Drawing from the experiments conducted for the AIDCPT (AI, Decoloniality and Creative Poetry Translation) project, this paper focuses on the impact of text generative AI and LLMs in studying low resource languages within a decolonial context and understands the impact that intervention of text generative AI has had on different contexts of language use, with special focus on translation, and knowledge production of low resource languages. It is apparent that the intervention of AI has produced new ways of using linguistic skills for oral language-cultures that do not have a significant presence in the lettered world. My experiments so far have captured the manner in which careful prompt engineering and ongoing dialogue with the machine help in working with low-resource languages and complex situations of language difference and contestation. This paper delineates the strategies adopted to translate poems emerging from the Miya poetry movement in contemporary Assam, educate the LLM in context, questioning its assumptions about language, and uploading materials such as an alphabet script, audio-visual tools to make it learn the importance of the latent heterogeneity within plurilingual language-worlds.With the intervention of AI, the human experience of translation, which shapes and is further shaped in the process of establishing a relationship with an other, and the modalities of language experience becomes complex. Acts of ‘doing’ language, through writing, reading, talking, listening are intervened by the arching presence of AI that can participate in acts which were earlier contingent upon human experience. This paper aims at mediating into the networks of AI as sites of learning and knowledge production, and that of cultural exchange which is facilitated in the shared socialites of language use in everyday speech as well as creative writing. With AI and LLMs intervening into this site, questions regarding the production, acquisition and dissemination of knowledge become inevitable. While the contribution of AI and LLMs in research and academic practices is undeniably important, this paper intends to rethink the manner in which these models acquire existing knowledge and generate responses, thus engaging with the technicalities of prompt engineering and AI training along with concerns of ethics and representation.

This paper will begin by introducing the AI-Decoloniality and Creative Poetry Translation project (AIDCPT) based in the Oxford Comparative Criticism and Translation Research Centre at Oxford University (https://occt.web.ox.ac.uk/ai-decoloniality-and-creative-poetry-translation). Large Language Models construct language variety in a different way from older tools: instead of an array of separate standard languages, they represent language as a something more like a continuum of difference. This new ontology of language difference is manifest in the well-known ability of LLMs to imitate a range of styles; it also enables them to participate in translanguaging and other kinds of non-standardised linguistic interaction. The AIDCPT project explores how this capacity can support creative, decolonial translation practices. The paper will then present one such practice: the imposition of constraints, via prompt engineering, which can force an LLM to produce more creative and linguistically varied translations than it otherwise would. I will end by considering what is meant by ‘creativity’ in this case.

Generative AI tools have amplified the potential for proliferation inherent in the translation process, creating both challenges and opportunities in the field of literary translation. To explore some of these issues, I will focus on the translation into Spanish of the poetry collection Piano in the Dark (2023), by the American writer Nancy Naomi Carlson, winner of the Oxford-Weidenfeld Prize 2022. Through creative interaction practices with the Large Language Models (LLMs) ChatGPT, Gemini and Claude, I will analyze the translation capabilities of these tools based on the corpus of poetic texts. These AI tools offer translation options and insights into the poems that can be useful for translators. However, their output sometimes reveals biases and stereotypes. My aim is to identify effective strategies that may guide these models to contribute to the literary translation process and study how the opportunities these tools offer may be maximized, while addressing the ethical and ideological considerations tied to their use.

In the 2003 sf novella by Liu Cixin 刘慈欣, “Poetry Cloud” 诗云, an advanced AI system, in the shape of a clone of Tang-dynasty poet Li Bai 李白, generates poetry. Yiyi 伊依, a literature teacher, challenges the clone Li Bai to generate poems of the same value as the original Li Bai. Naturally, their interpretation of what poetry is differs greatly: Yiyi thinks that poetry is the output of individuals and their human experience within their environment. The clone Li Bai believes that technology, and its ability to store and connect information suffice to create and surpass Li Bai’s original poetic compositions.

Since the publication of “Poetry Cloud”, the binary opposition between AI and human creativity continues to be a main issue of contention, even though much of our understanding of how the world connects and creates its cultural products has been relying on data gathering combined with computational analysis.

In this paper I aim to investigate some conceptual implications and practical possibilities of human-LLM poetry translation. The idea is to make poetry translation a mode of inquiry that draws more visibly and more widely on knowledges and practices across different linguistic, cultural, and literary domains. My specific interest is to experiment with posthuman multilingualism that highlights algorithmic, synaesthetic aspects of the relationship between words and sound, and creates synergies that are culturally transgressive or have boundary-shifting effects.

The resulting multiverse should consist of an extremely fluid, constantly deconstructing and reconstructing system, where distinct languages rub against each other, re-assorted, interrelated.