JP7705322B2 - Text analysis device and text analysis method - Google Patents

Description

The present invention relates to a text analysis device and a text analysis method.

During the development process of various systems that process software programs, verification is performed to ensure the quality of the system as to whether the system meets the required specifications. One example of a technology that supports such verification work is a technology that creates a machine-readable description of the required specifications by combining selectable natural language text segments, and automatically generates tests based on the description to verify whether the system meets the required specifications.

JP 2008-171391 A

In this way, in order to determine how to verify whether a system satisfies the requirement specifications, it is desirable that the requirement specifications can be interpreted uniquely and without ambiguity from the text in which the requirement specifications are described. However, the text describing the requirement specifications contained in the specifications, etc. created in the upstream process of system development is more abstract than the software design documents, etc. created in the downstream process, and often uses ambiguous expressions. In this regard, the above-mentioned example of the prior art proposes generating the description of the requirement specifications themselves in a machine-readable format. However, it is more desirable to be able to properly interpret the requirement specifications based on the normal description of the requirement specifications created in the normal system development process, rather than on the description in such a special format. In addition, being able to properly interpret the requirement specifications based on such text is effective not only in the system verification work, but also in the stage of analyzing the requirement specifications of the system in the upstream process.

Therefore, in one aspect of the present invention, it is an object of the present invention to make it possible to appropriately interpret the requirements specifications of a system based on a requirements document that describes the requirements specifications of the system.

In one aspect of the present invention, input data including a requirement sentence describing the requirement specifications for the system to be developed is read, a sentence structure analysis of the requirement sentence is performed, and the semantic role of a target phrase included in the requirement sentence in the requirement specifications is identified. In addition, a dictionary database in which dictionary information in which definitions corresponding to roles are associated with each role for phrases related to the system is searched using the target phrase as a key, and the dictionary information corresponding to the target phrase is extracted from the dictionary database. Furthermore, the definition corresponding to the role identified for the target phrase is extracted from the dictionary information. Then, output data is generated and output in a mode in which the definition corresponding to the role of the target phrase can be displayed together with the requirement sentence.

According to one aspect of the present invention, it becomes possible to appropriately interpret the requirements specifications of a system based on a requirements document that describes the system requirements.

FIG. 1 is a diagram illustrating an example of a writing analysis device according to a first embodiment. FIG. 11 is a diagram showing an example of request text input data. FIG. 4 is a diagram illustrating an example of a dictionary database. FIG. 10 is a diagram illustrating an example of a synonym database. FIG. 11 is a diagram showing an example of an output mode of request text output data. 11 is a flowchart illustrating an example of a text analysis process according to the first embodiment. 13 is a flowchart showing the details of a dictionary search process. FIG. 11 is an explanatory diagram showing a specific example of data in the text analysis process. FIG. 11 is an explanatory diagram showing a specific example of data in fuzzy search and sentence replacement processing. FIG. 13 is a diagram illustrating an example of a writing analysis device according to a second embodiment. 13 is a flowchart illustrating an example of a text analysis process according to the second embodiment. FIG. 2 is a diagram illustrating a hardware configuration of a computer that constitutes the writing analysis device.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.
<Technical area to which the present embodiment is applied>
First, in order to facilitate understanding of the present embodiment, an example of a technical field (background) to which the present embodiment is applied will be described. The present embodiment relates to development support for various systems in which a program is executed on a computer (including a microcomputer). As an example, in the development of an in-vehicle system, development is performed based on a development process that applies a model called a V-shaped development model based on a process model such as A-SPICE (Automotive Software Process Improvement and Capability dEtermination). In the V-shaped development model, development is performed in the order from upstream to downstream, that is, (1) requirement (requirement) analysis for the system to be developed, (2) system architecture design, (3) software requirement analysis, (4) software architecture design, and (5) software detailed design. Then, for each of the processes (1) to (5), verification work is performed in the order from downstream to upstream, such as (6) software unit verification, (7) software integration test, (8) software qualification test, (9) system integration test, and (10) system qualification test, thereby further improving the quality of the system.

In the development process described above, for example, in the upstream process (1), a requirements specification or the like is referenced, which describes the requirements (requirements) for the system. The requirements documents that describe the requirements and are included in such requirements specifications are often more abstract and vague than detailed design documents created in the software detailed design process (5). Therefore, for example, in process (1), it can be difficult to perform requirements analysis correctly, and a large portion of it depends on the skills of the analyst, etc. In addition, when creating tests to confirm the suitability of the system in process (10) to verify whether process (1) was performed correctly, it can be difficult to determine what kind of tests can be used to verify whether the system meets the requirements.

In this embodiment, by analyzing the requirements text created in the upstream process of the development process of such a system, the ambiguity of the requirements text is reduced, and the requirements specifications can be clearly interpreted from the requirements text. This improves the accuracy of the analysis and verification work of the requirements specifications, thereby realizing support for system development.

First Embodiment
[Configuration of the text analysis device]
1 is a diagram showing an example of a writing analysis device in the first embodiment. The writing analysis device 1 is a computer, and includes a control unit 10 that executes a program, and a storage unit 20 in which data used to execute the program is stored. The control unit 10 includes a text structure analysis unit 11, a dictionary search unit 12, a definition extraction unit 13, and a text output unit 14, the functions of which are realized when the program is loaded from the storage unit 20 and executed. The storage unit 20 also includes at least request text input data 21, a dictionary database 22, a synonymous phrase database 23, and request text output data 24, in addition to data of the program body executed by the control unit 10 (not shown).

The sentence structure analysis unit 11 reads out the request sentence input data 21 from the storage unit 20, and performs a sentence structure analysis process on the request sentence included in the request sentence input data 21. The sentence structure analysis unit 11 identifies the semantic role of a phrase (which may include one word or a phrase combining multiple words) included in the request sentence in the request specification described in the request sentence through the sentence structure analysis process. The phrase for which such a role is identified is, for example, a noun or a noun phrase, but is not limited to this. In this embodiment, such roles are classified into three roles: "condition", "observation target", and "expected action". In the following description, the phrase for which the role is identified in the request sentence is called a target phrase.
The dictionary search unit 12 accesses the dictionary database 22 and extracts dictionary information corresponding to each of the target words and phrases obtained by the sentence structure analysis unit 11 .
The definition extraction unit 13 extracts, from the dictionary information extracted by the dictionary search unit 12, a definition according to the role of each target word or phrase analyzed by the sentence structure analysis unit 11.
The sentence output unit 14 generates and outputs request sentence output data 24, which is data in a format capable of displaying a definition according to the role of the target word extracted by the definition extraction unit 13 together with the request sentence.

Requirement text input data 21 is data in which requirement text included in requirement specifications for system development is stored in a format that can be read by a computer (e.g., CSV (Comma Separated Values)). Figure 2 shows an example of requirement text input data 21 in table format. Requirement text input data 21 includes, for example, a requirement ID for identifying each requirement text, the requirement text, and the chapter to which the requirement text belongs in the requirement specifications.

The dictionary database 22 is a database in which definitions (explanations of specific contents related to the term) are associated with terms related to the system to be developed. FIG. 3 shows an example of the data structure of the dictionary database 22. As shown in FIG. 3, in the dictionary database 22, a definition is associated with each term as dictionary information for each semantic role in the requirement specification as described above. In this embodiment, as described above, such roles are classified into three roles, namely, "condition", "observation target" and "expected action". Note that the classification of roles is not limited to this example, and can be set arbitrarily depending on the contents and characteristics of the requirement sentence in the system. In addition, it is not necessary to associate all of these three role definitions with all terms, and only one of the three definitions may be associated with some terms. In addition, as shown in FIG. 3, in the dictionary database 22, the likelihood that the term corresponds to the role is associated with each role. The dictionary information of the dictionary database 22 can be registered arbitrarily in all of the system development processes described above. In particular, by registering dictionary information in the dictionary database 22 in the downstream process of the system development process, it is possible to provide more detailed definitions for words and phrases. Such a dictionary database 22 can be created in a format such as JSON (JavaScript Object Notation).

The synonym database 23 is a database in which synonyms are associated with words related to the system to be developed. As shown in FIG. 4, in the synonym database 23, one or more synonyms and the likelihood that the word can be read as the synonym are associated with each word. Data in the synonym database 23 can be registered in advance before the start of system development, or can be registered at any time during any of the system development processes described above.

The request sentence output data 24 is data in a form that allows a definition corresponding to the role of the target phrase to be displayed together with the request sentence. As an example, the request sentence output data 24 may be configured as HTML (Hyper Text Markup Language) data that includes the request sentence. FIG. 5 shows an example of the request sentence output data 24 generated in HTML data format and output on a web browser. By incorporating JavaScript, CSS (Cascading Style Sheets), etc. into the HTML data, the request sentence output data 24 can be implemented to display the request sentence on a web browser and dynamically display the definition corresponding to the role of each phrase as a comment.

[Text analysis processing]
FIG. 6 is a flowchart showing the text analysis process executed in the text analysis device 1 in the first embodiment.
In step 1001 (denoted as S1001 in the figure, and the same applies below), the sentence structure analysis unit 11 reads out the request sentence input data 21 from the storage unit 20. Then, the sentence structure analysis unit 11 performs a sentence structure analysis process on the request sentence included in the request sentence input data 21.

The text structure analysis process includes, for example, the following three processes.
(1) Morphological Analysis The sentence structure analysis unit 11 performs morphological analysis by dividing the request sentence into morphemes, which are the smallest units that have meaning in a language, and determining the part of speech and inflection of each morpheme. Existing technology can be used for the morphological analysis. As specific examples, analysis tools such as MeCab and JUMAN are provided, and an existing corpus can be used as a dictionary to be used in the analysis process.

(2) Syntax Analysis Next, the sentence structure analysis unit 11 performs syntactic analysis (dependency analysis) to analyze the relationship between each word in the request sentence based on the result of the morphological analysis. Existing technology can be used for the syntactic analysis, and for example, a tool based on a machine learning algorithm can be used. As such a tool, for example, there is SyntaxNet, which is a recurrent neural network (RNN) model built using TensorFlow and is specialized for natural language understanding (NLU).

(3) Role analysis of words (semantic analysis)
The sentence structure analysis unit 11 identifies the role (whether it is "condition", "observation target", or "expected action") of the target phrase contained in the request sentence based on the results of the morphological analysis and syntactic analysis. The role classification at this time is to be consistent with the role classification that can be set in the dictionary information of the dictionary database 22. Such role analysis processing can be realized, for example, by deep learning based on a machine learning algorithm. As a specific example of an existing technology that realizes such deep learning, a neural network framework written in Python can be used.

As an example of processing by deep learning, it is possible to prepare teacher data in which the roles of words used in a request sentence are classified into "condition", "object of observation", and "expected action", and to perform a learning process using the teacher data to generate a learning model. In addition, the input layer of deep learning can include, for example, the target phrase itself, the grammatical structure of the request sentence, the chapter to which the request sentence belongs in the request specification, and keywords included in the request sentence. Information indicating the grammatical structure of the request sentence is, for example, information indicating what words are before and after the target phrase. As a specific example, the information can be used for analysis such as, if there is a particle such as "in" immediately after the target phrase, the role is likely to be "condition". In addition, the output layer can be information indicating the role of the above-mentioned phrase, that is, one of "condition", "object of observation", and "expected action". Although not shown in FIG. 1, the teacher data and learning model data used in the deep learning are also stored in the storage unit 20 of the text analysis device 1.

In step 1002, the dictionary search unit 12 accesses the dictionary database 22 and performs a process of searching and extracting dictionary information corresponding to each of the target words and phrases included in the request sentence. The details of the dictionary search process will be described later. The dictionary information extracted from the dictionary database 22 by the dictionary search unit 12 includes information indicating the roles of the target words and phrases in the request sentence, i.e., definitions corresponding to all or part of "conditions", "object of observation", and "expected action", as shown in FIG. 3, and the likelihood of each.

In step 1003, the definition extraction unit 13 extracts a definition from the dictionary information for each of the target phrases included in the request sentence, for which dictionary information has been extracted from the dictionary database 22, that corresponds to the role of the target phrase identified by the sentence structure analysis unit 11.

In step 1004, the sentence output unit 14 generates and outputs request sentence output data 24 capable of showing the definition of a target phrase, from among the target phrases included in the request sentence, for which a definition according to the role has been extracted, together with the request sentence. As described above, the sentence output unit 14 can, as an example, display the request sentence on a web browser as shown in FIG. 5, and dynamically display the definition according to the role of each target phrase as a comment. This allows the user to refer to the definition according to the role of each target phrase together with the request sentence. At this time, the sentence output unit 14 can also output the likelihood set in the dictionary database 22 for the definition according to the role of the phrase, also as shown in FIG. 5.

FIG. 7 is a flow chart showing the detailed contents of the dictionary search process in step 1002 of the text analysis process shown in FIG.
In step 1011, the dictionary search unit 12 accesses the dictionary database 22 and performs a search using the target word or phrase contained in the request sentence as a key.
In step 1012, the dictionary search unit 12 determines whether the search results in a hit, that is, whether dictionary information corresponding to the target word or phrase exists in the dictionary database 22. If the dictionary information exists, the process proceeds to step 1013 (Yes), and if not, the process proceeds to step 1014 (No).
In step 1013, the dictionary search unit 12 extracts dictionary information corresponding to the target word or phrase from the dictionary database 22.

In step 1014, the dictionary search unit 12 performs a fuzzy search of the dictionary database 22 for the target phrase. Specifically, the dictionary search unit 12 searches the synonymous phrase database 23 using the target phrase as a key. If the search returns a hit, that is, if the synonymous phrase database 23 contains data of a synonymous phrase corresponding to the target phrase, the dictionary search unit 12 extracts a synonymous phrase and the likelihood of replacing the target phrase with the synonymous phrase from the synonymous phrase data. In addition, in the fuzzy search, if a corresponding phrase is registered in the synonymous phrase database 23 for a part of the phrase, the part of the phrase may be replaced with a synonymous phrase. In addition, if the likelihood value is equal to or less than a predetermined threshold value, the synonymous phrase may not be extracted. The dictionary search unit 12 then performs a re-search of the dictionary database 22 using the extracted synonymous phrase as a key.
In step 1015, the dictionary search unit 12 determines whether or not dictionary information corresponding to the target phrase exists in the dictionary database as a result of the search. If the dictionary information exists, the process proceeds to step 1013 (Yes), and if not, the process proceeds to step 1016 (No).

In step 1016, the dictionary search unit 12 performs a sentence replacement process on the request sentence. Specifically, the dictionary search unit 12 uses the existing syntax analysis technology described above to estimate the dependency relationship between words and changes the order of words so that the meaning of the sentence does not change. For example, assume that the granularity of a certain word is larger than that of the word in the dictionary information contained in the dictionary database 22 (that is, the word is a higher concept than the word in the dictionary information contained in the dictionary database 22). In this case, a modifier word that is in a dependency relationship and modifies the word may exist in another position in the request sentence that is not consecutive to the word. In this case, the modifier word is replaced in a position consecutive to the word to form a single combined word, so that the word has a more specific meaning. At this time, the particles before and after the word whose position has been changed may be changed as necessary. In this way, when multiple words are combined into one word, or when each of these consecutive words is converted into a synonym and then combined into one word, the word may match a word contained in the dictionary database 22.

After performing the sentence replacement process in step 1016, the dictionary search unit 12 again proceeds to step 1011, and searches the dictionary database 22 again after the sentence replacement process has been performed. Note that the number of times that the processes in steps 14 to 16 are performed when the search in the dictionary database 22 does not return a hit may be counted, and if the number of times exceeds a predetermined threshold, the dictionary information search process may be terminated.

[Example of data for text analysis processing]
Fig. 8 is an explanatory diagram showing a specific data example for the sentence analysis process described with reference to Fig. 6. Here, the request sentence with the request ID "ID-001" in the request sentence input data 21 shown in Fig. 2, "The vehicle speed in the evacuation driving mode should be limited to 50 km/h or less", is the target of processing.
As a result of the sentence structure analysis unit 11 performing morphological analysis and syntactic analysis of the request sentence in the sentence structure analysis process (S1001), the request sentence can be divided into, for example, "The vehicle speed in the evacuation driving mode is limited to 50 km/h or less." Among these, the part "in the evacuation driving mode" can be analyzed as a modifier (M) because of the description "in." The part "The vehicle speed is" can be analyzed as a subject (S) because of the particle "wa." The part "50 km/h or less to" can be analyzed as a complement (C) because of the particle "ni." Furthermore, the part "limit" can be analyzed as a predicate (V) because the part of speech is a verb.

Then, based on the analysis results, the role of each word is further identified. For example, the role of "evacuation driving mode" in the request sentence can be identified as "condition" because it is included in the modifier and because of the description "in". The role of "vehicle speed" in the request sentence can be identified as "object of observation" because it is included in the subject. The roles of "50 km/h or less" and "restrict" in the request sentence can be identified as "expected action" because they are included in the complement and predicate.

Next, the dictionary search unit 12 searches the dictionary information for each word. Specifically, the dictionary database 22 is accessed, and the dictionary database 22 is searched using each word as a key to extract dictionary information (S1002, S1011 to S1016). In this specific data example, dictionary information is extracted for each of "evacuation driving mode" and "vehicle speed" from the dictionary database 22 shown in FIG. 3. Each piece of dictionary information includes "conditions," "object of observation," and "expected actions," which are classifications of the roles of words in this embodiment, as well as definitions and probabilities associated with each of these roles.

Then, the definition extraction unit 13 extracts a definition and a probability according to each role from the dictionary information extracted for each of "evacuation driving mode" and "vehicle speed". In this specific data example, since the role of "evacuation driving mode" is identified as "condition", the definition "If both diagnosis ○○ and ×× are NG, flag △△ becomes evacuation driving mode" and the probability "85 percent" are extracted from the dictionary information. In addition, since the role of "vehicle speed" is identified as "observation target", the definition "RAM value: ○○ (calculation period xx ms) Received value: XXX (CAN) Raw value measurement method: △△△" and the probability "75 percent" are extracted from the dictionary information.

Furthermore, the text output unit 14 generates and outputs request text output data 24 capable of displaying the extracted definitions for "evacuation driving mode" and "vehicle speed" together with the request text. As described above, FIG. 5 shows an example of the output of the request text output data 24. In this output example, the request text "Vehicle speed in evacuation driving mode should be limited to 50 km/h or less" is displayed on the screen displayed on the computer display. In the request text, the words "evacuation driving mode" and "vehicle speed" from which definitions corresponding to their roles have been extracted are underlined. When the user places the mouse pointer over these words, the role of the word, its definition corresponding to its role, and its likelihood are dynamically displayed as a comment.

Next, the fuzzy search (S1014) and sentence replacement (S1016) in the dictionary search process described with reference to FIG. 7 among the sentence analysis processes will be described with reference to the specific data example shown in FIG. 9. Here, the request sentence "The inverter must be switched off when a shutoff request is received" is the request sentence with request ID "ID-002" in the request sentence input data 21 shown in FIG. 2, and is the target of processing. Meanwhile, as shown in FIG. 3, the phrase "power module shutdown" is registered in the dictionary database 22. Also, as shown in FIG. 4, the synonym database 23 registers synonyms and probabilities for the phrases "inverter" and "switch off".

When a fuzzy search is performed on the request sentence, synonyms can be extracted for each of the words "inverter" and "switch off" in the synonym database 23, but words corresponding to each of the synonyms are not registered in the dictionary database 22. If a sentence replacement process is performed on the request sentence, the request sentence can be changed to "The inverter switch off must be performed when a shutoff request is received" by analyzing the dependency relationship. If a fuzzy search is then performed again on the sentence after the sentence replacement, the synonyms "power module" for "inverter" and "shutdown" for "switch off" can be extracted for the word "inverter switch off". In this way, the word "power module shutdown", which is a combination of these, can be recognized in the request sentence. It is then possible to extract the definition of the word "power module shutdown" from the dictionary database 22.

In this case, the likelihood of "power module shutdown" being read as a synonymous phrase can be set to "35 percent" obtained by multiplying "50 percent" of "power module" and "70 percent" of "shutdown," which are set in the synonymous phrase database 23 shown in Fig. 6. Furthermore, when a fuzzy search is performed in this manner, the likelihood displayed together with the definition according to the role of the phrase in the output example of the request sentence output data 24 shown in Fig. 5 may be set to, for example, a value obtained by multiplying the value of the likelihood of the definition according to the role by the value of the likelihood of being read as a synonymous phrase.
Furthermore, when such a sentence replacement process or fuzzy search process is performed, the sentence output unit 14 may generate the request sentence output data 24 so as to display both the original text of the request sentence and the text converted in the sentence replacement process or fuzzy search.

[Effects of the first embodiment]
In this manner, in this embodiment, a text analysis device that analyzes a request text describing a requirement specification for a system to be developed performs a text structure analysis of the request text and identifies the role of a target phrase included in the request text. In addition, a dictionary database 22 that stores dictionary information in which definitions corresponding to roles are associated with each role for phrases related to the system is searched using the target phrase as a key, and dictionary information corresponding to the target phrase is extracted from the dictionary database 22. Then, a definition associated with the identified role for the target phrase is extracted from the dictionary information, and the request text is output in a form that allows the definition corresponding to the role of the target phrase to be displayed. This allows a user involved in system development to refer to the definition corresponding to the role of each phrase along with the request text.

Here, being able to refer to definitions according to the role of each term in this way has the following effect. That is, even for the same term, the meaning (target) indicated by the term may differ depending on, for example, the context in each request sentence. For this reason, even if it is possible to simply refer to a definition uniquely set for a term, that definition may not necessarily be appropriate for each request sentence. In this regard, in the present embodiment, definitions according to roles are associated with terms in the dictionary database 22 for each role. Then, for target terms included in a request sentence, a definition according to the role identified for the target term is extracted. This allows the user to refer to an appropriate definition according to the content of the request specification.

Therefore, in this embodiment, it is possible to reduce the ambiguity of requirement statements contained in requirement specifications, which are more abstract and often use ambiguous expressions compared to software design documents created in downstream processes of system development. This allows the user to properly interpret the requirement specifications based on the requirement statements. This improves the accuracy of the analysis and verification work of the requirement specifications, and realizes support for system development.

For example, in the case of the V-shaped development model described above, after downstream development processes, a large amount of dictionary information containing detailed definitions according to the roles of words related to the system being developed is registered in the dictionary database 22. This improves the hit rate in searches of the dictionary database 22, particularly in the verification process of the requirements analysis (system eligibility confirmation test) that is carried out afterwards. This makes it possible to extract definitions according to the roles of more target words for the requirements text, improving the accuracy of the verification work of the requirements analysis.

Here, as an example of a system to be developed to which the present embodiment is applied, a mechatronics system that senses and controls the dynamics of the controlled object, such as a control device for an automobile powertrain, will be considered. In this case, the required specifications that the control device must satisfy include the behavior of the powertrain of the controlled object, such as the engine and transmission of the automobile. For example, the driver's accelerator operation is required to achieve the expected acceleration when the throttle opening and fuel injection amount are adjusted, and a sensory evaluation of the ride quality is also required. A requirement document that describes such required specifications in a document inevitably contains ambiguity. On the other hand, in the actual development process, downstream processes such as the design and detailed design of the system architecture that is created by considering when and how to control which sensors and actuators are specifically created. However, because the system is complex, the relevance between the design contents for the individual components of the system created in these downstream processes and the original required specifications tends to be weak. Therefore, there have been cases where it is difficult to determine how to specifically verify whether the original required specifications that should be verified are satisfied. In such a situation, according to this embodiment, it becomes possible to refer to the specific definition set in the downstream process for the target term included in the requirement specification according to its role. This makes it easy to understand the relationship between the requirement specification and each component of the system, such as which function of which component in the detailed design etc. needs to be realized to satisfy the requirement specification.

In this embodiment, roles are classified into "conditions," "objects to be observed," and "expected behavior," and definitions are extracted according to these roles. This makes it possible to clarify what conditions are presupposed by the requirements specifications described in the requirements text, what the objects to be observed are, and how the objects to be observed should behave. This classification of roles is particularly effective for interpreting requirements specifications for systems that include control devices.

In addition, in this embodiment, when dictionary information associated with the target phrase does not exist in the dictionary database 22, synonyms associated with the target phrase are extracted from the synonym database 23. Then, dictionary information associated with the synonyms is extracted from the dictionary database 22, and a definition associated with the role identified for the target phrase is extracted from the dictionary information. Then, the definition of the target phrase is displayed, and the likelihood of the synonyms associated with the target phrase is also displayed. As a result, even if the target phrase is not registered in the dictionary database 22, it is possible to extract a definition associated with the role of the target phrase as long as the synonyms are registered in the dictionary database 22. Therefore, it is possible to improve the hit rate in searching the dictionary database 22 as a result, and the ambiguity of the request sentence can be further reduced.

Furthermore, in this embodiment, a process of replacing the request sentence, that is, estimating dependency relationships between words and changing the order of the words so as not to change the meaning of the sentence, is performed. As a result, as described above, for example, two words in different positions may be changed to adjacent positions, and the words may be linked and identified, resulting in a match with a single word registered in the dictionary database 22. As a result, it is possible to improve the hit rate in searches of the dictionary database 22, and further reduce the ambiguity of the request sentence.

When registering dictionary information in the dictionary database 22, further substructures may be set for the definition of each role according to its contents. As a specific example, in the development of an in-vehicle system, the calculation points for the phrase "vehicle speed" may be "sensor value, backup sensor value, internal calculation value (motor rotation speed x gear ratio x tire diameter), and CAN communication reception value." In this case, for example, each of these calculation points may be set as a substructure of the definition corresponding to the role of "observation target." Then, when the sentence output unit 14 generates the request sentence output data 24, for example, it may be possible to refer to all of the contents of these substructures when displaying the definition of the phrase. It may also be possible for the user to select one of the pieces of information from these substructures.

Second Embodiment
[Overview of the second embodiment (differences from the first embodiment)]
Next, a second embodiment will be described. In the first embodiment, there is one dictionary database, and it is assumed that dictionary information is registered mainly during the development process. In this case, in the verification stage for the upstream process of the system development process described above (for example, process (10) in the case of the V-shaped development model described above), sufficient dictionary information is registered after the downstream process. However, in the upstream process itself before entering the detailed design stage (for example, process (1)), sufficient dictionary information has not yet been registered.

For this reason, in the second embodiment, a dictionary database that has already been created in a system that has been developed in the past is stored in the memory unit of the text analysis device, and the dictionary database used in a system that has high affinity with the system to be developed is reused. As a result, in the second embodiment, a large amount of dictionary information containing detailed definitions is registered in the dictionary database from the start of system development. Therefore, even in the upstream process before entering the detailed design stage, detailed definitions according to the role of target words included in the requirement text can be referenced, thereby realizing development support.

[Configuration of the text analysis device]
10 shows the configuration of the writing analysis device 1 in the second embodiment. Unlike the first embodiment, the writing analysis device 1 in the second embodiment further includes a dictionary selection unit 15 in the control unit 10. Also, the storage unit 20 further includes past request sentence data 25. The past request sentence data 25 can be stored in the same number as the number of other systems that have been developed in the past. Also, in the second embodiment, multiple dictionary databases 22 and synonymous phrase databases 23 can be stored. The other components are similar to those in the first embodiment, so description thereof will be omitted.

The dictionary selection unit 15 identifies a system that has a high affinity with the system to be developed this time from among the previously developed systems based on the request sentence input data 21 and the request sentence past data 25 of systems previously developed, and then selects the dictionary database 22 and the synonym database 23 used in that system.
The past request text data 25 is, for example, one or more pieces of data having a structure similar to that of the request text input data 21, and is data in which request texts for one or more systems that have been developed in the past are stored for each system.

The dictionary database 22 has the same structure as in the first embodiment, but in the second embodiment, there may be multiple dictionary databases. Each dictionary database 22 is associated with past requirement text data 25 that stores requirement specifications of other systems for which the dictionary database 22 has been used in the past. Each dictionary database 22 stores data that has already been created in the development process of the other systems.

The synonymous phrase database 23 also has a structure similar to that of the first embodiment, but there may be a plurality of synonymous phrase databases in the second embodiment. Similarly to the dictionary database 22, each synonymous phrase database 23 is associated with past requirement text data 25 that stores requirement specifications of other systems for which the synonymous phrase database 23 has been used in the past. Each synonymous phrase database 23 stores data that has already been created in the development process of the other systems.
The synonymous phrase database 23 does not necessarily have to be the selection target of the dictionary selection unit 15, but one synonymous phrase database 23 may be used for all systems, and further synonymous phrases may be registered and accumulated as necessary. Alternatively, the synonymous phrase data itself in the synonymous phrase database 23 may be used in common for all systems, and only the likelihood of each synonymous phrase may be set to a different value depending on the characteristics of the system. In this case, the dictionary selection unit 15 selects likelihood data that has a high affinity with the system currently being developed.

[Text analysis processing]
FIG. 10 is a flowchart showing a text analysis process executed in the text analysis device 1 according to the second embodiment.
In step 2001, the dictionary selection unit 15 identifies a system that has a high affinity with the system to be developed this time based on the request sentence input data 21 and the request sentence past data 25 of other systems that have been developed in the past. For example, the dictionary selection unit 15 can identify a system that has a high affinity by the following method. First, the dictionary selection unit 15 reads out each of the request sentence input data 21 and the request sentence past data 25. Then, the dictionary selection unit 15 can perform a semantic analysis (latent semantic analysis) on the request sentences included in the request sentence input data 21 and the request sentences included in the request sentence past data 25. Furthermore, the dictionary selection unit 15 compares the analysis result of the request sentence included in the request sentence input data 21 with the analysis result of the request sentence included in the request sentence past data 25. The dictionary selection unit 15 performs this process on all the request sentence past data 25 stored in the storage unit 20, and identifies one of the request sentence past data 25 based on each comparison result.

More specifically, for example, as an example of the results of semantic analysis of a requirement sentence, a numerical value is obtained for the frequency of use of each word in a vector space based on the granularity of nouns and noun phrases. The dictionary selection unit 15 can calculate the logical sum of the analysis results of the requirement sentence contained in the requirement sentence input data 21 and the analysis results of the requirement sentence contained in each of the previous requirement sentence data 25, and perform a gap analysis between the two based on this logical sum. Then, by identifying the previous requirement sentence data 25 that has the smallest gap with the requirement sentence input data 21, i.e., the highest similarity, it is possible to identify a system that has a high affinity with the current target system for development.

Then, the dictionary selection unit 15 selects the dictionary database 22 associated with the identified past request sentence data 25, i.e., the dictionary database 22 used in a system that has a high affinity with the system being developed this time. Similarly, the dictionary selection unit 15 can also select a synonym database 23 used in a system that has a high affinity with the system being developed this time.

Steps S2002 to S2005 are similar to steps S1001 to S1004 of the sentence analysis process in the first embodiment, and therefore will not be described here. In the second embodiment, steps S1011 to S1016 of the dictionary search process shown in FIG. 7 are also executed in the same manner.

[Effects of the second embodiment]
According to the second embodiment, the dictionary database used in the text analysis process can be a dictionary database that has already been created in a previously developed system that is similar to the system to be developed. As a result, in the second embodiment, a dictionary database in which a large amount of dictionary information including detailed definitions is registered can be used from the start of system development. Therefore, it becomes possible to refer to detailed definitions of target words and phrases included in requirement texts according to their roles not only in the verification stage of system development but also in upstream processes before entering the detailed design stage, making it possible to realize support for system development.

When identifying a system similar to the system under development, a predetermined threshold for determining the degree of similarity may be set. If the degree of similarity is lower than the predetermined threshold (which also has the same meaning as when the result of the gap analysis is higher than the predetermined threshold), a new dictionary database 22 may be generated without reusing the existing dictionary database 22. The same applies to the synonym database 23.
Furthermore, if there is only one request sentence past data 25 stored in advance in the storage unit 20, and one dictionary database 22 or one synonym database 23 used in the past system, there is no need to perform a process to identify a similar system. Even in this case, if the degree of similarity is lower than a predetermined threshold value as described above, the dictionary database 22 or the synonym database 23 may not be reused.

<Hardware Configuration>
12 shows an example of the hardware configuration of a computer constituting the writing analysis device 1 in each of the above embodiments. This computer includes a processor 910, a memory 920, a storage 930, a portable storage medium drive device 940, an input/output device 950, and a communication interface 960.
The processor 910 includes a control unit, an arithmetic unit, an instruction decoder, etc., and the execution unit executes arithmetic and logical operations using the arithmetic unit according to the instructions of the program decoded by the instruction decoder and in response to the control signal output from the control unit. The processor 910 includes a control register that stores various information used for control, a cache that can temporarily store the contents of the memory 920 that has already been accessed, etc. The processor 910 may be configured to include multiple CPU (Central Processing Unit) cores. The above-mentioned control unit 10 is configured by the processor 910 executing a program.

The memory 920 is a storage device such as a RAM (Random Access Memory), and is a main memory in which the programs executed by the processor 910 are loaded and data used in the processing of the processor 910 is stored. The storage 930 is a storage device such as a HDD (Hard Disk Drive) or a flash memory, and stores programs and various data. The above-mentioned storage unit 20 corresponds to the storage 930, for example. The portable storage medium drive device 940 is a device that reads data and programs stored in the portable storage medium 970. The portable storage medium 970 is, for example, a magnetic disk, an optical disk, a magneto-optical disk, or a flash memory. The processor 910 executes the programs stored in the storage 930 or the portable storage medium 970 while cooperating with the memory 920 or the storage 930. The programs executed by the processor 910 and the data to be accessed may be stored in other devices that can communicate with the computer.

The input/output device 950 is an input means and an output means such as a touch panel, a keyboard, a display, etc., and accepts operation commands by user operations, etc., while outputting the results of processing by the information processing device. The communication interface 960 enables data communication with the outside. The components of the computer described above are connected by a bus 980.

＜Other＞
The above-described embodiments of the present invention are merely some of the possible implementations within the technical scope of the present invention, and do not limit the technical scope of the present invention. Furthermore, the functional configurations and physical configurations in each embodiment are not limited to the above-described aspects, and for example, each function or physical resource can be integrated and implemented, or conversely, can be further distributed and implemented, and further, some of the configurations can be added, deleted, or replaced with other configurations.

1...Text analysis device, 11...Text structure analysis unit, 12...Dictionary search unit, 13...Definition extraction unit, 14...Text output unit, 15...Dictionary selection unit, 21...Request text input data, 22...Dictionary database, 23...Synonymous phrase database, 24...Request text output data, 25...Request text past data

Claims (5)

A text analysis device that analyzes a requirement text describing a requirement specification for a system to be developed,
a sentence structure analysis unit that reads input data including the request sentence, analyzes the sentence structure of the request sentence, and identifies the semantic role of a target word or phrase included in the request sentence in the request specification;
a dictionary search unit that searches a dictionary database, in which dictionary information in which a definition according to a role is associated with each role for a word related to the system, using the target word as a key, and extracts the dictionary information corresponding to the target word from the dictionary database;
a definition extraction unit that extracts the definition according to the role identified for the target phrase from the dictionary information;
and a sentence output unit that generates and outputs output data in a format that allows the definition of the target phrase according to the role to be displayed together with the request sentence. The text analysis device according to claim 1, wherein the role is one of a condition, an observation target, and an expected action in the requirement specification. when the dictionary information corresponding to the target phrase is not present in the dictionary database, the dictionary search unit searches a synonym database in which synonyms of the phrase related to the system and likelihoods of replacing the phrase with the synonyms are associated with the phrase, using the target phrase as a key, extracts the synonyms corresponding to the target phrase from the synonym database, searches the dictionary database again using the synonyms as a key, and extracts the dictionary information corresponding to the synonyms from the dictionary database;
The definition extraction unit extracts the definition according to the role identified for the target phrase from the dictionary information corresponding to the synonymous phrase;
The text analysis device according to claim 1 , wherein the text output unit displays the definition extracted for the synonymous phrase and also displays the likelihood. a dictionary selection unit that reads out the input data including the request sentence and one or more past data each including a request sentence describing a requirement specification for one or more other systems developed in the past for each of the other systems, performs a semantic analysis to analyze the meaning of the request sentence included in the input data and the request sentence included in each of the past data, identifies one of the past data having a high similarity to the input data based on a result of comparing the analysis result of the request sentence included in the input data with the analysis result of the request sentence included in each of the past data , and selects a dictionary database created in the development of the other system and associated with the identified one of the past data,
The text analysis device according to any one of claims 1 to 3, wherein the dictionary search unit searches the dictionary database selected by the dictionary selection unit using the target phrase as a key, and extracts the dictionary information corresponding to the target phrase from the dictionary database. A text analysis method for analyzing, by a computer , a requirement text describing a requirement specification for a system to be developed, comprising the steps of:
The computer reads input data including the request sentence, performs a sentence structure analysis of the request sentence, and identifies a semantic role of a target word or phrase included in the request sentence in the request specification;
The computer searches a dictionary database, in which dictionary information in which definitions according to the roles are associated with terms related to the system, using the target term as a key, to extract the dictionary information corresponding to the target term from the dictionary database;
The computer extracts , from the dictionary information, the definition corresponding to the identified role for the target phrase;
generating and outputting output data in a mode that allows the computer to display the definition of the target phrase according to the role together with the request sentence;
A method for analyzing text , comprising :