My robotic friend? – My (belated) foray into Machine Translation POst Editing (MTPE)

 

As I wrote a few weeks ago, I have made the strategic decision to focus on my competitive advantage – Hebrew to English legal translation. The practical significance of that decision is that I must maximize its potential. Thus, this last week I took on a project involving editing machine translation of an insurance contract. I had previously avoided such projects due to their idiot-savant nature. The project confirmed many of my concerns but, in contrast, demonstrated the advantages of working on machine translation. I discovered that they could indeed be satisfying, both financially and emotionally.

In explanation of the term, machine translation does not necessarily refer to AI engines such as ChatGPT and may include older methods such as Google Translate. The term machine translation designates the initial use of a digital linguistic tool that translates the source text by applying similar patterns in a database, whether vetted and closed or open, Internet-based. Machine translation has existed for several decades, initially through translation memories developed by translators, agencies and companies and expanding to more sophisticated ones based on neural networks. The European Community has developed one of the most specific and sophisticated ones based on previous translations of all EU laws into all of the languages of the community. The open machine translations, notably Google Translate and AI, use statistical probability to choose the most probable translation available on the Internet. The quality of machine translation varies depending on the algorithm, language combinations and sources.

The resulting translation generally resembles one produced by an idiot-savant, which requires neither pure translation nor pure editing. To explain, if a human produces a poor text, it is far more economical in time and energy to retranslate from scratch. Simply put, the editor does not trust anything the original translator did. On the other hand, an editor, identifying an excellent translation, trusts the resulting text in terms of content and merely makes tweaks to improve the language. Furthermore, the editor learns to find a pattern of these mistakes and focus on them. In any case, two pairs of eyes are always better than one, regardless of the skill level. By contrast, machine translation, in my limited experience, produces highly uneven and unpredictable results. One sentence can be perfect, even better than one the editor could write. The next one can be a complete disaster and require complete rewriting. Even more difficult, a given translation may appear correct but closer analysis shows small but significant errors. It requires careful attention to identify those issues. Thus, machine translation is not consistent in quality nor are its mistakes predictable.

In the text I did, the translation engine, DeepL, produced a mixed bag. On the one hand, there were very few content mistakes, i.e., a reader could correctly understand the meaning of the vast majority of the provisions, albeit with a bit of effort and a few terminology errors. On the other hand, it was clear that a human translator had not produced the text. Here is a partial list of the error types:

1.     Articles (he vs. it)

2.    Modals (misuse of “shall” to indicate future instead of legal obligation)

3.    Literal translation of phrases (has the right to instead of may)

4.    Inconsistent capitalization (company and Company)

5.    Translation of the name of the Company

6.    Keeping sentence in the passive (The premium will be paid… vs the Policyholder must pay……

7.    Misplaced adjective (the benefits retained vs the retained benefits)

Thus, the machine translation, while accurate, was not correct.

Upon completion of the project, I decided that I would take on more such projects. Granted, it required great attention, with many breaks, to catch the issues and improve the text. However, the original text was better in some ways than that produced by far too many human translators. Moreover, as I knew that no human was responsible for it, I did not get annoyed. Since I had priced the project by projected time after viewing the translation beforehand (which turned out to be fairly accurate) and offered two different quotes, light and heavy editing, the compensation was more than acceptable. Most importantly, the final text read well, always a satisfying result. Thus, I will now take on more such projects. Maybe robots could be our friends.

AI and the future of freelance translating – a perspective

 

Freelance translators, like many other professionals, see dark clouds. The media feeds stories on the ever-improving ability of AI to translate. Strangers innocently ask why anybody needs translators anymore. Customers and income decrease month to month. It is all quite depressing but not necessarily a full or accurate picture in the long term. On the contrary, paid translation needs are actually expanding. Moreover, the market niches that AI is destroying have been in decline for over a decade due to technological changes. In practice, AI changes the translation business but not only does it not eliminate freelance business but can even provide an opportunity to expand. It is reasonable to be cautiously optimistic despite all the apparent omens.

In terms of current trends for language service providers, which includes both agencies and freelancers, the future seems quite optimistic. Based on the total volume of the worldwide agencies, demand for linguistic services continues to increase steadily. Experts predict that the value of these services will increase approximately 28% from 2024 to 2027 to around 90 billion USD. World trade and the needs of international commerce will continue to feed the demand. To be fair, international agencies are taking a lion’s share of business with freelancers struggling with downward pressure on their rates. B2B business, without agencies, requires more marketing effort, skill and confidence, which many freelancers lack. Yet, in practice, there is a steady demand for translators.

It is important to note that translation technology, which includes but is not limited to AI, shapes which niches will remain and even expand and which ones will decline and disappear. For ten years, machine translation of all types has automated the translation process. Computer Assisted Translation (CAT) and translation memory began defining the work process over 15 years ago. Machine translation, most notably Google Translation, has made simple translation accessible and free to the average person for almost 20 years. More specialized translation memories, in particular neural translation in recent years, make it possible to effectively translate large masses of specialized legal and other material in a short time. There is less and less work available for a general translator because of the plethora of no-cost and sufficiently effective alternatives. By contrast, these machine translations, including AI, struggle to produce effective results when the message goes beyond mere understanding but requires precision or a human touch.  Some fields suffering from a lack of proficient human translators include medical, marketing, legal and technical translation. Furthermore, the need for official certification of government documents for court and bureaucratic purposes creates a steady market for certified translators of all types. Specialists can find lucrative niches.

The various language technologies have changed the whole panorama of translation in terms of methods and tasks. The use of CAT tools is a requirement for many projects and has significantly increased productivity and shaped its rates. Machine translation serves as a basis for many initial drafts, either in terms of suggestions or complete translation. AI can instantly produce a large-scale translation, albeit of highly uneven quality. Thus, the translator’s work may involve editing machine translation, actual translation or both. Clearly, not every freelancer wishes to be involved in editing but those that accept it and do it efficiently and effectively are in demand. By contrast, those freelancers that completely reject technology find their market shrinking. The name of the game is constant adaptation.

Thus, it is clear that translation is not only not a dying profession but instead one with a future. Technology will shape its future, as it has done in its past and present. Specialized and flexible translators can find an opportunity to make a living. The most difficult period is the transition during which the advantages and limitations of each new digital tool emerge and define the market. AI is not the end of human translators just as Google Translate and its cousins were not. They merely shaped the profession. It is most probable that for the foreseeable future human translators will continue to handle those tasks where it is important to fully convey the meaning of one language in another language and where approximation is not sufficient as well as ensure that machine translation does not create unnecessary or even dangerous mistranslations. Many current AI uses will return to human translation as issues arise from AI translation.  I am cautiously optimistic about the future of translation despite AI.

Talking about the elephant in the China shop – should translation buyers use AI/machine translation?

 

[elephant]

The most common question non-translators ask (in one form or another) after I say that I am a professional translator is whether anybody needs me anymore. Likewise, when translators gather, the hot topic is the impact of AI on the business in general and the person in specific. AI carries the image of a drum roller machine, flattening anything in its path and whose existence many prefer to ignore out of dread. Going beyond this fear, it is legitimate to consider the role of machine translation, whether of the simple Google translation, more complex neural network or Hogarthian ChatGPT type. More specifically, I will address the blunt question when a translation buyer should pay money and employ a human translator.

As a matter of introduction, each of the three main machine translations types in their various guises attain their translation results in slightly different ways. The simplest, Google Translation and its cousins, search for the most common translation of the term as it appears in bilingual texts in the Internet and any online glossaries. The results are free but not always very relevant for the context. Neural translation and other more specialized methods selectively pick corpuses, including approved high-level bilingual texts such as from the European Community or the UN, and search for terms based on the type of the texts, e.g. legal and financial. This data base is more precise but is time-consuming to build. However, the results are generally far more relevant Accordingly, this method often involves some buyer cost to cover the development costs. Finally, AI translation uses a probability algorithm based on a broad internet data base and uses prompts to fine tune the results. The costs vary on the engine. Accurate results are somewhat dependent on the ability of the person to write prompts and specify the desired result. Even in the best circumstances, AI results tend to be a bit idiot-savant, i.e., ranging from brilliant to imbecile.

Here are some pointers regarding the appropriate use of machine and human translators:

·   To state the obvious, there is no need to pay a human translator to translate most texts for private use. If the goal is to understand more or less the content of an email or website text, any of the three types of machine translation will produce a sufficiently clear result, albeit occasionally with amusing terminology errors.

·   For longer texts, when time is a premium, it is possible to use a neural network or ChatGPT to produce a reasonable translation for a small internal audience. The purpose of such translations is essentially to share information, reducing the impact of any inaccuracies.

·   For mass translations where the cost of human translation is prohibitive, such as in identifying the content of  a large volume of legal documents or producing descriptions for a multilanguage low-cost online site that does not have the profit margin to employ human translators, machine translation might be a solution although the poor quality of the less expensive options may negatively affect results.

·   For texts aimed to make an impression on the public, including marketing and menus, businesses should employ a human translation. In these cases, it is not sufficient to transmit the facts. There is a need to persuade and impress. The public often equates the care invested into the text with the care invested in the product. Seller, beware.

·   For texts with legal and medical consequences, among others, translation buyers should avoid using machine translation. The consequences of a poorly translated legal brief or medical device instruction manual far outweigh the cost of a proper technical translator.

·   In some cases, government authorities require human translation and a signed certificate of accuracy. They do not accept self or machine-made equivalents. In these cases, read the requirements very carefully. For more information on what exactly a certified translation is, see here.

In summary, if you desire more than to attain information, employ a human translator. The costs of the poor results will far exceed any savings from free or low-cost machine translation. In other words, to talk about the elephant that is machine-translation, it should not be let into a china shop of fragile words as it tends to be somewhat clumsy but it is acceptable to let it into the gym as long as you clean it up afterwards, no bull.

Staring at the technological future and present in the eyes – lessons from the 2023 conference of the Israel Translators Association

 

[owl*]

Nothing is more frightening than change and the unknown. This past week, the Israel Translators Association hosted several industry and technological leaders, who addressed the future of translation in the face of fast-developing technology. The participants in the conference gained an understanding of the impact of new developments in the field and practical approaches to future growth. The level of the presentations, both in terms of content and delivery, was extremely high without any significant overlapping, thus providing a perspicuous and broad view of the situation. The lecturers included Keith Brooks (To Grow, or Not to Grow, That is No longer the Question, But the Imperative), Zvi Gordon (Technology in the translation industry: current picture and a look at the future),  Katia Jimenez (Understanding the progress of artificial intelligence in language), Kirti Vashee (The changing translation technology landscape), Nora Díaz (Translation and Interpretation Technology: The Basics and Beyond) and Rafa Lombardino (Language Professionals: Technology is On Our Side), to name just a few. My conclusion from these lectures was that not only was technology already here and a part of the industry, it does not actually threaten translation or translators, nor does the new wonderchild, ChatGPT. Just as importantly, translators can and should study the emerging technology and harness it to their specific needs.

The dominant message was that the various forms of machine translation had already established a strong presence in the industry but not necessarily at the expense of human translators. Specifically, certain domains have fully adopted every-improving version of machine translation and use them almost exclusively. Interestingly, these domains are places where human translation could not tread, i.e., where the sheer volume, potential cost and limited context made machine translation the ideal tool. For example, companies such as Airbnb, Ali Express and Amazon need to localize thousands of words on a daily basis. By contrast, in areas where context, accuracy and style count, such as in law, medicine and marketing, the worldwide volume of human translation keeps on expanding. Referring to the human element, Ellen Elias-Bursać said in her lecture on interpreting and translating during the Hague War Crimes trials that the translators had to stand in court and justify their translation. Where ever there is that potential, whether in court or in front of any person, human translation and interpretation is the best option.

As for the latest craze, reaching 1 million users in only four months, ChatGPT is very interesting but not actually a threat to translation. Specifically, the speakers noted that the machine compiled text from existing corpus without discrimination of accuracy, bias or style of content. In other words, it was a random generator of text, somewhat regulated by the limitations imposed by the party entering the request. It can be valuable for identifying grammar mistakes or improving style or vocabulary, especially for non-native writers. However, as its output is not specific or accurate enough for most translating assignments, its value as a translator is quite limited.

In practice, the speakers consistently spoke of the need for translator to investigate any technology that may improve their work content or process. Clearly, it is possible to be successful without applying most if not all of the technological tools, but future success will probably depend on selective use of modern methods as well as understanding their advantages and limitations. The message for translators was that the future involved not only selectively adopting new technology but also changing the attitude towards it from complete fear to measured understanding.

As a demonstration of that approach, one of the speakers mentioned fully automated cars, noting that some major companies have stopped investing money on their development but are actively using them for moving goods from one building to another on a clearly defined route. The moral here is that not that the technology is not valuable but its effective use is limited by its lack of ability to make complex judgments. Likewise, where judgment is vital, human translators will have work. Yet, those that better leverage existing and emerging technology will have an efficiency advantage. To identify the appropriate tools, more than ever, translators need to keep an open mind in regards to technology. It is far less frightening than it would seem and may be quite benevolent. The future is in our hands as we learned at the ITA 2023 conference.

* Picture captions help the blind fully access to the Internet.

Picture credit

Machine translation and Orwell

For many translators, machine translation is a combination of nuclear war and global warming. There is a sense that it will wipe them out from the map but they hope that it won’t happen in this generation. This week, I skimmed through two articles discussing MT. My thoughts were led not the future of the profession but to the future of English.

The first article, written by Florian Faes and cited by Slator in its weekly newsletter, discusses the linguistic differences between texts translated by MT as compared to human translation after editing. Among the writer’s conclusions, albeit on a limited literary sample, was MT texts tend to have a higher resemblance to their originals in terms of structure even when this structure differs in the target language as well be “simpler and more normalized”.

In another article, also cited by Slator, Jochen Hummel, the creator of the Trados computer assisted translation (CAT) tool, declared that his tool would no longer be used in the future but instead all human translation would be based on MT. In other words, the opus of previously written text will standardize our language. As I see it, what was will be in a much stronger form than today and subject to manipulation by corporate and governmental organization.

These two observations led me to recall Orwell’s 1984. For those who have forgotten or simply never read the book, he described a world where the government controlled everybody (Big Brother). Interestingly, one of its main tools was its control of language. English vocabulary had been reduced to the bare minimum. For example, a negative was expressed by adding “un” to the positive, e.g., unhungry. All texts were online (yes, this book was written 1949) and amended as political winds changed so that the public never had any proof of any change in policy or thinking. Looking at North Korea in 2019, Orwell would be appalled but not shocked.

Given that English is the dominant language of communication worldwide for the foreseeable future and assuming that machine language, however “artificial” it sounds initially, becomes the statistical and controlling norm, it is not hard to imagine a world in which people’s thoughts are expressed in intentionally simplified language and form. It would be undoing Dickens and Shakespeare, to name a few.

It is clear that MT has its place and will not disappear. It also clear that MT will have a huge influence on translation project management. Looking at the development of chat language, which is also simplified in many ways, it can be argued that this language is no less rich and individual. However, I still fear the gradual “poorification” of English not as a result of government action but instead due to industrial pressures.  I hope Orwell was wrong in this prediction and that 1984 will never come.

Digital Idiot Savant

The world of translation for both the general public and professionals is the midst of a revolution.  Machine translation has taken off.  Google Translate may be its most public form but far from its most important use. Corporations such as Nestle and Amazon are using and developing better forms of machine translation. 

To explain the process, phrases and sentences are compared with company-prepared glossaries, known Internet-accessible translations and grammar rules to create translated documents. Of course, as anybody that has ever used Google Translate can testify, the results are sometimes ludicrous but more and more often quite satisfactory.

Recently, I post-edited a very long machine translation of a complex tender offer in French.  I felt I was dealing with an idiot savant in the sense that genius and stupidity were randomly mixed. While for confidentially reasons I cannot provide specific examples, I can say that a perfect translation of a complex sentence would often be followed by an irrelevant translation of a simple sentence. The same word would be translated differently in consecutive sentences. The grammar ranged from Oxford correct to awful first year ESL student. In short, unlike human translation, there was no rhyme and reason to the quality of the translation.

This required me to treat each sentence as completely isolated in terms of my confidence level in the translation. When editing human translation, it is a bit like observing the driver ahead of you: you quickly get a sense of whether to trust or avoid him/her. Here, my mind had to refuse to trust any translation based on the previous segments. Even harder psychologically, I could not even say to myself “what an idiot” or “what a good translator” because the translator was digital. All in all, it was a very different editing experience.

Many translators fear that machine translation is the end of the profession. The probable truth is the opposite. Translation is one of the fastest growing professions in the world thanks to the world-village phenomenon, among other reasons. It is clear that machine translation handles certain jobs, especially large masses of text and very standard email messages, much more efficiently and cost-effectively than human translation. However, technical translation of all kinds, including medical and legal, requires the human brain both with and without computer help. As we have all experienced, there is nothing more intelligent and stupid than a computer.