JP2021179673A - Sentence generation device, sentence generation method and sentence generation program - Google Patents

Abstract

To convert a sentence into an appropriate form.SOLUTION: A sentence generation device having a processor that executes a program and a storage device that stores the program executes: acquisition processing of acquiring one or more summary elements that are word strings constituting an input sentence; and generation processing of generating an output sentence by referring to a sentence generation template constituting a sentence template with one or more designation elements designating at least a part of the word strings and by performing the designation with the designation elements in the sentence generation template capable of designating at least a part of the word strings of the summary elements acquired by the acquisition processing.SELECTED DRAWING: Figure 1

Description

The present invention relates to a sentence generator, a sentence generation method, and a sentence generation program for generating a sentence.

There is a need to transform a sentence into an appropriate form, such as summarizing a sentence or generating a question from a sentence. As a technique for summarizing sentences for each viewpoint, for example, there is a text summarizing device of Patent Document 1 below.

In this text summarizing device, the subtree holding ZDD creation unit constructs the ZDD. The viewpoint information extraction unit calculates the degree of relevance to the viewpoint for each word contained in the document for each of the plurality of viewpoints, and based on the degree of relevance calculated for each of the viewpoints for each word, a plurality of viewpoints in the document. Calculate the ratio of viewpoints. The part represented by the ZDD constructed by the viewpoint subtree holding ZDD generation unit based on the upper limit of the length according to the ratio of the viewpoints, the relevance of the viewpoints of the words corresponding to each of the nodes, and the length of the words. A ZDD representing a subset of a subtree of a set of trees is generated as a viewpoint ZDD corresponding to the viewpoint. The optimal subtree calculation unit searches for the optimal path that satisfies the upper limit of the length of the summary among the paths of the common ZDD that represent the subset common to the subset of the subset represented by the viewpoint ZDD, and the searched path represents the path. Output the subtree. The summary output section outputs a summary of the input text based on the subtree.

Further, as a technique for extracting a summary word from a sentence, for example, there is an information processing apparatus of Patent Document 2 below. This information processing device has an acquisition unit that acquires a document consisting of one or more sentences, a basic score assigning unit that weights each word contained in the sentence according to its case element, and a predicate in the sentence. It includes a score update unit that updates the weight of each word included in the sentence, a summary word selection unit that selects the summary word of the document based on the weight of each word, and an output unit that outputs the selected summary word. ..

Japanese Unexamined Patent Publication No. 2018-81463 Japanese Unexamined Patent Publication No. 2019-20445

When a sentence is summarized for each viewpoint as in the text summarizing device of Patent Document 1, it is limited by the length of the sentence and the number of words in the sentence, but in the actual sentence, the length of the noun varies. It is not appropriate because the length of the sentence does not match the amount of information contained in the sentence. Further, since the text summarizing device of Patent Document 1 determines the length of each subtree by proportionally dividing it by the ratio of viewpoints based on the upper limit of the length of the summary, the amount of information for each viewpoint is controlled by grammatical conditions. Can not do it. In addition, there are practically no numerical restrictions on the length of sentences in the summarization of minutes and the generation of dialogue materials. Further, since the information processing apparatus of Patent Document 2 generates a summary word expressing the subject of the sentence in one word, the sentence cannot be generated or output. This applies not only to sentence abbreviations such as summaries, but also to the generation of question sentences and answer sentences, and the generation of headlines.

An object of the present invention is to convert a sentence into an appropriate form.

The sentence generator which is one aspect of the invention disclosed in the present application is a sentence generator including a processor for executing a program and a storage device for storing the program, and the processor constitutes an input sentence. Acquired by the acquisition process with reference to the acquisition process for acquiring one or more summary elements that are word strings and the sentence generation template that constitutes a sentence template with one or more specified elements that specify at least a part of the word string. It is characterized in that the generation process of generating an output sentence is executed by designating at least a part of the word string of the summarized element with the designated element in the sentence generation template that can be specified.

According to a typical embodiment of the present invention, a sentence can be transformed into an appropriate form. Issues, configurations and effects other than those described above will be clarified by the description of the following examples.

FIG. 1 is an explanatory diagram showing an example of sentence generation by the sentence generation device according to the first embodiment. FIG. 2 is a block diagram showing a hardware configuration example of the sentence generator. FIG. 3 is an explanatory diagram showing an example of an extraction rule DB (database). FIG. 4 is an explanatory diagram showing an example of the sentence generation template DB. FIG. 5 is a flowchart showing an example of a sentence generation processing procedure by the sentence generation device according to the first embodiment. FIG. 6 is an explanatory diagram showing an example of the tree structure analysis result of the input sentence 100 by the tree structure analysis (step S501) according to the first embodiment. FIG. 7 is an explanatory diagram showing the acquisition result of the summary element by the acquisition of the summary element according to the first embodiment (step S502). FIG. 8 is an explanatory diagram showing the result of pruning by modifying the summary element according to the first embodiment (step S503). FIG. 9A shows an extraction result example 1 by extracting a word string from a summary element (step S504). FIG. 9B shows an extraction result example 2 by extracting a word string from a summary element (step S504). FIG. 9C shows an extraction result example 3 by extracting a word string from a summary element (step S504). FIG. 9D shows an extraction result example 4 by extracting a word string from a summary element (step S504). FIG. 9E shows an extraction result example 5 by extracting a word string from a summary element (step S504). FIG. 9F shows an example 6 of the extraction result by extracting the word string from the summary element (step S504). FIG. 10 is an explanatory diagram showing the connection result in the connection (step S505) by the sentence generation template according to the first embodiment. FIG. 11A is an explanatory diagram showing Example 1 of the conversion of the inflected flexion (step S506) according to the first embodiment. FIG. 11B is an explanatory diagram showing Example 2 of the conversion of the inflected flexion (step S506) according to the first embodiment. FIG. 12A is an explanatory diagram showing a specific example 1 of sentence generation by a sentence generation device. FIG. 12B is an explanatory diagram showing a specific example 2 of sentence generation by the sentence generation device. FIG. 12C is an explanatory diagram showing a specific example 3 of sentence generation by the sentence generation device. FIG. 13 is an explanatory diagram showing an example of detecting the tense according to the second embodiment. FIG. 14 is an explanatory diagram showing an example of detection of the negative form according to the second embodiment. FIG. 15 is an explanatory diagram showing an example of detection of the hypothetical form according to the second embodiment. FIG. 16 is an explanatory diagram showing an estimation example of the anaphora relationship according to the third embodiment. FIG. 17 is an explanatory diagram showing an example of complementing the zero pronoun according to the third embodiment. FIG. 18 is an explanatory diagram showing an example of an English input sentence. FIG. 19 is an explanatory diagram showing an example of the extraction rule DB according to the fourth embodiment. FIG. 20 is an explanatory diagram showing an example of the sentence generation template DB according to the fourth embodiment. FIG. 21 is an explanatory diagram showing an example of the tree structure analysis result of the input sentence by the tree structure analysis (step S501) according to the fourth embodiment. FIG. 22 is an explanatory diagram showing the acquisition result of the summary element by the acquisition of the summary element according to the fourth embodiment (step S502). FIG. 23 is an explanatory diagram showing the result of pruning by modifying the summary element according to the fourth embodiment (step S503). FIG. 24 is an explanatory diagram showing the connection result in the connection (step S505) by the sentence generation template according to the fourth embodiment.

In this specification, as an example, sentence shortening in the case of automatically creating minutes from transcribing remarks at a meeting will be described. Here, the minutes are documents that compactly summarize the decisions made at the meeting, the opinions and questions raised at the meeting, etc., and are indispensable for information sharing and recognition.

The minutes summary requires that the important types of information, such as decisions, be summarized in an easy-to-read format, rather than a thorough record of what was said at the meeting. In addition, since it is important to be able to grasp the contents of the meeting in a short time, it is important to use as few unnecessary words as possible in the summary of the minutes.

Further, as a method of document summarization, it is conceivable to identify an important sentence and use the sentence as it is, but the spoken sentence is often long, and one sentence may contain a plurality of information. For this reason, it is necessary to create a short sentence using only the parts that correspond to important types of information.

From this, the sentence generation device according to each of the following embodiments first extracts the kind of information to be recorded in the minutes, and combines the results to generate a sentence suitable for the minutes. This will automate the preparation of minutes as much as possible and reduce human work. In each of the examples shown below, the creation of minutes will be taken as an example, but it is useful for shortening sentences such as summaries, as well as for simplifying sentences such as generating question sentences and answer sentences, and generating headings. If so, it is not limited to the minutes. Hereinafter, the sentence generator according to each embodiment will be described with reference to the attached drawings.

<Sentence generation example>
FIG. 1 is an explanatory diagram showing an example of sentence generation by the sentence generation device according to the first embodiment. FIG. 1 is an example of generating a simplified sentence from the voice of a meeting and creating minutes. (1) First, the person in charge of preparing the minutes manually transcribe the audio of the meeting. This transcription may be done by an existing speech-to-text converter that converts the recorded speech of the conference into sentences.

(2) The sentence generation device applies an extraction rule from the sentence obtained in (1) to execute relation extraction. Relationship extraction is the extraction of the elements of a particular relationship between things from a sentence. For example, since there are essential elements for each type of information, if you want to extract the elements related to the content of the decision from the sentence, an extraction rule specialized for extracting the decision is applied.

When executing relation extraction, the sentence generator analyzes the sentence of the minutes generation source by tree structure analysis (processing to generate the tree structure of the sentence by morphological analysis and dependency analysis), and corresponds to the extraction rule which is a tree structure pattern. The element to be used is extracted from the tree structure analysis result. For example, the input sentence 100, which is the source of the minutes, says, "Since the operation is in the future, I would like to hear the opinions of the parties concerned tomorrow. ] Is the character string. The input sentence 100 is a sentence including an instruction.

The statement generator applies an extraction rule that extracts summary elements. The summary element is a word string that summarizes the word strings constituting the input sentence 100 by an extraction rule. A word string is a character string in which one or more words are continuous. As one of the instruction extraction rules applicable when the content of the input sentence 100 indicates an instruction, the sentence generation device applies, for example, the instruction extraction rule ir1. As a result, "about operation" is extracted as summary element # a1, "opinions of the parties concerned" as summary element # a2, and "I want you to listen tomorrow" as summary element # a0. (Indicated by a dotted rectangle in FIG. 1).

(3) The sentence generation device generates a sentence flattened by the sentence generation template. Specifically, for example, the sentence generator applies the summary elements # a1, # a2, # a0 extracted in (2) to the sentence generation template tp1. The sentence generation template tp1 is "# a1 is # a2 # a2. Case # a0.lemma". Summary elements # a1, # a2, # a2 specified in the sentence generation template tp1. case, # a0. Lema is referred to as a designated element in order to distinguish it from the summary element extracted in (2).

The statement generator extracts "operation" as the designated element # a1 from "about the operation" of the summary element # a1, and "also the opinions of the parties" of the summary element # a2 as the designated element # a2. Extract the "opinions of the parties" and specify the designated element # a2 from the "opinions of the parties" of the summary element # a2. The particle "mo" is extracted as a case, and the summary element a0. Extract the basic form "listening" of "listening" indicating the content word (verb) of the summary element # a0 as lemma.

Then, the sentence generation device uses the extracted designated elements # a1, # a2, # a2. case, # a0. Applying lemma to the sentence generation template tp1, the input sentence 100 is shortened (hereinafter, the output sentence) 101 to generate "operation listens to the opinions of the parties concerned". The output statement 101 is generated for each input statement 100. However, if the input sentence 100 does not correspond to the extraction rule, the summary element is not extracted and cannot be applied to the sentence generation template, so that the output sentence 101 may not be generated. Further, if the corresponding extraction rule or sentence generation template is different, the content of the output sentence 101 is also different. The output sentence 101 is output in a state in which the extraction rule and the sentence generation template tp are associated with each other. In this example, since the output statement 101 is made from the instruction extraction rule, it is output with the "instruction" label attached.

(4) The person in charge of creating the minutes selects or rearranges the plurality of output sentences 101 obtained in (3) with reference to the label associated with the output sentence 101, and the minutes 102. To create. As a result, it is possible to reduce the labor of identifying the information to be left in the minutes from the remarks made at the meeting and compiling it into an easy-to-read sentence.

<Hardware configuration example of sentence generator>
FIG. 2 is a block diagram showing a hardware configuration example of the sentence generator. The statement generator 200 includes a processor 201, a storage device 202, an input device 203, an output device 204, and a communication interface (communication IF) 205. The processor 201, the storage device 202, the input device 203, the output device 204, and the communication IF 205 are connected by the bus 206. The processor 201 controls the sentence generator 200. The storage device 202 serves as a work area for the processor 201. Further, the storage device 202 is a non-temporary or temporary recording medium for storing various programs and data. Examples of the storage device 202 include a ROM (Read Only Memory), a RAM (Random Access Memory), an HDD (Hard Disk Drive), and a flash memory. The input device 203 inputs data. The input device 203 includes, for example, a keyboard, a mouse, a touch panel, a numeric keypad, and a scanner. The output device 204 outputs data. The output device 204 includes, for example, a display, a printer, and a speaker. The communication IF 205 connects to a network and transmits / receives data.

The statement generator 200 may be a server of a client-server system. In this case, the client transmits the input sentence 100 to the sentence generator 200, and the sentence generator 200 transmits the output sentence 101 to the client. Further, the sentence generator 200 may be a stand-alone type. For example, the statement generator 200 may be a smartphone or a personal computer in which a statement generator is installed. Whether it is a client server system or a stand-alone type, the sentence generator 200 generates, for example, an input sentence 100 from an input voice (may be a text input of the input sentence 100), and outputs an output sentence 101 by voice (screen display). It can be used for chatbots. Further, the statement generation program installed in the statement generation device 200 may be introduced from the storage medium of the external device via the network and / or via the portable storage medium.

<Database>
FIG. 3 is an explanatory diagram showing an example of an extraction rule DB (database). The extraction rule DB 300 is a database that stores various extraction rules. The extraction rule DB 300 is stored in the storage device 202 of the sentence generation device 200. Further, the extraction rule DB 300 may be stored in a computer capable of communicating with the sentence generation device 200.

The extraction rule DB 300 may be used, for example, to distinguish between the instruction extraction rules ir1, ir2 ... (when ir1, ir2, ... Are not distinguished, simply expressed as ir) and the determination extraction rules dr1, ... (Simply written as dr), opinion extraction rule or1, ... (simply written as or when not distinguishing or1, ...) and question extraction rule qr1, ... (simply written as qr when not distinguishing between qr1, ...) ) And the common extraction rules cr1, ... (If no distinction is made, simply express as cr).

The decision extraction rule dr is an extraction rule that can be applied when the content of the input sentence 100 indicates a decision. The opinion extraction rule or is an extraction rule that can be applied when the content of the input sentence 100 indicates an opinion. The question extraction rule qr is an extraction rule that can be applied when the content of the input sentence 100 indicates an opinion. The common extraction rule cr is an extraction rule that can be applied regardless of the content of the input sentence 100.

The input sentence 100 itself is not given an index indicating what kind of content the input sentence 100 indicates. For example, if the instruction extraction rule ir1 is applied to the input sentence 100, the summary element related to the instruction is extracted from the input sentence 100, and if the decision extraction rule dr1 is applied to the input sentence 100, the summary element related to the decision is applied to the input sentence 100. Extracted from. The same applies to the opinion extraction rule or1 and the question extraction rule qr1.

Here, the details of the instruction extraction rule ir1 will be specifically described. The "((id. Instruction_01))" in the first line indicates the index of the instruction extraction rule ir1.

The second and sixth lines "(# a0.Suf = tai & .POS = verb * ... *)" are from the input sentence 100 with the phrase including the verb and the auxiliary verb "tai" as the summary element a0. The extraction condition C1 to be extracted. "*" Indicates that any phrase to which the parent phrase is related and received may exist. In the case of the input sentence 100, "I want you to listen to tomorrow" is extracted as the summary element # a0.

The third line ".lemma = koto & .POS = noun & .POS2 = non-independent *" is a condition that becomes a child node of the extraction condition in the second line, but is not an extraction condition. Specifically, for example, it is a detection condition D for detecting that a non-independent noun "koto", which is a content word (lemma), exists as a child node having a node of # a0 as a dependency. In the case of the input sentence 100, the existence of "koto" is detected.

“About # a1.case = & .POS = noun *” on the 4th line is the extraction condition C2 that is a child node of the detection condition D on the 3rd line. Specifically, for example, a noun and a particle. A phrase containing "about" such as "about" is extracted from the input sentence 100 as the summary element # a1. In the case of the input sentence 100, "about operation" is extracted as the summary element # a1.

The sixth line "# a2.case = mo & .POS = noun *" is an extraction condition C3 for extracting a phrase including a noun and a particle "mo" from the input sentence 100 as a summary element a2. In the case of the input sentence 100, "the opinions of the parties concerned" is extracted as the summary element # a2.

Further, the common extraction rule cr1 is a deletion rule (pruning condition) for specifying a word string to be deleted from the target summary element # a0 and deleting it, so-called pruning. The phrase that becomes the common extraction rule cr1 is a phrase that has a word classified as an adverb in the morphological analysis at the beginning. Subtrees with this phrase as their roots are pruned. In the case of "I want you to hear tomorrow" of the summary element # a0, "Tomorrow" is deleted, that is, the branch is pruned, and the summary element # a0 becomes "I want you to hear".

In this way, by applying the extraction rule, the sentence generation device 200 can extract the summary element by using the modification relation (dependency analysis result) between the clauses of the input sentence 100 and the morphological analysis result as clues. Become.

FIG. 4 is an explanatory diagram showing an example of the sentence generation template DB. The sentence generation template DB 400 is a database that stores sentence generation templates tp1, tp2, tp3, tp4, tp5, ... (If tp1, tp2, tp3, tp4, tp5, ... Are not distinguished, simply tp). The sentence generation template DB 400 is stored in the storage device 202 of the sentence generation device 200. Further, the sentence generation template DB 400 may be stored in a computer capable of communicating with the sentence generation device 200.

The sentence generation template tp is a template for generating an output sentence 101 by extracting a word string of a designated element from the extracted summary element and setting it at the position of the component in the sentence generation template tp. In the case of the sentence generation template tp1, the designated elements # a1, # a2, # a2. case and # a0. lemma is applied.

Although not shown, the sentence generator 200 refers to a word dictionary. The word dictionary stores information necessary for morphological analysis, such as part of speech, basic form, inflected form, and reading, for each word. The word dictionary is stored in the storage device 202 of the sentence generator 200. Further, the word dictionary may be stored in a computer capable of communicating with the sentence generator 200.

<Sentence generation processing procedure>
FIG. 5 is a flowchart showing an example of a sentence generation processing procedure by the sentence generation device 200 according to the first embodiment. The sentence generation device 200 includes tree structure analysis (step S501), acquisition of summary elements (step S502), modification of summary elements (step S503), extraction of word strings from summary elements (step S504), and concatenation by a sentence generation template. (Step S505) and conversion of the inflected word (step S506) are executed to end a series of processes. Specifically, the processing of the tree structure analysis (step S501) to the conversion of the inflected flexion (step S506) is executed by the processor according to the sentence generation program stored in the storage device 202, for example. Hereinafter, steps S501 to S506 will be described in order.

FIG. 6 is an explanatory diagram showing an example of the tree structure analysis result of the input sentence 100 by the tree structure analysis (step S501) according to the first embodiment. The tree structure analysis result 600 is data to be extracted by the extraction rule. In the tree structure analysis (step S501), the sentence generation device 200 analyzes the input sentence 100 in a tree structure. Specifically, for example, the sentence generation device 200 executes morphological analysis and dependency analysis on a phrase-by-phrase basis for the input sentence 100. As a result, the tree structure analysis result 600 is obtained.

FIG. 7 is an explanatory diagram showing the acquisition result of the summary element by the acquisition of the summary element according to the first embodiment (step S502). In the acquisition of the summary element (step S502), the sentence generator 200 applies, for example, all the extraction rules of the extraction rule DB 300 to the tree structure analysis result 600 to extract the summary element. Specifically, for example, the sentence generator 200 specifies the extraction condition C1 in the second row of the instruction extraction rule ir1 from the tree structure analysis result 600, and sets the summary element # a0 corresponding to the extraction condition C1 as “Listen tomorrow”. I want you to do it. "

Further, the sentence generation device 200 specifies the extraction condition C2 in the third row of the instruction extraction rule ir1 from the tree structure analysis result 600, and extracts "about operation" as the summary element # a1 corresponding to the extraction condition C2. Further, the sentence generation device 200 identifies the extraction condition C3 on the fourth line of the instruction extraction rule ir1 from the tree structure analysis result 600, and sets the summary element # a2 corresponding to the extraction condition C3 as "the opinions of the parties concerned. Is extracted. Further, the sentence generation device 200 detects that the input sentence 100 satisfies the detection condition D.

Further, in the acquisition of the summary element (step S502), the sentence generation device 200 executes pruning when the extracted summary element corresponds to the common extraction rule cr1. The sentence generator 200 may acquire the summary elements # a0 to # a2 from another computer capable of communicating with the sentence generator 200.

FIG. 8 is an explanatory diagram showing the result of pruning by modifying the summary element according to the first embodiment (step S503). In FIG. 8, “tomorrow”, in which the morphological analysis result can be adverbed, is pruned from the summary element # a0. Hereinafter, unless otherwise specified, in the first embodiment, the summary element # a0 is "please listen" after pruning.

9A-9F show examples 1 to 6 of extraction results by extracting a word string from a summary element (step S504). In the extraction of the word string from the summary element (step S504), the sentence generator 200 extracts the word string from the summary elements # a0 to # a2 according to the summary element (hereinafter, designated element) constituting the sentence generation template. .. The specified elements include, for example, #an (n is an integer greater than or equal to 0), #an. lemma, # an. case, # an. suf, # an. There is a tense.

For example, #an is a specification element that extracts the content words of the root clause contained in the summary element #an (for example, verbs, nouns, adjectives, excluding particles and auxiliary verbs) and the words of the child nodes. # An. lemma is a designated element that extracts the content word of the clause of the root from the summary element #an.

# An and # an. The difference from lemma is whether or not to extract the word of the child node. For example, if the designated element # a2 is applied to the "opinion of the parties" of the summary element # a2, the child node "of the parties" of the root "opinion" is also included in the "parties". Opinion of "is extracted. On the other hand, the designated element # a2 in the "opinion of the parties" of the summary element # a2. When lemma is applied, only the root "opinion" is extracted without including the child node "parties" of the root "opinion".

# An. Case is a designated element that extracts a specific particle of the root clause from the summary element #an. At this time, particles that do not have a significant effect on the meaning of the sentence, such as the sub-particle "etc." and the sub-particle "only", are not included in the specific particles and are not extracted. # An. The suf is a designated element that extracts a specific auxiliary verb of the root clause from the summary element #an. # An. Tense is a designated element that extracts the auxiliary verb representing the tense of the root clause from the summary element #an.

In the case of the sentence generation template tp1, the specified element is # a1. lemma, # a2. lemma, # a2. case, # a0. lemma.

As shown in FIG. 9A, the sentence generator 200 extracts the noun "operation" from the "operation" of the summary element # a1 by the designated element # a1. That is, "about" and "ha" are excluded from the extraction target because they are determined to be particles by morphological analysis.

As shown in FIG. 9B, the sentence generator 200 extracts the noun phrase "opinions of the parties" from the "opinions of the parties" of the summary element # a2 by the designated element # a2. .. That is, "etc." and "mo" are excluded from the extraction target because they are determined not to be content words by morphological analysis.

As shown in FIG. 9C, the sentence generator 200 has the designated element # a2. "Mo" is extracted by case. "Etc." is also a particle, but it is not included in the extraction because it is not a specific particle mentioned above.

As shown in FIG. 9D, the sentence generator 200 has the designated element # a0. With lemma, the basic form "listening" of the verb "listening", which is the content word of the root phrase, is extracted. That is, since it is determined by morphological analysis that the two "te" are particles and the "tai" is an auxiliary verb, and that "ya" and "kake" are not content words, they are excluded from the extraction target. ..

9E and 9F show the extraction of word strings by designated elements not included in the sentence generation template tp1. Only in FIGS. 9E and 9F, the past tense "I want you to hear" is taken as an example of "I want you to hear" of the summary element # a0. Since the sentence generator 200 "wants to hear" the summary element # a0, the designated element # a0. The auxiliary verb "ta" that represents the past is extracted by tense. In the case of FIG. 9F, the sentence generator 200 has the designated element # a0. The suf extracts the basic form "tai" of the auxiliary verb "taka" that expresses hope.

FIG. 10 is an explanatory diagram showing the connection result in the connection (step S505) by the sentence generation template according to the first embodiment. In the concatenation by the sentence generation template (step S505), the sentence generator 200 has the "operation", "ha", "opinion of the parties", extracted by the extraction of the word string from the summary element (step S504). By applying "mo" and "listening" (see FIGS. 9A to 9D) to the sentence generation template tp1, "operation also listens to the opinions of the parties concerned" is generated as the consolidation result 1000.

In the conversion of the inflected conjugation (step S506), the sentence generation device 200 converts the inflected conjugation for the concatenation result 1000 in the concatenation by the sentence generation template (step S505), and outputs the output sentence 101.

FIG. 11A is an explanatory diagram showing Example 1 of the conversion of the inflected flexion (step S506) according to the first embodiment. In the case of the concatenation result 1000, the verb "listening" is located at the end of the sentence, and the basic form of the verb "listening" is the same as the imperfect form, so the conjugation ending "ku" of the verb "listening" is not converted. As a result, the sentence generation device 200 outputs "the operation also listens to the opinions of the parties concerned" as the output sentence 101.

FIG. 11B is an explanatory diagram showing Example 2 of the conversion of the inflected flexion (step S506) according to the first embodiment. In the case of the connection result 1100 of FIG. 11B, the basic forms of the verbs "listen" and "summary" are connected. In such a case, the verb "listen" is not located at the end of the sentence, and the verb "summary" exists after the verb. , And add the conjunctive particle "te" to connect with the verb "summary".

Since the verb "summary" is located at the end of the sentence and the basic form of the verb "summary" is the same as the imperfect form, the conjugation ending "ru" of the verb "summary" is not converted. As a result, the sentence generation device 200 outputs "the operation also listens to the opinions of the parties concerned and summarizes" as the output sentence 1101.

<Specific example of sentence generation>
12A to 12C are explanatory views showing specific examples 1 to 3 of sentence generation by the sentence generation device 200. FIGS. 12A to 12C show the difference in the output sentence when different sentence generation templates are applied to the same input sentence 1211 "Prime Minister XX announced that X left the country in a hurry yesterday before dawn".

FIG. 12A is an example in which the sentence generation template tp2 for shortening the sentence is applied as in the output sentence 101 described above. The relationship to be extracted by the applied extraction rule is "announcement". In the acquisition result 1212 of the summary element extracted by the extraction rule, # a1 is "Prime Minister XX", # a2 is "X left the country in a hurry the day before yesterday", and # a0 is "announced before dawn yesterday". Therefore, when the sentence generation template tp2 is applied, "Prime Minister XX announced that X has left the country the day before yesterday" is output as output sentence 1213.

FIG. 12B is an example in which the sentence generation templates tp3 and tp4 for generating the question sentence and the answer sentence are applied. The relationship to be extracted by the applied extraction rule is "announcement". In FIG. 12B, the same acquisition result 1212 as in FIG. 12A is used. Therefore, when the sentence generation template tp3 is applied, the output sentence 1223 will be "What did Prime Minister XX announce before dawn yesterday?" ] Is output. In addition, when the sentence generation template tp4 is applied, the answer sentence "The day before yesterday X left the country in a hurry" is output as the output sentence 1224.

FIG. 12C is an example in which the sentence generation template tp5 for generating a headline is applied. The relationship to be extracted by the applied extraction rule is "departure". In the acquisition result 1232 of the summary element extracted by the extraction rule, it is assumed that # a1 is "X" and # a0 is "busy departure". Therefore, when the sentence generation template tp5 is applied, "X departure" is output as the output sentence 1233.

As described above, according to the first embodiment, by outputting the output sentence in which the input sentence is simplified, the sentence generation can be automated as much as possible and the human labor can be reduced.

The second embodiment will be described. In the second embodiment, in the first embodiment, when the input sentence contains a compound sentence, the tense of the summary element #an extracted from the compound sentence and the negative form and the hypothetical form in the summary element #an are detected, and the summary element # This is an example of establishing consistency between an. In Example 2, since the differences from Example 1 will be mainly described, the description of Example 1 will be omitted. In Example 2, a compound sentence will be described as an example, but the same applies to a compound sentence.

FIG. 13 is an explanatory diagram showing an example of detecting the tense according to the second embodiment. The sentence generator 200 executes the tree structure analysis (step S501) on the input sentence 1300, so that the input sentence 1300 receives the preceding sentence "A promptly distributes the material related to the in-house system" and the following sentence "break." Later, B made a presentation using the material. 』Detects that it is a compound sentence. Further, the sentence generation device 200 outputs the extraction result 1301 by executing the acquisition of the summary element (step S502). The extraction result 1301 includes the summary element 1311-1315 as the summary element #an.

In the modification of the summary element (step S503), the sentence generator 200 compares the summary elements 1313 and 1315 including the verbs "distribute" and "explain" of each sentence of the compound sentence, and the summary element 1313 is a verb. It is detected that the auxiliary verb representing the tense of "distribute" does not exist, and the auxiliary verb "ta" representing the tense of the verb "explain" exists in the summary element 1315. In this case, the meaning of the verb "distributed" in the summary element 1313 is not the present tense "distributed" but the past tense "distributed". Therefore, the sentence generator 200 adds the auxiliary verb "ta" representing the past to the summary element 1313 and "promptly distributes" in order to make the tense of the verb "distribute" of the summary element 1313 of the preceding sentence into the past tense. I corrected it to. "

After that, the sentence generator 200 extracts the word string from the subsequent summary element (step S504), concatenates by the sentence generation template (step S505), and converts the inflected flexion (step S506) based on the modified tense. ) Is executed. As a result, the sentence generation device 200 can read the intention of the input sentence 100 and express the output sentence 101 in an appropriate tense.

FIG. 14 is an explanatory diagram showing an example of detection of the negative form according to the second embodiment. The sentence generator 200 executes a tree structure analysis (step S501) on the input sentence 1400, and the input sentence 1400 follows the preceding sentence "The attendees of the meeting take out the materials related to the in-house system to the outside." Sentence "Please do not leak the contents of today's meeting to anyone outside the company. 』Detects that it is a compound sentence. Further, the sentence generation device 200 outputs the extraction result 1401 by executing the acquisition of the summary element (step S502). The extraction result 1401 includes summary elements 141 to 1415 as summary elements #an.

In the modification of the summary element (step S503), the sentence generator 200 compares the summary elements 1413 and 1415 including the verbs “take-out” and “leakage” of each sentence of the compound sentence, and the summary element 1413 is the verb “take-out”. It is detected that there is no auxiliary verb representing the denial of the verb "leakage" and the auxiliary verb "not" representing the denial of the verb "leakage" is present in the summary element 1415. In this case, the correct meaning of the verb "take out" of the summary element 1413 is not the affirmative form "take out" but the negative form "do not bring out". Therefore, in order to make the verb "take-out" of the summary element 1413 of the preceding sentence a negative form, the sentence generator 200 adds the auxiliary verb "not" indicating negation to the summary element 1413 and modifies it to "do not take out". do.

After that, the sentence generator 200 extracts the word string from the subsequent summary element (step S504), concatenates by the sentence generation template (step S505), and converts the inflected flexion (step S506) based on the modified tense. ) Is executed. In the conversion of the inflected flexion (step S506), "do not take out" is converted to "do not take out". As a result, the sentence generation device 200 can read the intention of the input sentence 1400 and express the output sentence in an appropriate usage.

FIG. 15 is an explanatory diagram showing an example of detection of the hypothetical form according to the second embodiment. By executing the tree structure analysis (step S501) for the input sentence 1500, the sentence generation device 200 can display the input sentence 1500 "If the data of the materials related to the in-house system is taken home and saved in the personal computer at home". , Detects that it is a compound sentence of the preceding sentence "Take the data of the material related to the in-house system home" and the following sentence "If you save it on your home computer". Further, the sentence generation device 200 outputs the extraction result 1501 by executing the acquisition of the summary element (step S502). The extraction result 1501 includes the summary elements 151 to 1515 as the summary element #an.

In the modification of the summary element (step S503), the sentence generator 200 compares the summary elements 1512 and 1513 including the verbs “take-out” and “save” of each sentence of the compound sentence, and the summary element 1512 is replaced with the verb “take-out”. It is detected that there is no connecting particle that represents the assumption of "take-out", and that the summarizing element 1513 has the connecting particle "ba" that represents the assumption of the verb "save". In this case, the meaning of the verb "take-out" of the summary element 1512 is not the assertive form of "take-out" but the hypothetical form of "take-out". Therefore, in order to make the verb "take-out" of the summary element 1512 of the preceding sentence into a hypothetical form, the sentence generator 200 adds the connecting particle "ba" representing the assumption to the summary element 1512 to "take-out". Correct.

After that, the sentence generator 200 extracts the word string from the subsequent summary element (step S504), concatenates by the sentence generation template (step S505), and converts the inflected flexion (step S506) based on the modified tense. ) Is executed. In the conversion of the inflected flexion (step S506), the connection result "if you take it home and save it on your home computer" is converted to "if you take it home or save it on your home computer". As a result, the sentence generation device 200 can read the intention of the input sentence 1500 and express the output sentence in an appropriate usage.

As described above, according to the second embodiment, it is possible to match the tenses, negative forms, and hypothetical forms of a plurality of verbs in a compound sentence to optimize the output sentence. In Example 2, the verbs in the compound sentence have been described as an example, but the same applies to adjectives and adjective verbs.

Example 3 is an example in which the correspondence analysis is executed for the input sentence in Example 1 or Example 2. Anaphora analysis is a process of estimating referents of anaphora such as pronouns and demonstratives, and complementing omitted noun phrases (zero pronouns). In Example 3, since the differences from Example 1 and Example 2 will be mainly described, the description of Example 1 and Example 2 will be omitted.

FIG. 16 is an explanatory diagram showing an estimation example of the anaphora relationship according to the third embodiment. The sentence generation device 200 distributed the material related to the in-house system by "A distributed the material related to the in-house system" of the advance input sentence 1601. "And the subsequent input sentence 1602," After that, he explained the in-house system. ], The tree structure analysis (step S501) is executed. In the tree structure analysis (step S501), "A" is identified as a proper noun of a person's name, and "he" is identified as a pronoun.

Further, the sentence generation device 200 outputs the extraction result 1610 regarding the preceding input sentence 1601 and the extraction result 1620 regarding the subsequent input sentence 1602 by executing the acquisition of the summary element (step S502). The extraction result 1610 regarding the type-ahead sentence 1601 includes the summary elements 1611 to 1613 as the summary element #an. The extraction result 1620 regarding the subsequent input sentence 1602 includes the summary elements 1621 and 1622 as the summary element #an.

In the modification of the summary element (step S503), the sentence generator 200 identifies the anaphoric relationship between the extraction result 1610 for the preceding input sentence 1601 and the extraction result 1620 for the subsequent input sentence 1602 by performing the anaphoric analysis. .. Specifically, for example, the sentence generator 200 specifies the pronoun "he" of "A" in the summary element 1611 of the extraction result 1610 for the preceding input sentence 1601 from the extraction result 1620 for the subsequent input sentence 1602. The sentence generator 200 converts the pronoun "he", which has an anaphoric relationship with "A", into "A". Thereby, "he is" of the summary element 1621 is corrected to "A is" of the summary element 1631.

After that, the sentence generation device 200 extracts the word string (step S504), concatenates with the sentence generation template (step S505), and converts the inflected flexion (step) based on the extraction result 1610 and the corrected extraction result 1620. S506) is executed. In this way, the sentence generation device 200 can convert a pronoun into a proper noun from the anaphoric relationship and generate an appropriate output sentence.

FIG. 17 is an explanatory diagram showing an example of complementing the zero pronoun according to the third embodiment. The sentence generation device 200 distributed the material related to the in-house system by "A distributed the material related to the in-house system" of the advance input sentence 1601. "And the subsequent input sentence 1702," After that, the in-house system was explained. ], The tree structure analysis (step S501) is executed. In the tree structure analysis (step S501), "A" is identified as a proper noun of a personal name.

Further, the sentence generation device 200 outputs the extraction result 1610 regarding the preceding input sentence 1601 and the extraction result 1720 regarding the subsequent input sentence 1602 by executing the acquisition of the summary element (step S502). The extraction result 1610 regarding the type-ahead sentence 1601 includes the summary elements 1611 to 1613 as the summary element #an. The extraction result 1720 regarding the subsequent input sentence 1702 includes the summary element 1622 as the summary element #an.

In the modification of the summary element (step S503), the sentence generator 200 identifies the anaphoric relationship between the extraction result 1610 for the preceding input sentence 1601 and the extraction result 1720 for the subsequent input sentence 1702 by performing the anaphoric analysis. .. Specifically, for example, the sentence generator 200 specifies that the subject of the succeeding sentence is the subject "A is" of the preceding sentence because the subject does not exist in the succeeding input sentence 1702. Therefore, the sentence generator 200 adds the summary element 1611 including the subject "A ga" in the extraction result 1610 to the extraction result 1620 as a zero pronoun. This complements the zero pronoun.

After that, the sentence generation device 200 extracts the word string (step S504), concatenates by the sentence generation template (step S505), based on the extraction result 1610 and the extraction result 1720 after completion to which the summary element 1611 is added. And the conversion of the inflection ending (step S506) is executed. In this way, the sentence generator 200 can complement the zero pronoun to generate an appropriate output sentence.

The fourth embodiment will be described. In Examples 1 to 3, the sentence generator 200 has described an example in which a Japanese output sentence is generated for a Japanese input sentence. In the fourth embodiment, an example in which the sentence generation device 200 generates an English output sentence with respect to an English input sentence will be described. In Example 4, since the differences from Examples 1 to 3 will be mainly described, the description of Example 1 will be omitted.

FIG. 18 is an explanatory diagram showing an example of an English input sentence. The input sentence 1800 is in Japanese, "I would like to hear the opinions of the parties concerned as the operation is in the future. 』It is an English translation.

FIG. 19 is an explanatory diagram showing an example of the extraction rule DB 300 according to the fourth embodiment. The extraction rule DB 1900 is a database that stores extraction rules that can be used when the language of the input sentence 1800 is English. The extraction rule DB 1900 is a database that stores various extraction rules. The extraction rule DB 1900 is stored in the storage device 202 of the sentence generation device 200. Further, the extraction rule DB 1900 may be stored in a computer capable of communicating with the sentence generator 200.

The extraction rule DB 1900 is similar to the extraction rule DB 300 of FIG. 3, for example, the instruction extraction rules ire1, ire2 ... If you do not distinguish dr1e, ..., simply write it as dr), the opinion extraction rule or1e, ... (If you do not distinguish or1e, ..., simply write it as ore), and the question extraction rules qr1, ... (If not distinguished, simply written as qre) and the common extraction rule cr1e, ... (If not distinguished, simply written as cr) are stored.

Here, the details of the instruction extraction rule ir1e will be specifically described. In the second line, ".type = root & .POS = VBN & .lemma = discus", the type of dependency (type) is the root, and the part of speech is the past participle form (VBN) of the verb, and its participle. It is a description that the content word (verb) specifies "discuss". As a result, the sentence generator 200 can acquire an English input sentence related to the discussed content as a summary element.

The third line "# a1.type = nsubjpass & .POS = NN *" is a noun (nsubjpass) whose dependency type (type) is the subject of the passive voice as a child node of the description in the second line. , Is an extraction condition for extracting the summary element #a1 whose part of speech is the singular form (NN) of a noun.

The fourth line ".type = aux & .POS = MD & .lemma = will" is a child node of the description in the second line, and the type of dependency (type) is an auxiliary verb (aux) related to the verb. It is a description that specifies the content word "will" whose part of speech is an auxiliary verb (MD).

In the sixth line, "# a0.type = advcl & .POS = VB & .lemma = like", the type of dependency (type) is an adverb clause modifier (advcl), and the part of speech is the original form of the verb (VB). This is an extraction condition for extracting the content word "like", which is, as a summary element # a0.

The 8th line ".type = nsubj & .POS = PRP & .lemma = I" is a noun (nsubj) whose dependency type (type) is the subject of the active voice as a child node of the extraction condition in the 6th line. In addition, it is a description that specifies the content word "I" whose part of speech is a pronoun (PRP).

The 9th line ".type = aux & .POS = MD & .lemma = world" is a child node of the extraction condition on the 6th line, and the dependency type (type) is an auxiliary verb (aux) related to the verb. , Is a description that specifies the content word "world" whose part of speech is an auxiliary verb (MD).

".Type = xcomp & .POS = VB" on the 10th line is the original form (VB) of the verb whose dependency type (type) is an open phrase complement (xcomp) as a child node of the extraction condition on the 6th line. It is a description to specify.

The 11th line ".type = aux & .POS = TO & .lemma = to" is a child node of the description on the 10th line, and the type of dependency (type) is an auxiliary verb (aux) related to the verb. It is a description that specifies the content word "to" whose part of speech is the TO infinitive (TO).

The 12th line "# a2.type = dobj & .POS = N. **)" is a child node of the description on the 10th line, and the dependency type (type) is the direct object (dobj). , Is an extraction condition for extracting the summary element # a2 which is a noun.

Further, the common extraction rule cr1e is a pruning condition for performing so-called pruning, in which a word to be deleted is specified from the target summary element # a0, as in the common extraction rule cr1. The "(type_pruned.dbj)" of the common extraction rule cr1e is a conditional expression that the type of dependency (type) is a direct object (dobj), and "(POS_prund.IN)" has a part of speech as a preposition or a subordinate. Indicates that it is a conjunction (IN). That is, the common extraction rule cr1e is a pruning condition in which the object (dobj) and the preposition or the subordinate conjunction are deleted directly from the summary element # a0.

FIG. 20 is an explanatory diagram showing an example of the sentence generation template DB 400 according to the fourth embodiment. Similar to the sentence generation template DB400 of FIG. 4, the sentence generation template DB2000 is a database that stores, for example, the sentence generation templates tp1e, tp2e, tp3e, ... be. The sentence generation template DB 400 is stored in the storage device 202 of the sentence generation device 200. Further, the sentence generation template DB 2000 may be stored in a computer capable of communicating with the sentence generation device 200.

The sentence generation template tp is a template for generating an output sentence by setting the extracted summary element at the position of the component in the sentence generation template pe. In the case of the sentence generation template tp1e, the summary elements # a1, # a2, # a0 are applied.

<Sentence generation processing procedure>
In the sentence generation processing procedure example by the sentence generation device 200 according to the fourth embodiment, the sentence generation device 200 includes the tree structure analysis (step S501), the acquisition of the summary element (step S502), and the modification of the summary element (step S502) shown in FIG. Step S503), concatenation by the sentence generation template (step S505), and conversion of the inflected flexion (step S506) are executed to end a series of processes.

FIG. 21 is an explanatory diagram showing an example of the tree structure analysis result of the input sentence 1800 by the tree structure analysis (step S501) according to the fourth embodiment. The tree structure analysis result 2100 is the data to be extracted by the extraction rule. In the tree structure analysis (step S501), the sentence generation device 200 analyzes the input sentence 1800 in a tree structure. Specifically, for example, the sentence generator 200 executes morphological analysis and dependency analysis on a word-by-word basis for the input sentence 100. As a result, the tree structure analysis result 2100 is obtained.

In the tree structure analysis result 2100 of FIG. 21, in addition to the one shown in FIG. 19, "det" is a determiner, "prep" is a preamble, "pobj" is an object of a pronoun, "DT" is an article, and "mark". "Is a conjunction," NNS "is a noun (plural)," rcmod "is a relative clause, and" WP is a Who pronoun (single form).

FIG. 22 is an explanatory diagram showing the acquisition result of the summary element by the acquisition of the summary element according to the fourth embodiment (step S502). In the acquisition of the summary element (step S502), the sentence generator 200 applies, for example, all the extraction rules of the extraction rule DB 1900 to the tree structure analysis result 2100 to extract the summary element. Specifically, for example, the sentence generator 200 specifies the extraction condition C1e in the third row of the instruction extraction rule ir1e from the tree structure analysis result 2100, and sets the abstract element # a1 corresponding to the extraction condition C1e as the passive voice of the verb. Extract "the operation" which is the subject of "discussed".

Further, the sentence generation device 200 identifies the extraction condition C2e on the sixth line of the instruction extraction rule ir1e from the tree structure analysis result 600, and sets the summary element # a0 corresponding to the extraction condition C2e as “so I world like to her the opinion”. "of this who are involved in the project" is extracted. Further, the sentence generation device 200 identifies the extraction condition C2e on the 12th line of the instruction extraction rule ir1e from the tree structure analysis result 2100, and sets the summary element # a2 corresponding to the extraction condition C2e as "the opinion of this who are incorporated in". "the project." Is extracted.

Further, in the acquisition of the summary element (step S502), the sentence generation device 200 executes pruning when the extracted summary element corresponds to the common extraction rule cr1e.

FIG. 23 is an explanatory diagram showing the result of pruning by modifying the summary element according to the fourth embodiment (step S503). In FIG. 23, the direct object “the opinion of this who are involved in the project” defined in the common extraction rule cr1e and the preposition or the subordinate conjunction “so” are deleted from the summary element # a0. , Set the summary element # a0 to "I world like to hair". Hereinafter, unless otherwise specified, in Example 4, the summary element # a0 is referred to as "I world like to hair" after pruning.

In the fourth embodiment, after the modification of the summary element (step S503), the sentence generator 200 executes the tree structure analysis word by word in the tree structure analysis (step S501), so that the word string is extracted from the summary element (step). Concatenation by the sentence generation template (step S505) is executed without executing S504).

FIG. 24 is an explanatory diagram showing the connection result in the connection (step S505) by the sentence generation template according to the fourth embodiment. In the concatenation by the sentence generation template (step S505), the sentence generation device 200 applies the summary elements # a0 to #a2 to the sentence generation template tp1e, and as a concatenation result 2400, the sentence generation device 200 obtains “About the option, I wound to hear”. "the option of this is who are involved in the project." Is generated.

In the conversion of the inflected flexion (step S506), the sentence generator 200 converts the inflected flexion for the concatenation result 2400 in the concatenation by the sentence generation template (step S505) and outputs an output sentence. As the conversion of the inflected flexion (step S506), there is a process of adding or deleting the s (es) of the triad to the verb according to the subject, but since it is not necessary in the concatenation result 2400, the sentence is generated. The device 200 outputs the connection result 2400 as an output statement.

In this way, even if the language of the input sentence 1800 is English, as in the first embodiment, by outputting a simplified output sentence of the input sentence 1800, sentence generation is automated as much as possible and human beings are human. Work can be reduced. Further, regarding the detection of tenses, negative forms and hypothetical forms between compound sentences and the modification of the summary element regarding the detection result (step S503), the sentence generator 200 according to the fourth embodiment follows the process shown in the second embodiment. Can be executed. Further, the sentence generator 200 according to the fourth embodiment also executes the anaphora analysis (anaphora and zero pronoun) and the modification of the summary element related to the anaphora analysis result (step S503) according to the process shown in the third embodiment. Can be done.

Further, in Examples 1 to 4 described above, an example in which the language is Japanese or English has been described, but for other languages having a common grammar, Examples 1 to 4 are described as long as the grammar is common. Can be applied.

Further, the sentence generation device 200 according to the above-described first to fourth embodiments can be configured as described in (1) to (11) below.

(1) In the sentence generator 200 including the processor 201 for executing the program and the storage device 202 for storing the program, the processor 201 is one or more summarizing elements (# a0) which are word strings constituting the input sentence 100. , # A1, # a2) and the sentence generation template tp that constitutes a sentence template with one or more specified elements that specify at least a part of the word string. Specifying elements (# a1, # a2, # a2.case) in the sentence generation template tp that can specify at least a part of the word string of the summary element (# a0, # a1, # a2) acquired by (step S502). , # A0.lemma), the generation process (step S505) for generating the output statement 101 is executed.

By setting the sentence generation template tp in advance, the input sentence 100 can be converted into an appropriate form. As a result, the creation of the output sentence 101 based on the input sentence 100 can be automated as much as possible, and the human work can be reduced.

(2) In the sentence generation device 200 of the above (1), the processor 201 executes a tree structure analysis process (step S501) that analyzes the tree structure of the input sentence 100 and outputs the tree structure analysis result, and acquires the process (step S501). In S502), the processor 201 uses the extraction rule ir for extracting the summary elements (# a0, # a1, # a2) based on the tree structure analysis result 600 output by the tree structure analysis process (step S501). , The abstract elements (# a0, # a1, # a2) are extracted from the input sentence 100 (S502).

By setting the extraction rule ir and the sentence generation template tp in advance, the input sentence 100 can be converted into an appropriate form by using the word string of the summary element extracted from the input sentence 100 by the extraction rule ir. .. As a result, the creation of the output sentence 101 based on the input sentence 100 can be automated as much as possible, and the human work can be reduced.

(3) In the sentence generator 200 of the above (1), the processor 201 executes a correction process (step S503) for modifying the summary elements (# a0, # a1, # a2), and in the generation process (step S505). , Processor 201 refers to the sentence generation template tp, and specifies in the sentence generation template tp that can specify the word string (“I want you to listen”) of the summary element after the correction process (step S503). The output statement 101 is generated by specifying it in the element (# a0.lemma).

Thereby, the input sentence 100 can be converted into a more simplified and appropriate form than the input sentence 100 by using the modified summary element. Therefore, it is possible to automate the creation of the output sentence 101 as much as possible and reduce human work.

(4) In the correction process (step S503) in the sentence generator 200 of the above (3), the processor 201 summarizes based on the deletion rule (common extraction rule cr1) that deletes a part of the word string of the summary element. Delete some word strings ("Tomorrow") of the element ("I want you to listen").

As a result, unnecessary words can be deleted before the sentence generation template tp is applied, so that the input sentence 100 can be converted into an appropriate form that is simpler than the input sentence 100. Therefore, it is possible to automate the creation of the output sentence 101 as much as possible and reduce human work.

(5) In the sentence generation device 200 of the above (3), in the correction process (step S503), the processor 201 is the first tense word (“distributed”) that constitutes the preceding sentence of the compound sentence included in the input sentence 100. The tense of the first predicate and the tense of the second predicate are different for the first summary element 1313 including the above and the second summary element 1315 including the second predicate (“explained”) constituting the subsequent sentence of the compound sentence. In this case, the tense of the first predicate is changed to the tense of the second predicate.

As a result, the tense matching is executed before the sentence generation template tp is applied, so that the input sentence 100 can be converted into a form having a clearer meaning. Therefore, it is possible to automate the creation of the output sentence 101 as much as possible and reduce human work.

(6) In the sentence generation device 200 of the above (3), in the correction process (step S503), the processor 201 is the first predicate (“take out”) that constitutes the predicate of the compound sentence included in the input sentence 100. The first predicate is affirmative and the first predicate is affirmative with respect to the first summary element 1413 including 2 If the predicate is in the negative form, change the first predicate to the negative form.

As a result, the matching of the negative expressions is executed before the sentence generation template tp is applied, so that the input sentence 100 can be converted into a form having a clearer meaning. Therefore, it is possible to automate the creation of the output sentence 101 as much as possible and reduce human work.

(7) In the sentence generation device 200 of the above (3), in the correction process (step S503), the processor 201 is the first predicate (“take back”) that constitutes the predicate of the compound sentence included in the input sentence 100. The first predicate is affirmative and the second is about the first summary element 1512 including the compound sentence and the second summary element 1513 including the second predicate (“if saved”) constituting the subsequent sentence of the compound sentence. If the predicate is a hypothetical form, change the first predicate to the hypothetical form,

As a result, the matching of the hypothetical expressions is executed before the sentence generation template tp is applied, so that the input sentence 100 can be converted into a form having a clearer meaning. Therefore, it is possible to automate the creation of the output sentence 101 as much as possible and reduce human work.

(8) In the sentence generation device 200 of the above (3), in the correction process (step S503), the processor 201 inputs the first word string (“A”) of the specific nomenclature included in the type-ahead sentence 1601 and the type-ahead. It is detected that the second word string (“he”) of the anaphora contained in the subsequent input sentence 1602 following the sentence 1601 matches, and the second word string (“he”) is changed to the first word string (“he”). Change to "A").

As a result, the anaphora analysis is executed before the sentence generation template tp is applied, so that the input sentence 100 can be converted into a form in which the anaphora relationship is clear. Therefore, it is possible to automate the creation of output sentences as much as possible and reduce human work.

(9) In the sentence generation device 200 of the above (3), in the correction process (step S503), the processor 201 uses the word string (“A”) of a specific noun included in the first input sentence 1601 or a corresponding word (“A”) thereof. It is detected that "he" or "she") is not included in the second input sentence 1602 following the first input sentence 1601, and the summary element 1611 containing the word string of a specific noun ("A") is inserted. , Add to the set of one or more summary elements of the second input sentence 1602.

As a result, since the correspondence analysis is executed before the sentence generation template tp is applied, the input sentence 100 can be converted into a clear form by complementing the word string of the zero pronoun. Therefore, it is possible to automate the creation of output sentences as much as possible and reduce human work.

(10) In the sentence generator 200 of (1) above, when the processor 201 concatenates two verbs (“listening” and “summarizing”) in the output sentence 101, the processor 201 is the leading verb (“listening”). The conversion process (step S506) for converting the inflectional ending (“ku”) of is executed.

As a result, the input sentence 100 can be converted into an appropriate form that is grammatically correct and does not give a sense of discomfort. Therefore, it is possible to automate the creation of the output sentence 101 as much as possible and reduce human work.

(11) In the sentence generator 200 of (1) above, the processor 201 executes a conversion process (step S506) for converting the conjugation flexion of the verb in the predicate based on the relationship between the subject and the predicate in the output sentence 101. do.

As a result, the input sentence 100 can be converted into an appropriate form that is grammatically correct and does not give a sense of discomfort. Therefore, it is possible to automate the creation of the output sentence 101 as much as possible and reduce human work.

It should be noted that the present invention is not limited to the above-mentioned examples, but includes various modifications and equivalent configurations within the scope of the attached claims. For example, the above-described embodiment has been described in detail in order to explain the present invention in an easy-to-understand manner, and the present invention is not necessarily limited to those having all the described configurations. Further, a part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Further, the configuration of another embodiment may be added to the configuration of one embodiment. In addition, other configurations may be added, deleted, or replaced with respect to a part of the configurations of each embodiment.

Further, each of the above-mentioned configurations, functions, processing units, processing means, etc. may be realized by hardware by designing a part or all of them by, for example, an integrated circuit, and the processor 201 performs each function. It may be realized by software by interpreting and executing the program to be realized.

Information such as programs, tables, and files that realize each function is recorded in a memory, a hard disk, a storage device such as an SSD (Solid State Drive), or an IC (Integrated Circuit) card, an SD card, or a DVD (Digital Versail Disc). It can be stored in a medium.

In addition, the control lines and information lines show what is considered necessary for explanation, and do not necessarily show all the control lines and information lines necessary for mounting. In practice, it can be considered that almost all configurations are interconnected.

cr common extraction rule dr decision extraction rule ir instruction extraction rule or opinion extraction rule qr question extraction rule tp sentence generation template 300 extraction rule DB
400 Statement generation template DB
100 Input statement 101 Output statement 102 Minutes 200 Statement generator 201 Processor 202 Storage device 600 Tree structure analysis result 1900 Extraction rule DB

Claims (13)

A statement generator comprising a processor that executes a program and a storage device that stores the program.
The processor
Acquisition processing to acquire one or more summary elements that are word strings that make up an input sentence, and
The above can specify at least a part of the word string of the summary element acquired by the acquisition process by referring to the sentence generation template that constitutes the sentence template by one or more specification elements that specify at least a part of the word string. A generation process that generates an output statement by specifying it with the specified element in the statement generation template,
A statement generator characterized by executing. The sentence generator according to claim 1.
The processor
The tree structure analysis process that analyzes the tree structure of the input sentence and outputs the tree structure analysis result is executed.
In the acquisition process, the processor extracts the summary element from the input statement based on the tree structure analysis result output by the tree structure analysis process using the extraction rule for extracting the summary element.
A sentence generator characterized by that. The sentence generator according to claim 1.
The processor
Execute the correction process to correct the summary element,
In the generation process, the processor refers to the sentence generation template and specifies the word string of the summary element corrected by the modification process in the designated element in the sentence generation template that can be specified. Generate output statement,
A sentence generator characterized by that. The sentence generator according to claim 3.
In the correction process, the processor deletes a part of the word string of the summary element based on the deletion rule of deleting a part of the word string of the summary element.
A sentence generator characterized by that. The sentence generator according to claim 3.
In the correction process, the processor includes a first summary element including a first term that constitutes a preceding sentence of the compound sentence included in the input sentence, and a second predicate that includes a second predicate that constitutes the succeeding sentence of the compound sentence. When the tense of the first predicate and the tense of the second predicate are different from each other, the tense of the first predicate is changed to the tense of the second predicate.
A sentence generator characterized by that. The sentence generator according to claim 3.
In the correction process, the processor includes a first summary element including a first syntactic word constituting the preceding sentence of the compound sentence included in the input sentence, and a second predicate including a second predicate constituting the succeeding sentence of the compound sentence. With respect to the summary element, if the first predicate is affirmative and the second predicate is negative, the first predicate is changed to negative.
A sentence generator characterized by that. The sentence generator according to claim 3.
In the correction process, the processor includes a first summary element including a first syntactic word constituting the preceding sentence of the compound sentence included in the input sentence, and a second predicate including a second predicate constituting the succeeding sentence of the compound sentence. With respect to the summary element, when the first predicate is a definite form and the second predicate is a hypothetical form, the first predicate is changed to the hypothetical form.
A sentence generator characterized by that. The sentence generator according to claim 3.
In the correction process, the processor includes a first word string of a specific nomenclature included in the first input sentence, a second word string of anaphora included in the second input sentence following the first input sentence, and the second word string of the anaphoric words included in the second input sentence. Detects that the second word string is anaphoric, and changes the second word string to the first word string.
A sentence generator characterized by that. The sentence generator according to claim 3.
In the correction process, the processor detects that the word string of a specific noun included in the first input sentence or its anaphora is not included in the second input sentence following the first input sentence. A summary element containing the word string of the particular noun is added to the set of one or more summary elements of the second input sentence.
A sentence generator characterized by that. The sentence generator according to claim 1.
The processor
When two verbs are concatenated in the output sentence, a conversion process for converting the conjugation flexion of the first verb is executed.
A sentence generator characterized by that. The sentence generator according to claim 1.
The processor
Based on the relationship between the subject and the predicate in the output sentence, the conversion process for converting the conjugation flexion of the verb in the predicate is executed.
A sentence generator characterized by that. A statement generation method executed by a statement generator having a processor that executes a program and a storage device that stores the program.
The sentence generation method is
The processor
Acquisition processing to acquire one or more summary elements that are word strings that make up an input sentence, and
The above can specify at least a part of the word string of the summary element acquired by the acquisition process by referring to the sentence generation template that constitutes the sentence template by one or more specification elements that specify at least a part of the word string. A generation process that generates an output statement by specifying it with the specified element in the statement generation template,
A statement generation method characterized by executing. To the processor
Acquisition processing to acquire one or more summary elements that are word strings that make up an input sentence, and
The above can specify at least a part of the word string of the summary element acquired by the acquisition process by referring to the sentence generation template that constitutes the sentence template by one or more specification elements that specify at least a part of the word string. A generation process that generates an output statement by specifying it with the specified element in the statement generation template,
A sentence generation program characterized by executing.

Images