JP5630138B2 - Sentence creation program and sentence creation apparatus - Google Patents

Description

  The present invention relates to a sentence creation program and a sentence creation apparatus.

  There has been proposed a technique for creating a sentence having a more complex syntax than a simple sentence type sentence when a simple sentence type sentence and a phrase are input.

  As a technology related to this, Patent Document 1 accepts modified clauses that are input separately from the body of the text when the body of the text is input, and parses the body and the modified clause, and modifies them. Multiple candidate phrases that can be modified by the phrase are extracted from the text, the candidate phrase that best matches the modified phrase is selected as the modified phrase, and the optimal insertion position of the modified phrase to qualify the selected modified phrase and A sentence creation device that determines a utilization form and inserts it in the text is disclosed.

JP-A-11-259480

  An object of the present invention is to provide a sentence creation program and a sentence creation apparatus for creating a sentence from a character string when an input character string does not constitute a sentence.

[1]
A receiving means for receiving a character string;
Dividing means for dividing the character string received by the receiving means into words;
Expansion means for expanding the word divided by the dividing means by a predetermined method to generate an extended character string;
A first value is calculated for each extended character string using a value of a semantic preservation rate that is associated in advance with the predetermined method used by the extension means, and the extension means is calculated based on the first value. First estimating means for estimating a degree of coincidence between the meaning of the extended character string expanded by and the meaning of the accepted character string;
An evaluation means for evaluating the validity of the method used by the extension means that generated the extension character string based on the first value;
A sentence creation program for functioning as output means for outputting the extended character string as a sentence candidate created from the accepted character string based on the evaluation result of the evaluation means.

[2] As the predetermined method, the extension means inserts a word between the words divided by the dividing means, changes the utilization form of the divided words, or replaces the divided words with synonyms. The sentence creation program according to [1], in which the extended character string is generated by using a plurality of predetermined methods included in a combination of these predetermined methods in order.

[3] The sentence creation program according to [2], wherein the expansion unit sets an upper limit on the number of the predetermined methods used for the combination of the predetermined methods or the number of the combinations.

[4] The first estimation means estimates the first value by using each value of the semantic preservation rate previously associated with the predetermined method included in the combination combined by the extension means. The sentence creation program according to [2] or [3].

[ 5 ] The extended character string is parsed to calculate a second value, and the computer is further functioned as second estimation means for estimating the likelihood of the extended character string as the syntax based on the second value. Which
The evaluation means evaluates the validity of the method used by the extension means that generates the extended character string based on the first value and the second value, and any one of [1] to [ 4 ] The sentence creation program described in.

[ 6 ] A third value is calculated based on a difference in the number of characters between the extended character string and the character string or the number of the extension methods used by the extension means, and the extended character string is calculated based on the third value. And further causing the computer to function as a third estimating means for estimating the processing cost required for the editing,
Said evaluating means, said first value, to [5] to evaluate the validity of the second value and wherein said expansion means is used that generated the extension string based on said third value The sentence creation program described.

[ 7 ] The dividing unit divides the character string received by the receiving unit into words, and when the divided word partially matches a word included in a word dictionary prepared in advance, The sentence creation program according to any one of [1] to [ 6 ], wherein the divided words are replaced.

[ 8 ] Accepting means for receiving a character string;
Dividing means for dividing the character string received by the receiving means into words;
Expansion means for expanding the word divided by the dividing means by a predetermined method to generate an extended character string;
A first value is calculated for each extended character string using a value of a semantic preservation rate that is associated in advance with the predetermined method used by the extension means, and the extension means is calculated based on the first value. First estimating means for estimating a degree of coincidence between the meaning of the extended character string expanded by and the meaning of the accepted character string;
An evaluation means for evaluating the validity of the method used by the extension means that generated the extension character string based on the first value;
A sentence creation device comprising: output means for outputting the extended character string as a sentence candidate created from the accepted character string based on the evaluation result of the evaluation means.

According to the invention which concerns on Claim 1 or 8, when the character string input does not comprise a sentence, a sentence can be created from a character string.

  According to the invention according to claim 2, as a predetermined method, insertion of a word between divided words, change of a utilization form of the divided word, replacement of the divided word with a synonym Or, and combinations thereof can be used.

  According to the invention of claim 3, an upper limit can be set for the number of generated extended character strings.

  According to the fourth aspect of the invention, the first value can be calculated using a value obtained based on a predetermined method.

According to the invention of claim 5, it is possible to evaluate the validity of the method said expansion means is used that generated the extended string in consideration of the likelihood of a syntax.

According to the sixth aspect of the invention, it is possible to evaluate the validity of the method used by the extension means that generates the extension character string in consideration of the processing cost required for editing.

According to the seventh aspect of the present invention, a sentence can be created from a character string even when an incomplete word is received.

FIG. 1 is a block diagram illustrating a configuration example of a sentence creation device. FIG. 2 is a schematic diagram illustrating an example of semantic preservation rate information. FIG. 3 is a schematic diagram showing an input character string input to the sentence creation device. 4A to 4G are schematic diagrams showing an input character string divided by the character string dividing unit and an extended character string expanded by the character string expanding unit. FIGS. 5A to 5G are schematic diagrams showing the likelihood of the input character string divided by the character string dividing means and the extended character string extended by the character string extending means. 6A to 6F are schematic views showing the evaluation values of the extended character string calculated by the extended character string evaluation means. FIG. 7 is a flowchart illustrating an operation example of the sentence creation device.

(Configuration of sentence creation device)
FIG. 1 is a block diagram illustrating a configuration example of a sentence creation device.

  The sentence creation device 1 includes a CPU and the like, and controls each unit and executes various programs. A storage unit 11 includes a storage medium such as an HDD (Hard Disk Drive) or a flash memory and stores information. And an operation unit 12 such as a keyboard and a mouse, and a display unit 13 such as a liquid crystal display. The sentence creation device 1 is, for example, an electronic device such as a personal computer, a PDA, or a mobile phone, and creates a sentence from an accepted character string. The sentence creation device 1 may be a server device that does not include the operation unit 12 or the display unit 13, in which case the operation unit or display unit of the terminal device connected by a network or the like substitutes for these functions. To do.

  The control unit 10 executes a sentence creation program 110, which will be described later, so that the character string receiving unit 100, the character string dividing unit 101, the character string expanding unit 102, the meaning preservation rate estimating unit 103, the likelihood estimating unit 104, the extended character It functions as column evaluation means 105, sentence candidate output means 106, and the like.

  The character string accepting unit 100 may accept a character string input according to the operation of the operation unit 12 as text information or the like, or may obtain a character string prepared in advance. In addition, you may receive from the outside via the communication part which is not shown in figure.

  The character string dividing unit 101 divides the character string received by the character string receiving unit 100 into words or the like. The character string dividing unit 101 may recognize and divide a specific symbol such as a space included in the input character string as a specific means for dividing the input character string. It may be divided by performing morphological analysis or the like.

  The character string expansion means 102 inserts words and particles between words of the character string divided by the character string dividing means 101, exchanges each word, changes the utilization form of each word, deletes each word, deletes each word The character string is expanded by executing any one of the expansion methods such as changing the order of the above once or combining any expansion method and sequentially executing the methods. Hereinafter, the extended character string is referred to as an “extended character string”. The character string expansion unit 102 generates a plurality of extended character strings by changing the expansion method. Further, the character string expansion unit 102 may determine a predetermined upper limit value for the number of methods used for the combination of expansion methods to be executed, or may determine an upper limit value for the number of combinations of expansion methods. . Moreover, you may combine the method of expansion once or arbitrarily until the meaning preservation | save rate of the whole extended character string mentioned later falls below the predetermined lower limit.

  The meaning storage rate estimation unit 103 stores the meaning of the extended character string generated as a result of the character string expansion unit 102 expanding the character string with respect to the input character string. Estimate the “meaning preservation rate” that indicates the degree of. The meaning preservation ratio of the entire extended character string is calculated from the meaning preservation ratio quantitatively determined in advance for each of the expansion methods executed by the character string expansion means 102 based on the meaning preservation ratio information 111 described later. It is obtained from the product of the semantic preservation rates defined for each of the expansion methods.

  The likelihood estimation unit 104 estimates “likelihood” indicating how likely the syntax is as an extension character string as a sentence in the language. Likelihood may be calculated based on a probabilistic language model or may be calculated based on parsing by a parser. In the present embodiment, calculation is performed using a probabilistic language model based on bi-gram.

  The extended character string evaluation means 105 calculates an evaluation value of the extended character string based on the meaning preservation ratio estimated by the meaning preservation ratio estimation means 103 and the likelihood estimated by the likelihood estimation means 104.

  The sentence candidate output unit 106 outputs sentence candidates from a plurality of extended character strings based on the evaluation value calculated by the extended character string evaluation unit 105.

  The storage unit 11 is a sentence creation program 110 that causes the control unit 10 to operate as each of the above-described units 100 to 106, and a meaning that defines a predetermined semantic storage rate for each method in which the character string expansion unit 102 expands a character string. The storage rate information 111 and the sentence candidate information 112 output by the sentence candidate output unit 106 are stored.

  FIG. 2 is a schematic diagram illustrating an example of the semantic preservation rate information 111.

  The meaning storage rate information 111 includes an extension ID column 111a for identifying each extension method, an extension method column 111b indicating the specific contents of each extension method, and a meaning indicating the meaning storage rate of each extension method determined in advance. And a storage rate column 111c.

(Operation of sentence creation device)
Hereinafter, an operation example of the sentence creation device 1 will be described by dividing into (1) a character string expansion operation and (2) an expansion character string evaluation operation with reference to FIGS.

(1) Character String Expansion Operation First, the user operates the operation unit 12 of the sentence creation device 1 to input a desired character string.

  FIG. 7 is a flowchart illustrating an operation example of the sentence creation device 1.

  The character string accepting unit 100 accepts a character string input according to the operation of the operation unit 12 as text information or the like (S1).

  FIG. 3 is a schematic diagram showing an input character string input to the sentence creation device 1.

  The input character string 100 </ b> A has, for example, the content “kid writing book”, and spaces are inserted between “kid”, “writing”, and “book”, respectively.

  Next, the character string dividing unit 101 recognizes the space included in the input character string from the input character string 100A received by the character string receiving unit 100 and converts it into the words “child”, “write”, and “book”. Divide (S2).

  4A to 4G are schematic diagrams illustrating an input character string 100A divided by the character string dividing unit 101 and an extended character string expanded by the character string expanding unit 102. FIG.

  As shown in FIG. 4A, the character string dividing unit 101 divides the input character string 100A into words 101a to 101c.

  Next, the character string extending unit 102 extends the input character string 100A divided by the character string dividing unit 101 (S3). There are a plurality of methods for extending, and for example, an extended character string as shown below is obtained by the extension.

  As shown in FIG. 4B, the character string expansion unit 102 inserts a particle 102a having “GA” between the words 101a and 101b of the input character string 100A, thereby generating an extended character string 100B.

  As shown in FIG. 4 (c), the character string expansion means 102 inserts the particle 102b of “GA” between the words 101a and 101b of the input character string 100A, and the word 101b of “WRITE”. The expanded character string 100C is generated by replacing with the word 102c that is “written”.

  As shown in FIG. 4D, the character string expansion unit 102 inserts the particle 102d that is “ni” between the words 101a and 101b of the input character string 100A, and the word 101b that is “write”. The extended character string 100D is generated by replacing the word 102e that is “written” with the use.

  As shown in FIG. 4 (e), the character string expansion unit 102 inserts a particle 102f of “GA” between the words 101a and 101b of the input character string 100A, and uses the word 101b of “write”. The word 102g that is “written” is replaced with the word 101c that is “book” and the word 102h that is the synonym “book” is generated to generate the extended character string 100E.

  As shown in FIG. 4 (f), the character string expansion unit 102 inserts the particle 102j with “ga” between the words 101a and 101b of the input character string 100A, and uses the word 101b with “write”. The word 102k that is “written” is replaced with the word 101a that is “child”, and the word 102i that is the word “adult” other than the synonym is replaced to generate the extended character string 100F.

  As shown in FIG. 4G, the character string expansion unit 102 generates the extended character string 100G by inserting particles 102l and 102m having “ga” between the words 101a and 101b of the input character string 100A. .

  Next, the semantic storage rate estimation unit 103 calculates the semantic storage rate for the input character string 100A of the extended character strings 100B to 100G generated as a result of the character string expansion unit 102 extending the character string, based on the semantic storage rate information 111. Estimate (S4).

  As shown in FIG. 4B, since the input character string 100A is expanded by the method of inserting “ga” in the extended character string 100B, “other than independent words” is displayed in the expansion method column 111b of the semantic preservation rate information 111. From the meaning storage rate column 111c, α = 0.95. Since no other method is used, the semantic preservation rate of the entire extended character string 100B is A = 0.95.

  Further, as shown in FIG. 4C, the expanded character string 100C is expanded in the input character string 100A by the method of inserting “ga” and the method of switching to another utilization form. Corresponds to “insertion of words other than self-supporting words” and “replacement to another utilization form after utilization” in the expansion method field 111b of the meaning storage ratio field 111c with α = 0.95 and α = 0.95. Become. From the product of these semantic preservation rates, the semantic preservation rate of the entire extended character string 100C is A = 0.95 × 0.95 = 0.9025.

  By the same calculation method as described above, the semantic preservation rate A of the extended character strings 100D to 100G is obtained to the values shown in FIGS.

  Next, the likelihood estimation means 104 estimates “likelihood” indicating how likely the extended character strings 100B to 100G are as sentences (S5).

  5A to 5G are schematic diagrams illustrating the likelihood of the input character string 100A divided by the character string dividing unit 101 and the extended character string expanded by the character string expanding unit 102. FIG.

  The likelihood β is calculated using a bilingual probabilistic language model. For example, as shown in FIG. 5G, the likelihood of the extended character string 100G is an unnatural continuation of “ga”. Since it is a sentence, β = 0.0000000001, which is a smaller value than the likelihood β of the extended character strings 100B to 100F shown in FIGS.

  Next, the extended character string evaluation means 105 calculates the evaluation value X of the extended character string from the product of the semantic preservation ratio A estimated by the semantic preservation ratio estimation means 103 and the likelihood β estimated by the likelihood estimation means 104 ( S6).

  FIGS. 6A to 6F are schematic views illustrating the evaluation value of the extended character string calculated by the extended character string evaluation unit 105. FIG.

  As shown in FIG. 6A, the evaluation value of the extended character string 100B is calculated as X = A × β = 0.95 × 0.006 = 0.0005 from the product of the semantic preservation rate A and the likelihood β. Is done. By the same calculation method as described above, the evaluation values of the extended character strings 100C to 100G are obtained as the values shown in FIGS.

  The sentence candidate output means 106 outputs sentence candidates from a plurality of extended character strings based on the evaluation value calculated by the extended character string evaluation means 105 (S7). Since a sentence with a large evaluation value stores a meaning and is more likely as a sentence, the sentence candidate output means 106 ranks the sentence candidate information 112 in the order of the extended character strings 100D, 100C, 100B, 100E, 100F, and 100G. And stored in the storage unit 11.

  Further, the sentence candidate output unit 106 may display the sentence candidate information 112 on the display unit 13 with priority given to a sentence having a high evaluation value.

[Other embodiments]
The present invention is not limited to the above embodiment, and various modifications can be made without departing from the gist of the present invention. For example, the estimation of the semantic preservation ratio by the semantic preservation ratio estimation means 103 may not be necessarily based on the meaning preservation ratio information 111 given in advance, but may be calculated dynamically. For example, for the exchange of words by the character string expansion unit 102, the similarity between the word before the exchange and the word after the exchange may be used as the estimated value of the semantic preservation rate. Here, the synonym between words is dynamically calculated by referring to a database storing synonyms together with synonyms for words or using a thesaurus having a network structure describing relationships between words. Such means can be considered.

  The sentence creation device 1 may be combined with a natural sentence search program that performs a natural sentence search using an input sentence (not a word). A user selects a candidate of a sentence having a high probability that the user intends from sentence candidates created by the sentence creation device 1 by inputting a keyword in the natural sentence search, and the natural sentence without inputting the natural sentence. A search can be performed.

  The extended character string evaluation unit 105 calculates the evaluation value of the extended character string based on the editing cost in addition to the semantic preservation rate estimated by the semantic preservation rate estimation unit 103 and the likelihood estimated by the likelihood estimation unit 104. May be. Here, the “editing cost” is a value obtained by subtracting the number of characters of the character string before expansion from the number of characters of the expansion character string (however, it is 0 when the value is negative), or the number of expansion methods to be executed. It is obtained from the value based on Further, the upper limit value determined for the number of methods used for the combination of expansion methods executed by the character string expansion unit 102 or the number of combinations of expansion methods may be determined based on the editing cost.

  In addition, when the character string received by the character string receiving unit 100 includes an incomplete word “Compu” such as “Store Compu”, the character string dividing unit 101 prepares “Compu” in advance. A search such as a forward match may be performed on the dictionary, and a word that matches the search result, for example, “computer” may be treated as a word included in the character string and output to the character string expansion unit 102.

  The sentence creation program 110 can be provided by being stored in a storage medium such as a CD-ROM, and can be downloaded from a server device connected to a network such as the Internet to a storage unit in the device. Good. In addition, part or all of the character string receiving unit 100, the character string dividing unit 101, the character string expanding unit 102, the semantic preservation rate estimating unit 103, the likelihood estimating unit 104, the extended character string evaluating unit 105, and the sentence candidate output unit 106. May be realized by hardware such as ASIC. Note that each step shown in the operation description of the above embodiment can be changed in order, omitted or added.

DESCRIPTION OF SYMBOLS 1 ... Sentence preparation apparatus, 10 ... Control part, 11 ... Memory | storage part, 12 ... Operation part, 13 ... Display part, 100 ... Character string reception means, 100A ... Input character string, 100B-100G ... Extended character string, 101 ... Character Column dividing means, 101a-101c ... word, 102 ... character string sorting means, 102a ... particle, 102b ... particle, 102c ... word, 102d ... particle, 102e ... word, 102f ... particle, 102g ... word, 102h ... word, 102i ... Word, 102j ... Particle, 102k ... Word, 102l ... Particle, 103 ... Meaning preservation rate estimation means, 104 ... Likelihood estimation means, 105 ... Extended character string evaluation means, 106 ... Sentence candidate output means, 110 ... Sentence creation Program, 111... Semantic preservation rate information, 111a. Extended ID column, 111b. Extension method column, 111c. Semantic preservation rate column, 112. Sentence candidate information

Claims (8)

Computer
A receiving means for receiving a character string;
Dividing means for dividing the character string received by the receiving means into words;
Expansion means for expanding the word divided by the dividing means by a predetermined method to generate an extended character string;
A first value is calculated for each extended character string using a value of a semantic preservation rate that is associated in advance with the predetermined method used by the extension means, and the extension means is calculated based on the first value. First estimating means for estimating a degree of coincidence between the meaning of the extended character string expanded by and the meaning of the accepted character string;
An evaluation means for evaluating the validity of the method used by the extension means that generated the extension character string based on the first value;
A sentence creation program for functioning as output means for outputting the extended character string as a sentence candidate created from the accepted character string based on the evaluation result of the evaluation means.   The extension means uses, as the predetermined method, insertion of a word between words divided by the dividing means, change of a utilization form of the divided word, or replacement of the divided word with a synonym. Alternatively, the sentence creation program according to claim 1, wherein the extended character string is generated by sequentially using a plurality of predetermined methods included in a combination of these predetermined methods.   The sentence creating program according to claim 2, wherein the expansion unit sets an upper limit on the number of the predetermined methods or the number of the combinations used for the combination of the predetermined methods. The first estimation unit estimates the first value by using a value of a semantic preservation rate associated with the predetermined method included in the combination combined by the extension unit. The sentence creation program according to 2 or 3. Parsing the extended character string to calculate a second value, and further causing the computer to function as second estimating means for estimating the likelihood of the extended character string as a syntax based on the second value. There,
Said evaluating means, as claimed in any one of claims 1 to 4 for evaluating the validity of the first value and wherein said second value said expansion means generates the extension string based on is used Sentence creation program. A third value is calculated based on the difference in the number of characters between the extended character string and the character string or the number of the expansion methods used by the expansion means, and the extended character string is edited based on the third value. Further causing the computer to function as a third estimating means for estimating the processing cost required,
It said evaluation means, according to claim 5 for evaluating the validity of the method of the first value, the second value and said extension means generates the extension string based on said third value is used Sentence creation program. The dividing unit divides the character string received by the receiving unit into words, and when the divided word partially matches a word included in a word dictionary prepared in advance, the divided word with the matched word The sentence creation program according to any one of claims 1 to 6 , wherein A receiving means for receiving a character string;
Dividing means for dividing the character string received by the receiving means into words;
Expansion means for expanding the word divided by the dividing means by a predetermined method to generate an extended character string;
A first value is calculated for each extended character string using a value of a semantic preservation rate that is associated in advance with the predetermined method used by the extension means, and the extension means is calculated based on the first value. First estimating means for estimating a degree of coincidence between the meaning of the extended character string expanded by and the meaning of the accepted character string;
An evaluation means for evaluating the validity of the method used by the extension means that generated the extension character string based on the first value;
A sentence creation device comprising: output means for outputting the extended character string as a sentence candidate created from the accepted character string based on the evaluation result of the evaluation means.

Images