WO2011047608A1

WO2011047608A1 - Forming method of patterned bilingual sentence pair and forming device thereof

Info

Publication number: WO2011047608A1
Application number: PCT/CN2010/077772
Authority: WO
Inventors: 张龙哺
Original assignee: 北京东方爱译科技有限责任公司
Priority date: 2009-10-20
Filing date: 2010-10-15
Publication date: 2011-04-28
Also published as: CN102043773B; CN102043773A

Abstract

A forming method of a patterned bilingual sentence pair, wherein, the patterned bilingual sentence pair has a patterned unit at least in a translation part; a translation unit content and corresponding identification information of the original text are in the patterned unit; the method includes: step S1: selecting a word from a sentence of the original text; step S2: judging whether the grammatical attributes of the word comply with the replaceable word conditions; if the judgment result in step S2 is "yes", then step S3 is executed: the identification information of the word and the translation content of the word are composed to be a replaceable unit, and the replaceable unit is written into the translation part; if the judgment result in step S2 is "no", then step S4 is executed: the translation content of the word is written into the translation part. And a corresponding forming device is provided.

Description

Patterned double sentence pair forming method and forming device thereof

Field of invention

The present invention relates to the establishment and accumulation of intelligent translation knowledge in the field of computer translation technology, and more particularly to a method for forming a dual statement pair and a forming apparatus thereof. Background of the invention

Since the 1930s, people have come up with the idea of machine translation. With the development of computer technology, various types of computer translation systems and technologies have emerged, such as ED (Electronic Dictionary), MT (Machine Translation), TM (Translation Memory), IT (Interactive Translation) and CAT (Computer Aided Translation). ) and many more.

These systems use different methods for language conversion work on certain aspects of natural language. Among them, the electronic dictionary can only translate or find words.

Traditional MT technology converts languages based on grammar rules, which are written by language experts and written by the programmer in the translation program. It can only be added and modified by programmers. Due to the richness and flexibility of the language, it is impossible to cover all linguistic phenomena with a small number of grammatical rules. Therefore, traditional MT technology cannot obtain good translation quality, especially for sentences with long sentences and complex sentences.

With the rapid increase in the speed of computer operations and the storage capacity of recording media, a statistical-based translation technique, Translation Memory Technology (TM), was introduced in the 1990s. The basic idea is to store two pairs of statements in large quantities. For the original sentence that has been translated or stored, as long as the corresponding translation is extracted, accurate translation results can be obtained. Therefore, TM technology points to a high-quality and accurate translation direction for computer translation technology.

Figure 1A shows a conventional translation scheme using TM translation technology. Among them, the TM translation mode compares (matches) the input original sentence with the original part of the corpus double statement pair. If the match is matched or the specified match rate is satisfied, the translated portion of the double sentence pair is output as the TM translation result.

Figure 1B shows an example of a sentence pair recorded by a conventional sentence-to-record method. That is, the original text is recorded in the left part, and the translation is recorded in the right part, separated by a separator. Among them, the original text and the translation are regular text content, namely words (words), punctuation marks, and so on. There is no other information to help translate than the separator between the original and the translation. Therefore, this kind of The role of sentence pairs is very limited. That is to say, in addition to the same sentence can give accurate translation results, it can not give accurate translation results for similar sentences.

Therefore, with traditional TM technology, it is necessary to accumulate all possible sentences and translation sentence pairs. However, due to the flexibility and richness of the language, and the arbitrariness of each author's writing, it is almost impossible to accumulate all the sentences in a certain translation language pair. This is because the amount of sentences is infinite or immeasurable. In practice, we have accumulated hundreds of thousands of sentence pairs in a certain profession, and spent a lot of manpower and financial resources, but in the translation test, only a few thousand coverage. Therefore, TM computer translation technology has encountered huge obstacles. Thus, people in turn think of the benefits of traditional MT technology, that is, to cover more sentences with a small number of grammatical rules or sentence patterns. Or combine MT technology with TM technology to form a multi-strategy translation technology.

In addition, some TM improvements have emerged in recent years, such as the use of sentence patterns in TM technology solutions, the purpose of which is to cover more sentences with the sentence patterns stored in the sentence library. The principle is to abstract the translated example sentence into a sentence pattern. In the translation, the sentence to be translated is first parsed and abstracted into a grammatical tree structure, and then the sentence is used to create the translation with the sentence to be translated (translation ). This method actually returns to the old path of traditional MT technology. The first reason is that abstracting an example sentence into a grammatical sentence is a time-consuming and laborious task, and it cannot be automated. At present, no practical sentence accumulation methods and tools have been seen.

The inventor of the present application, based on years of research on human brain translation thinking and foreign language learning and memory, proposed a set of systems for simulating human brain memory and storing translation knowledge, namely Bodian intelligent knowledge base system, and Corresponding Super Intelligent Computer Translation Technology (TM++). In the intelligent knowledge base system, the sentence pair is not a simple original text plus translation form, nor is it an abstract sentence pattern as described above, but an instance-based patterned sentence pair. The advantages of the model sentence to the translation theory are: 1. To embody and instantiate the complex abstract grammar for easy understanding and implementation; 2. In the intelligent translation technology, the pattern sentence is very suitable because it is The translation examples and the translated sentence patterns are integrated, which not only preserves the uniqueness of the specific translated sentence pairs but also the universality of the translated sentence patterns. Figures 2A-2C and 3A-3C show some examples of instance-based patterned sentence pairs (referred to as patterned sentence pairs).

The inventors of the present application have also invented various methods, devices and systems to achieve the intelligentization Summary of the invention

SUMMARY OF THE INVENTION The object of the present invention is to provide a method of forming a patterned sentence pair and a forming apparatus therefor.

With the method of forming the patterned sentence pairs and the forming apparatus thereof, the patterned sentence pairs can be formed and accumulated quickly and efficiently. The knowledge base of the intelligent knowledge base can be accumulated for all users, and the machine can automatically generate and accumulate intelligent translation knowledge while the user translates. This completely frees the traditional translation software from the language experts to develop translation rules or sentence patterns and is written or updated by software professional technicians, and will greatly accelerate the development and speed of the intelligent knowledge base. Therefore, it provides a viable technical solution for the early realization of high quality fully automatic machine translation.

DRAWINGS

Figure 1A is a block schematic diagram of a conventional TM computer translation technology solution.

Figure 1B shows an example of a traditional sentence pair.

2A-2C and 3A-3C show examples of patterned sentence pairs in the present invention.

Figure 4 shows an example of additional information for a patterned sentence pair.

Figure 5 is a flow chart showing a first embodiment of a method of forming a patterned sentence pair of the present invention.

Figure 6 is a flow chart showing a second embodiment of the method for forming a patterned sentence pair of the present invention.

Figure 7 is a flow chart showing a third embodiment of the method of forming a patterned sentence pair of the present invention.

Figure 8 is a flow chart showing a fourth embodiment of the method for forming a patterned sentence pair of the present invention.

Figure 9 is a flow chart showing a fifth embodiment of the method of forming a patterned sentence pair of the present invention.

Figure 10 is a flow chart showing a sixth embodiment of a method of forming a patterned sentence pair of the present invention.

Figure 11 is a block diagram showing a first embodiment of the patterning sentence pair forming apparatus of the present invention. Figure 12 is a block diagram showing a second embodiment of the patterning sentence pair forming apparatus of the present invention. Figure 13 shows the user interface interface of a moded dual statement pair forming apparatus of the present invention.

Fig. 14 schematically shows an example of a word unit. The specific embodiments of the present invention will be described in detail below with reference to the drawings. Implementation

Before describing a specific embodiment of the present invention, the modematic sentence pairs in the present invention will be described in detail.

In general, a two-state sentence pair includes: a source sentence expressed in a first language (referred to as a first-language original sentence), and a corresponding target sentence in a second language (referred to as a second-language translation sentence). The first-language original sentence is sometimes referred to more simply as the original text, and the second-language translated sentence is sometimes referred to more simply as the translation, because the second-language translated sentence is usually the translation result of the first-language original sentence.

For the original or original sentence, it can be a simple sentence, a complex sentence, or a phrase, a phrase, a short sentence, and so on. In summary, the original sentence referred to in this application has no limitation on its length or structure.

The method for forming a modular sentence pair of the present invention can be used in a computer translation system, and is particularly useful for forming and maintaining a sentence library in a computer translation system. Of course, it can also be used in other fields, such as corpus collection and collation.

Figures 2-3 illustrate various embodiments of patterned sentence-to-recording in the present invention.

In the modeled two-statement pair in the present invention, the original language, that is, the first language is recorded in the first portion, and the translation, that is, the second language, is recorded in the second portion. In the case of storing in the form of a file, the first part and the second part may be in the same file, for example, the first part and the second part are in the same line, separated by a specific separator, as shown in FIG. 2A; Or the first part and the second part are respectively in two adjacent rows, for example, the first part is an odd line, and the second part is an even line, as shown in FIG. 2B. In addition, in the case of storing in the form of a file, as shown in FIG. 2C, each of the first part and the second part may exist in a separate file, between the first part and the second part of the same sentence pair Have a corresponding relationship, such as they are in the same row.

In the case of storage in the form of a database, the first part and the second part can be in the same form. For example, the first part and the second part are in different column units of the same row, as shown in Fig. 3A. Or the first part and the second part are respectively in two adjacent rows, for example, the first part is an odd line, and the second part is an even line, as shown in Fig. 3B.

In the case of being stored in the form of a database, as shown in FIG. 3C, each of the first part and the second part may exist in a single form, and there is a correspondence between the first part and the second part of the same sentence pair. Relationships, such as they are in the same row. The patterned sentence pair described in the present invention has a regular unit and a pattern unit at least in any of the first portion and the second portion. In the patterning unit, the content of the own language unit and the corresponding language unit information are recorded in a predetermined format.

Specifically, in the translation recorded in the second portion, a patterning unit is used in addition to the conventional unit. The conventional unit means an immutable part, that is, a conventional translation, for example, "Yes", "Buy", "One Piece" in Figure 2-3 (may also be considered "buy a piece" is a conventional unit or Regular unit block). The patterning unit means a part that can be replaced, that is, the part of the translation can be replaced by other content, for example, in Figure 2-3: {\he|0|he/}, {\his|5|his /}, {\wife|6|Wife/}, {\gold watch|3|Gold watch/}.

A sentence pair having the patterning unit is referred to as a patterned sentence pair. The number of the conventional unit and the modular unit and the positional relationship therebetween may be arbitrary, which are determined by the structure of the sentence and the need for translation. A modular sentence pair typically has one or more regular units, one or more modular units. The positional relationship between the conventional unit and the modular unit may be mutually inter-phased, or several conventional units or modular units may be connected to each other.

A patterned sentence pair can all be a modular unit. E.g:

{\lazy I adj/} {\boy | n | /} {\! | f /} - {\lazy 101 lazy earning spoon /} {\boy 111 male {\! 121! /} The patterning unit has a predetermined format. The purpose of using the predetermined format is to enable the translation units in the patterning unit to be replaced. The patterning unit can include information such as the corresponding original unit, part of speech, attribute, and serial number in the sentence as needed to make accurate and desirable replacement. An example of a patterning unit is as follows: "He I he I pronoun |0", {\gold watch | 3 | Gold watch noun|Items /}, {\wife|6|Wife /}. Among them, "he", "golden watch", "wife" are the contents of the translation unit; "he", "gold watch", "wife" are the contents of the original unit; "pronoun" and "noun" are the part of the original unit It can also be the part of speech of the translation unit; "0", "3", "6" are the word numbers of the original unit in the original sentence. For computer programming, the serial number starts with "0".

The various information in the patterning unit can be separated by a specific separator, such as the characters "1", ", or spaces, or tabs, etc. The purpose is to be able to use the patterned sentence pairs when translating Better identification and processing. Each modular unit can be identified by a specific symbol pair, such as: "{" and "}", "{\" and "/}", etc., so that the patterning unit can be easily Identified.

In the example of Figure 2A, the second part of the translation has a patterning unit: {\he|0|he/}, {\his | 5 | His /}, {\wife | 6 | Wife /}, {\gold watch | 3 | Gold Watch /}, etc. Although the patterning unit is not clearly marked in the first part, those words or phrases of the first part referred to in the patterning unit in the second part of the translation are implied as replaceable. They are the 0th he, the 5th his, the 6th wife, and the 3rd gold_watch.

Further, in the original text of the first part of the record, in addition to recording the regular unit, the patterning unit may also be recorded. The conventional unit means an immutable part, that is, a conventional text. The patterning unit means a part that can be replaced, that is, the part of the original text can be replaced by other contents. Similarly, the number of conventional units and modular units can be arbitrary, depending on the structure of the sentence and the need for translation.

The patterning unit is recorded in a predetermined format. The purpose of the patterning is to enable the original text units in the patterning unit to be replaced. The patterning unit can include information such as part of speech, attributes, and the like of the word or phrase as needed to make accurate and desirable replacements. Preferably, the additional information in the patterning unit in the first partial text source is preferably complementary to the additional information in the patterning unit in the second partial translation. In addition, the patterning unit in the first part of the original text is preferably generated simultaneously with the corresponding patterning unit in the second part of the translation.

An example of labeling a word or phrase that can be replaced by a patterning unit in the first part, the original text, is shown in Figures 2B and 2C. As shown in the figure, the labeling method is as follows: {\he | pron/}, {\gold_watch | noun/}, {\his | prond | /}, {\wif e | noun | /}. Of course, other methods of labeling may be used, the purpose of which is to facilitate identification and replacement.

Further, in the original sentence of the first portion, in the patterning unit, the original unit content and the corresponding translation unit information may also be recorded in a predetermined format. The translation unit information includes: content of the translation unit and information such as part of speech, attribute or serial number of the translation unit, or any combination of the above various information.

In order to make better use of the above-mentioned pattern sentence pairs, some additional information can also be recorded therein, such as: the total number of units of the sentence, the modification mark, the quality level, the user name, the update date, the language number, etc., as shown in FIG. Shown. The additional information may be placed at the beginning, end, or other location of the patterned sentence pair as long as it has a correspondence with the patterned sentence pair. "29|N|2|L.g _an 88|031121|01" in Fig. 4 is a specific example of the attachment information.

As can be seen from the above description, the patterned sentence pair in the present invention is both a translation example and a translation model. Therefore, it can retain the uniqueness of the specific translated sentence pairs and the universality of the translation mode. Using this pattern of sentence pairs, you can make regular input sentence sentences The matching translation, in order to guarantee the specific translation requirements of the specific sentence, the pattern matching matching translation of the input original sentence, and the more advanced intelligent translation, the related content can refer to other related inventions of the present invention.

The method for forming a modular sentence pair of the present invention does not require abstraction of the translated double sentence pair (the abstraction operation requires a lot of grammar thinking and induction and a large number of rules), and only needs to add some existing information, so This method is easily implemented by a computer. For example, in the process of translating with translation software, the interactive translation (IT) module or the auxiliary translation module (CAT) is used to collect the information needed by the patterning unit and form the required patterning unit, and then write them into the model sentence pair. Just fine. DETAILED DESCRIPTION OF THE INVENTION Various specific embodiments of the present invention will now be described in detail with reference to Figures 5-12. Referring first to Figure 5, there is shown a first embodiment of a method of forming a patterned sentence pair in accordance with the present invention.

In this embodiment, a method for forming a dual statement pair according to the present invention includes the following steps: Step S1: Select a word in the original sentence. The word can be a word, a phrase, or a phrase.

Step S2: Determine whether the grammatical attribute of the word meets the condition of the replaceable word. The replaceable word conditions may be specified and judged according to part of speech, for example: nouns, adjectives, pronouns, numerals, etc. are predetermined as replaceable words. Then, if the part of speech of a word is a noun, an adjective, a pronoun or a number, the grammatical attribute of the word conforms to the condition of the replaceable word. Of course, the alternative word condition can also be specified and judged according to the attribute of the word, for example: a word whose attribute is "object", "person", "time" or "place" is defined as a replaceable word.

If the result of the determination in step S2 is "YES", then step S3 is performed: the identification information of the word and the translation content of the word are combined into a patterning unit, and written to the translation portion. The identification information may include information such as the content of the original text unit and the part of speech, the attribute or the serial number of the original text unit, or any combination of the above various kinds of information, as needed. See the description of the modular unit above for more details.

If the result of the determination in step S2 is "NO", then step S4 is executed: the translation content of the word is written to the translation portion.

As shown in FIG. 6, a second embodiment of a method for forming a dual statement pair according to the present invention is different from the first embodiment shown in FIG. 5 in that the result of the determination in step S2 is "NO". In the case, the execution step S5 is further performed to determine whether there is a special control character or instruction. Flexible control of the formation of the patterning unit can be achieved by setting special control symbols or commands. With it, words that do not conform to replaceable words can be modeled outside of the predetermined rules.

If the result of the determination in step S5 is "YES", then step S3 is performed: forming the identification information of the word and the translated content of the word into a replaceable unit, and writing it to the translation part; If the result of the determination in S5 is "NO", then step S4 is performed: the translation content of the word is written to the translation portion.

Referring to Fig. 7, there is shown a mode double statement pair forming method of a third embodiment of the present invention, wherein the word corresponds to a word unit.

Before the step S1, the execution step SO: forms the word unit of the original sentence into a word unit. The step S 1 is specifically: selecting one translation in the word unit.

The specific method of forming the word unit of the original sentence may be a method of looking up the dictionary, that is, using the original word to search the dictionary or the sentence, and obtaining the corresponding translation (interpretation), part of speech, attribute, association and the like. In the word unit, the word serial number information of the word in the original sentence is also included.

Further, the SO operation can be performed on all the words in the original sentence to form an array of word units.

As shown in FIG. 8, a method for forming a pattern double statement pair according to the fourth embodiment of the present invention is different from the third embodiment shown in FIG. 7 in that: the result of the determination in step S2 is "NO". In the case, the execution step S5 is further performed to determine whether there is a special control character or instruction.

Referring to Figure 9, there is shown a patterning two-statement pair forming method in accordance with a fifth embodiment of the present invention, the method comprising:

Find alternative words in the original sentence; the basis can also be the various predetermined criteria or conditions discussed above.

Finding a translation with the replaceable word in the target sentence;

The identification information of the word is added at the translation to form a patterning unit.

The original text identification information includes information of the original unit and the part of the original unit, the attribute or the serial number of the sentence, or any combination of the above various kinds of information.

Referring to FIG. 10, a method for forming a patterned double sentence pair according to a sixth embodiment of the present invention is shown. The method includes:

Find alternative words in the original sentence; the basis can also be the various predetermined criteria or conditions discussed above. Finding a translation with the replaceable word in the target sentence;

The identification information of the word and the translated content are combined into a patterning unit, and the original translation content is replaced by the original translation content.

Referring to FIG. 11, there is shown a first embodiment of a device for forming a patterned double sentence pair according to the present invention, wherein a patterned double sentence pair has a patterning unit in at least a translation portion; in the patterning unit, Having a translation unit content and corresponding original text identification information; the device includes:

a judging module, configured to determine whether the grammatical attribute of the word conforms to the replaceable word condition; the modularizing unit forming module, configured to form the identifying information and the content of the word into a modular unit;

a writing module, configured to write a translation or a patterning unit of the word to the translation part; and a word unit forming module, configured to form the word unit, and the forming method may be to look up the dictionary.

Referring to Figure 12, there is shown a second embodiment of a moded dual statement pair forming apparatus of the present invention, wherein the patterned double sentence pair has a patterning unit in at least the translation portion; in the patterning unit, Having a translation unit content and corresponding original text identification information; the device includes:

a writing module, configured to write a translation or a patterning unit of the word to the translation part; and, a word unit forming module, configured to form the word unit, the forming method may be to search the dictionary; the word unit forming module may be All words in a sentence perform a word unit forming operation to form an array of word cells.

Figure 13 shows a user interface interface of a moded dual statement pair forming apparatus of the present invention. Among them, the original sentence 〃 We see the wonderful translation result of the system Each word in "TM++ technology." is displayed in the interactive translation area (the upper half of the figure) and forms a unit of words. In the figure, the third (4th word) word unit is specifically shown. Multi-translation (interpretation). In the interactive translation area, when a certain translation is clicked with a mouse, the device for forming a modular double-statement pair of the present invention can be triggered, and the method for forming a dual-state sentence pair according to the present invention To form a modular sentence pair.

Fig. 14 schematically shows an example of a word unit. While various aspects, embodiments, and embodiments of the present application have been described in detail above, the invention of the present application is not limited thereto. Various changes, modifications, or modifications can be made by those skilled in the art. Such changes, modifications, and modifications are intended to be included within the scope of the present invention without departing from the spirit and scope of the invention.

Claims

Claim

A method for forming a schema double statement pair, wherein the mode double statement pair has a patterning unit in at least a translation part; and in the patterning unit, having a translation unit content and corresponding original text identification information; Methods include:

Step S1: Select a word in the original sentence;

Step S2: determining whether the grammatical attribute of the word meets the condition of the replaceable word;

If the result of the determination in step S2 is "YES", then step S3 is performed: forming the identification information of the word and the translated content of the word into a replaceable unit, and writing it to the translation part; If the result of the determination in S2 is "NO", then step S4 is performed: the translation content of the word is written to the translation portion.

2. The method according to claim 1, wherein, in the case where the determination result in step S2 is "NO", the processing step S5 is further performed to: determine whether there is a special control character or instruction;

3. A patterned double sentence pair formation method according to claim 1, wherein said word corresponds to a word unit;

Before the step S1, executing step so: forming a word of the original sentence into a word unit; the step S1 is specifically: selecting one of the word units.

4. The method according to claim 3, wherein the step SO is specifically: forming a word unit by using a dictionary method;

5. The method according to claim 1, wherein the original text identification information includes information of a source unit and a part of a word unit, an attribute or a serial number of the original unit, or any of the plurality of types of information. combination.

A method for forming a pattern double statement pair, wherein the patterned double statement pair has a patterning unit in at least a translation part; and in the patterning unit, having a translation unit content and corresponding original text identification information; Methods include: Find alternative words in the original sentence;

Finding a translation with the replaceable word in the target sentence;

A method for forming a dual statement pair, wherein the mode double statement pair has a patterning unit in at least a translation part; and in the patterning unit, having a translation unit content and corresponding original text identification information; Methods include:

Find alternative words in the original sentence;

Finding a translation with the replaceable word in the target sentence;

8. The method according to claim 7, wherein the original text identification information includes information of a source unit and a part of a word unit, an attribute or a serial number of the original unit, or any of the plurality of types of information. combination.

9. A device for forming a schema double statement pair, wherein the patterned double statement pair has a patterning unit in at least a translation part; and in the patterning unit, having a translation unit content and corresponding original text identification information; The device includes:

A write module for writing a translation or a patterning unit of a word to a translation part.

10. The device for forming a modular double sentence pair according to claim 9, further comprising: a word unit forming module for forming a word unit, the forming method may be a dictionary; the word unit forming module may be used for the original sentence All words in the word unit are formed to form an array of word cells.