JP2013016011A - Synonym dictionary generation device, method therefor, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a synonym dictionary generation technique that enables generation of high-accuracy synonym dictionary on the basis of both a document text and a voice text.SOLUTION: A synonym dictionary generation technique includes the steps of: calculating similarity between a context including a reference vocabulary referenced when creating a synonym dictionary and a context including a related vocabulary related to the reference vocabulary; calculating similarity between a notation of the reference vocabulary and that of the related vocabulary; calculating similarity between a pronunciation of the reference vocabulary and that of the related vocabulary; using each calculated similarity of the context, the notation, and the pronunciation to obtain each synonymy index for the reference vocabulary and the related vocabulary; and determining whether or not the related vocabulary is a synonym of the reference vocabulary on the basis of a level of the synonymy index.

Description

  The present invention relates to a synonym dictionary generation method, a synonym dictionary generation device, and a program thereof that determine synonyms between words, register and associate synonym words, and generate a synonym dictionary.

  A synonym dictionary is a dictionary that collects words that have the same meaning but different notations. For example, when a single word is input as a search word in an information search, the synonym dictionary can easily search for information intended by the user by using the synonym dictionary to complement the search word and search. Used to be able to.

  Patent Document 1 is known as a conventional synonym dictionary generation method. Patent Document 1 discloses a synonym dictionary generation system that generates a synonym dictionary by defining the degree of association between words based on search behavior of a specific user.

JP 11-31168 A

  The conventional synonym dictionary creation method creates a synonym dictionary based on document text, and does not assume that a synonym dictionary is created based on speech text such as a speech recognition result. Therefore, when it is determined whether or not a plurality of words are synonymous based on speech text including word dropout, insertion, recognition error, etc., the accuracy is considered to deteriorate. The document text means text information originally created as a document such as a newspaper, magazine, or web, and the voice text means a single talk by a single speaker (lecture, speech, etc.) or by two speakers. Dialogue means text information obtained as a result of performing voice recognition on voice data recording conversations of three or more speakers, and text information originally created based on voice.

  An object of the present invention is to provide a synonym dictionary generation technique capable of creating a synonym dictionary with high accuracy not only based on document text but also based on speech text.

  In order to solve the above problems, according to the first aspect of the present invention, a context similar to a context including a reference vocabulary used as a reference when creating a synonym dictionary and a context including a related vocabulary related to the reference vocabulary are similar. To calculate the similarity between the reference vocabulary and the related vocabulary, calculate the similarity between the reference vocabulary reading and the related vocabulary reading, and determine the probability that the reference vocabulary and the related vocabulary are synonyms. The synonymous index shown indicates that the context of the reference vocabulary and the context of the related vocabulary are more likely to be more similar, and the notation of the reference vocabulary and the related vocabulary are more likely to be more likely. , To indicate that the reading of the reference vocabulary and the reading of the related vocabulary are unlikely to be similar, and use the calculated context, notation, and similarity of readings to indicate a synonymous index for the reference vocabulary and related vocabulary. Asking Related vocabulary based on the size of the synonymous indicator determines whether a synonym of the reference vocabulary.

  According to the synonym dictionary generation technology according to the present invention, there is an effect that a synonym dictionary with high accuracy can be created not only based on document text but also based on speech text.

The functional block diagram of the synonym dictionary production | generation apparatus 11. FIG. The figure which shows the processing flow of the synonym dictionary production | generation apparatus 11. FIG. The figure which shows the example of data memorize | stored in the memory | storage part. 4A shows an example of data stored in the vocabulary information storage unit 16 (combination of reference vocabulary and related vocabulary), and FIG. 4B shows an example of data stored in the vocabulary information storage unit 16 (vocabulary information). . The figure which shows the example of data memorize | stored in the synonym information storage part. The figure which shows the example of data (vocabulary information and a concept vector) memorize | stored in the vocabulary information storage part 16. FIG.

  Hereinafter, embodiments of the present invention will be described. In the drawings used for the following description, constituent parts having the same function and steps for performing the same process are denoted by the same reference numerals, and redundant description is omitted.

<Synonym Dictionary Generation Device 11 According to First Embodiment>
A synonym dictionary generation device 11 according to an embodiment of the present invention will be described with reference to FIGS. 1 and 2. The synonym dictionary generation device 11 may be configured by a known computer including program execution means such as a CPU, storage means such as a memory and a hard disk storage device, input means such as a keyboard and a mouse, and display means such as a monitor. Alternatively, it may be configured as a dedicated device for synonym dictionary generation including similar means.

  As shown in FIG. 1, the synonym dictionary generation device 11 functionally includes a related vocabulary acquisition unit 12, a text information storage unit 13, a context acquisition unit 14, a context similarity calculation unit 15, a vocabulary information storage unit 16, The notation similarity calculation unit 17, the reading similarity calculation unit 18, the part-of-speech similarity calculation unit 19, the synonym determination unit 20, the synonym information storage unit 21, and the storage unit 22 are configured.

<Processing flow>
The outline of the processing flow of the synonym dictionary generation device 11 will be described with reference to FIG. The synonym dictionary generation device 11 acquires the reference vocabulary from the input means or another device (s201) and stores it in the storage unit 22 (see c01 in FIG. 3). The vocabulary is a concept made up of words and their meanings. In this embodiment, the vocabulary is expressed by word notation, reading and parts of speech. Information including word notation, reading, and part of speech is called vocabulary information. The reference vocabulary is a vocabulary used as a reference when creating a synonym dictionary, and is a vocabulary to be searched when searching for synonyms in the synonym dictionary. In other words, a system or a user who uses a synonym dictionary can search the synonym dictionary using the reference vocabulary as a key, and acquire synonyms of the reference vocabulary.

  The related vocabulary acquisition unit 12 receives the reference vocabulary from the storage unit 22, acquires the related vocabulary related to the reference vocabulary from the vocabulary information storage unit 16, and combines the notation, reading, and part of speech of each of the reference vocabulary and the related vocabulary. It is acquired (s202) and stored in the storage unit 22 (c02, c03, c04 in FIG. 3). The related vocabulary is a vocabulary related to the reference vocabulary, in other words, a vocabulary that is a candidate for a synonym of the reference vocabulary, or a vocabulary that is determined whether or not it is a synonym for the reference vocabulary. For example, an existing synonym dictionary may be stored in the vocabulary information storage unit 16. In that case, the reference vocabulary and one or more related vocabularies for the reference vocabulary are stored in combination (FIG. 4A), and further, the vocabulary information storage unit 16 stores vocabulary information (FIG. 4B).

  Next, the context acquisition unit 14 receives the reference vocabulary and the related vocabulary from the storage unit 22, acquires the context including the reference vocabulary and the related vocabulary from the text information storage unit 13 (s203), and stores it in the storage unit 22 ( C05 in FIG. 3). In the present embodiment, the context means an arrangement of words or a set of words (hereinafter, also referred to as “word string”), and (I) speech text, (II) a set of words obtained from the speech text, or ( III) Conjunctive data (details will be described later). Hereinafter, a context including the reference vocabulary is referred to as a reference vocabulary context, and a context including the related vocabulary is referred to as a related vocabulary context.

  Next, the context similarity calculation unit 15 receives the reference vocabulary context and the related vocabulary context from the storage unit 22, and acquires the similarity between the reference vocabulary context and the related vocabulary context (hereinafter referred to as “context similarity”) (s204). ) And stored in the storage unit 22 (c06 in FIG. 3).

  Further, the notation similarity calculation unit 17, the reading similarity calculation unit 18 and the part-of-speech similarity calculation unit 19 receive the reference vocabulary and related vocabulary notation, reading and part-of-speech from the storage unit 22, respectively, and the notation, reading and part-of-speech similarity Characteristics (hereinafter referred to as “notation similarity”, “reading similarity”, and “part of speech similarity”) are acquired (s206) and stored in the storage unit 22 (c07, c08, c09 in FIG. 3).

  The synonym determination unit 20 receives the context similarity, the notation similarity, the reading similarity, and the part-of-speech similarity from the storage unit 22, and uses these values to obtain a synonym index for the reference vocabulary and the related vocabulary (s207). ). The synonym index is an index indicating the probability that the reference vocabulary and the related vocabulary are synonyms, and indicates that the similarity of the context of the reference vocabulary and the related vocabulary is more probable. This is an index indicating that the notation of the vocabulary and the related vocabulary are more likely to be similar, and the reading of the reference vocabulary and the reading of the related vocabulary are more likely to be similar.

  Further, the synonym determination unit 20 determines whether or not the related vocabulary is a synonym of the reference vocabulary based on the size of the obtained synonym index (s208).

  When it is determined that it is not a synonym, the synonym determination unit 20 ends the process.

  When the synonym is determined to be a synonym, the synonym determination unit 20 uses the reference vocabulary and its related vocabulary (hereinafter, the related vocabulary determined to be a synonym is treated as a “synonym”), and the synonym for the reference vocabulary. Are combined with the context similarity, the notation similarity, the reading similarity, the part-of-speech similarity, and their synonyms and stored in the synonym information storage unit 21 (see s209, FIG. 5).

  The synonym dictionary generation device 11 outputs information stored in the synonym information storage unit 21 or a part thereof (which may be information including at least a reference vocabulary and synonyms) as a synonym dictionary.

  Hereinafter, the processing content of each part is demonstrated.

<Related Vocabulary Acquisition Unit 12 and Vocabulary Information Storage Unit 16>
The related vocabulary acquisition unit 12 acquires at least one related vocabulary of the reference vocabulary from the vocabulary information storage unit 16 using the reference vocabulary. The related vocabulary acquired here may be (1) an existing synonym dictionary (see FIGS. 4A and 4B), or (2) a highly related word based on a co-occurrence relationship in a large amount of text information. There may be. Note that the large amount of text information may be voice text stored in the text information storage unit 13 or other document text. The case (2) will be described. In the case of (2), the related vocabulary acquisition unit 12 stores vocabulary information and a “concept base” (see FIG. 6, Reference 1).
[Reference Document 1] JP 2009-277099 A

  This “concept base” is a database consisting of pairs of words and concept vectors corresponding to the words for the purpose of determining similarities between words and searching for words of the same concept. A corpus concept base created from a corpus is known. The “concept vector” of a predetermined word is calculated according to the frequency of co-occurrence with each of a plurality of predetermined co-occurrence words within the range (eg, sentence) to which the predetermined word belongs. As co-occurrence words in the corpus concept base, words that appear frequently in the corpus are used, each word is a row, co-occurrence words are columns, and the co-occurrence frequency of words and co-occurrence words is a matrix component. Create a co-occurrence matrix. In the corpus concept base, a matrix in which the dimension of the column of the co-occurrence matrix is compressed by singular value decomposition, and the row vector of each row of the compressed matrix is a concept vector. The concept base created in this manner has the property that the higher the similarity between words, the closer the distance between the concept vectors of the words, so it is effective in determining similarity between words. In other words, it can be determined that the closer the concept vector between two words is, the higher the similarity between the two words is.

  Prior to using the above-described concept base, a conventional technique is constructed based on a large amount of text information and stored in the vocabulary information storage unit 16. The related vocabulary acquisition unit 12 acquires the concept vector of the reference vocabulary from the vocabulary information storage unit 16 and obtains a concept vector having a distance close to the concept vector. For example, the concept vector or cosine similarity that maximizes the cosine similarity A large number of top concept vectors are obtained, and a word corresponding to the concept vector is used as a related vocabulary and acquired together with vocabulary information.

  For example, a list of related vocabulary created from a similar word search based on the concept vector of the reference vocabulary “set top box” is {“suspend”, “tuner”, “remote control”, “STB”, “medium”,. ....

<Context acquisition unit 14 and text information storage unit 13>
A large amount of text information is stored in the text information storage unit 13, and the context acquisition unit 14 acquires the reference vocabulary context and the related vocabulary context from the text information storage unit 13.

The text information storage unit 13 includes, for example, a hard disk storage device, and stores a plurality of (I) speech texts generated by a plurality of speech recognition servers (not shown) connected to the net as contexts. A plurality of sets of words obtained from (II) speech text may be stored as context. A plurality of (III) “collocation data” may be stored (see References 2 and 3).
[Reference 2] JP 2010-117764 [Reference 3] Yuichiro Terada et al., “Development of Japanese collocation data”, Fukuoka University Engineering Bulletin, 2007, September, 79, p.53-57

  (II) will be described. Humans may extract keywords from speech text and create a set of them. Also, based on the word co-occurrence probability related to the vocabulary included in the speech text, a word string including the vocabulary generated by the statistical model (that is, a set of words having a high word co-occurrence probability) is created as a context. Also good. With such a configuration, a word set having a conceptual group of words and a statistically sufficient number of samples can be acquired as a context.

  (III) will be described. “Collaborative data” is a database composed of a set of word headings and word strings following the words, focusing on the nature of word chains for the purpose of searching for contexts that define semantic units and include vocabularies. “Conjunction data” for a given word includes probabilistic binding between the word and the word string connected to it (probability of co-occurrence between element words), lexical integrity ( It is classified by the degree of idioms (difficulty of constructability principle). The presence or absence of the property as a collocation is determined by a statistical feature quantity in addition to the introspection of the organizer. A word string having a word chain group and a statistically sufficient number of samples can be acquired as a context. According to this aspect, it is possible to perform synonym determination with higher accuracy in the following synonym determination process in consideration of the word chain group and sample number of the reference vocabulary context or related vocabulary context, and high-quality synonyms can be determined. A word dictionary can be generated.

  Further, the context stored in the text information storage unit 13 is periodically added / updated.

  The context acquisition unit 14 acquires the reference vocabulary context and the related vocabulary context from the text information storage unit 13. The vocabulary context acquired here may be any one of (I) to (III). For example, in the case of (III), a local word including a reference vocabulary or a related vocabulary and having a probability of concatenation of 0.90 or higher and a convergence of 0.60 or higher among the numerical values representing probabilistic binding. It may be a chained sequence, or may be a vocabulary set in which the number of word interruptions is 1 or less among numerical values representing lexical unity.

<Context Similarity Calculation Unit 15>
The context similarity calculation unit 15 calculates the similarity between the reference vocabulary context and the related vocabulary context. For example, when the context acquired by the context acquisition unit 14 is a speech text, the context similarity calculation unit 15 divides it into morphemes using a predetermined morpheme analysis algorithm. Moreover, when it is collocation data, it is assumed that it is divided into morphemes. Next, the similarity is calculated for each of the reference vocabulary context and the related vocabulary context. For example, context similarity is calculated based on the co-occurrence relationship of all vocabularies in the reference vocabulary context and the co-occurrence relationship of all vocabularies in the related vocabulary context. Specifically, the concept vector for each divided morpheme is acquired from the vocabulary information storage unit 16, and the sum of the cosine similarity of the word concept vectors of each morpheme is normalized to obtain the reference vocabulary context and the related vocabulary context. Similarity.

<Notation similarity calculation unit 17>
The notation similarity calculation unit 17 calculates the similarity between the reference vocabulary notation and the related vocabulary notation. For example, the notation similarity calculation unit 17 divides the vocabulary notation acquired by the vocabulary information storage unit 16 for each character. Specifically, it is extracted as an individual element for each minimum unit for describing a vocabulary on a program or medium such as an encoded character or character code. Next, the similarity of the notation of the reference vocabulary and the related vocabulary is calculated based on the matching rate of each character. For example, the reference vocabulary and the related vocabulary are divided for each character, the generated code string is regarded as two patterns, the code is regarded as an individual element, and the reference vocabulary and the related vocabulary are associated with each other. The matching (DP matching) by dynamic programming (Dynamic Programming) is used as a method for efficiently calculating the similarity, and the similarity between the notation of the reference vocabulary and the related vocabulary is calculated as a normalized matching rate.

<Reading similarity calculation unit 18>
The reading similarity calculation unit 18 calculates the similarity between the reading of the reference vocabulary and the reading of the related vocabulary. For example, the reading similarity calculation unit 18 divides the vocabulary reading acquired by the vocabulary information storage unit 16 into phonemes. Note that phonemes are generally described in one unit of vowels, repellent sounds, and prompting sounds, and other units are described in two units of consonants and vowels. Next, the similarity between the phoneme notation of the reference vocabulary and the related vocabulary is calculated based on the matching rate of each phoneme. For example, DP matching is used in the same manner as the notation similarity calculation unit 17 to normalize the matching rate and calculate the similarity between the reading of the reference vocabulary and the related vocabulary.

<Part of speech similarity calculation unit 19>
The part of speech similarity calculation unit 19 calculates the part of speech similarity between the part of speech of the reference vocabulary and the related vocabulary. Here, the part of speech is assumed to be located on a semantic system that is divided into a tree shape from a large classification with all the parts of speech as roots (see Reference 4).
[Reference 4] Satoshi Shirai, Yoshifumi Oyama, Satoru Ikehara, Masahiro Miyazaki, Akio Yokoo, "On the Japanese Vocabulary System", Information Processing Society of Japan. IM, November 1998, Vol.1998 No.106, p.47-52

  For example, the similarity between the part of speech of the reference vocabulary and the related vocabulary is calculated based on the distance between each part of speech. Based on the major classification common to both the part of speech of the reference vocabulary and the related vocabulary in the part of speech system, the sum of the hierarchical differences between both parts of speech is defined as the distance of part of speech. For polysemy, use the smallest value. The reciprocal number is normalized to calculate the similarity between the part of speech of the reference vocabulary and the related vocabulary.

<Synonym determination unit 20 and synonym information storage unit 21>
The synonym determination unit 20 obtains a synonym index for the reference vocabulary and the related vocabulary using the context similarity, the notation similarity, the reading similarity, and the part-of-speech similarity, and the related vocabulary is determined based on the size of the synonym index. It is determined whether it is a synonym. Next, the synonym determination unit 20 outputs a combination of the reference vocabulary and the related vocabulary determined to be synonymous with the reference vocabulary, and stores it in the synonym information storage unit 21.

  For example, context similarity, reading similarity, notation similarity, and part-of-speech similarity are constant value 1 when the context, reading, notation, and part-of-speech for both the reference vocabulary and related vocabulary match, respectively, Normalization is performed so that the value 0 is calculated, and a synonym index is obtained by weighting and combining each value.

Specifically, the synonym index increases as context similarity, notation similarity, and part-of-speech similarity increase. On the other hand, when the reading similarity is large, if the notation similarity and the part of speech similarity are small, it can be regarded as an erroneous word due to misrecognition. The synonymous index of the reference vocabulary and the related vocabulary is represented by an expression in which each similarity is linearly combined as an example.
S _vocab (u, v) = S _context (u, v) + β ・ S _POS (u, v) + γ ・ S _describe (u, v) + δ ・ S _pronounce (u, v)
(0 ≦ S _context (u, v), S _POS (u, v), S _describe (u, v), S _pronounce (u, v) ≦ 1, β ≧ 0, γ> 0, δ <0) ( 1)
Here, with respect to the reference vocabulary u and the related vocabulary v, S _context , S _POS , S _describe and S _pronoun represent context similarity, part-of-speech similarity, notation similarity and reading similarity, respectively. β, γ, and δ are weighting factors. | Β |, | γ |, and | δ | are preferably smaller than 1.

As another example, a synonym index in which a sigmoid function is introduced in order to further emphasize notation similarity and reading similarity with respect to context similarity and part-of-speech similarity is expressed by the following expression.
S _vocab (u, v) = (S _context (u, v) + β ・ S _POS (u, v)) × s _α (S _describe (u, v) -S _pronounce (u, v)) (α> 0 (2)
Here, s _α is a sigmoid function of gain α. For example, α takes a value of about 3.0 to 5.0.

  “Determining whether or not a synonym is a synonym based on the size of the synonym index” means, for example, (i) synonymous with a related vocabulary corresponding to the synonym index exceeding the threshold when the calculated synonym index exceeds the threshold κ (Ii) Find synonymous indices for multiple related vocabularies, and if the synonymous index is the largest and significantly larger than the synonymous indices of multiple related vocabularies excluding the maximum value, it corresponds to the maximum value It is determined that the related vocabulary is a synonym.

  An example of how to obtain α, β, and threshold value κ in equation (2) is shown. β is the weight of part-of-speech similarity with respect to context similarity, and a small number of documents and synonyms so that the discrimination performance between the synonym-related reference vocabulary, related vocabulary, and irrelevant vocabulary is maximized. The value is determined by a learning set consisting of a collection of vocabulary. At this time, the threshold value κ is defined as the boundary value defined in the first term on the right side depending on the presence or absence of the synonymous relationship. For example, the second term on the right side is set to 1, and the weighting factor β is varied to experimentally obtain the weighting factor β and the threshold value κ that can accurately separate values for identifying synonyms and non-synonyms in the learning set. Next, α uses a learning set whose related vocabulary is known to be a result of misrecognition, and determines a value that minimizes synonym determination errors due to misrecognition. Specifically, a speech transcription document and a speech recognition result document are used as target documents for extracting synonyms from an existing small synonym dictionary.

  Next, the reference vocabulary and the related vocabulary determined to be synonyms are registered in the synonym dictionary. Specifically, the meaning information acquired from the vocabulary information storage unit 16 and the context similarity, notation similarity, notation similarity, reading similarity, part-of-speech similarity, and synonym index are stored in the synonym information storage unit 21. (See FIG. 5).

  A user of the synonym dictionary generation device 11 inputs a necessary amount of reference vocabulary. Then, the synonym dictionary generation device 11 obtains a synonym for the input reference vocabulary and stores it in the synonym information storage unit 21. The user can use the stored information as a synonym dictionary.

<Effect>
According to the synonym dictionary generating apparatus 11 described above, the synonym characteristic that is likely to appear in a similar context and the recognition error characteristic that the similarity of notation and part of speech is low as opposed to reading similarity are used. Thus, it is determined whether the reference vocabulary and the related vocabulary are synonyms based on the context similarity, reading similarity, notation similarity, and part-of-speech similarity. For this reason, it is possible to accurately determine whether the reference vocabulary and the related vocabulary are synonyms based on the speech text without being affected by word dropout, insertion, recognition error, etc., and generate a synonym dictionary Can do.

<Other variations>
In the present embodiment, each data is exchanged via the storage unit 22, but data may be directly transmitted and received between each unit without using the storage unit 22.

  Note that the text information storage unit 13 stores (I) document text, (II) a set of words obtained from the document text, and (III) “collocation data” obtained from the document text, instead of the speech text. May be. A synonym dictionary can also be created based on document text.

In the present embodiment, when performing synonym determination processing (s208, s209), the part-of-speech similarity is used, but it is not essential. The synonym determination processing may be performed using at least context similarity, reading similarity, and notation similarity. In this case, the vocabulary is expressed by word notation and reading. At this time, the synonym dictionary generation device 11 may not include each part related to the part of speech similarity (the part of speech similarity calculating unit 19 or the like), and the part of speech or the part of speech similarity in the processes related to the part of speech similarity (s202, s206, s207, and s209). Processing), data (data relating to part of speech and part of speech similarity among the data stored in the vocabulary information storage unit 16, the synonym information storage unit 21, and the storage unit 22) (and thus the vocabulary). Information includes reading and notation, not part of speech.) The synonym determination unit 20 calculates a synonym index by the following formula.
S _vocab (u, v) = S _context (u, v) + γ ・ S _describe (u, v) + δ ・ S _pronounce (u, v)
(0 ≦ S _context (u, v), S _describe (u, v), S _pronounce (u, v) ≦ 1, γ> 0, δ <0) (3)
Or
S _vocab (u, v) = S _context (u, v) × s _α (S _describe (u, v) -S _pronounce (u, v)) (α> 0) (4)
The above formula is obtained by _setting S _POS = 0 in the formulas (1) and (2). Since the part of speech is not used as a material for determining whether or not it is a synonym, the accuracy may slightly decrease. However, since it is considered that the characteristics of recognition errors can be used only from notation and reading, an almost comparable accuracy can be expected, and the amount of calculation can be reduced.

  The present invention is not limited to the above-described embodiments and modifications. For example, the various processes described above are not only executed in time series according to the description, but may also be executed in parallel or individually as required by the processing capability of the apparatus that executes the processes. In addition, it can change suitably in the range which does not deviate from the meaning of this invention.

<Program and recording medium>
The above-mentioned synonym dictionary generation device can also be operated by a computer. In this case, each process of a program for causing a computer to function as a target device (a device having the functional configuration shown in the drawings in various embodiments) or a processing procedure (shown in each embodiment) is processed by the computer. A program to be executed by the computer may be downloaded from a recording medium such as a CD-ROM, a magnetic disk, or a semiconductor storage device or via a communication line into the computer and executed.

Claims (7)

A context that includes a reference vocabulary used as a reference when creating a synonym dictionary, and a context similarity calculation unit that calculates a similarity between contexts including related vocabulary related to the reference vocabulary;
A notation similarity calculator that calculates the similarity between the notation of the reference vocabulary and the notation of the related vocabulary;
A reading similarity calculator for calculating the similarity between the reading of the reference vocabulary and the reading of the related vocabulary;
A synonym indicator for a reference vocabulary and related vocabulary indicates the likelihood that the reference vocabulary and the related vocabulary are synonyms, and that the context of the reference vocabulary and the context of the related vocabulary are more likely to be similar. Show that the notation of the reference vocabulary and the notation of the related vocabulary are more likely to be similar, and that the reading of the reference vocabulary and the reading of the related vocabulary are less likely to be similar, Whether a synonym index for the reference vocabulary and the related vocabulary is obtained using the calculated context, notation, and reading similarity, and whether the related vocabulary is a synonym of the reference vocabulary based on the size of the synonym index A synonym determination unit for determining whether or not,
Synonym dictionary generator. The synonym dictionary generation device according to claim 1,
Context, notation and each _S context similarity reading _{(u, v), S describe} (u, v) and _{S pronounce} (u, v) and then, to the synonymous indicator _{S vocab} (u, v) and, s _alpha Is a sigmoid function of gain α, and the synonym determination unit determines the synonym index as
S _vocab (u, v) = S _context (u, v) × s _α (S _describe (u, v) -S _pronounce (u, v))
Asking,
Synonym dictionary generator. The synonym dictionary generation device according to claim 1,
A part-of-speech similarity calculator that calculates the similarity between the part of speech of the reference vocabulary and the part of speech of the related vocabulary;
The similarities of context, notation, reading, and part of speech are S _context (u, v), S _describe (u, v), S _pronounce (u, v), and S _POS (u, v), respectively, and the synonymous index is S _vocab (u, v), s _α is a sigmoid function of gain α, β is a weighting factor, and the synonym determination unit determines the synonym index as
S _vocab (u, v) = (S _context (u, v) + β ・ S _POS (u, v)) × s _α (S _describe (u, v) -S _pronounce (u, v))
Asking,
Synonym dictionary generator. The synonym dictionary generation device according to any one of claims 1 to 3,
Using the reference vocabulary, a related vocabulary acquisition unit that acquires a related vocabulary related to the reference vocabulary;
A text information storage unit for storing a large amount of text information;
A context acquisition unit for acquiring a context including the reference vocabulary and a context including the related vocabulary from the text information storage unit;
The synonym determination unit outputs a combination of a reference vocabulary and a related vocabulary determined to be synonymous with the reference vocabulary.
Synonym dictionary generator. The synonym dictionary generation device according to claim 4,
The reference vocabulary, the related vocabulary determined to be synonymous with the reference vocabulary, the vocabulary information of the reference vocabulary and the related vocabulary, the similarities between the reference vocabulary and the related vocabulary, and the synonyms A synonym information storage unit for storing the index;
Synonym dictionary generator. A context similarity calculation step of calculating a similarity between a context including a reference vocabulary used as a reference when creating a synonym dictionary and a context including a related vocabulary related to the reference vocabulary;
A notation similarity calculating step of calculating the similarity between the notation of the reference vocabulary and the notation of the related vocabulary;
A reading similarity calculation step of calculating the similarity between the reading of the reference vocabulary and the reading of the related vocabulary;
A synonym index that indicates the likelihood that a reference vocabulary and related vocabulary are synonyms indicates that the similarity of the context of the reference vocabulary and the context of the related vocabulary is more likely, and the notation of the reference vocabulary and the related vocabulary The similarity of the notation is more likely, the reading of the reference vocabulary and the reading of the related vocabulary are not likely to be similar, and the similarity of the calculated context, notation and reading A synonym determination step of determining whether or not the related vocabulary is a synonym of the reference vocabulary based on the size of the synonym index, and obtaining a synonym index for the reference vocabulary and the related vocabulary using gender; including,
Synonym dictionary generation method.   The program for functioning a computer as a synonym dictionary production | generation apparatus in any one of Claim 1-5.

Images