Server device, method, and program for generating semantic map

Abstract

The present invention generates a semantic map that functions as a bilingual dictionary. A server device 20 comprises: a communication device 23 that is connected to a client device 30; and an arithmetic processing device 21 for generating a semantic map in response to a request from the client device 30, and transmitting the semantic map to the client device 30 through the communication device 23. The semantic map includes a plurality of nodes and a plurality of links. Each link connects two nodes among the plurality of nodes. Each node is associated with a word, and a link connecting two nodes represents a translation relationship between two words that are respectively associated with the two nodes.

Description

Server device, method and program for generating a semantic map

[0001] The present invention relates to a server device, a method, and a program for generating a semantic map.

[0002] A semantic network is a model of knowledge representation, in which knowledge is represented using a graph. For example, words are represented as nodes in the graph, and the relationship between two words is represented by links connecting the nodes. This is a model of human associative memory. For example, words associated with a particular word can be highly related or less related, and in a semantic network, each word is linked according to the strength of these associations. By using such a semantic network to define, for example, the associations between words or the causal relationships between words, computers that cannot understand the meaning of words can formally process language.

[0003] In conventional semantic networks, for example, links are formed between words with similar meanings, or between causal relationships between words, but no network has been proposed that links translation relationships (edges). Such networks are expected to be useful for representing knowledge with rich functionality, such as bilingual dictionaries and thesauruses.

[0004] Therefore, an object of the present invention is to propose a technique for generating and searching a semantic network that functions as a bilingual dictionary, a thesaurus, or the like.

[0005] To solve the above-mentioned problems, the present invention provides a unique semantic map. This semantic map includes a plurality of nodes and a plurality of links, each link connecting two of the plurality of nodes, each node corresponding to a word, and the link connecting two nodes representing the translation relationship between the two words corresponding to each of the two nodes. To explore this network, the present invention also provides a server device including a communication device connected to a client device and a processing unit that generates a semantic map in response to a request from the client device and transmits the semantic map to the client device via the communication device.

[0006] According to the present invention, it is possible to generate and search a semantic network that functions as a bilingual dictionary, a thesaurus, or the like.

[0007] FIG. 1 is a diagram showing the hardware configuration of a system for generating a semantic map according to an embodiment of the present invention. FIG. 2 is an explanatory diagram showing the configuration of a semantic network according to an embodiment of the present invention. FIG. 3 is an explanatory diagram showing the configuration of a semantic space according to an embodiment of the present invention. FIG. 4 is an explanatory diagram showing the configuration of a semantic map according to an embodiment of the present invention. FIG. 5 is a diagram of a table showing the number of times each word appears in a bilingual corpus. FIG. 6 is a diagram of Table 1 showing the number of times each word in two different languages ​​appears in a bilingual corpus. FIG. 7 is a diagram showing an example of a bilingual corpus, which is a table showing the number of times one of two words in different languages ​​is translated into the other in the bilingual corpus. FIG. 8 is a flowchart showing the processing flow of a server device generating a semantic map.

[0008] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Here, the same reference numerals denote the same components, and redundant description will be omitted.

[0009] 1 shows the hardware configuration of a system 10 for generating a semantic map according to an embodiment of the present invention. The system 10 includes a server device 20 and a client device 30 that are connected to each other via a communication network 40.

[0010] The server device 20 is a web application server for generating a semantic map, and includes a processing unit 21, a main memory 22, a communication unit 23, and a storage unit 24.

[0011] The arithmetic processing device 21 is an integrated circuit including, for example, a central processing unit (CPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or a graphic processing unit (GPU).

[0012] The storage device 24 is a computer-readable recording medium such as a disk medium (e.g., a magnetic recording medium or a magneto-optical recording medium) or a semiconductor memory (e.g., a volatile memory or a non-volatile memory). Such a recording medium may also be called, for example, a non-transitory recording medium. The storage device 24 stores a program 25 for generating a semantic map and a bilingual corpus 26.

[0013] The program 25 is a computer program having instructions for causing the processor 21 to execute the method for generating a semantic map according to this embodiment. The program 25 is read from the storage device 24 into the main memory 22 and executed by the processor 21.

[0014] The bilingual corpus 26 is document data that collects bilingual expressions. A bilingual expression means a pair of an expression in one language and its corresponding expression in another language. For example, bilingual expressions include not only the pair of the Japanese word "inu" (dog) and its corresponding English word "dog," but also the pair of the Japanese sentence "I am looking forward to meeting you" and its corresponding English sentence "I am looking forward to seeing you."

[0015] The communication device 23 is, for example, a communication interface that performs HTTP (Hyper Text Transfer Protocol) communication with the client device 30 via the communication network 40. The communication network 40 is, for example, an IP (Internet Protocol) network that includes a combination of wired and wireless networks.

[0016] The client device 30 is a terminal device that receives a semantic map from the server device 20, and includes, for example, a personal computer with a web browser installed, a smartphone, etc. The client device 30 includes a processing unit 31, a main memory 32, a storage device 33, a display device 34, an input device 35, and a communication device 36.

[0017] The hardware configurations of the processing unit 31, main memory 32, storage device 33, and communication device 36 of the client device 30 are similar to the hardware configurations of the processing unit 21, main memory 22, storage device 24, and communication device 23 of the server device 20, respectively.

[0018] The input device 35 is, for example, a keyboard or a mouse that allows a user to input instructions (e.g., an instruction to generate a semantic map) to the server device 20 through the client device 30. The display device 34 is, for example, a liquid crystal display or an organic electroluminescence display that displays the semantic map received from the server device 20.

[0019] Next, a semantic network, a semantic space, and a semantic map according to an embodiment of the present invention will be described with reference to Figures 2 to 4. Figure 2 is an explanatory diagram showing the configuration of a semantic network according to an embodiment of the present invention. Figure 3 is an explanatory diagram showing the configuration of a semantic space according to an embodiment of the present invention. Figure 4 is an explanatory diagram showing the configuration of a semantic map according to an embodiment of the present invention.

[0020] In an embodiment of the present invention, a semantic network is defined as follows (1) and (2): (1) Each node corresponds to a single word. The weight of a node increases as the number of times a word corresponding to the node appears in the bilingual corpus 26 increases. (2) Each link connects one node corresponding to a word in a language with another node corresponding to a word in a language different from the given language. The two words corresponding to the two nodes connected by the link are in a relationship in which one is a translation of the other (hereinafter referred to as a "translation relationship"). The weight of a link increases as the number of times a translation between the two words corresponding to the two nodes connected by the link appears in the bilingual corpus 26 increases.

[0021] In the semantic network shown in Fig. 2, symbol Ni represents the i-th node, and symbol Lij represents the link connecting the i-th node and the j-th node. For convenience of explanation, the example shown in Fig. 2 shows a case where i and j are integers ranging from 1 to 5, but the values ​​of i and j may be any positive integer.

[0022] In the example shown in FIG. 2 , the English word "beautiful" is associated with node N1, and the Japanese word "utsukushii" is associated with node N4. Link 14 connects node 1 and node 4. The weight of node N1 increases as the number of times the word "beautiful" appears in the bilingual corpus 26 increases. The weight of node N4 increases as the number of times the word "utsukushii" appears in the bilingual corpus 26 increases. The weight of link 14 increases as the number of times the translation between the two words ("beautiful" and "utsukushii") connected by link 13 appears in the bilingual corpus 26 increases.

[0023] Note that the words "lovely," "wonderful," and "nice" correspond to nodes 2, 3, and 5, respectively, and the weight settings for these nodes and the weight settings for the links between the nodes are similar to the weight settings for node 1 and the weight settings for the link between nodes 1 and 4 described above.

[0024] Here, Fig. 5 is a diagram of a table T1 showing the number of times each word in a certain language (e.g., English) appears in the bilingual corpus 26. In the example shown in Fig. 5, the number of times the word "beautiful" appears in the bilingual corpus 26 is NE1. The weight of node N1 is set so that it increases as the value of NE1 increases. The same applies to other words besides "beautiful".

[0025] 6 is a diagram of table T2 showing the number of times each word in a language (e.g., Japanese) different from the above-mentioned certain language (e.g., English) appears in the bilingual corpus 26. The example shown in FIG. 6 shows a case where the number of times the word "beautiful" appears in the bilingual corpus 26 is NJ1. The weight of node N4 is set to be heavier as the value of NJ1 increases. The same applies to other words besides "beautiful."

[0026] 7 is a diagram of table T3 showing the number of times one of two words in different languages ​​has been translated into the other in the bilingual corpus 26. In the example shown in FIG. 7, the number of times the English word "beautiful" has been translated into the Japanese word "utsukushii" in the bilingual corpus 26 is NT1. The weight of the link 14 increases as the value of NT1 increases. The same applies to combinations other than the combination of "beautiful" and "utsukushii."

[0027] In an embodiment of the present invention, the semantic space is defined as follows: (3) A repulsive force virtually acts between two different words in the same language. The more frequently one of two different words in the same language appears in the bilingual corpus 26, the stronger the repulsive force acting from one of the two words to the other. (4) An attractive force virtually acts between two words in different languages. The more frequently one of two words in different languages ​​is translated into the other in the bilingual corpus 26, the stronger the attractive force acting from one of the two words to the other. (5) The coordinates of a node corresponding to a word are the position where the repulsive and attractive forces are balanced.

[0028] The reason why a virtual repulsive force acts between different words in the same language, so that the more frequently used a word is, the more it repels other words, is based on the idea that "words with the same meaning are unnecessary." The reason why a virtual attractive force acts between words in different languages, so that the more frequently translated words are, the more strongly they attract each other, is based on the idea that "meaning should be preserved in translation." The coordinates of the node corresponding to a word are the position where the virtual repulsive and attractive forces acting on the word are balanced, so the distance between words can be visually expressed as the difference in meaning between the words.

[0029] In the semantic space shown in FIG. 3 , symbol Fr34 indicates the repulsive force acting from the word "wonderful" corresponding to node N3 to the word "beautiful" corresponding to node N4. The more frequently the word "wonderful" appears in the bilingual corpus 26, the stronger the magnitude of the repulsive force Fr34. Similarly, symbol Fr43 indicates the repulsive force acting from the word "beautiful" corresponding to node N4 to the word "wonderful" corresponding to node N3. The more frequently the word "beautiful" appears in the bilingual corpus 26, the stronger the magnitude of the repulsive force Fr43.

[0030] On the other hand, symbol Fa14 indicates the attractive force acting from the word "beautiful" corresponding to node N1 to the word "utsukushii" corresponding to node N4. The more times the word "beautiful" is translated into the word "utsukushii" in the bilingual corpus 26, the stronger the magnitude of the attractive force Fa14. Similarly, symbol Fa41 indicates the attractive force acting from the word "utsukushii" corresponding to node N4 to the word "beautiful" corresponding to node N1. The more times the word "utsukushii" is translated into the word "beautiful" in the bilingual corpus 26, the stronger the magnitude of the attractive force Fa41.

[0031] For ease of explanation, Fig. 3 shows, as an example of a repulsive force, hypothetical repulsive forces Fr34 and Fr43 acting between two different words in the same language, "wonderful" and "beautiful," but repulsive forces also act between two different words in the same language other than these words. Similarly, Fig. 3 shows, as an example of an attractive force, hypothetical attractive forces Fa14 and Fa41 acting between two different words in different languages, "beautiful," but attractive forces also act between two different words in different languages ​​other than these words.

[0032] As shown in Fig. 4, the semantic map is generated using a semantic network (Fig. 2) that defines information on the connection relationships between each node and each link, and a semantic space (Fig. 3) that defines information on the coordinates of each node calculated from the position where the attractive and repulsive forces acting virtually on the word corresponding to each node are balanced. The semantic map is generated by matching the coordinates of each node on the semantic network to the coordinates of each node defined by the semantic space, while maintaining the connection relationships between each node and each link defined by the semantic network.

[0033] Here, with reference to FIG. 8, a method for generating tables T1, T2, and T3 shown in FIGS. 5 to 7 will be described. FIG. 8 shows an example of the bilingual corpus 26. In this example, a pair of the English sentence "I ate an apple" and the Japanese sentence "I ate an apple" is shown as a bilingual expression. The English sentence "I ate an apple" can be divided into multiple words, "I," "ate," "an," and "apple." Similarly, the Japanese sentence "I ate an apple" can be divided into multiple words, "watashi," "wa," "ringo," "wo," and "taeta." The words "I," "ate," and "apple" can be associated with the words "watashi," "ringo," and "taeta," respectively. Therefore, table T3 shown in FIG. 7 can be generated from these correspondences. Furthermore, table T1 shown in FIG. 5 can be generated by counting the number of times each word in a certain language (e.g., English) appears in the bilingual corpus 26. Similarly, by counting the number of times each word in a language (e.g., English) and a different language (e.g., Japanese) appears in the bilingual corpus 26, table T2 shown in FIG. 6 can be generated.

[0034] Next, the flow of processing by the server device 10 to generate a semantic map will be described with reference to Fig. 9. The processing unit 21 executes the program 25, and the semantic map is generated through the processing of steps 901 to 905, which will be described later.

[0035] First, in step 901 , the processing unit 21 reads the bilingual corpus 26 from the storage unit 24 .

[0036] Next, in step 902, the processing unit 21 extracts correspondences between words in a certain language and words in a language different from that language for each bilingual expression included in the bilingual corpus 26, and generates table T3 shown in Fig. 7. The processing unit 21 also generates table T1 shown in Fig. 5 by counting the number of times each word in a certain language (e.g., English) appears in the bilingual corpus 26. Similarly, the processing unit 21 generates table T2 shown in Fig. 6 by counting the number of times each word in a language (e.g., Japanese) different from the certain language (e.g., English) appears in the bilingual corpus 26.

[0037] Next, in step 903, the processing unit 21 generates a semantic network that defines the information on the connection relationships between each node and each link from the information in tables T1, T2, and T3 based on the definitions (1) and (2) above.

[0038] Next, in step 904, the processing unit 21 generates a semantic space that specifies the coordinate information of the node calculated from the position where the attractive and repulsive forces acting virtually on the word corresponding to each node are balanced from the information in tables T1, T2, and T3, based on the definitions (3) to (5) above.

[0039] Next, in step 905, the processing unit 21 generates a semantic map by matching the coordinates of each node on the semantic network with the coordinates of each node defined by the semantic space generated in step 904, while maintaining the connection relationships between each node and each link defined by the semantic network generated in step 903.

[0040] In response to a request from the client device 30, the processing unit 21 generates a semantic map through the processing of steps 901 to 905, and transmits the generated semantic map to the client device 30 via the communication unit 23.

[0041] Next, we will explain how to calculate the coordinates of the nodes corresponding to a word from the attractive and repulsive forces that virtually act on the word. １ , l ２ , ...}, and W （ｌ） Let us consider a set of words {W １ （ｌ） , W ２ （ｌ） , ...}. The coordinates of the node corresponding to word w in the d-dimensional space are set to Xw.

[0042] As shown in equation (1), the coordinates of the node corresponding to word w are calculated by the energy loss E due to the repulsive force acting virtually on word w. ｒｅｐ and the energy loss E due to the virtual attractive force acting on the word w. ａｔｔ Tonowa E ｔｏｔａｌ This can be calculated under the condition that E is minimized when the attractive and repulsive forces acting on the word are balanced. ｔｏｔａｌ is the smallest.

[0043] Here, E ｒｅｐ can be written as equation (2), and E ａｔｔ can be written as in equation (3).

[0044] The function rep in equation (2) is a function that outputs the energy loss caused by the repulsive force that virtually acts on the word w, and this function can be written as in equation (4). Furthermore, the function att in equation (3) is a function that outputs the energy loss caused by the attractive force that virtually acts on the word w, and this function can be written as in equation (5).

[0045] In equation (4), a repulsive force acting virtually on a word is modeled after the repulsive force generated by an electrostatic charge. ｕ and q ｖ are the values ​​of the electric charges when the word u and the word v are regarded as virtual static charges. ｕThe value of q increases as the number of times that the word u appears in the bilingual corpus 26 increases. ｖ The value of ∑ i = ...

[0046] In equation (5), the elastic force of a spring is modeled as the attractive force acting virtually on the word. ｕ，ｖ represents a spring constant when a virtual spring is assumed to be connected between the word u and the word v. The more times one of the words u and v has been translated into the other in the bilingual corpus 26, the greater the k ｕ，ｖ The value of becomes larger.

[0047] In a semantic map according to an embodiment of the present invention, multiple words from two or more different languages ​​are arranged in a space, with words that have been translated more frequently being connected by shorter links. Each word is placed in a predetermined position in the space according to its meaning, and the length of the link connecting two words is adjusted depending on the number of times those two words have been translated in the past. If sufficient data on the number of translations cannot be obtained, it is also assumed that the edge weights and coefficients of attraction and repulsion will be designed at the discretion of experts in each language. (For example, an unweighted graph is also conceivable in which edges are connected if the translation relationship is listed in a Japanese-English dictionary and an English-Japanese dictionary, and edges are not connected if the translation relationship is not listed; this is also within the scope of the present invention.)

[0048] The semantic map according to the embodiment of the present invention can aggregate information distributed across multiple dictionaries into a single map. For example, the semantic map according to the embodiment of the present invention can serve as a Japanese thesaurus, an English thesaurus, an English-Japanese dictionary, and a Japanese-English dictionary, and can therefore simultaneously fulfill the roles of these four dictionaries.

[0049] The semantic map according to the embodiment of the present invention makes it possible to grasp the nuance of one of two different words in two different languages ​​connected by a link from the other word. For example, if a user is proficient in one of two different languages ​​and unfamiliar with the other, the user can understand the nuance of the word in the unfamiliar language by referring to the semantic map according to the embodiment of the present invention, using the sense of the proficient language.

[0050] The semantic map according to the embodiment of the present invention allows a user to understand how frequently two different words in two different languages ​​have been translated in the past, thereby enabling the user to grasp the overall picture of the concepts expressed by these words by considering not only the translation relationship between the two different words in two different languages ​​but also the translation relationship between the two different words and other words that have similar meanings to the two different words.

[0051] As described above, a user can use a semantic map according to an embodiment of the present invention as a thesaurus (e.g., a Japanese thesaurus, an English thesaurus, etc.) or a bilingual dictionary (e.g., an English-Japanese dictionary, a Japanese-English dictionary, etc.), intuitively understand the nuances of words in the language they want to learn by utilizing the sense of the language they use on a daily basis, and grasp the overall picture of the concepts expressed by multiple words from the translation relationships between these words.

[0052] The above-described embodiments are intended to facilitate understanding of the present invention and are not intended to limit the present invention. The present invention may be modified or improved without departing from the spirit thereof, and equivalents thereof are also included in the present invention. In other words, designs that are appropriately modified by a person skilled in the art to the embodiments are also included within the scope of the present invention as long as they comprise the features of the present invention. Furthermore, the elements of the embodiments can be combined to the extent technically possible, and such combinations are also included within the scope of the present invention as long as they comprise the features of the present invention.

[0053] REFERENCE SIGNS LIST 10... System 20... Server device 21... Processing device 22... Main memory 23... Communication device 24... Storage device 25... Program 26... Bilingual corpus 30... Client device 31... Processing device 32... Main memory 33... Storage device 34... Display device 35... Input device 36... Communication device 40... Communication network

Claims

1. A server device comprising: a communication device connected to a client device; and an arithmetic processing device that generates a semantic map in response to a request from the client device and transmits the semantic map to the client device via the communication device, wherein the semantic map includes a plurality of nodes and a plurality of links, each link connecting two nodes among the plurality of nodes, each node corresponding to a word, and the link connecting the two nodes representing the translation relationship between the two words corresponding to each of the two nodes.

2. A server device as described in claim 1, wherein the calculation processing device is configured to calculate repulsive and attractive forces virtually acting on the word based on a bilingual corpus of two different languages, and to calculate the coordinates of the node corresponding to the word on the semantic map as a position where the repulsive and attractive forces are balanced, the repulsive force being a force virtually acting between two different words of the same language, and the more frequently one of the two words appears in the bilingual corpus, the stronger the magnitude of the repulsive force acting from one of the two words to the other word, and the attractive force being a force virtually acting between two words of different languages, and the more frequently one of the two words is translated into the other in the bilingual corpus, the stronger the magnitude of the attractive force acting from one of the two words to the other word.

3. A server device as described in claim 2, wherein the calculation processing device is configured to calculate the coordinates of the node under the condition that the sum of the energy loss caused by the repulsive force virtually acting between two different words in the same language and the energy loss caused by the attractive force virtually acting between two words in different languages ​​is minimized.

4. A method comprising: a server device generating a semantic map in response to a request from a client device; and the server device transmitting the semantic map to the client device, wherein the semantic map includes a plurality of nodes and a plurality of links, each link connecting two nodes among the plurality of nodes, each node corresponding to a word, and the link connecting the two nodes representing a translation relationship between two words corresponding to each of the two nodes.

5. A program that causes a server device to execute the following operations: generating a semantic map in response to a request from a client device; and transmitting the semantic map to the client device, wherein the semantic map includes a plurality of nodes and a plurality of links, each link connecting two nodes among the plurality of nodes, each node corresponding to a word, and the link connecting the two nodes representing the translation relationship between the two words corresponding to each of the two nodes.

Images