ブリスベーン学会

　　　　　　　　NTT Information Network Systems Laboratories　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　　

　　　　　　　　　　　　　　　Satoru　IKEHARA　　　　　　　　　　　　　

○ Thank you very much, Mr. Chairman, for your kind words of introduction.

○ I am also thankful to Professor Rigsby and the Organizing Committee▼for the

　 opportunity to visit this beautiful country▽and to get to know my

　 Australian and other colleagues better.

○ I am particularly honored to be here▽to speak at this Anniversary of your

● Today I would like to speak about CPT and MLT method.

① I think it would be appropriate to start by some introductory remarks about

　 myself▼and the research efforts exerted on NLP by our Laboratories.

② Almost 20 years have elapsed▽since I joined the Electrical Communication

・ The first 10 years were spent in research work on▼Form. Al.Mani. and traffic　 theory, with emphasis on QN.　

・ It is therefore 10 years▽since I started research work on NLP.

① NTT currently has 13 laboratories.　

② Research on CLs have been conducted for long time▼at basic research

③ Research on NLP were started▽at the Laboratory in Yokosuka 10 years ago.　

・ I was involved in this research effort from the very beginning.　

・ From a staff of three in the beginning,▼the number of researchers has grown

④ In the area of NLP,▽I have undertaken research efforts in▼

　　　・J text to speech,　　　　・key word extraction,

　　　・error detection and correction of J text,　　

　　　・and statistical approaches for voice recognition.

・ This research effort continues on today.　

⑤ I am now engaged mainly in J-to-E MT.　　

-----------------------------------------------------------------------------

① While research work on CLs and NLP resemble each other,▼they are not the

　 same and it would be adviseable to handle them separately.　

② My current research efforts have been concentrated on NLP and not on CLs.

・ It is not research on how a NL can be transcribed by a theoretical model.

・ It is research on how can J, that is actually in everyday use, can be

　 translated into understandable E.　

・ My research efforts are directed toward realization of this method.

③ Upon starting research work on MT,▼a major problem was a considerable gap

④ It can be stated that▽if we apply the results of CLs to actual every day

　 words,▼there arise many exceptional cases and appropriate processing is

-------------------------------------------------------------------------------

× It may be that if CLs would achieve further progress and development, these

　 problems would cease to exist.　

× But can we afford to bide our time and wait for such an event before

　 proceeding any further with our MT system ?

× Our Laboratory is an institution engaged in information processing.　

× In conducting MT research, regardless of the method, the final objective is

　 to make possible the translation of actual words.　

× To await the development of CLs, would make it difficult to start MT research.

× We would be denied the budget for this purpose.　

× It was no easy task for us to obtain approvals for conducting research on MT.

--------------------------------------------------------------------------------

① NL is something like a custom or habit▼that has spontaneously grown up

・ It reflects the very human society▽in which it was born.　

② It is no easy task to explain each individual phenomenon▼without conflict by　 a unified model.　

③ If Ls are a social custom, there should be a method of processing that

④ Researches of NLP require the pursuit▼of every one of these words.

① I have decided to begin research work▼from the viewpoint of NLP and not CLs.

② Accordingly,▽I have sought a new approach amidst the results of

　 conventinonal linguistics,▼promoting a new methodology.　

③ The background for my research work is▼the CPT of L by Motoki Tokieda,▼who

　 followed up on the work▽initiated in the 18th century by Norinaga Motoori.

・ Our MT research efforts have been based on this theory▼and has resulted in

　 a MLT method▼and a test product J-to-E experimental MT system.

① Today,▼I would like initially to introduce the outline and experimental

　 results▼of our MT system in the half of my talk.　

② In the latter half,▼I propose by describing our thinking and methods▽making

　 comparisons with the thinking behind CLs.

-------------------------------------------------------------------　　　

① Let us start with the relation between our puporse and the ccurrent methods

② As you are aware,▽MT methods have advanced▼and many commercial systems have

③ Our major concern, at this stage, is J-to-E translation.

④ JL is very different from European Ls.

・ The quality of J-to-E translation is not good▼compared to French to E or

・ Current methods can be classified as a word-to-word translation,▼or literal　　translation.

⑥ JL itself has been enhanced by importation of European culture in the M-Era.

・ Therefore, JL has certain expression patterns▼that are easily translatable

・ These parts can be translated by current methods.

○ However,▽conventional J expressions are required to be rewritten into easily

⑤ That is, pre-E is inevitable▼in order to apply the current MT system to

⑦ The purpose of our research is to realize communication services with

　 translation.　　　　・ We hope to realize direct communications.

・ However,▼current methods are not satisfactory in realizing such services▼

⑧ Thus, our aim is to achieve translation without pre-E.

⑨ Let's look at the manner in which pre-E ia conducted▽to know about the

　 limitations of current methods and their causes.

-------------------------------------------------------------------　　

① How should original sentences be rewritten before translation ?

② Let us look at the examples in this figure.

・ The four essential points of pre-E are shown.

③ First, every word should have only one meaning.

・ This example shows the J verb ｢kakeru｣▼having, in this case, six different

　 meanings,▽pour, place, build, sit, mop, and play.

・ Such J words need to be rewritten into other words▼which correspond to a E

　 word,▼but there are cases which present difficulties in rewriting.

④ Second, structures of sentences and expressions should be simple.

・ In this example, the adjective ｢utsukushii｣　meaning beautiful,▼is placed

　 in front of the pronoun (I, me or my).

・ This can resault in two interpretations,▼"beautiful me" and "beautiful

・ To avoid these ambiguities,▼modifiers should be placed immediately in front

　 of the word to be modified,▼as in this rewritten example.

⑤ Third, supplementation of elliptical subjects and objects will be required.

⑥ Fourth, idiomatic expressions and metaphores need to be rewritten into other　　expressions.

⑦ It can be stated that▼these requirements have derived from limitations of

----------------------------------------------------------------

① Let's consider the translation methods.

② Conventional translation methods can be classified into two types, the pvt-M

③ Among researchers of MT, there have been arguments as to which method is

○ Among the translation systems developed by private companies, there are many

・ However, there have been none that can be stated to have truly realized the

・ This is because universal intermediate L that can be commonly applied to many　 L cannot be designed.

○ NL has primarily originated by reflecting the perpsectives and thinking of

　 the group of people using such L.　

・ It would appear unreasonable to consider a intermediate L that would be

　 commonly applicable to every L which differs in thinking and perspective.　

○ When a translation between Ls which are very close each other is considered,　　or if a rough ranslation is to suffice, the pvt-M may be satisfactory.　

・ But in either case, I think pvt-M has a basic misunderstanding of Ls in its

① The trs-M can be stated as being more realistic compared to the pvt-M,

　 because it does not assume that intermediate L is universal.

② However, if intermediate L can be thought of as a meaning expression such as　 deep structure, the same difficulties as in the case of the pvt-M would arise.

③ The meanings of surface structure cannot necessarily be retained by deep

-----------------------------------------------------------------　　　　

○ Let us consider the manual translation method.

① In the expression that is to be translated, the state of the objects as

　 recognized through the eyes of the speaker is tied with the expressions.

・ And at the sane time, the awareness of the speaker toward the object is also

② To tie the speaker's recognition with the expression, a rule of words,

　 i.e. linguistic norm, is used.　

・ This norm differs with every L, such as in E and in J.

○ A human translator experiences the speaker's recognition through the

・ He is required to know both the state of object and speaker's recognition

・ These contents are expressed in the framework of the target L▼on behalf of

① This figure shows▽what is objective in this path and the what is subjective

② Thus,▼translation is conducted through a re-grasping of the speaker

③ A simulation of this process is the MLT method which we proposed.

-------------------------------------------------------------------　　　

② The source L is J, and the target L is E.

③ The term "Multi-Level" indicates that▼translation is being conducted at

① This system is divided up into 2 major conversion paths.　

② The one is the conversion of the speaker's sense or what is subjective.　

③ The other is the path for conversion of the description for the object.

・ And this path splits up into 3 more paths,▼depending on the abstraction

④ Thus, the MLT method has currently four translation paths.　

○ The major difference between this method and the conventional methods is▼

　 whether the speaker's recognition and the object are regarded separately

-------------------------------------------------------------------　　　　　　

○ To realize a MT system, a knowledge of L is essential.

○ NLP can be stated as being a battle with ambiguity▼from the beginning to end.

・ In the translation process,▼various types of ambiguities arise,▼but this is

　 regarded as being a lack of knowledge that is required.　

① To overcome such ambiguity, it is necessary to study what knowledge is

　 lacking and to have such knowledge established as the rule or as a dictionary.

② It is important to consider the type of ambiguity and relationship▼with

　 the knowledge corresponding to it.　

○ In our MT system,▼we place importance on knowledge for MP▼for the purpose

　 of solving ambiguity of sentence structure▼and of the meaning of individual

○ We have collected 2 major types of information▼and have compiled them into

　 a form of dictionary▽as shown in the figure.

① When the speaker expresses an object, abstraction is conducted and the object　 is conceptualized.　

② Details will be mentioned later, but in L expressions, the concept of object

　 is expressed separately in terms of a concept of substance and a concept of

○ Linguistic conventions which combine concepts to expressions have therefore

　 been compiled into 2 separate dictionaries.

○ The one is knowledge related to semantic use of words.　

・ This has been compiled in a SWDic.　

① The other is knowledge related to meanings of expression structures.　

② The meanings of expressions are not necessarily represented by the sum of the

　 meaning of every word in the expression.　

③ In fact, it would appear thet the meanings of linguistic expressions actually　 used cannot be explained　solely by the meaning of each word used in the

④ I believe that▼expression structures need to be considered▽as units of

⑤ Based on such thinking, the abstraction of expression structures centered

　 around declinable words and compiled as units of meanings▼has resulted in

⑥ This is one of the important points asserted in our system▼and I shall be

　 dealing with this factor later in this presentation.　

○ I would like to state clearly that▼the knowledge we have discussed here is

　 a L knowledge▼and not a worldwide knowledge such as general commonn sense

-------------------------------------------------------------------　　　　　　

○ As discussed in the foregoing,▼we have compiled a SWD and a SSD.

○ The meanings of words and their translation are listed in ordinary

・ But it does not serve to have the computer know▽how they are to be used.

・ For example,▼the word "school" is used as a "place"▽and also as an

○ As a means of desscribing these knowledge,▼we have compiled the use of the

① It is important to consider precision of knowledge description▼in relation

② For example,▽an algorithm which was valid in a world of 1,000 words▼may not

　 necessarily be valid in a world of 100,000 words.　

・ Experiences indicate that▼increase of the number of words causes rapid

③ It is a rule of science that▼quantity gives rise to changes in quality.　

④ We need to keep this rule in mind, especially in NLP.

① Seeking to establish practical method, we have collected words including

　 proper nouns normally used day to day and have compiled a dictionary of

　 about 400,000 words for our system.

① Thus, it is necessary for SA system to work with this environment.

② We began with about 500 categories.

・ But this has resulted in that translation of verbs cannot be precisely

③ With the consideration of our aim and environment,▼we set up 2,800

　 categories for general noun SAs▼and 200 categories for proper noun SAs.

① Working with these SAs,▼we have compiled a SWD consisting of 400,000 words▼

　 and a SSD consising of 15,000 sentence forms.　

② The SSD is in a Valentz pattern form▼and the index words are verbs and

------------------------------------------------------------------------------

① This figure shows the nodes▽for the top 4 levels▼of the SA system for

・ General nouns are branched off into concrete and abstract,▼with concrete

　 branching off into subjects, places, and concrete objects.

② There is a total of 2,800 category names.

③ Here, I have taken one example at the deepest level.　

・ The depth of category amounts to 12 levels at the deepest level.

----------------------------------------------------------------

① Next we have the SA system for proper nouns.

② Attributes for proper nouns are divided into 4 types,▼the names of persons,

　 of places, of organizations and other.　

------------------------------------------------------------------------------

× Names of persons will stop at about the 4th level, but names of places go

　 further reflecting on the difference in use of names of persons as opposed

× Names of places are used in the form of "nesting" such as prefecture country,　 ,county, towns and aza (local sub-division) thus accounting for depth of categorization.

-------------------------------------------------------------------------------

① The number of nodes amounts to 200 and the maximum depth is 9 level.

② An example of the deepest node is shown here.

----------------------------------------------------------------------------

① Now, take a look at how these SAs are located in the SWD.　

② This figure shows the contents of one record within the SWD.　

② One record consists of 350 Bytes and some 200 to 300 various types of

　 information have been input into a record.　

③ SAs of words have been registered in the final 15 fields.

① In the SWD, each word has been given a lower level of attribute.　

② Conversely, with the SSD, higher level attributes have been used.

③ This is due to consideration for retaining the relationship between higher

・ Expression structures should be defined by generalized format as far as

------------------------------------------------------------------

○ At this point,▼we would like to see the differences of description

　 capabilities▼between of our SA systems and other similar systems.

○ Case 1 is a typical example of a J-to-E MT system which has recently been

・ The number of semantic categories are between 30 to 50.

○ Case 2 is an example of the dictionary of word concept▼being planned by the

　 Japan Electronic Dictionary Research Institute (EDR)▼established by MITI

　 (Ministry of International Trade and Industry) of Japan.

○ Case 3 is an example of our translation system ALT-J/E.

① A comparison was made of the three cases.

・ And, the capabilities of describing the translation rules for verbs are

② In the comparison, the case of ALT-J/E has been assumed to be 100%.　

③ The dictionary in the case of EDR is just in the planning stage so that the

　 values for Case 2 have been obtained from experiments conducted in our system　 based on similar conditions.

○ The results of this comparison are as follows.

○ The capabilities of description was 31% for case 1 and 59% for case 2.

① Those results indicate that▼when precision levels of SAs are low, essential

② Our experiences also indicates that,▼in J-to-E MT,▼precision level for some

　3,000 types of SAs are necessary▼for differentiation of translation for verbs.

③ However, this is not sufficient for translation of nouns.

・ I think that some other different system from the SA is required for

　 differentiation of translation in the case of nouns.

------------------------------------------------------------------

○ Let me show▽what kind of translation has become possible by our system.

① The first example is differentiation in the translation of the J verb ｢kakeru｣.

・ This is the same example as shown in the beginning of Pre-E.

② The verb in the J sentences are all the same ｢kakeru｣.

・ But, the translation by ALT-J/E shows▼poured, made, caused and other

③ J verbs in general▽need to be translated into many different E words.　

④ In the case of ｢kakeru｣,▼there are about 15 different ways of translation

　 for general use▼and to include idiomatic usage,▽there will be a need to

　 consider some 80 types of translations.　

○ The verb with the largest number of variations would be ｢suru｣▼with over 300

① The second example shows differentiation in the translation of the noun ｢mure｣.

② The same word ｢mure｣ is used 8 times in the J text▼but the translations are

③ In this case,▼the E word varies depending on what is referred to.

------------------------------------------------------------------------------

× According to a native E speaker who was interviewed, the word "bevy" here is　　used only in the case of beautiful women.　

× It was also pointed out that "mod" sounds strange.　

× It would appear that we need to change our rules somewhat.　

× Here, it would suffice if our audience can appreciate that differentiation

　 needed in the translation of nouns has also progressed to certain level.

---------------------------------------------------------------------------

○ The next is an example of an automatic rewriting function of J sentences.

② The literal translation of this sentence is＝（英語を読む）

・ I think this E may be understood,▽but it does not sound like proper E.　

② The reason for this can be seen clearly▼as having excessive number of verbs.

○ J is a L of "circumstantial logic",▼and it necessarily uses many verbs.　

・ In contrast, E is a L of "substantial logic"▼and is said to prefer

① Thus, to reduce the verbs,▼a partial rewriting of the J text in the figure

② This rewriting is performed by the translation system automatically.　

③ After automatic partial rewriting,▼we get the translation as =（英語を読む）

・ I think the quality of E has improved.

--------------------------------------------------------------------------------

×① It goes without saying that it is our goal to think through J texts that

　　are difficult to translate and to translate these into E that is readily

×② Yet it is another alternative to change Japanese text that is difficult

　　 into one that is easily translatable automatically.　

×③ This method is particularly convenient when a translation function that is

　　 already completed can be applied without having to establish the function

------------------------------------------------------------------------------

○ The next example is supplementation of subjects and objects.

① The J convention is to refrain from mentioning▼what is already known to the

② Lengthy, redundant expressions are not popular.　

③ Particularly, the subject and object are omitted unless absolutely necessary.

・ Because▽these can generally be judged by the partner▼by joshi and auxiliary

① In contrast,▼in E,▽both subject and object are generally necessary.　

② There are cases▼when the subject is missing and the object is turned into

　 the subject▼and rewriting into the passive voice.　

③ But this is not always appropriate.　

① Here are two sentences which are linked to one another. （日本語を読む）

② The first sentence has the subject and object,▼but the second sentence has

③ Therefore,▼ALT-J/E has supplemented the first subject by the object of the

④ And the latter subject is supplied by the subject of the first sentence.

○ I have shown some of the new functions that have been realized by ALT-J/E.　

○ I believe that,▼with the realization of these new functions,▼the

　 possibility of MT without pre-E▼is now real possibility.

－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－

① The foregoing is the initial portion of my presentation.

・ I have introduced the MLT method and our MT system ALT-J/E.

② In the latter half of my presentation,▼I would like to mention details of

③ If there should be any questions on the discussions up to this point, I shall　 be happy to answer them.

----------------------------------------------------------------------------

① I shall begin the latter portion of my presentation.

② I have already mentioned that▼our research efforts are being conducted from

③ Let me discuss the thinking behind our MT system.

① The background to our research lies in the CPT of L.　

② This theory was advocated some 50 years ago▼by a J linguist, Motoki Tokieda.

③ Tokieda took after the J grammar▼which was established some 200 years ago by　 Norinaga Motoori,▼and introduced the CPT from a standpoint of▼criticizing

　 the philology of Ferdinand de Saussure.

① First, I wish to present the differences▼between the concepts of CLs and of

② I shall follow by discussing▽how the CPT is related to our MLT method.

-------------------------------------------------------------------------------

① Let's consider the background of the CLs, first.

② In considering logics,▽there has been two types of logics since the olden

② This type of logics asserts that▼actual existence should be thought of as a

・ This type of logics has appeared in the era of ancient Greece.

② This logics insisted on purity and handling of an ideal world.

・ And it cut itself off▽from the realistic world.

○ Symbolic logics and computational logics are derived from the latter, formal

① There appears an effort to apply CLs to NL.　

② But NL is a L▽that deals with the realistic world.

③ This would seem to be the cause of various problems.

・ For example,▼Frege's principle, which is compositional semantics and

　 constitutes a major assumption in CLs,▼is not valid in a NL.

----------------------------------------------------------------------------

-------------------------------------------------------------------Fig.15

① Let us look at the differences between the J and Ind-European Ls.

② The Ind-European Ls have sentence patterns such as 5 patterns▼which are

　 relatively independent from contents.

・ These Ls can be classified as a suitcase type L.

・ Structualism was developed▼from the viewpoint of importance of forms and

・ Chomsky took notice of the importance of contents to overcome the limitation　　of structualism.

③ In contrast,▼JL is relatively free from forms▽and does not have a rigid

・ J can be classified as a 「furoshiki」(wrapping cloth) type L.

・ In other words, the shape of contents appears in a form of expression.

④ Thus, contents are important in J processing.

⑤ In J, there are 20 different types of grammars.

○ Here, I would like to take up two▽out of the four.

・ This grammar is strongly influenced by European philology.

・ And this grammar was adopted for use in school▽by the Ministry of Education

・ Another grammar is Tokieda grammar▽which was derived from Norinaga Motoori.

⑥ As nature is a complexed body of processes▼that develops with conflicts as

　 the motivating force,▼so also is L.

⑦ Tokieda established the CPT of L▽based on this idea.

----------------------------------------------------------------- Fig. 16

① This figure shows the differences▼between the CPT and Chomsky's concept.

② CPT asserts that▼a L should be considered as a complex body of processes▼

　 comprised of objects, speaker recognition and expressions.

・ This process is explained by the reflection theory.

④ #Recgnts are combined into expressions using norms,▽namely the rule of Ls.

・ #Ling-norms are national rules▼which have been fostered in each community.

・ These norms are referred to as a grammar▼in the broad meaning of the word.

⑤ In contrast,▼Chomsky interpreted L in the form of a dual structure,▼a

　 surface structure and a deep structure.

⑥ The existence of deep structure has been assumed▼as the meanings of surface　　structure.　

・ This is certainly a strange theory▽where because meanings are assumed for

・ It would further appear that there would be difficulty in explaining the

・ More recently,▼he appears to be attempting explanation based on　mystic

・ He says▼L cannot be explained without assmuming that▼humans are born with a

⑦ Well, results are achieved from causes▼and these results, in turn, become

　 the causes for the next series of results.　

・ Nature is a composition of such processes▼and it is the conflicts and

　 discrepancies which activate these processes.

・Conflict is energy.　=== ・I believe this is the thinking followed by science.

・Language is a spiritual by-product of humans.　

・There is room for various theories to arise▼based on how the human spirit is　 to be considered.

----------------------------------------------------------------- Fig. 17

○ Let us now see▽what happens▼when we regard L as a causal sequence▽between

○ The world of objects consists of a speaker and other objects.

・ Objects are composed from substances, attributes and relations.

③ The speaker recognizes the object world▼and at the same time, recognizes

・ In recognition of objects,▽conceptualization takes place.　

○ There are two occasions regarding recognition of self.

･ One is the occasion where▼speaker represents himself directly in the

　expression without conceptualizing himself.

･ On other ocassions,▼he conceptualizes himself and represent it as an object.

○ When he conceptualizes himself,▼the phenomena of self disunion take place.

・ A separate imaginary self other than the actual self appears.

・ And the self conceputualizes the actual self.

○ Recognition related to subjects and objects is represented▼by two types of

④ SEs represent speaker emotions and intentions.

・ These expressions are expressed in the JL▼by post positionals and adverbs.

⑤ OEs represent conceptualized objects.　

・ And these expressions are generally expressed by nouns and verbs.

⑥ This relation is somewhat different in E.

・ Inflections usually express SEs in E.

------------------------------------------------------------------ Fig.18

① Let us take a look at this example.　　　== 日英単語対を読む＝＝

② This sentence means▼"He will also want to go to America".

・ A similar view is witnessed in the grammar of Port Royal in France▽300

⑤ The difference is an important factor in the translation method.

・ I plan to mention this later in relation to the MLT method.

------------------------------------------------------------------

● Here, let us proceed to the problem of meanings.

① Recently, the importance of semantic processing in NLP is▼being strongly

・ Much efforts have been devoted to semantic or meaning processing,▼but it is

　 felt,▽in many cases, the definition of meaning is ambiguous.

○ I would like to point out here what we regard as meaning.

② Let us consider the structure of Ls.　

・ If we assume that meaning is a substance,▼it is one of either "object",

　 "recognition" or "expression"▼and the fourth element of "interpretation".

・ If we say, at this point, that▼meaning is born out of forms,▼we would

　 arrive at a "formal semantic theory".

③ Next, if we say that forms are born out of contents,▼this would result in a　　"object semantic theory" or "recognition semantic theory".

④ If we assume that objects and recognitions are meanings,▼how should we

　 explain the sentence that has been erroneously written.　

・ This would result in a sentence that is incorrect▽in which something that is　 opposite is written,▼yet stating that the meaning is correct.

⑤ Further,▼the object and recognition can never remain the same indefinitely

・ Then, the meaning of the sentence will change independently from expressions.

⑥ Motoki Tokieda chose eclecticism and took meanings to be listener's reactions.

・ Tsutomu Miura took after the Tokieda grammar,▼but improved the theory and

　 advocated the "relation meaning theory".

⑦ He proposed that▼structural processes from object to recognition are

・ He explained the meanings of linguistic expression as the relationship▼

　 between object, recogntion and expression.

⑧ This kind of relationship is concrete and specific.　

・ As long as the expression exists,▼so will the meanings.

・ When expression is cancelled out,▽so will all relationships,▼and the

⑨ The concept of regarding relationships as meaning▼resembles recent situation　 semantics,▼but actually, the two are entirely different.

⑩ In situation semantics,▽the meanings of expressions▼and the meanings re-

lated to the location in which the expression is placed▼are completely mixed up.

・ The Miura grammar features a clear distinction▼between linguistic expression　 and expression of locations.

----------------------------------------------------------------- Fig. 20

○ How do we go about MP▼from the viewpoint of "relation meaning theory".

○ Meaning is the relationship between object, recognition and expression.　

○ SE involves the emotions and intentions of the speaker,▼whereas OE expresses　 the status of the object▼as viewed from the eyes of the speaker.

・ Thus, MU would involve re-tracing such relationship▼and to relive and

　 experience the status of object and recognition.

④ Here, let us think of MP in two steps.

⑤ The first step would be the process▼which designates the rules or

　 conventions used in linguistic expression.

⑥ Conventions regarding Ls are complex and liable to be construed in many ways.

・ The listener must identify the convention used by the speaker.　

・ I would wish to define this act of identifying the convention as "MA".

⑦ The second step would be the process of identifying the recognition of the

　 speaker▼and the status of the object that is tied in with the expression.

・ Here, we would like to refer to this action as "MU".

⑧ Therefore, MP consists of two processes,▼"MA" and "MU".

⑨ The knowledge that is required for MA would be linguistic knowledge,▼that is,　 conventions or norms regarding Ls.　

・ In contrast,▼MU would require knowing the status of objects▼and this would

　 require general knowledge, that is, worldwide knowledge and knowledge in

⑩ Let us go over the next example. ------------------------- Fig. 21　　　　　 ----------------------------------------------------------------------------

① This diagram is an example of MA.　　　=日本語と英語を読む=

○ The meaning of the sentence is ------------.

○ Let us think about the convention involved in these words.　

○ The meaning of the word "takai" is expensive, high, noble, loud among others▼　 but here, it is used with the meaning "expensive".　

○ "Abura" has various conventions as shown in the figure,▼but here the

　 convention expressing the meaning of oil is used.

○ Thus,▼MA is identification of the convention▼which the speaker actually

　 used among numerous conventions▽that exist in the L.

・ MA cannot be achieved▼by merely staring at the words one by one.

① Let us look at the lower sentence.　==日本語を読む＝＝＝

② Look at this portion of the sentence.　背の高い meaning the "back is high".

・ In E, one word "tall" is used to convey this meaning.　

・ But looking at back and high separately▼will not bring the concept of tall.

③ The same can be said of the other expression.

・ Sell and oil jointly means idle away one's time in this case.

④ Thus,▼conventions regarding words can never be decisive▼if they are thought

② I said that▼the meanings of expressions are the relationship between objects,　 recognition and expressions.

③ Structrures of objects reflect themselves into speaker recognition.　

・ These structures are also combined to expressions.

① This means that▼the structure of an expressinon is a part of the meaning.　

・ Structures cannot be separated from meanings because structures also

② Thus, in MA also,▼there is the need to think the relationship between

・ More specifically,▼it is important to grasp the structure of expressions as

------------------------------------------------------------------ Fig.22

① As discussed in the foregoing,▼MU involves the identification of the

　 speaker's recognition▼and the status of object as tied in with the

② To know the speaker's recognition and the status of object,▼the listener

　 must be equipped with a certain world in his mind.

・ This world must have certain elements in common with the world▼which the

・ To restructure a world▽corresponding to the expressions of the speaker,▼he

　 will be required to link up the elements of the speaker's expressions▼and

③ In other words,▼MU can be considered as linking the elements of the

　 linguistic expressions▼and within the world model of the listener.　

・ This action of linkage will structure a new portion in the listener's world.

④ In contrast to liguistic knowledge for MA,▼MU will require worldwide

⑤ Compared to linguistic knowledge,▽worldwide knowledge is massive▼and

　 considerable amount of difficulties will be involved in research.　

・ We are therefore conducting our reseach effort based on the following

----------------------------------------------------------------- Fig. 23

① We shall discuss MP in MT by dividing it into MA and MU.

② The relationship between our MT research and MP is as follows.

○ In MT, the contents of source text to be translated are diversified▼and it

　 would be difficult indeed to prepare a worldwide knowledge.

⑤ MT can be acceptable▼even if the computer cannot entirely understand the

　 contents of the text▼as long as the translation results can be finally

⑥ Replacing the linguistic conventions used in the JL▽by E conventions▼and

　 leaving it up to the judgment of the average person.　

・ This is the basic stance assumed by our research efforts in MT.　

○ First,▼we want to realize a translation based on MA.

○ Next,▼we will extract the expressions and concepts▽which could not be

　 translated by MA▽and attempt a MU with a limited world model.

○ In contrast,▼with areas such as telephone number information and database

　 inquiries,▼the target world can be contracted into a comparatively small

・ We are structuring world models and conducting research in MU for such fields.

----------------------------------------------------------------- Fig. 24

① The method of translation based on the foregoing MA▼is the MLT method.

② I have previously mentioned the need to observe differences▼between SE and

③ And, I have also pointed out the necessity of consideration▼about the

④ The abovementioned translation method has been structured based on these

⑤ The first point is considered as follows.

・ J and E are very different in presenting SEs.

・ Therefore,▽SEs are segregated from OEs▼and converted into E.

⑥ The second point is resolved by abstraction of expression structures.

・ Three levels of abstraction have been proposed.

・ The first is related to a level of idiomatic expressions.

・ The second concerns specific structures such as loosely coupled words.

・ The last concerns most loosely coupled structure that can be represented by

⑦ From the foregoing,▽MLT method consists of two major structures.　

○ From the viewpoint of linguistic knowledge,▼linguistic conventions are

　 divided into those▼which are related to SEs and those related to OEs.

⑧ Of the two,▼the knowledge related to SEs are smaller in scale compared to

　 OEs▼and have been solved in terms of rules or programming.　

・ But the OEs being massive,▼we have decided to compile them in the form of a

⑨ We shall next explain the method of summarizing linguistic knowledge▼related　 to objectve expressions.

----------------------------------------------------------------　Fig. 25

① I have already explained that▼the target world of linguistic expression is

　 structured by the object▼consisting of substances, relations and attributes▼　 and by the subject which is the speaker himself.　

・ Here,▼the objectified speaker himself is identical with substances.　

・ Thus,▼knowledge for OE can be considered as consisting of three types,▼

　 substances, relations and attributes.

② In terms of linguistic expression,▼substances and relations can be expressed　 by nouns.　

・ Thus,▼the knowledge concerning use of nouns has been summarized in a SWD.

③ Regarding attributes,▼there are a dynamic type and a static type▼which are

　 represented by verbs and adjectives, respectively.　

・ Thus, these have been summarized in a SSD.

④ As mentioned in the earlier portion of my presentation,▼linguistic knowledge　 has been summarized based on these ideas.

× As an item similar to the SWD, there is thesaurus.　

× In E, the thesaurus by Roget is famous and there are several examples in

× These have perceived the similarity between the meanings of individual words

　 and have made classifications accordingly.　

× Our word semantic dictionary features the viewpoint on which the word is

　 based and used and has thus added a semantic role.

----------------------------------------------------------------- Fig. 26

① I have presented an outline of the J to E MT system, ALT-J/E▼and the related　 conceptual background.

② The current state of NLP research at our research laboratory▼can be summed

③ First, morphological analysis▼which constitutes a considerable stumbling

・ I think the complexity is not less than that of syntax analysis or meaning

④ Technology involved has been almost completed.　

・ A remarkably high level of precision and accuracy has been achieved.　

・ Syntax analysis and MA are being undertaken in the course of MT research.

・ High levels have been achieved,▼but I do not believe,▼it is satisfactory

　 yet to be applied to newspaper article translation.

④ Research on MU has been initiated in question answering.

・ This will require more time and effort.

----------------------------------------------------------------　Fig. 27

① The main issues confronted by MT research▼and future research targets are as　 presented in the figure.

② Current subjects are ．．．．=＝No.1 を読む＝=

・ Our major research efforts are being concentrated in these areas.

③ Our future study efforts will be concentrated on ．．．．＝=Ｎｏ．２を読む＝＝

・ Preparations for these studies are now under way.

－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－－

○ Today,▼I have mentioned the thought and method of our MT system.

○ I am hoping that these ideas will become one of the bases of non-literal

・ The problem of non-literal translation has begun to attract attention▼as

　 witnessed by the fact that▼it was taken up at the IJCAI Workshop in Sydney

○ If sharing our experiences in dealing with this system▼would be of interest

　 to you,▼it would be our pleasure to do so.

○ This concludes my presentation,▽and once again,▼my heartfelt appreciation

--------------------------------------------------------------------------------

図２０はここからスタート（４０分番組の時）

● Here, let us proceed to the problem of meanings.

･ I can't discuss details of meaning theory here.

･ However, we have adopted Miura's theory which assert meanings are

　 relationship between object, recognition and expression.

･ This theory is called "relation meaning theory".

⑨ The concept of regarding relationships as meaning▼resembles recent

　 situation semantics,▼but actually, the two are entirely different.

⑩ In situation semantics,▽the meanings of expressions▼and the meanings

　 related to the location in which the expression is placed▼are completely

･ The Miura grammar features a clear distinction▼between linguistic

　 expression and expression of locations.

I have not discussed the details of structural meaning of expression.

However, I must point out here the necessity of consideration about the

meanings of structure as the second point of our translation method.