JPH05113996A - Natural language processing system - Google Patents

Abstract

PURPOSE:To obtain the natural language processing system improving conve nience so as to correct modification relation with simple operations and to simultaneously display all the analyzed results used for the correction. CONSTITUTION:An input sentence is analyzed, and the analyzed result is displayed in a modification relation drawing having a clause as a minimum unit (100). A user desires the correction and fetches the change information of the clause unit in the inputted and instructed modification relation (107-109). Based on this change information of the modification relation, the modification relation is corrected and displayed again (112-115, 102 and 103). When the user executes the input operation of instructing fixing, the analyzed result is fixed to the modification relation displayed at that time (110 and 116). Thus, since the correction is executed while displaying the analyzed result with the clause as the minimum unit, a display picture is reduced and the user facilitates the instruction of the correction as well.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language processing system, and more particularly to a user interface function after obtaining an analysis result for an input sentence.

[0002]

2. Description of the Related Art A natural language processing system such as a machine translation system, an automatic abstract creation system, or a natural language question answering system analyzes a natural language sentence.

In such a syntax analysis, input natural language sentences are collectively processed to obtain an analysis result. However, natural language sentences have ambiguity that cannot be correctly judged because they depend only on grammar and logic.

As a method for removing the influence of such ambiguity and obtaining a correct analysis result, information for evaluating a plurality of analysis results considered for the ambiguity is prepared in advance and based on this information. There is a method of evaluating the analysis result. There is also a method of removing the influence of ambiguity and obtaining a correct analysis result by a proofreading operation or an editing operation by the user. In the case of the former method, since the amount of information prepared in advance is large, many natural language processing systems adopt the latter method.

As the above-mentioned method of entrusting the user to remove the ambiguity of the analysis result, for example, (1) a plurality of obtained syntactic analysis results are sequentially displayed and the user selects the optimum syntactic analysis result ( JP-A-1-134570), and (2) a method of allowing the user to select or modify the dependency relationship of the syntax (JP-A-59-137132).
Japanese Patent Laid-Open No. 61-278970).

[0006]

In the case of the method (1) described above, there are cases where the correct parsing result does not exist among the obtained plural parsing results. From this point,
The method (2) in which the user makes the dependency relationship clear is preferable.

However, even with the method (2), there is a drawback that the analysis result used for correction by the user may not fit on the display screen (Japanese Patent Laid-Open No. 59-137132), and there is a problem of correct dependency. It has a drawback that the designation and input operations are complicated (Japanese Patent Laid-Open Nos. 59-137132 and 61-278970).

The present invention has been made in consideration of the above points, and it is possible to make a correction selection of the analysis result for the input sentence by a simple operation, and to display all the analysis results to be used for the correction selection. It aims to provide a good natural language processing system.

[0009]

In order to solve such a problem, the natural language processing system of the present invention has the following processing.

That is, the input natural language sentence is analyzed, and the analysis result is displayed in a dependency relationship diagram in which the bunsetsu is the minimum unit, and the dependency relationship instructed by the bunsetsu unit is displayed. Based on the change information import process that captures the change information and the modification information that modifies and displays the modification relationship based on the captured modification information of the modification relationship, it is displayed at that time in response to the input to instruct confirmation. The confirmation processing for confirming the analysis result in the dependency relationship that has been performed is performed.

Here, when the target system is a natural language processing system for machine translation, it is preferable to change the analysis result display processing and the confirmation processing as follows and to provide the following new processing.

That is, in the analysis result display processing, in addition to the analysis result, the translated word is displayed in association with the clause in the dependency relationship diagram. Further, a translation word change information fetching process for fetching the change information of the displayed translation word candidate and a translation word correction process for correcting and redisplaying the translation word candidate based on the fetched translation word candidate change information are provided. Further, the confirmation processing not only confirms the analysis result in the dependency relationship displayed at that time, but also determines the translation word in the translation word candidate displayed at that time, in response to the input instructing the confirmation.

[0013]

When the analysis result is displayed with each word included in the input natural language sentence as the minimum unit, the display area may become too large to be displayed on one display screen. It is difficult for the user to accurately recognize the relationship (syntax structure), and there is a possibility that the correction cannot be input accurately.

Therefore, in the present invention, the analysis result with the bunsetsu as the minimum unit is displayed for correction by the user.

That is, the input natural language sentence is analyzed, and the analysis result is displayed in a dependency relationship diagram with the clause as the minimum unit. Upon seeing this, the user who recognizes the necessity of correction indicates the change information of the dependency relationship in phrase units, and the system takes in this information. This allows the system to
The dependency relationship is corrected and displayed again based on the fetched modification information on the dependency relationship. When it is determined that the initially displayed analysis result is valid, or when it is determined that the analysis result has become valid by one or more corrections, the user performs an input operation to confirm this, and In response, the system confirms the analysis result in the dependency relationship displayed at that time.

Here, "bunsetsu" means a grammatical phrase when a Japanese sentence is an input sentence, and an input sentence in a language in which there is no grammatical term "bunsetsu". ,
The dependency relationship within the group does not become ambiguous,
It also means a set of words (including the words themselves) suitable for the user to determine the dependency relationship.

When the target system is a natural language processing system for machine translation, the analysis result and the translation word candidate are displayed on the same screen, and the analysis result correction processing and the translation word candidate correction processing are performed. If it can be executed, the two correction processes are not completely separated, which is convenient for the user.

Therefore, in the case of such a system,
In the analysis result display process, in addition to the analysis result, the translated word is displayed corresponding to the phrase in the dependency relationship diagram,
When the change information of the displayed translation candidate is given by the user, based on the translation information of the translation candidate,
When the user inputs an instruction to fix and redisplay the translated word candidate and confirms it, the translated word candidate that is displayed at that time is not only fixed in the dependency relationship that is displayed at that time It is preferable that the translated word is fixed.

[0019]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment in which the present invention is applied to a Japanese-to-English machine translation system will be described in detail below with reference to the drawings.

FIG. 2 shows an example of the configuration of a computer system to which this embodiment is applied. In FIG. 2, this computer system 1 is composed of a CPU 2, a keyboard 3, a display 4, a main memory 5, and a magnetic disk device 6. The magnetic disk device 6 stores an original sentence analysis program, a user interface program, a translated sentence generation program, a dictionary of source and target languages (including parsing grammatical rules), a bilingual dictionary and display screen data. The various programs described above are loaded into the main memory 5 and executed by the CPU 2. The dictionary, bilingual dictionary, display screen data, etc. used by the CPU 2 may be loaded into the main memory 5, and when the capacity of the main memory 5 is insufficient, the dictionary is used while being stored in the magnetic disk device 6. .. The keyboard 3 is for inputting input sentences, modification information on dependency relationships, correction information for translation word candidates, and the like, and the display 4 is for displaying analysis results and the like.

The following keys are provided in terms of the function of the keyboard 3 when it is used for inputting modification information related to dependency. That is, a translation word selection key group (consisting of two) 3a for changing a translation word candidate to a previous one or a subsequent one, and a modification destination for modifying a modification target clause in a modification relationship to a preceding one or a next one. A phrase selection key group (consisting of two) 3b, a target phrase selection key group (consisting of two) 3c for changing a modification source phrase (target phrase) in a dependency relation to a previous one or a subsequent one, and a user A confirmation key 3d for instructing the confirmation of the dependency relationship and the translated word,
A multi-branch selection key group (so-called numeric keypad) 3e used for selecting a predetermined translation word candidate displayed collectively and selecting a predetermined dependent phrase from a plurality of numbered phrases , And a prefix key 3f for instructing to use the multi-branch selection key group 3e for selection from translation word candidates displayed collectively.

The relationship between the prefix key 3f and the multi-branch selection key group 3e will be described in detail. When the multi-branch selection key group 3e is pressed after the prefix key 3f is pressed, the translated word candidates (collectively described later) are displayed. As described above, each candidate is displayed with a serial number added), and when the multiple branch selection key group 3e is directly pressed without pressing the prefix key 3f, This means that the related destination clause is individually specified from among the displayed clauses (as will be described later, each clause is displayed with a serial number added thereto).

This embodiment is characterized by the analysis process up to the semantic analysis of the original sentence, and such process is realized by software. Therefore,
Characteristic processing will be described in detail with reference to the flowchart of FIG.

First, the input sentence is analyzed, translation word candidates are generated, and the analysis result and translation word candidates are displayed (step 100). Here, the analysis is performed so as to obtain the dependency relationship (syntax parse tree) with the clause as the minimum unit. Various methods have already been proposed as an analysis method using a clause as the minimum unit, and any method may be applied. For example, a word dictionary contains information about independent words and adjunct words, and independence words and adjunct words of words obtained by morphological analysis are connected to obtain clauses, after which grammatical rules such as phrase structure rules are added. Apply to get the analysis result with bunsetsu as the minimum unit. Also,
For example, after obtaining the analysis result with each word as the minimum unit,
By applying the phrase structure rules that indicate clauses, the structural elements on the lower level side are put together to obtain the analysis result with the clause as the minimum unit.

Depending on the input original sentence, a plurality of analysis results can be obtained due to the ambiguity in the grammar, and all the analysis result information is stored and displayed in any one.
Only perform the analysis results. The plurality of pieces of analysis result information are stored, for example, in such a manner that there is a plurality of related destination clause information for a certain clause.

FIG. 3 shows the structure of the display screen by this processing. That is, the phrase display unit 20 showing each phrase by sequentially arranging serial numbers from the leading-side phrase of the input sentence and showing the dependency relation of each phrase, for example, in a tree structure (however, nodes are shown by simple line connection) From the dependency relation display unit 21 shown in, the translation word display unit 22 that displays the translation word candidates for each phrase, and the input auxiliary display unit 23 that displays the translation word candidates of a certain phrase with a serial number all together. Become.

It should be noted that the input auxiliary display unit 23 also displays a message or the like that teaches the user various operating methods.

The reason why the analysis result with the bunsetsu as the minimum unit is displayed is that the display area is smaller than the analysis result with the word as the minimum unit. In addition, the analysis result using the bunsetsu as the minimum unit makes it easier for the user to judge the validity of the dependency relationship than the analysis result using the word as the minimum unit.

When the analysis result and the translated word are obtained and displayed on the display 4 as described above, the first bunsetsu is set as the focused bunsetsu (entrance bunsetsu), and this focused bunsetsu is displayed separately from other bunsetsu (step 101, 102). For example, it is displayed in a color different from that of other clauses or is displayed in reverse video.
After that, the dependency relationship related to the target phrase is specified and displayed (step 103). That is, the dependent phrase of the target bunsetsu is clarified by color display or reverse display, and the dependent line (for example, two main lines) connecting the focused bunsetsu and the related target phrase is separated from other related lines (for example, one main line). Display separately. Further, the translated word of the target phrase in the translated word display unit 22 is displayed by being colored and highlighted so as to be distinguished from the translated words of other phrases, and the translated word of the focused phrase is collectively displayed in the input assisting unit 23 (step 104). ..

In this way, when the target bunsetsu, the related bunsetsu, and the translation candidate which are currently targeted are clarified and displayed, the process proceeds to the process of discriminating the key pressed by the user (steps 105 to 110).

When it is detected that the translation word selection key 3a for changing the translation word candidate to the previous one or the subsequent one is pressed by the user, the translation word of the target phrase is selected as the previous translation word candidate or the backward translation word in accordance with the operation. After changing to the translated word candidate, the process returns to step 104 relating to the specific display process of the translated word candidate (step 111). Further, when it is detected that the prefix key 3f and any one of the multiple branch selection keys 3e are pressed by the user, the translation word candidate of the target phrase is changed to the translation word candidate corresponding to the multiple branch selection key 3e, and then the translation word candidate The process returns to step 104 related to the specific display process (step 1
11). The above is the process regarding the change of the translation word candidate.

When it is detected that the dependent phrase selection key 3b for changing the dependent phrase to the previous or next rank is pressed by the user, the dependent phrase of the noticed phrase is changed according to the operation. After performing the re-analysis process after changing the phrase to a different phrase, the process returns to step 103 related to the specific display process of the dependency destination phrase and the dependency line (steps 112 and 11).
3). In addition, the user may select one of the multiple branch selection keys 3
When it is detected that e is pressed immediately, after fixing the dependent clause of the target clause to the clause corresponding to the multi-branch selection key 3e and executing the re-analysis processing, the specific display processing of the dependent clause and the relation line Return to Step 103 relating to (Step 1
14, 113). The above is the processing relating to the modification of the related clause.

The reanalysis process will be described in more detail. If the first or second order is designated, it means that one of the plurality of analysis results obtained by the initial analysis has been designated, and this is taken out from the stored contents.
Further, in the case of designation by the multi-branch selection key 3e, it is judged whether or not it is one of a plurality of analysis results obtained by the initial analysis, and if a positive result is obtained, it is extracted from the stored contents. When a negative result is obtained, the analysis is performed from the beginning with the phrase specified as the dependent phrase of the target phrase fixed. In addition, in the re-parsing performed when the dependent clause that is not obtained in the initial analysis is specified, in addition to the method of analyzing the entire sentence, the part affected by the change in the dependent clause is re-parsed. Is also good.

In the above-described detection of the pressed key, when it is detected that the user has pressed one of the attention phrase selection keys 3c for changing the attention phrase (entry source phrase) to the previous one or the latter one, the attention phrase is detected. Is changed to the previous phrase or the subsequent phrase according to the operation, and then the process returns to step 102 relating to the specific display processing for distinguishing the target phrase from other phrases (step 115).

When it is detected that the confirmation key 3d is pressed by the user, the analysis result and the translated word are fixed to the contents displayed at that stage (step 116).

In this way, the processing of the characteristic part of the embodiment is completed, and after the semantic analysis is executed, the process proceeds to the generation of the sentence (translated sentence) of the target language.

There is no limitation on the number of key inputs for the correction operation, and the above-described processing corresponding to the pressed key may be repeated several times.

Next, the processing explained above will be described in detail by an example sentence "As internationalization progresses, the number of companies expanding overseas will increase."

First, when this sentence is analyzed, the clauses "internationalization", "as it progresses", "going overseas", "expanding",
The analysis result and the translation word indicating the dependency relationship with respect to “company is” and “increase.” Are obtained, and the analysis result and the translation word candidate are displayed.

Here, it is assumed that the analysis result shown in FIG. 4 (A) was initially obtained. That is, it is assumed that the analysis result obtained is that it is correct that the destination of the phrase “as it goes” is the phrase “increase.” But that the destination phrase is “advance”. Further, it is assumed that "(as) improves", which is not appropriate as a translation of the phrase "as we go", is obtained. The part in parentheses in the translated word is
The part that may or may not be displayed by the application system.

Immediately after the display of such an analysis result, as shown in FIG. 4A, the first clause "Kokusaika" is the noticeable clause, and this clause "Kokukaika" is the noticeable clause. A specific display is made, and the translation candidate of this phrase “Kokusaika” is displayed differently from the translation candidates of other phrases, and the phrase “as you go” is specified as the related phrase of this phrase “Kokusaika”. The bar line connecting these two clauses “internationalization” and “as it progresses” is displayed separately from other bar lines.

The user who confirms such display contents is
As for the attention phrase “Kokusaika”, the obtained dependency relation and translation candidate are valid, so that the attention phrase (dependence source phrase) is changed to the next phrase “as it progresses”. The attention phrase selection key 3c to be changed to the one is pressed.

As a result, as shown in FIG. 4B, the phrase "as you go" is displayed as a particular phrase, and the translation word candidate of this phrase "as you go" is different from the translation candidates of other phrases. Is displayed and the phrase “advance” is specified and displayed as the destination phrase of this phrase “as you go”, and the entanglement line connecting these phrases “as you go” and “advance” is another Is displayed separately.

The user who confirms such display contents is
A correction operation is performed by recognizing that the related part of the phrase “as you go” is not valid and that the translated word of the phrase “as you go” is not valid.

Either the modification operation of the dependent phrase or the translation word candidate may be executed first, but here it is assumed that the modification operation of the dependent phrase is performed.

Here, the user uses the multi-branch selection key 3e ("6" in the numeric keypad in this example) or the dependent-destination phrase selection key 3b for designating the dependent-destination phrase candidate of the next rank, to increase the phrase. Will be instructed. As a result, reanalysis is performed in which it is fixed that the clauses “as you go” and “increase.” Have a dependency relationship, and the results are displayed.

FIG. 4C shows the result of the re-analysis. As in the case before the re-analysis, the phrase "as you go" is displayed as a particular phrase, and the translated word of this phrase "as you go". The candidate is displayed differently from the translation candidates of other bunsetsu, and the bunsetsu "increase." Is identified and displayed as the related phrase of this bunsetsu "as we go". The bar line connecting "." Is displayed separately from other bar lines.

With such a display content, when the dependency relation of the target phrase “as we go” becomes appropriate, the translated word “(as) impr” of the phrase “as we go”.
oves ”correction processing. The user selects, for example, the translation word selection key 3a for displaying the next translation word candidate, the translation word candidate “(as) advance” desired to be displayed as the translation word candidate.
Press repeatedly until "s" is reached. Alternatively, for example, after pressing the prefix key 3f, the desired translation word candidate “(as) adva” is selected from the translation word candidates displayed collectively.
nces ", and a multi-branch selection key (any numeric keypad) equal to the serial number assigned to the translation candidate 3
Press e. Regardless of which operation method is adopted, the translation word candidates are changed from “(as) improves” to “(as)
changed to “advances” and displayed as shown in FIG.

When the display content shown in FIG. 4D is checked and it is determined that the dependency relationship (analysis result) and the translation word candidate are all valid, the user presses the confirmation key 3f to analyze the analysis result and the translation word. To confirm.

As described above, according to the above-mentioned embodiment,
Since the analysis result is displayed with the clause as the minimum unit, the display area of the analysis result can be made small, and the entire display can be displayed even when the input sentence is a long sentence. Further, since the dependency relationship is corrected in units of phrases, the user can easily give an instruction. Incidentally, it is difficult for many users that the user recognizes a strange point in the analysis result from the analysis result using the word as the minimum unit and gives an instruction for the correction content. Furthermore, the method of instructing the correction content is also related to the original clause (attention clause).
Also, it is simple and easy to operate because it only requires a specific operation of the related clause.

In addition, since the translation of the translated word can be corrected in the series of correction processing of the analysis result, the support by the user can be efficiently executed.

Although the present invention is applied to the machine translation system in the above embodiment, the present invention can be applied to various natural language processing systems that require analysis processing of input natural language sentences. You can When applied to a system other than the machine translation system, naturally, the processing of displaying and correcting the translation word candidates is eliminated.

Further, although the case where the inputted natural language sentence is a Japanese sentence is shown in the above-mentioned embodiment, the present invention can be applied to the analysis of a sentence of another language. Here, in the case of sentences in other languages, there is no grammatical element called bunsetsu, but if the present invention is applied with one set of word groups (including only one word) similar to this as the minimum unit. good. That is, one set of word groups may be used as the minimum unit, in which the dependency relation or phrase structure in the set does not have a pseudo, and the dependency relation of each set is easily recognized. For example, the English sentence "I saw the man in the
park with the telescop
e. , "I", "saw", "the ma"
n ”,“ in the park ”,“ with th
The analysis result may be displayed with the "e telescope" as the minimum unit and used for correction. Note that the term "bunsetsu" in the claims also includes such a language other than Japanese.

[0054]

As described above, according to the present invention, it is possible to display the analysis result in which the bunsetsu is the minimum unit and to specify the bunsetsu to correct the bunsetsu. It is possible to realize a user-friendly natural language processing system capable of performing correction selection of analysis results for natural language sentences and displaying all analysis results provided for correction selection at once.

[Brief description of drawings]

FIG. 1 is a flowchart showing a user interface process in an example system.

FIG. 2 is a block diagram showing a computer system to which the embodiment is applied.

FIG. 3 is an explanatory diagram showing a display screen structure in the example system.

FIG. 4 is an explanatory diagram showing a user operation and a specific change of a display screen.

[Explanation of symbols]

1 ... Computer system, 2 ... CPU 2, 3 ... Keyboard,
4 ... Display, 5 ... Main memory, 6 ... Magnetic disk device, 3a ... Translated word selection key, 3b ... Dependent phrase selection key,
3c ... attention phrase selection key, 3d ... confirmation key, 3e ... multiple branch selection key, 3f ... prefix key.

Claims (2)

[Claims] 1. An analysis result display process of analyzing an input natural language sentence and displaying the analysis result in a dependency relationship diagram with a clause as a minimum unit, and a dependency relationship input instructed in terms of a clause. The change information fetching process for fetching the change information, the modification process for revising and displaying the dependency relation based on the fetched change information for the dependency relation, and the display at that time in response to the input for confirming A natural language processing system characterized by having a determination process for determining the analysis result in the dependency relationship that has been used. 2. In a natural language processing system for machine translation, a natural language sentence of an input source language is analyzed, and the analysis result is displayed as a dependency relationship diagram with bunsetsu as a minimum unit, and this relation is also displayed. Analysis result display processing that displays the translated words corresponding to the clauses in the dependency relationship diagram, relationship change information import processing that captures the modification information of the dependency relationship that is input by clause unit, and modification of this imported dependency relationship Based on the information, the relationship correction process that corrects the dependency relationship and displays it again, the translation word change information capture process that captures the change information of the displayed translation word candidate, and the translation information of the translation word candidate that was captured, The translation word correction process that corrects and redisplays the translation candidate, and the analysis result is confirmed in the dependency relationship that was displayed at that time in response to the input to confirm In addition, a natural language processing system for machine translation, characterized by further comprising a confirmation process for confirming a translation word in the translation word candidate displayed at that time.