JP5473533B2 - In-context exact (ICE) match - Google Patents

Description

  The present invention relates generally to content processing, and more specifically to content content-related source content to simplify or otherwise facilitate content-related translation and other processing functions. Relating to ensuring accurate translation matching.

  The role of translation has changed away from simple handwriting from source text to target language, as information becomes more accessible on a global basis, especially given the emergence and quick use of the Internet and the World Wide Web. It was. Translators must now ensure timely and accurate deployment of translated content to designated locations and customers. Thus, the growing demand for content translation has urged many companies to develop tools to automate and help part of the translation process. Assuming that the translator seeks to translate the content as soon as possible, the translation becomes more efficient, with greater flexibility in software functionality and ability to save previous translations for future use. can do. Accordingly, tools have been created that store translations, including translation blocks and / or segments (“translation memories” or “TMs”), in computer memory.

  A translation memory, also known as a translation database, is a collection of entries in which source text is associated with corresponding translations in one or more target languages. The translation memory includes a database that stores the source / target language pairs of text segments that can be retrieved for use with current text and text that will be translated in the future. Typically, TM is used in a translation tool, and when a translator “opens” a segment, the application looks up the database for equivalent source text. The result is usually a list of matches ranked using a score that represents the percentage of similarity between the source text in the document and the source text in the TM. A translator or a different TM system provides a target text segment that is paired with a lookup segment to ensure that the final result is a high quality translation.

  A number of computer-aided translation (“CAT”) tools are available to assist translators, including bilingual and multilingual dictionaries, grammar checkers, spell checkers, and terminology software. At the same time as using other CAT tools, we go one step further by matching the original source document stored in the database with the updated or revised document via exact and fuzzy matching. An exact match (100% match) is a match where there is no difference between the source text in the document and the source text in the TM (or there is no difference that cannot be processed automatically by the tool). A fuzzy match (less than 100% match) is a match where the source text in the document is very similar to the source text in the TM but is not exactly the same. Replicated exact matches are often treated as fuzzy matches. The TM system is used to help translators and stores human translator text in a database for future use. For example, a TM can be utilized when a translator translates the original text and stores the paired source and target segments using a translation memory. The translator can then reuse the stored text to translate a revised or updated version of the text. Only segments of the new text that do not match the old text must be translated. An alternative is to translate the original text using a manual translation system or a different CAT system. The TM system is then used by the translator to revise or update the text produced by the translator or other CAT system, aligning them and storing them in the TM database for current and future work. Can be translated. The translator can then proceed to translate only the new text segment using the TM as described above.

  There are numerous benefits to using TM. The translation can proceed much faster, avoid unnecessary re-entry of existing translations, and / or allow the translator to change only certain parts of the text. TM also allows better control of the quality of translation. In the related art, TM has been used to speed up translation steps in very large projects. For example, a software company may need to publish version 1 of its software product and translate the accompanying documentation. The documentation is broken down into sentences and translated with all sentence pairs taken into the TM. Two years later, the company releases version 2 of the software. Although the documentation has changed significantly, there are significant parts that are similar to the original documentation. This time, when the translator translates the documentation, the translator's work is reduced through taking advantage of exact and fuzzy matches from the TM. As shown by this example, TM is typically used as an aid within the pipeline process. The related art also has some limitations on the use of TM.

  Automatic use of translations that use exact matches (without their validation) can produce incorrect translations. This is because there is no verification of the context in which the new segment is used compared to the context in which the original segment was used. This is the difference between reuse and recycling. In the related art, the TM system is a recycling system. For web content, and here for many types of content, it is common for documents to be translated, after which minor changes are made, and then translated again. For example, a web document listing product benefits may be translated, and then new benefits may be added, so the document will need to be translated again. In the related art, TM reduces the effort of translating a document a second time. If the source text is identical to one or more entries in the TM, there is an exact match for most sentences. The translator then verifies that the correct exact match was selected for each by evaluating the suitability of the match for the contextual information. However, the related art does not provide content context determination. In addition, in the related art, automated to select the best exact match for a given segment correctly, or to validate whether a given exact match is an appropriate match for the context in which it is applied. There is no process. Therefore, the translator is required to validate the match. The fact that translators need to perform an action on every sentence when it is validated and perhaps only a few words may have been changed Or it is significantly inefficient considering that it can be translated differently under context.

  In view of the foregoing, there is a need in the art for an automated process that accurately validates whether a given exact match is an appropriate match with respect to the context in which it is applied.

According to a first aspect of the invention, there is provided a method for determining a matching level of a plurality of source texts stored in a translation memory for a lookup segment to be translated, comprising:
Determining all exact matches of lookup segments in multiple source texts;
If at least one exact match is determined, each exact match is an in-context exact (ICE) match of the lookup segment if the context of the lookup segment matches the context of the respective exact match Determining that the context includes at least two levels;
A method is provided in which at least two levels include a source usage context level and a target usage context level. Thus, if multiple exact matches are found when the translation memory is searched for the occurrence of a particular lookup segment, the present invention allows disambiguation between multiple exact matches. Disambiguation is performed according to the context of the lookup segment compared to the respective context of the exact match. A segment is ICE matched if at least one context level of a segment matches the context level of the lookup segment. ICE matches tend to be more relevant than non-ICE matches. ICE matches can be identified according to different levels of context, such as a source usage context level of the source language and / or a target usage context level of the target language.

  In an embodiment of the present invention, when multiple ICE matches are determined, the step of determining an ICE match is to allow a more appropriate ICE match to take precedence over one or more other ICE matches. , Prioritizing each ICE match according to the degree of context matching. Thus, the present invention allows disambiguation of multiple exact matches, where higher priority ICE matches tend to be more relevant translations than lower priority ICE matches.

  In another embodiment of the present invention, an ICE match having both a source usage context level and a target usage context level that match both the source usage context level and the target usage context level of the lookup segment is the lookup segment. Suppose that there is a higher degree of context matching than an ICE match with only a source usage context level that matches the source usage context level. The combined matching source usage context and target usage context tend to show more relevant matches than only the matching source target usage context, so the former takes precedence over the latter.

  In a further configuration of the invention, an ICE match having a target usage context level that matches the target usage context level of the lookup segment is higher than an ICE match having only a source usage context level that matches the source usage context level of the lookup segment. It is assumed that there is a degree of context matching. Target usage contexts tend to show more relevant matches than only matching source target usage contexts, so the former is given higher priority than the latter.

  In one configuration of the invention, the source usage context level includes a previous source usage context level and / or the target context usage level includes a previous target usage context level. The translation of the segment that precedes the segment currently being translated in the source language (current segment) and the segment that precedes the current segment in the target language are translated to a greater impact than the source and target segments that follow the current segment. There is a tendency to have a translation of an incoming segment, so the context of the segment preceding the current segment is considered important for disambiguation purposes.

  In other configurations of the present invention, the source usage context level includes a post-source usage context level and / or the target context usage level includes a post-target usage context level. The translation of the segment following the currently translated segment of the source language (current segment) and the segment following the current segment of the target language may have an impact on the translation of the currently translated segment, and therefore The context of the segment following can be considered either in addition to or instead of considering the context of the segment preceding the current segment.

  In some embodiments of the invention, at least two levels include structural context levels. (See the “Definitions” section in the detailed description below for the meaning of structural context). Thus, the structural context level can be used in ICE match disambiguation in addition to the usage context level.

In the configuration of the present invention, when multiple ICE matches are determined, the step of determining an ICE match may allow a more appropriate ICE match to take precedence over one or more other ICE matches. Prioritizing each ICE match according to the degree of context matching,
The source usage context level includes a preceding source usage context level, the target context usage level includes a preceding target usage context level,
An ICE match having a preceding source usage context level and / or a preceding target usage context level that matches the preceding source usage context level and / or the preceding target usage context level of the lookup segment matches the structural context level of the lookup segment It is assumed that there is a higher degree of context matching than an ICE match with only structural context levels. Translations that use matching preceding source usage context levels and preceding target usage context levels tend to be more relevant than translations that use only structural context matching, so the former is given higher priority than the latter.

  In an embodiment of the present invention, the step of determining an ICE match is such that each exact match is an ICE match of the lookup segment only if the multiple context levels of the lookup segment match the context level of the respective exact match. Indicates that An ICE match that matches a lookup segment at multiple context levels is more likely to be a better match than an ICE match that only matches a lookup segment at one context level, and thus some of the present invention In this embodiment, only ICE matches with multiple context level matches are considered for disambiguation, i.e., in such embodiments of the invention, multiple exact matches are only one context level. Is not disambiguated if it matches the context level of the lookup segment (eg, structural context only or source usage level only).

  In an arrangement of the present invention, the step of determining an ICE match is when the at least one usage context level and the structural context level of the lookup segment match at least one usage context level and a structural context level of each exact match. Only indicates that each exact match is an ICE match of the lookup segment. An ICE match that has both a usage context level and a structural context level match tends to be an associated translation of the look-up segment, so such an ICE match is considered important during disambiguation .

  In another configuration of the invention, the at least one usage context level comprises a preceding target usage context level. The segment that precedes the currently translated segment of the target language is generally a very good indication of the associated translation, and therefore, the preceding target usage context level match is preferentially prioritized. In a preferred embodiment of the present invention, disambiguation is performed on the basis of a target usage context level combined with a source usage context level.

  In yet another configuration of the invention, the at least one usage context level includes a preceding source usage context level. The segment that precedes the currently translated segment of the source language is generally a good indication of the associated translation, and therefore prior source use contact level matches can be given priority instead.

  In an embodiment of the present invention, when multiple ICEs are determined to match, the step of determining an ICE match may allow a more appropriate ICE to take precedence over one or more other ICE matches. Including prioritizing each ICE match according to the degree of context match. Prioritizing may include a plurality of different prioritizing steps that can be considered sequentially to find the most appropriate match for the lookup segment. The first step gives first priority to ICE matches having a source usage context level, a target usage context level, and a structural context level that match the source usage context level, the target usage context level, and the structural context level of the lookup segment. Steps may be included. The second step can include second priority over ICE matches having a source usage context level and a target usage context level that match the source usage context level and the target usage context level of the lookup segment. Preferably, the source context level is a preceding source usage context level and the target usage context level is a preceding target usage context level. The third step is that the look-up segment precedes the ICE match that has only either the post-source usage context level or the post-target usage context level that matches the post-source usage context level or the post-target usage context level. A third priority may be included for ICE matches that have either the previous source usage context level or the previous target usage context level of the lookup segment that matches the source usage context level or the previous target usage context level. The fourth step determines the fourth ICE match with any usage context level that matches the usage context level of the lookup segment, rather than the ICE match with only the structural context level that matches the structural context level of the lookup segment. Overriding steps may be included. The fifth step includes fifth priority over ICE matches having a structural context level that matches the structural context level of the lookup segment over ICE matches having a different structural context level of the lookup segment. Can do. The sixth step may include a sixth priority over an ICE match having a position closest to the position of the lookup segment in the asset.

  The location within the asset can be determined on the basis of the segment number, where the segments within the asset can be numbered sequentially and the number can be used as the respective segment identifier. Alternatively, the position can be determined based on the line number or word number in the asset. In addition, the structural usage context is based on the position in the asset, that is, based on whether a match occurs within a “heading”, “table cell”, “paragraph”, “footnote”, etc. It can be interpreted as indicating the correct position.

  If a preferred step yields a definitive result, the disambiguation procedure ends; otherwise, the procedure proceeds to successive steps to determine the definitive result in the form of a preferred match. The same is true until it is reached. Other embodiments of the present invention may include some or all of the above preferred steps in a different order, and either other instead of or in addition to the above preferred steps. It may be combined with a step.

  Embodiments of the invention further include allowing the user to select ICE matches based on prioritization. Thus, an ICE match can be displayed to the user rather than automatically selected and inserted into the translated document, and the user can select an ICE match based on the displayed rank of the ICE match. You can choose. Further, ICE matches can be ranked and displayed according to the number of matching context levels and / or the type of matching context levels (source, target, later, predecessor, structure, etc.).

  In an embodiment of the present invention, the lookup segment includes a plurality of lookup segments that are substantially identical in content, and determining the ICE match includes determining an ICE match for each lookup segment. Thus, multiple segments can be processed together in batch or substantially simultaneously to reduce the overall processing required. Applying a heuristic-based approach to help select the “best” translation of a segment so that the next segment's ICE level can be determined when a batch of segments to be translated is processed Yes (when the target usage context is taken into account). One such approach may be the use of best ICE matches or single exact matches, optionally combined with a fallback mechanism such as segment location, last translation date, asset metadata, etc. The heuristic-based approach is useful in scenarios where there is little or no interaction with the translator, in which case the translation system makes sure that the disambiguated matches (ICE matches or other matches) are adequate for the segment I can't be sure if it is a good match.

  In some configurations of the present invention, at least one lookup segment has a different ICE match than at least one other lookup segment. Typically, one lookup segment has a different ICE match, which can be useful in the process of identifying an ICE match for each lookup segment.

According to a second aspect of the invention, a system for determining a matching level of a plurality of source texts stored in a translation memory for a lookup segment to be translated,
A means of determining all exact matches of lookup segments in multiple source texts;
If at least one exact match is determined, determine that each exact match is an in-context exact (ICE) match of the lookup segment if the context of the lookup segment matches the context of each exact match And the context includes at least two levels,
A system is provided in which at least two levels include a source usage context level and a target usage context level.

According to a third aspect of the present invention, a method for storing translation units of source text and target text in a translation memory comprising:
Assigning a context to a translation unit, the context comprising at least two levels, at least two levels comprising a source usage context level and a target usage context level;
Storing a context with the translation unit.

According to a fourth aspect of the present invention, a system for storing translation units of source text and target text in a translation memory,
Means for assigning a context to a translation unit, wherein the context comprises at least two levels, at least two levels comprising a source usage context level and a target usage context level;
And a means for storing the context along with the translation unit.

According to a fifth aspect of the present invention, a plurality of source texts for comparison with a lookup segment;
A context identifier for each source text,
And a translation memory is provided in which the context identifier includes a source usage context portion and a target usage context portion.

  In some embodiments of the invention, the source usage context portion and / or the target usage context portion comprises a pre-use context portion and / or a post-use context portion. In other embodiments of the invention, the context identifier comprises a structural context portion. Thus, context identifier data can be stored in association with the translation unit in the translation memory so that the context associated therewith can be easily identified, such as during retrieval of the lookup segment from the translation memory. A translation unit may have multiple usage context identifiers and / or multiple structure related context identifiers.

According to a sixth aspect of the present invention, there is a client-side system for interacting with a translation system including a translation memory,
Means for assigning a segment identifier to a segment translated by a translation system, wherein the segment identifier indicates a source usage context and a target usage context of the segment;
Means for communicating segment identifier assignments for storage as part of a translation memory.

  In the arrangement of the invention, the segment identifier comprises a structural context. Thus, segments can be identified according to their structural context.

According to a seventh aspect of the present invention, there is provided a method for determining a matching level of a plurality of source texts stored in a translation memory for a lookup segment to be translated,
Determining all fuzzy matches of lookup segments in multiple source texts;
Determining that each fuzzy match is an in-context fuzzy match of the lookup segment if the context of the lookup segment matches the context of the respective fuzzy match if at least one fuzzy match is determined; Is provided.

  Thus, the present invention also allows disambiguation between multiple fuzzy matches. Disambiguation is performed according to the context of the lookup segment compared to the respective context of the fuzzy match. A segment is an in-context fuzzy match if at least one context level of a segment matches the context level of the lookup segment. In-context fuzzy matches tend to be more relevant than non-contextual fuzzy matches. Preferably, the context includes a structural context.

  In-context fuzzy matches can be identified according to structural context levels and / or different levels of context, such as metadata. Similarly, different context-level prioritization can be applied to ICE matches at the time of disambiguation between in-context fuzzy matches.

  The present invention can be applied to fuzzy matches having the same degree (also referred to as fuzzy matching score) fuzzy matching, eg, the same percentage of fuzzy matches. One example application may be disambiguation between multiple fuzzy matches, each having the same degree of fuzzy matching of 97%. Another example application may be disambiguation between multiple fuzzy matches, each having the same degree of fuzzy matching of 99%. The term “identical” herein can be exactly the same or nearly identical, for example, a fuzzy match that is the same when rounded to the nearest percentage value or such. be able to.

  Instead of or in addition to prioritizing fuzzy matches with the same degree of fuzzy matching, prioritize fuzzy matches within a certain range (or score band) of fuzzy matching, eg 99 to 95% Or between 85-94%, where prioritization can be performed using structural contexts such as headings, paragraphs, table cells, etc.

  The present invention can also be applied to fuzzy matches with different degrees of fuzzy matching, and a first fuzzy match (eg, 98% fuzzy match) with a lower degree of fuzzy matching than a second fuzzy match (eg, 99% fuzzy match). Match) is still a better match for the lookup segment because the second fuzzy match does not have a context level that matches the lookup segment, but the first fuzzy match has a context level that matches the lookup segment be able to. This can occur despite the first fuzzy match having a lower degree of fuzzy matching fuzzy than the second fuzzy match.

  The invention can also be applied to hybrid scenarios where both exact and fuzzy matching contexts are considered. Here, for example, 99% of the fuzzy match is a better match than the exact match because the exact match does not have a context level that matches the lookup segment, but the fuzzy match has a context level that matches the lookup segment. Can think. Again, different levels of context matching and different context level prioritization can be applied to such hybrid configurations.

According to an eighth aspect of the present invention, there is provided a system for determining a matching level of a plurality of source texts stored in a translation memory for a lookup segment to be translated,
Means for determining all fuzzy matches of lookup segments in multiple source texts;
Means for determining that each fuzzy match is an in-context fuzzy match of the lookup segment if the context of the lookup segment matches the context of the respective fuzzy match when at least one fuzzy match is determined; Is provided.

  According to a ninth aspect of the present invention there is provided computer software adapted to carry out the method according to any one of the first, third or seventh aspects of the present invention.

  The foregoing and other features of the invention will become apparent from the following more specific description of embodiments of the invention.

  Embodiments of the present invention will be described in detail with reference to the drawings, in which like designations represent like elements.

1 is a block diagram showing a computer system using an ICE matching translation system according to the present invention. FIG. 2 is a flow diagram illustrating one embodiment of an operational methodology for the system of FIG. 2 is a flow diagram illustrating one embodiment of an operational methodology for the system of FIG. FIG. 3 illustrates two entries in an exemplary translation memory. FIG. 4 illustrates an example source asset that includes the entry of FIG. 6 is a flowchart illustrating an embodiment of translation memory generation according to the present invention. 6 is a flow diagram illustrating an alternative embodiment of translation memory generation according to the present invention.

  This detailed description is for convenience only and includes the following headings: Definition, II. General overview, III. System overview, IV. Operating methodology, and Includes conclusions.

I. Definitions “Asset” means a content source that defines a constrained set of related content or grouping of text segments, eg, by context, usage, size, etc. In general, an asset is associated with a document, such as a hypertext markup language (HTML) file, a Microsoft® Word® document, or a simple text file. However, some assets do not correspond to file system files. Assets can actually be defined from a column in a database table or a structure in an extensible markup language (XML) repository. Regardless of how they are physically represented, all assets share a common purpose: defining a constrained set of related content that can be accessed, manipulated, and ultimately translated To do. Assets can include content, formatting information, and internally structured data that depends on the nature of the asset.

  “Source asset” refers to the asset from which the lookup segment is derived.

  A “segment” includes translatable chunks of content, such as phrases, sentences, paragraphs, and the like. A segment represents the minimum unit of translation work. In practice, a segment can represent a paragraph, a sentence, or even a sentence fragment. A segment is usually not a single word, but single word segments can be used.

  “Source text” refers to the text in the translation memory corresponding to the original (source) language that is the language to be translated. The source text is compared to the lookup segment from the asset during the match lookup process to find a match.

  “Target text” includes the translation of the source text for a particular locale, ie, the target text is half the translation memory (TM) unit.

  A “translation memory” (abbreviated TM) includes a repository containing TM entries. A TM can contain TM entries for any number of locales. For example, the TM may include entries such as English to French, Greek to Russian, Albanian to Turkish, and the like.

  A “TM unit” is a translation pair stored in a translation memory that maps source text to target text (also known as translation unit, TM entry, or record). A TM is specific to a given translation pair, which includes a source text locale / target text locale pair, usually associated with the asset that made the translation pair. In effect, TM units represent previous translations that can be reused later. In addition, each TM unit according to the present invention includes a context portion that identifies the context of the associated source text / target text pair. A translation unit is usually bilingual, but can be in multiple languages and can include metadata such as context information, usage counts, creation dates, and the like.

  “Exact match” means source text that includes source text that is exactly the same as the look-up text from the instant asset coming from the translation memory. As used herein, an exact match also includes a 100% match, which is similar to an exact match, but does not necessarily result from an exact match because of the differences that exist within the translation memory unit. is not. For example, a match can be: 1) the difference between unscored white space characters, ie the use of a different type of white space character from TM-based white space prevents it from being selected as an exact match, 2) A configured penalty that effectively ignores certain differences between the source text and the lookup text, or 3) via it to correct the difference between the TM match and the original lookup text Because of one of the reasons for segment repair that can apply repair heuristics, a match can be scored as 100% without being an exact match. Segment repair can be performed, for example, by use of placeable elements, automatic replacement techniques, and automatic localization techniques described in US Pat. No. 7,020,601.

  “Context” means a discourse that surrounds a text segment and helps determine its interpretation. Context can include different levels as used herein. For example, the context can include a usage context level, an asset context level, a structural context level, and / or a segment identifier. Each different context may require different language depending on the intended audience of the content.

  “Usage context” refers to a discourse that surrounds a segment and affects how the present invention derives an appropriate translation of the content. The usage context may be a source usage context with consideration of the source language text surrounding the translated text. The usage contact may be the target usage context with consideration of the target language text surrounding the translation of the text to be translated. Usually, the usage context is defined with surrounding content that provides insight into the meaning of the segment being translated. The usage context can have a level with respect to the text preceding the specific segment and the text following (after) the specific segment, with respect to whether the context relates to the text surrounding the translated text in the source or target language Can also have a level. Combining the pre-use context level and post-use context level with the source use context level and the target use context level, the forms of the pre-source use context level, pre-target use context level, post-source use context level, and post-target use context level To create additional context levels.

  “Asset Context” refers to the discourse about the asset environment in which the segment resides, ie, the background and perspective framework of the overall content in which the text segment appears.

  "Structural context" refers to a document structural unit in which a segment appears, such as a paragraph, heading, table of contents item, list, or table cell.

  “In-context exact (ICE) match” for a lookup segment means that the source text is an exact match and must share at least one context level with the TM unit that produces the match.

  “Lookup text” refers to a segment of text from which a source asset is searched for TM matches.

  A “segment identifier” (SID) includes a label that defines the usage context in which a given segment must be translated and is associated with the content when the TM unit or content is created. The SID provides context identification for a given segment. The SID can include a marker tag that defines a segment boundary. As described below, a SID is an alternative or extension of being based on usage context or surrounding segments.

II. General Overview The present invention provides, among other things, a method, system, and program product for determining a matching level of a plurality of source texts stored in a translation memory to a translated lookup segment. The present invention produces a high quality match for source content from previously stored translations in a translation memory (TM). In the related art, the best match available was an exact match, i.e. a match where the source text was identical to one or more entries in the TM. However, there is no automated process that accurately selects the best exact match for a given segment or verifies whether a given exact match was an appropriate match for the context to which it is applied. Specifically, a segment may be translated differently under different circumstances or contexts. Appropriateness of exact matching requires evaluation of contextual information, which depends on content usage (defined by the sentence or segment surrounding it), structural context, asset context (depending on the intended audience) And / or metadata such as domain identifiers, customer names, or other supplied or automatically generated or extracted from previous translation projects It can be based on information.

  Region identifiers may relate to a wide range of subjects where exact matches are related to “Electrical Engineering”, “Agriculture”, “Computer Science”, and the like. Taking the word “bank” as an example, this can be disambiguated if the region identifier indicates a financial region rather than a geographic region.

  The present invention does not replace the exact matching process. Rather, the present invention reduces the need for manual validation and creates a new match-up beyond exact match by using a true reusable TM system that helps people in creating the most valuable TM possible. Provides a level. Specifically, an embodiment of the present invention determines all exact matches of a lookup segment within a plurality of source texts, and if at least one exact match is determined, In a translation memory to a translated lookup segment by determining that each exact match is an in-context exact (ICE) match of the lookup segment if the context matches the context of the exact match A matching level of a plurality of stored source texts is determined. Thus, ICE match determination determines the appropriateness of an exact match based on the context of the lookup segment. Source text that is an exact match and has a matching context is referred to as an “in-context exact (ICE) match”. An ICE match is considered superior to an exact match in that it ensures that the applied translation is appropriate with respect to the context in which it is used. An ICE match is a translation match that guarantees a high level of relevance thanks to a previously translated match in the same context as the currently translated segment.

  The present invention 1) determines the appropriateness of an exact match as a high quality (no re-examination) match for new content, 2) selects the best contextual match for a given lookup segment, 3) Context information is used to ensure that previously assigned translations of previously translated content are always recoverable. For new content, the present invention utilizes context information to find a high quality match from the TM based on the segment usage context. The present invention also ensures that the same content is always translated in the same way given the context at the asset and / or structure and / or context level.

  Assume that for a given lookup segment translation, the source document is translated and all segments are stored in the TM. If the exact same source document is passed again, it will be an exact match, including all of its contents, and the resulting translated document will be exactly the same as the originally translated document. become. This behavior is simple and expected. However, this can only be guaranteed as a result of using contextual information. To further understand the importance of this guarantee, consider a source document in which the same exact sentence is repeated twice in two different places. The second use is translated differently because it may not have the same implicit intention as the first use. Here again, consider that the same document is utilized for TM. Should the duplicated sentence have the same translation or should it be as different as the original document? Without taking into account the context of their use, these sentences are most likely to be translated in the same way by the TM. However, according to the present invention, context is taken into account, whereby the two sentences continue to have different translations as long as their usage context dictates so.

  With respect to the restoration of previously translated text segments, the present invention also ensures that the translation of a new document does not affect the ability to restore the translation of a previously translated document. Provides a way to ensure that the translation is fully reproducible by ensuring that the document yields the same translation as that stored by the translator. Once again, consider two identical documents. The first document is translated and the result is stored in TM. When a second document is leveraged for the same TM, the document is represented as fully translated using ICE matching. The usage context is the same as the usage context of the first document. Next, consider that the second document targets a different audience. The source language text is not modified in this example because it is considered appropriate for both audiences. However, translation into the target language requires some changes. The translator updates the translation for this document and stores the result in the TM. Over time, copies of both translated documents are requested again. For reasons of space, the original translated document was deleted. None of the source documents have changed, and therefore still contain the same source text. The present invention facilitates the reproduction of original translated documents, each identical to the originally translated document (which were not themselves identical). Despite the same source document content, the present invention can leverage asset context information to ensure that document specific translations are recoverable.

  The present invention can be utilized as part of a content management system such as Idiom WorldServer ™ or as a separate system. For example, WorldServer ™ is a web-based application that enables companies to leverage content in the context of the entire globalization process while leveraging established web architecture, content management, and workflow systems. . The content management system simplifies the multiple complexities that arise, for example, from the deployment of a global web strategy, and allows a company website to efficiently support multiple countries as well as different languages, locations, and cultures. The content management system provides structure and processes for collaboration between site administrators, web developers, content owners, translators and editors, streamlined processes, synchronized global web strategies, and Bring a coordinated global web team. The translator uses the content management system to see what the translator needs to translate. With WorldServer ™, translators can either export content that requires translation to a third-party editing tool or use a translation workbench to perform the actual translation. Translators can be individual contributors, including users who have adapted content but have not translated and / or reviewers that review the content. The content management system stores the translated phrase in the TM for later recall.

III. System Overview Referring to the accompanying drawings, FIG. 1 is a block diagram of an in-context exact match translation system 100 according to the present invention. Although the system 100 is illustrated as a separate system, it should be understood that this can be implemented as part of a larger content management system or translation system, such as Idiom's WorldServer ™. In this regard, the description of system 100 includes certain functionality of the translation system, but other functionality may be omitted for clarity of explanation. Further, although system 100 is illustrated in a client-server (eg, web-based) environment, it should be appreciated that other configurations are possible.

  System 100 is illustrated implemented as computer program code on a computer 102. To this extent, the computer 102 is shown including a memory 112, a processing unit 114, an input / output (I / O) interface 116, and a bus 118. Further, computer 102 is shown in communication with external I / O devices / resources 120 and storage system 122. In general, processing unit 114 executes computer program code, such as system 100, stored in memory 112 and / or storage system 122. While executing computer program code, processing unit 114 may read and / or write data to / from memory 112, storage system 122, and / or I / O device 120. Bus 118 provides a communication link between each of the components in computer 102, and I / O device 120 can be any device that allows a user to interact with computer 102 (eg, a keyboard, pointing device, Display, etc.).

  Alternatively, the user can interact with another computing device (not shown) that communicates with computer 102. In this case, the I / O interface 116 may be any device (eg, network system, network adapter, I / O) that allows the computer 102 to communicate with one or more other computing devices over a network. O port, modem, etc.). A network can include any combination of various types of communication links. For example, the network can include addressable connections that can utilize any combination of wired and / or wireless transmission methods. In this case, the computing device (eg, computer 102) can take advantage of conventional network connectivity, such as token ring, Ethernet, WiFi, or other conventional communication standards. Further, the network can include one or more of any type of network, including the Internet, a wide area network (WAN), a local area network (LAN), a virtual private network (VPN), and the like. When communication occurs over the Internet, connectivity can be provided by conventional TCP / IP socket-based protocols, and the computing device can establish an Internet service provider to establish connectivity to the Internet. Can be used.

  Computer 102 only represents various possible combinations of hardware and software. For example, the processing unit 114 can include a single processing unit, or can be distributed across one or more processing units at one or more locations, eg, on a client and a server. Similarly, the memory 112 and the storage system 122 can reside in one or more physical locations. Memory 112 and storage system 122 may be any combination of various types of computer readable and / or transmission media including magnetic media, optical media, random access memory (RAM), read only memory (ROM), data objects, and the like. Can be included. The I / O interface 116 may include any system that exchanges information with one or more I / O devices. In addition, it should be appreciated that one or more additional components not shown in FIG. 1 (eg, system software, math coprocessing units, etc.) can be included in computer 102. To this extent, the computer 102 can include any type of computing device, such as a network server, desktop computer, laptop machine, handheld device, mobile phone, pager, personal digital assistant, and the like. However, if computer 102 includes a handheld device or the like, one or more I / O devices (eg, a display) and / or storage system 122 may be located within computer 102 rather than externally as shown. It should be understood that

  As described further below, the system 100 includes an exact match determiner 130, an in-context exact (ICE) match determiner 132, a hash algorithm 133, a fuzzy match determiner 134, a translation memory TM generator 136, a segment retriever 138, and other systems. A component (Sys. Comp.) 140 is shown. The ICE match determiner 134 includes a context identifier 142, a match evaluator 144, and an ICE match prioritizer 146. Other system components 140 may include other functionality necessary for the operation of the present invention, but are not explicitly described herein. For example, other system components 140 may include automated translation systems and / or content management system functionality such as that provided by Idiom's WorldServer ™.

  Although not shown in FIG. 1 for clarity, it is understood that the client-side system 150 can include a structure similar to the computer 102 and includes program code that provides the functionality described below. I want.

  Also shown in FIG. 1 is a translation memory 128 (hereinafter “TM 128”) for use by the system 100. As shown in FIG. 3, TM 128 includes a plurality of TM entries 148 including stored target texts 162, 164 previously translated into a specific language for specific source texts 152A, 152B, respectively. (Only two are shown for simplicity of illustration). For example, the stored source text “global enterprises” 152A has been translated into multiple German translations 162 or target text, and the stored source text “team of visionaries” 152B has multiple French translations 164 or target texts. Has been translated into Each source text 152A and 152B is for comparison with a lookup segment. Each TM unit 148 also includes a context identification 166 (labeling only two for clarity of illustration). In one embodiment, context identification 166 includes an indication of different context levels, such as source usage context portion 168 and asset context portion (AC) 170. The asset context portion 170 includes an asset code that identifies a particular asset to the system 100, such as “33333”, for example. Other context levels can be provided, such as a structural context level. In some cases, asset context portion 170 may be omitted. In this embodiment of the invention, the usage context levels are a pre-source usage context level and a post-source usage context level. In other embodiments of the present invention, the usage context level may alternatively or in addition include a pre-target usage context level and / or a post-target usage context level.

  In one preferred embodiment, each usage context 168 includes a preceding source usage context (UC) hash code 172 and a subsequent source usage context (UC) hash code 174. The preceding UC hash code 172 is generated using a hash algorithm 133 based on the text stream generated by the preceding segment in which each source text appears next to it during translation. Similarly, the post-UC hash code 172 is generated using a hash algorithm 133 based on the text stream generated by the subsequent (post) segment in which each source text appears next to it during translation. Hash algorithm 133 includes any currently known or later developed hash algorithm that can convert a text stream into a unique numeric identifier (the illustrated hash code has been simplified for clarity of illustration). ) Thus, each UC hash code indicates a unique usage context level for each source text. In an alternative embodiment, only one usage context hash code can be used for a particular source text 152 and the preceding and following segments.

  If the lookup segment 154 is assigned to a context at creation time, the context identification 166 is generated using the user-specified SID described above, rather than the hash algorithm 133 or a combination of this and the use of a user-specified SID. Here, one of the hash algorithm 133 and the user-specified SID can be given priority over the other.

  It should be understood that the specific codes used herein are for illustration only.

IV. Operational Methodology Turning now to FIGS. 2A-B, a flowchart of one embodiment of the operational methodology of the present invention will now be described. The discussion of FIGS. 2A-B takes place in conjunction with FIGS.

A. Preliminary Steps Beginning with FIG. 2A, as a preliminary step PS, in one embodiment, a look-up segment 154 is directly linked or some type of network, eg, a client computer system 150 linked by a LAN, WAN, or the Internet. Is loaded into the ICE matching system 100. For example, lookup segment 154 may be loaded via a translation workflow application server (not shown), such as Idiom's WorldServer ™, which ICE matching system 100 may be part of. it can. Lookup segment 154 can be loaded as part of a larger asset. In this case, the system 100 may segment a larger asset in any form now known or later developed to create a lookup segment 154, for example as provided by Idiom's WorldServer ™. It can be carried out. Segmentation is a process through which asset content is analyzed and published as translatable segments. The size of the segment depends on the segmentation rule, which can be user defined.

B. General Methodology Steps S1-S12 represent an analysis for each lookup segment 154.

  In a first step S 1, all exact matches of lookup segments 154 in the plurality of source texts 152 in TM 128 are determined by exact match determiner 130. The exact match determiner 130 can function in the same way as most conventional TM systems, and the exact match determiner of conventional TM systems measures the appropriateness of translations stored in TM 128 using a string comparison algorithm. Where the score is given based on how well the two strings match. A score of 100% usually indicates that an exact match has been found, or the difference is automatically resolved (eg, “repaired” using placeable elements, auto-localization techniques, or auto-replacement techniques). Indicates the match to be obtained. For example, as shown in FIG. 3, the lookup segment “global enterprises”, when translated into German, has two exact matches: 1) global Wesen, 2) global Internetnemen, and 3) global Gescheefte. Should bring. The look-up segment “team of visionaries”, when translated into French, is 1) e'quipe de visionnaires, 2) group devisionnaires, 3) bande des visionnaires, and 4) group de foulol4 It should be. Multiple exact matches may exist in TM 128 for each lookup segment 154. This is because the meaning of multiple translations and a statement in a given language for any given segment is derived not only from the word but also from the context in which it is used. is there. Thus, each previous translation may result in multiple target text translations 162, 164 of a particular source text 152 and thus the same lookup segment 154.

  In step S2, a determination is made as to whether at least one exact match has been determined, i.e., has been found in TM128. If not, in step S2, processing proceeds to step S3, where fuzzy match determiner 134 has a fuzzy match for lookup segment 154 in any currently known or later developed form. Determine. All fuzzy matches for lookup segment 154 are reported in step S4. “Report”, as used herein, can mean display of results to a user, transfer of results, and / or storage of results. Although not shown, if no fuzzy match is found, conventional automatic translation can be initiated.

  If YES in step S2, in step S5, ICE match determiner 132 determines whether each exact match is an in-context exact (ICE) match of lookup segment 154. As noted above, “ICE match” means that the source text 152 must be an exact match, which also shares a common context with the lookup segment 154. In other words, an exact match with a context identification 166 that matches the context identification of the lookup segment 154 is an ICE match. In one embodiment, the context in this determination includes only the usage context. However, other context matching levels can be used, as described below.

  Step S5 includes two substeps. First, substep S 5 A, context identifier 142 identifies the context of lookup segment 154. In one embodiment, context identifier 142 identifies a context based on the segments surrounding lookup segment 154 in its source asset. In this case, a hash algorithm 133 is implemented to determine the usage context of the lookup segment 154 by calculating a lookup segment (LS) predecessor UC hash code and a lookup segment (LS) post-UC hash code. The Again, the hash algorithm 133 includes any currently known or later developed hash algorithm that can convert a text stream into a unique numerical identifier. Referring to FIG. 4, an exemplary source asset 180 is shown that includes a lookup segment 154A in the form of “team of visionaries”. The LS predecessor UC hash code is formed based on the immediately preceding segment 190. For example, as shown in FIG. 4, the LS predecessor UC hash code is calculated for “Idioms was found in January 1998 by a team of visions”. Similarly, the post-LS UC hash code is calculated for the segment 192 that follows directly, that is, “team of visionaries who recognized the need for an enterprise-class software product that meets the weight.” The asset context of the source asset 180 can be identified by the context identifier 142 based on the asset hash, which is based on the identification of the particular asset's system, eg, asset name, location in the system, and the like.

  In an alternative embodiment, the context identifier 142 identifies the context of the lookup segment 154 according to the segment identifier (SID) associated with the lookup segment 154, which is the segment translated as described above. Contains a label that defines the usage context that must be present. The SID can include a marker tag that defines a segment boundary. Preferably, the SID is associated with the source text 152 and / or lookup segment 154 during segment creation, ie, by the content creator. However, the SID can be associated or overwritten with the source text 152 and / or lookup segment 154 at a later time, for example, by a previous content translator.

  In sub-step S5B, the context identification 166 stored with each candidate to determine whether the ICE match evaluator 144 was used in the same context as the lookup segment 154, ie, whether each exact match is an ICE match. Is used for each exact match of the lookup segment 154. The degree of context matching required for an exact match to be considered an ICE match can be predetermined. In one embodiment, the ICE match evaluator 144 determines that each exact match is an ICE match of the lookup segment 154 only if each context level of the lookup segment 154 matches the context level of the respective exact match. Indicates. For example, if the context includes a usage context level and an asset context level, the decision step can determine that both the usage context level and the asset context level of the lookup segment Only if both match, it can be shown that each exact match is an ICE match of the lookup segment. As a further example, if the context includes a usage context level and a structural context level, the determining step can include determining whether the usage context level and the structural context level of the lookup segment are the exact matching usage context level and structure, respectively. It can be shown that each exact match is an ICE match of the lookup segment only if it matches both the target context level. The usage context level can include one or more of a preceding source usage context level, a post source usage context level, a post target usage context level, and a post target usage context level.

Example Referring to FIG. 3, assume that an exemplary lookup segment 154 includes the text “team of visionaries”, which must be translated into French. Also assume that the lookup segment “team of visionaries” has a LS pre-UC hash code of 333 and a post-LS UC hash code of 4444 and assumes an asset context of 666666 (all hash codes of FIG. (It should be understood that it has been simplified for clarity of illustration). As shown in FIG. 3, the exact match determiner 130 has four exact matches of the look-up segment “team of visionaries” when translated into French: 1) e'quipe devisionnaires, 2) group devisionnaires 3) bandes des visionnaires and 4) groups de futurology should be determined. The ICE match evaluator 144 reviews the exact match and, as shown in FIG. 3, when the lookup segment “team of visionaries” is translated into French, the source text “group de visionnaires” Since it has the asset context 170 and the usage context (hash code) 172, 174, it should be determined that they have the same context. Therefore, “group devisionnaires” should be an ICE match. The other source text should not qualify as an ICE match because it does not have at least one context code for the lookup segment “team of visionaries”.

  In an alternative embodiment, the ICE match evaluator 144 may determine that each exact match matches the ICE match of the lookup segment 154 even if only some context levels of the lookup segment match the context level of each exact match. It can be shown that.

Example Referring to FIG. 3, assume that an exemplary lookup segment 154 includes the text “global enterprises”, which must be translated into German. Also assume that the lookup segment “global enterprises” has a 1234 LS pre-UC hash code and a 4321 post-LS UC hash code and assumes an asset context of 7890. As shown in FIG. 3, the exact match determiner 130 has three exact matches of the lookup segment “global enterprises” when translated into German: 1) global Wesen, 2) global Internetnemen, and 3 ) Should determine global Geschaefte. Assuming that only one usage context level is required for an exact match to become an ICE match, the ICE match evaluator 144 reviews the exact match and, as shown in FIG. When “global enterprises” is translated into German, the source texts “global Wesen” and “global Unternehmen” each have one UC hash code that matches one of the LS UC hash codes, so they have the same context. It should be judged. That is, “global Wesen” has the same previous (preceding) source UC hash code as the look-up segment, and “global Untemehmen” has the same post-source UC hash code as the look-up segment. The other source text should not qualify as an ICE match because it does not have at least one context level for the lookup segment “global enterprises”. Details of how the system 100 prioritizes multiple ICE matches is described below.

  If no ICE match is determined, i.e. NO in step S6, all exact matches are reported in step S7. Thereafter, in step S8, the exact and fuzzy matches, ie the matches from steps S3-4, can be validated by the user in any currently known or later developed form. In this case, exact matches and fuzzy matches are retrieved into their respective caches and made available to the translator by the client computer system 150, where the translator is given the source asset 180 content and each exact match. Such an exact match must be validated and each fuzzy match updated to match the source asset 180 content to ensure that is the best match.

  If an ICE match is determined, ie, YES in step S6, as shown in FIG. 2B, in step S9, the ICE match ranker 146 determines whether a plurality of ICE matches are found. If only one ICE match is determined, in step S10, that single ICE match is reported. After the ICE match is automatically reported, the system 100 allows retrieval of the target text 162, 164 via the segment retriever 138.

Embodiments of the present invention relate to improving ICE match disambiguation using structural context levels. Consider the following first example with a structured context level.
Source 1 (s1) Structured context = Heading text “Print the document”
Target 1 (t1) Structured context = Heading text “Das Document ausdruken”
In this case, in the heading structural context (identical for both the source and target segments in the translation unit), the German translation uses the infinitive (non-imperative) form in the translation.

Here, the following second example with a structural context level that is in the form of an instruction list so that the same English sentence is used in the same document, but this time the imperative form is required in German Please consider.
Source 2 (s2) Structural context = instruction list “Print the document”
Target 2 (t2) Structural context = instruction list “Drucken Sie das Document aus”

  When the above exemplary sentence is stored in a translation memory or other such data store in accordance with an embodiment of the present invention, not only the different translations, but also its structural context (i.e., heading vs. instruction list) Stored in translation memory. Thereafter, whenever the same sentence appears in the same structural context, a translation appropriate to the current structural context can be preferentially proposed.

  The structural context level can be used instead of or in addition to the usage context level (predecessor, after, source, and / or target) and / or asset context level.

C. Multiple ICE Match Prioritization Referring to FIG. 2B, steps S11-12 represent an optional step that addresses the situation where multiple ICE matches are determined in step S5, ie, YES in step S9. In one embodiment (not shown), the ICE match determiner 130 can simply allow the user to select an ICE match from a list of ICE matches. However, this is not preferred because it overrides one purpose of ICE matching, namely that it is not necessary to validate an exact match. In the preferred embodiment shown in FIG. 2B, when multiple ICE matches are determined, the ICE match prioritizer 146 prioritizes (ranks) each ICE according to the degree of context match in step S11. As explained above, the “degree of context matching” can be determined in advance. This step prioritizes each ICE match degree of context matching, and in step S12 either presents the ICE match to the user for selection or automatically selects the highest ranking ICE match. Do. It should be understood that various formulas for prioritizing multiple ICE matches are possible depending on the number of context levels. As an alternative to or in addition to one or more fixed formulas, optimal (or otherwise acceptable) prioritization can be reached through the learning process, where Observed results from a given training corpus, or output from previous translation projects over a period of time or parts from live translation projects, are used to adjust prioritization of different context levels.

  The following example illustrates one embodiment of prioritizing multiple ICE matches.

Example The context includes usage context level and asset context level, and the lookup segment “team of visionaries” is translated into French using TM 128 of FIG. 3 based on source document 180 as shown in FIG. Assume that it must be translated. In this case, “team of visions” is based on previously stored translations and has four exact matches: 1) e'quipe devisionnaires, 2) group devisionnaires, 3) banded devisionnaires, and 4) have group de futurology. Also assume that the lookup segment “team of visionaries” has a pre-LS source UC hash code 333, a post-LS source UC hash code 4444, and an asset code 666666. It is also assumed that only one context level needs to match the context level of the lookup segment in order for an exact match to be indicated as an ICE match by the ICE match determiner 132. In this case, each exact match is an ICE match. Specifically, 1) “e'quipe de visionnaires” has a matching previous source UC hash code and asset code, and 2) “group de visionnaires” has all matching context levels, 3) “band des visionnaires” has a matching asset code, 4) “group de futurology” has a matching post source UC hash code.

  It should be appreciated that by definition, ICE matches are given higher priority than non-matching lookup segments (ie, segments that require manual or machine translation), fuzzy matches, and exact matches that are not ICE matches. One prioritization rule for ICE match is shown below. In this rule, the rankings are listed in reverse order of priority (eg, the higher the number, the higher the priority), the usage context (UC) levels include the preceding UC level and the subsequent UC level, and some examples The prioritizing step includes:

1. Source usage context (UC) matches, target UC) matches, and structural context matches take precedence over source usage context (UC) matches and target usage context (UC) matches.
Have both the same source UC hash code and target UC hash code as the lookup segment and the same structural context match than the ICE match that has only the same source UC hash code and target UC hash code as the lookup segment ICE match takes precedence.

2. Source UC matches and target UC matches take precedence over source-only UC matches.
An ICE match that has both the same source UC hash code and the target UC hash code as the lookup segment takes precedence over an ICE match that has only the same source UC hash code as the lookup segment.

3. A preceding UC match takes precedence over a subsequent UC match.
Matches the previous source usage context level or the previous target usage context level of the lookup segment, rather than an ICE match that has only either those levels that match the later source usage context level or the later target usage context level of the lookup segment ICE matches with any of those levels are preferred.

4). A full source usage context (UC) match takes precedence over a partial source usage context (UC) match.
Assuming that an exact match with only one source UC that matches the hash code of the lookup segment is considered an ICE match (referred to as a “partial match”), the previous source UC that is identical to the lookup segment rather than the partial match ICE matches that have both a hash code and a post source UC hash code (referred to as "full source UC match") are preferred. In other words, an ICE match that has both a previous source UC level and a subsequent source UC level that match the previous source UC level and the subsequent source UC level of the lookup segment is the previous source UC level or the subsequent source UC level of the lookup segment. Overrides an ICE match that has only one of the previous source UC level and the later source UC level. For example, ICE match 2) “group de visionnaires” is preferred over all others because it has a pre-UC hash code (333) and a post-UC hash code (4444) that match the lookup segment.

  Similarly, prioritization can be given to full target UC matches rather than partial target UC matches.

5. ICE matches from the same asset as the lookup segment take precedence over matches from other assets.
In this case, multiple ICE matches cannot be distinguished by prioritizing the full match over the partial match described above, i.e., the first priority is non-deterministic and is looked up based on the asset code. ICE matching from the same asset as the segment takes precedence. In other words, if the first priority is non-deterministic, an ICE match from the same asset as the lookup segment takes precedence over an ICE match from a different asset. In this example, ICE match 1) "e'quipe de visionnaires" and 4) "group de futurology" are both partial ICE matches, but ICE match 1) "e'quipe de visionnaires" is the lookup segment From the same asset “666666” and this should be preferred.

6). If multiple ICE matches from the same asset are determined for a lookup segment, the ICE match that has a position closer to the location of the lookup segment within the asset takes precedence.
This prioritization addresses the situation where the lookup segment 154 exists in multiple locations within a single source asset 180, resulting in multiple ICE matches for a particular asset. For example, FIG. 4 shows a source asset 180 that includes two occurrences of the lookup segment “team of visionaries” 154A, 154B. In this case, the ICE match prioritizer 146 evaluates the position within the asset of a particular lookup segment and determines the ICE match that is closest to the position within the asset of the lookup segment rather than other ICE matches from the same asset. Prioritize. In other words, if the second preference is non-deterministic, the ICE match that has the closest location to the location of the lookup segment in the asset takes precedence over the other ICE matches. This assessment of location can be repeated for any number of iterations of the lookup segment within a particular asset.

  Thus, prioritization can involve first looking for a match of the source usage context level and the target usage context level combined with the structural context level to remove the ambiguity of multiple ICE matches. If this first step is not sufficient for disambiguation during an ICE match, it may use a match at both the source usage context level and the target usage context level as well as a match at the source context level. it can. The pre-use context level can then be used instead of the post-use context level, which can be followed by a full source usage context or a full target usage context rather than a partial source usage context or a partial target usage context. An asset context level can be used if the above steps fail to disambiguate between multiple ICE matches. If this is still insufficient for disambiguation, the location of the lookup segment can be used.

  Asset metadata can be used in prioritization of multiple ICE matches for disambiguation in any combination and / or priority order with the above or similar steps. Similarly, when removing fuzzy matching ambiguities, different prioritization schemes and priority orders from the above or similar steps can be used.

  The above prioritization steps and ordering of steps are given by way of example, and different combinations and orders of prioritization steps can be used.

  After prioritization is complete, at step S12, the ICE match prioritizer 146 ranks the user in any currently known or later developed form, eg, via the graphical user interface of the client computer system 150. Allows selection of ICE matches based on, or automatically selects the highest priority ICE match. After an ICE match is selected, the system 100 allows retrieval of at least one target text 162, 164 via the segment retriever 138.

  Because of the high level match quality provided by ICE match, source text that is determined to be ICE match does not need to be reviewed or validated by the translator. These source texts can be automatically accepted, thus reducing translation cycle time and resulting in lower translation costs. In addition, the system 100 addresses the situation where there are multiple lookup segments 154 in the single source asset 180 that are substantially identical in content. In this case, the system 100 can determine an ICE match for each lookup segment 154 based on the matching level. Typically, at least one lookup segment has a different ICE match than at least one other lookup segment to assist in this determination. Otherwise, multiple ICE matches can be reported to the user for selection as described above. The system 100 also facilitates translation of content sections, which can be used to retrieve matches and / or to separate content blocks within a single asset even when content segments are split or merged. It is repeated across different assets with minimal effort, including without limitation enabling translation.

  The operations described above can continue processing further lookup segments of the source asset 180 for TM 128, or provide output to the user after the entire asset is complete.

D. Target Usage Context Level The embodiments of the present invention described above mainly include a usage context level that is a source usage context level, that is, a usage context associated with text surrounding the translated text of the source language. However, other embodiments of the invention include a usage context level that is a target usage context level, ie, a usage context associated with the text surrounding the translation of the translated text of the target language.

Consider a current segment g with source s (g) and translation t (g), where t (g) is the desired translation of the lookup segment. If the translation memory contains multiple exact matches t ^* (g) for source s (g), consider the usage context of the lookup segment for disambiguation between multiple exact matches The usage context is used to select the translation t (g) within t ^* (g) that best fits within the current usage context. The segment that precedes the segment (in bilingual) is segment f, s (f) is the source of the preceding segment, t (f) is the target of the preceding segment, and the (following) segment following the current segment is If segment h, with source s (h) and translation t (h).

Thus, according to the embodiment of the invention described above, the leading source usage context level s (f) and the trailing source usage context level s (h) are used for disambiguation during the translation t ^* (s). Is done. However, other embodiments of the present invention may alternatively or additionally include target source usage context levels, ie, preceding and / or following target context levels t (f) and t (h) when available. Can be used to remove ambiguity between translations t ^* (g).

  The above difference between disambiguation using the source usage context level and disambiguation using the target usage context level will now be described by way of example.

A document
The cat sleeps. It is cut. It purrs.
3 segments in the form of three sentences.

In that case, it is assumed that a translation of the second segment into German, ie “It is cut.” In German, is desired, and further the translation of the first segment is already known (“Die Katze schlaft.”). Assuming that the potential translation t ^* (g) = {“Sie nis niedrich.”, “Er nis niedrich.”, “Es nis niedrich.”} According to the source usage context is It can only be based on the two segments surrounding the segment to be translated, namely the preceding segment s (f) = “The cat sleeps.” And the subsequent segment s (h) = “It purrs.”.

In contrast, between potential translations t ^* (g) = {“Sie ist niedrich.”, “Er nis niedrich.”} According to a combination of source usage context level and target usage context level Disambiguation is based on the segment s (f) = “The cat sleeps.” Preceding the translated segment in the source language, and the segment t (f) = “Die Katze” preceding the translated segment in the target language. It can also be based on “schlaft.”

  In many situations, the use of a combination of target and source usage context disambiguation can yield improved results over disambiguation based only on the source usage context. Other embodiments of the invention may use a target usage context and no source usage context.

  Note that t (h) is included primarily for completeness. In reality, the translators work sequentially through the document, so t (h) is not generally available until the following segment has been translated, so the translation of the following segment is often unknown, Therefore, it cannot be used for disambiguation.

E. Translation Memory Generation The presence of context information regarding TM entries is necessary for system 100 operation. Thus, embodiments of the present invention require that context information be stored with every new translation added to the TM. This makes it possible to efficiently compare the context information of the lookup segment with the context information of the previously translated segment without requiring access to the previously translated document.

  To this end, in another embodiment, the present invention allows context information to be stored with each translation when the translation is stored in TM 128, and therefore the translator has previously translated for the present invention to function. Provide a form that does not require you to keep files such as selected documents at hand. Turning to FIG. 5, the present invention also includes a method for storing translation pairs of source text and target text in TM 128. In the first step S100, the TM generator 136 is used to assign a context to the translation pair. The context can be assigned, for example, by the implementation of the SID described above during content creation, or through the operation of the hash algorithm 133 in the translation path. Next, in step S101, the TM generator 136 stores the context in the TM 128 together with the translation pair. As explained above, the context can include one or more of usage context levels (any combination of predecessor, post, source, target), structural context level, and asset context level.

  It should be understood that the TM generation described above can also be implemented on the client-side system 150 when an asset (segment) is created. In this embodiment, the present invention includes a client-side system 150 that interacts with a translation system that includes TM 128 (ie, system 100 with other content management system components 140). Turning to FIGS. 1 and 6, in this case, the client-side system 150 operates by providing a SID assigner 200 to assign a segment identifier (SID) to the segment 152 translated by the TM 128 (step S200). The SID indicates the usage context of the segment. The SID assigner 200 may allow a user to associate a predetermined SID, or the SID may be generated using a hash algorithm 133, for example. Additionally, the system 150 can include a communicator 202 that communicates SID assignments (step S201) to be stored as part of the TM 128, eg, by the TM generator 136 of the system 100.

V. CONCLUSION The invention described above gives translators the ability to perfectly match source content with TM content, reduces the need to validate source content with respect to TM, and creates a truly reusable TM system. By providing value to translators, thereby enabling a more efficient translation process.

  It should be understood that the order of steps described above is exemplary only. Up to this range, one or more steps can be performed in parallel, in a different order, such as when separated. Further, in various embodiments of the invention, one or more of the steps may not be performed.

  It should be understood that the invention can be implemented in hardware, software, propagated signals, or any combination thereof and can be partitioned differently than shown. All types of computer / server system (s) or other devices adapted to perform the methods described herein are suitable. The usual combination of hardware and software can be a general-purpose computer system with a computer program that performs the respective methods described herein when loaded and executed. Instead, a special purpose computer (eg, system 100) may be utilized that includes specialized hardware for performing one or more of the functional tasks of the present invention. The invention may also be implemented in a computer program product or propagated signal, which computer program product or propagated signal has all the respective features that enable the implementation of the methods described herein. And these methods can be performed when loaded into a computer system. A computer program, propagated signal, software program, program, or software may, in the current context, (a) convert to another language, code, or notation, and / or (b) recreate in a different material form Any, in any language, code, or notation, of a set of instructions intended to cause a system with information processing functionality to perform a specific function, either after or both of Means an expression. Further, it should be understood that the teachings of the present invention can be provided as a business method based on a subscription or fee. For example, the system and / or computer can be created, maintained, supported, and / or deployed by a service provider that provides the functionality described herein for a customer. That is, the service provider can provide the functionality described above.

  The present invention may be embodied in other specific forms without departing from its spirit or essential characteristics. It should be understood that the embodiments described above are merely exemplary and are not limiting with respect to the principles of the invention. Various other and other modifications and changes within the spirit and scope of the present invention can be made by those skilled in the art and thus fall within the meaning and scope of equivalence of the claims. All such changes are intended to be included within the scope of the claims.

100 exact match translation system in context 102 computer 112 memory 114 processing unit 116 I / O interface 118 bus 120 I / O device 122 storage system 128 translation memory 130 exact match determiner 132 ICE match determiner 133 hash algorithm 134 fuzzy match determiner 136 TM generator 138 Segment retriever 140 Other system components 142 Context identifiers 144 ICE match evaluator 146 ICE match prioritizer 148 TM entry 150 Client side system 152 Source text 154 Lookup segment 162 Target text 164 Target text 166 Context identification 168 Source usage Context part 1 0 Asset context portion 172 prior UC hash code 174 after UC hash code 180 source asset 190 directly preceding segment 192 immediately follows the segment 200 SID assigner 202 communicator

Claims (29)

A computer-implemented method for determining a matching level of a plurality of source texts stored in a translation memory for a translated lookup segment, comprising:
Determining all exact matches of the lookup segment in the plurality of source texts;
When at least one exact match is determined, the respective exact match is an in-context exact (ICE) match of the lookup segment when the context of the lookup segment matches the context of the exact match A step of determining
The context includes at least two levels;
It said at least two levels, see contains a source usage context level and a structural context level,
The ICE match includes a match between the structural context level of the lookup segment and the structural context level of the source text;
Method.   If multiple ICE matches are determined, the step of determining an ICE match is in accordance with the degree of context matching so that a more appropriate ICE match can take precedence over one or more other ICE matches. The method of claim 1, comprising prioritizing each ICE match.   An ICE match that has both a source usage context level and a structural context level that match both the source usage context level and the structural context level of the lookup segment matches the source usage context level of the lookup segment The method according to claim 1 or 2, wherein there is a higher degree of context matching than an ICE match with only source usage context levels.   An ICE match having a structural context level that matches the structural context level of the lookup segment has a higher degree of context matching than an ICE match that has only a source usage context level that matches the source usage context level of the lookup segment. The method according to claim 1, wherein the method is said to be present. The at least two levels include a target usage context level;
5. The method of any one of claims 1-4, wherein the source usage context level comprises a previous source usage context level and / or the target usage context level comprises a previous target usage context level. The at least two levels include a target usage context level;
6. The method of any one of claims 1-5, wherein the source usage context level comprises a post-source usage context level and / or the target usage context level comprises a post-target usage context level. The at least two levels include a target usage context level;
If multiple ICE matches are determined, the step of determining an ICE match is in accordance with the degree of context matching so that a more appropriate ICE match can take precedence over one or more other ICE matches. Prioritizing each ICE match,
The source usage context level includes a preceding source usage context level; the target usage context level includes a preceding target usage context level;
An ICE match having a preceding source usage context level and / or a preceding target usage context level that matches a preceding source usage context level and / or a preceding target usage context level of the lookup segment is a structural context level of the lookup segment. 7. A method according to any one of the preceding claims, wherein there is a higher degree of context matching than an ICE match with only a matching structural context level.   The step of determining an ICE match is such that each exact match is an ICE match of the lookup segment only if multiple context levels of the lookup segment match the context level of the exact match. The method according to claim 1, wherein   The step of determining an ICE match only if at least one usage context level and structural context level of the lookup segment matches at least one usage context level and structural context level of each exact match, 9. A method as claimed in any one of the preceding claims, indicating that each respective exact match is an ICE match of the lookup segment.   The method of claim 9, wherein the at least one usage context level comprises a preceding target usage context level.   The method of claim 9, wherein the at least one usage context level comprises a preceding source usage context level. The at least two levels include a target usage context level;
If more than one ICE match is determined, the step of determining an ICE match may be performed according to the degree of context match so that a more appropriate ICE match may take precedence over one or more other ICE matches. Prioritizing each ICE match,
The step of prioritizing comprises:
Priority is given to an ICE match having a source usage context level, a target usage context level, and a structural context level that match the source usage context level, the target usage context level, and the structural context level of the lookup segment. The method according to claim 1.   If the first priority step is non-deterministic, secondly select an ICE match having a source usage context level and a target usage context level that match the source usage context level and the target usage context level of the lookup segment. The method of claim 12, comprising a priority step.   14. The method of claim 12 or 13, wherein the source usage context level is a preceding source usage context level and the target usage context level is a preceding target usage context level.   If the second priority step is non-deterministic, only either the post-source usage context level or the post-target usage context level that matches the post-source usage context level or the post-target usage context level of the lookup segment ICE matches that have either a previous source usage context level or a previous target usage context level that matches the previous source usage context level or the previous target usage context level of the lookup segment over the ICE matches that have The method of claim 13, comprising steps.   Any usage context level of the lookup segment over an ICE match that has only a structural context level that matches the structural context level of the lookup segment if the third priority step is non-deterministic The method of claim 15, comprising: fourth prioritizing an ICE match having a usage context level that matches.   If the fourth priority step is non-deterministic, it matches the structural context level of the lookup segment rather than an ICE match that has a different structural context level than the structural context level of the lookup segment. 17. The method of claim 16, comprising the fifth priority of ICE matches having a structural context level that   18. The method of claim 17, wherein the step of prioritizing fifth comprises the step of prioritizing an ICE match that has a position closest to the position of the lookup segment in an asset if it is non-deterministic. .   The method of any one of claims 2-18, further comprising allowing a user to select the ICE match based on prioritization. The lookup segment includes a plurality of lookup segments that are substantially identical in content;
20. A method according to any one of the preceding claims, wherein the step of determining an ICE match comprises determining an ICE match for each lookup segment.   21. The method of claim 20, wherein the at least one lookup segment has a different ICE match than at least one other lookup segment. A system for determining a matching level of a plurality of source texts stored in a translation memory for a lookup segment to be translated,
Means for determining all exact matches of the lookup segments in the plurality of source texts;
When at least one exact match is determined, the respective exact match is an in-context exact (ICE) match of the lookup segment when the context of the lookup segment matches the context of the exact match And means for determining
The context includes at least two levels;
It said at least two levels, see contains structural context level and a target usage context level,
The system wherein the ICE match includes a match between the structural context level of the lookup segment and the structural context level of the source text . A computer-implemented method for storing translation units of source text and target text in a translation memory comprising:
Assigning a context to the translation unit, the context comprising at least two levels, the at least two levels comprising a source usage context level and a structural context level;
Storing the context together with the translation unit;
The structural context level of the source text is used to determine an in-context exact (ICE) match with the structural context level of the translated lookup segment;
Method. A system for storing translation units of source text and target text in a translation memory,
Means for assigning a context to the translation unit, the context comprising at least two levels, the at least two levels comprising a source usage context level and a structural context level;
Look including a means for storing the context with the translation unit,
The structural context level of the source text is used to determine an in-context exact (ICE) match with the structural context level of the translated lookup segment;
system. The translation memory is
Multiple source texts for comparison with lookup segments,
25. The system of claim 24, comprising a context identifier for each source text, wherein the context identifier includes a source usage context portion and a target usage context portion.   26. The system of claim 25, wherein the source usage context portion and / or target usage context portion comprises a pre-use context portion and / or a post-use context portion.   27. A system according to claim 25 or 26, wherein the context identifier comprises a structural context portion. A client-side system that interacts with a translation system that includes a translation memory,
Means for assigning a segment identifier to a segment translated by the translation system, the segment identifier indicating a source usage context and a target usage context of the segment;
Means for communicating the segment identifier assignment for storage as part of the translation memory;
The segment identifier comprises a structural context;
The structural context of the segment is used to determine an in-context exact (ICE) match with the structural context of the source text of the translation memory in the translation system;
Client side system.   24. Computer software adapted to perform the method of claim 1 or 23.

Images