JP5048001B2 - Concatenated cost adjustment apparatus, method, and program - Google Patents

Description

  The present invention relates to an apparatus, a method, and a program for adjusting a connection cost in a morphological analysis dictionary.

  2. Description of the Related Art Conventionally, a morphological analysis technique is used in which a sentence written in a natural language is divided into word strings, and the part of speech and the utilization form of each word are determined. In this morphological analysis, a sentence is divided into word strings by referring to a morphological analysis dictionary including a word list with part-of-speech information.

  At this time, a method is often employed in which a concatenation cost that represents the difficulty of connecting words as a numerical value is given and a word string that minimizes the sum of the concatenation costs is output as an analysis result (see, for example, Patent Document 1). .

  Furthermore, a method (minimum cost method) is adopted in which an occurrence cost that expresses the difficulty of appearance of the word itself as a numerical value is given and a word string that minimizes the sum of the concatenation cost and the occurrence cost is output as an analysis result. For example, as shown in FIG. 9, when a sentence “Fill until coming” is divided into word strings, a plurality of patterns are extracted. Among these, the combination in which the sum of the concatenation cost (value indicated by the arrow in the figure) and the occurrence cost (value indicated in the frame in the figure) is the minimum (in this example, “7400”) Is output as an analysis result.

  Moreover, the method of morphological analysis is not limited to these. In particular, in a language such as Japanese in which the boundary position of a word is not explicit, it is necessary to perform word boundary identification and part of speech identification at the same time. Under such circumstances, various analysis methods have been proposed so far (see, for example, Non-Patent Document 1), and the accuracy of morphological analysis results has been improved.

  However, in any of the methods, the accuracy of the analysis result depends on the accuracy of the morphological analysis dictionary to be used, and generation and adjustment of the dictionary are important issues. Thus, for example, methods for grouping words and smoothing with statistical values have been proposed (see, for example, Non-Patent Document 2).

JP 2002-157244 A

"Japanese morphological analysis using Conditional Random Fields", Information Processing Society of Japan Research Report, Natural Language Processing Study Group Report, 2004-NL-161- (13) "Extended HMM model for statistical Japanese morphological analysis", Information Processing Society of Japan Research Report, Natural Language Processing Study Group Report, 2000-NL-137-6

  Here, as described above, when the method using the connection cost is adopted, it is desired that the connection cost between the words is registered with high accuracy in the morphological analysis dictionary. Since this concatenation cost is calculated based on a certain structured document data (corpus), an improvement in accuracy can be expected as the amount of the corpus increases.

  However, in order to reduce the labor for preparing a huge corpus and the load of calculation processing, in reality, a sufficient amount cannot be given. If the amount of corpus is not sufficient, there may be a variation in a plurality of concatenated costs that should converge to an equivalent cost.

  Such variation in concatenated cost, even if a known smoothing method is used, the improvement effect still depends on the amount and accuracy of given learning data (corpus). Therefore, there is a possibility that the contents of the morphological analysis dictionary may deviate from the contents of linguistic knowledge known from experience and statistics.

  Therefore, an object of the present invention is to provide an apparatus, a method, and a program capable of leveling statistical variations in connection costs based on language knowledge.

  The present invention provides the following solutions.

(1) In the morphological analysis dictionary used in the minimum cost method, a concatenated cost adjustment device that adjusts the statistical variation of the concatenated cost between words,
Based on a predetermined corpus, calculation means for calculating a concatenation cost between the words and a concatenation cost of a part of speech pair that is a combination of a predetermined part of speech to which each word belongs;
Receiving means for receiving an input for designating a combination of part of speech pairs in which the concatenation cost of the part of speech pairs should be equivalent;
A concatenated cost adjusting device comprising: adjusting means for adjusting concatenated costs between the words so that the concatenated costs of the part-of-speech pairs included in the specified combination are close to each other;

  According to such a configuration, the connection cost adjustment device is a combination of a connection cost between words and a predetermined part of speech to which each word belongs, based on a predetermined corpus in the morphological analysis dictionary used in the minimum cost method. Calculate the concatenation cost of part-of-speech pairs, accept input specifying part-of-speech pair combinations that should have equivalent part-of-speech pair concatenation costs, and use words to bring the concatenation costs of part-of-speech pairs included in the specified combination closer to each other Adjust the connection cost between.

  Thus, the concatenated cost adjustment device accepts as input a combination of part of speech pairs (connection group) that can be given based on existing linguistic knowledge related to the connection between parts of speech and that the concatenated cost of part of speech pairs should be equivalent. Can do. And the said connection cost adjustment apparatus adjusts the connection cost between the words stored in the morphological analysis dictionary by comparing the connection cost of a part-of-speech pair within this connection group. Variations in connection costs can be leveled with language knowledge.

  In addition, the inaccuracy of concatenation costs (joint costs between words and concatenation costs of part-of-speech pairs) due to insufficient amount of corpus can be mitigated by leveling the dispersion of concatenation costs of part-of-speech pairs. There is.

(2) The accepting means further accepts an input designating an adjustment strength of the connection cost between the words by the adjusting means,
The said adjustment means determines the adjustment amount of the connection cost between the said words based on the adjustment intensity received by the said reception means, The connection cost adjustment apparatus as described in (1) characterized by the above-mentioned.

  According to such a configuration, the concatenated cost adjusting apparatus further receives an input designating the adjustment strength of the concatenated cost between words, and determines the adjustment amount of the concatenated cost between words based on the received adjustment strength. .

  Accordingly, the connection cost adjusting apparatus can determine an adjustment amount for leveling the connection cost of the part-of-speech pairs based on the input adjustment strength. Therefore, for example, the leveling level can be varied based on the adjustment strength set for each connection group.

  (3) The adjustment unit multiplies the adjustment strength by a difference between each connection cost of the part of speech pair and an average value of the connection cost of the part of speech pair in the combination received by the reception unit, and The connection cost adjustment device according to (2), wherein an adjustment amount of a connection cost between words is determined.

  According to such a configuration, the connection cost adjusting apparatus can level the connection cost of the part of speech pairs so as to approach the average value of the connection costs of the part of speech pairs. At this time, the average value is approached at the same ratio according to the adjustment strength, and therefore the adjustment can be made without changing the average value.

  (4) The adjustment unit multiplies the adjustment strength by a difference between each of the connection costs between the words and an average value of the connection costs of the part-of-speech pairs in the combination received by the reception unit, The connection cost adjustment device according to (2), wherein an adjustment amount of a connection cost between words is determined.

  According to such a configuration, the connection cost adjusting apparatus can level the connection cost of the part of speech pairs so as to approach the average value of the connection costs of the part of speech pairs. Furthermore, since different adjustment amounts are determined according to the adjustment strength for each of the concatenation costs between words, there is a possibility that variations in the concatenation costs between words within the same part-of-speech pair can be alleviated.

(5) The accepting means further accepts an input for designating a tolerance of the joint cost of the part-of-speech pairs that does not require adjustment by the adjusting means,
The adjusting means does not adjust the concatenated cost between the words related to the part-of-speech pair when the concatenated cost of the part-of-speech pair included in the combination received by the receiving means is within the tolerance from the average value in the combination. The connected cost adjusting apparatus according to any one of (1) to (4), wherein:

  According to such a configuration, the concatenated cost adjusting apparatus further accepts an input designating a tolerance of the concatenated cost of the part-of-speech pair that does not require adjustment, and the concatenated cost of the part-of-speech pair included in the received combination is within the combination. If the average value is within the tolerance, the concatenation cost between words related to the part-of-speech pair is not adjusted.

  As a result, the concatenated cost adjusting apparatus accepts the tolerance of the concatenated cost of part-of-speech pairs that do not need to be adjusted, and does not adjust the concatenated cost of parts-of-speech pairs and the concatenated cost between words by allowing variations within this tolerance. . Therefore, it is possible to achieve leveling that can withstand practical use while reducing unnecessary processing load in a connection group that does not require the connection costs of part-of-speech pairs to be exactly the same.

(6) In a morphological analysis dictionary used by a computer in a minimum cost method, a method of adjusting statistical variation in concatenated costs between words,
Based on a predetermined corpus, a calculation step of calculating a concatenation cost between the words and a concatenation cost of a part of speech pair that is a combination of a predetermined part of speech to which each word belongs;
An accepting step of accepting an input designating a combination of part of speech pairs in which the concatenation cost of the part of speech pairs should be equivalent;
Adjusting the concatenation cost between the words so that the concatenation costs of the part-of-speech pairs included in the specified combination are close to each other.

  According to such a configuration, the same effect as in (1) can be expected by executing the method.

  (7) A program that causes a computer to execute the method according to (6).

  According to such a configuration, the same effect as in (1) can be expected by causing the computer to execute the program.

  According to the present invention, it is possible to level the statistical variation of the connection cost in the morphological analysis dictionary using language knowledge.

It is a figure showing the whole morphological analysis system composition concerning an embodiment of the present invention. It is a figure which shows the hardware constitutions of the dictionary adjustment server which concerns on embodiment of this invention. It is a figure which shows the function structure of the dictionary adjustment server which concerns on embodiment of this invention. It is a figure which shows the connection cost statistics table which concerns on embodiment of this invention. It is a figure which shows the connection group table which concerns on embodiment of this invention. It is a figure which illustrates the method of determining the adjustment amount of the connection cost which concerns on embodiment of this invention. It is a flowchart which shows the process of the control part which concerns on embodiment of this invention. It is a flowchart which shows the process which adjusts the connection cost between the words which concern on embodiment of this invention. It is a figure explaining the minimum cost method concerning a prior art.

  Hereinafter, an example of an embodiment of the present invention will be described with reference to the drawings.

[Entire system configuration]
FIG. 1 is a diagram showing an overall configuration of a morphological analysis system 1 including a dictionary adjustment server 10 (concatenated cost adjustment device) and related elements according to the present embodiment. The morpheme analysis system 1 includes a dictionary adjustment server 10 and a morpheme analysis server 20, and each server cooperates via a dictionary DB 40.

  The dictionary adjustment server 10 receives the corpus DB 30 as an input, calculates a concatenated cost between words, and registers it as dictionary data for morphological analysis in the dictionary DB 40. At this time, in order to improve processing efficiency, the amount of the corpus DB 30 is often kept small. As a result, the reliability of the calculated connection cost is not sufficient. Therefore, the dictionary adjustment server 10 adjusts registration data in the dictionary DB 40 based on a connection group definition described later provided from the outside, and improves the accuracy of the morpheme analysis processing by the morpheme analysis server 20.

  The morphological analysis server 20 refers to the dictionary DB 40 registered and adjusted in this way, decomposes the input document into word strings, and outputs the result as an analysis result. The morphological analysis server 20 is connected to a communication line such as the Internet and provides a service in response to a request from an information terminal such as a PC (Personal Computer) or a PDA (Personal Digital Assistant).

  The morphological analysis system 1 is described as a configuration including two servers and the dictionary DB 40, but is not limited to this, and may be integrated into the same server as appropriate. Alternatively, the servers may be further distributed for each function.

[Hardware configuration]
FIG. 2 is a diagram illustrating a hardware configuration of the dictionary adjustment server 10 according to the present embodiment. The dictionary adjustment server 10 includes a CPU (Central Processing Unit) 310 (a plurality of CPUs such as a CPU 320 may be added in a multiprocessor configuration), a bus line 200, and a communication I / F (I / F: An interface) 330, a main memory 340, a BIOS (Basic Input Output System) 350, an I / O controller 360, a hard disk 370, an optical disk drive 380, and a semiconductor memory 390. The hard disk 370, the optical disk drive 380, and the semiconductor memory 390 are collectively referred to as a storage device 410.

  The control unit 300 is a part that controls the dictionary adjustment server 10 in an integrated manner, and appropriately reads and executes various programs stored in the hard disk 370 (described later), thereby cooperating with the hardware described above. Various functions related to are realized.

  The communication I / F 330 is a network adapter when the dictionary adjustment server 10 transmits / receives information to / from the morphological analysis server 20 of FIG. 1 and other information terminals via the network. The communication I / F 330 may include a modem, a cable modem, and an Ethernet (registered trademark) adapter.

  The BIOS 350 records a boot program executed by the CPU 310 when the dictionary adjustment server 10 is started, a program depending on the hardware of the dictionary adjustment server 10, and the like.

  A storage device 410 such as a hard disk 370, an optical disk drive 380, and a semiconductor memory 390 can be connected to the I / O controller 360.

  The hard disk 370 stores various programs for causing the hardware to function as the dictionary adjustment server 10, programs for executing the functions of the present invention, and the like. The dictionary adjustment server 10 can also use an external hard disk (not shown) as an external storage device.

  As the optical disk drive 380, for example, a DVD-ROM drive, a CD-ROM drive, a DVD-RAM drive, or a CD-RAM drive can be used. In this case, the optical disk 400 corresponding to each drive is used. A program or data can be read from the optical disk 400 by the optical disk drive 380 and provided to the main memory 340 or the hard disk 370 via the I / O controller 360.

  The computer in the present invention refers to an information processing device including a storage device, a control unit, and the like. The dictionary adjustment server 10 includes an information processing device including a storage device 410, a control unit 300, and the like. The information processing apparatus is included in the concept of the computer of the present invention.

[Function configuration]
FIG. 3 is a diagram illustrating a functional configuration of the dictionary adjustment server 10 according to the present embodiment. The control unit 300 of the dictionary adjustment server 10 includes a cost calculation unit 11 (calculation unit), a connection group reception unit 12 (reception unit), and a cost adjustment unit 13 (adjustment unit). DB14 and connection group DB15 are provided.

  The cost calculation unit 11 reads dictionary data from the dictionary DB 40, and further calculates and updates the concatenation cost between words related to the connection between words based on the corpus DB 30. Furthermore, the cost calculation unit 11 calculates the concatenation cost of the part-of-speech pair related to the connection between the parts of speech by statistically processing the concatenation cost between the generated or updated words. The dictionary DB 40 stores concatenated costs between words. The cost calculation unit 11 classifies these words into one of a plurality of parts of speech set in advance, and calculates a statistical value for each part of speech. Specifically, for example, an average value of connection costs between words is calculated. The calculated statistical value (joint cost of part-of-speech pairs) is stored in the statistical DB 14.

  FIG. 4 is a diagram showing a concatenated cost statistical table in which the statistical value of the concatenated cost between words according to the present embodiment, that is, the concatenated cost of the part of speech pairs is stored. The concatenated cost statistical table is stored in the statistical DB 14 and stores a statistical value (for example, an average value) of concatenated costs between words classified by part of speech and their utilization forms. For example, the statistical value of the concatenation cost (the concatenation cost of a part-of-speech pair) in the case of a part-of-speech pair in which “sentence” is followed by “named self” is “1377”. Similarly, “2052” is displayed when “Noun” is followed by “sentence”, and “975” is obtained when “noun” is followed by “adjective basic form”.

  The connection group reception unit 12 receives an input of a combination (connection group) of a plurality of part of speech pairs whose connection cost of the part of speech pair should have an equivalent value, and stores the combination in the connection group DB 15. The connection group is designated by linguistic knowledge based on empirical rules and past statistical information, and the statistical variation of the connection cost in the dictionary DB 40 is adjusted based on this connection group.

  FIG. 5 is a diagram showing a connection group table in which connection groups according to the present embodiment are stored. The connection group table is stored in the connection group DB 15 and stores combinations of part-of-speech pairs (connection groups) in which the concatenated cost of part-of-speech pairs should be equivalent to the identification ID (connection group ID), strength, and tolerance. .

  Here, the strength indicates the degree to which the connection cost of part-of-speech pairs is brought closer to each other by adjustment, and the amount of adjustment increases as the strength value increases. In addition, the tolerance indicates an allowable range regarding the variation in the connection cost of the part of speech pairs within the connection group. If the difference from the average of the connection costs of the part of speech pairs is within this range, the adjustment for the part of speech pair is not performed. It is unnecessary.

  The cost adjustment unit 13 refers to the connection group table (FIG. 5), and determines the concatenation cost between words stored in the dictionary DB 40 so that the concatenation costs of the part of speech pairs are brought close to each other for the part of speech pairs in each connection group. adjust. At this time, the cost adjustment unit 13 determines the adjustment amount of the concatenated cost between words in each part-of-speech pair based on the above-described strength and tolerance.

  FIG. 6 is a diagram illustrating a method for determining the adjustment amount of the connection cost according to the present embodiment. In this example, three part-of-speech pairs A, B, and C are set in the connection group. Then, the average of the concatenated cost of part-of-speech pairs is separated in the order of part-of-speech pairs B, C, and A, and the concatenated cost of part-of-speech pair B is within the tolerance from the average.

  In this case, the part-of-speech pair A and the part-of-speech pair C are adjusted in a direction approaching the average value at a ratio according to the designated strength (for example, “0.5”). However, the part-of-speech pair B is within the tolerance, so no adjustment is made and the concatenation cost of the part-of-speech pair is maintained.

[Processing flow]
FIG. 7 is a flowchart showing processing of the control unit 300 in the dictionary adjustment server 10 according to the present embodiment.

  In step S1, the control unit 300 calculates a concatenated cost between words based on the corpus DB 30, and updates the dictionary DB 40.

  In step S2, the control unit 300 reads dictionary data including concatenated cost information between words from the dictionary DB 40, statistically processes the concatenated cost between words for each of the classified part of speech pairs, and calculates the concatenated cost of the part of speech pairs. To do.

  In step S3, the control unit 300 determines whether a connection group is designated. Specifically, it is determined whether or not the user (administrator) of the dictionary adjustment server 10 selects any or all of the connection group IDs registered in the connection group DB 15 and inputs an adjustment instruction. To do. If this determination is YES, the process proceeds to step S4, and if the determination is NO, the process ends.

  In step S4, the control unit 300 adjusts the concatenation cost between words in the dictionary data for the connection group specified in step S3 (detailed in FIG. 8). Thereafter, the control unit 300 repeats calculation and adjustment using the corpus DB 30 or another corpus as an input in response to a request from the user who confirmed the adjustment result. Thus, by repeating a series of processes, it is possible to converge to a more reliable connection cost.

  FIG. 8 is a flowchart showing a process (step S4 in FIG. 7) of adjusting the concatenation cost between words in the dictionary data among the processes of the control unit 300 in the dictionary adjustment server 10 according to the present embodiment. This process is executed for each designated connection group.

  In step S11, the controller 300 calculates the average value of the concatenated cost of part-of-speech pairs in the connection group to be adjusted designated in step S3 (FIG. 7).

  In step S12, the control unit 300 selects one part-of-speech pair as a target for adjusting the connection cost from the connection group to be adjusted.

  In step S13, the control unit 300 refers to the concatenated cost statistics table (FIG. 4), and acquires the concatenated cost of the part-of-speech pair selected in step S12. Then, the control unit 300 calculates a difference between the acquired concatenated cost of part-of-speech pairs and the average value calculated in step S11, and determines whether this difference is within an allowable difference. If this determination is YES, the process proceeds to step S15, and if the determination is NO, the process proceeds to step S14.

  In step S14, since it is necessary to adjust the concatenation cost between words related to the selected part of speech, the control unit 300 determines the difference between the concatenation cost of the part of speech pair calculated in step S13 and the average value. An adjustment amount of a predetermined ratio (strength) is determined. And the control part 300 updates the connection cost between the words stored in dictionary DB50 which concerns on the selected part of speech pair with the determined adjustment amount.

  In step S15, it is determined whether or not adjustment has been performed by selecting all the part-of-speech pairs in the designated connection group (step S12). If this determination is YES, the process ends. If the determination is NO, the process returns to step S12 to select the next part-of-speech pair.

  As described above, according to the present embodiment, by accepting the designation of the connection group, the statistical variation of the concatenation cost of the part of speech pairs in the connection group can be leveled by language knowledge. That is, it is possible to alleviate connection costs and inconsistencies with low reliability resulting from a shortage of corpus, etc., and improve the accuracy of morphological analysis.

  As mentioned above, although embodiment of this invention was described, this invention is not restricted to embodiment mentioned above. The effects described in the embodiments of the present invention are only the most preferable effects resulting from the present invention, and the effects of the present invention are limited to those described in the embodiments of the present invention. is not.

  In the above-described embodiment, the connection cost of a part-of-speech pair is compared with the average value in the connection group, and the adjustment amount of the connection cost between words included in the part-of-speech pair is collectively (as the same value). ) Decided, but not limited to this. For example, the amount of adjustment may be determined for each concatenated cost between words by comparing the concatenated cost between each word and the average value of the concatenated cost of part-of-speech pairs in the connection group. Thereby, the connection cost between words in dictionary data can be finely adjusted.

  In the above-described embodiment, the concatenated cost between words is adjusted for all the part-of-speech pairs included in the connection group and all the word pairs included in the part-of-speech pairs. Absent. For example, a word pair whose concatenation cost between words is not less than a predetermined value may be excluded from the calculation of the concatenation cost of the part of speech pair, or a part of speech pair whose concatenation cost of the part of speech pair is not less than the predetermined value may be excluded from the average value calculation. Good. As a result, it is possible to prevent the accuracy of the adjustment result from being lowered due to the influence of data with particularly low reliability.

  Further, in this case, the concatenated cost between words that are not targeted may be replaced with the average value of the concatenated cost of part-of-speech pairs calculated during the adjustment process or the concatenated cost of the part-of-speech pair after adjustment. As a result, the concatenation cost between words with low reliability can be adjusted to a more probable value.

  In the above-described embodiment, the calculation result by the cost calculation unit 11 is written in the dictionary DB 40 and further updated by the cost adjustment unit 13, but the present invention is not limited to this. For example, the calculation result or dictionary data including the calculation result may be temporarily stored in the dictionary adjustment server 10 and reflected in the dictionary DB 40 after adjustment.

  Moreover, although the above-mentioned embodiment demonstrated as what the connection cost between words was defined in dictionary DB50, it is not restricted to this. For example, the present invention can be applied even to a dictionary that defines only the joint cost of a part of speech pair. In this case, the concatenation cost of part-of-speech pairs is the concatenation cost in the dictionary data itself, and the dictionary data includes the adjustment amount determined from the difference from the average of the concatenation costs of the part-of-speech pairs statistically processed using the corpus DB 30. Update the concatenation cost of part-of-speech pairs.

DESCRIPTION OF SYMBOLS 1 Morphological analysis system 10 Dictionary adjustment server 11 Cost calculation part 12 Connection group reception part 13 Cost adjustment part 14 Statistics DB
15 Connection group DB
20 Morphological analysis server 30 Corpus DB
40 Dictionary DB
300 Control unit 410 Storage device

Claims (7)

In the morphological analysis dictionary used in the minimum cost method, a concatenated cost adjustment device for adjusting the statistical variation of the concatenated cost between words,
Based on a predetermined corpus, calculation means for calculating a concatenation cost between the words and a concatenation cost of a part of speech pair that is a combination of a predetermined part of speech to which each word belongs;
Receiving means for receiving an input for designating a combination of part of speech pairs in which the concatenation cost of the part of speech pairs should be equivalent;
Adjusting means for adjusting the concatenation cost between the words so that the concatenation cost of the part-of-speech pairs included in the designated combination is close to each other ;
The accepting means further accepts an input designating an adjustment strength of the connection cost between the words by the adjusting means;
The adjustment means multiplies the difference between each of the connection costs of the part of speech pairs and the average value of the connection costs of the part of speech pairs within the combination received by the reception means by the adjustment strength, and Determine the amount of adjustment for articulated costs
A connected cost adjusting device characterized by the above . In the morphological analysis dictionary used in the minimum cost method, a concatenated cost adjustment device for adjusting the statistical variation of the concatenated cost between words,
Based on a predetermined corpus, calculation means for calculating a concatenation cost between the words and a concatenation cost of a part of speech pair that is a combination of a predetermined part of speech to which each word belongs;
Receiving means for receiving an input for designating a combination of part of speech pairs in which the concatenation cost of the part of speech pairs should be equivalent;
Adjusting means for adjusting the concatenation cost between the words so that the concatenation cost of the part-of-speech pairs included in the designated combination is close to each other ;
The accepting means further accepts an input designating an adjustment strength of the connection cost between the words by the adjusting means;
The adjusting means multiplies the difference between each of the concatenated costs between the words and the average value of the concatenated cost of the part-of-speech pairs within the combination received by the receiving means by the adjustment strength, and Determine the amount of adjustment for articulated costs
A connected cost adjusting device characterized by the above . The accepting means further accepts an input for designating a tolerance of the joint cost of the part-of-speech pairs that does not require adjustment by the adjusting means;
The adjusting means does not adjust the concatenated cost between the words related to the part-of-speech pair when the concatenated cost of the part-of-speech pair included in the combination received by the receiving means is within the tolerance from the average value in the combination. articulating cost adjustment device according to any one of claims 2 to claim 1, characterized in that. In a morphological analysis dictionary used by a computer in a minimum cost method, a method of adjusting statistical variation of concatenated costs between words,
Based on a predetermined corpus, a calculation step of calculating a concatenation cost between the words and a concatenation cost of a part of speech pair that is a combination of a predetermined part of speech to which each word belongs;
An accepting step of accepting an input designating a combination of part of speech pairs in which the concatenation cost of the part of speech pairs should be equivalent;
The articulation cost of parts of speech pair is included in the specified combination as close to each other, seen including a an adjustment step of adjusting the articulation costs between said word,
The accepting step further accepts an input designating an adjustment strength of the concatenated cost between the words in the adjusting step,
The adjustment step multiplies the adjustment strength by a difference between each of the connection costs of the part of speech pairs and an average value of the connection costs of the part of speech pairs in the combination received in the reception step, The amount of connection cost adjustment for
A method characterized by that . The program which makes a computer perform the method of Claim 4 . In a morphological analysis dictionary used by a computer in a minimum cost method, a method of adjusting statistical variation of concatenated costs between words,
Based on a predetermined corpus, a calculation step of calculating a concatenation cost between the words and a concatenation cost of a part of speech pair that is a combination of a predetermined part of speech to which each word belongs;
An accepting step of accepting an input designating a combination of part of speech pairs in which the concatenation cost of the part of speech pairs should be equivalent;
The articulation cost of parts of speech pair is included in the specified combination as close to each other, seen including a an adjustment step of adjusting the articulation costs between said word,
The accepting step further accepts an input designating an adjustment strength of the concatenated cost between the words in the adjusting step,
The adjustment step multiplies the difference between each connection cost between the words and the average connection cost of the part-of-speech pairs within the combination received in the reception step by the adjustment strength, The amount of connection cost adjustment for
A method characterized by that . The program which makes a computer perform the method of Claim 6 .

Images