Linguists in the AI era: From resistance to renaissance

Geneva is a city of historical and current interplay between linguistics and technology. Historically, in Geneva, Ferdinand de Saussure laid the basis of scientific linguistics and many concepts behind current Large Language Models (LLMs). Nowadays, it is a city of hundreds of translators and interpreters translating among six UN languages in thousands of international meetings. Geneva’s School of Translation and Interpretation is a widely renowned linguistic teaching and research centre.

All of these, in addition to cutting-edge technological developments, make Geneva a unique place to address the impact of AI on the linguistics profession as ChatGPT and other LLMs begin to take over translation and interpretation jobs.

Will machines render human expertise obsolete? The answer is both yes and no. While AI will automate many language tasks, the linguistic profession is not vanishing—it is transforming. At the heart of this evolution lies a tension between syntax and semantics. Linguists who engage with this duality stand to thrive, becoming essential architects in shaping AI’s capacity to navigate human language.

Syntax vs. semantics: The core of AI’s blind spot

Take two sentences: “Vandals destroyed the shop” (active) and “The shop was destroyed by vandals” (passive). Structurally distinct, they share identical meanings. AI systems, such as large language models, excel at parsing syntactic patterns—tracking word order, grammatical roles, and token frequencies. However, meaning remains elusive. Semantics requires understanding why the passive voice shifts focus to the shop’s fate, or how cultural context shapes metaphors like “a storm of protests”.

This divide mirrors linguistics’ intellectual history. Early 20th-century scholars, like Ferdinand de Saussure, framed language as a structural system, prioritising syntax. Decades later, Noam Chomsky revolutionised the field by arguing that syntax alone couldn’t explain language’s creativity or how humans infer meaning from ambiguity. His transformational grammar reinstated semantics as indispensable.

Today’s LLMs embody this unresolved debate. They generate fluent text by mimicking syntactic patterns in vast datasets but often miss nuance, irony, or intent. A model might mechanically translate “break a leg” into another language, oblivious to its idiomatic meaning. It cannot grasp why a diplomatic statement uses the passive voice to obscure accountability or when doing so risks ethical compromise.

Linguists as AI’s bridge builders

This gap is where linguists hold unique value. Their expertise in both structure and meaning positions them to address AI’s limitations. Three opportunities emerge:

Training models beyond data scarcity: LLMs struggle with low-resource languages lacking extensive digital corpora. Linguists can encode syntactic rules and semantic frameworks into these systems, ensuring smaller languages are not marginalised in AI developments.

Annotating for depth, not just data: Machines learn from labelled text like layers of meaning analogous to mediaeval palimpsests. Linguists can play a critical role by annotating texts to capture irony, politeness strategies, or domain-specific semantics (e.g., legal “reasonable doubt” vs. colloquial use). Such work transforms raw data into tools for teaching AI contextual sensitivity.

Responsible and ethical AI design: In high-stakes domains—law, medicine, diplomacy—miscommunication carries real consequences. Linguists can audit AI outputs, ensuring translations preserve intent or that automated summaries avoid distorting nuance.

A call to action: Adaptation over resistance

Resisting AI’s rise proves as futile as resisting the printing press. Instead, linguists might pivot by focusing on four strategic shifts:

Strategic upskilling: Acquiring a basic understanding of AI and skills such as data annotation and syntax-based analysis of AI.

Specialising in tacit knowledge: Prioritising domains where semantics dominate—transcreating marketing content, interpreting cultural subtext, or curating ethically sensitive datasets—capitalises on irreplaceable human judgement.

Collaborative partnerships: Working alongside AI developers to refine model training, troubleshoot semantic errors, or design adaptive language technologies, especially in under-represented languages.

Championing ethical priorities: Advocating for transparency in AI’s decision-making and pushing to preserve linguistic diversity ensures technology respects, rather than erases, cultural nuance.

Diplo’s experience in AI and diplomatic language

Diplo transcribed more than 1000 UN and other meetings. As we have been training and adapting our AI tools, we noticed the following main challenges in AI transcribing:

Accents

Problem: Struggles with non-native or heavily accented speech;

Solutions:

• training for typical accents (e.g. Indian, Slavic, Spanish, French)
• training for the accent of frequently featured speakers – e.g. secretaries general and directors general of IOS, presidents, and ministers of foreign affairs.

Numbers/Dates

Problem: Frequent errors in figures, especially with zeros (e.g., “March 16, 2000” instead of “2022”);

Solutions: Additional training of the model for different ways of presenting numbers and dates.

Proper nouns and names

Problem: Names of countries, geographical toponyms and people are often misinterpreted;

Solutions: Additional training, starting with the most frequently used names (e.g. names of countries and cities).

Technical terms

Problem: Mistranslations of specialised language for telecommunications, AI, and privacy;

Solutions: Develop customised dictionaries and retrain the basic AI model.

Conclusion: Linguistics’ new frontier

Chomsky once critiqued AI’s purely statistical approach as ‘rote mimicry’. By bridging syntax and semantics, linguists can transform LLMs from pattern-recognition engines into tools capable of thoughtful communication. This is not about job preservation—it is about elevating language itself.

The linguistic professions are not disappearing. They are entering a renaissance. Those who embrace syntax and semantics as twin compass points will ensure AI amplifies—rather than flattens—human expression. In doing so, their work can get new relevance for tech changes ahead of us.

Geneva and its diverse and rich community of translators and interpreters can play a vital role in charting linguistic preparations for the AI era.