5.4 Translation memory management systems
A translation memory (also known as a 'bi-text') is a file consisting of a table of matches between the segments of two texts, one of which is the translation of the other. In other words, a translation memory is a table matching discourse segments in two different languages. The matches are generally, but not necessarily, based on whole sentences. Depending on the system, set phrases, paragraphs, sections or even whole chapters may also be segmented and matched. Segments which are deemed to be 'equivalent' (in that they 'match') can then be substituted for each other whenever one of them appears in a document being translated (particularly when the translation on hand concerns an updated version of an existing document).
Any segment in a new text which appears in an existing match table will be spotted, and translated as in that table. The user is therefore given the impression that the system 'remembers' that the segment has already been translated though the term bi-text would seem more appropriate to describe the actual content and form of the memory.
The translation memory system requires three separate software applications:
- an aligner, to create the memory files,
- a format and tag management system,
- a translation memory management system.
5.4.1 Aligners
Aligners are software applications which can 'align' a source text and its translation, by identifying 'matching' segments in both texts, which will then form the basis for the tables that make up the translation 'memory'.
The aligner 'aligns' or establishes an equivalence between segments deemed to have the same meaning in two different languages, therefore requiring a digital version of both the source and the target text. The alignment can also be automated, with the translator checking and validating the memory. Each 'bi-text' (i.e. the source text and its translation) is then archived and indexed for future use (e.g. by client name, product or process name, file code, domain, or any other relevant indicator). An aligner can be used to rapidly build up a stock of translation memories and get a return on the quite significant investment that a translation memory system requires.
Nowadays, the word 'aligner' can also refer to whoever in the department or company is in charge of creating translation memories by processing existing translations along with their sources. This is often a task given over to students on translation courses during their work placement. The task is actually more interesting and useful from a training point of view than is usually thought, because it forces the operator to check the matches produced by the system, therefore providing an opportunity to learn by example.
As new translations proceed, translators generate new memories or expand existing ones because, the moment the system comes in operation, translating amounts to systematically declaring matches between segments in the source material and the translated material. The inventory of translation memories therefore increases every time a translator produces a new translation, because, as we said, this automatically adds to the memory. The stock will increase all the more rapidly if there are a large number of translators using the same
system via a network (an intranet). The system's efficiency and profitability increases in line with the repetitive nature of the documents to be translated.
A common standard for translation memories has now been developed as TMX to allow translation memories created under different applications and existing or future TM management systems to become compatible.
The probability is that translation memories shall soon be offered for sale. Those will apply to stereotyped documents (contracts), stereotyped operations (assembly and disassembly of machines, maintenance operations, sequences of operations), and sometimes even be domain-specific.
Major work providers and major service providers in the translation business have created or assembled huge translation memories which are maintained and amplified on an ongoing basis. All their employees and subcontractors access, use, and complement or update those memories - just like they used to access, use, and complement dictionaries. TMs (translation memories) are dictionaries of a kind and will necessarily, like any other dictionary, find their way onto the market.
There should therefore shortly emerge a new business segment and job profile having to do with translation memory creation (not to be confused with the design and development of translation memory systems) starting with general memories relating to clearly defined areas (translation of balance sheets, contracts, operating instructions, etc.) These will automatically translate the core sections of each type of document, and each translator will add his own customized tables according to the specificities of his own materials.
5.4.2 Format and tag management applications
Like any other automatic processing system, translation memories are likely to generate formatting and tagging problems.
Some formats may need to be converted before the translation can take place (but most systems accommodate a wide variety of formats) and, whenever relevant, format tags or other kinds of tags have to be preserved in the source text. In the worst case scenario, the file has to be converted for translation into a format the application can recognise, and then converted back into the format required by the client once it has been processed.
In the most recent translation memory management systems, the user is not even aware of the pre-translation processing or conversions as these are carried out automatically.
5.4.3 Translation memory management systems
Translation memory management systems (which exploit existing memories and create new ones every time a new translation is carried out) work, in fact, very much like word processing applications.
As the translator moves through the translation, the application identifies those segments of the document that are in the translation memory, picks up the 'translation' from that same translation memory, and integrates it directly into the translation, or asks the translator to confirm.
Technically, this means that the application compares 'significant' character chains in the source text with the segments in the source language recorded in the memory, and, if such segments are found, restores the recorded translation in the target language - 'significant' means they have been segmented according to predetermined parameters, not that they are necessarily 'meaningful'. The translation memory system either suggests perfect 'matches' (i.e. segments in the B language that are exactly equivalent to segments in the A language) or 'fuzzy' matches (i.e. memory segments which only partly correspond to source text segments). The operator can increase the efficiency of the system by setting match thresholds (a 75% match, for instance, means that 75% of the two character chains are identical). As this type of application does not make use of artificial intelligence, the match is based on a degree of coincidence between the character chains in the text to be translated and those in the translation memory, not on semantic criteria.
The translation memory systems currently available on the market are incapable of recognising either that two different segments in the same language convey the same meaning, or that two morphologically identical character chains convey different meanings. All they can do is blindly exploit segments that have been declared and confirmed as equivalent in meaning. This means that such equivalences should normally be carefully checked, especially where recurrence is high, because the match tables may contain glaring errors. Many conscientious translation service providers have undertaken extensive revision of all the phraseology contained in their existing translations. The expansion of translation memories should therefore rapidly lead to the recognition of phraseology as a vital part of translation studies - 'phraseology' being considered here both as (a) the study of discourse and enunciation stereotypes (b) the sets of phrases and formulations specific to a particular speaker, type of document, discourse, professional field, etc.
What must be borne in mind is that:
- translation memory management systems are only worth the expenditure of time (and money) if they have extensive memories to draw upon. Empty memories are no use and some kind of initial investment in terms of text alignment is mandatory unless one is prepared to start building from scratch.
- translation memory systems are useful with repetitive materials - either single documents containing a large number of recurrent phrases or paragraphs, or sets or series of documents that are largely identical. They were originally used for updating successive versions of product documentation or translating
different documents delivered with one same product. The first ones were developed by major companies with large-scale translation needs. Today, all major companies with significant translation needs tend to have their own translation memories, which they then require their translators to use.
Given that (a) a large part of the product documentation needing to be translated is in the IT area and that (b) such documentation has been designed and disseminated in electronic format, it is no surprise that translation memory management systems have been specifically designed to help with the translation (localisation) of Web pages, software packages, video games, and the like.
It must finally be mentioned that machine translation applications are now being sold as part of a translation memory system package (cf. TranscendTM, included with Star Transit™) and that TM management systems are sold under the name of CAT (Computer-assisted translation) tools.
The leading TM brands are SDL Trados™, Star TransitTM, Similis™, Dej'aVu™, Wordfast™, Catalyst™ (for localisation), IBM's Translation Manager™, Trans Suite 2000™, MetaTexis™, TR-Aid™, Pasolo™, Mneme™ by Logos, Maestro™, by Lionbridge, OmegaT (free, Java), MultiTrans™, ACROSS™.
5.5 Specialist software
Specialist software is now rapidly becoming an integral part of the translator's everyday environment. This includes both comprehensive software packages and specific applications designed for one or several particular tasks.
5.5.1 Comprehensive software packages
Specialist software packages are now available for many forms of special-purpose translation, notably in areas such as subtitling (e.g. Monal Virtual by LVT) or software localisation (e.g. Locale, Robohelp, Catalyst, Locstudio) Such tools are designed to allow the user to carry out all of the necessary operations smoothly and comfortably. Failing this, the translator would have to carry out manually the somewhat complex and tricky processes of, say, disassembling and re-assembling the medium, and carrying out the functional qualification tests.
Such packages can free the translator from the constraints of having to rely on other specialists, and in particular the IT specialists who had to come in to process the material before and after the translation phase proper.
Subtitling packages, for instance, allow the translator to carry out a number of automatic or semi-automatic operations on an ordinary workstation. The translator can now:
- detect segments and time-codes,
- do all necessary tagging,
- generate the sub-titles,
- run simulations,
- and even actually burn the sub-titles.
Localisation packages can be used on that same workstation to:
- protect (or 'freeze') formats and tags,
- implement all the necessary conversions,
- extract items calling for special treatment (e.g. menus and menu options),
- call up the contents of existing translation memories,
- carry out the translation, and integrate it,
- carry out any back any conversions or compiling which may be required,
- implement qualifying tests for the translated version (including homogeneity testing and consistency checks).
5.5.2 Dedicated software applications
Many applications are used to carry out specific functions, such as text tagging, file management, spell-checking, revision, editing, etc. Applications such as these are usually trying to cash in on the growing and relatively lucrative translation market. Any function likely to generate added value is therefore liable to be automated or semi-automated at some stage, particularly if the application is expected to reach a market extending beyond the circle of language professionals. This is a perfect illustration of the reversal of traditional marketing principles, whereby a generic product designed for the general public was then adapted to specialist uses, including the translation market. It has now become clear that a return on investment can be generated by specialist software tools, in the knowledge that they usually find their way to much wider markets and, hopefully, to the public at large.
5.6 Voice recognition software
Voice recognition systems (or dictation systems as they are sometimes called), were obviously not specifically designed for translators. As would seem obvious to anyone who has ever walked into a hospital, they were originally designed for the medical profession. They will no doubt be instrumental in bringing about the third major revolution in translation practice after word processing and translation memories. If it were not for the fact that they require very high speed CPU clocks and large RAM capacity, they would already be far more widely used. The efficiency of the system is in fact quite spectacular. More important still, such systems can generate significant productivity gains and free the translator from the slavery of the keyboard and the computer screen. That revolution is on the march right now.
From the translator's point of view, voice recognition amounts to a dictation system, with the translator dictating translations just as a company director used to dictate letters or a doctor reports. The system works on a probabilistic basis, processing extended phonemic sequences and calculating the most probable corresponding word forms. The translator dictates at (slightly lower than) normal speed, using a microphone. With a short time lag, the system then displays the dictated text, and the translator can dictate any corrections he thinks must be made.
The use of voice recognition systems requires quite heavy preparation. Like any other user, the translator first of all has to master the use of the application basics. He then needs to configure and 'instruct' the system in order to make the best use of its functionalities. To this aim and purpose, the translator must:
- carry out a series of dictations so that the application will 'recognise his voice',
- dictate, then correct, as many of his own standard type documents as possible, to build up and customize the system's dictionaries, and also to fine tune the algorithms in order to reduce the risk of ambiguity and error,
- create and record the voice macro commands which will be used to control all the standard operations, particularly those used in word processing (i.e. stop, start, select, correct, create table, indent, etc.),
- learn the application's basic rules (how to dictate figures, acronyms, etc.),
- learn to use the system from within his regular word processor - e.g. 'dictate directly into Word'.
Every time the user dictates a new translation, all the dictionaries and algorithms are automatically upgraded and enhanced.
Voice recognition software is, in fact, a kind of 'vocal word processor', where the microphone replaces the keyboard and recognition errors are made instead of typing errors. Recognition rates improve with volume and soon become more than adequate and even impressive when the material is specialised and repetitive.
Voice recognition can provide all the benefits of a word processor in terms of page outlay, error correction, cutting and pasting, editing, formatting, tables, window management, styles, sorting, and so on, with all the commands being activated by the operator's voice. The translator can also use it to process translated material which has already been 'put through' a translation memory system, as he can then dictate the segments that need replacing or completing and carry out any corrections required.
It should also be said that most voice recognition systems allow the translator to listen to the translation (via a voice synthesis system). This means that in future, a translator may receive dictated source material which will then be digitized, processed by a translation memory system, and translated using voice recognition technology, pending the introduction of voice translation memories, no doubt in the very near future.
The impetus behind the move towards the use ofvoice recognition technology in translation is of course linked to the search for ever greater productivity gains. Translators will need to get used to this new way of working and will have to learn to dictate efficiently, to implement corrections quickly so as to avoid error repetition, and adopt completely new translation techniques. This means talking at normal speed, speaking clearly but without unduly stressing the words, avoiding 'noise', using a top of the range microphone, finding the optimal position for the microphone and always positioning the microphone the same way when dictating. It also means learning to take advantage of the differences between oral and written translation, i.e. the fact that oral translation 'flows' better and is more coherent, and overcome the drawbacks of dictating a translation without the support of the keyboard, which requires a lot of practice.
Some translators do readily move over to voice recognition: this is particularly true of those who never type their own translations (the few who are still around are obviously keen 'VR' enthusiasts) or of those who have poor keyboard skills and have to keep their eyes riveted on their fingers when typing. Productivity gains are less spectacular for those who are good typists and who have completely mastered window management and advanced word processing skills. For these, the key advantage would be in being freed from the workstation. The prospect of being able to use truly mobile digital voice recorders, as provided with the more advanced voice recognition software, is often the clinching argument.
For the older generation of translators, voice recognition technology is the revenge of the good old Dictaphone which used to be the only way to productivity gains in pre word processor days. If the Dictaphone can be seen as a voice recording system without the recognition (that was provided by the typist), voice recognition systems are in a sense Dictaphones with a built-in text typing function (thus saving on the cost of the typist).
Be that as it may, most translators - with the exception of those who have never mastered the keyboard - carry on doing what they have always done, i.e. they use the new technology whenever it proves really useful, and continue to use the tried and tested tools of the trade whenever the gains are doubtful.
5.7 Translation management software
Translation management software is designed to organise and manage the whole translation chain. It usually covers a number of the following functions:
- prospect management,
- client management,
- management of agreements, briefs, specifications and procedures,
- contract forming,
- management of partner relations,
- generation of estimates,
- work load and work flow scheduling,
- pre-translation operations,
- quality control,
- quality management,
- billing and collection,
- comparisons between estimated and real time and cost,
- post-mortem project analyses.
Workflow management software is becoming a standard component of the workstations of freelancers working for the bigger translation companies. The latter have developed systems that feed in the material to be translated by the translator, indicate schedules and deadlines, and monitor translator's progress.
5.8 Machine translation systems requiring human intervention
Machine translation systems which require some kind of human intervention are generally known as 'semi-automatic' translation systems. They automatically substitute confirmed terminology and phraseology and 'suggest' translations for segments for which no previous approved translation exists. They are similar in that respect to translation memory systems. Their most remarkable feature is that they usually offer the human translator a choice of possible translations that have to be confirmed.
In practice, translators working with such systems very often decide to 'disable' some of their functions on the grounds that they can no more bear to make the same choices over and again, repeatedly, and to correct mistakes they have already corrected hundreds oftimes.
6. The Impact of information technology
Clearly, information technology and dedicated applications are now having a major impact on the profession, and are beginning to create a rift between those who are able and willing to make full use of the resources available, and those who are not and leave the added value bits of the translation service provision to the work providers or agencies, on whom they become increasingly dependent because they never get the money to invest in the technology. They are, quite often, the older generations, who find it rather difficult to keep pace with the evolutions of ICT. The rift also means there is a growing gap between the two groups in terms of remuneration for their work. And a finer distinction is now also beginning to appear within the 'technology users', between those who are quite satisfied with using the basic combination of'word processor + translation memory + terminology management system + Internet' on the one hand, and those who offer more elaborate services and process specific types of materials and media, by using more sophisticated software systems (subtitling applications, localisation applications, translation project management systems, and an infinite variety of customized software). Again, the latter come off best in terms of added value and remuneration.
chapter 14
From manual to automatic
Introduction
Information technology has had a tremendous impact on translators' information, practices, and markets. Depending on circumstances, the impact has been unreservedly favourable, mitigated, or potentially detrimental.
1. The friendly revolution
No one would question the view that translators have enormously benefited on many counts from the computerisation of their workstation and working environment.
Intranets. For translators, as for many other professionals, the intranet is the best way to share resources and experience, to be in constant touch with colleagues, to access other staff within the company or within the group for information or advice on specific points as well as for quick revisions. It is also, basically, the best and quickest way to create, use and constantly update shared resources.
E-mail. Email is another essential tool for translators as it allows all kinds of file transfers, reduces the risks associated with barely legible faxes, establishes immediate communication and reinforces the links between mailing list subscribers. In the guise of mailing lists and forums, it is also a powerful source of information about all kinds of freeware, some of which can be a real goldmine for those who 'know something about computers'.
The Web. The Web has now become a kind of vital umbilical cord linking translators to a wealth of on-line documentation. It allows the translator to have access to specialist forums and discussion groups related to specific subject areas or materials, to put their questions to domain specialists, to consult colleagues and fellow-professionals, strengthen collaborative networks, to access the vast array of terminology bases and banks now available (with the proviso that too much of the terminology available on the Web is of mediocre quality, and that its availability encourages all kinds of amateur translators to try their hand at technical translation, simply because they have '''found a fantastic dictionary on the Internet') and gain immediate access to all types of particularly helpful documentation (in particular, foreign language documentation from subsidiaries or partners in other countries, documentation from competitors working in the same or similar fields and producing similar products, manufacturers' documentation, parts lists and so on).
More importantly, the Web has well and truly changed the perception of translation and translation practices. In pre-Internet days, every translator was deeply convinced of the importance of 'knowing the subject' and getting hold of the necessary linguistic raw material (e.g. the terminology and phraseology) before attempting to translate. But the obstacles and the time needed to get hold of the relevant sources and documentation were such that hardly anyone was able to comply with that requirement. Failing access to informers, the required documentation and dictionaries would usually come days after the translation had been handed in. Even when they did have the time, translators were not sure where to start looking and even less so whether what they were looking for actually existed. Everyone knew full well what needed to be done, but few had the wherewithal to do it.
Today, with a click or two of the mouse (thanks Google and others!) translators can know whether the material has been translated (in part or in full), get all the information needed to elucidate the meaning of an unknown term or concept or to learn about the subject, check on the validity of their hypotheses, find models and preformatted translations or phrases - and much more. Definitions, drawings, charts, pictures and all kinds of representations are just one click removed.
It takes no more than a few seconds to find in-depth information (a key Web site, for instance) on the subject of the translation and it is usually a matter of minutes (or an hour or two, as the case may be) to learn what it is vital to know. There no longer is any excuse for bypassing any episode of the full documentation stage.
When confronted with a terminology problem, translators can access major terminology databases within seconds (or minutes if the network and database are very, very, busy), as well as a host of dictionaries, glossaries and word banks for almost every conceivable language and subject area. There is in fact very little that translators cannot find, nowadays, in the terminological and phraseological treasure trove that the Web has now become. And if they get stuck, fellow list subscribers will be glad to oblige.
For the first time ever, whenever the search is conducted in the target language, translators now have access to thematic terminology (i.e. terminology relating to an entire subject area or domain, as opposed to 'pinpointed' terminology relating to particular items found in the translation). The documentation corpus in the target language will most probably contain all the relevant terminology in its 'natural environment' so to speak, and provide detailed knowledge of which terms are really used and in which context. We have shifted from 'source-to-target terminography' to 'target-to-source terminography': instead of listing the terms in the source document and looking up their equivalents in the target language, translators can now mine out a parallel document in the target language that exactly 'mirrors' the source document in the sense that it deals with exactly the same subject and link up the terminology in that target language document with the terminology identified in the source document.
The documentation available on the Web also contains all the phraseological stereotypes that used to be impossible to find (unless, by sheer luck one came across a document dealing with the same subject in the target language and read it through). Today, given a relevant corpus in PDF format and the standard Acrobat® explorer, the phraseological concordances are (just about) immediate. This means that the new resources now available over the Web are such that phraseological accuracy has come well within the translator's reach.
The Web resources also allow the translator to check a 'hunch', by entering the term as a key word in any search engine to find out, using the search results as a knowledge corpus, if the term or phrase actually exists, whether it is accurate and what exactly it refers to. Any term can thus be shown in context, with all its different usages, and the translator can then determine whether that term can be used in the target language in identical conditions to those in which it appears in the source text.
And if the translator still has doubts or has failed to find some of the terminology or phraseology needed, the Web can also be used to contact domain specialists or experts, informers, colleagues and anyone else who might be able to supply the answer.
Дата добавления: 2015-09-29; просмотров: 19 | Нарушение авторских прав
| <== предыдущая лекция | | | следующая лекция ==> |