JP3348872B2 - Japanese morphological analyzer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

TECHNICAL FIELD The present invention relates to a Japanese morphological solution 析装 location,
And more particularly, to a morpheme solution 析装 location or syntax analysis apparatus of the Japanese. For example, the following parts for Japanese documents: (1) keyword extraction in a retrieval system, (2) language statistical information extraction in a document creation system, (3) proofreading information extraction in a proofreading support system, (4) Index word extraction, (5) applied to the Japanese sentence morphological analyzer of the Japanese speech synthesis system.

[0002]

2. Description of the Related Art A morpheme dictionary and an adjacent morpheme part-of-speech connection table are provided. A morpheme character string is cut out based on the morphological dictionary for Japanese sentences mixed with kana-kanji characters, and a morpheme part-of-speech candidate is assigned to the character string. A Japanese morphological analysis method for decomposing a morpheme string by examining adjacent morpheme parts of speech based on a known technique is known. For example,
"Illustrated natural language processing" (Shoichi Noguchi et al., "Processing morphemes" P23-34) and JP-A-56-17467.

For example, in order to perform a morphological analysis of the following Japanese sentence, "He traveled to the wide United States in search of her," a morpheme as shown in FIG. 5 is registered in at least a morphological analysis dictionary. If there is an adjacent morpheme part-of-speech connection table as shown in FIG. 7, a morphological analysis result as shown in FIG. 7 can be obtained (hereinafter, for the sake of simplicity, description will be made based on the simplest longest matching method among conventional morphological analysis methods) However, this does not preclude a more accurate priority selection method such as the two-phrase longest match method). In other words, of those that match the morpheme string in the morpheme dictionary from the beginning of the target Japanese sentence,
The morpheme character string having the longest character string length is cut out together with the morpheme part of speech, and the morpheme part of speech is divided into a list of connectable morphemes while checking the connection possibility of the adjacent morpheme speech. In addition, for example, in some cases, it is not possible to determine whether the "thing" of "delivering a message" is a "form noun" or a "final particle" indicating an instruction. Absent.

The morphological analysis is based on the premise that the morphological dictionary has all morphemes (words) included in the target Japanese sentence. Therefore, a dictionary is usually required to have morphemes of tens of thousands of words or more. However, morphemes that are not actually in the dictionary often appear,
In order to remedy the analysis failure due to this, a process called unregistered word processing is required. For example, JP-A-63-6
6672 and JP-A-63-163962. In addition, since many unregistered words occur when a Japanese document in an unknown field is targeted, a technical term dictionary of several thousand to several tens of thousands of words must be prepared.

Although the Japanese morphological analysis technology itself is at a high level, as described above, since a dictionary having a large capacity is required, it is difficult to mount the technique on a notebook personal computer having a limited disk capacity. In addition, for documents in unknown fields, work to register new words in advance occurs.
This is a factor that hinders the use of this technology. On the other hand, Japanese sentences written with kanji and kana are not a few exceptions, but in general, hiragana is a particle, an auxiliary verb, or an inflected ending, while a character string consisting of kanji or katakana is a noun, And the stem of adjectives are known, and this property is used in unregistered processing. It is described in P33 ,, of the known document.

[0006]

An object of the present invention is to provide an apparatus for performing a morphological analysis on a subset of Japanese sentences mixed with kanji and kana without a large-size dictionary. to provide a device that extends, also possible to provide a device to improve the accuracy, or one in which that you provide a device for obtaining a sequence of words from the morphological analysis results obtained was made as the feature .

[0007]

To achieve the above object, the present invention provides (1)
Japanese decompose the Japanese statement of kanji and kana in the morpheme string
The morphological analyzer stores at least a morpheme of a non-independent word consisting of only a hiragana character string and its morpheme part of speech.
Morpheme dictionary and adjacent morpheme parts of speech can be connected
Morphological part-of-speech connection table that stores
An input means for inputting a subset of the Japanese sentence;
Cuts out a character string of the same character type from the input sentence
Output means and when this cut out character string is Hiragana
Matches the morphological dictionary with the longest match method.
Morphemes and their morpheme parts of speech
And, on the other hand, if the extracted character string is non-hiragana
A morpheme part-of-speech candidate assigning means for assigning a plurality of specific morpheme part-of-speech candidates for each character type of the character string;
Referring to the morpheme part-of-speech connection table, these morpheme part-of-speech candidates
Connectivity checking means for checking connectivity of the
The morphological analysis result of the sentence can be connected by the connectivity checking means.
Morpheme sequence determined to be functional, and
(2) In the above (1), the morphological dictionary further includes:
A morpheme of an independent word that is often described only by a hiragana character string , and a morpheme having a morpheme part of speech other than the specific morpheme part of speech candidate, which is composed of a character string composed of only the same character type such as kanji or katakana, and a kanji character string Morphological parts of speech correspond to morphemes followed by hiragana character strings.
Allowed to be stored, further, (3) in (1) or (2), the morpheme dictionary, this morpheme dictionary
Of the morphemes of the character string consisting only of the same non-Hiragana character type that is not stored, the morpheme part-of-speech form that is rarely used
The morpheme is also stored, and the morpheme part of speech candidate assigning means
From the specific morpheme part-of-speech candidates to be given when the extracted character string is a non-hiragana,
Be excluded, further, (4) the (1), and (2)
Have the suffix dictionary for storing fraud and mitigating risk morphemes consisting of kanji character string that can be attached to the end of the Kanji character string (3), the
The extracting unit extracts a character string having the same character type.
When, if the end of the string is matched with the stored form <br/> Taimoto the suffix dictionary, matching this string
Morphemes of it has to cut off the string, further, (5) the (1), (2), (3) or (4) to be had you <br/>, the morpheme dictionary, a plurality of morphemes And the arrangement of morpheme parts of speech in the order of those morphemes
Was stored in the connection checking means, when checking the connection with the morpheme adjacent to the composite morphemes, the adjacent shaped
References Taimoto part of speech connection table, the arrangement of the composite morpheme for the previous morpheme performed by the morpheme part of speech for the leftmost morpheme arrangement of the multi <br/> case morphemes for morphemes after of and TURMERIC line by the morphological part of speech for the top-right of the morpheme, and further, to the morphological part of speech together a sequence of (6) morpheme part of speech
The word composition rule that derives a word by replacing, and the above (1) , (2), (3), (4) or (5 )
Apply the sequence of split morphemes to this word composition rule
It is characterized by having a word generation means for deriving words
Things . Hereinafter, a description will be given based on examples of the present invention.

[0008] First, a description will be given in Example 1. In the present invention, in contrast to the conventional morphological analysis method described in the section of the prior art, the morphological dictionary has only morphemes of non-independent words consisting only of hiragana character strings, and his noun broad adjective: stem (wide) American noun One-stage verb: stem (seek) Travel Sa variant: stem Without a kanji or katakana character string such as in the morphological dictionary, for these strings, "noun", "adjective: stem", "adjective verb: stem" , "Godan Verb: Stem",
In this method, morpheme analysis is performed by giving "upper verb: base", "lower verb: base", "sa varistor: stem", etc. as morpheme part-of-speech candidates.

Note that, for katakana character strings, "five-step verb: stem", "upper-step verb: base", "lower-step verb: base", etc. do not need to be set, and are given according to the character type. The morpheme part of speech candidates are different. In the range of this example sentence, "lower one-stage verb: base" which results in a morpheme part of "request" is introduced, and the next morpheme part of speech is added to "me". The one-step verb of Ma-line lower: Send, the one-step verb of Ma-line lower: Indefinite ending, the one-step verb of Ma-line lower: Conjunctive ending In addition, in the adjacent morpheme part-of-speech connection table, "Ma-line lower-stage verb: Sent" etc. Following "single-stage verb: base", "ma-line-lower-stage verb: sent" can be connected to "lower-stage-verb: adnominal-type ending", "lower-stage-verb: final-type ending", and the like. Thereby, a morphological analysis result as shown in FIG. 1 is obtained.

In the morphological analysis method of the present invention, a kanji character string or a katakana character string is cut out as a unit.
As shown above, "lower one-step verb: stem" is composed of "lower one-step verb: base" in the kanji part and the sending part of Hiragana "□ line lower one-step verb: send", "□ lower line one-step verb: unprepared. Infinitive endings "and" □ line lower verb: continuous use endings "(□: a, ka, mo, etc.). The same applies to the upper verb. In addition, the stem "beautiful" of the adjective "beautiful" and the stem "quiet" of the adjective "quiet" are treated as follows.・ "Beauty": "Shiku utilization adjective: base"("Beauty") + "Shiku utilization adjective: send"("shi") ・ "Quiet": "Kada utilization adjective: base"("static") ) + "Kada Utilization Adjective: Send"("ka")

FIG. 2 is a diagram showing a processing flow of the Japanese morphological analysis method according to the present invention. It should be noted that alphabetic character strings, symbol character strings or numeric character strings, and reading marks “,”, punctuation marks “.”, And the like are handled by processing for each character type. step 1 : Extract a target character string (S: Sentence). step 2 : Determine whether the first character string of S is a hiragana or a non-hiragana. step3 : In step 2, if Hiragana is used, the morpheme of the longest matching Hiragana character string is cut out using a morpheme dictionary, and the same morpheme part of speech is set. step 4 : In step 2, if non-hiragana, a character string of the same character type is cut out. step 5 : Next, morpheme part-of-speech candidates are set for each character type. step6 : Delete the part of speech candidate by checking the connection between the previous morpheme and the morpheme part of speech candidate. step7 : Extract the remaining character string of the cutout. step 8 : Determine whether the character string is empty. If it is not empty, the process returns to step 2, and if it is empty, the process ends.

The morpheme analysis method of the present invention does not handle the following morphemes. Morphemes such as nouns and verbs expressed only in hiragana character strings; morphemes such as connectives and adverbs; morphemes in which a hiragana character string continues after a kanji character string. Also, the ambiguity of the following morpheme part of speech cannot be resolved.・ "Noun" + "case particle (ni)" and "adjective verb: stem" +
"Adjective verb: Conjunctive ending" (Ni) ・ "Godan verb: Stem" + "Sa line 5-dan verb: Conjunctive ending"
(Shi) and "Sa variant: Stem" + "Sa behavioral verb: Conjunctive ending" (Shi), etc. Despite the shortcomings, etc., a large dictionary is required for a certain range of subsets of Japanese sentences mixed with Kanji and Kana. Morphological analysis results In the present invention,
Compound words in which morphemes of the same character type are linked cannot be divided, and are treated as one morpheme. Most of these compound words are composed of nouns, for example, "morphological analysis" in which "morpheme" and "analysis" are linked. Treating it as one word is not necessarily unjust.

[0013] Next, a description will be given in Example 2. The second embodiment extends the range of a subset of Japanese sentences mixed with kanji and kana that can be handled by having the next morpheme in the morpheme dictionary, which is not handled in the morphological analysis method. (1) A morpheme of an independent word consisting only of a hiragana character string (2) A morpheme consisting of a character string consisting only of the same character type such as kanji or katakana and having a morpheme part of speech other than the specific morpheme part of speech candidate (3) Kanji characters A morpheme consisting of a sequence of hiragana characters

The above (1) refers to nouns, verbs, adjectives, and adjectives that are often described in hiragana, such as "hiragana" (noun) and "aru" (five-step verb). Further, these are other independent words not dealt with in the first embodiment. Add the adverb "aru", the conjunction "and", and the adverb "slow". The above (2) is an adverb noun “yesterday” or a formal noun “between” which is treated differently from ordinary nouns, or other independent words not treated in the first embodiment, as in the latter of the above (1). , Adverbs “why”, “in the unlikely event” and the like. The above (3)
Is composed of kanji and hiragana, such as the parallel verb "and", the noun "mutual", the adverb "mayichi" or the stem "floating" of the five-stage verb "floating". It is clear that the above expansion extends the range of the subset of Japanese sentences that can be handled with kanji and kana.

FIG. 3 shows a Japanese morphological analysis method according to the present invention.
It is a figure showing the processing flow of Example 2 of the method.step1 : Extract the target character string (S).step2 : Cut morpheme of longest matching character string by extended dictionary
Then, the same morpheme part-of-speech candidate is set.step3 : Judge whether the extraction is successful or unsuccessful. Success
If so, the process proceeds to step 8. step4 : In step 3 above, if unsuccessful, the first sentence of S
Judge whether the character string is Hiragana or non-Hiragana.step5 : In step 4 above, if hiragana, hiragana
Longest match with the Kanagawa dictionary and the Hiragana independent dictionary
Cut out the morpheme of the Hiragana character string, and
Set the candidate words.step6 : Same if non-Hiragana in step 4
Extracts character string.step7 : Set morpheme part of speech candidates for each character type.step8 : By checking the connection between the previous morpheme and the morpheme part-of-speech candidate
Delete part-of-speech candidates.step9 : Extract the remaining character string after clipping.step10 : Determine whether the character string is empty. In the sky
If not, the process returns to step 2; if empty, the process ends. Here, the morphological dictionary of Example 1 is
, Hiragana independence dictionary containing the morpheme of (1)
Dictionary containing the morphemes of (2) and (3)
However, it is not always necessary to use another dictionary.

In the processing shown in FIG. 3, before the processing of the hiragana character string or the non-hiragana character string that branches depending on whether the beginning of the character string in the first embodiment is a hiragana character or not, a morpheme extraction process using an extended dictionary is added. Conventional processing is performed only when this processing fails. In the processing of the Hiragana character string, both the Hiragana attached word dictionary and the Hiragana independent word dictionary are searched, and the longest morpheme of both is adopted. If both are the same length, the merged result of both is set.

[0017] Next, a description will be given in Example 3. In the morphological analysis method described above, possible morpheme part-of-speech candidates are assigned to kanji strings or katakana strings (in the second embodiment, kanji strings or katakana strings that are not in the extended dictionary).
In the third embodiment, taking into account the number of morphemes that are the morpheme POS, the morpheme of the morpheme POS having a small number is registered in the extension dictionary, and the processing efficiency and accuracy of the morpheme POS are improved. In the above morphological analysis method, for example, a kanji string (in the second embodiment, a kanji string not in the extended dictionary)
For "noun", "adjective: stem", "adjective verb:
Stem, "5-stage verb: stem", "Upper-stage verb: base",
"Lower verb: base", "variable verb: stem", etc. were set uniformly.

However, for example, "sa variant: stem" and "adjective verb: stem" usually originate from a Chinese word such as "survey" or "perfect", so that there are very few one-kanji characters. By utilizing this property, for one kanji string, instead of assigning “sa var: stem” or “adjective verb: stem” as a morpheme part-of-speech candidate, one kanji “sa var: stem”
And "adjective verb: stem" in the extended dictionary. Examples of these are “love” and “easy”, but since the total number is extremely small, the size of the dictionary is hardly affected.

[0019] In addition, the "one-step verb" and the "five-step verb"
It is known that the number of "Upper one-step verbs" is considerably smaller than the number of "Lower one-step verbs", so it is reasonable to register "Upper one-step verbs: stem" in the extended dictionary and remove them from morpheme part-of-speech candidates for Kanji strings. . That is, although the size of the dictionary is slightly increased by the above-described processing, the set morpheme part-of-speech candidates are eliminated, and the efficiency is improved, and the accuracy of the analysis of the registered morphemes can be expected to be improved. . Note that, in the processing flow, step 7 “setting of morpheme part-of-speech candidate for each character type” in FIG. 3 is changed based on the above.

[0020] Next, a description will be given in Example 4. In the morphological analysis method of the present invention, a compound word in which morphemes of the same character type are connected cannot be divided in principle, but as described at the end of the first embodiment, this is not necessarily unreasonable. However, when a general noun is not continuous, such as "at the time of analysis" or "thorough", it is desirable to divide the noun. If divided, the former can be recognized as an adverb noun and the latter as an adjective verb,
The morphological analysis result can be a grammatically valuable result. In the present embodiment, when a morpheme suffixed to a morpheme is attached to a morpheme of the same character type (in this case, a kanji character), such as “time” or “target”, a morphological analysis method of dividing these is used. I will provide a. That is, in addition to the extended dictionary, a suffix dictionary having these morpheme expressions and their character string lengths is provided.
After `` cutting out a character string of the same character type '', check whether the suffix dictionary has a morpheme at the end of the cut-out character string, for example, by cutting off the matching suffix from the end of the cut-out character string. Realize. As a result, the character string cut off in the next loop comes first, and is preferentially cut out by the extended dictionary.

[0021] Next, a description will be given in Example 5. The fifth embodiment is a method that is effective when limited to the morphological analysis method based on the longest match strategy. In general, this method does not handle consecutive compound words of the same character string, so that the effect of the two-phrase longest-match method is not necessarily large compared to the simple longest-match method. However, for example, the next two-character hiragana character string may be of two types, two characters and one character, respectively, and the simplest longest matching method may cause the latter to drop.・ Noni: Conjunctive particle, quasi-body particle (no) + case particle (ni) ・ So: Conjunctive particle, quasi-body particle (no) + case particle (de) ・ More: case particle, la line five-step verb ): Stem + La line five-stage verb: Conjunctive ending ・ Nara: Auxiliary verb (da): Hypothetical form, La line five-stage verb (Naru): Stem + La line five-stage verb: Unfinished ending

According to the present invention, a character string having a morpheme sequence can be registered as a morpheme part-of-speech sequence as in the latter case.
A connection check with the morpheme part of speech of the front or back morpheme is performed by the leftmost or rightmost morpheme part of the row of the morpheme part of speech, and when this row is finally determined,
Dividing according to this arrangement compensates for the defect caused by the longest match that occurs in this morphological analysis method. Specifically, in each of the morphological dictionaries (Hiragana attached word dictionary, Hiragana independent word dictionary, extended dictionary), for example, it is allowed to set the arrangement of the set of the morpheme part of speech and the number of characters as follows.・ More: La line five-stage verb (by): stem {1} + La line five-stage verb: Conjunctive ending {1} The preceding morpheme is "La line five-stage verb (by): stem", the back morpheme What is "La five-step verb:
A connection check is performed using the continuation form ending, and when this is determined, it is divided according to the number of characters shown in {}. ・ Yo: La line five-stage verb (Yo): stem ・ Ri: La line five-stage verb: It is a continuation form ending. These processes are performed in step 8 in the process flow of FIG.
This may be performed in “deletion of part-of-speech candidate by checking connection of previous morpheme with morpheme part-of-speech candidate”.

Next, said even real Example 1 "thing" and "to deliver the message" Do "Format nouns" or indicating a command "final particle", the morphological analysis of conventional morphological analysis methods and the present invention Even with the method, the ambiguity of morpheme parts of speech cannot always be resolved. The above example is a problem that occurs regardless of the method, but in addition, since the method of the present invention does not use a sufficient morphological dictionary, the ambiguity of morpheme part-of-speech is reduced compared to a case where a sufficient dictionary is provided. Poor ability is considered. This embodiment is used in the case where ambiguity remains in the result of morphological analysis, in order to compensate for the problem of ambiguity that cannot be resolved by the original morphological analysis and the lack of ambiguity of the method due to an insufficient dictionary. Get accuracy by inquiring the person. As an example of the processing, after the processing of the entire target sentence shown in FIG. 3 is completed, the morpheme result is sequentially inspected, and if there is a plurality of morpheme part-of-speech, a message to that effect is issued, and the remaining morpheme part-of-speech candidates are One of the selections may be requested, and the morpheme part of speech may be determined according to the input.

Next, two methods for avoiding the above-mentioned inquiry to the user without significantly lowering the accuracy will be described. One is that if one of the frequently remaining combinations of ambiguity is fairly frequent, it will be automatically selected without asking the user. this is,
A rule describing this combination and its selection side (and the situation before and after it as necessary) may be processed according to the rule immediately before the inquiry to the user. For example, the following rule can be considered.・ Noni: a: <conjunctive particle>, b: <quasi-morphic particle (no) + case particle (ni)> → a (-1: <>, + 1: <>) Indicates that if the above two a and b remain as morpheme part-of-speech candidates, the connective particle of a is selected irrespective of the morpheme before and after (-1, + 1).

Another is to avoid the ambiguity resolution itself if the application using the result of the morphological analysis of the present invention does not cause a problem even if the specific ambiguity combination is not resolved. For example, "tsu" means "ta line",
"La row" and "wa row" are "five-step verbs: concatenated suffix endings", but "bought" means that "buy" is a "wa row five-step verb" unless it is in the dictionary. It is not possible to decide whether to go. However, this ambiguity does not necessarily need to be determined if it is only necessary to detect "verb" + "auxiliary verb (ta)".
Therefore, there is no inconvenience even if the inquiry is avoided. In this case, the same rule as described above may be described, and the rule may be processed just before the inquiry to the user. For example, the following rule can be considered.・ T: a: <Ta quintuple verb: Conjunctive spelling ending>, b: <La quintuple verb: Conjunctive suffix ending>, c: <W : <>, + 1: <>) This rule specifies that if the above three a, b, and c remain as morpheme part-of-speech candidates for the morpheme "tsu", regardless of the morpheme before and after (-1, + 1) This indicates that the inquiry of the morpheme part of speech to the user is avoided.

[0026] Next, a method for a sequence of morphemes obtained in the morphological analysis method before Symbol to words line (see FIG. 4). For example, the following word composition is performed. Beauty: Shiku conjugation adjective: base → beautiful: adjective (beautiful): conjunctive form c: shiku conjugative adjective: send: adjective: conjunctive ending This is obtained by setting and applying the following word composition rules. Shiku inflection adjectives: base + sikh inflection adjectives: sent → adjective: stem Adjective: stem + adjective: X-type ending → adjective (1): X-form

On the left side of the above rule, the parts of speech of morphemes (words) are arranged by separating them with "+", and on the right side, morpheme (word) parts of speech to be replaced with them. If the arrangement of the left side matches the morpheme sequence, the notations are joined for the morphemes in the range, and the part of speech replaces that of the right side. In the second rule, variables are used. X on the left side is
The value is obtained by matching with the morpheme part of speech, and set to X on the right side. Also, the value of the notation of the morpheme of the first term on the left side is taken as 1 on the right side. In the following, 2,3 ... is the second 2,3 ... on the left side
This is the value of the term morpheme notation.

The above-mentioned “Beauty: adjective adjective: base” + “shi: adjective adjective: send” + “ku: adjective conjunctive ending” is obtained by applying the first rule to “beautiful: adjective: stem”. + "Ku: adjective: conjunctive ending"
Then, by applying the second rule, it becomes "beautiful: adjective (beautiful): continuous form". In addition,
In the application of the second rule, X has a value of “continuous use” and 1 has a value of “beautiful”. In addition, if a rule of X: continuous form + noun → noun is set, it is possible to recognize a derivative word such as “cut” as one word.

The Japanese morphological analyzer according to the present invention is realized by providing the input unit for inputting a Japanese sentence and the output unit for outputting a result in the morphological analysis method described above. There are a plurality of input methods for inputting Japanese sentences, such as cutting out sentences from a Japanese document file and interactively inputting Japanese sentences. However, the input method is not limited to any of these. It does not prevent you from preparing. Also,
The format and output medium of the result output means are not specified.

[0030]

As apparent from the above description, the present invention has the following effects. (1) with respect to a large subset of the kanji and kana Japanese, unknown words registered consuming large hard disk capacity and manpower
Avoid the record, at a low cost, follow in a very small size of the dictionary
From the existing string level processing to morphological analysis level processing
Can be raised up . This facilitates the realization of an application using the morphological analysis result on a notebook personal computer with a limited hard disk capacity. Further, even when morphological analysis of a large number of documents in unknown fields is required, it is possible to use the morphological analysis result without performing an operation such as registering unknown words.
Furthermore, it is possible to extract the original form and derivative words of the inflected words that are not registered in the dictionary, and it is possible to collect words.

[Brief description of the drawings]

FIG. 1 is a diagram showing a morphological analysis result based on a Japanese morphological analysis method according to the present invention.

FIG. 2 is a diagram illustrating a processing flow according to the first embodiment of the present invention.

FIG. 3 is a diagram illustrating a processing flow according to a second embodiment of the present invention.

FIG. 4 is a diagram showing a morphological analysis result of the present invention.

FIG. 5 is a diagram showing a conventional morphological analysis dictionary (part).

FIG. 6 is a diagram showing a conventional adjacent morpheme part-of-speech connection table (part).

FIG. 7 is a diagram showing a conventional morphological analysis result.

Claims (6)

(57) [Claims] The Japanese statement of claim: 1. A kanji and kana in the morpheme string
In the Japanese morphological analyzer that decomposes, the number of non-independent word morphemes consisting only of Hiragana character strings and their morpheme parts of speech are reduced.
At least the stored morpheme dictionary and the adjacent morpheme part of speech
Adjacent morpheme part-of-speech connection table that stores whether connection is possible
And a subset of Japanese sentence mixed with Kanji and Kana
A character string consisting of the same character type as the input means and this input sentence
And the extracted character string
In the case of Chinese characters, the morphological dictionary and the longest match method are used.
Match and form matched morphemes and their morpheme parts of speech
As a part-of-speech participant, the extracted character string is
In this case, the morpheme part-of-speech candidate assigning means assigns a plurality of specific morpheme part-of-speech candidates for each character type of the character string.
With reference to the adjacent morpheme part-of-speech connection table,
A connectivity checking means for checking the connectivity of POS candidates
The morphological analysis result of the input sentence is
Japanese morphological analysis apparatus characterized by a morpheme string is determined to be able to connect in stages. 2. The Japanese morphological analyzer according to claim 1,
In addition , the morphological dictionary further includes only hiragana character strings.
It consists of a morpheme of an independent word that is often written as, and a character string consisting of only the same character type such as kanji or katakana,
The specific morpheme having a morphological part of speech other than morphological word class candidates, hiragana strings continuous morpheme and each lish morphological analyzer day you characterized by storing in correspondence a morphological part of speech after a kanji character string. 3. The Japanese morphological analysis according to claim 1 or 2.
In the apparatus, the morphological dictionary is stored in the morphological dictionary.
Of 憶that is not a string of only the same character type of non hiragana morphemes, the number used is small morphological parts of speech in the form
The morpheme part-of-speech candidate assigning means,
From a particular morpheme word class candidates provided when out string is a non Hiragana, the number of used small morphological parts of speech
Lish morphological analyzer day it said the exclusion. 4. A Japanese morpheme according to claim 1, 2 or 3.
In analyzing apparatus, has a suffix dictionary for storing a morpheme composed of kanji character string that can be attached to the end of the kanji character string, wherein
The extracting unit extracts a character string having the same character type.
When, if the end of the string is matched with the stored form <br/> Taimoto the suffix dictionary, matching this string
The lish morphological analyzer day it said that you have to cut off the morpheme string. 5. The Japanese form according to claim 1, 2, 3 or 4
In the morphological analyzer, the morphological dictionary includes a compound morpheme in which a plurality of morphemes are arranged, and a morpheme part of speech in the order of the morphemes.
If storing the sequence of in correspondence, the connection checking means for checking the connection with the morpheme adjacent to the composite morpheme
If the double is a reference to the adjacent morphemes part of speech connection table for the previous morpheme performed by the morpheme part of speech for the leftmost morpheme arrangement of the composite morphemes for morphemes after
If morpheme line TURMERIC and the morphological analyzer day said by the morphological part of speech for the top-right of the morpheme sequence. 6. A summary of the sequence of form Taimoto part of speech morphological part of speech
And a Japanese morphological analyzer according to claim 1, 2, 3, 4, or 5 .
Apply the morpheme sequence divided by
A word generating means for deriving words by causing
Japanese morphological analyzer .