What Does It Mean for a Text Analysis Tool to Be 'Objective'?

The Myth of the Neutral Tool

There is a seductive story told about computational methods in the humanities: that algorithms, unlike human readers, are free from bias. Feed a corpus into Voyant Tools, run a TF-IDF analysis, produce a word cloud — and what you get back is, in some sense, just what is there.

I spent several years questioning that story. The result was my 2023 article in Digital Scholarship in the Humanities, where I proposed what I call Hermeneutical Postphenomenology as the appropriate framework for understanding what computational tools actually do when applied to texts.

Don Ihde and the Human–Technology Relation

Postphenomenology, developed by the philosopher Don Ihde and extended by Peter-Paul Verbeek, studies how technologies mediate human experience. Ihde distinguished several types of human–technology relation. In an embodiment relation, the technology recedes from attention — we perceive through it (glasses, a cane). In an interpretive relation, we read a representation the technology produces (a thermometer reading, a brain scan). Digital humanities tools create a hermeneutic relation with texts — but they also embed assumptions about what counts as a meaningful unit, what relationships are worth tracking, what gets visualised and what gets silenced.

What This Means in Practice

Consider Voyant's Cirrus (word cloud). Words are sized by frequency. Already, that is a theoretical claim: that frequency is a proxy for significance. Stopwords are removed by default — but which stopwords? The list reflects assumptions rooted in English-language processing. When I began working with Yoruba-language corpora, these assumptions became visible in ways they are not when working with English texts. The tool's seams showed.

Or consider topic modelling. Latent Dirichlet Allocation (LDA) identifies statistically co-occurring word clusters and calls them "topics." The number of topics is a parameter chosen by the researcher. The algorithm is doing something, but what it produces is not a fact about the text — it is a co-production of algorithm, parameters, corpus, and interpreter.

Hermeneutical Postphenomenology as a Critical Practice

My proposal is not that we abandon computational methods. It is that we approach them with hermeneutical postphenomenological awareness — an understanding that every tool is a perspective, not a window. This awareness has practical consequences: transparency about tools and their assumptions, cross-validation across multiple methods, and sustained attention to what the tool does not see.

Towards a More Honest Digital Humanities

The best digital humanities scholarship is already doing this — reflexively, rigorously, with awareness of its own mediation. The framework I propose names and systematises what good practitioners already know: that the tool is not a neutral observer, and that this is not a problem to be overcome but a condition to be understood.