Device for translating phrases of several words from the first language to the second


(57) Abstract:

The invention relates to a system for translating phrases from a first language to a second language, in particular, but not exclusively, to such a system, which generates speech output in the second language of the speech input in the first language. The device comprises a means of storing data 1, input means 2, the output of the tool 3 tool 4 speech recognition processing means 5. 10 C.p. f-crystals, 1 Il.

The present invention relates to a system for translating phrases from a first language to the second, in particular, but not exclusively, to such a system, which generates speech output in the second language of the speech input in the first language.

For many years there is a search machine that could quickly and automatically to transfer, in particular oral, from one language to another. However, even despite the huge advances in computing, pattern recognition and intesely speech in recent years, these machines remain the subject of fantasy and dreams.

In the field of automatic translation of the text were carried out intensive research work with computing systems. In addition to the very limited purposes (for example, perevalne, to replace human translators. Problems of interpretation are compounded by errors in speech recognition, the additional information contained in intonation, stress, etc., as well as inaccuracies of speech itself.

Unfortunately, existing packages automatic translation of texts somehow inefficient and do not meet the requirements of systems speech translation. Most of these packages have been developed to help professional translators and the result of such packages must be edited before it will take a reasonable view of the language has been translated. Most of these packages is a dialog or live and work in the slow mode, batch processing, and none of them are able to work in real-time voice translation. In addition, the software packages used for translation, fixed, because idioms and other exceptions to the rule may cause an error on exit: the user has no guarantee that the resulting text is translated correctly. Existing systems are also very heavy use of the CPU, which makes them expensive to operate and, therefore, unsuitable d the th translation which eliminated these drawbacks.

According to the present invention a device for translating phrases from a first language to a second language (see drawing) containing a means of storing data 1, which contains a set of phrases in the second language, the input means 2 for receiving the phrase in the first language, the output means 3 for output in a second language one of the phrases from the specified set of phrases in the second language tool 4 speech recognition to determine which of the phrases specified set corresponds to the input phrase; processing means 5, triggered under the influence of the specified characterizing means to control the output medium and to ensure output of the same phrases from the set that corresponds to the input phrase.

Such a device provides a very fast translation. Spent time equal to the time to identify (a characterization of the output phrase and search for "answer" in a second language.

The device can be implemented to give the user performing the input, confirmation that she recognized/understood the system correctly, which is especially important in the translation of oral speech.

When the user receives the confirmation, the EP consists only of previously executed accurate translations.

The device also provides quick translation for a few second languages; essentially, you must add only the additional storage device that contains sets of phrases in each of the additional second languages.

In the drawing, is shown a block diagram of principal components of the device of the present invention.

The invention is based on our recognition that there is an opportunity to describe and capture the semantic content of a large number of separate phrases using a much smaller number of keywords. By an appropriate selection of keywords you can use existing commercially available devices, speech recognition, which can recognize significantly fewer words than are contained in a sufficiently large set of phrases, for characterization and differentiation of a large set of phrases.

The operation of the automated transfer generally relies on the ability of these key words to conduct a proper distinction between the phrases. The greater the degree of separation of phrases is achieved with work, the higher the tolerance of the device to recognition errors and to various inconsistencies introduced by the estate To each of the words N phrases in accordance with the frequency of occurrence of words in phrases.

2. To select the M most frequent words as the initial list of keywords, where M is the number of words vocabulary speech recognition.

Then determined by the presence or absence of each keyword in every sentence. Is the number of phrases (E), certain key words.

4. Let i=I.

5. The key word temporarily excluded from the list, and calculates a new value E(E').

6. Temporarily excluded keyword is assigned the value e-E', it will be the indicator of the deteriorating working after upgrade key words and, therefore, its contribution to overall system efficiency. In fact, this indicator is used for each keyword has contributed to the separation of the maximum possible number of pairs of phrases, but without duplicating the functions of other keywords.

7. Temporarily excluded keywords restored and the process is repeated for each of the M keywords.

8. The word with the lowest index is excluded from the current list of keywords.

9. To replace the deleted words used (M+i)-th most frequent word, and then calculates a new value that is

< i is incremented and the process repeats from step 5 until while M+i>K, then the process ends.

Otherwise, (M+i)-th word is dropped, the coefficient is incremented and the process repeats from step 9 until such time as M+i>K, then the word, dropped in step 8 is replaced and the process is terminated.

The final list of keywords contains the optimal set of M individual keywords to identify phrases.

Separation phrases may further iteration, starting with M the best words obtained in the previous iteration. To determine the sequence of word candidates can use other heuristic methods, in addition to the frequency of the sort, especially if there is a priori linguistic information. In addition, there is the likelihood that words that appear in the lower part of the frequency list, will not play a significant role in the separation of the phrases so the search in the lower half or in the lower two thirds of the frequency list may be superfluous.

Sometimes it happens that most of the phrases are different and the value E becomes very close to zero in fairly early stages of the search. In these cases, it is possible to calculate on the basis that the phrases are separated by more than the minimum number of keywords, and to some extent, will provide protection against errors in speech recognition.

When the search becomes clear that several classes of phrases cannot be split up, if not to expand the dictionary of keywords. These "clusters" or groups of phrases tend to differ from each other only in one word or chain of subordinate words (e.g., date in a business letter) and represent candidates that have been obtained automatically for the preparation of subsahara keywords (see below).

Obviously, the recognition of the individual keyword does not take into account the order of words and the additional sense that it can contain. The presence or absence of steam (or other multiple groups of keywords with different separation between them can therefore also be useful to improve the efficiency of the single set of keywords. In speech recognition, this gives the advantage that the efficiency can be no extension of the recognition dictionary. If the device is used for working with texts, additional benefits can be obtained due to the generation of keywords to include punctuation, parts of words and combination with the my bed" and "sell this bed".

The use of pairs of keywords increases the value of individual words-components, if you are able to resolve other conflicts. Find pairs of words that are not necessarily adjacent, but separated by different amounts in other words, again begins with frequency sorting. Word pairs with both components in M keywords, compiled from the sorted list, if they resolve the outstanding contradictions phrase. The final list of single keywords and pairs of keywords is estimated as above, and calculates the overall score contradictions phrase that is

Now begins the search for the best way of working couples where one or both members of a couple of keywords not included in the current list of keywords. Candidates in the following few words are taken from the top of the frequency sorted list and attached to the bottom of the keyword list. Individual word pairs that are not yet present in the list, also joined, and an equal number of the least effective individual keywords erased. This can lead to abrasion of other pairs, if part of the word is no longer present in the list. Calculated a new value of E (E'). If there is improvement and E'<E, last modification list is bathing in the frequency list are selected following pairs like when searching for single words, couples can select and other evristicheskij methods.

It should be noted that some keywords play an important role when participating in multiple groups of words together than separately.

This method applies to the formation of large groups of keywords (containing more than 2 words), but with decreasing frequency influence on the resolution of contradictions phrases noticeable only in a very large set of phrases.

The number of calculations performed to search for key words increases with the number of keywords and phrases. They can be reduced, first, by implementing the algorithm on a subset of phrases that are conflicting or close to be contradictory, keywords and their estimated coefficients obtained in this way provide a more efficient sorting "candidates" in keywords for the main algorithm, which will work with a more complete set of phrases.

When speech recognition is the use of some words that are not included in the set of keywords can lead to numerous cases of false recognition of key words, such as the appearance of the word "I" can lead to recognition as a key which starts the search keywords, when the subsequent identification of phrases the actual separation of the phrases will not be affected. Moreover, since the frequency of occurrence of synonyms will be higher if they are taken together, compared with single words, usually with their detection will be contacted by a greater amount of information contained in the phrase.

The use of keywords can be extended to parts of keywords, i.e. phonemes that also occur with greater frequency and which carry most information about the different phrases than whole words. Moreover, the identification of certain parts of words in continuous speech is often not as difficult as identifying full of words, and therefore preferred in the translation system, in which the incoming speech. In the present description, the term "key word" for multiplicity is used to denote the whole of keywords or parts of keywords.

Phrases of different classes differ from each other only subordinate phrases and clauses, which may contain details such as dates, times, prices of items, names, or other groups of words. It may happen that the dictionary speech recognition will be sufficient to unisciti these sub-patterns.

Moreover, it is quite possible that the overall vocabulary necessary to separate classes of phrases and subordinate bodies, will contain many words that will be confused with each other. The advantage of the method according to the present invention is that the source statement (or any transformation of the original utterance) can be entered into the buffer and the recognition process can be repeated after you have defined the class phrase. Thus, the OCR engine should not substitute a generic dictionary with its many potential errors related to confusion in words. This task rests with the user. It should be noted that the speed of the second detection process is not limited by the speed of the source statements, and in principle this is the second recognition can be performed faster than real-time, and therefore will not make a noticeable delay. Iteration recognition can be performed as many times as necessary to identify the desired phrase and its subordinate bodies. Thus, it is possible to "embed" the process of recognition, phrase, characterized in several separate steps, with each step involves different dictionaries Clacton. This is because the position of chain words are defined as such only if there are several alternatives for reducing them close contextual dependence. In addition to the contextual importance it is assumed that there are dependencies between words that are inside and outside of the potential sub-chains, and, therefore, should be the goal for keywords, which consists in the separation of the phrase as a whole without the use of words within the chain. This is illustrated in phrases containing changing the date, where rarely change the words, except for the date (further studies should demonstrate the justice or injustice of the assumption that such independence from context in General invariant between languages and can be used for endless expansion of translated phrases).

This particular aspect of the present invention also has significant advantages when used in the translation of the text, where the computational cost of the process when you search in large dictionaries can be dramatically reduced through the use of such a hierarchy of smaller dictionaries and phrasebooks. Some subordinate phrases do not require translation, and often in such cases, C is penetrated statements, which contain references to such labels as proper names: "Call to phone Mr. Smith, please." As before, the device can determine the class of phrases together with the position of the words in the cupboard, which correspond to the reference to the label. The processing of these words, referring to the label at the time of transfer, is simply the transmission source of the acoustic signal in the appropriate place statements in the object language. Clearly, it is desirable that the synthesized voice on the object language consistent with the speaker's voice and speech synthesizers based on the input text, it is necessary to set certain parameters of speech, in order to achieve the best match possible voices (young, old, male, female, etc).

To ensure that the user was sure that the object language will be given the right phrase, the device determines whether a sentence in the input language will be repeated. To ensure this function, the device has means 1, which contains the full set of phrases in the input language.

Preferably, in the device of the phrases are stored as text in the code AS CII, for example, because it reduces the need for memory of all who is from memory and transmitted to text-speech Converter and a speech synthesizer. Using the code AS CII requires one byte per character, so the memory volume of 0.5 M can accommodate about 10,000 phrases. Thus, the device provides a translation of about 10,000 phrases, requires a minimum of approximately 1 megabyte, which is easy to implement on your hard disk. The use of two symmetrically constructed devices ensures the implementation of the bilateral relations. The benefit here is that each device recognizes and synthesizes words in the language of the operator of this device. Communication with the second device is carried out according to the Protocol, which defines the phrase and the content of subordinate phrases. The Protocol language-independent and, therefore, allows you to send messages without having to identify the object language. In addition, it allows people who speak several languages, to obtain simultaneously the translation output only device.

Example. Demonstration system connected to the telephone network, was used to demonstrate the effectiveness of the approach type "PhraseBook". In the demonstration system was used devices speech recognition "Wotan" speech synthesizer "Infovox and computer IBM PC XT.

The speech analyzer "Wotan" can raspoznavat for system management, you can as key to select up to 60 words. Words, the control system should not appear in the input phrases, so it is preferable to use, instead of voice commands.

Memory phrases contains 400 English words and their French equivalents. In the English phrases contains about 1100 different words. This number of phrases usually found in the typical phrase book for business people. Once the program has been run selection of keywords, described above, was selected 60 words that have successfully shared by all phrase. Of 400 phrases only 33 are separated by one word, while these 32 phrases are 16 pairs.

Demonstration system for recognition keywords finds an appropriate phrase, verbally confirms it, and outputs through a speech synthesizer with its French equivalent.

It is important to note that text-speech synthesizer is not required for the present invention. It is highly desirable, for example, to synthesize speech on objective language from pre-recorded or coded words and phrases. Here the advantage is that such speech can be recorded by the user acoustically and can match any webrocket in text-speech synthesizer in the languages of the major countries, where such technology is unlikely to get hardware implementation in the near future, for example, for languages Hindi and Arabic.

In addition to interpreting the present device is applicable to systems of translation, writing-oral or oral and written translation. May be particularly useful in the automation of clerical work, where it is easy to implement managed voice machine, automatically displaying text in a foreign language. In fact, this car will use a speech analyzer, the above program and management system, but output text in a second language will be to a printer or telex or other communication device. It is easy to enter the standard phrase routine business correspondence in different languages.

1. Device for translating phrases of several words from the first language to the second contains the input means and the speech recognition, characterized in that it introduced a means of storing data, processing means and output means, and means memorization of data is stored first set of N phase in a second language input tool for the transmission of speech phrases of several words in the first language to the tool rozpoznawanie certain keywords or parts of keywords in the first language, moreover, R M and less than the total number of different words N phrases in the first language corresponding to the first set of phrases in the second language, the processing means operable connected with speech recognition, a means of storing data and output means and is arranged with the possibility of controlling the recognition by the speech recognition any R keywords or parts of keywords in the input sentence and to determine the results of the recognition of keywords or parts of keywords, or combinations thereof, which phrase from the first set of phrases corresponds to the input phrase and with the ability to control the output of the appropriate phrase from the means of storing data through the output means.

2. The device under item 1, characterized in that the means of storing data stored second set of N phrases in the first language, in which each of the N phrases in the second set corresponds to one of N phrases in the first set, the processing means is configured to determine in the process depending on the recognized keywords or parts of keywords, or combinations thereof, which phrase from the second set corresponds to the input sentence, and to indicate to the user the mouth of the ri that the processing means can also prevent the output of the corresponding phrase in the second language until you receive confirmation, the definition of done correctly.

3. The device under item 1 or 2, characterized in that it contains remote from each other of the first and second terminals and operable terminals are connected by a data transmission channel, the first terminal includes the aforementioned input means and processing means, the second terminal includes a means of storing data, where the first set of phrases, and the output means in the first terminal contains a means for generating messages for transmission to the second terminal through the transmission channel and the message indicates what the phrase from the first set of phrases corresponds to the input phrase.

4. The device according to PP.1 to 3, characterized in that it contains another means of storing Q keywords, where Q <> M, while the processing means is configured to control the process of recognition by the speech recognition any Q keywords.

5. The device according to PP. 1 to 4, characterized in that the output means provides speech output.

6. The device under item 5, characterized in that it has a voice synthesizer, which feeds on the output means of speech signals.

7. The device according to PP.1 to 6, characterized in, mo, as part of a phrase in a second language.

8. The device according to PP.1 to 7, characterized in that it has means for education text output in a second language.

9. The device according to PP.1 to 8, characterized in that it contains a storage device for storing further set of phrases on each of the many second languages.

10. The device under item 1, characterized in that each phrase of the first set of N phrases is uniquely characterized by keyword, part of the keyword or combination thereof.

11. The device according to p. 10, characterized in that the number of R keywords contains only a sufficient number of words or parts of keywords.


Same patents:

FIELD: computer science, possible use for processing information composed in accordance to context-dependent grammar rules.

SUBSTANCE: method includes transforming input information in text processor to semantic text profiles, which are then sent to situational processor, which forms a set of situations development schemes, generates semantic profile for circuit of analytic count, performs comparison and determines likeness of semantic text profile and semantic profile of analytic count circuit, determines events - initiators of entity values changes and inputs them to series of cause-consequence graphs, computes new values of entities, on basis of which forms different situation development schemes, which are inputted into visualization processor, which forms end data set for displaying selected data.

EFFECT: improved information capacity and precision of news events analysis by selecting in process of analysis of semantic profile of text.

7 dwg

FIELD: technology for organization of multi-language messages in an intellectual network for determining mode of broadcast relay of spoken message.

SUBSTANCE: method contains following stages: setting of field "type index" in identification codes of spoken message and setting of "type index table" for each concrete spoken message, meant for broadcast relay, reading of index number from field "type index" with following search in "type index table" to determine mode of broadcast relay on basis of speech attributes of concrete spoken message in "type index table".

EFFECT: substantially increased amount of supported attributes.

2 cl, 11 tbl

FIELD: translations, transparent translation (automated, without user intervention and unnoticeable for user) of information in network environment to one or several target languages.

SUBSTANCE: source language is recognized using language identifier means or by analysis of information presented in electronic format. Target language is defined by reading user profile. Source and target languages are compared. If they do not match, needed translation is made automatically. Translated information is presented to user in electronic format.

EFFECT: automatic presentation of information for user in preferred language.

9 cl, 5 dwg

FIELD: video telephone translation systems.

SUBSTANCE: video telephone translation system takes call from terminal of calling user and the call is subject to translator's registration spreadsheet to select number of translator's terminal who is able to make translation from language of calling user to language of called user, and connects terminal of calling user, terminal of called user and translator's terminal. Translation system also has function of video image and sound transmission being required among terminals. Sound from translator is transmitted to either calling user or to called user after translator's terminal is recognized. Conversation partner's sound is suppressed or interrupted when sound of translator is detected by means of sonic synthesizer, which provides higher speed and precision of translation service. Calling user doesn't need to look for a translator in advance and to have consultations concerning negotiation with partner before. System and method are also available urgently.

EFFECT: minimized over-employment of translator; improved efficiency of operation.

20 cl, 22 dwg

FIELD: method and device for using translation data segment.

SUBSTANCE: information contains read segment of translation data, which includes global translation parameters for producing translation of aforementioned information to target language. Translation data are analyzed to determine translation parameters used during translation of information. Translation data preferably includes at least one of the following: information translation pointer, or pointer at translation device for translating information, or element pointing at individual translator who is a specialist in the area of translation of aforementioned information.

EFFECT: possible transparent translation of information in network environment.

2 cl, 3 dwg

FIELD: technology for supporting a set of languages on Web-server for inbuilt systems.

SUBSTANCE: web-server contains source files, meant for browsing by remote browsers, located on one or more terminals, having direct or mediate connection to web-server. In accordance to method, files are browsed to detect unique text phrases depending on language; one unique identifier is generated for each unique text phrase and each text phrase is replaced with unique identifier appropriate for it in new files, each one of which matches one of source files; text phrases are transferred to required languages and translated text phrases are saved in text phrase files; on receipt of request for original file, appropriate new file is browsed to detect unique identifiers, and appropriate text phrases are extracted from files, translated to language of browser used by requesting terminal.

EFFECT: lesser memory space required for storing elements of Web-server related to content.

9 cl, 4 dwg

FIELD: technical means of informatics and computer science, possible use for solving problems of automated comparison and analysis on basis of transformation of non-structured stream of input data to object form.

SUBSTANCE: system contains user control device, device for storage and checking of certificates and centralized control device, consisting of device for registration of system user, decoding block, encoding block, device for controlling operation process, device for storage, classification and searching for ordered data, device for forming and storing data about system users, device for registration of ordered data and device for controlling and analyzing operation.

EFFECT: expanded functional capabilities and increased level of protection from unsanctioned access to system.

8 cl, 11 dwg

FIELD: information technology.

SUBSTANCE: system for communication between people includes a sending computer system, receiving computer system and a web-site for translation. The sending system is capable of detecting the first natural acceptable language associated with sender and the second natural acceptable language associated with recipient. Languages being different, the system is capable of sending the information into translation system for translation. The address of the receiving system is sent together with the above information. The information is translated and sent by the translation system to the recipient. The first and the second languages coinciding, the information is sent directly to the recipient. The recipient compares language of the incoming information with the recipient language and, if needed, sends the information to the translation web-site for translation and rollback.

EFFECT: wider possibilities of transferring messages for translation.

56 cl, 9 dwg

FIELD: information technology.

SUBSTANCE: present invention relates to translation from one language to another. Wireless mobile device service is provided with a service for selective downloading information from the server. The downloaded information includes the architecture of the translation, which has means of translating, independent of language, and at least one language dependent translation database. The translation data base which depends on the language includes translation templates and dictionary translations. It can also load specialised database for the chosen city or cities in the world. Translation between languages is carried out through the application of translation database in the means of translating which depends on the language and translation database, specific to a particular city. For simplification of translation process the means of translation implements a user controlled scheme for replacing terms.

EFFECT: invention increases quality and convenience of text translation in a natural language.

24 cl, 5 dwg

FIELD: physics; computer engineering.

SUBSTANCE: invention relates to machine translation. The method of providing information to an automatic machine translation system for increasing translation accuracy involves reception of initial text. A sample translation corresponding to the initial text is received from the automatic machine translation system. Correctional information is also used, which is configured to correct at least one error in the sample translation. At the final step the automatic machine translation system is provided with information for reducing probability of repetition of the error in the next translations made by that system. The said information is embedded in the automatic machine translation system.

EFFECT: invention increases machine translation accuracy.

38 cl, 11 dwg