JPS62184572A - Retrieving system for dictionary of cooperative compound word in word division device - Google Patents

Abstract

PURPOSE:To enable a cooperative compound word to be retrieved, when a character string segmented from an input characater string coincides with the first entry, and the second entry thereafter are present, by retrieving whether the character string represented at the entry is present in a remaining character string or not. CONSTITUTION:An input character string 1 is inputted to a word division device 2, and a word division is performed referring to a dictionary 3, and a result is outputted as a word division output 4. At the dictionary 3, as for the coopera tive compound word, a cooperative compound word 31 in which a word that cooperates with the second entry thereafter other than the first entry is designat ed, is posted, and when the character string coincides with the first entry, and the second entry is present, and the word designated at the second entry is present in the input character string, the cooperative compound word is added in a division candidate word. In this way, the translation of a text including the cooperative compound word can be performed exactly and rapidly.

Description

[Detailed description of the invention] 〔table of contents〕 overview Industrial applications Conventional technology Problems that BA is trying to solve Means to solve problems action Example Effect of the invention 〔overview〕 Word segmentation of an input sentence in machine translation, etc.

In word segmentation of an input sentence containing a compound @ (consonant compound word) that has a meaning different from the original meaning of the word due to the consonance of distant single-chain comrades, a dictionary in which the heading of the consonant compound word is composed of multiple entries. established.

When a consonant candy is specified in the dictionary, it is now possible to look up the dictionary contents corresponding to the consonant compound word by checking whether the consonant candy is in the input string. What has been done 1. [Industrial Application Field] The present invention relates to a dictionary search method for vocative compound words in a simple division device during zero machine translation.

When performing machine translation using an automatic translation device.

The sentence to be translated is divided into words, a rate @i is attached to each divided word, and the sentence is analyzed based on that. In such machine translation, when zero sentences are analyzed based on the output results of a simple segmentation device, it is necessary to create consonant compound words (consonance between distant words that have a meaning different from the original meaning of the word) in the input sentence. When there is @), it is necessary to search the dictionary contents corresponding to the meaning as a consonant compound word, not in the dictionary 1'3 valley of each word that makes up the word, and there is an efficient search for this purpose. A method is required.

[Conventional technology]

The basic method of conventional single-segmentation devices is to sequentially cut out parts of an input character string that match dictionary entries from the beginning, and at the same time add the contents of the dictionary.

Figure 6 shows an overview of this conventional single division device and dictionary.
It is a figure shown as an example used for English candy → Japanese translation. O6
In the figure, 51 indicates an input character string, 52 a single division device, 53 a dictionary, and 54 a word division output.

Dictionary 53 has each word and its corresponding translation.
Mature candy such as pr give up J is entered. In this case, "give up J" means input string 5.
"I will give it up 5o" shown in 1.
In the word "on J", it is placed between "itJ" and has the meaning of "giving up" in response to each other, but in this case, as shown in each word output 54, the meaning as a consonant compound word is Instead, only the dictionary 1'3 valleys as individual words forming the consonant compound word #f can be obtained.

In the end, if there is an IJ in the same character string but different dictionaries, or if there is an IJ at the beginning, the same character string has different lengths)
There may be IJ, so if there are multiple candidates,
All possible divisions are shown when a single image is selected using an appropriate evaluation function or a connection check with previous and subsequent candidates.

In any case, r give it up J
In the case of a consonant compound word such as ``give upJ,'' it is impossible to accurately translate it using the conventional division method using a single division device. The response processing is performed by

[Problems that the invention is supposed to solve]

As mentioned above, in the case of the conventional device, 9-sentence analysis was used to translate vocative compound words, but this had the problem of increasing the burden of sentence analysis.

In this case, there is a problem in that the contents of the dictionary become complicated because the information on the consonant compound word needs to be included in the information on the original word.

This invention was made in order to improve such problems, and it is a simple configuration that can search for an appropriate translation of a vocative compound word without using 9-sentence analysis. The purpose is to provide a dictionary search method for words.

[Means for solving problems]

In order to solve the above-mentioned problems, in the present invention, O1
As shown in the figure, an input character string 1 is input to a character string division device 2, and is divided into words while referring to a dictionary 3, and the result is outputted as a word division output 4.

In the dictionary 3, in addition to the first entry word, a consonant compound word 31 in which consonant words are specified from the second entry onward is described.

If the first entry matches, and if the word specified in the second entry and subsequent entries is found among other input characters, the consonant compound word is determined to be a split candidate single chain.

[Effect]

This makes it possible to accurately and quickly translate sentences containing complex combination bets.

[Embodiment] Next, the operation of an embodiment of the present invention will be described with reference to FIGS. 2, 3, and 4.

Figure 2 is a diagram showing part 9 of the search engine for ordinary words and vocative compounds in a dictionary, and Figure 3 is.

Fig. 4 shows an algorithm for searching all division candidate words including the character string from the beginning of the input character string (partial sentences that are not single-stranded). The progress of searching for division candidates for up 5oon J is shown.

In the dictionary 3, as shown in FIG. 2, for consonant compound words, in addition to the first entry word, consonant words from the second entry onwards are listed. for example.

[For give J, see “Entry section 21 when using give J alone.
Entry section 22 of iveaway J. It also consists of an entry section 23 for "give up J" as a consonant compound word and an entry section 24 for the idiom "givein J."The number placed before the word in each entry section indicates the number of characters in the word that follows. Therefore, rgive
In the J-only entry section 21, only "4" is written in front of r give J, and rOJ is placed after it, indicating that there is no entry after "give J." On the other hand, in the case of vocative compound words, as shown in entry sections 22.23, "4" is placed before the first entry, give, and "4" is placed before the second entry, away or up, respectively. r4J and r2J indicating the number of characters are placed respectively.

In addition, as shown in the entry section 24, the idiom ``half'' is entered in such a way that the space between words is included in the number of characters. Therefore, in the case of ``give in J'', the first number is ``7''. It is filled in.

String, rIw item 1 give it up 5oon
The search method of the present invention will be fully explained using an example of Jf. "
Since it is the same as the conventional method up to IwillJ, r
I will explain what happens after give it up J. First, the partial string length (including spaces) from g to p.
1Oino, and set the search character length n=o (step ■).

Next, set the search character length H=t, and after confirming that n is within Ho (step ■), n:1, that is.

Go to search to see if "g" is entered in the dictionary.

At this time, "g" is searched, but "g" does not have a second entry (see Dictionary 3 in Figure 1), so (step ■)
It is registered as is as a 6-division candidate (step ■).

Then, I go back to step ■ again, but there is no entry corresponding to the length of 1 character.

Return to step (2) and increase the character length by +1. n
If n is 2 or 3, there is no corresponding entry for ``giJ'' or ``girJ.'' When n becomes 4, entry section 2 is
1 "givetJ" is searched. This is because it does not have a second entry (step ■).

It will be added as a candidate word. For n = 4, [givezJ in the entry section 22 is further searched. This has rawayJk as the second entry, so
Search for "away J'e" from other input strings, but it does not exist, so it is not a candidate word (step ■)
.

Next, r give s J in the entry section 23 is searched.

Since there is also a second entry rupJ, the process moves to step (2) and a search is made to see if rupJ exists in the input character string. At this time, in the input string.

rupJ exists, so it is added as a candidate word through steps ① and ②.

Thereafter, there is no corresponding entry for n equal to 4, and there is no corresponding entry for n equal to or greater than 5, so the candidate that includes the character string from the beginning of this partial character string is "g".

There are three: r give J and r give up J. this.

Figure 4 summarizes the results.

Among the candidates selected in this way, “g” is the following “iv”
It was dropped because it did not connect with the cut-out word candidate of "e", and t
Ta. Using the longest match method, which is the best known evaluation method, r give J (character length 4) and si ``give
up J (character length 6) is preferentially selected.

FIG. 5 shows an embodiment of the single division device used in the dictionary search method according to the present invention.
IQ, dictionary 301 division result storage section 40° character string storage section 5
Configure from 0. OPU 10f': To.

Word division control unit 11. Character string extraction section 12. It has a dictionary search reference section 130 and a division candidate storage section 14.

As mentioned above, the dictionary 30 is registered with the vocative compound @ having a first entry and a second entry.

When a character string is input from the input unit 20, it is stored in the character string storage unit 50. Under the control of the control section 11, one character string is cut out into the minimum unit of quick and dark by the string cutting section 12. This word is compared with the dictionary 30 using the method described above, one division candidate is selected using the method described above, and this is stored in the division candidate storage section 14. Then, apply all evaluation functions such as the longest match method to this candidate.

The results are stored in the division result storage section 40.

In the above description, an example of translation from English to Japanese has been described, but the present invention is of course not limited to these languages.

〔Effect of the invention〕

In the present invention, it is possible to search for vocative compound words, which has not been done with conventional rate word segmentation devices, and it is also possible to describe information on vocative compound words in the dictionary independently of the rate candy that makes up the vocative compound word. Therefore, efficient translation can be performed.

[Brief explanation of drawings]

FIG. 1 is a schematic diagram for explaining the present invention. FIG. 2 is a diagram showing the search entry section of the dictionary of the present invention. FIG. 3 is a diagram showing a search algorithm for dividing candidate words. FIG. 4 is a diagram showing an example of the search progress for split candidate single chains. FIG. 5 is a diagram showing an example of a single-piece dividing device used in the present invention. FIG. 6 is a diagram showing an outline of a conventional single division device and dictionary.

Claims (1)

[Claims] In a dictionary search method in a single division device for machine translation that divides an input character string into words and searches for corresponding translations, 3) is provided, and if the character string cut out from the input character string matches the first entry and there is a second entry or later, it searches for the character string shown in that entry to see if it exists in the remaining input character strings. A dictionary search method for vocative compound words in a word segmentation device, characterized in that: