This paper reflects on the current transformations of information retrieval (IR) systems, recommendation systems (RS), and automated literature review tools. Despite their potential impact on innovation and discoverability in science, the role of these systems remains largely invisible. The integration of AI systems into all phases of IR suggests that new, discrete research practices —such as those described by Clavert and Muller —are becoming entrenched without full awareness of their impact on knowledge production. In this paper, I will update the stakes of scholarly literature discoverability through a state-of-the-art review of AI research assistants, assessing their influence on the scientific ecosystem and emerging discrete research practices. I will also propose avenues for reflection, design suggestions, and architectural frameworks for recommendation and information retrieval systems tailored to address these major challenges.

The reliability of AI Large-Language Models used in academic research is a growing topic in the recent AI conversational landscape. A 2024 study by Liao et al. reported that 81% of survey respondents had already incorporated LLMs into their research workflow. Two common problems cited with LLM-based research are "hallucinations" (plausible but nonetheless fictional results) and so-called "temporal shifts" that can occur when LLMs are trained on one corpus from a particular period of time but are deployed on texts from a different time period (Ushio; Rosin). These temporal problems -- shifting word definitions and different discursive language practices -- can thwart the LLM's ability to generate useful results. This paper is an attempt to understand and rectify those two problems by building a local (that is, not cloud-based) retrieval-augmented generation (RAG) system using the open source model, Ollama.

Our test-case corpus is a selection of historical brewing texts. We have chosen this corpus for two reasons: 1) very few of these texts are available online as either plain text or XML files and so are not easily accessible for off-the-shelf LLM training; and 2) this well-bounded corpus highlights both hallucinations and temporal shifts quite dramatically. The discourses surrounding the brewing industry have shifted considerably over the XXX years of our corpus, making these texts a robust and well-bounded arena for experimentation.

The language variations in this corpus are numerous and vexing enough for humans -- from the oft-repeated historical difference between "beer" (with hops) and "ale" (without) to a more recent distinction between "ale" (using top-fermenting yeast) and "lager" (using bottom-fermenting yeast). Or the confusing use of the terms "porter" and "stout" that were once merely adjectives but have evolved into distinctively different beer styles. Moreover, across the centuries, the backdrop of science has also added and subtracted terminology: from 17th-century Newtonian physics (the "break") to 18th-century chemistry (which saw fermentation as an acid-base reaction) to 19th-century biology (when yeast was first identified as an organism that digested sugar and produced alcohol as a by-product). The description of the brewing process is always framed by a time-shifting understanding of the science behind the art.

A significant limitation for this project has been acquiring reliable plaintext versions of historical texts. Our corpus is still small, but is growing quickly and we hope very soon to have a much larger, and more robust, corpus to serve as the repository for our local LLM.

Translation Studies (TS) appears to occupy a peripheral position in the Humanities and Social Sciences, as evidenced by the fact that the CFP focuses on the theorization of translation without explicitly considering the discipline that squarely focuses on translation – and its corollaries such as the untranslatable. Indeed, the CFP doesn’t account for the many voices from contemporary TS that have explored the untranslatable and the relationships between translation and (physical/digital) materiality (Littau 2016), translation and embodiment (Ivancic and Zepter 2022), translation and cultural specificity (Al-Ghadeer et al. 2025), translation and online/digital media (Desjardins 2017, 2019, 2020), translation and the Internet of Things (Desjardins 2025), and translation in the era of artificial intelligence (AI). In fact, in the Routledge Handbook of Translation Technology and Society (Baumgarten and Tieber 2025), a chapter is dedicated to TS and the Digital Humanities (Tanasescu 2025). Therefore, to contextualize our presentation, we overview some key contemporary debates located at the TS/DH nexus, including the concept of the untranslatable/translatable, both metaphorically and literally. How might we conceptualize what is translatable/untranslatable in an era marked by surveillance capitalism, algorithmic curation, and non-human communication/translation? This leads into the presentation’s second part and focus, which is multilingual, translational, and algorithmic trends within the online creator economy, with specific attention given to Canadian creators and influencers. The content created for social platforms is key to DH, including as the source text for cultural analysis and analytics. Questionnaire data from a SSHRC-funded project on translation and multilinguality within the Canadian creator economy will be presented, elucidating multilingual practices among Canadian creators: do they use AI to translate their content? Do they self-translate? What do they find resists translation or facilitates translation and what novel strategies might they employ? Further, given that online engagement and monetization are often premised on ‘posting for the algorithm’, what algorithmic strategies are employed? Building on McCulloch’s work in Because Internet (2019), linguist and influencer Adam Aleksic’s book states: “[…] the internet has ushered in an unprecedented linguistic upheaval. We’re entering an entirely new era of etymology heralded by the invisible forces driving social media algorithms”. If, as Aleksic’s work suggests, “communication is changing in both familiar and unexpected ways,” from the use of emojis to the way different generations talk about taboo subjects, how do translation and multilinguality operate in these (arguably) transnational settings and what lessons might we draw from these multimodal and multilingual encounters? With a focus on systemic inequities, how does translation—linguistic, cultural, and algorithmic—shape how new content is created in digital platforms? Finally, we hope to dispel the idea that English is the turnkey language for engagement on the social internet, which runs parallel to ideas presented in the preface of the Dictionary of Untranslatables (Cassin et al. 2014) (English version of le Dictionnaire des intraduisibles), including “exploring the places where languages touch” and revealing “the limits of discrete national languages and traditions”. After all, as tech journalist Taylor Lorenz (2025) reports: “the global [multilingual] influencer era is upon us”.