JP3605735B2 - Natural language semantic analysis processor - Google Patents

Description

[0001]
[Industrial applications]
The present invention relates to a device that sequentially stores information of an input sentence as context information and performs a semantic interpretation, and in particular, a natural language semantic analysis process that performs a semantic interpretation so that context information can be naturally stored in a car navigation interactive process or the like. It concerns the device.
[0002]
[Prior art]
In the past, many studies on dialogue have focused on the intent and topic recognition and use methods (eg, Atsushi Takeshita: "Recognition of topics using interaction structure of dialogue", IPSJ Natural Language Processing Research Group 87-10 (1992)). However, at present, it is hard to say that there is still a stable technology for the context construction technology that is the premise of such recognition.
[0003]
As a problem in context construction, the role of semantic interpretation is to correctly recognize the meaning of the input sentence without depending on the style, organize and integrate it into the context, and extract information necessary for problem solving. The "context" constructed as a result of the sentence interpretation must have a uniform format so that the subsequent problem solving process can be started stably. On the other hand, when looking at actual sentences, a so-called “blurring” often appears in a stylistic style such that a given content is given in one sentence or a plurality of sentences. Even within one sentence, the parse tree that is the input of the semantic interpretation processing has considerable diversity in the types of nodes and branching, so sentences with the same semantic content are May be represented by a different syntactic structure.
[0004]
In this case, if the same contents are allowed to be represented by different structures, it is impossible to stably associate and integrate meanings. In order to integrate sentence contents semantically and construct one context, it is necessary to absorb the structural diversity of stylistic styles and parsing trees. For this purpose, (1) the semantic expression does not have a structure that fluctuates depending on the complexity and the semantic content of the sentence, and (2) the semantic interpretation result of the sentence is expressed in the context of the context. It is necessary to be positioned in a fixed place.
[0005]
In the conventional natural language processing method, the contents of an input sentence are generally arranged in the form of a logical expression, a verb frame, a semantic net, or the like.
[0006]
[Problems to be solved by the invention]
However, for example, a verb frame has an internal structure of a predicate + a surface case (or a deep case), and there are many synonymous expressions having different expression forms due to variations in verbs and case names. For example, even in a sentence where the same departure place represents Tokyo, there is a variety of expressions such as “depart from Tokyo” or “go from Tokyo”. There is a similar problem in the case of a semantic net or the like. These expressions are, in extreme terms, just a formal transformation of the parse tree into another kind of structure with similar complexity, and like the parse tree, have the problem of expression diversity, From the viewpoint of accumulation of words, it was not appropriate as a semantic expression.
[0007]
Therefore, an object of the present invention is, in particular, to provide a natural language semantic analysis processing device that has developed a semantic analysis method capable of naturally storing context information using, for example, a conversation for car navigation as a theme.
[0008]
[Means for Solving the Problems]
In order to achieve the above object, a natural language semantic analysis processing device of the present invention includes a parsing unit that obtains a parse tree of an input sentence,
A semantic analysis unit that performs the following processes a) to e):
a) Analyzed by the parsing unitWasDecompose the substructure tree into a set of branches of [predicate + case particle phrase] for each clause included in the parse tree,
b) the aboveFor "case particle + noun attached to case particle" corresponding to "case particle phrase" of each branch obtained in a)Case Particles and Nouns Attached to Case ParticlesWhenIn between, "attribute noun" and "sai", which indicate the class of "noun attached to case particle", specified by "case particle"WhenTo be converted to [predicate + case particle + attribute noun + "name" + noun attached to case particle],
c) FurtherConverted in b) aboveDecompose "is" into a relative clause structure of "equal to" and convert it into [predicate + case particle + attribute noun + relation pronoun + case particle "ha" + "equal" + case particle "ni" + case particle. Noun]
Invert attribute nouns and relative pronouns to make "equal" the main verb, [predicate + case particle + relation pronoun + attribute noun + case particle "ha" + "equal" + case particle "ni" + noun attached to case particle ]
d)Result converted in c) aboveAfter formalizing as <<<< predicate + case particle + relational pronoun >> + attribute noun> = <noun attached to case particle>]On the left side[<< predicate + case particle + relative pronoun> + attribute noun>]The noun corresponding to the attribute of the phenomenon concept specified by the predicate is replaced with one noun as a parameter, and the right-hand side <noun attached to the case particle> is used as a parameter value, and is stored as context information of a “parameter parameter value” pair.
e) Further, The "parameter parameter value" pair accumulated in the process d) is compared with the "parameter parameter value" pair previously accumulated by the process d), and the attribute of the parameter is determined based on the collation result. The context information determined byWrite processDo and accumulate,
A control unit for controlling the syntactic analysis unit and the semantic analysis unit and controlling the input sentence acquisition and the output sentence output.
[0009]
It further includes a map information operation unit that performs map information operation based on map data, and in the semantic analysis unit, when a noun pointing to an entity on a map or a clause that refers to a movement phenomenon appears, the map information operation unit is Identifying noun referents and simulating the movement process by referring to map data via a computer can complement the flow of phenomena and spatial connections when spatio-temporal information is involved, leading to the problem of handling continuity. This is preferable because it can be partially addressed.
[0010]
The control unit controls input and output of a sentence, and also controls the syntax analysis unit and the semantic analysis unit, a dialogue control unit that stores control rules, and a problem solving rule that stores problem solving rules. It is preferable to have a unit in that a dialog system for car navigation can be constructed.
[0011]
[Action]
According to the present invention, a parse tree analyzed by a parsing unit is divided into clause units by a semantic analyzing unit, input and analyzed for each branch of [predicate + case particle phrase], and a parameter value pair uniquely determined for each phenomenon is determined. By organizing in the form of, the variation of the branch caused by the predicate can be absorbed, the structure can be made uniform, and the context can be accumulated.
[0012]
【Example】
Hereinafter, the present invention will be described in detail with reference to an embodiment shown in the drawings. FIG. 1 is a block diagram showing a configuration of a natural language semantic analysis processing device according to an embodiment of the present invention. FIG. 2 is a diagram showing a concept of a phenomenon of a clause content model by the natural language semantic analysis processing device of the present invention. FIG. 3 is a diagram showing an example of a context model by the natural language semantic analysis processing device of the present invention.
[0013]
First, a basic method for solving the problem of the present invention will be described. Since the basic unit of the proposition level that constitutes a sentence is a clause, it is natural that the meaning of the input sentence is interpreted in units of clauses. In this case, the factors of structural diversity of clause-level expressions are (a) diversity of combinations and numbers of branches that modify predicates and nouns, and (b) diversity of meanings of predicates and nouns. is there. In the case of a parse tree or a semantic expression containing structural diversity of the same type, it is impossible to escape from the structural diversity caused by the above (a) and (b). [Predicate + case particle phrase]], and consider that each branch is interpreted independently of the other branches. As a result, the predicate is interpreted a plurality of times, but this is acceptable because the branches are interpreted individually. Also, in practice, there may be some semantic dependency between the branches. If there is a dependency between the parameters, an additional procedure for referring to or calculating the value of another parameter according to the dependency will be given to the parameter as described later. Details will be described later.
[0014]
However, when the clause is decomposed into branches, the problem of the diversity of the meanings of the predicates and nouns in (b) remains. Therefore, each branch [predicate + caseParticle+ CaseParticleNoun to [<<<Predicate + caseParticle+Related Pronouns>+ Attribute noun>=<CaseParticleThe noun>], And<<<Predicate + caseParticle + Relational Pronoun>+ Attribute noun>Is uniquely determined by the phenomenon, and a noun word that does not depend on the predicate, for example, if “goes from”, is interpreted as “departure place”. This makes it possible to absorb the variation of the branch caused by the predicate and to make the structure uniform.Here, [] indicates the range or unit of the partial syntax tree to be described. <> Represents a unit of a phrase or a clause. In the expression << predicate + case particle + relational pronoun + attribute noun>, the inner <> represents a group of adnominal modifier clauses, and the outer <> represents a group of noun phrases headed by an attribute noun. The left side of = represents the nominative component of "equals", and the right side represents the qualification component.
[0015]
In the present invention, a noun representing this "predicate + case + attribute noun" is called a parameter, and a noun attached to the case is called a parameter value. When there is a dependency between parameters, an additional procedure for referring to or calculating another parameter value according to the dependency is provided for the parameter. That is, according to the present invention, the contents of the node can be represented by a flat structure that is structurally simplified as shown in FIG. Here, a frame corresponding to a predicate surrounding a parameter value group is prepared to represent a unit called a clause, but this frame is not essential as context information. In such a semantic expression format, both the sentence “Departing from Tokyo” and the sentence “Going from Tokyo” can be arranged as “Departure place” = “Tokyo”. In addition, by determining the position of each of these parameters in the context field in advance, it is possible to uniquely determine where in the context information of each branch of the input clause should be written. Become. As described above, (1) the semantic expression does not have a structure that fluctuates depending on the complexity and semantic content of the sentence, and (2) the semantic interpretation result of the sentence is expressed in the context of the context. The condition that the content is positioned at a fixed place for each content is satisfied, and the context information is accumulated by stably writing the content of the clause at a specific position in the context.
[0016]
For example, in the case of the sentence "Turn left at the Ihei intersection, drive southwest along the Numata highway along the Hinoemata river, and go to the innermost Hinoemata hot spring in Aizu in about an hour from Yaso on the Aizu plateau." A context model as shown is formed.
[0017]
In the above, the problem involving spatiotemporal information has been ignored, but in actual sentences, the spatiotemporal section of the event, such as "I want to leave Tokyo, pass Utsunomiya and go to Nikko." May be described fragmentarily, and in integration, it may be necessary to supplement the flow of phenomena and spatial connections between them. Although it is currently difficult to solve such a problem of handling spatiotemporal continuity in general, in the present invention, spatial information has map information holding road, area, facility information, and the like. I will deal with it in part. If a sentence related to movement or presence is input, a simulation or referent search is performed on the map to attempt integration. In the case of the above-mentioned example sentence "I want to leave Tokyo and go to Utsunomiya and go to Nikko", a route from Tokyo to Nikko is obtained by simulation, and based on that, it is integrated as one movement phenomenon.
[0018]
Next, an embodiment in which the natural language semantic analysis processing device of the present invention is configured as a navigation interactive system will be described.
[0019]
In FIG. 1, a navigation interactive system according to the present embodiment includes an input device 1 for inputting a sentence, an output device 2 for outputting a sentence and outputting map information, and a language for performing syntax analysis and semantic analysis of the input sentence. It comprises a processing unit 10, a map management unit 20 for managing map information, and a control unit 30 for controlling the input device 1, the output device 2, the language processing unit 10 and the map management unit 20 as a whole.
[0020]
The input device 1 is a device that takes in a sentence under the control of the control unit 30, and is, for example, a part that inputs a sentence by voice recognition and performs a morphological analysis process. The output device 2 is a part that outputs a sentence or map information under the control of the control unit 30. For example, the output device 2 is a part that outputs a sentence by voice synthesis or that displays a sentence or a map on a display. . The language processing unit 10 analyzes a syntax of a sentence input by the input device 1 based on a grammar analysis dictionary 12 for syntax analysis, and obtains a parse tree. Is analyzed for each branch of the predicate + case particle phrase based on the semantic information dictionary 14 and arranged in the form of parameter value pairs, and a noun pointing to an entity on a map or a clause referring to a movement phenomenon appears. In reference to the map information of the map management unit 20, identification of referents of nouns and simulation of the movement process are performed, and the contents of each parameter value pair are collated one after another in a context field, or context information defined by parameters is used. The semantic analysis unit 13 performs a process of writing (overwriting) in the above-mentioned fixed location. The map management unit 20 includes a map information operation unit 21 that performs map information operation based on map data 22 that stores map information. The control unit 30 controls the input device 1, the output device 2, the syntax analysis unit 11, the semantic analysis unit 13, and the map information operation unit 21, and controls a control rule unit 32 for activating a control rule. An interactive control unit 31 fetches an input sentence by interactive processing based on a problem solving rule unit 33 for activating a problem solving rule and controls output of an output sentence and map information.
[0021]
Next, a description will be given of an example in which the natural language semantic analysis processing device according to the present embodiment absorbs variations in branches caused by predicates and makes the structure uniform.
[0022]
FIG. 4 is a diagram showing a syntax tree (dependency structure) of an example sentence 1 of the first embodiment, FIG. 5 is a diagram showing a syntax tree (dependency structure) of an example sentence 2 of the same first embodiment, and FIG. 6 is a first embodiment. FIG. 9 is a diagram showing an example of organizing the syntax tree of Example sentence 1 into a parameter value pair.
[0023]
In the first embodiment, the example sentence 1 is equivalent to "go from Tokyo to Utsunomiya on the Tohoku Expressway." The example sentence 2 is "depart from Tokyo and use the Tohoku Expressway to go to Utsunomiya." I have. As shown in FIGS. 4 and 5, the dependency structure trees of these two example sentences 1 and 2 are different from each other due to differences in predicates (units of reference) used and cases of the predicates. Therefore, it is difficult to directly determine the semantic equivalence of two sentences from this data. In order to absorb these differences in the surface structure, an equivalent transformation is performed for each predicate + case particle phrase as shown in FIG. 6, and the contents of each branch of the dependency structure tree are arranged in a uniform structure. For such an equivalent deformation, reference is made, for example, to Takagi and Ito: "Processing of Natural Language", Maruzen, 1987.
[0024]
Hereinafter, as a first embodiment shown in the drawings, the following example sentences are used for each branch.[Predicate + caseParticle+ CaseParticleThe noun]To[<<Predicate + caseParticle+Related Pronouns>+ Attribute noun>=<CaseParticleThe noun>]A method of performing a synonymous transformation to absorb variations of branches caused by predicates and making the structure uniform will be described.
[0025]
Now, the input sentence “Go from Tokyo to Utsunomiya on the Tohoku Expressway” in the example sentence 1 is analyzed by the parsing unit 11 into a dependency structure tree shown in FIG. Can be The semantic analysis unit 13 analyzes this structural tree for each branch of [predicate + case particle phrase] and arranges the sentence “go from Tokyo” into the form of “departure place = Tokyo” as shown in FIG. . As for the nounization by interpretation of the noun phrase "equal (is)", the validity is guaranteed in the above-mentioned document, so the dependency structure is not actually transformed, and the meaning of the predicate is not considered. For each feature, the corresponding relationship between the consecutive modification components (case, attribute, etc.) that can be received and the noun to be obtained by interpretation is stored in the semantic information dictionary 14, and the conversion process is performed mechanically. FIG. 8 is an example of a noun obtained as an interpretation result by such a combined modifier component and a predicate semantic feature. If the predicate is "(is)", which indicates an assertion, it indicates the speaker's judgment on an arbitrary phenomenon or the class of an entity, parameter value, etc., and therefore, as in other predicates that directly refer to the phenomenon, Parameters cannot be extracted directly from. When "(is)" represents a judgment on the parameter value of a phenomenon, the noun attached to the nominative indicates the concept to be mapped to the parameter of the phenomenon, and the noun attached to the nominative indicates the value. At this time, a general noun is often used as the nominative noun, and is often not directly associated with the parameter name. For example, the "charge is 700 yen." "Charge" and the corresponding parameter "local transfer required charge" prepared as the parameter of "movement" do not directly match. Therefore, in this case, a correspondence is separately defined.
[0026]
The other case particle phrases are semantically analyzed by the semantic analysis unit 13 in the same manner, and “go to Utsunomiya” → [destination = Utsunomiya]
"Go by Tohoku Expressway" → [Road used = Tohoku Expressway]
It is arranged with. At this time, if a noun indicating an entity on a map such as Tokyo, Utsunomiya, or the Tohoku Expressway or a clause referring to a movement phenomenon appears, the map data 22 is referred to by the map information operation unit 21 of the map management unit 20. Identification of referents in nouns and movement processessimulationIs performed.
[0027]
By arranging in this manner, the interpretation of the example sentence 1 "Go from Tokyo to Utsunomiya on the Tohoku Expressway" can be summarized in a very flat form as shown in FIG.
[0028]
Next, in the case of the example sentence 2 "Leave Tokyo and use the Tohoku Expressway to go to Utsunomiya.", The structural tree shown in FIG. 5 is arranged for each section in a format similar to that of the example sentence 1. The results are summarized as shown in FIG. 9, and the obtained interpretation results of each section are further organized. In this case, since the structure is uniform and the position of the parameter is uniquely determined in the context field for each phenomenon, it can easily be placed in a fixed place on the context information regardless of the dependency structure and semantic content. The writing process can be performed (overwriting). As a result, the interpretation result of the example sentence 1 is exactly the same as the interpretation result of the example sentence 2. As described above, in the embodiment of the present invention, the processing of writing information in a predetermined place on a uniform data structure is performed (overwritten), so that the meaning of the entire sentence can be changed in a form appropriate for the context. It can accumulate.
[0029]
As described above, in the dialogue system of the present invention, the contents of a clause can be represented by a flat structure that is structurally simplified as much as possible, as shown in FIG. Both "do" and "go from Tokyo" can be arranged as "departure point" = "Tokyo". In addition, by determining in advance all the positions of these parameters in the context field, it is possible to uniquely determine where in the context information of each branch of the input clause should be written. Has been described above.
[0030]
In the example sentences 1 and 2 of the first embodiment described above, the problem involving spatiotemporal information has been ignored, but in actual sentences, for example, "I want to leave Tokyo and pass Utsunomiya to Nikko." As mentioned above, the spatio-temporal cross section of an event may be described only in a fragmentary manner, and in integration, it may be necessary to complement the flow of phenomena and the spatial connection between them. It is currently difficult to solve handling problems in general. In the interactive system device of the present invention, the spatial information is provided with the map information operation unit 21 of the map management unit 20 and the map data 22 to have the map information holding the road / area / facility information. Partially addressed. That is, when a sentence relating to movement or presence is input, integration is attempted by, for example, performing a simulation or referent search on the map using the map information operation unit 21. For example, in the case of the above-mentioned example sentence "I want to leave Tokyo and pass through Utsunomiya to Nikko.", A route from Tokyo to Nikko is obtained by simulation, and based on that, it is integrated as one movement phenomenon. Therefore, the semantic analysis is accumulated and used.
[0031]
Although the example of the interpretation of the example sentence 2 "Leaving Tokyo and going to Utsunomiya using the Tohoku Expressway" has been described above, if the semantic analysis / semantic expression method of the present invention is used, what style is described in one sentence? It is possible to obtain a stable semantic interpretation result without being affected by the fluctuation of the surface structure of the sentence such as described in (1). Further, in this interpretation model, by determining all parameters existing in one phenomenon in advance, information is always stored in a fixed position according to its meaning (what parameters are described). Determines the storage position), so that the semantic analysis result of the input sentence can be easily collated or written (overwritten) on the same type of context information described in the same format, and It is possible to smoothly collect data necessary for processing even in some problem solving processing. However, for example, when a route is described, a description is given in a form of repeating the description of the route fragments and describing the entire route as a whole. In this case, the parameter used to represent the root fragment will appear repeatedly multiple times, and the position of the parameter cannot be uniquely identified in the context. In such a case, (1) If any other parameter value obtained from the clause in the interpretation of the input sentence is already positioned in the context, the corresponding parameter in the root fragment containing it is included. Set the value of the interpretation result to (2) substitute the value for the corresponding parameter in the route fragment that was just talked about (the parameter value was set), or (3) the route that was talked just before If the value of the corresponding parameter in the fragment is already filled, processing such as substituting the value for the corresponding parameter in the next route fragment is performed.
[0032]
Further, for the same reason, there may be an ambiguity as to whether the repeated modification component of a certain predicate refers to a parameter that defines the entire route or a parameter that defines a fragment of the route. In such a case, for example, at the time of interpreting the input sentence, it is assumed that a parameter that specifies a fragment of the route is referred to for the time being, and then compared with the context information, and as necessary while referring to the map information. Processing such as re-interpreting the parameters as defining the entire route is required.
[0033]
For example, "Go from Tokyo to Utsunomiya."
"Go to Nikko from Utsunomiya."
When interpreting the two sentences, for example, the following processing is performed. First, based on the interpretation of the first sentence, the local departure place = the Tokyo local arrival place = Utsunomiya. When comparing with the context at this stage, since nothing is written in the context, the above parameters and parameter values are directly substituted into the context as the description of the first route fragment, and this defines the parameters of the entire route. In consideration of the possibility that the route is present, the departure place = Tokyo destination = Utsunomiya is also substituted for the parameters “departure place” and “destination” that define the entire route. Next, the local sentence = Utsunomiya local destination = Nikko is obtained by the analysis of the second sentence. Matching this with the context, the second sentence is the second successive route because the fragment of the route has already been recorded and its local destination and the value of the local departure from the second sentence match. And assigns a value to a parameter in the corresponding context. At the same time, for the parameter that defines the entire route, the value is updated to “destination = Nikko” considering that “Nikko” may be the destination of the entire route. Thereafter, when it is determined that the input of the sentence describing the route has been completed, the destination = Nikko is determined. Also, for example, as the third sentence, "It takes three hours from Tokyo to Nikko."
Is input, the local departure place = the Tokyo local destination = the Nikko local movement required time = 3 hours as the semantic analysis result. When this is compared with the context, it is found that "Tokyo" and "Nikko" are the "departure point" and "destination" of the entire route, so "local travel time" is interpreted as "travel time" of the entire route It is determined that the time is to be changed, and the time required for movement = 3 hours is substituted for the “time required for movement” parameter in the context. By the way, if "takes 2 hours" is input after the first sentence, since the local departure place and the local destination are not specified, it is determined that the route fragment mentioned immediately before is described. "2 hours" is substituted for the "local movement required time" in the route fragment of the above. In the case where the input sentence is a question sentence, the semantic analysis is performed in the same manner as in the case of the declarative sentence, while the ambiguity of the parameter is resolved by collating with the context information. The processing written above is not performed (overwritten) (the pronoun or abbreviation component in the question sentence may be adapted and complemented by the context information).
[0034]
Next, based on the above-described concept, a result of the following example sentence 3 as the second embodiment will be described.
"Leave Tokyo at 8am. First, head to Utsunomiya on the Tohoku Expressway. You will reach Utsunomiya in about 100 minutes. From Utsunomiya to Nikko, use the Nikko Utsunomiya Road. The fare is 770 yen. Distance is about 25km, about 20 minutes It will arrive at Nikko around 10 o'clock. "
Is input, a context as shown in FIGS. 10 to 12 is constructed in the same manner as in the first embodiment. However, as described above, the “fee” and the “distance” are respectively associated with the “local travel required fee” and the “local travel distance” by a predefined table. The “stroke” is interpreted as “move”, and each union modifying component is converted into a “move” parameter corresponding to the “move” predicate. “About 20 minutes” is set to the value of the “local movement required time” parameter using the unit noun “minute” as a clue.
[0035]
As a result of the above, for example, even when a question sentence “How long does it take to go from Tokyo to Nikko?” Can be answered with a very simple algorithm. That is, the content of this question sentence is collated with the context information constructed by the previously input sentence, and finally, the interpretation result shown in FIG. 13 is obtained. By adopting this form, it can be seen that the “time required for movement” in the context is an object of the inquiry, and the procedure of the problem solving rule unit 33 for obtaining the required time required for movement is started. In order to obtain the required travel time, the “local travel required time” may be summed. That is, in the case of the above example, the required required travel time is obtained as "120 minutes" by adding the required local travel time between Tokyo and Utsunomiya for 100 minutes and the required local travel time between Utsunomiya and Nikko for 20 minutes. Therefore, it becomes possible to answer the question based on it.
[0036]
As described above, since the storage location of the information is determined by its meaning, it is possible to stably identify an appropriate procedure necessary for solving the problem and collect data necessary for solving the problem. .
[0037]
As described above, the dialogue system apparatus of the present invention can perform a dialogue process as shown in FIG. 14, for example. In other words, in response to the user's question "I want to go to the sightseeing spot A." When the user instructs "L intersection" in response to this instruction, the system answers "Yes, I understand." Next, when the user asks "I want to depart at 6:00 am and arrive at 5:00 pm", the system answers "Yes, I understand." Then, when the user asks "Tell me the road.", The system responds, "There is one road. First, proceed northeast on National Road No. 1 and go to the E intersection. Next, National Road No. 1 To the north and go to the H-intersection. Finally, follow National Route 1 to the northeast. " In this interactive system device, a user inputs by voice from the input device 1, and the system is output by voice from the output device 2 by voice synthesis or displayed on a display.
[0038]
According to the above framework, it is a commercially available sightseeing drive guidebook as experimental data (1) "Nikko / Nasu / Shiobara, go and drive and stay overnight, Obunsha (1993)" and (2) "Blue Guide Drive" Nikko / Kinugawa / Nasu "and" 382 "related to" movement "appearing in" Jitsugyo no Nihonsha "(1994). About 70% of the sentences were composed of an appropriate context model using this framework. I confirmed that I can do it.
[0039]
Although the preferred embodiment of the present invention has been described, various modifications and changes can be made without departing from the spirit of the present invention.
[0040]
【The invention's effect】
As described above, in the natural language semantic analysis processing apparatus of the present invention, the parse tree analyzed by the syntactic analysis unit is divided and input into clause units by the semantic analysis unit, and the [predicate + case particle phrase] branch is input. By analyzing each parameter and arranging them in the form of parameter value pairs, variations in branches due to predicates can be absorbed, the structure can be made uniform, and context can be accumulated.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of a natural language semantic analysis processing device of the present invention.
FIG. 2 is a diagram showing a concept of a phenomenon of a model of a clause content by the natural language semantic analysis processing device of the present invention.
FIG. 3 is a diagram illustrating an example of a context model by the natural language semantic analysis processing device of the present invention.
FIG. 4 is a diagram showing a syntax tree of an example sentence 1 of the first embodiment by the natural language semantic analysis processing device of the present invention.
FIG. 5 is a diagram showing a syntax tree of an example sentence 2 of the first embodiment by the natural language semantic analysis processing device of the present invention.
FIG. 6 is a diagram showing the arrangement of the syntax tree of the example sentence 1 of the first embodiment in the form of parameter value pairs by the natural language semantic analysis processing device of the present invention.
FIG. 7 is a diagram showing a semantic analysis result and context information of an example sentence 1 of the first embodiment by the natural language semantic analysis processing device of the present invention.
FIG. 8 is a diagram showing an example of a noun obtained as an interpretation result by a combined modifier component and a predicate semantic feature by the natural language semantic analysis processing device of the present invention.
FIG. 9 is a diagram showing a semantic analysis result and context information of an example sentence 2 of the first embodiment by the natural language semantic analysis processing device of the present invention.
FIG. 10 is a diagram illustrating a semantic analysis result and context information of an example sentence 3 of the second embodiment by the natural language semantic analysis processing device of the present invention.
FIG. 11 is a diagram illustrating a semantic analysis result and context information of an example sentence 3 of the second embodiment by the natural language semantic analysis processing device of the present invention.
FIG. 12 is a diagram illustrating a semantic analysis result and context information of an example sentence 3 of the second embodiment by the natural language semantic analysis processing device of the present invention.
FIG. 13 is a diagram showing a result of semantic analysis of a question sentence by the natural language semantic analysis processing device of the present invention.
FIG. 14 is a diagram illustrating an example of an interactive process performed by the natural language semantic analysis processing device of the present invention.
[Explanation of symbols]
1 Input device
2 Output device
10 Language processing unit
11 Syntax analyzer
12 grammar analysis dictionary
13 Semantic analysis part
14 semantic information dictionary
20 Map Management Department
21 Map information operation unit
22 Map data
30 control unit
31 Dialogue control unit
32 Control Rule Section
33 Problem Solving Rule Division

Claims (3)

A parser for obtaining a parse tree of an input sentence;
The next a) to e units sections included in the semantic analysis unit and a) parse tree is analyzed by the syntax analysis unit performs a process of), the partial structure tree of the predicate + case particle Clause Decompose into a set of branches,
b) corresponds to the "case particles Clause" of each branch obtained by said a) for the "nouns attached to case particles + case particle" and "case particle" between "nouns attached to case particle", is specified by the "case particle", get to represent a class of "noun to get the case particle" and "attribute noun" to complement the "that" [predicate + case particle + attribute noun + "of" + case particle Noun],
c) Further, the word "" converted in the above b) is decomposed into a relational clause structure of "equal to", and the relation [predicate + case particle + attribute noun + relation pronoun + case particle "ha" + "equal" + Case particle "ni" + noun attached to case particle]
Invert attribute nouns and relative pronouns to make "equal" the main verb, [predicate + case particle + relation pronoun + attribute noun + case particle "ha" + "equal" + case particle "ni" + noun attached to case particle ]
d) The result converted in c) is formalized as <<<< predicate + case particle + relational pronoun >> + attribute noun >> = <noun attached to case particle>], and then the left-hand side is <<<< predicate + The case particle + relative pronoun> + attribute noun>] is replaced with a single noun corresponding to the attribute of the phenomenon concept specified by the predicate, and is used as a parameter. The right-hand side <noun attached to the case particle> is used as a parameter value. Store as context information of "parameter parameter value" pair,
e) Further , the "parameter parameter value" pair accumulated in the process d) is compared with the "parameter parameter value" pair previously accumulated by the process d), and based on the result of the collation. Writes and accumulates the context information determined by the attribute of the parameter to a certain position ,
A natural language semantic analysis device, comprising: a control unit that controls the syntax analysis unit and the semantic analysis unit and controls the input sentence capture and the output sentence output. It further includes a map information operation unit that performs map information operation based on the map data. 2. The natural language semantic analysis processing apparatus according to claim 1, wherein the reference of the noun is identified and the movement process is simulated by referring to the map data via the interface. The control unit controls input and output of a sentence, and also controls the syntax analysis unit and the semantic analysis unit. The natural language semantic analysis processing device according to claim 1 or 2, further comprising a unit.

Images