AI Literacy: The concept of suitability and core translation skills

Chapter 3 AI Literacy: The concept of suitability and core translation skills Ramon Inglada Heriot-Watt University, United Kingdom The release of ChatGPT in November 2022 reignited the debate on the future of the translation profession and, consequently, on how translator training programmes should change and adapt. Similar to discussions that occurred with the introduction of neural machine translation a few years earlier, some argued that core translation skills, such as language knowledge, translation ability and cultural expertise that have long been essential components of many translator training programmes, had now been rendered obsolete by the arrival of generative AI (GenAI). Inspired by the concept of machine translation literacy, a case will be made that these core skills (together with some other complementary skills, such as selection and assessment) are absolutely essential in order to ascertain whether the content produced by GenAI tools is not simply accurate or inaccurate, but, more importantly, suitable for the requirements of any given translation project, as specified in the relevant translation brief. Examples will be given of specific AI-based tasks (such as translation of content, terminology extraction, multilingual glossary creation and machine translation postediting) in which the suitability of the results cannot be determined without recourse to so-called traditional core translation skills. 1 Machine translation literacy and artificial intelligence literacy The concept of machine translation (MT) literacy, introduced by Bowker & Buitrago Ciro (2019), has attracted a lot of interest, both in academia and in Ramon Inglada. 2026. AI Literacy: The concept of suitability and core translation skills. In JC Penet, Joss Moorkens & Masaru Yamada (eds.), Teaching translation in the age of generative AI: New paradigm, new learning?, 49–64. Berlin: Language Science Press. DOI: 10.5281/zenodo.17641068 Ramon Inglada the language industry, as well as in other sectors. Furthermore, Bowker’s MT Literacy project1aims to educate users on the dos and don’ts of using MT output and its infographics have been very popular on social media. It also highlights the importance of confidentiality and privacy when using programs such as DeepL and Google Translate, with an emphasis on different use cases of free online MT systems. The project’s focus on enhancing digital literacy skills and the responsible, ethical, and sustainable use of MT systems can be used as the basis to further develop and adapt the concept of Artificial Intelligence (AI) literacy. Long & Magerko (2020: 2) defined AI literacy as “a set of competencies that enables individuals to critically evaluate AI technologies; communicate and collaborate effectively with AI; and use AI as a tool online, at home, and in the workplace”. In this chapter, this definition will be used as the basis for a different definition specifically created with the translator training context in mind. This approach aligns with (Krüger 2024) framework for AI literacy in translation, which emphasises the need for translators to develop competencies in understanding and effectively using AI technologies in their work. It is undeniable that the release of ChatGPT in November 2022 generated a lot of hype. It also reignited the debate on the future of the translation profession and, consequently, on how translator training programmes should change in order to adapt to this latest technological advancement. Similar to discussions that occurred with the introduction of neural machine translation (NMT) systems a few years earlier, some argued that core translation skills that have long been essential components of translator training programmes had now been rendered obsolete by the arrival of GenAI. However, it could be argued that this is an oversimplistic approach. This chapter advocates for technological advancement, as it can bring many benefits to society and offer solutions to some of the greatest challenges humanity currently faces. In the field of translation, when used sensibly as part of welldesigned workflows, computer-assisted translation (CAT) tools and MT can be a great asset for professional translators. The same applies to GenAI chatbots based on large language models (LLMs), such as OpenAI’s ChatGPT, Microsoft Copilot and Google Gemini. However, given how recent this technology is and the pace at which it evolves, it could be argued that the translation industry might struggle to agree on what a ‘sensible’ and ‘well-designed’ AI-based workflow constitutes in the field of professional translation. In any case, it is always worth keeping in mind that no technology-based solution is infallible. This obviously applies to ChatGPT (and to the other LLMs) and it could be said that, to 1https://sites.google.com/view/machinetranslationliteracy/ 50 3 AI Literacy: The concept of suitability and core translation skills a certain extent, even ChatGPT itself reminds us to be vigilant and use our own judgment and common sense. At the time of writing, this is the message that appears underneath ChatGPT’s chat interface: ChatGPT can make mistakes. Consider checking important information. 2 Core translation skills The purpose of this chapter is not to provide an exhaustive list of core translation skills, describing what they are and why they can be considered as key skills all translators should have, as many others have already done this very effectively in the past. Instead, it will focus on three very general skills which, arguably, most translation scholars would generally agree are essential for all professional translators (and which, therefore, should feature prominently in all translator training programmes). This is also done for the sake of simplicity as, ultimately, these skills are mere examples. This selection of core skills has been based on the reference standards for translator training set out in the European Master’s in Translation (EMT) Competence Framework (2022) and on the translation competence model created by PACTE (2003). Different translator training programmes, institutions and settings could naturally select other skills that they would consider as essential. These chosen three core skills are: • Linguistic knowledge • Translation ability • Cultural expertise Core translation skills such as the ones mentioned above should not be allowed to disappear from translator training programmes, but rather should now be considered more important than ever, precisely because of the advent of GenAI. This is not to say that translator training curricula should not be adapted: they should, particularly as concepts such as augmented translation and human in the loop are becoming increasingly important, and discussions around translators now becoming ‘language specialists’ or ‘language experts’ are happening with increasing frequency. However, the debate should not be about these core skills becoming obsolete, but about how essential they are when GenAI is used in translation. 51 Ramon Inglada 3 The concept of suitability over correctness When translators or translation students use ChatGPT or other similar tools to assist them in their translation tasks, the focus should not be on whether the AIgenerated output is ‘correct’ (something that could also entail a degree of subjectivity), but rather on whether such output is suitable (or fit-for-purpose) for any given set of translation requirements, as described in the translation brief, for instance in terms of tone, register, target audience, vocabulary and text type. The priority should be placed on the concept of ‘suitability’ of AI-generated content, rather than on its ‘correctness’. This cannot be successfully achieved if the aforementioned core skills (as a minimum, linguistic knowledge, translation ability and cultural expertise) are not present and well honed. Translators would struggle to decide whether AI-generated content is suitable for any given translation project (and its set of unique needs) if they do not know how to translate (and they have not developed their core translation skills). Using GenAI chatbots is easy; using their output critically requires thought. 4 New and complementary skills It could also be argued that not only are core translation skills still essential, but that they should also be complemented with the ‘new’ skills of selection and assessment. These are obviously not new skills — after all, many translators have been already selecting and assessing translation memory matches for decades — but they have now become fundamental when using GenAI in translation. Professional and trainee translators need to be able to select among the alternatives offered by ChatGPT and other similar tools. They also need to be able to critically assess the suitability of the output being presented to them. Therefore, a renewed emphasis should be given to selection and assessment as core skills to be integrated into and developed in translator training programmes. The reasons why ‘traditional’ core language, translation and cultural expertise skills should still be a pivotal part of translator training programmes have already been discussed, in combination with the reasons why they should also be complemented by the not-so-new skills of selection and assessment. However, is there any additional new skill that should be added to this list of core skills for the training of translators in the new era of GenAI? The answer is a resounding yes. This skill is prompting (also known as prompt engineering). Prompt engineering can be defined as the process of designing, crafting, and refining inputs to elicit specific responses from a GenAI model, aiming to optimize interaction outcomes through careful consideration of the prompts (Bozkurt 2024). The model 52 3 AI Literacy: The concept of suitability and core translation skills then generates a response based on the input it receives. Effective prompting is crucial for obtaining relevant, coherent and suitable answers. Translator training programmes should seek to integrate prompting skills into their curricula, so that future translators can ensure a greater degree of suitability in the output generated by GenAI. When adapting existing translator training programmes (or when designing new offerings), an emphasis should be placed on processes aimed at the careful creation of effective prompts (or sequences of prompts) with different roles/personalities, different levels of complexity or even different degrees of creativity. 5 AI Literacy in translation: a simple definition Based on all the principles discussed before, AI Literacy for Translation could be defined as the combined application of a definite set of basic core skills in order to maximise the usefulness and relevance of GenAI output and ensure its suitability in relation to the requirements set out by the translation brief in any given translation task. The basic core skills are linguistic knowledge, translation ability and cultural expertise, complemented by selection and assessment. All these skills should be applied to GenAI output produced as a result of effective prompting. The overarching notion that underpins the application of AI Literacy in translation is that of suitability, rather than ‘correctness’, of the generated output. 6 AI Literacy applied to practical language and translation tasks Five examples will now be provided of translation or translation-related tasks, chosen to represent some of the tasks that professional translators carry out on a regular basis (and which, as such, are also commonly observed within the confines of the translation classroom). The concept of AI Literacy will be applied to these examples. This means that, in all cases, an argument will be presented to emphasise the importance of relying on the three generic core skills of linguistic knowledge, translation ability and cultural expertise (presented earlier in this chapter) in order to select the most appropriate GenAI output and assess the suitability of this output. An example of the prompt (or prompts) used to request the completion of the task in hand by the GenAI chatbot (ChatGPT based on GPT 3.5) will also be provided for each example. 53 Ramon Inglada In the first example, ChatGPT itself was used to request the creation of a sentence related to renewableenergy containing some grammar issues and awkward word groupings. In the other four examples, the three initial paragraphs from the English version of the Wikipedia article on the topic of renewable energy (from July 2024) were used as the reference source text. These five examples are: sentence reformulation, translation of content, terminology extraction, multilingual glossary creation and MT post-editing (MTPE). 6.1 Example 1: Sentence reformulation For our first example, I will start with one of the simplest tasks LLMs can be used for. This task has also been chosen as an introductory example as anecdotal evidence seems to suggest that this represents one of the primary ways in which GenAI chatbots are commonly utilised, especially by non-native speakers of the language they are engaging with. ChatGPT can be a valuable tool for reformulating badly written sentences. By providing a poorly constructed sentence to ChatGPT, users can receive suggested revisions that address spelling mistakes, grammar errors, improve clarity and enhance overall readability. ChatGPT leverages its model of language patterns and grammar rules to generate alternative phrasing that is more coherent and idiomatic. Users can input sentences from various fields, such as renewable energy in our example, to receive tailored suggestions for improvement. With ChatGPT’s assistance, individuals can refine their writing skills, produce clearer communication and convey their messages effectively across diverse contexts. In this example, the following prompt was used to ask ChatGPT to create the sentence I would use as a basis for reformulation (see (1)): (1) Prompt used to elicit a poorly constructed sentence from ChatGPT 3.5. (Created by Ramon Inglada) This was the sentence resulting from the prompt above: (2) ‘The sun is giving us muchly energy, so we should put many solar panels to catch it all’. The following prompt in (3) was then used to request an improved version of the sentence, correcting any grammar issues and changing any unnatural word collocations: 54 3 AI Literacy: The concept of suitability and core translation skills (3) Sentence reformulation prompt in ChatGPT. (Created by Ramon Inglada) This was ChatGPT’s reply to our request: (4) ‘The sun provides us with abundant energy, so we should install numerous solar panels to capture it efficiently’. It could certainly be argued that this version corrects existing grammar issues and uses more idiomatic collocations to convey the intended meaning more clearly. However, it would only be possible to do that if our knowledge of the English language is advanced enough. If I now try to apply the concept of AI Literacy in this first example, it is immediately evident that I would need to rely on one of the aforementioned core skills (in this case, knowledge of the English language) firstly to realise that the original sentence contains some issues, and secondly to assess the suitability of the suggested ‘improved’ version provided by the LLM. Furthermore, in this specific example, ChatGPT decided that three elements in the original sentence needed to be changed. These are ‘muchly energy’, ‘many solar panels’ and ‘catch it all’. If my English language skills were insufficient, I could simply accept all three suggested improvements, assuming (or even hoping) that the resulting sentence is now much more suitable for my needs. However, I could also decide to select only some of these changes, so I would only keep those that I (and not ChatGPT) consider as suitable. I could even accept them all and then go on to further modify them to end up with a final sentence which would be the collaborative result between me and the LLM. This process exemplifies the concept of “centaur tasks” as described by Mollick (2023), where there is a clear division of labour between human and AI, leveraging the strengths of each. Once more, the decision on whether this final sentence is the most suitable option for our needs is something that can only be achieved if our core skills have been developed enough. 6.2 Example 2: Translation of content This second example is intended to cover the use of LLMs as MT tools. While LLMs were not in principle designed to be used as MT providers, this is certainly one of the many tasks these tools can perform. It is also probably one of the first uses that comes to mind when one thinks about the potential uses of GenAI in 55 Ramon Inglada the field translation. There is, indeed, ample anecdotal evidence of the use of LLMs as MT tools. (5) below is an example of a prompt that could be used to ask ChatGPT to translate a piece of text. Yamada (2023) explored how incorporating the translation’s purpose and the target audience into prompts affects the quality of translations generated by ChatGPT. Therefore, in this specific example, the chatbot will be given a concrete personality, the language combination to be used will be English into French, the content to be translated is related to renewable energy, and the translation is to be published in a government report. (5) Content translation prompt in ChatGPT. (Created by Ramon Inglada) If I were to go ahead and enter this prompt in ChatGPT followed by a text for translation, the LLM would promptly generate a translated version of the content in question (in French in this case). As mentioned earlier, the three initial paragraphs from the English version of the Wikipedia article on the topic of renewable energy have been used as the source text. At this point, I would have two options. Option 1 would simply entail accepting the generated translated content at face value, without conducting any checks and without verifying the suitability of the said content (depending on a series of factors that could include overall quality, register, tone, intended audience, publication media and so on). Option 2 would involve operationalising the concept of AI Literacy. Using my core skills, I could perform all the checks mentioned beforehand. I could select other potential translation alternatives requested to the LLM for any passages that I might not be satisfied with (for instance, asking ChatGPT to rephrase any given part of the translated text). Finally, I could also assess the overall suitability of the generated output. The relevance and usefulness of both the initial translation-generating prompt and any additional prompts entered in the chatbot (for instance to check for alternative formulations or to change the level of formality in the translated text) would be greatly dependent on the use of effective prompting techniques. 6.3 Example 3: Terminology extraction One of the potential uses of GenAI tools that was quickly identified as potentially highly interesting for translators, interpreters, terminologists, researchers and other professionals was that of a terminology extraction tool. Users can request ChatGPT to create a list of key terms from a given passage of text. Users can ask 56 3 AI Literacy: The concept of suitability and core translation skills for a specific number of terms to be identified, and they can even request the tool to present the results in a table format, to facilitate further work. An example of a prompt that could be used for this purpose (using a text in the field of renewable energy, including all the requirements mentioned above and also a personality) is shown below: (6) Prompt: You are a professional terminologist. Please extract a list of 15 key terms from the following text in the field of renewable energy and present them in table format. Please also include a definition. (7) The first four terms related to renewable energy and their definitions, as provided by ChatGPT (Created by Ramon Inglada) As can be seen in (7), when using this prompt and providing ChatGPT 3.5 with a text to perform the terminology extraction on, the LLM will very quickly produce a convenient table containing the specified number of terms in the field in question and also their definitions. Once again, I could simply accept the automatically created selection of terms and their corresponding definitions as relevant and suitable for our needs (whatever they might be). However, I could also use critical thinking and in this instance, again, apply the concept of AI Literacy. The speed and convenience with which ChatGPT has created this list of terms and corresponding definitions is astounding (and several orders of magnitude faster than performing a similar process manually), but how can any user ensure the suitability of the generated terms and definitions if the core skills are lacking? How can I select the candidate terms that might indeed be useful for our needs in any given task? How can the suitability (and even accuracy) of the definitions provided be assessed? And how can I specify a concrete number of terms to be 57 Ramon Inglada EMT. 2022. European master’s in translation competence framework 2022. Tech. rep. Brussels: European Commission. 1–12. https://commission.europa.eu/ system/files/2022-11/emt_competence_fwk_2022_en.pdf. Krüger, Ralph. 2024. Outline of an artificial intelligence literacy framework for translation, interpreting and specialised communication. Lublin Studies in Modern Languages and Literature 48(3). 11–23. DOI: 10.17951/lsmll.2024.48.3.1123. Long, Duri & Brian Magerko. 2020. What is AI literacy? Competencies and design considerations. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (CHI ’20), 1–16. Honolulu, USA: Association for Computing Machinery. DOI: 10.1145/3313831.3376727. https://doi.org/10.1145/3313831. 3376727. Mollick, Ethan. 2023. On-boarding your AI Intern. (25 November, 2024). https: //www.oneusefulthing.org/p/on-boarding-your-ai-intern. PACTE. 2003. Building a translation competence model. In Fabio Alves (ed.), Triangulating translation: Perspectives in process oriented research (45), 43–66. Amsterdam, Netherlands: John Benjamins. DOI: 10.1075/btl.45.06pac. Waldo, Jim & Soline Boussard. 2024. GPTs and hallucination: Why do large language models hallucinate? Queue 22(4). 19–33. DOI: 10.1145/3688007. Yamada, Masaru. 2023. Optimizing machine translation through prompt engineering: An investigation into ChatGPT’s customizability. In Masaru Yamada & Félix do Carmo (eds.), Proceedings of Machine Translation Summit XIX, Vol. 2: Users Track, 195–204. Macau, China: Asia-Pacific Association for Machine Translation. https://aclanthology.org/2023.mtsummit-users.19. 64