In-house translation – outside the box

Tag: HITL

  • 7 thoughts on how the Human in the Loop approach trivialises the translation profession

    7 thoughts on how the Human in the Loop approach trivialises the translation profession

    The Human in the Loop approach trivialises the translation profession in many ways. Here are just seven of them, based on some recent discussions on LinkedIn in various threads of comments on posts.

    1. It fails to set out the qualification/expertise required to be the human in the loop responsible for ensuring the quality of the output. The human may not necessarily possess the genuine level of language skills in the target language to assess the quality of MT output. In particular, non-translators should not be allocated the task of doing the heavy lifting for e.g. appraising the quality of an MT solution.
    2. Advocating a “good enough” approach that is not “fit for purpose” nurtures the narrative that anyone with a better than rudimentary grounding in two languages is a translator. Translators frequently given the inner eye-roll upon hearing qualification of source/target language ability via statements like, “My brother’s son grew up in the States” or “I spent 18 months in Brussels”. In both cases there is an exposure to the language, but there is no guarantee that they have any formal training or education as a translator.
    3. By relegating the human in the loop to a role that is often no better than lightly edited MTPE for derisory rates, it means practitioners may only scrape a minimum hourly wage. Translators on minimum wage will as a rule not be as committed or engaged as if they were being remunerated commensurate to their skill. Similarly, this approach is also a classic example of how translation has been uberised to the extent of reducing translation to a commoditised service.
    4. The real cost of GenAI translation is questionable: the amount of energy consumed on LLM training and to use the LLM to “translate” is not apparent – skewing the real cost-saving. If the output is of such poor quality that it has to be (re)translated from scratch, the cost is high than using human translation in the first place. Where models use multiple alternatives for single sentences e.g. where there are QA steps also performed with an LLM, the number of prompts used may be vastly more than one per sentence. The metric of comparing the among used for a search machine request cf. a prompt is no as reliable as it was, as many search providers now offer AI summaries automatically, driving up the energy consumption of a search request.
    5. Linguistic colonialism and extensive use of pivot languages unrealistically trivialises LOTE (Languages Other Than English) (h/t Sarah Swift). Frequently models are trained between the source language and English, but less frequently directly between to LOTEs. Translation using English as a pivot language may overlook similarities/proximities, or even the abundant supply of translators in a language pair between geographically neighbouring languages (e.g. consider translation directly from Slovak to Hungarian, as opposed to an MT route using a pivot language, where a Slavic language to a Finno-Ugric language is performed via a Romance language pivot). Pivoting via English introduces unnecessary lexical gaps.
    6. MT’s “confidently wrong” approach is at distinct odds with a professional translator’s “cautiously right” approach. The latter will use context and question wordings – and refer the source text back to the author, whereas the former will always offer a translation. If you want to test this, take a provision from a law with nested clauses and remove parts of the verb phrases. The MT tool will nevertheless always offer a translation (based on probability) without context, whereas a human can at best guess from the surrounding context, or state that there is a problem with the source text.
    7. There is even a risk that MT might create a false veneer that translation is a job that “anyone can do” – thereby reinforcing the false impression that translation is little more than a mechanical task of replacing words, rather than a expertise-based service delivered by highly specialised trained professionals.
  • Who’s in/on the lead as we head into 2024?

    Who’s in/on the lead as we head into 2024?

    The debate about the future of (human) translation and changing role of translators is the biggest topic in translator circles. 2023 has been the year of the (unstoppable?) march of machine translation. Within a year of bursting onto the scene as an unknown, OpenAI’s chatbot, ChatGPT, can apparently also translate. Human translators increasingly face tighter, more competitive markets. Many are not even consulted about their replacement by MT solutions, but maybe grudgingly offered PEMT work. And there are talks of tightened budgets and gloomy outlooks of recession. So are the days of out-and-out translators numbered?

    The Chartered Institute of Linguists, which I recently joined, has released a white paper: CIOL Voices on AI and Translation. It addresses some initial reflection and major concerns. The White Paper points to a shift in professions: today’s professional translators will be the future’s language experts and consultants. Sometime new job titles are dismissed as a case of “old grapes in new bottles”?

    The introduction to the White Paper concludes:

    […} we can ensure that linguists remain at the forefront of AI integration in our field – the essential expert ‘humans in the loop’.

    Steve Doswell, Linguist, consultant and Chair of CIOL Council in CIOL Voices on AI and Translation

    The use of “expert ‘humans in the loop’” is telling here. Without attaching the “expert”, it implies that an involved human may not be an linguistic expert. This ties in with concerns about the need for human judgement in using MT and LLMs for translation. It remains essential that users clearly understand their responsibility, as well as the pitfalls of using unsupervised MT. In-house language units must have an active role in training and onboarding users. Their involvement in the decision-making regarding the adoption of such approaches remains essential. It is not an out-and-out IT decision – even if the technological nature of the solution, means IT must be on board. There is some very sensitive messaging in moving from a “human translation” approach to “human in the loop” if bypassing the intermediate “machine in the loop” stage.

    Potential for upskilling and job crafting

    This presents possibilities for upskilling and job crafting – both useful tools for in-house staff retention. New remits might help retain senior staff members wishing to have a change from day-in-day-out translation. Any in-house solution will need dedicated language technologists. Language technologists are the new translators in terms of language services recruitments. Central banks and financial market supervision authorities have been hiring people with this profile for several years.

    It is also important to remember that for any solution to work to its full potential, will need dedicated staff. The quick and dirty approach might be to outsource, but such solutions, although quicker to implement, may not allow the desired level of control. An attractive interface is one thing, but there might not be the possibility to tweak the temperature of the underlying model, or to train it to your specifications – which is beneficial to extract the maximum benefit for your use case. However, this training isn’t possible on the fly – it needs a long-term training concept and commitment. And naturally potential succession management issues need handling too. These issues may be due to sabbaticals, secondments, retirements and maternity leave. Entrusting an entire solution on a single set of shoulders is also an operational risk.

    In this case, human involvement is still in more of an expert capacity – training and refining the engine, and ironing out the wrinkles. (Rinse and repeat as required!) Other tasks include managing new versions of software and interfaces, or plugins to CAT environments and maintenance. With an outsourced solution the situation is not so clear cut. This brings us back to the issue of the position of the human expert in the loop – and whether human or machine is subordinate – in the translation process as a whole, and the problems with the terms used.

    Driven loopy – the expert/machine/human in the loop/lead.

    I first heard of “human in the loop” mentioned at the 2021 edition of the Translating Europe Forum (TEF). TEF is the European Commission’s annual translation *industry* event. Over the last two years, I have lost count of the amount of discussions I have had with other people, about it. The problem it throws up lies in the interpretation of the role of the human.

    Moving further back, human-in-the-loop in 2012 was a classification for autonomous weapons systems. In that context, a human must instigate the action of the weapon. Human-on-the-loop is a classification whereby a human may abort an action. Lastly, and most terrifyingly, human-out-of-the-loop is the classification for no human action is involved. In this case human-in-the-loop does not imply that the human is subordinate to the machine.

    An intermediate stage exists between human translation and human in the loop: “machine in the loop“. In that case the machine is subordinate to the human, or more likely an expert. Both “machine in the loop” and “human in the loop” are weasly terms. Both fail to mention the role of human expertise – which is why some prefer “human at the core” or “human in the lead“. Additionally, one experienced colleague recently pointed out on LinkedIn that anything “human” omits to say anything about their expertise. This is why I actively try to opt for “expert in the lead” (should that maybe be EITL or XITL?).

    There can be a lot of difficulties in explaining the delicacy of the situation to lay colleagues – they see a binary situation: human translation or machine translation.

    After all, If you are not in the lead, but only in the loop, then you are effectively “on the lead”. And naturally there is the issue of the subsequent drift from human in the loop to human on/out of the loop. In that situation, we’re in the territory of fully autonomous self-driving vehicles.

    Resistance is futile?

    AI technology is clearly here to stay. While there is a certain hype cycle, it is not just a passing fad. The truth is that its limitations are well recognised: AI/MT cannot be used unsupervised in many settings. There are possibilities that the enhanced use of technological assistance might also open up new seams for translation (MT is a suitable use case for e.g., translating Airbnb and travel site reviews where a gist translation is what is needed). Humans will remain an integral part for training the underlying systems. Otherwise, at some point there will only be synthetic data to train systems that require high quality human data for training. Increased efficiency needs to be offset against the lack of job satisfaction that some will experience from being relegated to post-editing.

    Resistance to the advancing AI/MT tide is futile – both in-house and as freelancers. The battle to fight is in educating and countering assumptions that the lay public holds of machines being better, faster and cheaper. People need to understand the real risks and costs. However, part of this battle will also be to ensure that the current cohort of translators/language consultants/language technologists in the making learn the skills they will need for the career of the future. Many university courses adapt to the changing times at a pace observed in glacial creep. This is where professional associations come in – both in upskilling existing linguists, but also in supporting the next generation as it begins its journey.