JP3091540B2 - Morphological analysis method for Japanese sentences - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a morphological analysis method for a Japanese sentence, and more particularly to a morphological analysis method for a Japanese sentence using a longest match method.

[0002]

2. Description of the Related Art Conventionally, in a morphological analysis method of a Japanese sentence using a longest match method (a method of selecting a word having a longest notation length from candidate words extracted from a dictionary), a character string to be analyzed is Search the word dictionary to extract candidate words,
If the longest notation length is selected from the connectable candidate words and, for example, the sentence “Looks like walking” is subjected to morphological analysis by the longest matching method, [Walk / Like / Fir / Eru: Here, "Momi" is a conjugation form of the verb "Mom".] There is a drawback that "Momiru" is not correctly divided, and is divided into "Momi" and "Eru". .

[0003] Further, when there are a plurality of words having the longest notation length, the preceding element in the candidate word list is prioritized, so that the word to be selected is used for word extraction during dictionary search. There is also a disadvantage that the correct word is not always selected depending on the order.

[0004] For example, as a result of performing a dictionary lookup for the sentence "can be connected", (1)
(1) "Connect" (noun) which is the preceding element in the candidate word list is selected when "connect" (noun) and (2) "connect" (the stem of the verb "connect") are created Is done.

Next, the remaining character string "possible" is looked up in a dictionary, and a list of connectable candidate words is given as (1)
Since "possible" (noun) is created and there is only one candidate, this word (1) "possible" (noun) is selected. Here, the adjective verb "possible" and the auxiliary verb "possible" are looked up in a dictionary, but the immediately preceding word "connection" (noun) is determined to be unconnectable and is not added to the candidate word list.

Next, the remaining character string "da" is looked up in a dictionary, and as a list of connectable candidate words, (1)
Since "da" (auxiliary verb) is created and there is only one candidate, this word (1) "da" (auxiliary verb) is selected.

As a result, the division is erroneously performed as follows.

[0008] Connection (noun) / possible (noun) / da (auxiliary verb).

[0009]

SUMMARY OF THE INVENTION The present invention has been made in order to eliminate the above-mentioned drawbacks. In the word selection processing, errors in the division position caused by selecting the word having the longest notation length are reduced. It is another object of the present invention to provide a morphological analysis method that can select a more correct word than before when there are a plurality of candidate words having the longest notation length.

[0010]

According to the morphological analysis method for a Japanese sentence of the present invention, a Japanese sentence using a longest match method for selecting a word having the longest notation length from candidate words extracted from a dictionary is used. In the morphological analysis, there is provided a candidate word deletion condition table that describes an erroneous combination of a notation character string of the candidate word and a character string that follows the candidate word, and the word having the longest notation length is the candidate word deletion condition. It has candidate word deletion processing means for deleting the word from the candidate words when the conditions in the table are met.

Further, the morphological analysis method for a Japanese sentence according to the present invention is a morphological analysis method for a Japanese sentence using the longest matching method, wherein a usage type of a word and a priority corresponding to the usage type are paired. When there are a plurality of words having the longest notation lengths in the type priority table and the words retrieved from the dictionary, the word having the longest notation length is referred to by referring to the utilization type priority table. And a utilization word selection processing means for selecting a word having the utilization type having the highest priority from the above.

[0012]

[0013]

According to the present invention, in a morphological analysis of a Japanese sentence using the longest match method, a candidate word deletion condition table describing an erroneous combination of a written character string of a candidate word and a character string succeeding the candidate word. If the word having the longest notation matches the condition in the candidate word deletion condition table, the word is deleted from the candidate words.

According to the present invention, in a morphological analysis of a Japanese sentence using the longest matching method, a utilization type priority table in which a utilization type of a word and a priority corresponding to the utilization type are paired; If there are a plurality of words having the longest notation length in the words that have been extracted from the dictionary, the priority is selected from the words having the longest notation length by referring to the utilization type priority table. Select the word with the highest usage type.

[0015]

[0016]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment using three example sentences in Japanese will be specifically described below.

FIG. 1 is a block diagram of a morphological analysis process embodying the present invention. (1) is a dictionary lookup unit, (2) is a word selection unit, and (3) is a word list creation unit.

(4) is a word dictionary storing information necessary for morphological analysis such as headwords and connection information.

(5) is a connection table that describes whether adjacent words can be connected or not. The dictionary lookup unit (1) searches the word dictionary (4) for the remaining character strings from the input Japanese sentence mixed with kanji and kana to the word separator selected by the word selection unit (2). At the same time, a "candidate word list" that can be connected is created with reference to the connection table (5).

(6) is a candidate word deletion condition table that describes an erroneous combination of a written character string of a candidate word and a character string that follows the candidate word.

(7) is a utilization type priority table in which the utilization type of the word and the priority corresponding to the utilization type are paired.

(8) is a part-of-speech priority table in which the part of speech of a word and the priority corresponding to the part of speech are paired.

In the process according to the present invention, the word selection unit (2) refers to the candidate word deletion condition table (6), the utilization type priority table (7), and the part of speech priority table (8) to search the dictionary. A word is selected from the "candidate word list" created in part (1).

The word list creating section (3) adds the word selected by the word selecting section (2) to the "word list".

FIG. 2 shows a word selector (2) according to the present invention.
6 is a flowchart showing the operation of the first embodiment. Hereinafter, each step of word selection will be described.

[Step A] If the candidate word list is empty (step B), go to step C if it is not empty.

[Step B] The word selection of the immediately preceding word is re-executed by the backtracking process, or the range of the unknown word is determined by the unknown word processing and the unknown word is set as the selected word, and the word selection is terminated. .

[Step C] If the word having the longest description in the candidate word list matches the condition in the candidate word deletion condition table (6) (step D), it does not match (step D). Move to E).

[Step D] The word having the longest description in the candidate word list is deleted from the candidate word list, and the process returns to step A.

[Step E] If there are a plurality of words having the longest notation length in the candidate word list (Step G)
If not, the process proceeds to step F.

[Step F] After selecting the longest word in the candidate word list as the selected word, the word selection ends.

[Step G] Referring to the utilization type priority table, if there are a plurality of words having the utilization type with the highest priority among the words having the longest notation length (step I), If it does not exist, the process moves to step H.

[Step H] After selecting the word having the highest usage type as the selected word, the word selection is terminated.

[Step I] Referring to the part-of-speech priority table, a word having the part of speech having the highest priority among words having the highest usage type is selected as a selected word.
Terminate word selection.

FIG. 3 shows an example of the candidate word deletion condition table (6) in this embodiment. Here, a plurality of pairs of a character string of a candidate word and a character string that follows the candidate word are described.

FIG. 4 shows a utilization type priority table according to the present embodiment, in which utilization types and priorities are paired. The priorities are quantified, and the higher the numerical value, the higher the priority.

FIG. 5 shows a part-of-speech priority table according to the present embodiment, in which part-of-speech and priority are paired. The priorities are quantified, and the higher the numerical value, the higher the priority.

The following three morphological analysis processes of the Japanese sentence will be described in order, centering on the operation of the word selector (2). (A) "Looks like walking" (b) "Connectable" (c) "We handle 3"

[Embodiment 1] First, the case of Japanese sentence (a) will be described.

When the word selection up to "like walking" is completed,
After "walking" and "you" are added to the "word list",
Dictionary lookup unit (1) for the remaining character string "Momiru"
As a result, as shown in FIG. 6, "momi" (conjunctive form of the verb "momu") and "mo" (adjunct particle) are obtained as a "candidate word list". Move on to processing.

Since the "word candidate list" is not empty (step A), it is checked whether or not the word "mum" having the longest description matches the condition of the candidate word deletion condition table (FIG. 3) (FIG. 3). Step C). In step C, the notation of the candidate word having the longest notation matches any one of the notation character strings in the candidate word deletion table (FIG. 3), and the succeeding character corresponding to the matched notation character string. Check if the sequence matches the front part of the string that follows the longest candidate word. In the case of this example, the notation “Momi” of the candidate word having the longest notation matches the first written character string “Momi” in the candidate word deletion table (FIG. 3), and the corresponding postscript character Since the column "eru" matches "eru" which follows "fir" in the example sentence (a) (step C), "fir" is deleted from the candidate word list (step D).

Returning to step A, since the "word candidate list" is not empty, it is checked whether the word "mo" having the longest description matches the condition of the candidate word deletion condition table (FIG. 3). In this case, it does not fit (step C). Next, in the candidate word list, since the word having the longest notation is only “mo” and a plurality of words do not exist (step E), the word “mo” having the longest notation in the candidate word list is used. "As a selected word, and terminates the word selecting section (2) (step F). Next, in the word list creation unit (3), the selected word "mo"
(Sub particle) is added to the word list, and the process returns to the dictionary lookup unit (1).

Next, as a result of performing the processing of the dictionary look-up unit (1) on the remaining character string "seeable", as shown in FIG. 7, "seeable" as the "candidate word list" (the verb "seeable" Adjunct / Termination) and "Mie" (noun) and "Mie"
(A verb "verb") is obtained, and the process proceeds to the word selecting unit (2).

Since the "word candidate list" is not empty (step A), it is checked whether the word "seeable" having the longest notation length satisfies the condition of the candidate word deletion condition table (FIG. 3). If not, it does not match (step C). Next, in the candidate word list, the word having the longest notation length is only "visible" and does not exist (step E).
The word selection unit (2) ends with the word “see” having the longest description in the candidate word list as the selected word (step F).

Next, the selected word "seeable" is added to the word list by the word list creating section (3), and the end of the sentence is reached.
The morphological analysis processing ends.

[Second Embodiment] Next, the processing in the case of the example sentence (b) "Connectable" will be described.

First, the character string “connectable” is processed by the dictionary lookup unit (1). As a result, as shown in FIG. 8, “connect” (noun) and “connect” (The stem of the verb “connect”) is obtained, and the process proceeds to the word selection unit (2).

Since the "word candidate list" is not empty (step A), it is checked whether the word "connection" having the longest notation length satisfies the condition of the candidate word deletion condition table (FIG. 3). If not, it does not match (step C). next,
Since there are a plurality of words (“connect” (noun) and “connect” (verb)) having the longest notation in the candidate word list (step E), the process proceeds to step G.

As for the inflection type of the word having the longest notation in the candidate word list, "connection" (noun) is "no conjugation" and "connection" (verb) is "stem". Here, referring to the utilization type priority table (FIG. 4),
The priority of “no use” is “1”, and the priority of “stem” is “3”, so the priority of “stem” is higher. In this example, since there is no word having the highest usage type with the highest priority among the words having the longest notation length (step G), the word “connect” having the highest usage type is used. Is used as a selected word, and the word selecting section (2) is terminated (step F). Next, the selected word “connect” is added to the word list by the word list creation unit (3), and the process returns to the dictionary lookup unit (1).

Next, as a result of processing the remaining character string "possible" by the dictionary lookup unit (1), "possible" (auxiliary verb) is obtained as a "candidate word list" as shown in FIG. Then, the process proceeds to the processing of the word selection unit (2). In the dictionary lookup unit (1), the noun “possible” is also looked up in the dictionary,
The immediately preceding word “connection” (the stem of the verb “connect”) is determined to be unconnectable, and is not added to the candidate word list.

Since the "word candidate list" is not empty (step A), it is checked whether the word "possible" having the longest description matches the condition of the candidate word deletion condition table (FIG. 3). In this case, it does not fit (step C). Next, in the candidate word list, the word having the longest notation length is only “possible” and does not exist (step E).
The word “possible” having the longest description in the candidate word list is selected as a selected word, and the word selecting unit (2) is terminated (step F). Next, the selected word “possible” is added to the word list by the word list creation unit (3), and the ending of the sentence is reached.

[Embodiment 3] Next, the processing in the case of the example sentence (c) "Three handles" will be described.

After the selection of the words up to “3” is completed, the remaining character string “I will handle this book” is processed by the dictionary lookup unit (1). As a result, as shown in FIG. As a "list", "book" (noun), "book" (prefix), and "book" (numeric) are obtained, and the process proceeds to the word selection unit (2).

Since the "word candidate list" is not empty (step A), it is checked whether the word "book" having the longest description matches the condition of the candidate word deletion condition table (FIG. 3). If not, it does not match (step C). Next, there are a plurality of words ("book" (noun), "book" (prefix), and "book" (numerical number)) in the candidate word list that have the longest expression length (step E). Then, go to step G.

The use type of the word having the longest notation in the candidate word list is “no use” for any “book”.
And the priority corresponding to the utilization type is the same. Therefore,
Since there are a plurality of words having the highest usage type among the words having the longest notation length (step G),
Move to step I. In Step I, referring to the part-of-speech priority table (FIG. 5), a word having the part-of-speech having the highest priority among words having the highest usage type is set as a selected word. According to the part of speech priority table (FIG. 5), the priority of the noun is “1”, the priority of the prefix is “2”, the priority of the classifier is “3”, and the priority of the classifier is the highest.
Therefore, in step I, the word selector (2) ends with the classifier "book" as the selected word. Next, the word selection unit (3) adds the selected word "book" (the classifier) to the word list, and returns to the dictionary lookup unit (1).

Next, as a result of performing the processing of the dictionary look-up unit (1) on the remaining character string “handle”, as shown in FIG. 11, “handle” (noun) and “handle” as “candidate word list” "Handling" (conjunctive form of the verb "handle") "take" (conjunctive form of the verb "take") is obtained, and the process proceeds to the word selection unit (2).

Since the "word candidate list" is not empty (step A), it is checked whether the word "handling" having the longest notation length satisfies the conditions of the candidate word deletion condition table (FIG. 3). If not, it does not match (step C). Next, in the candidate word list, there are multiple words with the longest notation length ("handling" (noun) and "handling" (verb))
Since it exists (step E), the process proceeds to step G. The usage type of the word having the longest description in the candidate word list is “handling” (noun) is “no usage”, and “handling” (verb) is “continuous”. Referring to the utilization type priority table (FIG. 4), the priority of “no utilization” is “1” and the priority of “continuous type” is “2”, so that the priority of “continuous type” is higher. . In this example, since there are no words having the highest usage type having the highest priority among the words having the longest notation length (step G), the word “handle” having the highest usage type is used. As the selected word, the word selecting section (2) ends. Next, the selected word "handle" is added to the word list by the word list creation unit (3), and the process returns to the dictionary lookup unit (1).

Next, as a result of processing the remaining character string "mas" by the dictionary lookup unit (1), "mas" (auxiliary verb) is obtained as a "candidate word list" as shown in FIG. Then, the process proceeds to the word selecting unit (2).

Since the "word candidate list" is not empty (step A), it is checked whether the word "mas" having the longest description matches the condition of the candidate word deletion condition table (FIG. 3). In this case, it does not fit (step C). Next, since the word with the longest notation in the candidate word list is only "masu" (step E), the word "masu" with the longest notation in the candidate word list is selected. Then, the word selection unit (2) is terminated (step F). Next, the selected word "mas" is added to the word list by the word list creation unit (3), and the ending of the sentence is reached.

[0060]

As described above, according to the present invention, in the word selection process in the morphological analysis of a Japanese sentence by the longest match method, an error in the division position caused by selecting the word having the longest notation length is eliminated. In the case where there are a plurality of candidate words having the longest notation length, it is possible to obtain an effect that a correct word can be selected as compared with the related art.

[Brief description of the drawings]

FIG. 1 is a block diagram of a morphological analysis process embodying the present invention.

FIG. 2 is a diagram illustrating a flowchart of an operation of a word selection unit.

FIG. 3 is a diagram showing a candidate word deletion condition table.

FIG. 4 is a diagram showing a utilization type priority table.

FIG. 5 is a diagram showing a part of speech priority table.

FIG. 6 is a diagram showing the contents of a candidate word list.

FIG. 7 is a diagram showing the contents of a candidate word list.

FIG. 8 is a diagram showing contents of a candidate word list.

FIG. 9 is a diagram showing contents of a candidate word list.

FIG. 10 is a diagram showing contents of a candidate word list.

FIG. 11 is a diagram showing the contents of a candidate word list.

FIG. 12 is a diagram showing contents of a candidate word list.

[Explanation of symbols]

 1 ・ ・ ・ Dictionary lookup unit 2 ・ ・ ・ Word selection unit 3 ・ ・ ・ Word list creation unit 4 ・ ・ ・ Word dictionary 5 ・ ・ ・ Connection table 6 ・ ・ ・ Candidate word deletion condition table 7 ・ ・ ・ Usage type priority Degree table 8 ・ ・ ・ Part of speech priority table

Claims (2)

(57) [Claims] In a morphological analysis of a Japanese sentence using a longest matching method for selecting a word having the longest notation length from dictionary-searched candidate words, a notation character string of the candidate word and the candidate word A candidate word deletion condition table that describes an incorrect combination with the following character string is provided, and when the word having the longest notation matches the condition in the candidate word deletion condition table, the word is removed from the candidate word. A morphological analysis method for Japanese sentences having candidate word deletion processing means to be deleted. 2. A morphological analysis method for a Japanese sentence using the longest matching method, wherein a utilization type priority table in which a utilization type of a word and a priority corresponding to the utilization type are paired; If there are a plurality of words having the longest lengths in the above, refer to the inflection type priority table and select the inflection type having the highest priority from the words having the longest notation length. A morphological analysis method for Japanese sentences that has a utilizing word selection processing means for selecting words to be possessed.