JPH07129596A - Natural language processor - Google Patents

Abstract

PURPOSE:To easily obtain an analyzed result desired by a user by accurately analyzing various sentences by means of a small scale system without making the capacity of a dictionary larger. CONSTITUTION:A dictionary retrieving and dictionary control part 4 extracts a candidate work group through the use of only a standard independent work dictionary 2a and a standard attached word dictionary 2b in the case of a usual text sentence and a morpheme analysis part 5 executes morpheme analysis. In the case of the text sentence of a specified field, an extension independent word dictionary 2c and an extension attached work dictionary 2d which suit with a specified field are added in addition to it. In the case of a different field, the extension independent word dictionary 2c and the extension attached word dictionary 2d are replaced and in the case when there are plural dictionaries to be retrieved, a priority order is given to each dictionary so as to give a more precise judging reference at the time of selecting a word in morpheme analysis processing. In this constitution, a dictionary 2 can be constituted to the absolute minimum.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language processing device such as a machine translation system, a Japanese text-to-speech synthesis system, or an electronic computer that handles mixed kanji and kana sentences.

[0002]

2. Description of the Related Art Conventionally, a machine translation system that analyzes a sequence of input characters when a text sentence is input and then translates it, and a voice synthesis system that synthesizes a voice by synthesizing parameters according to a predetermined rule are known. ing. The voice synthesis system is used, for example, in a welfare application in which a visually impaired person listens to and understands a text sentence without a helper, and also in a situation where it is difficult to visually obtain character information. In such a natural language processing apparatus, it is expected that the input text sentence will be analyzed as accurately as possible.

As described above, in the morphological analysis of a mixed kanji / kana sentence in the natural language processing device, a dictionary search is performed by using the independent word dictionary, the auxiliary word dictionary, and the user dictionary, and the optimum word is selected from the search results. is there. In order to perform this morphological analysis accurately, the number of vocabularies recorded in the dictionary and the method of selecting the optimum word from a plurality of candidate words are important.

The larger the number of vocabularies recorded in the dictionary, the better the result of morphological analysis. Therefore, a natural language processing apparatus for analyzing newspapers or the like has a dictionary with a scale of 500,000 words. Further, when the sentence to be analyzed is limited to a specific field, only the vocabulary of the field is increased and a method using a minimum dictionary is adopted. On the other hand, as for the method of selecting from a plurality of words, in addition to the rules for connecting words, the importance of each vocabulary recorded in the dictionary and the determination criteria such as the longest matching principle are used.

[0005]

By the way, in the above-mentioned conventional natural language processing apparatus, the dictionary becomes large,
It is difficult to realize in a small-scale system other than a large-scale dedicated system for analyzing newspapers. Further, the method of increasing the vocabulary of a specific field has a problem that it is not suitable for analyzing sentences in other fields because it becomes a dedicated system for the specific field. Further, in a natural language processing device using a plurality of dictionaries, the number of words to be selected increases, so that an appropriate selection criterion is required, but there is a problem that the selection criterion is difficult to set.

Therefore, the present invention provides a natural language processing apparatus capable of accurately analyzing various sentences in a small-scale system without increasing the dictionary capacity and easily obtaining the analysis result intended by the user. It is intended to be provided.

[0007]

In order to achieve the above object, the natural language processing apparatus according to the invention of claim 1 stores a general vocabulary in a natural language processing apparatus for performing morphological analysis of a kanji / kana mixed sentence. Equipped with a standard dictionary and a plurality of detachable extended dictionaries that store only vocabulary peculiar to a specific field, and a dictionary reading means for reading the peculiar vocabulary and a search target character string for the kanji / kana mixed sentence A combination of the standard dictionary and the plurality of extended dictionaries,
A morpheme is searched for in the combined dictionary, a search and extraction unit that extracts a candidate word group of the search target character string, the Kanji / Kana mixed sentence, and the candidate word group that is the result of the search and extraction unit. And a morphological analysis means for performing analysis.

In the natural language processing apparatus according to the present invention, the standard dictionary includes a standard independent word dictionary in which general independent words are stored and a standard auxiliary word in which general auxiliary words are stored. It is characterized in that the plurality of extended dictionaries are constituted by a plurality of extended independent word dictionaries in which only independent words unique to a specific field are stored.

Further, in the natural language processing apparatus according to the third aspect of the present invention, there is provided a priority order setting means for setting the priority order in the standard dictionary and the extended dictionary, and the search / extract means is a searched candidate word group. The priority set by the priority setting means is given to each of the above, and the morphological analysis means assigns a predetermined priority to a word having a high priority among the candidate word groups extracted by the search and extraction means. Giving,
It is characterized by performing morphological analysis.

Further, in the natural language processing apparatus according to the present invention, the plurality of extended dictionaries are composed of a storage medium such as an IC card, a magnetic disk, a magneto-optical disk or an optical disk.

[0011]

In the present invention, in the case of a general kanji / kana mixed sentence, only the standard independent word dictionary and the standard adjunct dictionary are used, and in the case of a kanji / kana mixed sentence of a specific field, in addition to that, it is suitable for the specific field. The extended independent word dictionary and the extended auxiliary word dictionary are added. In the case of different fields, at least the extended independent word dictionary is replaced. Therefore, the dictionary can be configured to the minimum required.

When there are a plurality of dictionaries to be searched,
By giving a priority to each dictionary, a more accurate determination criterion can be given when selecting a word in the morphological analysis process. In addition, the standard dictionary is composed of a standard independent word dictionary in which general independent words are stored and a standard auxiliary word dictionary in which general auxiliary words are stored. A more accurate judgment criterion can be given by constructing from a plurality of extended independent word dictionaries in which only unique independent words are stored. Further, the plurality of extended dictionaries may be composed of a storage medium such as an IC card, a magnetic disk, a magneto-optical disk, an optical disk.

[0013]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a speech synthesis system to which the natural language processing device of the present invention is applied. In the figure, 1
Is an input unit for inputting a mixture of Kanji and Kana, such as a keyboard, OCR (optical character reading device),
It consists of a magnetic disk. The dictionary 2 is, for example, I
A standard independent word dictionary 2a, which stores a general independent word, and a standard auxiliary word dictionary 2b, which stores general auxiliary words (indicating words other than non-independent words), including a storage device such as a C memory and a magnetic disk. And 2 are the minimum configurations, and specific fields (for example,
It has a plurality of extended independent-word dictionaries 2c in which only independent words in the computer field and the like, and an extended auxiliary-word dictionary 2d in which only auxiliary words in a specific field are recorded. Each dictionary stores the spelling of a word that is a reference of a morpheme, its accompanying information (for example, reading, part-of-speech information, accent, etc.).

The sentence analysis unit 3 comprises a dictionary search / dictionary management unit 4, a morpheme analysis unit 5 and a phonetic symbol generation unit 6.
First, the dictionary search / dictionary management unit 4 searches for words included in the kanji / kana mixture sentence input from the input unit 1 according to the spelling of the words stored in the dictionary 2 or the attached information thereof,
The search target dictionaries are switched and the priority order is changed.

Here, the above-mentioned dictionary 2 and dictionary search /
The configuration of the dictionary management unit 4 will be described with reference to FIG.
As described above, the dictionary 2 is composed of the standard independent word dictionary 2a, a plurality of extended independent word dictionaries 2c1, 2c2, ..., And the standard auxiliary word dictionary 2b, and a plurality of extended auxiliary word dictionaries 2d1, 2d2 ,. Also, the dictionary search / dictionary management unit 4
Dictionary search unit 4a, dictionary management unit 4b, independent word dictionary management table 4
It is composed of c and the attached word dictionary management table 4d. The standard independent word dictionary 2a, a plurality of extended independent word dictionaries 2c1,
2c2, ... Are managed based on the independent word dictionary management table 4c, and the standard auxiliary word dictionary 2b and the plurality of extended auxiliary word dictionaries 2d1 and 2d2 are managed based on the auxiliary word dictionary management table 4d.

The independent word dictionary management table 4c and the auxiliary word dictionary management table 4d manage the currently used dictionary and the priority order of the dictionary, and the dictionary management unit 4b follows the instruction of the user and becomes independent. Each of the word dictionary management table 4c and the attached word dictionary management table 4d is changed. For example, in the case of an instruction to use a new dictionary, it is added to the independent word dictionary management table 4c or the auxiliary word dictionary management table 4d. Further, when an instruction to stop using a certain dictionary is given, the dictionary is deleted from the independent word dictionary management table 4c or the adjunct word dictionary management table 4d. Further, when changing the priority order, the priority order of the independent word dictionary management table 4c or the auxiliary word dictionary management table 4d is changed. The dictionary search unit 4a, when the dictionary search target character string is given,
The morpheme analysis unit 5 shown in FIG. 1 is searched by searching the dictionaries registered in the independent word dictionary management table 4c and the adjunct word dictionary management table 4d, prioritizing the searched dictionaries in the dictionary search results.
Supply to.

Next, the morphological analysis unit 5 includes the kanji / kana mixed sentence input from the input unit 1 and the dictionary search / dictionary management unit 4.
The morphemes are analyzed based on the word group retrieved in step S1, and converted into kana character strings, and then decomposed into words and phrases. That is, in Japanese, words are not separated into words like English, so for example, a word such as "US industry" is "US / industry / world" or "US / domestic / industry". Can be divided into two types. Therefore, the morphological analysis unit 5 refers to the dictionary 2 and uses the continuity of words and the statistical property to input a text input into a word,
The bunsetsu is decomposed into bunsetsu, and the boundaries of words and bunsetsus are detected.

The phonetic symbol generator 6 also generates a phonetic symbol string based on the data from the morpheme analyzer 5. Next, the voice unit storage unit 7 is composed of a storage device such as an IC memory and a magnetic disk, and stores voice units. The voice unit is composed of waveform data used when generating a synthetic sound represented by each CV unit. The voice unit data used for this waveform synthesis has the following configuration.

Regarding the voiced part of the voice unit data, an impulse corresponding to one pitch and a unit response waveform extracted using the complex cepstrum analysis in the voiced part of the real voice are set as one set, and this set is set as one voice. The unit data is stored by a required pitch, and the unvoiced part of the voice unit data is formed by cutting out the waveform of the unvoiced part of the actual voice and storing it. Therefore, when the voice unit data is in CV units, when the consonant part C of one voice unit CV is an unvoiced consonant, a plurality of sets of a cutout waveform of an unvoiced portion and an impulse and a unit response waveform form one voice. When the unit data is configured, and when the consonant part C of one voice unit CV is a voiced consonant, one voice unit data is configured by only a plurality of pairs of impulses and unit response waveforms.

Next, the voice synthesis rule unit 8 generates a voice synthesized waveform pattern and a pitch pattern from the data obtained from the phonetic symbol generation unit 6, the information of the voice unit storage unit 7, the phoneme / prosodic rule and the like. obtain. That is, the voice synthesis rule unit 8
The voice unit data loaded from the voice unit storage unit 7 is synthesized in the order according to the text input to obtain a synthesized voice waveform in which there is no intonation. Also, the voice synthesis rule unit 8
The text input is divided into appropriate lengths based on a predetermined prosody rule to detect cuts (ie, consisting of pauses). Thus, for example, as shown in FIG.
When the text "Give a beautiful flower from Mr. Yamada" is input as the text input (Fig. 3 (A)), the text input is "pretty", "hana", "from Yamada san", After being decomposed into "I got it", a pose is detected between "Hana" and "From Yamada-san" (Fig. 3 (B)).

Further, the speech synthesis rule unit 8 detects the accent of each phrase based on the prosody rule and the basic accent of each word. That is, the accent of a Japanese phrase alone can be expressed sensuously in two levels, high and low, in units of kana characters (hereinafter referred to as mora). At this time, the accent position of the phrase can be distinguished according to the content of the phrase. For example, edges, chopsticks, and bridges are 2-mora words, and can be categorized into 0 type without accent, 1 type with accent position in the first mora, and 2 type with accent position in 2nd mora. it can. Thus, in this embodiment, the speech synthesis rule unit 7 classifies each phrase of the text input into 1 type, 2 type, 0 type, and 4 type (FIG. 3 (C)). Detect a pose.

Further, the voice synthesis rule unit 8 generates a basic pitch pattern representing the intonation of the entire text input, based on the accent and pause detection results. That is, in Japanese, the accent of a bunsetsu can be sensuously expressed in two levels, while the actual intonation is characterized by gradually decreasing from the position of the accent (FIG. 3).
(D)).

Furthermore, in Japanese, when the bunsetsu becomes one sentence in a row, the intonation gradually decreases from one pose to another (FIG. 3 (E)).
Therefore, the speech synthesis rule unit 8 generates a parameter representing the intonation of the entire text input for each mora based on the characteristics of the Japanese language, and then the intonation changes smoothly as if a human uttered. Then, parameters are set by interpolation between mora. Thus, the voice synthesis rule unit 8
Synthesizes the parameters of each mora and the interpolated parameters in the order according to the text input (hereinafter referred to as the pitch pattern), and thus the pitch pattern representing the intonation of the voice read out from the text input (FIG. 3 (F)). Can be obtained.

Next, the voice synthesizing unit 9 has a voice synthesizing rule unit 8
A voice waveform is synthesized based on the data (synthesis waveform data and pitch pattern) obtained from the above. The waveform synthesizing process is as follows. In the voiced part of the synthetic speech, impulses in the synthetic waveform data are arranged based on a pitch pattern, and a unit response waveform corresponding to each of the arranged impulses is superimposed on each impulse.

In the unvoiced part of the synthetic voice, the cut-out waveform in the synthetic waveform data is used as it is as the waveform of the desired synthetic voice. As a result, it is possible to obtain a synthetic sound in which the intonation changes according to the change in the pitch pattern. Therefore, since the impulse is used for the sound source information in the synthesized sound, even if the pitch period of the synthesized sound expands or contracts, it has almost no effect on the sound source information, and even if the pitch pattern changes greatly, the spectrum It is possible to obtain a high-quality arbitrary synthesized voice that is similar to human voice without causing distortion in the envelope. Next, the output unit 10 outputs the voice synthesis unit 9
The voice waveform synthesized in step (3) is output to, for example, a speaker, a magnetic disk, or the like.

Next, the structure of the natural language processing apparatus in this embodiment will be described with reference to FIG. FIG. 4 is a schematic diagram showing the configuration of the natural language processing device. In the figure, the natural language processing device 11 internally includes a standard independent word dictionary 2a, a standard auxiliary word dictionary 2b, and an independent word dictionary management table 4
c and the auxiliary word dictionary management table 4d, and the extended independent word dictionaries 2c1, 2c2, ... And the extended auxiliary word dictionary 2d.
1, 2d2, ... Are read by the extended dictionary reading device 12.

Here, the extended independent word dictionary 2c1 is added,
An example of the procedure for updating the independent word dictionary management table 4c will be described below. 1) The natural language processing device 11 registers the built-in standard independent word dictionary 2a and standard auxiliary word dictionary 2b in the independent word dictionary management table 4c and the auxiliary word dictionary management table 4d, respectively. At this point, the standard independent word dictionary 2a and the standard auxiliary word dictionary 2b are
Both have a priority of "1". 2) The extended independent word dictionary 2c1 is used as the extended dictionary reading device 12
To insert. 3) Detecting dictionary insertion into the extended dictionary reading device 12,
Read the extended independent word dictionary 2c1. 4) It is determined that it is the extended independent word dictionary 2c1, and the independent word dictionary management table 4c is updated.

When the extended dictionary has a mechanism giving priority over the standard dictionary, the priority of the standard independent word dictionary 2a is changed to "2", and the priority of the extended independent word dictionary 2c1 is "2". 1 ”is additionally registered in the independent word dictionary management table 4c. Also, in the case of a natural language processing device having a plurality of dictionary reading devices built-in, each dictionary reading device is given a priority order, and the priority order of the extended dictionary is made to follow the priority order given to the dictionary reading device. Good. In this case, a mechanism for switching the priority of each dictionary reading device may be provided.

For example, as shown in FIG. 3, in the natural language processing device, the priority of the upper dictionary reading device 12a is "1", and the lower dictionary reading device 12b is "2".
It is assumed that "3" is given to the built-in dictionary reading device (not shown). When the extended dictionary is not inserted in the extended dictionary reading devices 12a and 12b, there is no dictionary other than the standard dictionary, so the priority of the standard dictionary is set to "1". on the other hand,
When the extended dictionary is inserted in the lower dictionary reading device 12b, the priority of the extended dictionary is set to "1" and the priority of the standard dictionary is set to "2".

Next, the operation of the natural language processing apparatus of this embodiment will be described with reference to FIG. FIG. 5 is a flowchart showing an example of the morpheme analysis processing of the natural language processing device 11 in this embodiment. In the above-mentioned configuration, first, the extended independent word dictionaries 2c1, 2c2, ..., The extended auxiliary word dictionaries 2d1, 2d1, 2 are depending on the field of the input text sentence.
The extended dictionary reading device 12 reads d2, .... Next, by the dictionary search / dictionary management unit 4, the built-in standard independent word dictionary 2a, standard auxiliary word dictionary 2b, extended independent word dictionary 2c1, 2c2, ..., And extended auxiliary word dictionary 2d.
1, 2d2, ... Priority is set according to the procedure described above.
It is set in the independent word dictionary management table 4c and the auxiliary word dictionary management table 4d.

Next, the text sentence input from the input unit 1 is set in the independent word dictionary management table 4c or the auxiliary word dictionary management table 4d by the dictionary search unit / dictionary management unit 4 of the sentence analysis unit 3. According to the priority order, the dictionary 2 (standard independent word dictionary 2a, standard auxiliary word dictionary 2b, extended independent word dictionary 2c
1, 2c2, ..., and extended auxiliary word dictionaries 2d1, 2d
2, ...) are searched in accordance with the spelling of the words stored in (2, ...) And the attached information, and as a result, the word group is extracted.

Next, the morpheme analysis unit 5 performs morpheme analysis on the kanji / kana mixed sentence input from the input unit 1 and the word group searched by the dictionary search unit 4a, and converts it into a kana character string. After that, it is decomposed into words and phrases. Here, the morphological analysis for obtaining the morphological analysis result from the dictionary search result will be described with reference to FIG. 7. FIG. 7 is a flowchart showing an example of a morpheme analysis procedure of this embodiment for obtaining a morpheme analysis result from a dictionary search result.

First, in step S1, a gazing point for morpheme connection determination is set. When this morphological analysis process is executed for the first time, the gazing point is placed at the beginning of the sentence. Next, in step S2, a morpheme candidate group at the beginning of a sentence is acquired and a hypothesis group is generated. Then, in step S3, the gazing point is shifted by one, and it is determined whether or not the gazing point is the end of a sentence. If the gazing point is not at the end of the sentence in step S3, the process proceeds to step S4.

In step S4, it is determined whether or not there is a hypothesis that needs to be connected to the morpheme starting from the gazing point. Then, the determination result in step S4 is "N
If “O”, that is, if there is no hypothesis that needs to be connected to the morpheme starting from the gazing point, step S3
Returning to the above, while searching for a hypothesis that requires connection with a morpheme starting from the gazing point while shifting the gazing point forward by one, and the judgment result in step S4 is "YES", that is, the morpheme starting from the gazing point. If there is a hypothesis that needs to be connected (the gaze point at this time is called a connection point), the process proceeds to step S5, and a morpheme group starting from the gaze point is acquired.

Next, in step S6, it is determined whether or not each hypothesis whose gazing point is a connection point can be connected to the morpheme group acquired in step S5. If there is a connectable morpheme, the process proceeds to step S7, and the hypothesis is updated by the morpheme. If there are multiple connectable morphemes, a new hypothesis is generated. On the other hand, if there is no connectable morpheme in step S6, the process proceeds to step S8 and the hypothesis is deleted.

Hereinafter, the above-mentioned processing is performed for all applicable hypotheses. Then, in step S3, when the gazing point comes to the end of the sentence, the analysis ends, and the process proceeds to step S9 to output the analysis result. If it is necessary to narrow down the hypotheses because there are multiple hypotheses, step S
During the arbitrary processing of 3 to S8, narrowing down is performed by applying various norms such as the longest matching principle.

Next, in the phonetic symbol generator 6 shown in FIG. 1, a pronunciation is given to each morpheme which is the above-mentioned analysis result. That is, the reading and accent types of each morpheme are converted into phonetic symbols. In this way, the sentence analysis unit 3
The text input is analyzed based on a predetermined dictionary 2, converted into a kana character string, decomposed into words and phrases, and then a basic accent is detected for each word, and these are synthesized into a speech synthesis rule unit 8
Output to.

The detection results of the words and the boundaries of the clauses and the basic accents are processed by the speech synthesis rule section 8 in accordance with a predetermined phonological rule, and synthetic waveform data representing a speech in which the text input is read aloud without inflection is generated. To be done. Further, the detection result of the word and phrase boundaries and the basic accent is processed by the speech synthesis rule unit 8 in accordance with a predetermined prosody rule, and a pitch pattern representing the intonation of the entire text input is generated. The pitch pattern is output to the voice synthesizing unit 9 together with the synthetic waveform data, and after the synthetic sound is generated based on the pitch pattern and the synthetic waveform data, the output unit 10 produces the sound by the speaker. Alternatively, the composite waveform data may be stored in the magnetic disk.

According to the above-mentioned configuration, in the case of a general text sentence, only the standard independent word dictionary and the standard auxiliary word dictionary are
In the case of a text sentence of a specific field, in addition to that, an extended independent word dictionary and an extended auxiliary word dictionary suitable for the specific field are added. The extended auxiliary word dictionary may not be needed. In the case of different fields, the extended independent word dictionary and the extended auxiliary word dictionary (the extended auxiliary word dictionary may not be necessary) are replaced. In this way, the dictionary can be made to have the minimum necessary configuration.

If there are a plurality of dictionaries to be searched,
As a matter of course, each vocabulary in each dictionary is given a priority for each dictionary, so that a more accurate determination criterion can be given when selecting a word in the morphological analysis process. Further, by providing an extended auxiliary word dictionary in addition to the standard auxiliary word dictionary, it is possible to perform accurate morphological analysis even for a phrase peculiar to a specific field.

[0041]

According to the present invention, a standard dictionary in which general words are stored and a plurality of extended dictionaries in which words in a specific field are stored are prepared, and a combination method of the standard dictionary and the extended dictionary is analyzed. It is possible to analyze various sentences with a relatively small-scale system by changing it according to. Also, when using multiple dictionaries, give priority to each dictionary, and update the priority of each vocabulary obtained by dictionary search based on that priority and the priority given to each vocabulary, and then The use of for the morphological analysis has an advantage that the analysis result intended by the user can be easily obtained.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of a natural language processing apparatus that is an embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a dictionary search / dictionary management unit and a dictionary in the embodiment.

FIG. 3 is a schematic diagram showing a procedure of generating a basic pitch pattern in the embodiment.

FIG. 4 is a perspective view showing an external configuration of a natural language processing device in the embodiment.

FIG. 5 is a flowchart showing an example of a morphological analysis process of the Japanese language processing apparatus in the embodiment.

[Explanation of symbols]

1 input unit 2 dictionary 2a standard independent word dictionary (standard dictionary) 2b standard adjunct word dictionary (standard dictionary) 2c extended independent word dictionary (extended dictionary) 2d extended adjunct word dictionary (extended dictionary) 3 sentence analysis unit 4 dictionary search / dictionary Management unit (search extraction unit, priority setting unit) 5 Morphological analysis unit (Morphological analysis unit) 6 Phonetic symbol generation unit 7 Speech unit storage unit 8 Speech synthesis rule unit 9 Speech synthesis unit 10 Output unit 12 Extended dictionary reading device (dictionary (Reading means)

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Office reference number FI technical display location G10L 5/02 J

Claims (4)

[Claims] 1. A natural language processing device for performing morphological analysis of kanji and kana mixed sentences, a standard dictionary storing general vocabulary, and a plurality of detachable extension dictionaries storing only vocabulary peculiar to a specific field. , A dictionary reading means for reading the specific vocabulary, and a combination of the standard dictionary and the plurality of extended dictionaries in accordance with the search target character string of the kanji / kana mixed sentence, and searching the combined dictionary And a morphological analysis unit that performs morphological analysis on the candidate word group of the search target character string, the Kanji / Kana mixed sentence, and the candidate word group that is the result of the search and extraction unit. A natural language processing device comprising: 2. The standard dictionary includes a standard independent word dictionary in which general independent words are stored and a standard auxiliary word dictionary in which general auxiliary words are stored, and the plurality of extended dictionaries are 2. The natural language processing device according to claim 1, wherein the natural language processing device comprises at least a plurality of extended independent word dictionaries in which only independent words peculiar to a specific field are stored. 3. A priority order setting means for setting a priority order to the standard dictionary and the extended dictionary, wherein the search and extraction means is set by the priority order setting means for each of the searched candidate word groups. A morphological analysis is performed by assigning a priority, and the morpheme analysis unit gives a predetermined priority to a word having a high priority order among the candidate word group extracted by the search and extraction unit. Item 1. The natural language processing device according to item 1. 4. The natural language processing apparatus according to claim 1, wherein the plurality of extended dictionaries are storage media such as IC cards, magnetic disks, magneto-optical disks, and optical disks.