JPH10124515A - Natural language sentence generation system, method and storage medium storing program for executing the same - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a natural language sentence generation system, method and storage medium storing program for executing the same with which natural language sentences can be generated corresponding to the personality of baby, child, aged and sofisticated girl. SOLUTION: Semantic structure is inputted (S201) and when there is a concept (S202, YES), S203 is started to retrieve the concept of a baby dictionary. When there is a correspondent concept (S204, YES), S205 is started to retrieve a baby vocaburary. When there is no correspondent concept (S204, NO), S209 is started to retrieve the baby concept while using a system dictionary (S209) and when there is a correspondent concept (S210, YES), S205 is started. When there is no correspondent concept (S210, NO), a system concept is retrieved (S211). Thus, the natural language sentence can be generated by using the baby dictionary while taking priority over the system dictionary.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language sentence generation system and method for generating a sentence in a natural language from a semantic structure, and a storage medium storing a program for implementing the method.

[0002]

2. Description of the Related Art In a conventional natural language sentence generation method, the wording of a generated sentence is changed depending on the type of a sentence to be generated. For example, "paper style", "manual style"
Specify the type of sentence, such as "Polite", "Zonzai", etc.
For example, the end-of-sentence expression such as "Naru-tona" is switched.

[0003]

However, in a response system using voice or the like, the responding side (voice synthesis system) is given personality such as male (voice), female (voice), robot, etc. In many cases, the sound quality of voice can be switched. In the future,
It is considered that ages such as children and old people and personalities such as roundness are also given.

[0004] The difference between these personalities of sound quality is "attentive",
It is often the case that only the switching of the sentence end expression based on the treatment expression such as “Zonzai” cannot achieve the expression and the naturalness of the generated sentence is lost. For example, "Please buy a book"
If the sentence at the end of the sentence is written in Kansai dialect, it means "I have bought a book", but that alone is not enough.
Must be said "this way".

[0005] The present invention has been made under such circumstances, and a natural language sentence generation system and method for generating a natural language sentence tailored to the personality of an infant, a child, an old man, a young lady, and the like, and a method thereof. It is an object to provide a storage medium storing a program to be realized.

[0006]

In order to achieve the above object, the present invention provides a natural language sentence generation system according to the following (1).
(7), the natural language sentence generation method is configured as in the following (8), and the storage medium is configured as in the following (9).

(1) Input means for inputting a semantic structure, first storage means for storing a system dictionary, second storage means for storing a personality dictionary, and a semantic structure input to the input means Sentence generation means for generating a sentence by using the personality dictionary with priority over the system dictionary.

(2) An input unit for inputting a semantic structure, a first storage unit for storing a system dictionary, and a second storage unit for storing a plurality of personality dictionaries assuming a plurality of sentence generating entities having different personalities. Storage means, a personality dictionary selecting means for selecting one personality dictionary from the plurality of personality dictionaries, and the personality dictionary selected by the personality dictionary selecting means based on the semantic structure input to the input means. A natural language sentence generation system comprising: a sentence generation means for generating a sentence by using the dictionary preferentially.

(3) The natural language sentence generation system according to (1) or (2), further comprising a registration means for inputting a concept, a vocabulary, and a generation rule and registering the input into a system dictionary and / or a personality dictionary.

(4) There is provided control means for inputting a personality dictionary assuming a sentence generating entity of a new personality, storing the personality dictionary in the second storage means, and selecting the stored personality dictionary by the personality dictionary selecting means. The natural language sentence generation system according to (2).

(5) The natural language sentence generation system according to any one of (1) to (4), wherein the concept dictionary included in the personality dictionary has a concept hierarchy such as upper and lower levels.

(6) The concept dictionary included in the personality dictionary does not have a concept hierarchy such as upper and lower levels.
The natural language sentence generation system according to any one of (1) to (4).

(7) The natural language sentence generation system according to any one of (1) to (4), wherein the personality dictionary includes a format for describing the function words and the function words.

(8) A method for generating a natural language sentence for a natural language sentence generation system, comprising the steps of:
A natural language sentence generation method that generates sentences using the personality dictionary with priority over the system dictionary.

(9) A storage medium storing a program for realizing the method according to (8) via a CPU.

[0016]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below in detail with reference to embodiments of a "natural language sentence generation system".
In this embodiment, a stand-alone type in which components are concentrated in one place is assumed. However, the present invention is not limited to this type. For example, a type in which components are distributed on a wired or wireless network. Can be implemented. The "natural language sentence generation system" of the claims includes the stand-alone type and the type distributed on the network.

[0017]

FIG. 1 is a block diagram showing a system configuration of a "natural language sentence generation system" according to an embodiment. In the present embodiment, an infant is taken as an example of the personality of the sentence generation subject.

In FIG. 1, reference numeral 101 denotes a semantic structure input unit;
102 is a sentence generation unit, 103 is a sentence output unit, 104 is an association search unit, 105 is a personality management unit, 106 is a personality registration unit,
107 is a system concept dictionary, 108 is a system vocabulary dictionary, 109 is a system generation rule, 110 is an infant concept dictionary, 111 is an infant vocabulary dictionary, and 112 is an infant generation rule. The system concept dictionary, the system vocabulary dictionary, and the system generation rules are collectively referred to as a system dictionary, and the concept dictionary, the vocabulary dictionary, and the generation rules that assume a sentence generation subject of a specific personality such as an infant are collectively referred to as an individuality dictionary. The system dictionary is a dictionary that assumes a general sentence generator.

FIG. 2 is a flowchart showing the procedure of the processing for generating an infant sentence according to the present embodiment.

FIG. 3 is a flowchart showing a procedure for registering the personality dictionary of the sentence generator according to the present embodiment.

FIG. 4 is an explanatory diagram of the system dictionary and the infant personality dictionary.

FIG. 5 is an explanatory diagram of sentence generation when the system and the infant sentence generation entity are used.

FIG. 2 shows the processing when the sentence generation subject is an infant. The processing of these flows is performed by a CPU (not shown).

In S201, the input of the semantic structure is accepted.
An input such as the semantic structure of the example of generating a sentence in FIG. 5 is accepted.
In S202, in order to perform processing for each concept in the semantic structure, it is determined whether an unprocessed concept remains. When there are no more unprocessed concepts, S
Proceed to 214. The processing for each concept is performed in S203.
Do from.

In S203, the concept is searched in the infant concept dictionary. For example, the concept "collie" in FIG. In S204, it is determined whether or not the concept exists. If there is, go to S205, otherwise go to S209.

For example, since the concept "Cory" in FIG. 5 does not exist in the infant concept dictionary, the process proceeds to S209. If the child concept is not found, the child concept is searched using the system concept in S209. For example, the concept "collie" in FIG. 5 is searched in the system concept dictionary, and the superordinate concept "dog" is obtained.
Since "dog" is in the infant dictionary, the concept "dog" of this infant dictionary is extracted. Next, in S210, it is determined whether an appropriate concept has been extracted. If it can be taken out, the process proceeds to S205. If it cannot be taken out, the process proceeds to S211.

When the concept of the infant is extracted, a search for the infant vocabulary is performed in S205. For example, the child vocabulary "Wanwan" is extracted from the concept "dog" of the infant dictionary in FIG. S2
At 06, it is determined whether there is a corresponding vocabulary, and if there is, the process returns to S202 to repeat the process for the next concept.

If there is no corresponding vocabulary in S206,
Proceeding to 207, a corresponding vocabulary is searched using the system concept dictionary and the system vocabulary dictionary. For example, the infant concept "run" in FIG. 5 has no corresponding infant vocabulary. Therefore, the system vocabulary “run” is extracted from the system concept “run”. In S208, it is determined whether or not the corresponding vocabulary can be extracted. If the vocabulary can be extracted, the process returns to S202 to repeat the processing for the next concept. If not, S2
Proceed to 13.

If the concept of the infant cannot be extracted in S210, the system concept is searched in S211. Next S21
In step S207, it is determined whether or not the vocabulary has been extracted. If the vocabulary has been extracted, the process proceeds to step S207 to search for a system vocabulary corresponding to the concept. If it cannot be taken out, the process proceeds to S213.

In S213, it is determined that no suitable vocabulary has been found, and a meaningless word is generated. As the meaningless word, a character string such as "Mummumu", "Eto", "Ah", "Nuu" may be used, or a silent character string may be generated.

When the process is completed for all the concepts in S202, the process proceeds to S214. In S214, a search for an infant generation rule is performed. This extracts from the semantic structure a production rule that matches the structure. In FIG. 5, the generation rule “S, V” is extracted. In S215, it is determined whether or not there is a rule. If there is, the process proceeds to S216,
If not, the process proceeds to S218. In S216, the vocabulary is applied to the generation rule. In FIG. 5, “Wanwan” is applied to the position of the subject “S” in the position of the subject “S”, and Hiragana notation of “run” is applied to the position of the verb “V”, and “Wanwan,
Get the character string At the time of this fitting, you may apply as it is "run".

If there is no rule in S215, S21
In step 8, a search for a system generation rule is performed. In S219, the presence or absence of a rule is determined. If there is, the process proceeds to S216, where vocabulary is applied to the generation rule. If not, proceed to S220.
In S220, the searched vocabulary is randomly arranged to generate a character string.

When a character string as a sentence is generated in S216 or S220, the generated sentence is output in S217 and the processing is terminated.

As described above, in this embodiment, when generating a sentence, the personality dictionary is used in preference to the system dictionary.
Generate the required statement.

FIG. 3 shows a procedure for registering a new vocabulary or the like in the infant personality dictionary.

In step S301, an input of a new vocabulary or the like is accepted, and it is determined whether or not the vocabulary is registered. For vocabulary, S
Proceed to 302, otherwise proceed to S309.

In S302, the vocabulary input is accepted, and S
At 303, the vocabulary is registered. In step S304, a key input for designating what the vocabulary corresponds to is accepted. This key may be a vocabulary or concept of synonyms, synonyms and broader terms.

In step S305, based on the key input, the corresponding concept is listed up from the system concept dictionary.
Is displayed and the selection is made. In S307, the selected concept is input, and registered in the concept dictionary in S308. If there is a correspondence with the vocabulary dictionary, it is also registered, and the process ends.

If it is not registered in the vocabulary dictionary in S301, it is determined in S309 whether it is registered in the concept dictionary. In the case of registration in the concept dictionary, the process proceeds to S304.

If it is not registered in the concept dictionary, S3
At 10, it is determined whether or not the generation rule is registered. If the generation rule has been registered, the generation rule is registered in step S311 and the processing ends.

If it is not a registration of a generation rule in S310, the process ends without doing anything.

As described above, according to the embodiment, a dictionary of system concepts, vocabulary, and generation rules (system dictionary)
In addition, by having concepts, vocabulary, and generation rules as personality dictionaries, it is possible to generate natural language sentences that match personalities of infants, children, the elderly, young women, etc., and to use personality in response systems using voice. And a natural language sentence that matches naturally with the sound quality for giving

Further, a new concept, vocabulary, and generation rule can be edited and registered, and the capability of the system can be expanded.

(Modification of Embodiment) a. In the above embodiment, the individuality of the sentence generating entity is “infant”. However, regarding the individuality of various sentence generating entities such as “Young lady”, “Edo kid”, “Kansaijin”, “Shogashiji”,
A personality dictionary may be created and switched according to the user's preference.

B. In the above embodiment, a new vocabulary is registered for a personality dictionary that has already been created, but a new personality dictionary is created using this as another personality,
The user may customize it.

C. Alternatively, a new personality dictionary may be newly created, a user may create a unique language system, and a user-specific sentence may be generated.

D. In the above embodiment, the concept dictionary of the personality dictionary has a concept hierarchy such as upper and lower levels. However, the concept dictionary may not be provided, and the hierarchy may be referred to only the concept hierarchy of the system dictionary.

E. In the above-described embodiment, the generation of the end-of-sentence expression is not described in the sentence generation rules.
About "mood expression" such as will and desire to talk,
In order to cope with the problem, a format for describing the function word and a function for registering the function word as a personality dictionary may be provided.

[0049]

As described above, according to the present invention,
A natural language sentence can be generated according to the personality of an infant, a child, an old man, a young lady, and the like.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a system configuration according to an embodiment;

FIG. 2 is a flowchart showing a processing procedure of infant sentence generation;

FIG. 3 is a flowchart showing a processing procedure for registering an individual dictionary;

FIG. 4 is an explanatory diagram of a system dictionary and an infant personality dictionary.

FIG. 5 is an explanatory diagram of sentence generation.

[Explanation of symbols]

 101 Semantic structure input unit 102 Sentence generation unit 107-109 System dictionary 110-112 Infant dictionary

Claims (9)

[Claims] An input unit for inputting a semantic structure; a first storage unit for storing a system dictionary; a second storage unit for storing a personality dictionary; and a semantic structure input to the input unit. Sentence generation means for generating a sentence by using the personality dictionary with priority over the system dictionary. 2. An input means for inputting a semantic structure, a first storage means storing a system dictionary, and a second storage means storing a plurality of personality dictionaries assuming a plurality of sentence generators having different personalities. Storage means, a personality dictionary selecting means for selecting one personality dictionary from the plurality of personality dictionaries, and a personality dictionary selected by the personality dictionary selecting means based on the semantic structure input to the input means. A natural language sentence generation system, comprising: a sentence generation means for generating a sentence by using it more preferentially. 3. The natural language sentence generation system according to claim 1, further comprising a registration unit for inputting a concept, a vocabulary, and a generation rule and registering the input into a system dictionary and / or a personality dictionary. 4. A personality dictionary assuming a sentence generating entity of a new personality is input and stored in a second storage means, and a control means is provided for enabling the stored personality dictionary to be selected by the personality dictionary selecting means. 3. The natural language sentence generation system according to claim 2, wherein: 5. The natural language sentence generation system according to claim 1, wherein the concept dictionary included in the personality dictionary has a concept hierarchy such as upper and lower levels. 6. The natural language sentence generation system according to claim 1, wherein the concept dictionary included in the personality dictionary does not have a concept hierarchy such as upper and lower levels. 7. The natural language sentence generation system according to claim 1, wherein the personality dictionary includes a format describing a function word and a function word. 8. A natural language sentence generation method for a natural language sentence generation system, wherein a sentence is generated based on a semantic structure by preferentially using an individuality dictionary over a system dictionary. Statement generation method. 9. A storage medium storing a program for realizing the method according to claim 8 via a CPU.