© 2026 Lisa Baer-Tsarfati

Research

My research asks a simple, uncomfortable question: how does power learn to speak convincingly?

I study language as an infrastructure of governance, an operating system through which institutions define legitimacy, discipline behaviour, and make inequality feel reasonable. Working at the intersection of intellectual history, sociolinguistics, and digital humanities, I treat discourse not as a mirror of society but as one of its most effective technologies: a means of classifying authority, managing dissent, and stabilizing power over time.

Across both historical and contemporary projects, I examine how concepts such as ambition, harm, risk, responsibility, and fairness are produced through language, and how those concepts are converted into rules, policies, and technical systems. My work is concerned less with what institutions say they value than with how language quietly determines what they are able to see, justify, or ignore.

The Method Was Always the Point: Computational Approaches & Digital Humanities

Historical meaning does not announce itself. Words shift, concepts migrate, and the moral landscape of a period has to be reconstructed from within — from the patterns of how language actually moved, clustered, and cohered across texts, genres, and decades. Close reading alone cannot do that at scale. You need a method that can.

The computational approach at the heart of my research uses Latent Semantic Analysis to map the semantic field of key concepts across large historical corpora. LSA works by tracking statistical patterns of word co-occurrence — which terms travel together, which repel each other, which occupy contested middle ground. Applied to a curated corpus of 225 Scots- and English-language texts circulating in early modern Scotland, it revealed how ambition clustered consistently with pride, treason, avarice, and covetousness, while maintaining contested proximity to honour, virtue, and glory — precisely the sites where legitimate and illegitimate authority were actively being negotiated.

What makes this methodologically significant is not just what it found, but what it is. LSA is grounded in the same distributional semantics and vector-space mathematics that underpin modern word embedding models — the same mechanics that power large language models today. This is not coincidence. It is continuity. The same logic that allows a generative AI system to learn patterns of association from text was used here to recover how early modern writers learned, repeated, and reshaped the moral lexicon of their time.

That continuity is now central to my contemporary research. Understanding how these models work — not as neutral tools but as systems that inherit and reproduce the statistical patterns of the language they are trained on — requires exactly the kind of critical humanistic training that historical discourse analysis develops. The archive got larger. The stakes got higher. The question stayed the same.

When Ambition Becomes a Crime: Historical Research

Condemnation and Control: Ambition in Scotland, 1550–1625

Ambition is not a personality flaw. It never was. It is a charge, a diagnostic label applied to claims on power that someone has decided are illegitimate.

That argument is at the heart of this research. In early modern Scotland, ambition functioned as a cultural technology of governance: a way of identifying political overreach, explaining instability, and authorizing intervention without having to dismantle the political order itself. Condemnations of ambition intensified precisely when authority was most uncertain—during the Scottish Reformation, royal minorities, factional conflict, and contested sovereignty. When the rules of power were up for grabs, the language of ambition put them back.

Drawing on a corpus of sermons, moral treatises, proclamations, diplomatic correspondence, and legal records, and excavated using the computational methods described above, this research traces how ambition moved across moral discourse, political writing, and legal practice simultaneously. In each domain it did slightly different work,  framing overreach as vice, justifying competing theories of authority, shaping the mechanisms through which dissent was managed and suppressed. But across all three, the function was consistent: ambition translated structural conflict into personal moral failure, making inequality feel like justice and control feel like order.

That is not a sixteenth-century problem. It is a template.

The Insult of Ambition: Gender, Authority, & Political Language

In cultural discourse, ambition was heavily feminized, associated with unruly women, effeminacy, and moral instability. The ambitious woman was a stock figure of anxiety: threatening not just because of what she wanted, but because her wanting exposed the arbitrariness of the rules that said she couldn’t have it.

Yet in legal and administrative practice, the language used to discipline ambition was strikingly ungendered. Institutions were often more pragmatic than the polemicists who spoke for them. The rhetorical hysteria about female authority and the measured, bureaucratic mechanisms through which ambition was actually suppressed operated in parallel, related but distinct systems, doing different kinds of work.

That disjunction matters. It explains both the intensity of cultural anxiety surrounding female rule and the relative absence of ambition as a formal prosecutable offense in church court records. Gendered rhetoric was doing ideological work. Institutional practice was doing governance work. The two are not the same, and conflating them obscures how power actually operates.

Old Ghosts, New Machines: Ongoing & Future Research

 

The rhetoric of control doesn’t disappear. It simply finds new targets.

The same logic that turned ambition into an accusation, that translated structural conflict into personal moral failure, that made inequality feel like justice and dissent feel like deviance is embedded into the systems governing contemporary life. Content moderation policies define “harm” without defining whose harm counts. Algorithmic risk assessments encode existing hierarchies as neutral outputs. Ethical frameworks invoke “fairness” and “responsibility” while quietly determining who bears the cost of both.

These are not new problems wearing new clothes. They are old problems running on new infrastructure, and they are significantly harder to contest when the mechanisms are automated, opaque, and presented as objective.

My contemporary research examines how moral and political categories — harm, risk, fairness, responsibility, misuse — are produced through language in policy documents, ethical frameworks, and governance standards, and how those definitions become actionable within AI systems and data governance infrastructure. This includes a genealogy of the colonial logics embedded in AI design: how systems built to perform helpfulness and availability simultaneously erase the knowledge traditions of the communities whose labour and language trained them. It includes an investigation into what is lost when computational systems mediate cultural memory at scale: the provenance, the ambiguity, the interpretive disagreement that humanistic inquiry treats as data and automated summarization treats as noise. And it includes an audit of how AI safety systems govern speech: specifically, how they reproduce institutional patterns of minimization and misclassification when they encounter critical, situated, or emotionally complex language from users they were nominally built to serve.

The question is the same one I brought to the sixteenth-century sources. Not what institutions say they value. How language quietly determines what they are able to see, justify, and ignore.

The archive got larger. The stakes got higher. The tools got faster. The question stayed the same.

Receipts, Footnotes, & Further Trouble: Explore the Research

For readers interested in engaging more deeply with my work, the following materials provide further detail and examples: