Features Analysis of a Patent Act Based on Legal Condition–Effect Structure: Conversion of Law Texts into Logical Formulas for a Learning Support System

Abstract

With the development of information and communications technology, intellectual property education is expected to become increasingly important. However, it has been shown that intellectual property education is insufficiently conducted in university education faculties of engineering. Therefore, we previously focused on engineering students learning logic circuits to create an e-learning system that displays articles as logic circuits. Evaluation experiments demonstrated the learning effect of this system, but it could not handle all intellectual property laws because it required manually converting law texts into logic circuits in the problem creation process. To address this problem, in this paper we examine a method for converting intellectual property laws into logical formulas. Previous studies have converted law texts into logical formulas to perform consistent tests of law texts, whereas this study aims at more human-comprehensible visualizations. We divide law texts into constituents based on a legal condition–effect structure and convert them into propositional calculus. This paper presents the results of an attempt to convert 101 law texts of typical patent acts in intellectual property law into logical formulas. From the results of integrating the logical formulas into our law text learning system, we describe future policies.

1 Background

1.1 Importance of Intellectual Property Law Education

With the development of information and communications technologies, sometimes called the Fifth Industrial Revolution, many technologies using AI and IoT are being put into practical use. In the future, new technical ideas will be created in a 5G environment. Therefore, to strengthen the international competitiveness of domestic industries, it is necessary to develop human resources who can generate and protect intellectual property. As a country that lacks natural resources, this is considered particularly important in Japan, and industry expects students to acquire knowledge about intellectual property. The Japanese government has thus established the Intellectual Property Strategy Headquarters, a Cabinet department that establishes strategies for promoting utilization of intellectual property and enhancing intellectual property education [1].

1.2 Intellectual Property Education in Faculties of Engineering

In addition to government and industry, universities too recognize the importance of intellectual property education. However, the results of a survey of faculty of engineering syllabi at Japanese national universities revealed that current intellectual property education is insufficient [2]. This is likely because the curriculum is generally limited to exercises such as experiments, and intellectual property is only partially treated in ethics classes.

Therefore, the authors have been developing a learning support system for intellectual property law for engineering students. To facilitate use by anyone without restrictions on time and place, we developed our e-learning system as a web application and introduced it in actual classes.

2 Previous Research

We considered a mechanism for promoting understanding of law texts as needed for faculty of engineering students not specializing in law. Considering that such students learn logic circuits, we developed a system that can display law texts as logic circuits [3]. However, in an e-learning system without an instructor, it is difficult for learners to notice mistakes on their own. To address this problem, we focused on error-based simulation and improved the system so that when learners make mistakes, an animation that can intuitively indicate the error is displayed [4]. Figure 1 shows the system interface, which displays a logic circuit that utilizes the fact that Article 2 Paragraph 1 of the Patent Act (“‘Invention’ in this Act means the highly advanced creation of technical ideas utilizing the laws of nature”) can be represented as a propositional formula: “(1) Utilizing the laws of nature ∧ (2) Technical ideas ∧ (3) Creation ∧ (4) Advanced → (5) Invention.” The learner can thus see the requirements for “invention” by connecting (1) through (4) to the logical conjunction element and connecting that element to (5).

We introduced this system into an actual course on intellectual property law and conducted an evaluation experiment regarding its learning effects. Felder’s learning-style-based analysis [5], which defines the learning styles of science students, suggests that this system would be particularly effective for visual learners, rather than for verbal learners.

However, this system required manually converting law texts into logic circuits in the problem creation process, making it difficult to comprehensively handle all of intellectual property law. Therefore, in this study, we examined a method for converting intellectual property law into logical formulas and showing the results of integrating the converted formulas into the system.

Fig. 1.

Interface of the law text learning system using logic circuits [3].

3 Purpose

Some studies on converting law texts into logical formulas have been conducted in the field of expert systems, where law texts are converted into logical formulas based on predicate logic to mechanically perform consistency tests for the texts. In other words, the conventional conversion method does not assume learning and is not a visualization method for ease of comprehension. Moreover, no method for converting predicate logic into logic circuits has been established. Therefore, this study focuses on the legal condition–effect structure of law texts to convert text into a propositional calculus, then logic circuits. As a first step, in this paper we show the results of converting the Patent Act to logical formulas and their integration into the system.

4 Analysis Method

This study was carried out according to the following procedure.

• 1. Get law text data.

• 2. Preprocess the data.

• 3. Select target texts to be analyzed.

• 4. Conduct morphological analysis and dependency analysis.

• 5. Examine constituent identification patterns based on the legal condition–effect structure [6].

• 6. Identify and tag the constituents by pattern matching.

• 7. Create formulas by regarding clauses or phrases enclosed in tags as propositional variables.

• 8. Judge the correctness of the result.

• 9. Integrate formulas into the system.

The legal condition–effect structure is a logical structure in which the law text consists of five constituents: a subject part, a condition part, a target part, a content part, and a regulation part. Figure 2 shows an overview.

Fig. 2.

Outline of the legal condition–effect structure [6].

Law texts consist of an effect part that expresses rights and obligations, and a requirement part that expresses the conditions under which the effect part is applied. The requirement part further comprises a subject part, which is the subject of the law text, and a condition part, which is the condition of the law text. In addition, the effect part comprises a regulation part, which is a predicate, a target part, which is an object, and a content part. The law text thus becomes a logical formula, “Subject ∧ Condition → Target ∧ Content ∧ Regulation.” Law texts can consist of any combination of these five constituents, and there are cases where multiple constituents are contained or some are omitted.

4.1 Acquisition and Preprocessing of Text Data

Law text data is obtained from the administrative information portal site “e-Gov,” operated by the Japanese government. The raw data is an extensible markup language (XML)-format file, so we extract only the textual part for conversion into a comma-separated values (CSV)-format file. When there is parenthesized text for supplementary explanations, it is necessary to separately analyze the range of the parentheses and the dependency relationship between the main text and the parenthetical text. Since the establishment of such an analysis method is beyond the scope of this study, we removed parenthetical text in this study.

Another issue is that law texts, including the Patent Act, often make special use of the Japanese kana syllabary. For example, modern Japanese uses a small tsu character (っ) to express a double consonant and other small kana characters to express specific sounds. Law texts, however, always use full-size kana, such as the text “mono no hatsumei ni atte wa” (物の発明にあつては, which should be written as 物の発明にあっては in modern Japanese) in Article 2, Paragraph 3, Item 1 of the Patent Act. Because this hinders accurate morphological analysis, in preprocessing we convert such special kana usage to modern usage. In this study, pretreatment is performed according to the contents of Table 1.

Table 1. Kana usage correspondence table for preprocessing.

4.2 Selection of Target Texts and Manual Conversion of Logical Formulas

When converting the Patent Act text into propositional logical formulas, we first manually created the logical formulas and asked an expert to confirm their correctness. The Patent Act has 290 articles, excluding those deleted through amendments and supplementary provisions. We manually selected 34 articles (comprising 101 texts) that are considered particularly important and frequently used in intellectual property management skill tests, and converted them into logical formulas. Table 2 shows the selected texts and their captions.

Next, based on the result of manual conversion from law texts to logical formulas, we examined the identification pattern of the constituents. Figures 3 and 4 show examples of the law texts that were manually converted into logical formulas.

Table 2. .

Fig. 3.

Well-formed logical formula of Article 78, Paragraph 2 of the Patent Act.

Fig. 4.

Patent Act Article 168, Paragraph 6.

4.3 Analysis of Articles and Examination of Constituent Identification Patterns

The morphological analyzer uses JUMAN [7], which is suitable for analyzing proper nouns and technical terms, and KNP [8] for dependency analysis. We examined identification patterns of the constituents by comparing the manually converted logical formulas with the results of text analysis. Constituents are tagged using the identified identification patterns by comparing the manually converted logical formula with the results of text analysis. The subject part in the law text is enclosed by the tags <Subject> </Subject>, and the condition part is enclosed by the tags <Condition> </Condition>. The target part, content part, and regulation part are collectively enclosed by the tags <Effect> </Effect> as the effect part to prevent the logic circuit from becoming too complicated.

5 Results and Discussion

5.1 Result of Constituent Identification

Using the results of morphological analysis and dependency analysis for 34 articles, we extracted morphemes with high frequency of appearance and their combinations as constituent identification patterns. Table 3 shows the identification patterns and the number of appearances.

Table 3. Identification patterns and number of appearances.

Table 4 shows the results of identifying the text constituents by pattern matching based on Table 3.

Table 4. Results of identifying the constituents.

Successful Examples of Constituent Identification.

As a result of the constituent identification of 101 documents in 34 Articles, about half of the constituents could be identified as intended (see Table 4). Figures 5 and 6 show examples in which constituent identification was performed as intended.

Fig. 5.

Constituent identification results for Article 168, Paragraph 6 of the Patent Act.

Fig. 6.

Constituent identification results for Article 39, Paragraph 7 of the Patent Act.

Failure Examples of Constituent Identification.

Figures 7 and 8 show examples in which the constituents could not be identified as intended.

Fig. 7.

Constituent identification results for Article 77, Paragraph 3 of the Patent Act.

The law text in Fig. 7 is based on a context meaning “exclusive licenses can be transferred only if the conditions are met.” Therefore, an “exclusive license” is considered to be a constituent that originally corresponds to Content, not Subject. In this way, when the noun corresponding to the content part was used as the subject for a passive verb, we found that an erroneous judgment was made in our identification pattern. In addition, we considered that the combination of morphemes “ni kagiri” (に限り, “only when”) can be used for identification of the condition part, but this was undefined as an identification pattern because the number of appearances was relatively small.

Fig. 8.

Constituent identification results for Patent Act Article 2, Paragraph 3.

In the case of the law text in Fig. 8, the context is that “the following acts are the implementation of the invention,” so the subject part should be “tsugi ni kakageru koi” (次に掲げる行為, “the following acts”), and the effect part should be “hatsumei ni tsuite jisshi” (発明について実施, “implementation of the invention”). However, this interpretation was difficult to identify with our simple rules. Therefore, it is considered that a separate feature for specifying the text of a description format similar to Article 2, Paragraph 3 of the Patent Act is required.

5.2 Integration with Learning Support System

Figures 9 and 10 show the results of integrating the logical formulas into the system. Figure 9 shows the example of integrating Article 168, Paragraph 6 of the Patent Act, in which the constituents were identified as intended, into the system. Figure 10 shows the example of integrating Article 77, Paragraph 3 of the Patent Act, which failed to identify constituents, into the system.

Fig. 9.

Example correct logic circuit (Patent Act Article 168, Paragraph 6).

Figure 9 shows that the proposed method can integrate a law text with a simple logical structure into the system. When the effect part is very long, however, it is difficult to understand the content. Therefore, it is considered necessary to devise measures such as dividing the target and content parts according to the numbers of characters and phrases in the effect part.

Fig. 10.

Example incorrect logic circuit (Patent Act Article 77, Paragraph 3).

Figure 10 shows that “sen’yo jisshi ken” (専用実施権, “exclusive license”) is part of the requirement part, because it was mistakenly identified as the subject part. In addition, “tokkyo ken ja no syodaku o eta baai oyobi sozoku sonota no ippan keisyo no baai” (特許権者の承諾を得た場合及び相続その他の一般承継の場合, “where the consent of the patentee is obtained or where the transfer occurs as a result of general succession including inheritance”) is included in the effect part. From the above, this logic circuit is incorrect. Figure 11 shows a logic circuit when conversion to a correct logical formula can be performed.

Fig. 11.

Correct logic circuit (Patent Act Article 77, Paragraph 3).

It is necessary to modify the algorithm in the future so that the logic circuit shown in Fig. 11 can be created. Other future tasks include increasing the number of rules for identifying constituents and improving accuracy, such as adding part-of-speech and dependency information to existing rules. In addition, when conjunctions such as “mata wa” (又は, “or”), “katsu” (かつ, “and”), and “oyobi” (及び, “and/or”) are included, as in Fig. 7, it is necessary to divide the parallel relationship so that it is easy to understand.

6 Conclusion

We investigated a method for automatically converting law texts into propositional calculus to support a law text learning system using logic circuits. As a result of analysis based on legal condition–effect structures, it was possible to tag law texts having relatively simple structures, as intended.

In the future, to improve the identification accuracy of constituents, we will consider identification patterns that utilize morpheme category information, etc., and perform machine learning using manually converted texts as learning data to automatically formulate logical formulas. We will also divide the condition part in consideration of parallel relationships such as “oyobi” (及び, “and/or”) and “katsu” (かつ, “and”).