JPH11224251A - Method and device for machine translation - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize a machine translating method which meets both requirements of accurate translation and high speed translation. SOLUTION: A source language character string is inputted for a certain range or group according to a time series (step S1) and character strings in the certain range are collated against an example dictionary (step S2). The record of a source language pattern including a character string of the source language is retrieved through the above collation. Then one or more target language patterns of the retrieved record are combined to compose a translation sentence, which is outputted (step S4). If there is deficient information (step S3), the records of the source language pattern corresponding to the information and target language pattern are added as new records to the example dictionary (step S5). Then the updated example dictionary is used to repeat the processing after the collation operation.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] The present invention relates to a method for converting a first language into a second language.
In particular, the present invention relates to a machine translation method and apparatus that is intended for use in a form that requires real-time translation processing such as simultaneous interpretation.

[0002]

2. Description of the Related Art With the spread of computer networks in recent years, the distribution of text documents (hereinafter referred to as "electronic documents") handled on computers has become active, and even users who use computers by themselves have been described in foreign languages. Opportunities for receiving digitized documents from overseas and transmitting digitized documents in foreign languages to foreign countries are also increasing. This situation further induces a demand for incorporating and using a translation function in a document processing device or a portable information processing device, and is not limited to a digitized document. For example, a document in the form of audio data (hereinafter, referred to as a document) It is expected that the translation of a telephone document (hereinafter referred to as a voice document) will be further developed. The machine translation device used in such a form has a requirement for accurate translation (= first requirement) conventionally required of a machine translation device and a requirement for high-speed translation in real time. The second requirement is important.

The translation process of a conventional machine translation device can be roughly classified into three types: an analysis unit, a conversion unit, and a generation unit. The analysis unit analyzes an input sentence described in a first language (hereinafter, referred to as a source language) and creates a syntax tree reflecting the structure of the source language. The conversion means converts the syntax tree of the target language created by the analysis means into a syntax tree reflecting a structure of a second language (hereinafter, referred to as a target language). The generation unit generates a translation of the target language from the syntax tree of the target language created by the conversion unit.

However, in the above method or means, the analyzing means waits until the input sentence of the source language is completely input to create the syntax tree of the source language, and the converting means operates the analysis tree which operates before that. The source language parse tree is converted into the target language parse tree by waiting until the means generates a complete parse tree of the input sentence. Each unit cannot start its operation without waiting for the completion of the operation of the unit at the previous stage, such as creating a translation of the target language from the syntax tree of the target language after waiting until it is completely created .

That is, according to the conventional machine translation apparatus, each means cannot be operated independently because it is constituted by means depending on the order. Therefore, the improvement of the speed is not improved. There is a limit in realizing a machine translation apparatus that can perform translation only within the range and satisfies the above-described second requirement of performing high-speed translation in real time.

In order to solve such a problem, for example,
There has been proposed a machine translation apparatus as disclosed in "Document: JP-A-8-171566 (machine translation apparatus)". This apparatus includes a plurality of modules having processing interruption means for interrupting the processing, and the processing interruption means of each module determines whether or not the processing of the module can be interrupted based on a predetermined rule, and determines that the processing can be interrupted. In this case, the processing result obtained by the module up to that point is sent to another module that receives the processing result of the module, and if it is determined that the processing cannot be interrupted, the processing of the module is continued. Also, each module of this apparatus is provided with an output result adjusting means for adjusting the processing result to be sent to another module so that the consistency with the processing result already sent to the other module is maintained. With such a configuration, it is intended to satisfy the above-described second requirement that translation is performed at high speed in real time.

[0007]

With the machine translation apparatus disclosed in the above document, it is possible to realize a translation apparatus that is generally faster than earlier machine translation apparatuses.
However, considering the first requirement of accurate translation, the timing of the processing interruption is restricted by the semantic validity of the data input to the module. Will not be resolved, and if you try to perform a semantically perfect translation, you will not be able to interrupt the translation process. Therefore, in this machine translation apparatus, a time for determining the interruption is added to the processing time of the earlier machine translation apparatus, and in the worst case,
It is expected that the processing speed will be even slower than earlier machine translation devices.

For example, when translating Japanese into English, the Japanese predicate is located at the end of the sentence, and it cannot be determined whether the intention of the sentence is negative without looking at the usage of the predicate at the end of the sentence. Unless all of the Japanese in the input sentence is analyzed, an English sentence, which is the target language, cannot be generated. In other words, even with the machine translation apparatus disclosed in the above-mentioned document, the order dependency of the translation process is not eliminated, so that the translation is performed in real time and at high speed while satisfying the first requirement of accurate translation. A machine translation device that can also satisfy the second requirement has not been realized.

In view of the above, realization of a machine translation method and apparatus capable of satisfying the first requirement for accurate translation and also satisfying the second requirement for high-speed translation in real time. Was desired.

[0010]

The present invention employs the following structure to solve the above-mentioned problems. <Structure of Claim 1> A step of inputting source language data in a predetermined range from an input document according to a time series, and using an example dictionary for storing a set of a source language pattern and a target language pattern, Searching for a set of source language patterns including information of data; combining one or more target language patterns of the searched set to synthesize a translated sentence;
In addition, if there is insufficient information when performing the synthesis, the step of adding a set of the source language pattern and the target language pattern corresponding to this information to the example dictionary as a new set is repeatedly executed to execute the input document. Is a machine translation method characterized by translating.

<Explanation of Claim 1> According to the invention of claim 1, if there is a missing part at the time of creating a translated sentence, the translation processing is performed while adding the missing part as a new example, and there is no missing part. The data is output as a translation. As a result, there is an effect that translation can be performed in real time and at high speed while eliminating the order dependency in the translation process and satisfying the condition of performing accurate translation.

The input document is, for example, an electronic document.
It may be an audio medium. The source language and the target language are, for example, Japanese and English, but any language is applicable. Also, as a result of the search, a plurality of sets of source language patterns including information of a certain range of data may exist. Furthermore, when synthesizing a translated sentence, one or more target language patterns may be combined.

<Structure of Claim 2> A timing gate for inputting source language data from the input document in a fixed range according to time series, an example dictionary for storing a plurality of records each comprising a set of source language pattern and target language pattern, Matching means for matching a certain range of source language data input from the timing gate with a source language pattern in the example dictionary and searching for a source language pattern record including information matching the certain range of data, If a translated sentence is synthesized by combining one or more target language patterns of the searched records, and there is insufficient information when performing the synthesis, the set of the source language pattern and target language pattern corresponding to this information is A machine translation device comprising a predictive conversion unit for adding a new record to an example dictionary as a new record.

<Explanation of Claim 2> The invention of claim 2 is an invention of a machine translation apparatus which realizes the machine translation method of the invention of claim 1. The timing gate inputs, for example, a character string in the source language at regular intervals. The collating means collates a certain range of the character string input to the timing gate with the source language pattern of the example dictionary, and searches for a record of the source language pattern including the certain range of the character string. Here, the source language pattern including a certain range of character strings is defined as a case where the certain range of the character string completely matches the source language pattern, and a case where the certain range of the character string matches a part of the source language pattern. Is included.

The predictive conversion means synthesizes a translated sentence using one target language pattern of the record searched by the matching means or a combination of a plurality of target language patterns. Here, when there is missing information when synthesizing a translated sentence, it means that there is a portion where a certain range of character strings does not match the source language pattern in the record searched by the matching means, and as a result, the target language pattern This is when there is information that cannot be specified. Therefore, the predictive conversion unit adds such information as a new record to the example dictionary.

The collation means performs collation of the subsequent source language character strings including the added record. Therefore, in the source language character strings to be newly collated, the collation means matches the source language pattern of the added record. If there is something to do, the predictive conversion means
This is used to synthesize and output a translated sentence.

As described above, if there is no missing information as a translated sentence, the missing information is added as a new example while the translation result is output as it is, and subsequent matching processing is continued. While satisfying the condition that translation can be performed, the condition that an accurate translation can be obtained can be satisfied.

Also in the second aspect of the invention, the input document may be a voice medium or the like other than the character string, and the source language and the target language may be any language.

[0019]

Embodiments of the present invention will be described below in detail with reference to the drawings. << Specific Example >> FIG. 1 is a flowchart showing a specific example of the machine translation method of the present invention. Prior to this description, the machine translation apparatus of the present invention will be described.

<Structure> FIG. 2 is a diagram showing the structure of a specific example of the machine translation apparatus according to the present invention. The apparatus shown in the figure comprises a timing gate 2, an input buffer 3, a matching means 4, a predictive conversion means 5, and an example dictionary 6. In the figure, 1 is an input document consisting of a source language character string, and 7 is a target language character string. Is shown.

The timing gate 2 is a module for inputting source language data from the input document 1 in chronological order. The input buffer 3 is a buffer for temporarily storing unprocessed data among the data input by the timing gate 2. The matching unit 4 is a module that matches a source language character string as input data with an example record in the example dictionary 6 and searches for an example that matches the source language character string. The prediction conversion unit 5 is a module that converts a source language character string into a target language character string while predicting missing information based on the example searched by the matching unit 4. The example dictionary 6 is a dictionary in which examples used for operations of the matching unit 4 and the predictive conversion unit 5 are registered.

FIG. 3 is an explanatory diagram showing the structure of the example dictionary 6. Here, FIG. 3A shows the configuration of the example dictionary 6, and FIG. 3B shows the actual configuration example of the example dictionary 6.

As shown in FIG. 3A, the example dictionary 6 has two fields (101, 10
An arbitrary number of records of the format composed of 2) are stored.

The first field 101 of the example dictionary 6 describes a pattern to be collated with a source language character string input through the timing gate 2 (hereinafter referred to as a source language pattern). The source language pattern includes a symbol (hereinafter, referred to as an arbitrary pattern) indicating matching with an arbitrary word represented by a numeral “＄” and a condition of an arbitrary pattern represented by “= ｛. , Attribute conditions) are allowed.

The description of the attribute condition is "part of speech: variant" for convenience in the configuration example of FIG. 3B, but the attribute serving as the condition may be listed according to the purpose. The numeral portion of the arbitrary pattern is given to identify the collation position of the arbitrary pattern. Therefore, when a plurality of arbitrary patterns are described in the source language pattern, the numerical portions of the arbitrary patterns are distinguished and described.

For example, when there is one arbitrary pattern in the source language pattern to be matched with an arbitrary noun phrase,
It is described as “{1 = {noun phrase}” (shown at 103 in the figure). Also, the arbitrary pattern that matches the verb's intact form is
It is described as "{1 = {verb: unpredictable}".
04).

On the other hand, in the second field 102 of the record of the example dictionary 6, a pattern (hereinafter, referred to as a target language pattern) for converting a portion conforming to the source language pattern into a target language is described. The notation of the target language pattern is the same as that of the source language pattern. However, it is possible to change the attribute condition at the time of conversion. For example, after converting an arbitrary pattern that is a noun in the source language into a noun in the target language, the noun in the converted target language is converted to a verb in the target language. In the case of conversion to the form, the number part of the arbitrary pattern is made common, and “＄ 1 = ｛noun｝” is described in the source language pattern and “＄ 1 = ｛verb: complete｝” is described in the target language pattern (see FIG. Medium, 105).

<Operation> Next, the machine translation method of the present invention will be described with reference to the flowchart of FIG. First, the timing gate 2 inputs a source language character string from the input document 1 in chronological order (step S1). Then, this data is stored in the input buffer 3, and the collating means 4 extracts the source language character string stored in the input buffer 3, compares it with the example in the example dictionary 6, and includes the original language character string. An example is searched (step S2).

Thereafter, the predictive conversion means 5 combines one or more target language patterns of the example searched by the collating means 4, synthesizes and outputs a translated sentence (step S4). Here, the predictive conversion means 5 determines whether or not there is missing information when performing synthesis (step S3). If there is no missing information, the process proceeds to step S4. On the other hand, if there is missing information in step S3, a set consisting of a part of the source language pattern and a part of the target language pattern corresponding to the missing information is added to the example dictionary 6 as a new example (step S3). S5), returning to step S2.

By repeating such operations, the source language character string of the input document 1 is output as a translation of the target language.

Next, the operation of the timing gate 2 to the predictive conversion means 5 will be described for each module. FIG.
5 is a flowchart showing the operation of the timing gate 2. First, the timing gate 2 sequentially inputs a source language character string from the input document 1 and stores it in the input buffer 3 until a predetermined time has elapsed (step S).
101, S102). If it is determined in step S102 that the predetermined time has elapsed, it is determined whether a source language character string exists in the input buffer 3 (step S103), and a state in which the source language character string has not been stored in the input buffer 3 yet. If so, the timing gate 2 is in a state of waiting for the input document 1 to arrive, so the process returns to step S101 and repeats the input operation until the next fixed time elapses.

In step S103, if the source language character string exists in the input buffer 3,
Is transferred to the matching means 4 (step S104).

Thereafter, the timing gate 2 initializes the input buffer 3 by deleting the source language character string stored in the input buffer 3 (step S10).
5). Finally, the timing gate 2 checks whether the input of all the data of the input document 1 has been completed (step S106), and if there is any uninput data, returns to step S101 and repeats the above-described processing. On the other hand, if it is determined in step S106 that all data of the input document 1 has been input, the timing gate 2 ends the operation.

It should be noted that even if the operation of the timing gate 2 is completed, the new input document
If the setting is made such that the system is restarted when arrives, the processing from step S101 can be repeated as appropriate.

Next, the operation of the matching means 4 will be described. FIG.
5 is a flowchart showing the operation of the matching means 4. The matching unit 4 receives the source language character string from the timing gate 2 in synchronization with the operation of the timing gate 2 shown in step S105 described above, and temporarily stores the character string to be searched provided in the matching unit 4. Add / store in a storage area (hereinafter referred to as a search buffer) (step S20)
1).

Next, the character string stored in the search buffer at this time is compared with the source language pattern of a record (hereinafter, referred to as an example record) stored in the example dictionary 6 (step S202). Here, it is checked whether or not there is an example record that matches the search character string (step S203). It should be noted that the determination as to whether or not matching is made in step S203 is true when the search character string partially matches the source language pattern of the example record (= when the search character string is included in the source language pattern). I do.

If it is determined in step S203 that there is no matching example record, the process proceeds to step S203.
Returning to 201, the processing of steps S201 to S203 described above is repeated. As a result, a new source language character string is input from the timing gate 2 and stored in the search buffer. Thereafter, the collating operation is repeated with the character string in the new search buffer.

If it is determined in step S203 that there is a matching example record, the matching unit 4 transfers the matching example matching record (a plurality of example records may exist) to the predictive conversion unit 5. (Step S2
04).

Finally, the collating means 4 initializes the search buffer by erasing the character string stored in the search buffer (step S205), and ends the operation.
Even if the collating unit 4 has completed the operation, if it is set to restart when a new source language character string arrives from the timing gate 2, step S 201.
Subsequent operations can be repeated as appropriate.

Next, the operation of the predictive conversion means 5 will be described.
FIG. 6 is a flowchart showing the operation of the predictive conversion means 5. First, the predictive conversion unit 5 receives an example record matching the search character string from the matching unit 4 in synchronization with the operation of the matching unit 4 shown in step S204 (step S3).
01). Here, a translated sentence is synthesized according to the target language pattern of the received one or more example records (step S302). As a result, if a complete translation cannot be generated due to lack of information (step S303), a new example record including this missing information portion is created and registered in the example dictionary 6 (step S304).

After step S304, the prediction conversion means 5
Returns to step S301, receives a new search character string and an example record matching the search character string from the matching unit 4, and repeats the processing from step S301.

If a complete translation is generated in step S303, the predictive conversion means 5 outputs this translation as an output document 7 (step S305), and ends the operation. Note that, even when the operation is completed, the predictive conversion unit 5 outputs a new search character string
If the setting is made to restart when an example record that matches the search character string arrives, the operation after step S301 can be repeated as appropriate.

The operation of the machine translation apparatus as described above can be realized by control by a program of a computer serving as the machine translation apparatus. Therefore, the machine translation apparatus of the present invention is realized by means of recording the program on various recording media such as a floppy disk and a CD-ROM, and then installing the program on a general computer or downloading the program via a network. be able to.

Hereinafter, the above operation will be further described using a specific example as the input document 1. Here, it is assumed that a Japanese document "(Original)" It is said that a company that cannot transmit high-quality information will not grow "is input and that this Japanese document is translated into English. Will be explained. The example dictionary 6 contains
It is assumed that the record shown in FIG. 3B is stored.

First, the timing gate 2 inputs a part of the original sentence to the input buffer 3 for a fixed time (for example, several seconds) (steps S101 to S103). As a result, "high quality information" is assumed to be stored in the input buffer 3. Next, the timing gate 2 transfers the source language character string “high quality information” stored in the input buffer 3 to the matching unit 4 (step S104).

In synchronization with this operation, the collating means 4 receives the source language character string "high quality information" from the timing gate 2 and adds / stores it in the search buffer (step S2).
01). Next, the matching unit 4 checks the search character string “high quality information” stored in the search buffer against the source language pattern of each example record in the example dictionary 6 (step S202).

As a result, record 103 and record 10
6 match the source language pattern (step S20).
2), the search character string "high quality information", the matched record 103 "/ {1 = {noun phrase} / {1 = {noun phrase} /"", and the record 106" / high quality information / High-q
uality information / ”is transferred to the predictive conversion means 5 (step S204).

In synchronization with this operation, the prediction conversion means 5
Receive the above character string and two records (step S
301). Next, the predictive conversion unit 5 synthesizes a translated sentence according to the target language patterns of the two received records (step S302). In this case, the target language pattern “{1 = {noun phrase}”
The target language pattern “high-quality information” of No. 6 is applied, and the translated sentence 1 “high-quality information” is synthesized. Since there is no missing information in the translated sentence 1 (step S303), it is output as it is as the output document 7 (step S305).

As described above, since a fixed time elapses during a series of operations, the timing gate 2 newly inputs a part of the original sentence to the input buffer 3 (steps S101 to S1).
03). As a result, "cannot send" is displayed in input buffer 3
Shall be stored in

Next, the timing gate 2 transfers the source language character string "cannot be transmitted" stored in the input buffer 3 to the collating means 4 (step S104). In synchronization with this operation, the collation means 4 receives the source language character string "cannot be transmitted" from the timing gate 2 and adds / stores it in the search buffer (step S201).

Next, the collating means 4 determines that the search character string "cannot be transmitted" stored in the search buffer is "not sent".
The source language pattern of each example record is compared (step S202). As a result, record 105 and record 1
07 match the source language pattern (step S20).
3), this search character string "cannot be transmitted", and the matching record 105 "/ {1 = {noun} cannot} 2 = {noun} is / have to be {1 = {verb: complete}, unless ot"
Herwise {2 = {noun} / "and the record 107" / transmission / transmission / "are transferred to the predictive conversion means 5 (step S204).

In synchronization with this operation, the predictive conversion means 5 receives the character string and two records (step S3).
01). Next, the predictive conversion unit 5 synthesizes a translated sentence according to the target language patterns of the two received records (step S302). In this case, the target language pattern “{1 = {noun phrase}”
7 is applied to the target language pattern "transmission", and the applied "1" part "transmission" is converted to the verb completed form "transmitted", and "have to be transmitted"
d, unless otherwise {2 = {noun} ”is synthesized. Since there is $ 2 as missing information in this translated sentence (step S303), "/ {2 = {noun} is / {2 = {noun} /""including this missing information portion is used as an additional record as an example dictionary 6. (Step S30)
4). Then, “have t
o be transmitted, unless otherwise ”is retained as translation 3.

As described above, since a certain time elapses during a series of operations, the timing gate 2 newly inputs a part of the original sentence to the input buffer 3 (steps S101 to S1).
03). As a result, it is assumed that “the company does not grow” is stored in the input buffer 3.

Next, the timing gate 2 transfers the above-mentioned source language character string "the company does not grow" stored in the input buffer 3 to the collating means 4 (step S10).
4). In synchronization with this operation, the collation means 4 receives the source language character string "company does not grow" from the timing gate 2 and adds / stores it in the search buffer (step S20).
1).

Next, the collating means 4 collates the search character string "company does not grow" stored in the search buffer with the source language pattern of each example record in the example dictionary 6 (step S202). As a result, since the source language patterns of the records 104, 108, 109, and the additional record newly registered as the record including the lack information are collated (step S203), this search character string "the company does not grow". And the matched record 10
4 “/ ＄ 1 = ｛verb: not yet possible / not be able to ＄
1 = {verb: prototype} / ”and record 108“ / company /
company "and record 109" / grow up
/ ”And additional record“ / ＄ 2 = {noun} is / ＄ 2 =
Is transferred to the predictive conversion means 5 (step S2).
04).

In synchronization with this operation, the predictive conversion means 5 receives the character string and four records (step S3).
01). Next, the predictive conversion unit 5 synthesizes a translated sentence according to the target language patterns of the received four records (step S302). In this case, the target language pattern “company” of the record 108 is applied to the target language pattern “＄ 2 = {noun}” of the additional record, and the target language pattern “not beable to ＄ 1 =
Apply the target language pattern “grow up” of record 109 to {verb: original form} and “company not be able to grow”
The translated sentence 4 “up” is generated. Further, a translated sentence 5 is generated together with the translated sentence 3 held in the previous process (step S302), and "have to be transmitted, u" is generated.
"nless otherwise company not be able to grow up" is output (step S305).

In the same manner, the remaining source language character string “said” is also recorded in the record 110 as “, it is said”.
The translated sentence 6 is converted into the translated sentence 6, and as a result, the translated sentence 1, the translated sentence 5, and the translated sentence 6 are sequentially output, and the following translated sentence is obtained.

(Translation) "high-quality information hav
e to be transmitted, unless otherwise company not b
e able to grow up, it is said.

<Effects> As described above, according to the above specific example, a translated sentence is generated while predicting a shortage by collation from the source language to the target language. In the case where と does not satisfy the constraint, it is possible to predict a requirement that satisfies the constraint, so that order dependency of the translation process can be resolved, and translation can be performed quickly and accurately in real time.

This will be described below by taking the above (original text) as an example. For example, in a machine translation apparatus as shown in a conventional document, a sentence such as the above (original sentence) cannot be correctly translated unless one unit is translated up to the end of the sentence. Cannot be satisfied with the first requirement of translation into However, when such processing is performed, the second requirement of high-speed translation cannot be satisfied. On the other hand, for fast translation,
Once the above (original text) is separated by "cannot send",
It is possible that two sentences, "cannot transmit high-quality information" and "the company is said to not grow", may be translated separately. For example, "high-quality information can
not be transmitted. '' and `` it is said that company c
annot growup. ", and an accurate translation such as the above (translation) in this specific example cannot be obtained.

<Usage Form> In the above specific example, the machine translation apparatus for translating a Japanese document into an English document at high speed has been described. However, the language to be translated may be any language. Although the input / output medium has been described as an electronic document, the machine translation method and apparatus of the present invention can be applied to an audio medium having a strong time constraint as the input / output medium.
For example, such a device can be applied to a voice translation device, a translation telephone, and the like.

[Brief description of the drawings]

FIG. 1 is a flowchart showing a specific example of the machine translation method of the present invention.

FIG. 2 is a configuration diagram showing a specific example of the machine translation device of the present invention.

FIG. 3 is an explanatory diagram of a configuration of an example dictionary in the machine translation device of the present invention.

FIG. 4 is a flowchart showing the operation of a timing gate in the machine translation device of the present invention.

FIG. 5 is a flowchart showing the operation of the matching means in the machine translation device of the present invention.

FIG. 6 is a flowchart showing the operation of the predictive conversion means in the machine translation device of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input document 2 Timing gate 3 Input buffer 4 Matching means 5 Predictive conversion means 6 Example dictionary 7 Output document

Claims (2)

[Claims] 1. A step of inputting source language data in a predetermined range from an input document according to time series, and using an example dictionary for storing a set of a source language pattern and a target language pattern, Searching for a set of source language patterns including the information of the target language pattern, combining one or more target language patterns of the searched set to synthesize a translated sentence, and when there is insufficient information when performing the synthesis. Translating the input document by repeatedly executing a step of adding a set of a source language pattern and a target language pattern corresponding to the information to the example dictionary as a new set. 2. A timing gate for inputting source language data in a fixed range from an input document according to a time series, an example dictionary for storing a plurality of records each consisting of a source language pattern and a target language pattern, and an input from the timing gate. Data of a certain range of the source language is compared with the source language pattern of the example dictionary,
A matching unit for searching for a record of a source language pattern including information matching the data in the certain range; and combining one or more target language patterns of the record searched by the matching unit to synthesize a translated sentence, and If there is insufficient information at the time of performing, the predictive conversion means for adding a set of a source language pattern and a target language pattern corresponding to this information to the example dictionary as a new record. Translator.