JPH0345421B2 - - Google Patents

Description

[Detailed Description of the Invention] [Field of Application of the Invention] The present invention relates to an analysis method for analyzing language, and particularly to a language analysis method that efficiently analyzes sentences written in natural language. .

[Background of the invention]

In conventional language analysis methods, such as the method discussed in Chapter 3 of Makoto Nagao's "Machine Processing of Language" (1984), processing is performed on a sentence-by-sentence basis, even when multiple sentences are analyzed in sequence. , the analysis results up to the preamble were not used, and even if a clause that was the same as a clause that had already appeared appeared, all the analysis processing, such as division into words, had to be performed from the beginning. Therefore, when a plurality of sentences are sequentially analyzed, processing means proportional to the number of sentences are required, resulting in a problem of poor processing efficiency.

[Purpose of the invention]

An object of the present invention is to solve the problems of the conventional method and to provide a language processing method that improves analysis efficiency when sequentially analyzing a plurality of sentences.

[Summary of the invention]

The language processing method of the present invention stores analysis results in a storage medium in a language processing device that analyzes or translates natural language, and when analyzing the next sentence, performs an analysis process that uses the analysis results up to the previous sentence. The above objective is achieved by partially omitting .

[Embodiments of the invention]

Hereinafter, one embodiment of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram of an automatic translation system from a first language to a second language, which is a language processing system according to an embodiment of the present invention. here,
For convenience, the first language will be Japanese and the second language will be English. As shown in FIG. 1, the language processing system according to the present invention includes a processing device 1, a processing program, a storage medium [1] 2 in which analysis results are stored, a storage medium [2] 3 in which dictionary data is stored, It is composed of a storage medium [3] 4 in which input text to be processed is stored, a display device 5, and a keyboard 6. This language processing system sequentially extracts sentences in the storage medium [3] 4, translates them using the dictionary data in the storage medium [2] 3, and outputs the results to the display device 5.

Figures 2, 3, and 4 show the flow of the analysis method according to the present invention.

Figure 5 shows an example of an input sentence, Figure 6 shows an example of stored analysis results, and Figure 7 shows an example of word division processing of an input sentence (a is division into substrings, b is word division). FIG. 8 is a diagram showing another storage format of the analysis results. As shown in Figures 6 and 8, in this example, the analysis result is that the sentence clause is stored as a unit of the auxiliary verb string, and the stored information is expressed as the in-sentence representation of each unit (hereinafter referred to as the header). This is dictionary data such as character strings), the in-sentence notation of the words that make up each unit, parts of speech, and conjugation information (conjugations, types of conjugations), etc. In addition, the 6th,
The code consisting of two English letters in Figures 7 and 8 is a part-of-speech code.

In order to automatically translate Japanese into English, the first step is to divide the solidly written sentences with kanji and kana into words. As an embodiment of the present invention, this word division process is described in the second,
This will be explained according to Figures 3 and 4.

It is now assumed that the sentences shown in FIG. 5 are stored in the storage medium [3] 4, and the analysis process has just finished for the first sentence. At this point, the analysis result of the first sentence is stored on the storage medium [1] 2 in the format shown in Figure 6.
is stored in Then, analysis processing of the second sentence is executed (101).

At the beginning of analysis, the sentence to be processed is separated into subsequences that can be used as stored analysis results and subsequences that cannot (102). Specifically, a substring that matches the heading character string of the stored analysis result is regarded as a usable substring of the analysis result. As a result, the second input sentence becomes as shown in FIG. 7a.
However, the shaded part is the available subsequence of the analysis result. Next, among the unprocessed subsequences in the separated subsequences, pick up the first one (this is called subsequence a) (103), and if subsequence a is a subsequence for which analysis results can be used, (104), the process (105) in FIG. 4 is executed, and if not, the process (106) in FIG. 3 is executed. After repeating this process until there are no unprocessed subsequences (107), the analysis results (the analysis results of FIG. 7a and FIG. 7b) are stored in the storage medium [1]2 (108). To store the analysis results,
This is done with a clause or auxiliary verb string as one unit, but storage processing is not performed for units of analysis results that have the same heading literary string. The above process is repeated until there are no more unprocessed sentences (109).

Next, the process will be explained with reference to FIG. This process is for subsequences for which stored analysis results are not available, and first,
Word extraction based on the principle of longest match (this word is W) is performed for the substring (201). For example, "and" is cut out for the first subsequence in FIG. 7a. However, if the extracted word is a word with a conjugation, the inflection is also included in the extraction.
Next, it is checked whether the extracted word can be connected to the preceding word (202). If it is possible to connect,
Words after the extracted word are extracted using the longest match principle (203), and the possibility of backward connection is checked using that part of speech as a criterion (204). If connection is possible, the word is certified (205). If forward connection or backward connection is not possible, the extraction and recognition of word W is rejected, another word is extracted from the same subsequence (207), and the check for forward connection possibility is restarted. Also,
If the process in 207 is impossible, the extraction and recognition of the word immediately before word W is rejected, and the process in 207 is redone (208). However, if the word W is the first word in a sentence, the word division process is assumed to have failed because the process cannot be redone. This process is repeated until there are no more unparsed character strings (206), and the process ends. Regarding the first subsequence in Figure 7a, by the process, "and" is a conjunction, and ","
is interpreted as a comma.

The processing will be explained with reference to FIG.
This process is for subsequences for which stored analysis results are available. First, the possibility of forward connection is checked based on the part of speech of the first word of the subsequence obtained from the stored analysis results (301). Regarding the second subsequence in FIG. 7a, checking the connectability of "," and "Taro" corresponds to the process 301. If it is possible to connect, the word is extracted from the subsequent subsequence using the longest match principle, its part of speech information is obtained, and the possibility of backward connection with the last word of the subsequence is checked (303). Regarding the second subsequence in Figure 7a,
The possibility of connecting “ha” and “chestnut” will be checked. If the connection is possible, the process ends. However, if the forward connection is not possible, the analysis result corresponding to the subsequence is rejected and the process is executed (304). In addition, if backward connection is not possible, reject the analysis result corresponding to the last word of the subsequence,
(305).

If it is not possible to store all the analysis results due to storage medium capacity limitations, you can prioritize the analysis results based on the time they were obtained and ensure that new analysis results are always retained. Processing efficiency is improved compared to the case without .

Furthermore, if the analysis results are stored in the format shown in Figure 8, that is, if the analysis results are expressed using information indicating the storage addresses of information that should become the constituent elements of the analysis results, then the same element can be , there is no need to secure a single storage area, the storage efficiency of analysis results is improved, and the overall processing efficiency is also improved.

〔Effect of the invention〕

The embodiments of the present invention have been described above, but according to the present invention, it is possible to omit the analysis process for the same character string, so it is possible to improve the efficiency of the analysis process for a plurality of sentences. In particular, when analyzing sentences with many repeated expressions or sentences with a habit of ending sentences,
You can get a big effect.

[Brief explanation of drawings]

Figure 1 is a block diagram of the entire language processing system according to the present invention, Figures 2, 3, and 4 are diagrams showing the flow of analysis processing according to the present invention, Figure 5 is a diagram showing an example of an input sentence, and Figure 6 is a diagram showing the flow of analysis processing according to the present invention. Diagram showing an example of stored analysis results,
FIG. 7 is a diagram showing the execution of word division, and FIG. 8 is a diagram showing another type of storage format for the analysis results. 1...Central processing unit, 2...Storage medium [1],
3...Storage medium [2], 4...Storage medium [3], 5
...display device, 6...keyboard.

Claims (1)

[Claims] 1. A language analysis method for analyzing a sentence containing kanji and kana, which comprises means for inputting a sentence containing kanji and kana, means for dividing the sentence containing kanji and kana into its constituent words, and words divided by the dividing means. A phrase or an auxiliary verb string in which at least one of the connectable words immediately before and after the word is connected as a headword, and means for storing information regarding the word and the connectable word; means for separating a partial character string that matches the headword stored by the storing means from the continuously inputted sentence containing Kanji and Kana; dividing the remaining character string other than the character string into words constituting the remaining character string by the dividing means, and storing the predetermined information by the storing means based on the division result. A language analysis method characterized by: