JPH0658674B2 - Blank generation method when generating translated text - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Outline] A blank generation method for generating a blank when a translated sentence is generated. A blank is generated by referring to a blank generation table based on an adjacent number extracted from an adjacent morpheme of a translated sentence. Information to determine whether or not to generate a blank by simple control, and to create a blank between adjacent morphemes and a dictionary created in advance by assigning adjacent numbers to morphemes. And a blank generation table that stores in advance, for the morphemes generated by morphological / syntactic analysis of the input sentence, the dictionary of the language to be translated is searched, the corresponding morphemes and adjacent numbers are extracted, and after translation, The adjacent numbers are extracted from the adjacent morphemes of the sentence and the blank is generated by referring to the blank generation table based on these numbers. It is configured to insert a blank between the translated morphemes.

[Industrial application field]

The present invention relates to a blank generation method that generates a blank when a translated sentence is generated.

[Conventional technology]

Machine translation and Q & A systems need to automatically produce natural English sentences. In the past, in order to create a sentence, it was necessary to consider the morphemes (stems and inflection endings) that compose a word, as well as the blanks between the words, and output the blanks.

[Problems to be Solved by the Invention]

Therefore, there is a problem that the control becomes complicated. In addition, since it is determined whether to put a blank before a word or after a word based on a grammatical rule, there is a problem that a blank is forgotten or overprinted. Further, there is a problem that a grammatical rule for blank judgment is required, and the number of grammatical rules in a portion for generating a translated sentence such as English increases.

An object of the present invention is to determine whether to generate a blank by referring to a blank generation table based on an adjacent number extracted from an adjacent morpheme of a sentence after translation, and to generate a blank by simple control. I am trying.

[Means for solving the problem]

 Means for solving the problems will be described with reference to FIG.

In FIG. 1, a morpheme generation unit 3-1 generates a translated morpheme by referring to the dictionary 4. At this time, the adjacent number also given to the morpheme is taken out from the dictionary 4.

The blank generation unit 3-2 refers to the blank generation table 5 and performs blank generation.

The dictionary 4 is created in advance by assigning adjacent numbers to morphemes.

[Action]

According to the present invention, as shown in FIG. 1, the morpheme generator 3-1 generates a morpheme translated by referring to the dictionary 4, extracts the adjacent number from the dictionary 4, and the blank generator 3-2 translates the morpheme. The blank generation table 5 is referred to on the basis of the adjacent number between them, it is determined whether or not a blank is generated between the morphemes, and the blank is inserted.

Therefore, it is possible to insert a blank between post-translational morphemes by a simple control of determining the occurrence of a blank by referring to the blank generation table 5 based on the adjacent number of the post-translational adjacent morpheme.

〔Example〕

Next, the occurrence of blanks when translating Japanese into English will be sequentially described in detail with reference to FIGS. 1 to 3.

In FIG. 1, an input unit 1 is for inputting a sentence to be translated (input sentence) such as Japanese. For example, the input sentence “He eats an apple.” Is input.

The Japanese analysis unit 2 is for morphologically dividing the Japanese input from the input unit 1 (FIG. 2), parsing the Japanese (FIG. 2), and further generating a concept network (FIG. 2).

The English sentence generation unit 3 is composed of a morpheme generation unit 3-1 and a blank generation unit 3-2, and generates an English sentence corresponding to the concept network analyzed by the Japanese analysis unit 2.

The morpheme generator 3-1 is a conceptual network (Fig. 2).
The subject, the verb, and the like are sequentially taken out, and a morpheme translated into English is generated by referring to the dictionary 4 (FIG. 2). At this time, the adjacent number (FIG. 2) assigned to the English morpheme is also taken out and stored.

The blank generation unit 3-2 generates a blank between the morphemes by referring to the blank generation table 5 from the post-translational morpheme sequence generated by the morpheme generation unit 3-1 based on the adjacent number of the adjacent morphemes. It is determined whether or not to perform, and a blank is inserted when it is determined to generate. (Fig. 2,).

An English sentence composed of these generated morphemes and blanks after translation is printed on paper by a printer or displayed on a display (English output).

Next, the operation of the configuration shown in FIG. 1 will be specifically described according to the order shown in FIG.

In FIG. 2, is divided into morphemes. It is typed in Japanese, as shown on the right. He eats an apple.

Is divided into morphemes (words) as underlined.

Are put together into clauses by parsing. This means that the morphemes that have been morpheme-divided by are grouped into clauses as shown.

Analyzes the relationship between the subject and the object from the dependency.

Generates a concept network. Based on the result of analyzing the dependency of, a network is created with the concept of a Japanese input sentence as shown in the figure, centering on the verb "eat", subject "he", object "apple", etc. It means to do.

Generates a morpheme translated into English. This means that, for example, the subject "he" is extracted from the Japanese concept network, and the corresponding English morpheme "He" is extracted by referring to the dictionary (English dictionary) 4.

Retrieves the adjacent number. This is because, for example, when an English morpheme “He” is taken out from the dictionary 4, the adjacent number (L,
R) is taken out.

Refers to the blank generation table 5 and determines "presence" or "absence" of occurrence of blank as indicated by an arrow. This is a dictionary in English (English dictionary) 4
The right adjacent number (R) of the adjacent preceding morpheme (word) in the morpheme string extracted from the row index (R
C) and the left adjacent number (L) of the morpheme (word) to be added later is used as a column index (LC), and the RC and LC are used to refer to the blank generation table 5 in FIG. This means that “none” is determined as indicated by the arrow in the figure.

Produces a translated sentence. This means that a blank is inserted between the morphemes determined to be “present” in the morpheme sequence generated in (3) to generate a translated sentence as illustrated.

Next, the procedure for generating a blank in the translated text of FIG. 2 will be specifically described using the blank creation table 5 of FIG.

(1) "He" at the head is a copy of the morpheme "He" shown in FIG.

(2) The adjacent number R “4” of the morpheme “He” and the adjacent number L “6” of the next morpheme “eat” are used as the matrix index, and “⊚” (indicates that there is a blank) shown in FIG. Find out. This "◎" indicates that there is a blank, and a blank is inserted after He in the translated text.

(3) The second "eat" is the morpheme "ea" of FIG.
It is a copy of "t" as it is.

(4) The adjacent number R "26" of the morpheme "eat" and the adjacent number L "23" of the next morpheme "s" are used as matrix indices, and "○" (representing no blank) shown in FIG.
Find out. By this "○", there is no blank, and no blank is inserted between eat and s of the translated sentence.

(5) The third “s” is a copy of the morpheme “s” shown in FIG. By this, ea in (4)
Since no blank is inserted after t, the result is
It becomes "eats".

(6) The adjacent number R “4” of the morpheme “s” and the adjacent number L “6” of the next morpheme “apple” are used as matrix indices, but since they are the same as in (2), Insert a blank between s and apple.

(7) Similarly, the translated sentence of FIG. 2 is generated.

FIG. 2 shows an example of a blank generation table. R on the vertical axis in the figure
C represents the row index number, and LC on the horizontal axis represents the column index number. Morphological types are assigned to these numbers, respectively, as shown in the figure. "◎" in the figure
And "●" means that there is a blank, and "○" means that there is no blank.

〔The invention's effect〕

As described above, according to the present invention, the blank generation table 5 is based on the adjacent numbers of the translated adjacent morphemes.
Since the configuration that determines the occurrence of blanks is adopted by referring to, it is possible to quickly determine whether or not to insert a blank between translated morphemes (words) with a simple control called table lookup. it can. This eliminates the need for the conventional grammar rules for blank generation in the translated sentence generation unit, reduces the number of grammar rules for translated sentence generation, and generates blanks by referring to the grammar rules. In this case, it is possible to prevent the blank from being left out or the blank being put out too much.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an embodiment of the present invention, FIG. 2 is an operation explanatory diagram of the present invention, and FIG. 3 is an example of a blank generation table. In the figure, 2 is a Japanese analysis unit, 3-1 is a morpheme generation unit, 3-
2 is a blank generation unit, 4 is a dictionary, and 5 is a blank generation table.

Claims (1)

[Claims] 1. A blank generation method for generating a blank when a translated sentence is generated, a dictionary (4) created in advance by assigning adjacent numbers to morphemes, and information on whether or not to generate a blank between adjacent morphemes. A blank generation table (5) in which is stored in advance, and for the morphemes generated by morpheme / syntactic analysis of the input sentence, the dictionary (4) of the language to be translated is searched to find the corresponding morphemes and adjacencies. Extract the numbers, extract the adjacent numbers from the adjacent morphemes of the translated sentence,
Based on these, it is determined whether or not a blank is generated by referring to the blank generation table (5), and a blank is inserted between the morphemes after translation. Blank generation method.