How should large language models be regulated?
A 134-page report from the University of Michigan’s Gerald R. Ford School of Public Policy warns researchers and policymakers alike against the potential dangers of…
→ Continue ReadingINTERVIEW BY CAMERON RASMUSSON
I
n the modern world, peace of mind means everything, and nothing gives language service buyers peace of mind quite like knowing their language partners are among the most qualified in their field. That’s where professional standards come in. It’s the simplest shorthand for demonstrating the mastery of complicated skills, best practices, and concepts, which raises the questions: How are these standards developed? Who decides what is important and what isn’t? What is the process by which standards are developed and approved?
Well, who better to answer those questions than two individuals who have done the work themselves? Dr. Christopher Kurz and Ilona Wallberg contributed to multiple sets of standards for the International Organization for Standardization (ISO). Their latest effort is a new forthcoming set of standards: ISO 5060.
Advertisement
Having worked in the translation industry since 2000, Kurz holds a degree in translation and a Ph.D. in quality management. His former roles include technical lead translator at SDL (now RWS) and translation project manager for Ferrari, BMW, and Deutsche Bahn. Since 2016, he has been heading the Translation Management department at ENERCON, a leading manufacturer of wind turbines. He is the project leader of ISO 5060 and has been a member of the ISO group of experts in TC37/SC5/WG1 Translation since 2011. He has also been involved in developing ISO 17100, ISO 18587, and ISO 11669.
Ilona Wallberg is managing director of beo Gesellschaft für Sprachen und Technologie mbH and is responsible for quality management and human resources. Prior to that, she worked for Siemens for more than 22 years. At the German standards organization, Deutsches Institut für Normung (DIN), she heads the German mirror committee for translation, interpretation, and related technology. Since February 2023, she has chaired the advisory board of DIN’s Terminology Committee. At ISO TC37, she represents German interests in all sub-committees as head of delegation.
The new and upcoming ISO 5060 – Evaluation of Translation Output aims to help translation buyers and translation service providers (TSPs). The goal is to provide a methodological framework for evaluating translation output professionally and objectively. Quite often, we witness that translations are judged by personal preferences and in a highly subjective way. The new standard wants to level the playing field with regard to an objective, unbiased, fair, and analytic evaluation in order to get a clear and reliable picture of the translation quality.
One of the core values of ISO 5060 is that translation output can only be judged in relation to the translation requirements defined in the project specifications. This links to ISO 17100 and the revamped and upcoming ISO 11669 – Translation Projects General Guidance.
But regarding your question, ISO 5060 is intended for the entire language and translation industry — for the output of human translation, machine translation (MT) and post-editing, MT-only projects, and academia.
In recent years, we witnessed that translations were evaluated according to SAE J2450 or the LISA QA model. Each of these models is useful and has its purpose and reasoning. However, the common feedback from the translation industry was the lack of and the demand for a universal translation quality evaluation standard that is neither overly complex nor so much stripped to the bone that it provides no methodological foundation and features merely a basic scorecard.
The feedback from the industry convinced us that such a standard was missing and that it was something the professional translation industry was longing for. And that’s the reason we started the project in early 2020.
Advertisement
The standard is now in the “final draft international standard” (FDIS) stage. The text has been finalized and sent to the ISO committee experts for review. It is the last stage in which technical changes to the text can be made to a draft of a standard. After passing the FDIS stage and the corresponding ballot, it will take about two to three months until the standard will be published. So we are talking about mid-2024.
The ISO committee TC37/SC5/WG1 Translation developed the standard. At the moment, the membership features more than 30 countries from all over the world and approximately half a dozen liaison members from the translation industry. The experts’ group comprises clients, TSPs, members of academia, supra-national institutions, and non-governmental organizations (NGOs). So, we can boldly claim that we listened to the ideas and received input from all stakeholders from the industry and that we took care of their needs.
The most important guideline for developing an ISO standard is the ISO Directives – Part 2. It’s an 80-page compendium that sets forth all rules, procedures, and stages for developing an ISO standard. The time for developing an ISO standard is 36 months.
In short, a project team comprising a project leader and an editor approaches the relevant ISO experts’ group with a New Work Item Proposal, which is a document outlining the basic ideas for a new standard. It should feature a clear scope, so that everybody involved gets an idea of what the future standard might be about and to ensure everybody is on the same page.
If the proposal is accepted, the draft is honed and fine-tuned in different stages in which all countries involved in this ISO committee can comment on the draft and propose their ideas and changes. This process is usually done during regular virtual meetings with the experts that each country nominates. A ballot takes place at certain stages, and the standard requires a certain approval rate for it to continue to the next stage. Together with the vote, member countries are asked to hand in comments and suggestions to improve the text. These comments are then discussed with the group of experts until the text reflects the ideas and reasoning of all experts actively involved in the process.
So, it’s a very democratic and open procedure in which all experts can voice their opinions, ideas, or doubts. On the other hand, this requires diplomatic and sensitive handling of the discussions to keep everybody on board and get the full buy-in from all the experts. This is also why an ISO standard always reflects the experts’ consensus. Consensus, however, does not require unanimity.
In 2012 and 2019, two prior projects with a similar scope were cancelled. These two abandoned projects did indeed influence the development of ISO 5060. Instead of going for the absolute maximum, we settled — with the consent of the ISO experts — for an approach of minimum complexity. We decided to opt for this route to make the standard as broadly applicable as possible and not limit its window of applicability.
Other standards influencing this are ISO 17100 and the extremely interesting, soon-to-be-published ISO 11669. ISO 17100 was most influential for ISO 5060, as it focuses on the definition of and adherence to translation requirements during the production of a translation. This provided the yardstick for ISO 5060: If a requirement is stated in the translation brief or the translation specifications and is not complied with, this non-conformity has to be regarded as an error. If, on the other hand, something desired by a translation requester or a client is not stated in the translation specifications, its absence cannot be regarded as an error.
So, it’s basically all about translation specifications. Each error is counted and assigned to an error category. Depending on its severity, different penalty points are assigned to an instance of an error, resulting in an error score and a subsequent quality rating for the evaluated translation.
Firstly, let me say that ISO 5060 is not a requirements standard, but a guidance standard. It is, therefore, explicitly not intended for certification. But you can use it for your daily work and to help you set out a reliable framework. And, of course, you can state that you work according to this standard.
The users of the standard can vary from large TSPs to smaller TSPs, clients, institutions, freelancers, and academia.
Advertisement
Artificial intelligence (AI) and large language models (LLMs) will be a new field of standardization. There are so many things that should and could be standardized in that field that the development of standards will play an important part in the future.
Regarding the current standards, the experts’ group will start on the overhaul of ISO 18587 – Post Editing of Machine Translation. We recognized that this standard is outdated and that the use of MT and the application of post-editing has drastically changed since its publication in 2017. Therefore, the experts’ group will work on this standard to mirror the current state of technology.
There will also be a new standard for the development of source output that is intended for translation. Here, international experts will guide the users through the jungle of phrases and signs that might cause misunderstandings or additional costs during translation or localization. “ISO 18968 – Translation-Oriented Writing” is the working title, and it is due for publication in the summer of 2025.
In the wind energy sector and numerous other sectors, the application of standards is the norm. In some sectors, the application of standards is even binding. We would be happy if the application of standards were to become more frequent and more accepted in our industry, too.
We also hope our industry’s sometimes negative view of standards will change over the years, as we believe the negativity is undeserved. Standards are a means to provide guidance and sometimes set requirements in order to achieve a common ground and a common understanding. But there is nothing compulsory about ISO standards in our industry. It’s up to each TSP whether to work according to a standard.
As a project team, we hope that ISO 5060 will be accepted and welcomed with open arms by the translation and language industry. We have all put endless hours of work into the standard, and we are excited that it now has the look and feel of a good standard. We are very grateful for every kind of feedback from the industry and hope that ISO 5060 will be well received.
Cameron Rasmusson is a senior writer and editor for MultiLingual Media.
Related Articles
A 134-page report from the University of Michigan’s Gerald R. Ford School of Public Policy warns researchers and policymakers alike against the potential dangers of…
→ Continue ReadingULG Translation Business Unit demonstrates commitment to quality with a leading level of global certifications. ULG continues to lead the language services industry with multiple global…
→ Continue ReadingXLIFF Version 2.1 has reached an important milestone in its development. On February 7, the OASIS XLIFF TC members approved the second Committee Specification Draft and…
→ Continue Reading