Argumentation device and argumentation method

Abstract

An argumentation device (2) is constituted to comprise: a first search window acquisition unit (12) for acquiring a first search window indicating an analysis range when estimating each argumentation element in a text including a plurality of argumentation elements; and a second search window acquisition unit (13) for acquiring a second search window indicating an analysis range in the text when estimating an argumentation relationship between the plurality of argumentation elements. Furthermore, the argumentation device (2) is provided with: an argumentation element estimation unit (14) for analyzing the text within a range included in the first search window while shifting the position where the first search window is applied to the text, and estimating each of the argumentation elements on the basis of the analysis result of the text; and an argumentation relationship estimation unit (15) for applying the second search window to the text, analyzing the text within the range included in the second search window, and estimating, on the basis of the analysis result of the text, an argumentation relationship between the plurality of argumentation elements estimated by the argumentation element estimation unit (14).

Description

Argumentation device and argumentation method

[0001] The present disclosure relates to a proof device and a proof method.

[0002] There is an argumentation device that infers multiple argument elements contained in a text and infers the argumentation relationships between the multiple argument elements. For example, Non-Patent Document 1 discloses an argumentation device that analyzes a text and, based on the results of the text analysis, infers multiple argument elements and infers the argumentation relationships between the multiple argument elements.

[0003] Muhammad Tawsif Sazid and Robert E. Mercer. 2022. “A Unified Representation and a Decoupled Deep Learning Architecture for Argumentation Mining of Students' Persuasive Essays.” In Proceedings of the 9th Workshop on Argument Mining, pages 74-83, Online and in Gyeongju, Republic of Korea. International Conference on Computational Linguistics.

[0004] The argumentation device disclosed in Non-Patent Document 1 has a problem in that the range of text analysis when inferring argument elements and the range of text analysis when inferring argument relations are both fixed. For example, if the range of text analysis when inferring argument elements is too broad, the efficiency of inferring argument elements may deteriorate, and if the range of text analysis when inferring argument elements is too narrow, the accuracy of inferring argument elements may deteriorate. For example, if the range of text analysis when inferring argument relations is too narrow, the accuracy of inferring argument relations may deteriorate.

[0005] The present disclosure has been made to solve the above-mentioned problems, and aims to provide an argumentation device that can accept both the range of text analysis when inferring argument elements and the range of text analysis when inferring argumentation relationships.

[0006] The argumentation device according to the present disclosure includes a first search window acquisition unit that acquires a first search window indicating an analysis range for inferring each of the argument elements in text containing a plurality of argument elements, and a second search window acquisition unit that acquires a second search window indicating an analysis range for inferring an argument relation between the plurality of argument elements in the text. The argumentation device also includes an argument element inference unit that analyzes the text within the range included in the first search window while shifting the position at which the first search window is applied to the text, and infers each of the argument elements based on the analysis results of the text, and an argument relation inference unit that applies the second search window to the text, analyzes the text within the range included in the second search window, and infers the argument relation between the plurality of argument elements inferred by the argument element inference unit based on the analysis results of the text.

[0007] According to the present disclosure, it is possible to accept both the range of text analysis when inferring argument elements and the range of text analysis when inferring argument relations.

[0008] FIG. 1 is a configuration diagram showing a proof apparatus 2 according to a first embodiment. FIG. 2 is a hardware configuration diagram showing the hardware of the proof apparatus 2 according to the first embodiment. FIG. 3 is a hardware configuration diagram of a computer in the case where the proof apparatus 2 is realized by software, firmware, or the like. FIG. 4 is a flowchart showing an argumentation method, which is a processing procedure of the proof apparatus 2. FIG. 5 is an explanatory diagram showing the input and output of a language model. FIG. 6 is an explanatory diagram showing the relationship between an original vector, a DS vector, and a first search window. FIG. 7 is a configuration diagram showing a proof apparatus 2 according to a second embodiment. FIG. 8 is a hardware configuration diagram showing the hardware of the proof apparatus 2 according to the second embodiment. FIG. 9 is an explanatory diagram showing groups grouped by the argumentation relation estimation unit 16. FIG. 10 is a configuration diagram showing a proof apparatus 2 according to a third embodiment. FIG. 11 is a hardware configuration diagram showing the hardware of the proof apparatus 2 according to the third embodiment. FIG. 12 is an explanatory diagram showing element labels indicating each argument element.

[0009] In order to explain the present disclosure in more detail, embodiments of the present disclosure will be described below with reference to the accompanying drawings.

[0010] Embodiment 1. Fig. 1 is a configuration diagram showing an argumentation device 2 according to embodiment 1. Fig. 2 is a hardware configuration diagram showing the hardware of argumentation device 2 according to embodiment 1. In Fig. 1, a text output device 1 outputs text containing a plurality of argumentation elements to argumentation device 2. Argumentation device 2 includes a text acquisition unit 11, a first search window acquisition unit 12, a second search window acquisition unit 13, an argumentation element estimation unit 14, and an argumentation relationship estimation unit 15. A display device 3 displays a plurality of argumentation elements and argumentation relationships on a display (not shown) as the estimation results of argumentation device 2.

[0011] The text acquisition unit 11 is realized by, for example, a text acquisition circuit 21 shown in Fig. 2. The text acquisition unit 11 acquires text containing multiple argument elements from the text output device 1. The text acquisition unit 11 outputs the text to each of the argument element estimation unit 14 and the argument relation estimation unit 15.

[0012] The first search window acquisition unit 12 is realized, for example, by the first search window acquisition circuit 22 shown in FIG. 2. The first search window acquisition unit 12 acquires a first search window indicating an analysis range for inferring each argument element from a text containing multiple argument elements. The first search window acquisition unit 12 outputs the first search window to the argument element estimation unit 14. The first search window may be provided to the first search window acquisition unit 12 from outside via a man-machine interface (not shown), for example, or may be provided to the first search window acquisition unit 12 from outside via a network interface (not shown).

[0013] The second search window acquisition unit 13 is realized, for example, by the second search window acquisition circuit 23 shown in FIG. 2 . The second search window acquisition unit 13 acquires a second search window indicating an analysis range when estimating an argumentation relationship between multiple argumentation elements in a text. The second search window is, for example, a search window with a wider analysis range than the first search window. The second search window acquisition unit 13 outputs the second search window to the argumentation relationship estimation unit 15. The second search window may be provided to the second search window acquisition unit 13 from outside via a man-machine interface (not shown), for example, or may be provided to the second search window acquisition unit 13 from outside via a network interface (not shown).

[0014] The argument element inference unit 14 is realized, for example, by the argument element inference circuit 24 shown in FIG. 2 . The argument element inference unit 14 acquires text from the text acquisition unit 11 and acquires a first search window from the first search window acquisition unit 12. The argument element inference unit 14 analyzes the text within the range included in the first search window while shifting the position at which the first search window is applied to the text. The argument element inference unit 14 infers each argument element based on the results of the text analysis. The argument element inference unit 14 outputs the inference results of each argument element to the argument relation inference unit 15 and the display device 3, respectively.

[0015] The argumentation relationship estimation unit 15 is realized, for example, by the argumentation relationship estimation circuit 25 shown in FIG. 2 . The argumentation relationship estimation unit 15 acquires text from the text acquisition unit 11 and acquires a second search window from the second search window acquisition unit 13. The argumentation relationship estimation unit 15 also acquires estimation results of each argumentation element from the argumentation element estimation unit 14. As the estimation results of the argumentation elements, for example, estimated labels or vectors representing words held by the argumentation element estimation unit 14 may be acquired. The argumentation relationship estimation unit 15 applies the second search window to the text and analyzes the text within the range included in the second search window. Based on the analysis results of the text, the argumentation relationship estimation unit 15 estimates argumentation relationships between the multiple argumentation elements estimated by the argumentation element estimation unit 14. The argumentation relationship estimation unit 15 outputs the estimation results of the argumentation relationships to the display device 3.

[0016] 1, it is assumed that each of the components of the argumentation apparatus 2, namely, the text acquisition unit 11, the first search window acquisition unit 12, the second search window acquisition unit 13, the argument element inference unit 14, and the argument relation inference unit 15, is realized by dedicated hardware as shown in Fig. 2. In other words, it is assumed that the argumentation apparatus 2 is realized by a text acquisition circuit 21, a first search window acquisition circuit 22, a second search window acquisition circuit 23, an argument element inference circuit 24, and an argument relation inference circuit 25. Each of the text acquisition circuit 21, the first search window acquisition circuit 22, the second search window acquisition circuit 23, the argument element inference circuit 24, and the argument relation inference circuit 25 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination thereof.

[0017] The components of the argumentation device 2 are not limited to those realized by dedicated hardware, and the argumentation device 2 may be realized by software, firmware, or a combination of software and firmware. Software or firmware is stored as a program in the memory of a computer. A computer refers to hardware that executes a program, and includes, for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a central processing unit, a processing unit, an arithmetic unit, a microprocessor, a microcomputer, a processor, or a DSP (Digital Signal Processor).

[0018] 3 is a hardware configuration diagram of a computer when the argumentation device 2 is realized by software, firmware, etc. When the argumentation device 2 is realized by software, firmware, etc., programs for causing a computer to execute the processing procedures of the text acquisition unit 11, the first search window acquisition unit 12, the second search window acquisition unit 13, the argument element estimation unit 14, and the argumentation relation estimation unit 15 are stored in memory 31. Then, a processor 32 of the computer executes the programs stored in memory 31.

[0019] 2 shows an example in which each of the components of the argumentation apparatus 2 is realized by dedicated hardware, while Fig. 3 shows an example in which the argumentation apparatus 2 is realized by software, firmware, etc. However, this is merely an example, and some of the components in the argumentation apparatus 2 may be realized by dedicated hardware, and the remaining components may be realized by software, firmware, etc.

[0020] Next, the operation of the argumentation device 2 shown in Fig. 1 will be described. The text output device 1 outputs, for example, the following text to the argumentation device 2 as text containing a plurality of argumentation elements: [Text] [...] the significance of competition is that how to become more excellence to gain the victory. Therefore, it is always said that competition makes the society more effective. [...]

[0021] 4 is a flowchart showing the argumentation method, which is a processing procedure of the argumentation device 2. The text acquisition unit 11 of the argumentation device 2 acquires text containing multiple argument elements from the text output device 1 (step ST1 in FIG. 4). The text acquisition unit 11 may output the acquired text as is to each of the argument element estimation unit 14 and the argument relation estimation unit 15, but here the text acquisition unit 11 performs morphological analysis of the acquired text and outputs vectors indicating each word to each of the argument element estimation unit 14 and the argument relation estimation unit 15.

[0022] Specifically, as shown in FIG. 5 , the text acquisition unit 11 provides each word included in the text to a language model, acquires vectors indicating each word (hereinafter referred to as "original vectors") from the language model, and outputs each original vector to the argument element estimation unit 14. FIG. 5 is an explanatory diagram showing input and output of the language model. Since the language model itself is a well-known model, detailed description will be omitted. In the example of FIG. 5 , for example, when the language model is given the word "the", it outputs an original vector indicating "the", and when the language model is given the word "significance", it outputs an original vector indicating "significance".

[0023] The first search window acquisition unit 12 acquires a first search window from an external device (step ST2 in FIG. 4). The first search window indicates an analysis range for inferring each argument element in a text containing multiple argument elements. The analysis range indicated by the first search window may be, for example, a range including two words, a range including three words, or a range including four words. The first search window acquisition unit 12 outputs the first search window to the argument element estimation unit 14.

[0024] The second search window acquisition unit 13 acquires a second search window from the outside (step ST3 in FIG. 4). The second search window is a search window that indicates the analysis range when estimating the argumentation relationship between multiple argumentation elements in a text that includes multiple argumentation elements. The analysis range indicated by the second search window is, for example, a range that includes all words contained in the text acquired by the text acquisition unit 11. The second search window acquisition unit 13 outputs the second search window to the argumentation relationship estimation unit 15.

[0025] The argument element estimation unit 14 acquires original vectors representing each word from the text acquisition unit 11 and acquires a first search window from the first search window acquisition unit 12. When acquiring text itself from the text acquisition unit 11, the argument element estimation unit 14 simply identifies each of the multiple words contained in the text and acquires original vectors representing each word. The argument element estimation unit 14 analyzes the text within the range included in the first search window while shifting the position at which the first search window is applied to the text. The argument element estimation unit 14 estimates each argument element based on the results of the text analysis (step ST4 in FIG. 4). The argument element estimation unit 14 outputs the estimation results of each argument element to the argument relation estimation unit 15 and the display device 3. The argument element estimation process performed by the argument element estimation unit 14 will be described in detail below.

[0026] If the analysis range indicated by the first search window includes, for example, three words, the argument element estimation unit 14 first applies the first search window to the first to third original vectors from the top of the multiple original vectors, as shown in FIG. 6 . In this case, an original vector indicating the word “the,” an original vector indicating the word “significance,” and an original vector indicating the word “of” are provided to a machine learning model (not shown). The machine learning model performs downsampling to compress these three original vectors into one and outputs a downsampled vector (hereinafter referred to as a “DS vector”) related to these three words. The DS vector potentially contains information on the three words “the significance of,” and the initial “the” potentially contains information indicating that it is a “premise.” The downsampling process itself is a well-known technique, so a detailed description will be omitted. FIG. 6 is an explanatory diagram showing the relationship between the original vector, the DS vector, and the first search window.

[0027] As the machine learning model, for example, a convolutional neural network (CNN) is used. In the learning process of the machine learning model, for example, a backpropagation algorithm is used. For example, the machine learning model is given training data in which the class "premise" is assigned to the word "the", and learns that the word "the" is an argument element called "premise". For example, the machine learning model is given training data in which the class "assertion" is assigned to the word "effective", and learns that the word "effective" is an argument element called "assertion".

[0028] Next, the argument element estimation unit 14 applies the first search window to the second to fourth original vectors from the top in the figure, among the multiple original vectors, by shifting the position at which the first search window is applied to the text by one. In this case, an original vector indicating the word "significance," an original vector indicating the word "of," and an original vector indicating the word "competition" are provided to the machine learning model. The machine learning model performs downsampling to compress these three original vectors into one, and outputs a DS vector related to these three words. The DS vector potentially contains information on the three words "significance of competition," and the leading "significance" potentially contains information indicating that it is a "premise."

[0029] The argument element estimation unit 14 shifts the position at which the first search window is applied to the text by a fixed distance, and ultimately applies the first search window to the original vectors from the third lowest to the bottom among the multiple original vectors in the figure. The fixed distance shift may be, for example, a shift by one position. In this case, an original vector representing the word "society," an original vector representing the word "more," and an original vector representing the word "effective" are provided to the machine learning model. The machine learning model performs downsampling to compress these three original vectors into one and outputs a DS vector related to these three words. The DS vector potentially contains information on the three words "society more effective," and the leading word "society" potentially contains information indicating that it is an "assertion."

[0030] Next, the argument element estimation unit 14 performs upsampling on the multiple DS vectors. Specifically, all DS vectors are provided to, for example, a second-layer CNN. The CNN performs upsampling on the multiple DS vectors and outputs upsampled vectors (hereinafter referred to as "US vectors") indicating each word contained in the text. The US vectors are similar to the original vectors in that they are vectors on a word-by-word basis, but differ from the original vectors in that they contain argument information indicating that the words are "premise" or "assertion," etc. The argument element estimation unit 14 may perform skip connection to add the original vectors to the US vectors output from the CNN.

[0031] The argument element estimation unit 14 refers to the argument information included in the US vector and estimates what kind of argument element each word included in the text is. The argument element estimation unit 14 outputs the estimation results of each argument element to both the argument relation estimation unit 15 and the display device 3.

[0032] The argumentation relationship estimation unit 15 acquires text from the text acquisition unit 11 and acquires a second search window from the second search window acquisition unit 13. The argumentation relationship estimation unit 15 also acquires vectors representing words, such as estimated labels, original vectors, US vectors, or DS vectors, from the argumentation element estimation unit 14 as estimation results for each argumentation element. The argumentation relationship estimation unit 15 applies the second search window to the text and analyzes the text within the range included in the second search window. Based on the results of the text analysis, the argumentation relationship estimation unit 15 estimates argumentation relationships between the multiple argumentation elements estimated by the argumentation element estimation unit 14 (step ST5 in FIG. 4 ). The argumentation relationship estimation process itself is a well-known technique, and therefore a detailed description thereof will be omitted. For example, regarding the argumentation relationship between an argumentation element that is a "premise" and an argumentation element that is an "assertion," the argumentation relationship estimation unit 15 estimates that the argumentation element that is the "assertion" supports the argumentation element that is the "premise." The argumentation relation estimation unit 15 outputs the result of estimating the argumentation relation to the display device 3 .

[0033] In the first embodiment described above, the argumentation apparatus 2 is configured to include a first search window acquisition unit 12 that acquires a first search window indicating an analysis range when inferring each argumentation element in text containing multiple argumentation elements, and a second search window acquisition unit 13 that acquires a second search window indicating an analysis range when inferring an argumentation relationship between the multiple argumentation elements in the text. The argumentation apparatus 2 also includes an argumentation element inference unit 14 that analyzes the text within the range included in the first search window while shifting the position at which the first search window is applied to the text, and infers each argumentation element based on the text analysis results, and an argumentation relationship inference unit 15 that applies the second search window to the text, analyzes the text within the range included in the second search window, and infers the argumentation relationship between the multiple argumentation elements inferred by the argumentation element inference unit 14 based on the text analysis results. Thus, the argumentation apparatus 2 can accept both the text analysis range when inferring argumentation elements and the text analysis range when inferring argumentation relationships.

[0034] Furthermore, in the first embodiment, the argumentation device 2 is configured so that the second search window acquisition unit 13 acquires, as the second search window, a search window with a wider analysis range than the first search window. Therefore, the argumentation device 2 can suppress a deterioration in the accuracy of inferring argumentation relations that occurs when the analysis range of the text when inferring argumentation relations is too narrow.

[0035] Embodiment 2 In embodiment 2, an argumentation device 2 will be described in which the argumentation relation estimation unit 16 groups adjacent identical argumentation elements, among the multiple argumentation elements estimated by the argumentation element estimation unit 14, into one group, and estimates argumentation relations between multiple different groups as argumentation relations between the multiple argumentation elements.

[0036] Fig. 7 is a block diagram showing an argumentation apparatus 2 according to a second embodiment. In Fig. 7, the same reference numerals as in Fig. 1 indicate the same or corresponding parts, and therefore detailed explanations will be omitted. Fig. 8 is a hardware block diagram showing the hardware of the argumentation apparatus 2 according to the second embodiment. In Fig. 8, the same reference numerals as in Fig. 2 indicate the same or corresponding parts, and therefore detailed explanations will be omitted. The argumentation apparatus 2 shown in Fig. 7 comprises a text acquisition unit 11, a first search window acquisition unit 12, a second search window acquisition unit 13, an argumentation element inference unit 14, and an argumentation relation inference unit 16.

[0037] The argumentation relationship estimation unit 16 is realized, for example, by the argumentation relationship estimation circuit 26 shown in FIG. 8 . The argumentation relationship estimation unit 16 acquires text from the text acquisition unit 11 and acquires a second search window from the second search window acquisition unit 13. The argumentation relationship estimation unit 16 also acquires estimation results of each argumentation element from the argumentation element estimation unit 14. Similar to the argumentation relationship estimation unit 15 shown in FIG. 1 , the argumentation relationship estimation unit 16 applies the second search window to the text and analyzes the text within the range included in the second search window. Similar to the argumentation relationship estimation unit 15 shown in FIG. 1 , the argumentation relationship estimation unit 16 estimates argumentation relationships between the multiple argumentation elements estimated by the argumentation element estimation unit 14 based on the results of the text analysis. However, the argumentation relation estimation unit 16 groups adjacent argumentation elements of the same kind among the multiple argumentation elements estimated by the argument element estimation unit 14 into one group, and estimates argumentation relations between multiple different groups as argumentation relations between the multiple argumentation elements based on the results of text analysis. The argumentation relation estimation unit 16 outputs the estimation results of the argumentation relations to the display device 3.

[0038] 7, it is assumed that each of the components of the argumentation apparatus 2, namely, the text acquisition unit 11, the first search window acquisition unit 12, the second search window acquisition unit 13, the argument element inference unit 14, and the argument relation inference unit 16, is realized by dedicated hardware such as that shown in FIG. 8. That is, it is assumed that the argumentation apparatus 2 is realized by a text acquisition circuit 21, a first search window acquisition circuit 22, a second search window acquisition circuit 23, an argument element inference circuit 24, and an argument relation inference circuit 26. Each of the text acquisition circuit 21, the first search window acquisition circuit 22, the second search window acquisition circuit 23, the argument element inference circuit 24, and the argument relation inference circuit 26 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC, an FPGA, or a combination of these.

[0039] The components of the argumentation device 2 are not limited to those realized by dedicated hardware, and the argumentation device 2 may be realized by software, firmware, or a combination of software and firmware. When the argumentation device 2 is realized by software, firmware, or the like, programs for causing a computer to execute the processing procedures of the text acquisition unit 11, the first search window acquisition unit 12, the second search window acquisition unit 13, the argument element estimation unit 14, and the argumentation relation estimation unit 16 are stored in memory 31 shown in Fig. 3. Then, a processor 32 shown in Fig. 3 executes the programs stored in memory 31.

[0040] 8 shows an example in which each of the components of the argumentation apparatus 2 is realized by dedicated hardware, while Fig. 3 shows an example in which the argumentation apparatus 2 is realized by software, firmware, etc. However, this is merely an example, and some of the components in the argumentation apparatus 2 may be realized by dedicated hardware, and the remaining components may be realized by software, firmware, etc.

[0041] Next, the operation of the argumentation apparatus 2 shown in Fig. 7 will be described. However, apart from the argumentation relation estimating unit 16, the argumentation apparatus 2 is the same as that shown in Fig. 1. Therefore, only the operation of the argumentation relation estimating unit 16 will be described here.

[0042] The argumentation relationship estimation unit 16, like the argumentation relationship estimation unit 15 shown in Fig. 1, applies a second search window to the text and analyzes the text within the range included in the second search window. Like the argumentation relationship estimation unit 15 shown in Fig. 1, the argumentation relationship estimation unit 16 estimates argumentation relationships between the multiple argument elements estimated by the argument element estimation unit 14 based on the results of the text analysis. However, as shown in Fig. 9, the argumentation relationship estimation unit 16 groups adjacent argument elements of the same type into one group among the multiple argument elements estimated by the argument element estimation unit 14. Fig. 9 is an explanatory diagram showing the groups grouped by the argumentation relationship estimation unit 16.

[0043] Based on the results of the text analysis, the argumentation relationship estimation unit 16 estimates argumentation relationships between multiple different groups as argumentation relationships between multiple argument elements. Specifically, the argumentation relationship estimation unit 16 identifies, among the multiple groups, for example, a group that groups together argumentation elements that are "premises," a group that groups together argumentation elements marked with "O," which indicates that the argumentation elements are neither "premises" nor "assertion," and a group that groups together argumentation elements that are "assertion." As the argumentation relationship between the group that groups together argumentation elements that are "premises" and the group that groups together argumentation elements that are "assertion," the argumentation relationship estimation unit 16 estimates that the group that groups together argumentation elements that are "assertion" supports the group that groups together argumentation elements that are "premises." The argumentation relationship estimation unit 16 outputs the estimated result of the argumentation relationship to the display device 3.

[0044] In the above-described second embodiment, the argumentation device 2 is configured so that the argumentation relation estimation unit 16 groups adjacent identical argumentation elements, among the multiple argumentation elements estimated by the argument element estimation unit 14, into one group, and estimates argumentation relations between multiple different groups as argumentation relations between the multiple argumentation elements based on the results of text analysis. Therefore, the argumentation device 2 can accept both the text analysis range when inferring argumentation elements and the text analysis range when inferring argumentation relations, and can efficiently infer argumentation relations between multiple argumentation elements.

[0045] Embodiment 3 In embodiment 3, an argumentation device 2 will be described in which the argumentation relation estimation unit 18 estimates argumentation relations between a plurality of argumentation elements estimated by the argumentation element estimation unit 17, based on the results of analyzing the text within the range included in the second search window and the element labels indicating the respective argumentation elements estimated by the argumentation element estimation unit 17.

[0046] Fig. 10 is a configuration diagram showing an argumentation apparatus 2 according to embodiment 3. In Fig. 10, the same reference numerals as in Fig. 1 indicate the same or corresponding parts, and detailed explanations thereof will be omitted. Fig. 11 is a hardware configuration diagram showing the hardware of the argumentation apparatus 2 according to embodiment 3. In Fig. 11, the same reference numerals as in Fig. 2 indicate the same or corresponding parts, and detailed explanations thereof will be omitted. The argumentation apparatus 2 shown in Fig. 10 comprises a text acquisition unit 11, a first search window acquisition unit 12, a second search window acquisition unit 13, an argumentation element inference unit 17, and an argumentation relation inference unit 18.

[0047] The argument element inference unit 17 is realized, for example, by an argument element inference circuit 27 shown in FIG. 11 . The argument element inference unit 17 acquires text from the text acquisition unit 11 and acquires a first search window from the first search window acquisition unit 12. Similar to the argument element inference unit 14 shown in FIG. 1 , the argument element inference unit 17 analyzes the text within the range included in the first search window while shifting the position at which the first search window is applied to the text. Similar to the argument element inference unit 14 shown in FIG. 1 , the argument element inference unit 17 infers each argument element based on the results of the text analysis. The argument element inference unit 17 outputs the inference results of each argument element to the argument relation inference unit 18 and the display device 3, and also outputs element labels indicating each argument element to the argument relation inference unit 18.

[0048] The argumentation relationship estimation unit 18 is realized, for example, by an argumentation relationship estimation circuit 28 shown in FIG. 11 . The argumentation relationship estimation unit 18 acquires text from the text acquisition unit 11 and acquires a second search window from the second search window acquisition unit 13. The argumentation relationship estimation unit 18 also acquires the estimated vector results of each argumentation element and element labels indicating each argumentation element from the argumentation element estimation unit 17. Similar to the argumentation relationship estimation unit 15 shown in FIG. 1 , the argumentation relationship estimation unit 18 applies the second search window to the text and analyzes the text within the range included in the second search window. The argumentation relationship estimation unit 18 estimates argumentation relationships between the multiple argumentation elements estimated by the argumentation element estimation unit 17 based on the text analysis results and the element labels indicating each argumentation element. The argumentation relationship estimation unit 18 outputs the estimated argumentation relationship results to the display device 3.

[0049] 10 assumes that each of the components of the argumentation apparatus 2, namely, the text acquisition unit 11, the first search window acquisition unit 12, the second search window acquisition unit 13, the argument element inference unit 17, and the argument relation inference unit 18, is realized by dedicated hardware as shown in FIG. 11. That is, it assumes that the argumentation apparatus 2 is realized by a text acquisition circuit 21, a first search window acquisition circuit 22, a second search window acquisition circuit 23, an argument element inference circuit 27, and an argument relation inference circuit 28. Each of the text acquisition circuit 21, the first search window acquisition circuit 22, the second search window acquisition circuit 23, the argument element inference circuit 27, and the argument relation inference circuit 28 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC, an FPGA, or a combination of these.

[0050] The components of the argumentation device 2 are not limited to those realized by dedicated hardware, and the argumentation device 2 may be realized by software, firmware, or a combination of software and firmware. When the argumentation device 2 is realized by software, firmware, or the like, programs for causing a computer to execute the processing procedures of the text acquisition unit 11, the first search window acquisition unit 12, the second search window acquisition unit 13, the argument element estimation unit 17, and the argumentation relation estimation unit 18 are stored in memory 31 shown in Fig. 3. Then, a processor 32 shown in Fig. 3 executes the programs stored in memory 31.

[0051] 11 shows an example in which each of the components of the argumentation apparatus 2 is realized by dedicated hardware, while Fig. 3 shows an example in which the argumentation apparatus 2 is realized by software, firmware, etc. However, this is merely an example, and some of the components in the argumentation apparatus 2 may be realized by dedicated hardware, and the remaining components may be realized by software, firmware, etc.

[0052] Next, the operation of the argumentation apparatus 2 shown in Fig. 10 will be described. However, apart from the argumentation element estimation unit 17 and the argumentation relation estimation unit 18, the argumentation apparatus 2 is the same as that shown in Fig. 1. Therefore, only the operation of the argumentation element estimation unit 17 and the argumentation relation estimation unit 18 will be described here.

[0053] The argument element estimation unit 17 acquires text from the text acquisition unit 11 and acquires a first search window from the first search window acquisition unit 12. Similar to the argument element estimation unit 14 shown in FIG. 1 , the argument element estimation unit 17 analyzes the text within the range included in the first search window while shifting the position at which the first search window is applied to the text. Similar to the argument element estimation unit 14 shown in FIG. 1 , the argument element estimation unit 17 estimates each argument element based on the results of the text analysis. The argument element estimation unit 17 outputs the estimation results of each argument element to the argument relation estimation unit 18 and the display device 3, and also outputs element labels indicating each argument element to the argument relation estimation unit 18, as shown in FIG. 12 . FIG. 12 is an explanatory diagram showing element labels indicating each argument element. The element label indicates, for example, that the argument element for the word “the” is a “premise,” and indicates, for example, that the argument element for the word “significance” is a “premise.” Furthermore, the element label is, for example, a label of "O" indicating that the argument element for the word "Hence" is neither a "premise" nor an "assertion," and is, for example, a label indicating that the argument element for the word "effective" is an "assertion."

[0054] The argumentation relationship estimation unit 18 acquires text from the text acquisition unit 11 and acquires a second search window from the second search window acquisition unit 13. The argumentation relationship estimation unit 18 also acquires the estimation results of each argument element and element labels indicating each argument element from the argument element estimation unit 17. Similar to the argumentation relationship estimation unit 15 shown in FIG. 1 , the argumentation relationship estimation unit 18 applies the second search window to the text and analyzes the text within the range included in the second search window. The argumentation relationship estimation unit 18 recognizes what kind of argument element each word is based on the element labels indicating each argument element, and then estimates the argumentation relationships between the multiple argument elements estimated by the argument element estimation unit 17 based on the analysis results of the text. The argumentation relationship estimation unit 18 outputs the estimation results of the argumentation relationships to the display device 3.

[0055] In the above-described third embodiment, the argumentation device 2 is configured so that the argumentation relation estimation unit 18 estimates argumentation relations between the multiple argumentation elements estimated by the argumentation element estimation unit 17, based on the results of analyzing the text included in the second search window and the element labels indicating the respective argumentation elements estimated by the argumentation element estimation unit 17. Therefore, the argumentation device 2 can accept both the range of text analysis when inferring argumentation elements and the range of text analysis when inferring argumentation relations, and can efficiently estimate argumentation relations between the multiple argumentation elements.

[0056] In addition, the present disclosure allows for free combination of the respective embodiments, modification of any of the components of the respective embodiments, or omission of any of the components of the respective embodiments.

[0057] The present disclosure can accept both the range of text analysis when inferring argument elements and the range of text analysis when inferring argument relations, and can be used in argumentation devices and argumentation methods.

[0058] 1 Text output device, 2 Argumentation device, 3 Display device, 11 Text acquisition unit, 12 First search window acquisition unit, 13 Second search window acquisition unit, 14, 17 Argument element estimation unit, 15, 16, 18 Argumentation relationship estimation unit, 21 Text acquisition circuit, 22 First search window acquisition circuit, 23 Second search window acquisition circuit, 24, 27 Argumentation element estimation circuit, 25, 26, 28 Argumentation relationship estimation circuit, 31 Memory, 32 Processor.

Claims

1. An argumentation device comprising: a first search window acquisition unit that acquires a first search window indicating an analytical range when inferring each argument element from text containing multiple argument elements; a second search window acquisition unit that acquires a second search window that indicates an analytical range when inferring an argument relation between the multiple argument elements from the text; an argument element inference unit that analyzes text within the range included in the first search window while shifting the position at which the first search window is applied to the text, and infers each argument element based on the analysis results of the text; and an argument relation inference unit that applies the second search window to the text, analyzes text within the range included in the second search window, and infers argument relations between the multiple argument elements inferred by the argument element inference unit based on the analysis results of the text.

2. The argumentation device according to claim 1, characterized in that the second search window acquisition unit acquires, as the second search window, a search window with a wider analysis range than the first search window.

3. The argumentation device according to claim 1 or 2, characterized in that the argumentation relationship inference unit groups adjacent identical argumentation elements among the multiple argumentation elements inferred by the argumentation element inference unit into one group, and infers argumentation relationships between multiple different groups as argumentation relationships between the multiple argumentation elements based on the analysis results of the text.

4. An argumentation device as described in any one of claims 1 to 3, characterized in that the argumentation relationship estimation unit estimates argumentation relationships between multiple argumentation elements estimated by the argumentation element estimation unit based on the analysis results of the text within the range included in the second search window and element labels indicating each argumentation element estimated by the argumentation element estimation unit.

5. A method of argumentation in which a first search window acquisition unit acquires a first search window indicating an analysis range when inferring each argument element from text containing multiple argument elements; a second search window acquisition unit acquires a second search window indicating an analysis range when inferring an argument relation between the multiple argument elements from the text; an argument element inference unit analyzes the text within the range included in the first search window while shifting the position at which the first search window is applied to the text, and infers each argument element based on the analysis results of the text; and an argument relation inference unit applies the second search window to the text, analyzes the text within the range included in the second search window, and infers the argument relation between the multiple argument elements inferred by the argument element inference unit based on the analysis results of the text.

Images