JPH10187699A - Document processor and its method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To estimate and automatically register a meaning classification of an unknown word by the document processor which converts an inputted character string while properly referring to a dictionary containing meaning classifications of words. SOLUTION: A program module 252 has program modules 261 to 264 as subroutines. Here, the program module 261 extracts an unknown word and its postpositional particle of Japanese and the module 262 extract a verb having modification relation with the extracted known word; and the module 263 takes a verb valence pattern corresponding to the verb extracted from a verb valence dictionary (dictionary containing meaning classifications of nouns corresponding to respective verbs and their postpositional particles) and estimates a meaning classification of the unknown word from the verb valence pattern matching the postpositional particle of the unknown word and the module 264 registers the meaning classification of the unknown word.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document processing apparatus and method, and more particularly to a document processing apparatus and method for creating a sentence mixed with kanji and kana by a kana-kanji conversion mechanism.

[0002]

2. Description of the Related Art Conventionally, a document processing apparatus such as a Japanese word processor generally inputs a sentence mixed with kanji and kana using a kana-kanji conversion mechanism. The kana-kanji conversion mechanism is a mechanism for converting a reading sequence input by referring to a kana-kanji dictionary into kanji. This Kana-Kanji dictionary is a dictionary in which part-of-speech information is described for each word, and the Kana-Kanji conversion mechanism analyzes the reading sequence to create possible phrase candidates, combines them to determine conversion candidates, Posted in likely order. Then, the operator selects a desired candidate from the posted conversion candidates.

[0003] For example, if the reading sequence "I am using a cat" is input, the kana-kanji conversion mechanism is "Japanese",
"Cotton", "I", "other magazines", "other magazines", "sleeping",
"Cat,""cat,""child,""child,""buying,""keeping,""winning,""selfish,
The phrase candidates such as “I stay” and “Enter” are created, and conversion candidates of “I am buying a cat” and “I have a cat” are presented in this order. in this case,
Buy is generally more frequent than keep,
"I have a cat" is rarely presented as the first candidate.

[0004] To increase the conversion efficiency, a conversion method using verb valency has been proposed. This method captures the dependency relationship between a noun's semantic classification and a verb in a sentence as a valency pattern (a pattern of the relationship between a verb and a word having a predetermined semantic classification) and converts the reading sequence into kana-kanji characters Sometimes, examine the dependency relationship of the words included in the reading sequence, compare it with the valence pattern, and if there is a matching valence pattern, preferentially present conversion candidates according to the valence pattern It is a technique to do.

[0005] For example, if the valency pattern of "buying <object>" and "keeping <animal>" is registered in the verb valence dictionary, an input of "cats" is
“Keep a cat” is presented as a first candidate from the relationship “cat = <animal>”.

[0006]

However, in the above-mentioned conventional example, the semantic classification is unknown for unknown words that are not registered in the dictionary (for example, in the above example, “cat = <<
Animal> ”is unknown) and thus cannot be compared with the valency pattern. Therefore,
When an unknown word is included, the kana-kanji conversion mechanism cannot present an appropriate conversion candidate as the first candidate. Therefore, in order to make the conversion candidate appropriate, it is necessary for the operator to register the semantic classification of unknown words in addition to reading, writing, and part of speech.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and it is an object of the present invention to provide a document processing apparatus and method for estimating the semantic classification of unknown words and automatically registering them.

[0008]

In order to solve the above-mentioned problems, a document processing apparatus according to the present invention is a document processing apparatus for converting an input character string while appropriately referring to a dictionary containing meaning classifications of words. An unknown word extracting means for extracting an unknown word whose semantic classification is unknown from the converted character string; a dependency word extracting means for extracting a dependency word having a dependency relationship with the extracted unknown word; A semantic classification estimating means for estimating the semantic classification of the unknown word from the relationship between the particle and the dependency word, and an unknown word registering means for registering the estimated semantic classification in the dictionary together with the unknown word. Features.

According to a preferred embodiment of the present invention, for example, the dependency word is a verb, and the semantic classification estimating means determines a semantic classification of at least one word that may have a dependency relationship with each verb. A valence dictionary means for storing, for each verb, a particle suffixed to the word having the semantic classification, and referring to the valence dictionary means, estimating the semantic classification of the unknown word from the particles of the unknown word; It is desirable to do.

A document processing apparatus according to the present invention operates based on a program on a memory medium, and converts an input character string while appropriately referring to a dictionary containing meaning classifications of words. The memory medium includes a procedure code of an unknown word extracting step for extracting an unknown word whose semantic classification is unknown from the converted character string, and a dependency for extracting a dependency word having a dependency relation with the extracted unknown word. The procedure code of the word extraction step, the procedure code of the semantic classification estimation step of estimating the semantic classification of the unknown word from the relationship between the particle of the unknown word and the dependency word, and combining the estimated semantic classification with the unknown word And a procedure code for an unknown word registration step of registering in the dictionary.

Further, the document processing method according to the present invention is a document processing method for converting an input character string while appropriately referring to a dictionary containing the meaning classification of words. An unknown word extraction step of extracting an unknown unknown word, a dependency word extraction step of extracting a dependency word having a dependency relation with the extracted unknown word, and a relationship between the particle of the unknown word and the dependency word A semantic classification estimating step of estimating the semantic classification of the unknown word; and an unknown word registration step of registering the estimated semantic classification in the dictionary together with the unknown word.

According to a preferred embodiment of the present invention, for example, the dependency word is a verb, and the semantic classification estimating step includes the step of determining the semantic classification of at least one word that may have a dependency relationship with each verb. It is preferable that the semantic classification of the unknown word is estimated from the particles of the unknown word by referring to a valency dictionary in which particles suffixed to the word having the semantic classification are stored for each verb.

[0013]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram illustrating a system configuration of the document processing apparatus according to the present embodiment. The CPU 110 is a microprocessor that performs calculations, controls, and the like for character processing. The CPU 110 includes a control bus (CB) 211, a data bus (DB) 212, and an address bus (AB) 213.
Control each device. The CB 211 is a control bus including a read signal, a write signal, an interrupt signal, and the like. The DB 212 is a data bus that transfers data between the CPU 110 and each device. AB212 is
An address bus for specifying a control target of the CPU 110.

The ROM 120 is a read-only memory, and stores a program to be supplied to the CPU 110 according to the embodiment. This program includes an operator rating system, a document processing program, and the like. The details of the document processing program will be described later.

The RAM 130 is a random access memory functioning as a work memory for the CPU 110.
BUF131, OBUF132, DIC133, KDI
C134, TBUF135, and program load area 136. IBUF 131 is an input buffer for temporarily storing key input data (reading sequence) input by keys, OBUF 132 is an output buffer for temporarily storing the result of kana-kanji conversion, and DIC 133 is a word for performing kana-kanji conversion. The dictionary, KDIC134, is a verb valency dictionary describing verb valency patterns, TBUF
Reference numeral 135 denotes a text buffer for storing a document being edited. Further, depending on the embodiment, the load area 136 is
An area for loading a document processing program or the like stored in the ISK 140 and supplying the program to the CPU 110.

A KB 150 is a keyboard including a keyboard interface. In addition to alphabet keys, hiragana keys, katakana keys, and other character symbol input keys,
Various function keys for giving an instruction to the document processing apparatus, such as a conversion key, a determination key for determining the conversion, a copy key, and a deletion key, are provided.

The DISK 140 is an external storage unit including a hard disk and a floppy disk. The disk 140 stores document data edited on a text buffer (TBUF) 135, and stores a document editing program described later on a floppy disk or the like. The program can be used to transfer the program to the load area 135 of the RAM 130. The stored document data can be read and used on a text buffer (TBUF) 135 as needed. In the case where the document processing program is supplied from the DISK 140, it is not necessary to store the document processing program in the ROM 120 redundantly.

The cursor register (CR) 160 has a CR
This register stores address information indicating the cursor position displayed by T190. The CPU 110 can change the cursor position by rewriting the address information in the cursor register (CR) 160.

Display buffer memory (DBUF) 170
Is a buffer memory for holding data (excluding cursors) displayed by the CRT 190,
Develops a bitmap image on the display buffer memory (DBUF) 170 with reference to the character generator (CG) 200 and the like.

CRT controller (CRTC) 180
Supplies an image signal to a CRT 190 serving as a display device and supplies a synchronization signal and the like to control the CRT 190 based on information supplied from a cursor register (CR) 160 and a display buffer memory (DBUF) 170. .
The CRT 190 is a display device using a cathode ray tube or the like.
The character generator (CG) 200 is a CRT 19
It generates or holds pattern data of characters and symbols displayed in 0.

In the above configuration, when an operator performs a key input through the keyboard (KB) 150, the keyboard (KB) 150 issues an interrupt request to the CPU 110. At this time, the execution of the CPU 110 shifts from the main routine to an interrupt routine for processing a key input, that is, a key input determination routine to be described later.

Next, the configuration of the document processing program according to this embodiment will be described. FIG. 2 is a diagram showing a conceptual configuration example of the document processing program, and its substance can be stored in the ROM 120 or the DISK 140 as described above.

Reference numeral 250 denotes a program module related to a key input determining process for determining a key input and performing a corresponding process based on the key input, and includes program modules 251 to 253 as subroutines. 251 is a program module for converting input characters into kana-kanji characters, 2
Reference numeral 52 denotes a program module relating to a determination process for performing learning when the kana-kanji conversion is determined, and reference numeral 253 denotes a program module relating to an editing process according to a normal technique such as moving, deleting, inserting, and copying a document.

The program module 252 has program modules 261 to 264 as subroutines. 261 is a program module for extracting an unknown word and its particles, 262 is a program module for extracting a verb having a dependency relationship with the extracted unknown word, and 263 corresponds to a verb extracted from a verb valence dictionary (KDIC) 134. A program module 264 for extracting the verb valency pattern and estimating the semantic classification of the unknown word from the verb valency pattern that matches the particle of the unknown word is a program module 264 for registering the semantic classification of the unknown word.

FIG. 3 shows an input buffer (IBUF) 131.
2 is a diagram showing a conceptual configuration example of an output buffer (OBUF) 132. FIG. The first two bytes of both buffers are buffer size information, and are numerical values obtained by subtracting 1 from the number of characters stored in the buffer and doubling the number. “//” at the end of the input buffer (IBUF) 131 means that the conversion key has been pressed there. Each character is composed of 2 bytes per character. For example, "JIS X02
08 ”code.

FIG. 4 is a diagram showing a conceptual configuration example of the word dictionary (DIC) 133. As shown, the word dictionary (DIC) 133 includes “reading”, “notation”, “speech”,
It is composed of "word likelihood" and "semantic classification" fields.

The "reading" stores the reading of the word, the "notation" stores the notation of the word, and the "part of speech" stores the part of speech of the word. In the “word likelihood”, information indicating likelihood of the word itself, such as frequency information of use, is stored as values 1 to 5. A likelihood value of 5 means most likely and is interpreted as suspicious as the value decreases. Since the likelihood value 0 means that it cannot be considered at all, it does not exist as a word likelihood value. In the “semantic classification”, semantic classifications such as <organization>, <animal>, <food>, <human>, etc. are stored (generally plural). However,
The semantic classification is described when the word is a noun.

FIG. 5 shows a verb valence dictionary (KDIC) 13.
FIG. 4 is a diagram illustrating a conceptual configuration example of No. 4; As shown in the figure, the verb valence dictionary (KDIC) 134 includes a “verb” and fields of “first slot” to “fifth slot”.

The "verb" stores a verb serving as a heading. The “first slot” to the “fifth slot” store verb valency patterns of “verbs”. In each slot, information indicating the likelihood in the verb valency pattern of each slot is stored as “likelihood” in addition to the noun “semantic classification” and “particle”. The example of the verb valency pattern shown in the figure is “<human> keeps <animal>”,
"<Human> buys <thing>".

The input buffer (IBUF) 13
If an unknown word is included in the reading sequence input to No. 1, since the semantic classification of the unknown word is obviously unknown, an appropriate conversion candidate can be obtained by comparing the reading sequence with the verb valency pattern. Can not do. Therefore, in order to obtain an appropriate conversion candidate when a similar reading sequence is input next time, it is necessary to learn the semantic classification.

In this embodiment, when a reading string containing an unknown word is input (when the unknown word is included in the conversion result), the unknown word is converted based on the conversion result created according to the instruction of the operator. Automatically learns the semantic classification of.

<Principle of Learning of Semantic Classification> FIG. 6 is a diagram showing the principle of learning of semantic classification. As an example, the reading string "I am rat" is input buffer (IB
A case will be described in which the meaning classification of the word “rat” (mouse) is input to the UF) 131 and is unknown.

In this example, when the conversion result finally determined by the operation of the operator is "I have a rat", the particle "o" suffixed to the unknown word "rat" is obtained.
And searches the verb valency dictionary (KDIC) 134 for the particle “o” in the verb valency pattern of the verb “Kai” in the sentence. Then, if there is one that matches the condition, it can be specified that the semantic classification of the unknown word “rat” is “<animal>” from the correspondence with the verb valency pattern. By registering this semantic classification in the word dictionary (DIC) 133 together with readings, notations, etc., the next time a reading sequence including “mouse” is input, “mouse” is “<animal>”. Therefore, kana-kanji conversion can be performed efficiently.

Hereinafter, an operation example of the document processing apparatus will be described with reference to a flowchart.

<Operation Example of Key Input Determination Routine> FIG.
5 is a flowchart showing the flow of processing of a key input determination routine. In the key input determination routine, when a key input occurs via the keyboard (KB) 150,
The transition is made by generating an interrupt request from the keyboard (KB) 150 to the CPU 110.

First, in step S701, a key input from the keyboard (KB) 150 is fetched. Step S
At 502, it is determined what key the input key is, and the process branches based on the result. That is, if the input key is a conversion key ("//" above), the process proceeds to step S703. If the input key is a determination key, the process proceeds to step S704.
If it is another key (a key other than the conversion key and the confirmation key), the process proceeds to step S705.

Step S703 (conversion processing) and step S704 (confirmation processing) will be described later. In step S705, based on the key input, an editing process according to a normal technique such as moving, deleting, inserting, or copying a document is performed. Then, when the processing of step S703, S704 or S705 is completed, the process returns to the main routine.

<Example of Operation of Conversion Processing Routine> FIG. 8 is a flowchart showing the flow of the conversion processing in step S703. In step S801, the input buffer (IBU
F) Analyze the read character string stored in 131 and create a phrase candidate group. For example, when the reading column “I am a cat” is input, the phrase candidates are, for example, “Wa”,
"Cotton", "I", "other magazines", "other magazines", "sleeping",
"Cat,""cat,""child,""child,""buying,""keeping,""winning,""selfish,
It is like "I'm here", "I'm going".

In step S802, a first candidate determination process for determining a first conversion candidate from the created document candidate group is performed. In step S803, a conversion result is created based on the determined first candidate, and the output buffer (OBUF) 13
2 to return to the key input determination routine.

FIG. 9 is a flowchart showing the flow of the first candidate determining process in step S802. First, in step S901, the maximum likelihood sentence likelihood is initialized to, for example, a minimum value (for example, -32767) that is allowed in processing.

In step S902, step S801
One phrase candidate sequence (for example, “I have a cat”, “I buy a cat”, etc.) is extracted from the phrase candidate group created in the (phrase candidate creation process).

In step S903, a verb valency pattern applicable to the extracted phrase candidate string is retrieved from the verb valency dictionary (KDIC) 134, and the "verb valency likelihood" which is the sum of likelihoods included in the pattern is retrieved. Is calculated (details will be described later).

In step S904, the sentence likelihood indicating the likelihood of the target phrase candidate sequence is calculated by using the general phrase likelihood, the inter-phrase likelihood, the example likelihood, and the above-mentioned verb binding likelihood. Is calculated from

In step S905, step S904
It is determined whether or not the sentence likelihood (current sentence likelihood) calculated in (2) is larger than the maximum sentence likelihood (likelihood). If it is likely, the process proceeds to step S906; otherwise, the process proceeds to step S908. Go directly to.

In step S906, step S904
The maximum sentence likelihood is updated based on the current sentence likelihood calculated in step S907, and in step S907, the target phrase candidate sequence is stored (if it is already stored, it is updated).

In step S908, it is determined whether or not another clause candidate string exists.
Returning to step 902, the above processing is repeated for another phrase candidate sequence, and a phrase candidate sequence with the maximum sentence likelihood is found.
On the other hand, when the above processing is completed for all the phrase candidate strings, the process returns to the conversion processing routine.

FIG. 10 is a flowchart showing the flow of the valency likelihood sum calculation processing in step S903. First,
In step S1001, the sum of likelihood of valence is initialized to zero. In step S1002, the verb is extracted from the phrase candidate sequence, and the verb valency pattern (KDIC) 134 is searched for a verb valency pattern relating to the verb. In step S1003, a noun having a dependency relationship with the verb extracted from the phrase candidate sequence and a particle of the noun are searched.

In step S1004, step S10
03, it is determined whether or not a noun having a dependency relation with the extracted verb and a particle of the noun could be found, and if the noun could not be found (the processing for all applicable nouns was performed). In this case, the process is terminated, and the process returns to the first candidate determination process routine.

In step S1005, the verb valence pattern corresponding to the verb extracted from the phrase candidate sequence and the semantic classification of the noun extracted from the phrase candidate sequence (word dictionary (DIC)
133) and its noun particle and extract the likelihood of the matching slot. Then, the extracted likelihood is added to the sum of valence likelihoods, and step S1003
Return to Hereinafter, the next noun and its particle are searched from the phrase candidate sequence, and the processing from step S1003 is repeated until all the nouns are processed.

<Operation Example of Determination Processing Routine> An operation example of the determination processing routine will be described below. FIG. 11 is a flowchart illustrating the flow of the determination process in step S704. First, in step S1101, "semantic classification unknown word processing" is performed in accordance with the <principle of learning of semantic classification>. In step S1102, another learning process (for example, a relationship between reading and notation) is performed based on a normal technique.
Return to the key input determination routine.

FIG. 12 is a flowchart showing the flow of the process of unknown words in the meaning classification in step S1101. First,
In step S1201, the word dictionary (DIC) 133 is referred to to determine whether the output result of the kana-kanji conversion, that is, the output result on the output buffer (OBUF) 132 includes an unknown word whose semantic classification is unknown. If an unknown word is not included, the process returns to the finalization process routine because it is not the target of the following process. On the other hand, if an unknown word is included, the process proceeds to step S1202, and the unknown word and a particle suffixed to the unknown word are extracted.

In step S1203, a verb having a dependency relationship with the extracted unknown word is output to an output buffer (OBUF).
132 from the conversion result. Step S1204
Then, referring to the verb valence dictionary (KDIC) 134,
Obtain the verb valency pattern for the extracted verb. In step S1205, the particle extracted in step S1202 is compared with the particle in each slot of the verb valency valence pattern obtained in step S1204. If there is a slot in which both match, the semantic classification of the slot is determined in step S1206. It is presumed that the unknown word is a semantic classification of the previously extracted unknown word, and the unknown word is referred to a word dictionary (DI).
C) 133, the process returns to step S1201, and the same processing as described above is performed for the next unknown unknown. On the other hand, if there is no matching slot, the semantic classification of the previously extracted unknown word cannot be estimated according to the above example, so that the unknown word is not registered, and step S1201 is not performed.
And the same processing as above is performed for the next unknown post.

As described above, when there is an unknown word relating to the semantic classification in the conversion result, the semantic classification of the unknown word is determined by referring to the verb valency pattern relating to the verb having a dependency relationship with the unknown word. Estimating and registering eliminates the need for the operator to register unknown words every time.

<Another Processing Example Regarding Unknown Word> In the above description, an example of realizing estimation of the semantic classification of an unknown word by comparing the particle of the unknown word with the corresponding verb valency pattern will be described. As described above, when there is no verb valence pattern that matches the particle of the unknown word, for example,
The semantic classification of an unknown word can be estimated as follows. That is, for example, if the particle suffixed to an unknown word is "de", when the unknown word has a specific particle, such as by assigning a semantic classification of "<place>", the particle is uniquely identified from the particle A semantic classification that can be derived, or a semantic classification that is likely to be associated with the particle can also be given.

The present invention can be applied to a system including a plurality of devices (for example, a host computer, an interface device, a reader, a printer, etc.), and can be applied to a single device (for example, a copier, Facsimile machine, etc.). Further, an object of the present invention is to provide a storage medium storing a program code of software for realizing the functions of the above-described embodiments to a system or an apparatus, and a computer (or CPU or MPU) of the system or apparatus to store the storage medium. Needless to say, this can also be achieved by reading and executing the program code stored in the program.

In this case, the program code itself read from the storage medium realizes the functions of the above-described embodiment, and the storage medium storing the program code constitutes the present invention.

As a storage medium for supplying the program code, for example, a floppy disk, hard disk, optical disk, magneto-optical disk, CD-ROM, CD
-R, a magnetic tape, a nonvolatile memory card, a ROM, or the like can be used.

When the computer executes the readout program code, not only the functions of the above-described embodiment are realized, but also the OS (Operating System) running on the computer based on the instruction of the program code. ) May perform some or all of the actual processing, and the processing may realize the functions of the above-described embodiments.

Further, after the program code read from the storage medium is written into a memory provided on a function expansion board inserted into the computer or a function expansion unit connected to the computer, based on the instructions of the program code, It goes without saying that the CPU included in the function expansion board or the function expansion unit performs part or all of the actual processing, and the processing realizes the functions of the above-described embodiments.

[0060]

As described above, according to the present invention, the semantic classification of an unknown word is estimated and automatic registration is performed, so that the operator can learn the semantic classification without registering the unknown word. And the operability can be improved.

[0061]

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a system configuration of a document processing apparatus according to an embodiment.

FIG. 2 is a diagram illustrating a conceptual configuration example of a document processing program.

FIG. 3 is a diagram illustrating a conceptual configuration example of an input buffer and an output buffer.

FIG. 4 is a diagram showing a conceptual configuration example of a word dictionary.

FIG. 5 is a diagram showing a conceptual configuration example of a verb valence dictionary.

FIG. 6 is a diagram showing the principle of learning of semantic classification.

FIG. 7 is a flowchart illustrating a flow of processing of a key input determination routine.

FIG. 8 is a flowchart illustrating a flow of a conversion process.

FIG. 9 is a flowchart illustrating a flow of a first candidate determination process.

FIG. 10 is a flowchart illustrating a flow of a valence likelihood sum calculation process.

FIG. 11 is a flowchart illustrating the flow of a confirmation process.

FIG. 12 is a flowchart showing a flow of semantic classification unknown word processing.

[Explanation of symbols]

 131 input buffer 132 output buffer 133 word dictionary 134 verb valency dictionary 135 text buffer 140 external storage unit

Claims (5)

[Claims] 1. A document processing apparatus for converting an input character string while appropriately referring to a dictionary storing semantic classification of words, wherein the unknown word extracting an unknown word whose semantic classification is unknown from the converted character string Extraction means; dependency word extraction means for extracting a dependency word having a dependency relationship with the extracted unknown word; and estimating the semantic classification of the unknown word from the relationship between the particles of the unknown word and the dependency word. A document processing apparatus comprising: semantic classification estimating means; and unknown word registering means for registering the estimated semantic classification in the dictionary together with the unknown word. 2. The dependency word is a verb, and the semantic classification estimating means includes a semantic classification of at least one word that may have a dependency relationship with each verb, and a particle suffixed to the word having the semantic classification. 2. A valence dictionary means for storing each of the verbs for each verb, and referring to the valence dictionary means to estimate a semantic classification of the unknown word from a particle of the unknown word. Document processing device. 3. A document processing apparatus that operates based on a program on a memory medium and converts an input character string while appropriately referring to a dictionary containing semantic classifications of words. A procedure code of an unknown word extraction step of extracting an unknown word whose semantic classification is unknown from the extracted character string, and a procedure code of a dependency word extraction step of extracting a dependency word having a dependency relationship with the extracted unknown word; A procedure code for a semantic classification estimation step of estimating the semantic classification of the unknown word from the relationship between the particle of the unknown word and the dependency word; and an unknown word registration for registering the estimated semantic classification in the dictionary together with the unknown word. A document processing apparatus, comprising: a procedure code of a process; 4. A document processing method for converting an input character string while appropriately referring to a dictionary containing semantic classification of words, wherein the unknown word extracting an unknown word whose semantic classification is unknown from the converted character string. An extraction step, a dependency word extraction step of extracting a dependency word having a dependency relation with the extracted unknown word, and a semantic classification of the unknown word is estimated from a relationship between the particles of the unknown word and the dependency word. A document processing method comprising: a semantic classification estimating step; and an unknown word registering step of registering the estimated semantic classification together with the unknown word in the dictionary. 5. The dependency word is a verb, and the semantic classification estimating step includes a semantic classification of at least one word that may have a dependency relationship with each verb, and a particle suffixed to the word having the semantic classification. 5. The document processing method according to claim 4, wherein a semantic classification of the unknown word is estimated from the particles of the unknown word by referring to a valence dictionary containing the following for each verb.