JPS60191360A - Language processing system - Google Patents

Abstract

PURPOSE:To improve the input efficiency of the language information by deciding the degree of intensity of connecting relation between pieces of language information according to a transistion table containing the connecting relations set previously and selecting a sentence coincident with a prescribed standard out of obtained sentence candidates. CONSTITUTION:The characters supplied from a keyboard 22 are read by a keyboard input circuit 20 and stored to an input character string memory 1 as well as to a display character string memory 8. Then these characters are displayed to a display device 32 through a display control circuit 30. While a processor 10 collates the memory 1 with a dictionary memory 3 and stores the words coincident with each other to a data memory 5. The memory 1 of the following word is compared with the memory 3, and coincident words are stored in the memory 5. Then the memory 1 of the next following word is collated with the memory 3, and the coincident words are stored to the memory 5. The preceding word is connected to the word of this time within the memory 5, and the connection class is stored to the memory 5 with reference to a transition table memory 4. Then a sentence candidate 6-5 is obtained from the degree of intensity of connecting relation between words for the contents of the memory 5, and a sentence coincident with a standard is stored to the memory 8.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Application of the Invention] The present invention particularly relates to a language processing method suitable for decomposing a character string into its constituent unit words.

[Background of the invention]

Conventionally, as a method for decomposing a character string into words, there is a method disclosed, for example, in Japanese Patent Publication No. 53-29504, ``Device for Conversion from Phonetic Symbol String to Written Symbol String.''

This method compares the input string with a dictionary, finds -(2) matching words, checks whether a linguistic connection can be established with the previously extracted word, retains the words that hold the word as candidates, and selects the candidate. If there are multiple words, this method selects the longest word and processes the next part of the input character by repeatedly checking it against the dictionary and checking whether a connection is established. In this method, if the connection is not established during the connection test, the word is selected from the other words extracted this time, and the next part of the character string is checked against the dictionary and the connection check is repeated. If all of the words extracted this time fail, one is selected from the previously extracted words, and the next part of the character string is checked against the dictionary and the connection is checked again. However, in this method, the longest word is selected from the extracted words, and processing is performed while determining one word. , the word (heading: kare, spelling: dead leaves, part of speech: noun) is selected, and the words (heading: kare, spelling: 2 he, part of speech: pronoun) and (heading: ha, spelling: ha, part of speech: modal particle) are selected. ) is not selected.

(3) As another method, [Makino et al.: “Separation of solid text and Kana-Kanji conversion - Separation using the two-clause longest match method -
”, Journal of Information Processing Society of Japan, 7, 1,979°VoQ, 2
0, &4.337-345).

This method selects the longest two clauses from among the possible candidates obtained by checking the dictionary and checking the connection, and processes the first half of the two clauses as the longest one. It is. However, with this method, for the input character string "Here is a pheasant", (heading: scree, writing: dead leaves, part of speech: noun) (heading: pheasant, writing: article, part of speech: noun) (heading: , notation: , part of speech:
(Heading: Gare, Writing: He, Part of Speech: Pronoun) (Heading: HA, Writing: HA, Part of Speech: Pronoun) also holds true.
(Heading: Pheasant, Description: Article 1 Part of speech: Noun) (Heading:
The problem is that it is difficult to select the correct connection between the words ``he'', ``ha'', ``article'', and ``wo''.

[Purpose of the invention]

(4) The present invention has been made in response to the above-mentioned problems, and includes a device for converting Japanese kana character strings into sentences mixed with kanji, and a device for translating one language into the language of another country. In order to improve the precision of a processing device, it is an object of the present invention to provide a language processing method that more accurately decomposes a character string into its constituent unit words.

[Summary of the invention]

In order to achieve the above object, the present invention consists of an analysis process that analyzes an input character string and breaks it down into single word candidates, and a selection process that selects a valid word candidate as an A candidate from among the word candidates. When analyzing character strings, in order to suppress the occurrence of word candidates as much as possible, we assign a grade to the connection relationships between words, and from sentence candidates obtained by connecting words with the strongest connection relationship, we select the most characteristic word. Selecting sentences with a large number of sentences, checking the consistency with sentence pattern information and deleting sentences that do not match.
It is characterized by selecting the sentence with the largest number of independent words and selecting the sentence with the highest average appearance frequency of each word.

(5) [Embodiment of the Invention] Hereinafter, an embodiment of the present invention will be described in detail using the drawings.

FIG. 1 is an overall configuration diagram showing an embodiment in which the language processing method according to the present invention is applied to a Japanese word processor.

The processing unit [110 is a microprocessor, which executes the program stored in the program memory 15 on the pass line 5.
0 and execute through pass line 5.
0, input string memory 1, display string memory 8
, input string position memory 2-7, data record number memory 8, data memory 5, sentence candidate memory 6-5, collation memory 6-8, dictionary memory 3, transition table memory 4, sentence pattern table memory 7, target It controls data transfer with word list memory 6-7, term list memory 6-6, and independent word list memory 6-9. The keyboard input circuit 2o reads the position on the keyboard when the character keyboard 22 is pressed by the operator,
The corresponding code is stored in the input character string memory 1 and the display character memory 8, and when a character (6) conversion key (not shown) on the keyboard 22 is pressed, an interrupt is made to the processing device 10. The display control circuit 30 is a CR
The display string memory 8 is refreshed and displayed on the display device 32 composed of T. The file control circuit 40 transfers data between the file medium in the file device 42 and each memory.

When a character is input from the character keyboard 22, the keyboard input circuit 2
0, stored in the input character string memory l and the display character string memory 8, and displayed on the display device 32 by the display control circuit 30. By executing the program stored in the program memory 15, the processing device 10 collates the input character string memory 1 with the dictionary memory 3, stores the matching word in the data memory 5, and then stores the next word after that word. The input character string memory l is checked against the dictionary memory 3, the matching word is stored in the data memory 5, the previous word and the current word are connected in the data memory 5, and the transition table memory 4 is referred to for connection. The grade of is stored in the data memory 5, and the same processing is performed to check the input character string memory 1 with the dictionary memory 3 and when all (7) word-to-word connections are completed, the contents of the data memory 5 are stored. From the connection relationship between words, a path that can be established as a sentence is investigated and stored in the sentence candidate memory 6-5, and this sentence candidate memory 6-
Regarding the contents of 5, sentences are selected by comparing the number of specified feature words, sentence pattern matching, comparing the number of independent words, and comparing the frequency of appearance.
Stored in display character string memory 8. The contents of this display character string memory 8 are displayed on the display W32. Further, the selected sentence is stored in the file device 42 by the file control circuit 40 as necessary.

The language processing method of the present invention is capable of processing various character strings as input character strings. explain.

FIG. 2 shows the contents of the input character string memory in FIG. 1, and shows, for example, the input character string 60 "He called the pheasant." is stored in the input character string memory. In Figure 2, "shi" 62 and r<]J64 are symbols added before and after the input character string 60 in order to make processing uniform, (8) and r171J66 indicates that there is no information. It is a symbol.

FIG. 3 is a diagram showing an example of the dictionary memory 3 in FIG. 1. The dictionary memory 3 includes dictionary headings 3-1, dictionary notations 3-2, dictionary attributes A3-3, dictionary attributes B5-4. It consists of dictionary feature word designation items 3-5. Dictionary notation 3-2
is written with another character string or symbol string corresponding to the dictionary heading 3-1. Dictionary notation 3-2 is dictionary heading 3-
It may be a notation character string mixed with kanji corresponding to 1, or it may be a phonetic symbol, and it can be set arbitrarily to suit the purpose of horizontal use. The dictionary attribute A3-3 stores classification numbers when words are classified based on their grammatical properties. For example, in the dictionary attribute A3-3, value 21 is the stem of a five-step verb, value 303 is a parallel particle, value 4 is a one-character Japanese noun, value 30 is the adnominal form of the eaves one-step conjugation, and value 6 is a pronoun, value 5 is a Japanese noun with two or more characters, value 1 is a Chinese noun, value 210 is the assertive auxiliary verb "da", value 220
is the past auxiliary verb "da", value 301 is the modal auxiliary (9) word, value 27 is the verb stem of the 5-stage conjugation of the M line, value 26 is the verb stem of the 5-stage conjugation of the B line, and value 9 is the numeral word. , the value 205 is the conjunctive form of the passive auxiliary verb, the value 300 is the storage side, and the value 250 is
The conjunctive conjugation endings of the 5-stage conjugated verbs in the ``ba'' and ``ma'' lines have a value of 10.
00. Value 10o1, value 1002, and value 1003 each have a corresponding symbol ">".

r<]”, r. Represents J, r, J.

Dictionary attribute B5-4 includes, based on the frequency of word appearance, etc.
A numerical value is assigned in advance and stored, and the smaller the numerical value, the higher the appearance frequency.

A value 0 or 1 is stored in advance in the dictionary feature word designation 3-5, and the value 1 is set when the word is a feature word.

The contents stored in the dictionary memory 3 shown in FIGS. 3(a) and 3(b) become part of a dictionary record related to processing the input character string "He called the pheasant." Also, Figure 3 (
The content stored in the dictionary memory 3 in c) is an example of a part of a dictionary record related to processing the input character string "Gabushiki Gaisha".

FIG. 4 is a diagram showing an example of -(10) of the transition table memory 4 in FIG.
4-2, next state 4-3, and transition table connection grade 4-4. Transition table heading A4-1, transition table heading B4-
2. Next state 4-3, the values from transition table connection class 4-4 are n, m, n', Q, respectively. This means that in state n, the word with dictionary attribute A3-3 m is next When this happens, the state moves to state n', indicating that the connection class in state n' is Q. The connection grade 4-4 means that the smaller the value, the higher the probability of connection.

For example, record (#20) in transition table memory 4
For example, the value of transition table heading A4-1 is 1101 (
When the verb stem of the M-line five-stage conjugation, which is indicated by the value 27 in the transition table heading B4-2, comes after the state (indicating a state in which a case particle or dependent young fish is connected after a noun), the value of the next state 4-3 2270 (corresponding to the state in which the verb stem of the M-line five-stage conjugation comes after a case particle or a modal particle), which means that the conjunctive grade 4-4 at that time is 1.

The transition table shown in FIGS. 4(a) and 4(b) is stored in the memory 4 (11) P( The stored content is the input character string ``He called the six.''
This is part of the data in the transition table memory 4 that is referenced when processing.

FIG. 5 is a diagram showing the analysis process 2 of the processing device 10 in FIG. End determination process 2-4, result existence determination process 2-5, and connection process 2-6
It consists of

Analysis processing 2 of the processing device 10 will be explained using FIG.

In the initial setting process 2-1, the input character string memory 1 and the dictionary memory 3 are collated, and the content corresponding to the heading 3-1 of the dictionary record that matches the input character string memory 1 is stored in the data memory 5. FIGS. 6(a) and 6(b) show data memory 5
This is a diagram showing the contents of data heading 5-1 and data notation 5.
-2, data attribute A3-3, data attribute B5-4, data feature word specification 5-5, data status 5-6, data connection grade 5-7, input character string position 5-8, number of characters 5-9, child It consists of the following items: pointer 5-10, presence/absence 5-11, and extended 5(12)-12. The results of the comparison between the input character string memory 1 and the dictionary memory 3 are the dictionary heading 3-1 and dictionary notation 3-2 of the dictionary record (#22) of the dictionary memory 3.
, dictionary attribute A3-3, dictionary attribute B5-4, and dictionary feature word specification 3-5 are the data heading 5-1 and data notation 5-2 of the data record (#1) in the data memory 5.
, data attribute A3-3, data attribute B5-4, and data feature word specification 5-5. Next, data status 5-6 of data record (#1), data connection grade 5
-7, input character string position 5-8, number of characters 5-9, child pointer 5-10, presence/absence 5-11, extended 5-12, 0, 1, 1, 1, respectively. [2], 0°0 is stored. Furthermore, the initial setting process 2-1 includes the input character string position memory 2-7.
The value 2 is stored in the data record number memory 2-8, and the value 1 is stored in the data record number memory 2-8.

The collation process 2-2 collates the contents of the input character string memory 1 after the position indicated by the input character string position memory 2-7 with the dictionary memory 3, creates a new data record, and stores the contents of the dictionary in the dictionary memory 3. The contents of the dictionary record in dictionary memory 3 that matches heading 3-(13)
, data attribute A3-3, data attribute B5-4, and data feature word specification 5-5.

The data records created by the verification process 2-2 after the execution of the initial setting process 2-1 are the portion from data record (n2) to data record (#7) in FIG. 6(a). Next, the matching process 2-2 is performed at the input character string position 5 of the data records (Kan 2) to (#7) that were worked on this time.
-8 stores the value 2 of input character string position memory 2-7,
The character number 5-9 stores the number of characters for the character string of the data heading 5-1, the child pointer 5-10 stores the word, and the number 5-11 stores the value 0 for the last record created this time. , store the value 1 in the other records created this time,
The value 0 is stored in extended 5-12.

Furthermore, the verification process 2-2 includes data record number memory 2
Child of data record (#1) pointed to by -8 'Pointer 5
-10, set the value 2 as the first number of the data record (14) created this time, and also set the value 2 as the starting number of the data record (#1) created this time.
), store the value 1 in extended 5-12.

Note that the dictionary memory 3 to be collated by the collation process 2-2 contains nouns,
It may be divided into a part for independent words such as verbs and a part for attached words such as particles, auxiliary verbs, and conjugated endings.

Next, the connection evaluation process 2-3 uses the value of the data state 5-6 of the data record (ttl) indicated by the data record number memory 2-8 and the data record (#2) to stored this time.
The transition table header A4-1 and the transition table header B4-2 in the transition table memory 4 are collated using each value of the data attribute A3-3 of (#7) as a collation element, and a transition table record where both match is found. The next state 4-3 and transition table connection grade 4-4 are stored in the data state 5-6 and data connection grade 5-7 of (#2) to (#7) of the data record created this time, respectively. Here, when comparing the transition table heading A4-1 and transition table heading B4-2 in the transition table memory 4, if there is no matching transition table record, a value of 0 is set to the data state 5-6 of the corresponding data record. , data (15) - Stores the value 4 in data connection grades 5-7. For example, in the first execution of the connection evaluation process 2-3, the value O of the data state 5-6 of the data record (#1) indicated by the data record number memory 2-8, and the value O of the data state 5-6 of the data record (#2) stored this time.
The transition table heading A4-1 and the transition table heading B4 of the transition table memory 4 are set using the value 21 of the data attribute A3-3 as a matching element.
-2, the value 2210 of the next state 4-3 of the matching transition table record (#7) and the value 1 of the transition table connection grade 4-4 are created this time, and the data of the data record (#2) is created. Stored in state 5-6 and data connection class 5-7. Similar processing is performed for data records (#3) to data records (#7).

In the data record (#3), the value 0.
Since there is no transition table record matching 303, the value 0 is stored in data state 5-6 and the value 4 is stored in data connection grade 5-7.

The extension candidate extraction/termination determination process 2-4 performs the process shown in the flowchart of FIG. First, the extended data record 5-12 in the data memory 5 is already created (16) with the value O. Data connection grade 5-12 is selected from among all completed data records.
Select the data record number of the data record with the smallest value of 7 and store it in the data record number memory 2-8 (
Step 100). If there are multiple applicable data records, the one with the smaller data record number is selected. Next, in order to check whether the end has been reached, the data header 5-1 of the data record pointed to by the data record number memory 2-8 is checked by "<" (step 110), and this data header 5-1 is checked by "<". If not, add the value of the number of characters 5-9 to the value of the input character string position 5-8 of the data record indicated by the data record number memory 2-8, and store the resulting value in the input character string position memory 2-7. Store (step 140). Also, when the data heading 5-1 is "〈", the data record indicated by the data record number memory 2-8 has the same data connection grade 5-7, and the extended data record 5-12
is 0 and the data header 5-1 is other than "<" (step 120). If there is a data record (17) that satisfies these conditions, the data record number is stored in the data record number memory 2-8 (step 130);
Add the value of the number of characters 5-9 to the value of the input character string position 5-8 of the data record indicated by the data record number memory 2-8, and store the resulting value in the input character string position memory 2-7. 1.40), proceed to the next result determination process 2-5.

For example, in the case of the first execution of the extension candidate extraction/completion determination process 2-4, the value 2 is stored in the data record number memory 2-8.
The value 3 is stored in the input character string position memory 2-7, and the process proceeds to result existence determination processing 2-5.

Result existence determination processing 2-5 is performed by data record number memory 2.
-8 indicates a data record whose data status 5-6, data connection grade 5-7, input string position 5-8, number of characters 5-9 are the same, and extended 5-12 is the value 1. Check if there are others, and if there are, connect process 2
Proceed to -6 and find the child pointer 5 of the data record found.
The value of 10 and the extended value of 5-12 are added to the child pointer (18
) After storing in 5-10 and extended 5-12, the extension candidate extraction/end determination process 2-4 is executed. If there is no other data record mentioned above, collation process 2-2 is executed. In the first execution of result existence determination process 2-5, since there is no other data record as described above, matching process 2-2 is performed.
Execute.

A data record (#8) is created by executing the second verification process 2-2, connection evaluation process 2-3, and extension candidate extraction/termination determination process 2-4, and the data record number memory 2
The value 4 is stored in -8 and the value 4 is stored in the input character string position memory 2-7. After the result determination process 2-5, the third series of processes is executed, a data record (#9) is created, the value 5 is stored in the data record number memory 2-8, and the input character string position memory 2- The value 4 is stored in 7.

After the result determination process 2-5, the series of processes described above is executed for the fourth time, a data record (#12) is created from the data record (#10), and the value 6 is stored in the data record number memory 2-8. The value 4 is stored in the input character string position memory 2-7. Favorable result (19) After the process 2-5, the series of processes described above is executed for the fifth time, data records (#13) to (#15) are created, and the value 7 is stored in the data record number memory 2-8. , the value 5 is stored in the input character position memory 2-7. After the result determination process 2-5, the sixth series of processes is executed,
Data records (316) to (318) are created, and the value 11 is stored in the data record number memory 2-8 and the value 5 is stored in the hexagonal character string position memory 2-7. After the result determination process 2-5, the seventh series of processes is executed,
(#21) is created from data record (#19),
The value 12 is stored in the data record number memory 2-8, and the value 5 is stored in the input character string position memory 2-7. After the execution of the seventh extension candidate extraction/end determination process 2-4, the result existence determination process 2-5. Connection processing is executed in the order of 2-6,
Value 19 in child pointer 5-10 of data record (#1)
Data record (#19) by setting
connection is made, and in the 8th time, (#21) has already been created from the data record (319) as an extension light of the data record (#11), so the matching process 2-2, The extension candidate extraction/termination determination process 2-4 is executed without executing the connection evaluation process 2-3, and the data record number memory 2-8 has a value of 13 and the input character string position memory 2-7 has a value of 13. 5 is stored. The above process is repeated until data records (#38) are created.

In the execution of the last extension candidate extraction/completion determination process 2-4, the value 38 is stored in the data record number memory 2-8, and the data header 5-1 of the data record indicated by the data record number memory 2-8 is "<'' (step 110), and furthermore, the data record indicated by the data record number memory 2-8 and the data connection grade 5.
-7 is the same, extended 5-12 is 0 and data heading 5
It is checked whether there is a data record in which -1 is other than "<" (step 12o), and if there is no data record that satisfies these conditions, the analysis process 2 is terminated.

The resulting connection relationship of data records is shown in FIG. 20 (21). In FIG. 20, the numbers indicate data record numbers, and the letters indicate the notation of Day 7 records. Also,
At the branch of the route, the bottom side shows the first one, and the right side shows the child.

Here, FIG. 19 shows another embodiment regarding the extension candidate extraction/termination determination process 2-4. The extension candidate extraction/completion determination process 2-4 extracts the data record with the smallest value of data connection grade 5-7 (step 700), and extracts the data record with the smallest value of data attribute B5-4 from among them. is extracted (step 710), and it is determined whether the extension ends (steps 720 to 730). If it is the end, the next selection process 6 is executed, and if it is not the end, the result existence determination process 2-5 is executed.

FIG. 8 shows the selection process 6 of the processing device 10 in FIG. 1, which consists of an expansion process 6-1, a characteristic process 6-2, a deletion process 6-3, and a comparison process 6-4.

The expansion process 6-1 first stores the first data record (#1) of the data memory 5 on the stack (22) (step 200). Next, check whether the value of extended 5-12 of the data record pointed to by the stack is 1 (step 210), and if the value is 1, store the value of the data record's child interface 5-10 in the stack.Steps 260, 270 ), by repeating this (
Steps 21.0, 260, 270° 280), store the route in the stack. When the data header 5-1 reaches a data record with "<", the contents of the stack, that is, the number string of the corresponding data record, are stored in the character candidate memory 6-5, and the next candidate is found (steps 280, 290).
. If the value of extended 5-12 is 0 (step 210) and the value of 5-11 is 1, the value of extended 5-12 is determined for the next data record (step 22).
0,250,210). If the value of the 5th presence/absence 5-11 is 0, the process ends when it is no longer possible to return to the previous data record (P = 1), and returns when it is possible to return to the previous data record (P = 1).
Step 220°230.240).

FIG. 10 is a diagram (23) showing the contents of the sentence candidate memory 6-5, and each sentence candidate record has data record numbers 6-5-1 to 6-5-99 in the order of forming the sentence. Stored.

Next, as shown in FIG. 11, the feature processing 6-2 performs data feature word designation 5-5 among the data records pointed to by the data record number included in the sentence candidate records on the sentence candidate memory 6-5. The number of data records with value 1 is calculated for each sentence candidate record and stored in the number of feature words 6-5-100 (step 300), leaving the sentence candidate record with the largest number of feature words 6-5-100 and leaving the others. Delete (step 31
0).

In the case of Figure 10, the number of feature words 6-5-100 are all 0.
Therefore, feature processing 6-2 has no effect and all sentence candidate records are left, but for example, "Kabushiki Kaishiya"
In this case, as shown in the dictionary memory 3 of FIG. Designation 20) (Dictionary heading: ka, Dictionary notation: ka, Dictionary attribute A:
303, dictionary attribute B: 1, dictionary feature word specification: 0) (24
) (Dictionary heading: Ishiya, Dictionary notation: Doctor, Dictionary blank A:
5. Dictionary attribute B: 1, dictionary characteristic word specification 20) and (dictionary heading: Kabushiki, dictionary notation: stocks, dictionary attribute A: 1, dictionary attribute B: 1, dictionary characteristic word specification: 0) (dictionary heading: Kabushiki, dictionary notation: stocks, dictionary attribute B: 1, dictionary characteristic word specification: 0) : Kaijiya, Dictionary notation: Company, Dictionary attribute A: 1, Dictionary attribute B
:1, Dictionary feature word specification: 0) and (Dictionary heading: Kabushiki Kaishiya, Dictionary notation: Co., Ltd., Dictionary attribute A: 2, Dictionary attribute B: 1, Dictionary feature word specification: 1). In this case, the feature word processing 6-2 selects the third word with the dictionary feature specification.
The second sentence candidate, ie, "Corporation Corporation" is selected.

Next, the deletion process 6-3 includes the term list memory 6-6 shown in FIG. 13 and the target word list memory 6-7 shown in FIG.
, sentence pattern table memory 7, and collation memory 6 shown in FIG.
-8 is used to perform the processing shown in FIG.

FIG. 13 is a diagram showing an example of the term list memory 6-6, and shows the classification number of the dictionary attribute A3-3 of the word corresponding to the term. For example, class numbers (25) are written that correspond to verb classes such as 5-column verbs, adjectives, adjective verbs, etc.

FIG. 14 is a diagram showing an example of the target word list memory 6-7, in which the character strings written in the dictionary headings 3-1 of particles and particle equivalent words included in the sentence pattern table memory 7 and their dictionary attributes A
A 3-3 value pair is shown.

FIG. 15 is a diagram showing an example of the sentence pattern table memory 7, in which a sentence pattern table record consisting of a sentence pattern table heading 7-1 and a sentence pattern component column 7-3 is stored. The character string of the dictionary entry 3-1 and the value of its dictionary attribute A3-3 are stored in the sentence structure component column 7-3 as the dictionary entry 3-1 of the particle and particle equivalent when the word forms a sentence. The character string and the value of dictionary attribute A3-3 are shown.

The deletion process 6-3 is shown in the flowchart of FIG.

For each sentence candidate record in the sentence candidate memory 6-5 shown in FIG. Step 425), the sentence pattern table record corresponding to the term is taken out from the sentence pattern table memory 7 shown in FIG. 15 and stored in the reference memory 6-8 (step 450).

The contents of the collation memory 6-8 are the same as the sentence pattern table memory 7 shown in FIG. 15, and will therefore be omitted.

Next, the data header 5-1 and data attribute A3- of the data records pointed to by the data record numbers 6-5-1 to 6-5-99 of each sentence candidate record on the sentence candidate memory 6-5 are
3 is included in the target word list memory 6-7 shown in FIG. 14 (step 475). If it is included, check whether the data header 5-1 and data attribute A3-3 of the data record are included in the sentence structure component column 7-3 of the sentence pattern table record on the verification memory 6-8 (step 410). , if there is a matching word, add 1 to the number of matches 6-5-102 in the sentence candidate record, and if there is no matching word, add - to the number of mismatches 6-5-1.01 (step 4
85,495).

For example, in the sentence candidate record (#3) shown in FIG. Corresponding to expression ko (27) 1 do (#2), the sentence structure component column 7-3 of this sentence pattern table record (#2) is ``wa J r301J'' and ``ga J
It is r300J. On the other hand, the data heading 5-1 and data attribute A3-3 of the sentence candidate record (#3) that are included in the target word list memory 6-7 are "ha", "301", and "301" of the data record ($13). ” and data record (Kan 2
6) "wo" is "300". Therefore, "J r301J" in data record (#13) matches, but "J r300J" in data record (#26) does not match "J r300J " in collation memory 6-8, and sentence candidate record (#3) The number of discrepancies 6-5-1.01 is the value 1,
The number of matches 6-5-102 has a value of 1.

If there is a sentence candidate record in which the number of mismatches 6-5-101 has a value of 0, leave that sentence candidate record and create sentences whose number of mismatches 6-5-101 has a value of 1 or more in other sentence candidate records. The candidate record is deleted from the sentence candidate memory 6-5 (steps 500, 505).

If it is not included in the target word list memory 6-7.

Skip the above process.

(28) In FIG. 12, a sentence pattern table record is additionally written to the collation memory 6-8 so that it can be processed even when a single sentence has multiple predicates (step 45).
0).

In the sentence candidate memory 6-5 in FIG. 10, sentence candidate records 3 and 6 are deleted by this deletion process 6-3.

As shown in FIG. 16, the comparison process 6-4 calculates how many of the sentence candidate records in the sentence candidate memory 6-5 include the contents of the independent word list memory 6-9 shown in FIG. Step 600
), leave the sentence candidate record containing the least amount and delete the others (step 610), then select the sentence candidate record with the smallest average value of the data attribute B5-4 of the data record (step 620), the corresponding data record The data representation 5-2 of is stored in the output character string memory 8 (step 630).

Note that the independent word list memory 6-9 in FIG. 17 stores attributes A of independent words such as word classes and verb classes (29).

In this embodiment, the word notation 5-2 is stored in the output character string memory 8, but the information stored in the output character string memory 8 is the content of the sentence candidate memory 6-5 or the information stored in the sentence candidate memory 6-5. Any of the contents of the data memory 5 may be used, and is selected depending on the application of the present invention.

According to this embodiment, by using the present invention in a Japanese word processor, kana character strings can be converted into multiple phrases or phrases without having to break up each phrase or instruct conversion on a phrase-by-clause basis as in the past. Since it is possible to input plain text as it is and convert it into Japanese text mixed with kanji, it is possible to improve the efficiency of Japanese input.

Although this embodiment will be explained using Japanese processing as an example, it goes without saying that the present invention can also be applied to processing foreign languages such as English.

Furthermore, although this embodiment relates to a kana-kanji conversion input section of a Japanese word processor, the present invention also relates to a translation device that converts from one language to another, (30) inputting voice and printing a character string. It is obvious that the present invention can also be applied to voice typewriters and the like.

〔Effect of the invention〕

As is clear from the above description, according to the present invention, it is possible to accurately decompose a character string into its component words, and also to recognize the grammatical properties of the words with high accuracy. Linguistic appropriateness can also be evaluated, improving the accuracy of the kana-kanji conversion input section of word processors, translation devices that convert from one language to another, voice typewriters that print character strings by inputting voice, etc. The effect is that manual corrections are reduced when using the device.

[Brief explanation of drawings]

FIG. 1 is an overall configuration diagram showing an embodiment of the language processing method according to the present invention applied to a Japanese word processor;
Figure 2 is a diagram showing an example of the input character string string memory and its input character book in Figure 1;
4 shows an example of the transition table memory in FIG. 1 (31), FIG. FIG. 7 is a flowchart showing the extension candidate extraction/termination determination processing procedure in FIG. 5, FIG. 8 is a flowchart showing the selection processing procedure, and FIG. A flowchart showing the processing procedure of the expansion process, FIG.
FIG. 11 is a thumb chart showing the processing procedure of the feature processing in FIG. 8, FIG. 12 is a flowchart showing the processing procedure of the deletion processing in FIG. 8, and FIG. FIG. 14 is a diagram showing an example of the target word list memory in FIG. 8; FIG. 15 is a diagram showing an example of the sentence pattern table memory in FIG. 1; FIG. 16 is a flowchart showing the processing procedure of the comparison process in FIG.
7 is a diagram showing an example of the independent word list memory in FIG. 8, FIG. 18 is a diagram showing an example of the output string memory in FIG. 1, and FIG. 19 is a diagram showing an example of the extension candidate extraction/termination determination process in FIG. 5. FIG. 20, a flowchart showing an example of another implementation (32) of the processing procedure, is a diagram showing the connection relationship of data records. 1... Input string memory, 2... Analysis processing, 2-1
...Initial setting processing, 2-2...Verification processing, 2-3.
...Connection evaluation process, 2-4...Extension candidate extraction/end judgment process, 2-5...Result existence judgment process, 2-6...Connection process, 2-7...Input string position Memory, 2-8・
...Data record number memory, 3...Dictionary memory,
3-1...Dictionary heading, 3-2...Dictionary notation, 3-
3...Dictionary attribute A, 3-4...Dictionary attribute B, 3-5
...Dictionary feature word specification, 4...Transition table memory, 4-1
...Transition table heading A, 4-2...Transition table heading B,
4-3...Next state, 4-4...Transition table connection grade,
5...Data memory, 5-1...Data heading, 5
-2...Data notation, 5-3...Data attribute A, 5
-4...Data attribute B, 5-5...Data feature word specification, 5-6...Data status, 5-7...Data continuity grade, 5-8...Input character string position , 5-9...number of characters, 5-10...child pointer, 5-11...th presence/absence,
5-12...Extended, 6...Selection processing, 6-1.
...Development processing, 6-2...Characteristic processing, 6-3...(
33) Deletion processing, 6-4... Comparison processing, 6-5... Candidate memory, 6-5-1 to 6-5-99... Data record number, 6-5-100... Characteristics Word count, 6-5-10
1...Number of mismatches, 6-5-102...Number of matches, 6-
5-103...Number of independent words, 6-6...Word list memory, 6-7...Target word list memory, 6-8...
Verification memory, 6-9... Independent word list memory, 7...
・Sentence pattern table memory, 7-1...Sentence pattern table heading, 7-3・
...Sentence pattern component string, 8...Display character string memory, 10
... Processing device, 50... Pass line, 15... Program memory, 30... Display control circuit, 32...
Display device, middle 20...Keyboard input circuit, 22...Character keyboard, 40...File control circuit, 42.
... Fu (34) I3 Figure CC) Hole/4-Concave Otsu-4 JP-A-Sho GO-191360 (17) ? lq diagram □

Claims (1)

[Claims] 1. A step of creating word information from a dictionary file using dictionary information that matches each word of an input character string, and a step of creating word information from a dictionary file based on a transition table in which the strength of connection relationship is determined in advance. a step of determining the strength of a connection relationship; and a step of selecting a sentence that meets a predetermined criterion from sentence candidates obtained by linking the word information based on the strength of the connection relationship. Processing method. 2. The language processing method according to claim 1, wherein the step of selecting a sentence selects a sentence with the largest number of characteristic words from among the sentence candidates. 3. The step of selecting the sentence includes checking the match between the sentence candidate and sentence pattern information that defines possible particles for a certain term, and deleting sentences that do not match from the sentence candidate. The language processing method described in Scope 1. 4. The language processing method according to claim 1, wherein the step of selecting the sentence selects a sentence that is independent from the sentence candidates and has the least number of words. 5. The step of selecting a sentence is to select the sentence candidate with the most characteristic words from the sentence candidates, check the match between the selected sentence candidate and the sentence pattern information, delete the sentence candidates that do not match the sentence pattern information, and delete the remaining sentence candidates. 2. The language processing method according to claim 1, wherein a sentence candidate having a small number of independent words is selected from the selected sentence candidates, and a sentence candidate having the highest average appearance frequency of each word is selected from the selected sentence candidates.