JPH0492966A - Numerical quantity expression processing system - Google Patents

Abstract

PURPOSE:To accurately identify the structure of a phrase in a text processing, to execute reading through the use of the structure and to accurately decide appropriate expression in the language of an opposite party in translation by arranging and discriminating an ordinal numeral function and a noun function. CONSTITUTION:Since the characteristic and the meaning interpolation of a construction are arranged and described when a Japanese word is related to a numeral in the dictionary of the numeral, the noun and the suffix of the Japanese word, the expression of a numerical quantity can accurately be analyzed and generated by using the dictionary. The examples of the contents in the dictionary are shown in the diagram. Since the corresponding expression in English is arranged and described in the dictionary, it can be made to accurately correspond to the expression of the numerical quantity in English. Thus, a numerical quantity expression processing system has the dictionary where the characteristic and meaning interpolation of the construction when the Japanese word is related to the numerical are described in the dictionary of the ordinal numeral, the noun and the suffix of the Japanese word and the Japanese word including the expression of the numerical quantity is analyzed by using the information in a natural language processing system in which a natural language is set to be input and the structure and meaning are approved. Thus, a Japanese sentence related to the ordinal numeral can accurately be processed. Furthermore, such system is applied to a machine translation system, for example.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method for processing quantitative expressions in natural language.

[Conventional technology]

Japanese, a natural language, requires a particle to count nouns. Counter words are metric words that appear immediately after number words, such as ``tsu'' in ``one'' and ``two,'' and ``-person.''
Typical examples include ``person'' in ``two''.

Until now, there has been no processing method that can uniformly and correctly analyze, identify, and interpret linguistic expressions related to particle words.

[Problem to be solved by the invention]

As mentioned above, there is no processing method that can uniformly and correctly analyze, recognize, and interpret linguistic expressions related to classifiers. This is because Japanese particle nouns have the following properties.

(i) Discrimination between nouns and particle words In Japanese, number words are usually connected to nouns through particle words. For example, when counting animals, we need the particle ``human'', as in ``three dogs,'' and the number word and noun are not directly connected, as in ``three big''. The ``teri'' in ``three'' is a typical particle, and ``teri'' alone does not function as an independent noun. In this way, nouns and particles generally have completely different functions.

However, an ordinary noun is sometimes placed immediately after a number word and used as a metric, such as ``bag'' in ``-bag of tomatoes.'' In order to correctly identify the meaning of this expression and make it correspond to appropriate English, it is necessary to distinguish between the noun function and particle function of ``bag''. If we assume that "bag" in "-bag of tomatoes" is just a normal noun, then in English, according to the word order of normal nouns, rtomatos of a b
The correct English expression for ``-bag of tomatoes'' is ra bag of.
``tomatos'', and as in Japanese, the metric word is placed before the noun with the preposition ``Of'' in between.

Conventionally, it was unclear how to distinguish between particles and nouns,
We were unable to properly handle this phenomenon.

(ii) Correspondence with English In Japanese, ``jin'' is used to count people, and ``hon'' is used to count pencils, but there are no equivalent expressions in English.

It simply connects a number word and a noun directly. That is, "-man" is raIlanJ, and "two pencils" is "rtwo pencils."

As shown in 2, English has less well-developed classifiers than Japanese.

However, when counting paper, just as in Japanese we say '-sheets of paper,' in English we say 'a 5 sheets of paper.'
Use the expression equivalent to the number particle "paper".

In the past, in what cases were there discrepancies between Japanese and English?
Since the issue of what kind of cases should be dealt with has not been sorted out, appropriate dictionary descriptions have not been possible, and a processing method for correctly handling such phenomena has not been fully established.

SUMMARY OF THE INVENTION An object of the present invention is to provide a quantitative expression processing method that eliminates such drawbacks and accurately processes expressions related to classifiers.

[Means to solve the problem]

The present invention uses a natural language processing method that receives natural language as input and identifies its structure and meaning. The feature is that it uses information from the dictionary section to analyze Japanese words that include quantitative expressions.

[Effect]

In the particle processing method of the present invention, Japanese particles, nouns,
Since the suffix dictionary organizes and describes the syntactical features and semantic interpretation of the word when it is related to a numeral, by using the dictionary, quantitative expressions can be correctly analyzed and generated. Figure 4 shows an example of dictionary contents. Since the corresponding English expressions are also organized and described in the dictionary, it is possible to correctly correspond to English quantitative expressions.

In the following, linguistic phenomena related to Japanese classifiers, which form the background of the present invention, will be explained in detail using FIGS. 2 and 3. The dictionary contents of classifier-related words as shown in FIG. 4 are created based on the analysis shown in FIGS. 2 and 3.

First, the types of classifiers will be explained using FIG.

FIG. 2 is a classification diagram of classifiers for explaining the contents of the dictionary of the present invention. The numbers in FIG. 2 represent classification symbols for classifiers. This classification symbol is described in a dictionary of classifiers as shown in FIG.

In FIG. 2, the meaning content of the number particles is first roughly divided into two categories: classification symbol 1, which indicates "abundance", and classification symbol 2, which indicates "order".

The "abundance" expressed by the number particle is the "total amount of a collection of things" in classification symbol 11, classification symbol 12.
``Types that make up the whole'' (e.g. type, street) Classification code 13 ``Attributes of things'' (e.g. cm, age) Classification code 14 ``Frequency of actions'' (e.g. times, circumference, It is divided into four types (shield).

Furthermore, there are three ways to count the "total amount of a collection of things": In other words, the classification symbol 111, "Count the thing itself" (e.g., person, individual, animal, piece), the classification symbol 112, "count in groups" (e.g., group, foot, step, layer), the classification symbol 113, " count using a measuring device” (e.g., cups,
They can be categorized into three types: bags, boxes, and plates).

There are restrictions on how number particles can be used depending on the meaning and appearance of the noun that follows them. For example, the classifier of classification symbol 111 is: (a) Neutral..."tsu" (b) Connotes the shape of a noun..."book""piece""sheet"
1 grain" (c) Semantic classification of nouns..."person""book""table"
It can be seen that there are classes such as ``one leg''.

Similarly, the classification symbol 112 includes: (a) Neutral...'Group' r set' (B) Number of constituent elements...'Pair''Pair' (semantic classification of ten nouns) 'Foot''Pair' (c) Shape of gathering
...There are "heavy", "tier", "layer 31 bundle", etc.

Next, with reference to FIGS. 2 and 3, points that the prior art does not deal with regarding Japanese particle words will be explained.

First, we will explain the point that the prior art does not deal with regarding the above-mentioned (i) "discrimination between nouns and classifiers". Third
The figure shows the classification of words necessary for distinguishing between particles and nouns. In FIG. 3, number word-related words, which had not been clearly distinguished in the past, are classified into six types of word groups based on their positional relationship with number words.

In terms of positional relationship with number words, Consider two positions.

Normally, in Japanese, nouns are connected to number words through number particles. For example, when counting animals, the number word ``ni'' is required, as in ``three dogs,'' and the number word and noun are not directly connected, as in ``three big.'' This is the syntactic property of typical nouns.

On the other hand, ``to'' in ``三人'' is a typical number particle, and is always placed immediately after a number word, with no other elements intervening between it and the number word.

In this way, nouns and particles generally appear in completely different positions. In Figure 3, this difference is shown in the ``ordinary noun'' in the first line and the ``ordinary particle'' in the sixth line. In FIG. 3, the symbol ◯ indicates that it can occur in that positional relationship, and the symbol X indicates that it does not occur in that positional relationship.

Previously, all words that could be placed immediately after a number word were considered to be number words, and there was no clear classification. However, there are some words that behave differently from regular nouns and regular particles. There are four types of such word groups, shown in lines 2 to 5 in FIG. 3. All four types can be placed immediately after a number word.

First, there are words that can be used both as nouns and particles, such as ``bag'' in ``-bag of tomatoes.''

Such a word group is typically found in the classification symbol 113, a noun having the meaning of "container" in the classification shown in FIG. When ``bukuro'' is counted by itself, it is counted as an ordinary noun with a particle, such as ``one bag.''"
- When it comes immediately after a number word, such as "fukuro", "fukuro"
We are not counting ``bag'', but using ``bag'' as a particle and counting the following nouns, such as ``tomato.'' Although there is no difference in the meaning of this group of words when used as a noun or as a particle, there is a difference in their syntactic behavior as quantitative expressions. When placed immediately after a number word,
In other words, when used as a particle, even when translated into English, unlike a normal noun, ra bag
It is placed before a noun via the preposition "rof", as in "of tomatoes". This group of words is shown in the fourth line of Figure 3.

There is also another group of words similar to ``bukuro.'' 2nd in Figure 3
These are words such as ``word'', ``sentence'', ``column r'', and ``row'' shown in the rows. Although these words are nouns like ``bukuro,'' they can be placed immediately after a number word. but,"
Unlike the ``bag'' class, when counting by itself, it can be expressed as ``-word'' without using a ``single word'' and a particle (with the same meaning, without using a particle. In other words, this word group is Although it is placed immediately after a number word, it does not function as a particle.The semantic feature of this word group is that it is itself a component of something larger.When counting the number of nouns, it is generally As in the classification symbol 111 of ``counting itself'' in the classification in Figure 2, counting is done through a particle such as ``one j,''``two,'' but when the noun has the above-mentioned properties, the particle It can be counted without using .

In addition, as another word group, "
There are word groups such as ``book'', ``person'', ``table'', and ``head''. These words are similar to ``bukuro'' (bag) in that they can appear both as nouns and as particles, but their meanings when used as nouns and particles are completely different.

For example, ``hon'' is used as a noun to refer to a book, but as a particle, it is used to count long and narrow nouns and has nothing to do with books. Furthermore, the pronunciation of a word often differs between when it is a noun and when it is a particle.

For example, ``person'' is ``hito'' as a noun, but ``su'' or ``nin'' as a particle.

Furthermore, there is another group of words for which it has been unclear to distinguish them from classifiers, such as t, listed in the third line of FIG. "Magazine" to represent a magazine, "Gakko" to represent a school, and "" to represent a company.
This is the case with the suffixes such as ``sha'' and ``paper'', which refers to newspapers. For example, these words are placed immediately after a number word, such as ``sanshi'' and ``shiko'', and unlike normal nouns, ``shi'' and ``school'' do not function as nouns. . Because of this property, it is often used as a number particle such as ``tsu'' in ``hitsuto'', but it is different from a particle in meaning. It can be said that they are similar to the above-mentioned ``words'' and ``bun'' in that number particles are used to count other nouns, whereas ``shi'' and ``school'' count themselves. .

``Word'', ``bun'', etc. stand on their own as nouns, whereas ``magazine'', ``school'', etc. differ in that they are less syntactically independent; however, they are metrically independent. What they have in common is that they do not function as a word or particle. This can also be seen from the fact that, for example, when the word ``school'' is translated into English, the word ``school'' corresponds to the noun ``r 5chool''.

As mentioned above, the words related to particle numbers, which were unclear in the past, are classified as shown in Figure 3, and are written in the dictionary according to this classification. Can discriminate correctly. An example of dictionary contents is shown in FIG.

Next, regarding (ii) “correspondence with English” mentioned above,
Points not covered by the prior art will be explained.

By categorizing them as shown in Figure 2, we can also organize the correspondence between Japanese and English expressions. Classification symbol 1 has the largest difference in expression.
This is the ``order'' of ``counting things'' classification symbol 2 of 11. Among the classification symbols 111, there are no particles in English that correspond to "person", "individual", etc., which imply that the noun is a discrete individual. However, even with the same classification symbol 111, when a noun refers to a non-discrete thing, the Japanese "-"
rashet of paper
There are words like J that have a particle function. Classification symbol 1
"Group", "Container", "Unit", "Type", "
An expression equivalent to the particle "time" also exists in English. The English particle equivalent expression for "gumi" is r two gr.
The expression corresponding to ``container'' is ``r two kgof'', the English expression corresponding to ``kind'' is ``r two kgof,'' and ``times.'' An example of the equivalent is r two times J.

Another big difference between Japanese and English is the expressions related to "two orders." As is well known, in English, ordinal numbers are used to express order, and in Japanese, a number is superimposed on the base number.

In addition, in Figure 2, the so-called "r unit" is defined as volume, capacity, weight (classification symbol 113) and other attributes (classification symbol 13).
). This is the semantic "total amount of a collection of things"
It is based on the difference between "attributes of the thing itself". In English, this difference in meaning is reflected in the difference in syntax, such as weight, etc., where the quantity comes before the preposition of, as in "r 3kg of sugar," and other attributes, as in ran output of 3011VJ. Uses a syntax in which the quantity is placed at the end.

As explained above, in the present invention, the conventionally unorganized classifier-related words are organized and their behavior is described in the dictionary, so that Japanese sentences related to classifiers can be processed correctly.

〔Example〕

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing one embodiment of the present invention.

The quantitative expression processing method shown in FIG.
It is equipped with

In such a quantitative expression processing method, the morphological analysis unit 1
0 divides the input Japanese sentence into words.

The quantitative expression recognition unit 20 recognizes the structure and meaning of a quantitative expression when a quantitative expression is included in a sentence. At this time, the dictionary contents of the classifier words described in the dictionary 30 are referred to. The processing procedure of this quantitative expression recognition section 20 is shown in FIG.

As shown in FIG. 5, the results of the morphological analysis of the input sentence are obtained (step S1), the dictionary contents of each morpheme are retrieved from the dictionary 30 (step 32), and the part of speech of the word immediately after the number word is determined. Check (step S3).

If this word is a particle (step S4), the classification code is rllllJ, and it is checked whether it is used for a discrete noun (step s5). If r YES J in step S5, the number-quantity-noun A semantic structure is created (step S6). Further, in the case of rNOJ in step S5, a semantic structure of numeric value-quantity-one quantifier-one quantity-noun is created (step S7).

In step S3, if the part of speech of the word immediately after the number word is a suffix that lines up with the number word (step S8), a semantic structure of numerical value - loss amount one suffix is created (step S9).

In step S3, if the part of speech of the word immediately after the number word is a noun that is aligned with number (step 510), a semantic structure of number - loss amount one suffix is created (step 5ll).

In step S3, if the part of speech of the word immediately after the number word is neither a particle nor a suffix along with a number word, nor a noun along with a number word, the input sentence is recognized as a syntax (step S3).
12).

In this way, the quantitative expression recognition section 20 creates a semantic structure and passes it to the quantitative expression generation section 40.

The dictionary 30 is connected to the morphological analysis section 10, the quantitative expression recognition section 20, and the quantitative expression generation section 40. This dictionary 30
is a dictionary of Japanese particles, nouns, and suffixes. In addition, a dictionary of Japanese number words, nouns, and suffixes organizes and describes syntactical features and semantic interpretations when the words are related to number words. An example of dictionary contents is shown in FIG. The contents of the dictionary in FIG. 4 are described according to the classification of the present invention shown in FIG. 2 and FIG. 3. Figure 4 shows Japanese surface words, their parts of speech, semantic functions, and the corresponding English parts of speech and surface. Classification codes are assigned to the semantic functions of the classifiers according to the classification shown in Figure 2.

The quantitative expression generation section 40 receives the semantic structure of the quantitative expression certified by the quantitative expression certification section 20. Quantitative expression generation unit 40
outputs English using the information on the English side of the dictionary contents.

The procedure of such processing in the quantitative expression generating section 4o is shown in FIG.

As shown in FIG. 6, the semantic structure is received from the quantitative expression recognition unit 20 (step 521), and the dictionary 3° is extracted (
Step 522), and check whether a particle is included (Step 523). If rYESJ is determined in step s23, the words are generated in the order of number word, particle noun, preposition Of, and noun (step 524). If "NO" in step S23, the words are generated in the order of numerals and nouns (step 525).

In this way, the quantitative expression generation unit 40 generates and outputs English.

Next, the operation of this embodiment will be explained.

As an example, consider a case where "-book pencil" and "-bag pencil" are input.

First, when "-book pencil" is input, the morphological analysis unit 1
As a result of the analysis using 0, the words are divided into 1/book/of/pencil/. This result is sent to the quantitative expression recognition section 2o. The quantitative expression recognition unit 20 looks up the dictionary 3o.

The dictionary contents for "book" in the dictionary 30 are as shown in FIG. 4, and it is described that there is a classifier and a noun with the same notation, "1 book". Through the process shown in FIG. 5, the quantitative expression recognition unit 20 recognizes that the number word "-" and "hon" are adjacent to each other in this input sentence, so that "hon" is a classifier. Therefore, the quantitative expression recognition unit 20 interprets that the semantic structure of this input sentence is as follows.

This structure represents that the quantity of pencils is "1".

The quantitative expression generation unit 40 that has received this structure generates English based on the information in the dictionary 30 through the processing shown in FIG. The sign "-" is marked in the English dictionary for the particle "r book" in Figure 4, but this means that there is no English word corresponding to the particle "hon", and there is no word corresponding to "hon". means that it is unnecessary in English. Based on this dictionary information, the quantitative expression generation unit 40 does not produce a word equivalent to ``one pencil'', but instead outputs a word equivalent to ``one pencil'' or a penc.
Generate il.

Next, a case will be described in which "-bag of pencils" is input. As in the case above, 1/bag/of/pencil/ is morphologically analyzed. When we look up the dictionary 3° in the quantitative expression recognition unit 2o, we find that the dictionary for ``1 bag'' has two types: nouns and classifiers, as shown in Figure 4. Quantitative expression certification part 2
Since the number word "-" and "bukuro" are adjacent to each other in the input sentence, "0" is recognized as a particle word. As shown in Figure 4, from the semantic function column of the dictionary, the particle word ``bukuro'' is [
You can see that it is used as a measuring device when "counting using a measuring device." Therefore, the following semantic structure is created.

This structure allows the quantity of pencils to be peeled off and used as a measuring device.
It represents IJ. The quantitative expression generation unit 4o, which has received this semantic structure, looks up the dictionary 3o and learns that the English word corresponding to "bukuro" is "number noun j.In the case of English, the word order of the number noun is There are two types of expressions, and in the case of the expression 13 "Referring to the attributes of the thing itself" in the classification in Figure 2, for example, ran output of 3011
As in 1VJ's "301Ilv", the number noun is a preposition "
However, in the case of classification 113 "Counting using a measuring device" in Figure 2, the arrangement is reversed and the fractional noun comes before the noun. In this case, the dictionary information shows that "bag" is the latter, so the quantitative expression generation unit 4 places "r bag" corresponding to "bag" in front of the noun, and converts "a bag of pencils" into "a bag of pencils". Output.

In this way, the quantitative expression processing method of this embodiment is a natural language processing method that receives natural language as input and identifies its structure and meaning. It has a dictionary that describes the syntactical features and semantic interpretation of related cases, and uses this information to analyze Japanese words that include quantitative expressions.

As a result, in this embodiment, the conventionally unorganized classifier-related words are organized and their behavior is described in the dictionary, so that Japanese sentences related to classifiers can be processed correctly. Further, such a method is applied to, for example, a machine translation system.

(Effects of the Invention) As explained above, the present invention organizes and distinguishes particle functions and noun functions, thereby enabling accurate identification of phrase structures in text processing, reading using them, and target language recognition in translation. Students will be able to accurately determine appropriate expressions for

[Brief explanation of drawings]

Figure 1 is a block diagram showing one embodiment of the present invention, Figure 2 is a diagram for explaining the classification of classifiers used in the present invention, and Figure 3 is a diagram for explaining the distinction between classifiers and nouns. FIG. 4 is a diagram showing an example of dictionary contents such as classifiers, nouns, and suffixes described in the dictionary 3o in FIG. 1, and FIG. Flowchart for explaining the operation of FIG. 6 is a flowchart for explaining the operation of the quantitative expression generation section 40 in FIG. 10...Morphological analysis unit 20...Quantitative expression recognition unit 30...Dictionary 40...Quantitative expression generation unit

Claims (1)

[Claims] (1) In a natural language processing method that receives natural language as input and identifies its structure and meaning, a dictionary of Japanese parts of speech contains a dictionary section that describes syntactic features and semantic interpretation when each part of speech is related to a number word. A quantitative expression processing method characterized in that it analyzes Japanese language containing quantitative expressions using information from the dictionary section.