JP3393173B2 - Machine translation equipment - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translation device for translating a conversational sentence in a first language into a second language by utilizing a pre-registered conversational example sentence.

[0002]

2. Description of the Related Art Nowadays, when there are many opportunities to go abroad for sightseeing and the like, many conversational example books that collect many conversational expressions have been published so as not to bother with conversations in foreign countries. Also,
A machine translation apparatus has been proposed which digitizes these conversational example collections, selects conversational example sentences by a key operation or the like, and displays the translated sentences of the selected conversational example sentences. For example, JP-A-6
The invention of No. 325072 corresponds to that.

Further, in this machine translation device, a variable portion is provided in advance for conversational example sentences, and the same example sentences are used even if some words in the input free sentences are replaced.
You can increase the variation of example sentences while saving memory.

Further, there has been proposed a machine translation device for automatically selecting an optimum conversational example sentence by freely inputting a sentence to be translated and displaying the translated sentence. This machine translation device also has a variable part, which automatically generates a sentence close to the input sentence by automatically replacing it with the input phrase.
It is configured to output.

According to this method, a large number of conversational example sentences and phrases that characterize each example sentence are grouped into synonyms and stored. The former is called a fixed sentence and the latter is called a semantic feature. In addition, variable semantic features are called variable semantic features.

In this method, semantic features are extracted from an input sentence, and a fixed sentence having a pattern close to the combination thereof is searched. At that time, it is necessary to determine how much the input sentence and each fixed sentence match as a pattern of a combination of semantic features.

In the conventional technique, the degree of coincidence between the input sentence and each fixed sentence is calculated by matching each word unit, and only the fixed sentence above a certain threshold is displayed.

[0008]

However, when the input sentence and a certain fixed sentence do not match only the variable semantic features and all other parts of the sentence match, the matching degree is calculated by the conventional method. In some cases, the degree of coincidence does not exceed the threshold value and the machine intuitively feels that the meaning is close, but the machine does not display.

As an example, it is assumed that there is an input of "I was disconnected from the tour operator". On the other hand, it is assumed that there is no fixed sentence with the exact meaning of the sentence, and the fixed phrase "I was separated from a friend" was the closest one. Here, it is assumed that the part corresponding to “friend” is a variable semantic feature, and the feature “tourist” is not registered.

According to the present invention, when the sentence constituents other than the variable semantic features are almost the same and the variable semantic features are not the same, the word and its translation in the word dictionary built into the device are used. , By combining the bilingual sentence, the fixed sentence and the translated sentence whose variable semantic features do not match are displayed. In this way, when the word dictionary is used to replace the variable semantic feature of a fixed sentence, it is necessary to find out which part of the sentence constituent element corresponds to the variable semantic feature.

[0011] As an example, "I missed the tour with the tour operator"
Suppose there is an input sentence such as. On the other hand, it is assumed that there is no fixed sentence with the exact meaning of the sentence, and the fixed phrase "I was separated from a friend" was the closest one. here,
It is assumed that the part corresponding to "friend" is a variable semantic feature, and the feature "tourist" is not registered. At this time, it is necessary to search for a partial character string that can be replaced with “friend” from the character string “at the attendant and the station” that does not match the fixed phrase.

According to the present invention, when a word dictionary is used to replace variable semantic features of a fixed sentence, a set of connectable character strings is defined immediately after the variable semantic feature, so that a variable meaning can be changed from a sentence constituent element. Find the part that corresponds to the feature.

Further, even if a set of character strings that can be connected immediately after the variable semantic feature is used to narrow down the sentence constituent elements corresponding to the variable semantic attribute, the word dictionary is still included in the character string that does not match the fixed sentence. There may be multiple registered words.

As an example, it is assumed that there is an input sentence "Do you have a thigh?" On the other hand, it is assumed that there is no fixed phrase with the exact sentence meaning, and the fixed phrase "Do you have an apple?" Was the closest sentence. Here, it is assumed that the part corresponding to “apple” is a variable semantic feature, and neither the feature “peach” nor the feature “sumomo” is registered. Furthermore, it is assumed that the word "sumomo" and the word "peach" are registered in the word dictionary. At this time, it is necessary to search for a partial character string that can be replaced with "apple" from the character string "Sumomoha" that does not match the fixed phrase.

In the present invention, the word dictionary is used to replace the variable semantic features of a fixed phrase, and when there are a plurality of words registered in the word dictionary in a character string that does not match the fixed phrase,
By selecting the longest character string, the part corresponding to the variable semantic feature is searched for from the sentence constituent elements.

Further, a set of character strings that can be connected immediately after the variable semantic feature is used to narrow down the sentence constituent elements corresponding to the variable semantic attribute, and the length of the character string registered in the word dictionary is the most. Even if you narrow down to long ones, there may be multiple candidates for the part corresponding to the variable semantic feature in the character string that does not match the fixed phrase.

As an example, it is assumed that there is an input sentence "Please carefully pack in a wooden box." On the other hand, it is assumed that there is no fixed phrase with the exact sentence meaning, and the fixed phrase "Please package in a case" was the closest sentence.
Here, it is assumed that the part corresponding to the “case” is a variable semantic feature, and the feature “wooden box” is not registered.

Further, it is assumed that the word "wooden box" and the word "polite" are registered in the word dictionary, and the part of speech of "wooden box" is a noun and the part of speech of "polite" is an adjective verb. . At this time, it is necessary to search for a partial character string that can be replaced with “case” from the character string “polite in a wooden box” that does not match the fixed phrase.

In the present invention, when the word dictionary is used to replace the variable semantic features of a fixed sentence, the part of speech of the variable semantic feature is registered and registered in the word dictionary in a character string that does not match the fixed sentence. When there are a plurality of words, the word corresponding to the variable semantic feature is searched for from the sentence constituent elements by selecting the word having the same part of speech among the words in the word dictionary.

[0020]

[Means for Solving the Problems] <Means for Claim 1> Parallel translation in which a sample sentence of a conversation in a first language, a sentence in a second language that is a translation thereof, and a semantic feature are registered as a pair. The database, the semantic feature dictionary that stores and defines the semantic feature and the notation to be extracted as the semantic feature, the variable semantic feature dictionary that has the same semantic feature and has replaceable words registered, and the bilingual database An input unit for inputting an input sentence for searching; a searching unit for searching the bilingual database, the semantic feature dictionary, and the variable semantic feature dictionary based on the input sentence input from the input unit; Accordingly, with respect to the example sentence retrieved based on the translation database and the semantic feature dictionary, and replacement means for performing a substitution based on the previous SL variable semantic feature dictionary, the search means and replacement means An example sentence extracted by the input unit, an analysis unit for calculating the degree of coincidence with the input sentence input from the input unit, an output unit for outputting a translated sentence of the example sentence based on the analysis unit, and a first language an apparatus and a word dictionary that can be from the word of searching a translated word in the second language,
In the analyzing means, the variable semantic feature in the input sentence
In the corresponding portion, if there is a character string that is not registered in the semantic feature dictionary and the variable semantic feature dictionary as the semantic feature and is registered in the word dictionary, the character string becomes the variable semantic feature. The degree of agreement is calculated as a match,
The replacement means converts the character string into a translated word registered in the word dictionary to solve the above problem.

[0021] The said translation database <means according to claim 2>, which can be connected string is registered immediately following string with each semantic feature in the analysis means, variable semantic feature of the input sentence the portion corresponding to said the semantic feature dictionary Contact and variable semantic feature dictionary is not registered as the semantic feature, wherein the word dictionary when there is Rubun string is registered, of the sense element of When the connectable character string immediately after and the character string immediately after the character string match, the degree of matching is obtained by assuming that the character string matches the variable semantic feature, and the replacing means sets the character string to the word. The above problem can be solved by converting the translated words registered in the dictionary.

<Means of Claim 3> In the analysis means, the portion corresponding to the variable semantic feature in the input sentence is not registered as the semantic feature in the semantic feature dictionary or the variable semantic feature dictionary, If the word dictionary with a character string registered, the word of the string
With respect to the longest character string registered in the dictionary, the degree of coincidence is obtained as a match with the variable semantic feature, and the replacing means converts the character string into a translated word registered in the word dictionary. Solve the problem.

[0023] in the word dictionary <means of claim 4>,
In addition to words and translation of the second language in the first language, and the word part of speech in the first language is registered, the said translation database are part of speech registration of each semantic feature,
In the analyzing means, the variable semantic feature in the input sentence
If there is a character string that is not registered in the semantic feature dictionary or the variable semantic feature dictionary as the semantic feature in the corresponding part , but is registered in the word dictionary, the sentence
Of string, translation of the string word class matches the part of speech of the variable semantic feature, determine the degree of coincidence as the matches to the variable semantic feature in the replacement unit, registered with the character string in the word dictionary The above problem is solved by converting to.

[0024]

DETAILED DESCRIPTION OF THE INVENTION The present invention will be described in detail below with reference to the drawings. In the following example, Japanese is used as the first language and English is used as the second language. It is not particularly limited to these languages.

FIG. 2 shows an appearance of a machine translation device according to an embodiment of the present invention. A large number of Japanese and English sentence examples are stored as a pair, and these can be displayed on the liquid crystal display 2. Further, this apparatus uses handwritten character recognition by the dedicated pen 15 as an interface. A display other than liquid crystal such as CRT can be used, and the input means can be other means such as voice or keyboard input.

FIG. 1 shows an outline of a circuit configuration of such a machine translation device. The machine translation device is a CPU (central processing unit) for performing the central functions of various controls 1
Are arranged. The CPU 1 includes a ROM 6, a working memory 5, a display driver 3, and a bus 18 such as a data bus.
It is connected to each of the input recognition units 4.

The ROM 6 is a large-capacity read-only memory that stores programs for performing various controls of the machine translation device and a large number of phrases. The work memory 5 is a random access memory for temporarily storing data required for executing a program. The display driver 3 is a driver for displaying bilingual sentences on the liquid crystal display 2. The input recognition unit 4 recognizes the input data from the handwriting input board 14 of FIG.

In the ROM 6, the parallel translation database 7, the semantic feature dictionary 8, the variable semantic feature dictionary 9, and the word dictionary 1 are also included.
9 is stored. For example, the parallel translation database 7 has addresses 0 to 50000, and the semantic feature dictionary 8 has addresses 50001 to 7
0000, variable semantic feature dictionary 9 has addresses 70001-10
0000, the word dictionary 19 has addresses 100000 to 3000
It is stored as 00.

The CPU 1 is divided into an analysis unit 10, a search unit 11, a replacement unit 12, and a control unit 13 according to the nature of operation. The analysis unit 10 analyzes the input data recognized by the input recognition unit. The search unit 11 searches the parallel translation database 7, the semantic feature dictionary 8, and the variable semantic feature dictionary 9 stored in the ROM 6. The replacement unit 12 replaces the variable semantic feature element.
The control unit 13 performs control other than the above, such as control of screen display.

An embodiment of the data structure of the bilingual database 7, the semantic feature dictionary 8, the variable semantic feature dictionary 9, and the word dictionary 19 will be described below. These are common to claims 1 to 4 except a part.

The parallel translation database 7 defines a large number of useful sentence examples which are desired to be translated and transmitted to local people when traveling abroad.

Example 1 How much is it to {New York} How much is it to {New York}? How much is {place name} Example 2 How much is it from {New York} to {Chicago} How much is it from {New York } $ 1 $ to {Chicago}
$ 2 $? How much is from {place name} to {place name}?

Here, the first line of the sentence example is the Japanese sentence to be displayed, the second line is its translated sentence, the English sentence displayed together with the Japanese sentence as the translation result, and the third line are associated with this bilingual sentence. It is a set of semantic features. This is called a semantic feature sequence. Also, the part enclosed in {} is a variable semantic feature.

The semantic feature is a character string that serves as a keyword when searching for a sentence example, and is often assigned to each bunsetsu of each Japanese sentence. In sentence example 1, "How much is up to New York?"
"How much" and "is" are the semantic features of this sentence.
To be precise, these are names of semantic features, and what character strings are defined as these semantic features is described in the semantic feature dictionary 8.

Further, "New York" in the sentence example 1 is a variable semantic feature. When various place names such as "London" and "Oxford" are entered in this part of the sentence example 1, the sentences that can be translated are increased, which is very effective. Therefore, the Japanese character strings that can be entered in this part and their translations are defined in the variable semantic feature dictionary 9, and the set is treated as one variable semantic feature called "place name".

In the bilingual database 7, the name of the variable semantic feature is described by enclosing it in {} in the semantic feature column part, and one element thereof is written in {} in the Japanese sentence on the first line. Also, write the translated word in {} in the English sentence on the second line. The Japanese and English texts in {} are defined as parallel translations in the variable semantic feature dictionary 9. These also serve as representative values of the variable semantic features in this example sentence. That is, if there is no character string corresponding to the variable semantic feature in the input, this element is displayed.

For example, sentence example 1 has one variable semantic feature named "place name", the representative value of the variable semantic feature in this sentence is "New York", and its translation is "New York". When there are multiple variable semantic features in a sentence, the order of appearance of variable semantic features in the Japanese sentence and the order of appearance of variable semantic features in the sequence of semantic features match.

Similarly, sentence example 2 has two variable semantic features named "place name", and the representative value of the first "place name" variable semantic feature in this sentence is "New York",
Its translation is "New York." The representative value of the second "place name" variable semantic feature is "Chicago", and its translation is "Chicago".

Also, $ 1 $ and $ 2 $ after the {} symbol in sentence example 2
Indicates that the "place name" variable meaning attribute corresponds to the first and second variable meaning features in Japanese sentences.

That is, the first "place name" variable semantic feature containing "New York" is {New York} $ 1 $,
The second "place name" variable semantic feature containing "Chicago" corresponds to {Chicago} $ 2 $. Moreover, when distinguishing a non-variable semantic feature from a variable semantic feature, it is called an invariant semantic feature.

In the semantic feature dictionary 8, names of semantic features and Japanese character strings to be extracted as the semantic features are defined. For example, "from,""to,""howmuch,""is?"
The meaning feature is defined in the meaning feature dictionary 8 as follows, from [semantic feature dictionary 8] through to how much is it?

This means that the Japanese character string on the right side is defined as the semantic feature of the name on the left side.
For example, the character string “how much” is defined as the semantic feature of the name “how much”.

In the variable semantic feature dictionary 9, names of variable semantic features and Japanese character strings as elements included therein,
The corresponding English translations are defined. For example, the variable meaning feature element “place name” and its set of translations are defined in the variable meaning feature dictionary 9 as follows: [Variable semantic feature dictionary 9] Place name New York Chicago Chicago London London Oxford Oxford .

This defines a Japanese character string on the left side of the second and subsequent lines (such as "New York") as an element of a variable semantic feature called "place name", and translates it into an English character string on the right side of the character string. It means to be defined as (“New York” etc.).

The word dictionary 19 defines Japanese headwords, parts of speech, and English translations. For example, the parts of speech and translations of the words "school" and "tourist" are defined in the word dictionary 19 as follows: [word dictionary 19] school noun school tour operator noun a tour conductor.

This means that the part of speech is the noun and the English translation is school for the Japanese character string "school". The word dictionary does not need to be newly created for the machine translation device, and an existing word dictionary or electronic dictionary may be used.

The operation of this machine translation device will be described below. The following is a description of the processing common to the claims. Embodiments of the individual claims are described below. The outline of the basic operation of this machine translation device is that the user inputs the translated sentence, detects the semantic feature from the sentence, detects the fixed sentence from the semantic feature, and replaces and displays the variable semantic feature in the fixed sentence. To do. Hereafter, the translated sentence input by the user is the input sentence,
Japanese sentences stored in the machine translation device are called standard sentences. When we say a pair of a fixed phrase and its translated sentence in English, we call it a translated sentence.

The operation of this machine translation device will be described with reference to the flowchart of FIG. The description will be given in detail with respect to each step of the flowchart of FIG. Take, for example, the process of translating the sentence "How much is it to London?" (Hereinafter, this sentence is referred to as input sentence 1).

<Step A1> Control unit 13 of CPU 1
Displays the “translated sentence input screen” of FIG. 4 on the liquid crystal display 14 through the display driver 3.

<Step A2> In the translated sentence input screen of FIG. 4, the user manually inputs "How much is it to London?" On the handwriting input board 16 with the dedicated pen 15. FIG. 5 is a diagram in the middle of inputting, and FIG. 6 is a state in which the inputting is completed.

<Step A3> The user presses the interpreter icon 17 in FIG. This input sentence is recognized by the input recognition unit 4 and stored in the work memory 5. Then, it moves to step A4.

<Step A4> Analysis unit 10 of CPU 1
Is the input sentence 1 stored in the work memory 5 in step A3
Then, a partial character string is created and stored in the work memory 5. Then, the process proceeds to step A5.

The partial character strings of the input sentence 1 are as follows. Loron Rondo London London to London: How much: Store these in the working memory 5.

<Step A5> Search unit 11 of CPU 1
However, in step A4, the semantic feature dictionary 8 and the variable semantic feature dictionary 9 stored in the ROM 6 are searched with the partial character string of the input sentence 1 stored in the work memory 5 as a key.
If there is a character string that is defined as a semantic feature in the semantic feature dictionary 8 among the partial character strings, the search unit 11 of the CPU 1 stores the character string and the name of the semantic feature in the working memory 5 as a pair, and step A6 Move on to. If not, the process proceeds to step A16. Also, at this time, substrings are searched in order from the front,
By saving the order, the appearance order in the input sentence of each semantic feature is also saved.

The process of searching the semantic feature dictionary 8 and the variable semantic feature dictionary 9 with the partial character string "how much" of the input sentence 1 as a key will be described below. The search unit 11 of the CPU 1 is a ROM
The semantic feature dictionary 8 stored in 6 is read line by line from the address 50001 and stored in the work memory 5.
On the left side of each line of the semantic feature dictionary 8, the name of each semantic feature is
The character string defined as the semantic feature is described on the right side.

In the semantic feature dictionary 8, [Semantic Feature Dictionary 8] How much It is supposed to be stored in this way.

Now, the character string "How much" is the semantic feature dictionary 8
Since it is defined inside, the retrieval unit 11 of the CPU 1 stores the semantic feature name “How much” on the left side thereof and the partial character string “How much” of the input sentence in the working memory 5 again as a pair.
Since the character string "how much" is not defined elsewhere in the semantic feature dictionary 8, the search of the semantic feature dictionary 8 is completed.

Next, the search unit 11 of the CPU 1 searches the variable semantic feature dictionary 9 with "how much" as a key. Since the character string "how much" does not appear in the variable semantic feature dictionary 9 now, the search using "how much" as a key is completed and the next partial character string is set as a key.

The process of searching the semantic feature dictionary 8 and the variable semantic feature dictionary 9 using "London" as a key will be described below. It is assumed that the character string "London" is now stored in the working memory 5.

The search unit 11 of the CPU 1 checks whether the character string "London" is defined in the semantic feature dictionary 8 stored in the ROM 6. Since the character string "London" does not appear in the semantic feature dictionary 8 now,
Next, the variable semantic feature dictionary 9 is searched.

In the variable semantic feature dictionary 9, [Variable semantic feature dictionary 9] Place name New York Chicago Chicago London London Oxford Oxford Suppose there is such a description.

The character string "London" is the variable semantic feature dictionary 9
Since it is defined in the above, the search unit 11 of the CPU 1 stores the name “place name” of the variable semantic feature including it and the translation “London” of the element in the working memory 5. Since the character string "London" is not defined elsewhere in the variable semantic feature dictionary 9, the search of the variable semantic feature dictionary 9 ends. The above operation is performed for all the partial character strings of the input sentence 1. It is assumed that the following variable / invariant semantic features are detected from the input sentence 1.

[Semantic feature corresponding to substring] "London" = "place name": variable semantic feature "up" = "up": invariant semantic feature "how much" = "how much": invariant semantic feature "is" = “?”: Invariant Semantic Feature Here, the right side of = is the name of the semantic feature, and each character string on the left side means that it is detected as a variable / invariant semantic feature of the name on the right side.

The translated word "London" of the variable semantic feature element "London" is also stored. [Variable Semantic Feature Element] “London” = “London” Here, the right side of = is a translation of the element of the variable semantic feature, and means that the phrase on the right side is detected as a translation of the phrase on the left side.

Further, it is assumed that the respective semantic features detected from the input sentence are detected in the order from the top to the bottom of the above correspondence table. This order is stored in the working memory 5. In this example, since the semantic feature has been searched, the next step A6
Move on to.

<Step A6> Search unit 11 of CPU 1
However, the parallel translation database 7 stored in the ROM 6 is searched using the variable / invariant meaning feature name stored in the work memory 5 in step A5 as a key. If there is a variable / invariant semantic feature name that appears in the bilingual database 7, the fixed sentence in which the semantic feature appears is stored in the working memory 5 together with the English sentence and the semantic feature string of the translated sentence, and the step Move to A7. If not, the process proceeds to step A16. The details of step A6 will be described by taking the input sentence 1 as an example.

First, the CPU 1 searches the parallel translation database 7 with one of the semantic feature sets stored in the work memory 5 in step A5 as a key. First, the parallel translation database 7 is searched using the semantic feature “place name” as a key. Now, in the parallel translation database 7, as shown below, Suppose there is such a description.

In the sentence example 1, the "place name" semantic feature is included in the semantic feature sequence. Therefore, the search unit 11 of the CPU 1 stores the sentence example 1 in the work memory 5. In the parallel translation database 7, the "place name" semantic feature is not included in any other part. Therefore, the search using the "place name" as a key is ended.

The retrieval unit 11 of the CPU 1 performs the above-mentioned work for all the semantic features stored in the work memory 5. Now, assume that the following sentence examples 1 and 3 are retrieved and stored in the work memory 5 after the retrieval for all the semantic features is completed.

[0070] Sentence example 3 is searched by the semantic feature “how much”. The above is the description of step A6 by way of example. In this example, since the fixed phrase is searched, the process proceeds to step A7.

<Step A7> Analysis unit 10 of CPU 1
Selects one sentence stored first in the fixed phrases stored in the work memory 5 in step A6. A flag is set for the selected fixed phrase. Here, the sentence example 1 is selected and a flag is set.

<Step A8> Analysis Unit 10 of CPU 1
Calculates the degree of coincidence between the fixed phrase selected in step A7 or A12 and the input sentence. Here, the calculation is performed using the number of semantic features of the standard sentence that match the semantic features of the input sentence. Then, the pair of the fixed phrase and the degree of coincidence is stored in the work memory 5, and if the degree of coincidence is equal to or higher than a certain threshold value, the process proceeds to step A9, and if it is lower than the threshold, step A1
Move to 3. Here, the threshold is 200 points.

Now, the sentence example 1 is selected. Semantic features of sentence 1 "place name""upto""howmuch""isit?"
Are all in the input sentence. The number of coincidences is multiplied by 100, and the degree of coincidence between this fixed phrase and the input sentence is set to 400 points. Therefore, the sentence example 1 and 400 points of coincidence are stored in the work memory 5. Since the matching score of 400 points in the sentence example 1 exceeds the threshold value of 200 points, the process proceeds to step A9.

<Step A9> Analysis unit 10 of CPU 1
Determines whether the fixed sentence stored in the work memory 5 in step A8 has a variable semantic feature. If yes, go to step A10; if no, go to step A11.
Since sentence example 1 has a variable semantic feature "place name", step A
Go to 10.

<Step A10> Replacement Unit 1 of CPU 1
2 replaces the variable semantic attribute element in the Japanese sentence of the fixed phrase stored in the work memory 5 with the variable semantic feature element in the input sentence 1. Further, the variable semantic feature element in the English sentence is replaced with the translated word of the element stored in the work memory 5. The translated texts are stored in the work memory 5 in the replaced form. afterwards,
Move to step A11. Now, the element “New York” is included in the “place name” variable semantic feature of sentence example 1. on the other hand,
The “place name” variable semantic feature element detected from the input sentence 1 is “London”. At the same time, the translated word "Lo" for "London"
“Ndon” is also stored in the work memory 5. Therefore, C
The replacement unit 12 of PU1 replaces “New York” with “London” and replaces “New York” with “London” in the sentence example 1. It looks like this:

[0076] This form is stored in the working memory 5, and the process proceeds to step A11.

<Step A11> Analysis Unit 1 of CPU 1
0 checks whether or not there is a fixed text in the working memory 5 that has no flag. If so, step A12
If it does not exist, the process proceeds to step A14. Since the example 3 is not flagged, the process moves to step A12.

<Step A12> Analysis Unit 1 of CPU 1
For 0, one fixed phrase in the working memory 5 in which no flag is set is selected and the flag is set. Then, the process proceeds to step A8. Here, the sentence example 3 is selected, a flag is set, and the process proceeds to step A8.

<Step A8> Analysis unit 10 of CPU 1
Calculates the degree of coincidence of sentence example 3. Among the semantic features of sentence example 3, what is in the input sentence is "how much".
It is two of "is?" The remaining "today", "exchange", and "rate" do not exist in the input sentence. Further, the sentence example 3 does not have two semantic features of “place name” and “up to” in the input sentence. Therefore, -50 points, which is obtained by multiplying the number of matching 2 by 100 and the number of non-matching 5 by 50, is set as -50, which is the degree of matching with the input sentence of this fixed phrase.
Therefore, the sentence example 2 and the degree of coincidence of −50 points are stored in the working memory 5. Since the matching score of −50 in sentence example 3 is lower than the threshold of 200, the process moves to step A13.

<Step A13> Analysis Unit 1 of CPU 1
In 0, the fixed sentence stored in the work memory 5 in step A8 and having the degree of coincidence less than the threshold is deleted from the work memory 5. After that, the process proceeds to step A11. Now, the sentence example 3 whose degree of coincidence is −50 is deleted from the working memory 5,
Move to step A11.

<Step A11> Analysis Unit 1 of CPU 1
0 senses that there is no fixed phrase in the working memory 5 that has not been flagged yet, and proceeds to step A14.

<Step A14> Controller 1 of CPU 1
3 is a liquid crystal display 14 through the display driver 3.
Is switched from the free text interpreter input screen of FIG. 6 to the parallel text display screen of FIG. Then, it moves to step A15.

<Step A15> Controller 1 of CPU 1
Reference numeral 3 is a liquid crystal display 14 for displaying Japanese and English sentence examples stored in the work memory 5 through the display driver 3.
Are displayed in parallel as shown in FIG. At this time, the variable semantic features are underlined and displayed. In addition, when displaying, fixed phrases are displayed in descending order of matching. At present, only the sentence example 1 is stored in the working memory 5. The Japanese and English sentences are displayed as shown in FIG.

The above is the explanation of the operation of the machine translation device taking the input sentence 1 as an example. Next, Figure 3 which was not traced above
The steps in the flowchart will be described. If no semantic feature or standard sentence is found in step A5 or A6, it is determined that the translation is impossible and that fact is displayed. The process is step A16 below.

<Step A16> Control unit 1 of CPU 1
3 is a liquid crystal display 14 through the display driver 3.
Is switched from the free text interpreter input screen of FIG. 6 to the untranslatable screen of FIG. The above is the common operation of the machine translation device of the present invention.

In the present embodiment, the retrieved sentence is displayed in parallel translation, but it is also possible to display only the Japanese sentence or the English sentence. Also,
4 to 9 are examples of the screen design to be displayed.
Various designs are possible depending on the application. Further, in the present embodiment, as the degree of coincidence between the input sentence and the fixed phrase, a number obtained by subtracting 50 from the number of non-matching semantic features is subtracted from 100 from the number of matching semantic features. In addition to this, various methods such as dividing by the number of semantic features in the input sentence to obtain a relative evaluation of 100 points can be considered.

Embodiments will be described below with respect to each claim. Similar to the above-described basic embodiment, each embodiment will be described by taking as an example a process of displaying a bilingual sentence for a certain input sentence. Hereinafter, an embodiment of the machine translation apparatus according to claim 1 will be described. The process of translating the input sentence by the machine translation device according to claim 1 is the same as that of the basic embodiment except step A8. Therefore, here, only the difference in step A8 will be described in detail. Since the operation flow is the same as that of the basic embodiment, it will be described based on the flowchart of FIG.

Now, in steps A1 and A2, the user inputs "I was disconnected from the tour operator", and in steps A3 and
It is assumed that the following sentence example is retrieved in A6. Example 4: I got separated from {friend} I got separated from {my friend}. {Person} was separated Here, it is assumed that the "person" variable semantic feature does not have an element "tourist". At this time, step A of the basic embodiment
When the degree of matching with the input sentence of sentence example 4 is calculated based on 8,
Of the semantic features of sentence example 4, “person”, “hagure”, and “ta”, “hagure” and “ta” are in the input sentence. The remaining "personal" variable semantic features do not exist in the input sentence. Moreover, the sentence feature 4 does not have the semantic feature of “passenger” in the input sentence.

Therefore, 100 points, which is obtained by multiplying the number of matching 2 by 100 and subtracting the number of not matching 2 by 50, is 100, which is the degree of matching with the input sentence of this sentence example 4. .

However, since the degree of coincidence of 100 points in sentence example 4 is lower than the threshold value of 200 points, it is erased in the next step A13. In this claim, the above is different from the fixed sentence for the input sentence in which only the portion corresponding to the variable semantic feature is different and the portion corresponding to the variable semantic feature is registered in the word dictionary. Instead of simply calculating it as a disagreement,
The corresponding translated text can be displayed.

In the present claim, in step A8 of the basic embodiment, "if there is a variable semantic feature in the fixed sentence,
The element corresponding to it does not exist in the input sentence, and the only semantic feature that does not match the semantic feature of the fixed sentence in the input sentence and the character string that does not match the semantic feature of the fixed sentence are registered in the word dictionary. If it is, the variable semantic features that do not match in the fixed sentence and the semantic features that do not match in the input sentence are regarded as matching and the matching score is calculated, and the character string is stored in the word dictionary. A process of "creating a translated sentence by replacing the registered translated word" is added.

Now, in the sentence example 4, there exists a variable semantic feature “person” that does not match the input sentence. Further, in the input sentence, there is only the character string "tourist" that does not match the semantic feature of the fixed sentence. Furthermore, in the word dictionary 19, "tourist"
And its translation are defined. Therefore, this "person" and "tourist" are considered to match and are calculated.

Then, the sentence example 4 and the input sentence match each other in all semantic features. Multiplying the number of matching 3 by 100, the degree of matching with the input sentence of sentence example 4 becomes 300 points.
The sentence example 4 and the degree of coincidence of 300 are stored in the work memory 5.

At this time, the replacing unit 12 replaces the variable semantic feature element in the Japanese sentence of the fixed form sentence stored in the working memory 5 with the "tourist", and further replaces the variable semantic feature in the English sentence with "a tour conductor". The bilingual sentence is stored in the work memory 5 in a form replaced with. The threshold of 20 is the matching score of 300 in sentence example 4.
Since the score is over 0, the process moves to step A9. The subsequent steps are similar to those of the basic embodiment. Steps A10 to A15
In, the sentence example 4 is displayed on the liquid crystal display 4 as shown in FIG. 13, and the series of translation operations is completed.

Next, an embodiment of a machine translation device including the claim contents described in claim 2 will be described. In the present embodiment, the following parallel translation database in which a connectable character string is added immediately after each semantic feature is used in the parallel translation database of the basic example.

Example 1 ': How much is {New York}? How much is it to {New York}? {Place name} / {Up to}} How much is it to {New York}? Example 4': {Friend} got separated from {my friend}. {Personal name} / {and and, and, even, and even} were separated. Here, in {} that follows /, the set of connectable character strings is registered immediately after the variable semantic feature. There is.

The process of translating the input sentence by the machine translation device according to the second aspect is the same as that of the basic embodiment except step A8. Therefore, steps B1 to B4 that replace step A8 will be described here with reference to the flowchart of FIG. The other steps are the same as those in FIG.

Now, in steps A1 and A2, the user inputs "I missed the tour operator and the station", and then step A
Suppose that sentence example 4 ′ is retrieved from the parallel translation database in 3-A6. Here, it is necessary to search for a partial character string that can be replaced with the semantic feature “friend” from the character string “at the attendant and the station”. The following steps B1 to B4 are processing therefor.

<Step B1> In step B1, there is a variable semantic feature in the fixed sentence, an element corresponding to it does not exist in the input sentence, and the semantic feature that does not match the semantic feature of the fixed sentence is unique in the input sentence. Then, it is checked whether or not the condition that a character string that does not match the semantic feature of the fixed phrase is registered in the word dictionary is satisfied. If the condition is satisfied, the process proceeds to step B2, and if not, the process proceeds to B4. Now, the fixed sentence 4'has a variable semantic feature "person", there is no element corresponding to it in the input sentence, and the word dictionary has a substring "Tourist" at the "Tourist and station". station"
There is a headword. These partial character strings are called semantic feature candidate character strings. Since the semantic feature candidate character string exists, the process moves to step B2.

<Step B2> It is checked whether the set of character strings that can be connected immediately after the variable semantic feature includes the character string that appears immediately after the variable semantic feature candidate character string in the fixed sentence and is included. The matching degree is calculated by regarding a semantic feature candidate character string having a character string as a semantic feature, and the process proceeds to step B3.

Now, to a set of connectable character strings immediately after the variable semantic feature {and, and, and even, and up to},
Character string "to" immediately after the semantic feature candidate character string "tourist"
Therefore, the "passenger" is regarded as a semantic feature, and the degree of coincidence is calculated in the same manner as in the basic embodiment. Then, the process proceeds to step B3.

<Step B3> A translated sentence is created by replacing the character string that is the semantic feature with the translated word registered in the word dictionary. Now, the variable meaning feature of the English sentence of the fixed phrase 4'is translated into the translation "a tour condu" of the "tourist" registered in the machine-enabled dictionary.
The bilingual text replaced with “ctor” is stored in the working memory 5.

<Step B4> If the calculated coincidence is greater than or equal to a certain threshold, the process proceeds to step A9, and if it is less than the threshold, the process proceeds to step A13. Here, if the threshold value is calculated as 200, the degree of coincidence becomes 300, which exceeds the threshold value, and therefore the process proceeds to step A9. With this, we were able to find a substring that could be replaced with the semantic feature "friend" from the character string "At the trainer and at the station" and replace the bilingual translation. The subsequent steps are similar to those of the basic embodiment. In steps A10 to A15, the sentence example 4'is displayed on the liquid crystal display 4 as shown in FIG. 14, and the series of translation operations is completed.

Next, an embodiment of a machine translation device including the claim contents described in claim 3 will be described. Claim 3
The process of translating the input sentence by the machine translation device described in (1) is the same as that of the basic embodiment except step A8. Therefore,
Here, steps C1 to C4 that replace step A8
Will be described with reference to the flowchart of FIG. The other steps are the same as those in FIG.

Now, in steps A1 and A2, the user inputs "Is there a thigh?"
It is assumed that sentence example 5 is retrieved from the parallel translation database in 6. Here, it is necessary to search the character string "sumomo" for a partial character string that can be replaced with the semantic feature "food". The following steps C1 to C4 are processing therefor.

[0106] It is assumed that such an example is stored in the sentence example 5.

<Step C1> In step C1, there is a variable semantic feature in the fixed sentence, an element corresponding to it does not exist in the input sentence, and the only semantic feature that does not match the semantic feature of the fixed sentence is in the input sentence. Then, it is checked whether or not the condition that a character string that does not match the semantic feature of the fixed phrase is registered in the word dictionary is satisfied. If the condition is satisfied, the process proceeds to step C2, and if not, the process proceeds to C4.

Now, the fixed sentence 5 has the variable semantic feature "tabemono", there is no element corresponding to it in the input sentence, and the partial character string "sumomo" of "sumomo" in the word dictionary,
There is an entry word "peach". Since the semantic feature candidate character string exists, the process moves to step C2.

<Step C2> In step C2, the longest semantic feature candidate character string is selected, and the process proceeds to step C3. Currently, the semantic feature candidates are "sumomo" and "peach", and "sumomo" is longer, so "sumomo"
Is regarded as a semantic feature, and the degree of coincidence is calculated by the same method as in the basic embodiment. Then, the process proceeds to step C3.

<Step C3> In step C3, the character string as the semantic feature is replaced with the translated word registered in the word dictionary to create a translated sentence. Now, the bilingual sentence in which the variable semantic feature of the English sentence of the fixed sentence 5 is replaced with the translated word “a plum” of “Sumomo” registered in the machine-enabled dictionary is stored in the working memory 5.

<Step C4> In step C4, if the calculated coincidence is not less than a certain threshold value, step A9 is executed.
If not, move to step A13. Here, if the threshold value is calculated as 200, the degree of coincidence becomes 300, which exceeds the threshold value, and therefore the process proceeds to step A9. With this, we were able to find the substring that can be replaced with the semantic feature "tabemono" from the character string "sumomo" and replace the bilingual translation. The subsequent steps are similar to those of the basic embodiment. In steps A10 to A15, the sentence example 5 is the liquid crystal display 4
Is displayed as shown in FIG. 15, and the series of translation operations is completed.

Next, an embodiment of a machine translation apparatus including the contents of claim 4 will be described. In the present embodiment, the following bilingual database in which parts of speech of each semantic feature are added to the bilingual database of the basic example is used.

[0113] It is assumed that such an example is stored.

Here, the part of speech in the word dictionary of variable semantic features is registered after “:” after the variable semantic features. The process of translating the input sentence by the machine translation device according to claim 4 is the same as that of the basic embodiment except step A8. Therefore, steps D1 to D4 that replace step A8 will be described here with reference to the flowchart of FIG. The other steps are the same as those in FIG.

It is assumed that the user inputs "Please carefully pack in a wooden box" in steps A1 and A2, and the sentence example 6 is retrieved from the parallel translation database in steps A3 and A6. Here, it is necessary to search for a partial character string that can be replaced with the semantic feature “case” from the character string “polite in a wooden box”. The following steps D1 to D4 are processing therefor.

<Step D1> In step D1, there is a variable semantic feature in the fixed sentence, there is no element corresponding to it in the input sentence, and the only semantic feature that does not match the semantic feature of the fixed sentence is in the input sentence. And matching of the condition that a character string that does not match the semantic feature of the fixed sentence is registered in the word dictionary is performed. If the condition is satisfied, the process proceeds to step D2, and if not, the process proceeds to D4.

Now, the fixed sentence 6 has a variable semantic feature “case”, there is no element corresponding to it in the input sentence, and the word dictionary “subject to the wooden box” is the partial character string “wooden box”. , There is a headword "Polite". Since the semantic feature candidate character string exists, the process moves to step D2.

<Step D2> In step D2, a character string whose part-of-speech matches the part-of-speech of the variable semantic feature is selected from the semantic feature candidate character strings, and the process proceeds to step D3. Now, the part-of-speech of the variable semantic feature “case” in the fixed sentence 6 of the bilingual database is a noun, while the semantic feature candidates are “wooden box” and “polite”, and the part-of-speech of “wooden box” is the variable semantic feature “ Case"
Since it matches the part of speech of, the "wooden box" is regarded as a semantic feature, and the matching degree is calculated by the same method as in the basic embodiment. Then, the process proceeds to step D3.

<Step D3> In step D3, a translated sentence is created by replacing the character string that is the semantic feature with the translated word registered in the word dictionary, and the process proceeds to step D4. Now, the parallel translation sentence in which the variable semantic feature of the English sentence of the fixed sentence 5 is replaced with the translated word “a wooden box” of “wooden box” registered in the machine-enabled dictionary is stored in the working memory 5. Then, the process proceeds to step D4.

<Step D4> In step D4, if the calculated coincidence is not less than a certain threshold value, step A9
If not, move to step A13. Here, if the threshold value is calculated as 200, the degree of coincidence becomes 300, which exceeds the threshold value, and therefore the process proceeds to step A9. With this, we were able to find a partial character string that can be replaced with the semantic feature "case" from the character string "polite in a wooden box" and replace the bilingual translation. The subsequent steps are similar to those of the basic embodiment. In steps A10 to A15, the sentence example 6 is displayed on the liquid crystal display 4 as shown in FIG. 16, and the series of translation operations is completed.

[0121]

<Effect of the invention><Effect of claim 1> In the invention according to claim 1, when the sentence constituents other than the variable semantic features are almost the same and the variable semantic features are not the same, the word dictionary is changed. It is possible to extract the parallel translation of the character string corresponding to the semantic feature, and reliably display the sentence close to the sentence input by the user and the parallel translation.

<Effect of Claim 2> In the invention according to Claim 2, when the sentence constituents other than the variable semantic features are almost the same and the variable semantic features are not the same, the variable semantic features are extracted from the word dictionary. If there is more than one such candidate semantic feature, the set of connectable character strings immediately after the variable semantic feature includes the character string immediately after the candidate semantic feature. By selecting the feature candidate, it is possible to reliably display the sentence close to the sentence input by the user and its parallel translation.

<Effect of claim 3> In the invention according to claim 3, when the sentence constituents other than the variable semantic features are almost the same and the variable semantic features are not the same, the variable semantic features are extracted from the word dictionary. If there is more than one such semantic feature candidate, the longest semantic feature candidate is selected from among the semantic feature candidates, and the sentence close to the user's input sentence and its translation are selected. It can be displayed reliably.

<Effect of Claim 4> In the invention according to claim 4, when the sentence constituents other than the variable semantic features are almost the same and the variable semantic features are not the same, the variable semantic feature is extracted from the word dictionary. If there are multiple candidates for such semantic features that are similar to the sentence entered by the user, select the semantic feature candidate that has the part of speech that matches the part of speech of the variable semantic feature. And its translation can be displayed without fail.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of a conversational sentence translation device of the present invention.

FIG. 2 is an external view of an embodiment of the conversational sentence translation device of the present invention.

FIG. 3 is a flowchart of an embodiment of translation processing of the present invention.

FIG. 4 is an example of a display screen in an embodiment of the translation processing of the present invention.

FIG. 5 is an example of a display screen in the embodiment of the translation processing of the present invention.

FIG. 6 is an example of a display screen in the embodiment of the translation processing of the present invention.

FIG. 7 is an example of a display screen in the embodiment of the translation processing of the present invention.

FIG. 8 is an example of a display screen in the embodiment of the translation processing of the present invention.

FIG. 9 is an example of a display screen in the embodiment of the translation processing of the present invention.

FIG. 10 is a flowchart of an embodiment of translation processing of the present invention.

FIG. 11 is a flowchart of an embodiment of a translation process of the present invention.

FIG. 12 is a flowchart of an embodiment of translation processing of the present invention.

FIG. 13 is an example of a display screen in the embodiment of the translation processing of the present invention.

FIG. 14 is an example of a display screen in the embodiment of the translation processing of the present invention.

FIG. 15 is an example of a display screen in the embodiment of the translation processing of the present invention.

FIG. 16 is an example of a display screen in the embodiment of the translation processing of the present invention.

[Explanation of symbols]

1 CPU 2 LCD display 3 Display driver 4 Input recognition section 5 Working memory 6 ROM 7 Bilingual database 8 Semantic feature dictionary 9 Variable Semantic Feature Dictionary 10 Analysis Department 11 Search Department 12 Replacement part 13 Control unit 14 LCD display 15 dedicated pen 16 Handwriting recognition board 17 interpreter icon 18 bus 19 word dictionary

Claims (4)

(57) [Claims] 1. A parallel translation database in which an example sentence of a conversation in a first language, a sentence in a second language that is a translation thereof, and a semantic feature are registered as a pair, and a semantic feature and its semantic feature are extracted. A semantic feature dictionary that defines and stores the notation, a variable semantic feature dictionary that has the same semantic feature and is registered with replaceable words, and an input unit that inputs an input sentence for searching the bilingual database A search means for searching the bilingual database, the semantic feature dictionary, and the variable semantic feature dictionary based on the input sentence input from the input means; and the search means for creating the parallel translation database and the semantic feature dictionary. to sentence retrieved based, a substituted unit that performs replacement on the basis of the previous SL variable semantic feature dictionary, and example sentences extracted by the retrieval means and substituting means, the input An analysis unit that calculates the degree of coincidence with the input sentence input from the column, an output unit that outputs the translated sentence of the example sentence, and a translated word of the second language from the word of the first language based on the analysis unit. an apparatus and a word dictionary that can, in the analyzing means, the variable semantic feature in the input sentence
If the corresponding portion has a character string that is not registered in the semantic feature dictionary and the variable semantic feature dictionary as the semantic feature and is registered in the word dictionary, the character string becomes the variable semantic feature. A machine translation device characterized in that a degree of coincidence is obtained as a match, and the replacing means converts the character string into a translated word registered in the word dictionary. Wherein is wherein the translation database, the string can be connected after the string straight with each semantic features registered
In the analysis means, the variable semantic features in the input sentence are
In the corresponding part , the semantic feature dictionary is added as the semantic feature.
Contact the and the variable semantic feature dictionary is not registered, but the case in the word dictionary which is Rubun string is registered, immediately after the string and the character string connectable immediately after the sense element resistance When the character strings match, the degree of matching is obtained by regarding the character strings as matching the variable semantic features, and the replacing means converts the character strings into translated words registered in the word dictionary. The machine translation device according to claim 1. 3. A said analyzing means, said input sentence Yes
A portion corresponding to the variable semantic feature, is the semantic feature dictionary and the variable semantic feature dictionary as the semantic features is not registered, the the word dictionary a character string registered
In this case, for the longest character string registered in the word dictionary among the character strings, the degree of coincidence is determined as a match with the variable semantic feature , and the character string is calculated by the replacing means. The machine translation device according to claim 1, wherein the machine translation device converts the translation words registered in the word dictionary. The method according to claim 4, wherein said word dictionary, in addition to words and translation of the second language in the first language, part of speech registration word in the first language
Are, in the bilingual database, are part of speech registration of each semantic feature in the analysis means, the variable semantic feature in the input sentence
If there is a character string that is not registered in the semantic feature dictionary or the variable semantic feature dictionary as the semantic feature in the corresponding part , but is registered in the word dictionary, the sentence
Of string, translation of the string word class matches the part of speech of the variable semantic feature, determine the degree of coincidence as the matches to the variable semantic feature in the replacement unit, registered with the character string in the word dictionary The machine translation device according to claim 1, wherein