Computer-Aided Translation Lynne Bowker and Des Fisher 1.

Introduction Computer-aided translation (CAT) is the use of computer software to assist a human translator in the translation process. The term applies to translation that remains primarily the responsibility of a human, but involves software that can facilitate certain aspects of it. This contrasts with MACHINE TRANSLATION (MT), which refers to translation that is carried out principally by computer but may involve some human intervention, such as preor post-editing. Indeed, it is helpful to conceive of CAT as part of a continuum of translation possibilities, where various degrees of machine or human assistance are possible. After recognizing that fully automatic MT was a formidable challenge, researchers gradually turned their attention to CAT in the 1960s, starting with the creation of term banks, which used computers to store large volumes of structured information (see TRANSLATION AND TERMINOLOGY). Advances in computing and computational linguistics in the late 1970s and early 1980s spurred the development of modern CAT tools, which rely on computers not only for storing information, but also for actively searching and retrieving it. Visionaries such as Martin Kay (1980), among others, conceived of tools that became the backbone of CAT. However, it was not until the mid1990s that these tools became widely commercially available. Since then, the rapid evolution of technology has made CAT accessible, affordable, popular and even necessary to help translators in this globalized information age tackle the enormous volumes of text

to translate in ever shorter turnaround times (Esselink 2000; Lagoudaki 2006; GLOBALIZATION AND TRANSLATION; LOCALIZATION).

2. CAT Tools Since many translators are avid technology users, a wide range of tools could fall under the heading of CAT. However, this term is typically reserved for software designed specifically with the translation task proper in mind, rather than tools intended for general applications (e.g. word processors, spelling checkers, e-mail). The most popular and widely marketed CAT tool in use today is the Translation Environment Tool (TEnT)sometimes referred to as a translators workstation or workbenchwhich is in fact an integrated suite of tools. Individual components differ from product to product; however, the main module around which a TEnT is generally constructed is a translation memory (TM), which, in turn, most often functions in close association with a terminology management system (TMS).

2.1. Translation Memory Tools A TM is a tool that allows users to store previously translated texts and then easily consult them for potential reuse. To permit this, the source and target texts are stored in a TM database as bitexts. An aligned bitext is created by first dividing the texts into segments which are usually sentencesand then linking each segment from the source text to its corresponding segment in the translation. When a translator has a new text to translate, the TM system first divides this new text into segments and then compares each with the contents of the TM database. Using

pattern-matching, the TM system tries to identify whether any portion of the new text has been previously translated as part of a text stored in the TM database. When the TM system finds a match for a given segment (see Table 1), it presents the translator with the match from the TM database, allowing him or her to see how that segment was translated last time and to decide whether that previous translation can be usefully integrated into the new translation (see Figure 1). In keeping with the CAT philosophy, where tools are designed to assist rather than replace the translator, he or she is never forced to accept the matches presented by the TM system; these are presented only for consideration and can be accepted, modified or rejected as the translator sees fit.

Exact match Full match Fuzzy match

A segment from the new text is identical in every way to a segment stored in the TM database. A segment from the new text is identical to a segment stored in the TM database save for proper nouns, dates, figures, etc. A segment from the new text has some degree of similarity to a segment stored in the TM database. Fuzzy matches can range from 1% to 99%, and the threshold can be set by the user. Typically, the higher the match percentage, the more useful the match; many systems have default

Sub-segment match Term match No match

thresholds between 60% and 70%. A contiguous chunk of text within a segment of the new text is identical to a chunk stored in the TM database. A term found in the new text corresponds to an entry in the termbase of a TM systems integrated TMS. No part of a segment from the new text matches the contents of the TM database or termbase. The translator must start from scratch; however, this new translation can itself be sent to the TM database for future reuse.

Table 1. Types of matches commonly displayed in TMs.

New segment to translate Fuzzy match and corresponding translation retrieved from TM database

Instructions for reconstitution and intramuscular administration. EN: Instructions for reconstitution and subcutaneous administration.

FR: Directives pour la prparation du produit et ladministration sous-cutane. Figure 1. An example of a fuzzy match retrieved from a TM database.

2.2. Terminology Tools While TM systems are at the heart of TEnTs, they are typically integrated with terminology tools, which can greatly enhance their functionality. A terminology management system (TMS) is a tool that is used to store terminological information in and retrieve it from a termbase. Translators can customize term records with various fields (e.g. term, equivalent, definition, context, source), and they can fill these in and consult them at will in a standalone fashion. Retrieval of terms is possible through various types of searches (e.g. exact, fuzzy, wildcard, context). However, termbases can also be integrated with TM systems and work in a more automated way. For instance, using a feature known as active terminology recognition, a TMS can scan a new text, compare its contents against a specified termbase, and automatically display records for any matches that are found. If desired, users can further automate this step by having the system automatically replace any terms in the text with their target-language equivalents from the termbase. Consistency and appropriateness are maintained by using termbases specific to certain clients or text types.

2.3. Other TEnT Components TEnTs include more than just TMs and TMSs. Table 2 summarizes the functions of other common TEnT components. For more detailed descriptions, see Bowker (2002), Somers (2003a/b) or Quah (2006), as well as the entries for CORPORA IN TRANSLATION STUDIES, LOCALIZATION, MACHINE TRANSLATION and TRANSLATION AND TERMINOLOGY. Component Concordancer string and displays these in context. Compares a new text to translate with the contents of a specified TM Document analysis module Machine translation system Project management module Quality control module Term extractor database or termbase to determine the number/type of matches, allowing users to make decisions about which TM databases to consult, pricing and deadlines. Generates a machine translation for a given segment when no match is found in the TM database. Helps users to track client information, manage deadlines, and maintain project files for each translation job. May include spelling, grammar, completeness, or terminologycontrolled language-compliance checkers. Analyzes (bi)texts and identifies candidate terms. Table 2. Some common TEnT components. Brief description Searches a (bi)text for all occurrences of a user-specified character

3. Impact on translation With much touted benefits of increased productivity and improved quality being hard to resist, the incorporation of CAT into the translation profession has been significant: CAT tools have been adopted by many translation agencies, governments, international

organizations, transnational corporations and freelance translators (e.g. Joscelyne 2000: 94; Jaekel 2000: 159-160; Lagoudaki 2006). However, organizations and individuals must take into account their unique needs and, in light of these, must evaluate the costs and benefits of CAT before adopting it. The introduction of a new tool will almost certainly impact the existing workflow, and the use of TMs in particular can affectboth positively and negativelythe translation process and product. Given the way TMs operate, any gain in efficiency depends on the TMs ability to return matches. Texts that are internally repetitive or that are similar to others that have already been translated (e.g. revisions, updates and texts from specialized fields) will tend to generate useful matches. Texts that are less predictable, such as literary works or marketing material, will not. Once matches are found, simply being able to automatically copy and paste desired items from the TM database or termbase directly into the target text can save translators typing time while reducing the potential for typographic errors. However, significant gains in productivity are usually realized in the medium to long term, rather than in the short term, because the introduction of CAT tools entails a learning curve during which productivity could decline. Moreover, a number of translators find these tools so challenging to use that they simply give up on them before realizing any such gains (Lagoudaki 2006). With regard to quality, CAT still depends on human translators; should a client have a TM that it wants used for its work, the translator has no control over its contents. Furthermore, the segment-by-segment processing approach underlying most TM tools means that the notion of text is sometimes lost (Bowker 2006). For example, translators may be tempted to stay close to the structure of the source text, neglecting to logically split

or join sentences in the translation. To maximize the potential for repetition, they may avoid using synonyms or pronouns. Moreover, in cases where multiple translators have contributed to a collective TM database, the individual segments may bear the differing styles of their authors, and when brought together in a single text, the result may be a stylistic hodgepodge. Although such strategies may increase the number of matches generated by a TM, they are likely to detract from the overall readability of the resulting target text. CAT also affects the professional status of translators, their remuneration and their intellectual property rights. For instance, some clients ascribe less value to the work of the translator who uses CAT tools. If CAT is faster and easier than human translation, clients may ask to pay less for it. When a TM offers exact matches or fuzzy matches, should the translator offer a discount? Clients may be even more demanding if they use their own TM to pre-translate a text before sending it to a translator. Yet, even exact matches do not equate to zero time spent; a translator must evaluate the suggested sentences and make adjustments depending on the communicative context. Traditional fee structures (billing by the word or page) may no longer be appropriate. In many instances, these are being replaced by hourly charges. In addition, the movement away from desktop TM systems towards Web- and server-based access to TM databases is giving more control to clients and less to translators, particularly freelancers. Under this model, translators are restricted to using the clients databases and do not have the option of adding matches to or retrieving them from their own TMs, which makes it difficult for translators to build up their own linguistic assets (Garcia 2008: 62).

Ethical, financial and legal questions surround the ownership and sharing of CAT data. The contents of a TM are source texts and translations, the ownership of which presumably remains with the client. However, when collected as a database, control and ownership are thrown in doubt. Translators may wish to exchange or sell a TM or termbase. However, confidentiality of the information may be demanded by the client or original owner (e.g. for copyright protection). Currently, translators often deal with this problem through contracts and nondisclosure agreements. Variations in laws across jurisdictions further complicate matters in a global translation market. In the future, it is hoped that legal remedies, clients licensing of databases to translators and good-faith negotiation of agreements that serve both translators and clients can establish best practices and put the industrys collective linguistic knowledge to good use (Topping 2000; Gow 2007: 175, 188).

4. Emerging possibilities CAT is here to stay, and translators will continue to adapt to its presence and its increasing importance. However, CAT will undoubtedly continue to develop at a rapid pace. Advances in TMs may include the introduction of linguistic analysis and the ability to recall the surrounding context of matching segments, as well as that of their corresponding translationsalready a feature of certain TMs. Moreover, in the case of fuzzy matches, current TM systems readily identify differences between the new sentence to be translated and the source text segment retrieved from the TM database; however, future enhancements may also indicate which elements of the corresponding target segment may need to be preserved or modified. Sharing across different products will become easier as

standards such as Translation Memory eXchange (TMX), TermBase eXchange (TBX) and XML Localization Interchange File Format (XLIFF) become more widely adopted (Savourel 2007: 37). In addition, the availability of open source products will make it easier for more translators to access CAT tools, while Web- and server-based access to TM databases is also becoming an increasingly common model (Garcia 2008). The Internet provides several atypical possibilities for CAT. For instance, crowdsourcing translations involves leveraging the knowledge and free labour of the crowd the general public. Collaborative translationoften undertaken as part of a volunteer effortis similar in that multiple translators participate, but it could be limited to selected or professional ones. Tools such as wikis and weblogs can serve these purposes (Bey et al. 2006; NETWORKING AND VOLUNTEER TRANSLATORS and WEB AND TRANSLATION). There will be no shortage of innovation while millions of technologically savvy people, speakers of dozens of languages, contribute daily to the information sphere that is our world. However, none of this will obviate the need for professional translators, equipped with specialized skills and knowledgeand even CAT toolsto continue their indispensable work into the next decades.

References Bey, Youcef, Boitet, Christian and Kageura, Kyo. 2006. BEYTrans: A Wiki-based environment for helping online volunteer translators. In Topics in Language Resources for Translation and Localisation, Elia Yuste Rodrigo (ed.), 135-150. Amsterdam/Philadelphia: John Benjamins.

Bowker, Lynne. 2002. Computer-Aided Translation Technology: A Practical Introduction. Ottawa: University of Ottawa Press. Bowker, Lynne. 2006. Translation Memory and Text. In Lexicography, Terminology and
Translation: Text-based Studies in Honour of Ingrid Meyer, Lynne Bowker (ed.), 175-187. Ottawa: University of Ottawa Press. Esselink, Bert. 2000. A Practical Guide to Localization. Amsterdam/Philadelphia: John Benjamins.

Garcia, Ignacio. 2008. Power Shifts in Web-based Translation Memory. Machine Translation 21(1): 55-68. Gow, Francie. 2007. You Must Remember This: The Copyright Conundrum of Translation Memory Databases. In Canadian Journal of Law and Technology 6(3): 175-192. Jaekel, Gary. 2000. Terminology Management at Ericsson. In Translating into Success: Cutting-edge strategies for going multilingual in a global age, Robert C. Sprung (ed.), 159-171. Amsterdam/Philadelphia: John Benjamins. Joscelyne, Andrew. 2000. The Role of Translation in an International Organization. In Translating into Success: Cutting-edge strategies for going multilingual in a global age, Robert C. Sprung (ed.), 81-95. Amsterdam/Philadelphia: John Benjamins. Kay, Martin. 1980. The Proper Place of Men and Machines in Language Translation. Research Report CSL-80-11, Xerox Palo Alto Research Center, Palo Alto, CA. Reprinted in Readings in Machine Translation, Sergei Nirenburg, Harold L. Somers and Yorick A. Wilks (eds.), 221-232. Cambridge, MA: The MIT Press.

Lagoudaki, Elina. 2006. Imperial College London Translation Memories Survey 2006. Translation Memory systems: Enlightening users perspective. Online at: http://www3.imperial.ac.uk/pls/portallive/docs/1/7307707.PDF Lange, Carmen Andrs and Bennett, Winfield Scott. 2000. Combining Machine Translation with Translation Memory at Baan. In Translating into Success: Cutting-edge strategies for going multilingual in a global age, Robert C. Sprung (ed.), 203-218. Amsterdam/Philadelphia: John Benjamins. Quah, Chiew Kin. 2006. Translation and Technology. Houndmills, UK/New York: Palgrave Macmillan. Savourel, Yves. 2007. CAT Tools and Standards: A Brief Summary. MultiLingual 18(6): 37. Somers, Harold. 2003a. The translators workstation. In Computers and Translation: A Translators Guide, Harold Somers (ed.), 13-30. Amsterdam/Philadelphia: John Benjamins. Somers, Harold. 2003b. Translation memory systems. In Computers and Translation: A Translators Guide, Harold Somers (ed.), 31-47. Amsterdam/Philadelphia: John Benjamins. Topping, Suzanne. 2000. Sharing Translation Database Information. Multilingual Computing and Technology 11 (5): 59-61.