by the user is extracted from the dictionary, combined, and presented to the user as a provisional value of the dictionary information of the collocation candidate, and the provisional value is 3. The collocation registration method according to claim 1, wherein the user is made to input information as to whether or not the collocation is appropriate, and if the provisional value is appropriate, the provisional value is registered in a dictionary as dictionary information of the collocation candidate. 6. In the collocation candidate extraction step, it is checked whether each collocation candidate extracted from the text has already been registered in the dictionary, and if it has been registered, it is deleted from the collocation candidates. How to register. 7. A dictionary in which dictionary information of words and collocations is registered; a collocation candidate extraction rule file that stores collocation candidate extraction rules for extracting collocation candidates from text; and a collocation candidate extraction rule file for extracting collocation candidates from text based on the dictionary information and collocation extraction rules. a collocation candidate extraction means for extracting collocation candidates using a combination of words, a collocation candidate display means for displaying the extracted collocation candidates on a display device, and a collocation candidate specifying means for allowing a user to specify one of the displayed collocation candidates. and an inclusion relationship display means for selecting other collocation candidates that have an inclusive relationship with the collocation candidate specified by the user, and displaying these collocation candidates on a display device in a format that allows the inclusion relationship to be understood; A collocation registration device comprising: information input means for allowing a user to input predetermined information; and dictionary registration means for registering a collocation candidate designated by the user as a collocation in a dictionary together with the input information. 8. A KW that extracts sentences containing collocation candidates specified by the user from the text and presents them to the user in the form of a KWIC list.
8. The collocation registration device according to claim 7, further comprising IC presentation means. 9. The information input means estimates a central word according to the pattern of collocation candidates specified by the user, extracts dictionary information of the central word from the dictionary, and presents it to the user as a provisional value of the dictionary information of the collocation candidate. The dictionary registration means includes provisional value presentation means and confirmation input means for allowing the user to input information as to whether the provisional value is appropriate, and when the dictionary registration means receives an input that the provisional value is appropriate, 9. A collocation registration device according to claim 7, wherein said provisional value is registered in a dictionary as dictionary information of said collocation candidate. 10. The information input means includes a central word specifying means for specifying one of the words constituting the collocation candidate specified by the user as a central word, and extracts dictionary information of the designated central word from the dictionary and uses it as the collocation candidate. provisional value presentation means for presenting the provisional value of the dictionary information to the user; and confirmation input means for prompting the user to input information as to whether the provisional value is appropriate; 9. The collocation registration device according to claim 7 or 8, wherein the provisional value is registered in a dictionary as dictionary information of the collocation candidate when there is an input that the collocation candidate is appropriate. 11. Temporary value presenting means, wherein the information input means extracts dictionary information of each word constituting the collocation candidate specified by the user from the dictionary, combines it, and presents it to the user as a provisional value of the dictionary information of the collocation candidate; and confirmation input means for prompting a user to input information as to whether the provisional value is appropriate, and the dictionary registration means selects the provisional value as the collocation candidate when there is an input that the provisional value is appropriate. 9. The collocation registration device according to claim 7 or 8, wherein the collocation registration device registers in a dictionary as dictionary information of . 12. The dictionary is a bilingual dictionary for language translation, and includes translated words as dictionary information, and the information input means extracts the translated words of each word constituting the collocation candidate specified by the user from the dictionary, combines them, and combines them into the above-mentioned A provisional value presentation means for presenting a provisional value of a translation of a compound word candidate to a user, and a confirmation input means for allowing the user to input information as to whether or not the provisional value is appropriate; 12. The collocation registration device according to claim 7, wherein the provisional value is registered in a dictionary as dictionary information of the collocation candidate when there is an input that the value is appropriate. 13. The collocation candidate extraction means includes a registration check means for checking whether each collocation candidate extracted from the text has already been registered in a dictionary, and a collocation candidate narrowing means for deleting registered collocation candidates from the collocation candidates. The collocation registration device according to any one of claims 7 to 12.