Recent Trends in Machine Translation”

Written by: Abir HASSSAN- Bassima KHALED- Rania FOUANI   I. Abstract Miller (2008) in his article titled “Recent Trends in Machine Translation” indicated that In the last two years machine translation (MT) has embarked on a voyage into the future, spurred by the presence of personal computers on individual desktops throughout the world and, more recently, universal access to electronic text on-line. This impressive growth has led to many new trends, including major changes in the profile of the user. Apace with this trajectory has come better communication and increased collaboration between all the groups concerned—MT researchers, developers, users, and watchers. The International Association for Machine Translation (IAMT), together with its three regional associations created in 1991, has fostered this convergence by creating opportunities-workshops, conferences, publications—through which to share the latest information in this dynamically growing field. II. Commentary A. New Dimensions in MT Service Delivery a. The dreams of yesteryear's visionaries are finally coming true. Machine translation (MT) has launched on an unparalleled surge of growth—a historic shift in the way it is being used and a phenomenal increase in the number of people who rely on it. We now have MT software that is viable, affordable, and runs on virtually any 1990s desktop. Today there are more than 500 vendors of MT software for the personal computer around the world, and among them they put out well over 1,000 products.3 One of the vendors, Global link, sells its extensive line of software in at least 6,000 stores in North America alone, and at present Europe is its fastest-growing market. The ubiquity of the desktop computer with access to the Internet has given momentum to an unprecedented growth in MT user ship. We now have MT on-line, accessible through e-mail, client server arrangements, Internet service providers, and a growing number of other sites on the Internet. The on-line phenomenon is changing our whole way of thinking about machine translation. Together, these two developments-the abundance of low-cost MT in shrink-wrapped boxes, coupled with MT on-line—are turning machine translation into an everyday commodity that is within the reach of virtually anyone with a late-model personal computer. The sudden shift in MT use and the dramatic increase in its user ship have also brought a sea change in the profile of the user. Because of its widespread availability, MT has been forced, appropriately or not, to graduate from the days when a system's caretakers had to nurture it constantly in order for it to perform acceptably. It now stands on its own, and, by and large, its new users must fend for themselves, whether by customizing the system and/or learning how to post edit, or accepting the output as it is. While all these changes are taking place in MT use, other exciting trends are also redrawing the entire map of the field of machine translation itself. Many languages, especially the more challenging ones, are being tackled and added to the vendors' repertoires. In the new game of "plug-and-play," MT engines are now being made inter faceable with a variety of other software functions. Speech translation is making steady progress. Off-the-shelf tools are speeding up research and development. Creative partnerships are being forged between and within the commercial and academic communities. Systems of different philosophies are being joined together. Indeed, on all fronts MT research is accelerating its ongoing march toward distant horizons. It's safe to say that never in the history of this field has so much happened within such a short period. b. We have all witnessed the explosive expansion of the World Wide Web, the Internet service providers, and, most recently, the intranets. Not many of us, however, are aware of the extent to which machine translation is being swept along in this tide. Already on-line access is causing MT use to grow at an unprecedented rate. As of September 1996, low-cost machine translation in one form or another was available at some 30 on-line sites in cyberspace. It comes in a variety of forms and modalities. MT vendors are also currently gearing up for intranets. On-line purchase is yet another way to go. As with many other kinds of software, the vendors make it easy to order an MT package on- line. In fact, we predict that within a few years the shrink- wrapped box will have yielded almost entirely to on-line sale/purchase arrangements. This growing use of MT on-line cannot be dismissed as casual curiosity. Unlike software purchased off the shelf, for which no direct measurements are possible, on-line access is documented automatically, and therefore patterns can be discerned. For example, the records for CompuServe's production translation service show a number of repeat large-volume users. The statistics (ibid.) reveal that about 85% of the requests are for raw MT—a much larger percentage than had been anticipated. What could not be determined automatically was whether the raw translation was being used for gisting purposes only or whether it was being post edited for further use. To discover more about its subsequent fate, Flanagan conducted a market survey which revealed that the CDTS is used mostly for business and technical purposes where assimilation- quality MT is sufficient (ibid.). The bottom line is that the customer is willing to pay for this service. In the World Community Forum, although there is no direct evidence of the extent to which the machine translations are being relied on, at least one fact can be reported: the Forum's sysop is inundated with complaints on the rare occasions when the MT system goes down. In these circumstances, only a fully automatic process capable of handling very large volumes of text with near-real-time turnaround can provide the translation capacity required by on-line markets. Flanagan (ibid.) also points out that the on-line culture favors rapid and shallow assimilation of information. For these reasons, MT is an ideal fit. B. The New User/Consumer Profile Now that we have seen the new trends in MT from the point of view of the general public, we should look at the perspective of the user and the end consumer. 1. Acceptance: Purpose of translation (a new typology). Traditionally MT usage has been classified according to its purpose. It is considered to be either translation for dissemination, or translation for information purposes only, also known as "gisting" or assimilation. At this point we would like to add a third category and at least two types of each. Type 1 represents the more direct use of MT in one of its natural niches, while type 2 is a further development that requires greater human intervention at some point in the process. Problems: Heavy post editing, judgment calls are time- consuming, domain drift, hence need for improved quality; few systems perform well in this arena; linguistic development investment difficult to target. What it takes: post editing aids; very large and sensitively coded lexicon(s), easy to update (better a combined dictionary than "topical glossaries"); parser and rule base; filters and translation memory also helpful. 2. Assimilation: a. "Raw" MT for gisting, sometimes automated post editing; broad range of subjects. Problems: Quality tends to be poor. What it takes: Very large and judiciously coded lexicon(s), easy to update (better a combined dictionary than "topical glossaries"); parser and rule base. b. Problems: Lack of public awareness of this option; shortage of suitable post editors, translators often not able to relax standards. What it takes: Good quality; large and richly coded dictionaries.  III. Conclusion and Recommendation To sum up, MT Service Delivery has new dimensions where the machine translation launched on an unparalleled surge of growth. In addition to a new User/ Consumer Profile. It is essential for a translator to know that as high-level executives begin to see the huge value and market enabling power of translating large amounts of relevant content, we can expect to see that translation will be viewed as a much more strategic core competence. As this happens, translation professionals could become facilitators and enablers of many key conversations between global enterprises and their customers. The skills required will include the following (and many are just emerging so this is a great opportunity for innovators and leaders): 1. customization of MT systems for specific business purposes; 2. corpus analysis and assessment skills; 3. evolutionary approaches to making high value content multilingual; 4. rapid quality assessment skills; 5. linguistic steering of automated translation systems; 6. community and crowd collaboration management and administration to do a variety of linguistic work; 7. more structured approaches to post-editing MT to enable rapid error identification and correction; 8. continuously evolving and learning MT systems that produce on-going improvements in translation quality; 9. much better and more robust data interchange standards will likely develop Systran: Since September 11, 2001, the warlike spirit which blows on Washington seems to have swept these scruples. Reverso: Since September 11, 2001, the warlike spirit which blows on Washington seem to have swept (annihilated) these scruples. Human translation: Since 11 September 2001 the warmongering mood in Washington seems to have swept away such scruples.   VI. References Anoun, H. (2006). Towards a Logical Approach to Nominal Sentences Analysis in Standard Arabic. In Proceedings of the Eleventh ESSLLI Student Session. Badawi, A. Elsaid, L. Mike, and G. Carter, G. (2004). Modern Written Arabic: A Compre-hensive Grammar. Routledge. Baerman, M. et al (2006). the Syntax-Morphology Interface. A Study of Syncretism. Cambridge Studies in Linguistics. Cambridge University Press. Bar, H. and Yoad, W. (2005). Choosing an Optimal Architecture for Segmentation and POS-Tagging of Modern Hebrew. In Proceedings of the ACL Workshop on Computational Approaches to Semitic Languages. Ann Arbor, Michigan: Association for Computational Linguistics. (pp. 39, 46) Bar, H. and Yoad, W. (2005). Choosing an Optimal Architecture for Segmentation and POS-Tagging of Modern Hebrew. In Proceedings of the ACL Workshop on Computational Approaches to Semitic Languages.