AI and I – an attempt at perspective

 

I admit to feeling somewhat hostile about the idea of using AI in my work as a professional translator.Thus, I felt (and still feel) a need to ascertain whether my internal resistance is stubbornness or intuition, two similar but not identical impulses. Consequently, I invested the time to listen to Doron’s Tzur excellent 2-hour presentation to the Israel Translators Association on March 26, 2025 on the subject of AI for translators. He gave an excellent introduction to the approach of AI as well as its practical use in translation, providing examples using several different AI applications. Granted, this presenting does not make an expert on the subject but the lecture did provide me with some updated data. I thus feel more at ease discussing my perspective. I now understand that an AI-based translation method does not fit me in terms of task/time allocation, the resulting standard and the produced style. Yet, while I do not see AI as a practical tool in the initial translation phase, I definitely can see its value in the editing stage, when it is possible.  In short, my personal answer is: “No, but”.

One of the Doron’s first points is that AI can and does change the time relationship between translation and QA, which include both finding errors and improving style. Specifically, he noted that the initial translation process, what he called the white paper stage, using AI takes minutes, if not less, allowing translators to focus their energy and skill and improving the text, rendering it both accurate and human. It should be noted that many translators, including myself, actually prefer the creative stage, the writing of the first draft, over the editing stage. An important passion in translation is the joy of playing with words, preferably in at least two languages. Thus, as demanding and time-consuming as translation can be, the work is fascinating and satisfying. By contrast, editing is a technical skill requiring great attention to detail and great discipline. Consequently, as I have no great desire to specialize in editing, I prefer the traditional method of translation, i.e., without AI.

Another point made by Doron is that AI is becoming or has become the new standard. He noted that it is possible to produce a workable translation of even long texts in hours, not days. I certainly do not dispute his assessment of time requirements or creation of a standard. However, I do challenge the worthiness of the standard. Written communication, especially in English, necessarily involves polishing and repolishing. Each QA session reveals more underlying issues, theoretically infinite but in practice limited by delivery deadlines. Clearly, any translation produced and edited in three hours lacks that polish if even it suffers from no concrete language error. Thus, it is possible to produced rushed translation but the issue is to what standard.  I am aware that, historically, people have come to accept lower quality standards, e.g. cars and clothes, but I find that disturbing.

Finally, AI, because it is based on statistics, not intelligence (Doron’s words), produces a clear but somehow flawed text based. To paraphrase Doron, average is never good as the 50% of the other texts are better.  The actual level of finishing depends on the skill and knowledge of the editor, many of whom are not even native speakers. Consequently, the AI-produced and human-edited translation often is correct but off, like a fluent but non-native speaker telling a story. Everything is clear but a real person would not write that. For many purposes, this artificialness is not an issue, in particular when a person only requires the translation to ascertain the main idea or find a detail. However, if the writer wishes to move the reader to buy, be inspired or any other action, this dissonance ruins the effect. People believe writers and speakers because they are authentic even if their facts are not quite accurate. Effective written communication must reflect the writer’s voice, not the statistical average. Therefore, I feel that using an AI-produced translation does not serve the needs of most if not all my customers’ needs.

One of the most attractive features of AI was its editing capacity. The ability of AI to identify errors of all types, including context-based spelling errors on some applications, is highly useful and a significant improvement on Word’s Spellcheck and even Grammarly, to name a few non-AI applications. However, to use these AI-tools efficiently requires consistent effort and time investment in attaining the computer skills and keep up with the weekly changes and newest versions. Moreover, translators cannot upload many legal, commercial and medical documents  due to confidentiality issues. I would love to upload the first draft 18,000 words I am working onto AI but will not do so because of the fear of it entering the public realm. Finally, it should be noted that most older people and quite a few younger people are not native to technology.  Yet, for many translators, using AI for editing makes sense both in terms of effectiveness and economics.

If I view the translation economic equation as time/effort as compared to income, at this moment at least, I feel that it is correct decision for me is to ignore AI but to keep an eye on it. AI is not the end of translation just as Computer Aided Translation and Google translations did not destroy the profession. Clearly, AI will change the industry but in what manner and which degree, nobody can say. It remains for each translator, or any other professional for that matter, to examine and decide, as Kirstie and Phil would say, to love it or leave it. I am leaving it for the moment.

Paine-ful wisdom – on translators, editors, between them and beyond them

 

[The Odd Couple]

This week, I sent the following email to a trusted project manager (PM) after reviewing the changes her editor had made to my translation: “I reviewed the edits and noticed that the editor found a few minor errors [ones that do not affect understanding], introduced a few minor errors, improved the phrasing in a few places and mainly made preferential changes. The most important thing is that the customer receives the best possible translation.” The translation involved was a two-page bank declaration in convoluted Hebrew (aren’t all bank documents in convoluted language?) on a poor PDF delivered in 24 hours.

This unpleasant interaction between translator and editor, with the PM in between, led me to reconsider the Odd Couple-like nature of the relationship between technical translators and editors. Limited by time constraints, translators must convert a text in the source language, often not very easy to read due to the quality of the PDF, into a Word document in the target document acceptable for use by the customer. This task involves reading the document many times in several forms, including bilingual and monolingual as well as on screen and on paper, all within a short period of time. Not only is it difficult to read the document “freshly” every time without sufficient time gaps, another pair of eyes will always find ways to improve the translation. Thus, the even the most professional translations are not perfect. 

By contrast, the editor’s task is read the monolingual document, comparing it when necessary to the source document, identify any errors and refine the existing text until it shines. They, of course, apply their judgment and opinion of correct language, which may not be identical to the translators’ view. The task is far more focused and involves many less read-throughs. Details matter, as Felix Ungar would say, without consideration of the emotional reaction of the translator that had put so much effort in producing the text. The result is that many translators feel that editors abuse them, as reflected in a typical emotional response: ”This editor had to prove his/her existence.” Clearly, there is generally little love between technical translators and editors.

However, if the goal of the translation process is to produce the most faithful and well-written document for the customer, i.e., to provide value, both the translator and editor are essential to the process. No one pair of eyes can attain that goal regardless of the amount of knowledge and experience. In the best possible world, the PM understands this.  In my case, she will continue to work with me as she has done for many years as I provide a solid translation on or before the deadline, exactly what is required of me. Likewise, she will continue to depend on that editor to improve translations in order to retain her customers. Unfortunately, this global understanding of the goal is far from universal. Still, it is important for translators to keep it in mind when receiving "massacred" versions of their translation. In fact, the words of Thomas Paine ring true: “The real man smiles in trouble, gathers strength from distress, and grows brave by reflection.”

Cleaning costs – the uniqueness of editing

 

[text with editing marks*]

Certain tasks are standardized in terms of process and price. For example, an oil change essentially involves the same materials and time almost regardless of the actual car model and mileage. By contrasts, others, including house cleaning and language editing, involve unique situations in terms of initial situation, personal skill and purpose, rendering it impossible to state how long the task should take without knowing all facts. Thus, for such processes, the service provider, needs to carefully consider all factors before providing a quote while the service buyer should not expect a standard rate.

While I will be discussing specifically language editing, I will first compare it to a more common service, house cleaning. The issue of how long it should take to clean a flat of 100 m² (1076 sq. f.) is actually quite complex. First of all, the first factor is the initial state of cleanliness and order. There is a marked difference between the condition of a flat that has been regularly cleaned as compared to one on which the contractors have just finished renovation. It can range from 2 hours to 1 full day even. I say that from experience. Likewise, cleaning workers vary considerable in their skills and speed. Speed and quality are often but not necessarily related, meaning slower work sometimes translates into more thorough results. Finally, the required time is often dependent on the desired result. Making the place livable, preparing for a visit from your mother-in-law and staging the house to put on sale require a significantly different level of finishing. The more polishing, literal and figurative, the more time is needed.

Editing, a form of cleaning if you look at it from a certain point of view, is similar. First, no two people write the same way, have the same knowledge or spend the same time of time and effort in preparing the text. On one extreme, quite rare, are skilled writers seeking a second pair of eyes to identify any remaining minor issues. In this case, the editor has very few mechanical issues to deal with and serves essentially as an important final QA check. On the other extreme are texts written by non-natives that may not know how to write properly in their own language let alone the language of the text. They are wise enough to recognize that the text is unacceptable and needs to be revised. In this case, the task of the editor is far more comprehensive, involving identifying grammatical and syntactical errors, rephrasing ideas and even sometimes reorganizing whole sections. This process involves multiple readings, ideally with each run-through focusing on specific issues. The time investment is significant in terms of the ratio between time and words.

Clearly personal approach affects the amount of time required for editing. Editors vary in the speed in which they process text, with some quickly identifying errors while others working slowly to fully analyze the text. There is some connection between thoroughness and speed, i.e., slower editors catch more issues on a given read-through than faster editors but additional directed readings can compensate for any shortcomings.  Likewise, editing requires concentration. While a few people can maintain full focus for long periods of times, most editors require various frequencies of breaks in order to maintain their focus. The longer the text, the greater the impact of these necessary breaks. To put it simply, it is essentially impossible to read 10,000 words in one reading session. Not all editors are created equal.

Finally, different text purposes involve different levels of polishing, radically affecting editing time. On the fast end of the scale, if a document is an internal document distributed to a limited number of persons, the language must be clear and accurate but it does not have to be particularly elegant. A document with a large and/or potentially wide audience, such as a published article may affect perception of the party ordering the editing and should reinforce a positive image. Finally, editing documents intended to persuade must go beyond correct language and effectively express the intended message, which sometimes involves transcreation, i.e., completely rewriting of the text, a creative and long process. Such documents include sales material and website text. Thus, the editor needs to know the purpose of the document in order to estimate the required level of polishing and time.

Curiously, industry standards range from 1000 to 2000 words per hour. However, as explained above, “standards” are not relevant for editing. In practice, each document is unique and quoted according to its base level, the required skill of the editor and the level of polishing needed. When editors do not take these factors into consideration, they often find themselves selling themselves short and unable to take on further work due to extended time on the current project or produce inferior work. Editing purchasers should appreciate a carefully produced price proposal, understanding that an editor that invests time to understand its texts and meeting its needs will probably produce a desired result. Since every text, every circumstance and every editor are unique, so will be the costs of editing. One size does not fit all.

* Pictures captions help the blind fully access the Internet.

Picture credit

In reconsideration – Karen Tkaczyk on editing and proofreading

 

[edited document*]

As part of a series of podcasts, Paul Urwin of Proz.com interviewed Karen Tkaczyk, an experienced editor and translator. As the podcast is not a formal workshop but instead a series of questions and answers, Karen was unable to go into depth on many of the important issues she raised. I wish to add some thoughts on three matters that are of importance to both translators and editors, specifically the obligation to proofread, the time involved and criteria for editorial intervention.

[eye chat & glasses]

Both Paul and Karen were very polite and understated in their statements that translators must include proofreading in their translation process. At minimum, this QA involves spellcheck and a quick rereading. Ideally, they recommended use of a QA tool, whether one intrinsic to the CAT tool and/or an external program such as Xbench. Ideally, a second pair of eyes, as required by ISO standards, would read the text but freelancers often find this step impractical due to time or budgetary factors. However, clearly failure to apply any proofreading process leads to substandard work. In my experience, far too many translators, even experienced ones, expect someone else to perform this step. While it is very challenging at best to produce a perfect translation, that difficulty does not excuse translators from the obligation to strive to do so as much as circumstances allow. Proofreading and editing are an essential step in a proper translation process.

[results, not excuses]

The next issue is how long this QA should take. In their discussion, Paul and Karen mentioned three percentages, all relevant. If the translator of the specific documents reviews the document, it should take around 20% of the translation time. An external review of a properly proofed translation should take around 30-40% while for poorly written texts, revision can reach 50% or even, in the worst cases, justify a retranslation. These numbers are relevant to both translators and agencies. For translators, rates and deadlines need to reflect this time. An overly tight deadline may not allow proper proofreading, which results in a poor translation. Regardless of whether the translator informs customers of such a risk, they are often unhappy with the result, resulting in a lost client. It is far better business practice to refuse impossible deadlines. In freelancer relations with agencies, many project managers are not very realistic on how long it takes to properly review translations, especially long ones that require several sittings. Often but not always education and insistence develop their awareness of the required QA time. Again, it is far better to deliver early than to make excuses for late delivery or fail to properly review the text. In summary, linguists must insist on sufficient time for proofing but the amount is variable.

[3-face customer satisfaction]

The final matter discussed was the essence of the revision process itself. Both Karen and Paul made the point that the task of reviewers was to identify errors, not to produce their ideal text. To clarify, as no two people write the same way, no two people produce the exact same translation of the identical text. One translation may be partially or completely superior to the other in some criteria but both express the content and serve the purpose. Therefore, the appropriate role of reviewers is not to impose their phrasing or linguistic tendencies but to identify and correct issues that would interfere with the message of the text. Admittedly, there is a fine line between mandatory and preferential changes but editors need to make evert effort to be on the right side of that line. Not only do unnecessary changes poison the atmosphere between translators and reviewers, they waste everybody’s time and money. Overzealous editors often spend hours unnecessarily rewriting acceptable translations and then cause the anguished translators to spend hours justifying their original choices. Time is money. The host mentioned that one cause of this extraneous effort was editors’ need to justify their existence. As a frequent reviewer of translation, I have never had any negative feedback in writing that the translation is fine and making at most a few minor changes. I can also attest how difficult it is to switch hats from translator to editor. However, the effort of distinguishing the tasks is much less than the effort required by the all parties in the case of unnecessary rewriting. Reviewing is a fundamentally different task from translation.

Overall, I enjoyed the podcast and hope to find time to view more in the future. Both new and experienced translators and reviewers can profit from a review of the fundamentals and discussion of the specifics. The podcast reinforced the notions that translation review is a necessary, time-consuming and distinct task. Any translator can profit from listening to it in full and reconsidering their attitude and approach to proofing and editing.

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

Picture credits: Pixaby

Comma Use

Comma Use

As a professional editor, many if not most of my changes involve removing or adding commas.  Since punctuation use is different for each language, I’ll try to provide a rule of thumb for Engish.

Without getting too technical, I can give you two approaches to sensing whether a comma isnecessary or not, one for the intuitive English learners and one for those who like short, simple rules.

A comma is the written equivalent of a pause in speech.  We pause for good reasons, not randomly.  We might want to add interesting, but not necessary information, as in “Sally’s cousin … the one that just took the bar exam … is coming to visit.”  We stop to distinguish an incomplete or linked idea from the main idea:  I’ll let you get on the bus … even if you don’t have the fare. A pause is used to communicate the direct connection between a name and title: Mr. Jones … the current CEO … has just received a bonus.  By contrast, we don’t pause between the important parts of the sentence, unless we are trying to build suspense.  My … new… job … pays… 10,000 dollars … a week sounds like someone has a speech impediment or has run up 12 flights of stairs.  We also don’t pause if the description is required to understand the sentence.  My neighbor with the AK17 in the living room is angry at me tells me that the rest of the neighborhood is not so bad.  I do have other neighbors, still. What does this mean for comma use?  If you would naturally stop to breathe, place a comma.  If not, avoid a comma.

For those who don’t trust their ears and like hard and fast rules.  Don’t place a comma between the subject and verb and the verb and objects.  If the information is subordinate or extra, insert a comma.

By the way, American and some British usage places a comma before the and in a list, i.e. cats, dogs, and parrots.  However, it is not a mistake to omit it.