NL8702359A - LANGUAGE ANALYSIS DEVICE. - Google Patents

Description

* '. } 4 7 N.O. 34769 1

Language analysis Device

The present invention relates to a language analyzer and more particularly relates to a language analyzer for analyzing natural languages, which analyzer can be used in automatic translation devices.

The language analysis devices known from the prior art have a number of problems as described below.

When analyzing morphemes in a sentence, for example words etc., it is important to assess whether a particular word is used alone or whether it is used as a compound word or in an expression where it is linked with another word or with in other words, and this also has a significant impact on the outcome of the analysis.

Conventional analyzers employ a system for performing the analysis both in cases where consecutive words are assumed to be part of an expression as well as in independent words and ultimately an appropriate translation is selected based on the result thereof or uses a system in which such consecutive words are preferably judged as an expression. The former system requires a long processing time while the latter system has a high chance of producing an incorrect translation.

In morphological analysis, performed when generating a translation, a phrase or similar information of the morpheme, such as a word, is obtained from a glossary. Because the majority of common words such as nouns and verbs can be included in a glossary, they can be easily looked up for information.

However, because expressions such as indications for length 30 (m), speed (m / s), acceleration (m / s ^) and other units exist in many different types, it is not efficient to include them all in the glossary because the memory capacity for storing the glossary information is increased in a wasteful manner.

35 In a morphological analysis of a sentence, the numerical expression in a given language need not always correspond to that in other languages. For example, the basic concept of counting with numbers, that is, the positional motion, is different in 8702356 2 Λ i f% the Japanese language and in European languages such as the English language.

For example, if an English expression for a numerical value "a hundred and two thousand two hundred and four" is simply decomposed into its constituent parts and is merely replaced by the respective corresponding expressions in the Japanese language, then the analysis would it reads "100 and 2200 and 4". This should eventually translate correctly as "102.204", ie "a hundred and two thousand and four" in the Japanese language. In the morphological analysis performed on a translation, an entered sentence is divided into word list reference units by means of boundary attributes such as spaces, commas, and periods, and the word list is addressed by the words list reference units to obtain a phrase information and other information. . For example, in such a case, a proper noun can be used in two or 15 more different meanings depending on the context. For example, the expression "Osaka City" is used to denote a group as a subject as, for example, in "Osaka City has decoded ..." and also in the meaning of the place as an object as in ".... in Osaka City ". Since so far only one meaning has been stored for each of the proper names, these differences in meaning cannot be taken into account, which reduces the parsing accuracy during the analysis.

For example, in the English sentence ".... in the Central Park John Willson had a ...", the sentence cannot be correctly parsed unless it is recognized that a separation must be made between "Central Park" and John Willson "in this context. Likewise in the English sense" .... in Boston Mr. Baker was .... ", it should be recognized that a separation should be made between" Boston "and" Mr. Barker. ”However, in conventional systems, a series of such proper nouns is misrecognized as a single associated combination of words.

For example, when analyzing morphemes in an English sentence, a sequence of words beginning with capital letters is generally analyzed as a proper noun. However, if the words beginning with the capital letter continue, it is not always correct to recognize them as a single proper noun expression. There are cases when they actually form a number of proper noun phrases that sometimes occur consecutively. When parsing morphemes during translation translation, phrase information and other information is obtained by looking up a glossary. 870 2 3 rc '4 · I' ï 3 list. Because in such a case most common nouns, verbs etc. can be stored in the glossary, they can be easily retrieved to obtain the information.

However, since there are many different types of proper nouns, it is impossible to put them all in a glossary. Therefore, those proper names that are not registered in the glossary cannot be recognized as proper names.

If an array of characters, combined in a particular pattern, is present, then there is a high probability of erroneous parsing if a regular word unit operation, which is applied to common sentences, is applied for such an array of characters and this could potentially result in a meaningless translation.

For example, in the case of the English array of characters "sun-15 day, Jan 26, '80" that displays a timeframe, translation into Japanese creates only a series of nouns and numbers such as "Sunday, 26, Jan ., '80 ".

Furthermore, because derivative words are not estimated on the phrase and on the semantic nature, translation depending on the case cannot be obtained.

The content-free grammar, the so-called cfg grammar used for analyzing a language, has the drawback that many unnecessary solutions are generated that ultimately cannot be used in a bottom parsing process and a top-down analysis process in the sentence. Many of these redundant solutions are seemingly recognized as errors when they are actually read. However, because a structure transformation or translation formation is also performed for these redundant solutions and the correctness of the result is judged in respective processing steps, they result in a large amount of redundant processing time.

For example, the English word "let" has both the meaning of an instruction and an invitation and therefore the decomposition must be performed for the respective possibilities which reduces efficiency. Furthermore, it is difficult to select one of them.

35 In addition, in the English language, a number of words can be combined by hyphens such as "take-care-of-him attitude" to freely form an adjective group. However, it is difficult to treat it with the usual degradation grammar.

40 The same situation also applies to additive interrogation. Although the 8702 35 § 't * 4 form of the additive interrogation in the English language is very limited, it requires extremely complicated processing within the usual analysis methods. Furthermore, it is not easy to determine which verb the additional question related to.

The present invention has been made in light of the above-mentioned problems.

A first object of the present invention is to provide a language analyzer with which to assess the degree of coupling between two consecutive words and to judge whether they are an expression or not, based on the result thereof.

A second object of the present invention is to provide a language analysis device capable of morphological analysis of an input sentence containing an assembly of the array of characters as in dimensioned units, without all of these character arrays in a glossary must be saved.

A third object of the present invention is to provide a language analyzer that can perform an appropriate morphological analysis on an expression containing numerical values.

A fourth object of the present invention is to provide a language analyzer capable of translating a proper noun into a meaning associated with its contaxt.

A fifth object of the present invention is to provide a language analyzer with which morphemes can be correctly parsed with respect to an expression containing a series of consecutive proper names.

A sixth object of the invention is to provide a language analysis device with which an unregistered proper name can be processed and with which an appropriate analysis of a proper name can be performed taking into account a relationship with phrases that occur before or after.

A seventh object of the present invention is to provide a language analyzer that allows proper analysis of morphemes with respect to an array of characters having one particular meaning by being combined according to a particular rule.

An eighth object of the present invention is to provide a language analyzer capable of estimating the grammatical nature, 8702359 if -τ 5 meaning, etc. of a word recognized as being a derivative, according to a predetermined certain rule.

A ninth object of the present invention is to provide a language analyzer with which to recognize the structural properties of an entered sentence and to perform a decomposition according to these properties.

The first object of the invention can be met by a language analyzer, comprising: a glossary memory in which glossary data is stored including morpheme data for words, compound words and phrases, and a parser for performing a morphological analysis for an input sentence with reference to said vocabulary memory, wherein said vocabulary memory contains data indicative of the degree of coupling between each of the words belonging to compound words or phrases and references said parser to the vocabulary memory for the respective words are present in the entered sentence and, if some word list data is found for a word combining with other words, then the combination of words with a higher degree of matching is selected by referring to the degree of matching information.

The second object of the invention can be met by a language analyzer comprising: input means for inputting a character array in a predetermined language, a basic glossary memory used to search for the input character array and containing basic data stored, and a parser for analyzing the input character array by searching the basic glossary wherein said parser addresses the basic word list memories using the input character amy and if thereby finds a portion of said character array, then the basic glossary memory is addressed in the same manner with other parts of said character array to thereby analyze the entire character array.

The third object of the present invention can be accomplished by a language analysis. Device comprising: a glossary memory in which glossary data is stored for each glossary reference unit, and a parsing memory for dividing an input phrase into 5 words. and for performing a morphological analysis of said words list reference units with reference to the glossary memory, said glossary memory at the glossary data having distinguishing indicia indicating whether the words list reference units represent numbers and said parser referring to the word list memory using the respective words list reference units present in the entered sentence and if said distinctive indication is present in the word list found, then the word list is reference unit with which said distinguishing indication is found combined with a glossary reference unit which is present near said vocabulary reference unit and with which another distinguishing indication is found, the numerical values represented by the two 20 word list units are calculated together into a single numerical value and the words ice reference units are transformed into a single parsing unit.

The fourth object of the present invention can be met by a language analyzer comprising: 25 input means for inputting a character array in a predetermined language, a glossary memory used for looking up said input character array, search means for searching in said dictionary memory 30 using the input character array, and type information providing means for providing type information in a character array not registered in the dictionary memory and in a character array whose type information is not recorded in said dictionary memory, but does belong to the 35 character arrays entered, the type information providing means providing a number of type information to said character array which does not have type information.

The fifth object of the present invention can be fulfilled by a language analyzer comprising: 40 a glossary memory in which glossary data is stored for all glossary reference units, and parsing means for dividing an entered phrase in words! ice reference units and to perform a morphological analysis by searching the word ice memory using the 5 word ice reference units, the word data in the ice memory having distinctive information specifying the position of a word ice unit stands for a proper name if a number of proper names can occur in a consecutive series of proper names, and searches the parsers in the dictionary memory with the respective words list reference units, present in the entered sentence and, if the distinguishing information is present in the found dictionary data, the words ice reference unit with which the distinguishing information has been found is combined with the words ice reference unit directly adjacent to said words! ice reference unit and which has a meaning other than proper noun in a single parsing unit according to the position specified by the distinguishing information.

The sixth object of the invention can be met by a language analyzer comprising: input means for inputting a character array in a predetermined language, a glossary memory used for looking up said character array input via the input means, and type information parsing means that search the dictionary memory using the input character array and parse the type information of said character array, the type information parsing means parsing the type information of the character array taking into account the type information of the character arrays before and after the said character array. The type information parsing means are for analyzing the type information of a number of rows of characters by collectively arranging them in character arrays.

The seventh object of the present invention is met by a language analyzer comprising: a glossary memory in which glossary data is stored for all glossary reference units, and parsing means for dividing an input phrase into glossary reference units and performing a morphological

870235S

* 8 analysis for said words ice reference units by searching said words ice memory wherein said parsing means distinguish a series of words! Ice reference units that have a specific semantic element, form a composite unit that has a specific meaning associated with a particular rule, and transforms a sequence of words with ice reference units with said specific semantic element into a single parsing unit.

The eighth object of the present invention can be fulfilled by a language analyzer in which the grammatical nature, the semantic nature or the translated word is estimated as being a word not included in the glossary, which is related to the morphology. fairy properties are recognized as a derivative such as a suffix · The ninth object of the present invention can be met by a language analyzer, comprising: a first parser for performing a morphological analysis on an input phrase in a predetermined language, a second parser for performing a sentence analysis in said language based on the result of the morphological analysis in the first parser, words memory storing words data of said language used for the analysis by said first and second parsing means, and control means for searching the said words memory and for performing the parsing procedure in the first and second parsers, said first parser searching in the words memory, distinguishing the structural arrangement by distinguishing the features of the entered sentence in said language in regard to its shape and estimates the attitude which may result from the decomposition and the structural role of said arrangement functioning in said sense, and said second decomposition agent the surface layer structure of the phrase in said sense analyzes language using a grammatical rule based on the estimated attitude and role, and analyzes the possible subordinate relationship of the constituent parts of the said sentence.

Using the language analysis device according to the present invention, an automatic translation can be produced with high 8702359, f, speed and with high quality.

A more detailed explanation of the language analyzer according to the present invention will now be given with reference to nine embodiments illustrated in the figures.

However, the present invention is not limited to these embodiments.

Figures 1 to 10 illustrate a first embodiment of the language analyzer according to the present invention used for automatic translation from English to Japanese.

Figure 1 is a functional block diagram illustrating an exemplary embodiment of a detailed structure for a morphological analysis section.

Figure 2 is a functional block diagram illustrating the entire structure.

Figure 3 shows an explanatory view illustrating an example of the structure of a word list file provided with a highest preferred flag.

Figure 4 is a flow chart illustrating an example of a morphological analysis process.

Figure 5 is a flow chart illustrating an example of the input processing in the morphological analysis process.

Figure 6 is an explanatory view illustrating an example for transforming an input character array.

Figure 7 is an explanatory view illustrating an example of a word list search process.

Figures 8A through 8D show flow charts illustrating an example of the contradiction elimination for the highest preferred flag during the morphological analysis.

Figure 9 is an explanatory view illustrating an example of the contents of a recovered glossary information buffer after the glossary search process.

Figure 10 is an explanatory view illustrating an example of the contents of the recovered glossary information buffer as a result of performing a contradiction elimination for the highest preferred flag.

Figures 11 to 16 illustrate the second embodiment of the present invention.

Figure 11 is a block diagram of the embodiment.

Figure 12 shows in a number of diagrams an example of the data storage in the glossary.

8702358 10 Λ,

Figure 13 shows a diagram illustrating an example of data storage in a basic unit dictionary.

Figure 14 shows a diagram illustrating an example of the data storage in a glossary information retention table.

Figure 15 shows a flow chart illustrating the operation of the device.

Figure 16 shows a flow chart illustrating the unit recognition.

Figures 17 to 29 illustrate the third embodiment 10 of the language analyzer according to the present invention used for automatic translations from English to Japanese.

Figure 17 is a functional block diagram illustrating an example of the detailed structure for performing the morphological analysis.

Figure 18 is a functional block diagram illustrating the entire structure.

Figures 19A and 19B are flow charts illustrating an example of the morphological analysis.

Figure 20 is a flow chart illustrating an example 20 of the collective arrangement of a money value symbol and a unit in the morphological analysis.

Figures 21A and 21B are flow charts illustrating an example of hyphenated numbers in the morphological analysis.

Figures 22A and 22B are flow charts illustrating an example of the processing of consecutive numbers during the morphological analysis.

Figures 23A and 23B are flow charts illustrating an example of the collective aggregation with a previous numerical value during the morphological analysis.

Figure 24 shows an explanatory view illustrating an example of the structure of a glossary file provided with a numeric flag.

Figure 25 shows an explanatory view illustrating an example of an input character array.

Figures 26A through 26D are explanatory views illustrating the contents of the glossary information retention table obtained from the glossary using the input character array illustrated in Figure 25 in subsequent processing steps.

40 Figure 27 is an explanatory view illustrating another 8702350 11 4 -

f 'X

exemplary embodiment of an input character array.

Figure 28 is an explanatory view illustrating the contents of a money value symbol table, position notation table, and decimal point table in the glossary.

Figures 29A through 29D are explanatory views illustrating an example of the glossary information retention table obtained from the glossary using the input character array shown in Figure 27 in various processing steps.

Figures 30 to 36 illustrate the fourth embodiment 10 of the present invention.

Figure 30 is a block diagram of the embodiment.

Figure 31 is a diagram illustrating an example of the data stored in another glossary.

Figure 32 is a flow chart illustrating the operation of the entire device.

Figure 33 is a flow chart illustrating the processing of a proper name registered in the glossary.

Figure 34 is a flow chart illustrating processing of a proper noun not registered in the glossary.

Figure 35 is a flow chart illustrating a processing that yields standard type information.

Fig. 36 is a diagram illustrating an example in which data stored in the glossary information retention table is varied after the input sentence has been processed.

Figures 37 to 46 illustrate the fifth embodiment of the language analyzer according to the present invention used for automatic translations from English to Japanese.

Figure 37 is a functional block diagram illustrating a detailed structure for the morphological analysis section.

Figure 38 is a functional block diagram illustrating the entire structure.

Figure 39 is an explanatory view illustrating an embodiment of the glossary file structure.

Figure 40 is a flow chart illustrating an example 35 of the morphological decomposition of a proper name.

Figure 41 is a flow chart illustrating an example of the collective aggregation of proper names, registered in the glossary, during the morphological analysis.

Figures 42, 43 and 44 are flowcharts illustrating 40 an example of a processing which depends on position information of the proper name during the analysis of a proper name.

Figure 45 is a flow chart illustrating an example of a collective aggregation of proper nouns not registered in the glossary during the morphological analysis.

Figures 46A through 46F are explanatory diagrams illustrating the contents of the glossary information retention table referenced by the glossary for the example of an input character array in various processing steps.

Figures 47 to 52 illustrate the sixth embodiment 10 of the present invention.

Figure 47 is a block diagram of an embodiment.

Figure 48 is a diagram illustrating an example of the data stored in the reference dictionary.

Figure 49 is a flow chart illustrating the operation of the entire device.

Figure 50 is a flow chart illustrating the processing of glossary-registered proper names.

Figure 51 is a flow chart illustrating the processing of proper names not registered in the glossary.

Figure 52 is a diagram illustrating an example in which data stored in the glossary information retention table after input sentence processing is varied.

Figures 53 to 57 illustrate the seventh embodiment of the language analyzer according to the present invention used for automatic translations from English to Japanese.

Figure 53 is a functional block diagram illustrating an embodiment of the detailed structure of the morphological analysis section.

Figure 54 is a functional block diagram illustrating the entire structure 30.

Figures 55A and 55B are flow charts illustrating an example of the morphological analysis.

Fig. 56 is an explanatory view illustrating an example of the contents of the information table in section 108 providing the morph processing information.

Fig. 57 is an explanatory view illustrating an example of the contents of a match table 128.

Figures 58 to 63 illustrate the eighth embodiment of the present invention.

40 Figure 58 is a block diagram for explaining the entire structure.

Fig. 59 is a block diagram for explaining an example of derivative word processing using a prefix.

Figure 60 is a diagram illustrating an example of the derivative word processing by means of a suffix.

Figure 61 is a block diagram of some details obtained by synthesizing Figures 58, 59 and 60.

Figure 62 is a block diagram showing further details for an entire unwritten word processing section in Figure 1061.

Figure 63 is a block diagram for explaining an embodiment of an automatic translation device in which the present invention is applied.

Figures 64 through 90 illustrate the ninth embodiment 15 of the language analyzer according to the present invention used for translating from English to Japanese.

Figure 64 is a functional block diagram illustrating the entire structure.

Figure 65 is a functional block diagram collectively illustrating the recognition function of the structural arrangement of an English input sentence as a block.

Figure 66 is a flow chart illustrating an example of the collective arrangement of a block in the English input sense.

Figure 67 is a flow chart illustrating details of the word lookup process in the processing flow.

Figure 68 is an explanatory view illustrating an example of the glossary of English words or phrases stored in the glossary.

Fig. 69 is an explanatory view illustrating an example of the table data for the block initial state, final state, and target and role estimation states stored in the analyzer rule file.

Figure 70 is an explanatory view illustrating an example of the collective arrangement of a structure.

Figure 71 is an explanatory view illustrating an example of a collective arrangement for a block.

Figure 72 is an explanatory view illustrating English information and word information arranged together in a block.

8702359 f t 14

Figure 73 is a flow chart illustrating an example of the analysis operation performed in an analysis section.

Fig. 74 is an explanatory view similar to that of Fig. 68 illustrating an example of an entry in a 5 word phrase glossary in case this exemplary embodiment is provided with the function of estimating identical cases.

Fig. 75 is an explanatory view similar to that of Fig. 69 illustrating an example for the starting and ending state of a block and a table of the block preparation information in case this embodiment includes a function of estimating identical cases.

Figure 76 is a functional block diagram similar to that of Figure 64 illustrating the entire structure of a modification of this embodiment.

Figure 77 is a functional block diagram similar to that of Figure 76 for aggregating the let information parsing functions in the modified embodiment of Figure 76.

Fig. 78 is an explanatory view illustrating an exemplary embodiment of the glossary information containing let information in English words and phrases stored in the glossary of the modified embodiment.

Figures 79 and 80 are explanatory views similar to that of Figure 72 illustrating an example of the block information and the word information in which English sentences with let information are collectively combined in a block.

Fig. 81 is a flow chart similar to that of Fig. 73 illustrating an example of the series of steps for obtaining a collective compilation of the let's input from the English input.

Figure 82 is a flow chart similar to that of Figure 73 illustrating an example of the analysis processing if let information is present, performed in the analysis section in the modified embodiment.

Figures 83A and 83B are flow charts illustrating an example of the analysis flow of let information in the entered English sense.

Figure 84 is a flow chart illustrating an example of the flow of analysis for torque keywords in the English sense.

Figure 85 is an explanatory view similar to that of Figure 72 for illustrating an example of the block information and the word information 8702359/15 combined together from the English input sentence with hyphenated words.

Figure 86 is a functional block diagram similar to that of Figure 64 illustrating the entire structure of another modified embodiment.

Figure 87 is a functional block diagram similar to that of Figure 65 in which the additive interrogation analysis function in the English input sentence is collectively assembled in the modified embodiment shown in Figure 93.

Figures 88 and 89 are explanatory views, similar to that of Figure 72, illustrating an example of block information and word information, brought together from an English sentence containing an additive interrogation in a block.

Figures 90A and 90B are flow charts illustrating an example of an analysis flow in additive interrogation in the English input sense.

The first embodiment of the present invention will now be described.

Figure 2 illustrates the entire composition of the first embodiment, in which the language analyzer of the present embodiment is used in an automatic translation machine from English to Japanese. Of course, the present invention can effectively be used not only in such an automatic English to Japanese translation device, but can also be used for any other language analyzer in which the sentences of the input language are mainly analyzed when translating. a certain language in another language.

The embodiment illustrated in the drawing has an input section 1010 through which an English text 1012 to be translated into Japanese is input. Input section 1010 includes, for example, a keyboard with character keys such as alphanumeric keys or function keys, an optical character reading unit (OCR unit) for reading English-written text and / or a file memory device for reading an English text stored 35 on a memory medium such as a magnetic disk.

The English text entered from the input section 1010 is read into a pre-editing section 1014 in which translation pretreatment is performed. Mainly sentence recognition and unknown word processing are performed in this section. This functions as a part of a morphological analysis.

8702358 t y 16

The pre-edited English data along with the information obtained during the pre-editing is transferred to a morphological analysis section 1016. Section 1016 divides the sentence by reference to a glossary 1018, analyzes the morphemes of the En-5 gelse sentence, performs various types of classifications such as processing unknown words, proper nouns, time expressions, number expressions and the like and performs whole sentence processing such as searching for fixed expressions and recognizing identical cases. The rules for the morphological analysis are stored in the 10 analysis rule file 1036.

The English data, after the morphological analysis, together with the glossary information obtained during the morphological analysis, is transferred to an analysis section I 1020. The analysis section I 1020 is a functional section that performs an analysis on the surface structure of an English sentence by means of a apply grammatical rule to the English data and detect all structural possibilities.

The English data after the analysis in the analysis section I 1020 together with the decomposition information thereof is fed to the analysis section II 1022. In this section, a solution is selected by performing a syntactic analysis on the result of the first analysis I on the surface layer . In this way, a plausible syntax analysis for the English sentence is prepared to form its structure. These analysis rules are also stored in the analysis rule file 1036.

25 The English data is transferred after the analysis as data for the syntax analysis to a structure transformation section 1024. The structure transformation section 1024 creates a corresponding structure for the Japanese sentence from this syntax analysis, ie an intermediate structure of the English sentence and converts it to a Japanese 30 underlying structure from which Japanese can be easily translated.

The syntax analysis data showing the Japanese underlying structure and thus subjected to a structure transformation is sent to a translation generating section 1026 in which the translated sentence is generated. This is a functional section for generating a Japanese sentence from the Japanese tree structure.

The Japanese zing data translated in this manner, i.e. the data of the translated sentence, is supplied to a post-digest section 1030.

The post-editing section 1030 modifies the translated data of the sentence 40 by referring to the glossary 1018 while applying information 8702359 <* '* 17 used in the translation operation to obtain a more natural Japanese sentence »The Japanese sentence data are transferred to an output section 1032 and then output as the translated Japanese phrase 1034 by the output section 1032. The output section 1032 includes, for example, a printer, a display and / or a file memory device such as a magnetic disk memory.

The sequence of successive translation operations is controlled by a control section 1038 that controls the entire device. The glossary 1018 contains the glossary entries for the words of the English and Japanese languages, and in this embodiment are not only the vocabularies, but also various information such as interrelationship, that is, words that occur together, meanings, multiple and singular forms , phrases, etc., are included. The analysis rule file 1036 contains the data of the rules for morphological analysis and syntactic analysis.

The control section 1038 is coupled to an operation display section 1040. The operation display section 1040 includes operation keys that allow an operator to issue various instructions 20 to the present device, such as translation instruction keys or cursor keys, a display unit or indicator with which the input English text, the Japanese sentence resulting from the translation, intermediate data such as glossary information, various instructions for the operator, etc. can be made visible. Most of the operation display functions 25 may be implemented to be housed in a keyboard if it is present in the input section 1010 or in a display unit that may be present in the output section 1032.

In Figure 1, a detailed structure of the morphological analysis section 1016 is illustrated. The analysis section 1016 has an input device 1100 such as a keyboard for the input section 1010 and an input interface circuit 1104 which interfaces with the input document file 1102. The input interface circuit 1104 includes an input character array buffer into which the data of the English character array is stored. the form of encoded data is input, for example 35 in the form of ASCII from the input unit 1100 or from the input document file 1102 and in which the character array data is temporarily stored. The input character array may be the pre-edited array from the pre-editing section 1014.

The morphological analysis section 1016 includes, as illustrated 40 in the figure, a processing section 1106, a word search section 8702359 T 3 18 1108, a contradiction elimination rule processing section 1110, and a polling section 1112. The processing section 1106 is an analysis function section for performs a morphological analysis and includes a retrieval glossary information buffer, ie, a glossary information retention table 1120 (see Figure 9). The morphological analysis is performed by orderly controlling the glossary recovery from the start of the input character array in accordance with the retrieval key character array and storing the obtained glossary information from the glossary lookup section 1108 accordingly in the glossary information retention buffer 1120 and by processing a preferred degree in accordance with the highest preferred flag as will be described later.

The glossary lookup section 1108 is a functional section for looking up glossary information by searching the glossary 1018 based on the recovery key character array originating from the processing section 1106, after which the found glossary is transferred to the processing section 1106.

The glossary 1018 contains grammatical information such as phrase information and inflections of the word in each entry entry, as well as a highest preferred flag as shown in Figure 3 for an example entry entry. The glossary is referred to as a word list file with the highest preferred flag. "Highest Preferred Flag" is a flag indicating the degree of linkage between words in a compound word or an expression that is part of the glossary data, where a "0" indicates a weak or no link while a "1" represents a In this case, the use as an expression is provided for a compound word or an expression to which a strong link is assigned, while on the other hand the possibility of using the words as individual words is contemplated in parallel.

As shown in Figure 3 as an example, each of the input data in the glossary 1018, respectively. arranged by compound word, expression, and individual words that form part thereof, with no difference being made between the individual words and the 35 compound words or expressions. Furthermore, every inflection is part of the entry data. If there are several forms of inflection, they are resp. defined as different input data. The type of bend is shown in the bend section. The situation is similar for the phrase information, the registration of several 40 phrases and the information contained therein is possible. As a further information, the countability or non-countability of a noun, the transitability or non-transitability of a verb, or a translated word and the like is recorded.

For example, "Get" is the indefinite mood of a verb, and the 5 highest preferred flag is "0". The phrase "get up" is an expression with a verb in some sense and its highest preferred flag is "1". Furthermore, the preposition group "up to" has the highest reference flag "1", but a group of nouns "white house" has the highest preferred flag "0" as a compound word, so the latter shows that the link between the words is low . In the figure, the symbol "indicates an empty character.

In this manner, the glossary information obtained in the glossary lookup section 1108 contains the highest preferred flag. If the highest preferred flag of identical character arrays or overlapping character arrays for both is set to "1", then such contradiction must be eliminated. The contradiction elimination rule processing section 1110 takes care of the elimination of the contradiction and this process is performed with reference to the contradiction elimination rule for the highest preferred flag stored in the analysis rule file 1036.

The contradiction elimination rule is applied in the present embodiment in the order from (1) to (3) with which a preferred selection is made.

(1) Phrase or word whose phrase is equivalent to a working word.

(2) Compound word, phrase, or word with many constituent words.

(3) Compound word, phrase or word situated in the first part of the sentence.

The use for the word selected in this way, ie the analysis unit, displayed as the active information in the information buffer 1120 for the received word list information in the processing section 1016. The active information shows that the analysis unit is valid or effective if it is "1" and shows that its capabilities are not used if it is "0".

The control section 1112 is a functional section for controlling and controlling the operation and processing in each of the functional sections of the morphological analysis section 1016. The section may be included in the control section 1038 for controlling the entire device.

8702359 r ί 20

A result of the morphological analysis is transferred to the analysis section I 1020 via the output coupling circuit 1114. If the result is not directly transferred to the analysis section I 1020, it is stored in the analysis input file 1116 and in the analysis glossary information file 1118.

Since in this embodiment, all words, compound words or sentences, starting with the excision position of the words ice reference unit, are used in the morphological analysis, words ice information is obtained for the individual words constituting a compound word or for a expression disregarded as a composite "unit" in accordance with the highest preference flag. That is, the degree of coupling between the words in the sentence is judged while referring to the highest preferred flag of the words ice information obtained in the morphological analysis. Compound words or expressions assume that those who are judged with a strong link are used as an expression in the sentence, and if not, the possibility of being used as separate words is also assessed in parallel.

Such processing with the highest preferred flag is performed by the flow chart shown in Figure 4. The data for the input character arrays is received from the input section 1010 (1200), with the input character array being cut out as words ice reference unit with which the glossary file 1018 with the highest preferred flag (1201) can be searched, the glossary 1018 searches accordingly, which is executed up to the last position of the sentence represented by the data of the input character array ( 1202), after which contradiction on the highest preferred flag is eliminated (1204) and the result of the morphological analysis is carried out to the analysis section I10 (1205).

In the input processing block 1200, the data is first read from the input document file 1102 or in the input unit 1100 input into the input character array buffer or input torque circuit 1104 (1210, see Figure 5). For example, the data for the input character array is input in the form of ASCII codes, and if the symbol EOF (end of file) is read, the processing section 1106 writes the zero code into the input character array buffer at the last position.

Then, the input character array (1211) is reformed by the processing 8702359 i 21 section 1106. For example, if two or more characters corresponding to a space are found in succession, they are arranged in a single empty character for correction. For example, a character corresponding to a space is an empty character (represented by the symbol U), tabulator, line break (represented by the symbol, etc.). These characters corresponding to spaces between the beginning and the first appearing character, which is not equal to a space with a corresponding character in the input character array buffer is eliminated.

10 The input character array "ü UIliwill uuget u up ^ toUgo ÜtoUa U

whited house ... "is transformed as shown in figure 6 into" I üwill üget f-lup ^ to ^ go ^ toUa ^ white ^ house ^ .... (zero code) ". The position of the zero- code indicates the last position in the buffer.

The words ice reference limiters used for the cropping operation 1201 of the word reference units are placed at the position of alphabetic characters, numeric characters, quotation marks and characters other than hyphens and rest marks, as well as apostrophes following blank characters. The processing section 1106 has an initial pointer which refers to the glossary 20 and is initially set to the beginning of the buffer.

The glossary lookup section 1108 then searches the glossary file 1018 along with the highest preferred flag using the character array starting with the character indicated by the initial pointer up to the character preceding the next limiting value, which characters the recovery key character array to shape. The input data in the glossary and the recovery key character array are then compared and if both are identical, the glossary information is adopted (1203). A match is established if the entire character array of the vocabulary entry data corresponds to at least a portion of the character array from the beginning of it and if the portion is just a word reference delimiter, or an apostrophe, or rest character just after this portion. For example, as shown in Figure 7, if the start pointer points to the first character "g" in the recovery key character array "get U upon * -)", then "get" and "get upon" vote from the glossary entry data corresponding to that.

The recovered glossary information is stored in the recovered glossary information buffer 1120 of the processing section 40 1106. During readout, the correctly assessed start position is also 8702359

T I

22 and the end position of the character array is stored. This in an orderly manner specifies the position of the characters in the input buffer starting at the beginning. An accumulation area for the active information is made available in the recovered word list information buffer 1120, which is information indicating whether or not the recovered word list information is effective for the subsequent operation, and each is set to "1" in this step.

Then, the initial pointer is moved further to the next glossary reference unit and placed at the character immediately after the limiter that appears first after the current position of the initial pointer, scanning the character array from left to right. The search in the glossary is therefore carried out successively. In the above example, as a character at the beginning of a word ice reference unit, first the 15 'T' is taken for the word "1", then "w" for "will" and then "g" for "get". When the initial pointer passes the zero code, it is recognized as the last position (1202).

Figure 9 shows an example of the glossary information obtained in this way from the input English character array described above.

An explanation will be given for the contradiction elimination operation 1204 performed by the contradiction elimination rule processing section 110, referring to the contradiction elimination rule file for the most preferred file 1036 in conjunction with Figures 8A through 8D. The flow chart shown in Figures 8A and 8B illustrates the processing in case the positions for the words in which the highest preferred flags are set overlap, while the flow chart in Figures 8C and 8D illustrates the processing in which an analysis unit, which i.e., elements with the highest preferred flag, by transforming active information into "0" is eliminated. In these flowcharts, "<=" represents a substitution, represents a reference, and "P-» x "represents the content x stored in the input data designated by the respective pointer value p.

First, a group of words each with the highest preferred flag equal to "1" whose positions in the sentence overlap is detected (steps 1220-1223). Then, the highest preferred flag elimination rule is applied to each of the detected groups and the effective groups are selected therefrom (steps 40 1224-1235).

8702359 i'm 23

For example, in the above-described embodiment, the highest preferred flag "1" is set at "get up" at the start position "8" and the end position "13" and at "up to" at the start position "12" and the end position "16" for the character array "get ^ up ^ to" as shown in figure 9 where the positions of the characters are overlapping. Then, the selection rule (1) described above is applied first and it is judged whether it is a verb or not referring to the phrase of the conservation pointer psave and the phrase of the pointer p (1224). Because in this example 10 is similar to a verb, the combination "get up" is selected.

If rule (1) is not satisfied then rule (2) is applied (1228) and the length (lens) of the character array with reference to the input data of the conservation pointer psave and the length 15 and of the character array with reference to compared the input data of the pointer p. If rule (2) is also not satisfied, rule (3) is applied (1229) and the starting position with reference to the starting position of the conservation pointer psave and the starting position with reference to the starting position of the 20 pointer p are compared with each other .

If any of the contradiction elimination rules (1) through (3) are satisfied when applied in this order, then the active information of the input data that was not satisfactory, that is, is not effective, is set to "0" (1232) while in other instances the active information, ie the effective input data, is left at "1" (1231) which was the existing state. Such application of the contradiction elimination rule is performed sequentially while the pointer p progresses (1234, 1235) to the last position for each of the successful input data and the active information is maintained at "1" only for the effective input data. The processing for the above example is illustrated in Figure 10. For the input entry "up to", for example, the active information is set to "0".

Next, the overlapping units, even partially overlapping units, both of which have active information with the highest preferred flag "1" are detected (1236-1241) and their active information is set to "0" (1242, 1249). The application of such a contradiction elimination rule is executed orderly for each of the input data while the pointer p 8702359 τ i 24 advances stepwise (1243, 1248) to the last position and the active information of ineffective input data is set to " 0 ". The active information for “get” and “up” is therefore set to “0” for example for the input data “get up” (figure 10). Om-5 that for all input data “white”, “white house” and “house” all highest preferred flags are "0", even if the positions overlap, their active information is maintained at "1".

When processing is performed in this manner, just before the last position (zero code), the contents of the input buffer of the input torque circuit 1104 and of the recovered words ice information buffer 1120 are output through the output torque circuit 1114 to the analysis section I 1016. The contents of the recovered words ice information buffer 1120 are output only for input data for which a "1" is noted for the active information. For example, the contents of the input-15 buffer can be written into the analysis input file 1116, while the contents of the retrieved words List information buffer 1120 can be written into the analysis dictionary information file 1118. Because in this case, both the active information and the highest preferred flag being issued is the structure of the analysis word list information file 118 identical to that of the recovered words list information buffer. The method can also be performed so that the active information and the highest preferred flag are not issued.

The present invention will now be described more particularly with reference to the second embodiment.

Figure 11 illustrates the second embodiment of the language analyzer of the present invention used in an automatic English to Japanese translation device.

This embodiment has an input section 2014, in which data 30 is input from an input unit 2010 or an input document file 2012. The input unit 2010 comprises, for example, a keyboard with character keys such as alphanumeric keys or function keys, and an optical character reader for reading on paper captured English text. The input document file 2012 is a memory unit 35 with English text recorded on a memory medium such as a magnetic disk.

The input section 2014 includes an input character array buffer 2014a and houses the English input sentence input from the input unit 2010 or from the input documents file 2012 in the input character 40 terarray buffer 2014a. The input section 2014 reads the entry in the input character 8702359 £ Ί.

25 terarray buffer 2014 stored sentence and feed it to the processing section 2016.

The processing section 2016 is a functional section that performs the morphological analysis for the entered sentence that originates from the 5th entry section 2014 by looking up in a glossary file. The processing section 2016 includes a glossary information retention table 2016a and stores the information obtained by looking up a glossary file 2022 or a basic unit glossary file 2026 to be described later in a glossary information retention table 2016a.

The processing section 2016 searches the glossary with a back-win key character array obtained as a unit from the character array that makes up the entry sentence, entered from the entry section 2014. The recovery key character arrays are formed in an orderly manner starting with the first character in the character array that makes up the input sentence in accordance with a predetermined formation rule. For example, the input sentence is divided from the beginning of the sentence in an orderly manner by delimiters, for example, spaces, commas, etc., and the divided character arrays are resp. used as a 20 key recovery array. In this case, character arrays that express units such as m, km, m / s, respectively. also shaped as a recovery key character array. The processing section 2016 sends the recovery key character array obtained from the character array constituting the input sentence to the glossary lookup section 2020.

25 The glossary lookup section 2020 provides the glossary 2022 search based on the recovery key character arrays emitted from the processing section 2016. In the glossary 2022, input data and grammatical information such as phrases are stored in the manner shown in Figure 12. If there is an input 30 is present in the glossary 2022, then the glossary lookup section 2020 reads out the phrase, information, etc. for that entry and passes it to the processing section 2016. If no entry is found in the glossary 2022 as a result of the glossary search 2022, the glossary lookup section 2020 reflects this situation to the processing section 2016.

The processing section 2016 stores phrase information, etc., obtained with the glossary lookup section 2020 in a glossary information retention table 2016a. If there is no entry for the recovery key character array in the glossary 2022, then the processing section 2016 outputs the recovery key character array to a unit recovery section 70242 X8 26 section.

The unit recognition section 2024 searches a fundamental unit glossary 2026 based on the recovery key character array emitted by the processing section 2016. The input data in the fundamental unit is stored in the basic unit glossary 2026 in the manner shown in Figure 13. If the input data is present if in the basic unit glossary 2026, this input is read out by the unit recognition section 2024. If there is no entry 10 in the glossary fundamental unit 2026, then the recovery key character array is divided into a number of character arrays as described later, and Tater 2026 fundamental unit is searched again. Then, if the fundamental unit input data is present in one of the following search processes in the fundamental unit glossary 2026, a number of unit information is obtained from the fundamental unit input data. If no fundamental unit input data is present during any of the multiple search processes, an indication is placed in the glossary, indicating that this data has not been registered.

20 The unit recognition section 2024 transfers the basic unit input data, the composite unit information and information, indicating that the word is not registered in the glossary, to the processing section 2016. The processing section 2016 stores these information, originating from the unit recognition section 2024. the 25 glossary information retention table 2016a. The glossary information retention table 2016a thus retains the entry data for the retrieve key character array ^ and the grammatical information such as the phrase obtained during the search in the glossary 2022 or the basic unit glossary 2026 based on the recovery key character array.

After the data is stored in the words! List information retention table 2016a, the processing section 2016 provides this data together with the entered phrase to the output coupling circuit 2018. The output coupling circuit 2018 provides the input sentence and the data for the morphological analysis, originating from the processing section 2016 , to an output device 2030 such as a printing unit or a display unit, or to a memory file 2032 such as a magnetic disk memory.

Alternatively, it is possible to directly input the input sentence and the morphological analysis data, issued by the processing section 2016, into analysis means (not illustrated) in order to carry out the analysis of the input sentence in these analysis means and 87 0 2 5 o &

t A

27 further prepare a translated sentence based on the analysis.

The control section 2028 is intended to control the operation of each of the functional sections in the present device and can preferably be realized using a microprocessor.

The operation of the present device will be explained with reference to the flow chart shown in Figure 15.

First of all, the English input sentence is read from the input device 2010 or the input document proof 2012 in this input section 2014 (2100). The sentence read in the input section 2014 is stored in the input character array buffer 2014a. The phrase stored in the input character array buffer 2014a is then read and delivered to the processing section 2016.

If the input sentence has been entered in the processing section 2016, the glossary recovery units are cut (2102). That is, the character array that constitutes the input sentence is distributed according to predetermined rules in recovery key character array is as the units searched in the glossary 2022 or the basic glossary 2026 unit starting successively. start of the character array. Thereafter, it is judged whether or not the recovery key character array is present (2104) and if it is present then the recovery key character array is sent to the glossary lookup section 2020.

If the recovery key character array has been sent to the glossary lookup section 2020, then the glossary lookup section 2020 will search the glossary 2022 for the recovery key character array (2106). It is judged whether or not the recovery key character array is present in the input data of the glossary 2022 as shown in Figure 12 and, if the input data is present, the grammatical information such as the phrase stored in the glossary 2022 is read out and the read out data is sent to the processing section 2016 and stored in the words! ist information retention table 2016a (2110). Thereafter, the flow chart returns to step 2102 and a new words ice recovery unit is cut out.

If no entry is present in the glossary 2022, the glossary lookup section 2020 returns the recovery key character array to the processing section 2016, and the processing section 2016 sends the recovery key character array to the unit recognition section 2024, in which the unit recognition is performed (2112).

40 In a case where the recovery key character array, sent 870 2 3 5 9 ï 28 to the words! List lookup section 2020 consists of common words such as nouns and verbs, since entries for most of them are present in the glossary 2022, grammatical information such as phrases, read from the glossary 5 2022 and the data is sent to the processing section 2016 and recorded in the glossary information retention table 2016a. As described above, entries for common words such as verbs and nouns have been generated but no entries have been created for character arrays that express the 10 units in the glossary 2022. Therefore, in the case where the recovery key character array is a character amy which expresses a unit, such as kg or m / s, since there is no input data in the glossary 2022, the flowchart proceeds to step 2112 for the unit recognition.

The unit recognition operation in step 2112 will be explained with reference to Figure 16.

If the recovery key character array for which no entry in the glossary 2022 is present when searching the glossary 2022 is sent from the processing section 2016 to the one-20 recognition section 2024, the pointer P is set to the character at the beginning of the recovery key character array in the unit recognition section 2024 (2200).

Thereafter, the unit recognition section 2024 searches the base unit glossary 2026 on the character array starting with the character with the pointer P set (2201). During this search, it is judged whether the fundamental unit, for which an entry is present in the basic unit glossary 2026, appears as a complete character array in the character array starting with the character with the pointer P set or not and whether it begins with the character character with pointer P set or not. Namely, it is determined whether or not a character array comprising a number of characters, starting with the character to which the pointer P is set, corresponds to any of the fundamental units for which an entry is present in the fundamental unit glossary 2026 For example, in the case where the characters to which the pointer P is set are k, m, s, etc., there are input data in the basic glossary 2026 for these individual characters starting with the character where the pointer is P is set as shown 40 in Figure 13.

8702359

Sr * 29

The unit recognition section 2024 judges whether or not the input data is present in the basic unit glossary 2026 as a result of the search process in the basic unit glossary 2026 (2204) and, if the input data is present, then the pointer P becomes the length of the recognized fundamental unit (2208) is placed further. Therefore, in the case where the fundamental unit is a k, m, s, etc., the pointer P is moved further over a character and then set to the next character in the recovery key character array.

The unit recognition section 2024 judges whether or not a further character terarray, starting with the character with the pointer P set, is present (2208). If such a further character array is present, the flowchart returns to step 2202, again searching the basic glossary unit 2026 with the character array beginning with the character with the pointer P set. The section then judges whether or not the input data is present in the basic glossary unit (2204) as a result of the search process in the basic glossary unit 2026 and, if the input data is present, the pointer P is further placed over the length of the recognized fundamental unit.

If, in step 2208, the character array, starting with the character in which the pointer P is set, is no longer present, the search in the basic unit glossary 2026 is terminated, i.e. the recognition of the composite unit is successful.

For example, in the case where the recovery key character array, sent to the unit recognition section 2024 equals km / s, representing a unit, the entry is not present in the basic unit glossary 2026 because km / s is itself a composite unit. The pointer P is first set to k (2200) and k is found in the glossary basic unit 2026 confirming the presence of the input data (2202).

Thereafter, the pointer P is set to m (2206) and m is found in the basic glossary unit 2026 (2202) confirming the presence of the input data in the same manner. Since the unit recognition section 2024 considers a slash (/), a dot (.), Etc. as part of a unit, the pointer P is then set to s skipping "/" in km / s (2206). Thereafter, s is found in the basic unit glossary 40 2026 confirming the presence of the entry data in the same manner (2202). As Input data is found for both k, m and s when searching the basic unit glossary 2026, as a result, it is judged that km / s is a character array expressing a unit. In this manner, in a case where input data is present in the basic unit glossary 2026 for all characters that are part of the recovery key character array, or in a case where data is present in the basic unit glossary 2026 for all characters except symbols such as oblique dashes, dots, etc., which are considered to be part of the unit, judging that the recovery key character array is a character array representing a unit.

When the unit recognition section 2024 has finished searching the basic unit glossary 2026 and has successfully recognized the composite unit, the unit information thus obtained is transferred to the processing section 2016 in which it is stored in the words! Ice retention table 2016a (2210).

Unit recognition is completed in this way.

If no entry is present in the basic glossary unit 2026 as a result of searching the basic unity glossary 2026 on the character array starting with the character with the pointer P set, then in step 2204 it means that the character array cannot be recognized as a fundamental unit or as a composite unit. The unit recognition section 2024 therefore sends the information, indicating that the character array is a word not registered in the glossary, ie, indicating that the word does not represent a unit, back to the processing section 2016, where it is stored in the words list information retention table. 2016a of the processing section 2016, which completes the unit recognition.

Referring again to Figure 15, the unit recognition (2112) has ended, then the flowchart returns to step 2101 and again a word reference unit cutout is performed by the processing section 2016.

After cutting out the words! Ice reference unit, the processing section 2016 judges whether a cut unit is still present or not (2104) and, if no cut unit, ie no recovery key character array, is present, the information is stored in the words ijstlist information retention table 2016a, issued to the output device by the output torque circuit 2018 (2114). The analysis of the input sentence has thus been completed.

8702359 ΐ f 31

As described above, according to this embodiment, the English input phrase is divided into recovery key character arrays and a plain word list 2022 is first searched, and if there is no input data in the word list 2022 then unit recognition is performed. In the unit recognition, the recovery key character arrays are divided, indicated by the pointer P, and for each of the sub-character arrays, the basic unit of vocabulary 2026 is searched. Those that are defined in the basic unit of vocabulary 2026, or those that are compiled from a series of arrays registered in the basic unit glossary 2026, are then judged as character arrays representing units.

Therefore, since it is possible to perform a unit recognition even for a character array, which expresses a composite unit by combining fundamental units stored in the basic unit glossary 2026, the analysis can be performed in accordance with a variable meaning of the units. In addition, since it is only necessary that the glossary fundamental unit 2026 contains fundamental units, for example, k, m, s, 20 ...., etc., and do not need composite units as combinations thereof, for example km, km / s, etc. the capacity of the glossary can be reduced.

The third embodiment of the present invention will now be explained.

Figure 18 illustrates the entire structure of the third embodiment in which the language analyzer according to the present invention is used for automatic translation from English to Japanese.

This embodiment includes an input section 3010 by means of which an English text 3012 to be translated into Japanese is input. Input section 3010 may include, for example, a keyboard with character keys such as alphanumeric keys or function keys, an optical character reader (OCR) for reading English text recorded on paper and / or a file memory device for reading English text, registered on a memory medium, such as a magnetic disk memory.

The English text entered from the input section 3010 is read into a pre-editing section 3014 in which translation pretreatment is performed. In this case, sentence recognition and unknown word processing is mainly performed. This functions 8702359 * * 32 as part of the morphological analysis.

The English data is transferred after the preregistration along with information obtained during the preregistration, to a morphological analysis section 3016. The section 3016 analyzes the morphemes 5 of the English sentence while each time it is distributed with reference to the glossary 3018, performs different classifications such as operations on unknown words, proper nouns, time expressions, numbers, etc. and performs operations on the entire sentence such as searching for standard expressions and adjectives. The morphological analysis rules are stored in the analysis rule file 3036.

The English data is transferred after the morphological analysis along with the glossary information obtained from the morphological analysis to an analysis section I 3020. The analysis section I 3020 is a functional section that analyzes the surface layer structure of the sentence by releasing a grammatical rule on the Investigate English data and the possibilities in the structure.

The English data that has been subject to the analysis in analysis section I 3020 is transferred together with the analysis information to an analysis section II 3022. In this section, a solution is selected by applying a syntactic analysis based on the result of the analysis of the surface layer structure by the structure decomposition I, thereby producing a plausible syntax analysis for the English sense to form its structure. These analysis lines are also stored in analysis line 25 file 3036.

The English data pertaining to the analysis, syntactic analysis data is transferred to a structure transformation section 3024. The structure transformation section 3024 prepares a syntactic analysis of a corresponding Japanese sentence from the structure which is an intermediate structure of the English sentence in order to provide the Japanese sentence to the underlying Japanese structure from which a Japanese sentence can be easily translated.

The syntax data thus transformed indicating the Japanese underlying structure is transferred to a translation forming section 3026 in which the translated sentence is generated. This is a functional section for creating a Japanese sentence from the syntax data of the Japanese sentence.

The Japanese sing data, edited as a translated sentence, i.e., the translated data, is transferred to a post-digest section 303030. The post-digest section 3030 modifies the translated data by searching 8702359 1 k 33 using a glossary 3018 using information that was also used during the translation process to complete a more natural Japanese sentence. The Japanese sentence data is transferred to an output section 3032 and hence output as the translated Japanese sentence 3034. The output section 3032 includes, for example, a printer, a display unit and / or a memory unit such as a magnetic disk memory.

The flow chart of a series of translation operations is controlled by a control section 3038 which takes care of the control of the entire device.

The glossary 3018 contains glossary data for English and Japanese words in this embodiment, which includes various information such as interrelation, i.e., interrelationships, meanings, multiple or singular form, phrase, etc., in addition to the vocabulary. Furthermore, the analysis rule file 3036 contains line data for the morphological analysis and for the syntactic analysis.

The control section 3038 is coupled to an operation display section 3040. The operation display section 3040 includes, for example, operating keys such as translation indicator keys or cursor keys with which an operator can give various instructions to the present device and an indicator that visually observes the entered English text, as well as the Japanese text resulting from the translation and intermediate data such as the glossary information, and which further provides various indications to the operator. Many of these operation indication functions can be performed so that they can be included in a keyboard contained in the input section 3010 or on a display unit located in the output section 3032.

Referring to Figure 17, an example of a detailed structure for processing numbers in the morphological analysis section 3016 is shown. Section 3016, of course, includes other functional analysis sections, but only those sections that are directly relevant to an understanding of the present invention are illustrated here. The morphological analysis is performed by instructed to search a glossary starting from the beginning of the input character array successively in accordance with recovery key character arrays and by processing the glossary information obtained from the words list lookup section 3104 in accordance with a numerical flag which is further will be described.

The morphological analysis section 3016 includes an input processing 40 section 3100 for receiving and processing data from the input character array of the preprocessing section 3014. The input processing section 3100 includes an input character array buffer that receives the English character array data at the input in the input processing array. form of encoded data, for example in ASCII code and which stores the character array-5 data at a time.

The input character array data, temporarily stored in the input processing section 3100, is sent to a unit excision section 3102 that divides the input character array data into word list recovery units, such as words. The unit excision section 10 3102 is a functional section that distinguishes between the word reference units that together form the recovery key character array for searching the glossary 3018 in the glossary recovery section 3104. The glossary referring term used for cutting out the words ijstreferen-15 unit, is placed at the position of an English character, a numeric character, an apostrophe, a character other than a hyphen and a rest sign, as well as at an aprostrop that follows an empty character. This is stored in a limiter table 3108 and used as a reference when cutting out the word reference reference unit in the unit cutout section 3102.

The glossary 3018 specifically contains information for obtaining the cutting units. As shown with the example of the input information in Figure 24, for each entry data, there is stored in the glossary data, ie words, grammatical information such as phrase, as well as a distinctive indication that the word represents a digit, ie a numeric flag and numeric value information related to the numerical value for which the word stands.

As illustrated in the figure, both single and multiple forms are described side by side for each of the input data of the glossary 3018 and each forms its own input data.

The numeric flag indicates that the word represents a number if the flag has a "T" set. As further information, for example, countability and non-countability for a verb, identification for transitive or non-transitive verbs, translated words, etc. are recorded. For example, referring to "thousand", because this is a standalone word representing a number, the numeric flag above is set to "1" and the numeric value is indicated as "1000". Furthermore, with reference to 8702359 ï Jü 35 "thread", this is a noun but not a noun indicating a number, the numeric flag is set to "0".

The recognition of a number is performed by its 5 numeric flag in case the word is registered in the glossary 3018, for example, as "one" or "thousand". Also unregistered words, for example, a sequence of numeric characters such as "123", two groups of consecutive numbers with a "rest" character in between such as a small number, such as "10.2" and a 10 sequence of numeric characters with a comma in between, for example "1,000,000 "are also recognized as numbers. In the present description, the term "numeric character" is generally used to include not only Arabic numerals but also the spelled out expression such as "thirteen".

As shown in Figure 28, the glossary 3018 includes a monetary unit table 3018a in which various monetary symbols are registered, a notation symbol table 3018b in which notation symbols "(space)" are registered, and a decimal point table 3018c in which the decimal point ". " etc. are registered. The ta-20 bells for the notation symbols and decimal points are provided because it is used as the notation symbol or used for the decimal point in Japanese or English, the space being used mainly or as the notation symbol and used for the decimal point in other European languages such as French and German and 25 should therefore take into account the use of the symbols in the different languages.

The glossary lookup section 3104 is a functional section that divides the glossary information into pieces by searching the glossary 3018 based on the recovery character array input from the unit excision section 3102 and transmitting it to the processing sections 3110, 3112, and 3116.

The sequential arrangement of numerical characters is performed by the following two treatments. First, if words are recognized as a digit, as described above, and in the search, the next word reference unit is also recognized as a digit, then they are arranged together to be synthesized into a number. This operation is repeated as long as consecutive digits occur. For example, "30 thousand" is converted to "30000" and "1.5 million" into "1500000". If after 40 the numeric expression continues with insertion of "and" then 6702359 36 all digits to the left of "and" corresponding to digits of numerical values indicated by the pointer to the right of "and" will be equal to "0" taking into account the meaning of the numeric expression and are synthesized into a number. For example, "one 5 hundred and thirty" is synthesized in "130", while "30 thousand and two hundred" is synthesized in "30200".

After such recognition of a figure, the further required local analysis is performed. In this process, a series of analysis units is actuated by the analysis actuation information for each of the 10 analysis units, collectively arranged in a single analysis unit based on a local analysis rule. A money value symbol and a numeric value, for example, "¥ 1,000" are arranged together as "1000 yen", and a numeric value with a unit of "1.5 km" is aggregated together to 1.5 kilometers ".

These rankings are performed in processing sections 3110-3122. The processing section 3110 is a functional section for collectively arranging a number together with a money value symbol or a unit. The processing section 3112 is a functional section in which the numbering of the number is performed. Furthermore, the processing section 3114 is a functional section for processing numbers connected by a hyphen. Furthermore, the processing section 3116 is a functional section for processing subsequent numeric characters.

With reference to the number after the rank with a money-25 value symbol or a unit, the combination of the money-value symbol and the numeric value into a single noun is performed in the processing section 3118. Furthermore, the combination of the unit and the numeric value into a single noun executed in the processing section 3120. In the case of a number subjected to a numbering operation, a hyphenated number and a continuous continuous number, further processing for combining with the previous numeric value is performed in the processing section 3122. The words list information for the input character array is stored after completion of such an operation in the sorted information word list buffer, that is, the words list information retention table 3124.

The results of the morphological analysis are transferred from the glossary information retention table 3124 to the analysis section I 3020.

The processing by the numerical flag is performed in the following order as shown in Figures 19A and 19B. The data for the input character array is received in the input processing section 3100 where the input processing is performed (3200). Thereafter, the unit excision section 3102 intersects the input character array into the words ice reference units to be searched in the glossary 3018 (3201). The word ice recovery section 3104 sorts the word list 3018 accordingly (3203) and if a word ice entry entry is present (3204) then the numeric flag is examined (3205). If the numeric flag is not set because the word does not indicate a 10 number, then the words! List information is accumulated in the glossary information retention table 3124. If a "1" is set for the numeric flag, the number is numbered in the processing! ngs section 3112 (3206) and the composition is formed (3207) with the preceding numerical value in the processing section 3122. If these operations are performed on the last part of a sentence, indicated by the input character array data (3202), then the composite arrangement is (3209) with the money value symbol or unit output in processing sections 3118, 3120, and the result of the morphological analysis is fed to the analysis section I 3020 (3210).

If no entry is found in the glossary in step 3204 and if the element is hyphenated (3212), then the processing for a hyphenated number (3213) is performed in the processing section 3114. If it is first character is not a hyphen character but a money value symbol (3214) then only the money value symbol is included in the words! list information retention table 3124 (3216) and the money value symbol is omitted (3217) from the words! list reference unit. If the first character is not a money value symbol (3214) then 30, the processing for the consecutive numeric characters 3215 is performed in the processing section 3116. This operation is performed up to the last position (3202).

The money value symbol assembly 3209 and the unit is outputted in the processing section 3110 by the processing flow chart shown in Fig. 20. First, in the initial processing 3220, the processing start pointer is set at the beginning of the buffer. If the element indicated by the pointer is not a numerical value (3221), the pointer is moved forward in steps (3226). If the character has a numerical value 40, but it has no preceding money value symbol and no next unit, the pointer is also moved forward step by step (3222, 3224). Processing is continued up to the last position of the glossary reference unit (3227).

If the character is a numeric value (3222), then the 5 money value symbol and the numeric value are arranged together in a single noun (3223). The money value symbol and numeric character, for example "¥ 1,000" are combined in a single noun. Furthermore, if the preceding character is not a money value symbol and the successive character is a unit, then the numerical value and the unit are arranged together in a single noun (3225). A numeric character and a unit such as "1.5 km", for example, are combined into a single noun. Processing is performed up to the last position of the glossary recovery unit (3227).

The processing for hyphenated numbers is performed in the processing section 3114 by the processing flow chart shown in Figures 21A and 21B. First, the hyphenated word reference unit is stored in the buffer during the initial preprocessing 3230. Furthermore, the numeric value "0" is held and the hyphen in the original word reference unit is changed to a space.

Then the words list reference unit is cut out (3231) to search the glossary (3235). If no entry is found as a result of the glossary search process, ie the word is not registered in the glossary (3236), the entire glossary reference unit is held as a word not registered in the glossary in the glossary information retention table 3124 ( 3237).

If an entry is obtained as a result of the glossary search (3236), it is examined whether its numeric flag is a "1" or not. If the numeric flag is a "1", it means that the character is not a numeric character and the entire glossary reference unit is retained as a word not registered in the glossary in the glossary information retention table 3124 (3237).

If a "1" is set in the numeric flag for the glossary! " Then, the number is converted into a numeric expression in the processing section 3012 based on the input data 40 (3239). Then, the numeric converted 8702359 (ii 39) into a numerical value is added to the currently stored numeric value (3240) and the result of the concatenation is held (3241). For example, "two" in "twenty-two" is added to the "twenty" in the foregoing "3020" to "3022" The processing is carried out up to the last position of the words list reference unit (3232).

When the last position is reached in this stepwise process, the flowchart proceeds to the processing 3233 in step 3232, and the retained numerical value is substituted as a numeric value for the entire hyphenated reference unit. Thereafter, the numerical value compound 3207 with the preceding numerical value is output.

An explanation will now be given of the processing of consecutive numerical characters 3215 which is performed in processing section 3116 with reference to Figures 22A and 22B. In these flow charts, the symbol "=" represents a substitution. First, the initialization 3250 is performed in which the retained numeric value val-save is set to "0", the parameter "i" to "1", and the pointer is set to the beginning of the character array of the 20 word reference unit .

Then it checks whether the character * p, indicated by the pointer p, is a numeric character (3251), a notation character (3252) or a decimal point (3253) and if none of these, the entire character array is stored as a word not registered in the word list in the word list information retention table 3124 (3255). If it is a decimal point (3253) then the parameter "i" is multiplied by 10 (3254) and step 3258 is performed. In step 3258, the numeric value num (* p) for the character * p is added to the retained numeric value val-save to obtain a new numeric value to be retained. The numeric value num (* p) is a value that refers to the character (* p) as a numeric value.

After step 3251 or 3252, if the character is a numeric character or a notation character, step 3257 is performed. In step 3257 35, the retained numeric value val-save is multiplied by 10, adding the numeric value num (* p) for the character * p to obtain a new numeric value to be retained.

After these operations, the pointer is stepped forward (3259) and the operation is repeated up to the last position 40 of the glossary reference unit (3260). If this is the last position of the character array, then the numeric value for the entire character array is converted to the retained numeric value (3261), and the composite arrangement 3207 with the previous numeric value is obtained in the processing section 3122. By processing 5, successive numeric characters such as, for example, "1,000.5" are analyzed as a numerical value "1000.5".

The composite arrangement 3207 with the preceding value is obtained in the processing section 3122 as follows. First, the pointer of the words ice table is set to a previous position of the words ice reference unit (3270). If there is nothing in this position, it means that the first position in the retention table indicates the numeric value, and then the numeric value for the running words! List reference unit is registered in the glossary retention table 3124 (3284). The recording position is the position that follows the position currently designated by the pointer P.

If a word is present in the preceding position in step 3271, and if the entry indicated by the pointer p is not "and" (3272) and the pointer p does not indicate a numerical value (3273), then the numerical value for the instantaneous words list reference unit registered next to the position currently designated by the pointer p in the glossary retention table 3124 (3284). For example, in the example of "To him two ....", "two" is now registered as a numeric value "2".

If, in step 3273, the pointer points to a numeric value, then the numeric value p-5> v for the input data, designated by the pointer p, is multiplied by the numeric value v- for the current words! List reference unit to obtain a new numeric value p ^ v for the input data indicated by the pointer p (3274). For example, in the case of "two thousand", "2x1000 = 2000" is output to arrange the whole "two thousand" in a single data, then the end position for the instant words list reference unit is set to the end position for the pointer input data p, i.e. the p-end position 35 (3282).

If, in step 3272, the input indicated by the pointer p is equal to "and", the pointer p is transferred to the preceding glossary reference unit (3275). If this is not the last position (3276) and if it is a numerical value 40 (3277), then the numerical value v-nu of the current words- 8702358 41 c 4t list reference unit is transferred and rounded to the most significant digit, which is set to a value vl. For example, if the numerical value v-nu of the current glossary reference unit is "8", "8.1", "98" or "11", then the value of v1 resp. equal to "10", "10", "100" or "100".

Then, it is checked whether the remainder obtained by dividing the numerical value p-v for the input data, indicated by the pointer p, by v1, i.e. mod (p-v, v1) is or is not "0". If it is not equal to "0", the pointer -10 pointer is incremented (3283), and the numerical value for the current glossary reference unit is registered at a position next to the position indicated by the current pointer p in the glossary retention table 3124 (3284). For example, in the case of "I and two", "two" is now registered as a numeric value "2".

If the remainder is "0" in step 3279, then the numeric value v-nu for the current glossary reference unit is added to the numeric value p-^ v for the input data, designated by the pointer p, to create a new numeric value p - ^ v for the input data indicated by the pointer p (3280). In the case of "two thousand and two", for example, at this stage "two thousand" was already ranked in "2000". Then, for "two", the value "2" is added by adding 3200 to finally obtain the whole portion in "2002". Then, the information "and", designated by the pointer p + 1, is removed from the information retention table 3124 (3281) and the flowchart proceeds to step 3282.

An explanation will be given on the basis of an example. For example, if the glossary is performed for the input character array "To him two thousand and twenty-two—" as shown in Figure 25, then the glossary entry data-30 information is written into the glossary information retention table 3124 as shown in Figure 26A. For example, for "him", the starting position is "4", the ending position is "6", and the phrase is a pronoun. In numerical processing, it is first judged for "two" that the numeric flag is equal to "1" (3205) and the numerical value therefor is equal to "2". Since the character preceding "two" in this character array is not a numeric value, it is stored directly in table 3124 (3206, 3208, 3284).

Then the pointer is raised and continues to process "thousand". The numeric flag is "1" and the numeric value is 40 "1000" (3205, 3206). In addition, because the numeric value of the preceding 8702359 s * 42 character is "2" (3207, 3273), the multiplication: 2x1000 is performed (3274), and the result is stored in table 3124 (see figure 26B). For the next "and", the glossary information is temporarily accumulated as it is in Table 3124 (see Figure 26C).

The pointer is moved further to process "twenty-two". Because these hyphenated words are not found in the words ice entry data (3212), "20x2 = 22" is output by the appropriate processing unit 3213 (3237, 3239-3241). Since the preceding word is "and" (3272) and the preceding numeric value is "2000" (3277), the numeric value "22" is rounded to the most significant digit in "100" (3278) and a partial operation (3279) is performed. Since the remainder is "0", the addition 3280 between "2000" and "22" is performed. The information for "and" is eliminated (3282) from the retention table 3124 and the result of the addition "2022" is held as a numeric value in the table 3124, recognizing "two thousand and twenty-two" as "2022". The composite arrangement of the previous numerical value has been done in 3207 in this manner.

Another example will be shown. As shown in Figure 27, the decomposition is performed for the input character array "You said was $ 1,000.5 thousand -". ". $ 1,000.5" is not registered in the glossary 3018. The first character is a money value symbol 25 which can be recognized as a value value symbol from the words! ice input data. This is independently recorded in the retention table 3124 (3214, 3216, Figure 29A).

Thereafter, "1,000.5" is converted to a numeric value "1000.5" by sequential numeric character processing 3215. Since the preceding character is the symbol and no numeric value, the numeric value is recorded as it is (3280-3273, 3284, Figure 29B).

The next word "thousand" is a number and its numerical value is "1000". Since the preceding character is a numeric value 35 (3272, 3273), a calculation is performed: "1000.5x1000 = 1000500" (3274) (Figure 29C).

After the words ice recovery has ended in this manner, the content retained in the words ice information retention table 3174 is sequentially examined. Because the money value symbol is present 40 immediately prior to the numerical value "1000500" both 8702353 43

-V

t arranged together and "41000500" is formed as a single independent input data (3209, 3221-3223: Figure 29D).

An explanation will now be given of the fourth embodiment of the present invention.

Figure 30 shows the fourth embodiment of the language analysis device according to the present invention used as automatic English-Japanese translation.

This embodiment includes an input processing section 4014, and in this input processing section 4014, data is input from an input device 4012. The input device 4012 includes, for example, a keyboard with character keys such as alphanumeric keys and function keys, an optical character registration device for reading an English recorded on the paper text or a file format for a magnetic disk.

The input processing section 4014 has an input character array buffer 4014a and houses the input English phrase input from the input unit 4012 into the input character array buffer 4014a. The input processing section 4014 reads out the input phrase stored in the input character array buffer 4014a and supplies it to a unit-cutout section 4016. The unit-cutout section 4016 is a functional section which transmits the input reference word units 4014, from the input processing sections 4014a. isolates. A delimiter 4018 contains the delimiters such as spaces and commas.

The unit excision section 4016 reads the limiter from the limiter table 4018 and divides the input sentence sent from the input processing section 4014 into character arrays as units for searching a reference glossary 4020 by dividing the sentence where limiter is present. The sub-character arrays are entered in the words ice lookup section 4022.

The words list lookup section 4022 performs the search in the reference word list 4020 for the input sentence divided into the word list reference units transmitted by the unit excision section 4016. The reference word list 4020 stores input data for its English character arrays, nature of the information, etc., as shown, for example, in Figure 31. The reference dictionary 4020 also contains, in addition to the proper nouns as shown in the figure, character arrays for other phrases, for example for verbs and adjectives. The registration of proper nouns as a phrase in the figure is used in the yet to be described processing of proper nouns, but not the meaning as with conventional 8702358 44 y * 4 grammatical proper names. The type information further points to those represented by a proper name which cannot always be limited to a single one because, as will be described later, a proper name can represent a variety of types depending on the application case.

In the glossary lookup section 4022, the reference dictionary 4020 on the character array is searched, divided into the glossary reference units, and if the character array is a proper name, the array is output to the proper name processing section 4024 in which the proper name processing is performed. is carried out in the manner to be described later. If it is not a proper name, it is issued to the processing section 4036 and held in the glossary information retention table 4036a of the processing section 4036.

The proper name processing section 4024 includes a preceding sentence signal processing section 4026, a preceding proper name processing section 4028, a section 4030 for processing the proper name in itself, a processing section 4032 for the previous proper name and the proper name in itself, and a section 4034 containing that information. about the standard type.

20 The preceding sentence processing section 4026 judges whether the character array, prior to the character array entered and searched by the glossary lookup section 4022, is the end of a sentence or not and, if the preceding character array is the end of a sentence, the capital letter at the beginning of the character array to be processed is converted into a lowercase letter, and forwarded to the glossary lookup section 4022 whereby the glossary lookup sections 4022 again addresses the reference glossary 4022. The character array, which is not retrieved even on a second attempt, is judged as an unregistered proper name and is sent to processing section 4036 and stored in the glossary! information retention table 4036a. Further, if the preceding character array is not at the end of a sentence, it is supplied to the processing section 4036 as a proper name whose type information is unknown and recorded in the glossary information retention table 35 4036a as will be described later.

The preceding proper name processing section 4028 parses the type information of the previous character array sent by the section 4026 for processing the end of the preceding sentence and outputs the result to the section 4030 for processing the 40 proper name by itself. The section 4030 for processing the proper name 8702359 * J-45 by itself examines the type information of the proper name to be parsed and, as will be described later, if the type information of either the proper name or the previous proper name is not is registered then the proper name and the previous proper name 5 are together analyzed by the registered information for their other and held in the glossary information retention table 4036a in the processing section 4036.

The processing section 4032 for the previous proper name and the proper name in itself examines the common part with the type information of the proper name and the previous proper name to be analyzed, analyzes this proper name with the common part and outputs the result to the processing section 4036 and stores this in the glossary information retention table 4036a in the processing section 4036.

The section 4036 providing the standard type information supplies type information to a proper name after reading from the glossary information retention table 4036a, sent via the unit excision section 4016 to the glossary lookup section 4022 if no type information appears to be present as a result of the search process. in the reference glossary 4020 through the glossary recovery section 4022. Because it is often the case that a proper name is of different type depending on the situation of use, all type information deemed necessary is provided. For example, "person, place, group and others" are provided. After the proper name is provided with the type information 25, the section 4034 for providing the standard type information sends the data to the processing section 4036 and stores it in the glossary information retention table 4036a in the processing section 4036.

The processing section 4036a provided with the glossary information retention table 4036a stores the data sent from the processing sections 4032 for the previous proper name and the proper name by itself, with the standard type information supplying section 4034 or the glossary lookup section 4022, in the glossary information retention table 4036a, and then reads out the stored data and outputs it to the analysis section 4038. The analysis section 4038 performs the analysis for the entered phrase after it has been subjected to the morphological analysis and reads from the glossary information retention table 4036a.

The operation of this device will be explained with reference to the flow chart of Figure 32.

First of all, an English input sentence of the input device 8702359 jf% 46 4012 is read into the input processing section 4014 (4100). The phrase read into the input processing section 4014 is loaded into the input character array buffer 4014a. The input sentence loaded in the input character array buffer 4014a is read out to the unit cut-out section 4016.

Once the input sentence has been entered, the unit excision section 4016 reads the limiter from the boundary table 4018 to cut out the glossary reference units (4102). That is, the character arrays constituting the input input sentence are sequentially divided beginning at the beginning of the character arrays in recovery key character arrays, as units to be searched in the reference dictionary 4020, by division on those parts where the delimiters such as a space and a colon is present. The section assesses whether the last classified words! List reference unit, ie the last recovery key character array, has already been found (4104) and if there are more recovery key character arrays (not yet the last one), the recovery key character array is sent to the glossary lookup section 4022.

When a recovery key character array is sent to the word list search section 4022, the word list search section 4022 searches the reference glossary 4020 to recover the recover key character array (4106). The section judges whether or not the recovery key character array is present in the input data of the reference word list 4020 as shown in Figure 31 (4108) and, if an input data is given, the phrase stored in the reference word list 4020 is read and is evaluated whether or not the recovery key character array is a proper name (4110).

If the recovery key character array is not a proper name, then the glossary lookup section 4022 sends the readout data from the reference glossary 4020 to the processing section 4036 and registers it in the glossary information retention table 4036a (4112). If the data is stored in the glossary information retention table 4036a, the input data indicating that the data is stored in the unit excision section 4016 and the data for the just-previously recovered key character array is input from the processing section 4036. Then, the flowchart returns to step 4102 and another word reference unit is cut out in the unit cutout section 4016.

If, in step 4110, the recovery key character array is a proper name, 40 then the glossary lookup section 4022 transmits the data from the proper name read from the 8702359 47 4 Λ - Τ reference glossary 4020 (simply referred to as a proper name in the following) to the proper name processing section 4024 together with the proper name processing section 4024. data from the previous recovery key character array input from the glossary information retention table 4036a 5 into the processing section 4036 through the unit excision section 4016 to the glossary lookup section 4022, and the processing of the glossary proper name is performed in the proper name processing section 4024 (4124).

The processing of the word name registered in the glossary 10 is explained with reference to the flow chart of Figure 33.

The data sent from the glossary lookup section 4022 to the proper name processing section 4024 is passed through the preceding sentence processing section 4026 to the preceding proper name processing section 4028. During processing of the proper name registered in the glossary, section 4026 does not function to process the end of the preceding sentence.

The preceding proper name processing section 4028 judges whether the entry, prior to the proper name, is a proper name not registered in the reference dictionary 4020 or not, that is, whether it has been involved in the processing of the previous to be described later. proper name not registered in the glossary or not (4100). In the case of an unregistered proper name, the processing section judges the entire part of the proper name and of the previous unregistered proper name as a proper name with type information for a proper name (4002), sends the data to processing section 4036 and stores it in the glossary! information retention table 4036a (4218).

If the section 4028 for processing the previous proper name in step 4200 judges that the input data, prior to the proper name, is not registered as a proper name, then it is judged whether the input data, prior to the proper name, is a registered proper name in the reference dictionary 402 is or is not (4204). If the entry data is a registered proper name immediately prior to the proper name, then it is judged whether the type information for the previous proper name is unknown or not, ie whether it is not registered in the reference dictionary 4020 or not (4206).

In case the type information of the preceding proper name is unknown, the flowchart proceeds to step 4202 and the entire portion of the proper name in itself and the immediately preceding 8702359 48 S * X $ proper name is judged as a proper name with type information (4202), and the processing section 4028 for the previous proper name sends the data to the processing section 4036. The data sent to the processing section 4036 is recorded in the words-5 list information retention table 4036 (4218).

If the preceding proper name processing section 4028 judges in step 4206 that the type information for the preceding proper name is not unknown, that is, it is registered in the reference words -1 list 4020, then the data from the section 4028 is processed for the processing. of the previous proper name sent to section 4030 for processing the proper name by itself. The proper name processing section 4030 alone decides whether the proper name type information is unknown or not (4208). If the type information for the proper name is unknown, then the section 4030 for processing the proper name per se judges the entire part of the proper name in itself and the immediately preceding proper name as a proper name with type information of a proper name (4210 ) and sends the data to the processing section 4036. The data sent to the processing section 4036 is recorded in the words list information retention table 4036a.

If the proper name processing section 4030 judges per se that the proper name type information is not unknown per se, i.e. it is registered in the reference dictionary 4020, then the proper name processing section 4030 sends the 25 data to the processing section 4032 for the previous proper name and the proper name in itself. The processing section 4032 for the previous proper name and for the proper name in itself judges whether there is a common type in the type information between the own name in itself and the immediately preceding proper name (4212) and if there is a common type then the entire portion of the own name in itself and of the immediately preceding proper name is judged as a proper name with the common type information (4214) and the data is transferred to the processing section 4036. The data sent to the processing section 4036 is recorded in the words! list information retention table 4036a (4218).

If there is no common type in the type information of the proper noun in itself and the immediately preceding proper name, the processing section judges that the proper name is a proper name with the type information recovered from the reference dictionary 40 4020 and is different from the immediately preceding proper name 0/0 2 359 ί * · 49 (4216) and sends the data to the processing section 4036. The data sent to the processing section 4036 is recorded in the words list information retention table 4036a (4218).

Reference is again made to Figure 32. If in step 4108 no te-recovery key character array is present in the input data of the reference words, list 4020 then it is judged whether the first character of the recovery key array is capitalized or not (4116) and is not capitalized, then the words op ice lookup section 4022 judges the recovery key character array as an unregistered word and sends it to the processing section 4036 for registration in the words! list information retention table 4036a (4118).

If the first character is uppercase, the data for the recovery key character array is sent along with the data for the previous recovery key character array from the glossary lookup section 4022 to the proper name processing section, where processing of the proper name not registered in the glossary is performed (4120 ).

With reference to Figure 34, it will be explained how processing of a proper noun not registered in the glossary takes place.

The recovery key character array data together with the preceding recovery key character array data is sent to the sentence end processing section 4026 and the sentence end processing section 4026 judges whether the end of the preceding input data is is a candidate before the end of the sentence or not (4300). The assessment of whether this is a candidate for the end of the sentence or not is performed by judging whether the end of the preceding entry is a candidate for the end of the sentence such as a separate period (.), Etc. or not.

If the end of the preceding entry is a candidate for the end of the sentence, data from the processing section 4026 for processing the end of the preceding sentence is sent to the section 4028 for processing the previous proper name and the section 4028 for processing the preceding proper name, the preceding input data judges as the end of the sentence (4302), converts the first character in the recovery key character array to a lowercase letter and sends it to the dictionary lookup section 4022.

40 The glossary lookup section 4022 takes care of the search in refe 8702359

> V

50 interest glossary 4020 on the recovery key character array, converted to lowercase (4304) and judges whether or not there is an entry in the reference glossary 4020 (4306). If an entry is present, the glossary lookup section 5 4022 sends the data obtained from the reference glossary 4020 to the processing section 4036 and adds it to the glossary information retention table 4036a (4308). If no entry is present, then the glossary lookup section 4022 capitalizes the first character in the received character array and sends it as an unregistered proper name to the processing section 4036 for registration in the glossary information retention table 4036a (4310).

If, in step 4300, the end of the preceding sentence processing section 4026 judges that the end of the preceding input is not a candidate for the end of the sentence, then the data from the end of processing section 4026 is not the preceding sentence is supplied to the section 4028 for processing the preceding proper name, and the section 4028 judges the preceding input as not being the end of the sentence (4312).

The data is passed from the section 4028 for processing the previous proper name to the section 4030 for processing the proper name per se, and the section 4030 judges the recovery key character array as a proper name whose type information is unknown (4314) .

The proper name processing section 4030 in itself sends the data back to the previous proper name processing section 4028 and the section 4028 performs the processing on the glossary registered proper name (4316). The processing of the proper name registered in the glossary is the same as that shown in Figure 33.

If, again with reference to Figure 32, the cut-out glossary reference unit is at the end of the array in step 4104, then the glossary lookup section 4022 sends an appropriate indication signal to section 4034 which provides the standard type information and the section 4034 reads out the information registered in the glossary information retention table 35 4036a in the processing section 4036 and provides the pronoun with the standard type information (4122).

With reference to Figure 35, an explanation will now be given of providing a proper name with standard type information.

40 In the section 4034, which provides the standard type information, 8702359 51 λ i i i, a pointer is first set at the beginning of the data in the glossary information retention table 4036a (4400). That is, the pointer is set to the input data at the beginning of the input sentence, which is divided into input data, respectively. are provided with information by searching in the reference dictionary 4020. Subsequently, it is judged whether the input data designated by the pointer is a proper name or not (4402) and if it is a proper name then it is judged whether the type information of the proper name is known is or is not (4404). If it is not a custom name, the flow chart proceeds to step 4408 and the pointer is advanced to the next input data.

If in step 4404 the type information for the proper name is unknown, the standard type information is provided (4406). When providing the standard type information, it is appended to a proper name whose type information is unknown as shown in the lower part of Figure 36. In the lower part of the figure, for example, the proper name is "Johnson". type information is unknown, provided with all types of type information, ie "person, place, group, and others". By providing all the type information to the proper name whose type information is unknown, it is possible to provide space so that analysis of the proper name in any number of types is possible in the following syntactic analysis.

If in step 4404 the proper name type information is not unknown, then the flowchart proceeds to step 4408 and the pointer is advanced to a next input.

It is judged whether the pointer input is end or not (4408) and if it is not end, the flowchart returns to step 4402 and judged whether the next entry is proper or not. When the input data designated by the pointer is at the end, the provision of the standard information is terminated.

After termination of providing the standard type information to the proper name, the data recorded in the words list information retention table 4036a is output from the processing section 4036 to the syntax analysis section 4036 (4124), thereby completing the morphological analysis in this embodiment.

With reference to an example of an entered sentence, the operation of the device discussed above will now be explained.

40 The explanation will be given in case the input sentence "In 8702353 iiit 53 <λ> rt has input in the glossary 4020 and it is a proper name (4110) processing of a proper name registered in the glossary is performed (4114). the flowchart continues to figure 33. Because "Station" in the previous section is not a non-registered proper name (4200) but a registered proper name (4204) and because the type information (place, group) is not unknown (4106 ), and further because "Mr." has the type information of "person" and is not unknown (4208), it is checked whether there is a common part or not in the type information between "Station" in the preceding 10 section and "Mr." (4212) Because "Station" has the information "place, group", while "Mr." has the information "person" and there is no common information, "Mr." is only registered as a proper noun with the type information "person" (42 16).

15 Then the flowchart returns to Figure 32 and becomes the reference! Glossary 4020 addressed on "Walter" (4106). Since an entry for "Walter" is present in the ref glossary 4020 (4108) and it is a proper name (4110), the treatment for a proper name registered in the glossary is performed (4114). The flow chart continues to Figure 33. Because "Mr." in the preceding part, it is not an unregistered proper name (4200) but a registered proper name (4204), and because the type information is "person" and not unknown (4206) and because the type information for "Walter" which is equal to "person, place, group" is also not unknown (4208), an agreement between the type-information (4122) is examined. Because in both cases type information "person" is present, "Mr. Walter" is registered together as a single noun with type information "person" (4214).

Then, "with" is searched in the reference dictionary 30 4020 and because there is an entry (4108) and this is not a proper name (4110), the data obtained from the reference dictionary 4020 is registered in the dictionary information retention table 4036a (4112).

Furthermore, "Johnson" is searched in reference glossary 4020 (4106). Since there is no entry for "Johnson" (4108) and 35 because the first character is uppercase (4116), treatment for glossary proper noun is performed (4120). Then the flowchart continues to Figure 34. Since "with" in the preceding section is not a candidate for the end of the sentence (4300), it is judged that "with" is not the end of the sentence 40 (4312), "Johnson" is considered a proper name with unknown type 8702358 £ 54 ai information (4314) and treatment is being performed for a proper name registered in the glossary (4316). Then, the flowchart proceeds to Figure 33. Since "with" is not an unregistered proper name (4200) nor a registered proper name (4204) in the previous section, "Johnson" is separately registered as a proper name whose type information is unknown.

After the above operations, the standard type information is provided to the proper name as shown in Figure 35.

The pointer is set to "In" being the first words-10 list reference unit (4400). Since this is not a proper name (4402), the pointer is moved further (4408) and set to "Tokyo Station". Since "Tokyo Station" is a proper name (4402) and the type information is not unknown because the entire "Tokyo Station" section has been recognized as a place, group in the previous operation for the registered proper name (4404) becomes the pointer moved further (4408) and set to "Mr. Walter".

Because "Mr. Walter" is also a proper name (4402) and because the type information is not unknown (4404), the pointer is moved further (4408). Since "with" is not a proper name (4402), the pointer is moved further (4408). Because "Johnson" is a proper name (4402) and because its type information is unknown (4404), standard type information is provided (4406) and "Johnson" is provided with the type information "person, place, group, and others "as shown in figure 36.

As described above, in this embodiment, an English input sentence is divided into recovery key character arrays, which are then searched in the reference dictionary 4020 and if there is an entry as proper name in the reference dictionary 4020, processing for a registered proper name is performed. When processing the registered proper name, the preceding recovery key character array is considered, and if the preceding recovery key character array is a proper name, the type information of the previous recovery key character array and of the proper name as object is examined. If there is no type information then other type information is provided, while if any type information is present for both of them the common part is considered the type information for these proper names. It is therefore possible to appropriately provide a proper name without type information with suitable type information, as well as appropriately limiting type information provided to more suitable type-40 information. This allows a more effective analysis in the following syntax analysis and leads to a correct translation.

Furthermore, if the first character of a character array not registered in reference dictionary 4020 is uppercase and the preceding character array has been judged to be the end of the 5 sentence, and because the uppercase letter is converted to a lowercase letter and thus the referent. if list 4020 has been addressed again, it is also possible to search for the character array at the beginning of the sentence in the dictionary 4020. If, furthermore, a character array starting with a capital letter occurs in a part that is not the beginning of the sentence, then 10 this is judged as a proper name and the proper name type information is provided by means of a proper name with registered type information if it occurs before or after. Therefore, a proper noun not registered in the reference dictionary 4020 can be analyzed to some extent.

Furthermore, because a proper name not provided with type information is provided with all necessary type information and the unnecessary type information is removed during the processing of the word, it is possible to use a proper name, the type information of which is not known or analyze an unregistered proper name.

Because multiple type information is added to a given proper name and because suitable type information is selected depending on the type information of the proper name before or after it is possible to select suitable type information while analyzing a proper name with various types of type information depend on the relationship with others before or after, so that an effective analysis of the input sentence is possible.

In the following, a fifth embodiment of the present invention will be explained.

Figure 38 illustrates the entire structure of the fifth embodiment of the language analyzer of the present invention used in an automatic translation machine from English to Japanese.

This embodiment includes an input section 5010 and an English text 5012 to be translated into Japanese is entered through this section. For example, the input section 5010 may include a keyboard with character keys such as alphanumeric keys or function keys, an optical character reader (OCR reader) for reading an English text recorded on paper, and / or a file memory device for reading an English text is recorded on a memory medium such as a 40 magnetic memory medium.

6702358 56 y y y £

The English text entered from the input section 5010 is read into a pre-editing section 5014 in which translation pretreatment is performed. In this case, mainly sentence recognition and unknown word processing are performed. This functions as part of the morphological analysis.

The pre-edited English data, together with the information obtained during the pre-editing, is transferred to a morphological analysis section 5016. Section 5016 divides the data into sentences searching in a glossary 5018, analyzes English mor-10 femen, processes unknown words, proper names, various arrangements such as time expressions and number expressions, and performs operations on the entire sentence such as searching for fixed expressions and sayings. The morphological analysis rules are stored in the analysis rule file 5036.

15 The English data after the morphological analysis is transferred together with the words icy information obtained during the analysis to an analysis section I 5020. The analysis section I 5020 is a functional section that analyzes the surface structure of the sentence by applying grammatical rules to the English data and that traces all 20 structural possibilities.

The English data subjected to structure analysis in analysis section 5020 is sent together with the analyzed information to analysis section II 5022, where a solution is selected from the result of the structure analysis of the surface structure in analysis section I by applying a syntax analysis.

A plausible analysis structure for the English sentence is prepared in this way and its structure is formed. The analysis rules syntax are also stored in the analysis rule file 5036.

The English data, after submission to the syntax analysis, is transferred as syntax structure data to a structure transformation section 5024. In the structure transformation section 5024, a syn-tax structure for the corresponding Japanese sentence is prepared from the syntax structure which is an intermediate structure of the English sentence and which has been transformed into the Japanese underlying structure, from which the Japanese sentence can be easily translated.

The syntax structure data indicating the Japanese underlying structure and thus subjected to the structure transformation is transferred to a translation forming section 5026 in which a translated sentence is generated. This is a functional 40 section that generates a Japanese sentence from the Japanese structure.

8702 35 9 57 A *

1 V

The Japanese data resulting from the translation translation, that is, the data of the translated sentence, is sent to a post-digest section 5030. The post-digest section 5030 modifies the translated data by recovering a glossary 5018 using information that is obtained in the translation process in order to complete a more natural Japanese sentence. The Japanese sentence data is transferred to an output section 5032 and output to it as a translated Japanese sentence 5034. The output section may, for example, include a printer, a display screen or a file memory device such as a magnetic disk memory.

The flow of a series of such translation operations is controlled by a control section 5018 which controls the entire device.

The glossary 5018 contains glossary data for English and 15 Japanese words and includes various information such as interrelation, that is, an indication of whether words occur together, as well as meanings, singular or plural form, phrase, etc. in addition to the vocabulary itself . Furthermore, the analysis rule file 5036 contains the data of the morphological analysis rules and syntactic analysis rules.

The control section 5038 is connected to an operation display section 5040. The operation display section 5040 has operation keys with which the operator of the present device can give various indications, for example a translation indication key, cursor key, etc., and contains a display unit or indication screen with which the entered English text, the Japanese text as a result of the translation, intermediate data such as word list information, various indications for the operator and the like can be made visible. The display section may be configured so that most of the operation indication functions are included in the keyboard if position is selected at the input section 5010 or in a display panel if position is selected at the output section 5032.

Referring to Figure 37, a detailed structure intended for processing a proper name in the morphological analysis section 35 5016 will be considered as an example. Of the analysis section 5016, that portion is illustrated that is of direct importance for a better understanding of the present invention, although there are, of course, further functional analysis sections. The morphological analysis is performed by instructing the glossary search from the beginning of the 40 input character array in succession in accordance with

8702 35 S

i * e 58 the recovery key character array and the processing of the words obtained! To perform ice information in accordance therewith by the words ice lookup section 5104 on the basis of the positional information of the proper name, as will be described later.

The analysis section 5016 has an input processing section 5100 for receiving data from the input character array input from the preprocessing section 5014 and performing the input processing. The input processing section 5100 includes an input character array buffer into which the English character array data is input in the form of code data, for example, in ASCII, and in which the character array data is temporarily accumulated.

The input character array data temporarily accumulated in the input processing section 5100 is transferred to a unit excision section 5012 in which the data is divided into word reference units such as words. The unit excision section 5102 is a functional section that distinguishes between the word reference units that form the recovery key character arrays with which the glossary 5018 in the dictionary recovery section 5104 is subsequently addressed. The words -20 list reference limiters used in the excision operation for the words! ice reference units are placed at the position of the English character, number, apostrophe, characters other than hyphens and rest characters, as well as at and apostrophe following an empty character. They are stored in a limiter table 5108 and are referenced in the unit excision section 5102 during the cutting out of the words ice reference unit.

More specifically, the reference glossary 5018 contains the information for making the cutouts. For example, as shown in Fig. 38 for the example of an input information, grammatical information such as phrase information is present for each of the words ice reference units. The phrase information contains an indication for the noun whether it is an ordinary noun or a proper noun. For the proper name, a distinctive indication indicates how the position in the sentence is limited, ie positional information is stored for a proper name. This will be described in more detail later.

Other information is also recorded, for example whether a noun is countable or not countable, the distinction between a transitive or a non-transitive verb, its translation, etc.

8702359

* A

59

There are four types of positional information for a proper name, i.e. the patterns "0" - "3" in the present embodiment. The pattern "0" indicates a proper noun without a positional restriction, for example "City" or the name of a person "Walter". The pattern "1" 5 indicates that the proper name, for example "Mr." stands for a single proper noun or a series of proper nouns, that is, a proper noun arranged in a singular group of proper nouns. The pattern "2" indicates that it is a proper name such as "Station" or "Bay" located at the end of a single proper name, or a proper-10 name that is arranged as a word in a group of proper names and that differs of the pattern "3" to be described. The pattern "3" indicates that it is a proper name such as "River" in "the Sumida River", which is the same as pattern "2" but belongs to a defined preposition "the" arranged at the beginning of the proper name 15 in a group of proper names.

The word list lookup section 5104 is a functional section that obtains glossary information by searching the word list 5013 based on the recovery key character array input from the unit excision section 5102 and transmitting it to the word list list retention table 5124, the positional information processing section 5110 and the section 5112 determining the end of a preceding sentence.

The processing based on the patterns "0" - "3" in accordance with the positional information of the proper names, derived from the glossary 5018, are performed by the proper name processing sections 5114, 5116 and 5118. The processing of the proper names are for the pattern 1 in the proper name processing section 5114, for the patterns 2 and 3 in the proper name processing section 5116 and for the pattern 0 in the proper name processing section 5118, respectively. 5114.

In this embodiment, the proper names are arranged collectively using as a key a word that is part of a group of proper names grouped into a single proper name expression and in which a positional restriction occurs in the arrangement. Thus, even if a number of proper nouns occur continuously, these can be correctly arranged together with their context without being erroneously always considered simply a single group of proper nouns. The edits intended for this purpose are performed in the proper name edit sections 5114, 5116 and 5118. Proper names are registered to a certain extent 40 in the reference glossary 5018. Such proper names listed in the glossary are 8702355% £ 60 subject to analysis in the positional information processing section 5110 and in the proper noun editing sections 5114, 5116 and 5118. They form an editing section for glossaries registered nouns, the proper nouns not registered in the glossary 5018 are analyzed in section 5112 where the end of the preceding sentence is and in the section 5118 if the proper name has the pattern “O”. These are the edit sections for the proper names not registered in the glossary.

The operation of a proper name is performed by the following 10 two steps. First of all, a proper name is recognized in the input character array. In the case of a word registered in the glossary 5018, this is done because the proper name is indicated in its morph change information. Furthermore, in the case of a word not registered in the glossary 5018, this is done because the character at the beginning is an English capital letter, for example, "John" or "U.S." etc.

A group of proper names is then arranged collectively in order to consider the entire section as a single proper name.

When it is recognized as a proper name from the glossary information and if the next word reference unit is also a proper name, the entire portion is collectively synthesized into a single proper name.

For example, "M. Weber" is analyzed in its entirety as a single proper name. The result of the analysis is a candidate for grouping the ideomatic expression including the proper names in the local analysis.

The necessary local analysis is then carried out. In this case, a sequence of analysis units is actuated by the morpheme actuation information for each of the analysis units collectively arranged in an analysis unit based on a local analysis rule, for example, sorting by type. For example, "Mr. Brown" is collectively ranked in "Brown shi". Furthermore, words that form part of an area name 35 are also held together. For example, "Lake Biwa" is arranged together into "Biwako." Likewise, words that are part of the group name are also arranged collectively. For example, "Yale University" is analyzed as "Yale Daigaku".

In the case of the proper name "Mr. -" or "Lake -", 40 contextually, there is always an ending just before 8702359 i 61 r x this word. In the English sentence "With Tom Mr. Brown went to -" there is therefore a certain ending between "Tom" and "Mr." as for the context. Therefore, if "Tom Mr. Brown" has been merged into a single own name, then an error has occurred in the following analysis. For example, the 5 proper noun "University" is always followed by an ending. In the English phrase "At Yale University Tom is ----", it is recognized that there is a pause between "University" and "Tom". In this embodiment, information about the position where the respective proper names undergo a positional restriction in a sequence of proper names is stored in the glossary 5018 as the above described positional information, i.e. as the patterns "0" - "3" .

The composite configuration is created using processing positions 5110, 5112, 5114, 5116 and 5118 using these positional information. The glossary information for the input character array is stored in the retrieved glossary information buffer, ie, the glossary information buffer, after the completion of these operations. holding table 5124.

The result of the morphological analysis is transferred from the glossary information retention table 5124 to the analysis section I 20 5020.

The processing of the proper name position information is performed through the sequence shown in Figure 40. The input processing is performed by receiving the data from the input character array in the input processing section 5100 (5200). Thereafter, the unit cut-out section 5102 cuts the input character array into glossary reference units to address the glossary 5018 (5201).

The glossary lookup section 5104 addresses the glossary 5018 accordingly (5203), and if a glossary entry data is present (5204), the phrase thereof is examined (5205). If the phrase is not a proper noun, the processing of the proper name is not performed in this embodiment, but the glossary! information accumulated in the glossary information retention table 5124 (5206). If it is a proper name, then the treatment for a glossary-registered proper name 5207 is performed in the positional information processing section 5110 and in the proper name processing sections 5114, 5116 and 5118. If these treatments are performed up to the last position of the sentence, indicated by the data for the input character array (5202), the result of the morphological analysis is output to the sentence structure analysis section I 5020 40 (5210).

£ 8702359 * 62

If no entry is found as a result of the dictionary search in step 5204 and if the element starts with a capital letter (5212) then it is recognized as a proper name not registered in the dictionary and treated as a non in the 5 glossary registered proper name in 5213, in section 5112 that reviews the preceding section and in section 5118 which covers the proper name. If the initial character is not capitalized, then the word, because it is not registered in the glossary 5018, is stored in the words list information retention table 5124 as an unregistered word (5214). The operation is continued to the last position (5202).

The treatment of glossaries registered in the glossary 5207 is performed in processing sections 5110, 5114, 5116 and 5118 in the flow chart shown in Figure 41. First, reference is made to the obtained positional information contained in the glossary information (5220) . Thereafter, the proper name processing 5221 for the pattern 0 is performed if the pattern "0" is indicated, the proper name processing 5222 is performed for the pattern 1 if the pattern "1" is indicated, and the proper name processing 5223 is performed for the patterns 2 , 3 if the patterns "2" resp. "3" are indicated.

The proper name processing 5221 for the pattern 0 is performed in the processing section 5114. The processing is applied to a proper name without positional restriction. To begin with, if a part-25 prior to the words list reference unit in question is an unregistered proper name (5230), then the entire part is merged into a single proper name with the proper position name information "1" and stored in the words! list preservation table 5124 (5233). If the preceding part is a proper name with positional information pos "1" (5231) then the treatment is performed in the same manner.

If the preceding part is a proper name with positional information "O" (5231) then the whole part is merged into a single proper name with positional information of "O" and stored in the words! List information retention table 5124 (5235). the preceding part is not a proper name with positional information "0", then the entire part is stored only as a proper name with positional information "0" in the glossary information retention table 5124 (5134).

40 The proper name processing 5222 for the pattern 1 is shown at 8702359 63

* X

T t described in the manner described. This processing is applied to a proper name such as "Mr.11 situated at the beginning of a single proper name or at the beginning of a proper name belonging to a group in a consecutive series of multiple proper names. As to start the part prior to the glossary reference unit in question is an unregistered proper name (5240), then the word is transformed into an unregistered word (5241) If it is an unregistered proper name, the word is stored only as an own name with the positional information pos "1" in the glossary information retention table 5124 (5242).

Now an explanation will be given of the proper name processing 5223 for the patterns 2, 3 with reference to Figure 44. The processing is performed, for example, on proper names such as "Station" or "River" which are located at the end of a single proper name or which as 15 proper names belong to a series of consecutive multiple proper names. First of all, if the part that is an unregistered proper name (5250) prior to the glossary reference unit in question, then they are arranged together with the preceding words in a single proper name with the positional information pos-self associated therewith as the positional proper name information pos and stored in the words list information retention table 5124 (5225). Furthermore, the treatment is performed in the same manner as the preceding part is a single proper name with positional information "1" (5251).

If the preceding part is not a custom name with positional information "0" (5252) then it is concatenated into a custom name expression with the associated positional information pos-self and stored in the glossary information retention table 5124 (5257).

In step 5252, the preceding portion is a proper name with positional information "0", then the corresponding positional proper name information is pos-self checked (5253) and the treatment 5255 is performed if the pattern is "2". If the own positional proper name information pos-self is equal to the pattern "3" then it is further checked whether the element prior to the glossary reference unit is equal to "the" or not. If it is not the defined word "the", the treatment 5255 is performed. If it is "the", then the group between "the" and the element under consideration is merged into an own name expression with the positional own name information "3" and stored in the glossary information retention table 5124 (5256).

40 For a word that starts with a capital letter and is recognized as 8702358 64 * * '* · 5 an unregistered word for which no entry is found in the reference glossary 5018 as a result of the glossary search 5203, the flow chart continues through the steps 5204 and 5212 with treatment 5213 performed in section 5112 in which end of the preceding sentence is judged. First of all, if the preceding portion of the glossary reference unit in question is not a candidate for the end of a sentence, the proper name processing 5221 for the pattern 0 is performed in the processing section 5118 as described above.

10 The preceding section may run for the end of a sentence in the following four cases. The first is the case where a separate one is present. Then there is the case where the preceding data has a dot at the end and the positional information of the proper noun is not "1". This case includes, for example, an abbreviation "U.S.A.". Furthermore, there is the case of one or the sequence of a point and an apostrophe ". And the sequence of a point and a quotation mark. The last case is that of the beginning of the input character array buffer.

20 If the preceding part belongs to any of the above four cases, then the preceding part is recognized as a candidate for the end of the sentence (5261) and the word list is used again after converting the capital letter of the word to a lower case letter (5262). As a result of this search process, if a word entry entry data is obtained (5263), it is stored in the words entry retention table 5264 (5264). If not, it is recorded as an unregistered proper noun in the words! List retention table 5264 with the initial character unchanged as capital letter (5265).

30 An explanation will be given by example.

For example, if the glossary is addressed on an input character array “Along the Sumida River Paul and Mr. Gold Smith get used —— ", then the glossary entry information is first written in the words! List information retention table 5124 as shown in Fig. 35 46A. For example, referring to" the ", the start position in the sentence is" 7 "and the end position is" 9 " "and as a phrase, this word is a definite article. No entry is found for the word" Along "at the beginning of the input character array during the glossary search 5203 and this word is therefore considered unregistered. Because the preceding portion can meet the 8702359 65, Λ Μ τ * prerequisites for a sentence ending and thus it can be at the beginning of the input buffer (5260), the capital "A" is converted to a lower case at the beginning and again a search process 5262 performed in the glossary on "along".

5 Then the pointer is moved to continue processing "Sumida". This word is not registered in the glossary 5018 in this embodiment. Since the preceding portion is not a candidate for the sentence ending, the flowchart proceeds to the proper name processing 5221 for the pattern 0. As shown in Figure 46A, a proper name for the phrase and "0" for the positional proper name information are found.

The following glossary reference unit "River" is a proper name with positional proper name information "3". The preceding part is a proper name with positional proper name information "0" and the preceding part is equal to "the". As explained above, "the Sumida River" is merged into a single proper name through steps 5250-5254 and stored with the positional proper name information "3" in the glossary information retention table 5124 (Figure 46B).

Thereafter, the next glossary reference unit "Paul" is an unregistered proper name with positional proper name information "0" for which the treatment 5213 is performed. Although the preceding word is a proper name, since the positional information therefor is "3", it is not aggregated but the word list information is accumulated as it is in Table 5124 (Figure 46C). The normal operation is performed on the "and" suffix that follows.

The next word "Mr." is a proper name with positional information "1", and it is accumulated as it is in the glossary information retention table 5124 (Figure 46D). Although the preceding section is a proper name "Paul", because there is a division between the words immediately before "Mr.", this word can be preserved as such in the words list information retention table 5124.

The word "Gold" is further a proper name not registered in the glossary, and the treatment 5213 is applied thereto. Because the word "Mr." preceded by the positional information having "1" are both joined together and the entire portion is provided with the positional information "1" as a single proper name (Fig. 46E). Then, the same treatment is performed for the next word "Smith" (Figure 46F). The subsequent "used" is one

S70 2 35 S

66>? The past tense of a verb and then the normal analysis is applied.

As described above for the present embodiment, the proper names are concatenated using as key a word that is part of a group of proper names formed together into a single proper name and subject to a positional constraint if they are combined into a single proper name. In this way, even if a continuous number of proper names occurs, it is possible to find the correct collective composition in the context without obtaining an incorrect collective composition simply by dividing them into groups of proper names. For example, in the above example, "the Sumida River" is analyzed as a group of proper names separated from the subsequent "Paul". Furthermore, "Mr. Gold Smith" is also analyzed as a group of 15 associated proper nouns.

Next, an explanation will be given of the sixth embodiment of the present invention.

Figure 47 illustrates the sixth embodiment of the language analyzer according to the present invention used for automatic translation from English to Japanese.

The present embodiment has an input processing section 6014 and data is input into the input processing section 6014 from an input unit 6012. The input unit 6012 includes, for example, a keyboard with character keys such as alphanumeric keys or function keys, an optical character for reading English text. recorded on paper and a readout unit of a magnetic disk memory.

The input processing section 6014 has an input character array buffer 6014a and contains the English input sentence input from the input unit 6012 into the input character array buffer 6014a. The input processing section 6014 reads out the input sentence stored in the input character array buffer 6014 and outputs it to the unit excision section 6016. The unit excision section 6016 is a functional section which cuts out the words list reference units from the input sentence received from the input processing section 6014 with using the limiter table 6018. The limiter table 6018 contains the limiter such as space, comma, etc. The unit excision section 6016 reads the limiter from the limiter table 6018 and divides the sentence from the input processing section 6014 into character arrays as units for 40 addressing a reference glossary 6020 by changing the sentence 870235 »67 4, J.

T T

parts in parts, where limiters are present. The sub-character arrays are entered in the glossary lookup section 6022.

The glossary lookup section 6022 addresses the reference glossary 6020 using the input phrase from the unit excision section 6016 and divided into the glossary reference units. For example, the reference glossary 6020 contains, as shown in Figure 48, input data for the character arrays, phrases thereof, type information, etc., from the English sentence. The reference dictionary 6020 additionally contains the proper names shown in the figure, which are character arrays that form other phrases, for example verbs, adjectives, etc. The proper name in this figure means that proper name which is a registered and later to be used proper name, but does not refer to the usual grammatical proper noun. Furthermore, the type information indicates what the proper noun expresses, and no restriction is made to a single expression.

The glossary lookup section 6022 searches the reference glossary 6020 for the character array distributed in the glossary reference units and if the character array is a proper name, it is output to the proper name processing section 6024 for the proper proper name treatment to be described later. If not a proper name, it is issued to a processing section 6036 and stored in the words! ice information retention table 6036a in the processing section 6036.

The proper name processing section 6024 includes a section 6026 for processing the preceding sentence end, a section 6028 for processing a previous proper name, and a section 6030 for processing a proper name as such.

The preceding sentence ending section 6026 judges whether a character array prior to the input character 30 array received from the glossary lookup section 6022 is at the end of a sentence or not and if the preceding character array is at the end of the sentence then it is sent to the glossary lookup section 6022 after capitalizing the capital letter at the beginning of the character array to be processed, after which the glossary-lookup section 6022 again addresses the reference glossary 6020. The character array not found on the second attempt is judged as an unregistered proper name and sent to the processing section 6036 and stored in the words list information retention table 6036a. Furthermore, if the character array preceding the entered 40 character array is not at the end of a sentence, it is supplied 8702350 68: »* · ï * is entered to the processing section 6036 as a proper name whose type information is unknown and registered in the glossary information retention table 6036a as described later.

The preceding proper name processing section 6028 analyzes the type information of the preceding character array transmitted from the preceding sentence processing section 6026 and outputs the result to the processing section 6036. proper noun as such. The section 6030 of the proper name processing as such checks the type information for the proper name to be decomposed and if the type information is not registered with that of the proper name and of the previous proper name then, as later described the proper name with the previous proper name merged together by the registered type information of the other and the result is stored in the glossary information retention table 6036a in the processing section 6036.

The processing section 6036 includes the glossary information retention table 6036a and contains the data transmitted from the section 28 for processing the previous proper name or the glossary lookup section 6022 in the glossary information retention table 6036a and 20 then reads out and stores the data stored in this manner these to the structure analysis section 6038. The analysis section 6038 searches the syntactic analysis on the input sentence, read from the glossary information retention table 6036a, and subjected to the morphological analysis.

The operation of the device will now be explained with reference to the flow chart illustrated in Figure 49.

To begin with, an English input sentence from the input unit 6012 is read into the input processing section 6014 (6100). The input sentence read in the input processing section 6014 is stored in the input character array buffer 6014a. The sense stored in the input character array buffer 6014a 30 is read in the unit cutout section 6016.

Once the input sentence has been entered, the unit cut-out section 6016 reads the limiter from the boundary table 6018 to perform the cutting of the glossary reference units (6102). The character arrays constituting the input input sentence are successively shared from the beginning in the recovery key character arrays that form the units with which the reference glossary 6020 is addressed by dividing at the positions where the delimiters are present such as spaces, commas, etc. It is judged whether or not the termination into glossary reference units, i.e. the recovery key character-40 ray's, has been terminated (6104) and if there is still a recovery 8702359 69 <-t T f key character array present (no end), the recovery key character array sent to the list lookup section 6022.

If the recovery key character array is sent to the glossary lookup section 6022, then the glossary lookup section 6022 accesses the reference glossary 6020 using the recovery key character array (6106). It is judged whether or not the recovery key character array is present as input data in the reference word list 6020 as shown in Figure 48 (6108) and if an input data is present then the phrase information stored in the reference word list 6020 is read out and assessed whether the recovery key character array is a proper name or not (6110).

In case the recovery key character array is not a proper name, the glossary lookup section 6020 sends the information read from the reference glossary 6020 to the processing section 6036 in which it is stored in the glossary information retention table 6036a (6112). If the data is stored in the words! ice information retention table 6036a, a characteristic indicating that the data is stored with the data of the immediately preceding recovery key character array is input from the processing section 6036 into the unit excision section 6016. In this manner, the flowchart returns to step 6012 and again becomes an excision. for a word list unit performed in the unit cutout section 6016.

If in step 6110, the recovery key character array is a proper name, then the glossary lookup section 6022 sends the proper name read from the reference words list 6020 (simply referred to as the proper name hereinafter) along with the data of the previous recovery key character array input from the glossary information retention table 6036a the processing section 6036 through the unit excision section 6016 to the glossary lookup section 6022, and to the proper name processing section 6024 where the processing of the proper name registered in the glossary is performed (6114).

Next, an explanation will be given of the processing of a glossary-registered proper noun with reference to the flow chart of Figure 50.

The data sent from the glossary lookup section 6022 to the proper name processing section 6024 is input from the preceding sentence end processing section 6026 to the preceding proper name processing section 6028. In the processing of the proper name registered in the glossary, section 6026 plays no role in processing the preceding sentence end.

8702358 70 «

In the preceding proper name processing section 6028, it is judged whether or not the recovery key character array, prior to the proper name, is a proper name not registered in the reference dictionary 6020, i.e. whether it is a proper name that is subject to processing. of a proper name registered in the glossary (6200) or not, as will be described later. If an unregistered proper name is concerned, the entire proper name with the previous unregistered proper name is judged as a proper name with the proper name type information (6202), and the data is sent to the processing section 6036 and stored in the words List information retention table 6036a (6214).

In the preceding proper name processing section 6028, if the recovery key character array prior to the proper name is judged as an unregistered proper name, it is judged in step 6200 15 whether or not the recovery key character array prior to the proper name is a proper name registered in the reference dictionary 6020 (6204). If the recovery key character array prior to the proper name is a registered proper name, then it is judged whether the type information of the previous proper name is unknown or not, ie whether it is not registered in the reference dictionary or not 6020 (6206).

If the previous proper name type information is unknown, the flowchart proceeds to step 6202, wherein the entire proper name with the previous proper name is considered a single proper name with the proper name type information (6202), and the preceding proper name processing section 6028 sends the data to the processing section 6036. The data sent to the processing section 6036 is stored in the words list information retention table 6036a (6214).

In the preceding proper name processing section 6028, if the previous proper name type information is judged as unknown, that is, if it is judged as registered in the reference dictionary 6020, the data is transferred from the processing section 6028 from the previous single proper name to section 6030 for proper name processing. In the proper name processing section 6030, it is judged whether or not the proper name type information is unknown (6208). In case the type information for the proper name is unknown, the proper name processing section 60330 judges the entire portion of the proper name with the previous 40 proper name as a proper name with the type information of the previous 8702359 71

A A

r> own name (6210) and sends the data to the proper name processing section 6036. The data sent to the processing section 6036 is recorded in the glossary information retention table 6036a (6214).

5 In the proper name processing section 6030, if it is judged that the proper name type information is not unknown, that wi! say if it is registered in the reference words! list 6020, the proper name judged as a proper name with the type information originating from the reference dictionary 6020, independent of the previous proper name (6212) and the data is sent to the processing section 6036. The data transmitted to the processing section 6036 is recorded in the glossary information retention table 6036a (6214).

Referring again to Figure 49, if there is no recovery 15 key character array as input in the reference glossary 6020 in step 6108, it is judged whether the first character in the recovery key array is capitalized or not (6116) and judged capitalized. the glossary lookup section 6022 recovers the key character array as an unregistered word, sends it to the processing section 6036 and stores it in the glossary information retention table 6036a (6118).

If the first character is uppercase, the data for the recovery key character array together with the data from the previous recovery key character array is sent from the glossary search section 6022 to the proper name processing section 6024 in which processing for a proper name not registered in the glossary is performed (6120 ).

An explanation of the processing of a non-glossary proper name will now be made with reference to Figure 51.

The data for the recovery key character array, together with the data of the preceding entry data recorded in the glossary information retention table, is sent to the section 6026 for processing the preceding sentence end if it is judged that the end of the preceding input data in the glossary information retention table is a candidate for the sentence ending or not (6300). This end-of-sentence candidate is made by judging whether the end of the preceding entry entry the glossary information retention table is an end-of-sentence candidate 40, such as a separate period (.), Etc., or 8702358 72 & i A '4 not.

If the end of the preceding entry data recorded in the glossary information retention table is a candidate for the end of the sentence, the data from the end sentence of the preceding sentence processing section 6026 is sent to the section 6028 for processing the end sentence. preceding proper name and section 6028 judges the preceding entry recorded in the glossary information retention table as the end of the sentence (6302), and transmits it to the words list lookup section 6022 after changing the capital letter 10b at the beginning of the recovery key array in a lowercase letter.

The words list lookup section 6022 searches the reference words list 6020 for the recovery key character array transformed with a lowercase letter (6304) and judges whether there is an entry in the reference word list 6020 (6306). If an entry is found, then the words list lookup section 6022 sends the data from the reference glossary 6020 to the processing section 6036 and stores it in the glossary information retention table 6036a (6308). If no entry is present, then the words list lookup section 6022 capitalizes the first character of the recovery key character array and sends the word as an unregistered proper name to the processing section 6036 and stores it in the glossary information retention table 6036a (6310). ).

In step 6300, if the preceding sentence end processing section 6026 judges that the end of the previous entry in the glossary information retention table is not a candidate for the end of the sentence, the data from section 6026 is fed to the section 6028 for processing the previous own name and section 6028 judges that the previous entry data registered in the words list information retention table is not the end of the sentence (6312). The data is sent from the section 6028 for processing the previous proper name to the proper name processing section 6030, and the section 6030 judges the recovery key character array as a proper name whose type information is unknown 35 (6314).

The proper name processing section 6030 returns the data to the section 6028 for processing the previous proper name and the processing of the proper name registered in the dictionary is performed in the processing section 6028 (6316). The processing of the proper name registered in the 40 word list is the same as that shown in figure 8702359 73 * Λ figure 50.

Referring back to Fig. 49, when the glossary reference unit excision is terminated in step 6104, the data recorded in the words ice information retention table 6036a is output from -5 from the processing section 6036 to the structure analysis section 6038 (6122) that performs the morphological analysis in this document. embodiment is completed.

The operation of the present device described above will now be explained with reference to an exemplary input sentence.

The explanation will be given with reference to Figure 52 in which, for example, an input phrase "In Tokyo Station Mr. Walter -" is entered. First, the input processing 6100 for reading the input sentence into the input processing section 6014 is performed. Then, the glossary excision units are cut (6102) by dividing the input sentence into the respective words by the spaces. First, the reference dictionary 6020 searches for "In" (6106). There is no input for "In" in the reference glossary 6020. If processing of nouns not registered in the glossary is continued, the previous part is recognized as the beginning of the sentence (the beginning of the file ) "In" transformed into "in". Because "in" has an entry in the reference dictionary 6020 and because it is not a proper name "6110), the data 6020 from the reference dictionary is recorded in the glossary information retention table 6036a (6112).

Next, the reference glossary 6020 is queried on "Tokyo" (6106). Since there is no entry for "Tokyo" in the reference dictionary 6020 (6108) and because the first character is uppercase (6116), processing for a non-glossary proper name is performed (6120). As a next step 30, continue with Figure 51. Since the preceding part is equal to "In" and is not a candidate for the end of the sentence (6300), "In" is recognized as not belonging to the end of the sentence ( 6312), "Tokyo" is recognized as a proper name whose type information is unknown (6314) and processing for a glossary registered proper name is performed (6316). The next step is continued with figure 50. Since the preceding "In" is not a registered proper name (6200) nor an unregistered proper name (6204), "Tokyo" is registered in itself as a proper name with its own type information, ie say a proper noun of which the type-in-40 formation is unknown (6216).

870235S

74 • T, 1 Λ $

The next step is continued with figure 49 and in the reference word list 6200 a search is made for "Station" (6106). Since there is an entry for "Station" in the reference dictionary 20 (6108) and this is a proper name (6110), processing for a proper name registered in the dictionary is now performed (6114). Next, proceed to Figure 50. Since the previous "Tokyo" is not registered as a proper name (6200), the entire portion "Tokyo Station" is registered as a proper name with as type information "place" associated with "Station" (6202).

Then, in the reference dictionary 6200 in Figure 49, a search is made for "Mr." (6016). Because there is an entry for "Mr." is present in the reference glossary 6020 and because it is a proper name (6110), processing for a proper name registered in the glossary is performed (6114). Then continue with figure 50.

The preceding "Station" is not an unregistered proper name (6200) but a registered proper name (6204) and the type information "place" is not unknown (6206). Because "Mr." if type information has "person" which is not unknown (6208), "Mr." individually registered as a proper name with "person" as type information (6212).

Returning again to Figure 49, the reference glossary 6020 is examined for "Walter" (6016). Because "Walter" has an entry in the reference dictionary 6020 (6108) and because it is a proper name (6110), processing for the proper name registered in the dictionary is performed (6114). Then continue with figure 50. Because the previous "Mr." is not an unregistered proper name (6200) but a registered proper name (6204) whose type information "person" is not unknown (6202), while "Walter" type information is unknown (6208), "Mr. Walter" merged-30 and registered as a proper name with as type information "person" (6210).

As already described above, in the present invention, the English input sentence is divided into recovery key character arrays, which first query the reference glossary 6020. If an entry is present as a proper name in the reference dictionary 6020, processing for a registered proper name is performed. The processing for a registered proper name is carried out while taking into account the previous input data, registered in the words list information retention table. If the previous entry, registered 87 0 2 359 75 <Α · - r τ in the words! List information retention table, is a proper name, then the type information is checked and the current proper name is processed. If either type information is missing, then the other type information is added, while if both are ahead of type information, they are recognized as individual proper names with respective inherent type information.

It is therefore possible to provide a proper name without type information with suitable type information and also to limit the type information that is present in a more correct manner. This allows for a much more effective analysis in the following structural analysis and also an adequate translation.

If further from the character array not registered in the reference dictionary 6020, the first character is uppercase and if the preceding character array is judged as belonging to the end of the sentence, the reference dictionary 6020 is searched again after changing the uppercase letter to lowercase and it is therefore possible to query the reference glossary 6020 also for a character array at the beginning of the sentence. Furthermore, if a character array starting with a capital letter appears in a part that does not belong to the beginning of the sentence, it will be judged as a proper name and this proper name will be provided with the type information associated with a proper noun that precedes it. or follows it. In this way, a proper name not registered in the reference glossary 6020 can be decomposed to a certain extent.

Reference is then made to the seventh embodiment of the present invention.

Figure 54 illustrates the entire structure of the seventh embodiment of the language analyzer according to the present invention used in an automatic translation device from English to Japanese.

This embodiment has an input section 7010 through which an English text 7012 to be translated into Japanese is input. Input section 7010 may include, for example, a keyboard with character keys such as alphanumeric keys or 35 function keys, an optical character reader (OCR reader) that reads an English text, recorded on paper, and / or a file memory device for reading an English text that is recorded in a memory medium, such as on a magnetic disk.

The English text, entered through input section 7010, is entered into a pre-editing section 7014 in which the pre-treatment for the 6702359 76% translation is performed. In this case, mainly sentence recognition and unknown word processing are performed. These functions are part of the morphological analysis.

The pre-edited English data, along with the information obtained in the pre-editing, is transferred to a morphological analysis section 7016. Section 7016 analyzes the morphemes of the English sentence while it is distributed by reference to dictionary 7018, performs various types of formats such as operations of a unknown word, proper noun phrases, time, numbers, etc. and performs 10 operations on the entire sentence such as searching for fixed expressions and sayings. The morphological analysis rules are stored in the analysis rule file 7036.

The English data subjected to the morphological analysis, together with the glossary information obtained from the analysis, are transferred to a structure analysis section I 7020. The structure analysis section I 7020 is a functional section that analyzes the surface layer structure of the sentence while grammatical rules are applied to the UK data in order to detect all structural possibilities.

The English data subject to the analysis in the structure analysis section I 7020 are sent together with the analysis information to a structure analysis section II 7022 in which a solution is selected from the result of the analysis of the surface layer structure in the structure analysis section I by application of a syntax 25 analysis. A plausible analysis structure of the English sentence is prepared in this way and the structure is manufactured. The analysis rules are also stored in the analysis rule file 7036.

After submission to the analysis, the English data is transferred as data from the analysis structure to a structure transformation section 7024. The structure transformation section 7024 prepares a corresponding Japanese structure from the structure that forms an intermediate structure with the English structure and transforms it into an underlying Japanese structure. structure from which a Japanese sentence can be easily translated.

35 The analytical structure data that the transformed in this way

Japanese underlying structure tones are transferred to a translation generating section 7026 in which the translated sentence is generated. This is a functional section in which the Japanese sentence is generated from the Japanese analysis structure.

40 The data for the sentence translated into Japanese, that is, the 8702359 77At translated sentence data is sent to a post-digest section 7030. The post-digest section 7030 modifies the translated phrase data with reference to the glossary 7018 using the information used in the translation process in order to form a more natural 5 Japanese sentence. The Japanese sentence data is transferred to an output section 7032 and hence output as the translated Japanese sentence 7034. The output section may include, for example, a printing unit, a display unit and / or a file memory device such as a magnetic disk memory.

The sequence of translation steps is controlled by a control section 7038 with which control is performed over the entire device.

The glossary 7018 stores the glossary data of the English and Japanese words in this embodiment. Stored therein are the vocabularies, as well as the interrelationships, ie connecting relationships and various information such as meaning, singular or multiple form, phrase, etc. Furthermore, control data for the morphological analysis and the syntax analysis are stored in the analysis rule file 7036 .

The control section 7038 is connected to the operation display section 7040. The operation display section 7040 includes operation keys that enable various instructions from an operator at the present device, for example, a translation instruction key and a cursor key, and a display unit or a display that displays the entered English sentence text, the Japanese text as a result of the translation and any intermediate data such as glossary information and various instructions can be made visible to the operator.

The device can be configured so that most operation display functions can be included in the keyboard if it is mounted on the input section 7010 or in a display unit 30 if it is located on the output section 7032.

In Figure 53, a detailed structure of the morphological analysis section 7016 is shown in particular for digit processing. Those parts that are directly relevant to an understanding of the present invention have been shown, but the analysis section 7016 has, of course, other functional analysis sections.

The section 7016 has an input processing section 7100 for receiving and processing the input character array data input from the precursor section 7014. The input processing section 7100 includes an input character array buffer in which the English character-40 array is input in code form, for example, as ASCII code data 8702359; * 78 that temporarily accumulates the character array data.

The input character array data temporarily accumulated in the input processing section 7100 is fed to a unit excision section 7102 which divides the data into word list units such as words. The unit excision section 7102 is a functional section that distinguishes vocabulary reference units that are part of the character array, which units are used to access the vocabulary 7018 by the subsequent vocabulary search section 7106. Glossary reference delimiters used in the vocabulary reference unit cut-out process. are placed at the position of an English character, a numeric character, an apostrophe, characters other than a hyphen and a rest character, as well as an apostrophe following an empty character. They are stored in the limiter table 7104 and used during the cutout of the glossary reference unit in the unit cutout section 7102.

The glossary 7018 specifically includes information for looking up a cut unit. Furthermore, the glossary 7018 contains morphological analysis information such as the names of months, calendar-20 days, base numbers that only have numerical meaning, ordinary numbers, units for expressing grams and the like, time, the or commas (,), points (.), etc.

The word list lookup section 7106 is a functional section that accesses the word list 7018 to extract word list information based on the character array input from the unit excision section 7102 which is transferred to the section 7108 to provide the morphological information. analysis information.

The section 7108 for providing the morphological analysis information contains information (see also Figure 56) indicating that a sequence of characters with morphological properties has a time meaning such as hour, year, month, etc. while further providing specific information to the character array recognized as containing a base or time meaning in the words! ice-pop-up section 7016.

35 For example, "year" provides the information "numeric number".

The character array, provided with the analysis information in the information providing section 7108, is further subjected to the required local analysis.

40 In this case, a unit group of the words! Ice reference 8702359 Λ I · I Ϋ 79 unit, such as a word, is actuated by joining the morpheme information together into a unit using the local analysis rules. For example, "name of the month", "numeric expression" are added together to "name of the month + numeric expression", i.e. "Oct." and "18" are grouped into "Oct.18". In addition, collective combinations are made such as "November the 2nd" is combined to "name of the month + the + numeric expression", "22 March" to "numeric expression + name of the month," the 23rd May "to" the + numeric expression + name 10 of the month "," the 11th of June "to" the + numeric expression + or + name of the month "," '86, Jan. 27. Mon. "to" year +, + basket and day +, + calendar day "," sunday, 26, jan. 1986 "for" calendar day +, + month and day +, + year "," 11:30 am "to" numeric value: numeric value + am (or pm) "or" name of month + year "," name of 15 month + or + year ", etc.

The local analysis operation is performed in an initial value setting section 7110, a matching derivative section 7112, a unit cut-out section 7114, a morpheme processing information supplying section 7118, the lookup sections 7116 and 7120 and the processing sections 7122 and 7124 as well as a match table 7128 indicating a morpheme processing indication which is a distinctive reference table which recognizes that a sequence of units contains numerical numbers and time factors as shown in Fig. 57 for a composite unit of time factors that satisfy a certain rule. The initial value setting section 7110 sets the initial value of a counter n which counts the number of vocabulary reference units that are matched in successively looking up vocabulary reference units as a unit group as described above in the matching agreement. section 7112.

The matching indicating section 7112 searches the matching table 7128 for each of the words reference reference units to find a match. The unit excision section 7114 distinguishes the glossary reference units, designated "p", at the end of the glossary search in the glossary search section 7106 from the glossary reference units that form the character arrays after the glossary reference units are completed in the glossary search process using the glossary search units. counter n.

The look-up section 7116 is a functional section with a similar function to that of the words ice-look-up section 7106, which addresses the 8702 359 80, woorden a <* glossary 7018 to extract glossary information therefrom based on the cut-out in the unit cut-out section 7114. character array which information is transferred to the morpheme processing information providing section 7118. The morpheme processing information providing section 7118 has the same function as the morpheme processing information providing section 7108, in which further specific information is added to each of which is recognized as an ordinary number or as a time factor in the lookup section 7116.

The look-up section 7120 and the processing sections 7122 and 7124 merge a sequence of word list reference units up to "p + n" obtained from the match-looking section 7112 via the processing in the morpheme processing information generating section 7118 in a single word reference unit. Thereafter, the record is stored in the words list information retention table 7126 which forms a buffer for storing the words list information obtained during the search.

The result of the morphological analysis is transferred from the word list retention table 7126 to the syntax analysis section I 7020.

In the following, an explanation is given of the morpheme processing information in accordance with the present invention referring to the flow charts in Figures 55A and 55B.

For example, it is believed that the following character array 25 is input into the input processing section 7100 (7300). Input character array: "..26 Jan., '80 hey .."

The unit excision section 7102 divides the input character array into glossary reference units to address the glossary 7018 (7302). "26" in the input character array is cut out as a unit for use as a glossary reference unit. It is judged whether the cutout of glossary reference units has been terminated or not for this input character array and if it has been terminated then the operation is terminated (7304) while if it is not yet terminated the flowchart proceeds to the next step 7306.

The glossary 718 is addressed at "26" in the input character array and is found glossary information indicating that "26" is a "base" base (7306). Next, a morpheme processing information indicating that "base, base" has a morpheme property, i.e., consists of a sequence of numeric 8702359 81 L r r numbers, is treated as a grouped base (7308). It is judged whether the group that provided the glossary information includes the morpheme processing information or not in step 7308 (7310). If it is then provided, the flowchart 5 proceeds to step 7315 for further processing based on the local analysis regimes while on non-provisioning, the group is recorded in the glossary information retention table 7126 (7312) and the flowchart returns to step 7302. Therefore, "26" becomes , provided with the morph editing information, supplied to step 7314.

The operation in step 7314 is performed in accordance with the flow chart shown in Figure 55B.

First, an initial value "0" is set in the counter n which counts the number of matching glossary reference units when the glossary reference units are used by the match lookup section 7112 (7410). Furthermore, because the glossary reference unit, completed with the glossary search by the glossary lookup section 7016, is set to "p", the match table 7128 is addressed by the p + n_de (n = 0) glossary reference unit, that is, "26" from the matching lookup section 7112 (7412). Because "26" is provided with the morpheme processing information indicating that it is a base in step 7308 and because these combinations "each have a base" are present at the beginning of it and are present after the second entry for combination in the match table 7128 (see also figure 57) 25, the word reference unit "26" becomes equal and thus match is found with the information in the match table 7128. In this case, match is obtained for Ms-Me when setting the second matching configuration as "Ms", where the latest data of the combination with "base" at the beginning is equal to "Me" in the matching table 7128.

Based on the result of the match in the match table 7128 at the p + nth (n = 0) glossary reference unit, the match state is judged (7414) and if the match is decided, the flowchart proceeds to step 7416, 35 while if no agreement is established, the flowchart proceeds to step 7424.

If match is determined then a "1" is set in the counter n to perform an excision by the vocabulary excision unit at p + 1 (n = 1) in the input character array. The cutout is performed in the same manner as in step 7302. The 8702358 * * 82 glossary 7018 is addressed on "Jan.", which is cut as a glossary reference unit from the character array following "26" to provide the morpheme processing information (7420 7422). This operation is performed in the same manner as in steps 7306 and 7308.

By repeating the procedure for steps 7412 through 7422 as above, the flow chart proceeds to "Jan. 26," 80 he ". However, since “he” does not call for a match when looking in the match table 7128 in step 7412, the flowchart proceeds from step 7414 to step 7424. This means that while the data up to “Jan 26, * 80” has been translated into "base, month, year" in the adjustment table 7128, no adjustment is found for "26, Jan., '80 he".

Further, if the sentence in the input character array has ended, for example, with "Jan. 26, '80," that is, if no further excision can be made from a subsequent words list reference unit, then the flow chart proceeds from step 7418. proceed to step 7424.

If no adaptation is recognized in step 7414, it is judged whether the contents of the counter n is not greater than 1 or not (7424) and if it is not greater than 1, it is registered as single words! glossary information retention table 7126 (7434).

If it is not less than 1, then the adjustment procedure 25 is performed with p + n (n = 3) ie "he" in "Jan 26, '80 he" is considered "end of sentence" with which the end of the phrase is indicated (7426, 7428). If no match is found, the flowchart proceeds to step 7434. If match is found, "Jan 26, '80", which is p- (p + nl) for the 30 words list reference unit, is merged according to the result. of the concatenation corresponding to Ms in the match table 7128 and the result is recorded in the glossary information retention table 7126 (7430).

Thereafter, it is assumed that the words ice reference units have all been treated to the (p + 1-n) -th unit and (p + n-1) is reset to "p" (8432).

In the following, the eighth embodiment of the present invention will be explained.

Figure 63 illustrates the entire structure of the eighth embodiment of the language analyzer according to the present invention; & 4 *> 1 if 83 thing applied to a device for automatic translation from English to Japanese.

In Figure 63 are shown the input section 8001, the English text 8002, the pre-editing section 8003, the morphological analysis section 8004, the structure analysis section I 8005, a sentence structure analysis section II 8006, the operation display section 8007, the glossary 8008, the analysis rule file 8009, the control section. 8010, the structure transform section 8011, the translated sentence generating section 8012, the post-digest section 8013, the output section 8014 and the Japanese sentence 8015. The present translation apparatus, as illustrated in the figure, has the input section 8001 through which the English text 8002 to be translated into Japanese is entered. Input section 8001 may include, for example, a keyboard with character keys such as alphanumeric keys and function keys, an optical character reader (OCR reader) for reading English text and / or a file memory device for reading English text registered on a memory medium such as a magnetic disk medium.

The text input through the input section 8001 is read into the pre-editing section 8003, where translation pretreatment 20 is performed. In this case, sentence recognition and processing of unknown words are mainly performed. These functions are part of the morphological analysis.

The pre-edited English data, along with information obtained in the pre-editing, is sent to the morphological analysis section 8004. The section 8004 distributes the data by reference to the glossary 8008, analyzes the English morphemes, performs various types of classifications such as editing of unknown words , expressions of own names, time expressions and numbers, and performs operations on the whole sentence such as additive interrogation and recognition of standard expressions. The analysis rules are stored in the analysis rule file 8009.

After submission to the morphological analysis, the English data is transferred along with glossary information obtained during the morpholotic analysis to the structure analysis section I 8005. The structure analysis section I 8005 is a functional section that decomposes the surface layer structure of the sentence by loosening a grammatical line leave on the English data and look up all structural possibilities.

After submission to the analysis procedure in the structure analysis section I 8005, the English data together with the analysis information 8 7 o 2 3 5 ê 84. » * if $ sent to structure analysis section II 8006, where a solution is selected from the result of the surface structure analysis! through the structure analysis section I by applying a syntax analysis. In this way, a plausible analysis structure 5 of the English description is prepared to build its structure. These analysis rules are also stored in the analysis rule file 8009.

The English data subjected to the structure analysis is fed as data for the analysis structure to a structure transformation section 8011. The structure transformation section 8011 prepares a corresponding Japanese analysis structure which forms an intermediate structure of the English sentence in order to convert it into a Japanese underlying structure of from which the Japanese sentence can be easily translated.

The data for the analysis structure, indicating the Japanese underlying structure and transformed in this manner, is sent to a translation generating section 8012 in which the translated sentence is generated. This is a functional section for generating a Japanese sentence from the Japanese analysis structure of the sentence.

The Japanese phrase data thus translated, that is, the data for the translated sentence, is sent to the post-editing section 8013. The post-editing section 8013 modifies the translated data by a glossary 8008 look-up process using the information provided. was used in the translation procedure to obtain a more natural Japanese sense. The Japanese sentence data is transferred to output section 8014 and output as translated Japanese sentence 8015 through output section 8014. The output section 8014 includes, for example, a printing unit, a display unit and / or a file memory device such as a magnetic disk memory device.

The flow chart of the series of translation operations is controlled by a control section 8010 which controls the entire device. The glossary 8008, in the case of the illustrated embodiment, contains glossary data for the English and Japanese 35 words, in which, in addition to the actual vocabulary, various information is present such as interrelation, i.e. words that occur together, meanings, singular or multiple forms, phrases, etc. Furthermore, the analysis rule file 8009 contains line data for the morphological analysis and for the syntax analysis of the English sentence.

40 The control section 8010 is connected to the operation display section 8702358 85 <J.

T t 8007. The operation display section 8007 includes operation keys for providing various instructions from an operator to the present device, for example, a translation instruction key, a cursor key, etc., a display unit or indicator unit with which the input is entered. English sentence, the Japanese sentence as a result of the translation, intermediate data such as glossary information and various instructions can be made visible to the operator. The device may also be configured such that the majority of these operation display functions are realized in a keyboard which is present in the input section 8001 or at the display unit located at the output section 8014.

The embodiment of the present invention, as indicated above, is used as an automatic translation device and is designed such that if a derived word is present in the English text 8002, a grammatical type, a semantic type, a translated word, etc. are estimated depending on the circumstances under which that derivative word is recognized as such by a prefix or the like, thereby increasing the reliability of the translation analysis result obtained. A prefix dictionary is used to estimate the glossary information for unknown words in the morphological analysis. Three types of processing are required, i.e. an operations for the prefix, the prefix for the suffix, and an estimated processing of the suffix. However, the type of data essentially comprises two types, i.e. prefix estimate data and suffix estimate data.

First, an explanation will be given of the prefix estimate data and suffix estimate data described above.

(1) Prefix estimate data.

If the prefix portion of a word not registered in the glossary matches what is described below and the remainder is contained in the glossary, the word is treated in accordance with its stem in the glossary. It is possible to add an internal attribute in the dictionary to the series of original internal attributes and to add a Japanese prefix to the original translated word. For example, for the input word "electrochemical", an entry in the words list information retention table "electrochemical" may be generated in accordance with the glossary data for the prefix "electro" and for the word list entry "chemical" 8702358 86 y ♦ * ï as in the the following is shown: prefix input word vocabulary input data electro chemical 5 input word input data in the glossary information retention table electrochemical electrochemical

It is noted that the entry in the words ice information retention table inherits all the words ice information of which the word stem is similar.

(2) Suffix estimate data

If the suffix portion of a word not registered in the glossary matches what is described below and the remaining portion is contained in the glossary, then this word is registered by providing a new glossary entry according to the information described in the suffix glossary . In this case, the first translated word in the words! Ice data, corresponding to the portion of the word retrieved and used as the translated word in the new word data.

For example, for the input word "controler", an input data "controler" is registered in the words! List information retention table based on the suffix input data "- (e) r" and on the word "entry" data entry "control".

Suffix given words] ice start data (verb) - (e) r control sift noun verb 30 input word input data in the words! Ice information retention table controler check noun 35 Next, an explanation will be given of the processing of a derivative and the processing from an unknown word.

(1) For a word not registered in the word list, if a prefix or suffix is present at the beginning or end of a word and if the remainder of the word 40 is registered in the word list, the English phrase information will be 8702359 87 t-t T 7 internal properties and the Japanese translated word synthesized based on the glossary information and the suffix information.

(2) The prefix and suffix are resp. listed in a series and can be edited independently of a program.

5 (3) First, possibilities for the prefix are tried and if this fails, the possibilities for the suffix are tried. If both are present, no attempt is made.

(4) If a word is unsuccessful in an estimation attempt 10, processing of the end portion is performed as an unknown word.

More specifically, the following will explain the eighth embodiment with reference to the figures.

Figure 58 is a block diagram for explaining an embodiment 15 of the present invention. In the figure, the input processing section 8020, the unit cutting section 8021, the limiter table 8022, the derivative processing section 8023, the reference words 1 list 8024 and the glossary information retention table 8025 are shown. First, an English sentence is read in from the input 20 input unit comprising an input document file, a keyboard, an OCR device, etc. Then, the glossary reference units are cut out in the glossary reference unit excision section, each time referring to the limiter table and, if the end is not yet reached, a search is performed in the 25 glossary performed using the reference glossary.

If an entry is found as a result of the search procedure, the search procedure result is recorded in the glossary information retention table, while processing for a derivative is performed if no entry is found.

Figure 59 is a block diagram for explaining an embodiment of the derivative processing using a prefix.

The figure shows the matching section 8030 between the prefix and the prefix dictionary, the prefix dictionary 8031, a glossary lookup section 8032 for word search except the prefix portion, a match section 8033 for the prefix dictionary and the prefix dictionary. phrase in the entry, a glossary information preparation section 8034 based on the prefix estimate, and a derivative processing section 8035 based on the prefix.

40 Figure 60 is a block diagram illustrating an embodiment. 8702359 88. » £ £ ï illustrates the derivative processing by means of the suffix. In the figure, the matching section 8040 between the end portion and the suffix glossary, the suffix words! Ice 8041, a processing section 8042 for completely unknown not registered words, a glossary lookup section 8043 for glossary search on a portion excluding the suffix portion, a matching section 8044 for correspondence between the phrase of the stem and the phrase of the entry in the suffix, a glossary information preparatory processing section 8045 from the suffix estimate, a glossary lookup section 8046 for performing the words list lookup procedure by adding the stem change in the suffix to the portion except the suffix portion, and a processing section 8047 for processing unregistered words based on the suffix estimate. First, a search is made for correspondence between the end part and the suffix dictionary. If no match is found, the word is processed as a completely unknown unregistered word, while if match is found, a word search is performed for the part except the suffix. If no entry is found as a result of the glossary search, then a glossary search is performed in which the stem change in the suffix glossary is added to the part except the suffix. If no entry is found as a result, processing for an unregistered word is performed via the suffix estimate. If, on the other hand, an entry is found, then a match is sought between the phrase of the entry and the phrase of the stem in the suffix dictionary in the same manner as when an entry is present as a result of the glossary search for the part containing exception of the suffix. If an agreement is reached, then the word list information preparation processing is performed by the suffix estimate, while if no agreement is found, the processing for an unregistered word is performed by the suffix estimate.

Figure 61 shows a block diagram showing details of the whole structure obtained by synthesizing the parts of Figures 58, 59 and 60 40 and Figure 62 illustrating details of section 8042 for processing 8702359 89 <·· * T 1 completely unknown unregistered words shown in Figure 61. The processing for completely unregistered words 8042 shown in Figure 61 includes a processing section 8050 for estimated information preparation for a noun and a section 8051 for estimated information preparation for a verb. However, since the parts in Figures 61 and 62 have already been described in more detail, further clarifications thereof will be omitted.

The ninth embodiment of the device according to the invention will now be described.

Figure 64 illustrates the entire structure of the ninth embodiment of the present invention used in an automatic translation device from English to Japanese.

This embodiment has an input section 9010 through which an English text 9012 to be translated into Japanese is input. The input-15 input section 9010 may include a keyboard with character keys such as alphanumeric keys or function keys, an optical character reader (OCR reader) for paper-recorded English text and / or a file memory device for reading English text recorded on a memory medium, such as a magnetic disk memory.

The English text, entered through input section 9010, is read into a pre-editing section 9014 in which translation pretreatment is performed. In this case, mainly sentence recognition and unknown word processing are performed. These functions are part of the morphological analysis.

The pre-edited English data, along with information obtained in the pre-editing, is transferred to a morphological analysis section 9016. The section 9016 divides the sentence using the glossary 9018, parses the English morphemes, performs various classifications such as processing unknown words, proper nouns, time expressions, number expressions, etc. and also performs operations on the entire sentence such as searching for sayings and standard expressions. The morphological analysis rules are contained in the analyzer gel file 9036.

35 The English data subjected to the morphological analysis are transferred, together with the words ice-cream information obtained from the analysis, to an analysis section I 9020. The analysis section I 9020 is in this embodiment a functional section with which the surface layer structure from bottom to top and from left to right is analyzed by applying the cfg grammar rule to the English data 8702 35 9 S '* 90 and detecting all structural possibilities.

The English data will be sent to analysis section II 9022 together with analysis information after analysis In analysis section I 9020. The section selects a solution from the result of the analysis of the surface layer structure in analysis section I by adding a syntax analysis to suit. In this way, a plausible English sentence analysis structure is prepared to form its structure. These analysis lines are also stored in the analysis line file 9036.

10 After submission to the analysis, the English data is fed to the structure transformation section 9024 as analysis structure data. In the structure transformation section 9024, the corresponding analysis structure of the Japanese sentence is prepared from an analysis structure which is an intermediate structure between the English sentence and transformed into a Japanese sentence. underlying structure, from which the Japanese sentence can be easily translated.

The analysis tree data indicating the Japanese underlying structure and transformed in this manner is added to a translation generating section 9026 where-20 is generated in the translated sense. This is a function that generates a Japanese sentence from the Japanese analysis structure. First, a sentence structure is generated in which the analysis structure is changed by changing the order so that agreement is reached with the Japanese language, and then a morpheme generation is performed to generate the translated sentence from top to bottom and from left to right in the analysis structure of the sentence.

The Japanese sentence data generated in this way, that is, the translation data, is sent to a post-digest section 9030. The post-digest section 9030 modifies the translation data by a glossary search process 9018 using the information used during the translation process. to get a more natural Japanese sense. The data for the Japanese sentence is transferred to an output section 9032 and hence, as a translation, the Japanese sentence 9034 is output via the output section 9032. The output section 9032 comprises, for example, a printing unit, display unit and / or a file memory device such as a magnetic disk memory.

A flow chart of the series of translation operations is controlled by a control section 9038 which takes care of the entire control of the device. The word list file 9018 contains word 87.0 2 o o 9 91 X -Λ.

List of entries for English and Japanese words and the analysis rule file 9036 contains line information for the morphological analysis and the syntax analysis in this embodiment.

The control section 9038 is coupled to an operation display section 5 9040. The operation display section 9040 includes operation keys for providing various instructions to an operator of the present device, for example, a translation instruction key or cursor key, and a display screen or indication option that displays entered English text, Japanese sentences. of the translation, intermediate data such as glossary information and various instructions can be made visible to the operator., It may also be designed so that the display screen functions are obtained using the keyboard if it is placed at the input section 9010 or with a display screen that is located at the exit section 9032. 15 Via the analysis section I 9020, the cfg grammar rule is applied to the English sentence in a way from bottom to top and from right to left for the English data after the morphological analysis in order to find all possible solutions. gene for the sentence structure. The solutions are generally understood in the form of an analysis structure. This shows a relationship for words or groups present in each of the sentences that are related in a subordinate or cohesive relationship such as a modifying relationship or a case relationship, for example a subordinate relationship such as parent, child, grandchild, etc. Each of the words or groups is located at the node of the analysis structure.

In this embodiment, prior to the syntax analysis, a distinction is made according to the form and vocabulary of a sentence to assess collective rankings in the sentence structure. In connection with the sentence structure, reference is made to "unit" and "block".

The "unit" is a group of words that form the minimum unit for the translation process, which unit is treated identically to a word in the analysis procedure, and the glossary information for each of the constituent elements contained therein is not used.

A "block" is a structural composition in which the analysis is preferably performed for the internal part rather than the external part, and which is treated in an equivalent manner as a unit with respect to its external part. This could be, for example, an expression, a group, etc. as well as one

87 0 2 3 b S

92 T -f R £ part corresponding to intermediate symbols used in the cfg grammar. Furthermore, there may be a nested structure, i.e. a block may contain a further block in itself. Furthermore, the concept of the block may also include a sentence, paragraph 5, multiple sentences, each of which is considered a block. Processing prefers partial analysis and is referred to hereinafter as "partial analysis", thereby eliminating some redundant from the above-described structural solutions and improving the efficiency of the analysis process resulting in a more plausible analysis result.

Two types are defined for a block in this embodiment. One is a symbol in the cfg line which is referred to as "target" in the present description and which must be placed within the block 15 as a result of the analysis performed on each of the constituent elements, i.e. a symbol representing the structure whether the properties of a block are described. The other type is a symbol in the cfg line referred to as "role", which is added to a block when performing the analysis of the exterior of a block in a phrase, group or phrase in which the block is present, 20 that is, a symbol indicating the relationship between the block and other blocks.

For example, in the English sentence "I said," White house isn't white "", the target is a phrase and the role is a noun (a phrase). Although purpose and role are identical in most cases, they can sometimes differ from each other as in this example.

In the embodiment shown in Figure 64, if the English sentence structural configuration is recognized as a block, then the functional sections for estimating the target and its role are summarized in the structure as shown in Figure 65. From this figure it appears that the structural composition of the English phrase data, pre-edited in the pre-editing section 9014, is distinguished in the morphological analysis section 9016 using the glossary 9018 and the analysis rule file 9036.

35 The glossary 9018 contains words! List information for English words and phrases. For example, as shown in Figure 68, input data has been generated for all variations of each word in this embodiment and all possible information has been developed. For example, as phrase information, a large number of phrase information can be provided as shown in the figure. It will be clear that the 8702355.

93 1 Λ Τ> manner in which glossary 9018 has been compiled is not limited to this one embodiment.

The analysis rule file 9036 contains data for the top position indicating the beginning of a block, the end position indicating the end of it, and block operation information that provides the block with a target and a table role. An example of this is shown in Figure 69. For example, a block starts at ", concatenation" and ends at the end of a sentence. A block is therefore formed starting from the beginning of the sentence 10 until immediately before the concatenation and its purpose is a phrase, while its role is a phrase. Furthermore, another block is formed from the concatenation to the end of the sentence in which both the purpose and the role are equal to a phrase.

Furthermore, a block starts at "relative preposition" and ends at 15 or at the end of the sentence. In this case it is possible to allow a number of end states at an initial state. In case the block is terminated by, a cluster from immediately preceding the relative prefix to the next forms a block in which the target is a phrase and in which the role is an adverb or adjective. This means that the cluster functions as an adverb or adjective sentence. If the end coincides with the end of the sentence, a cluster forms a block from immediately preceding the relative preposition to the end of the sentence, in which the target is a phrase and in which the role is an adverb or adjective. This is in accordance with the conditions for forming a group, phrase or phrase appearing in common modern English sentences. In the figure, a space is indicated by the "" symbol.

It is noted that in the morphological analysis section 9016, the English text, entered from the pre-editing section 9014, is first of all divided into sentences that are considered translation units. In this case, misspellings or unregistered words are detected. The glossary 9018 is addressed for each of the phrase units and the glossary information for each of the constituent parts is retrieved. Various composition modes are used in accordance with these glossary information.

Figure 66 shows a flowchart for the collective composition of a block performed in the morphological analysis section 9016. First, a pointer indicating the readout position for an English sentence is set to the beginning (9100). The starting position does not coincide with the word at the beginning but coincides with the (thinking-

870 2 yyS

i S

94 figurative) end of the immediately preceding sentence. The word cut operation 9101 is performed at that position. As shown in Fig. 67, in a word cut operation 9101, a word is cut out by incrementing the position one place each (9111) except when the end of the sentence is reached (9110) and the glossary 9018 is addressed on this word ( 9112) in order to obtain the word information (9113).

In this manner, if the word information is obtained in the word cut operation 9101, then the table 9036 for the start and end 10 states of the block is addressed to judge whether there is any correspondence with the start position or not (9102). In this manner, steps 9101 and 9102 are repeated until the word correspondence with the home position is detected.

If agreement with the starting position is reached, then the next word and the following words with the required number are retrieved and addressed according to the starting position of the block (9104). In this case, if necessary, the glossary is addressed for each of the words. The position indicator is not moved forward.

If agreement is reached with the start position of the block in step 9104, a word corresponding to the block end position, regardless of the start position, is addressed (9105). Steps 9104-9106 are repeated until a word corresponding to the end position is found. If a word matches the end condition (9106) then a cluster including the word is recognized as a block and the block is written (9107). More specifically, a block is prepared for judging that the block operation condition has been fulfilled in the position where the end condition is first met. Next, with reference to the block-30 edit information table 9036, the position for the word, indicated by the pointer at the position where its progress was stopped in the operation 91Q3, is defined as the start position for the block, and the position of the word satisfies the end condition that first appears next is defined as the end position of the block. At the same time, the purpose and role of the block is also registered.

As a result of such block recognition, if "..., concatenation ..." occurs in an English sentence, for example, as shown in Figure 70, the cluster from the beginning of the sentence to 40 the portion before is recognized as one block, while the job-

8/0 2 3 y S

X 1 »

IV

95b from ", conjunction" to the end of the sentence is recognized as another block. In the figure, () indicates the interior of a block. In this block, both the purpose and the role are a sentence. Furthermore, from the word after the concatenation to the end of the sentence, the cluster forms another block, in which in both cases the purpose and the role are also a sentence.

Alternatively, the cluster can be defined as a block from the join to the end of the sentence. In that case, the goal is a sentence and the role is an adverb.

The block can also be defined starting from the position that does not have ",". Furthermore, a punctuation or the like can be excluded from the analysis objective to find the information contained in the block.

Likewise, "... relative preposition ...", ", relative preposition ..." can be recognized as a block. In this block, the goal is a phrase or phrase and the role is an adverb or adjective.

The block can of course be present in a nested structure.

For example, if the English sentence has a structure "(beginning of sentence) ..., concatenation, ..., relative preposition ... (end of sentence)" 20 as shown in figure 71, for example, the cluster starts from ", concatenation" to the end of the sentence is a block BL1-BL1, in which ", relative preposition ...," is present as another block BL2-BL2.

In this way, in the morphological analysis section 9016, a distinction is made between the characteristic of the sentence with respect to the form and the vocabulary to be able to distinguish the structural composition as a block. In addition to such block recognition, the morphological analysis section 9016 also performs various classifications such as expressions for proper nouns, derivatives, unknown words, abbreviated words, numerical expressions, time expressions, abbreviated words, apostrophes 30 ('), as well as searching for words and fixed expressions around the morphological prepare analysis data.

The English sentence subjected to the morphological analysis in this manner is transferred together with the analysis information to the analysis section I 9020. Figure 72 shows an example of the starting data. 35 The figure shows the result of the English sentence "I said," White house is'nt white "" is entered from the input section and decomposed into the morphological analysis section 9016. Block 1 starts at the word position 4 and ends at position 10 , with both the target and the role being optional in this case. Likewise, block 2 starts at position 40 and ends at position 6, in which the target is a group of self-contained 8702; i 96 ί * i> nouns while the role is its own name »ie the block" White house is'nt white "contains in itself another block namely" White house "as nest. In one block, that is say the smaller block "White house" each of the internal assemblies function as a proper noun, while taking the position of noun phrase relative to its environment ie "is'nt white". White house can be treated as a unit.

Along with such block information, word information is retrieved from the glossary 9018 and added and sent from the analysis section 9016 to the analysis section I 9020.

The morphological analysis section I 9020 provides an analysis of the surface layer structure of the English sentence by applying a context-free grammar rule, stored in the analysis rule file 15 9036 to find all possible analysis structures. In this case, if a block is embedded, then the partial analysis described above is performed with preference being given to the local analysis. This can improve the effectiveness and accuracy of the analysis. More specifically, the block inclusion relationship is found from the positional information for the block. The inner block is therefore analyzed. The block completed with the analysis is considered to be unit and its interior is not further processed.

In this way, the analysis area is gradually extended to the outer blocks. Finally, the entire sentence has been analyzed. The analysis is performed on the basis of a cfg grammatical rule in a way from bottom to top and from left to right in the English sense.

The analysis is performed in this way because it maintains all grammatically allowed possibilities.

Figure 73 shows an example of such an analysis processing flow chart. First of all, based on the English data, provided by the analysis section I 9020, all structural clauses for a sentence are recognized as blocks and the purpose and role thereof are estimated (9120). The method of joining is illustrated in figure 70. If after that no block is present in such a composition (9121), then the sentence is analyzed (9125), and only a sentence that is collectively composed as a symbol is selected and the analysis of the sentence is ended (9126). Since operations 9125 and 9126 are included in operations 9121-9124 if the processing system is used as a block for treating the entire sentence, they are therefore unnecessary.

87Q2o: Q

97 α -ι I l

If a block is present, the inner block is first analyzed (9122). In the example shown in Figure 71, the interior of the block BL2-BL2 is analyzed. Although various solutions are generally obtained during the analysis, out of these 5 solutions is selected that solution in which the block is collectively arranged as a cfg symbol and which satisfies the purpose of the block. In this case, all those blocks with a symbol and as an optional target the block are selected. Thereafter, the blocks selected in this way are treated as a single composition with the block (9124) as the roll. In a block with an optional role, the role of the symbol collected in processing 9123 is defined as the role. Operations 9121-9124 are repeated consecutively.

In this way, in the example of Figure 71, the interior of the block BL2-BL2 is first analyzed and then the interior of the BL1-BL1 is analyzed. In this case, the block BL2-BL2 is treated in the same way as a single word, and each of its constituent parts is not analyzed.

When the data defining the structural configuration and the interdependent relationships are obtained in this way, they are sent to the analysis section II 9022. The data can be easily recognized in the form of an analysis structure in the manner described above. The data is further transformed into the Japanese sentence structure in the structure transform section 9024, and in the translation generating section 9026, the translated sentence is generated for each of the nodes contained therein. The node operation in the analysis structure is performed from top to bottom and from left to right.

The translated sentence generated in this way is subjected to a post-processing in the post-processing section 9030, visualized in the operation display section 9040 and printed, for example, as a Japanese sentence 9034 in the output section 9032.

In this manner, in accordance with this exemplary embodiment, the English sense of shape and vo-cabular is distinguished to discriminate the structural composition as a block. For the block, the purpose, which may be the analysis result, and the structural role with which the block functions in relation to the outside world are estimated. The surface layer structure of the English sentence is then analyzed by applying a context-free grammatical rule to detect all possible analysis structures at 40. This makes it possible to reduce the number of useless solutions 8702359 * 3 98 and to improve the analysis efficiency as well as to obtain a more reliable analysis result.

Since there are various patterns for adjective expressions, it is difficult to recognize them during the parsing, especially during a contrast-free type analysis. Since, with reference to the above, it is generally difficult to perform adjective recognition after the parsing, an ambiguous translation is inevitable. Furthermore, if a rule were drawn up that would allow recognition, then there would be a risk that no adjective would be recognized as an identical case or a very large number of possible combinations would arise. That is, an unnecessary local analysis is performed between the parts contained in the adjective and other parts.

Taking into account the above, in the present invention, the burden on the processing in the analysis step is mitigated by recognizing the adjective through the characteristic of the sentence taking into account the form or semantic nature of the words. An estimate of the adjective is performed by recognizing the following pattern as a block.

20 For the English sentence structure "* u, relative pronoun, the relative pronoun is recognized by providing the phrase code of the word with a certain code, for example" R ". In this case, the interior surrounded by V is recognized as a block provided that this does not intersect a block or unit indicated in the rationale and that it does not contain an "and" or "or" in the part after the second pronoun relative to the English sentence structure. ", the interior is surrounded by V ' and V considered as a block. The period can be any other symbol that can be used for the end of a sentence.

To perform such an adjective expression estimate, the glossary 9018 is constructed to include the meaning information of the words. The meaning information illustrates the distinction between items, place, person, etc. as shown in figure 74. Also for the block operation conditions, table 9036 is arranged as shown in figure 75 so that the beginning of the block is recognized by " proper name (person), noun (person) "as the initial condition and the beginning of the block is recognized by" proper name (person), article U noun (person) ". Thus, it is possible to estimate the adjective expression from the morpheme properties and the 8702359 J_ i I 1 99 semantic properties without performing an analysis and performing the analysis in accordance with the estimate of the adjectives in the example shown in Figure 64 for other operations.

5 In the English sense, there are groups that carry very special information and are used only in a very limited way. If these are analyzed in the same way as the usual groups, then they are decomposed in the sentence in a completely different way and it is difficult to recover the original nature of the sentence by the analysis. This often results in loss.

The expressions "let's" or "let us" immediately after a punctuation etc. are analyzed as an imperative with the causal verb "let" and should be parsed as a group with an inviting nature "let". Furthermore, "Let" has various applications such as transitive work-word and applications as a noun and is not limited to use in invitations such as those with an auxiliary verb. Therefore, the decomposition must be performed for the respective possibilities to reduce the efficiency. Furthermore, it is difficult to give preference to use as an invitation in the analysis result because there is no difference between use as a causal verb and inviting verb only in the sentence structure itself and it is difficult to distinguish between them as only the sentence structure is known.

The wasteful loss during the analysis can be reduced by excluding "let's" or "let us" immediately after punctuation from the decomposition. Furthermore, by separating these terms from the fundamental word usage, i.e. the causal word usage, the semantic analysis can be easily performed.

If "please", "let's" or "let us" appears at the beginning of a block, then a flag is set for the block information and no unit information is provided for each of these cases. For example, the English phrase "let's go to school" is processed as <go to school> <linked with let's y.

To perform such processing of "let", a section 9200 containing let information is interposed between the morphological analysis section 9016 and the analysis section I 9020 in a modification of this embodiment. Figure 77 illustrates the respective sections together. In this figure, the same elements as those shown in Figure 64 are indicated by the same reference numerals.

Furthermore, the glossary 9018 is configured to store the 40 let information for the word. As shown in fi 8702359 i x 100? Figure 78 provides the Let Information with a "0" for common words, a "1" for "let's" and "let us" and a "2" for "please".

The let information processing section 9200 has a function of receiving the result of the morpholophic analysis together with the English input sentence from the analysis section 9016 and adding the let information as additional information to the word information during the analysis as shown in Fig. 79. In this case, a block is placed before the sentence. In the example shown in the figure, block 0 is equal to (start: 1, end: 10, target: sentence, role: sentence). That is, the block in this example contains a sentence in addition to a clause, group, etc. In this case, the concept of the block also includes a paragraph and the entire sentence, each of which can be considered a block. Furthermore, "the root of a transitive verb (provided with an 's)" is described as a phrase in "let's" for the word information, and the let information equals "1".

As shown in Figure 81, an operation is performed to prepare the block of the sentence 9300 prior to the start of the block's collective assembly at the English input sentence. The subsequent processing may be the same as in the flow-20 scheme shown in Figure 66. For example, in the English sentence "I said," Let's us go to school "", the block becomes 0 (beginning: beginning of the sentence, ending : end of sentence, role: sentence, purpose: sentence) formed.

As shown in Figure 81, in the structural syntax analysis section I 9020, each of the structural compositions is recognized as a block 25 based on the English data provided to them and its purpose and role estimated (9120). If the block is not present in the composition, decomposition is terminated. If blocks are present in the input sentence, the innermost block is analyzed first (9122). Although various solutions are generally obtained during the decomposition, only the collectively arranged solution as a cfg symbol is selected therefrom (9123). The subsequent operations are the same as shown in Figure 73.

Such a let information processing is performed in the let information processing section 9200 in accordance with the processing flowcharts shown as an example in Figs. 83A and 83B. First, a pointer is set at the beginning of the block (9330) to check the word position at the beginning of the block (9331). If the let information equals "0", the 40 pointer is stepped further (9339) to pass to the next word.

If the let information is not "0", the preceding glossary reference unit is checked (9322). If this is not punctuation or if the pointer is not at the beginning, then the pointer is moved step by step (9339) to move on to the next word.

If the preceding glossary reference unit in the check is a punctuation, or if the pointer indicates the beginning, then the innermost block of the layer containing the word is cleared (9333).

If subsequently the let information equals "1" (9334), which means that it is "let's" or "let us" immediately after the punctuation, then the role of the marked block is recognized as "inviting sentence" ( 9336). If the information equals "2" which equals 15 "please", the role of the highlighted block is recognized as a "interrogative phrase" (9335). Thereafter, the target of the marked block is recognized as an imperative (9337) and the word information designated by the pointer is omitted (9338). Then the pointer is moved forward (9339) step by step to move to the next word. Processing is performed up to the word in the last position (9340).

Fig. 80 shows examples of the decomposition result in which such a let information operation has been performed in the example to the above input sentence: "I said," Let's go to school "". The let-in-25 formation processing section 9200 eliminates, when the let information is added to the word information, the word information relating to the let information from the table and the block information is described as "imperative" for the target and "inviting phrase" for the role as shown in Figure 80.

If the vocabulary 9018 is addressed with all words or a number of words associated with a hyphen during the handling of such words in the English sense, if their entries are contained in the vocabulary 9018, the treatment is continued. For a hyphenated word that is not registered in the glossary 9018, the entire portion is treated as an unknown word, for example, as an adjective, this cannot be translated because the glossary information of each of the words, connected via a hyphen, cannot be used. Further, if the entry for the information of each of the hyphenated constituent parts 40 is contained in the glossary 9018 then 8702359 102 -r * * they cannot be ignored. In addition, if the decomposition is performed by decomposing such combinations in each of the constituent parts, then the method of coupling in the words is very versatile.

To solve the above-identified problem, the hyphenated words are parsed as an adjective in the sentence, and a parsing is performed only for the inner portion of the hyphenated word using the constituent parts of the 10 words connected by a hyphen and the result thereof is combined. This makes it possible to perform the hyphenated word parsing while using the information from each of its constituent parts. That is, for hyphenated words not registered in the glossary 909080, the entire portion is treated in the same manner as an adjective. The hyphenated nouns are referred to the dictionary and the parsing is performed in a closed form only for the inner part in the hyphenated words.

That is, if a hyphenated word is not registered in the vocabulary 9018, the block information related to the entire portion is broadcast as a block and the vocabulary reference is output for each of its constituent parts within the block to retrieve respective unit information without including the hyphen.

For the unregistered words in the words! Ice reference, an end estimate processing is performed as an unknown word processing.

Such hyphen processing can be performed in the example of the structure shown in Figure 64. In this case, the position of the word in the sentence is not expressed by the number applied to the word, but by the number of characters from the beginning of the sentence, that is, the character number.

Figure 84 illustrates an example of the processing of hyphenated words performed in the morphological analysis section 9016. For the entered English phrase that reads, for example, "The anti-war attitude is her open-door policy." the position pointer is moved in steps to retrieve a word (9135) and access the word list (9353). In this case, the hyphen 40 is not used as the word limiter. If there is an entry 8702358 103 * J- τ »zig (9353), then the word information is read (9359). This is repeated until the end of the sentence.

If no entry is found as a result of the dictionary search (9352) and if it is not a word 5 containing a hyphen (9354) then the word information is read (9359) while if it is a word with a hyphen the block with the hyphen is inscribed 9355. In the hyphen block, the starting position also becomes the starting position of the words connected via the hyphen and the end position also becomes the ending position for the words connected via the hyphen. The objective is optional and the role is that of adjective / noun. Thereafter, the hyphen is removed to obtain each of the constituent words (9356) separately and the respective constituent words address the glossary (9357). 15 The information found as a result of the glossary search (9358) is recorded. In the case of output of word information in steps 9359 and 9358, it is processed as phrase = a word not registered in the word list in the case of a word not registered in the word list.

Figure 85 shows examples of the block information and word information of the English block that is present together in the block as a result of the processing of the English input sentence in this example. In this example, the hyphenated words "anti-war" are registered in the glossary 9018 and the words "open-door" are not registered in the glossary. Therefore, the input for the hyphenated words "anti-war" is recorded as information for the word. However, the hyphenated words "open-door" are resolved as "open" and "by" and written as information for the words themselves and block 1 (start: 30, ein-30 de: 38, target: optional, role : adjective / noun) is included as information for the block.

Although the form of the English additive interrogation is very limited, its processing is very complicated in the conventional decomposition process. Furthermore, it is not easy to determine the verb to which the additive interrogation refers.

After recognizing that there is an additive interrogation based on the properties of the sentence with respect to its shape, taking into account the above, the sentence is treated as an information related to the structural configuration associated with it. which allows the verb that was central to the additive interrogation to be specified. That is, the portion of the additive interrogation in the English sense is found as a structural pattern and the decomposition is performed while the portion of the additive interrogation is considered to be lesser information with a particular type of associated structural arrangement.

In the present embodiment, a unit or block is described in the form of a symbol (a starting point indicating that it is an end point of a unit or block).

10 In the morphological analysis, the input sentence text is transformed and the recognition of the block is also performed. In the present embodiment, "a quotation mark" is designated "Q" and a parenthesis is designated "P". The following formats are used: for "..." is written / (Q1 ... /) 1, 15 for "..." is written / (Q "... /)", for (...) is written (/(P.../)}, for ... is written /(P.../), for ... is written /(P.../) and for [...] is written written [/(P.../)].

Block recognition is performed in the same way.

The opening symbol and the ending symbol of a block are only applied in the context in which the block is opened or closed by these symbols. The part immediately before the start signal and immediately after the end signal must be different from the alphanumeric symbol. The above symbols that do not correspond to it are treated as mere symbols. The blocks can sometimes be nested, provided they don't cross.

If, when processing the additive queries, the following phrases follow at a time when the pointer is at 30, then the cluster up to u? U is omitted as a unit and a block flag is set. That is, the form of the additive interrogative sentence contains: ", (auxiliary verb) + (personal pronoun)" ", (auxiliary verb) n't + (personal pronoun" 35 ", (auxiliary verb) + (personal pronoun) + not ".

Other types of auxiliary verbs include: am, is, are, was, were, do, does, did, have, has, had, will, shall, would, should, can, cannot, could, may, might, must, ought , won't, shan't, need, dare, used. The types of personal pronouns or pronouns include -40 ten I, you, he, she, it, we they.

8702 * * 1 »105

These are used as information for the innermost layer of the block to which they belong. In the English sense, for example; you said so, didn't you1? the whole part is recognized as a block seen in the structural composition in [you said so,] <with additive 5 interrogation>. Similarly, in the English sense: I said, "you said so didn't you?", The quoted question "you said so didn't you?" recognized as a block 1 with respect to the structural configuration, and further, the entire portion is recognized as a block 2 with respect to the structural configuration. That is, [I said, [you 10 said so,] <linked to the additive questioning>.

The abbreviated word such as "didn't" is treated after the development of a fully spelled form in accordance with a predetermined table. For words with a large number of abbreviated shapes, all these shapes are output.

To perform such an additive interrogation processing, an additive interrogation processing section 9210 is interposed between the morphological analysis section 9016 and the syntax analysis section I 9020 in another modification of the present embodiment. Figure 87 illustrates these respective sections together.

In this figure, elements identical to those shown in figure 64 are designated by the same reference numerals.

The additive interrogation processing section 9210 receives the result of the morphological analysis along with the input English sentence from the analysis section 9016 and, as shown in Fig. 88, 25, composes a block for the sentence. In the examples shown in the figure, block 0 is equal to (start: 1, end: 12, target: sentence, role: sentence). In this case, the word is represented by the number of the word in this modified embodiment. For example, in this embodiment, the block includes a sentence next to the clause and the group. In this case, the concept of the word also includes a paragraph and the entire sentence, each of which can be considered a block.

The collective composition of the English input sentence block including the additive interrogation may be the same as in the above flowchart of Figure 81. That is, the sentence block processing 9300 is prepared. performed before starting the operation. For example, in the English sentence: I said, "it is good, isn't it", block 0 (start: start of sentence, end: end of sentence, role: sentence, goal: sentence) is formed.

In the syntax analysis section I 9020, the analysis is performed according to the same flow chart shown in Figure 92.

8702358 w 106

The analysis in the additive interrogation processing section 9210 is explained with reference to Figures 90A and 90B. First, a pointer is set on the word at the beginning of the word information (9370). If it is not a comma, the pointer continues in step 5 (9384) and this is repeated until the end of the sentence (9371). It is then checked whether the word following the comma is a word that belongs to the α group or a word that belongs to the 3 group, where the pointer remains where it stood (9373, 9379). It is defined in this case that those words, which contain an auxiliary verb or are a verb in the phrase and do not appear in the negative form, are words that belong to the α group, while those words that contain the negative form of a auxiliary verb or the negative form of a verb in the phrase contain words that belong to the β group. If the word does not belong to either group, then the pointer continues step by step (9484) and the procedures are repeated until the end of the sentence (9371).

In case the word belongs to the α group, the pointer advancing step 9384 is performed if the word following the words of the α group is not a pronoun. If it is a pronoun 20 then it is checked whether the next word is "not" or not (9375) and if it is not "not" it is examined whether the word following the pronoun is a question mark or not (9377). If it is not a question mark, then the pointer advance step 9384 is performed. If it is a question mark, the target is rewritten as "negative phrase" and the role is changed to "additive interrogative phrase" for the innermost block (9378), and the is removed from the word information table (9383 The innermost block means that block that meets the condition: start position ^ (position for ",") and also meets the condition: end position> (position for 30 "7") for the block position and with the minimum of ( end position - start position).

If the word following the pronoun equals "not" in step 9375, it is checked whether the word following "not" is a question mark or not (9376). If it is not a question mark, then the pointer advance step 9384 is performed. If it is a question mark, the target for the inner block is rewritten to "affirmative phrase" while the role is changed to "additive interrogative phrase" (9382), and is removed from the word information table (9383).

40 If, in step 9379, the word following the comma is a word that S / 0 Z oo9 107 4 * r ι · belongs to the β group, then the pointer advance step 9384 is performed if the word following the β group is not a pronoun. If it is a pronoun, then it is examined to determine if the subsequent word is a question mark (9381) or not and the pointer advance step 9384 is performed if it is not a question mark. If it is a question mark, the target of the innermost block is rewritten to "affirmative phrase", while the role is changed to additive question mark "(9382), and is removed from the word information table (9383). Then the pointer 10 is incremented moved forward (9384) and the procedure is repeated until the end of the sentence.

For example, Figure 88 shows the information for the block and for the words obtained from the morphological analysis section 9016 to the add-on interrogation processing section 9210 for the English sentence: I said, "It 15 is good, isn't it" which was described above described. The block formation for block 1 is (start: 4, end: 12, target: optional, role: optional). If this block is subjected to an additive polling operation in the additive polling processing section 9210, the information for the block 1 is rewritten in (start: 4, end: 12, target: confirmatory sentence, role: additive interrogative sentence) , and at the same time, the word information related to the additive interrogation 8-11 is removed.

0702358

Claims (30)

If the separating indication is present in the words found, then the words ice reference unit with which the said distinguishing indication has been found is combined with a words ice reference unit which is present near the said words ice unit and with which another distinguishing indication is found, the numeric values represented by the two words list reference units are calculated together into a single numeric value and the words list reference units are converted into a single parsing unit. A language analysis device, comprising: a word list memory in which word list data is stored including morpheme data for words, compound words and sentences, and a parser for performing a morphological analysis for an entered sentence with reference to the said phrase. words! ice memory, wherein said words ice memory contains data indicative of the degree of coupling between each of the words belonging to compound words or sentences and said parser refers to the word memory for the respective words contained in the entered sentence and, if a number of glossary data is found for a word that combines with other words, then the combination of words with a higher degree of coupling is selected by referring to the degree of coupling data. 2. Language analysis device as claimed in claim 1, wherein if the degree of coupling data indicates a high degree of coupling for a number of combinations between a word taken out of the word list and other words, the parser produces preferred data with a given preference for one of the said number of combinations according to a certain rule. 3. Language analyzer, comprising: 25 input means for inputting a character array in a predetermined language, a basic glossary memory used to search for the input character array and storing basic data, and a parser for analyzing of the input character array by searching the basic dictionary where said parser addresses the basic dictionary memory using the input character array and if thereby a portion of said character array is found, the basic dictionary memory is similarly addressed with other parts of said character array to thereby analyze the entire character array. The language analysis device according to claim 3, wherein the basic vocabulary memory is provided with a fundamental unit 40 vocabulary memory containing data stored therein with which dimension 87 0. V le units are expressed and said parsing device accesses the fundamental unit glossary memory using said input character array to thereby determine by parsing whether said character array expresses a dimensional unit or not. 5. Language analysis device according to claim 4, wherein the parsing device accesses the fundamental unit of vocabulary memory using the character array and, if said character array consists exclusively of a combination of character arrays stored in the fundamental unit of vocabulary memory with which a unit is expressed then, as a result of the search procedure, the character array is judged as a unit expression. 6. Language analysis device according to any one of claims 3 to 5, wherein the parsing device comprises a pointer, which pointer is set at a character at the beginning of the input character array and, if a part of the character array is found when looking up in said basic glossary memory using said character array starting with the character with the pointer set, then the pointer is set to a character array following the portion of the recovered character array, and said basic dictionary memory is addressed again using the following character array, with the pointer now set. The language analysis device according to any one of claims 3 to 6, wherein the input character array is searched in the usual word list memory and not in the basic word list memory. 8. Language analyzer comprising: a glossary memory in which glossary data is stored for each word list reference unit, and a parsing memory for dividing an input phrase into word list reference units and performing a morphological analysis of said words list reference units. wherein reference is made to the vocabulary memory, said vocabulary memory being provided with the vocabulary data with distinguishing indicia indicating whether the vocabulary reference units represent numbers and said parser referring to the vocabulary memory using the respective vocabulary reference units 40 contained in the entered sentence and if the aforementioned 8702359 The language analyzer of claim 8, wherein the parser combines said parser, if accompanied by a word reference unit that expresses a money value symbol or a dimensional unit, together with the numerical value into a single parser. 10. Language analysis device comprising: input means for inputting a character array in a predetermined language, a word list memory which is used for looking up the said input character array, search means for searching the said words list memory by means of the input character array, and type information providing means for providing type information in a character array not registered in the word list memory and in a character array whose type information is not recorded in said word list memory, but we! belongs to the input character arrays, the type information providing means providing a number of type information to said character array which does not have type information. 11. A language analyzer according to claim 10, wherein said analyzer further comprises: unit excision means for dividing the input character array into glossary search units, and parsing means for analyzing the character array provided by the unit excision means is divided and searched by means of search means, together with the preceding character array. 12. Language analysis device according to claim 11, wherein if type information is present in said character array or in the character array preceding it and no type information is present in the other character array, the parsing means comprises said character array without type information provided with the type information of the other 8702359 ί I Γ ΐ character array. Language analyzer according to claim 11, wherein, if character information is present in each character array and in the preceding character array, the parsing means provides a common type information as well as type information to both the character array and the preceding character array. A language analysis device according to any of claims 2 to 4, wherein the character array and the preceding character array, analyzed by said parsing means, are proper names. 15. Language analysis device according to claim 11, wherein if the character array starts with a capital letter and the preceding character array is at the end of a sentence, the parsing means convert the upper letter at the beginning of said character array to a lower case letter and then again searching the word list memory by means of the search means and, if no record is found in the word list memory during said search process, said character array is analyzed as an unregistered proper name. 16. Language analyzer comprising: a glossary memory in which glossary data is stored for all glossary reference units, and parsing means for dividing an input phrase into glossary reference units and performing a morphological analysis by searching the glossary memory using the glossary reference units wherein the glossary data in the glossary memory includes distinguishing information specifying the position of a glossary reference unit representing a proper name if a plurality of proper names can occur in a consecutive sequence of proper names, and searches the parsing means in the glossary reference units, respectively , present in the entered sentence and, if the distinguishing information is contained in the found glossary data, the glossary reference unit with which the distinctive information is found is combined with the glossary reference unit which is directly adjacent to said glossary reference unit and which has other than proper name meaning in a single parsing unit according to the position specified by the distinguishing information. 8702355 τ V * * 17. Language Analysis Device comprising: input means for inputting a character array in a predetermined language, a glossary memory used for looking up said character array input through the input means, and type information parsing means searching in the dictionary memory with parse using the input character array and the type information of said character array, the type information parsing means decomposing the type information of the character array taking into account the type information of the character arrays before and after said character array. The language analysis device of claim 17, wherein the type information parsing means parses the type information for a number of character arrays by collective combination of said number of character arrays. The language analysis device according to claim 17, wherein if type information is present for said character array or for the character array preceding this character array and the type information is not present in the other array, the type information is The character array which is not provided with type information provided with the type information of the other character array. The language analysis device according to any of claims 17-19, wherein the character array parsed by said type information parsing means is a proper name. 21. Language analyzer comprising: a glossary memory in which glossary data for all glossary reference units are stored, and parsing means for dividing an input phrase into glossary reference units and performing a morphological analysis for said glossary reference units by searching said glossary memory wherein said parsing means distinguish that a series of glossary reference units having a specific semantic element constitutes a composite unit having a specific meaning associated with a given rule and converts a sequence of glossary reference units with said specific semantic element into a single parsing unit. The language analyzer according to claim 21, wherein the vocabulary memory contains data for distinguishing 40 vocabulary reference units with a particular semantic element, said70 pars having a discrimination matching table to distinguish that a series of glossary reference units with said specific semantic element constitutes a composite unit expressing a specific meaning under certain rules, and searches the parsers in said word list memory using the respective words list reference units from the input sentence, and converting a set of glossary reference units with said specific semantic element into a single parsing unit in accordance with the discrimination agreement table if the glossary reference units are distinguished by said specific sem antic element. 23. Language analysis device in which the grammatical nature, the semantic nature or the translated word is estimated as being a word not registered in the glossary, which is recognized for its morpheme properties as a derivative such as a suffix. 24. Language analyzer, comprising: a first parser for performing a morphological analysis on an entered sentence in a predetermined language, a second parser for performing an analysis of the sentence in said language based on the result of the morphological analysis in the first named parser, glossary memory in which wordlist data of said language is used which is used for the analysis by said first and second parser means, and control means for searching the said dictionary and for executing the parsing procedure in the first and second parsing means, wherein said first parsing means searches in the vocabulary memories, distinguishes the structural arrangement by distinguishing the characteristics of the phrase entered in said language with respect to its form and estimating the attitude that may be the result of the o decomposition and of the structural role of said arrangement which functions in said sentence, and said second parsing agent analyzes the surface structure of the sentence in said language using a grammatical rule based on the estimated posture and role, and analyzes the possible subordinate relationship of the constituent parts of the said sentence. The language analysis device of claim 24, wherein the two-way if the parsing means performs the analysis of the ranking with preference over others if the ranking is present in the sense in said language. 26. A language analysis device according to claim 24, wherein the first parser distinguishes the properties of the sentence in the given language with respect to its shape and meaning and estimates its adjective expression based on the distinguished properties. The language analyzer according to claim 24, wherein the dialing language is the English language and said glossary memory is provided with word list data, distinguishing information distinguishing predetermined words containing "let's and let us", and said first parser, if searching the said glossary memory to obtain the said distinguishing information, exclude "let's" from the purpose of the analysis in the second parser and "let us" exclude from the purpose of the analysis in said second parser if punctuation is present in the previous section. The language analysis device of claim 27, wherein the first parsing means estimates the attitude as an instruction and the role as an invitation to the ranking contained in the excluded portion. 29. A language analyzer according to claim 24, wherein the first parser, if no glossary data is obtained in the glossary memory search for a number of hyphenated words, searches the glossary for each of the number of words, estimates all words, consider a configuration, and estimate the attitude of said configuration as an adjective group. 30. A language analysis device according to claim 24, wherein the predetermined language is the English language and the first parser distinguishes the properties of the English sentence in its form, distinguishes the additive interrogator based on said distinguished property, a 35 estimates for the whole of the said additive interrogation group considered as a configuration whose attitude is that of an additive interrogation group and excludes it from the analysis in the second parser. ********* 8702359