JPH05225229A - Processing system for expressing degree in natural language - Google Patents

Abstract

PURPOSE:To process a phenomenon called 'implication of conversion,' the semantic relation between a negative sentence and a source positive sentence, the possibility of a question sentence and the answer and the modification possibility of a degree modifier adverb, for example, or the like on a computer by exactly describing and calculating the language action of degree expression. CONSTITUTION:The meaning of a certain degree expression is expressed as a discrete relative position in a frame formed by one group of expressions paired with that expression. When expressing the meaning of a certain degree expression, both of ideographic meaning and meaning logically implicated in that degree expression are expressed. Further, when expressing the meaning of an expression adding any adjective and any degree modifying element to that adjective, the meaning is expressed while being divided into two levels of a level expressing the meaning of the adjective itself and a level expressing the meaning of the expression added the modifying element.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a natural language processing system for analyzing a natural language, that is, Japanese or English on a computer, expressing its meaning, performing inference, and generating a result.

[0002]

2. Description of the Related Art In natural language, that is, Japanese and English, there are many expressions that express the degree of things. For example,
The expressions "relative" such as "all,""many," and "some," the frequency such as "always,""frequently," and "sometimes" and adjective pairs such as "high" and "low" This is a typical example. Such degree expressions exist beyond the distinction between Japanese and English languages.
Within each language, transcendental parts of speech such as adjectives, adverbs and pronouns exist.

Linguistic research has pointed out that there is a phenomenon called "conversational implication" in the degree expression (H. Paul Grice, 1975. "Logic").
and Conversation "In P.Col
e and J. Morgan, "Speech ac
ts'pp. 45-58. Symtax and Se
mantis3. New York: Academi
c Press. (H. Paul Grice, "Logic and Conversation," P. Cole and Jay El Morgan, "Syntax and Conversation"
Semantics, Volume 3, "Speech Acts" is included.
1972)).

[0004] For example, when talking about the quality of the test, when I said, "I solved 3 questions." I literally accepted that I had answered exactly 3 questions, not 4 or 2. This is the usual interpretation of the case. However, even if you actually answered 4 questions and said "3 questions were answered," it is not necessarily a lie in theory. Because 4
It is because he / she surely has 3 questions since he / she answered. A phenomenon called "conversational implication" is that, in spite of this logical implication, in reality, it is usually understood that only three questions have been solved. Similar phenomenon is not limited to the case of numbers like this example,
Seen between "some people" and "everyone",
“Pretty tall” and “ve”
It is also seen between the expressions "ry tall".

As described above, expressions representing partial quantities such as "all" and "some", adjective pairs such as "high" and "low", and numbers such as "3" and "4" have common properties as degree expressions. , Conventionally, a uniform method of expressing it has not been devised, but it has been handled individually.

First of all, expressions representing partial quantities such as "all" and "some" are basically considered in correspondence with logical symbols "A (upside down)" and "E (upside down)" in logic. (Laurence R. Horn, 1
972 "On the Semantic Prop
rties of logical operator
s in English "Ph.D destroy
ation, University of Calif
ornia, Los Angeles. (Lawrence ・
Earl Horn, "On the Semantic Properties of Logical Operators in English," Doctoral Thesis, University of California, Los Angeles, 1972).

That is, it has been pointed out that the meanings of expressions such as "all" and "some" in natural language are different from the meanings of these logical symbols. Nevertheless, an effective expression method instead of this logical symbol is It wasn't devised. In addition, "many" in natural language
The meaning of "few (only ...)" cannot be associated with a logical expression.

Next, regarding adjectives, in conventional linguistic and computer science research, the meanings of adjectives are expressed in correspondence with continuous quantities. For example, in Fuzzy logic, the meanings of the expressions “high” and “very high” regarding height are expressed by a continuous graph (Lofti A. Zadeh, “A”).
fuzzy-set-theoretic inte
rpretation of linguistic h
edges ”Journal of cybernet
ics. 1972. (Lofty A. Zade, "A
Father Set Theoretic Interpretation of Linguistic Heads "Journal of Cybernetic, 1972)).

[0009] Such an idea is basically also applied to linguistics, and "tall" and "pretty tall" are represented by a continuous graph as shown in FIG.
It expresses the meaning of the degree expression (James
D. McCawley, "Everythiwg th
at Linguistics have AlwaysW
anted to Know anout Logic
but surehamed to ask "C
hicago: The University of
Chicago Press. 1981. (Everything That Linguistics Hub Or Wes Wanted Toe, By James Dee Macquarie
No About Logic But Were A Shamed To Ask, University of Chicago Press, 1981)).
In FIG. 3, "5'3""(5
I'm using a graph that uses the actual height of the height, such as 3 '.

Regarding the number, for example, in the case of "3", it is a usual method to simply express it as the number "3" or to express it as a point on the number line in the same way as it does with the number. Is. Thus, in the case of both adjectives and numbers, degree expressions have traditionally been treated in a continuous framework.

There is no conventional research on the idea of expressing the degree expression discretely. There was only one report that the applicant of this patent applied it to some degree expressions (Kaicei Shinichiro, Muraki Kazuto "Modeling degree expressions" The Institute of Electronics, Information and Communication Engineers, NLC89-19, 1989. Year).

[0012]

As described above, in the past, a property that is widely common to degree expressions, for example, a linguistic phenomenon such as "conversational implication" has been studied in detail, while "all"
There was no way to uniformly display various degree expressions such as partial quantities such as "something", adjectives, and numbers. Also, there was no framework for uniformly handling "many", "all", "some", and so on. Furthermore, there was no method for calculating the meaning change when making a negative sentence from an affirmative sentence containing a degree expression and the possibility of answering a question containing a degree expression on a computer.

Further, there is no method for comparing and examining the correspondence of the meaning of the degree expression between the respective languages, and it is sometimes impossible to judge whether or not the correspondence is appropriate.

An object of the present invention is to provide a processing method of a degree expression in natural language for solving these problems.

[0015]

A natural language processing system for analyzing a degree expression in a natural language according to the present invention is a natural language processing system which analyzes Japanese or English on a computer, expresses its meaning, makes an inference, and generates a result. "All""many"
An expression that represents a partial quantity such as "some" or "high"
Adjective pairs such as "low" or "3" and "4"
On a computer, the meaning of a degree expression that indicates the degree-related part of the meaning of a quantity in natural language is represented as a discrete relative position in the framework formed by a group of expressions that form a pair with that expression. It is configured to express the literal meaning of a certain degree expression and the meaning logically implied by the degree expression.

Further, according to the processing method of the degree expression in the natural language of the present invention, when expressing the meaning of one degree expression and the expression in which the degree modifier is added to the degree expression, the level representing the meaning of the original degree expression is used. , It is configured to express the meaning by dividing it into two levels, that is, a level indicating the meaning of the expression with a modifier.

[0017]

As shown in FIG. 1, according to the present invention, the meaning of one degree expression in natural language, for example, "al" in English.
When describing the meaning of the expression "l" (corresponding to "all" in Japanese), a group of expressions paired with the expression, that is, "all", "many", "some", "few"
It is expressed as a relative position within the framework formed by "no". By adopting such an expression method, it is possible to express a meaning that cannot be expressed by one word in natural language. For example, when describing the meaning of the adjective pairs "tall" and "short", "" tall "and" sho "
Although it is necessary to describe the meaning of a state that is not “rt”, such a state can be easily described by using the expression method of the present invention.

In the present invention, as shown in FIG. 1 (e), when expressing the meaning of one degree expression, there are two meanings: the literal meaning and the meaning logically implied by the degree expression. Think of meaning as the meaning of one degree expression, and express both of them. By doing so, it is possible to express the "conversational implication" in the degree expression in linguistics, and by using this expression method, the meaning of the affirmative sentence including the degree expression corresponds to the negative sentence. It is possible to calculate the meaning of and the possibility of an answer to a question sentence including a degree expression.

Further, in the present invention, as shown in FIG. 2, when expressing the meaning of one degree expression and the meaning of the expression in which the degree modifier is attached to the degree expression, the meaning of the degree expression without the original modifier element The meaning is expressed by dividing it into two levels, that is, a level indicating the expression and a level indicating the meaning of the expression with the modifier. For example, as shown in FIG. 2B, a level indicating the adjectives "tall" and "short",
In addition, the degree modifiers "very" and "a li"
Expressions such as "little", that is, "ver
The meaning of the degree expression is expressed by a structure having different levels indicating the meaning of "y tall" and "a little tall". By expressing in this way, it is possible to calculate the possibility of answering a question sentence including a degree expression with a degree modifier. This kind of calculation treats the degree expression as continuous like the conventional method, and expresses the original degree expression and the expression with degree modification on the same number line. That was something I could not explain.

[0020]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, the present invention will be described with reference to the drawings.

The present invention can be used to represent the meaning of a degree expression in any language such as Japanese or English, but the case of English will be mainly described here as an example.

In the following, first, (A) a method of uniformly describing various degree expressions in natural language using the expression method of the present invention will be described. Next, (B) the modification relation between the degree expression and the degree modifier is expressed using this method. Next, (C) "conversational implication" in the degree expression in linguistics is expressed by the method of the present invention. Next, a method of calculating the meaning of the corresponding negative sentence from the meaning of the positive sentence including the (D) degree expression will be described. Next, by using this method to describe question sentences including (E) degree expressions and their answers, it is shown that phenomena that could not be explained in the past can be explained. next,
(F) A method of inferring from expressions including degree expressions using this expression method will be described.

(A) First, a method of uniformly describing various degree expressions in natural language using the expression method of the present invention will be described. Specifically, "all""many"
Degree expressions that express partial quantities such as "some", "few" and "no", adjective pairs such as "tall" and "short", and parts of the meanings of numbers such as "3" and "4" related to degrees are expressed in the same manner .. This description is the premise of the description that follows.

FIG. 1 is an explanatory diagram showing an embodiment of the expression method of the present invention. FIG. 1A shows the basic form of the expression method of the degree expression in the present invention. In this figure, five windows are arranged in a row to form a unit. Each window corresponds to the degree represented by each degree, but is represented by the relative position in the row.

FIGS. 1 (b), 1 (c) and 1 (d) are specific examples in which various basic expressions are expressed using this basic form. FIG. 1B shows “all”, “many”, “some”, “fe”.
Figure 1 (c) shows the meaning of "w" and "no", which is a relative quantity with respect to the whole, in the meaning of the English adjectives "tall" and "short", which represent "tall" and "short" Fig. 1 (d) shows the meaning of the part related to "," and the meaning of the part of "3" when "I solved three problems".

In the case of FIG. 1 (b), all the five windows shown in the basic form are used, but in the case of FIG. 1 (c), "high state", "neither high nor low state", and "low state" are shown. Therefore, I use three of the five windows. In the case of FIG. 1D, the number specifically described in the sentence (for example, “3
The number of questions "0" that could not be solved, centered on "3" of "Question" (represented by NO in the figure), the state where one or two questions were solved (represented by 3> in the figure), In order to distinguish five states, which are not all but four questions solved (represented by 3> in the figure) and all solved (represented by ALL in the figure), all five windows of the basic form are used. In each drawing of FIG. 1, each window is shaded for the sake of visual clarity. In the expression system of the present invention, the relative position of the windows indicates the relative degree of each expression, so the diagonal lines added to the windows are originally redundant, but as in the case of FIG. There is a case where is not used, so in order to clearly indicate which window is used in such a case, a figure with diagonal lines will be used in the explanation.

When actually implementing this expression on a computer, the information corresponding to the diagonal lines of each window may not be used. This is because if you know in advance which window it will use when a concrete expression is given, then you can express the relative degree of each expression only by the relative position of the window, and use that. This is because various calculations as described below can be performed. That is, when it is realized on a computer, the part relating to the degree can be expressed by the number of bits corresponding to the number of windows used.

(B) Next, with reference to FIGS. 2 and 4, the description will be given of expressing the modification relation between the degree expression and the degree modifier with this method. This description is the premise of the description that follows.

It is known from linguistic studies that there are the following types of degree modifiers (Takao Yagi, "Degree Expression and Comparative Structure", Daishukan Shoten, 1987). In the case of English, the following words are examples.

A. Amplification word ... very, extrem
ely etc. B. Relaxation word ... pretty, other, som
ewhat etc. C.I. Weakening word ... a little, slightly
Etc. D. Proximity words ... almost, neighbory, etc. E. Complete word ... completely, absoluu
Although these word groups such as tery modify expressions to various degrees, there are semantic restrictions on the modification method. For example, the close word "a
"most" means "almost all" and "almos"
“All” and “no” can be modified like “t no”, but “many”, “some” and “feo” cannot be modified. Similarly, the amplification word “very” can modify “many” and “few”,
"All", "some" and "no" cannot be modified. FIG. 4 shows the limitation of the semantic modification of the degree modifier. That is, the amplification word, the relaxation word, the weakening word, the completion word, and the proximity word described above modify which of the five windows according to the method of the present invention. In the figure, the symbol + indicates that modification is possible, and the symbol-represents that modification is not possible.

By reflecting the modification relation between the degree expression and the degree modifier as shown in FIG. 4, the meaning of the degree expression and the meaning of the expression having the degree modifier attached to the degree expression are used in the method of the present invention. An example expressed using is shown in FIG. Figure 2
In (a), five windows (“all”, “many”, “so”)
(corresponding to “me”, “few”, “no”), degree modifiers “completely”, “almost”, “ver”
It represents the meaning when “y”, “pretty”, “altitle”, etc. are attached. The symbols A, B, C, D, and E in the figure represent expressions made up of amplification words, relaxation words, weakening words, closing words, and proximity words, respectively. Further, in FIG. 2B, the degree modifiers “very”, “pretty”, and “a lit” are added to the adjective pairs “tall” and “short”.
"tle" and the like indicate the meaning.

In FIGS. 2A and 2B, the level representing the meaning of the degree expression and the level representing the meaning of the expression in which the degree modifier is added to the degree expression are expressed separately. This is one of the discrete features of the expression method of the present invention, which is different from the conventional one. This will be described with reference to FIG.

An expression ("pretty tall") in which an adjective "tall" in English is accompanied by one of the degree modifiers (for example, "pretty") is "tall".
However, it means that it is not "very tall". In FIG. 2B, “very tall” and “pretty ta” are displayed below the window corresponding to “tall”.
There is a window representing "ll", which represents such a situation. The important points here are "tall" and "pre".
It is at a different level from "tty tall", and it is not possible to compare which is more "tall". On the other hand, since "pretty tall" is at the same level as "very tall", the two can be directly compared. This is "" pretty ta
"ll" is a part of "tall", and "very ta"
It should be compared with "ll".

Another important point in FIG. 2 is, as expressed in FIG. 2 (b), the expression "pretty tall" which is the same as "tall" with a degree modifier.
And "a little tall", the former is "ta
It is located under the window "ll" and the latter is "not tal".
It is located under the window "l and not short". This is because the former with a relaxation word is included in the concept of "tall", while the latter with a weakening word is "not talland".
It is included in the concept of "not short", but the latter with weakened words is "not tall and n".
It is included in the concept of “ot short”.

Here, too, there is a discontinuity between expressions, which cannot be seen by the conventional expression method. The validity of such discrete expressions is shown by the fact that it is possible to explain the peculiar phenomenon that occurs between the question sentence containing these expressions and the answer to them. This will be described later.

(C) Next, the "conversational implication" in the degree expression in linguistics is expressed by the method of the present invention. As a specific example, an utterance when talking about how many questions were solved in the test, “I solved three of
the problems. Is used to express the meaning of the part relating to the degree of the number "3", using this method. In FIG. 1 (e), two rows that form a set of five windows are used. In this method, the basic form of the degree expression is used in two columns to express the "implication of conversation" in linguistics. The first column shows the literal meaning of this example sentence, that is, the number of solved problems is just "3", and there is not more or less than that. The second column shows the possibility that this example sentence has logical implications.

If the number of solved problems is greater than 3, then "3
The statement "I asked you" is not a mistake, considering only the relationship of numbers. However, in the normal case, if you actually solve 4 or more questions, you do not say "3 questions have been solved". The part of the difference between the literal meaning and the logical possibility, that is, the part where the windows of "all" and ">3" in the first row are blank represents the "implication of conversation" in linguistics. .. The meaning of conversational implications is not always clear in affirmative sentences, but plays an important role in negative and interrogative sentences as explained below.

(D) Next, a method of calculating the meaning of the corresponding negative sentence from the meaning of the positive sentence including the degree expression will be described.
Specifically, the affirmative sentence “I solved three
ofthe problems. "I didn't solve three of
Take the meaning of “the problems” (3 questions not solved, 3 questions not solved, 3 questions not solved) as an example. Generally, in the case of a negative sentence containing a degree expression, there can be several different interpretations, but the interpretation can be systematically obtained from the meaning of the original positive sentence using the method of the present invention.

First, the negative sentence "I didn't sol"
The first meaning of "ve three of the problems" is that there are three unsolved problems. FIG. 5A shows a method of calculating the corresponding Japanese expressions by calculation. In the case of this interpretation, the original affirmative sentence “I solved three
of the problems. There is no need to change the expression in the case of “”, that is, FIG. 1 (d). However, if the role of the window is changed from one that represents the number of problems that can be solved to one that represents the number of problems that cannot be solved. good.

The negative sentence "I didn't solve"
The second meaning of "three of the problems" is that the number of solved problems did not reach three. In Japanese, it corresponds to the meaning of the expression "I could not solve 3 questions". Figure 5 shows how to calculate this meaning from the meaning of the original affirmative sentence.
It is (b). The calculation method for this interpretation is as follows.

(1) First, the meaning of the original affirmative sentence (see FIG. 1).
Make the negation of each of the two columns representing (e)). Specifically, the on / off state of the window is just reversed.

(2) Next, the common part of the two columns is taken. Specifically, the windows that are commonly turned on are selected.

The result thus obtained is the second column of the negative interpretation to be sought, that is, the logical possibility. Further, in order to obtain the first column of the negative interpretation to be obtained, that is, the literal meaning, the window of (3) ALL or NO is turned on for the state of the second column showing the logical possibility just obtained. If so, turn it off.

The entire two columns thus obtained express the meaning of the negative sentence to be obtained. Usually, if you give a specific number (3 in this case) and say "I couldn't solve 3 questions", it means that I solved some questions, but the number did not reach the specific number. The first column in FIG. 5B shows this meaning. However, the same statement "I could not solve 3 questions"
However, it may mean that the number of solved problems was not one. The second column shows the possibility of such a meaning.

The negative sentence "I didn't solve"
The third meaning of "three of the problems" is to mean "I did not solve 3 questions, but I solved 4 questions (5 questions or all questions)."
FIG. 5C shows a method of calculating the expression expressing this meaning from the expression of the meaning of the original affirmative sentence.
The calculation method for this interpretation is as follows.

(1) First, each of the two columns representing the meaning of the original affirmative sentence is negated. Specifically, the on / off state of the window is just reversed.

(2) Next, the difference between the two examples will be taken. Specifically, select a window in which only one is turned on.

The result thus obtained is the second column of the negative interpretation, that is, the logical possibility. Further, in order to obtain the first column of the negative interpretation to be obtained, that is, the literal meaning, as in the case of FIG. 5B, (3) ALL or N
Turn the windows of O on and off.

The two columns thus obtained express the meaning of the negative sentence. If you use the phrase "I didn't solve 3 questions" by using a specific number (3 in this case), and usually say that the number of solved problems was greater than that specific Means that there is a problem that could not be solved. The first column in FIG. 5C represents this meaning. However, it is possible to include the case where all the problems have been solved by the same expression "not solved three problems". The second column gives the meaning in such cases.

In this way, according to the expression method of the present invention, all meanings of the negative sentence including the degree expression, including the implications of the conversation in each case, correspond to the meaning of the positive sentence. It can be calculated from the expression.

(E) Next, a method of describing an interrogative sentence including a degree expression and its answer by this method will be described.
Specifically, with reference to FIG. 6, the affirmative sentence “I solved
three of the problems. "Did you solve thre
e of the problems? (Did you solve 3 questions?) And the answer to that. When asked in this way, if you actually answered four questions, both "yes" and "no" answers are possible in some cases.
That is, "Yes, in fact I solved
four. (Yes, I solved 4 questions) and “N
O, I solved four. (No, I solved 4 questions) is possible. This fact will be explained using this method.

The leftmost figure in FIG. 6 represents the meaning of the portion "3 questions" in this question sentence. On the other hand, the figure in the middle represents the meaning of the true situation “4 questions” to be answered. Since the leftmost figure is the meaning of the sentence actually uttered, as already mentioned, the literal meaning, that is, just "3
And the logically possible meaning, that is, "3
It is expressed in two columns with the meaning "more than a question". On the other hand, the figure in the middle represents a true situation, not a specific utterance, and is therefore represented by only one line.

The situation of answering a question is expressed by the operation of comparing one row in the left figure of FIG. 6 with the row of the middle figure. If the first column of the left graphic and the column of the middle graphic are compared, the answer is "NO"("No") because the two columns have no intersection. This answer is an answer when the literal meaning of the question is "just three questions" and the answer is given. If you compare the second column of the left figure and the figure in the middle, the answer is "Yes"("Yes") in this case, because the two columns have a common part, that is, the window of ">3". .. This answer does not mean that the meaning of the question is literally asking "whether you have exactly 3 questions", but is "if you have answered at least 3 questions", and is the answer when you answer it. is there.

Incidentally, the latter answer "yes" is usually the more natural answer. This indicates that in the question sentence, the second column, that is, the column representing the logical possibility, is more important than in the affirmative sentence.

Next, a question sentence including a degree expression with a degree modifier and an answer thereto are described using this method,
We show that we can explain phenomena that could not be explained in the past.
Specifically, the above-mentioned James D. McCawle
y, “Everything that Linguis
ts have Always Wanted to K
now anout Logic but there
ashamed to ask "Chicago: The
University of Chicago Pr
ess. 1981. (James Dee Macquarie, "Everything That Linguistics Hub Oluwes Wanted To No About About Logic But Were A Shamed To Ask"
It shows that the two phenomena described in the University of Chicago Press (1981) can be explained in a unified manner.

In this book, two examples of the following degree expressions in English are given. One is an example in which a relaxation word is used as a degree modifier. A questioner who doesn't know the height of a man says "Is he prety.
tall? When he asks, "Is he quite tall?", And the respondent thinks he is "very tall," he says, "Yes, he is very tall."
(Yes, very expensive) is a reasonable answer,
"(*) No, he is very tall."
(No, it's very expensive) is a phenomenon that is not a valid answer. Here, the symbol (*) indicates an invalid answer.

Another example is similar to the present example, but in the case where the weakening word is used as the degree modifier. "Is she a little fat?"
When she asks (is she a little fat?), If she thinks that the answer is "very fat," she asks, "(??) Yes, she is very fa.
t. "(Yes) also" (?) No, she is ver
y fat. "(No), neither answer is a phenomenon that is neither completely appropriate nor completely inadequate.
Here, the symbol (??) means that it is quite inappropriate but cannot be said to be completely inappropriate, and the symbol (?) Means that the degree of inappropriateness is low but a little inappropriate. The two phenomena will be described with reference to FIGS. 7 (a) and 7 (b). When considering this problem, the expression having the structure divided into two levels introduced in FIG. 2 is used as the expression of the meaning when the degree expression has a degree modifier.

The method of calculating the possibility of answering a question is described in the above-mentioned "Did you solve ofe of
the problems? It is basically the same as the case of ". There are two ways of answering: one is to think about the meaning of a question and to answer it, and the other is to answer a question literally. And, as in the previous example, the former is the usual answer.

The difference from the above-mentioned example is that the structure is 2 in FIG.
It is divided into two levels. As explained in FIG. 2, the upper level of the two levels is the relaxation word,
It is a level that represents the meaning of a degree expression that does not have a degree modifier such as a weakened word. The lower level is a level that represents the meaning of a degree expression with a degree modifier.

The leftmost figures in FIGS. 7 (a) and 7 (b) are "Is he pretty tall?" And "Is", respectively.
she a little fat? 2 ”, the lower level is expressed in two columns in this figure, unlike FIG. Of these two columns, the upper column represents the literal meaning of the question and the lower column represents the logical meaning of the question. This is similar to the case of FIG. For example, in FIG. 7A, the upper row of the lower level is literally "p
“Pretty tall”, and the lower column means the meaning that it logically implies, that is, “pretty tall”.
Represents all sides higher than.

First, the calculation method in the case of an answering method in which the meaning of a question is considered logically and the answer is given will be described.
In this sense, when we see whether there is an intersection between the meaning of the question and the situation of the answer, at the lower level, of the two columns, the lower one, that is, the question has a logical implication. Use only the column that represents the meaning. The upper level is also used for comparison. As mentioned above, higher levels such as "tall"
Is a window at the lower level, for example "pretty ta
Since it is a premise of the meaning of "ll", the agreement of the meaning at this higher level also affects the eligibility of the answer.

As a specific example, "Is he prety"
tall? FIG. 7 (a) shows a method of calculating the possibility of answering the question "," in consideration of its logical implication. The leftmost figure in the figure shows the meaning of the question, and the middle figure shows the answer situation, that is, "very tall". Compare these two figures to see if they have something in common.

First, in the upper level, the question "prett
In the case of "y tall", the answer situation is "very tall".
Also in the case of “l”, “tall” is premised, and they match. At the lower level, if the question is answered in the context of its logical implications, only the bottom row is compared to the answer situation.
Then, the two columns have a common part, that is, "veryta".
There is an "ll" window and the meanings match. In this way
The answer must be "Yes" because the meaning of the question and the situation of the answer match at both the upper and lower levels.

In FIG. 7, the rightmost part shows a method of calculating the contribution of the upper level and the lower level regarding the validity of the answer. Place the contribution of the upper level twice the contribution of the lower level. If they match at the upper level, give two ○,
If they do not match, give two ×. If there is a match at the lower level, give a one, and if not, give a one.

In order to calculate the validity of the final answer, the results of coincidence at the upper and lower levels are added. ○ 3 is perfectly valid, ○ 2 × 1 is slightly unsuitable (symbol (?) It is not completely unsuitable if it is 1 × 2, but it is quite unsuitable (state represented by the symbol (??)), ×
If it is three, it corresponds to being completely unsuitable (state represented by a symbol (*)). Above "pretty tal"
For the question "l", considering its logical implication, "ver
In the case of answering “y tall”, “Ye
The answer "s" is completely valid and the answer "No" is completely inappropriate. This illustrates the first example of the book listed above.

When a question is taken literally and answered, a different one from the example used for comparison in the above case is used. That is, the lower level uses the upper of the two columns, that is, the column that represents the literal meaning of the question. The upper level, that is, the meaning that is the premise of superficial expressions, is not used for comparison.

For the literal meaning of the question "pretty tall", "very tall"
When trying to answer "l", the answer is "No", which is equivalent to comparing only the upper row of the lower level with the lower level of the answer situation. In this case, since there is no common part, the answer is “No”. Using this method, we can explain the phenomenon that there are various possible answers.

Next, referring to FIG. 7B, "Is s
he a little fat? The possibility of an answer when answering "very fat" in consideration of its logical implication will be explained. As mentioned above, "Is
he pretty tall? As in the case of “,” the lower row of the upper level and the lower level of the leftmost figure is used for comparison with the figure in the middle.

In this case, the upper levels do not match. The question "a little fat" is "not fat"
While the meaning of "and not thin" (neither fat nor thin) is assumed,
This is because "very fat" is premised on the meaning of "fat" (thick). Therefore, "Y
As an evaluation of the validity when answering "es", two Xs are given for this level. For the evaluation of the “No” answer, just give two circles.

At the lower level, there is an intersection between the lower row, which represents the meaning that the question logically implies, and the lower level in the middle graphic. Therefore, as an evaluation of the validity in the case of answering "Yes", one is given to this level.
On the contrary, in the case of "No", one x is given.

The total of “Yes” and “No” of the validity evaluation for each answer of “Yes” and “No” is “Yes”.
X. This means that this answer is not completely unsuitable, but is rather unsuitable (state represented by the symbol (??)). For "No", the result is XX.
This means that this answer is not completely valid but is a little inappropriate (state represented by the symbol (?)). This calculation result accurately reproduces the judgment made by the English speaker in the above-mentioned book.

The phenomenon that the relevance of the answers to these two similar examples is different can be explained only by using the method of discretely expressing the degree as in the present invention. In the conventional method of treating the degree expression as a continuous quantity, “p
The difference between the case of "retty" and the case of "altitle" cannot be explained.

(F) Next, a method of inferring from expressions including degree expressions using this expression method will be described. Specifically, a phenomenon called asymmetry between “many” and “few” in inference will be described with reference to FIGS. 8 and 9.

An asymmetric phenomenon in inference is known in sentences including English "many" and "few" as shown in the following example sentences (Masayuki Ikeuchi, "Limited Expression of Noun Phrases", Taishukan Shoten 1985). Year).

(A1) Many girls are
both attractive and intel
agent. (A2) Many girls are attrac
five and many girls are i
ntelligent. (B1) Few girls are both at
tractive and intelligent. (B2) Few girls are attract
IVE AND FEW GIRLS are int
elligent. The example sentences A1 and B1 and A2 and B2 have exactly the same structure as a syntax, but there is a difference in the implication relation between A1 and A2 and between B1 and B2. A1 logically implies A2, but the reverse is not true. On the contrary, B2 logically implies B1, but the opposite is not true.

As shown in this example sentence, the behavior in logical reasoning is opposite between Many and Few, but there has been no method that can handle this well in a computer. 8 and 9 describe this phenomenon using the expression method of the degree expression of the present invention. 8 and 9
You can use this same calculation rule to calculate this seemingly asymmetrical logical entailment.

Att as described in (A1)
When there are many racitive and intelligent girls, it is logically correct that there are many attractive girls and many intelligent girls. In this case, all girls are att
It may be racitive, or all girls may be intelligent. But,
All girls are intelligent and at
The situation of being tractive must be excluded from the logical implication of (A1). In such a situation, instead of using "many" as in (A1), use "all" to "All girrl".
s are both attractive and
This is because it is expressed as "intelligent".

FIG. 8A shows this logical implication. In the figure, the part represented by "except" represents the above-mentioned case to be excluded.

Conversely, at as described in (A2)
There are many tractive girls intellig
In a situation where there are many ent girls, that does not mean that many girls have both of them. FIG. 8B shows this logical implication.

In the case of Many, there are the implications as described above, but in the case of Few, the implications are opposite as described above. However, using the expression method of the present invention,
These two can be handled uniformly by the same calculation method.

FIG. 9 shows the logical implication relationship in the case of "Few". In FIG. 8 and FIG. 9, calculation is performed according to the same principle as described below. When inferring from an expression (example sentences (A1) and (B1)) including the meaning of "attractive and intelligent" (represented by double diagonal lines in the figure) (that is, in FIG. 8A and FIG. 9A) In that case, to find its logical implication, first make two columns and turn on some more side windows at the same position as the source of each column or to the left of it. In the case of this example, each newly created column is "attra".
represents "active" or "intellegent". In the figure, this is represented by a single diagonal line. The case where both columns represent "ALL" is excluded from the logical implication. This principle is common to the case of "many" and the case of "few".

When inference is performed from expressions ((A2) and (B2)) in which sentences that separately describe "attractive" and "intelligent" are connected by and (that is, FIG. 8B), In the case of FIG. 9B), in order to obtain the logical implication, one column may be newly formed, and a window on the side at the same position as the original or on the right side may be turned on. This procedure is "many"
The case is common to the case of "few".

As described above, by using the expression method of the present invention, there is a possibility of modifying a modifier regarding a degree expression, a meaning of a negative sentence, a meaning of a question sentence and a possibility of a possible answer thereto and its validity. It can handle gender and certain types of logical reasoning in a unified manner.

[0084]

According to the present invention, the meaning of one degree expression can be expressed by a certain fixed number of bits. For example, "al
The meanings of “l”, “many”, “some”, “few”, and “no” can be represented by 10 bits because two columns of 5 bits can be used when the degree modifier is not considered.
Similarly, the meaning of the part "3" in the case of "solving three problems" can be expressed by using two 5-bit strings. The number of bits required to express the meaning of this part is always constant regardless of the specific number appearing in the sentence (“3” in this case).

In the case of adjectives such as "high" and "low", two 3-bit strings can be used when no degree modifier is considered. When considering a degree modifier, one 3-bit sequence and two 9-bit sequences can be used. Thus, as mentioned in the "Examples" section,
The meaning of a sentence including these degree expressions can be easily expressed on a computer, and the corresponding negative sentence, question sentence, and the answer can be calculated.

The present invention can also be used in a more detailed comparative study of the correspondence of the meaning of the degree expression between languages such as Japanese and English, which could not be achieved in the past. The actual amount indicated by a degree expression depends on the user and the situation of the expression. For example, even if “tall” is said to be the same, the specific frequency may differ greatly depending on the case. Therefore, when considering a degree expression in the other language that corresponds semantically to one degree expression in one language, the meaning of that degree expression is conventionally unclear, so it is said to be “tall”. Even though "tall" corresponds, it was hard to think of how accurately it corresponded. However, by using the expression method of the present invention, a wide range of linguistic phenomena related to degree expressions can be described and explained, and it is possible to examine whether or not expressions considered to correspond in each language have the same linguistic behavior. , Allows detailed comparison between expressions in different languages.

[Brief description of drawings]

FIG. 1 is an explanatory view showing an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing a description example of a degree expression.

FIG. 3 is an explanatory view showing a conventional example.

FIG. 4 is an explanatory diagram showing a modification relation of a degree expression.

FIG. 5 is an explanatory view showing a meaning expression of a negative sentence.

FIG. 6 is an explanatory diagram showing a relationship between a question sentence and an answer.

FIG. 7 is an explanatory diagram showing a relationship between a question sentence including a degree modifier and an answer.

FIG. 8 is an explanatory diagram showing an example of a sentence inference method.

FIG. 9 is an explanatory diagram showing another example of a sentence inference method.

Claims (2)

[Claims] 1. In a natural language processing system that analyzes Japanese and English on a computer, expresses their meanings, infers them, and produces results, "all,""many,""some"
Meaning of a degree expression that indicates a part quantity related to a degree in a natural language such as "high", "low", or an adjective pair such as "high" or "low" Is expressed on a computer as discrete relative positions in a framework formed by a group of expressions forming a pair with the expression, and the literal meaning of a certain degree expression and the degree expression logically connote it. A processing method of degree expression in natural language characterized by expressing meaning and. 2. The method of processing a degree expression in a natural language according to claim 1, wherein when expressing the meaning of one degree expression and an expression having a degree modifier in the degree expression, the meaning of the original degree expression is expressed. A processing method of a degree expression in a natural language, which is characterized by expressing the meaning by dividing it into two levels, a level and a level indicating the meaning of the expression with a modifier.