JPH01233619A - Natural language interactive processor - Google Patents

Abstract

PURPOSE:To interrupt interaction by providing a discourse plan recognizing means, an interaction plan recognizing means, a domain plan recognizing means and interaction state control means. CONSTITUTION:A discourse plan recognizing means 3 to recognize the general change of a subject, an interaction plan recognizing means 2 to recognize the give-and-take of the interaction, a domain plan recognizing means 4 to recognize the structure of the subject of the interaction and an interaction state control means 5 to control the state of the present interaction by a means indicating the subject to which respective speakers being participated to the interaction pay attention and a means to execute the state of the give-and-take of the interaction are provided. Thus, the interaction forming the filling up structure, the interaction involving the utterance of confirmation and chiming-in, the interaction in which the subject to which an interacter pays attention shifts, and moreover the interaction involving a corresponding expression and abbreviation can be interpreted.

Description

[Detailed Description of the Invention] [Industrial Application Field] This invention relates to a natural language dialogue processing device□, for example,
The present invention relates to a natural language dialogue processing device that is used in database searches and question answering systems using natural language, and processes natural language pronounced in a dialogue format.

[Prior Art] Various methods have been proposed and are being put into practical use as methods for processing dialogue using natural language. One method is “A Plan Recognition”.
Model for Subdialogue
s in Conversations”
by Litman and Al
len.

Cognitive 5science vol.
.

11 (1987). In this literature, the problem domain (
There are two plans: a domain plan that is determined depending on the domain (hereinafter referred to as the domain) and a discourse plan that is determined independently of the problem area, and a plan stack that is common to the two speakers and a plan that centrally manages the entire plan. The control unit attempts to recognize dialogue. A domain plan is, for example, to travel, you first buy a ticket, go to the train, and board the train. The knowledge that to buy a ticket is to go to the ticket office, pay money, and receive the ticket. A discourse plan is the knowledge of dialogue to continue a current topic, introduce a new topic, or identify parameters.

[Problem to be solved by the invention] However, in the conventional method described above, there is no way to manage dialogue exchange, so A: "Do you want to go see a movie?" B: "What kind of movie is it?" Air This is a movie titled ○O's original work. "B:
``Where?'' A: rXX movie theater'' B: ``Then I'll go.'' What is the embedded dialogue and Air deadline?'' B: ``It's December 25th.'' Air Okay.'' There was a problem in that it was not possible to interpret utterances expressing confirmation, such as "I understand."

In addition, in the conventional method, there is only one plan stack, so A (1): rPlease provide your address and name.'' B (
2): rOsakafu... Ward OO" (3) "
Can I bring my wife with me?'' A (4): Yes, that's fine.'' (5) There is also the problem of not being able to interpret dialogues where the interlocutor's focus shifts from the topic, such as ``Your name, please.'' there were. In other words, (1) above
In response to A's question, B thought that he had answered all of the questions with the utterance in (2), and moved on to the next topic in (3). However, A has not yet asked for his name, so after answering B's question (3) in (4), he asks for his name again in (5). If this cannot be interpreted, it cannot be interpreted that the name in (5) refers not to the wife's name but to B's name.

Therefore, the main purpose of this invention is to solve the problem of dialogues that have an embedded structure as seen in the above-mentioned dialogues, dialogues that include confirmation and concurrence utterances, dialogues in which the topic of attention of the interlocutor shifts, and furthermore, anaphoric expressions, An object of the present invention is to provide a natural language dialogue processing device capable of interpreting dialogue including omissions.

[Means for Solving the Problems] The present invention is a natural language dialogue processing device that recognizes and processes the content of dialogue performed in natural language, which receives as input the results of analyzing the syntax and meaning of natural language, representative form converting means for converting the input into a representative form expressing the same content as the input; discourse plan recognition means for receiving the output of the representative form converting means as input and recognizing global changes in the topic; an interrag shop run recognition means that receives the output of the form conversion means as an input and recognizes the interaction of the dialogue; a domain plan recognition means that receives the output of the representative form conversion means as an input and recognizes the structure of the topic of the dialogue; Based on the structure of the dialogue topic recognized by the plan recognition means, the current The system includes a dialogue situation management means for managing the dialogue situation, a plan control means for comprehensively managing the discourse plan recognition means, the interlag shop run recognition means, the domain plan recognition means, and the dialogue situation management means.

[Operation] The natural language dialogue processing device according to the present invention converts the result of analyzing the syntax and meaning of natural language into a representative form expressing the same content as the result, and based on the representative form,
Recognize dialogue. By representing the status of dialogue exchanges, it is possible to interpret dialogues that have embedded structures, and dialogues that include utterances of confirmation and agreement. Furthermore, by indicating the topic to which each speaker participating in the dialogue is currently paying attention, it is possible to interpret a dialogue in which the topic of attention of the speakers shifts. Furthermore, by recognizing the structure of the topic of dialogue, it is possible to interpret dialogue that includes anaphoric expressions and omissions.

[Embodiments of the invention] FIG. 1 is a schematic block diagram of an embodiment of the present invention.

First, the configuration of an embodiment of the present invention will be described with reference to FIG. The input information is the result of parsing the syntax and meaning of natural language. In other words, the input information includes information on crew words that indicate a change in topic, such as "By the way,""By the way,""By the way," and "A while ago," or information that there are no crew words, requests for approval/disapproval, and requests for explanation. , information on the type of utterance such as proposal, request, confirmation, answer to affirmative/no request, acceptance, refusal, reason, etc., information on who the speaker and listener are, information on the predicate that is the center of the utterance and its case elements. Contains.

The input information is given to the representative form converting section 1, and is converted into a representative form so that the predicate that is the center of the utterance and its case elements can be matched with the information described in the recognition section. this is,
This is because natural language can express one propositional content using multiple predicates and their case elements. For example, "I would like to participate in the conference,""I would like to register for the conference," and "I would like to participate in the conference" each represent the same content.

It is assumed that matching here is successful if there is no contradiction between the contents of the two pieces of information. For example, ((predicate send)(
Operationally A) (Beneficiary B) (Target Registration Form)) and
((Predicate Send) (Operation A) (Beneficiary B) (Target Participation Fee)) Since the value of "target" is different, matching will fail, but ((Operation A) (Beneficiary B) (Mode Hurry) Since there is no contradiction with )), matching is successful.

The result of the conversion by the representative form conversion section 1 is given to the interlag shop run recognition section 2. The interlag shop run recognition unit 2 recognizes the types of utterances, including affirmative/negative requests, explanation requests, and proposals.

If the request belongs to a "request" such as a request, an appropriate domain plan is selected via the discourse plan recognition unit 3. Specifically, a predicate that is the center of the input utterance and its case elements are converted into a representative form, and a predicate described in the domain plan and its case elements that are successfully matched are selected.

The discourse plan recognition unit 3 recognizes general changes in topics. The selected domain plan information and the information that a “request” was uttered are sent to the dialogue status management unit 5.
It is given to a means of expressing the situation of dialogue exchange provided in . If the type of utterance belongs to a "response" such as an answer to an affirmative/no request, acceptance, or refusal, the dialogue status management unit 5 searches for a corresponding "request" utterance from the means representing the dialogue exchange status. The utterance of the request is erased from the means representing the interaction. If the type of utterance is "confirmation", the utterance is considered to be a confirmation of the domain plan that was just completed, etc., depending on the situation. If the type of utterance is "other" such as reason, the dialogue status management unit 5 registers the utterance in the dialogue status management unit 5. The discourse plan recognition unit 3 obtains an appropriate domain plan from the domain plan recognition unit 4 based on the crew word information. The domain plan recognition unit 4 recognizes the structure of conversation topics. Interlag shop run recognition section 2. Discourse plan recognition unit 3. The plan control unit 6 performs overall management and generalization, such as correspondence between the arguments of each plan by the domain plan recognition unit 4 and the dialogue status management unit 5.

2 to 5A and 5B are diagrams showing specific examples of natural language dialogue processing according to an embodiment of the present invention,
In particular, FIG. 2 shows an example of input information, FIG. 3 shows an example of a domain plan, FIG. 4 shows an example of a discourse plan, and FIGS. 5A and 5B show an example of an interlag shop run. It shows.

First, input information will be explained with reference to FIG.

As mentioned above, the input information is the result of parsing the syntax and meaning of natural language, and includes dialogue such as yes/no requests, explanation requests, proposals, requests, confirmations, answers to yes/no requests, acceptance, rejection, reasons, etc. Type information, information on crew words that indicate a change in topic such as "By the way", "By the way", "A while ago", or information that there are no crew words, information on who the speaker and listener are, It contains information on the predicate and its case elements, which are the center of the utterance.For example, if speaker A utters to speaker B, "Please send me the registration form,"
The input is ((Utterance-Type request) (C1u
e-Word None) (Speaker A) (H
earer B)(Object((Type A
ction) (Predicate) (Obj
ect registration form))).

As shown in Figure 3, the domain plan includes HEADER slot, DECOMPO3IT ION slot, C
0N5IST-OF slot, PRED-STRUCT
Contains URE slot etc. The HEADER slot indicates the topic and
1) indicates a sequence of subtopics to be executed to realize the topic of the HEADER slot, and each subtopic has a temporal order relationship.

C0N5I5T-OF) indicates subtopics that constitute the topic of the HEADER slot, and each subtopic has no temporal order relationship.

PRED-STRUCTURE slot is HEADER
It shows the predicate that expresses the topic of the slot and the representative forms of its case elements. For example, to complete the registration procedure (REGI
5TRATION-PROCEDURE), obtain the registration form (GET-REGI STRATION-
FORM), return the registration form (RETEURN-R
EG I STRATION-FORM), pay the registration fee (PAY-REG I 5TRATION-FEE
). Also, the topic of returning the registration form (RETURN-REG I STRATI
ON-FORM) consists of a deadline topic (DEADLINE) and a return topic (RETURN-THE-FORM) for that C0N5 I 5T-OF slot.

The discourse plan shown in Figure 4 is used to recognize the overall flow of the conversation, and uses crew words such as ``by the way'', ``later'', 2 and ``just a while ago'' to recognize the direction of the flow of the topic. A domain plan that matches the input central predicate and its case elements (hereinafter referred to as predicate & cases) is selected from an appropriate search range according to the class of the crew word.

The interlag shop run shown in FIGS. 5A and 5B recognizes the exchange of utterances necessary for cooperative and natural "dialogue." Dialogue pairs such as question-answer, request-accept/reject, etc. are recognized in this plan.

This interlag shop run can be divided into an INFORMATION PLAN, which represents information exchange, and a CONFORMATION PLAN, which represents information confirmation. I NFORMAT I 0N-PL
AN consists of a dialogue pair such as a question and an answer (
GET2 VALUE-UNIT, ASK-VALUE,
INFORM-VALUE, etc.).

A GET-VALUE-UN IT plan describes a plan for obtaining the value of something.

For example, if speaker A wants to obtain deadline information from speaker B, A: "When is the deadline?" B: "December 25th"
The deadline is December 25th.Speaker A can ask speaker B and speaker B can answer, such as "It's the month of the month."

” (DE
corresponding to COMPO8ITION2). ASK-VA
The LUE plan is a plan that asks about the value of something, and corresponds to utterances such as "When is the deadline?" and "How much is the participation fee?" INFORM-VALUE
A plan is a plan that conveys the value of something to the other party, such as ``The deadline is December 25th.'' ``The participation fee is 10 yen.''
It corresponds to utterances such as ``It's 00 yen.''

FIG. 6 is a flow diagram for explaining the operation of one embodiment of the present invention.

Next, with reference to FIGS. 1 to 6, a specific operation of an embodiment of the present invention will be described.

The example shown in FIG. 6 shows an example of a conversation between two speakers A and B. First, initialization is performed in step (abbreviated as SP in the figure) SPI. Current plan pointer A indicates the domain plan that speaker A is currently paying attention to, and current plan pointer B indicates the domain plan that speaker B is currently paying attention to. As the initial values of these current plan pointers A and B,
Contains the first domain plan. Interlag Shop Runway List is a "request".

There are only ``request'' utterances in dialogue pairs formed by ``response'' utterances, and those for which ``response'' has not yet been uttered are retained. Here, empty is entered as the initial value. The related utterance list is "request".

Utterances other than "response" and "confirmation" are retained, but empty is set as the initial value. These current plan pointers A and B, the interlag shop run wait list, and the related utterance list are provided in the dialog situation management section 5 shown in FIG.

In STAGE SP2, the result of syntactic and semantic analysis is set as a variable u in memory.
If utterance is empty in step SP3, the process ends. In step SP4, utterance is converted into a representative form. This can be realized, for example, by using a correspondence table between assumed inputs and representative forms. This process is performed in the representative form converting section 1 shown in FIG.

In step SP5, the type of utterance of utterance is checked, and if it belongs to a "request" such as an affirmative/no request, an explanation request, a proposal, or a request, the crew word of utterance is checked in step SP6. When the crew word is “and” or “then”,
If it belongs to rNEXTJ, such as when there is no crew word, in step SP7, utteranC is selected from among the topics next to the current topic indicated by the current plan pointer.
You can find domain plans that match e's predicate & cases. Then, move the current plan pointer to the position of the found domain plan.

At this time, if the speaker of utterance is A, the current plan pointer A is used, and if the speaker is B, the current plan pointer B is used. Crew word is “
rFORW like "By the way" or "By the way"
If it belongs to ARDJ, in step SP8, predicate & cas from the topics after the current topic.
You can find a domain plan that matches es. Then, move the current plan pointer to the position of the found domain plan.

The crew word is rBACKWARD like “a while ago”
If it belongs to J, in step SP9, a domain plan that matches predicate & cases is found from the topics that have appeared so far. Then, move the current plan pointer to the position of the found domain plan.

In step 5P10, this "request" utterance is registered in the interlag shop run wait list.

If the type of utterance checked in step SP5 above belongs to a "response" such as an answer to an affirmative/no request, acceptance, or rejection, then in step 5P11, this "response" utterance is selected from the interlag shop run wait list. Dialogue pairs are created by selecting corresponding "request" utterances. If the type of utterance is "confirm", it is determined in step 5P12 whether or not there is a "confirm" utterance in the domain plan completed immediately before, and if there is not, this utterance is completed immediately before in step 5P13. If so, in step 5PI4, this utterance is set as an utterance to confirm the domain plan at the level above the domain plan completed immediately before. These operations in steps SP6 to SF3 are performed by the discourse plan recognition section 3 shown in FIG.

If the type of utterance checked in step SP5 is "other" such as reason, this utterance is registered in the related utterance list in step 5P15.

Steps SP5, 5PII to 5P15 are performed by the interlag shop run recognition unit 2 shown in FIG. 11, and steps 5PIO, 5PII.

After processing 5P13 and 5P14, step 5P
16-5P18, it is checked whether the related utterance list is empty. If the related utterance list is not empty, in the case of a "request" utterance, in step 5P19, the contents of the related utterance list are set to the related utterances of the previous domain plan, and the related utterance list is emptied.

When the process of step 5P1B is performed for "response" and "confirmation", in step 5P20, the related utterance list is emptied with the contents of the related utterance list as related utterances of the corresponding domain plan.

If you performed the process in step 5P14 with "Confirm",
In step 5P21, the contents of the related utterance list are set as related utterances of a domain plan one level higher than the corresponding domain plan, and the related utterance list is emptied.

In the above example, the type of utterance and the type of crew word are uniquely determined, but if there is a possibility of multiple types, processing will proceed assuming that it is one of them, and a contradiction will occur. Sometimes the process is performed by going back and selecting the next possibility.

In addition, although the above-mentioned embodiment showed the case of a conversation between two people, the present invention may be applied to a conversation of 3 Å or more.

In that case, it is sufficient to prepare as many current plan pointers as there are speakers.

The natural language dialogue processing apparatus shown in the above-described embodiments can be applied to general systems in which humans and computers or humans exchange information via computers. in particular,
An example of a system is shown below.

(1) Searching databases using natural language Just as humans exchange information through dialogue, computer systems and humans use databases while omitting topics and content that has already been stated and mutually understood. A system for searching can be created.

(2) Question answering system using natural language (1) above
In the same way, a computer system and a human being can make a yes/no judgment and present missing information while omitting dialogue and content that has already been stated and mutually agreed upon.

A system for exchanging information with a computer regarding advice, reasoning, etc. can be created.

(3) Dialogue translation using natural language A system in which when humans exchange information in different languages via a computer, the computer responsible for the translation process accurately recognizes the omitted parts in the conversation and performs the correct translation. Can be done.

[Effects of the Invention] As described above, according to the present invention, there is a discourse plan recognition means for recognizing global changes in topics, an interlag shop run recognition means for recognizing exchanges in dialogue, and a structure of conversation topics. a domain plan recognition means for recognizing the
By providing a means for indicating the topic to which each speaker participating in the dialogue is currently paying attention, and a dialogue situation management means for managing the current dialogue situation by means of controlling the dialogue exchange situation, Dialogues with embedded structures, dialogues that include confirmation and concurrence utterances, dialogues in which the topic of attention of the interlocutor shifts, and even anaphoric expressions.

Can interpret dialogue that includes omissions.

[Brief explanation of the drawing]

FIG. 1 is a schematic diagram of an embodiment of the present invention. FIG. 2 is a diagram showing an example of input information. FIG. 3 is a diagram showing an example of a domain plan. FIG. 4 is a diagram showing an example of a discourse plan. FIGS. 5A and 5B are diagrams showing an example of an interlag shop run. FIG. 6 is a flow diagram for explaining the operation of one embodiment of the present invention. In the figure, 1 is a representative form converter, 2 is an interlag shop run recognition unit, 3 is a discourse plan recognition unit, 4 is a domain plan recognition unit, 5 is a dialogue situation management unit, and 6 is a plan control unit. Patent applicant A.T.R. Co., Ltd. 3) IEADER: IIEGIsTRA
TION-PROCEDtlRE(speaker, h
earer, forDECOHPO5ITIOH:
GεT-REGISTIIATION-FORM(RE
TtlRN-REGISTRATION-F○PAY-
REGISTRATION-FEE(sHEADER:
RETLIRN-REGISTRATI
O)t-FOCONSIST-OF: DE
ADLINE (seeker, hearRE DINGURN
-THE-FORM(spaakerPRED-5choR
UCTIIRE: ((Type
Action) (Predicate 0kuru
)(＾gent 5peaker)(Be
naficativa hearer) (Object
t form) (Time
data) (DTime 6adline
OW) IER: 5peaker,
hearar, Iorm) INFORM-VALUE (spy, spl. DECOMPO5IT101+2: IIIFORM
-VALUECsp2. spy. Effectsl: Attach-
IntaraEfrecLsZ : At
touch-1nterauttarance2.
Variable (object, value),
domain-plan) Procedural Amendment (1) July 8, 1988

Claims (1)

[Scope of Claims] A natural language dialogue processing device that recognizes and processes the content of a dialogue conducted in a natural language, which receives as input the result of analyzing the syntax and meaning of the natural language, and uses the input as its input. representative form converting means for converting into a representative form expressing the same content as the representative form converting means; discourse plan recognition means for receiving the output of the representative form converting means as input and recognizing global changes in the topic; and output of the representative form converting means. Interrag shop run recognition means which receives as an input and recognizes the interaction of the dialogue; domain plan recognition means which receives the output of the representative form conversion means as an input and recognizes the structure of the topic of the dialogue; Based on the topic structure of the dialogue, the current dialogue situation is determined by a means of pointing to the topic to which each speaker participating in the dialogue is currently paying attention, and a means of expressing the dialogue exchange situation. and a plan control means for comprehensively managing the discourse plan recognition means, the interlag shop run recognition means, the domain plan recognition means, and the dialogue situation management means. Dialogue processing device.