2.4 Natural Language Processing Technology for Intellectual Document
Processing

Document processing technical special committee
Chairman Nomura 浩郷
(Professor of the Kyushu Institute of Technology)

2.4.1 Introduction

　Also compared with the former, importance has come to be attached to
various use of various written information by the spread of the
Internet. The request of wanting for information to know to come to
hand in a suitable form efficiently by easy operation has increased
suddenly. In order to respond to such a request, much more leap of
natural language processing technology is just going to desire, and
development of especially intellectual document processing technology
is indispensable. It is also regarded as it being one factor these
days that the brake to the development of IT industry seen in part
cannot fully correspond technically to these requests from a user. The
technology which solves this is processing technology which "is
intellectual" if it says at a word, and can be especially represented
with the word "intellectual information access" from the viewpoint of
the Internet use.

　Research of processing technology almost intellectual before a
half-century has been performed. However, even if it continues till
present with regrettable, it must be said that the attained technology
is far from calling it "intellectual."

first, the first problem -- "-- it is that some quantitive
definitions [ be / intellectual ] " are not performed The only
appropriate definition is called "test of a チュー ring" which is
qualitative and was proposed far at ancient times. Another proposal
proposed after this does not exist now, either.

The same is said of "understanding" which can be considered to be
also the language and this meaning of "being intellectual", a
"meaning", and "knowledge" on one side. The system of a discrimination
network, "understanding" by ミーンズ- and - analysis, and "problem
solution" was proposed in the 1960s. However, since these are too
simple and were [ or ] too much imperfect, there was no development in
what [ useful ]. Furthermore, "a semantic expression" by the "セマン
tick network" was proposed in the 1960s. However, this was the
expression form for explanation and was what carries out no
contribution to actual semantic processing. "Conceptual dependence
structure" was proposed and the trial of the actual semantic
processing which uses a conceptual primitive as an element was made in
the mid-1970s. However, it finished, without demonstrating the meant
effect, since there was a place when it is too much easy and
inadequate for a "meaning" or a definition of a "concept"
too. Although the "knowledge engineering" represented by the expert
system was proposed as different approach, by the time it contributes
to realization of high intellectual processing of capability, it has
not continued till the period by the hugeness of the knowledge which
should be built etc.

The 1960s are a time when grammar theory was modeled formally for the
first time. Since the main portions of the so-called sentence
composition structure which English has were suitable for formal
modeling by chance, this is produced. The portion about a meaning was
intellectual or a thing which reference is only ideally made and does
not contribute to a setup of the model of understanding at all as deep
structure.

As for technical gradual accumulation, the result of these researches
of the second problem is not necessarily having not become. Then, it
will enter in the 1980s and research and technical development which
attached the leg to the ground have come to be advanced. In the field
of natural language processing, when the so-called morphological
analysis system came to be offered as free software and it entered as
a step of processing late in the 1990s, a syntactic-analysis system
came to be similarly offered as free software. Although these do not
treat a "meaning" and "understanding" directly and they have come to
play the role of the leader who enables the measure for actual system
development as the fundamental processing element, they are good.

　The subject of intellectual document processing is in construction
and maintenance of extensive and detailed various kinds "knowledge",
and development of precise "intellectual processing system" using it
from these experiences. It needs to be tried hard for such research
and development preponderant and energetic over the long period of
time.

　Development of the practical use system consider that, on the other
hand, carried out processing which "is intellectual" as a result to
the object and the purpose which were limited also has a strong
request. The implementability is high, especially, in access to the
Internet, deep "semantic processing" is avoided, the portion which can
be formally treated from there is started, and it also becomes
important to reclaim a simple and efficient technique. The demand that
it is extensive, and a variety and the various Internet information
come to hand in a suitable form efficiently by easy operation since
the outline is sufficient increases increasingly. The approach of
applying gradually from the place where the technology of "semantic
processing" from such a viewpoint or "semantic understanding" is
realizable, providing practical use with a helpful system, and
developing a function gradually by the experience is also
important. As the concrete object, there are information retrieval,
information extraction, an information summary, etc., and text mining
for carrying out new discovery from information further occurs. The
side of multi-modal crossing リンガル is also added to these, and it
can be said that research and developing are one steady approach to an
intellectual information technology.

2.4.2 Construction of Electronic Knowledge

　Some trials which it was going to build in the form where the
knowledge for using for natural language processing etc. was processed
electronically occur. As linguistic knowledge, there are a history for
an electronic dictionary, an electronic thesaurus, electronic
ontology, and a conceptual definition, rank structure, coincidence and
concord, various kinds of syntax-charge receptacles, etc. There is the
so-called knowledge base which consists of a concept of object
dependence etc. as general knowledge.

About electronic knowledge, only the quantity which can attain the
purpose of processing is needed. On the other hand, if quantity
increases, it will become difficult to take the whole
adjustment. Moreover, it becomes difficult to use it correctly and
effectively. However, if it is impossible to realize processing
advanced without knowledge required for processing and it carries out
from the viewpoint, even if time and a labor are applied, they must
build electronic knowledge with sufficient quantity and quality. In
the present natural language processing, while these must submit to
the fact which cannot fully be being built, you have to recognize
having not established the technology of fully mastering these,
either.

　Construction of electronic knowledge targets construction of a
general-purpose thing over a long period of time, and although it
limited an object and the range of applied, it needs to build as 短
and a middle target. The demand conditions over electronic knowledge
are not necessarily clear. Moreover, the method about the composition
of electronic knowledge is not necessarily clear, either. Furthermore,
the policy which actually builds electronic knowledge is not
necessarily shown, either. An object and the range of applied are
considerably limited to the bottom of such a situation, and it is
necessary to actually build the electronic knowledge which can meet
the demand of the clear contents of processing, and to use
it. Although this may feed back experience and it may become
trial-and-error, it is necessary to pack the contents of electronic
knowledge, composition, and directions.

About the electronic knowledge which limited an object and the range
of applied, the demand in Internet access of these days has been
tight. For example, demand, such as B-to-C and e-commerce of B-to-B,
is very large, therefore it is strongly pressed for offer of the
electronic knowledge for the partial imitation another place. The
request to a thing applicable to offer of goods, matching of a demand,
the conclusion-of-a-contract act for dealing, etc. is strong. For that
purpose, technology, such as the information retrieval using
electronic knowledge, information extraction, and an information
summary, needs to be established. although it becomes a repetition of
an argument at this time, if the trade in borderless global society
takes that it is an international act into consideration -- crossing --
being processed リンガル is also clear moreover, Internet access until
it results not only in linguistic processing but in a contract --
setting -- the increase of convenience -- in order to offer efficient
processing -- the so-called multi -- the fusion to modal processing is
also required

When usual information retrieval also limits an object, realization
of high reference of accuracy is desired strongly. Moreover, to show a
reference result in the good form of readability is also desired
strongly. To arrange and show a lot of various information is also
desired strongly. In order to respond to these requests, it is
necessary to limit an object and the purpose and to realize the system
which is equal to practical use.

2.4.3 Modeling of Structure of Understanding

　Electronic knowledge should not be discussed with intellectual
processing or semantic understanding, and can evaluate neither
advantages and disadvantages nor an effect by independent [ it ]. make
it e-commerce -- make it Internet access of other everythings -- even
if a partner is man and it is a machine, it must be able to model
formally what the communication in human society is primarily Without
it, a means to realize intellectual processing and semantic
understanding on a machine does not become clear. Although it is easy
to argue about a meaning or understanding like the argument for an
argument in a vague emotion-form, this becomes only to kill time [ of
a chat over tea ].

　Generally communication is told that there is no purpose in many
cases. It comes out so, and if it becomes, most purposes of the
communication without the purpose may be attained only by [ certain ]
installing Eliza. There is usually the purpose in Internet access. It
is wanting for information to know to come to hand etc. It becomes the
purpose to establish the processing system for that which completes
the purpose most economically as technology of intellectual natural
language processing.

　It has argued about the argument about something as understanding
over a long period of time historically from the side [ be /
psychological / philosophical and epistemology-/ it ] etc. Those
details are set aside and it is thought that the formal model for
actually implementing on a machine from the engineering side is a
thing based on the 3 class model of the so-called memory which came
out of psychology. With three classes of memory, it is well known for
language, such as short-term memory, middle memory, and long-term
memory. Transposing this to the 3 class model of understanding, and
discussing it is performed. However, these arguments are not yet
applied to actual implementation, either.

　Understanding is not specified only by there being language
knowledge, general knowledge, or special knowledge. An abstract object
will not be able to be touched, a concrete phenomenon occurs
concretely, and the language phenomenon of corresponding to it will
assume that it is a processing object. Semantic understanding cannot
be correctly performed without the knowledge of a concrete external
situation when a language phenomenon occurs then. Or the knowledge of
such an external situation must be shared by the participant of
communication. The example of language expression often issued is I
saw a girlon the hill with telescope. Here, when only this sentence is
shown by the ambiguity of the charge point of the so-called PP
attachment, he cannot understand the meaning of this sentence
correctly by it. however -- the time of the knowledge of an external
situation being given or being shared -- very -- easy -- truth -- a
meaning can be identified That is, anyone can do the right semantic
understanding. In information access by the Internet, the cause of the
difficulty [ advanced natural language processing ] of realization is
in the point that the knowledge of such an external situation is
unsharable in many cases. This is the same also at the case of the
others between which the participant to communication cannot own the
knowledge of an external situation jointly.

　Required for semantic understanding another difficult problem is
that an intention of communication of those who have participated in
communication, or the state of the heart is unsharable. Such a
situation is the phenomenon in which you may try to set to
communication of everyday life. The subject known as an utterance act
is one of such examples.

　It originates also in it being troublesome that you make it fully
know against them in access to the Internet as for the problem where
the knowledge of an external situation and the state of the heart are
unsharable. In the information retrieval in the Internet, almost all
men actually desire for the target information to be able to come to
hand only by a small number of keyword being shown extremely. Only
AND/simple OR also shows the relation of the keyword at that
time. This is also because the present information retrieval system
receives a demand only in such a form. It can also be said that such a
system will not progress 1 step from the system of the 1960s if an
extreme way of speaking is carried out. Consequently, it is not rare
to also show URL which attains to no less than several 10,000 pieces
as a reference result occasionally. Now, that many men present
dissatisfaction must also submit.

　Access to the present Internet sends a demand by the keyboard input
except for special application. This is also one of the factors which
have lowered the convenience of the Internet. Moreover, it is one of
the factors to which the response from the Internet and having the
form centering on a character have also lowered convenience. the
so-called multi expected at the future when these are near -- if modal
communication is realized, there is also a portion solved

　Modeling and implementation of understanding of structure will be
what should pursue the realization neither according to an abstract
model nor a general-purpose mechanism, but should clarify the purpose
of processing further for concrete application, and should take in and
embody knowledge, an external situation, and the state of the heart.

2.4.4 Identification of Meaning

　The intellectual foundations of natural language processing are
identification of a meaning as already stated repeatedly. What
identification of the meaning has and defines poses a
problem. Supposing a definition of saying [ "being intellectual" ] is
the test of the チュー ring described previously, it is a time of, for
example, the user of the Internet having recognized that it is
"intellectual" to the response. However, although it was an
"intellectual" response when saying that it lost heavily on dealing of
the stock in the Internet since judgment of the Internet was trusted,
it cannot be made the judgment standard of how. It is the
responsibility of the application system utilized through the
Internet, and though the judgment to the object with an element
indefinite from the first is wrong as a result, it cannot make it the
cause of a machine. However, the Internet's presenting a collectively
intellectual aspect in the future is wished, and there is dilemma in
dealing with such a situation.

　The test of a チュー ring describes the phenomenon-side of
intellectual activity. therefore, this is described as a formal model --
it can kick and implementation is impossible The processing process of
natural language processing has the traditional main syntactic
analysis to a European and American language. This is because a part
of structural side of the so-called syntax structure which a European
and American language has can describe formally. The accuracy of
syntactic analysis was able to be considerably raised with the
research for here about ten years by including statistical or probable
information in it. However, probably, as for those who think that
intellectual natural language processing was realizable by this, there
is also no person who thinks that intellectual near natural language
processing is realizable at the extension in the future that there is
nothing.

　The trial which was going to realize identification of a meaning by
logical description in ancient times occurred. However, the sign-mark
of a meaning tended to express semantic structure, and since it was
carried out when correspondence with the actual world assigned a sign
mechanically, this was not connected with advanced natural language
processing, either. Although symbolization brings brief-izing and
strict-izing of processing, it is the talk of processing in the world
of a sign, and is not the talk of semantic understanding in the actual
world. Like the test of this symbolization of a チュー ring, probably,
the place which is common to the system of semantic processing of the
natural language tried by the present of what referred to as [ finish ]
having not necessarily succeeded was size, if connected to the
response. Namely, it can be thought now that the efficiency of
processing of the phenomenon in the actual world was increased by
symbolization.

　As stated until now repeatedly, the response can estimate the
justification of identification of a meaning. No, you may say that the
valuation basis of justification has only it. It can judge only by the
response whether he has understood the meaning, although it was [
whether the meaning has been identified and or not ] the same when put
in another way. Of course, a classification of a text, automatic
construction of a day base, etc. must also interpret the processing
performed within a system as what is contained in the act of a
response. When it goes in saying too far, only the justification of
the response from the viewpoint of the test of a チュー ring is a
judgment standard. However, the justification of the response is not
limited to the only contents. There is ambiguity by contingency,
special intention, etc. If it carries out and cuts, it will become the
most important subject how it is formalized and implemented.

　When displaying an analog figure on a display, it is digitized and
expressed as the present computer. If digitization exceeds the
discernment capability of man's retina and is made detailed at this
time, man will perceive as if he was an analog. If it learns from
this, the number of the kinds of a meaning or its response is made [
many / enough ] and man enables it to understand just like man's
response by the result, it will be expected that it can be judged that
identification of a meaning was completed for the time being. That is,
it may be judged that the test of a チュー ring was passed. It is
expected that there is much application in which a problem does not
occur practically in this range based on the opinion which this idea
should say also by human digital mechanism surely.

　How as identification of a meaning, it is formalized concretely and
these are implemented poses a problem. Patternizing can be considered
as the one method. Very many pattern expression is used in many fields
irrespective of natural language processing. Logic formula expression
and conceptual dependence structure expression is also considered to
be the special thing of pattern expression. If it goes about the word
of being special in saying too far to a frame being a Turing machine,
other things which are easy to treat mean that it is simpler than
it. The frame was the common form of this pattern expression, and has
been used by being called by the name of a rank frame as language data
for Japanese processing. A frame is a framework as the language shows,
and about processing and the contents of semantic identification,
nothing will have made reference only by saying [ frame ].

A rank frame expresses the semantic structure of the sentence which
carried out simple structure. About the sentence which carried out
compound structure, if a subordinate clause etc. is defined as being
one of rank elements, addition structure and the semantic structure of
a sentence which carried out compound structure although it embedded
and it became structure can also be expressed. The structure
expression consists of a semantic and syntax-dependency. The semantic
dependency is the same as that of a syntax-dependency.

Since a rank frame is a Turing machine, it can write in any
restrictions about the rank element and predicate. Furthermore, while
the semantic structure of a sentence is expressed, any contents and
restrictions about "a response" etc. can also be written in a rank
frame as addition information. Therefore, although it becomes very
complicated, the pattern which can pass the test of a チュー ring can
be prepared. A language knowledge, general knowledge, and external
situation, the state of the heart, etc. can be included in the
contents and the restrictions which are written in here.

What is necessary is just to prepare an acceptance machine by several
[ at most ] times as much fineness as the resolution of the retina, in
order to make a figure perceive in analog as the example was given
previously. What details describe a rank frame for semantic
understanding? By what diverseness should a response be prepared?
About them, it cannot imagine at all. It comes out and a certain thing
is not just going to have the room which will become very detailed,
therefore will become a lot of description to suspect. Therefore, as
mentioned above, it is required to limit an object and the purpose
small and to actually describe in it. However, it is in imagination
that it is what a labor requires stubbornly so that this work also
feels faint.

Identification of a meaning is performed by the pattern match. The
detailed processing at this time is the check of whether restrictions
are filled. Generally, a pattern match is processing which time
requires. Therefore, it may become practically unreal to perform
pattern match processing about the rank frame described in detail.

2.4.5 False Intellectual Processing

　If the test of a チュー ring is the basis of intellectual
processing, it will be thought that the realization takes most
time. And in the present condition that such a trial performed by the
orthodox approach is not performed at all, verification about the
realization possibility of the intellectual processing based on the
test of a チュー ring cannot be performed, either. Then, has the
future run aground on a reef, as for realization of intellectual
processing? it mentioned above -- as -- an orthodox approach -- not
implementing -- と -- if the test of a チュー ring can be passed
mostly as a result, intellectual processing is realizable apparently
It is that it limits the object and the purpose which have been
described above repeatedly, and puts a system in practical use.

　It is in large numbers so that almost all the things that are the
objects of natural language processings, such as information scan,
information retrieval, information extraction, an information summary,
information indie クシング, and formation of an information database,
are included as an example of application of limiting an object and
the purpose. Therefore, bearing the test of a チュー ring in mind,
though, an application system is created most effectively, and
experiencing [ in_which practical use is presented ] becomes
important.

　In information scan, if the fine feature of the grain of language
expression etc. is used in addition to the constant watch-methods,
such as a judgment of the important word by the Tf/Idf method used
now, the purpose attains in remarkable quality. [ many ] Here, a
detailed conceptual description described previously is not
necessarily needed. Therefore, processing becomes very light. However,
processing time becomes large suddenly as details, such as conceptual
description, increase processing of a pattern match.

　Other application of information retrieval, information extraction,
an information summary, information indie クシング, the formation of
an information database, etc. can use the same technique for
fundamentally. However, as for the responsibility of a check final
with things naturally, any application is left to a user.

　Another method of realizing false intellectual processing is
utilizing the restriction language model according to the
object. There is the feature in the way of writing according to the
kind of document. Moreover, even if it adds restriction to the way of
writing according to the kind of document, fault does not arise in
many cases. Neither a manual nor a law has many which are considered
such [ from the first ] a point is taken into consideration, and are
written. Clarifying a meaning by expressional unification-ization etc.
in how carrying out paragraph composition, how using a term, the order
that writes a rank element, and the end of a sentence etc. is actually
performed. By taking in such a feature to a language model, even if it
does not prepare complicated language knowledge, it is considered [
that false intellectual processing as a result is realizable in many
cases and ]. Moreover, even if it does not prepare knowledge about a
term in the form which was extensive and was complicated, it is
thought that false intellectual processing as a result can be realized
now by surface-processing.

　That whose reliance is most possible by the present processing is
based on surface-processing of language expression. Though the
accuracy whose syntactic analysis is about 80% is attained, for
realizing advanced intellectual processing in accuracy of this level,
it is too uneasy. Morphological analysis has the risk of bringing the
incorrect processing with the fatal mistake of 唯一つ in advanced
intellectual processing, although accuracy is good more. Simple
processing will be performed to the bottom of comprehension that it is
that by which the error is contained in the processing result, and
also avoiding the fault and dispute by incorrect processing will be
permitted by what it be supposed that a user is to blame for a final
check. And it is also demonstrating practicality sufficient with
information scan etc.

2.4.6 Man's Understanding Capability

　There is what is surprised in a decipherment of the latest gene
information, and an alteration of the kind by gene
manipulation. dangerous -- considering -- coming out -- being certain --
と -- だい said [ make / a biological meaning understanding machine /
will use technology, clone technology, etc. concerning these genes,
and / although considered, if formal modeling of man's intelligence
stops being in prospect ] -- it た考え may also appear on mind in SF
When a machine assumes the idea that he cannot be solved, in
calculation theory, man can solve all man's intelligence. There may be
a method that the bird which flies empty is seen and an airplane is
developed, and a method that empty is referred to as riding on the
bird which flies in the trial which is going to create and use the
intellectual machine which is helpful even if it takes a position on
human mechanism and it is a partial function of it. Although it is not
thought that the latter is desirable from it being cold high up in the
sky, and probably being unable to bear etc., it is one of these
methods surely to make a carrier pigeon carry a letter.

　When it is going to attain intelligence-ization of a machine, the
standard is not thought of other than the test of a チュー ring, as
stated repeatedly. It is thought that understanding in man is an
action-response and internal memory. The latter becomes the origin of
a response by the test of a チュー ring, the former concerning the
test of a チュー ring itself.

It will not mention that corporal activity says in connection with a
response etc., and it is thought that the phenomenon of understanding
is reacting pertinent to an input under the knowledge at the time of
communication, an external situation, and the state of the heart as
stated previously. At this time, knowledge is illustration-ization
under an external situation and the state of the heart. (インス
tongue ティエーション) It is carried out and it is thought that the
phenomenon of understanding produces it by assimilating to an
input. Semantic understanding is exactly such a phenomenon, but it is
also said that the phenomenon of understanding becomes being only one
form of the calculation theory-unification of many so-called
information under these preparations.

If the capability of understanding of man is considered in this way,
the modeling and implementation will be comparatively tinctured with
implementability. However, it is claimed that it will limit an object
and the purpose to as small a thing as possible, and will try actual
system construction as mentioned above since description of knowledge,
an external situation, and the state of the heart is considered to
become complicated composition in a huge quantity.

2.4.7 Accumulation and Share of Knowledge and Technology

　In research and development of natural language processing
technology, accumulation of knowledge, data, and technology may be
able to regard the situation which is not desirable of necessarily not
being recognized clearly as it being. Even research / development
project promoted on a large scale with great fanfare may be considered
that the side which cannot deny such touch is also felt. It may be
explained from the ability of the contents to necessarily even have
not performed the share of a result.

　As a concrete example, neither the dictionary nor the syntax rule is
produced commercially, for example. Each researcher and developer
overlap and has to perform the same work as other people in
himself. If the reason originates in the strike radio-and-TV G on
commerce, in order to lose such futility, they are developed and
managed in a neutral organization or a public organization, and it is
thought that it must be made to guide shared implementation.

　Standardization may also be securing one means realizing a share of
knowledge etc. The organization which defines and carries it out also
with this standardization is made open, and although built, you have
to prepare for a share of the future. If an example is taken by the
reality expected that various kinds of knowledge serves as a huge
quantity, such a measure will have to be realized under powerful
leadership.

　The appraisal method of the system studied and developed is
considered to be the factor which it has also become that it is not
clear that it is hard to make accumulation of knowledge or technology
recognize. I can think that an objective valuation basis applicable to
all from the viewpoint that the test of a チュー ring is the standard
of evaluation cannot be set up. However, the system which limited the
object and the purpose is considered that it can set up a quite
effective valuation basis.

　In the committee of the past of this committee, about ten years ago,
it inquired about the valuation basis about a machine-translation
system, and the valuation basis was actually set up. The result was
published as a report exceeding 100 pages, and also published the
English translation. The argument about a machine-translation
valuation basis and the international conference are held, and some
demands of copy sending of this report of ours are coming in the
foreign country today. Although our result was globally esteemed from
that time, since it had preceded the time when the concern in many
foreign countries increases no less than ten years, it is considered
to be the place and なつた thing by which the real value is
reappraised only after becoming by the end of today. Probably such
tradition must be inherited and usual natural language processing and
beginning research of the valuation basis of intellectual natural
language processing must also inquire.

2.4.8 Conclusion

　What is thought of was prosaically described about construction of
knowledge detailed about the title of the natural language processing
technology for intellectual document processing in Toyotomi, modeling
of semantic advanced understanding, realization of the false
intellectual processing system by simple processing, etc. These
explanation may be what gives an impression that it is only repetition
of longtime research etc. However, it will not be meaningless to
reexamine also including an old subject in the present condition that
a meaning, understanding, and knowledge and that language are easily
used too much over many years, and moreover the realization has not
fulfilled demand at all, either.

　It seems to be the only approach to realize intellectual technology
of natural language processing in the form which responds to the test
of a チュー ring directly. Therefore, language knowledge and general
knowledge must be processed electronically, the pattern for processing
must be built, and the processing system using it must be established.
since these knowledge serves as a huge quantity and it becomes
complicated composition -- first -- an object and the purpose --
limiting -- the inside of it -- intellectual processing -- truth --
you should take the tactics of challenging realization

　The request to establishment of the advanced natural language
processing technology for on the other hand the information which
wants to come and know the Internet time coming to hand in a suitable
form efficiently by easy operation has increased increasingly. Much
time and efforts are still required by establishment of intellectual,
very advanced natural language processing technology. However, there
is a lot of application to which being simple and false intellectual
processing can also respond to the present social request
partially. Therefore, it is important to pursue these both and to aim
at improvement in technology gradually.

　It is necessary to continue research not only about intellectual
natural language processing but about un-intellectual and formal
processing of the syntactic analysis mainly studied now. These are
because a possibility that it is useful to performing efficiently
processing of the thing in which formal modeling which can be started
out of intellectual processing is possible is size.