Model generation apparatus, method, and program

Abstract

This model generation device is provided with: a storage unit 11 that stores a word degree-of-familiarity database which receives, as input, a plurality of test words presented to a user and the user's answers regarding knowledge of the test words, and which stores a plurality of words and a plurality of degrees of familiarity associated respectively with the plurality of words, where a degree of familiarity is a measure of familiarity with a word; and a model generation unit 15 that obtains a model representing the relationship between a value based on the degree of familiarity associated with each test word and a value based on the probability that the user will answer that the user knows the test word, using the answers regarding the knowledge of the test words and the word degree-of-familiarity database stored in the storage unit 11.

Description

[Technical Field]

[0001] The disclosed technology relates to the technology for generating the model. [Background technology]

[0002] The total number of words a person knows is called their vocabulary size. A vocabulary size estimation test is a test that estimates that vocabulary size accurately in a short amount of time (see, for example, Non-Patent Document 1). The following is an outline of the estimation procedure.

[0003] (1) Select test words from the word familiarity database (DB) in order of familiarity, so that they are spaced at approximately equal intervals (for example, select one word every 1000 words). Familiarity (word familiarity) is a numerical representation of how familiar a word is. A higher familiarity indicates that the word is more familiar.

[0004] (2) Present the test words to the user and ask them to answer whether or not they know the words.

[0005] (3) Logistic regression analysis is performed to best explain these combinations of test words and answers. In this logistic regression analysis, the independent variable x is the total number of words in the word familiarity DB with a familiarity level equal to or higher than the familiarity level of each test word, and the dependent variable y is the probability that the user will answer that they know each word (e.g., 0 or 1). A logistic model (or logistic regression equation) is obtained as a result of the logistic regression analysis. An example of a logistic model is shown in Figure 12.

[0006] (4) In the logistic model obtained, find the value of x corresponding to y=0.5 and use this as the estimated vocabulary size. The estimated vocabulary size refers to the value estimated to be the user's vocabulary size.

[0007] This method uses a word familiarity database to accurately estimate a user's vocabulary size simply by testing whether or not they know selected test words. [Prior art documents] [Non-patent literature]

[0008] [Non-Patent Document 1] Tetsuo Kobayashi, Shigeaki Amano, and Nobuo Masataka, "The Current State and Future of Mobile Society," 2007, NTT Publishing, pp. 127-128. [Overview of the Initiative] [Problems that the invention aims to solve]

[0009] As shown in Figure 11, the number of words corresponding to each level of intimacy is not the same. In other words, the number of words varies depending on the level of intimacy.

[0010] Therefore, when the independent variable x in the logistic model is the total number of words in the word familiarity database that are more familiar than each test word, x tends to increase sharply around familiarity levels where many words are concentrated. In the example in Figure 11, x increases sharply around familiarity level 5 and around familiarity level 3.

[0011] As a result, conventional methods may not be able to generate an appropriate logistic model, potentially leading to failures or inaccurate estimations of the user's vocabulary size.

[0012] The disclosed technology aims to generate a model that can robustly and accurately estimate the user's vocabulary size. [Means for solving the problem]

[0013] One aspect of the disclosed technology is a model generation device which takes a plurality of test words and a user's response regarding their knowledge of the test words as input, where familiarity is an indicator representing familiarity with the words, and stores a word familiarity DB which stores a plurality of words and a plurality of familiarity values ​​corresponding to each of the plurality of words, and uses the response regarding knowledge of the test words and the word familiarity DB stored in the storage unit to generate familiarity corresponding to the test words or its non-monotonically decreasing function valueand the probability that the user answers that they know the test word or its non-monotonically decreasing function value A model representing the relationship with For each of the aforementioned test words, the familiarity value obtained from the word familiarity DB or its non-monotonic decreasing function value is used as the independent variable, and the value obtained by converting the answer regarding knowledge of the test word into a predetermined probability value is used as the dependent variable. Based on the relationship between the multiple associated independent and dependent variables, statistical regression analysis is performed. A model generation unit that obtains is provided.

Advantages of the Invention

[0014] According to the disclosed technology, it is possible to generate a model that can robustly and accurately estimate the vocabulary size of the user.

Brief Description of the Drawings

[0015] [Figure 1] FIG. 1 is a diagram showing an example of the functional configuration of a model generation device and a word selection device. [Figure 2] FIG. 2 is a diagram showing an example of the processing procedure of a model generation method and a word selection method. [Figure 3] FIG. 3 is a diagram showing an example of a logistic regression model. [Figure 4] FIG. 4 is a diagram showing an example of the functional configuration of an acquisition probability acquisition device. [Figure 5] FIG. 5 is a diagram showing an example of the processing procedure of an acquisition probability acquisition method. [Figure 6] FIG. 6 is a diagram for explaining an example of the generation of acquired word information. [Figure 7] FIG. 7 is a diagram showing an example of the functional configuration of a learning recommended word extraction device. [Figure 8] FIG. 8 is a diagram showing an example of the processing procedure of a learning recommended word extraction method. [Figure 9] FIG. 9 is a diagram showing an example of learning recommended words. [Figure 10] FIG. 10 is a diagram showing an example of the functional configuration of a computer. [Figure 11] FIG. 11 is a diagram showing an example of the correspondence between intimacy and the number of words. [Figure 12] FIG. 12 is a diagram for explaining the background art.

Modes for Carrying Out the Invention

[0016] Embodiments of the disclosed technology will be described below with reference to the drawings.

[0017] [First Embodiment] First, the first embodiment will be described. The first embodiment is a model generation apparatus and method, and a word generation apparatus and method.

[0018] As illustrated in Figure 1, the model generation device 1 of this embodiment includes a storage unit 11, a word selection unit 12, a presentation unit 13, a response receiving unit 14, a model generation unit 15, and a vocabulary size estimation unit 16. The model generation device 1 does not necessarily have to include the word selection unit 12, the presentation unit 13, the response receiving unit 14, the storage unit 11, and the vocabulary size estimation unit 16.

[0019] As shown by the dashed lines in Figure 1, the word generation device A1 is composed of a memory unit 11 and a word selection unit 12. The word generation device A1 may also include a presentation unit 13 and a response receiving unit 14.

[0020] <Storage section 11> The memory unit 11 has a pre-stored intimacy database (DB). The word intimacy DB is a database that stores pairs of M words (multiple words) and a predetermined intimacy level (word intimacy level) for each of those words. In other words, a word intimacy DB is stored that contains multiple words and multiple intimacy levels corresponding to each of those words.

[0021] The M words in the word familiarity database are ranked in order of familiarity (e.g., by familiarity). M is an integer greater than or equal to 2, representing the number of words included in the word familiarity database. There is no limit to the value of M, but for example, when measuring the vocabulary size of a native language, M should be 70,000 or more, and when measuring the vocabulary size of a second language (e.g., English for a native Japanese speaker), M should be 10,000 or more. This is because the vocabulary size of an adult Japanese person is said to be around 40,000 to 50,000 words, so around 70,000 words should cover the vocabulary of most people, including individual differences. On the other hand, in the case of a second language, the vocabulary size is often less than that of a native language, so it is thought that a smaller number of words than M for a native language can cover the vocabulary of most people. However, the number of words in a second language can vary greatly depending on how vocabulary is counted, such as how variations in spelling and derived words are handled. Therefore, depending on how vocabulary is counted, M may need to be 100,000 or more for a native language. Also, the estimated vocabulary size is limited by the number of words included in the standard word familiarity database. Therefore, when performing vocabulary estimation for individuals with large vocabulary sizes, which may be outliers, it is desirable to increase the value of M.

[0022] Word familiarity is an indicator that represents the level of familiarity with a word. Examples of indicators that represent word familiarity include: an indicator that represents how well a word is known (for example, the numerical representation of word familiarity introduced in Non-Patent Document 1); an indicator that represents how often a word is seen or heard; an indicator that represents how well a word is known; an indicator that represents how well a word can be written; and an indicator that represents how well a word can be used in conversation.

[0023] For example, words with a higher degree of familiarity are considered more intimate. In this embodiment, a higher numerical value representing familiarity indicates a higher degree of familiarity. However, this does not limit the present invention.

[0024] The memory unit 11 receives read requests from the word selection unit 12 and the model generation unit 15 as input and outputs the word corresponding to the request and the intimacy level of that word.

[0025] <Word Selection Section 12> Input: Problem generation request from user or system Output: N test words used for the vocabulary size estimation test. When the word selection unit 12 receives a question generation request from a user or system, it selects and outputs several test words w(1), ..., w(N) to be used in the vocabulary size estimation test from a set of ordered words contained in the word familiarity DB of the memory unit 11.

[0026] For example, the word selection unit 12 uses the word familiarity DB stored in the memory unit 11 to select multiple test words w(1), ..., w(N) from multiple words such that the intervals between familiarity levels corresponding to the test words are constant (step S12).

[0027] For example, the word selection unit 12 selects N words evenly from all the words included in the word familiarity DB of the memory unit 11 so that the familiarity of the selected words is at approximately constant intervals, and outputs the selected N words as test words w(1), ..., w(N).

[0028] For example, the word selection unit 12 selects words such that the intimacy interval is 0.1. For example, the word selection unit 12 may select a total of 61 words, such as word w(1) with intimacy level 1, word w(2) with intimacy level 1.1, ..., word w(60) with intimacy level 6.9, and word w(61) with intimacy level 7.

[0029] The familiarity levels of the test words w(1), ..., w(N) do not necessarily have to be at regular intervals; they just need to be selected evenly. For example, if familiarity levels around the boundary between whether a user knows the words or not are predicted based on past research, it is acceptable to select more words around the familiarity levels that you want to investigate intensively. In other words, there may be variations in the familiarity values ​​of the series of test words w(1), ..., w(N).

[0030] There is no restriction on the order of the test words w(1), ..., w(N) output from the word selection unit 12, but the word selection unit 12 outputs the test words w(1), ..., w(N) in order of increasing familiarity, for example.

[0031] The number of test words N may be specified by the question generation request or predetermined. There is no limit to the value of N, but for example, 50 ≤ N ≤ 100 is desirable. For sufficient estimation, N ≥ 25 is desirable. A larger N allows for more accurate estimation, but increases the burden on the user (subject) (step S12).

[0032] To reduce the burden on users and improve accuracy, for example, a test could be conducted multiple times (e.g., three times) with 50 words each, estimating the vocabulary size for each test, and then re-estimating the vocabulary size based on the responses from all tests combined. In this case, the number of words in each test can be reduced, lessening the burden on users, and providing results for each test helps maintain user motivation. Furthermore, combining the words from multiple tests for the final vocabulary size estimation can improve estimation accuracy.

[0033] By selecting multiple test words such that the intervals between their corresponding levels of familiarity are constant, the variability in familiarity can be reduced, making the logistic curve more likely to converge.

[0034] <Presentation part 13> Input: N test words Output: Instructions and N test words The presentation unit 13 receives N test words w(1), ..., w(N) output from the word selection unit 12. The presentation unit 13 presents the test words w(1), ..., w(N) to the user 100 (subject) according to a pre-set display format (step S13).

[0035] For example, the presentation unit 13 presents to the user 100, in a format for vocabulary size estimation tests, a predetermined instruction prompting the user 100 to input answers regarding their knowledge of test words, and N test words w(1), ..., w(N), according to a pre-set display format.

[0036] There are no limitations to this presentation format; this information may be presented as visual information such as text or images, as auditory information such as audio, or as tactile information such as Braille.

[0037] For example, the presentation unit 13 may electronically display the instruction text and test words on the display screen of a terminal device such as a PC (personal computer), tablet, or smartphone. In other words, the presentation unit 13 may generate screen information for display on a display and output it to the display.

[0038] Alternatively, the display unit 13 may be a printing device, and the instructions and test words may be printed on paper or the like and output. Alternatively, the display unit 13 may be a speaker of a terminal device, and the instructions and test words may be output as audio. Alternatively, the display unit 13 may be a braille display, and the instructions and test words may be displayed in braille.

[0039] The responses of 100 users regarding their knowledge of test words may either indicate "I know" or "I don't know" the test words (answering "I know" or "I don't know" for each rank of test word), or they may indicate one of three or more options that include "I know" and "I don't know". Examples of options other than "I know" and "I don't know" include "I'm not confident (about whether I know it)" and "I know the word, but I don't know its meaning". However, even if 100 users are given three or more options that include "I know" and "I don't know", the accuracy of vocabulary size estimation may not improve compared to when they are given only one option, "I know" or "I don't know". For example, if 100 users are given three options, "I know", "I don't know", and "I'm not confident", whether or not "I'm not confident" is selected depends on the personality of 100 user. In such cases, increasing the number of options will not improve the accuracy of vocabulary size estimation. Therefore, it is usually preferable to have 100 users answer the test words with a binary choice, such as "I know" or "I don't know".

[0040] However, instead of asking whether they "know" or "don't know," you could also ask for responses from the perspective of whether they "can create example sentences (using the test words)" or "cannot create example sentences," or whether they "can explain the meaning (of the test words)" or "cannot explain the meaning." Clarifying the perspective will change the estimated vocabulary size. For example, if the question is whether they "can create example sentences," you would be estimating the vocabulary size that the user believes they can use.

[0041] The following describes an example where 100 users are asked to answer either "I know" or "I don't know" for the test words.

[0042] Furthermore, while the test words are presented in order of increasing familiarity, for example, the presentation order is not limited to this, and the test words may be presented in a random order.

[0043] <Response Reception Department 14> Input: User's response regarding their knowledge of test vocabulary Output: User's response regarding their knowledge of test vocabulary User 100, who has been presented with instructions and test words, inputs their answers regarding their knowledge of the test words into the answer reception unit 14 (step S14).

[0044] For example, the response reception unit 14 is a touch panel of a terminal device such as a PC, tablet, or smartphone, and the user 100 inputs their response into the touch panel. The response reception unit 14 may also be the microphone of the terminal device, in which case the user 100 inputs their response by voice into the microphone.

[0045] User 100 may enter their answer into the answer reception unit 14 by clicking with a mouse or the like.

[0046] The response receiving unit 14 receives responses regarding knowledge of the input test words (for example, a response indicating that the test word is known, or a response indicating that the test word is not known), and outputs the response as electronic data. The response receiving unit 14 may output a response for each test word, output all the responses for one test at once, or output all the responses for multiple tests at once.

[0047] For example, if the response receiving unit 14 receives a response from user 100 indicating that they know the test word, it assigns the value 1 to the response regarding knowledge of that test word. On the other hand, if the response receiving unit 14 receives a response from user 100 indicating that they do not know the test word, it assigns the value 0 to the response regarding knowledge of that test word. These values ​​are output to the model generation unit 15.

[0048] <Model generation unit 15> Input: User's response regarding their knowledge of test vocabulary Output: Model The responses from user 100 regarding their knowledge of test words, output from the response reception unit 14, are input to the model generation unit 15.

[0049] The model generation unit 15 uses the responses regarding knowledge of the test words and the word familiarity DB stored in the memory unit 11 to obtain a model that represents the relationship between a value based on familiarity corresponding to a test word and a value based on the probability that user 100 will answer that they know the test word (step S15). The obtained model is output to the vocabulary size estimation unit 16.

[0050] The value based on familiarity corresponding to the test word may be the familiarity corresponding to the test word itself, or it may be a non-monotonically decreasing function of the familiarity corresponding to the test word (for example, a monotonically increasing function). For the sake of simplicity, the following example will illustrate the case where the value based on familiarity corresponding to the test word is the familiarity corresponding to the test word itself.

[0051] The value based on the probability that user 100 will answer that they know the test word may be the probability that user 100 will answer that they know the test word itself, or it may be a non-monotonically decreasing function of the probability that user 100 will answer that they know the test word (for example, a monotonically increasing function). For the sake of simplicity, the following example will illustrate the case where the value based on the probability that user 100 will answer that they know the test word is the probability that user 100 will answer that they know the test word itself.

[0052] There are no limitations on the model, but one example is a logistic regression model (logistic model). For simplicity of explanation, the following example uses a logistic curve y=f(x,Ψ) where the familiarity with each test word is the independent variable x, and the dependent variable y is the probability that 100 users will answer that they know each word. Ψ is a model parameter.

[0053] The model generation unit 15 refers to the word familiarity database stored in the memory unit 11 to obtain the familiarity corresponding to the test word w(n) that the user 100 answered as knowing, and denotes the obtained familiarity as x(n). This familiarity x(n) is the familiarity corresponding to the test word w(n).

[0054] The model generation unit 15 sets a point (x,y)=(x(n),1) where, for a test word w(n) that user 100 has answered that they know, the probability y that user 100 has answered that they know the test word w(n) is 1 (i.e., 100%), and the familiarity level corresponding to the test word w(n) is x(n).

[0055] Furthermore, the model generation unit 15 sets a point (x,y)=(x(n),0) where, for a test word w(n) that user 100 answered as unknown (or not known), the probability y that user 100 will answer as knowing the test word w(n) is 0 (i.e., 0%), and the familiarity level corresponding to the test word w(n) at that time is x(n).

[0056] The model generation unit 15 fits a logistic curve to each point (x,y) = (x(n),1) or (x(n),0) in n=1,...,N, and obtains a logistic curve y=f(x,Ψ) as a model that minimizes the error.

[0057] Figure 3 illustrates a logistic curve model y=f(x,Ψ). In Figure 3, the horizontal axis represents familiarity, and the vertical axis represents the probability (y) of knowing a word. The circles represent points (x,y)=(x(n),1) for test words w(n) that 100 users answered they knew, and points (x,y)=(x(n),0) for test words w(n) that 100 users answered they did not know (or did not answer they knew). "AIC" in Figure 3 represents the Akaike Information Criterion, and a smaller value indicates a better fit of the model. "n" in Figure 3 represents the number of test words.

[0058] Here, generation can also be rephrased as creation or construction. Therefore, the model generation unit 15 may be a model creation unit 15 or a model construction unit 15. Also, the model may be created or constructed.

[0059] <Vocabulary size estimation unit 16> Input: Model Output: Vocabulary size of 100 users The vocabulary size estimation unit 16 estimates the vocabulary size of the 100 users based on the model (step S16).

[0060] Below, we will explain three estimation methods as examples of how the vocabulary size estimation unit 16 estimates the vocabulary size of user 100.

[0061] (Estimation method 1) The vocabulary size estimation unit 16 obtains a predetermined value acquisition familiarity level, which is the familiarity level when the value based on the probability that user 100 will answer that they know a word is a predetermined value or close to a predetermined value. Examples of predetermined values ​​are 0.5 or 0.8. Of course, the predetermined value may be any other value greater than 0 and less than 1.

[0062] The vocabulary size estimation unit 16 then refers to the word familiarity database stored in the memory unit 11 to obtain the number of words with a familiarity level equal to or greater than a predetermined value, and sets the obtained number as the vocabulary size of user 100.

[0063] (Estimation method 2) The vocabulary size estimation unit 16 refers to the model and the word familiarity DB stored in the memory unit 11 to obtain the output value y(m) when the familiarity x(m) corresponding to the word w(m) contained in the word familiarity DB is input to the model. In other words, the vocabulary size estimation unit 16 calculates the value of y corresponding to the familiarity x(m) corresponding to the word w(m) in the model, and sets this calculated value as the output value y(m). The vocabulary size estimation unit 16 performs this process for each word w(m) (m=1,...,M) contained in the word familiarity DB to obtain the output value y(m) (m=1,...,M).

[0064] Then, the vocabulary size estimation unit 16 determines Σ m=1 M Calculate y(m) and use this calculated value as the vocabulary size for 100 users.

[0065] In this case, if word w(m) is a test word and a response regarding knowledge of the test word w(m) has been obtained, the vocabulary size estimation unit 16 may estimate the user 100's vocabulary size by taking into account the response regarding knowledge of the test word w(m).

[0066] For example, the vocabulary size estimation unit 16 sets y(m)=1 if the response regarding knowledge of the test word w(m) is "I know it," and y(m)=0 if the response regarding knowledge of the test word w(m) is "I don't know it." For words other than the test words, the output value y(m) obtained from the model as described above is used for y(m).

[0067] Then, the vocabulary size estimation unit 16 uses these y(m) to calculate Σ m=1 M Calculate y(m) and use this calculated value as the vocabulary size for 100 users.

[0068] By considering responses regarding knowledge of test vocabulary, a more appropriate vocabulary size can be estimated.

[0069] By estimating the vocabulary size based on a logistic model derived from the probability y that 100 users will answer that they know the test word, and the familiarity level x of the test word, the model converges more easily and the vocabulary size can be estimated more robustly than when the vocabulary size is directly set to x. Furthermore, even if the distribution of the number of words corresponding to each familiarity level differs significantly, abrupt changes in the estimated vocabulary size can be suppressed.

[0070] (Estimation method 3) The vocabulary size estimation unit 16 refers to the model and the word affinity DB stored in the memory unit 11 to obtain the output value y(i) when affinity x(i) contained in the word affinity DB is input to the model. In other words, the vocabulary size estimation unit 16 calculates the value of y in the model that corresponds to affinity x(i), and sets this calculated value as the output value y(i). The vocabulary size estimation unit 16 also refers to the word affinity DB stored in the memory unit 11 to obtain the number of words n(i) contained in the word affinity DB that correspond to affinity x(i). The vocabulary size estimation unit 16 performs these processes for each affinity x(i) (i=1,...,I) contained in the word affinity DB to obtain the output value y(i) (i=1,...,I) and the number of words n(i) (i=1,...,I). I is the number of types of affinity.

[0071] Then, the vocabulary size estimation unit 16 determines Σ i=1 I Calculate y(i) × n(i), and this calculated value will be used as the vocabulary size for 100 users.

[0072] If the level of familiarity is the same, the corresponding value of y will be the same. Also, there may be words with the same level of familiarity. For this reason, instead of estimation method 2, calculating for each level of familiarity, as in estimation method 3, can speed up the vocabulary size estimation calculation.

[0073] <Modification of the first embodiment> The word selection unit 12 may simply select multiple test words w(1), ..., w(N) from a group of words, rather than ensuring that the intervals of familiarity corresponding to the test words are at a fixed interval.

[0074] Furthermore, the model generation unit 15 may assume a response regarding knowledge of unpresented words and obtain a model that represents the relationship between a value based on familiarity with the test words and unpresented words, and a value based on the probability or assumption that the user 100 will answer that they know the test words and unpresented words.

[0075] Here, "non-presented words" are words from a group of words other than the test words. To facilitate the convergence of the logistic model, we assume responses for non-presented words that were not used as test words and use them in the model creation. Words close to the upper limit of familiarity are words that many people know, and words close to the lower limit are words that many people do not know. Therefore, if 100 users respond that they know the most familiar word among the test words, we assume that they also know non-presented words with a familiarity level higher than that. Conversely, if users respond that they do not know the least familiar word among the test words, we assume that they also do not know non-presented words with a familiarity level lower than that.

[0076] In other words, assuming that 100 users are presented with unpresented words, their responses regarding their knowledge of the unpresented words will be such that they know the words for which their familiarity level is higher than the maximum familiarity level of the test words, and they do not know the words for which their familiarity level is lower than the minimum familiarity level of the test words.

[0077] For example, if a user responds that they know a test word with a familiarity level of 6.5, we assume that they also know unpresented words with familiarity levels of 6.7 and 6.9. Similarly, if a user responds that they do not know a test word with a familiarity level of 2, we assume that they also do not know unpresented words with familiarity levels of 1.8 and 1.6.

[0078] In this way, by adding unpresented words (words not presented to user 100) and responses regarding knowledge of unpresented words, and then estimating the model, the model becomes more likely to converge, and a more appropriate model can be generated. This means that, for example, even if user 100 answers that they know most of the test words, or if user 100 answers that they do not know most of the test words, the model becomes more likely to converge, and a more appropriate model can be generated.

[0079] [Second Embodiment] A second embodiment will be described. The second embodiment is a device and method for acquiring acquisition probability.

[0080] The following section will focus on explaining the differences between the first embodiment and its modified form. Some points already discussed may be omitted.

[0081] As illustrated in Figure 4, the acquisition probability acquisition device 2 of this embodiment includes a storage unit 11, a model storage unit 21, a word extraction unit 22, a familiarity acquisition unit 23, an acquisition probability acquisition unit 24, and an acquired word information generation unit 25. The acquisition probability acquisition device 2 does not necessarily have to include the word extraction unit 22 and the acquired word information generation unit 25.

[0082] <Storage section 11> The memory unit 11 is the same as the memory unit 11 in the first embodiment.

[0083] The memory unit 11 stores a word affinity database containing multiple words and multiple affinity levels corresponding to each of those words. Here, affinity is an indicator representing the level of familiarity with a word.

[0084] <Model Memory Unit 21> The model memory unit 21 stores a model that represents the relationship between a value based on familiarity with each word and a value based on the probability that a certain person has acquired each word. Here, "a certain person" refers to the person who acquires the acquisition probability. "A certain person" may be user 100.

[0085] In this context, vocabulary acquisition means, in other words, knowing a word, being able to use a word, understanding a word, or being able to explain a word.

[0086] Examples of this model are models generated by the model generation device 1 of the first embodiment and a modified version of the first embodiment.

[0087] As shown by the dashed line in Figure 4, the acquisition probability acquisition device 2 may further include a model generation device 1 for generating a model to be stored in the model storage unit 21.

[0088] In other words, the acquisition probability acquisition device 2 may further include: (1) a word selection unit 12 that selects multiple test words from multiple words; (2) a presentation unit 13 that presents the test words to the user; (3) a response reception unit 14 that receives the user's response regarding their knowledge of the test words; and (4) a model generation unit 15 that uses the response regarding knowledge of the test words and the word familiarity DB stored in the memory unit 11 to obtain a model that represents the relationship between a value based on familiarity corresponding to a test word and a value based on the probability that the user will answer that they know the test word, and stores the obtained model in the model memory unit.

[0089] <Word Extraction Unit 22> Input: Text Output: Word The word extraction unit 22 extracts each word contained in the input text (step S22).

[0090] Each extracted word is output to the familiarity acquisition unit 23.

[0091] The text input to the word extraction unit 22 can be any text as long as it is readable by the word extraction unit 22, which is an information processing device. Examples of text include books such as textbooks and novels, newspapers and magazines, and text published on web pages.

[0092] The word extraction unit 22 extracts each word contained in the text by, for example, performing morphological analysis on the input text.

[0093] <Familiarity acquisition part 23> Input: Word Output: Words, familiarity The intimacy acquisition unit 23 receives each word extracted by the word extraction unit 22 as input. The intimacy acquisition unit 23 retrieves the intimacy corresponding to each word from the word intimacy DB stored in the storage unit 11 (step S23).

[0094] If the acquisition probability acquisition device 2 does not have a word extraction unit 22, each word contained in the text is input. In this case, the intimacy acquisition unit 23 acquires the intimacy corresponding to each word contained in the text from the word intimacy DB stored in the storage unit 11 (step S23).

[0095] Each word and the corresponding affinity level are output to the acquisition probability unit 24.

[0096] Furthermore, the affinity acquisition unit 23 and the word extraction unit 22 do not need to acquire affinity for words that are function words such as proper nouns, numerals, and particles. In other words, the word extraction unit 22 may acquire affinity only for words that are content words.

[0097] Function words such as numerals and particles are words that many people are familiar with. Therefore, by obtaining familiarity with these function words, in other words, by making these function words the target of processing, the proportion of estimated acquired words in the text calculated by the acquired word information generation unit 25 can be increased. Conversely, by not obtaining familiarity with these function words, in other words, by not making these function words the target of processing, the proportion of estimated acquired words in the text calculated by the acquired word information generation unit 25 can be decreased.

[0098] Furthermore, the affinity acquisition unit 23 may ignore words that are not included in the word affinity DB without acquiring their affinity. This allows the acquisition of acquisition probabilities to be performed appropriately even if the morphological analysis is incorrect.

[0099] <Acquisition probability acquisition unit 24> Input: Word, familiarity level Output: Words, acquisition probability The acquisition probability acquisition unit 24 uses at least the familiarity level corresponding to each word and the model stored in the model storage unit 21 to acquire the acquisition probability, which is the probability that a certain person has acquired each word (step S24).

[0100] The acquisition probability acquisition unit 24 obtains an output value when the affinity level corresponding to each word is input to the model, and uses the obtained output value as the acquisition probability corresponding to each word. In other words, the acquisition probability acquisition unit 24 calculates the value of y corresponding to the affinity level x for each word in the model, and uses the calculated value as the acquisition probability corresponding to each word.

[0101] If the model stored in the model memory unit 21 is a logistic curve y=f(x,Ψ) where the familiarity level corresponding to a word is the independent variable x and the dependent variable y is the probability that a person will answer that they know each word, then the acquisition probability acquisition unit 24 calculates the value of y=f(x,Ψ) corresponding to the familiarity level x corresponding to each word and takes that calculated value as the acquisition probability corresponding to each word.

[0102] The acquisition probability acquisition unit 24 may acquire the acquisition probability by considering the part of speech, word length, etc. For example, the acquisition probability acquisition unit 24 may acquire the acquisition probability by using the part of speech, word length, etc., as explanatory variables.

[0103] Each word and its corresponding acquisition probability are output to the acquired word information generation unit 25.

[0104] <Acquired word information generation unit 25> Input: Word, Acquisition probability Output: Acquired word information The acquired word information generation unit 25 generates acquired word information, which is information about the acquisition of words contained in the text, using the acquisition probability corresponding to each word (step S25).

[0105] Examples of acquired word information include at least one of the estimated acquired words in the text, the number of estimated acquired words in the text, or the percentage of estimated acquired words in the text.

[0106] The following explains examples of how to calculate the estimated acquired words in the text, the number of estimated acquired words in the text, and the percentage of estimated acquired words in the text.

[0107] (Estimated acquired words in the text) First, the acquired word information generation unit 25 estimates the vocabulary size of a given person. The vocabulary size can be estimated using the method described in the vocabulary size estimation unit 16 of the first embodiment. In order to estimate the vocabulary size, the word familiarity DB and the model from the model storage unit 21 may be input to the acquired word information generation unit 25 from the storage unit 11, as shown by the dashed line in Figure 4.

[0108] Next, the acquired word information generation unit 25 obtains GOISU(k), which is the number of words with an affinity level equal to or greater than the affinity level corresponding to each input word w(k). In order to obtain GOISU(k), the word affinity DB may be input from the storage unit 11 to the acquired word information generation unit 25, as shown by the dashed line in Figure 4.

[0109] Then, the acquired word information generation unit 25 sets a word whose GOISU(k) is less than or equal to the vocabulary size of a person having the GOISU(k) as an estimated acquired word in the text. Generally, the higher the intimacy, the smaller the GOISU(k). Therefore, it can be assumed that a person knows a word with a GOISU(k) that is less than or equal to the vocabulary size of that person.

[0110] An example of GOISU(k) is shown in FIG. 6.

[0111] (Number of estimated acquired words in the text) First, the acquired word information generation unit 25 estimates the vocabulary size of a person. The estimation of the vocabulary size can be performed by the method described in the vocabulary size estimation unit 16 of the first embodiment. To estimate the vocabulary size, as shown by the one-dot chain line in FIG. 4, the word intimacy DB may be input from the storage unit 11 and the model may be input from the model storage unit 21 to the acquired word information generation unit 25.

[0112] Next, the acquired word information generation unit 25 obtains the number of words GOISU(k) with a intimacy greater than or equal to the intimacy corresponding to each input word w(k). To obtain GOISU(k), as shown by the one-dot chain line in FIG. 4, the word intimacy DB may be input from the storage unit 11 to the acquired word information generation unit 25.

[0113] Then, the acquired word information generation unit 25 sets the number of words whose GOISU(k) is less than or equal to the vocabulary size of a person as the number of estimated acquired words in the text.

[0114] (Ratio of estimated acquired words in the text) The acquired word information generation unit 25 calculates, for example, a value determined by the following formula (1) or formula (2), and sets the calculated value as the ratio of estimated acquired words in the text.

[0115] (Σ k=1 K y(k)FREQ(k)) / Σ k=1 K FREQ(k)…(1) (Σ k=1 K y(k)DIFF(k)) / Σ k=1 KDIFF(k)...(2) Here, FREQ(k) is the number of times word w(k) appears in the text. Assuming the text is divided into multiple parts, DIFF(k) is the number of parts in which word w(k) appears. Examples of parts are predetermined units that make up the text, such as units, chapters, and sections. The entire text may also be used as the unit. K is the total number of words included in the text for which the acquisition probability has been obtained by the acquisition probability acquisition unit 24.

[0116] The acquired word information generation unit 25 counts FREQ(k) and DIFF(k) based on the input word. The acquired word information generation unit 25 uses the FREQ(k) and DIFF(k) obtained by counting to calculate a value determined by equation (1) or equation (2).

[0117] Figure 6 shows examples of FREQ(k) and DIFF(k).

[0118] Generally, words that are well-known tend to appear more frequently, while words that are unfamiliar tend to appear less frequently. Therefore, rare words will appear less often in a text than well-known words.

[0119] Therefore, the estimated proportion of acquired words in the text obtained by formula (1) using FREQ(k) is expected to be higher than the estimated proportion of acquired words in the text obtained by formula (2) using DIFF(k). Whether to use formula (1) or (2) is determined appropriately depending on what kind of information is needed as acquired word information.

[0120] The acquired word information generation unit 25 may use the ratio of estimated acquired words in the text (number of estimated acquired words in the text / K) as the proportion of estimated acquired words in the text. The number of estimated acquired words in the text can be determined using the method described in (number of estimated acquired words in the text).

[0121] [Third Embodiment] A third embodiment will now be described. The third embodiment is a learning recommendation word extraction device and method.

[0122] The following section will focus on explaining the differences between the first embodiment and its modified form. Some points already discussed may be omitted.

[0123] As illustrated in Figure 7, the learning recommendation word extraction device 3 of this embodiment includes a storage unit 11, a model storage unit 31, an acquisition probability acquisition unit 32, and a learning recommendation word extraction unit 33.

[0124] <Storage section 11> The memory unit 11 is the same as the memory unit 11 in the first embodiment.

[0125] The memory unit 11 stores a word affinity database containing multiple words and multiple affinity levels corresponding to each of those words. Here, affinity is an indicator representing the level of familiarity with a word.

[0126] <Model Memory Unit 31> The model memory unit 31 stores a model that represents the relationship between a value based on familiarity with each word and a value based on the probability that a certain person has acquired each word. Here, "a certain person" refers to the person from whom learning recommended words are extracted. "A certain person" may be user 100.

[0127] Examples of this model are models generated by the model generation device 1 of the first embodiment and a modified version of the first embodiment.

[0128] As shown by the dashed line in Figure 6, the learning recommendation word extraction device 3 may further include a model generation device 1 for generating models to be stored in the model storage unit 31.

[0129] In other words, the learning recommendation word extraction device 3 may further include: (1) a word selection unit 12 that selects multiple test words from multiple words; (2) a presentation unit 13 that presents the test words to the user; (3) a response reception unit 14 that receives the user's response regarding their knowledge of the test words; and (4) a model generation unit 15 that uses the response regarding knowledge of the test words and the word familiarity DB stored in the memory unit 11 to obtain a model that represents the relationship between a value based on familiarity corresponding to a test word and a value based on the probability that the user will answer that they know the test word, and stores the obtained model in the model memory unit.

[0130] <Acquisition probability acquisition unit 32> Input: Word Output: Words, acquisition probability The acquisition probability acquisition unit 32 receives a word set consisting of multiple words that are candidates for recommended words to learn.

[0131] The acquisition probability acquisition unit 32 uses at least the word familiarity DB stored in the memory unit 11 and the model stored in the model memory unit 31 to acquire the acquisition probability, which is the probability that a certain person has acquired each word included in the input word set (step S32).

[0132] The acquisition probability acquisition unit 32 obtains an output value when the affinity level corresponding to each word is input to the model, and uses the obtained output value as the acquisition probability corresponding to each word. In other words, the acquisition probability acquisition unit 32 calculates the value of y corresponding to the affinity level x for each word in the model, and uses that calculated value as the acquisition probability corresponding to each word.

[0133] If the model stored in the model memory unit 31 is a logistic curve y=f(x,Ψ) where the familiarity level corresponding to a word is the independent variable x and the dependent variable y is the probability that a person will answer that they know each word, then the acquisition probability acquisition unit 32 calculates the value of y=f(x,Ψ) corresponding to the familiarity level x corresponding to each word and takes that calculated value as the acquisition probability corresponding to each word.

[0134] The acquisition probability acquisition unit 32 may acquire the acquisition probability by considering the part of speech, word length, etc. For example, the acquisition probability acquisition unit 32 may acquire the acquisition probability by using the part of speech, word length, etc., as explanatory variables.

[0135] Each word and its corresponding acquisition probability are output to the acquired word information generation unit 25.

[0136] <Learning Recommended Word Extraction Unit 33> Input: Word, Acquisition probability Output: Recommended words for learning The learning recommendation word extraction unit 33 extracts learning recommendation words from the word set based on the acquired acquisition probability (step S33).

[0137] For example, the learning recommendation word extraction unit 33 may extract words whose acquired probability is close to a predetermined probability as learning recommendation words.

[0138] A given probability is a number greater than 0 and less than 1. An example of a given probability is 0.5.

[0139] The learning recommendation word extraction unit 33 may extract a predetermined number of words close to a predetermined probability as learning recommendation words.

[0140] If the predetermined probability is 0.5 and the predetermined number is 7, then, for example, the 7 words shown in Figure 9 are extracted as recommended words for learning. In Figure 9, ENTRY is the spelling of the word, PSY is the familiarity level, Prob is the acquisition probability, YN is information about the responses received from 100 users regarding these words (whether they know them or not), and Distance50 is the magnitude of the difference between the predetermined probability of 0.5 and Prob in this case.

[0141] In this example, since no responses indicating "I know" or "I don't know" have been received from the 100 users, "-" is displayed in YN. If the 100 users had responded that they knew the word, "1" would be displayed in YN; if the 100 users had responded that they didn't know the word, "0" would be displayed in YN.

[0142] The recommended vocabulary for learning will be presented to those who are eligible to have these vocabulary words extracted. The recommended vocabulary for learning may also be presented to those eligible to have these vocabulary words extracted in the format of the table shown in Figure 9.

[0143] The learning recommendation word extraction unit 33 may extract words that fall within a predetermined range that includes a predetermined probability as learning recommendation words.

[0144] The learning recommendation word extraction unit 33 may extract words of a predetermined part of speech whose acquired probability is close to a predetermined probability as learning recommendation words. Examples of predetermined parts of speech are verbs, nouns, and adjectives. The predetermined parts of speech may be two or more types. In this case, the learning recommendation word extraction unit 33 may extract words from each of the two or more types of parts of speech whose acquired probability is close to a predetermined probability as learning recommendation words.

[0145] Part-of-speech information may be stored in the word familiarity database. In this case, the learning recommendation word extraction unit 33 can refer to the word familiarity database to obtain the part of speech of a word and perform the above processing.

[0146] The learning recommendation word extraction unit 33 may also perform the above processing by referring to a dictionary containing words and their parts of speech stored in a memory unit (not shown) to obtain the part of speech of the words.

[0147] <Modified form of the third embodiment> The word set, consisting of multiple words that are candidates for recommended learning words, input to the acquisition probability unit 32 may be words included in a predetermined text. For this purpose, the recommended learning word extraction device 3 may include a word extraction unit 34, which will be described below.

[0148] <Word extraction unit 34> Input: Text Output: Word The word extraction unit 34 extracts each word contained in the input text (step S34).

[0149] Each extracted word is output to the acquisition probability unit 32 as a set of words that are candidates for recommended learning words.

[0150] The text input to the word extraction unit 34 can be any text as long as it is readable by the word extraction unit 22, which is an information processing device. Examples of text include books such as textbooks and novels, newspapers and magazines, and text published on web pages.

[0151] The word extraction unit 34 extracts each word contained in the text by, for example, performing morphological analysis on the input text.

[0152] [Differentiation] This disclosure is not limited to the embodiments described above, and various modifications and applications are possible without departing from the gist of this disclosure.

[0153] The various processes described in the embodiments may be performed not only in chronological order according to the sequence described, but also in parallel or individually as needed, depending on the processing capacity of the device performing the processes.

[0154] For example, data exchange between the components of the model generation device 1, the acquisition probability acquisition device 2, and the learning recommendation word extraction device 3 may occur directly or via a storage unit (not shown).

[0155] [Programs, recording media] The processing of each part of the above-mentioned devices may be implemented by a computer. In this case, the processing content of the functions that each device should have is described by a program. This program is then loaded into the memory unit 1020 of the computer 1000 shown in Figure 10, and the arithmetic processing unit 1010, input unit 1030, output unit 1040, display unit 1060, etc. are made to operate, thereby realizing the various processing functions of each of the above-mentioned devices on the computer.

[0156] The program describing this process can be recorded on a computer-readable recording medium. Computer-readable recording media are, for example, non-temporary recording media, specifically magnetic recording devices, optical discs, etc.

[0157] Furthermore, this program may be distributed, for example, by selling, transferring, or lending portable recording media such as DVDs or CD-ROMs on which the program is recorded. Alternatively, the program may be stored in the storage device of a server computer and distributed by transferring the program from the server computer to other computers via a network.

[0158] A computer executing such a program first stores the program recorded on a portable recording medium or transferred from a server computer in its own non-temporary storage device, the auxiliary recording unit 1050. Then, when processing is to be executed, the computer reads the program stored in the auxiliary recording unit 1050 into the storage unit 1020 and executes the processing according to the loaded program. Alternatively, the computer may directly read the program from the portable recording medium into the storage unit 1020 and execute the processing according to that program. Furthermore, each time a program is transferred to this computer from a server computer, it may sequentially execute the processing according to the received program. Alternatively, the above processing may be executed by a so-called ASP (Application Service Provider) type service, where the server computer does not transfer programs to this computer, but the processing function is realized only by execution instructions and result acquisition. In this embodiment, the program includes information used for processing by an electronic computer that is equivalent to a program (data that is not a direct instruction to the computer but has the property of defining the processing of the computer).

[0159] Furthermore, in this configuration, the device is configured by executing a predetermined program on a computer, but at least a part of these processes may be implemented in hardware.

[0160] For example, the word selection unit 12, presentation unit 13, response reception unit 14, model generation unit 15, vocabulary size estimation unit 16, word extraction unit 22, familiarity acquisition unit 23, acquisition probability acquisition unit 24, acquired word information generation unit 25, acquisition probability acquisition unit 32, learning recommended word extraction unit 33, and word extraction unit 34 may be configured by processing circuits.

[0161] Furthermore, the storage unit 11, model storage unit 21, and model storage unit 31 may be composed of memory.

[0162] The following additional information is disclosed regarding the embodiments described above.

[0163] (Additional note 1) Memory and At least one processor connected to the memory, Includes, The memory stores a word affinity database in which affinity is an indicator representing familiarity with a word, and which stores multiple words and multiple affinity values ​​corresponding to each of the multiple words. The aforementioned processor, Using the word familiarity database stored in the memory, multiple test words are selected from the multiple words such that the interval between familiarity levels corresponding to the test words is constant. Word selection device.

[0164] (Additional note 2) A non-temporary storage medium that stores a program executable by a computer to perform word selection processing, The aforementioned word selection process is, Intimacy is an indicator representing familiarity with a word. A word intimacy database is used, which stores multiple words and multiple intimacy values ​​corresponding to each of those words. Multiple test words are selected from the multiple words such that the intervals between the intimacy values ​​corresponding to the test words are constant. Non-transitory storage medium.

[0165] (Additional note 3) Memory and At least one processor connected to the memory, Includes, The memory stores a word affinity database in which affinity is an indicator representing familiarity with a word, and which stores multiple words and multiple affinity values ​​corresponding to each of the multiple words. The aforementioned processor, Taking multiple test words and the user's response regarding their knowledge of the test words as input, a model is obtained that represents the relationship between a value based on familiarity with the test words and a value based on the probability that the user will answer that they know the test words, using the response regarding knowledge of the test words and a word familiarity database stored in memory. Model generation device.

[0166] (Additional note 4) A non-temporary storage medium that stores a program executable by a computer to perform a model generation process, The aforementioned model generation process is: Taking multiple test words and the user's response regarding their knowledge of the test words as input, a model is obtained that represents the relationship between a value based on familiarity with the test words and a value based on the probability that the user will answer that they know the test words, using the response regarding knowledge of the test words and a word familiarity database. The aforementioned intimacy is an indicator representing familiarity with a word, and the word intimacy database stores multiple words and multiple intimacy values ​​corresponding to each of those words. Non-transitory storage medium.

[0167] (Additional note 5) Memory and At least one processor connected to the memory, Includes, The aforementioned memory contains, Intimacy is an indicator that represents familiarity with a word, and a word intimacy DB stores multiple words and multiple intimacy values ​​corresponding to each of those multiple words. A model that represents the relationship between a value based on familiarity with each word and a value based on the probability that a person has acquired each word. It is remembered, The aforementioned processor, The familiarity level corresponding to each word in the input text is obtained from the word familiarity database stored in the memory. Using at least the familiarity level corresponding to each acquired word and the model stored in the memory, the acquisition probability, which is the probability that a certain person has acquired each word, is obtained. A device for obtaining the probability of success.

[0168] (Additional note 6) A non-temporary storage medium that stores a program executable by a computer to perform a process for obtaining acquisition probabilities, The aforementioned process for obtaining the acquisition probability is as follows: Intimacy is an indicator representing familiarity with a word. The intimacy value for each word in the input text is obtained from a word intimacy database that stores multiple words and multiple intimacy values ​​corresponding to each of those words. A model representing the relationship between a value based on familiarity with each word and a value based on the probability that a person has acquired each word, and using at least the familiarity with each acquired word, the acquisition probability, which is the probability that a person has acquired each word, Non-transitory storage medium.

[0169] (Additional note 7) Memory and At least one processor connected to the memory, Includes, The aforementioned memory contains, Intimacy is an indicator that represents familiarity with a word, and a word intimacy DB stores multiple words and multiple intimacy values ​​corresponding to each of those multiple words. A model that represents the relationship between a value based on familiarity with each word and a value based on the probability that a person has acquired each word. It is remembered, The aforementioned processor, Using at least the word familiarity DB stored in the memory and the model stored in the memory, the acquisition probability, which is the probability that a certain person has acquired each word included in the input word set, Based on the acquisition probability obtained, recommended words for learning are extracted from the word set. A learning recommendation word extraction device.

[0170] (Additional note 8) A non-temporary storage medium that stores a program executable by a computer to perform a learning recommendation word extraction process, The aforementioned learning recommendation word extraction process is, Familiarity is an indicator representing familiarity with a word, and the acquisition probability, which is the probability that a certain person has acquired each word included in the input word set, is obtained using at least a word familiarity DB that stores multiple words and multiple familiarity values ​​corresponding to each of those words, and a model that represents the relationship between a value based on the familiarity of each word and a value based on the probability that a certain person has acquired each word. Based on the acquisition probability obtained, recommended words for learning are extracted from the word set. Non-transitory storage medium.

[0171] All documents, patent applications, and technical standards described herein are incorporated by reference to the same extent as if each individual document, patent application, and technical standard were specifically and individually described as being incorporated by reference.

Claims

1. Taking multiple test words and the user's response regarding their knowledge of those test words as input, Intimacy is an indicator representing familiarity with a word, and the system includes a storage unit that stores a word intimacy DB containing multiple words and multiple intimacy values ​​corresponding to each of those words, A model generation unit obtains a model that represents the relationship between the familiarity of a test word or its non-monotonic decreasing function value and the probability that the user answers that they know the test word or its non-monotonic decreasing function value, using the answers regarding knowledge of the test word and the word familiarity DB stored in the memory unit, by associating the familiarity or its non-monotonic decreasing function value obtained from the word familiarity DB for each of the test words as an independent variable, and the value obtained by converting the answers regarding knowledge of the test word into a predetermined probability value as the dependent variable, and performing a statistical regression analysis based on the relationship between the multiple associated independent and dependent variables. A model generation device that includes this feature.

2. A model generation apparatus according to claim 1, Among the aforementioned multiple words, any words other than the aforementioned multiple test words will be designated as non-presented words. Assuming that the aforementioned unpresented words are presented to the user, the user's responses regarding their knowledge of the aforementioned unpresented words will be such that they know the words for which their familiarity level is higher than the maximum familiarity level of the test words, and they do not know the words for which their familiarity level is lower than the minimum familiarity level of the test words. The model generation unit further uses the responses regarding knowledge of the unpresented words to generate a model representing the relationship between the familiarity or non-monotonic decreasing function value corresponding to the test words and the unpresented words, and the probability or non-monotonic decreasing function value that the user answers that they know the test words and the unpresented words. This model is obtained by associating the familiarity or non-monotonic decreasing function value obtained from the word familiarity DB for each of the test words and the unpresented words as independent variables, and the value obtained by converting the responses regarding knowledge of the test words and the unpresented words into predetermined probability values ​​as dependent variables, and then performing a statistical regression analysis based on the relationship between the multiple associated independent variables and dependent variables. Model generation device.

3. A model generation apparatus according to claim 1, A word selection section where multiple test words are selected from multiple words, A display unit that presents the aforementioned test words to the user, A response receiving unit that receives responses from the user regarding their knowledge of the test words, A model generation device that further includes the following.

4. A model generation apparatus according to claim 3, The word selection unit uses the word familiarity database stored in the memory unit to select multiple test words from the multiple words such that the interval between the familiarity levels corresponding to the test words is constant. Model generation device.

5. Taking multiple test words and the user's response regarding their knowledge of those test words as input, The model generation unit uses the responses regarding knowledge of the test words and the word familiarity DB to generate a model that represents the relationship between familiarity or its non-monotonic decreasing function value corresponding to the test words and the probability or its non-monotonic decreasing function value that the user answers that they know the test words. This model generation step involves associating the familiarity or its non-monotonic decreasing function value obtained from the word familiarity DB for each of the test words as an independent variable, and a value obtained by converting the responses regarding knowledge of the test words into predetermined probability values ​​as a dependent variable, and then performing a statistical regression analysis based on the relationship between the associated independent and dependent variables. The aforementioned intimacy is an indicator representing familiarity with a word, and the word intimacy DB stores multiple words and multiple intimacy values ​​corresponding to each of those words. Model generation method.

6. A program for causing a computer to function as each part of the model generation apparatus according to claim 1.

Images