JP6095487B2 - Question answering apparatus and question answering method - Google Patents

Description

  The present invention relates to a question answering apparatus that responds to a question and a question answering method.

  A question answering technique for presenting an answer to a question sentence in a natural language input by a user operation has been studied. For example, in response to the question “How tall is Mt. Fuji,” an answer “3776 m” is output. Such a technique is applied to an application for receiving a question by voice.

  In the above question answering technique, there are roughly two approaches. One is to use an ordinary web search engine and estimate the answer based on the text included in the web search result. For example, in response to the question “What is the height of Mt. Fuji?”, A noun is extracted and converted into a search term “Mt. Fuji height” and input to the search engine.

  Then, an answer “3776m” is estimated by statistical processing from the title and the summary sentence included in the search result page output by the search engine. Non-Patent Document 1 discloses the technology and variations thereof.

  The other approach is to accumulate knowledge in the database in advance and convert the question text into a query that can be queried to the database by linguistic analysis to obtain an answer.

  For example, in response to the question “What is the height of Mt. Fuji?”, The information of entity (object) = “Mt. Fuji” and property (attribute) = “elevation” is extracted and converted to SQL, which is a database query language, The answer “3776m” is obtained. Non-Patent Document 2 and Non-Patent Document 3 disclose the technology.

  In general, the former approach can answer a variety of questions although the answer accuracy is inferior, and the latter approach can answer with a high accuracy rate although the corresponding range of the question is narrow.

Kashiwazaki, Question Answering System (Natural Language Processing Series 2), Corona Ken Yoshimura, “Speaking concierge and language processing”, SLP93-4, IPSJ Spoken Language Processing, October 2012 Ken Yoshimura, "Knowledge search engine development that realizes Q & A function of concierge", Artificial Intelligence Society, 88th Artificial Intelligence Basic Research Group, January 2013

  However, the methods of Non-Patent Documents 1 to 3 cannot respond to more complicated questions. For example, there is a problem that it is not possible to respond to a question in which a question sentence is further included in one question sentence, such as a question “What is the birthday of the creator of △△ □ (work name)?”. Even if the search term “◯ △ □ author's birthday” is input to the search engine in the approach using the web search engine, the correct answer is rarely included in the search result. Further, even in an approach using a database, it cannot be simply decomposed into an entity and a property, and cannot be directly converted into a database query language.

  Therefore, in order to solve the above-described problems, an object of the present invention is to provide a question answering apparatus and a question answering method capable of answering a question sentence including another question.

In order to solve the above-described problems, a question answering apparatus according to the present invention includes a question sentence accepting unit that accepts a question sentence, a partial question extracting part that extracts a partial question that is a part of the question sentence, and an answer to the partial question. The partial question answer acquisition means to be acquired, the question replacement means for replacing the part of the partial question in the question sentence with the answer acquired by the partial question answer acquisition means, and the replacement question for acquiring the answer to the question sentence replaced by the question replacement means An answer acquisition means; an answer output means for outputting an answer to the question sentence; and a first check word list. The partial question extraction means includes a word in the first check word list in the partial question. Whether or not the partial question is appropriate is determined based on whether or not the partial question is appropriate, and the partial question answer acquisition unit acquires the answer to the partial question when the partial question is appropriate.

The question answering method of the present invention is a question answering method executed by a question answering device, a question sentence accepting step for accepting a question sentence, and a partial question extracting step for extracting a partial question that is a part of the question sentence; , A partial question answer acquisition step for obtaining partial question answers, a question replacement step for replacing the part of the partial question in the question sentence with the answer obtained in the partial question answer acquisition step, and a question sentence replaced by the question replacement step A replacement question answer obtaining step for obtaining an answer; an answer output step for outputting an answer to the question sentence; and a first check word list. In the partial question extracting step, the first check word is assigned to the partial question. Based on whether or not a word is included in the list, it is judged whether the partial question is appropriate, and the partial question answer is obtained. In step, when partial query is appropriate to obtain answers to the subquery.

According to the present invention, the partial question included in the question sentence is extracted, the partial question portion is replaced with the answer in the partial question, and the answer of the replaced question is obtained. It is possible to accurately answer questions including Further, since the question answering device determines whether or not the partial question is appropriate using the first check word list, the question answering device is not appropriate as the partial question (not forming the body as the partial question). Therefore, it is possible to avoid obtaining an answer as a partial question.

Further, in the question answering apparatus of the present invention, a question sentence accepting means for accepting a question sentence, a partial question extracting means for extracting a partial question that is a part of the question sentence, and a partial question answer obtaining means for obtaining an answer to the partial question A question replacement unit that replaces the portion of the partial question in the question sentence with an answer acquired by the partial question answer acquisition unit, a replacement question answer acquisition unit that acquires an answer to the question sentence replaced by the question replacement unit, and a question sentence Answer output means for outputting an answer to and a second check word list, and the partial question extraction means determines whether or not the second check word list is included in a part excluding the partial question of the question sentence. Based on the above, it is determined whether or not the partial question is appropriate, and the partial question answer obtaining unit obtains an answer to the partial question when the partial question is appropriate .

  In this case, the question answering device determines whether the partial question in the question sentence is appropriate using the second check word list (whether the question sentence is a question sentence having a partial question or not). Therefore, it is possible to avoid obtaining an answer as a partial question even though it is not appropriate as a partial question.

Further, in the question answering apparatus of the present invention, a question sentence accepting means for accepting a question sentence, a partial question extracting means for extracting a partial question that is a part of the question sentence, and a partial question answer obtaining means for obtaining an answer to the partial question A question replacement unit that replaces the portion of the partial question in the question sentence with an answer acquired by the partial question answer acquisition unit, a replacement question answer acquisition unit that acquires an answer to the question sentence replaced by the question replacement unit, and a question sentence An answer output means for outputting an answer to the question answering means, and when the partial question answer obtaining means obtains a plurality of answers, the replacement question answer obtaining means assigns the partial question part of the question sentence to each of the plurality of answers. Answers to the replaced question sentences are acquired, and the answer output means outputs a predetermined number of answers from the answers acquired by the replacement question answer acquisition means .

  In this case, the question answering apparatus can output a representative answer when there are a plurality of answers to the partial question.

Further, in the question answering apparatus of the present invention, a question sentence accepting means for accepting a question sentence, a partial question extracting means for extracting a partial question that is a part of the question sentence, and a partial question answer obtaining means for obtaining an answer to the partial question A question replacement unit that replaces the portion of the partial question in the question sentence with an answer acquired by the partial question answer acquisition unit, a replacement question answer acquisition unit that acquires an answer to the question sentence replaced by the question replacement unit, and a question sentence An answer output means for outputting an answer to and a whole question answer obtaining means for obtaining an answer to a question sentence when a partial question answer cannot be obtained . The answer output means is obtained by the whole question answer obtaining means. Output the answer.

  In this case, the question answering apparatus can output the answer of the entire accepted question sentence even when there is no question partially.

Further, in the question answering apparatus of the present invention, a question sentence accepting means for accepting a question sentence, a partial question extracting means for extracting a partial question that is a part of the question sentence, and a partial question answer obtaining means for obtaining an answer to the partial question A question replacement unit that replaces the portion of the partial question in the question sentence with an answer acquired by the partial question answer acquisition unit, a replacement question answer acquisition unit that acquires an answer to the question sentence replaced by the question replacement unit, and a question sentence Answer output means for outputting an answer to the answer, and the partial question answer acquisition means may take the answer from which the supplementary part is deleted as the answer to the partial question when acquiring the answer having the supplementary part. In this case, since the question answering device removes the additional part and replaces the question sentence, the answer can be obtained more accurately.

  According to the present invention, a partial question included in a question sentence is extracted, a partial question portion is replaced with an answer to the partial question, and another question is included to obtain the answer to the question. You can answer questions.

3 is a block diagram illustrating functions of a question answering apparatus 100. FIG. 2 is a hardware configuration diagram of the question answering apparatus 100. FIG. It is a figure which shows an example of the data currently hold | maintained in entity-property type | mold reply DB108. It is a figure which shows an example of the data currently hold | maintained in 1 question 1 type | formula answer DB110. It is a flowchart which shows the process of an entity-property type question answer. It is a flowchart which shows the reply process with respect to a question sentence. It is a flowchart which shows the reply process with respect to a question sentence.

  Embodiments of the present invention will be described with reference to the accompanying drawings. When possible, the same parts are denoted by the same reference numerals, and redundant description is omitted.

  FIG. 1 is a block diagram illustrating functions of the question answering apparatus according to the present embodiment. The question answering device is various information processing devices such as a server device and a terminal device.

  It is a block diagram which shows the function structure of the question answering apparatus 100 of this embodiment. As shown in FIG. 1, the question answering apparatus 100 includes a question accepting unit 101 (question sentence accepting unit), a partial question extracting unit 102 (partial question extracting unit), a check word list 103, and an answer obtaining unit 104 (partial question answer). Acquisition means, replacement question answer acquisition means, overall question answer acquisition means), question editing section 105 (question replacement means), answer generation section 106, entity-property type question answering section 107, entity-property type answer DB section 108, one A question-and-answer-type question answering unit 109, a question-and-answer-type answer DB unit 110, and an answer output unit 111 (answer output means) are configured.

  FIG. 2 is a hardware configuration diagram of the question answering apparatus 100. As shown in FIG. 2, the question answering apparatus 100 shown in FIG. 2 physically includes one or a plurality of CPUs 11, a main memory RAM 12 and ROM 13, and input devices 14 such as a keyboard and a mouse. The computer system includes an output device 15 such as a display, a communication module 16 that is a data transmission / reception device such as a network card, an auxiliary storage device 17 such as a semiconductor memory, and the like. Each function in FIG. 2 operates the input device 14, the output device 15, and the communication module 16 under the control of the CPU 11 by reading predetermined computer software on the hardware such as the CPU 11 and the RAM 12 shown in FIG. 3. In addition, it is realized by reading and writing data in the RAM 12 and the auxiliary storage device 17. Hereinafter, each functional block will be described based on the functional blocks shown in FIG.

  The question receiving unit 101 receives a question sentence including text information (including information obtained by converting voice input information into text). Specifically, the question accepting unit 101 accepts a question sentence from another device via a communication network, or accepts a question sentence input by the input device 14.

  The partial question extraction unit 102 extracts a partial question that is a part of the question sentence based on the morphemes. Specifically, the partial question extraction unit 102 performs morphological analysis on the input question sentence, and divides the question sentence into morphemes. And the partial question extraction part 102 extracts a partial question from the whole question sentence. The partial question extraction unit 102 defines a learning model that is a model learned so that partial questions can be extracted in advance, and extracts partial questions based on the learning model. The learning model here is a model that defines a question sentence pattern (a pattern based on a word used in a question sentence).

  Furthermore, when the partial question extraction unit 102 extracts the partial question, the partial question extraction unit 102 verifies the partial question and determines whether or not the partial question is appropriate. Details will be described later.

  The check word list 103 is a database that manages a list (check word list for partial questions) that holds words for determining whether or not a partial question is appropriate. Specifically, the check word list 103 holds two lists. Specifically, the check word list 103a which is the first check word list includes “elevation” “height” “size” “author” “author” “birthday” “prime” “president” “representative work” “ It is a list that includes words and expressions such as “Year of publication”, “In Japan”, “In the world”, “Best”. The check word list 103b, which is the second check word list, includes “elevation”, “height”, “size”, “author”, “author”, “birthday”, “prime”, “president”, “masterpiece”, “published year”, etc. A list containing words and expressions. The check word list 103a and the check word list 103b are lists of words for determining whether the sentence is a question-type sentence.

  The answer acquisition unit 104 receives any answer of the replacement question, which is a question obtained by replacing the entire question (the entire question received by the question receiving unit 101), the partial question, and the partial question portion of the whole question with the answer of the partial question. get. Specifically, the answer acquisition unit 104 outputs any of the whole question, the partial question, and the replacement question to the answer generation unit 106, and acquires the question answer from the answer generation unit 106.

  The answer generation unit 106 acquires a predetermined question sentence (any one of the whole question, the partial question, and the replacement question) from the answer acquisition unit 104, and outputs an answer to the question sentence to the answer acquisition unit 104. The answer generation unit 106 includes an entity-property type question answering unit 107, an entity-property type DB 108, a one-question-and-answer-type question answering unit 109, and a one-question-and-answer-type answer DB 110.

  The answer generation unit 106 outputs an answer result for output to the answer acquisition unit 104 based on the answer result by the entity-property type question answering unit 107 and the answer result by the one-question-type question answering unit 109. Specifically, when an answer is obtained by the entity-property type question answering unit 107, the answer generating unit 106 outputs the answer as an answer result for output, and the one question one type question answering unit 109 answers Is obtained as an answer result for output, and if no answer is obtained from either the entity-property type question answering unit 107 or the one-question-type question answering unit 109, the answer is The fact that it was not obtained (for example, “no answer”) is output as an answer result for output.

  When the entity-property type question answering unit 107 receives an entity-property type question, the entity-property type question answering unit 107 answers with reference to the entity-property type answer DB 108. Here, the entity-property question refers to a question that inquires about an attribute of a predetermined object.

  The entity-property type answer DB 108 is a database that manages entity-property type questions and answers. FIG. 3 shows a data example of the entity-property type answer DB 110. As shown in FIG. 3, the entity-property type response DB 110 manages information in which “entity” indicating an object, “property” indicating an attribute, and “value” indicating a value are associated with each other. .

  In the example of FIG. 3, when entity is “Mt. Fuji” and property is “elevation”, the value is “3776 m”. That is, the altitude of Mt. Fuji is 3776 m. Details of processing by the entity-property type question answering unit 107 will be described later.

  When the question-and-answer-type question answering unit 109 accepts a question-and-answer-type question, the answering unit 109 answers by referring to the question-and-answer-type answer DB 110.

  The question-and-answer-type answer DB 110 is a database that manages question-and-answer-type questions and answers. FIG. 4 shows an example of data in the one-answer-one-answer type DB 108. As shown in FIG. 4, the question-and-answer type answer DB 108 has “question” that means a question and “answer” that means an answer to the question. According to the data example shown in FIG. 4, the answer to the question “the highest mountain in the world” is “Everest”.

  The answer output unit 111 outputs the answer to the replacement question or the answer to the entire question acquired by the answer acquiring unit 104. Specifically, the answer output unit 111 transmits an answer to the external device that has transmitted the question text, or outputs the answer to the output device 15 included in the question answering device 100.

  Next, processing in which the answer generation unit 106 generates a question answer will be described. First, the answering process of the entity-property question by the entity-property question answering unit 107 will be described with reference to the flowchart shown in FIG.

  When the entity-property type question answering unit 107 first receives a question sentence (step S1), it executes morphological analysis (step S2). For example, in response to the question “Mt. Fuji is the height”, the entity-property type question answering unit 107 divides the morpheme unit as “Mt. Fuji // height / is”.

  Subsequently, the entity-property type question answering unit 107 extracts the entity. The entity-property type question answering unit 107 extracts using machine learning such as a conditional random field (CRF), for example. The entity-property question answering unit 107 extracts entities based on the part of speech of the morpheme and the relationship between adjacent morphemes.

  In the case of the above question example (“the height of Mt. Fuji”), the entity-property type question answering unit 107 extracts “Mt. Fuji” as an entity (step S3).

  Subsequently, the entity-property type question answering unit 107 determines property (step S4). Specifically, the entity-property type question answering unit 107 determines the property by machine learning such as multi-class SVM. The entity-property type question answering unit 107 determines whether or not the morpheme is property based on the part of speech of the morpheme or the relationship between adjacent morphemes, and identifies the property. When “elevation”, “author”, “birthday”, etc. are defined in advance as property, the entity-property type question answering unit 107 determines that “elevation” is “property” from the above-described question example.

  Subsequently, the entity-property type question answering unit 107 verifies the property (step S5). The entity-property type question answering unit 107 verifies whether or not the determined property is appropriate with reference to a predefined verification word list.

  For example, assume that words such as “elevation”, “height”, and “size” are defined as verification words corresponding to the “elevation” property. In the above question example, since “height” is included, the entity-property type question answering unit 107 determines that there is no problem as a verification result, and validates “elevation” of the property determination result.

  Then, the entity-property type question answering unit 107 searches the entity-property type answer DB 110 using the extracted entity “Mt. Fuji” and the property “elevation” as search keys to obtain an answer “3776m” (step S6). . The entity-property type question answering unit 107 outputs the obtained answer (step S7) and ends the process.

  Next, the question-and-answer type answering process executed by the question-and-answer-type question answering unit 109 will be described. As an example, when a question sentence is input, the question-and-answer type question answering unit 109 removes specific endings (eg, “ha”, “ka”, “do”) that are not related to the keyword. The question sentence from which the ending is removed is used as a query to search the one-answer-one-answer type DB 110 to obtain an answer. For example, when the question “the highest mountain in Japan” is input, the question-and-answer type question answering unit 109 removes the word “ha” from the end of the word, and then executes the query “the highest mountain in Japan”. The question-and-answer-type answer DB 110 is searched to obtain an answer “Mt. Fuji”. However, it is not always possible to obtain an answer. If no answer is obtained, the question-and-answer type question answering unit 109 outputs “no answer” as the answer result.

  Although the answer generation unit 106 has described the case of generating an answer for an entity-property type and one-question-and-answer type question, it may be configured to answer other types of questions. Specifically, a regular expression type question answer may be generated. For example, in the case of a calculation question that can contain various numbers such as “1 + 1 =”, a regular expression pattern (for example, “number + number”, etc.) is matched with the question, and an arbitrary calculation question is extracted. Addition calculation by numbers can be executed.

(Description of overall processing)
Next, the overall processing executed by the question answering apparatus 100 will be described using the flowchart shown in FIG. Here, the whole process refers to a process of receiving a question sentence, extracting a partial question from the question sentence, replacing the portion of the partial question with an answer of the partial question, and outputting the answer of the replaced question.

  First, when the question receiving unit 101 receives a question sentence (step S11), the partial question extracting unit 102 performs morphological analysis on the question sentence, and divides the question sentence into morphemes (step S12).

  For example, when a question sentence “What is the birthday of the author of △△△ ×” is received, the partial question extraction unit 102 determines that “○ △△ ×”, “no”, “author”, “no”, “birthday”, “ha”. ".

  Next, the partial question extraction part 102 extracts a partial question from the whole question (step S13). The partial question extraction unit 102 extracts a partial question using machine learning such as a conditional random field (CRF), for example. That is, the partial question extraction unit 102 defines a learning model that is a model learned so that partial questions can be extracted in advance, and extracts partial questions based on the learning model. For example, in the case of the above-described question sentence, the partial question extraction unit 102 determines that there is a partial question based on “no” as a particle and “author” as a noun often used in the question sentence. Then, the partial question extracting unit 102 extracts “the author of ○ ΔΔ ×” as the partial question in the above example of the question.

  If the partial question extraction unit 102 does not extract a partial question (step S14; NO), the process moves to step S22.

  When the partial question extraction unit 102 extracts a partial question (step S14; YES), the partial question extraction unit 102 verifies the partial question (step S15). Specifically, the partial question extraction unit 102 determines whether or not the partial question includes a word or expression included in the check word list 103a. In the above example, “author of ◯ ΔΔ ×” is a partial question, and therefore includes the word “author” and includes the word of the check word list 103a.

  Further, the partial question extraction unit 102 determines whether a question sentence other than the partial question includes a word or expression included in the check word list 103b. In the above-described example, “birthday” is a portion excluding the partial question, and the word “birthday” included in the check word list 103b is included in the portion. In addition, although the case where the check of the partial question and the check other than the partial question in the question sentence is performed using the two check word lists 103a and 103b of the check word lists 103a and 103b has been described, one check word list is used. It is also possible to check partial questions and check other than partial questions in a question sentence.

  As described above, the partial question extraction unit 102 determines whether or not the partial question is appropriate through the above two checks (that is, determines whether or not the question format has a partial question). In the case of the above-described example, the partial question extraction unit 102 determines that both the appropriate determination using the check word list 103a and the appropriate determination using the check word list 103b are appropriate. S16; YES), move to step S17.

  The partial question extraction unit 102 determines that the partial question is inappropriate when at least one of the appropriate judgment using the check word list 103a and the appropriate judgment using the check word list 103b is inappropriate. (Step S16; NO), the process moves to Step S22.

  The method for verifying the partial question is not limited to the above example. For example, when the partial question includes only the check word list 103a and the partial question includes the words and expressions of the check word list 103a, the partial question extraction unit 102 performs partial verification. You may make it judge that it is appropriate as a question.

  As another partial question verification method, when only the check word list 103b is prepared, and the portion excluding the partial question in the question sentence includes the words and expressions of the check word list 103b, the partial question extraction unit 102 You may make it judge that it is appropriate as a question.

  In step S17, the answer acquisition unit 104 outputs the partial question to the answer generation unit 106 and makes a partial question answer request (step S17).

  When the answer acquisition unit 104 acquires the answer to the partial question (step S18; YES), for example, when the answer “Yamada XX” is acquired for the above-mentioned partial question “the author of XX”, Move to step S19. If the answer acquisition unit 104 has not acquired an answer to the partial question (step S18; NO), the answer acquisition unit 104 moves to step S22.

  In step S19, the question editing unit 105 replaces “the author of ○ ΔΔ ×”, which is the portion of the partial question, with “Yamada ○○”, which is the answer to the partial question, for the question sentence.

  By the way, additional information may be included as an answer to the partial question. For example, it is assumed that an answer “[President] ○ △ × <Line feed code> (as of April 2013)” is obtained in response to the question “Head of □ ○ country”. Here, <new line code> indicates that a new line code is included. In addition, parentheses are included in the above answer as described in “(as of April 2013)”.

  In such a case, the question editing unit 105 may delete the answer part after the line feed code or delete the part in parentheses as the answer to the partial question. That is, the question editing unit 105 sets a part of the obtained answer by removing the additional part (the answer part after the line feed code and the part in parentheses). Thereby, the question edit part 105 can make the essential part of an answer into a partial answer.

  After the question editing unit 105 replaces the portion of the partial question in the question sentence with the answer of the partial question, the answer acquisition unit 104 outputs the replaced question to the answer generation unit 106 and sends an answer request to the answer generation unit 106. To obtain an answer to the question (step S20).

  If the answer acquisition unit 104 does not acquire an answer (step S21; NO), the process moves to step S22. If the answer acquisition unit 104 acquires an answer (step S21; YES), the process moves to step S24.

  In step S22, when a partial question could not be extracted (NO in steps S14 and S16), or when a partial question could be extracted, an answer to the partial question could not be obtained (NO in step S18) ), The answer acquisition unit 104 outputs the question accepted by the question accepting unit 101 to the answer generating unit 106 and makes an answer request (step S22).

  In step S22, when the answer acquisition unit 104 acquires an answer (step S23; YES), the process moves to step S24, and when the answer acquisition unit 104 does not acquire an answer (step S23; NO), the process ends. .

  When an answer is acquired in step S23 or step S21, the answer output unit 111 outputs the answer acquired by the answer acquisition unit 104 (step S24) and ends the process.

(Modification)
In the modified example, processing when there are a plurality of partial question answers will be described. FIG. 7 shows a flowchart of the answer acquisition process in the modification. The configuration of the question answering apparatus 100 is the same as that shown in FIG.

  Steps S31 to S37 are the same as Steps S11 to S18, respectively, and Steps S42 and S43 are the same as Steps S22 and S23, respectively.

  The partial question extraction unit 102 extracts partial questions. When the question sentence is “Year of Yamada XX's representative work”, a partial question “Yamada XX's representative work” is extracted, and the answer acquisition unit 104 acquires an answer to the partial question (step S31— S37).

  When the answer acquisition unit 104 has acquired an answer (step S38; YES), the question editing unit 105 has a plurality of (for example, N) answers to the partial question acquired by the answer acquisition unit 104. The question part is replaced with the answer of each partial question (step S39). That is, the question editing unit 105 generates question sentences after replacement for the number of answers to the partial questions (that is, N).

  The answer acquisition unit 104 outputs the question sentence generated by the question editing unit 105 to the answer generation unit 106 and makes an answer request (step S40). When the question editing unit 105 generates a plurality of question sentences, the answer acquisition unit 104 outputs all the question sentences to the answer generation unit 106.

  If no answer can be acquired (step S41; NO), the answer acquiring unit 104 moves to step S42.

  When the answer acquisition unit 104 acquires one or more answers (step S41; YES), the answer acquisition unit 104 moves to step S44 and outputs the acquired answer. At this time, a part of the obtained answers may be output without outputting all the answers.

  There may be more than one answer to the replacement question. Even in that case, M answers will be output for each replacement question. Therefore, N × M answers are output at the maximum.

  As described above, in the question answering apparatus 100, by allowing a plurality of answers to a partial question, even when an answer cannot be obtained from a question replaced with an answer of a partial question, the answer to another partial question is obtained. If the answer to the replaced question is obtained, the answer can be finally presented.

  Next, the effect of the question answering apparatus 100 of this embodiment will be described. In the question answering apparatus 100, the question receiving unit 101 receives a question sentence, and the partial question extracting unit 102 extracts a partial question that is a part of the question sentence. The answer acquisition unit 104 acquires the answer to the partial question, and the question editing unit 105 replaces the part of the partial question in the question sentence with the answer to the partial question. The answer acquisition unit 104 acquires the answer of the replaced question, and the answer output unit 111 outputs the answer of the replaced question.

  In this case, the question answering apparatus 100 extracts a partial question included in the question sentence, and replaces the partial question portion with an answer to the partial question. And since the question answering apparatus 100 acquires the answer in the replaced question as the answer to the original question, it can accurately answer a question sentence including another question.

  Therefore, the question answering apparatus 100 can accurately obtain an answer even for a complicated question that cannot be obtained by extracting a word in a question sentence and obtaining an answer using the word as a keyword.

  For example, when a user wants to know the date of birth of a certain author, even if the name of the author cannot be specified, a question (such as a representative work ○○ △) that includes information that can identify the author If the author's birthday) is presented, the question answering apparatus 100 can output the answer. That is, the question answering apparatus 100 can output an answer even for a complicated question (a relatively vague question).

  In addition, the question answering apparatus 100 has the check word list 103a, and the partial question extraction unit 102 determines whether or not the partial question is appropriate based on whether or not the word of the check word list 103a is included in the partial question. The answer acquisition unit 104 may acquire the answer to the partial question when the partial question is appropriate.

  In this case, the question answering apparatus 100 uses the check word list 103a to determine whether or not the partial question is appropriate, and thus obtains an answer as the partial question even though it is not appropriate as the partial question. You can avoid that.

  Further, the question answering apparatus 100 has a check word list 103b, and the partial question extraction unit 102 determines whether the partial question is based on whether or not the check word list 103b is included in a portion excluding the partial question of the question sentence. It may be determined whether or not the answer is appropriate, and the answer acquisition unit 104 may acquire the answer to the partial question when the partial question is appropriate.

  Also in this case, the question answering apparatus 100 determines whether the partial question in the question sentence is appropriate using the check word list 103b (whether the question sentence is a question sentence having a partial question or not). Therefore, it is possible to avoid obtaining an answer as a partial question even though it is not appropriate as a partial question.

  In the question answering device 100, when the answer acquiring unit 104 acquires a plurality of answers, the answer acquiring unit 104 acquires an answer to the question sentence in which the part of the partial question in the question sentence is replaced with each of the plurality of answers. The answer output unit 111 may output a predetermined number of answers from the answers acquired by the answer acquisition unit 104.

  In this case, the question answering apparatus 100 can output a representative answer when there are a plurality of answers to the partial question.

  In the question answering device 100, when the answer to the partial question is not obtained, the answer obtaining unit 104 may obtain the answer to the question sentence, and the answer output unit 111 may output the answer to the question sentence. In this case, the question answering apparatus 100 can output the answer of the entire accepted question sentence even when there is no question partially.

  In the question answering apparatus 100, when the answer acquisition unit 104 acquires an answer having an additional part such as parentheses as an answer to the partial question, the answer obtained by deleting the additional part may be used as the answer to the partial question.

  In this case, since the question answering apparatus 100 removes an additional part as an answer such as a bracketed part and replaces the question sentence, the answer can be obtained more accurately.

  In the question answering apparatus 100, when the answer acquiring unit 104 acquires an answer having an additional part such as a sentence after the line feed code as an answer to the partial question, the answer obtained by deleting the answer after the line feed code It may be an answer.

  In this case, since the question answering apparatus 100 replaces the question sentence by removing an additional part such as a sentence after the line feed code, the answer can be obtained more accurately.

(Other embodiments)
In the above-described embodiment, the case where the question answering apparatus 100 includes the answer generation unit 106 has been described. However, another apparatus (for example, a server or the like) may include the answer generation unit. In this case, the question answering apparatus 100 transmits a question sentence to an apparatus including an answer generation unit, and obtains an answer from the apparatus including the answer generation unit.

DESCRIPTION OF SYMBOLS 100 ... Question answering apparatus, 101 ... Question reception part, 102 ... Partial question extraction part, 103 ... Check word list, 104 ... Answer acquisition part, 105 ... Question edit part, 106 ... Answer generation part, 107 ... One question one answer type Question answering unit, 108 ... one question one answer type DB, 109 ... entity-property type question answering unit, 110 ... entity-property type answer DB unit, 111 ... answer output unit.

Claims (6)

A question sentence receiving means for receiving a question sentence;
Partial question extracting means for extracting a partial question that is a part of the question sentence;
A partial question answer obtaining means for obtaining an answer to the partial question;
A question replacement unit that replaces the portion of the partial question in the question sentence with the answer acquired by the partial question answer acquisition unit;
A replacement question answer obtaining means for obtaining an answer to the question sentence replaced by the question replacing means;
Answer output means for outputting an answer to the question sentence ;
A first check word list,
The partial question extraction means determines whether the partial question is appropriate based on whether the partial question includes a word of the first check word list,
The partial question answer acquisition means is a question answering apparatus that acquires an answer to a partial question when the partial question is appropriate . A question sentence receiving means for receiving a question sentence;
Partial question extracting means for extracting a partial question that is a part of the question sentence;
A partial question answer obtaining means for obtaining an answer to the partial question;
A question replacement unit that replaces the portion of the partial question in the question sentence with the answer acquired by the partial question answer acquisition unit;
A replacement question answer obtaining means for obtaining an answer to the question sentence replaced by the question replacing means;
Answer output means for outputting an answer to the question sentence;
A second check word list,
The partial question extraction means determines whether or not the partial question is appropriate based on whether or not the second check word list is included in a portion excluding the partial question of the question sentence,
The partial question answer acquisition means is a question answering apparatus that acquires an answer to a partial question when the partial question is appropriate. A question sentence receiving means for receiving a question sentence;
Partial question extracting means for extracting a partial question that is a part of the question sentence;
A partial question answer obtaining means for obtaining an answer to the partial question;
A question replacement unit that replaces the portion of the partial question in the question sentence with the answer acquired by the partial question answer acquisition unit;
A replacement question answer obtaining means for obtaining an answer to the question sentence replaced by the question replacing means;
Answer output means for outputting an answer to the question sentence,
When the partial question answer acquiring unit acquires a plurality of answers, the replacement question answer acquiring unit acquires an answer to the question sentence in which the portion of the partial question of the question sentence is replaced with each of the plurality of answers,
The answer output means, from among responses that the substitution question answer acquiring unit has acquired, and outputs a predetermined number of respondents, question answering system. A question sentence receiving means for receiving a question sentence;
Partial question extracting means for extracting a partial question that is a part of the question sentence;
A partial question answer obtaining means for obtaining an answer to the partial question;
A question replacement unit that replaces the portion of the partial question in the question sentence with the answer acquired by the partial question answer acquisition unit;
A replacement question answer obtaining means for obtaining an answer to the question sentence replaced by the question replacing means;
Answer output means for outputting an answer to the question sentence;
If the answer of the partial query is not obtained, and a whole question answer acquiring means for acquiring an answer of the question message,
The answer output means outputs the answer the whole question answer acquiring unit has acquired, question answering system. A question sentence receiving means for receiving a question sentence;
Partial question extracting means for extracting a partial question that is a part of the question sentence;
A partial question answer obtaining means for obtaining an answer to the partial question;
A question replacement unit that replaces the portion of the partial question in the question sentence with the answer acquired by the partial question answer acquisition unit;
A replacement question answer obtaining means for obtaining an answer to the question sentence replaced by the question replacing means;
Answer output means for outputting an answer to the question sentence,
The said partial question answer acquisition means is a question response apparatus which makes the answer which deleted the said supplementary part the answer of a partial question, when the reply which has a supplementary part is acquired. A question answering method executed by a question answering device,
A question sentence acceptance step for accepting a question sentence;
A partial question extracting step of extracting a partial question that is a part of the question sentence;
A partial question answer obtaining step for obtaining an answer to the partial question;
A question replacement step of replacing the portion of the partial question in the question sentence with the answer acquired in the partial question answer acquisition step;
A replacement question answer obtaining step for obtaining an answer to the question sentence replaced by the question replacing step;
An answer output step of outputting an answer to the question sentence ;
A first check word list,
In the partial question extraction step, it is determined whether or not the partial question is appropriate based on whether or not the word of the first check word list is included in the partial question.
In the partial question answer acquisition step, a question response method of acquiring an answer to a partial question when the partial question is appropriate .

Images