Research project · Philosophy · Linguistics · Cognitive Science · AI NSF SES-2336713

Human and Machine Translation
Cognitive, Linguistic, and Philosophical Perspectives

A philosophically oriented, practically informed study of translation technologies — from CAT memory to neural attention, LLMs, semantic opacity, and interlingua.

DIRECT · TRANSLATION MEMORY SYNTACTIC TRANSFER SEMANTIC TRANSFER analysis interlingua? generation SOURCE TEXT TARGET TEXT DEPTH OF ANALYSIS
Levels of representation in translation, after Vauquois. How far up this diagram do today's systems — and human translators — really operate?
The project

Translation from one language into another is a difficult and cognitively intense process. It draws on a broad set of linguistic and non-linguistic skills — search techniques, memory load control, multitasking, rapid mental set switching, and other executive functions — as well as the ability to keep in the air many syntactic, morphological, semantic, and pragmatic parameters of both languages, often interrelated in complicated ways. This occupation has long been the subject of translation studies, a diverse discipline blending into literary criticism on one side and neuroscience on the other. Philosophers, however, have rarely ventured into it.

This interdisciplinary project, supported by an NSF Scholar's Award (SES-2336713), seeks to fill that lacuna. Its objective is to explore the uneasy, complicated relationship between human and machine translation: a case study in the history and current state of translation technologies, conducted from the complementary perspectives of philosophy, linguistics, and cognitive science, and informed by first-hand knowledge of professional translation workflows. The project has two interrelated sides.

Theoretical · Meaning & Representation

How do translation systems represent meaning?

The theoretical side examines the representation of linguistic meaning in human, machine, and hybrid human–machine translation systems — from the translator's extended and distributed cognition to the semantically opaque vector spaces of neural models. A close look at translation technologies offers philosophy of language something it has never had: a live laboratory for the problem of meaning.

Practical · Evaluation & Workflows

How do humans and machines actually translate together?

The practical side studies the forms of human–machine symbiosis in technical (non-literary) translation and ways of improving them: rigorous evaluation and benchmarking of machine translation and LLM systems — including local, privacy-preserving models — and data-driven analysis of real professional workflows, combining automatic metrics with human judgment.

THEORY ⟷ PRACTICE

A better understanding of the cognitive, linguistic, and philosophical foundations of translation may suggest new ways of leveraging the strengths of humans and machines — while a close look at how they interact in real life may offer new insights into how physical systems represent linguistic meaning and, more ambitiously, what linguistic meaning consists in.

An open question

The advent of large language models has added more complexity to the picture: they can translate, among many other things — often remarkably well, without ever being taught how. What explains the origins of their translation abilities? Philosophers have approached the problem of meaning from many angles, but never in the context of these developments. This project takes them up.

Selected talks from the project
2026Translation Analytics for Freelancers II: Benchmarking Local LLMs — EAMT, Tilburg
2025On the Origins of the Translation Abilities of Large Language Models — AMTA (online)
2025Language Technologies and Interlingua — 11th EST Congress, University of Leeds
2025Translation Analytics for Freelancers I — MT Summit XX, University of Geneva
2024Human and Machine Translation: From Memory to Attention (and Back?) — Philosophy of Science Association, New Orleans
2024On the Potential Significance of 'Nine' in 'Canine': Quine Meets Deep Learning — Philosophy Colloquium, UC Davis
Acknowledgments

This project is supported by NSF Award SES-2336713. Earlier work was made possible by a Study in a Second Discipline Fellowship, which allowed me to audit classes in linguistics. My thanks to UGA's Office of the Senior Vice President for Academic Affairs and Provost, the Franklin College of Arts and Sciences, and the Departments of Philosophy and Linguistics. Any opinions, findings, and conclusions expressed here are those of the author and do not necessarily reflect the views of the National Science Foundation.