Translating With Feeling: Centering Translator Perspectives within Translation Technologies

Abstract: Rapid development of LLMs and similar automated approaches for translation tasks is increasingly affecting the landscape of translation technologies. As concerns about the outsourcing of translator work to these automated translation tools grow, it becomes increasingly crucial to gather insights from the translation community directly. To this end, we conduct an interview study with 19 professional translators working across 11 languages and 11 domains to understand their perspectives, experiences, and concerns with using translation technologies in their work. We find that translators are cautious when incorporating new tools into their workflow, with several expressing concerns machine translation (MT) and LLMs are infringing on the necessary human aspects and verification steps of translation, worried that these tools have potential for harmful downstream effects due to compromising the human aspect of translation work. These findings demonstrate the need to develop translation technologies that directly serve translators' needs rather than replacing human translation. This can be done by focusing more on the assistive, rather than the automating aspects of these tools.

• The paper establishes that translation is a creative, context-sensitive process where affect and stylistic fidelity outweigh literal correspondence.
• The paper employs semi-structured interviews and thematic coding to capture translators’ nuanced perspectives on automation and labor misalignment.
• The paper demonstrates that translators repurpose AI tools as assistive instruments, advocating for human-centered design to preserve professional agency.

Translator-Centered Perspectives on Translation Technologies

Introduction

"Translating With Feeling: Centering Translator Perspectives within Translation Technologies" (2604.00758) offers a qualitative investigation into the experiences, attitudes, and labor concerns of professional translators operating in a rapidly evolving landscape shaped by CAT tools and LLM-based machine translation systems. Departing from purely technical or evaluation-centric analyses of MT models, the study foregrounds the voices and nuanced practices of translators across 11 domains and languages, articulating granular misalignments between recent translation technologies and the craft of translation.

Methodological Framework

The authors employ semi-structured interviews with 19 certified, professional translators, ensuring domainal and linguistic diversity. Through multi-phase thematic and axial coding, common motifs and tensions in translator-tool interaction are systematically distilled. This methodology enables the capture of both shared concerns and context-specific divergences, revealing how translation expertise, tool adoption, and attitudes toward automation are deeply stratified across career trajectories and translator demographics.

Core Findings

Translation as Situated, Creative Labor

The work accentuates that translation is not a mechanistic, context-agnostic act. Instead, it is positioned as an inherently creative, situated, and interpretive process, requiring deep linguistic and domain competence. Translators privilege communicative intent, register, audience, stylistic fidelity, and cultural/idiomatic nuance over mere literal correspondence. Quality is variably defined across domains: legal and medical translators underscore risk, consistency, and terminological stability, while literary translators foreground affect, tone, and stylistic resonance.

Critically, translators conceptualize their work as affective, foregrounding the deliberate translation of emotion and nuance—"translating feelings, not just words." This is fundamentally misaligned with current MT objectives, which optimize for efficiency, fluency, and superficial adequacy.

Technology as Assistive, Not Replacive

CAT tools, domain-specific MT, and LLMs are widely used as assistive instruments. However, the affordances prioritized by technology developers—speed, batch consistency, terminology control—are frequently decoupled from translators' value metrics. Translators situate these tools downstream of their core analytic work: they serve to increase speed, paraphrase, or offer alternative phrasings, but never substitute for decisive, context-informed mediation by a human expert. LLMs are frequently regarded as "search tools" rather than translation agents, useful for lexical queries or corpus resource extraction particularly in under-resourced language pairs.

Distrust and Epistemic Concerns

A strong theme is professional skepticism, bordering on categorical distrust, toward LLMs and AI-based translation technologies. This is rooted in multiple axis: opacity by design, concern over privacy/confidentiality (especially in legal/medical domains), lack of traceable decision-making, and knowledge that LLM outputs are prone to hallucination and bias, including well-publicized phenomena such as gender bias transfer [stanovsky-etal-2019-evaluating]. Translators are particularly critical of the push toward full automation in translation, viewing it as a dual threat: epistemic marginalization—loss of control over meaning—and direct labor alienation, mirroring job displacement patterns observed in other creative fields [jiang2026professional].

Tool–Labor Misalignment

The paper underscores structural misalignment between translation technologies and human translation as labor. While CAT tools and LLMs optimize for operational metrics, translators operate within complex ecosystems, balancing client requirements, stylistic constraints, ethical and legal boundaries, and personal accountability. Automated translation systems, when mandated top-down by agencies or clients, erode translator autonomy. Freelancers and in-house translators experience varying degrees of leverage; the adoption of AI is correlated with labor precarity, aligning with macroeconomic analyses of automation-induced deskilling and job displacement [Woodruff2023HowKW].

Resistance and Agency

Despite automation discourses, translators actively exercise agency: they creatively integrate, resist, or circumvent translation technologies. LLMs are repurposed for creative augmentation or lexical exploration rather than full-pipeline translation. Experienced professionals carve durable boundaries around the use of AI systems, especially when client or end-user risk is non-trivial. Translators strategically assert their epistemic authority by controlling input, post-editing, and selecting specific workflows that retain human oversight and authorship.

Implications

For Translation Technology Developers

The research signals an urgent need for translation systems that foreground translator agency, transparency, and user control, rather than enforcing automation. Augmentation-centered design, rich explainability, and mechanisms for integrating fine-grained human feedback are deemed critical. Failure to reconcile tool affordances with translators' self-conceptions and labor conditions invites systemic resistance, underuse, or the outright rejection and circumvention of supposed "productivity" tools.

For the AI and HCI Research Community

The study provides robust empirical support for a paradigmatic shift away from automation-centric narratives toward collaborative and human-centered models in NLP and HCI [liebling-etal-2022-opportunities, Constantinides and Quercia 2025]. Incorporating epistemologies of situated knowledge [haraway2013situated] and creative-critical labor [grass2023translation] is essential for the deployment of MT in critical and sensitive domains.

Theoretical Concerns

Automation in translation is not a value-neutral process; it fundamentally reconfigures authority, raises epistemological issues, and risks the devaluation of creative labor. Translators’ distrust is reinforced by concrete risks: hallucinations, misalignments in nuance, and legal/ethical jeopardy. Consequently, theoretical models of socio-technical interaction in AI should account for context-dependent notions of quality, responsibility, and creative agency, moving beyond pipeline optimization.

Future Directions

Ongoing LLM advancements in translation accuracy, context sensitivity, and transparency may shift the boundaries of what is considered safely automatable. However, unless system designers prioritize participatory design and incorporate translator-centric values, AI risk exacerbating labor precarity, error propagation in high-risk domains, and socio-cultural erasure in low-resource language contexts [seyi, pasquinelli2023eye]. Countervailing design patterns could include systems-of-record for translation provenance, integrated feedback loops for iterative translator supervision, and mechanisms that make model errors and confidence levels directly actionable for end-users.

Conclusion

The study provides a rigorous, situated account of translation professionals' encounters with translation technologies. The findings contest the appropriateness of translation automation paradigms, underline the irreducible human, creative, and interpretive dimensions of translation, and highlight the labor risks and misalignments provoked by current AI trajectories. The trajectory of AI in translation hinges not on automation efficacy but on sustained, context-aware dialogue with those whose expertise, agency, and work it will continue to shape.