JPS6398072A - System for detecting ambiguous modifier relation - Google Patents

Abstract

PURPOSE:To correctly make a system operate, by detecting ambiguous modifier relation out of an inquiry text, displaying all of the modifier relation for which plural kinds of interpretation can be applied, and selecting the modifier relation intended by a user. CONSTITUTION:When a text in which a KANA (Japanese syllabary) and a KANJI (Chinese character) are mixed, is inputted, a clause segmenting/label attaching block 7 refers adjunct information stored in an adjunct information storing part 1, and divides one clause to the adjunct, and a character string from which the adjunct is eliminated, and decides a kind. Following that, an ambiguous modifier pattern detecting block 10 refers an ambiguous modifier pattern information storing part 8, and finds the modifier relation for which the plural kinds of interpretation can be applied. When the modifier relation is detected, a selection picture displaying and modifier relation qualifying block 11 refers a display picture information storing part 9, and displays the modifier relation for which the plural kinds of interpretation can be applied, on a screen, and urges selection to the user. And the modifier relation selected by the user is qualified, and the text in which the modifier relation intended by the user is adopted, can be reconstructed.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a natural language analysis section in a natural language dialogue system using a computer, and in particular, when the meaning cannot be uniquely determined due to ambiguous modification relationships, This invention relates to a method for accurately recognizing a modification relationship by clearly displaying it and asking the user about it.

[Conventional technology]

In order to understand the exact meaning of a query sentence in natural language, conventional devices predetermine the user's range and determine the modifying relationships between words within that range, as described in Japanese Patent Application Laid-Open No. 59-223883. It was designed to be uniquely recognized. However, even if a range is defined, the meaning of the words used by the user is not necessarily unique, and when the user intends a relationship different from the relationship defined by the system, consideration must be given to the means to adopt that relationship. It had not been done.

[Problem that the invention seeks to solve]

In the above-mentioned conventional technology, even if the range of users is defined, the meaning of the words used by the user is not necessarily -, and a modification relationship that is different from the modification relationship determined by the system is used.

There is no consideration given to the means by which the system adopts the qualification relationship when the user intends to do so.

Therefore, there was a problem in that the system could not accurately recognize the user's intended request, resulting in the system operating incorrectly.

The purpose of the present invention is to detect ambiguous modification relationships in a query sentence, display all possible modification relationships that can be interpreted in multiple ways on the screen, allow the user to select the modification relationship that he/she intends, and cause the system to operate correctly. The object of the present invention is to provide an ambiguous modification relationship detection method that can perform the following.

[Means for solving problems]

The above purpose is to identify modifiers and parallel directives in a query sentence, and use ambiguous modification pattern information to detect ambiguous modification relationships, that is, modification relationships that can be interpreted in multiple ways. This is achieved by displaying all of them on the screen and letting the user select the modification relationship he/she intends.

[Effect]

When the system detects that an ambiguous modification relationship exists in a query sentence, the system displays all of the resulting modification relationships that can be interpreted in multiple ways and allows the user to make a selection.

Thereby, the system.

It becomes possible to accurately recognize the user's intended action.

Further, since the ambiguous modification pattern information has a simple structure, it can be easily added and modified, and by arranging this information, it is possible to grow the system into an increasingly sophisticated system.

〔Example〕

An embodiment of the present invention will be described below with reference to FIGS. 1 to 16.

FIG. 1 shows the configuration of the ambiguous modification relationship detection method which is the main feature of the present invention. 1 is an adjunct word information storage unit, 2 is a
A modifier information storage section, 3, a parallel indicator information storage section, 4, a conjunction information storage section, and 5.

Parallel instruction adjunct word information storage unit; 6, ambiguous modification relationship recognition control functional block; 7, phrase extraction and labeling functional block; 8, ambiguous modification pattern information storage unit; 9, display screen information storage unit; 10 is an ambiguous modification pattern detection functional block, and 11 is a selection screen display and modification relationship recognition functional block. In addition, the hatched arrows in the figure indicate the flow of data, the double-lined arrows indicate references, and the single-lined arrows indicate the flow of control.

When a sentence containing kana and kanji (see FIG. 3) is input, a function block 7 is activated for segmentation and labeling. Step 7 cuts out one clause, then refers to the attached word information stored in the attached word information storage unit 1, and extracts the - clause,
Divide into an attached word and a string from which attached words are removed, and further,
Modifier information storage unit 2. Parallel directive information storage unit 3. Conjunction information storage unit 4. Referring to the parallel instruction adjunct word information storage unit 5,
The type is determined for each divided character string unit.

Subsequently, the @adjustment pattern detection function block 10 is activated. 10 refers to the ambiguous modification pattern information storage unit 8 to search for an ambiguous modification relation, that is, a modification relation that can be interpreted in a plurality of ways. Due to this.

If an ambiguous modification relationship is detected, a selection screen display and modification relationship recognition function block 11 is activated. 11 is
With reference to the display screen information storage section 9, modification relationships that can be interpreted in a plurality of ways are displayed on the screen, and the user is prompted to make a selection. Then, the modification relation selected by the user is recognized, and a sentence is reconstructed that adopts the modification relation intended by the user. The ambiguous modification relationship recognition control functional block 6 controls each of the above-mentioned functions as a whole. From FIG. 2 onwards, details of the functional blocks in FIG. 1 will be described.

FIG. 2 shows the hardware configuration of an embodiment of the present invention, in which 12 is a processor, and 13 is a device for displaying modification relationships that can be interpreted in multiple ways.

14 is a keyboard for inputting user responses; 15 is a keyboard for inputting user responses;
It is a storage device and consists of the following storage units.

16 is an attached word information storage unit, 17 is a program storage unit, 18 is a work area, 19 is an input/output canister, 21
9 is a modifier information storage unit, 31 is a parallel directive information storage unit, 41 is a conjunction information storage unit, 51 is a parallel directive adjunct word information storage unit, 81 is an ambiguous modification pattern information storage unit, 9
1 represents a display double-sided information storage section.

FIG. 3 shows an example in which a sentence containing kana and kanji is displayed, ``A beautiful person and a dog live together.''

FIG. 4 shows an example of displaying modification relationships that can be interpreted in a plurality of ways, which are detected when the sentence example of FIG. 3 containing kana and kanji is input.

FIG. 5 shows an example of a sentence reconstructed by adopting the modifying relationship intended by the user when the user selects the modification # relationship 1 in FIG. 4.

FIG. 6 shows an example of a format in which adjunct word information is stored in the adjunct word information storage unit 1.

FIG. 7 shows an example of a format in which modifier information is stored in the modifier information storage section 2. As shown in FIG.

FIG. 8 shows an example of a format in which parallel instruction word information is stored in the parallel instruction word information storage section 3.

FIG. 9 shows an example of the format in which connective word information is stored in the connective word information storage section 4. As shown in FIG.

FIG. 10 shows an example of a format in which parallel instruction adjunct word information is stored in the parallel instruction adjunct word information storage section 5.

FIG. 11 shows an example of a format in which ambiguous modification pattern information is stored in the ambiguous modification pattern information storage section 8. 8
consists of a display flag section and a pattern character string section, and the display flag section 1 matches that in the display screen information storage section 9. Further, the pattern character string section includes labels representing the Ni classes of the character strings divided in the functional block 7 (see processing steps 710 to 714 and 720 in FIG. 14, 'modification', 'contact').

There is ``Nami-ゝ'' and ``End 1.'' Hereinafter, the term label will be used in this sense. ) and a symbol that specifies the conditions for that label (No. 15)
See process step 1018 in Figure (b). Number l @
J. There is an alphabet. Hereinafter, symbols will be used with this meaning. ), separated by ``.'' When the example sentence in Figure 3 containing kana and kanji is input, the first line of 8 is adopted.

FIG. 12 shows an example of a format in which display screen information is stored in the display screen information storage section 9. 9 consists of a display flag section and a display format section, and 1 display flag is used for selecting a display format. In addition, the display format is 1 functional block 11, and "
” letters A and Bl.

B2...Bn is replaced with an appropriate character string, and the trailing part (〉) is removed, so that it can be output directly to the screen. 〈〉
The number inside represents the number of times modifiers are inserted during sentence reconstruction. When the example sentence in FIG. 3 containing kana and alphabetic characters is input, the first line of 9 is adopted.

FIG. 13 shows the contents of processing 1 of the ambiguous modification relationship recognition control function in PAD (processing flow diagram). When activated, the ambiguous modification relationship recognition control function block 6 reads a sentence containing mixed kana and kanji (601), activates the function block 7 shown in FIG. 1, and processes the extraction and labeling of one phrase. A leading number '1' is set in a counter representing the order of the character strings divided by the functional block 7 (602). In processing step 603, the array LABEL (I) in which the label is stored is compared with ``end'', and if it is not 1 end, processing steps 604 to 6 (1
8 and function block 10.11 are repeated. In processing step 604, it is determined whether LABEL (I) is 1 repair'. If it is 'repair', the process branches to 605, all arrays to be used are cleared, and function block 1o is to start. 10 is a flag MATC to check whether there is a modification relation that can be interpreted in multiple ways. Next, a flag MATC indicating whether an ambiguous modification relation is found
H (10, or 100 if an ambiguous modification relationship is detected)
', and if not detected, 101 is set. ) is compared with '100' (606), and if they are equal, the function block 11 is activated. 11 displays all detected multiple-interpretable modification relationships and allows the user to select the desired one. In processing step 607, the value of flag P indicating which clause is to be examined next is assigned to ■. In the previous processing step 604, if LABEL (I) is not 1 ', increase the value of ■ by 1 (608)
. Again, in processing step 603, the value of LABEL (I) is determined, and if it is 'end', this processing is ended.

FIG. 14 shows the processing contents of the phrase extraction and labeling functions in PAD. When the phrase extraction and labeling function block 7 is activated, a counter N becomes the subscript of the array 5ETSU that stores the divided character strings.
A leading number 11' is set in (701), and processing steps 703 to 719 are repeated as long as there is a clause in the input sentence (702).

First, one phrase is extracted from the input sentence (703).

Then, matching with the data in the adjunct word information storage unit 1 is performed from the end of the extracted phrase (704).

705) to determine if there is a match, and if there is a match, the matched adjunct is Sr':
Assign it to TSTJ (N+1) (706), and remove the matching adjunct from the clause to create a string of 5ETSU (N
) (707). If there is no match, leave a blank in SF! Assign to TSU(N+1)7
0B), assign the relevant clause to 5ETSU(N) (70
9).

Next, the modifier information storage section 2. Parallel directive information storage unit 3.
Refer to the conjunction information storage unit 4 to determine the type of 5HTSU(N) (710), and if it is a modifier, 'Shu' (711)
), for parallel words, 'parallel' (712).

If it is a conjunction, 'conjunction' (713) is substituted, and if it is another word, '1' (714) is assigned to LABEL (N).

Next, with reference to the parallel instruction attached word information storage unit 5, 5ET
SU (N+1) is difficult to distinguish between parallel directive adjuncts (715)
, in the case of a parallel demonstrative adjunct, assign 1-parallel' to LABEL (N+1) (716) and increase the value of N by 2 (717).
). If it is not a parallel directive adjunct, the value of counter N is set to 1.
Increase (71B).

Then, in processing step 719, 5ETSU(N) and 5
ETSU The string that combines (N+1) is 5ETSU
Assign to (N). Again, in processing step 702, it is checked whether there is a clause in the input sentence and if there is no clause, LABEL
1 end' is substituted for (N) (720) and the 1-step process is completed.

FIGS. 15(a), (b), and (c) show the processing contents of the ambiguous modification pattern search function in PAD. When the ambiguous modification pattern search function block 8 is activated, the value of ■ is assigned to P (1001).

The subscript IS of the array S for storing modifiers is set to 0' (1002). and.

Processing steps 1004 to 1007 are repeated while the value of LABEL(P) is 'correct', 'contact', or 'average' (1003). In processing step 1004, LABE
Compare the value of L (P) with 1 ', and if they are equal, IS
1005).

Assign the value of 5ETSU (P) to 5 (Is) (10
06). In processing step 1007, the value of P is increased by 1.

Again, determination is made in processing step 1003, and LAT'l
If EL (P) is neither 'modified' nor 'connected 9°1 parallel', set the flag MATCH to 0' (1
008).

Then, whether there is information (hereinafter referred to as data) stored in the ambiguous pattern information storage section 8 or M
While the value of TCH is (Ol), processing step 101
Repeat 0 to 1050 (too9).

In processing step 1010, an array PLABi that stores labels of data in 8 and an array PKI that stores symbols are created.
1' is set in the IP subscript of GO, and in processing step 1o11, data is read from 8, and in processing step 1012, the display flag in the data is assigned to flag FLAG representing the 1 display flag. Next, 1. in the data. ' as a delimiter and repeat processing steps 1014 and 1015 until the end (1013)
.

In processing step 1014, the label in the delimited character string is sent to PLABEL (I P ), and the symbol is sent to PKIGO.
Assign to (IP) (1014) and increase the value of IP by 1 (1015). Again, in processing step 1013, it is determined whether all character strings in the data have been separated, and when this is completed, O' is set in the subscript ISA of the array SAME that stores the same word, and the modifier and connective word are set.

0 in the subscript IT of the array T that stores words other than the parallel directive
', set 10' to the subscript ITtJ of the array TURN that stores the value indicating the last position of the parallel directive string, and set '100' to MATCH (1016
).

Next, while increasing the value of variable J indicating the order of PLABEL and PKI G○ by 1, processing step 10 is performed.
18 to 1047 are repeated IP times (1017). In processing step 1018, the value of PKIGO (J) is determined, and if it is a number (representing the number of consecutive labels), the process branches to processing step 1019, and 'umbrella' (the number of consecutive labels is 0 or more) If it is an alphabet (the same alphabet indicates that the corresponding character strings in 5ETSU are also the same), the process branches to step 1031, and if it is the same alphabet, it means that the corresponding character strings in 5ETSU are also the same.
Branch to.

When branching to processing step 1019, processing steps 1020 to 10 are performed while incrementing the value of counter JJ by 1.
Repeat 30. In processing step 1020, PLABE
L (J) and LABEL (P) are compared, and if they are equal, the process branches to step 1021. Processing step 1
At 021, determine the value of PLABEL (J), and if 'other is 1, increase the value of IT by 11022), 5ETS
Assign the value of U (P) to T (IT) (1023),
If 'average'.

The value of LABEL (P+1) is 102 compared to 6
4) If they are not equal, increase the ITU value by 1.0
25), Assign the value of IT to TURN (ITU).
026), an array lN5IERT (I
(1027).

When processing step 1021 is completed, the value of P is increased by 1 (1028). If they are not equal in processing step 1020, set MATCH to 10' (1029),
Extraction is performed from the edge return in processing step 1017, and the process branches to processing step 1048 (1030).

When processing step 1018 branches to processing step 1031, PLABHL (J) and I, ABEL (
P) and while equal, processing steps 1032-1
In processing step 1032, which repeats 035, PLA
Compare REL (J) and 1 and ′, and if they are equal, I
The value of T is increased by 1 (1033), and the value of T (IT) is assigned to 5ETSU (P) (1034).

When processing step 1032 is completed, the value of P is increased by 1 (1035).

When processing step 1018 branches to processing step 1036, PLABEL (J') and LABEL (P
), and if they are equal, processing steps 1037-10
46 processing is performed, and if they are not equal, processing step 10
After repeating step 17, the process branches to processing step 1048 (1036). In processing step 1037, P
Let's determine the value of LAnIEL (J)1037),'
If other', increase I'r(7) value by 11038)
, 5ETSU Assign the value of (P) to T(IT) (1
039). If it is 'average', in processing step 1040, L
Compare ABEL (P+1) and 'average', and if they are not equal, increase the value of ITU by 11041), assign the value of IT to TURN (ITU)1042), and change the value of P(7) to lN5ERT (I T U ) (104
3) When processing step 1037 is completed, increase the value of ISA by 1, L (1044), PKIGO (J) and 5
The string that combines ETsU(P) is SAME (ISA
) and increase the value of P by 1 (1046
).

After the repetition of processing step 1017, processing step 1048 determines that ISA is 1 or more and MA
It is determined whether TCH is '100' and if one condition is satisfied, the process branches to step 1049.

In processing step 1049, SAME (1) to SAM
Search for any two strings in E (ISA) that have the same first character but different character strings from the second onwards.104
9), if found.

Set MATCH to 'O'; 5 (1050).

Returning again to processing step 1009, the condition is determined, and if the condition is no longer satisfied, the process ends.

FIG. 16 shows the selection screen display and the processing contents of the ambiguous modification relationship recognition function in PAD.

When the selection screen display and ambiguous modification relationship recognition function block 11 is activated, it first selects a line from the 1 display screen information storage section 9 where the value of the display flag section is equal to FLAG (110
1). Then, change the character A in the display format of the line to A(1
) ~ 5 (IS) with a concatenated string (110
2), then.

Characters Bl, F12 . in the display format. ..., Bn,
A character string that is a concatenation of T (1) ~ T (TURN (1)), T (TURN (,1) + 1) ~ TCTU, respectively.
RN (2)),..., T(TUR
N (n-1)+1) to T (TURN(n)) are replaced with a concatenated character string (1103).

Next, the numbers in brackets (representing the number of insertions of modifiers used when reconstructing a sentence) in the display format are stored in order in the array KAISTJ, and then (1104). Then output the display format to the screen.

Waits for input from the user (1105). The number input by the user is substituted into INPUT (1106). KAIS
Compare U (TNr'tJT) and l OI.

If not equal, 1. While increasing the value of ■ by 1, process steps 1109 and 1110 are
) times (1108).

In processing step 1109, S E T S U (IN
SERT (J)) and 5ETSU (INSERT (J
) +1), S (1) , S (2) ,
..., insert S (IS). Then, the value of P is I
Increase to S (1110).

When the processing in processing step 1107 is completed, this processing ends.

The examples have been described in detail above.

According to this embodiment, since each functional block is realized by program logic, each functional block is
It can be made into an LSI and can be made faster. Furthermore, since each information storage section is independent, it can be modified individually.
Also, since the data creation is simple, it is easy to modify. Therefore, it is possible to improve the system to a more advanced one through repeated modifications.

〔Effect of the invention〕

According to the present invention, it is possible to detect modifying relationships that can be interpreted in multiple ways from an input sentence, clearly displaying them, and allowing the user to make a selection. This has the effect of making it possible to certify the

[Brief explanation of the drawing]

Fig. 1 is a functional block diagram constituting the present invention, Fig. 2 is a block diagram showing the hardware configuration of an embodiment of the present invention, Fig. 3 is an example of an input character string, and Fig. 4 is a block diagram of the present invention. According to
A display example of the generated multiple-interpretable modification relationship, Figure 5, is as follows:
An example of a sentence that accurately realizes the modifying relationship intended by the user, FIG. 6 is an example of the storage format of the adjunct word information storage section, FIG. 7 is an example of the storage format of the modifier information storage section, and FIG. 8 9 is an example of the storage format of the parallel instruction word information storage section, FIG. 9 is an example of the storage format of the conjunction word information storage section, FIG. 10 is an example of the storage format of the parallel instruction adjunct word information storage section, and FIG. An example of the storage format of the ambiguous modification pattern information storage section, FIG. 12 is an example of the storage format of the display screen information storage section, FIG. 13 is a PAD diagram of the warmth modification relationship recognition control process, and FIG. The labeled one is a PAD diagram of the process, Figures 15 (a), (b), and (c) are PAD diagrams of ambiguous modification pattern detection processing, and Figure 16 is a PAD diagram of selection screen display and ambiguous modification relationship recognition processing. It is. 1... Adjunct word information storage section, 2... Modifier information storage section, 3... Parallel instruction word information storage section, 4... Connecting word information storage section, 5... Parallel instruction adjunct word information Storage part, 6...
Conjunction information storage unit, 7... Ambiguous modification relationship recognition unit, 8.
... ambiguous modification pattern information storage section, 9 ... display screen information storage section, 10 ... ambiguous modification pattern detection section, 11.
・・Selection 13・・Table ho relief child L・・Keyboard 15 ・Note 1g table No. 3121 No. 5 Figure early 6th 7th day 8th mouth No. F I2I Figure 70 %tノ ? ] Bō 72 B No. 75 Kuji 1 No S ji] (α)

Claims (1)

[Scope of Claims] 1. In a natural language dialogue system using a computer, means for finding ambiguous modification relations, that is, modification relations that can be interpreted in multiple ways, from a user input sentence; prompts the user to select the intended qualification relationship,
An ambiguous modification relationship detection method characterized by providing a means for recognizing an accurate modification relationship. 2. An ambiguous modification relationship detection method according to claim 1, further comprising means for dividing an input sentence into phrases and attaching a label indicating the type of phrase. 3. An ambiguous modification relationship detection method according to claim 1, characterized in that the storage unit contains information representing a pattern of ambiguous modification relationships used when detecting ambiguous modification relationships. 4. In the method described in claim 1, the selection screen display format information when displaying the detected multiple-interpretable modification relationships and allowing the user to select the intended modification relationship,
An ambiguous modification relationship detection method characterized by having it in a storage section. 5. An ambiguous modification relationship detection method according to claim 2, characterized in that a plurality of pieces of information used for labeling a clause are stored separately. 6. In the method described in claim 3, an information section that defines the appearance pattern of phrases and a flag section that defines selection screen display format information are provided as information representing a pattern of ambiguous modification relationships. Features ambiguous modification relationship detection method. 7. In the method described in claim 4, as display format information when the user selects the modification relationship intended by the user, a flag part that defines selection screen display format information, a modifier character string in the input sentence, and a An ambiguous modification characterized by having a character string part that can be used as display screen information by inserting a modifier character string, and a numerical part that represents the number of times the modifier has been inserted during sentence reconstruction. Relationship detection method. 8. In the method described in claim 5, information used when labeling a clause includes information composed of a character string that can be a modifier, and characters that can be an independent word indicating a parallel relationship. The present invention is characterized by providing information composed of columns, information composed of character strings that can be used as words that indicate connection relationships, and information composed of character strings that can be used as adjunct words that indicate parallel relationships. An ambiguous modification relationship detection method. 9. The method described in claim 6, characterized in that a label indicating the type of clause and information specifying the appearance condition of labeled clauses are provided as information specifying the appearance pattern of clauses. , ambiguous modification relationship detection method. 10. In the method described in claim 9, the information specifying the appearance condition of labeled clauses includes information indicating the number of consecutive clauses of the same type, and that zero or more clauses of the same type occur in succession. and information indicating that two or more clauses are composed of the same character string.