“Our Word is Our Weapon”: Text-Analyzing Wars of Ideas from the French Revolution to

the First World War

Jeff Jacobs

Submitted in partial fulfillment of the

requirements for the degree of
Doctor of Philosophy
under the Executive Committee
of the Graduate School of Arts and Sciences

COLUMBIA UNIVERSITY

2022
 © 2022

Jeff Jacobs

All Rights Reserved

 Abstract
“Our Word is Our Weapon”: Text-Analyzing Wars of Ideas from the French Revolution to
the First World War
Jeff Jacobs

What are political thinkers doing with their words when they write a text, engage in
a debate, or give a speech? We propose a “computational political theory”, pairing recent
breakthroughs in computational linguistics with the hermeneutic practices of intellectual
history, as a set of tools for mapping out the political-discursive fields within which ideas
circulate. We begin by showing, via a series of historical case studies, how a particular class
of computational-linguistic algorithms called word embeddings are able to capture subtle
differences in how authors employ certain contested terms (liberty, freedom, sovereignty,
etc.) by explicitly modeling both the words and the contexts they’re used in across a corpus
of texts. We then demonstrate how the results of these embedding models can shed light on
important questions in the history of political thought, by performing two in-depth studies
of the origins and trajectories of Marxism from the 19th to the 20th century.
In the first study, we use these models to trace the construction of Marx’s thought out
of the raw intellectual materials of 18th and early-19th century philosophy. We combine
a new, comprehensive corpus of Marx’s complete works from 1835 to 1883 (N > 1200)
with a large sample (N = 250) of prominent 18th and early-19th century texts to measure
conceptual distance between Marx’s works and various schools of 19th-century thought (po-
litical economists, socialists, and Hegelian philosophers) over time. Two key breaks emerge in
Marx’s writings: (a) they become less Hegelian as he is exposed to Paris’ brand of working-
class-oriented socialism between 1843 and 1845, then (b) become more focused on issues of
political economy over the remainder of his life in London, from 1849 onwards.
Our second study turns from the origins to the illocutionary impacts of Marx’s published
works, assessing his influence on the broader socialist discourse of the 19th century using a
corpus of post-1850 socialist texts (N = 200). We find that Marx’s semantic trajectory is mir-
rored, with a lag, by changes in the semantic trajectory of European socialist thought. This
discourse shifts away from moralistic and Hegelian themes and towards a more positivistic
political-economic vocabulary, especially after Marx’s rise to public prominence in the wake
of the 1871 Paris Commune. Our findings thus trace out, within the computationally-inferred
ideological field of 19th-century socialist thought, how Marx’s unique blend of German phi-
losophy, French socialism, and British political economy defeated would-be competitors and
established his thought as the default language of European socialism by the time of Engels’
death in 1895.
This dissertation thus demonstrates the utility of modern context-sensitive language mod-
els as tools for historical research, providing a framework for their use in developing, testing,
and revising our understandings of key questions in the history of political thought.
 Table of Contents

Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii

Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi

Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi

Chapter 1: Meaning, Understanding, and Quantification in the History of Ideas . . . . 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Word Embeddings: The Geometry of Political Thought . . . . . . . . . . . . 9

1.2.1 The Historiography of Political Thought . . . . . . . . . . . . . . . . 12

1.2.2 From Computational Linguistics to the Cambridge School and Back . 18

1.2.3 Meaning and Context . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

1.2.4 Synchronic and Diachronic Analysis: Understanding the Langue-Parole

Distinction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

1.3 Theories of Influence, Past and Present . . . . . . . . . . . . . . . . . . . . . 62

1.3.1 Structure vs. Agency . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

1.3.2 Mapping and Evaluating the Theories . . . . . . . . . . . . . . . . . . 65

1.3.3 The Empirics of Influence: Historical Sketches . . . . . . . . . . . . . 71

1.4 Networks of Semantic Influence . . . . . . . . . . . . . . . . . . . . . . . . . 81

1.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

i
 1.A Graphical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

1.B Probabilistic Graphical Models . . . . . . . . . . . . . . . . . . . . . . . . . . 92

1.C Chronology of Major Texts and Events from McLellan 2007 . . . . . . . . . . 104

1.D Influence Model PGM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

Chapter 2: Communism in Context: The Genesis of Marxism, 1776–1848 . . . . . . . 109

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

2.1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

2.1.2 Background: Sources of Influence on Marx’s Thought . . . . . . . . . 110

2.1.3 The Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

2.1.4 The Corpora . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

2.2 Study I: Influences on Marx . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

2.2.1 German Philosophy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

2.2.2 French Socialism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

2.2.3 British Political Economy . . . . . . . . . . . . . . . . . . . . . . . . 141

2.2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

2.A Influence Study Corpora . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

2.A.1 Marx: Key Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

2.A.2 Influence Corpus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

Chapter 3: Forged in Battle: The Fall of Utopian Socialism and the Rise of Marxism,
1849–1914 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

3.2 Marx’s Interlocutors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

ii
 3.2.1 The Young Hegelians, 1843–1845: Bruno Bauer and Max Stirner . . . 157

3.2.2 “State Socialism” I, 1845–1849: Louis Blanc . . . . . . . . . . . . . . 157

3.2.3 Anarchism I, 1846–1849: Pierre-Joseph Proudhon . . . . . . . . . . . 158

3.2.4 Challengers to the Throne I, 1859–1860: Karl Vogt . . . . . . . . . . 161

3.2.5 “State Socialism” II, 1864–1883: Ferdinand Lassalle . . . . . . . . . . 162

3.2.6 Anarchism II, 1872–1883: Mikhail Bakunin . . . . . . . . . . . . . . . 163

3.2.7 Challengers to the Throne II, 1883–1884: Eugen Dühring . . . . . . . 165

3.3 Diachronic Embeddings: Marx and the Socialist Movement, 1849–1899 . . . 166

3.4 Semantic Leadership and Socialist Discourse . . . . . . . . . . . . . . . . . . 169

3.4.1 Quantitative Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 171

3.4.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

3.A Corpus Details: Marx’s Interlocutors . . . . . . . . . . . . . . . . . . . . . . 186

3.B Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188

3.B.1 Proudhon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188

3.B.2 Collections of Lassalle’s Works . . . . . . . . . . . . . . . . . . . . . . 194

3.B.3 Nietzsche’s Reading Records . . . . . . . . . . . . . . . . . . . . . . . 204

3.C Table Footnotes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

3.C.1 Table 3.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209

iii
 List of Figures

1.1 Graphical models of the relationships which context-sensitive text-analysis

algorithms can infer from sentences (1) through (4) above. . . . . . . . . . . 24

1.2 An illustration of how, after constructing the individual-sentence context net-

works in Figure 1.1, the algorithm can systematically merge the information
across sentences, eventually producing the final corpus-wide network shown
in Figure 1.3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

1.3 The final corpus-wide contextual network, combining all information from the
individual networks in Figure 1.1 (edge labels omitted for clarity). . . . . . . 25

1.4 A visualization of the “analogical math” which can be performed within word
embedding spaces, due to their ability to capture multiple types of word-word
relationships in the form of distances ranging over multiple dimensions: entity
vectors are typeset in fixed-width font (Entity), while relational vectors are
typeset in serif font (Relation). The vector from Russia to U.S., a relational
vector, is typeset in fixed-width font only to illustrate its mathematical form
(i.e., that it is computed via the subtraction two entity vectors). . . . . . . . 26

1.5 Visualizing three of the potential pathways by which one can move from the
Russia to the Kyiv vectors. . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

1.6 The word-level linguistic intuitions which, by constructing an appropriate ge-

ometric space, can be quantitatively measured by way of the Cosine Similarity
metric. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

1.7 An example of the higher-level properties of language which naturally emerge

when an embedding algorithm computes word-level contextual-semantic sim-
ilarities (that is, moves the points for words which occur in similar contexts
closer together) across a large corpus of text from many different languages.
(From Chi et al. 2020, Figure 6) . . . . . . . . . . . . . . . . . . . . . . . . . 30

iv
1.8 The mathematical operation (dimensionality reduction) by which we compress
the full set of word-context information in Hobbes’ Leviathan (the matrix on
the left) into a smaller matrix of two-dimensional vectors (the rows in the
right-side matrix) which best preserve contextual similarity. That is, similar
rows in the left-side matrix will be mapped to close-by points in 2D space,
with (x, y) coordinates as given in the right-side matrix. . . . . . . . . . . . 31

1.9 An example of the matrix decomposition procedure that word embedding

algorithms implement, to solve the problem of retaining information about
word-context relationships while reducing the M -dimensional representations
of each word down to 3 dimensions. (Example adapted from Kozlowski et al.
2019, Figure 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

1.10 A plot of the 3-dimensional word representations in Figure 1.9. . . . . . . . . 32

1.11 The Word-Context Matrix for the four-sentence Green Eggs and Ham corpus,
along with the two-dimensional embeddings computed for each word. . . . . 33

1.12 A plot of the two-dimensional vectors generated for the 8 unique words in the
Green Eggs and Ham corpus. . . . . . . . . . . . . . . . . . . . . . . . . . . 33

1.13 The same points as in Figure 1.12, but rotated clockwise so that the “love”
vector is aligned with the x-axis. . . . . . . . . . . . . . . . . . . . . . . . . . 34

1.14 The same points as in Figures 1.12 and 1.13, but this time rotated and rescaled
so that the vector from “cats” to “dogs” now serves as our x-axis. . . . . . . 36

1.15 A word embedding space constructed from the text of Hobbes’ Leviathan. . . 38

1.16 A visualization of BERT’s ability to compute numeric representations of the

same word in multiple different contexts. . . . . . . . . . . . . . . . . . . . . 40

1.17 A case where the information about the corpus (the Word-Context Matrix on
the left side) does not have sufficient statistical power to allow estimation of
the word and context vectors (the matrices on the right side). At a high level,
the algorithm is unlikely to be able to draw meaningful estimates for 7,680
parameters—the 3,840 entries in the word vectors matrix (the u parameters),
plus another 3,840 entries (not labeled) in the context vectors matrix V —
using only 25 input data points. Looking more closely at the structure of the
input data, we also see that it has no information from which to infer word
4’s context-sensitive relation to other words, as it appears in only one context
which is not shared by any other word in the corpus. . . . . . . . . . . . . . 42

v
1.18 100 randomly-generated points in 2D space, showing potential projections of
these points down into one dimension on both the x and y axes (the short
vertical and horizontal lines, respectively). . . . . . . . . . . . . . . . . . . . 44

1.19 The same data as plotted in Figure 1.18, but transformed via PCA so that the
horizontal axis represents the direction of greatest variance (the First Principal
Component) while the vertical axis represents the direction of second-greatest
variance (the Second Principal Component). . . . . . . . . . . . . . . . . . . 46

1.20 A two-dimensional embedding space trained on the WPA Slave Narratives

corpus, where we draw arrows connecting points for key racial terms (“white”
and “black”) and for terms denoting status within the slave system (“slave”
and “master”) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

1.21 The same points displayed in Figure 1.20, but transformed so that the hori-
zontal axis represents the black-to-white direction within the space, and the
vertical represents the slave-to-master direction. . . . . . . . . . . . . . . . . 48

1.22 The same embedding space as in Figures 1.20 and 1.21, with the vertical
axis now transformed to represent the man-to-woman direction, and with
additional key terms plotted to enable analysis of their interrelationships with
respect to these two sociologically-meaningful dimensions. . . . . . . . . . . . 49

1.23 A visualization of BERT’s ability to model the different contexts (and thus,
senses) in which words are used by different authors. In this case interference
falls squarely within the negative liberty (liberty as non-interference) clus-
ter, as a term central to distinguishing negative from other forms of liberty,
whereas absence falls between two clusters as it can be employed in both
republican (“absence of arbitrary power”) and negative-liberty (“absence of
interference”) contexts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

1.24 An illustration of how the logarithm (plotted in blue) “smoothes out” the
original corpus-wide frequencies (plotted in orange), where the horizontal axis
represents a word’s ranking in the corpus (n = 1 representing the most fre-
quent word, n = 2 the second most frequent, and so on) and the vertical
axis represents the frequencies of these words, measured in both original and
logarithmic units. In particular, note that the logarithmic plot decreases the
magnitude of the differences between the first ~30 words, and then slightly
magnifies the differences between the remaining ~70 words, relative to the
original frequency plot. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

1.26 A PGM where we’ve observed the person’s action (a = Go) but only have a
probability distribution over the weather w. . . . . . . . . . . . . . . . . . . 82

vi
1.27 The situation of one-way influence, where Leada1 →a2 (w) > Leada2 →a1 (w), al-
lowing us to infer that a1 ’s usage of w before 1848 had more influence on a2 ’s
usage over time than vice-versa. . . . . . . . . . . . . . . . . . . . . . . . . . 83

1.28 The situation of mutual influence, where Leada1 →a2 (w) = Leada2 →a1 (w), since
a1 ’s post-1848 usage of w exactly matches a2 ’s pre-1848 usage, and a2 ’s post-
1848 usage exactly matches a1 ’s pre-1848 usage. . . . . . . . . . . . . . . . . 85

1.29 A graphical model illustrating the initial decomposition which a historian of

the French Revolution might perform to make their analysis more tractable:
three nodes representing three historical entities—the Nobility, the Clergy,
and the Third Estate—and edges representing the interrelationships between
each pair of entities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

1.30 A graphical model illustrating the initial decomposition which a Marxist his-
torian of the French Revolution might perform to make their analysis more
tractable: three nodes representing three historical entities—the Bourgeoisie,
the Proletariat, and the Peasantry—and edges representing the interrelation-
ships between each pair of entities. . . . . . . . . . . . . . . . . . . . . . . . 91

1.31 A graphical model illustrating the initial decomposition which a “Great Man”
theorist might perform to make their analysis of the French Revolution more
tractable, with nodes for prominent individuals and edges between individuals
who are known to have interacted. . . . . . . . . . . . . . . . . . . . . . . . . 91

1.32 From Blei 2012, p. 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

1.33 A basic PGM, representing the relationship between w, the weather, and a,
the subsequent action of a person deciding whether to go out or stay in for
the night. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

1.34 A PGM representing the same situation as in Figure 1.33, except that the node
for variable a is now shaded, indicating a situation where we have observed
the person’s action (a = Go Out) but still only have a probability distribution
over the weather w. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

1.35 A first attempt at a PGM representing the data-generating process for an

observed word wd,i , the ith word in document d, within a text. . . . . . . . . 101

1.36 A PGM representing the Specify-Documents sub-process, in which the author

specifies a topic distribution θd . . . . . . . . . . . . . . . . . . . . . . . . . . 102

1.37 A PGM representing the Specify-Documents sub-process, using plate notation

to condense the repetition in Figure 1.36. . . . . . . . . . . . . . . . . . . . . 102

vii
1.38 A PGM representing the Specify-Topics sub-process, in which an author spec-
ifies a word importance score pt (w) for each word w and topic t. . . . . . . . 103

1.39 A PGM representing the Specify-Topics sub-process, using plate notation to

condense the repetition in Figure 1.38. . . . . . . . . . . . . . . . . . . . . . 103

1.40 The complete PGM representing the data-generating process for an observed
word wd,i , with the Specify-Documents and Specify-Topics sub-processes incor-
porated explicitly. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

1.41 A PGM representing the fusion of several individual models of French Rev-
olutionary entities, to explain the observed outcomes in the revolutionary
Legislative Assembly. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

1.42 The basic model of intellectual influence used to interpret diachronic shifts in
the word embedding spaces generated throughout the study, where each Xi,t
node represents an observed texts published by author i at time t, each Et
node represents the collection of historically-salient non-textual events which
occurred at time t, each Ri,t node represents a particular text read by au-
thor i at time t (the superscript j indexing among multiple texts read within
the same time period), and successive time periods are separated by dashed
vertical lines. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

2.1 Computational-linguistic “translations” of the hypothesis in Avineri 1968, that

key concepts deployed in Marx’s early and late works can both be mapped
onto a single set of concepts introduced in Hegel’s Logik. . . . . . . . . . . . 121

2.2 Computational-linguistic “translations” of the hypotheses in Avineri 1968. . . 122

2.3 Heatmap of Similarities between Hegel’s Texts . . . . . . . . . . . . . . . . . 129

2.4 Kernel Density Plot of Similarities between Hegel’s Texts . . . . . . . . . . . 130

2.5 Heatmap of Similarity Scores between Hegel’s Works and Randomly-Selected

Documents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

2.6 Kernel Density Plot of Similarity Scores between Hegel’s Works and Synthetic
Documents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

2.7 Heatmap of Similarity Scores between Hegel’s Works and Synthetic Documents133

2.8 Kernel Density Plot of Similarity Scores between Hegel’s Works and Synthetic
Documents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

viii
2.9 Marx’s similarity to Hegel, over time . . . . . . . . . . . . . . . . . . . . . . 134

2.10 Heatmap of semantic self-similarities between works of Proudhon . . . . . . 137

2.11 Kernel density plot of semantic self-similarities between works of Proudhon . 138

2.12 Heatmap of semantic similarities between Proudhon’s texts and randomly-

selected 19th-century French texts . . . . . . . . . . . . . . . . . . . . . . . . 139

2.13 KDE plot of semantic similarities between Proudhon’s texts and randomly-
selected 19th-century French texts . . . . . . . . . . . . . . . . . . . . . . . . 140

2.14 Heatmap of semantic similarities between Proudhon’s texts and randomly-

generated French-language texts . . . . . . . . . . . . . . . . . . . . . . . . . 140

2.15 KDE plot of semantic similarities between Proudhon’s texts and randomly-
generated French-language texts . . . . . . . . . . . . . . . . . . . . . . . . . 141

2.16 Marx’s similarity to Works of French Socialism, over time . . . . . . . . . . . 142

2.17 Heatmap of semantic self-similarities between works of political economy . . 143

2.18 KDE plot of semantic self-similarities between works of political economy . . 143

2.19 Heatmap of randomly-selected similarities . . . . . . . . . . . . . . . . . . . 144

2.20 KDE plot of randomly-selected similarities . . . . . . . . . . . . . . . . . . . 144

2.21 Heatmap of randomly-generated similarities . . . . . . . . . . . . . . . . . . 145

2.22 KDE plot of randomly-generated similarities . . . . . . . . . . . . . . . . . . 145

2.23 Marx’s similarity to works of Political Economy over time, major works . . . 146

2.24 Marx’s yearly volume of reading (computed based on the known or estimated
years of his reading notebooks) over the course of his lifetime. . . . . . . . . 147

3.1 The trajectories of Marx’s decade-by-decade centroids and the decade-by-

decade centroids of European socialist discourse, with respect to the static
Hegelian (defining the x-axis) and Political-Economic (defining the y-axis)
centroids. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

3.2 Rolling PE Similarity Scores for Marx’s Collected Writings . . . . . . . . . . 168

ix
3.3 A Principal Component Analysis (PCA) plot of the Political Economy cen-
troid with the top N = 25 words most distinctive to this subcorpus. . . . . . 170

3.4 A Principal Component Analysis (PCA) plot of the Hegelian centroid with
the top N = 25 words most distinctive to this subcorpus. . . . . . . . . . . . 170

3.5 Bakunin’s Literary Output, 1839–1876 . . . . . . . . . . . . . . . . . . . . . 175

3.6 Proudhon’s Literary Output, 1837–1868 . . . . . . . . . . . . . . . . . . . . 176

3.7 Lassalle’s Literary Output, 1840–1864 . . . . . . . . . . . . . . . . . . . . . . 177

3.8 The volume of correspondence between Marx and Lassalle, per year. Letters
from Marx to Lassalle are tabulated based on MECW Vols. 38–41. Letters
from Lassalle to Marx are tabulated based on Ferdinand Lassalle. Nachge-
lassene Briefe und Schriften. Herausgegeben von Gustav Mayer, Vol III: Der
Briefwechsel zwischen Lassalle und Marx, nebst Briefen von Friedrich Engels
und Jenny Marx an Lassalle und von Karl Marx an Gräfin Sophie Hatzfeldt
(Lassalle 1922). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178

3.9 The balance between T0 and T1 , in terms of the number of texts in each epoch,
given a cutoff date of 1849. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

3.10 PageRank scores for each author in the corpus . . . . . . . . . . . . . . . . . 181

x
 List of Tables

1.1 An intuitive schema for guessing what section of the New York Times a par-
ticular article was published in, using only the text of the article (i.e., by
counting the number of occurrences of each keyword in each article) . . . . . 10

1.2 The per-section accuracy scores for our word-count-based division of the New
York Times article corpus. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.3 Randomly-chosen words from Wikipedia articles on Thomas Hobbes and Karl
Marx, versus randomly-chosen words from across all Wikipedia articles . . . 95

1.4 The Conditional Probability Table for the PGM shown in Figure 1.33. . . . . 99

1.5 An example Conditional Probability Table for the Specify-Documents PGM

shown in Figures 1.36 and 1.37. . . . . . . . . . . . . . . . . . . . . . . . . . 102

1.6 An example Conditional Probability Table for the Specify-Topics PGM shown
in Figure 1.38. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

2.1 Books read by the “socialist study circle” founded by Moses Hess in Cologne
in 1842, according to Gregory 1983 . . . . . . . . . . . . . . . . . . . . . . . 128

3.1 A comparison of the dates of first reading for key political-economic texts, as
recorded in Proudhon’s and Marx’s respective reading notebooks. On Proud-
hon’s reading notebooks, see Appendix 3.B.1. On Marx’s, see Section 2.1.3.2.
Entries after Smith and Ricardo are listed in the order in which they appear
in Haubtmann 1982. Sources for each date of reading are given in Appendix
3.C. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

3.2 The balance between T0 and T1 , in terms of the number of texts in each epoch,
given a cutoff date of 1867. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180

3.3 PageRank scores, and the associated probabilities and counts, for each author
in the corpus, with a cutoff date of 1849. . . . . . . . . . . . . . . . . . . . . 181

xi
 Acknowledgements

First and foremost, I’d like to acknowledge Suresh for dragging me, kicking and screaming,
to the finish line, and being a model of what a responsible scholar with a Stakhanovite work
ethic looks like. And Hugo, for being a lap warmer and best friend literally up until the very
day this journey was complete. You two were the central pillars, holding up the foundation
of what was otherwise a chaotic, shaky edifice—my life over the past 8 years. I’d like to
anti-acknowledge Emily Delbridge (aka Emily Twines), for showing me the true depths of
evil and callousness that human beings are capable of. I’m still alive, and graduating, despite
your best efforts. I am infinitely thankful and lucky to have had Ellie Poling in my life the
past 2 years, accepting me with all of my wounds and imperfections and showing me the
depths of love and compassion that human beings are also capable of.
Aaliya for (among many many things) being there to help me get back on my feet literally
minutes after my release from the hospital, and Serena for giving up your evening that night
to quell my distress by watching a silly movie with me. Sara for friendship over those 72
chaotic hours, and Tina for a hug that saved my life. Lynda for housing and feeding me when
I couldn’t house or feed myself. Paxton and JT for trips out to Oakland despite the fact that
I’m the one with a car. Steve and Ellen for overflowing with love, care, and support, not
to mention housing when I was homeless with an old, sickly cat. Dan Gudenius and Alyssa
for reaching out to me in DC and pulling me out of my slump, aided by Alaina, Ashley,
Jay, Jimmy, John, Leo, Radha, and Shawna. James and Franny, and Matthew Austin, for
cheering me on and providing comic relief or a pat on the back when I most needed it. Sarah
Chase for always being there whenever I need a friend, shoulder, or literally anything else,
whether in Colorado, College Park, Bethesda, or Seoul (where you’re my one friend awake
at 4am!). Andrew, Brad, Brendan, Brian, Jeff K, and Noah for companionship and laughs
from the womb to the tomb. Leah for knowing me through every cringy era of my life and
still loving me. Mom for loving me in all ways that were possible, and David for making her
happy. Dad for life lessons, and Lisa for being a companion to him. Kelsey for putting more

xii
energy than I thought was humanly possible into crafting the armor that now protects me
and the flashlights I keep in all my drawers.
At Columbia, Cristy for the Hungarian Pastry Shop study sessions and for enduring
the white-supremacist hammer brought down upon you by Columbia. Alex and Ben M. for
being the rare friends who I could bounce dissertation ideas off of, no matter how silly or
ranting. [Other] Ben M. for helping me with the meditation resources, among other friendly
supportive conversations. Sinéad for opening my eyes to a completely new literature, and
thus new research pathways, on the hidden epistemological assumptions buried within most
historiographic endeavors. Anja for putting up with my grumpy commie banter in class/group
chats for so many years. Dylan Groves, for reaching out to me and showing me kindness and
compassion in the midst during my radio silence. Tom Leavitt for finding the time to work
on projects with me, and check on how I’m doing, despite so so many other things going on
in your life. Sahana Subramanyam for indulging my chaotic brain via chats about a million
completely random yet fascinating topics. Katie Ebner-Landy, what can I say... for being
the most intellectually curious and ambitious person I’ve ever met (i.e., my definition of
“intelligent”). Elliott for showing me so many cool ways to mix computers, social science,
and law together in a big bubbling cauldron, and same to Arin and Sid.
From Columbia SJP/GWC, Janine for always helping me keep academia in perspec-
tive, and for having the highest fight to body mass index I’ve ever seen. Tom for being
my sole computer science comrade, and always checking on me while I flailed around the
country/world these past few years, from Texas to Paris to Palestine. Nas and Ibtihal for
the Law School crying-with-laughter sessions in the midst of the coldest winters. Brenda for
teaching me how to channel emotions into freedom-fighting. Hassan for always pushing me
to question beliefs I had become complacent in. Aaron, Andrea C., Corben, Darializa, Geoff,
George, Hannalore, Isaac, Isobel, Jim, Julia, Kate J, Matt M., Max, Nadine, Naye, Sam,
Sarah, Seth P, Sheehan, Sherif, Shezza, Talia, Tara, Yousr, Yujin, Zach, and so many others
for putting in the tireless organizing work without letting it affect your spirit.

xiii
 Alicia for love in a time of hopelessness. Koel and Meghan for understanding and enduring
my inconsistency. Kana for showing me parts of NYC I never knew existed on dog walks.
Sarah (Not My Cousin) Jacobs for post-panic attack comfort and empathy. Çigdem and
Colleen for being the friendliest strangers ever to walk into Hungarian. Jasper and Seth for
keeping me going by always injecting perfect doses of new music into my life, at the perfect
times. Charleyne for the beautiful apartment and beautiful cat in Paris. Simone for the
immensely helpful comments on my last presentation before the defense.
At NYU, Andrea, Will, and Angela for being a co-instructional dream team. At Berkeley,
Yojo for always going absurdly above and beyond when taking care of my cats. Josh for
being a beacon of rad companionship in an otherwise mind-numbing work environment.
Dan, Ernie, Keeanga, Kitty, Laura, Ragina, Sid, Todd, and so many others for showing me
what lowercase-s socialism looks like in Oakland and what capital-S Socialism looks like
in Chicago. At Stanford, Mike Kim for matching (usually exceeding) my excitement about
algorithms, for vegan dinners and for opening some kind of portal to a made-up coffee shop
in Berkeley that I haven’t been able to find since. Julius Cheng and Steven for keeping me
sane in the face of our neoliberal Palo Alto predicament. Nick Isaacs for being a model of
a STEMer in touch with his feelings, always helping me in my journey to be one as well.
Rachel for the love and empathy through so many ups and downs. Cameron, Daniel, Kyle,
Kyler, Lashi, Neha, Raiyan, Stephen, Yifei, and many others for making Palo Alto bearable
on the Stanford side of ECR, and Chelsea, Elissa, Holman, Orchid, Reeta B., and Shelly for
making the Palo Alto side bearable. In Hong Kong, August, Christy, Sean, Honghao, Qifeng,
Weixiang, and others for introducing me to custard buns. In Illinois, Pate, Austin, Victoria,
LaDean, and others for putting up with me in my Lil Wayne era. In Beijing, Brendan,
Karthik, and Julian Drizzy for letting me teach you about Shaolin Shadowboxing/the Wu-
Tang Sword Style. In Santa Fe, Mike, Zackary, and others for lovely company in the middle
of the desert. Chris, Merry, Salin, and Syed for hostel camaraderie in the Boogie Down.
Josh, my first friend from my first day at UMD, for putting up with all my outbursts,

xiv
phases, successes, and disappointments since then. Jefferson and Chris for late-night talks
and enough ridiculousness to fill an entire whiteboard wall. Pat for introducing me to literally
hundreds of new artists and albums which still get me motivated when I need it the most. I
cannot wait to meet your beautiful new child. Rehan for the lax tips. Kelly for the joy you
beamed into me and the company during long periods when I was scared to be alone with
myself. Dr. Gasarch for being the definition of a charitable, selfless scholar—for all the time
you somehow found to mentor me and countless other DC-area math nerds, and for showing
us that academia meant far, far more than classes, exams, and grades (also for giving me
the nickname “Badass” of your own volition). I still owe you approximately $9K in lunches.
Aashish, Mike D., Andrew for being three of the funniest human beings I’ve ever met, and
for co-inventing Infinite Coca-Cola (why stop at finite profits?) with me. Alex G. for being
the cooler, counterfactual version of me if I had gone to school in PG instead of Moco. Pete
for teaching me to keep pushing through my mental health struggles by sharing what you
learned from yours. Allie for helping me heal from my first heartbreak. Ateeyeh Atefat for
friendship and a place to crash in Baltimore, and Leyla Korkut for the same in Bethesda.
In Palestine, Sadek for your self-made car sound system and your threats to “shut this
whole thing down”, Taylor for just an absurd and selfless work ethic keeping things running
smoothly on top of your normal job, and similarly for Alex. Laith for fighting all the fights
that life forced you into, and for taking my “every minute you sleep is a minute your enemies
are learning how to code” joke as only kind of a joke. Amro, Arori, Mo, and Tariq for
patiently teaching an ignorant American about four very different dimensions of Palestine.
Ghada for being the GOAT MC, bored by any beat slower than 200bpm. Romeo for bearing
with me while I was mentally unable to function in Nashville. Cooper for living literally 10
feet from me in Austin yet somehow never meeting me. Amartya, Ammar, Basil, Ben for
being a checken, David, Dina for letting me be your TA, Divya, Freya, Kareem for teaching
me “bad words!!”, Kyla, Laura, Mateo, Matthew, Maya, Misal, Comrade Babloo, Rekha,
Remi, Rory, Samah, Samed, Comrade Shelby, Sierra, Zaeda, Zunaira, and so many others.

xv
 Dedication

Dedicated to all who struggle contra el gran enemigo del género humano.

xvi
 Abbreviations

ADAV Allgemeiner Deutscher Arbeiter-Verein (General German Workers’ Association,

1863–1875)

CPT Conditional Probability Table

HMM Hidden Markov Model

IWMA International Workingmen’s Association (i.e., First International), 1864–1876

LDA Latent Dirichlet Allocation

PGM Probabilistic Graphical Model

SDAP Sozialdemokratische Arbeiterpartei Deutschlands (Social-Democratic Workers’ Party

of Germany, 1869–1875)

SPD Sozialdemokratische Partei Deutschlands (Social-Democratic Party of Germany,

1875–)

xvii
 Chapter 1: Meaning, Understanding, and Quantification in the
History of Ideas

Even the most abstract works of political theory

are never above the battle; they are always part
of the battle itself.

Quentin Skinner, Hobbes and Republican Liberty

(Skinner 2008), p. xv

The weapon of criticism cannot, of course,

replace criticism by weapons. Material force
must be overthrown by material force. But
philosophy also becomes a material force, once it
has gripped the masses.

Karl Marx, “Contribution to the Critique of

Hegel’s Philosophy of Right”, MECW 3, p. 182

Worte sind auch Taten.

(Words are also deeds.)

Wittgenstein, Philosophical Investigations

(Wittgenstein 1953), p. 155

1.1 Introduction

What do political thinkers do with their words when they perform a political speech
act—when they publish a treatise, give a speech, argue for a piece of legislation, and so
on? Since the 1960s a group of historians centered around Cambridge University professors
Quentin Skinner, J. G. A. Pocock, and John Dunn have changed the way we understand
several key political thinkers and texts1 , by consciously developing and applying a linguistic-
1
The foundational works of these scholars (discussed below) focus primarily on thinkers of the 16th
and 17th centuries: Machiavelli, Hobbes, and Locke in particular. As will become clear over the course

1
philosophical and context-sensitive methodology which places this question at the center
of historical inquiry (Skinner 1969, Pocock 1985). As part of this endeavor the Cambridge
School, as this group came to be known, placed a strong emphasis on the need for historians
to explicitly describe and justify the methodologies being employed in their studies2 Most
Cambridge School works therefore begin with extensive methodological introductions, in
which particular modes of inquiry are introduced and justified prior to their application in
the remainder of the work. Kenneth Minogue describes Skinner’s early work, for example, as

“primarily of interest to philosophers not for its excellent account of European

thought about the state but for the self-conscious philosophy which has gone
into it. It is a rare historian who pauses to get his philosophy in order before he
embarks on a major enterprise” (Tully 1988, p. 176)

Thus the first of the three volumes of Quentin Skinner’s Visions of Politics, for example,
is entirely devoted to method (providing 209 pages of methodological justification before
discussing the Renaissance and Thomas Hobbes in Volumes 2 and 3, respectively), while J.
G. A. Pocock’s Virtue, Commerce, and History begins with an intensive 36-page treatise on
methodology, applying lessons learned from prior investigations to refine his approach before
embarking on the inquiries of the remainder of the book.
It was not their emphasis on method as such, however, that set the Cambridge School
apart from other historiographic schools—indeed, many other postwar hermeneutic ap-
proaches (e.g., Derridean deconstructivism) have also accumulated a vast methodological lit-
of our argument, this focus on pre-industrial political thinkers is not incidental; there are fundamental
limitations to their approach which render studies of post-17th-century thought—i.e., thought in the era of
mass communication and literacy—infeasible without the aid of computational tools like those we introduce
in the coming sections.
2
Pocock, in fact, puts the point more strongly, arguing that the Cambridge School introduced this self-
reflection into the field of intellectual history:
“it was only in the middle 1960s, with the first appearance of writings by Quentin Skinner,
that historians of political thought began to state the logic of their own inquiry and pursue it
into fields where it encountered the philosophy of language.” (Pocock 1985, p. 3)
We explore this claim—placing the Cambridge School itself within its broader historical context to argue
that Popper, not Skinner, can be credited with introducing this focus (a focus whose lineage thus contains
important political consequences)—in Section 1.2.2 below.

2
erature3 . Rather, it was their focus on contextual analysis of texts—their call to de-emphasize
the “so-called ‘classic texts’ [...] and focus instead on the more general social and intellectual
matrix out of which [these] works arose” (Skinner 1978a, p. x)—that enabled their remark-
able impact on political theory and the history of political thought, and that motivates our
adoption and extension of their approach via computational-linguistic tools.
In fact, as we will argue in the remainder of this chapter, these computational tools
represent a radical breakthrough for historiographic practice, unlocking for the first time
the full potential of the Cambridge School approach for enriching our understanding of the
history of political thought. This immanent potential, we argue (building on the argument
of London 2016), was latent in the pre-computational formulation of the Cambridge School
approach, but unrealizable in practice due to the fundamental limits of human reading
comprehension. Cambridge School studies to this point, notwithstanding the massive impact
they’ve already had on the landscape of our historical knowledge, have been restricted to
instances in which the relevant context necessary to understand a given thinker or text was
on a scale small enough to allow one person (or a small team of people) to not only read
and process these pieces of context, but also to then synthesize what had been learned into
a coherent narrative or explanation. Hence Quentin Skinner explicitly describes his overall
historical project as one restricted to studying “the European tradition only during its most
formative days, which I take to have been the sixteenth and seventeenth centuries.” (Scott
and Keates 2001, p. 15) Cambridge School researchers have thus aptly demonstrated—in
studying clerical disputes in the early stages of the Protestant Reformation (Skinner 1978b),
for example, or the revival of Roman republican ideals in sixteenth century Florence (Pocock
1975)—the efficacy of this approach when the scope of the community of discourse under
examination can be feasibly circumscribed.
These two examples, however, and other impactful Cambridge School works, are note-
worthy perhaps for the very reason that they run up against the limits of human ability:
3
See Skinner 1990 or Tully 1988, for example, for conversations between Cambridge School practitioners
and scholars representing a wide range of alternative hermeneutic perspectives.

3
they examine “general social and intellectual matri[ces]” capacious enough to derive pro-
found insights into the history of political thought, but not so expansive as to render the
project infeasible for one or a few people. Pocock summarizes his goal in Virtue, Commerce,
and History, for example, as one of writing “a history of actors uttering and responding in a
shared yet diverse linguistic context.” Importantly, Pocock asserts, the subject of this history
is bounded not only by this shared context but also by the fact that its actors communicate
via the “internal” circulation of their writings, in the form of published manuscripts:

“We need not therefore apologize for the unrepresentative elitism of studying
only those readers whose responses were verbalized, recorded, and presented.”
(Pocock 1985, p. 18)

We argue that in general, but especially in the case of Marx and his desire to enact social
change on a mass scale (a desire succinctly and powerfully captured in his 11th Thesis on
Feuerbach4 ), the very nature of the political discourse we hope to study was transformed
over the course of the nineteenth century, in both its material and ideological dimensions.
Materially, as we discuss in more detail below, the accelerating development of commu-
nication and transportation technology combined with the spread of literacy gave rise to
international political-discursive fields which were no longer restricted to the small circle
of “literary elites” referred to by Pocock (e.g., the eighteenth-century French salonnières),
but now encompassed burgeoning mass publics engaging via new developments like workers’
newspapers and study groups. These infrastructural and institutional changes were inevitably
accompanied by ideological developments, in terms of e.g. the audiences that writers had in
mind when composing and issuing political speech acts: writers like Marx and Engels who
were interested in mobilizing these mass publics had to tailor both the content and style
of their writings—references, metaphors, idiomatic expressions, etc.—in such a way as to
appeal to a worker in Portugal just as well as one in Poland.
4
“Die Philosophen haben die Welt nur verschieden interpretiert, es kömmt drauf an sie zu verändern.”
(MEGA2 IV/3, p. 21; MEW Vol. 3, p. 7) (“The philosophers have only interpreted the world in various ways;
the point is to change it.”) (MECW 5, p. 5)

4
 Indeed, while late eighteenth-century developments like the hortatory printings of Marat
and other French Revolutionary pamphleteers represent early moves in this direction, their
scope was limited mainly to Paris or other prominent regions of France. As we will argue
below, the move from these parochial appeals to Marx and Engels’ address to the “workers
of the world” 60 years later nicely captures the magnitude of the expansion of scope we are
referring to here.
This historical trajectory thus leads us to the question of why—i.e., on the basis of
what epistemological principle—we should believe that the sufficiently-relevant context for
understanding a given thinker or text can be restricted to this individually-manageable size,
as more and more entrants contribute to these “shared yet diverse linguistic context[s]”?5 .
If a given historical event is complex enough, that is, if the collection of salient factors is so
vast as to make an individual close-reading-based study truly impossible, then to conduct
such a non-computational contextual study is to fall prey to the so-called drunkard’s search
fallacy: searching for your lost keys under a streetlight not because you dropped them near
the streetlight but because the area around the streetlight is the only spot illuminated enough
to see6 .
Even if we do believe that the relevant context can be restricted to this extent, however,
there are important historical-epistemological principles (such as the covering-law model,
5
It is important to note at this point that an alternative approach to intellectual history, the Begriffs-
geschichte approach of Reinhart Koselleck and others (mainly in the German-speaking world), does not avoid
this drawback. Though this approach centers concepts (begriff ) as the unit of study, rather than thinkers
or texts, within our framework this essentially amounts (we argue) to debating whether it is more fruitful
to focus on the white squares or black squares of a chessboard if we want to understand chess: though
Cambridge School practitioners start with individual thinkers or texts, they inevitably must engage with the
concepts used by these thinkers or texts. Similarly, though Begriffsgeschichte analyses start with concepts,
they inevitably must understand them by way of the individual thinkers and texts which shaped and altered
(or failed to shape/alter) these concepts. We do note, however, that there is a more well-formed literature
on the fruitful use of computational-linguistic tools for Begriffsgeschichte than there is for the Cambridge
School approach. See, e.g., Wevers and Koolen 2020.
6
The name of this observational bias refers to the following joke: A policeman sees a drunkard searching
for something under a streetlight and asks what the drunkard is looking for. He says he’s lost his keys, so
the officer offers to help and joins him in searching for the keys around the streetlight. After a few minutes,
the policeman asks him if he’s sure he lost them here. The drunk replies: “no, I lost them in the park across
the street”. When the policeman angrily demands to know, then, why they are searching here, the drunkard
calmly replies, “because this is where the light is!”.

5
or more general principles of replicability and Bayesian inference) which imply that the
conclusions we draw from some individually-studied context are meaningful precisely to the
extent that they are robust to the collection of additional evidence—in this case, evidence
derived from the computational study of texts beyond the human-readable frontier. In other
words, even if one rejects our stronger claim that these computational methods are necessary
for conducting Cambridge School-style studies of nineteenth- and twentieth-century political
thought, we hope to still present a compelling argument for their use in “checking one’s
work”—for example, verifying that nothing in the output flatly contradicts one’s findings—
after carrying out a non-computational study.
Although we maintain that even the historical explananda of the famous Cambridge
School studies—the Renaissance, the Protestant Reformation, or Pocock’s Machiavellian
Moment—are already too complex for any individual to fully assimilate their relevant con-
texts, for the purposes of this work we rely on a less strong claim, one that does not preclude
this possibility for cases of pre-industrial phenomena. Namely, we argue that by the time
of the French Revolution, and especially by the time of the advent and spread of Marx-
ism across Europe, literacy and communication technology had expanded the scope of the
relevant “social and intellectual matrix” underlying key political texts to the point that
a non-computational assimilation of this matrix (or, in other words, its comprehension in
totality by one individual) is no longer possible.
Indeed, as we move from the French Revolution to the 19th-century advent and spread of
Marxism, we enter an era where the conduct of political polemics undergoes a massive set of
changes. Not only are political speech acts increasingly conducted on a global scale, but also
increasingly able to penetrate below the level of a society’s elites, reaching and addressing
(or failing to address) the concerns of ordinary citizens who found themselves swept up in the
violent upheavals engendered by the Industrial Revolution, with its “constant revolutionising
of production, uninterrupted disturbance of all social conditions, [and] everlasting uncertainty
and agitation” (MECW 6, 487). The effect of these changes can be observed, for example,

6
in the Preamble to the Communist Manifesto—an archetypical example of an illocutionary
speech act in the form of a text—where it is explained that

Communists of various nationalities have assembled in London and sketched the

following manifesto, to be published in the English, French, German, Italian,
Flemish and Danish languages. (MECW 6, 481)

Thus we see that, in the eyes of the Communist League who issued the manifesto in
1848, the audience for their polemic was not restricted to the workers of London, nor was
it intended for the elites of England, France, Germany, Italy, Belgium, and Denmark, but
instead aimed to engage all workers across all of these nations7 .
The increasingly global character of these polemics highlights another dimension of the
intractability of the Cambridge School approach with respect to important historical phe-
nomena, namely, that the researcher is forced to read not just a vast number of texts, but a
vast number of texts across a vast range of languages. The challenge of explaining the 18th-
and 19th-century ideas which emerged from the discourse around freedom among creoles
of the colonies of North and South America, for example—a challenge taken up by Joshua
Simon in The Ideology of Creole Revolution—already brushes up against the limits of in-
dividual Cambridge School-style historical inquiry, requiring the assimilation of a massive
number of Spanish, English, French, and Portuguese texts. If we extended our investigation
of the spread of Marxism into the 20th century, for example, this problem would reach an
astronomical level of complexity, as the USSR’s ideological state apparatuses operated across
130 state-recognized domestic languages (Comrie 1981, p. 1) and over 60 foreign languages
into which Soviet propaganda was translated and exported (Jacobs 2021b).
The situation is not hopeless, however. By way of foreshadowing the possibilities enabled
by modern computational-linguistic methods, for example, Google’s Language-Agnostic BERT
7
Indeed, as discussed in Feuer 1963 and Gourevitch 2015 (pp. 185–189), German exiles would spread these
ideas beyond the borders of Europe in the aftermath of 1848–1849, transporting them across the Atlantic
and bringing them to bear on discussions around the “labor republicanism” propounded by radical U.S.
organizations like the Knights of Labor and (later) the Industrial Workers of the World (IWW).

7
Sentence Embeddings (LaBSE) model (Feng et al. 2022) is trained to encode semantic rela-
tionships among words and sentences across 109 different languages, and is able to reason
effectively across 35 additional languages due to “leaking” of semantic information from these
languages into the texts used to derive the 109-language model8 . Even more recently (and
indicative of the rapid progress in the field over just two years), Facebook Research’s “No
Language Left Behind” model (Heffernan et al. 2022) nearly doubles this number, allowing re-
searchers to analyze texts across over 200 languages via a single multilingual semantic space9 .
If, as we argue, researchers can use models like these to derive meaningful insights about the
way words and phrases are deployed in political speech acts, in spite of their lack of fluency
in the dozens of languages involved, it becomes possible to perform Cambridge School-style
contextual studies on a massive scale—like that of Soviet political discourse—while allowing
researchers to redirect their energies away from exhaustive close reading and language learn-
ing and towards the actual higher-level goal of the historical profession, namely, creatively
synthesizing the available historical data to deepen our understanding of political thought.
In the remainder of this chapter, then, with this ultimate goal in mind, we build to-
wards an understanding of the computational tools involved. Eschewing the gory details in
favor of just that minimal set of information which a historical practitioner would need to
know to perform their own computationally-aided inquiry, we emphasize throughout how the
computational models implement and/or extend established principles of empirical social sci-
ence which historians and political theorists already employ in their day-to-day practice. In
addition, to concretely illustrate these connections to existing practices, we pair each newly-
introduced tool or principle with examples of how they have been (or could be) fruitfully
applied to existing problems from the historical and political-theoretic literature.
In the next section we introduce the central focus of the chapter, semantic embedding
algorithms, and argue that they represent the quantitative implementation of precisely the
8
See the “Support to Unsupported Languages” section of Yang and Feng 2020 for details on these 35
languages and what it means for this linguistic information to “leak” into the model.
9
While Heffernan et al. 2022 provides details on how this model was trained, the model itself is available
via Facebook Research’s GitHub page.

8
“general social and intellectual matrix” which Cambridge School practitioners aim to analyze
over the course of a given study. After discussing how the most basic embedding algorithm—
the word2vec algorithm—can be extended to incorporate models of authorship as well as
influence over time, we conclude with a model which unifies these different extensions in
the form of embedding networks through which researchers can precisely estimate the degree
to which agents within a given social and intellectual matrix (e.g., authors, newspapers, or
institutions) influence one another over time.

1.2 Word Embeddings: The Geometry of Political Thought

Although computational algorithms for analyzing text have existed since the 1950s, un-
til recently the vast majority operated under a framework of drawing inferences from word
counts within and across documents drawn from a corpus. Indeed, though we will eventually
argue that these word-count-based approaches are insufficient for capturing semantic mean-
ing within political texts—since they fail to capture the contexts in which the words are
employed—they can nonetheless help researchers make basic inferences about the contents
of a text corpus before embarking on a full context-sensitive study. For example, when faced
with a new and unfamiliar corpus, a word-count-based approach could be helpful for quickly
dividing the corpus into sub-categories based on pre-specified keywords.
As an illustration, we scraped plaintext versions of all New York Times articles from May
2022, including metadata on what section the articles were published in (Sports, Business,
Arts, World News, etc.), and developed the intuitive schema for word-count based sorting
shown in Table 1.1.
With this schema in hand, we assigned each article a score for each potential section
based on how many times that section’s hypothesized keywords occurred in the article. By
categorizing each article based on the section with the highest keyword counts (so that, for
example, if the Food keywords appeared 5 times, the Technology keywords twice, and none
of the other keywords appeared in the article, it would be categorized as a Food article),

9
 Section Keywords
U.S. News state, court, federal, republican
World News government, country, officials, minister
Arts music, show, art, dance
Sports game, league, team, coach
Real Estate home, bedrooms, bathrooms, building

Table 1.1: An intuitive schema for guessing what section of the New York Times a particular
article was published in, using only the text of the article (i.e., by counting the number of
occurrences of each keyword in each article)

these guessed categories match the articles’ true NY Times sections an impressive 80.21%
of the time, a significant improvement upon the 20% accuracy one would get (on average)
via random guessing. Looking at the breakdown of this accuracy by section in Table 1.2,
however, we can begin to see the fundamental limitations of this approach.

Arts Real Estate Sports U.S. News World News Total

Correct 3020 690 4860 1330 1730 11630
Incorrect 750 60 370 1100 590 2870
Accuracy 0.801 0.920 0.929 0.547 0.746 0.802

Table 1.2: The per-section accuracy scores for our word-count-based division of the New York
Times article corpus.

While the Arts, Real Estate, and Sports sections use a vocabulary that is fairly unique to
those sections (thus, for example, the word “bedrooms” almost never appears outside of the
Real Estate section), distinguishing between U.S. News and World News is significantly more
difficult, as many of the words we could try to use to distinguish between them still occur
frequently in both sections. Knowing that the word “president” occurs frequently within an
article, for example, without knowing the surrounding context in which it is used, does not
help us very much in distinguishing between U.S. News and World News articles. And yet,
crucially, if we had a method for jointly capturing the number of occurrences of the word
and a breakdown of these occurrences by their contexts, suddenly our ability to distinguish
between these two categories would skyrocket. For example, if we knew the following about

10
two articles A and B:

• The word “President” appears 10 times in Article A, with 5 of these appearances

immediately preceding “Putin” and 5 immediately preceding “Clinton”, and

• The word “President” appears 10 times in Article B, with all 10 of these appearances
occurring immediately before “Clinton”,

we could now easily distinguish that Article A is likely to be a World News article while
Article B is likely to be a U.S. News article. With this insight, therefore—that purely word-
count-based approaches are insufficient for inferring meaning from text, but that the ability
to capture the contexts in which words appear brings us much closer to this goal—we turn
our attention away from these word-count-based models and towards a newer generation of
explicitly contextual language models.
Specifically, in the remainder of this section we introduce semantic embedding algorithms,
which enrich the analytical possibilities for text analysis by explicitly modeling the context
of a given lexical unit (word, phrase, sentence, etc.) alongside the lexical unit itself. By
incorporating this contextual information these algorithms can, as we’ll see below, learn the
meanings of words and phrases in a manner similar to that of human language acquisition—
a manner summarized in the so-called Firthian Hypothesis: “You shall know a word by
the company it keeps.” (Firth 1957, p. 11) First, however, to illustrate the relevance of these
considerations for the practice of political theory, we begin in the next section by outlining the
intellectual-historiographic debate about what role contextual knowledge ought to play in our
understanding of political-theoretic texts, especially in the aftermath of Quentin Skinner’s
provocative 1969 article “Meaning and Understanding in the History of Ideas” (Skinner
1969), which Pocock refers to retrospectively as “the manifesto of an emerging method of
interpreting the history of political thought” (Pocock 2009, p. 128; emphasis added), namely,
the method which became known as the Cambridge School approach.

11
1.2.1 The Historiography of Political Thought

Drawing on the linguistic philosophy of J. L. Austin, W. V. O. Quine, and the late

Wittgenstein10 , the structuralism of Ferdinand de Saussure, and the pragmatics of H. P.
Grice11 , scholars of the Cambridge School shifted much of contemporary historiography away
from the notion of “perennial questions” in political thought (Bevir 1994), and towards a
conception of historical texts as interventions into a particular, localized discourse.
Roughly speaking, an earlier school of intellectual historians (nowadays often associated
with Leo Strauss) viewed the “great minds” of history—Plato and Aristotle, Machiavelli and
Hobbes, Kant and Hegel—as engaged in a collective conversation on the “eternal questions”
of philosophy, such as the question of what constitutes a good life or a good society. As
summarized by Skinner, this school considered “a canon of leading texts” as “the only proper
object of research in the history of political thought”, given their unique contributions to “a
set of perennial questions definitive of political thinking itself.” (Skinner 1998, pp. 101–102).
Importantly—taking Strauss’ view as representative of this trend—it is only this echelon
of great minds who qualify as true political philosophers, since their philosophical work is
“animated by a moral impulse, the love of truth”, rather than the parochial desire to e.g. win
an argument or persuade members of the public to adopt their preferences (Strauss 1959).
In fact, under this conception, political philosophers must divorce themselves entirely from
such local or day-to-day political concerns, as “it is only when the Here and Now ceases to
be the center of reference that a philosophic or scientific approach to politics can emerge”.
In practice, then, the historian of political thought obtains the desired understanding of a
10
In the methodological volume of his 3-volume Visions of Politics, for example, Skinner explains that
“Among philosophers of language, my approach owes most to” Wittgenstein’s Philosophical Investigations
(Wittgenstein 1953) and Austin’s How to Do Things with Words (Austin 1962), along with Gottlob Frege’s
philosophy of language as summarized in Dummett 1973 (Skinner 2012, 161). The importance of Quine is
asserted earlier in the volume: “when I read [...] that the holism espoused by Quine and Wittgenstein ‘has
had little impact on the philosophy of history’, I feel that I have lived in vain”, to which he adds that his
“[Cambridge School] colleagues such as James Tully must feel the same.” (Skinner 2012, p. 5).
11
In the same methodological volume of Visions of Politics, Skinner describes his approach as “extending
J. L. Austin’s concept of a convention and [...] relating H. P. Grice’s theory of meaning to Austin’s account
of illocutionary acts.” (Skinner 2012, 133)

12
text by “reading it over and over again”, as

We can learn more about their arguments by weighing them over and over again
than by extending our knowledge of the circumstances in which they wrote.
(Plamenatz 1963, p. x)

A Cambridge School approach, on the other hand, rejects the notion that a text can be
understood outside of the day-to-day political issues of a thinker’s time, i.e., outside of the
circumstances in which they wrote. The claim is stated in its most direct form in Skinner’s
“Meaning and Understanding in the History of Ideas”:

There simply are no perennial problems in philosophy: there are only individ-
ual answers to individual questions, and as many different questions as there
are questioners. There is in consequence simply no hope of seeking the point of
studying the history of ideas in the attempt to learn directly from the classic au-
thors by focusing on their attempted answers to supposedly timeless questions.
(Skinner 1969, p. 50)12

For instance, to take an obvious example, the Cambridge School would view Machiavelli’s
The Prince not as uninterested philosophical reflections on just rule and the structure of a
just society, but instead as aiming to do something, to accomplish some desired end—in
this case, to convince the new Medici regime to employ Machiavelli as a political advisor,
after he had lost his patronage due to the fall of the previous regime. Or, to take a slightly
more controversial example13 , Ellen Meiksins Wood’s Citizens to Lords Wood 2008 challenges
Strauss’ conception of Plato’s Republic as an uninterested, non-partisan tract on perennial
12
The statement appears in slightly modified form in Skinner 2012, p. 88: “there are no perennial problems
in philosophy. There are only individual answers to individual questions, and potentially as many different
questions as there are questioners.”.
13
“Controversial” in the sense that practitioners of the Cambridge School approach, along with Wood
herself, view the approach of this work as differing in some respects from the “orthodox” Cambridge School
approach established in the works of e.g. Skinner, Pocock, and John Dunn (specifically, in its attempt to
infuse the “purely” discursive or illocutionary analyses of e.g. Skinner with insights drawn from historical
materialism).

13
questions of the good society, arguing instead that it cannot be understood outside of its
sociopolitical context. For example, in her discussion of Plato’s later years, she emphasizes
the political aspirations behind the 385 BC establishment of his Academy. Serving as a kind
of Hellenistic proto-think tank, “the political purposes of the Academy [were] unmistakable.
Its students—the sons of wealthy Athenians and foreign families—were educated in Platonic
politics and sent forth as consultants to rulers and cities throughout the Mediterranean
world” (p. 66). Therefore, just as e.g. the content of George Mason University’s economics
seminars held for US federal judges cannot be understood outside of the political project
of influencing legal outcomes (Ash et al. 2017), the content of Plato’s writings cannot be
understood outside of the project of trying to shape the political future of the Mediterranean
world.
Taking inspiration from the insights into Plato’s thought engendered by Wood’s contex-
tual analysis, we attempt to do the same in this work by contextualizing Marx’s thought, in
both its intellectual and material environment. Unlike in the case of Plato, however, where
we only have access to the thought of a few of his key interlocutors14 , by Marx’s time lit-
eracy, communication technology, and intellectual institutions were widespread enough to
make a deep reading of all his interlocutors impossible, thus necessitating the computational
“distant reading” methods detailed in the previous section.
In fact, we emphasize the links between our study and those of the Cambridge School
because we view the methods developed herein simply as ways to overcome an inherent limi-
tation of the latter’s research program as it has been conducted thus far. The genre-defining
Cambridge School studies—Skinner’s Foundations of Modern Political Thought, Pocock’s
Machiavellian Moment, and John Dunn’s The Political Thought of John Locke—all analyze
political thinkers from before the era of mass literacy and rapid global communication net-
works, the era that Marx was born into. The subjects of these studies typically wrote with
14
Importantly, we emphasize that this was not because he only had a small number of interlocutors, just
that we only have access to a small number of them due to the selective preservation and transmission of
historical texts (Shapiro et al. 1987), an effect compounded by the lower proclivity to record systematic
thought in the form of writing in Plato’s time, relative to Marx’s.

14
a small audience—a set of key political figures they hoped to influence—in mind. The first
volume of Skinner’s Foundations of Modern Political Thought, for example, traces the de-
velopment of the mirrors-for-princes genre across the independent city-states of Italy in the
thirteenth century, emphasizing how “the ambition to supply their rulers and magistrates
with practical advice on how best to conduct themselves provides the central theme” for
these works. Given this proscribed discursive community, restricted essentially to the au-
thors alongside the few prominent rulers and magistrates of their polities for whom they
wrote, Skinner is able to track incremental changes in the genre over time by examining
how political issues in these city-states (for example, the power struggles between the Em-
pire, the Pope, and their own desire for independence) gave rise to the novel ideologies
proffered in these texts. But, as Cambridge School works move towards examining later,
post-Renaissance communities of discourse, e.g. that of the English radicals to whom Locke
wrote from his exile in Holland15 ), they are increasingly hampered by the need to study
of larger and larger—and thus increasingly intractable—collections of texts, if they hope to
gain a representative understanding of the intellectual context of key texts like Locke’s.
So, while the limited scope of the most impactful Cambridge School works is what allows
their tractable contextual analyses via deep reading, it simultaneously erects an artificial
barrier whereby texts written for mass audiences are excluded from this mode of analysis,
an exclusion with no political-theoretic or historiographic justification. Thinkers like Kant,
Hegel, and Adam Smith, for example, who shaped political thought in the midst of the
industrial revolution’s earliest upheavals, therefore form a sort of “blind spot” in Cambridge
School analyses: should their works be analyzed as discussions restricted to a small set of
political and intellectual elites (so that understanding can be achieved via deep reading and
comparison of texts with the intrigues of these political and intellectual elite), or do they
represent the point at which historians need to start taking into account the burgeoning
mass publics of the subsequent centuries?
15
See, e.g., Ashcraft 1986.

15
 We hope, therefore, to point the way towards an expanded Cambridge School methodol-
ogy which employs both deep reading and computational inference to eliminate this artificial
time-barrier, thus allowing researcher to employ the power of the Cambridge School approach
towards understanding the full development of political thought, in uninterrupted sequence,
from antiquity up to the present.
Looking towards our studies in the following chapters, for example, a study of the de-
velopment and trajectory of Marx’s thought cannot plausibly restrict itself to studying the
discursive communities of one country, one ideological tendency, one language, or one “epoch”
of 19th century history. The key elements of his critique of capitalism were forged over the
course of a decade—the 1840s—in which a whirlwind of revolutionary upheavals and gov-
ernmental expulsions swept him from Germany to France, Belgium, and England, where he
finally settled after the abortive revolutions of 1848. Historians thus generally agree that his
thought represents a mixture of German philosophy, French socialism, and British political
economy, but disagreements begin to arise when the details of this mixture are interrogated:
What particular concepts did he absorb from each, and to what extent did he modify or
transform them? Did the influence occur gradually, through e.g. his day-to-day interactions
with workers in Paris? Or can we pinpoint particular moments when his reading of certain
texts immediately affected his thought? And, while we know that analyses at this level of
granularity have been carried out to great effect for earlier time periods—as we saw above for
the case of Skinner’s study of the step-by-step progression of the mirrors-for-princes genre—
is this type of detailed understanding still possible when the discourse is carried out across
so many countries, languages, and intellectual communities?
In the eyes of Cambridge School practitioners themselves, this indeed represents un-
charted territory. In a 2009 collection of reflective essays Pocock acknowledges these linguistic
limits, conceding that “most of the work produced in Cambridge and by those associated with
it has been concerned with [...] a history of political thought largely anglophone.” (Pocock
2009, p. 140) He then draws attention to the temporal limits as well, pointing out that

16
“Skinner has not yet ventured far into the eighteenth century”, and adding, “I myself cease
from inquiry about the year 1790” (ibid., p. 141). After discussing some promising historical
studies of the post-1790 development of political thought, he nonetheless concludes that “A
Skinnerian approach to the modern and the post-modern has not yet been tried.” (ibid.) This
is precisely the project we will embark upon in later chapters, but it is a project for which
we’ll need additional hermeneutic machinery if we want to manage the increased complexity
of nineteenth- and twentieth-century discourse, relative to the pre-1790 “neo-Latin culture
in which discourse was the preserve of established clerisies operating stable and continuous
languages.” (ibid.) Indeed, Pocock’s 1790 cutoff is particularly noteworthy given our moti-
vating interest, our aim of understanding how exactly the emergence of mass publics affected
the conduct of political theorization and argument. As Carla Hesse summarizes the effect of
the French Revolution on the publishing industry, for example,

“Between 1789 and 1793, [the Revolution] swept away the monopolies that France’s
preeminent cultural elites had over the means of producing and disseminating
ideas through the printed word” (Darnton et al. 1989, p. 82)

Pocock’s and Skinner’s decisions to restrict their close-reading-based studies to the era be-
fore this radical shift in European print culture are thus understandable: while the sweeping
changes described by Hesse already portended a massive expansion of discursive communi-
ties within France16 , the subsequent spread of French norms and institutions via Napoleon’s
incursions transformed this national trend into a continent-wide phenomenon by 1815, inau-
gurating “a modern age in which the printing press, polite conversation, and public opinion
were the defining discursive institutions.” (Goodman 1996, p. 5)
Thus, before beginning our post-1790 Cambridge School investigation of this “modern
age”, we argue in the next section that computational-linguistic methods developed in recent
16
Later in the essay Hesse describes one immediate effect of this “unprecedented expansion and democ-
ratization” of discourse in the proliferation of political, intellectual, and scientific journals: “The number of
journals produced in Paris skyrocketed from 4 in 1788 to 184 in 1789, and 335 in 1790.” (Darnton et al. 1989,
p. 92)

17
years enable us for the first time to tackle this increased complexity which “progressively
aris[es] as clerisies are replaced by intelligentsias and political discourse becomes increasingly
demotic” (Pocock 2009, p. 141). By identifying connections between the principles of Cam-
bridge School hermeneutics and the inner-workings of these computational tools in Section
1.2.2, we derive a neo-Cambridge School methodology for developing verifiable, reproducible
answers to Cambridge School-motivated context-sensitive questions—given a set of assump-
tions regarding how their terms should be operationalized—that enables us to bring this
approach to bear on developments in nineteenth- and twentieth-century political thought.
Importantly, our phrasing here (the “given a set of assumptions” qualifier) is intended to
highlight a key difference between our approach and the approach of what we’ll call “vulgar”
computational social science17 , namely, that the answers we generate are fundamentally
conditional on the set of assumptions that have been encoded in the form of the algorithms’
initial conditions and parameters. In other words, while we do attempt to justify the prior
assumptions we encode in the algorithms throughout, the key contributions lie not in these
assumptions or their justifications (which we hope can simply be seen as provisional and
contestable “encodings” of already-existing arguments from the literature) but rather in
the veracity of our methods with respect to their ability to “scaffold” all sides of a given
interpretive debate with verifiable and reproducible metrics. This endeavor can be considered
successful, for example, to the extent that it allows historians of political thought to adopt
a procedure akin to Rawls’ “reflective equilibrium” (Rawls 1951), where evidence generated
via our methods can be used to increase or decrease one’s confidence in a given interpretation
of a historical text or thinker.

1.2.2 From Computational Linguistics to the Cambridge School and Back

Our application of computational-linguistic methods to questions of intellectual history

in fact has a rich history, one which was unfortunately lost in the institutional divisions-of-
17
This terminology is meant simply to denote a model of the historiographic tendencies we will aim to
avoid, not to suggest that any particular researcher or work actually adopts this “vulgar” approach.

18
labor which followed. Our aim, in fact, is simply the re-integration of both the Cambridge
School historiographic approach and modern contextual computational-linguistic methods
with their “common ancestor”, a strand of thought catalyzed by the rapid turn away from
logical positivism and towards the analysis of natural language in the field of linguistic
philosophy over the course of the 1950s.
Rooted in the late Wittgenstein’s Philosophical Investigations (Wittgenstein 1953)18 and
culminating in J. L. Austin’s How to Do Things with Words (Austin 1962), this approach
broke sharply with the theretofore dominant logical-positivist strand of linguistics. Philoso-
phers like Austin argued that statements such as “I hereby declare you husband and wife”
or “Look out!”, despite not being translatable into logical true/false statements (and thus
rejected outright as “meaningless” by logical positivists19 ), ought to be analyzed by lin-
guists nonetheless in terms of their “locutionary effects”—what their utterances “do” in the
world when spoken—rather than their internal structure or consistency with other logical
statements.
In the whirlwind of philosophical activity sparked by the insights in Austin’s work, we
argue, the work of linguists interested in computation and of those interested in literary or
cultural aspects of language became “decoupled”. Indeed, one can see in the work of e.g.
Roman Jakobson the lack of separation between these two endeavors: simultaneously a pro-
fessor of Linguistics and Czech Literature, his 1948 book on the 12th-century Russian epic
The Tale of Igor’s Campaign (Grégoire et al. 1948) and his 1960 essay “Linguistics and Po-
etics” (Jakobson 1960) stand alongside highly-technical works like Child Language, Aphasia
and Phonological Universals (Jakobson 1941) and his 1957 MIT lecture on “Linguistics and
18
Though the genesis of this “turn” is indeed neatly traced back to the late Wittgenstein in most pop-
ular narratives, precedents for Wittgenstein’s contributions did certainly exist in the works of less-often-
acknowledged linguists and linguistic philosophers such as Roman Jakobson (who was, in turn, strongly
influenced by Ferdinand de Saussure’s 1916 Course in General Linguistics (Saussure 1916), as was the Cam-
bridge School’s J. G. A. Pocock).
19
Our narrative here, for the sake of brevity, papers over the subtleties of the interplay between logical-
positivist and natural-language philosophers. The oft-cited final proposition of the early Wittgenstein’s 1921
Tractatus Logico-Philosophicus (Wittgenstein 1921), however, that “whereof one cannot speak, thereof one
must remain silent,” broadly summarizes the perspective of early logical positivists with respect to non-
logically-analyzable linguistic utterances.

19
Physics” (Jakobson 1957).
The pathway leading from the logical-positivism-counterposed work of J. L. Austin,
Jakobson, and others to modern contextual computational linguistics is easy to trace: Noam
Chomsky, for example, explicitly viewed his own work as a “continuation” of Jakobson’s tech-
nical work, and his 1955 PhD dissertation established a framework for understanding which
linguistic structures were and were not capable of unambiguous parsing by a computer20 .
That same year, after attending the Harvard lectures published as How to Do Things with
Words, Chomsky met and began regularly corresponding with Austin, who in turn discussed
Chomsky’s dissertation in a series of meetings in 1957.
The diverging pathway leading from Austin and Jakobson to e.g. Quentin Skinner is also
straightforward, as the latter’s 1968 “Meaning and Understanding in the History of Ideas”—
considered the canonical reference for the central tenets of the Cambridge School—heavily
cites Austin’s works, and Skinner explicitly mentions Austin as a (perhaps the) key influence
in interviews conducted in the years since its publication. For the purposes of our argument,
however, we focus on a 1957 publication situated directly at the intersection of these two
strands, J. R. Firth’s “A Synopsis of Linguistic Theory, 1930–1955”. In this work, essentially
providing a motto and raison d’etre for both the Cambridge School and for contextual
computational linguistics, Firth made his famous assertion quoted above, that “you shall
know a word by the company it keeps”.
Following the pathway from Firth and Austin to subsequent linguistic philosophy, we can
interpret W. V. O. Quine’s “gavagai” thought experiment21 published three years later in
his 1960 Word and Object (Quine 1960) as driving Firth’s point home by illustrating the
indispensability of context for the construction of linguistic meaning and understanding.
Quine begins by asking the reader to imagine themselves trying to learn a foreign language
20
Regular expressions, for example, allow more powerful searching within and across web pages, as these
pages are scanned not for specific letters or digits but for patterns corresponding to “regular” context-free
grammars, one of the four levels of grammatical complexity within the Chomsky Hierarchy.
21
In recent computational linguistic work, the word “wampimuk” is often used in place of “gavagai”,
following Lazaridou et al. 2014.

20
by observation. Upon seeing a rabbit run by, they observe the native speakers of the language
pointing and saying “gavagai”. Though our instinct may be to infer thereby that “gavagai”
must mean “rabbit”, in fact with only this information to go by it may just as well mean
“furry thing”, “running animal”, or myriad other possibilities. It is only by observing multiple
different contexts in which “gavagai” is and is not used to denote various objects or events
that one can essentially narrow in on, or “triangulate”, the meaning of “gavagai”.
Contextual embeddings, the primary tool used throughout our study to measure and
compare the linguistic meanings of words, phrases, sentences, and documents, are constructed
precisely to “quantify” this Firthian notion of meaning: words22 are represented by points
in a geometric space such that the distance between these points is proportional to the
likelihood that they appear in similar contexts in a corpus. Under the hood, this entails
training a neural network to predict, given a particular word randomly sampled from a
sentence, what the surrounding words in the sentence are. Thus, for example, a neural net
which predicts “Scotia” when given the word “Nova” will likely do better on this task than
one which predicts “pineapple”, ceteris paribus. The information which the trained neural
network has stored for each word is then used to construct the geometric space: since the
optimized network placed a high probability on seeing “Scotia” in the context of the word
“Nova”, the distance between the geometric points corresponding to “Nova” and “Scotia” is
smaller than that between the points corresponding to “Nova” and “pineapple”.
Drawing the parallel between this and the New York Times corpus discussed above, for
example, upon seeing the word “Clinton” the neural net would probably achieve a high
accuracy score by predicting that “president” will appear somewhere in this context. Since
the same is true of seeing the word “Putin”, therefore, the resulting geometric space will
map “Clinton” and “Putin” to points very close together, representing the fact that they are
used in similar contexts. Similarly, the points for “Clinton” and “U.S.” will be close together,
as will the points for “Putin” and “Russia”.
22
We focus on word embeddings for simplicity in this section, but note that the principle generalizes to
phrases, sentences, and documents, as we discuss in more detail in Section 1.2.3 below.

21
 The multiple types of relations thus captured in the geometric space (since the simple
fact that “X often appears in context Y ” is still purposefully ambiguous with respect to
the precise nature of the relationship between X and Y ) will in fact give rise to higher-
order geometric relationships between points than just “close together” or “far apart”. The
aforementioned points, for example, will be arranged such that one can perform “analogical
math” on the terms:

# » # » # » # »
Putin − Russia + U.S. = Bush, (1.1)
# » # » # » # »
Washington − Bush + Moscow = Putin, (1.2)
# » # » # » # »
U.S. − Iraq + Russia = Ukraine, (1.3)

and so on. Thus we can use these entity vectors to solve for vectors representing relationships
between these entities—a technique that will become central to our studies in future chapters,
for example, when we analyze the different relationships Marx and Hegel ascribe to the
(state, civil society) entity pair23 :

# » # » # »
Leader-Of = Putin − Russia, (1.4)
# » # » # »
Invaded = U.S. − Iraq, (1.5)
# » # » # »
Capital-Of = Ukraine − Kyiv. (1.6)

This means that, despite not having this knowledge explicitly encoded into the system be-
forehand, the algorithms can infer these relationships from the fact that words for cap-
itals like “Washington” and “Tehran” are often used in similar contexts, metonymically,
to refer to states and/or administrations. In just the corpus of January–June 2006 New
York Times articles, for example, the following sentence alone provides crucial information
to the algorithm regarding the country-capital relationships Capital-Of(Tehran, Iran) and
23 # »
We denote these relational vectors using a sans-serif font (Relationship) to differentiate them from the
# »
entity vectors, which are instead rendered in a fixed-width font (Entity).

22
Capital-Of(United States, Washington), as well as the existence of diplomatic interactions
between the two states:

“The United States has said that if Iran suspended uranium enrichment, Washington
would join the Europeans in direct talks with Tehran.”

In fact, even in corpora without individual sentences like these directly establishing the
relationships among entities, their relationships can be inferred via the pooling of information
across multiple sentences. By synthesizing information from sentences like the following (also
from the 2007 New York Times corpus), for example, the same relationships can be inferred:

(a) “A British official, speaking anonymously because of the delicacy of the diplomatic
exchanges between London and Tehran...”

(b) “The Bush administration said Thursday that the release of 15 British sailors and
marines held by Iran for two weeks created no new openings in dealing with Tehran...”

(c) “By deceiving the nuclear agency about its activities, President Bush and British,
French and German officials say, Iran has given up whatever treaty rights it once
enjoyed.”

(d) “Asserting Washington’s determination to protect Japan and South Korea, its principal
allies in the region, Mr. Bush said the United States ‘will meet the full range of our
deterrent and security commitments.’”

Thus, considering these sentences in order, the algorithm could construct networks of mean-
ing for each sentence as illustrated in Figure 1.1, then combine the information from these
individual networks as shown in Figure 1.2, resulting in the synthesized-information network
shown in Figure 1.3, where the capital cities for the UK, US, and Iran are situated similarly
with respect to the Bush node.
The meaningful spatial positions of the learned vectors, moreover, give rise to a type
of transitive property allowing us to infer information that is never stated in the corpus,

23
 British Bush British

“dealing with”
London
Tehran
“diplomatic exchanges”

Tehran Iran
(a) “A British official, speaking anony- (b) “The Bush administration said Thursday
mously because of the delicacy of the that the release of 15 British sailors and
diplomatic exchanges between London and marines held by Iran for two weeks created
Tehran...” no new openings in dealing with Tehran...”
Bush British French German

“treaty rights” Washington

Iran Bush United States

(c) “By deceiving the nuclear agency about
its activities, President Bush and British,
French and German officials say, Iran has
given up whatever treaty rights it once en-
joyed.”
(d) “Asserting Washington’s determination
to protect Japan and South Korea, its prin-
cipal allies in the region, Mr. Bush said the
United States ‘will meet the full range of
our deterrent and security commitments.’”

Figure 1.1: Graphical models of the relationships which context-sensitive text-analysis algo-
rithms can infer from sentences (1) through (4) above.

# »
directly or indirectly. To see this, note how in Figure 1.4 one can travel from the Russia
# »
vector to the Kyiv vector via many different pathways, for example by:

# » # » # »
(a) Following the Invaded vector to Ukraine and then following the Capital-Of vector to
# »
Kyiv,

# » # » # » # »
(b) following the Capital-Of vector to Moscow and then following the Invaded vector to Kyiv,
or

# » # » # » # » # »
(c) Following the Leader-Of vector to Putin, the Invaded vector to Zelensky, the Leader-Of
# » # » # »
vector in the opposite direction to Ukraine, and finally the Capital-Of vector to Kyiv.

We illustrate this multitude of paths in Figure 1.5, where the pathways described by (a),

24
 British British

Bush London Bush London

“diplomatic “treaty
exchanges” rights”
“dealing
with” Tehran “treaty Tehran
rights”

Iran Iran
(a) The systematic merging of the networks (b) The network formed by incorporating the
in Figures 1.1a and 1.1b. information in Figure 1.1c into the network
of panel (a).

Figure 1.2: An illustration of how, after constructing the individual-sentence context net-
works in Figure 1.1, the algorithm can systematically merge the information across sentences,
eventually producing the final corpus-wide network shown in Figure 1.3.

British

Bush London

Washington Tehran

United States Iran

Figure 1.3: The final corpus-wide contextual network, combining all information from the
individual networks in Figure 1.1 (edge labels omitted for clarity).

(b), and (c) above are colored and labeled accordingly.

# »
For example, having learned the aforementioned Capital-Of vector based on the respective
contexts in which “U.S.” and “Washington” were used in the corpus, one could then add this
vector to the vector for the entity Poland to obtain the vector for Warsaw, even if the corpus
contained no explicit statements about Warsaw being the capital of Poland. Typically this
is made possible by the algorithm’s ability to pick up on subtle contextual patterns, like
the fact that capital cities of a given country are often the most frequently-mentioned cities

25
in stories about those countries. Thus, for example, the embedding for Ulaanbaatar could
occupy a similar position relative to Mongolia as Warsaw does to Poland, even without the
former appearing together in an explicit “capital of” context in the corpus, on account more
simply of their common co-occurrence.
It is in their ability to capture meaningful semantic relationships like these—an ability
extending even to relationships never explicitly mentioned within a corpus—that we argue
for understanding word embedding algorithms as methods for inferring and studying the
“geometry of political discourse”.

1 y
# » # »
Zelensky Hussein
0.9

0.8

0.7 # » # »
Putin Bush

0.6

0.5
# »
Leader-Of
# » # »
0.4 # » Ukraine Iraq
Invaded
0.3 # » # »
Kyiv Baghdad
0.2 # » # »
Russia # » − Russia
(U.S. # ») U.S.
# »
Capital-Of
0.1 # » # »
Moscow Washington
x
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 1.4: A visualization of the “analogical math” which can be performed within word
embedding spaces, due to their ability to capture multiple types of word-word relationships
in the form of distances ranging over multiple dimensions: entity vectors are typeset in fixed-
width font (Entity), while relational vectors are typeset in serif font (Relation). The vector
from Russia to U.S., a relational vector, is typeset in fixed-width font only to illustrate its
mathematical form (i.e., that it is computed via the subtraction two entity vectors).

While we will dive into the details of these embedding algorithms more deeply in Section

26
 1 y
# »
Zelensky
0.9
(c)2
0.8

0.7 # »
Putin

0.6 (c)3

0.5
(c)1
0.4 # »
Ukraine
(b)1 (c)
(b)2 # 4 »
0.3 Kyiv

0.2 # »
Russia
(a)1 (a)2
0.1 # »
Moscow
x
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 1.5: Visualizing three of the potential pathways by which one can move from the
Russia to the Kyiv vectors.

1.2.3 below, to see how they are able to capture these multi-layered semantic relationships,
for now we conclude by re-emphasizing how understanding contextual embedding algorithms
as computational implementations of Firth’s hypothesis thus “completes the loop” from com-
putational linguistics to the Cambridge School and back: the contextual analysis advocated
by Cambridge School scholars and carried out via a deep reading of a wide array of histori-
cal texts is precisely mirrored in the processes that generate these neural embedding spaces.
And, as touched on above (a consideration which we will also discuss in more detail below),
while embeddings restricted to the level of words are perhaps too lexically fine-grained to
allow a general understanding of a text in the sense advocated by the Cambridge School,
the move from contextual word embeddings up to contextual sentence embeddings in recent
years brings us significantly closer to the textual level of analysis, and offers hope that the

27
types of methods used to aggregate information from the word to the sentence level could
also bear fruit for the further aggregation of information up to the level of full texts within
a corpus.

1.2.3 Meaning and Context

1.2.3.1 Constructing Contextual Fields

The key computational tool used throughout the studies in the remainder of this work is
the contextual sentence embedding algorithm, a method for mapping linguistic units (words,
phrases, sentences, documents) into geometric spaces based on their semantic content. In
fact, there is no single “embedding algorithm”, but rather a wide range of algorithms which
transform various linguistic entities into points within a high-dimensional geometric space.
Although our unit of analysis herein is a document (e.g., a text, pamphlet, or letter), these
documents are in general too long for a single embedding algorithm to handle – state-of-the-
art embedding algorithms like BERT have an upper limit of 512 tokens that can be jointly
encoded into a single high-dimensional vector. Thus we instead compute a separate embed-
ding for each sentence within a text using SentenceTransformers (Reimers and Gurevych
2019), then combine these sentence vectors into a single document vector via mean pooling.
As a robustness check, however, we computed document-level embeddings via an experimen-
tal document embedding method called Longformer, described in Beltagy et al. 2020, and
obtained qualitatively similar results.
Turning to the issue of how exactly the semantic information in a text is given a ge-
ometric interpretation: at the most basic level, sentence embedding algorithms take every
word appearing in a corpus and represent them as points within a geometric space, such
that words which are used in similar contexts24 will be placed closer together in the space
24
Although “context” can be operationalized in different ways based on what information a user hopes to
extract, in our case the context of a word w in a sentence S is defined to be the set of n words appearing
before and after w in the sentence. For example, if S is “The sleepy grey cat likes salmon.”, and w is
“cat”, then the context of w with n = 2 would be the set {sleepy, grey, likes, salmon}.

28
than sentences which use dissimilar words and/or dissimilar contexts around these words.
In this way, computers can begin to reason about semantic relationships between words by
analyzing the geometric properties of these constructed spaces, as in Figure 1.6.

Figure 1.6: The word-level linguistic intuitions which, by constructing an appropriate geo-
metric space, can be quantitatively measured by way of the Cosine Similarity metric.

By implementing this micro-level property—that is, the property that words or phrases
which share similar contexts should be close together in the generated geometric space—
across an entire text corpus, what emerges at the macro level is a landscape of the words
and phrases in the corpus within which distinct regions can be identified which correspond to
our intuitions about semantic meaning in human language. For example, Figure 1.7 visualizes
a cross-lingual embedding space where each node is colored based on what part of speech
it represents, demonstrating how these embedding algorithms, despite focusing solely on
micro-level (i.e., word- or phrase-level) similarity properties, naturally give rise to geometric
spaces which capture macro-level properties of human languages.
To understand how exactly this text-to-geometric-space transformation works, Figure
1.9 illustrates how the task can be represented as a mathematical problem which, when

29
Figure 1.7: An example of the higher-level properties of language which naturally emerge
when an embedding algorithm computes word-level contextual-semantic similarities (that is,
moves the points for words which occur in similar contexts closer together) across a large
corpus of text from many different languages. (From Chi et al. 2020, Figure 6)

solved, produces precisely the desired semantic-capturing and context-sensitive geometric

space. Given the full set of contextual information about each word—that is, how many
times the word appears in each possible context—the aim is to compress the information for
each word into just 3 dimensions, such that words which appear in similar contexts will be
close together when visualized in 3D space. Visualizing the 3D space generated in the figure,
for example, word 5 will be very close to word n, as word 5 will be mapped to the coordinate
(x, y, z) = (.2, .0, .3), while word n will be mapped to (x, y, z) = (.2, .1, .3), so that they differ
only by 0.1 in the y coordinate, as visualized in Figure 1.10.
To see how powerful this dimensionality-reduction approach is, let’s consider an extremely
small corpus of four “documents”, where each document is actually just a single 3-4 word

30
 law power soveraign christ · · · justice x y
   
law 2257 376 560 74 ··· 100 law 0.52 7.38
   
power  376 1071 624 80 ··· 36   0.54
power  6.04
···
   
 560
soveraign 624 983 32 38  soveraign 0.35 10.92
 
≈ × V
 74 80 32 727 ··· 0   8.03 −6.53
   
christ   christ  
..  .
 .. .. .. .. .. ..  ..  .. ..
.  . . . . . 
 . 
 . .

justice 100 36 38 0 ··· 143 justice −0.22 10.50

Figure 1.8: The mathematical operation (dimensionality reduction) by which we compress

the full set of word-context information in Hobbes’ Leviathan (the matrix on the left) into
a smaller matrix of two-dimensional vectors (the rows in the right-side matrix) which best
preserve contextual similarity. That is, similar rows in the left-side matrix will be mapped
to close-by points in 2D space, with (x, y) coordinates as given in the right-side matrix.
context M
1
2
3
4
5
6
7
context
context
context
context
context
context
context

··· x y z
word 1 2 0 0 3 0 2 7 ··· 4 .3 .1 .0
   

word 2 3 1 0 6 0 0 2 ··· 0 .1 .3 .4
   
 
   
word 3 1 3 4 2 7 2 0 ··· 9 .0 .2 .1
   
   
   
.2 .2 .1 .3 .0 .2 .7 · · · .4
word 4 7

0 1 0 3 0 7 ··· 4 
 
.0 .9 .3 
  
 ≈  × .7 .0 .1 .0 .3 .5 .7 · · · .4 
word 5 0
 2 0 4 0 0 7 ··· 0 
  .2 .0 .3  
 
    .9 .8 .6 .3 .0 .1 .7 · · · .0
word 6 0
 9 3 2 1 3 0 ··· 0 
  .5 .0 .1 

    Context{zVectors
word 7 2
 0 0 1 0 5 1 ··· 3 
  .0 .9 .0 
 | }
3×M
.. .
. .. .. .. .. .. .. .. ..  
 .. .. .. 
. . . . . . . . . . 

. . .


word N 5 0 1 3 0 0 5 ··· 3 .2 .1 .3
| Word-Context
{z Matrix } Word
| Vectors}
{z
N ×M N ×3

Figure 1.9: An example of the matrix decomposition procedure that word embedding algo-
rithms implement, to solve the problem of retaining information about word-context relation-
ships while reducing the M -dimensional representations of each word down to 3 dimensions.
(Example adapted from Kozlowski et al. 2019, Figure 1)

31
 1 z

4
2
y
5n 1
6
x
1

Figure 1.10: A plot of the 3-dimensional word representations in Figure 1.9.

sentence:

1. Green eggs are sweet.

2. Ham is sour.

3. Cats love sweet food.

4. Dogs love sour food.

Taking the set of unique words25 , and defining the context of a word to be the full sentence
in which it appears, we obtain the Word-Context Matrix shown (along with the resulting
two-dimensional embeddings) in Figure 1.11. By examining the plot of the resulting two-
dimensional embeddings in Figure 1.12, we can see that despite having a corpus of only
four sentences (providing a total of 15 tokens) to infer information from, the method is
already able to detect multiple semantic relationships: First, we see that it has captured the
identical context-employment of green and eggs, by mapping them onto the same point in
the 2D space. Second, we see that it has used the x-axis to separate two triangles of points,
25
That is, the unique words after the function words “are” and “is” have been removed. Standard practice
in word embedding models is to remove all such function words, since they pertain to the syntax rather than
the semantic content of a given sentence. In other research applications, however, the syntactic structure of
the sentences may be exactly what one is hoping to capture with these embeddings, in which case all non-
function words are removed. See Mosteller and Wallace 1964 for a seminal application of this syntax-based
analysis in the social sciences.

32
one representing the semantic relations involving cat and the other the semantic relations
involving dog.

green

sweet
cats
dogs
eggs

love
sour
ham
x y
−0.308
   
cats 1 0 0 0 0 1 0 1 0.125
 −0.271 −0.316

0 1 0 0 0 1 1 0
  
dogs 


 −0.201
0 0 1 1 0 0 0 1 0.359
   
eggs 


green 0 0 1 1 0 0 0 1 −0.201 0.359
   
≈  × |{z}
V
   
ham 0 0 0 0 1 0 1 0 −0.107 −0.234

    2×8
love 1

1 0 0 0 2 1 1 −0.580
 
−0.191

   
 −0.378 −0.550
sour 0 1 0 0 1 1 2 0 
 
sweet 1 0 1 1 0 1 0 2 −0.510 0.483
| Word-Context
{z Matrix } | Word {z
Vectors }
8×8 8×2

Figure 1.11: The Word-Context Matrix for the four-sentence Green Eggs and Ham corpus,
along with the two-dimensional embeddings computed for each word.

sweet
green, eggs 0.5

cats x

love ham
dogs
-0.5
sour

Figure 1.12: A plot of the two-dimensional vectors generated for the 8 unique words in the
Green Eggs and Ham corpus.

Since the spaces produced by these word embedding algorithms are vector spaces in a
formal mathematical sense, we can use powerful tools from linear algebra to re-orient our

33
visualization from Figure 1.12, allowing us to build our intuition about various aspectsof the
corpus’ semantic relations. For example, looking at the original plot, we can observe that the
x-axis neatly separates most of the vectors into a cat quadrant (above the x-axis) and dog
quadrant (below the x-axis), but that the arrangement is a bit skewed such that the vector
for “love” lies within the dog quadrant, despite the fact that it actually perfectly bisects the
angle formed by the “cat” and “dog” vectors (that is, the orange line in the figure passes
through the midpoint of the dashed orange line joining “cats” and “dogs”, the point labeled
in the figure via an unfilled orange circle). Thus, as researchers, we may be able to get an
even better sense of the dog vs. cat relation by rotating the points clockwise until the vector
for “love” is exactly aligned with the x-axis, as pictured in Figure 1.13.

y
sweet
green, eggs 0.5

cats
love x
ham
dogs
sour -0.5

Figure 1.13: The same points as in Figure 1.12, but rotated clockwise so that the “love”
vector is aligned with the x-axis.

By performing this rotation we thus obtain a clean separation which corresponds to our
intuition with respect to the original corpus: words pertaining to cats in particular are in
the cat quadrant (above the x-axis), those pertaining to dogs in particular are in the dog
quadrant (below the x-axis), and the vector for the word “love”, since it is neutral with respect
to the cats-dogs split (i.e., since both cats and dogs are described as loving something), is
now located exactly on the boundary point between the quadrants. Although the pattern
which emerges could have been seen (via a tilt of the head) in the original plot, this ability

34
to re-orient the embedding space visualizations as needed becomes crucial for more complex
corpora.
As another example of a useful re-orientation, consider the case of a researcher who
studies the rotated plot in Figure 1.13, identifies the dog quadrant vs. cat quadrant pattern,
and wants to explore this axis further. By employing another linear-algebraic transformation
we can rotate and rescale the points so that the cats-to-dogs vector itself serves as our x-
axis, allowing us to examine where the remaining points fall on this cats-to-dogs spectrum,
as pictured in Figure 1.14. With this transformation, two salient splits now become evident:
first, the cat and dog regions from before are now split (by construction) by the y-axis, with
the cat region to its left and the dog region to its right. A second, new split can now be
seen as well: that between food words (above the newly-constructed x-axis) and taste words
(below the x-axis). The utility of this transformation is immediately apparent: with the four-
quadrant separation resulting from the combination of these two splits (that is, starting from
the upper-right quadrant and moving clockwise: dog-food words, dog-taste words, cat-taste
words, and cat-food words), one could analyze a much more complex corpus by seeing how it
“sorts” the words across the corpus into these categories. And indeed, computational linguist
Dan Jurafsky performed just this sort of analysis on a much larger corpus—a collection of
culinary descriptions from restaurant menus around the globe—as the basis for his award-
winning book The Language of Food: A Linguist Reads the Menu (Jurafsky 2014).
A key lesson can be drawn from this this example for the sake of studying political
theory26 , namely, that the effectiveness of these embedding algorithms for capturing semantic
relations in a text depends upon how systematically the text employs key terms (i.e., whatever
terms the researcher aims to study). That is, the algorithm’s success in inferring meaningful
semantic relations from just the four sentences in our constructed corpus was aided by the
fact that the sentences straightforwardly delineated these semantic relations: each sentence
26
Here we focus on political theory in general (i.e., viewed synchronically) rather than the history of
political thought (viewed diachronically) since a genuinely historical study will require a model of how these
embeddings evolve through time, which we indeed develop starting in Section 1.2.4 below.

35
 y

0.5

green, eggs ham

x
cats
dogs
sweet
sour
love
-0.5

Figure 1.14: The same points as in Figures 1.12 and 1.13, but this time rotated and rescaled
so that the vector from “cats” to “dogs” now serves as our x-axis.

contained a single subject-verb-object triad, with the subjects and objects shared across
sentences, enabling the straightforward construction of the network structures visualized in
Figures 1.12 through 1.14.
With this in mind, then, the most direct analogue of our hypothetical example within
political theory would thus be a political-theoretic text which similarly aims to system-
atically discuss key terms and their interrelationships. The obvious candidate in Western
political thought would be Hobbes’ Leviathan, a text so self-consciously systematic that
even a champion of contextualism like Skinner agrees with his Straussian critics in deeming
it “probably the most plausible candidate” for “a classic text which ‘makes eminent sense on
its own’” (Tully 1988, p. 104, quoting Parekh and Berki 1973, p. 173)27 . Leviathan stands
out from other texts, that is, in Hobbes’ attempt to construct an entirely self-contained
and internally-cohesive system of political terms, making minimal reference (polemical or
otherwise) to other texts28 .
27
The strength of the claim can be seen even more clearly by considering the full passage from which the
quoted snippet is taken: “unlike [Paine’s Agrarian Justice or Burke’s Reflections], Leviathan makes eminent
sense on its own, [...] circumscribed but not limited by its own historical conditions.” (Parekh and Berki
1973, p. 173)
28
Hobbes’ project in Leviathan thus stands in stark contrast to e.g. Marx’s in Critique of Hegel’s ‘Philos-
ophy of Right’ (discussed in the next chapter), which consists entirely of explicit references to excerpts from
Hegel, or his Herr Vogt (discussed in Chapter 3), which is infused with dozens of implicit references in the
form of puns and ironic statements which are unintelligible without knowledge of Vogt and his background.

36
 Thus, in Figure 1.15, we present a two-dimensional visualization of an embedding space
constructed from the original English text of Leviathan 29 . While perhaps not as cleanly-
structured as the manually-constructed Green Eggs and Ham example, we do observe the
algorithm distinguishing important clusters of words: key terms constituting the religious
aspects of the text are seen in the bottom-right corner (“christ”, “holy”, “lord”), while a set
of relevant vocabularic dyads can be seen throughout the space (“obey” and “fear”, “civill”
and “law”, “common-wealth” and “soveraign”, “authority” and “power”). The algorithm’s
formation of these clusters is already encouraging, and perhaps surprising, given the fact
that the standard applications of these word embedding algorithms in Computer Science
typically train them on enormous datasets with billions of sentences30 . To understand how
such a generated space could be used as a tool for historical, sociological, economic, or other
social-scientific research, we now turn to a review of existing works which have made effective
use of embeddings to study social-scientific phenomena. We emphasize throughout, however,
some key limitations of these works which our approach—by utilizing a newer generation of
explicitly context-sensitive embedding algorithms—is able to overcome.
Since the publication of the first widely-used word embedding algorithm, word2vec
(Mikolov et al. 2013b), in 2012, researchers in the social sciences have used these algo-
rithms to incorporate information from textual corpora into studies which previously were
restricted to using numeric or qualitative data. Recent studies have found, for example, that
contextual embeddings are able to capture salient properties of social class (Kozlowski et al.
2019), the ideology of political manifestos (Rheault and Cochrane 2020), and the influence
of economics on legal decisions (Ash et al. 2017). Of these three, the latter comes closest
29
An interesting comparative study—outside the scope of our demonstrative example, but pertaining
directly to some of Skinner’s arguments in Skinner 2008—could be performed by training an embedding
algorithm on both the original English and the later Latin edition of the text, to test whether any new
vocabularic innovations were introduced by way of the latter.
30
Google’s LaBSE model for example, mentioned above, is trained on the CommonCrawl corpus, which
contains over 6.4 PB (PB stands for “petabytes”, with one petabyte being equal to 1,000 gigabytes (GB))
of data. The text file for Leviathan is about 1 kilobye (KB), meaning that the algorithm was able to detect
these types of semantic relations even in a corpus 6,400,000,000,000 (6.4 trillion) times smaller than the
standard CommonCrawl corpus used to train this algorithm in industry settings.

37
 Figure 1.15: A word embedding space constructed from the text of Hobbes’ Leviathan.

to our work, in attempting to study the linguistic properties captured by word embeddings
using econometric methods for estimation of time-series effects.
A related literature, which predates the creation of word embedding algorithms, aims
to quantitatively capture the existence, direction, and magnitude of ideological influence
directly. Barron et al. 2018, for example, studies ideological influence across a time series of
French Revolutionary debate transcripts by introducing “transience” and “novelty” metrics,
which quantify how much the content of a given text is adopted by future texts, and how
much it differs from the content of earlier texts, respectively. Unlike the previously-mentioned
studies, however, this literature has yet to explore the new possibilities opened up by the
development of semantic embedding algorithms, instead opting for the more well-established
approach of probabilistic topic modelling (described here in Appendix 1.B). Although these
topic modelling algorithms are an effective tool for summarizing a corpus at a high level, they
are ill-suited for the task of tracing out the trajectory of particular terms or concepts over
time. Since we are interested in how Marx was able to cement his particular set of terms as

38
the vocabulary for later socialist discourse, embeddings are uniquely effective in allowing us
to look at exactly which contexts a given term was employed in by Marx, and how this differs
from the term’s typical contexts before and after Marx’s intervention. In other words, while
topic-modelling-based approaches can tell us that Marx’s writings influenced future socialist
discourse, embedding-based approaches can tell us how this influence operated – which terms
or concepts Marx utilized in particularly novel ways, and which of these illocutionary moves
were and were not effective in terms of influencing subsequent socialist thought.
Studies across both of these literatures, moreover, have yet to adopt the contextual sen-
tence embeddings we use herein, which utilize a more recently-developed embedding method
from 2019 known as BERT (Devlin et al. 2019). BERT, in essence, differs from previous embed-
ding algorithms in that it generates joint numeric representations for words and the contexts
in which they appear, rather than associating each word with a single embedding vector. This
means that, e.g., the word “bank” within the sentence “I took my money to the bank.”
would be given a different embedding from the same word within the sentence “I took a
nap on the bank of the river.”. With these context-differentiated vectors, then, we have
a powerful computational tool for grappling with a key difficulty in linguistically-informed
historical studies: the fact that “words denote and are known to denote different things at
the same time.” (Pocock 1985, p. 30)
As visualized in Figure 1.16, this approach to polysemy changes the unit of analysis in our
model: we are no longer modeling words themselves, but rather word-context pairs. In this
way, the context is explicitly integrated into the numerical representations we use to under-
stand an author’s use of language, providing researchers with a quantitative implementation
of precisely the type of Cambridge School-style contextual approach described above.
This improvement upon the original set of word embedding algorithms is therefore crucial,
we contend, for capturing the nuanced uses of language which occur frequently in political-
ideological polemics—the same concern which originally motivated the Cambridge School
shift from what an author is saying to what they are doing with their words. Marx’s 1860

39
 1 y
withdrawal2 river
bank2 stream1
0.8
addiction drugs swim

0.6

0.4
bank1 stream2
money teller
0.2
withdrawal1 revenue music

x
0.2 0.4 0.6 0.8 1

Figure 1.16: A visualization of BERT’s ability to compute numeric representations of the same
word in multiple different contexts.

polemic against Karl Vogt (Marx 1860), for example, makes use of several puns and pur-
poseful misspellings of the names of those he is attacking, thus evoking contexts which the
reader would not otherwise have read into the “standard” usage of the words. For example,
he calls a particular political opponent with the surname Ranigel “Ran-Igel”, likening him
to an “Igel”, the German word for hedgehog—we would therefore want to keep Marx’s use of
“Igel” in this sense separate from uses of “Igel” in general German texts. In general, due to
the harsh censorship of political writings under the regimes of Friedrich Wilhelm III and IV
(1797–1840 and 1840–1861, respectively)31 , pre-BERT text-analysis methods which are unable
to capture the wide variety of ironic, metaphorical, and figurative uses of words would thus
be unable to capture many important political illocutions throughout 19th century German
political discourse, which had to be performed almost exclusively by way of these figures of
speech.
31
See Rose 1978 for a detailed treatment of the effects of this censorship on Marx’s writings, and Prawer
1976 for a chronological examination of the explicit and implicit literary references in Marx’s writings.

40
1.2.3.2 Visualizing Contextual Fields

While the embedding spaces we’ve seen thus far have, for expository purposes, been
constructed as two- or three-dimensional spaces, these toy models are insufficient in practice
for analyzing the semantic content of a corpus. The rich variety of meanings which a set of
words can be infused with—even within a single text, much less across a collection of texts
in dialogue with one another—cannot be arranged neatly along two or three axes. For this
reason, even the most bare-bones embedding algorithms typically aim to embed words and
phrases within an extremely high-dimensional space: 300 dimensions in the original word2vec
algorithm, for example, or 768 in the case of BERT.
The tradeoffs between lower- and higher-dimensional spaces can be seen by comparing
the left and right sides of the decomposition equation shown in Figure 1.9. Each value in
the right-side matrix must be estimated from the corpus information encoded in the left-side
matrix. In the terminology of statistics, each additional term that must be estimated “costs”
an additional degree of freedom, which can be thought of as a measure of the “richness” of
the information provided by the corpus. If the word “alienation” only appears in the corpus
once or twice, for example, it will be difficult to generate a meaningful high-dimensional
vector for this term, as this vector’s semantic content is dependent upon (in fact, under the
Firthian hypothesis, is defined by) the range of contexts in which the word is seen. In Figure
1.17, for example, we can see that word 4 appears only one time in the corpus, in a single
context (context 5) in which no other words appear. In this case, therefore, BERT has no
information about this word’s relationship to any other word in the corpus, and thus no way
to infer where its 768-dimensional vector (the fourth row in the Word Vectors matrix on
the right-hand side of the equation, containing values u4,1 through u4,768 ) should be placed
relative to the other word vectors in the generated space. In cases like these, we say that the
dataset does not have a sufficient level of statistical power for estimation of embeddings via
this method32 .
32
We note that this term is used in relation to the method since, for example, the Word-Context matrix

41
 context 1
context 2
context 3
context 4
context 5

dim 768
dim 1

dim 2

dim 3

dim 4

dim 5

dim 6
···
0 7 1 2 0 u1,1 u1,2 u1,3 u1,4 u1,5 u1,6 ··· u1,768
 
word 1
 

word 2 0 7 3 6 0 u u2,2 u2,3 u2,4 u2,5 u2,6 ··· u2,768 

   
   2,1 
word 3 0 6 4 0 0 u
 ≈  3,1
u3,2 u3,3 u3,4 u3,5 u3,6 ··· u3,768 
 × |{z}
V
   

word 4 0 0 0 0 1  u4,1 u4,2 u4,3 u4,4 u4,5 u4,6 ··· u4,768 
    768×5
   
word 5 7 4 1 0 0 u5,1 u5,2 u5,3 u5,4 u5,5 u5,6 ··· u5,768
Word-Context | Word Vectors
{z }
| Matrix
{z } 5×768 = 3,840
5×5 = 25

Figure 1.17: A case where the information about the corpus (the Word-Context Matrix on
the left side) does not have sufficient statistical power to allow estimation of the word and
context vectors (the matrices on the right side). At a high level, the algorithm is unlikely to
be able to draw meaningful estimates for 7,680 parameters—the 3,840 entries in the word
vectors matrix (the u parameters), plus another 3,840 entries (not labeled) in the context
vectors matrix V —using only 25 input data points. Looking more closely at the structure of
the input data, we also see that it has no information from which to infer word 4’s context-
sensitive relation to other words, as it appears in only one context which is not shared by
any other word in the corpus.

For intuition regarding this tradeoff in language modeling, consider Quine’s gavagai
thought experiment described above. If we have only one or two observations of a foreign-
language speaker saying gavagai while pointing at a running rabbit, we won’t have a rich
enough set of contexts for this term to be able to differentiate whether, e.g., it refers to the
rabbit itself, to the rabbit’s fur, to a running animal in general, to creatures with tails, the
act of running itself, etc. With the few linguistic “degrees of freedom” available to us at
this point, we can only at best infer some vague properties of the term—perhaps that it has
to do with motion more so than stillness33 . As we observe the term being used more and
in Figure 1.17 could have a sufficient level of statistical power for estimation of much smaller (say, two- or
three-dimensional) word vectors.
33
Stretching the metaphor a bit for clarity: if our goal was to generate a mental word embedding space
for the new language, after these few observations we could (at most) conservatively move the gavagai point
slightly closer to the point for motion (or more accurately, given the inner workings of word embedding
algorithms, the point(s) for terms which our model hypothesizes as highly likely to represent or relate to the

42
more across a wider and wider range of contexts, however, we start accumulating enough
“linguistic degrees of freedom” to allow us to more and more confidently and precisely map
gavagai within the broader linguistic field34 of the foreign language.
Once we have accumulated enough observations from the corpus to feel confident in
our estimated embedding vectors, however, a second key question arises: how can we use
this high-dimensional embedding space to draw meaningful inferences about the language
of different authors or across different timespans within our corpus? Turning to the high-
dimensional vectors produced by the embedding algorithm, we may ask, what good are
(e.g.) points in a 768-dimensional geometric space when humans can only visualize points
in at most two or three dimensions? While we do have tools like the similarity metrics
discussed in Section 1.2 above, in a sense these only provide hints as to the full landscape
of the embedding space35 . We can do better using a set of visualization algorithms created
specifically to construct 2D and 3D representations of high-dimensional spaces.
The t-distributed Stochastic Neighbor Embedding (t-SNE) algorithm, for example, con-
structs a 2D (or 3D) plot with the goal of capturing the “neighborhood structures” present
within a higher-dimensional space, identifying distinct clusters of points which are closer to
one another than to the other points in the space. For intuition, one can imagine taking a
point in 3D space α and asking, if I construct a small sphere centered at this point and
grow it larger and larger, how soon will it come into contact with another point in the 3D
space β ? The t-SNE algorithm constructs corresponding points α◦ and β ◦ in 2D space such
that, if I construct a small circle centered at α◦ and grow it larger and larger, a similar
property will hold with respect to how long it takes for that circle to come into contact with
concept of motion).
34
In Section 1.2.4 below we will distinguish more precisely between linguistic utterances—the particular
instances where we’ve observed gavagai being used—and the broader linguistic field from which these utter-
ances were constructed (the Saussurean parole and langue, respectively). For now it suffices to understand
“linguistic field” as a working model of the foreign language which is built using data in the form of utterances
of gavagai.
35
To see this, consider e.g. the task of trying to reconstruct a subway map given only the distances between
stations. As illustrated for the Swiss rail system in Dokmanic et al. 2015, even sophisticated algorithms can
only produce an approximation of the underlying space (this process is nonetheless crucial for e.g. inferring
3D protein structures; see Vendruscolo et al. 1997)

43
β ◦ . In this way, nearby points in N -dimensional space remain nearby in the two-dimensional
visualization.
A separate visualization tool called Principal Component Analysis (PCA) also constructs
low-dimensional representations of high-dimensional spaces but, unlike t-SNE, focuses on
capturing the variance among points rather than their neighborhood structure. To this end
PCA can be thought of simply as projecting all the points in the space down to two dimen-
sions, from every possible angle, then choosing the most interesting of these projections with
respect to how spread out the points are. Thus, taking the scatterplot of points in Figure
1.18 as an example, we can see in the figure what these points in 2D look like when projected
down into one dimension, whether onto the x-axis (the short lines along the horizontal axis
at the bottom) or the y-axis (the short lines along the vertical axis on the left).

Figure 1.18: 100 randomly-generated points in 2D space, showing potential projections of

these points down into one dimension on both the x and y axes (the short vertical and
horizontal lines, respectively).

44
 However, this figure also illustrates a potential drawback of these “standard” projections
onto the x-axis or y-axis, namely, that they may not in fact be the most interesting projections
with respect to capturing the directions along which the points vary most. If the points
represented word embeddings learned from the WPA Slave Narrative Collection (Yetman
1967), for example, the reason for these points’ diagonal arrangement is likely to be far
more interesting to researchers than the points’ original (x, y) coordinates. Perhaps points
closer to the bottom-left of the plot are those used in the context of describing masters, and
those closer to the top-right are those used when describing fellow slaves, for example, while
points closer to the top-left corner are more likely to refer to women than those closer to the
bottom-right corner, regardless of their placement on the main diagonal.
The PCA algorithm captures this intuition, that the directions of highest variance may
be most salient to the researcher, by tossing out the original (x, y) coordinate system and
replacing it with a coordinate system based on Principal Components: the First Principal
Component axis is the axis along which the data varies the most, while the Second Principal
Component axis is the direction along which the data varies second-most. This process can
be continued, in fact (e.g., to generate a 3D PCA plot), up to the N dimensions of the original
set of points, revealing its power as both a dimensionality-reduction and data-reorientation
tool. In our working WPA Slave Narrative Collection example, for instance, our PCA plot
has the same dimensionality (2D) as the original plot, but re-orients the points so that the
x-axis is the axis of maximum variance and the y-axis that of second-highest variance, as
illustrated in Figure 1.19.
To see the usefulness of this method in practice, we start by computing two-dimensional
word embeddings for terms in the WPA Slave Narratives corpus, obtaining the space illus-
trated in Figure 1.20. And indeed we see the type of non-(x, y)-oriented pattern that we
hypothesized earlier: isolating key terms for race (“white” and “black”) and status within
the slave system (“slave” and “master”), we can observe that a move from bottom-right to
top-left corresponds to a move from “black” to “slave”, and that a nearly identical move in

45
Figure 1.19: The same data as plotted in Figure 1.18, but transformed via PCA so that the
horizontal axis represents the direction of greatest variance (the First Principal Component)
while the vertical axis represents the direction of second-greatest variance (the Second Prin-
cipal Component).

46
the same direction brings us from “white” to “master”. With this observation in mind, then,
we can choose our own horizontal and vertical axes and then analyze other terms with respect
to this newly-meaningful left-right or up-down orientation. For example, we generate Figure
1.21 with the same data but with the horizontal axis now representing the white-to-black
orientation we observed in the original plot.

Figure 1.20: A two-dimensional embedding space trained on the WPA Slave Narratives cor-
pus, where we draw arrows connecting points for key racial terms (“white” and “black”) and
for terms denoting status within the slave system (“slave” and “master”)
.

Interestingly, after performing this translation, we can observe a second emergent feature
of the word vectors’ orientation: although in the original plot the move from “woman” to
“man” seemed to be broadly similar to the move from “white” to “master” or “black” to
“slave”, once we differentiate the latter pairs by choosing them to be the horizontal and
vertical axes (respectively), we can see that in fact the “woman” to “man” direction is
diagonal with respect to this new orientation—as we observed in our theoretical example

47
 WPA Slave Narrative Embeddings: Black-White vs. Slave-Master Axes
0.20

0.15

0.10
← slave | master →

freedom
0.05 man money

0.00

woman
-0.05

-0.10

-0.15 slavery
cotton

-0.20
-0.30 -0.25 -0.20 -0.15 -0.10 -0.05 0.00 0.05 0.10 0.15 0.20
← black | white →

Figure 1.21: The same points displayed in Figure 1.20, but transformed so that the horizontal
axis represents the black-to-white direction within the space, and the vertical represents the
slave-to-master direction.

above, this indicates that perhaps our analysis could benefit even more from setting the
vertical axis to represent gender, as illustrated in Figure 1.22.
The arbitrariness of the (x, y) coordinates, and thus the impetus for employing PCA
as one possibility for deriving meaningful coordinates instead, is important enough to re-
emphasize here: since the word embedding algorithms don’t know the particular types of
meaning we hope to capture, the coordinates generated for each word have no a priori
human-interpretable meaning at all; they are chosen solely on the basis of placing words
closer together or further apart based on the similarity or difference between the contexts
they appear in. In a 2D space, for example, if gavagai’s initial embedding (typically randomly-

48
 Race vs. Gender in the WPA Slave Narratives Collection
0.20

0.15
money
0.10 land
master
overseer
← woman | man →

0.05
freedom
slave
0.00

-0.05 cotton

-0.10
child cook
-0.15
care

-0.20
clean
-0.25
-0.30 -0.25 -0.20 -0.15 -0.10 -0.05 0.00 0.05 0.10
← black | white →

Figure 1.22: The same embedding space as in Figures 1.20 and 1.21, with the vertical axis
now transformed to represent the man-to-woman direction, and with additional key terms
plotted to enable analysis of their interrelationships with respect to these two sociologically-
meaningful dimensions.

chosen) lies directly below that of bird and directly above that of fish, but the algorithm
observes gavagai being used in a context which shows it is neither of these animals, the
only way to increase its distance from both points is to move gavagai’s embedding left or
right, i.e., to modify its y coordinate. Since the same logic holds in cases where the three
points randomly start out e.g. in a horizontal line (gavagai’s embedding would have to be
moved up or down), or on the x-y diagonal (its embedding would have to be moved along
the off-diagonal), we can see that the coordinates themselves have no stable or human-
interpretable meaning on their own. It is up to the researcher to choose—on the basis of

49
their research goals and/or intuitions regarding their field of study—how to transform the
data to make sense of the embedding algorithm’s outputs. Thus we have another example of
how, as emphasized above, these algorithms do not automate social-scientific research, but
do provide a powerful addition to the researcher’s toolbox, to be used in conjunction with
their own domain knowledge and expertise.
There is an unexplored dimension of the WPA embedding examples thus far, however,
which may be crucial for a researcher’s understanding of the corpus, especially in the social
sciences and humanities. Namely: the spaces we’ve generated to this point have produced a
single point for each word, by averaging its uses in various contexts over all the interview
subjects, when in reality what social science researchers often care about is precisely how the
uses of these words differ between different individuals or groups. If a researcher is analyzing
the WPA Narratives with the goal of understanding 19th and early 20th century notions
of freedom in the U.S., for example, it would be crucial to differentiate between how the
term was used by slaves themselves versus how it was used by masters or overseers, just as
historians have analyzed how (e.g.) John C. Calhoun and Frederick Douglass infused terms
like “liberty” and “freedom” with very different semantic contents.
To this end, in the next section we describe a set of algorithms developed specifically
to model not only the different senses in which a word is used (which we saw above in our
explanation of the BERT algorithm), but also how different authors emphasize or de-emphasize
these senses. This brings us one step closer to our goal of understanding the history of political
thought as a history of illocutionary moves since, as we argue, these author-specific emphases
and de-emphases can be seen as textual manifestations of their attempts to steer discourse
in a certain direction: to steer popular understandings of “liberty” towards what we would
today call “negative liberty” (freedom as non-interference) in Hobbes’ case, for example, or to
steer socialist discourse away from utopian ideals and towards political-economic categories,
in Marx’s.

50
1.2.3.3 Author-Specific Embedding Spaces

As we gestured towards at the end of the previous section, in order to further pinpoint
the pathways through which political thought has evolved we’ll need a model which can keep
track not only of the different senses in which key terms are used, but also of the particular
individuals who use these terms in certain ways which shift their usage going forward. A
recently-developed extension of the original BERT algorithm (Welch et al. 2020) provides us
with exactly the tool we need in this case, in explicitly constructing an additional personalized
embedding for each (word, author) pairing on top of the main contextual embedding for
each word irrespective of author (i.e., relative to the entire corpus). This enables us to
capture not only the shared terms and figures of speech available to all authors within a
discursive community, but also the particular ways in which they are employed—or wielded—
by individual authors. Indeed, if one accepts (as we do) Skinner’s contention that historians
ought to treat these terms and figures of speech “less as statements about the world than
as tools and weapons of ideological debate” (Skinner 2012, p. 177), these author-specific
embeddings provide a crucial tool for explicitly modeling how particular word-weapons were
wielded by particular actors in particular wars of ideas, such as the battles for hegemony
over socialist discourse that we analyze in future chapters.
Importantly, this method does not generate a separate embedding space for every author,
since our entire goal is to be able to compare different authors’ vectors on some common
#» for
scale36 . Instead, in the author-specific BERT estimation, each author’s specific vector w i

a given term w is estimated as an offset relative to the vector for w in the MAIN vector
# ». Given a vector w
space, w # » representing the centroid of political-economic discourse
MAIN PE

within the broader ideological vector space (estimated via a procedure we detail in the next
section), for example, this allows us to instantly check whether an author A tends to use a
term w in a more political-economic context than some other author B, by checking whether
36
i.e., for authors A, B, and C, the distance dist(w# », w
# ») between A’s vector for some word w and B’s
A B
vector for w must be on the same scale as the distance dist(w # », w
# ») between A’s vector for w and C’s vector
A C
# » # »
for w, as well as the distance dist(wB , wC ) between B’s vector for w and C’s vector for w.

51
 1 y Non-Domination

libertyHarrington
0.8
arbitrary slavery

libertyPettit
0.6

ty No
berautonomy n-I
Li absence nte
i v e rfer
0.4 osi t enc
P e
libertyArendt libertyConstant
participation interference
0.2
libertyPlato libertyHobbes

x
0.2 0.4 0.6 0.8 1

Figure 1.23: A visualization of BERT’s ability to model the different contexts (and thus, senses)
in which words are used by different authors. In this case interference falls squarely within
the negative liberty (liberty as non-interference) cluster, as a term central to distinguishing
negative from other forms of liberty, whereas absence falls between two clusters as it can be
employed in both republican (“absence of arbitrary power”) and negative-liberty (“absence
of interference”) contexts.

# », w
dist(w # ») < dist(w
# », w
# »), or relative to the “average” usage of the term across the
A PE B PE

# », w
entire corpus, by checking whether dist(w # ») < dist(w
# », w# »).
A PE MAIN PE

Thus, while studies like Kozlowski et al. 2019 are able to discover shifts in the overall
discourse around social class during the 20th century, for example that different education-
related terms tended to become more dichotomous along the upper-class/lower-class axis, this
modification of BERT would allow them to identify which authors in particular were ahead of
the curve, adopting the newer senses of education-related terms earlier than others. Although
a fully-developed methodology for testing the causal impact of these early-adopters on the
broader discourse is outside the scope of this paper, we are still therefore able to identify a
necessary condition for causal influence—temporal precedence—and thus develop plausible
causal hypotheses based on which authors’ embeddings precede the overall embeddings in

52
terms of the movement of key terms along socially meaningful axes.
To this end, especially when we turn to diachronic embeddings below, we keep in mind
the notion of Granger Causality between two time-evolving data series Xt and Yt : if the
values of X up to some time t help predict future values of Y (Yt+1 , Yt+2 , . . .) better than
one could predict them solely on the basis of past values of Y , X is said to exert Granger
causality on Y . Thus, for example, if we consider X≤1848 to be data representing Marx’s
pre-1848 speech acts, Y≤1848 to be data representing language use by socialists besides Marx
before 1848, and Y>1848 to be data representing language use by socialists besides Marx after
1848, we can say that Marx’s speech acts exerted Granger causality on socialist discourse
to the extent that knowing X≤1848 helps us predict the post-1848 trajectory of socialist
discourse better than we could using Y≤1848 alone (i.e., pre-1848 socialist discourse excluding
Marx). Abstracting away from the details, then, when historians of political thought make
claims like “one cannot understand the trajectory of post-1848 socialist thought without
understanding Marx [or Hegel, or Babeuf, etc.]”37 they are (perhaps unknowingly) making
claims of Granger causality38 .
While testing for this type of causality using these author-specific embeddings is therefore
possible in theory, in practice the problem of moving from associational to causal analysis
is typically overshadowed by an even more fundamental problem, related to our discus-
sion of degrees of freedom in Section 1.2.3.2 above. While in that section we discussed how
the researcher must have enough text—with words ranging over a rich enough variety of
contexts—to allow estimation of all N entries within each word’s N -dimensional vector, the
problem becomes even more acute when estimating author-specific spaces. Now, if there are
M authors in the corpus, the researcher must have enough text from each author to estimate
37
Sholomo Avineri’s The Social and Political Thought of Karl Marx, for example (which we examine in
detail in Chapter 2), introduces its discussion of August von Cieszkowski with the claim that “one cannot
fully grasp Marx without recourse to Cieszkowski.” (Avineri 1968, p. 125)
38
On the applicability of this concept in the social sciences see the short entry in Lewis-Beck et al. 2003,
p. 439, or the more in-depth discussion in Hlaváčková-Schindler 2012. Within the humanities as such we
have been able to locate only one work, Wevers et al. 2020, which explicitly focuses on this conception of
causality.

53
all N ×M entries across the M vectors generated for each word, plus the additional N entries
in the N -dimensional “main” word vector. Methodologically, this is why the algorithm’s de-
velopers chose to estimate the single wMAIN vector for each word, and then learn each author’s
specific word embeddings wa as an offset of this MAIN vector: this allows the algorithm to
first pool all of the scarce information about w across the entire corpus towards estimating
wMAIN , then to estimate the author-specific embedding wa on the basis of how many degrees
of freedom remained (with respect to the word w) to “spend” on this additional estimation.
In practice this means that researchers must take care when interpreting cases where
an author’s embedding wa for some key word w seems not to differ much from wMAIN . While
ideally—that is, given enough textual data from each author—this indicates that the author’s
usage of w indeed did not differ substantially from the general usage of w across the corpus,
it could also mean simply that the algorithm did not have enough data on this author to
evaluate their usage, and thus “guessed” their usage to be at or near the corpus-average
usage, for lack of any better way to choose. This latter case corresponds to the heuristic
where, for example, if we only know that a man is Dutch, our best guess as to his height
would be just the average height of Dutch men (182.5 cm, or about 6 ft.).
This analogy also points the way towards a potential modification of the approach, how-
ever, which we use to minimize the chances that this collapse-to-the-average affects our re-
sults. Within the analogy, note that if we could obtain additional salient information about
this man, such as his membership within a subset of the Dutch population, we could narrow
the group over which we’re taking the heuristic average. For example, if we learned that he
was on the Dutch Olympic basketball team, we could do much better by guessing his height
to be the average height of this team’s members (about 6 ft. 7 in.), rather than that of the
Dutch population as a whole. Thus in our case, to address this collapsing-to-the-average
issue for authors with a small number of extant texts, we instead group individual authors
into “meta-authors” based on the discursive community they are generally associated with
in the historical literature—chosen to be as specific and/or exclusive as possible to minimize

54
the distortions introduced by algorithmic averaging39 . Thus, for example, the texts of Bruno
and Edgar Bauer, Arnold Ruge, Max Stirner, etc., are combined into one Young Hegelian
# » representing the centroid of Young Hegelian
meta-author in order to estimate a vector w YH

discourse (as defined by this author-to-group mapping) within the broader ideological space
of 19th-century German discourse. Importantly, however, this approach is not used to gener-
ate the political-economic and Hegelian vectors which serve as our orthogonal basis vectors,
for reasons we describe in the next section.

1.2.3.4 Discursive Fields as Embedding Clusters

# » and w
In the previous section we introduced two special vectors w # », representing
PE H

the centroids of political-economic and Hegelian discourse respectively, which we do not

compute via author-specific embedding estimation. Instead, to minimize the dependence (in
#
the statistical sense) between our two basis vectors and the vectors like w »
M arx for which

we want to observe movement over time, we compute these basis vectors as centroids of
word clusters which are derived independently via the cTFIDF measure40 , which generates a
ranking of all terms in the corpus on the basis of how “unique” they are to political-economic
texts relative to Hegelian texts (and vice-versa, by taking the N terms with lowest, rather
than greatest, cTFIDF scores).
The intuition we aim to capture with this measure is that a word wi is more “unique”
39
Considering the Dutch basketball player example again, this matches our intuition that the average
height of the Olympic team would provide a better guess than, for example, the average height of all Dutch
basketball players or athletes in general, as the former represents a much smaller height-related subset of
the population.
40
cTFIDF stands for the Class-Based Text-Frequency minus Inverse Document Frequency measure, formally
defined via the formula
 
N
cTFIDF(wi , cj ) = cf(wi , cj ) · log 1 + ,
f(wi )

where wi represents the ith word in the corpus, cj represents the jth class (grouping of documents), and N
represents the average number of words per class. cf(wi , cj ) measures the number of times word wi appears
across class cj ’s documents, while f(wi ) measures the number of times wi appears across the entire corpus.
In this section we provide intuition for how this formula is able to capture the uniqueness of a word wi to a
class of documents cj , relative to all other classes (which we collectively denote as c−j ).

55
to a class of documents cj to the extent that it appears frequently within cj ’s documents,
but infrequently within the documents of other classes. In the corpus of nineteenth-century
socialist texts we use in later chapters, for example, we may observe the German word das
appearing frequently in Marx’s writings, but it is not unique to Marx’s writings, since it
also appears frequently in the writings of the other authors in the corpus. In this case,
das has a high class frequency value relative to the class cMarx representing the collection
of Marx’s writings (denoted cf(das, cMarx )), but also a high corpus-wide frequency value
(denoted f(das)). However, if we consider the term Mehrwert (surplus value) instead, the
frequency with which this word appears in Marx’s writings is much higher than the average
frequency of this word across all nineteenth-century authors, so that cf(Mehrwert, cMarx ) is
high but f(Mehrwert) is low. Thus, as a simple measure of uniqueness, we can simply consider
the ratio of these two quantities:

cf(wi , cj )
Uniqueness(wi , cj ) = ,
f(wi )

so that

high
Uniqueness(das, cMarx ) = ≈ 1, while
high
high
Uniqueness(Mehrwert, cMarx ) = > 1,
low

matching our intuition regarding the uniqueness of a word to a particular author. Note that
the formula also allows us to measure how rare a word is for a particular author: if Marx
uses the term Utopie significantly less often than the other authors in the corpus, we’ll now
have a value for cf(Utopie, cMarx ) which is lower than f(Utopie), giving us

low
Uniqueness(Utopie, cMarx ) = < 1,
high

so that now we can qualitatively characterize an author’s use of a word wi in three ways:

56
 • If Uniqueness(wi , ca ) < 1, the word is used less often by the author than its average
usage across authors in the corpus

• If Uniqueness(wi , ca ) ≈ 1, the author uses the word wi at about the average rate it is
used by all authors in the corpus, and

• If Uniqueness(wi , ca ) > 1, the author uses the word wi more often than its average
usage across all authors in the corpus.

The formal cTFIDF formula, then, is simply this intuitive Uniqueness measure with one
modification enabling it to better capture a key empirical property of “natural” (human)
languages: namely, that word frequencies in these languages exhibit what’s called a “power
law” distribution. In a linguistic context, this “law” is really just an observed tendency, found
in nearly all known human languages, for the frequency of words to be inversely proportional
to their frequency rank in the language. In other words, if the frequency of the most common
word is f1 , the second most frequent word tends to occur 1/2 as often, the third most frequent
word 1/3 as often, an so on, such that the frequency of the nth most frequent word can be
written as fn = (1/n) · f1 .
In English for example, the most frequent word (“the”) occurs about f1 = 50, 000 times
per million words, the second-most frequent (“be”) occurs about half as often (f2 = (1/2)f1 ),
the third-most frequent (“and”) about one-third as often (f3 = (1/3)f0 ), and so on, with
this pattern (of the nth most frequent word occurring 1/n as often) continuing all the way
down to the rarest words in the language. Given this tendency, then, the frequencies of words
will exhibit massive (exponential) differences when comparing among high-frequency words,
but tiny differences when comparing low-frequency words, as shown by the orange points in
Figure 1.24. In the case of our Uniqueness measure, however, we don’t want the difference
between (e.g.) the first and second most frequent words to count 100 times more than the
difference between the 100th and 101st most frequent words. Thus, to prevent this linguistic
property from skewing our measurements, we replace the denominator with its logarithm,

57
ensuring a more gradual decrease in the denominator’s value as we measure uniqueness for
less frequent terms, illustrated by the blue points in Figure 1.24.

f (n) Raw Frequency

4 Log Frequency

n
10 20 30 40 50 60 70 80 90 100

Figure 1.24: An illustration of how the logarithm (plotted in blue) “smoothes out” the original
corpus-wide frequencies (plotted in orange), where the horizontal axis represents a word’s
ranking in the corpus (n = 1 representing the most frequent word, n = 2 the second most
frequent, and so on) and the vertical axis represents the frequencies of these words, measured
in both original and logarithmic units. In particular, note that the logarithmic plot decreases
the magnitude of the differences between the first ~30 words, and then slightly magnifies the
differences between the remaining ~70 words, relative to the original frequency plot.

With this modified, linguistic-theory-informed version of our Uniqueness measure in place

we can to capture, in a mathematically- and linguistically-principled manner, the central-
ity of certain terms to certain discursive genres: Hegelianism, political economy, Comtean
positivism, etc. In future chapters we therefore use this approach to anchor our axes, giv-
ing us a stable coordinate system—a common ruler—with which to measure differences in
word usage between authors as well as changes in usage over time within and across these
broadly-defined genres.
Specifically, the construction of these axes proceeds as follows: given a corpus C containing
words {w1 , w2 , . . . , wn } and a set of genres {G1 , G2 , . . . , Gm } (where each genre is associated

58
with a subset of the texts in the corpus which the researcher believes to be sufficiently
representative of that genre), the researcher can compute “genre-uniqueness scores” ui,j =
cTFIDF(wi , Gj ) for all (word, genre) pairings. With these scores in hand the researcher can
then rank, for a chosen genre G∗ , the N words most unique to G∗ by choosing the N words
with the highest cTFIDF(wi , G∗ ) values. Once word embeddings for all words in the corpus
have been generated, then, a centroid vector representing the genre G∗ within this space can
be computed as the weighted average of the word vectors for the N words found to be most
unique to G∗ . Once these centroid vectors have been computed, we can use them to orient
ourselves within the otherwise arbitrary41 geometric space. The movement of points through
this space, for example, which before corresponded only to non-human-interpretable changes
in the numeric values of a vector’s 768 coordinates, can now be meaningfully interpreted with
respect to one or more of these computed centroid vectors: a movement towards a centroid G
represents an increase in similarity to the semantics of this discursive genre, and vice-versa
for a movement away from G.
Social-scientifically, this interpretive lens enables us to analyze the linguistic field in
terms of comparative statics: holding some variables constant to enable the measurement of
variation in others with respect to this static value. The “Hegelian-ness” of various authors in
the corpus, for example, can be measured by comparing how their author-specific embedding
vectors vary in relation to the static Hegelian genre-centroid vector. Or, we can evaluate
an author’s “Hegelian-ness” over time (which we will indeed do for Marx in Chapter 2)
by tracking changes in the distances between their author-specific vectors and the static
Hegelian genre-centroid vector, over time.
This latter example, however, leaves some important questions unaddressed with respect
to the nature of the linguistic fields we are studying: while treating a centroid for Hegel’s
writings as a constant in order to analyze changes in Marx may be reasonable, since these
writings predated Marx’s own (i.e., Hegel was not in interlocutory or back-and-forth dialogue
41
Arbitrary, that is, with respect to the original x and y coordinates generated by the embedding algorithm.

59
with Marx in the same sense that Proudhon was), what do we do about the case where
one author’s speech acts feed into and have an effect on the very linguistic field from which
another author subsequently draws when constructing their own speech acts? To incorporate
time as a variable in our analyses, we need to be able to model these complex linguistic
feedback effects in a non-trivial way. That is, we need to explicitly augment our static
embedding model with a model of the temporal relationships between individual speech acts
and the linguistic fields from which they are drawn.
As the first step in this direction, then, we begin the next section with an introduction to
Saussurean structuralist linguistics. In particular, we discuss the distinction between langue
and parole introduced by this approach, corresponding to the distinction between what
we’ve been calling linguistic fields and speech acts thus far, and explain how this perspective
gives rise to a working model of the mutual interactions between these two co-constitutive
phenomena which we can carry over into our computational approach.

1.2.4 Synchronic and Diachronic Analysis: Understanding the Langue-Parole Distinction

Notwithstanding the progress in the field of semantic embeddings described at the end of
Section 1.2.2, resulting in algorithms more and more capable of capturing meaning holistically
across larger and larger linguistic units, there remain additional aspects of language—aspects
crucial to social-scientific analysis—which we have not yet discussed and which will require a
complementary set of tools to capture. To understand what the models thus far have lacked
(and thus what additional tools we’ll need to employ), however, we’ll need to understand the
Saussurean distinction between synchronic and diachronic approaches to language analysis.
Ferdinand de Saussure, in his Course on General Linguistics, differentiates between two
perspectives one can adopt when analyzing language: in the the first, which he calls the
synchronic perspective, the linguistic entities—texts, utterances, etc.—are treated as if they
exist in a single moment in time (thus roughly corresponding to the notion of cross-sectional
analysis in the sciences), so that one can consider their interrelationships within a stable,

60
static field of linguistic meaning. In other words, in the synchronic mode, one is concerned
with understanding a particular collection of utterances, but not with the dynamics by which
these utterances, once issued, subsequently changed the field of language from which future
utterances were generated (nor are they concerned with the other temporal direction: how
past utterances shaped the linguistic field which gave rise to the utterances of interest). In
Pocock’s phrasing, however, the linguistic field that Cambridge School historians study is
inherently, definitionally diachronic, and must be studied in a diachronic mode:

“the diversity of linguistic contexts that went to determine what could be said
but were at the same time acted upon by what was said.” (Pocock 1985, p. 2)

The models we’ve discussed thus far have all been essentially synchronic. The embedding
algorithms, for example, are trained with the sole objective of capturing the relationships
between words and the contexts they appear in across a given corpus, regardless of the
temporal order in which the word-context pairs appeared. In reality, however, if we want
our computational models to be able to aid in Cambridge School-style context-sensitive
historical inquiry42 , we should take heed of the inherently diachronic emphasis of e.g. Pocock’s
definition cited above: “a history of actors uttering and responding”.
This Saussurean jargon may, at first glance, seem like a distinction without a difference:
to obtain a diachronic model, one can simply train a sequence of synchronic models, one
for each time period. We argue, however, that this misses the point of language modeling
entirely: word embedding models are useful social-scientific tools because they implement a
powerful theory of linguistic discourse—namely, the Firthian Hypothesis that we shall “know
a word by the company it keeps” (Firth 1957, p. 11)—which enables researchers to see (via
the generated geometric spaces) the contours of how particular authors use particular words
to do particular things. By the same logic, then, we argue that a diachronic language model
is only useful or relevant for social-scientists to the extent that it can similarly implement
42
That is to say, contextual rather than textual analyses. If one was solely interested, for example, in
evaluating the internal coherence of a single text like Leviathan, the synchronic methods already introduced
would be sufficient, as we saw in the Leviathan embedding examples of Section 1.2.

61
an analytically-powerful theory of language change. Obtaining a sequence of trained word-
embedding spaces is therefore necessary but not sufficient for having a useful social-scientific
tool. It must be paired with a theory regarding the relationship between these time-slices,
just as pairing the outputs of word embedding algorithms with the Firth Hypothesis provides
us with a social-scientifically-grounded interpretive tool for understanding the relationship
between words in the embedding spaces.
To this end, in the next section we survey the political-theoretic literature on the notion
of influence—an often-asserted but rarely-interrogated phenomenon in the practice of intel-
lectual history—with an eye towards identifying what historians mean when they posit that
one event or thinker influenced another. Over the course of this survey, by pinning down this
meaning as precisely as possible, we derive a coherent theory of influence which can thus
serve as our lens through which to view and understand changes in embedding spaces over
time.

1.3 Theories of Influence, Past and Present

1.3.1 Structure vs. Agency

Within the community of intellectual history and the history of political thought, strict
battle lines are often drawn between those who center individual autonomy and free will
as the driving forces behind intellectual innovation, and those who instead focus on factors
outside of the individual such as material conditions, social institutions, or group psychology.
Grouping the latter categories under the rubric of “structures”, for example, Skinner sees
this as a growing concern among post-WWII scholars, to the point that “the question of
the relative weight to be assigned to agents and structures has indeed become a central
preoccupation with the younger generation of social theorists” (Skinner 1990, 18).
Perhaps especially in debates regarding our topics of study in the chapters that follow,
however—Marxism and 19th century socialist discourse more broadly—the adoption or dis-
missal of key arguments often does hinge upon whether or not one accepts the inviolability

62
of methodological individualism43 .
This general anxiety regarding the importance of methodological individualism manifests,
among intellectual historians, as an anxiety around whether historical explanations which
employ influence claims strip historical thinkers of their agency. In other words, the main
challenge for a theorist in either case is to find a conception of social phenomena which
can mitigate the incongruity between (a) looking out at the world and perceiving emergent
“social forces” which seem to act on individuals as objects despite these forces not having
agency or intentionality themselves, and (b) phenomenologically experiencing “selfness” as
a subject with the free will to act or not act upon the objects “out there”.
The history of Marxism for example, from Marx’s time to the present, is largely a history
of theorists trying to grapple with this antinomy: if the internal logic of capitalism implies
its inevitable self-destruction and replacement by socialism, and with its trajectory towards
this outcome determined by iron-clad “laws of motion” discovered by Marx, what exactly is
left for an individual socialist to do, free-will-wise (especially given the near-certainty of a
violent response by the agents of capital)? The complacency which this theory can therefore
induce, invoking images of socialists calmly sitting and waiting for capitalism to fall, has been
adduced to explain all manner of failures of Marxian socialism, but especially the failure of
the German SPD to seize power in the years leading up to the Reichstag Fire (a dynamic
we explore in more detail in Chapter 2).
Returning to the role of theory in the social sciences, in a paradigmatic case of “easier to
be a critic than a creator”, much ink has been spilled lambasting social theories on the basis
of both (a)—i.e., for granting too much autonomy to individuals and not acknowledging
the pressures exerted by external forces (for example in critiques of libertarian contract
theories)—and (b)—for denying the agency of the individual and treating them as a pawn
in the machinations of an external or abstract entity (for example in critiques of “vulgar”
or strongly-teleological historical materialism). In our view, however, these argumentative
43
For the case of debates among late-twentieth-century Western Marxists, for example, see Elster 1982,
Althusser 1968, Cohen 1978.

63
poles can serve as helpful boundary points delineating the two extremes of a spectrum of
theories of intellectual influence. The problem outlined at the end of the previous section,
of establishing a cogent theory of influence to serve as the interpretive machinery for our
diachronic embeddings, thus becomes more manageable; we can evaluate theories on the
basis of how well they adjudicate between these two critiques.
In fact, although Skinner frames it as somewhat of a zero-sum conflict in the quotation
cited above, Giddens 1979 analyzes aspects of the two “sides” that, we argue, can be made
complementary rather than contradictory. He points out, for example, how separate litera-
tures have already made great strides in understanding social behavior at various levels by
adopting one or the other approach. Social scientists, he argues, often err on the side of deter-
minism (with an extreme case being, e.g., vulgar historical materialism), while philosophers
err instead on the side of theorizing individual action without reference to social-relational
considerations:

“There exists a large philosophical literature to do with purposes, reasons and

motives of action; but it has to date made little impact upon the social sciences. In
some part this is understandable, because the philosophy of action, as developed
by British and American philosophers, has not paid much attention to issues
that are central to social science: issues of institutional analysis, power and social
change.” (Giddens 1979, p. 2)

In our reading of these facts, then, the historian’s task is not to choose between the two, or
to cleanly reconcile them, but rather to adopt the contributions of both literatures as part of
their historiographic toolbox. Hence, combining this interpretation of Giddens 1979 with the
later argument of Sperber 1996, we see no necessary contradiction in adopting a notion of
intellectual influence that remains committed to materialist and methodological-individualist
principles. Consider, for example, the following three points on an Ockham’s-razor-based
spectrum of theoretical complexity (i.e., a spectrum constructed via a notion of parsimony
with respect to the theories’ underlying assumptions):

64
 1. Physical theories: for example particle physics, which studies the smallest indivisible
units known to us (quarks),

2. Social theories: for example psychology, which studies the individual as a single cohesive
unit, and

3. Metaphysical theories: for example Hegel’s “world spirit” or the Deep-Ecological Gaia
Hypothesis, which assert non-observable agents whose dynamics give rise to observable
phenomena.

Considering the points themselves, we can gauge that (1) satisfies methodological individ-
ualism trivially (since particles aren’t thought of as having agency with respect to their
behavior) and (2) satisfies it non-trivially, but that (3) may not satisfy it unless it can be
decomposed into combinations of (1) and (2). Then, we can consider the spaces in between
the points, and posit e.g. that without any “objective” way to adjudicate among theories
lying between (2) and (3), as a rule of thumb we’d prefer the theory closer to (2). In the next
section, then, we choose from among various theories of influence in a similar way, evaluating
where they lie on a spectrum of strictness with respect to the scope of admissible influence
claims, subject to the constraint of not violating key historiographic prerogatives like that
of respecting individual agency.

1.3.2 Mapping and Evaluating the Theories

Though even the earliest historians implicitly grappled with the complexities of influ-
ence claims, it wasn’t until the 20th century that historians began explicitly constructing
and debating theories of intellectual influence. Assessing the state of the historical profes-
sion in his 1926–1928 lectures, for example, R. G. Collingwood laments that historiographic
methodology “is in point of fact almost wholly neglected by historians,” to the extent that
“the ordinary historian can give no account of the processes by which he extracts narrative
from sources” (Collingwood 1946, p. 389).

65
 While one can trace this historiographic self-consciousness back to e.g. the inquiries of
the Vienna School of logical positivism, the theories most relevant to our investigation were
formed in the aftermath of Karl Popper’s critiques of theretofore-unexamined aspects of
historical practice in The Open Society and its Enemies (Popper 1945). Popper massively
raised the stakes of the debate around historical influence by arguing that a set of historio-
graphic precepts, introduced by Plato but taken to extremes by Rousseau and Marx, in fact
provided crucial ideological support to the self-legitimation of totalitarian regimes. Though
Popper produced and presented these arguments, for the most part, in the context of pre-war
debates within the philosophy of science, it quickly spread to other fields in the aftermath
of World War II as scholars grappled with the horrors that the war had wrought.
Variations on this argument were introduced into political theory proper, for example, by
Hannah Arendt’s influential The Origins of Totalitarianism (Arendt 1951), while J. L. Tal-
mon’s The Origins of Totalitarian Democracy (Talmon 1952) centered Karl Marx’s thought
as the apotheosis of this immanently-totalitarian historiographic trend. This illocutionary
move to link Karl Marx’s thought as inexorably linked with the Soviet regime gained trac-
tion in less academic circles as well, as the Cold War heated up in the years following the
Berlin Airlift. Prominent US politicians were spurned into anti-communist action by Arthur
Schlesinger’s simplified version in The Vital Center (Schlesinger 1949), for example, and
perhaps most influential in this genre was Theodor Adorno’s co-authored The Authoritarian
Personality (Adorno et al. 1950), which gave this theory a micro-level social-psychological
interpretation.
We describe this post-war background so as to situate the emergence of the Cambridge
School within this broad atmosphere of suspicion and distrust attached to the perpetuation—
and even more so the creation—of sweeping “grand theories” of historical development. It is
in light of this environment that one can understand the skepticism embodied in the nearly
impossible-to-satisfy criteria for valid influence claims developed by Cambridge School prac-
titioners, which we argue are far too restrictive to be fruitfully employed towards interpreting

66
diachronic embeddings. Though their strictness was sometimes attenuated (but sometimes
increased) in later methodological works, here we consider the criteria given in Quentin
Skinner’s “Meaning and Understanding in the History of Ideas” as exemplary of the broader
Cambridge School conceptualization of intellectual influence.
In this work, Skinner straightforwardly defines the necessary conditions for a “minimally”
admissible44 influence claim of the form “A influenced B”:

(i) B is known to have studied A’s works;

(ii) B could not have found the relevant doctrines in any writer other than A; and

(iii) B could not have arrived at the relevant doctrines independently.

Our argument that these conditions are excessively strict, however, is based not so much in
the content of the conditions themselves than in the epistemic frame in which their necessity
is posited. Within an inferential system where statements are evaluated via laws of non-
probabilistic predicate logic, like that of the early logical positivists from whom Skinner
draws much of his inspiration45 , it may indeed be the case that the influence claim cannot
be evaluated as true unless all three conditions are satisfied. Using such a system to judge
a statement like “A influenced B” as made by a historian, however, makes sense only under
a draconian epistemic regime in which “truth” with respect to historical claims denotes
“certainty that all other potential explanations can be ruled out”, such that any status short
of this leaves the claim consigned to a dustbin of historiography, on the basis that “we’ll
never know”.
44
Though some discussions of these criteria use “acceptable” or “valid” rather than “admissible” to describe
them, the latter two terms introduce the assumption that the theory-maker would believe the influence claim
was true if it met their criteria, an assumption avoided in the use of “admissible”. Although this assumption
is probably the case for Skinner, because his conditions are so strict, most theory-makers seem to be outlining
something more subtle: conditions that, if they are not met, are not “eligible” for true/false inference in the
first place. Hence we use “admissible” vs. “non-admissible” throughout.
45
Here we have in mind, in particular, stronger forms of early logical positivism which contended that
statements are meaningful if and only if they can be converted into statements of predicate logic, but allowed
for the possibility that meaningful statements whose predicates require sense data for their evaluation (namely
synthetic, as opposed to analytic, statements) may be indeterminate in practice.

67
 Granting here that we indeed will never know the truth or falsity of an influence claim in
the sense implied this framework, however, we argue that (short of abandoning the historical
endeavor altogether) a feasible framework for rigorous inference from historical evidence is
still readily available in the form of the standard Bayesian inferential calculus. This approach,
shedding the methodological shackles imposed by the need to prove truth or falsity of a
historical claim, instead allows for varying degrees of belief in the truth of a proposition which
are—most importantly (in terms of the analytical grounding of our argument)—governed by
an “optimal”46 set of inference rules for updating these degrees of belief upon observing new
information. The move from the framework implied by Skinner’s criteria to our proposed
Bayesian framework can be summarized as changing the evaluative question the historian
asks, away from “Is proposition P true?” and towards “How strongly should we believe
proposition P to be true?” (Williamson 2010)—in other words, away from yes-or-no questions
and towards questions of degree.
Our proposal for adopting this degrees-of-belief approach is motivated by the observation
that many of the methods and presuppositions which social theorists import from statistics—
having been employed far more frequently and for far longer by the natural sciences—are not
up to the task of explaining social behavior as-is, i.e., that they will require significant revision
rather than cut-and-paste transposition from the natural to the social worlds. This point is
argued forcefully, for example, in William Connolly’s The Terms of Political Discourse, which
explicitly argues, “it is unlikely that models of explanation derived from natural science can
apply to social and political life in unrevised form” (Connolly 1974, p. 5) As a heuristic
for modifying these explanatory models to take account of social dynamics (drawing on the
pragmatist tradition), Sokal and Bricmont 1997 suggests that we look for inspiration in
the ways humans intuitively draw quite complex inferences every day, often without being
conscious of doing so. The Bayesian approach meets this challenge: for reasons outside of the
46
“Optimal” with respect to a collection of intuitive inferential axioms developed by statisticians, mathe-
maticians, and philosophers in the centuries since the posthumous publication of English minister Thomas
Bayes’ “An Essay Towards Solving a Problem in the Doctrine of Chances” (Bayes 1763).

68
scope of this work, this model better captures the way humans neurologically process and
store information, and use this stored information to improve future inferences. In evaluating
causal questions like “Does cilantro give me a stomachache?”, for example, human cognition
exhibits the degrees-of-belief approach of Bayesianism rather than the yes-or-no approach of
propositional-logic-based models like that employed in Skinner 1969.
While some historians find such an approach unsatisfactory on account of its dependence
on “subjective” probabilities rather than objective truths, we argue that this is in fact a point
of strength, not weakness, for the veracity of the approach. With respect to its statistical
foundations (i.e., when considered as a tool for information processing in general), Bayesian
inference is “no more inherently subjective than are likelihoods and the repeat sampling as-
sumptions required [by non-Bayesian approaches] for significance testing” (McElreath 2020,
35). Thus, we argue, when one takes it and applies it to historical phenomena, its strength
comes not from the fact that it eliminates uncertainty (no inferential procedure does) but
from the fact that it does the exact opposite: that it explicitly includes subjectivity, uncer-
tainty, and ambiguity as considerations which the model itself can reason about. “Because
the prior [the representation within a Bayesian model of the researcher’s background as-
sumptions] is an assumption, it should be interrogated like other assumptions: by altering it
and checking how sensitive inference is to the assumption.” (McElreath 2020) By way of this
sensitivity-checking process, moreover, historical researchers can explicitly test the robust-
ness of their findings. For example, among classical scholars, Plato’s Epistles are “among the
most disputed texts of antiquity” (Wohl 1998), with scholarly views regarding which of the 13
proposed letters are authentic spanning from “none” to “all”. Thus, if a historian who adopts
a Bayesian framework is concerned about whether their findings depend too strongly on the
contents of some or all of these letters, they can simply remove the questionable entries from
the set of admissible evidence encoded in their model’s prior, conduct the inference via the
same procedure as before, and see whether or not their previously-obtained results remain
significant.

69
 If one abstracts away from the particular Bayesian terminology of this conception, more-
over, it simply describes an approach to historical understanding which was consciously
articulated well before the advent of the Cambridge School. Collingwood’s 1926–1928 lec-
tures for example, mentioned at the beginning of this section, argue that “every narrative
that we can at any given moment put forward is only an interim report,” but that, impor-
tantly, “it does not not follow that there can be no solid advance in historical knowledge”
(Collingwood 1946). The Bayesian machinery, we argue, simply adds flesh to the skeleton
of this insight, allowing us to reason in a principled manner about what the most likely
narrative(s) are with respect to the currently-available evidence. Indeed, if we substitute the
phrase “most likely” for the term “true”, this approach satisfies the criteria Collingwood
proposes as the requirement for such “solid advance[s] in historical knowledge” to be made:
that “the principles [of historical inference] must be independently established a priori in
order that the narrative constructed by their means may be known to be true” (ibid., 389).
This move away from binary evaluation and towards a framework allowing for a continuum
of belief, therefore, better reflects the hermeneutic ideals of historians and readers of history
and gives us a richer procedure for evaluating influence claims than the “true only if no other
possible explanation” framework behind Skinner’s conditions.
It is worth noting, however, that although post-WWII skepticism led to their excessive
strictness, these criteria did serve as a much-needed corrective to a historiographic paradigm
that was only beginning to fall out of favor in 1969 (38 years after it was named and critiqued
by Herbert Butterfield): the paradigm of “Whiggish” teleological explanation. In short, works
in this paradigm typically construct a historical narrative wherein a well-ordered sequence
of “inevitable” events culminates in the form of an event the historian sought to explain,
which thus serves as the narrative’s climax. Examples abound, but Hegel’s presentation
of constitutional monarchy as the completion of a world-historical progression, or modern
hagiographies of US athletes and entrepreneurs valiantly pushing themselves beyond all
obstacles to their success, serve as examples of this tendency.

70
1.3.3 The Empirics of Influence: Historical Sketches

Having now introduced the models of inference we will employ, in this section we move
from the theory to the practice of historical inference. In a series of case studies, we develop
intuition around how the qualitative, interpretive claims made by political theorists can be
translated into quantitative models, thus building towards the more in-depth studies we
conduct in the chapters that follow.
If we hope to expand the horizons of historical research by bringing the power of well-
established quantitative methods to bear on our understanding of political texts, we first
need to understand how exactly these texts can be brought from the physical into the digital
realm, and get a sense for the landscape of already-digitized texts which can thus immediately
be used as inputs to our computational analyses.
Luckily for us, vast collections of historical texts and historically-relevant data have been
digitized in recent years, such as the millions of books made available by the HathiTrust Part-
nership47 —providing resources for which (we argue) supply currently far outpaces demand.
A computational study of influence in intellectual history could leverage these resources
to develop a set of initial hypotheses regarding who influenced whom—and what types of
mechanisms or relationships facilitated this influence—when could then be operationalized
as inputs to the embedding methods described above and evaluated based on their results.
In fact, we contend that many of the historical claims made in the Cambridge School
works discussed above suffer—despite their invaluable incorporation of contextual informa-
tion in the form of other texts—from a lack of attention to the non-textual metadata available
about the texts and authors under consideration. To take one prominent example, consider
Skinner’s analysis of E. M. Forster’s A Passage to India in the first volume of Visions of Pol-
itics. A seemingly insignificant part of Forster’s novel, its concluding sign-off—“Weybridge,
1924”—is considered in the context of literary conventions around sign-offs: James’ Joyce’s
47
In fact, the books in the HathiTrust library are not only digitized but also preprocessed via natural
language processing algorithms, enabling their use as direct inputs to many of the algorithms described and
used in this work.

71
“Trieste-Zürich-Paris” sign-off at the end of Ulysses, for example, is taken by Skinner as
an example of the sign-off’s role in “draw[ing] attention to the romantically nomadic life
of the author” (123). Thus, by signing A Passage to India off with the simple reference to
Weybridge—“the classic instance of a prosaic English suburb” in Skinner’s words (124)—
Forster is seen as doing something with words, namely, subtly mocking the “posturing” (ibid.)
of those who use this authorial convention to portray themselves as cosmopolitan, globe-
trotting intellectuals, traveling the globe to give a glimpse of its splendor to the unwashed
masses.
Bevir 1999 astutely points out, however, the fact that textual evidence on its own, from
Forster’s novel itself or the broader context of the sign-offs of contemporary novels, is not
sufficient to justify Skinner’s conclusion. Skinner may be right, but it also may simply be
that “he did write the novel in Weybridge in 1924, and maybe he intended simply to record
this fact.” Thus, to decide between these two hypotheses, “we must consider things other
than the linguistic context of the novel.” (86) Indeed, once equipped with non-linguistic
contextual data about Forster’s life, adjudicating between the two interpretations becomes
a fairly straightforward task: “If they discover that he wrote the novel in Cambridge and
India from 1922 to 1924 without ever visiting Weybridge, the case for understanding the
utterance as a parody would seem more or less conclusive. But if they discover that he wrote
the novel in Weybridge in 1924, the idea that his utterance was a parody would begin to look
rather more doubtful.” (ibid.) It is this example of the insufficiency of purely textual data
for contextual history, but the possibility of settling disputes via contextual metadata, that
motivates our sketch in the next section of bringing Wikipedia data to bear on discussions
of influence in the history of political thought.

1.3.3.1 Text-Mining Influence Claims

Though it has not, to our knowledge, made its way into the academic literature, a series
of projects from data-visualization hobbyists over the past few years have shed new light

72
on the “flow” of influence among philosophers by constructing diachronic plots constructed
from contextual metadata contained in Wikipedia pages for philosophers, intellectuals, and
political actors, in particular, from the “Influenced By” and “Influenced” sidebars often
found on the pages for these individuals. Beginning with a 2012 blog post titled “Graphing
the History of Philosophy”48 , the project was expanded to cover the entirety of Wikipedia
in 201449 , and this community has continued to explore the possibilities of what can (and
can’t) be learned from this data, whether automatically50 or in combination with manual
curation51 .
Inspired by this endeavor, and how the developed tools could aid studies of influence
in the history of political thought, we began by similarly scraping the data contained in
all instances of this sidebar across Wikipedia. Crucially, we were then able to augment this
base data with both the automatically-generated and manually-curated data generated by
the data-visualization projects described above, resulting in 5,033 hypothesized influencer-
influencee pairs52 .
Taking these influencer-influencee pairs as seeds, possibilities abound for “bootstrapping”
further pairs by computationally mining actual texts (both pedagogical and scholarly), using
the types of language models described earlier to identify:

(a) sentences, paragraphs, chapters, or other textual units53 where both thinkers in the
pair appear, as well as

(b) sentences which mention “influence”, “mentorship”, or other influence-related words,

which may or may not refer to any particular thinker.

As an example of a concrete step in this direction, we obtained JSTOR’s data54 for all
48
https://coppelia.io/2012/06/graphing-the-history-of-philosophy/
49
No longer on the web, but available via the Wayback Machine archive.
50
See, e.g., https://github.com/S4N0I/theschoolofathens
51
https://www.denizcemonduygu.com/philo/
52
This data is available at https://github.com/jpowerj/our-word-is-our-weapon
53
For the sake of brevity, but without loss of generality, we restrict our discussion to sentences for the
remainder of the section.
54
Specifically, via a combination of JSTOR’s data export tools and our own scraping code, we obtained

73
issues of the Journal of the History of Ideas from 1940 to present and all issues of the journal
History of Political Thought from 1980 to present55 . With this data in hand, in a manner
akin to Rawls’ method of reflective equilibrium56 , we refined the base hypotheses further and
further by

1. Analyzing random samples of the type (a), to discover alternative linguistic forms in
which influence claims are made, and then

2. Analyzing sentences of type (b) to reveal new influencer-influencee pairs that had not
previously been considered.

Once this procedure had been carried out to the point that diminishing returns overtook
the benefits—in our case, the admittedly subjective point at which an HPT expert could
attest that it covered a substantially wide range of the types of influence claims made in the
field—the data (i.e., the union of the influencer-influencee pairs themselves and the sentences
in our corpora which pertained to them) was used as the input to two separate machine
learning models. In the first, a binary classifier was trained to differentiate between influence
claim statements and all other statements, and (importantly) to provide confidence scores for
each classification, such that we were able to add additional statements to our collection of
influence claims by running the trained model on the full HPT journal corpus and extracting
claims classified above a minimum-confidence threshold as influence statements.
Then, to take a step from simply collecting these influence-statements towards under-
standing them, we trained a more complex multi-class classification model to categorize the
influence claims based on the mechanism(s) of influence they posit, from among the mecha-
nisms discovered via step (b) in the procedure discussed above. In this case, the model was
trained to differentiate between claims based on:
PDFs, OCRed text, N -grams (up to N = 3), and metadata (authors, dates, journal and issue names, etc.)
for each paper.
55
Also available at https://github.com/jpowerj/our-word-is-our-weapon
56
See Rawls 1951, which computer scientists would recognize as a very similar process to that of the
Expectation Maximization (EM) Algorithm in Machine Learning (where in the latter instance the hypotheses
happen to take the form of numerical estimates of a model’s parameters).

74
 1. Collaboration: for example, “Karl Marx and Friedrich Engels carried on one of the great
intellectual collaborations in the history of scientific research.” (Gandy 1979, p. ii)
=⇒ Collaboration(Karl Marx, Friedrich Engels)

2. Mentorship57 : for example, “the grandly-titled Tractatus Logico-Philosophicus was be-

gun in 1915 following Wittgenstein’s tenure at Cambridge under [Bertrand] Russell’s
mentorship” (Wright 2006, p. 72)
=⇒ Mentorship(Bertrand Russell, Ludwig Wittgenstein)

3. Reading: for example, “Not only had Marx read Hegel’s Logic in 1858, but we know
that he studied it once again in 1860.” (Dussel 2002, p. 195)
=⇒ Reading(Karl Marx, G.W.F. Hegel)

4. Shared-Intellectual-Community: employment or study at the same institution, for exam-

ple, as in “Schelling and Hegel first met at the Tübingen Seminary in 1790, and the
two young men shared a room there during their student years” (Levine 2006, p. 124)
=⇒ Shared-Intellectual-Community(F.W.J. Schelling, G.W.F. Hegel)

5. Shared-Geography: for example, “Lenin was an occasional visitor in Vienna. Hitler, like
Trotsky, had lived there for years.” (Morton 1990, p. 290)
=⇒ Shared-Geography(V.I. Lenin, Adolf Hitler)

6. Rivalry: for example, “the cultural philosophy of the Historical School and, more par-
ticularly, of Hegel’s rival at the University of Berlin, Friedrich Schleiermacher.” (Toews
1985, p. 7)
=⇒ Rivalry(G.W.F. Hegel, Friedrich Schleiermacher)
57
A recent quantitative work, published after the conclusion of our small-scale study, was able to scrape
37,157 mentor-mentee pairs from the ProQuest PhD Dissertation and Thesis databank (PQTD), “an official
record of advisor–student relationships taken from PhD theses” in biomedicine, chemistry, math, and physics
between 1960 and 2017, “supplemented with crowdsourced data from Academictree.org and the Mathemat-
ics Genealogy Project (MGP).” (Ma et al. 2020, 14077) The creation of a “ground truth” dataset of this
magnitude for the history of political thought (rather than the natural sciences) would, we argue, represent
a watershed moment for the empirically-principled study of the role played by influence in the development
of political thought.

75
 We note here that the goal is not to classify the hypothesized influencer-influencee pairs
using this schema—this would introduce a new set of complexities due to the confounded-
ness of our categories (specifically, the presence of one relationship makes the presence of the
others much more likely)—but only the statements making influence claims. After the state-
ments themselves have been classified by our model, we subsequently label the referenced
pairs with the relationship type learned for each statement.
We note this detail because it represents a situation which may be counter-intuitive for
those unfamiliar with what is called the “bias-variance tradeoff” in statistical inference: for
example, although one might have the intuition that the best model would be one able to
capture the multiple relations posited by a compound statement like “Hegel and Schelling
were university roommates, early friends and collaborators, and eventually rivals,” in practice
to the small-N nature of the study (relative to the datasets with billions of observations
regularly used by computer scientists in academic and commercial settings) means that the
greater statistical power of a simple model distinguishing between four classes far outweighs
any benefits of a more complex Multi-Label Classification model which, though it can handle
these compound statements, massively sacrifices statistical power by having to distinguish
between 24 = 16 possible combinations of classes for each statement. In fact, in our case, the
multi-label classification approach is unnecessary, as even compound statements like these
can be decomposed by way of linguistic dependency parses58 .
Then, by examining which linguistic features in the input statements best helped the al-
gorithm distinguish between these classes59 , we can understand in a linguistically-principled
sense how historians of political thought talk about influence, and (simultaneously) automat-
58
See Ash et al. 2020 for an example in computational social science, where this approach is used to
differentiate and separately model the subjects and objects of legal contract clauses.
59
The particular type of classifier we trained, called the Random Forests model (as implemented in the
scikit-learn library’s RandomForestClassifier package), produces an explicit decision tree which is con-
structed to optimally partition—in the sense of minimizing the entropy or Gini coefficients of successive
splits—the domain of the inputs given a set of pre-specified constraints like the maximum depth of the
desired tree. Hence one can “read off” the most helpful features directly from this tree, as (for example) its
topmost decision node will split the data based on the input feature most predictive of the output, the nodes
in its second level will split the data based on the input feature those most predictive of the output within
the subgroups created by the first split, and so on.

76
ically generate detailed timelines of posited influence paths, with each “link” labeled with
its mechanism.
As the final step in our derivation of a fruitful model of intellectual influence, we linked
these empirical findings with the Bayesian model described earlier, which allowed us to
derive measures of uncertainty with respect to each of the findings. Though this could be
accomplished in several ways (for example, as in the example of Plato’s Epistles above,
one could reduce the degree of certainty attached to some specific source to 1/2 the level
of all other sources), in our case we assigned weights representing our conception of the
relative strength with which each influence mechanism acts. For example, to encode the
supposition that the strength of the Shared-Geography relation tends to be lower than that
of the Collaboration relation60 in terms of explanatory power (as an anecdotal but certainly
noteworthy example, Hitler, Trotsky, Tito, Freud, and Stalin all lived in Vienna in the early
1910s61 , but several pairs within that group were most likely not aware of each other), we set
the priors in our model such that the effect on the model of observing a Shared-Geography
datapoint was scaled to 1/3 of the effect of observing any other type of datapoint62 .
The end result of inducing this prior, then, is that our lower confidence in claims of
influence based on Shared-Geography relations translates into a reduced posterior degree of
belief generated by the model for influence pairs which are heavily based on Shared-Geography
claims, relative to the degrees of belief that it would generate for these pairs if all four types
of claims were weighted equally. Thus we see that, as discussed above, adopting this Bayesian
60
Here we highlight another commonly-misunderstood strength of the Bayesian approach: though its critics
within social science and the humanities often decry the presumptuousness of “assigning a number” to an
abstract phenomenon like the strength of a type of influence, in fact what matters in nearly all of these
models is not the individual values of the numbers used to encode prior assumptions, only their relative
magnitudes. Mathematically, this fact can be derived in a manner similar to the econometric argument for
focusing on ordinal, rather than cardinal, utility in models of welfare economics (see, e.g., Harsanyi 1953 for
the basic argument, and Roemer 1996, 13ff., for an example of its relevance for conceptualizing, analyzing,
and comparing political-theoretic models of justice such as those of Rawls, Nozick, and G. A. Cohen).
61
See Morton 1990 (cited above for our example Shared-Geography statement) for details on their meetings
and near-meetings during the years 1913 and 1914.
62
Specifically, in terms of the mechanics of the Bayesian model, this scaling was achieved via Laplace
smoothing, which induces a particular prior but can be easily interpreted as telling the model to perform
inference as if it had observed each of the non-Shared-Geography datapoints three times, but the Shared-
Geography datapoints only once.

77
framework renders our modelling assumptions maximally transparent, to the point that it
in fact provides other researchers studying the same phenomena with a set of tools for direct
constructive criticism of the model: they can point to particular explicit assumptions in the
priors that they believe to be erroneous or invalid, remove them (for example, by weighting
all four types of influence claims equally), and see if our findings still hold.
Bringing together the different parts of the model we’ve described thus far, which we’ll use
as our working model of influence when interpreting the outputs of embedding algorithms
throughout the rest of the study, we obtain the Probabilistic Graphical Model pictured
in Figure 1.42, where the arrows in the diagram represent the pathways through which
prior information (that is, assignments of particular values to the variables represented by
the nodes) will “flow” through the model to produce posterior estimates of the unknown
quantities.

1.3.3.2 The Point is to Change It: Theoretical Innovation and Political Practice in the
History of Marxism

In this section we introduce the basics of how computational text-analysis algorithms can
be used to fruitfully study intellectual influence, by analyzing a corpus of prominent Marxist
texts from 1848 to the present and exploring what types of influence relations we can capture
among its texts. Specifically, we compare findings derived using an information-theoretic
method developed in Barron et al. 2018 with those of a probabilistic text-analysis method
developed in Gerow et al. 2018, both of which aim to quantify the novelty and influence
of particular texts within the corpus. We illustrate throughout how these computational
tools can be linked with social-scientific theories, by interpreting the algorithms and derived
quantities as capturing the creation and destruction of ideological “repertoires of contention”
within Marxist discourse, as defined in Tilly 2008.
Using this combination of computational and social-scientific models, we then move to
a confirmatory, rather than exploratory, analysis, testing a set of hypotheses regarding the

78
influence relationship between texts and the historical contexts in which they were published
(in the language of the model, the relationship between the Et and Xi,t+1 nodes). We utilize
statistical change-point estimation methods (Kulkarni et al. 2015) to test whether the Soviet
invasion of Hungary in 1956 sparked a statistically-significant shift in the novelty of texts in
the Marxist tradition. A significant positive shift would potentially indicate a search for new
approaches among Marxists who turned away from “official” Soviet communism as a result
of the invasion63 , whereas a significant negative shift could be interpreted as evidence of a
successful clampdown on critical discussion of the invasion by the Politburo.
The first step in our study was to develop inclusion criteria for a corpus of Marxist
texts sufficiently representative of both strands of thought and time periods in the history
of Marxism. While in our full study of the origins and diffusion of Marxism in the coming
chapters we focus more on these criteria, for the sake of this research sketch we simply adopt
the two-column timeline of major texts and major events in Marxism from McLellan 2007,
reproduced in Table 1.C.
With the texts we’d like to include in our corpus now established, the next task was to
determine which of them already exist in digital form, and which we would need to digitize
ourselves. By searching in HathiTrust, Archive.org, and ProQuest collections, we were able
to obtain already-digital versions of 90% of the texts in the corpus, leaving only 6 that we
had to digitize ourselves, by scanning the physical books borrowed from the library and then
running Optical Character Recognition software on the scanned .pdf files64 .
For the next step, we processed these digitized texts so as to extract from each book the
data necessary for input to the algorithms of Barron et al. 2018 and Gerow et al. 2018. In
the former case, the algorithm operates by (a) training a 500-topic LDA topic model on the
full corpus, without considering the timestamps of the texts, (b) using this corpus-wide topic
model to compute topic distributions for each time slice, and then (c) computing information-
63
See Hudson 1992, for example, for the finding that “the combined effect of the events of 1956 led to a
membership loss in the region of 7,000” for the Communist Party of Great Britain, mainly due to its official
support for the invasion.
64
Specifically, we used Version 15 of Abbyy FineReader Pro for Windows.

79
theoretic measures for pairs of time slices, for example between the topic distributions at
times t and t + 1.
The results of the study are summarized in Figures 1.25a and 1.25b, where each “block”
represents a quantized portion of the novelty-resonance space, so that for example the green
block lying on the dashed line in Figure 1.25b at the point where novelty and resonance both
equal 12 represents the fact that four documents had resonance and novelty scores between
12 and 12.333 (the quantization cutoff)65 . Three noteworthy outliers immediately become
apparent: Lenin’s State and Revolution and Trotsky’s The Revolution Betrayed stand out
as the two highest-resonance works across the corpus, while Stalin’s Economic Problems of
the USSR stands out as the lowest-resonance work, despite all three being comparable in
terms of their novelty with respect to previously-published works. The extraordinarily low
resonance score for Stalin’s last major publication before his death sheds light, for example,
on the severe impact the events of 1956 had on subsequent Marxist thought: on top of the
previously-discussed effects of the invasion of Hungary on the global communist movement,
Khrushchev’s “Secret Speech” of the same year—and the newfound possibilities for Com-
munists to criticize Stalin that came as a result—almost instantly destroyed the standing of
Stalin’s thought within this (admittedly Western-biased) corpus of Marxist literature. To see
this, note that Stalin’s earlier works were among the highest-resonance works in the corpus,
with most having a resonance score above zero (i.e., above the dotted line in Figure 1.25a),
with around the same novelty score as Economic Problems.
The massive difference in the impact of this latter work (operationalized through its
resonance scores), then, can plausibly be ascribed to some combination of his death and
the Secret Speech. While more work will need to be done, above and beyond this sketch,
to investigate this hypothesis in more detail (for example, if posthumous works of Stalin
published before the Secret Speech also display this property of extremely low resonance, we
65
In order to eliminate the potential effects of the texts’ lengths on their resulting scores, we partitioned
each text into equal-length chunks (of approximately twenty pages with respect to a standard printed English
translation, so that e.g. Lenin’s State and Revolution is split into exactly five chunks).

80
have evidence in favor of his death as the more impactful event with respect to the cessation
of Stalinist lines of thought among Western Marxists), it hopefully serves as another example
of how computational tools can aid—without replacing the role of humans—in text-focused
and context-sensitive historical research.

(a) Novelty and resonance scores for each (b) Transience and novelty scores for each
text in our corpus of influential Marxist text in our corpus of influential Marxist
texts. texts.

1.4 Networks of Semantic Influence

Although in previous sections we discussed how embedding spaces allow us to visualize

the lexical-semantic relationships among individual terms, sentences, documents, and au-
thors, the algorithms for constructing them operate only at the level of individual terms
and sentences, such that information is lost when the resulting term or sentence embeddings
are coarse-grained up to the document or authorial level. While we are many years away
from embedding algorithms that can directly incorporate and model semantics above the
sentence level66 , we are able build on recent work in cultural analytics which combines the
66
Even contextual sentence-level embedding algorithms are still in their infancy: while word embeddings
have been in wide use since 2014 (after Mikolov et al. 2013b, the canonical citation for word embedding-based

81
lexical-semantic modeling power of embeddings with explicit models of authorial influence,
allowing us to perform more sophisticated author-level comparisons than would be possible
through (e.g.) comparison of mean-pooled sentence embeddings.
In particular, we adopt the approach described in Soni et al. 2021 to construct a set of
Semantic Leadership Networks which directly model dyadic-temporal influence among major
sources of nineteenth-century European socialist thought. For example, to analyze the impact
of pre-1848 socialist thought on that of the post-1848 era, we can consider the network in
Figure 1.26, where each edge represents the hypothesized lexical-semantic influence of one
node before 1848 on the other after 1848.

Socialist Newspapers

Marx Bakunin

Lassalle Proudhon

Other European Socialists

Figure 1.26: A PGM where we’ve observed the person’s action (a = Go) but only have a
probability distribution over the weather w.

Then to estimate the weights on these edges, i.e., the degree of lexical-semantic influence
between a dyad of authors, we start by computing the “lead” of one author a1 over another
a2 with respect to the movement of their respective embeddings for a particular word w via
the following equation, a simplified version of equation (4) from Soni et al. 2021:

preaw1 · postaw2
Leada1 →a2 (w) = ,
preaw2 · postaw2

where the numerator represents the similarity between a1 ’s pre-1848 embedding for w and
works, showed their immense effectiveness on a wide range of NLP tasks), with libraries for their construction
available in nearly all programming languages by 2016, the first contextual sentence-level embedding library
was not made available until 2020 (a year after the canonical publication, Reimers and Gurevych 2019), and
at the time of writing no libraries are available for languages besides Python.

82
a2 ’s post-1848 embedding for w (both normalized to be of length one, an assumption we
relax for the following definitions), and the denominator represents the similarity between
a2 ’s pre- and post-1848 embeddings for w. In other words, considering our word similarity
function sim(·, ·) described in the previous section, we compute

sim(preaw1 , postaw2 )
Leada1 →a2 (w) = .
sim(preaw2 , postaw2 )

As outlined in more detail in Soni et al. 2021, this quantity operationalizes our concept of
author a1 ’s influence on author a2 over a given time span: it increases in proportion to how
much a2 ’s post-1848 usage of w is “pulled” towards a1 ’s pre-1848 usage. For example, Figure
1.27 visualizes a situation where a2 ’s post-1848 usage of a word w is drawn away from their
pre-1848 usage and towards a1 ’s pre-1848 usage: intuitively, this can be thought of as a2
being “convinced” to use w in the manner they observed a1 using it, thus dropping their
original pre-1848 conception of w over time.

0.8 preaw2
Hegelian

0.6

0.4
postaw2
0.2 preaw1 postaw1

0
0 0.2 0.4 0.6 0.8 1
Political Economic
Figure 1.27: The situation of one-way influence, where Leada1 →a2 (w) > Leada2 →a1 (w), allow-
ing us to infer that a1 ’s usage of w before 1848 had more influence on a2 ’s usage over time
than vice-versa.

Mathematically, assuming our embedding vectors are all in R[0,1]×[0,1] , we can obtain a

83
similarity metric sim that scales from 0 to 1 via the transformation

∥v2 − v1 ∥
sim(v1 , v2 ) = 1 − √ ,
2

where ∥v2 − v1 ∥ represents the Euclidean distance between the embedding vectors v1 and
√
v2 , and where 2 represents the maximum Euclidean distance possible between two points
in R[0,1]×[0,1] . Then we can compute the Euclidean distances as:

a2
q √
∥postw − preaw1 ∥ = (0.6 − 0.6)2 + (0.3 − 0.2)2 = 0.12 = 0.1,
a2
q q √
∥postw − preaw2 ∥ = (0.6 − 0.2)2 + (0.3 − 0.8)2 = (0.4)2 + (0.5)2 =
0.36 = 0.6,
a1
q q √
∥postw − preaw2 ∥ = (0.8 − 0.2)2 + (0.2 − 0.8)2 = 0.62 + (−0.6)2 = 0.36 + 0.36
√
= 0.72 ≈ 0.849,
a1
q √
∥postw − preaw1 ∥ = (0.8 − 0.6)2 + (0.2 − 0.2)2 = 0.22 = 0.2,

and thus compute our leadership scores for this situation as:

a a
∥postw2√−prew1 ∥ 0.1
sim(preaw1 , postaw2 ) 1− 2
1− √
2 0.929
Leada1 →a2 (w) = = a a = 0.6 ≈ ≈ 1.613,
sim(preaw2 , postaw2 ) 1− ∥postw2√−prew2 ∥ 1− √
2
0.576
2
a a
∥postw1√−prew2 ∥ 0.849
sim(preaw2 , postaw1 ) 1− 2
1− √
2 0.340
Leada2 →a1 (w) = = a a = 0.2 ≈ ≈ 0.400,
sim(preaw1 , postaw1 ) 1− ∥postw1√−prew1 ∥ 1 − 2
√ 0.859
2

so that we obtain the result that a1 ’s influence on a2 in this case was about 4 times stronger
than a2 ’s influence on a1 .
Importantly, however, the Lead function also allows us to identify instances of mutual
influence between a1 and a2 . In the most extreme case, visualized in Figure 1.28, a1 ’s usage
of w is influenced by a2 ’s usage to such an extent that the former’s post-1848 usage exactly
matches the latter’s pre-1848 usage and, with maximal influence also flowing from a1 to a2 ,
a2 ’s post-1848 usage exactly matches a1 ’s pre-1848 usage. In other words, a1 and a2 have
switched places—both authors drop their pre-1848 usage of w and instead adopt the other

84
author’s usage.

1
preaw1
0.8 posta2
w

Hegelian 0.6

0.4
postaw1
0.2 preaw2

0
0 0.2 0.4 0.6 0.8 1
Political Economic
Figure 1.28: The situation of mutual influence, where Leada1 →a2 (w) = Leada2 →a1 (w), since
a1 ’s post-1848 usage of w exactly matches a2 ’s pre-1848 usage, and a2 ’s post-1848 usage
exactly matches a1 ’s pre-1848 usage.

In this case, given the values shown in Figure 1.28, we can compute our Euclidean dis-
tances as

q
∥postaw2 − preaw1 ∥ = (0.2 − 0.2)2 + (0.8 − 0.8)2 = 0,
q q √
∥postaw2 − preaw2 ∥ = (0.2 − 0.8)2 + (0.8 − 0.2)2 = (−0.6)2 + 0.62 = 0.72 ≈ 0.849,
q
∥postaw1 − preaw2 ∥ = (0.8 − 0.8)2 + (0.2 − 0.2)2 = 0,
q q √
∥postaw1 − preaw1 ∥ = (0.8 − 0.2)2 + (0.2 − 0.8)2 = 0.62 + (−0.6)2 = 0.72 ≈ 0.849,

so that our leadership scores in this case are:

a a
∥postw2√−prew1 ∥ √0
sim(preaw1 , postaw2 ) 1− 2
1− 2 1
Leada1 →a2 (w) = a = a a
∥postw2√−prew2 ∥
≈ 0.849 ≈ = 2.5,
sim(preaw2 , postw2 ) 1− 1− √
2
0.400
2
a a
∥postw1√−prew2 ∥ √0
sim(preaw2 , postaw1 ) 1− 2
1− 2 1
Leada2 →a1 (w) = a = a a
∥postw1√−prew1 ∥
≈ 0.849 ≈ = 2.5,
sim(preaw1 , postw1 ) 1− 1− √
2
0.400
2

85
and thus we see that indeed the leadership score is able to capture the perfect mutual
influence between the two authors in this case.
Given this intuition around the leadership score with respect to a word, the process of
aggregating these scores up to the level of documents and authors becomes straightforward:
we take, for example, the network from Figure 1.26 above and assign edge weights such that
the weight on the edge from a1 to a2 is the number of times a1 led a2 , normalized by the
total number of times a1 led any author. Mathematically, we have

1[Leada1 →a2 (w) > Leada2 →a1 (w)]

P
w∈V
Weight(a1 → a2 ) = P ,
w∈V 1[Leada1 →a (w) > Leada→a1 (w)]
P
a∈A\{a1 }

where V represents our vocabulary, i.e., the set of all words in the corpus, A represents the
set of all authors in the corpus (so that A \ {a1 } represents the set of all authors in the
corpus besides a1 ), and 1[P ] is an indicator function, equal to 1 when the predicate P is true
and equal to 0 otherwise.
Once we have these weights assigned, the network analysis literature provides several
algorithms for computing the influence of one node over others in the network. In our case,
following Soni et al. 2021, we use the Pagerank algorithm, so that the set of influence scores
is the solution to the system of linear equations defined by

X
Influence(ai ) = α Weight(ai → aj ) · Influence(aj ) + β,
aj ∈A\{ai }

where α and β are set to be 0.85 and 0.15/|S| respectively, and |S| represents the number
of authors in the network, following Soni et al. 2021.
With this algorithm for constructing Semantic Leadership Networks we thus have a pow-
erful tool, rooted in linguistically-principled microfoundations, for measuring the relative
influences of the authors in our corpus of 19th-century socialist thinkers. In the next two
sections we introduce this corpus itself, arguing that it captures a wide swath of 19th century
socialist discourse and thus is sufficiently rich for our methods to infer characteristics of this

86
discourse as a whole.

1.5 Conclusion

We began by introducing the hermeneutic background and inferential principles of the

Cambridge School of historical analysis, and brought them into conversation with the com-
puter science literature on contextual word embeddings. We showed how the latter not only
implicitly implements the Cambridge School methodology in a quantitative form but is in
fact explicitly linked to it through a shared history, which bifurcated into computational-
linguistic and linguistic-philosophical strands in the 1950s, losing their shared language to
an extent in the intervening years.
After describing these implicit and explicit connections, we dove more deeply into the
details of how exactly computational embedding algorithms capture the contextual relation-
ships between words: by constructing a geometric space mapping each word to a point such
that points closer together share more contexts, or (in the case of the more recent BERT-style
models) mapping each word to a set of points capturing different senses in which the word
is used while still retaining the semantic meaning of distances between points.
We then took a quick detour through structuralist linguistics to explain the fundamental
distinction between synchronic and diachronic modes linguistic analysis, to point out how
these algorithms remain within the synchronic mode (modeling language as a single, static
geometric space), and thus need to be paired with a diachronic model to constitute a suffi-
cient tool for social-scientific analysis of language. Considering influence as the fundamental
diachronic relationship of interest to historians of political thought, we reviewed theories and
models of intellectual influence with the goal of identifying weaknesses in the standard Cam-
bridge School approach (that is, the approach described in Skinner 1969), and constructed
a new model able to overcome these weaknesses.
Next, to preview how this model can be employed towards fruitful analysis of historical
texts, we walked through a series of introductory case studies, each one highlighting a par-

87
ticular feature of the model which our studies in the remaining chapters will take advantage
of. In the mining-influence-claims case study, for example, we demonstrated a data-driven
method for constructing an ontology of influence claims in the format of our model, focusing
throughout on how assumptions regarding this ontology (for example, that Collaboration is
a stronger influential force than Shared-Geography) can be transparently encoded—and then
straightforwardly changed—as parameters within the model.
Finally, in the previous section, we introduced the Semantic Leadership Network model,
which explicitly combines the synchronic embedding algorithms described in Section 1.2
and the diachronic influence model described in Section 1.3 by constructing networks of
embeddings where each node represents the embedding of author a’s writing at time t and
each edge is a connection between a time-t node and a time-t+1 node, representing a potential
instance of temporal influence. We concluded this section with the final three pieces of the
puzzle, namely, (a) the mathematical formulation Leada1 →a2 (w), which measures author a1 ’s
influence on author a2 by way of their respective embeddings for a word w; (b) the method of
edge-weighting by which the network of word-level Leada1 →a2 (w) scores is coarse-grained up
to the level of author × time pairs (in our nomenclature, a network of Lead∗a1 →a2 scores); and
(c) the use of the PageRank algorithm and its variants to derive our final author-influence
scores.
By laying this groundwork—in particular, by walking through how the models emerge
naturally from philosophical and social-scientific considerations, not from arrogant preten-
sions of automating or “solving” the process of historical inquiry—we hope to have preemp-
tively quelled anxieties which may arise in the remaining chapters regarding what assump-
tions we’re hiding in the machinery of the models, or what types of conclusions we are and
are not comfortable drawing from their outputs. If we are successful in this endeavor, we
hope in turn to have nudged the discussion around works of computational social science
slightly away from these methodological fears and towards methodological possibilities for
understanding political thought.

88
1.A Graphical Models

In this Appendix we describe the details of the framework within which our theories of
influence are specified, known as Probabilistic Graphical Models. To begin with, however,
we describe the more basic notion of Graphical Models, building towards our discussion of
Probabilistic Graphical Models in the next section. When trying to understand a complex
historical phenomenon one which involves lots of “moving parts”—people, institutions, so-
cial relationships, and other related events—interacting to produce the event of interest,
researchers are faced with the daunting task of where to begin their investigation. In this
study we are indeed confronted with such a challenge, as we aim to understand the dynamics
of a decidedly complex set of historical phenomena: the intellectual origins of Marxism and
its subsequent spread across Europe and, later, the Third World.
Just as a mountain climber employs “rules of thumb” in deciding where to start their
climb, in our study the following rule of thumb determines the course our investigation: that,
to understand a complex historical phenomenon, the most fruitful way to make progress is
to (a) break it down into its constituent elements, and then (b) specify how these elements
work together to produce the phenomenon. This “decompositional” framework, we argue,
corresponds naturally to a particular statistical framework known as Probabilistic Graphical
Modeling. As we will argue below, this approach provides a method for transforming our in-
tuitions regarding events and their interrelationships into rigorous, measurable, and testable
statistical hypotheses.
A Probabilistic Graphical Model (PGM) is a statistical tool which, when paired with the
laws of statistical inference, formalizes the decompositional framework we outlined above.
Concretely, a PGM is a collection of nodes (drawn as circles), representing variables, and
edges (drawn as arrows), representing relationships of influence between nodes—relationships
which are then codified numerically in the form of “Conditional Probability Tables”. This
framework provides a level of abstraction that makes it invaluable for historical research, in

89
that a wide range of historical-analytical frameworks can be subsumed (and thus formalized
and tested) within it. Putting the “probabilistic” aspect aside for a moment, we can consider
using a graphical model to begin a historical inquiry into the origins and eventual trajectory
of the French Revolution, for example:

• “Mainstream” historians can begin modeling the event by considering nodes for the
Nobility, the Clergy, and the Third Estate, and edges for their pairwise interactions,
as in Figure 1.29.

• Marxist historians can instead consider the nodes to be the Bourgeoisie, the Proletariat,
and the Peasantry, with edges again representing pairwise interactions, as in Figure
1.30.

• “Great Men” theorists can model prominent individuals as nodes: one for Louis XVI,
another for Robespierre, a third for Napoleon, and so on, as in Figure 1.31. In this
case, we see that not all pairs of nodes have edges between them, since Louis XVI and
Napoleon did not (at least, in a standard interpersonal sense) interact as part of the
Revolution.

Nobility Clergy More Social Power

Third Estate Less Social Power

Figure 1.29: A graphical model illustrating the initial decomposition which a historian of
the French Revolution might perform to make their analysis more tractable: three nodes
representing three historical entities—the Nobility, the Clergy, and the Third Estate—and
edges representing the interrelationships between each pair of entities.

As these figures demonstrate, graphical models provide the researcher with a vast set
of modelling possibilities: nodes can represent people, classes, institutions, or other units of
observation, while edges can represent interpersonal interactions, economic relations, or any

90
 Bourgeoisie Landed Gentry More Social Power

Proletariat Peasantry Less Social Power

Urban Rural

Figure 1.30: A graphical model illustrating the initial decomposition which a Marxist his-
torian of the French Revolution might perform to make their analysis more tractable:
three nodes representing three historical entities—the Bourgeoisie, the Proletariat, and the
Peasantry—and edges representing the interrelationships between each pair of entities.

Louis XVI Mirabeau Robespierre P. Barras Napoleon

Marie Antoinette A. Barnave Mme Roland G. Babeuf

Time

Figure 1.31: A graphical model illustrating the initial decomposition which a “Great Man”
theorist might perform to make their analysis of the French Revolution more tractable,
with nodes for prominent individuals and edges between individuals who are known to have
interacted.

other dyadic relationships between nodes67 . In addition, given the graphical representation,
the position of the nodes in 2D space can also be used to represent divisions or axes of
differentiation: in the case of Figure 1.30, for example, moving left to right represents moving
67
In fact, although the probabilistic aspect of PGMs—the ability to derive statistical inferences regarding
the nodes and their interactions—requires edges to be dyadic, for the purposes of non-probabilistic modeling
one can introduce “hyperedges” representing a relationship between any number of nodes. The Clandestine
Cell Structure of modern insurgent groups, for example, can be modeled effectively using hyperedges, since
it is the holistic relationship among all members of a cell, rather than the individual dyadic relationships
between them, that is most relevant to the functioning of the structure:

Network Leader

Branch Leader Branch Leader

Alice Bob Bruce Derek Don

Arturo Andrew Bill Barry Betty Chris Cathy Dana Drew

91
from urban-situated to rural-situated entities, while moving from bottom to top represents
moving in the direction of increased social power. Similarly, in Figure 1.31, moving left to
right represents the passage of time (with nodes positioned based on when, in the course of
the revolution, their role was most prominent), and the nodes are divided vertically into two
groups based on whether or not the individual was the main political ruler of France at a
given time. With only a few nodes this spatial organization may seem like more trouble than
it’s worth, but it becomes immensely helpful as the model becomes more and more complex.
For example, if the Marxist researcher aimed to subsequently model the Soviet subdivision
of the Peasantry into bednyak, serednyak, kulak, and batrak classes, the batrak class could
be placed somewhere in the middle of the rural-urban axis, as members of this class often
migrated seasonally between urban and rural work.
As these examples show, Graphical Modelling provides a powerful tool for developing ex-
planatory theories of historical phenomena, in forcing the researcher to think simultaneously
about (a) what the relevant entities are and (b) what types of relationships exist among
them. Thus far, however, we have left out the “probabilistic” aspect of the model we will
eventually use throughout our studies, by remaining agnostic about what particular types of
relationships are admissible in these models. Now that we’ve shown the usefulness of general
Graphical Modelling for theory development, in the next section we show—by way of an
introduction to text-analytic topic models—how Probabilistic Graphical Modelling further
enhances the researcher’s toolkit, by enabling principled statistical inferences to be drawn
regarding the interrelationships between nodes in a Graphical Model.

1.B Probabilistic Graphical Models

While in the previous section we restricted our examples to general historical modeling,
in this section we turn to the use of textual data in conducting a historical study, and in the
process show how the probabilistic aspect of the Probabilistic Graphical Model allows us to
reason in a statistically-principled manner about what a given corpus of texts tells us about

92
the historical phenomena we’re interested in studying.
The first step in a researcher’s journey towards understanding a textual corpus is to de-
velop a rough schema of topics covered in the corpus, to figure out what each text is “about”,
and to group the texts accordingly. This thematic categorization, the act of transforming an
archive of texts into one partitioned into sections, often takes up a massive chunk of research
time and resources. Given this resource bottleneck, this transformation is precisely what one
of the first text analysis methods, Latent Semantic Analysis (LSA), was created to do, and
(somewhat miraculously) it does so without any input required from the researcher besides
having the texts in some digitized format.
Scale-wise, LSA is already leaps and bounds beyond human capabilities, but we can
do better: in 2003, David Blei, Andrew Ng, and Michael I. Jordan developed an extension
to LSA called Latent Dirichlet Analysis (LDA), which “zooms in” on each document and
actually learns distributions over topics for each token (word). That is, while LSA places each
document into a single category, LDA derives a more detailed summary of each document,
like “25% of this document is about computer science and 75% is about linguistics,” a more
realistic model given the tendency for most written documents to range across multiple topics
(for example, a news article introducing a new technology and then discussing its potential
societal impact). In fact, if a researcher does want a single category for each document,
they can simply choose the topic with the highest proportion: linguistics, in the case of our
example document.
For a researcher hoping to study taxation practices in ancien régime France, for example,
this means the difference between reading through every text in the archive and reading only
the specific subset of the texts which are known to discuss the topic of taxation. Depending
on the time and resources available to the researcher, for example, they can read the N
documents with the highest proportions devoted to taxation: the more resources are available,
the higher this N can be.
Both LSA and LDA fall into the category of “unsupervised” algorithms, given the lack

93
of user intervention in the topic-learning process. While this approach stays true to the idea
of “pure” exploratory data analysis, it is rare for a researcher to have absolutely no idea
what topics lie within an archive. More commonly, researchers come to the texts with a
rough set of topics in mind and want to see which texts fall within these topics, thus shifting
the nature of the research more towards confirmatory data analysis. In this case, they can
use a “semi-supervised” Labeled LDA algorithm like CorEx (Gallagher et al. 2017), which
allows them to “suggest” salient topics to the LDA algorithm before it runs. In our French
ancien régime example, Labeled LDA would allow researchers to suggest single keywords like
“impôts” (taxes) or keyword groups like {“gabelle”, “taille”} (the salt tax and the land tax,
respectively, two of the most onerous ancien régime tax burdens in the eyes of the peasantry),
thus nudging the LDA algorithm towards detecting taxation-related topics.
But how exactly are these algorithms able to detect topics within a corpus without
being given any domain-specific knowledge whatsoever? This seeming miracle is achieved,
in short, via a statistical model of the writing process: since texts tend to be structured—
into sentences, paragraphs, and chapters, for example—in such a way that a given section
(say, a paragraph) focuses on only one or a few topics selected from the much wider set
of topics existing across the corpus, we can infer that the more a pair of words co-occurs
within these sections the more likely they are to pertain to the same topic. At a high level,
therefore, we can scan over a corpus and discover its constituent topics simply by identifying
clusters of words that tend to occur close to one another. A topic then, in this sense, is
“coherent” to the extent that its words are more likely to co-occur than we would expect by
“random chance”—i.e., if the words were randomly selected one-by-one from a dictionary.
This captures the intuition that, for example, since the Wikipedia article for Thomas Hobbes
is about a particular topic, words which occur within this article (choosing eight at random,
programmatically) are more likely to be about the same topic than words chosen at random
across all words in all Wikipedia articles (again choosing eight at random, programmatically),
as illustrated in Table 1.3.

94
 Thomas Hobbes Karl Marx All Wikipedia
war economic wolf
civil rebellions chondrite
consent converted bawdy
among mode larynx
patriarchalists right-wing perchance
history countries cytosine
enter struggle pollinate
government industrialisation negligible

Table 1.3: Randomly-chosen words from Wikipedia articles on Thomas Hobbes and Karl
Marx, versus randomly-chosen words from across all Wikipedia articles

Concretely, for reasons which will be elucidated in the next section, we can transform this
intuition into a computationally-estimable model by specifying a “data-generating process”—
essentially a stylized story of how the author “chose” the sequences of words which appear
in the texts. For example, previewing what is to come, Algorithm 1 represents the basic
data-generating process underlying all topic models.
Despite the simplicity of this model, its definition of a topic as a grouping of words
precisely captures the word co-occurrence intuition from above: rather than choosing words
at random, as in the All Wikipedia column of Table 1.3, the author first chooses a topic, which
thus establishes the likelihood that particular words are chosen. In this way, continuing our
example, choosing the Thomas Hobbes topic increases the likelihood that the term “civil” is
chosen to be the next word in the text (relative to its frequency in the English language in
general), whereas choosing the Karl Marx topic would instead make “industralisation” more
likely to be chosen (again, relative to its overall frequency in English).
Once this data-generating process is specified, all that is left to do for a computer to
discover the latent topics within a corpus is to “run it backwards”: in essence, to divide
the words in the texts up into clusters which best fit this story. Putting “chondrite” and
“cytosine” (the first being a type of meteorite, the second a nucleotide found in DNA) into
the same cluster, for example, would not fit the story well, since these terms will rarely co-
occur in the same section of a text. If it instead placed “chondrite” into a cluster with other

95
Algorithm 1 Data-Generating Process for a Text Corpus

1. Choose hyperparameters. The author decides how many documents N they’d like to
write, as well as a set of K topics they’d like to write about in these texts. We assume
the author has a vocabulary of M words V = {w1 , w2 , . . . , wM } to draw on while
writing.

2. Specify topics. For each topic t ∈ {t1 , t2 , . . . , tK }, the author evaluates which words
in V are most and least pertinent. For example, if the topic is astronomy, “planet”,
“moon”, and “orbit” will be given high scores, while “koala” and “headache” will be
given low scores. Each topic t is thus assigned a corresponding set of importance scores
for each word: pt (v) represents the importance to topic t of word v. In our example,
the author may choose pt (planet) = pt (moon) = pt (orbit) = 0.3, and pt (koala) =
pt (headache) = 0.05.

3. Specify documents. For each document d ∈ {d1 , d2 , . . . , dN }, the author chooses

(a) how many words Nd they’d like to write for this document, thus creating blank
slots Sd = {sd,1 , sd,2 , . . . , sd,Nd }, and
(b) the distribution of topics θd they’d like to cover in this document. For exam-
ple, if d is an article on the Franco-Prussian War, they might choose θd to be
a balance between three topics: 40% France, 40% Prussia, and 20% warfare.
Mathematically, if the global set of topics T chosen in Step 1 was t1 = France,
t2 = Prussia, t3 = warfare, t4 = Buddhism, this balance would be represented by
θd = (0.4, 0.4, 0.2, 0.0).

4. Fill the documents with words. For each slot sd,i in a document d, the author chooses
a topic for this slot td,i based on the document’s topic distribution θd (the distribution
chosen in the previous step). They then choose a word w from their vocabulary V to
fill in this slot, with the probability of choosing a specific word wj proportional to that
word’s topic-td,i importance score ptd,i (wj ).

astronomy terms and “cytosine” into a separate cluster of microbiological terms, this would
better explain the tendency for words in the former to co-occur in texts about astronomy,
and for words in the latter to co-occur in texts about microbiology.
The fact that a computer is able to estimate this “best” division in mere seconds (i.e.,
without trying every single possible split of the vocabulary into separate word clusters, which
would be prohibitively time-consuming and computationally expensive) is due to the magic
of Probabilistic Graphical Models (PGMs), algorithms which allow researchers to specify the

96
parameters of their data-generating processes—any data-generating processes which adhere
to a loose set of constraints—and nearly instantly obtain estimates for the parameters within
these specifications. In the case of general topic models, for example, the basic parameters
would be (a) the distribution of topics across the corpus, and (b) the distribution of words
within each topic. By restricting the total number of topics K that we want the algorithm to
generate, we force it to choose a small set of coherent topics, based on the K word clusters
with the strongest co-occurrence patterns. In the next section we introduce the specific form
that these Probabilistic Graphical Models take, and walk through an example of how they
can be constructed around a set of hypotheses.

Figure 1.32: From Blei 2012, p. 3

Every such PGM is simply the formal mathematical representation of a data-generating

process, such as the text-generation process described above in Algorithm 1. So, if we wanted
to model the relationship between weather and a person’s choice of whether to go out and
party or stay in and watch a movie on a given Saturday evening, we could begin by proposing
the data-generating process given in Algorithm 2.
Now, given the description of a PGM given above (nodes as variables, edges as relation-

97
Algorithm 2 Data-Generating Process for Evening Plans

1. The person P looks out the window and observes the weather.

2. If the weather is sunny, P goes out to a party. Otherwise, P stays in and watches a
movie.

ships between variables), we can perform the move alluded to in the previous section: we can
convert our data-generating process into a PGM, by defining nodes (variables) and edges
(relationships) as follows:

1. A variable w which can take on values in {Sunny, Rainy}

2. A variable a which can take on values in {Go Out, Stay In}, and

3. An edge e from w to a which encodes the intuition that one is more likely to go out if
it’s sunny than if it’s rainy via the probability distribution P (Go Out | Sunny) = 0.75,
P (Stay In | Sunny) = 0.25, P (Go Out | Rainy) = 0.25, and P (Stay In | Rainy) = 0.75.

The resulting PGM, in graphical form68 , is presented in Figure 1.33, where the Condi-
tional Probability Table describing the edge from the w node to the a node is given in Table
1.4.

w a

Figure 1.33: A basic PGM, representing the relationship between w, the weather, and a, the
subsequent action of a person deciding whether to go out or stay in for the night.

PGMs can help us make inferences about the world in the face of incomplete information,
which is the situation in nearly every real-world problem. The key tool here is the separation
of nodes into two categories: observed (represented graphically as a shaded node) and latent
68
The “Graphical” in Probabilistic Graphical Model is not used in the same sense as the “graphical” we’re
used to from vernacular English. Capital-G Graphical denotes that the Probabilistic Model is represented
as a Graph, a well-defined mathematical object consisting of nodes and edges, which does not have to be
represented graphically (though it could be, like in our example here with circles and arrows). In fact, when
a computer program is estimating a PGM, it is by definition not in a graphical form—it’s in the form of 0s
and 1s, stored in the computer’s memory.

98
 Go Out Stay In
Sunny 0.75 0.25
Rainy 0.25 0.75

Table 1.4: The Conditional Probability Table for the PGM shown in Figure 1.33.

(represented graphically as an unshaded node). Thus we can now use our model as a weather-
inference machine: if we observe that the person we’re modeling is out at a party with us,
what can we infer from this information about the weather outside? We can draw this
situation as a PGM with shaded and unshaded nodes, as in Figure 1.34, and then use Bayes’
Rule to perform calculations over the network, to see how the observed information about
the person at the party “flows” back into the node representing the weather.

w a

Figure 1.34: A PGM representing the same situation as in Figure 1.33, except that the node
for variable a is now shaded, indicating a situation where we have observed the person’s
action (a = Go Out) but still only have a probability distribution over the weather w.

Keeping in mind that Bayes’ Rule tells us, for any two events A and B, how to use
information about P (B|A) to obtain information about P (A|B):

P (B|A)P (A)
P (A|B) = ,
P (B)

We can now apply this rule to obtain our new probability distribution over the weather,
taking into account the new information that the person has chosen to go out:

P (a = Go Out | w = Sunny)
P (w = Sunny | a = Go Out) =
P (a = Go Out)
P (a = Go Out | w = Sunny)
=
P (a = Go Out | w = Sunny) + P (a = Go Out | w = Rainy)

And now we simply plug in the information we already have from our conditional prob-

99
ability table to obtain our new (conditional) probability of interest:

(0.8)(0.5) 0.4 0.4

P (w = Sunny | a = Go Out) = = = ≈ 0.89.
(0.8)(0.5) + (0.1)(0.5) 0.4 + 0.05 0.45

We have learned something interesting: now that we’ve observed the person out at a
party, the probability that it is sunny out jumps from 0.5 (called the “prior” estimate of w,
i.e., our best guess without any other relevant information) to 0.89 (called the “posterior”
estimate of w, i.e., our best guess after incorporating relevant information).
Turning back to our analyses of the cahiers de dolèances in the next chapter, we can
now walk through the construction of our topic model in a similar fashion: note that, in the
previous example, we took information about an observed quantity—the presence or absence
of someone at our party—and used it to draw inferences about an unobserved quantity—
the weather. Analogously, for our topic model, we consider the words in the text to be our
observed data, and try to use this observed data to draw inferences about the unobserved
topics underlying the choice of words in a given text. Just as we used the presence of someone
at the party to infer a higher likelihood that it’s sunny out, our topic models will use the
presence of particular words in a cahier—say, “droits”, “liberté”, “égalité”—to infer a higher
likelihood that this grievance pertains to a particular topic—in this case, a topic involving
the concept of “individual rights”.
Given these parallels between our previous example and the basic structure of a topic
model, then, we can start to model the data-generating process for a single word in the text,
as illustrated in Figure 1.35. Note that it is identical to our example PGM in Figure 1.34,
except for the labeling of the variables: the node which represented the observed action a
now represents the observed word wd,i (the ith word in document d), and the node which
represented the unobserved weather w now represents the unobserved topic td,i .
However, by comparing this PGM with our full data-generating process in Algorithm 1
above, we can see that the former is not sufficient for representing the mechanics of this

100
 td,i wd,i

Figure 1.35: A first attempt at a PGM representing the data-generating process for an
observed word wd,i , the ith word in document d, within a text.

process. Whereas in the evening-plans model we assumed specific values for the conditional
probabilities of each action—for example, that P (Go Out | Sunny) = 0.75—in the topic
modeling case we want to estimate (rather than assume) the probabilities of words within
each topic itself. In other words, we now have two “layers” of unknown quantities: we want
to estimate both the document-topic distributions θd and the topic-word distributions pt (w).
The power of modelling via PGMs becomes apparent here since, just as we were able
to split the full data-generating process in Algorithm 1 into steps, we can start building
towards the full PGM by constructing “sub-PGMs”, each corresponding to one step in the
data-generating process. To start with, we can consider what the PGM for the Specify-
Documents sub-process would look like, and arrive at a sub-PGM like the one illustrated in
Figure 1.36.
The node for the observed word w is in dotted rather than solid outline here to denote
the fact that this PGM is incomplete on its own: it represents the step in which the author
specifies the relationships represented by edges here, i.e., specifies the likelihood of choosing
a given word for each of the K topics. An example of such a word-topic specification, which
would complete the definition of this sub-PGM, is given in Table 1.6. But the dotted-outline
w node here is not the same as the eventual observed-word node of our full data-generating
process, since in this full process the generation of a word depends on both the likelihood of a
word for each topic (as modeled here) and the particular topic which was chosen for the slot
in which the word will appear. In other words, it is only when the author has access to both of
these pieces of information—when they have (a) the Conditional Probability Table specifying
how to choose a topic for a given slot, and (b) the Conditional Probability Table specifying
how to choose a word from that chosen topic—that they can proceed to generating the word
itself. To address this incompleteness, we can separately model the process by which the topic

101
for each word-slot is chosen, the Specify-Documents sub-process of our full data-generating
process, as illustrated in Figure 1.36.

θd

td,1 td,2 ··· td,Nd −1 td,Nd

wd,1 wd,2 ··· wd,Nd −1 wd,Nd

Figure 1.36: A PGM representing the Specify-Documents sub-process, in which the author
specifies a topic distribution θd .

θd td,i wd,i

j = 1 . . . Nd
d = 1...M

Figure 1.37: A PGM representing the Specify-Documents sub-process, using plate notation
to condense the repetition in Figure 1.36.

Document
Topic Document 1 Document 2 · · · Document M
Astronomy 0.80 0.25 ··· 0.10
Marsupials 0.20 0.25 ··· 0.00
Music 0.00 0.50 ··· 0.00
Vegetables 0.00 0.00 ··· 0.90

Table 1.5: An example Conditional Probability Table for the Specify-Documents PGM shown
in Figures 1.36 and 1.37.

In words, this PGM diagram tells us that given a choice of topic td,i chosen for slot
sd,i , in conjunction with a choice of word-importance scores pt (w) for all words w relative
to topic t, we can determine exactly the probability of the word wd,i appearing in slot i

102
 p1 p2 ··· pK−1 pK

Figure 1.38: A PGM representing the Specify-Topics sub-process, in which an author specifies
a word importance score pt (w) for each word w and topic t.

pi w

i = 1, . . . , K

Figure 1.39: A PGM representing the Specify-Topics sub-process, using plate notation to
condense the repetition in Figure 1.38.

Topic
Word Astronomy Marsupials Music Vegetables
planet 0.30 0.03 0.01 0.06
moon 0.30 0.03 0.01 0.06
orbit 0.30 0.03 0.01 0.06
koala 0.02 0.41 0.01 0.06
possum 0.02 0.41 0.01 0.06
guitar 0.02 0.03 0.93 0.06
spinach 0.02 0.03 0.01 0.32
broccoli 0.02 0.03 0.01 0.32

Table 1.6: An example Conditional Probability Table for the Specify-Topics PGM shown in
Figure 1.38.

θd td,i wd,i pi

j = 1 . . . Nd i = 1, . . . , K
d = 1...M

Figure 1.40: The complete PGM representing the data-generating process for an observed
word wd,i , with the Specify-Documents and Specify-Topics sub-processes incorporated explic-
itly.

of document d. This walkthrough of constructing a PGM from scratch has illustrated two
powerful properties of PGMs that aided our modelling process: on the one hand, we have seen

103
how factoring the full data-generating process into sub-processes allowed us to ignore the
details of e.g. the Specify-Topics subprocess while we thought through the Specify-Documents
subprocess, and then gave us a natural way to join these parts together into a fully-specified
model of the whole (by joining the two PGMs at the w node, which both sub-processes
had only partially specified). Returning to our discussion of complex historical and social
processes in the beginning of the chapter, one can see how effective this factoring-and-
rejoining method could be for rendering studies of complex phenomena more manageable:
a team of researchers could establish a division of labor, with each individual researcher
modeling their particular sub-phenomena of interest or expertise via PGMs, then join their
respective PGMs at particular nodes of intersection. For example, in the French Revolution
case, a node representing the text of debates in the National Assembly could sit at the
intersection not only of individual models of the three estates, but also a model of the
Parisian crowd who often swayed these debates via cheering or booing from the gallery (a
dynamic which George Rudé’s The Crowd in the French Revolution (Rudé 1959) brought to
the forefront of French Revolution research69 ), as illustrated in Figure 1.41.

Nobility Clergy

Legislative
Assembly

Third Estate Crowd

Figure 1.41: A PGM representing the fusion of several individual models of French Revolu-
tionary entities, to explain the observed outcomes in the revolutionary Legislative Assembly.

1.C Chronology of Major Texts and Events from McLellan 2007

69
See also Rudé’s subsequent work, The Crowd in History Rudé 1964, for an even broader examination of
the role of the crowd in explaining historical social behavior.

104
Year Major Text Major Event
1883 Bebel, Women Under Socialism Death of Marx
Plekhanov, Socialism and the Political Founding of SDF
Struggle
1884
1885 Plekhanov, Our Differences Fabian Society founded
1886
1887 ILP founded
1888 Engels, Ludwig Feuerbach
1889 Founding of Second International
1890
1891 Erfurt Programme Erfurt Congress of SPD
1892
1893
1894 Plekhanov, On the Development of the Succession of Nicholas II in Russia
Monist Conception of History
1895 Engels, Preface to Class Struggles Death of Engels
China-Japan War
1896
1897
1898 Founding of RSDLP
1899 Lenin, Development of Capitalism in Rus- SPD Congress rejects Bernstein
sia
Bernstein, Evolutionary Socialism
Kautsky, Agrarian Question
1900 Luxemburg, Reform or Revolution British Labour Party founded
Boxer Rebellion in China
1901 Founding of Socialist Party of America
1902 Lenin, What Is To Be Done? 2nd Congress of RSDLP
Bolshevik-Menshevik split
1903
1904 Russo-Japanese War
1905 Revolution in Russia
Jena Congress of SPD
1906 Kautsky, Ethics
Luxemburg, Mass Strike
Trotsky, Results and Prospects
1907 Bauer, The Nationalities Question Stuttgart Congress of Second International
1908 Plekhanov, Fundamental Problems of
Marxism
Lenin, Materialism and Empiriocriticism
1909 Kautsky, Road to Power
1910 Hilferding, Finance Capital
1911
1912 Republic proclaimed in China
1913 Luxemburg, Accumulation of Capital
Stalin, National Question
1914 Outbreak of First World War
1915 Lenin, Philosophical Notebooks
Bukharin, Imperialism and the World Econ-
omy

105
Year Major Text Major Event
1916 Lenin, Imperialism
1917 Lenin, State and Revolution February and October Revolutions in Rus-
sia
USA Enters WWI
1918 Treaty of Brest-Litovsk
1919 Bukharin and Preobrazhensky, ABC of Abortive revolution in Germany
Communism
1920 Treaty of Versailles
1921 Bukharin, Historical Materialism Kronstadt Revolt
Beginning of NEP
1922
1923 Trotsky, New Course USSR established
Lukács, History and Class Consciousness Triumph of Fascism in Italy
Korsch, Marxism and Philosophy
1924 Stalin, Foundations of Leninism Death of Lenin
Northern Expedition of Chiang Kai-Shek
1925
1926 Gramsci, On the Southern Question
1927 Mao, Peasant Question in Hunan Trotsky and Zinoviev expelled from CPSU
1928
1929 Gramsci, Prison Notebooks (to 1935) Trotsky expelled from Russia
Start of Great Depression in the West
1930 Collectivization of agriculture in Russia
1931
1932
1933 Accession to power of Hitler
1934
1935 Lukács, Studies in European Realism The Long March
1936
1937 Trotsky, Revolution Betrayed
Mao, On Practice; On Contradiction
1938 Stalin, Dialectical and Historical Material-
ism
1939 Outbreak of Second World War
1940 Mao, On New Democracy
1941 Marcuse, Reason and Revolution
1942 Sweezy, Theory of Capitalist Development Battle of Stalingrad
1943
1944
1945 End of Second World War
1946
1947 Lukács, The Historical Novel
Adorno and Horkheimer, Dialectic of En-
lightenment
1948 Yugoslavia’s break with Soviet Union
1949 Stalin, Marxism and Linguistics People’s Republic of China proclaimed
Adorno et al., Authoritarian Personality
1950 Korean War
1951 Start of McCarthyism in USA
1952 Stalin, Economic Problems of USSR

106
 Year Major Text Major Event
Sartre, Communists and the Peace
1953 Death of Stalin
1954
1955 Marcuse, Eros and Civilization
1956 Krushchev’s Long Speech
Soviet Invasion of Hungary
1957 Baran, Political Economy of Growth
Sartre, Search for a Method
Mao, Correct Handling of Contradictions
1958 Great Leap Forward in China
1959 Cuban Revolution
1960 Sartre, Critique of Dialectical Reason Sino-Soviet Split
1961
1962 Cuban Missile Crisis
1963
1964 Marcuse, One-Dimensional Man Downfall of Krushchev
Togliatti, Testament
1965 Althusser, For Marx
Guevara, Socialism and Man in Cuba
1966 Baran and Sweezy, Monopoly Capital Beginning of Cultural Revolution
1967 Debray, Revolution in the Revolution
1968 Althusser, Reading Capital Soviet Invasion of Czechoslovakia
May events in Paris
1969
1970
1971 Fall of Lin Piao
1972 US withdrawal from Vietnam
1973 Althusser, Essays in Self-Criticism Allende government overthrown in Chile
O’Connor, Fiscal Crisis
1974 Braverman, Labour and Monopoly Capital
1975
1976 Fall of Mao
1977
1978 Cohen, Karl Marx’s Theory of History

1.D Influence Model PGM

107
 Xi,t−1 Xi,t Xi,t+1 ... Xi,T −1 Xi,T

... ET −1
Et−1 Et Et+1

j j j Ri,T −1
Ri,t−1 Ri,t Ri,t+1 ...

j = 1...Ni,t−1 j = 1...Ni,t j = 1...Ni,t+1 j = 1...Ni,T −1

Figure 1.42: The basic model of intellectual influence used to interpret diachronic shifts in
the word embedding spaces generated throughout the study, where each Xi,t node represents
an observed texts published by author i at time t, each Et node represents the collection of
historically-salient non-textual events which occurred at time t, each Ri,t node represents a
particular text read by author i at time t (the superscript j indexing among multiple texts
read within the same time period), and successive time periods are separated by dashed
vertical lines.

108
 Chapter 2: Communism in Context: The Genesis of Marxism,
1776–1848

For the bourgeois it is all the easier to prove on

the basis of his language the identity of
commercial and individual [...] as this language
itself is a product of the bourgeoisie.

Karl Marx and Friedrich Engels, The German

Ideology (MECW 5, p. 231)

Language, while it is a medium of intercourse, is

at the same time an instrument of struggle.

J. V. Stalin, “Marxism and Problems of

Linguistics” (Stalin 1950), p. 31

2.1 Introduction

2.1.1 Overview

How important were the preexisting ideas and concepts of German philosophy, French social-
ism, and British political economy to the formation of Karl Marx’s thought over the course of
the 1840s? And, once he had articulated his resulting critique of capitalism, how was he able
to cement its concepts and vocabulary as the hegemonic language of socialist movements
across Europe, as opposed to those propounded by other popular socialist writers of the era
like Pierre-Joseph Proudhon, Ferdinand Lassalle, and Mikhail Bakunin? What was he “do-
ing” when he intervened in the socialist discourse of the time via public speech acts—books,
articles, speeches, and so on—and how successful was he in this endeavor? In this chapter
and the next, we operationalize these questions in quantitative terms as empirically testable
hypotheses, then leverage recent advances in neural network-based text analysis methods to

109
attempt answers within an empirical rather than purely speculative or interpretive mode.
In the remainder of this section we present an overview of the literature on the devel-
opment of Marx’s thought and its subsequent spread throughout Europe, paying particular
attention to how competing claims within this literature can be translated into testable
propositions and evaluated using the methods described in Chapter 1. We then carry out a
computational decomposition of the texts thought to have influenced Marx’s thought, evalu-
ating their relative importance to its genesis and evolution from his 1836 poetry to his work
on the three volumes of Capital which occupied him from the late 1850s until his death in
1883.
We propose a method for identifying, given a set of representative texts, the contours
of a “discursive field”, by identifying clusters of terms which are most unique to this par-
ticular discourse relative to a larger semantic embedding space (in the studies, for example,
a space capturing the language of 19th-century European writing writ large). We then use
this method to operationalize the notion of how “Hegelian”, “French socialist”, or “political-
economic” an author’s writings are, and find (lending credence to what is commonly assumed
or asserted in studies of nineteenth-century radical political thought) that (a) Marx’s writing
becomes less “Hegelian” as he is exposed to Paris’ brand of working-class-oriented socialism
between 1843 and 1845, and (b) becomes more political-economy-oriented from this point
onwards, especially after his move to London in 1849, where he spent the remainder of his
life.

2.1.2 Background: Sources of Influence on Marx’s Thought

Lenin’s 1913 essay “The Three Sources and Three Component Parts of Marxism” (Col-
lected Works 13, pp. 23–28) is perhaps the most oft-cited assertion that Marx’s thought
developed at the intersection of German philosophy, French socialism, and British politi-
cal economy, but it is certainly not the earliest. Marx himself emphasized the necessity of
combining these regional strands of thought as early as 1844 in an article for Vorwärts!,

110
stating that “the German proletariat is the theoretician of the European proletariat, just as
the English proletariat is its economist, and the French its politician.”1 In fact, several of
Marx’s precursors had voiced a similar sentiment regarding the necessity of trans-national
philosophical and practical development: Ludwig Feuerbach had earlier proclaimed that “the
‘new’ philosophy, if it wished to be at all effective, would have to combine a German head
with a French heart,”2 while Moses Mess, an early convert to Communism (during his time
in Paris) who worked with Marx and Engels in Cologne as part of the Rheinische Zeitung
in 1843, had published a book in 1841 titled Die europäische Triarchie (The European Tri-
archy) (Hess 1841) emphasizing the necessity of precisely the tri-national combination under
consideration here (i.e., thus adding British political economy into the mix).
To this day, for better or worse, historians of 18th- and 19th-century Europe still often
adopt a similar three-way division of labor among regions of Europe as a broad organizing
principle: Great Britain thrusts the epoch of industrialization upon the continent, France
births and exports a slew of radical political ideas and practices, and the Germans observe
from the sidelines, “spinning webs of theory” which ostensibly reveal the metaphysical logic
behind the surface chaos of these sweeping changes. Hence, for example, these historians
often explain Hegel’s theories with reference to his (admiring) reflections on the French Rev-
olution, while the archetypically practical Bentham scoffs at the Rights of Man as “nonsense
upon stilts”. Though this trichotomy papers over the flows of mutual influence among all
three societies3 , it does help point out how Marx’s thought developed at the interstices of
vast global economic and political upheavals: a German whose formative years were spent
theorizing and then participating in revolution, he found himself suddenly washed up on
the shores of Great Britain after in 1849, where he would remain for the rest of his life—
1
Marx, “Critical Marginal Notes on the Article ‘The King of Prussia and Social Reform. By a Prussian’”,
Vorwärts!, No. 64, August 10, 1844. MECW 3, , p. 202.
2
McLellan 1973, p. 64.
3
In fact, even positing something called “Germany” as one of the three poles already brings us into
tendentious territory, as the Germany we know today was only unified in 1871—Hegel was born in Stuttgart
which was then part of the Duchy of Württemberg, for example, while Marx and Engels were both born in
Prussia.

111
theorizing a critique of capitalism, in good German fashion, but this time from within the
British heart of the global capitalist machine.
A vast literature already exists documenting this trajectory, and especially how these
life changes affected aspects of his thought. Hence, while we now review a portion this
literature, we emphasize that our selection of works is biased towards the polemical: we pay
particular attention to a set of key works which sought to change our scholarly and popular
understanding of Marx, whether successful or otherwise. For outlines of this trajectory in
broad strokes, however, one can consult Mehring 1918, Nicolaievsky and Maenchen-Helfen
1936, Berlin 1939, or McLellan 1973, or choose from among a wave of recently published
biographies by prominent Marx scholars: Sperber 2013, Stedman Jones 2016, and Liedman
2018.
In terms of more detailed, analytic treatments of this trajectory, for many decades the
four published volumes of August Cornu’s Karl Marx et Friedrich Engels: leur vie et leur
oeuvre 4 represented the most exhaustive account, though it covered only the period up to
1846. Recently, however, Michael Heinrich’s Karl Marx and the Birth of Modern Society 5 ,
by integrating important developments in Marx scholarship over the decades since Cornu’s
work with his own original research, can lay claim to the title of most exhaustive and up-to-
date biography available. To buttress the contents of these narrative works, one can consult
a key collection of reference works pertaining to Marx’s life, collectively referred to as the
Marx-Engels Cyclopedia 6 . The first volume, the Marx-Engels Chronicle (Draper 1985a) pro-
vides a day-to-day account of Marx’s and Engels’ reading, writing, meetings, travel, and so
on, spanning from Marx’s 1818 birth to Engels’ 1895 death. The second, the Marx-Engels
Register (Draper 1985b), contains a detailed listing of all known writings, cross-referencing
their dates of composition and publication in the Chronicle and listing all known republica-
4
Cornu 1955a, Cornu 1955b, Cornu 1955c, and Cornu 1955d.
5
Still a work in progress at the time of this writing, though the first volume, Heinrich 2018 (covering the
period up to 1841), is already available in English translation as Heinrich 2019.
6
We describe this collection in more detail in Section 2.1.3 below, where we introduce our novel digitization
and extension of these works as part of the Marx-Engels Digital Cyclopedia.

112
tions and translations of each work. The third volume, the Marx-Engels Glossary, contains
an entry for each entity (person, geographic location, historical event, etc.) mentioned across
the first two volumes, with appropriate cross-references. Thus, for example, those interested
in Hegel’s influence on Marx can start with the entry for “Hegel” within this third volume,
to obtain references to all points in the Chronicle where Marx or Engels read or mentioned
Hegel as well as all writings in the Register which mention or refer to Hegel.
For the trajectory of Marx’s thought in particular (and thus only referencing life circum-
stances when they help explain his writings), Elster 1986 and Musto 2020b are particularly
concise overviews which nonetheless manage to cover most key themes (classes, the state,
revolution, etc.), identifying important points of stability and points of change from the 1840s
until Marx’s death in the 1880s. The most rigorous analytical studies of the development
of his thought, however, which therefore most directly aided our construction of hypotheses
and interpretations of our findings, are found in White 1996, Hunt 1974, and Hunt 1984.
Finally, for situating Marx himself within the broader trajectory of Marxism that fol-
lowed, Hobsbawm 2011 provides a succinct but engaging overview, while McLellan 20077 goes
more in-depth, chronologically walking through developments in Marxism from Marx’s death
up through the contemporary work of scholars like Jürgen Habermas, Frederic Jameson, and
David Harvey.
For our goal of translating particular hypotheses about Marx’s life and thought into the
language of computational linguistics, however, we shift our focus away from these descriptive
overviews and towards works which explicitly aimed to alter our understanding of Marx.
Sometimes these took the form of new interpretations based on novel original research and/or
the uncovering of previously-unknown writings and letters, as in Nicolaievsky and Maenchen-
Helfen 1936, which revised our understanding of Marx’s life by incorporating information
from a cache of newly-discovered letters. More recently Bianchi 2018, focusing specifically on
7
Originally a sort of sequel to McLellan 1970, this work has been expanded multiple times since its
publication in 1979 to incorporate emergent strands of Marxist thought since the 1970s, with the most
recent 4th Edition bringing it up-to-date as of 2007.

113
the types of hypothesized influences on Marx which we will explore in the following sections,
challenged previously-held notions of Spinoza’s influence on the young Marx8 . Alternatively,
as in Althusser 1965, Althusser 1968, and Cohen 1978 for example, the intervention came
in the form of an argument for a new framework or interpretation which, if adopted, would
ostensibly allow us to understand Marx’s thought with greater clarity.
In the latter category, for example, Norman Levine’s The Tragic Deception Levine 1975
implores its readers to strictly separate Marx’s thought from that of Engels, arguing that the
former can only be truly understood once the latter’s editorial revisions have been excised
from our corpus. In this formulation we can already sketch the outlines of a computational
approach to this hypothesis, using the synchronic methods described in Section 1.2: if we
run an embedding algorithm on just those writings which were known to be Marx’s alone
(i.e., the writings in their “raw form”, before Engels’ editing), and another on the full corpus
of Marx’s and Engels’ combined works, how much do the resulting spaces differ, and in what
ways? Levine’s argument, for example, places a strong emphasis on how differences in their
backgrounds can be traced forward to eventual differences in their conceptions of key terms:

Unlike Marx, Engels was not influenced by Feuerbachian naturalism and conse-
quently was not only led to define communism differently than Marx, but also
the nature of the State and Bureaucracy. (Levine 1975, p. xv)

To start evaluating the veracity of this argument, then, one could perform a comparison of the
trained embeddings related to the State and Bureaucracy—those for “Staat”, “Bürokratie”,
“Staatsmacht”, for example—across the two embedding spaces. If these points do not shift
8
This extremely thorough investigation, conducted in the same analytic spirit as our own but without the
aid of computational tools, is thereby able to carefully scour Marx’s entire body of work for signs of Spinozan
influence or lack thereof, and thus detect subtle traces of Spinoza’s philosophy in Marx’s statements over
time that our method would miss. While this type of study is thus preferable to a computational approach if
one is interested in the influence of a single figure like Spinoza or Hegel (indeed, Bianchi 2018 and the analysis
of Hegel in Avineri 1968 both served as key models whose rigor we aimed to emulate using computational
tools), the strength of our approach comes from its ability to scale. The labor-intensive “manual” approach
falters when one considers assessing the full range of Marx’s influences: one can imagine, for example, the
amount of work that would be required to perform this type of analysis on just the books Marx read in 1851
alone (285 separate works, as shown in Figure 2.24), much less the 1,000+ books read over the course of his
life, and then to synthesize these individual findings into a coherent whole.

114
very much (relative to, say, the average shift observed across all words between the two
embedding spaces), one has grounds for decreasing their degree of belief in the proposition
“Engels’ contribution significantly altered the eventual conceptions which emerge from Marx
and Engels’ joint works”. If one does observe a significant shift between the two spaces,
however, an even more fruitful opportunity for further understanding emerges: what was the
nature of these shifts—which other key concepts did “Staat” move closer to or away from, for
example—and do they in fact reflect the details of Levine’s argument? In the excerpt quoted
above, for example, we see that this argument points to Engels’ lack of Feuerbachian influence
as an explanans for the differences between the two. Thus, when we examine the contours
of how the points for our key terms shifted, do we notice a pattern whereby the Engels-
inclusive space shifts the meaning of these terms away from their grounding in Feuerbachian
categories like “Wesen” (“Wesen des Christentums”, “Wesen des Glaubens”, “Wesen der
Religion”, etc.)? By conducting our inquiries into Marx’s thought in this manner, we have
access to a tool beyond our own intuition—namely, the constructed embedding spaces—to
which we can turn as a way of adjudicating between competing interpretations.
Although Levine’s claim is particularly interesting for our purposes due to the synchronic
mode in which it is presented9 , much of the discussion revolves instead around contending
claims of continuity versus change in Marx’s thought, with the claims of change often further
differentiated into claims of gradual change versus the assertion of sharp “points of disconti-
nuity”, i.e., drastic and sudden changes occurring in response to world events or to the dis-
covery of fatal flaws in his own theory. Most famously perhaps, Althusser 1965 and Althusser
1968 argue for the existence of an “epistemic break” in the development of Marx’s thought,
9
Synchronic in the sense that the primary claim concerns Marx’s thought as a whole with respect to
Engels’ modifications as a whole, rather than particular time-indexed changes in these bodies of thought. A
diachronic version of Levine’s hypothesis, for example, could be that Engels’ modifications were of greater
and greater magnitude over time, so that perhaps Engels’ contribution to the 1848 Communist Manifesto
did not alter Marx’s thought very much with respect to the final published version, but that his editing of
Volumes 2 and 3 of Capital strongly altered the content of Marx’s original drafts of these volumes. In this case
we would need to utilize the diachronic models described in Section 1.2.4, since a single non-time-indexed
Marx space and non-time-indexed Marx-Engels space would not allow us to differentiate e.g. Engels’ impact
on the Manifesto from his impact on the later volumes of Capital. We consider the evaluation of diachronic
hypotheses, derived from the literature on the development of Marx’s thought, later in this section.

115
an essentially clean separation between his early humanistic philosophy phase (manifesting
most clearly in his 1844 Economic and Philosophical Manuscripts) and his later “scientific”
political economy phase (manifested in the three volumes of Das Kapital).
Although this duality—the existence of both humanistic and political-economic strands in
his work—had been proposed earlier (for example by Robert Tucker in 1961, Tucker 1961),
Althusser’s assertion of a single point in time neatly separating the two strands, and his
consequent dismissal of the early Marx’s writings on the basis of their “unscientific” nature,
gave rise to a whirlwind of backlash among Marx scholars. Shlomo Avineri’s The Social and
Political Thought of Karl Marx Avineri 1968, for example, exhaustively documents a series
of parallels between Marx’s early and late writings on the topic of alienation, pointing to
their shared roots in Hegelian categories. To this end, Avineri argues, Marx’s 1843 Critique of
Hegel’s Philosophy of Right represents a kind of “ur-text”, in the sense that it established not
only (a) the vocabulary that the early Marx would use to critique e.g. the Young Hegelians,
but also (b) the problematics which would motivate his studies throughout his life, whether
he wrestled with them using his early humanistic or later political-economic vocabularies.
Indeed, Avineri perhaps makes an even stronger claim than this when he asserts that

All the main achievements, as well as dilemmas, of Marx’s later thought (like
the abolition of private property, of alienation and of the state) originate in this
work. (Avineri 1968, p. 3)10

Thus, by starting his analysis with an examination of this early work and identifying the key
Hegelian categories with which Marx seems to engage most deeply, Avineri’s first goal is to
“show how Marx, in his first confrontation with Hegel, could construct his materialist view
out of the Hegelian system itself” (Avineri 1968, p. 5). Once this has been established, how-
ever, his second goal is to bring Marx’s later works into the discussion—the works assumed
by Althusserian scholars to exist on the other side of a sharp “epistemological break” from
10
Joseph O’Malley seconds Avineri’s assessment in his editorial introduction to the 1970 English translation
of the work, stating that “In fact, the Critique of Hegel’s ‘Philosophy of Right’ developed into the whole
program of research and writing which occupied Marx for the remainder of his life”. (Marx 1970, p. xiv)

116
his early works—to argue that

Marx’s later writings merely articulate the conclusions at which he arrived at this
early stage of his intellectual odyssey. The various economic, social and historical
studies undertaken by Marx are but a corollary of the conclusions he drew from
his immanent critique of Hegel’s political philosophy. (Avineri 1968, p. 5)

It is significant for our purposes to note that Avineri’s investigation begins with a chap-
ter devoted exclusively to Hegel’s social and political thought, without reference to Marx.
Although a “pure” understanding of Hegel entirely divorced from its subsequent Young
Hegelian and Marxian developments is probably unattainable, we contend that Avineri’s ex
ante consideration of Hegel—focusing on Hegel’s thought as such rather than how it influ-
enced Marx—plays an important role in the larger work’s clarity in identifying this Hegelian
skeleton beneath the humanistic and political-economic surfaces of Marx’s oeuvre. Avineri’s
contribution to our understanding of Marx can thus be thought of as involving two steps,
which will be helpful to differentiate when we consider how to translate it into computational
terms: First, we can think of his initial study of Hegel’s thought as him converting the text
of Hegel’s writings into a set of ontologies with respect to the key categories Hegel uses to
describe his theories. Thus, for example, by reading the text of Hegel’s Philosophy of Right he
infers that Hegel’s theory concerns a particular set of relationships among a set of important
societal entities or institutions which collectively constitute an ethical order (Sittlichkeit):
the State (Staat), the Family (Familie), and Civil Society (bürgerliche Gesellschaft)11 .
Having constructed the understanding of Hegel’s thought described in the first chapter,
he then turns to a study of Marx’s thought, corresponding to the second step in our proposed
two-step approach. Beginning his investigation with Marx’s 1843 critique of the Philosophy of
Right, he now has the ability to not only infer Marx’s ontology and his arguments regarding
11
Avineri later expanded this chapter into a separate work, Avineri 1974, describing in detail how Hegel
theorized the nature of these entities and their relationships. Our discussion here draws heavily on this work,
especially Chapters 7 (“The Political Economy of Modern Society”) and 8 (“Social Classes, Representation,
and Pluralism”), as well as his earlier article, “The Hegelian Origins of Marx’s Political Thought” (Avineri
1967).

117
the relationships among its elements in this work (as he did for Hegel in the first step),
but also to compare the structure he sees in Marx’s thought with that of Hegel’s thought
proposed in the first chapter. It is precisely this comparative approach which allows him to
argue, for example, that Marx “construct[ed] his materialist view out of the Hegelian system
itself” (Avineri 1968, p. 5). By bringing this approach to bear on Marx’s full body of work,
then, Avineri is able to identify a pair of mappings—one for Marx’s humanistic vocabulary
and a second for his political-economic vocabulary—and to show how the ontological and
relational structures of Marx’s theories in either case mirrored those of Hegel’s.
As noted above, we present these details and disaggregate Avineri’s approach into these
two separate steps because, as in the case of Levine’s work, Avineri’s naturally points towards
a computational technique for examining the early-Marx/late-Marx/Hegel relationship. Sim-
ilar to the example in Section 1.2 where we inferred relations between entities (Leader-Of(x,y),
Invaded(x,y), and so on) by training an embedding space on the New York Times corpus, an
Avineri-esque investigation of Marx’s thought could be carried out in a computational mode
via an analogous two-step process. First, we could train an embedding space on Hegel’s
Philosophie des Rechts, taking note of the constellations formed by ontologically-related
points like those for Familie, Staat, and bürgerliche Gesellschaft. Then, for our second step,
if we can identify similar spatial patterns among key terms in an embedding space trained on
Marx’s early works, as well as one trained on Marx’s mature works, we have strong grounds
upon which to increase our degree of belief in Avineri’s hypothesis of the essential Hegelian
unity of Marx’s entire body of work.
Figure 2.2 illustrates just such a joint Marx-Hegel embedding space, with the caveat that
(for demonstration purposes) we’ve restricted the embedding algorithm to two dimensions
rather than the 300+ dimensional spaces which most embedding algorithms generate by
default. Despite the restricted low-dimensional output space we can already see that, for
example, Marx associates eigentum (property) more closely with sittlichkeit (ethical order)

118
than Hegel does12 :

# » # »
dist∠ (eigentumM , sittlichkeitM ) ≈ 0.073
# » # »
dist∠ (eigentumH , sittlichkeitH ) ≈ 0.110

and similarly for eigentum and rechts (right):

# » # »
dist∠ (eigentumM , rechtsM ) ≈ 0.057
# » # »
dist∠ (eigentumH , rechtsH ) ≈ 0.074

We can see Marx’s critique beginning to form, therefore, in that when Marx discusses
sittlichkeit in his Critiqe he is referencing the order which Hegel argues for in the Philosophie
des Rechts, a “harmonious” interplay between family, state, and civil society. Marx, however,
is here beginning to drive his critical wedge into this concept, by pointing out that what Hegel
calls “civil society” is in fact dominated by property owners, so that the “harmony” asserted
by Hegel is in reality a domination of the ethical order by the owners of property and later
on—a category he will develop in his mature political-economic writings—by the controllers
of capital.
In particular, in §303 of the Philosophie des Rechts, Hegel states straightforwardly that

“Der allgemeine, näher dem Dienst der Regierung sich widmende Stand hat un-
mittelbar in seiner Bestimmung, das Allgemeine zum Zwecke seiner wesentlichen
Tätigkeit zu haben”

(“The universal class, or, more precisely, the class of civil servants, must purely in
virtue of its character as universal, have the universal as the end of its activity”)

(Hegel 1820 §303, p. 473, English translation from Avineri 1974, p. 159),
12
In what follows we use the notation dist∠ (·, ·) to denote cosine distance, so as to differentiate this distance
metric from Euclidean distance, which we denote via distℓ2 (·, ·).

119
thus equating the “universal” (allgemeine) with the “class of civil servants” (Dienst der
Regierung sich widmende Stand). Marx, however, as early as his original critique of this
text but especially in the more developed Introduction to his critique published in the DFJ,
argues that Hegel’s universal class is in practice (in the Prussian state which Hegel is often
seen as cheerleading for) a class which manifests only a particular set of ruling interests,
and counterposes the proletariat—in fact, this represents the first appearance of this term
in Marx’s writings—as the true universal class:

“Wo liegt also die positive Möglichkeit der deutschen Emancipation?

Antwort: In der Bildung einer Klasse mit radikalen Ketten, einer Klasse der
bürgerlichen Gesellschaft, welche keine Klasse der bürgerlichen Gesellschaft ist,
eines Standes, welcher die Auflösung aller Stände ist, einer Sphäre, welche einen
universellen Charakter durch ihre universellen Leiden besitzt [...] Diese Auflösung
der Gesellschaft als ein besonderer Stand ist das Proletariat.” (MEGA2 I/2, pp.
181–182)

(“Where, then, is the positive possibility of a German emancipation?

Answer: In the formation of a class with radical chains, a class of civil society
which is not a class of civil society, an estate which is the dissolution of all
estates, a sphere which has a universal character by its universal suffering [...]
This dissolution of society as a particular estate is the proletariat.”) (MECW 3,
p. 186)

While these sketches focus on hypotheses in the literature which posit a relatively static
unity among Marx’s works (with respect to Engels’ works, in Levine’s case, or Hegel’s in
the case of Avineri), and hence fall under the rubric of synchronic analysis, the diachronic
methods introduced in Section 1.2.4 in fact unlock an even wider range of possibilities for ad-
judicating among competing interpretations of Marx’s thought. In broad outline: by running
these algorithms on collections of texts and analyzing how close or distant the embeddings

120
 1 y
# » # »
FamilieM StaatM

0.8
# »
GesellschaftM
# » # »
0.6 Dienst der AllgemeineM
# »
Regierung ProletariatM

0.4 # »
AllgemeineH
# »
GesellschaftH
0.2

# » # »
StaatH FamilieH x
0.2 0.4 0.6 0.8 1

Figure 2.1: Computational-linguistic “translations” of the hypothesis in Avineri 1968, that

key concepts deployed in Marx’s early and late works can both be mapped onto a single set
of concepts introduced in Hegel’s Logik.

for key terms are to one another, we can begin to trace out an author’s evolution in terms
of their trajectory through this space over time. For example, focusing in particular on hy-
potheses regarding Marx’s intellectual influences over the course of his life, we could delve
further into the yes/no question of “can we find structures in Marx’s thought homologous to
those of Hegel’s?” by considering the more nuanced question of “to what extent do Hegel’s
schema pattern Marx’s at different periods of his life?”
This question can then, in turn, lead us to formulate (for example) Lenin’s “three com-
ponent parts” hypothesis in the following terms: that the embeddings derived from texts of
the “early Marx” period (roughly from his first writings to the “Economic and Philosophic
Manuscripts of 1844”) would remain within the vicinity of a Hegelian vocabulary cluster—
whether derived from Hegel’s own texts or those of Young Hegelians like Bruno Bauer—but
afterwards would rapidly move away from this Hegelian cluster and towards clusters repre-
senting the vocabulary of French socialists (Louis Blanc, Pierre-Joseph Proudhon, Charles

121
 Figure 2.2: Computational-linguistic “translations” of the hypotheses in Avineri 1968.

Fourier, Sismondi) and, eventually, British political economists (Adam Smith, David Ri-
cardo, J. R. McCulloch). And, indeed, it is precisely this formulation that we employ in
the coming sections to study the relative influences exerted upon Marx’s thought by these
three anterior vocabularies over time. First, however, we introduce the datasets and corpora
we use throughout the study, emphasizing in particular how they were constructed via a
methodology aiming for maximal replicability and representativeness, thus allowing us to
bring these computational methods to bear on a set of otherwise interpretive, qualitative
questions of intellectual history.

2.1.3 The Data

Although the corpus of texts discussed in the next section serves as the direct input to
our influence-measurement studies, in this section we describe in detail the criteria upon
which texts were selected or excluded from this corpus. We then defend these criteria on
the basis of findings derived from “metadata” we collected on the readings, writings, and

122
interactions of key 19th-century European socialists, arguing that they enable us to construct
a representative sample of 19th-century European socialist discourse within which to situate
Marx’s writings over time.
A key contribution of our work is thus the open-sourcing of this collection of metadata,
in the form of the Marx-Engels Digital Cyclopedia Project, within which Marx’s and Engels’
reading and note-taking, writing, correspondence, and referencing of other texts can (in
almost all cases) be traced down to the exact day they took place. The datasets in the
collection, available for browsing and analyzing at https://marxdb.com, are described in the
sections that follow.

2.1.3.1 The Marx-Engels Digital Register

The Marx-Engels Digital Register (hereafter shortened as Digital Register), contains a

record for every known piece of writing by the two authors (N > 1000), linked to the full
text of the writing along with a substantial set of metadata:

• The title of the writing, in English

• The original title, if written originally in a language other than English

• Dates of writing and first publication (cross-referenced with the Digital Chronicle–see
below)

• The original language the text was composed in

• A listing of known translations

• The specific location of the writing within MEGA1, MEGA2, MEW, and MECW, and

• All known details regarding the relative contributions of Marx and Engels, in the case
of jointly-authored works.

123
 This dataset extends the original Marx-Engels Register (Draper 1985b), which itself built
upon and extended a large collection of previously-published bibliographies13 . The key con-
tribution of this dataset, beyond the the digitization and relational-database-structuring of
the out-of-print 1985 Register, is the inclusion of references to the volumes and pages of the
“New MEGA”, MEGA2. While only 16 volumes had been published by mid-1984, when the
original Register was finalized14 , an additional 56 volumes have been published in the subse-
quent years, bringing the total to 72 out of a planned 12215 . Thus, the Digital Register brings
the original references up-to-date with these dozens of subsequently-published volumes.

2.1.3.2 The Marx-Engels Digital Notebooks

The Marx-Engels Digital Notebooks (hereafter shortened as Digital Notebooks) contains

a record for every entry—i.e., every reference to another work—across Marx’s 168 extant
reading notebooks16 , along with the following metadata:

• How many notebook pages Marx filled with these excerpts, and

• The number of excerpts taken, and

• Links to the corresponding records in the Digital References (described in Section

2.1.3.7 below) containing metadata about the referenced works, along with the full
text of the works (when available).
13
The most comprehensive of these being Maximilien Rubel’s 1956 Bibliographie des oeuvres de Karl Marx
(Rubel 1956).
14
These published volumes being: I/1, I/10, I/22, I/24, II/1, II/2, II/3, II/5, III/1, III/2, III/3, III/4, IV/1,
IV/2, IV/6, and IV/7. See Draper 1985b pp. 206–207 for information on the progress of MEGA2 publication
at the time of compilation.
15
See https://marxdb.com/mega for our dataset containing detailed contents of all 122 planned volumes.
16
A listing of the contents of these notebooks, along with facsimiles of each hand-written page, are available
via the IISG website at this link.

124
2.1.3.3 The Marx-Engels Digital Chronicle

The Marx-Engels Digital Chronicle (hereafter shortened to Digital Chronicle), contains

a record for each known “event” in Marx’s and Engels’ lives17 . The events are cross-listed
with the other datasets as follows:

• Dates of writing and publication of articles, statements, etc. are linked to the corre-
sponding work in the Marx-Engels Digital Register,

• Dates on which entries were made in notebooks are linked to the corresponding entry
in the Marx-Engels Digital Notebooks,

• Dates of the sending and receipt of letters are linked to the corresponding entry in the
Marx-Engels Digital Correspondence, and

• Entities mentioned in the event (e.g., people to whom letters were written, or organi-
zations whose meetings were attended) are linked to the corresponding entries in the
Marx-Engels Digital Glossary.

2.1.3.4 The Marx-Engels Digital Correspondence

The Marx-Engels Digital Correspondence (hereafter shortened to Digital Correspon-

dence) contains a record for each known letter sent by or to Marx and/or Engels, along
with the following metadata:

• Date of sending and date of receipt,

• Full text of the letter (where available), and

• Links to the corresponding Digital Glossary entries for each author/recipient, as well
as for each entity mentioned in the letter itself.
17
See Draper 1985b for what types of events qualify for inclusion.

125
2.1.3.5 The Marx-Engels Digital Glossary

The Marx-Engels Digital Glossary (hereafter shortened to Digital Glossary) contains a

record for every “entity”—people, places, organizations, publications, and geographic/geopolitical
entities18 —referenced in all other datasets, along with metadata like links to the Wikidata
database id for each entry (which, in many cases, is subsequently linked to the same entity
in hundreds of additional databases—see, e.g., the entry for Karl Marx), where available.
As with the Digital Chronicle and Digital Register, this dataset builds upon and updates
the original 1985 Glossary with references to the subsequently-published volumes of MEGA2
and the newly-referenced entities therein.

2.1.3.6 The Marx-Engels Digital Bookshelf

The Marx-Engels Digital Bookshelf contains an entry for each book known to have been
owned by Marx or Engels over the course of their lifetimes, constructed based on the N =
1450 entries given in MEGA2 IV/32. Each record contains the full text of the book (when
available), along with metadata including whose bookshelf it was known to be in (Marx’s,
Engels’, Both, or Unknown) and whether or not marginal notes were known to have been
made by Marx, Engels, or a different, unknown reader.

2.1.3.7 The Marx-Engels Digital References

The Marx-Engels Digital References contains an entry for each work referenced by Marx
or Engels across all volumes of MEGA2, and is linked to the corresponding entries in the
Digital Bookshelf (for works which Marx or Engels were known to have owned) and Digital
Notebooks (for works from which Marx or Engels took reading notes and excerpts).
18
See https://marxdb.com/glossary/criteria/ for details regarding what types of entities qualify for inclu-
sion.

126
2.1.4 The Corpora

By analyzing the datasets described in the previous section, we were able to construct, and
then generate a representative sample of, the set of texts which constituted the “discursive
space” of 19th-century European socialism—the texts which the key interlocutors engaged
with or had in mind when conducting their correspondence, their polemical interventions,
and their political organizing.
This corpus was constructed via a bootstrapping procedure: starting from a dataset of ref-
erenced texts, which we call the first “layer” of the corpus—for example, those read by Marx
or Proudhon in a given year (the Digital Notebooks, described in Section 2.1.3.2 above)—we
then took a random sample of these texts and constructed a second-order contextual “layer”
representing all of the texts reference, in turn, by these first-layer texts.
Importantly, this bootstrapping procedure was conducted on the various referenced-texts
datasets with two distinct objectives in mind, resulting in two distinct (though partially
overlapping) subcorpora. The first, serving as the input to our study of influences on Marx
described below, was bootstrapped from the Digital Notebooks, in particular from the set
of texts which Marx was known to have read (whether or not he subsequently referenced
them explicitly in a written work or letter). The second, providing the input to our study
of Marx’s own influence on the broader European socialist movement (Chapter 3), was
instead bootstrapped from the Digital References, i.e., the set of texts which Marx explicitly
referenced in his works, letters, or notebook entries.
Although the bootstrapping method provides us with a substantial set of texts to start
our investigation with, we supplement these automatically-generated corpora with additional
texts which are known to be important but not included automatically due to their not
having been explicitly referenced by Marx or Engels. As discussed in Section 2.1.3.6 above,
for example, there exist dozens of books known to have been owned by Marx at some point—
many of which contain annotations in his handwriting—yet never referenced in any of his
writings. Entry 10 for example, Richard Alexander’s The Rise and Progress of British Opium

127
Smuggling, contains marginal notes written by Marx in ink on pages 37 and 38, yet was never
referenced in any known work, letter, or reading notebook.
The additional works added to our two corpora are listed in Appendices 2.A (for influ-
ences on Marx) and 3.A (for Marx’s influence), along with explanations for each inclusion.
For example, Gregory 1983 provides a list of works discussed by a “socialist study circle”
founded by Moses Hess in Cologne in 1842, which Marx frequented. In Table 2.1 we cross-
reference these works with the Digital References and Digital Bookshelf, revealing that Victor
Considérant’s Destinée sociale and Pierre Leroux’s De l’humanité should be included in the
corpus of books which potentially influenced the development of Marx’s thought, despite the
fact that Marx never explicitly referenced them in his writings, letters, or reading notebooks.
As an additional confirmation that Leroux’s work should be added, it is listed in Rubel 1975
as a work which Marx “probably read” (9). In this manner, in Appendices 2.A and 3.A, we
provide justification (via citations) for each inclusion of a text not originally captured by our
bootstrapping procedure, for transparency and replicability.

Work Referenced by Marx In Bookshelf

Cabet, Voyage en Icarie 7 times No
Considérant, Destinée sociale No No
Fourier, Théorie des quatre mouvements et des des- 15 times No
tinées générales
Hess, Die Europaische Triarchie 1 time No
Leroux, De l’humanité No Yes (766)
Proudhon, Qu’est-ce que la propriété? 25 times Yes (1080)
Weitling, Die Menschheit, wie sie ist und wie sie sein 5 times No
sollte
Weitling, Garantien der Harmonie und Freiheit 13 times Yes (1407)

Table 2.1: Books read by the “socialist study circle” founded by Moses Hess in Cologne in
1842, according to Gregory 1983

With our methods, data, and corpora now described, in the next section we begin with
our first of two studies on the development of Marx’s thought.

128
2.2 Study I: Influences on Marx

2.2.1 German Philosophy

2.2.1.1 Calibration

To help illustrate what exactly these methods are picking up when comparing two texts,
we start by examining the range of similarity scores between pairs of Hegel’s own texts in
Figure 2.3, to obtain the kernel density estimation illustrated in Figure 2.4. This provides
us with a rough upper bound of the similarity scores we can expect to see for a given text:
if the text’s Hegel-similarity score is within this range, that text is so similar to Hegel’s
texts as to be essentially indistinguishable from Hegel’s own writing, from a lexical-semantic
perspective.

Figure 2.3: Heatmap of Similarities between Hegel’s Texts

Next, to estimate a rough lower bound of the similarity scores we can expect to obtain, we
utilize two different methods based on two different notions of what it would mean to obtain
a “random” 19th-century German text. Computing a lower bound by obtaining such a text

129
 Figure 2.4: Kernel Density Plot of Similarities between Hegel’s Texts

and computing its Hegel-similarity captures the intuition that, if a text is indeed influenced
by Hegel to a significant degree, its Hegel-similarity score should also be significantly higher
than the “average” 19th-century German text.
However, although this “average” text may be easy to conceive of in theory, in practice
the selective transmission of historical documents (see, e.g., Shapiro et al. 1987) presents a
challenge. The presence of this implicit selection bias means that even a document selected
from a large historical corpus like the 181,732 19th-century German texts on HathiTrust19
may be more similar to Hegel’s works than the truly “average” piece of 19th-century German
writing, since the randomly-selected text was in fact one that ended up being preserved into
the 21st century rather than being relegated to the “dustbin of history”.
To counterbalance this potential bias, in addition to the collection of texts randomly
sampled from HathiTrust’s corpus, we also utilize a neural language model fine-tuned on a
balanced corpus of 19th-century German texts (described in Hosseini et al. 2021) to generate
19
The count for German-language texts between 1800 and 1899 as of the time of writing, obtained using
this search URL.

130
a separate collection of “randomly-generated” texts. In other words, in addition to the Hegel-
similarity scores of particular 19th-century German documents, we also obtain the Hegel-
similarity scores of a set of documents generated based on linguistic features averaged across
a balanced corpus of 19th-century German. In fact, although both approaches allows us to
get a sense of the lexical-semantic information captured by the similarity scores, we contend
that the Hegel-similarity scores for the synthetic documents better capture the similarity
between Hegel and an “average” text, especially given the massive dimensionality-reduction
(from millions of words down to a single 768-dimensional vector) which occurs regardless as
part of any document embedding method20 .
A heatmap of the Hegel-similarity scores for the randomly-selected documents are given
in Figure 2.5, and for the synthetic documents in Figure 2.7. The corresponding kernel
density plots are given in Figure 2.6 for the randomly-selected documents and in Figure 2.8
for the synthetic documents.

2.2.1.2 Results

With the upper bound and the two lower bounds from the previous section in hand,
we can now interpret the similarity scores in a framework salient to historians of political
thought:

• The lower bound ℓ represents the absence of influence, in the sense that if sim(A, B) ≈
ℓ, authors A and B bear no more similarity to each other than author A would have
on average with any randomly-chosen author C,

• The upper bound u represents “maximal” influence, since if sim(A, B) ≈ u, author

20
In other words, from a Shannon information-theoretic perspective, a coarse-graining transformation—the
embedding algorithm—applied to texts generated via a language model trained on a large, balanced corpus
(and thus ostensibly capturing linguistic properties which are themselves statistically representative of 19th
century discourse) will retain more information about this corpus than the same transformation applied to
a smaller set of particular texts randomly sampled from this corpus, all else held equal. We report similarity
scores for individual texts sampled from this corpus as well, however, to ensure that the model’s massive
averaging of linguistic properties does not result in noise—say, the errors of OCR software— overwhelming
the desired signal (the average lexical-semantic properties of 19th-century texts).

131
Figure 2.5: Heatmap of Similarity Scores between Hegel’s Works and Randomly-Selected
Documents

B has adopted author A’s lexical-semantic practices to the point that B’s writing is
“indistinguishable” from A’s.

Given these interpretations, then, we can interpret scores in between these two values as
existing on a gradient of increasing influence: if we arranged a collection of texts from lowest
to highest author-A-similarity score, and read through them in this order, we would observe
them becoming more and more “author-A-esque”, more and more distinguishable from other
books in terms of their resemblance to the texts of author A. With this interpretation in
mind, we move to the main analysis, a comparison of Marx’s work with Hegel’s over the
course of the former’s lifetime. A plot with Marx’s semantic similarity to Hegel (by work,
over time) is presented in Figure 2.9.
Interestingly, given that many scholars locate Marx’s departure from Hegel in his critical
engagement with the latter’s Philosophie des Rechts in 1843, in terms of semantic content
Marx’s engagement with Hegelian themes actually increases fairly dramatically after this
time. This makes sense, however, if we consider the period roughly from 1843 to 1846 as

132
Figure 2.6: Kernel Density Plot of Similarity Scores between Hegel’s Works and Synthetic
Documents

Figure 2.7: Heatmap of Similarity Scores between Hegel’s Works and Synthetic Documents

133
Figure 2.8: Kernel Density Plot of Similarity Scores between Hegel’s Works and Synthetic
Documents

Figure 2.9: Marx’s similarity to Hegel, over time

134
a period wherein Marx aimed to distance himself from the Young Hegelians precisely by
attacking their Hegelian-language-infused doctrines in Holy Family and German Ideology,
and then the period from 1847 onward as his transition into his final economist phase.
Supporting the former claim, that 1843–1846 was a period of direct engagement with
Hegelian philosophy with the aim of moving German philosophy “beyond” it, we have Marx’s
contention that he wrote the German Ideology in order “to settle accounts with our erst-
while philosophical conscience” (MECW 29, 264). Then, with these accounts having been
satisfactorily settled, the opening pages of the 1847 Misère de la Philosophie are the first in
which Marx explicitly identifies himself as an economist rather than a philosopher or political
journalist:

M. Proudhon has the misfortune to be uniquely misunderstood in Europe. In

France he has the right to be a bad economist, since he passes for a good Ger-
man philosopher. In Germany, he has the right to be a bad philosopher, because
he passes as a prominent French economist. Being ourselves both German and
economist, we have wished to protest against this dual mistake 21 (MECW 6, 109).

Another feature of Marx’s trajectory over time which emerges from our findings is the
partial recovery of his Hegelianism after the 1848 Revolutions, beginning with the 1852
publication of his 18th Brumaire and continuing up until the end of his life. Interestingly,
this is several years earlier than the typical point at which some studies posit a “return”
of Hegel to Marx’s intellectual purview, namely, the years leading up to the publication of
Capital Vol. 1.
For example, a key piece of evidence typically cited by scholars who contend that Capital
represents a re-infusion of Hegel into Marx’s mature political-economic thought is a letter
21
In Marx’s original French: “M. Proudhon a le malheur d’être singulièrement meconnu en Europe. En
France, il a le droit d’être mauvais économiste, parce qu’il passe pour être bon philosophe allemand. En
Allemagne, il a le droit d’être mauvais philosophe, parce qu’il passe pour être économiste français des plus
forts. Nous, en notre qualité d’Allemand et d’économiste a la fois, nous avons voulu protester contre cette
double erreur” (MEGA1 I/6, 19; Misère will also appear in its original French in the not-yet-published
MEGA2 I/6, ).

135
of 16 January 1858 to Engels where Marx mentions re-reading Hegel’s Logic and gaining
inspiration for his own treatment of political economy:

What was of great use to me as regards method of treatment was Hegel’s Logic at
which I had taken another look BY MERE ACCIDENT, Freiligrath having found
and made me a present of several volumes of Hegel, originally the property of
Bakunin 22 (MECW 40, 249, 1858.01.16).

While this sudden injection of Hegel’s work in 1858 would otherwise provide an explana-
tion for the large increase in Hegel-similarity between the publication of Wage-Labour and
Capital in 1849 and of the Grundrisse and Critique of Political Economy in 1858–1859, the
fact that 18th Brumaire predates Marx’s letter by six years yet exhibits nearly the same
level of Hegelianism as these latter works requires explanation, an explanation lacking in the
currently-available literature on Marx’s intellectual development.

2.2.2 French Socialism

2.2.2.1 Calibration

Turning to the second posited stage in Marx’s intellectual development, that of his in-
troduction to working-class-oriented socialism in Paris between 1843 and 1845, we begin by
again establishing our expectations as to what a particular numeric similarity score repre-
sents, qualitatively, to a historian of political thought.
Unlike in the case of Hegelianism, the intellectual strand of French socialism in Marx’s
time cannot straightforwardly be traced back to a single author or text. To ensure that the
upper bound we obtain here retains the same methodological-individualist interpretation as
that of the previous section, then, we restrict our calibration studies in this section to an
analysis of the collected works of Pierre-Joseph Proudhon, as arguably the best individual
22
In Marx reading Bakunin’s copies of Hegel, which contained marginal annotations likely to be Bakunin’s
(MEGA2 IV/32, 315) to which Marx subsequently added his own (ibid.), we have another identifiable vector
of potential influence between the two (see Section 3.2.6 for more discussion of Marx and Bakunin’s influence
on one another).

136
proxy for the current of French socialism. We then, for the final results described the next
section, expand this corpus to include a wide range of French socialist authors alongside
Proudhon.
A heatmap of the pairwise self-similarity scores between Proudhon’s works is given in
Figure 2.10, with the corresponding KDE plot given in Figure 2.11.

Figure 2.10: Heatmap of semantic self-similarities between works of Proudhon

Next, heatmaps of the lower-bound similarities between Proudhon’s works and randomly-
sampled 19th-century works in French are given in Figure 2.12, and those between Proud-
hon’s works and synthetic French documents in Figure 2.14. The corresponding kernel density
plots are given in Figure 2.13 and Figure 2.15, respectively. The randomly-selected works are
drawn from the set of French-language texts published between 1800 and 1899 and available
on HathiTrust, a total of 169,343 texts.

137
 Figure 2.11: Kernel density plot of semantic self-similarities between works of Proudhon

2.2.2.2 Results

With the lower and upper bounds for our similarity scores calibrated as described in the
previous section, we can now interpret the time series plot of similarities between Marx’s
works and works of French socialism given in Figure 2.16.
Here, unlike in our analysis of Hegelianism, we observe no clear pattern in terms of how
the influence of French socialist texts increases or decreases with respect to Marx’s major
relocations. In fact, if one counterposes the texts which obtain higher-than-average similarity
scores against those which obtain lower-than-average scores here, what emerges is essentially
a classification of Marx’s writings into those which comment on French literature and current
events and those which do not.
In Holy Family, for example, Marx devotes a large proportion of his writing to critiquing
Eugène Sue’s Les Mystères de Paris, which was published serially between 1842 and 1843
and quickly became widely popular throughout Paris just as Marx was acquainting himself
with the city. The Poverty of Philosophy similarly provides an extended critique of a popular

138
Figure 2.12: Heatmap of semantic similarities between Proudhon’s texts and randomly-
selected 19th-century French texts

French text, in this case Proudhon’s Système des contradictions économiques. After the 1848
Revolutions, the two texts which stand out as having high similarity with French socialist
texts are precisely the two in which Marx comments on French current events: the first, 18th
Brumaire, was a reflection on Louis Napoleon’s 1852 coup d’etat, and the second, The Civil
War in France, a response to the 1871 declaration of the Paris Commune.
On the one hand, these results seem prima facie to indicate an over-emphasis, in the
literature on Marx, on the impact that his stay in Paris had on the development of his
thought. There exists an equally plausible explanation for this lack of pattern, however, that
must also be considered: that the discursive lexical-semantic space of French socialism itself
was heterogeneous to the point that the only topic existing within the ideological intersection

139
Figure 2.13: KDE plot of semantic similarities between Proudhon’s texts and randomly-
selected 19th-century French texts

Figure 2.14: Heatmap of semantic similarities between Proudhon’s texts and randomly-
generated French-language texts

140
Figure 2.15: KDE plot of semantic similarities between Proudhon’s texts and randomly-
generated French-language texts

of all these discussions was a broad focus on the politics and literature of France.
In this case, then, we could conclude that the German philosophical discourse analyzed
in the previous section was more “focused” than French socialist discourse, given that the
similarity scores obtained in that study did not exhibit a high-low pattern corresponding
to whether or not the text happened to discuss e.g. the literature or politics of Prussia, of
Cologne, of Berlin, etc.

2.2.3 British Political Economy

2.2.3.1 Calibration

As we did for German philosophy and French socialism, here we first examine the range
of similarity measures our methods produce for a pair of books with an “established” rela-
tionship of influence: Adam Smith’s Wealth of Nations (1776) and David Ricardo’s On the
Principles of Political Economy and Taxation (1817), given that the latter is in large part

141
 Figure 2.16: Marx’s similarity to Works of French Socialism, over time

written as a response to Smith’s groundbreaking 1776 work23 .

A heatmap of the semantic self-similarity scores for the Political-Economic texts is given
in Figure 2.17, with the corresponding KDE plot given in Figure 2.18.
A heatmap of the similarities between works of political economy and randomly-selected
19th-century English texts is given in Figure 2.19, with the corresponding KDE plot given
in Figure 2.20. The same plots comparing the political-economic and synthetic English texts
are given in Figures 2.21 and 2.22, respectively.

2.2.3.2 Results

With our upper and lower bounds now established, we turn to analyzing the time series
of political-economic-similarity scores given in Figure 2.23. We find that the influence of
23
Unlike in the previous two sections, here we opt not to compare Smith’s Wealth of Nations with e.g. his
own earlier Theory of Moral Sentiments, on the grounds that our aim is not to capture “Smith-ness” writ
large, but rather “political-economy-ness” in the tradition established by Wealth of Nations, rather than the
less influential and more political-philosophy-oriented Theory of Moral Sentiments.

142
 Figure 2.17: Heatmap of semantic self-similarities between works of political economy

Figure 2.18: KDE plot of semantic self-similarities between works of political economy

political-economic discourse on Marx’s work increases significantly – by over 10% – from

his 1841 dissertation to his 1844 Manuscripts, then increases significantly but less rapidly

143
 Figure 2.19: Heatmap of randomly-selected similarities

Figure 2.20: KDE plot of randomly-selected similarities

in Poverty of Philosophy and reaches a peak in 1858-1867 with the Grundrisse, the Kritik
der Politischen Ökonomie, the three volumes of Theorien über den Mehrwert, and Kapital,

144
 Figure 2.21: Heatmap of randomly-generated similarities

Figure 2.22: KDE plot of randomly-generated similarities

Vol. 1, with the notable exception of the extremely non-political-economic Herr Vogt (an
anomaly we delve into in Section 3.2.4 below).

145
 Figure 2.23: Marx’s similarity to works of Political Economy over time, major works

This pattern mirrors—more so than the findings discussed in the previous two sections—
the “standard” narrative given in studies of the development of Marx’s thought: that his
exposure to key works of political economy in the months leading up to his 1843 move
to Paris quickly and permanently changed the trajectory of his thought. Again with the
exception of Herr Vogt, and even taking into account his project of “settling accounts with”
the Young Hegelians via the Holy Family and German Ideology, we find a sharp increase in
his similarity to works of political economy after 1843, with all works before this point lying
below the random-sample similarity threshold and all but two works afterwards lying above
it.
Interestingly, though, we do not find strong evidence for the influence of Marx’s mid-1850
admission to the British Museum library on the political-economic content of his writings
after this point. This lack of pattern, though it does not contradict the hypothesis that
Marx’s focus on political economy began in 1843, does shed significant doubt on the proposed
mechanism explaining why this increase occurred – if the reading of 4-5 works of political

146
Figure 2.24: Marx’s yearly volume of reading (computed based on the known or estimated
years of his reading notebooks) over the course of his lifetime.

economy in 1843 caused such a significant and permanent shift in his thought, why did the
incredible 332 works he read just in his first two years of British Museum library membership
(see Figure 2.24) fail to cause any shift at all? In the next section we synthesize our findings
here with the qualitative evidence discussed in Section 2.1.1 to attempt an answer to this
question, among others, which our findings give rise to.

2.2.4 Conclusion

The findings of the previous three sections, along with the observation in Section 2.2.1
that it was after his stay in Paris that Marx first began referring to himself as an “economist”
as well as the notion discussed in Section 2.1.2 that the primary engine driving the develop-
ment of Marx’s thought was his desire to engage in polemics with those he perceived to be
his ideological “enemies”, all come together to support the following conclusion.
It was not Marx’s reading of political-economic texts in and of itself that sparked his
novel insights into the inner-workings of capitalist political economy, but rather it was this
reading paired with the stimulating intellectual environment of Paris and his desire to distin-

147
guish himself within this environment via harsh polemics that explain his insights and their
“victory” in the subsequent 19th-century socialist war of ideas.
If it was simply his reading and synthesis of political-economic ideas that gave rise to
his insights, then (as mentioned in the previous section) we would expect his voracious
reading in the British Museum library to have accelerated his ideological shift even further.
However, this reading was performed in the sterile quietude of the British Museum library, in
the depressing emotional aftermath of the defeat of the 1848 Revolutions—an environment
which stood in stark contrast to Marx’s very recent memories of the eruption of class struggle
across the European continent. Indeed, as many of Marx’s biographers have noted, the era
from 1875 to 1883—after Marx’s last edits to Capital (resulting in the first French edition
of 1872–1875) up to the year of his death—represent another period of voracious reading
by Marx, in which he took literally thousands of pages of notes. And yet, after the 1875
circulation of his Critique of the Gotha Programme 24 , Marx’s public output essentially came
to a complete halt25 .
This conception of the necessary conditions for ideological innovation, and the ability to
measure their presence or absence via computational methods, accords with several recent
studies into the mechanics of ideological innovation and influence. However, we argue that
many of these studies are missing the identification of temporal events that enabled us to
identify moments of rapid ideological changes. For example, while the information-theoretic
approach of Barron et al. 2018 successfully captures the presence of, and measures the
magnitude of, these ideological shifts, but cannot explain why they occur at the particular
times when they are observed. Conversely, the alternative information-theoretic framework
used in Murdock et al. 2017 is able to pinpoint—essentially, to create a predictive model
24
Here we use “circulation” rather than “publication” since, for many years, this Critique existed only as
a private letter circulated among SPD elites such as August Bebel and Wilhelm Liebknecht.
25
The contents of these late-life studies by Marx remained mostly unpublished for over a century, and
excluded even from the planned publication of the first MEGA. However, the recent publication of these
studies as part of MEGA2 has sparked a new wave of studies, emphasizing not only their import for re-
interpreting Marx (Anderson 2010) but also for bringing his thought to bear on contemporary issues like
globalization (Davis 2018) and climate change (Saito 2017).

148
of—the reading patterns which correspond to moments of innovation in Darwin’s thought,
but is unable to capture the ways in which these innovations “diffuse” forwards in time: did
a given period of intensive depth-focused reading permanently shift his thought? Or, did it
simply result in a publication, after which point it faded or was replaced by yet another
shift? Our approach, we contend, bridges these gaps and thus provides a full-fledged toolkit
for analyzing continuity and change in political thought over time.

2.A Influence Study Corpora

2.A.1 Marx: Key Works

(See the Marx-Engels Digital Register for the complete listing of Marx’s and Engels’ writings)

Last Name First Name Title Year Language

1 Marx Karl Difference Between 1841 en
2 Marx Karl Difference 1841 fr
3 Marx Karl Differenz 1841 de
4 Marx Karl Prussian Censorship 1842 en
5 Marx Karl Censure Preusse 1842 fr
6 Marx Karl Zensor 1842 de
7 Marx Karl Critique of Hegel’s Philosophie des Rechts 1843 de
8 Marx Karl Critique of Hegel’s Philosophie des Rechts 1843 en
9 Marx Karl Jewish Question 1843 en
10 Marx Karl Judenfrage 1843 de
11 Marx Karl Jewish Question 1843 fr
12 Marx Karl Holy Family 1844 de
13 Marx Karl Holy Family 1844 en
14 Marx Karl Economic and Philosophic Manuscripts of 1844 en
1844
15 Marx Karl Manuscrits 1844 fr
16 Marx Karl Manuskripte 1844 de
17 Marx Karl Holy Family 1844 fr
18 Marx Karl The German Ideology 1846 de
19 Marx Karl The German Ideology 1846 fr
20 Marx Karl The German Ideology 1846 en
21 Marx Karl Das Elend der Philosophie 1847 de
22 Marx Karl Misere de philosophie 1847 fr
23 Marx Karl Poverty of Philosophy 1847 en
24 Marx Karl Communist Manifesto 1848 de
25 Marx Karl Communist Manifesto 1848 fr
26 Marx Karl Communist Manifesto 1848 en
27 Marx Karl Lohnarbeit 1849 de
28 Marx Karl Travail Salaire 1849 fr
29 Marx Karl Wage Labour and Capital 1849 en

149
 Last Name First Name Title Year Language
30 Marx Karl The Eighteenth Brumaire of Louis 1852 de
Napoleon
31 Marx Karl The Eighteenth Brumaire of Louis 1852 en
Napoleon
32 Marx Karl The Eighteenth Brumaire of Louis 1852 fr
Napoleon
33 Marx Karl Grundrisse 1858 de
34 Marx Karl Grundrisse 1858 en
35 Marx Karl Critique 1859 en
36 Marx Karl Critique 1859 fr
37 Marx Karl Kritik 1859 de
38 Marx Karl Herr Vogt 1860 de
39 Marx Karl Herr Vogt 1860 en
40 Marx Karl Mehrwert 1 1862 de
41 Marx Karl Mehrwert 2 1862 de
42 Marx Karl Mehrwert 3 1862 de
43 Marx Karl Theories of Surplus Value 1862 en
44 Marx Karl Lohn 1865 de
45 Marx Karl Salaire 1865 fr
46 Marx Karl Value, Price, and Profit 1865 en
47 Marx Karl Das Kapital Vol. 1 1867 de
48 Marx Karl The Civil War in France 1871 de
49 Marx Karl Civil War in France 1871 en
50 Marx Karl Guerre Civile 1871 fr
51 Marx Karl Le Capital Vol. 1 1867 fr
52 Marx Karl Critique of the Gotha Program 1875 de
53 Marx Karl Critique of the Gotha Program 1875 en
54 Marx Karl Critique Gotha 1875 fr
55 Marx Karl Das Kapital Vol. 2 1885 de
56 Marx Karl Capital, Vol. 1 1867 en
57 Marx Karl Das Kapital Vol. 3 1894 fr
58 Marx Karl Das Kapital Vol. 3 1894 de

2.A.2 Influence Corpus

Last Name First Name Title Year Language

1 Bentham Jeremy Deontologie 1834 fr
2 Blanqui Louis-Auguste Defense 1832 fr
3 Cabet Etienne Ikarien 1840 de
4 Considerant Victor Fourier 1841 de
5 Dante Divine Comedy 1320 de
6 Defoe William Robinson Crusoe 1719 de
7 Diderot Denis Souverain 1774 fr
8 Enfantin Barthélemy-Prosper National 1830 de
9 Engels Friedrich Umrisse 1844 de
10 Feuerbach Ludwig Wesen des Christentums 1841 de

150
 Last Name First Name Title Year Language
11 Fichte JG Reden 1808 de
12 Fourier Charles Theoie des Quatre Mouvements 1808 fr
13 Fourier Charles Reformplan 1838 de
14 Guizot Francois Civilisation 1838 fr
15 Hegel GWF Phänomenologie des Geistes 1807 de
16 Hegel GWF Phenomenologie 1807 fr
17 Hegel GWF Wisssenschaft der Logik 1816 de
18 Hegel GWF Espirit 1817 fr
19 Hegel GWF Geistes 1817 de
20 Hegel GWF Logic 1817 en
21 Hegel GWF Logik 1817 de
22 Hegel GWF Logique 1817 fr
23 Hegel GWF Naturphilosophie 1817 de
24 Hegel GWF Naturphilosophie 1817 en
25 Hegel GWF Philosophie de Nature 1817 fr
26 Hegel GWF Rechts 1820 de
27 Helvetius Claude Espirit 1758 fr
28 Hess Moses Europäische Triarchie 1841 de
29 Hess Moses Paris 1844 de
30 Hodgskin Thomas Verteidigung 1830 de
31 Holbach Baron Prejuges 1780 fr
32 Holbach Baron Natur 1788 de
33 Hume David Politiques 1752 fr
34 Mill James Elements of Political Economy 1821 en
35 Kant Immanuel Religion 1794 fr
36 Lahautière Richard Catechisme 1839 fr
37 Lamennais Felicite Grundriss 1841 de
38 Leibniz Gottfried Monadologie 1714 fr
39 Leroux Pierre De l’Humanite 1840 fr
40 Locke John Traite 1690 fr
41 Malthus Thomas Population 1798 en
42 Malthus Thomas Population 1798 fr
43 Malthus Thomas Population 1798 de
44 Marat Jean-Paul Esclavage 1792 fr
45 McCulloch JR Principles of Political Economy 1825 en
46 McCulloch JR Geld 1859 de
47 Montesquieu Spirit of the Laws 1730 de
48 Osiander JF Handelswerker 1840 de
49 Owen Robert Neue 1830 de
50 Owen Robert New View 1830 en
51 Paine Thomas Sens Commun 1776 fr
52 Paine Thomas Droits 1791 fr
53 Proudhon Joseph Grammaire Generale 1837 fr
54 Proudhon Joseph Celebration du Dimanche 1839 fr
55 Proudhon Joseph Eigentum 1840 de
56 Proudhon Joseph Property 1840 en
57 Proudhon Joseph Qu’est-ce que la propriété? 1840 fr
58 Proudhon Joseph Explications 1842 fr
59 Proudhon Joseph Ordre 1843 fr
60 Quesnay Francois Tableau Economique 1766 fr
61 Reimarus Hermann Samuel Ausfuhrliche 1797 de
62 Ricardo David Principes 1817 fr

151
 Last Name First Name Title Year Language
63 Ricardo David Principles of Political Economy and Taxa- 1817 en
tion
64 Ricardo David Grundgesetze 1838 de
65 Rousseau Jean-Jacques Contrat Social 1762 fr
66 Saint-Simon Henri Neues 1825 de
67 Saint-Simon Henri Nouveau 1825 fr
68 Say Jean-Baptiste Catechisme 1815 fr
69 Say Jean-Baptiste Katechismus 1827 de
70 Senior William Outline 1836 en
71 Sieyes Abbe Tiers Etat 1788 fr
72 Sismondi Simonde Pol Econ 1815 en
73 Sismondi Simonde Grundsatze 1827 de
74 Sismondi Simonde Pol Econ Selections 1847 en
75 Smith Adam Richesse des Nations 1-3 1776 fr
76 Smith Adam Richesse des Nations 1776 fr
77 Smith Adam Wealth of Nations 1776 en
78 Smith Adam Wohnung 1776 de
79 Stein Lorenz Socialismus 1842 de
80 Steuart Sir Okonomie 1767 de
81 Storch HF von Betrachtungen 1825 de
82 Strauss DF Life of Jesus 1835 en
83 Strauss DF Das Leben Jesu 1836 de
84 Voltaire Traite 1763 fr

152
 Chapter 3: Forged in Battle: The Fall of Utopian Socialism and
the Rise of Marxism, 1849–1914

In this study, we turn from an analysis of the earlier influences on Marx to an analysis of
Marx’s subsequent influence on the political discourse of his own time—in particular, his
influence on the rhetoric and practice of the European socialist movement from the 1850s
onward. We ask: what did he do with the concepts bequeathed to him by his influences, those
analyzed in Part I? How did he try to “steer” socialist discourse via his interventions, and
how successful was he in these various attempts? We begin by again construct an embedding
space, this time “tuned” to represent authorial position-taking within European socialist
polemics of the time, and argue that the movement of Marx’s discourse relative to this
general socialist discourse strongly supports the perspective that (a) his aim was to “pull”
socialists away from moralistic discourse and towards a more positivistic political-economic
discourse, and that (b) he was successful in this endeavor, as evidenced by the similarity
between the trajectory of Marx’s thought and the subsequent trajectory of general socialist
thought within this ideological (moralistic vs. political-economic) space. Then, to further
cement these claims, we move from a qualitative analysis of our diachronic embedding space
to a quantitative analysis of particular influence dyads—for example, to what degree do the
lexical-semantic changes observed in the language of socialist newspapers after 1848 mirror
the lexical-semantic structure of Marx’s pre-1848 writings?—and find that, indeed, Marx
emerges as the most influential node in the broader network of major nineteenth-century
socialist texts.

3.1 Introduction

Perhaps the single key driver of the development of Marx’s thought over the course of his
life was his almost obsessive impulse to engage in polemics with those he viewed as obstacles

153
to (in his early years) or rivals to (in later years) his assumption of the throne as the intellec-
tual leader of the 19th century European socialist movement. We have already encountered
this tendency in previous sections, when discussing the development of his thought with re-
spect to the “stationary targets” of his criticism—namely, the already-established big names
in political economy, early socialism, and philosophy such as Adam Smith, David Ricardo,
Henri de Saint-Simon, and Hegel. In this section we turn to Marx’s forward-facing polemics
against his various interlocutors, his critique of the “moving targets” vying with him for
intellectual influence over the rapidly-growing European socialist movements of the era.
Central to this section is the move from influences to interlocutors. While Marx could
isolate himself in his studies of the former for as long as he needed to develop cogent cri-
tiques, the European socialist movement moved at its own pace, requiring his critiques of
the latter to be not only cogent but timely as well. Engels, for example, urged Marx in
1845 to complete and publish his already-in-progress politial-economic tract as soon as pos-
sible, emphasizing that “people’s minds are ripe and we must strike while the iron is hot”1 .
The urgency heightened even more when, in the ensuing year, the prominent French social-
ist Pierre-Joseph Proudhon published his own such tract titled Système des contradictions
économiques. As Keith Tribe puts it in his insightful analysis of Marx’s subsequent polemics
with Proudhon, “the publication of Proudhon’s Système galvanised [Marx] into writing a
shorter work so that he might stake his claim to being Western Europe’s foremost ‘radical
political economist’. [Marx’s 1847] Misère de la philosophie is primarily a bid for market
leadership, and it certainly reads that way.” (Tribe 2015, p. 224)
Similarly, after having spent a decade in exile in London desperately working to maintain
his relevance within the growing socialist movement of his home country, the 1859 publication
of an attack on him in a German newspaper persuaded him to drop his political-economic
studies for nearly two years in order to craft a response which would quell the “foggy gossip
of the [1848] refugees”2 . The attack, penned by a minor figure from the 1848 Frankfurt
1
MEW Vol. 27, , p. 16.
2
MEW Vol. 30, , p. 17.

154
Assembly named Karl Vogt, cajoled Marx into working full-time to collect evidence against
Vogt, culminating in a 208-page work, Herr Vogt, for which he was never able to find a
German publisher. Thus, in the remainder of the section, we take seriously the centrality
of these “wars of ideas” in the development of Marx’s thought by turning to an analysis
of his illocutionary moves—what he was doing with these polemical interventions, and how
exactly he positioned himself within spaces of ideological debate, in a way which eventually
succeeded in crowding out all other contenders.
While some polemical episodes are not considered herein—for example, his 1847 polemic
against Karl Heinzen after the latter’s attack on Engels in the Deutsche-Brusseler Zeitung 3 ,
or his 1865 debate with his fellow International Working Men’s Association (IWMA) Gen-
eral Council member John Weston4 —we focus on four such episodes that were particularly
impactful, we argue, with respect to the development of Marx’s thought. We begin with his
first major “public break” (his break with the liberalism of his father and schoolmasters in
Trier, via his turn to Hegel in the late 1830s, only being known to us by way of his pri-
vate letter to his father cited previously), his polemics against the pure philosophizing of
the Young Hegelians between 1843 and 1845, culminating in his justly famous “Theses on
Feuerbach” which asserted his commitment to a philosophy of praxis – a commitment to
engaged philosophy which aimed not only to understand society but also to change it.
Next we move to the period from 1845 to 1849, characterized by the buildup to and
eventual defeat of the 1848 revolutions which swept across Europe. We focus especially
on his critique of “state socialists” like Louis Blanc here, since this became a central focal
point around which socialist polemics swirled after Blanc’s appointment to the revolutionary
provisional government in 1848, where he was expected to begin implementing his scheme
3
Culminating in the article “Moralising Criticism and Critical Morality. A Contribution to German Cul-
tural History. Contra Karl Heinzen”, published in the same newspaper in October and November of that
year. MECW 6, , p. 312.
4
Culminating in an 1865 address to the IWMA in which Marx introduced his mature political-economic
views (published long-form in the first volume of Das Kapital two years later), which was published posthu-
mously in the form of a pamphlet entitled Value, Price, and Profit.

155
for “national workshops” as outlined in his tract Organization of Labor (1840)5 .
In the years after the downfall of the 1848 revolutions and the inauguration of the First
International in 1864, Marx’s work becomes less attributable to polemics against individ-
ual rivals so much as against wrong-headed tendencies he observed in European socialist
thought. In particular, he took aim at the overly-sentimental strands of this thought which
had pinned hopes for victory in 1848 on wishful thinking (i.e., that the righteousness of the
workers’ cause entailed its inevitable triumph, or that the superior logic of left-wing Assem-
bly members would convince the ruling class to accept or even embrace socialism) rather
than a dispassionate analysis of the mechanisms by which power operated and how power
structures could be dismantled. Thus Bertram Wolfe explains Marx’s single-minded focus on
developing a “scientific” basis for socialism in these years—evinced by his meticulous reading
and note-taking from hundreds of scientific works in the British Library museum—as him
“seeking ‘scientific’ solace for the shipwreck of his dreams of the 1840s” (Wolfe 1969).
For many interpreters of Marx’s thought, though, it was precisely this rude awakening
and subsequent dive into political economy that enabled Marx’s most trenchant insights into
the workings of the capitalist system. Eric Hobsbawm, for example, asserts that “Marx rose
to the full heights of his genius as the historian of the collapse of the Second Republic and
the victory of Napoleon III”, referring to his first published work after the failings of 1848,
the Eighteenth Brumaire of Louis Napoleon (which began publication in December 1851). As
we will see below, 1851 indeed emerges as the most significant change-point in the semantic
content of Marx’s writings over the course of his life, even more so than works like Capital
usually presumed to be much more central to Marx’s “mature” thought. This presents a
puzzle for narratives of the development of Marx’s thought, which typically view his writings
from 1849 until his 1857 work on the Grundrisse as consisting mainly of journalistic grunt-
5
In reality, Blanc was almost completely hamstrung in these efforts from the start, and almost surely
doomed to failure. While public perception, shrewdly encouraged by his opponents, was that he had been
tasked with implementing the workshops, in fact he had only been appointed to head the “Luxembourg
Commission” where he labored to complete a report on the feasibility of his schemes, thus keeping him
preoccupied while forces of reaction and monarchical restoration worked to defeat the gains of 1848 outside
of these meetings. See, e.g., Agulhon 1983.

156
work performed for the sake of financial survival. Thus in the chapter’s concluding section
we argue for the veracity of these findings, emphasizing the strength of our computational
approach in its ability to detect important signals of developments in Marx’s political and
economic thought within the “noise” of this often dry day-to-day journalism.

3.2 Marx’s Interlocutors

3.2.1 The Young Hegelians, 1843–1845: Bruno Bauer and Max Stirner

Although he had contemplated critiques of prominent Hegelians like Georg Hermes6 or

Karl Rosenkranz as early as 1839, it was his critique of Hegel’s Philosophy of Right in 1843
that scholars typically point to as marking the beginning of Marx’s period of “committed”
polemics against the Young Hegelian milieu which he had initially identified himself as part
of. His first true engagement with a Hegelian “interlocutor”, however—as opposed to Hegel
himself, who had been dead for 12 years when Marx wrote the Critique of his Philosophy of
Right—came in the form of his writing “On the Jewish Question”, an explicit response to
Bruno Bauer’s intervention into debates around the political rights of Jews in Prussia. The
themes of this work were further expanded upon, and the scope of his polemics expanded to
include Max Stirner and Ludwig Feuerbach, in his first joint work with Engels in 1844, The
Holy Family. Marx considered this work to be his “final break” with the Young Hegelians,
with the exception of Feuerbach – his final break with the latter would not come until his
second joint work with Engels, the German Ideology, written in 1846 but never published in
Marx’s or Engels’ lifetime (in fact, not published until the 1930s).

3.2.2 “State Socialism” I, 1845–1849: Louis Blanc

Although a number of works have explored Marx’s views on the State in detail (most
notably Miliband 1965; Hunt 1974 and Hunt 1984; Draper 1977a; and Chapter 7 of Leipold et
6
In a July 9, 1842 letter to Arnold Ruge Marx writes, “I shall probably be drawn into a prolonged polemic
with the Cologne Hermes.” (MECW 1, p. 390) Later, in a letter thought to be composed on August 25, 1842,
Marx asks Dagobert Oppenheim to send him “all Hermes’ articles against the Jews.” (MECW 1, p. 391)

157
al. 2020), these investigations often devote much of their focus to Marx’s work after the rise
and fall of the Paris Commune in 1871, especially his 1875 Critique of the Gotha Programme.
In this section, however, we hope to draw more attention to Marx’s (admittedly more opaque
and imprecise) work from an earlier era, namely, the years leading up to the 1848 revolutions.
In his writings from this period one can already infer the main characteristics of Marx’s
burgeoning socialist conception of the state, especially in his criticisms of European “state
socialists” like Louis Blanc who envisioned a non-revolutionary path to socialism set into
motion by the introduction of universal suffrage and constitutional constraints on power.

3.2.3 Anarchism I, 1846–1849: Pierre-Joseph Proudhon

Pierre-Joseph Proudhon’s thought, as our quantitative analysis will corroborate, was

an eclectic and self-taught fusion of his engagement with theology and philology in the
1830s with his subsequent engagement—predating Marx’s own by a few years—with British
political economy and Hegelian philosophy. In fact, although details on their engagement are
scarce7 , some researchers accept Marx’s post-falling-out contention that he taught Proudhon
everything he knew about Hegel.
In fact, although many works on the early development of Marx’s thought contend that
his transition from Young Hegelianism to British political economy was sparked by Engels’
1843 engagement with the latter, as published in his “Umrisse zu einer Kritik der Nation-
alökonomie” (MEGA2 I/3, pp. 467–494; MECW 3, pp. 418–443), a good case could be made
for the hypothesis that Proudhon played a not-insignificant role. As a comparison of their
reading notebooks (Table 3.1) attests, Proudhon had already read many of the texts noted as
central to this Engels-to-Marx transmission three or four years before Marx began studying
them.
We therefore focus primarily on Proudhon as a key conduit of influence on Marx, from
the latter’s October 1843 move to Paris onwards. We argue, in fact, that the importance of
7
Scarce despite numerous research endeavors, typically in the same anarchists-versus-Marxists vein as the
studies on Marx and Bakunin described in more detail in Section 3.2.6 below).

158
 Text Proudhon Marx
Adam Smith, Wealth of Nations (1776) Oct 1841 Mar–Aug 1844 (a)
David Ricardo, Principles (1817) Oct 1841 1844 (b)
Charles Comte, Traité de la propriété (1834) 1839 — (c)
F. X. J. Droz, Propriété (1832) 1839 1845–1847 (d)
A. Destutt de Tracy, Économie politique (1823) 1839 — (e)
Adolphe Blanqui, Hist. de l’econ. pol. (1837) 1839 1845 (f)
Dugald Stewart, Esquisses de phil. morale (1793) 1839 1858–1862 (g)
J. B. Say, Cours complet de Écon. pol. (1828) 1839 Oct 1843–1845 (h)
A. A. Cournot, Principes mathém. d’Écon. pol. (1838) 1839 — (i)
P. Rossi, Cours d’Écon. pol. (1836) Jan 1840 1845 (j)
G. Garnier, Da la propriété (1792) Nov 1840 — (k)
A. Ciezkowski, Du crédit et de la circulation (1839) Oct 1841 — (l)

Table 3.1: A comparison of the dates of first reading for key political-economic texts, as
recorded in Proudhon’s and Marx’s respective reading notebooks. On Proudhon’s reading
notebooks, see Appendix 3.B.1. On Marx’s, see Section 2.1.3.2. Entries after Smith and
Ricardo are listed in the order in which they appear in Haubtmann 1982. Sources for each
date of reading are given in Appendix 3.C.

polemical rivalry as an accelerant to Marx’s intellectual endeavors—a dynamic we’ve empha-

sized throughout this chapter—means that Marx’s engagement with political economy was
probably far more influenced by his desire to “usurp” Proudhon’s bid for the role of leading
radical European economist than by his reading of Engels’ “Umrisse”8 . Indeed, one could
construct many feasible counterfactual trajectories of Marx’s thought based on polemical
works proposed by or to him, but never completed, as revealed through his correspondence:
from the critique of Hermes mentioned above, through Ruge’s request for an analysis of
German art in 1842, to his 1858 intimation to Lassalle that he hoped to write a pamphlet
revealing the “rational kernel” of Hegel’s dialectic. Thus, in a move we’ll see repeated time
and time again throughout Marx’s life (for example, after Karl Vogt’s 1859 attack on him
discussed in the next section), upon arriving in Paris and seeing the praise heaped upon
Proudhon’s Hegelian-terminology-infused social theories, Marx quickly dropped all previous
8
We formulate the point in this way—as a “bid” to “usurp” Proudhon—to highlight our indebtedness to
the interpretation given in Tribe 2015, especially with regard to Marx’s foreword to the Misère de Philosophie,
cited above.

159
engagements, put his planned polemics on hold, and devoted nearly all his energy to this
project of dethronement.
Though our focus is on Proudhon, for these reasons among others9 , the case should not be
overstated. As argued in Gregory 1983, for example, while Marx and Engels both probably
considered Proudhon to be “the” key figure in French socialism upon their arrival, they
nonetheless acknowledged several other French socialists as rivaling Proudhon in importance.
Marx (not yet a communist), in an article of October 1842, responds to critics accusing the
Rheinische Zeitung of communist leanings by contending that

such writings as those of Leroux, Considérant, and above all the sharp-witted
work by Proudhon, cannot be criticised on the basis of superficial flashes of
thought, but only after long and profound study.

A year later (in the same month of Marx’s move to Paris, October 1843) Engels, while
singling out Proudhon as “The most important writer” among French socialist thinkers, also
marvels at the fact that the “Icarian Communists” (followers of Etienne Cabet) can count
“half a million” among their ranks, and singles out the philosophical contributions of Pierre
Leroux and his newspaper the Independent Review.
Indeed, as William Clare Roberts argues in Marx’s Inferno (Roberts 2016), this small
group of thinkers (Leroux, Considérant, Proudhon, and Cabet) represented just the tip of the
iceberg of a much broader socialist movement taking shape among workers in Paris at that
time—a movement which Marx would be introduced to shortly after his arrival via face-to-
face engagement with these radical Parisian workers themselves (rather than the academic
philosophers who had served as his peers over most of the previous decade). Marx arrived in
the French capital at a time when French radical thought was developing a novel syncretic
9
There is also the unfortunate but unavoidable fact that Proudhon’s writings have been far more metic-
ulously collected and catalogued than those of most other 19th-century French socialists. Thus, while we do
perform our studies of French socialism on two corpora—one consisting only of Proudhon’s works but the
other containing a broader selection of important French socialist texts which may have come into Marx’s
purview—it is only with respect to Proudhon’s works that we can claim to have statistically-representative
coverage (statistically-representative because exhaustive, as documented in Appendix 3.B.1).

160
fusion of republicanism and socialism, with the former rooted in the ideals of the French
revolutionaries of 1789 and the latter drawing more on the subsequent proto-communist
radicals of 1796. That is, while the core French republican strand drew inspiration from the
early days of the Revolution, through figures such as Abbé Sieyes and Maximilien Robespierre
(Leipold et al. 2020), French socialism could be distinguished from this “pure” republican
tradition in terms of its shift of focus towards the later stages of the Revolution, especially
in the thought of François-Noël “Gracchus” Babeuf, leader of the failed 1796 “Conspiracy
of the Equals” coup which aimed to replace the Directory with an egalitarian proto-socialist
republic.
Notably, however, most of the progenitors of this republican-infused socialism did not
derive its tenets directly from Babeuf’s thought. They were galvanized instead by a second-
hand account of the coup published in 1828 by his co-conspirator Filippo Buonarroti, Histoire
de la Conspiration pour l’Égalité dite de Babeuf (History of Babeuf’s Conspiracy for Equal-
ity), which would quickly inspire 19th-century revolutionaries like Louis-August Blanqui to
devote the rest of their lives to clandestine revolutionary agitation, forming a slew of secret
societies which spread throughout France over the course of the 1830s. Thus, to capture this
additional layer of French socialist thought, we augment our corpus of Proudhon’s collected
writings with a corpus of French socialist writings like those of Buonarroti, Blanqui, Leroux,
and others, as described in Appendices 2.A and 3.A, and use this second corpus alongside
the Proudhon corpus throughout our study.

3.2.4 Challengers to the Throne I, 1859–1860: Karl Vogt

As chronicled by several of his biographers, Marx’s political-economic writing incurred

several major interruptions in the form of protracted polemics against other socialists whom
Marx viewed as potential opponents (or perhaps saboteurs) for hegemony over European
socialist discourse. In 1859 and 1860 for example, after the publication of his Zur Kritik, Marx
abruptly ceased working on his “Economics” and began a polemic with Karl Vogt, a minor

161
figure from the 1848 Frankfurt Parliament10 who had slandered him in a German newspaper.
The extent to which his need to strike back suddenly superseded all other concerns is aptly
described by David McLellan in his biography of Marx:

This quarrel, which occupied Marx for eighteen months, is a striking example
both of Marx’s ability to expend tremendous labour on essentially trivial matters
and also of his talent for vituperation. (McLellan 1973, 311)

Even Engels himself, normally supportive of Marx in all his endeavors, diplomatically begged
the latter not to allow the “Vogt affair” to interrupt his political-economic studies:

The prompt appearance of your second installment11 is obviously of paramount

importance in this connection and I hope that you won’t let the Vogt affair stop
you from getting on with it. [...] I am very well aware of all the other interruptions
that crop up, but I also know that the delay is due mainly to your own scruples.

Marx did not heed Engels’ plea, however, and pressed onwards with his 18 months of work
on what was to become the 208-page Herr Vogt. Characteristically, however, Marx was never
able to find a German publisher, thus defeating the entire purpose of the work in the first
place.

3.2.5 “State Socialism” II, 1864–1883: Ferdinand Lassalle

Ferdinand Lassalle, a German socialist intellectual and agitator12 played a variety of

seemingly-incongruous roles in the development of Marx’s post-1848 social and economic
10
We borrow this characterization of Vogt as a “minor figure” from the bulk of the literature on Marx
and on the specifically intellectual development of 19th-century German socialism in general. Vogt was less
“minor”, for example, with respect to the personal lives of the post-1848 exiles or to the development of
German science. His central role in the feud between Russian intellectual Alexander Herzen and German poet
Georg Herwegh is described throughout Carr 1933, while his role in 1850s German “materialist controversy”
is described in Beiser 2014, Ch. 2.
11
Referring to the “sequel” to Marx’s 1859 Contribution to the Critique of Political Economy (Zur Kritik
der Politischen Ökonomie), i.e., to the work that would eventually coalesce into the three volumes of Das
Kapital.
12
The rendering of his surname as “Lassalle” is actually a Gallicization of his family name, Lassal, a
spelling he promulgated early on to deflect attention away from his Silesian origins, as part of his goal to
establish himself as a radical intellectual in Paris starting in the mid-1840s.

162
thought. A study of MEGA2, for example, would give an impression of Lassalle as some-
one with whom Marx tried to remain on good terms via his direct correspondence, despite
harboring an intense disdain towards him (which comes out in his descriptions of Lassalle
in letters to Engels over the same time period), a balancing act which was “resolved” by
Lassalle’s early death after an 1864 duel.
If one takes into account late-19th-century developments in European socialism immedi-
ately before and after the 1875 establishment of the German SPD, however, it emerges that
in fact Lassalle’s immense posthumous influence outlived even Marx himself. It wasn’t until
Engels’ work moulding the ideology of the SPD over the 12 years following Marx’s death that
Lassalleanism was finally “defeated” as a viable competitor to Marxism among European so-
cialists. The magnitude of this shift can be seen in hindsight, for example, by considering
the anecdote cited in Reichard 1969 that “at the end of the protocol of the [SDAP’s] 1870
congress was a list of available literature: two works by Engels appeared, three by Marx, and
twenty-one by Lassalle” (259).

3.2.6 Anarchism II, 1872–1883: Mikhail Bakunin

Mikhail Bakunin, hailed by some as the father of modern anarchism, is typically cast
in the role of Marx’s main rival in the First International between 1868 (the year Bakunin
joined) to 1872 (when Bakunin and his followers were expelled by Marx), in socialist and
anarchist histories alike (see, e.g., Eckhardt 2016). Interestingly, however, Marx and Bakunin
crossed paths fairly regularly, in substantial ways, from 1840 onwards. To name just one
rarely-mentioned instance, Bakunin produced the first Russian translation of the Communist
Manifesto, which was published in the periodical Kolokol in London in 186013 . Starting the
narrative of the Marx-Bakunin relationship in 1864, therefore, ignores a great number of
interactions which impacted the development of Marx’s thought.
13
See Guillaume 1905, p. 283, cited in Favilli 1996. Bakunin had a number of path-crossings with Engels
over the years, as well. They were both in attendance, for example, at F. W. J. Schelling’s infamous 1841
lectures at the University of Berlin (Hunt 2010, 44–46).

163
 Bakunin moved from Moscow to Berlin in 1840 to enroll at the University of Berlin—the
same university where Marx had been studying since 1836—and quickly became a prominent
figure in the Young Hegelian movement alongside Marx, Bruno and Edgar Bauer, and Arnold
Ruge. Bakunin and Marx both, in fact, contributed articles to Ruge’s Deutsche Jahrbücher
fur Wissenschaft and Kunst in 1842, though Marx’s contribution (ironically, a commentary
on Prussian censorship restrictions) was censored by the Prussian government and only
published a year later in Switzerland. After the Prussian government banned this publication
outright in 1843, Marx and Ruge moved to Paris to co-found the Deutsche-Franzosische
Jahrbücher, with Bakunin joining them in the city that same year. After finally meeting in
person in 1844, Marx and Bakunin corresponded in a mostly-cordial fashion for decades, up
until the 1872 split of the First International. Even as late as 1871, for example, Bakunin
accepted a commission to produce the first Russian translation of Volume 1 of Das Kapital,
a work which he deeply admired, having earlier commented that “no other work that I know
of puts together such a profound, enlightening, scientific, decisive analysis” of the capitalist
economy.
The 1872 split and the four years leading up to it—an episode of Marx’s life which Alvin
W. Gouldner calls “the culminating conflict of [Marx’s] political life” (Gouldner 1982)—
have been exhaustively documented in two parallel literatures, which present two starkly
contrasting narratives. The first narrative, promulgated most heavily in the Soviet Union,
sees Marx effortlessly fusing theoretical insight with organizational prowess, keeping the
International sharply focused on its proletarian revolutionary aims despite the best efforts
of the saboteur Bakunin14 . The second narrative, promulgated by both Western anti-Soviet
historians and anarchists in nearly identical forms, sees Marx ruthlessly stamping out any and
all anti-authoritarian voices in the International, with Bakunin finally giving up on his noble
but quixotic efforts in 1872 to found the aptly-named Anti-Authoritarian International15 .
For the purposes of this work, however, it suffices to say that Marx viewed Bakunin
14
See Nicolaievsky and Maenchen-Helfen 1936, pp. 280–297, for a fairly inocuous example.
15
A stark example of this contrasting narrative can be found in Eckhardt 2016.

164
as a key rival for leadership of the European socialist movement. Public perception and
commentary on this movement, especially in the years leading up to the split, often compared
the two, adding fuel to Marx’s competitive fire. The Italian socialist newspaper La Plebe,
for example, characteristically referred to Marx as “Germany’s Bakunin” in a major article
of January 187216 . Hence, as is the case with nearly all of his works, Marx’s discursive
interventions throughout the era of the First International were driven primarily by polemical
concerns.
Just as it is inappropriate to begin the Marx-Bakunin narrative in 1868, it is also in-
appropriate to end it with the 1872 split. Two years after the split, in April of 1874, Marx
began reading Bakunin’s Staatlichkeit und Anarchie (Statism and Anarchy). By the time
he finished in January of 1875, he had copied 224 separate extracts into his notebook in
Russian, some spanning several pages. He provided extensive commentary on 39 of these,
breaking out of the extracts and writing paragraph-length or even page-length responses, in
addition to the shorter inline comments he made on nearly all of them (ranging from single
exclamation points to parenthetical definitions, translations, and quips)17 .

3.2.7 Challengers to the Throne II, 1883–1884: Eugen Dühring

When in the post-Kapital V1 years another contender for Marx’s historiographic crown
emerged, Eugen Dühring, Engels (having learned from the Karl Vogt episode) consciously
opted to take the lead and conduct the polemics himself so Marx could carry on with his
work on Kapital. Although Marx did end up contributing in a non-trivial way to the resulting
book Herr Eugen Dühring’s Revolution in Science (typically shortened as Anti-Dühring), it
was published under Engels’ name in 1884, a year after Marx’s death.
Unlike in the case of Vogt, however, Dühring was a worthy opponent, a major intel-
lectual figure in Germany who wielded great influence and thus directly challenged the
16
“Lettere da Berlino”, La Plebe, Jan. 5, 1872, cited in Favilli 1996, p. 32. See ibid. pp. 20–46 for an
in-depth discussion of the Bakunin-Marx rivalry and its relation to 19th-century Italian socialist thought.
17
Our calculations, based on MEW Vol. 18, , pp. 597–642.

165
recently-acquired gains in Marx’s prominence and notoriety after the rise and fall of the
Paris Commune. Dühring’s published works would have an immediate and substantial im-
pact on German political-philosophical discourse—albeit an impact fairly distant from the
epicenters of socialist discourse—with Friedrich Nietzsche being only one of many prominent
post-Hegelian thinkers who were profoundly influenced by Dühring18 .

3.3 Diachronic Embeddings: Marx and the Socialist Movement, 1849–1899

As can be seen in Figure 3.1, we indeed observe a time-lagged movement of the Socialist
centroids for each decade in the same direction as Marx’s, namely, in the direction moving
away from the Hegelian centroid and towards the political-economic centroid, from 1840
onwards. Interestingly, this comes after a move towards the Hegelian centroid between 1830
and 1840, which tracks closely the increasingly polemical debate among Hegelians which
eventually gave rise to distinguishable Left-, Right-, and Center-Hegelian tendencies by the
1840s19 . Marx for example, though he had engaged with Hegel’s texts directly as early as
1837, found himself thrust into the center of these debates after transferring from Bonn to
Berlin, with one of his early courses being taught by Eduard Gans, “the most important
Hegelian” of the era (Heinrich 2019, p. 160)20 .
To aid in the interpretation of this figure, i.e., what it means for a particular time-indexed
centroid to move within this ideological space, we present PCA plots of the Political Economy
centroid in Figure 3.3 and of the Hegelian centroid in Figure 3.4.
18
Nietzsche’s reading included nearly all of Dühring’s published works, some of which he read on multiple
occasions—see the supplemental dataset on Nietzsche’s known and conjectured reading described and linked
in Appendix 3.B.3. For a summary of the main trends in post-Hegelian German philosophy, see Beiser
2014, pp. 172–184 (“Dühring on the Value of Life”), where Dühring’s “important place in the history of
nineteenth-century philosophy” includes his role as “the founder of German positivism, the grandfather of
Schlick, Carnap, Neurath, and Reichenbach.” (p. 174)
19
Following Moggach 2006, however, we refer to members of this newly-polarized generation of Hegelians
as “New Hegelians”, to avoid the contentious exercise of classifying them neatly into such left, right, and
center camps.
20
Gans also, crucially, edited and appended unabashedly polemical editorial notes to the text of Hegel’s
Philosophie des Rechts, along with a Preface which “highlighted the liberal content of Hegel’s Philosophy
of Right and defended him against the accusation of having philosophically legitimized the Restoration.”
(Heinrich 2019, pg. 165) It is Gans’ critical edition which Marx would eventually read and use as the basis
for his Critique of Hegel’s ‘Philosophy of Right’, discussed above.

166
Figure 3.1: The trajectories of Marx’s decade-by-decade centroids and the decade-by-decade
centroids of European socialist discourse, with respect to the static Hegelian (defining the
x-axis) and Political-Economic (defining the y-axis) centroids.

The movement of the socialist discourse centroid from 1840 onwards, the era in which
Marx’s impact can be measured starting with his 1842 articles for the Rheinische Zeitung,
can best be understood by first isolating the time-series of Marx’s own trajectory with respect
to the Political Economy centroid. In Figure 3.2 we disaggregate the decade-by-decade split
used in Figure 3.1, by computing a series of Political Economy similarity scores based on
the 90-day rolling average of distances between each of Marx’s writings and the Political
Economy centroid. And, indeed, the plot reveals the importance of Marx’s Critique of Hegel’s
‘Philosophy of Right’ in his intellectual development: after hitting a global minimum with
Die Kommunalreform und die Kölnische Zeitung—discussed in Levine 2012 (pp. 82–83) as a
key text representing the clearest statement of Marx’s earliest Center-Hegelian phase—the
Critique marks the beginning of a rapid shift away from the Hegelian centroid (the negative
values in the plot) and towards the Political Economy centroid (the positive values). The rise
and fall of the 1848 revolutions accelerates this shift in Marx’s thinking even further, such
that his writings cross the zero-point (the point at which the writings resemble Hegelian and

167
 Figure 3.2: Rolling PE Similarity Scores for Marx’s Collected Writings

Political Economic semantics in equal measure) at almost the exact same time as the 1848
breakout of demonstrations in Vienna, and never again come close to the Hegelian similarity
of his early writings.
Another noteworthy feature of the plot can be seen by examining its trajectory before,
during and after the period (demarcated by vertical red lines) between the September 1867
publication of Capital Vol. 1 and the March 1871 proclamation of the Paris Commune. Given
the intense rise in Marx’s notoriety in the wake of the Commune (due to his leadership in the
IWMA), European radicals seeking out Marx’s most recent works at this point would find
Capital Vol. 1 and other heavily political-economic writings. However, as Shanin 1983 and
others have pointed out in recent decades, Marx’s studies did not come to an end with the

168
publication of Capital or The Civil War in France, but in fact swerved in a new direction,
especially after he taught himself Russian, trying to make sense of the economic development
of “peripheral” (relative to the “core” of Western and Central Europe) societies like Russia
and India.
Thus, while the plot shows a notable shift away from the Political-Economic centroid
and back towards the Hegelian centroid in his post-1871 writings, we hypothesize that this
in fact represents only the residual effect of what is in reality a rapid shift towards a third
centroid representing Historical-Anthropological discourse. In particular, the correspondence
of both Marx and Engels after 1871 reveals their deep engagement with works by figures like
Lewis H. Morgan, J. B. Phear, Henry S. Maine, and John Lubbock21 , among many others,
and the fairly significant changes in their thought which came as a result22 .

3.4 Semantic Leadership and Socialist Discourse

While the methods detailed in the previous two sections allow us to generate author-
specific embeddings and visualize the trajectory of these authors through a transparently-
defined and reproducible ideological space (the Hegelian vs. Political-Economic space, in our
case), these results remain impressionistic. The correlations we see in these visualizations, of
the socialist movement’s centroid trailing Marx’s over time, may in fact be the result of the
standard caveats when drawing causal inferences from statistical associations: for example,
21
We mention these four authors in particular as they are the only authors for whom Marx’s studies have
(partially) been made available, via Krader 1972, to English readers. Even in the German-speaking world
Marx’s post-1871 notes and excerpts are hard to come by, as the planned volumes of MEGA2 covering this
period have yet to be published, with the exception of MEGA2 IV/26 (notes on geological, mineralogical,
and agricultural works from March to September 1878) and MEGA2 IV/31 (notes on scientific works made
from mid-1877 to the beginning of 1883), neither of which contains any of the studies excerpted in Krader
1972. With the completion of the digitization of Marx’s notebooks by the IISH, however, these studies in
Marx’s original handwriting can be obtained via the IISH website (with notebooks B120 to B168 in particular
covering the period after 1871).
22
On this point see, for example, the series of lengthy drafts Marx composed for a letter to Vera Zasulich
(reprinted in Shanin 1983, pp. 97–126), who had written him asking for his perspective on the prospects for
communist revolution in Russia, but also his engagement throughout the 1870s and 1880s with writings of
prominent Russian intellectuals like N. G. Chernyshevskii (whose 1863 novel What Is To Be Done? also had
a significant impact on Lenin, inspiring his 1902 pamphlet of the same name), V. V. Bervi-Flerovskii, and
M. M. Kovalevskii, discussed in Wada 1981.

169
 PCA Plot of Political Economy Centroid
bank Type
Word
handel bank Centroid
Second Principal Component

england
trade
england
nation
nation land
land geld bevolkerung
ganz gold population
all hoch
PolEcon* nachfrage verhaltnis steigen
folge fall mehr high
boden demand relationship preis rise
consequence fall surplus gleich kapital arbeiter
weniger ground price
fewer zeit equal capital worker
time arbeit
labor
First Principal Component

Figure 3.3: A Principal Component Analysis (PCA) plot of the Political Economy centroid
with the top N = 25 words most distinctive to this subcorpus.

PCA Plot of Hegelian Centroid

Type
jesus Word
jesus Centroid
Second Principal Component

unmittelbar
directly
uberhaupt
bestimmen andern
in generaldenken
to determine to change zugleich to think christ
beziehung selbst
unterscheiden simultaneously christ
relationship Hegelian* self wahrheit
differentiate
bestimmtheit form inhalt begriff absolut truth religion
natur geist
bestimmtheit formcontent concept absolutenatur spirit religion
gott
einheit wirklichkeit bewusstsein god
unity awareness
dasein reality wesen
subjektiv abstrakt dasein being
subjective abstract
First Principal Component

Figure 3.4: A Principal Component Analysis (PCA) plot of the Hegelian centroid with the
top N = 25 words most distinctive to this subcorpus.

a third omitted source of influence may be driving both trajectories.

Thus, to move closer to a principled analysis of the causal impact of Marx’s writings on
the European socialist movement, we draw on recent work in the digital humanities which
leverages statistical and computational-linguistic methods to infer linguistic influence with
more precision than what is possible to infer solely from a time series of centroids.

170
 In particular, we construct “Semantic Leadership Networks” (Soni et al. 2021) for our
corpus of socialist texts, with different cutoff dates chosen to test different hypotheses regard-
ing influence. In our first experiment, for example, we use the end of the 1848 Revolutions
as a cutoff date, to measure the influence of the thought of “early Marx” on post-1848 Eu-
ropean socialist discourse. In our second, we instead choose the final publication of Capital
that Marx directly participated in (and the final publication of Capital in his lifetime), the
1872–1875 French edition, as the cutoff date, to measure instead the influence of Marx’s
mature political-economic theory on late-nineteenth-century European socialist discourse.

3.4.1 Quantitative Results

Although tracking the literary output of Marx’s interlocutors is generally a much more
difficult task than tracking Marx’s own23 , a few collections of key rivals exist which enable
us to “break apart” the Socialism vector into vectors for particular authors. With these
individual-author vectors, we can evaluate which authors in particular had trajectories which
drove the overall shift observed in the previous section.
Pierre-Joseph Proudhon’s writings are of special interest with respect to the computa-
tional nature of this work, since not only do we have his collected writings but also newly-
digitized scans of his notebooks, the Carnets, containing (as in the case of Marx) not only
drafts of his works at various stages but also the meticulous reading notes and extracts he
kept over the course of his life. An overview of this dataset, used throughout our analy-
sis in Section 3.4.1.1 below, is given in Section 3.2.3 above, with more details provided in
Appendix 3.B.124 . As for his main body of writings—those which were intended for public
consumption—several large collections have been compiled, starting as early as 1850 when
he was still actively publishing new works. This first 26-volume collection was completed
23
Ostensibly due to the fact that, unlike Marx, most 19th-century thinkers did not end up having a
global superpower collecting and propagating their texts—on this topic, see our discussion of the Soviet
“weaponization” of Marxism in Jacobs 2021b.
24
For an overview of the Carnets in general, see the section entitled “Notes et Annotations Diverses” in
the Appendix of Haubtmann 1982, pp. 1079–1098.

171
in 187225 . The most frequently-referenced collection of Proudhon’s works, however, and the
collection we utilize in our study, is known as the “Rivière edition”:

Oeuvres complètes de P.-J. Proudhon, nouvelle édition, ed. C. Bouglé et H. Moys-

set (Paris: Marcel Rivière, 1923–1959), 15 vols. in 19, in-8°,

consisting of 15 volumes published between 1923 and 195926 .

A comparison of Marx’s writings with those of Ferdinand Lassalle, somewhat in contrast
to the case of Proudhon, lets us analyze Marx’s speech acts as a self-consciously political
rather than economic actor. While (as argued in Section 3.2.3 above) Marx explicitly aimed
to distinguish himself as a “better” economist than Proudhon, it was Lassalle’s political
principles and their manifestations in e.g. the Gotha Programme of 1875 (and the Erfurt
Programme of 1891, in Engels’ posthumous efforts his behalf) that Marx explicitly worked
to repudiate and replace with his own.
Due to his prominence in the eyes of Second International-era SPD intellectuals like
Eduard Bernstein, new collections of Lassalle’s writings were compiled and published quite
frequently from 1865 (starting with J. P. Becker’s collection published just one year after
Lassalle’s death) up until the Nazi regime’s rise to power. Although some additional collected-
works projects were carried out in the GDR from 1949 onwards, three collected-works projects
from the SPD era remain basically the canonical reference texts for scholarship on Lassalle
to this day.
The most commonly-referenced collection, the Gesammelte Reden und Schriften (GRS),
was edited by Eduard Bernstein and published in 12 volumes from 1919 to 1920, while
a second collection, the Nachgelassene Briefe und Schriften (NBS), was edited by Gustav
Meyer and published in 6 volumes between 1921 and 1925, augmenting the corpus of the
earlier 12-volume project with (for example) posthumously-discovered letters and earlier
25
Scans of each volume are available through the Bibliotheque National de France’s Gallica portal at
https://catalogue.bnf.fr/ark:/12148/cb31154797t.
26
Digitized versions of every volume, with the exception of the first part of Volume 1, are available via
HathiTrust.

172
drafts of major speeches found in his notebooks27 . A third collection, Eduard Bernstein’s
1898 Reden und Schriften (RS) in 3 volumes, is referenced less often but remains influential
nonetheless due to its status as the canonical reference for Lassalle’s writings from the year
of its publication up until the 1920s (when, as the later volume’s title suggests, Bernstein’s
12-volume GRS supplanted the 3-volume RS)28 .
More recently, researcher Bert Andréas’ valuable bibliography (Andréas 1981b) contains
entries for all known writings of Lassalle, along with known translations and information on
differences (e.g., inclusion, exclusion, and modification of the original text) between subse-
quent editions and printings. Our Lassalle dataset, which pairs digital plaintext versions of
all his known writings with metadata on each text (e.g., date of writing and/or publication,
data on all known versions, on all known translations, etc.), thus uses the coding system of
Andréas’ bibliography as its index. The resulting diachronic corpus of Lassalle’s writings is
analyzed in Section 3.4.1.2 below and discussed in detail in Appendix 3.B.2.
Mikhail Bakunin’s complete works in German have been published in 3 volumes in an edi-
tion titled Gesammelte Werke edited by Max Nettlau, available for full viewing at HathiTrust.
In French, there also exists a 6-volume Oeuvres available at the Internet Archive29 .
However, as can be seen in Figure 3.5, which plots Bakunin’s published writings over time,
it is only in the period between 1868 and 1872 when Bakunin wrote to any significant degree,
with three minor exceptions: the first is his article Die Reaktion in Deutschland published
in Arnold Ruge’s 1842 Deutsche Jahrbucher (discussed in Section 3.2.6 above), the second
an 1848 speech in support of the revolutionary efforts in Poland, and the third his collected
correspondence. For example, the 6-volume French collection contains no writings outside of
27
Links to each volume are given in Appendix 3.B.2.
28
It is important to note, however, that (for reasons which are not made entirely clear in Bernstein’s
introduction) there are some texts in the RS which were not carried over into the GRS. For 6 of the 100
texts listed in Andréas’ bibliography, therefore, we had to scrape the plaintext by OCRing scans of the
original RS, which are of far lower quality than the available scans of the GRS and NBS. As explained in
Appendix 3.B.2, however, these 6 texts can be identified and excluded from any analysis by filtering out
texts whose source metadata variable is equal to "RS".
29
Volume by volume links to this French collection are as follows: Volume 1, Volume 2, Volume 3, Volume
4, Volume 5, Volume 6.

173
this period, while the Gesammelte Werke delves only slightly outside of this range, covering
the years 1865 to 1875 (with a single letter to Marx being the only inclusion from 1865, and
a total of 20 pages of post-1872 writings). Therefore, although he ran (and published) in the
same circles as Marx during the latter’s Young Hegelian phase, and although there is some
overlap in terms of whom they corresponded with before the era of the First International,
the dearth of written material outside of 1868–1872 means we do not have sufficient data
for our method to be able to track Marx’s influence on Bakunin. There exists, however, a
large body of interpretive work in the history of political thought on the mutual influence
between the two figures: in addition to the works cited in Section 3.2.6 which focus mainly
on the era of the First International, Thomas 1980 traces the interaction between Marx and
the anarchist movement more broadly over the course of Marx’s lifetime, while Marshall S.
Schatz’s Introduction to Bakunin 1990—a volume in the Cambridge Texts in the History of
Political Thought series, edited by Raymond Geuss and Quentin Skinner—situates Bakunin’s
thought within the context of both 19th century radical political thought and the political
upheavals across Europe during Bakunin’s lifetime.

3.4.1.1 Marx vs. Proudhon

As we hinted at in Section 3.2.3 above, Pierre-Joseph Proudhon’s written output is typi-

cally characterized as being either brilliantly syncretic or a chaotic jumble of contradictions,
depending on the evaluator’s tastes and political-theoretic proclivities. In this section we
apply the tools we’ve used throughout this section to evaluate the veracity of these two
views, to compare the trajectory of his thought with that of Marx’s, and to then draw a set
of conjectural hypotheses regarding how these results can shed some light on why Marx’s
thought “won out” over Proudhon’s in terms of how strongly they influenced subsequent
European socialist discourse.
Works such as Woodcock 1956 and Hoffman 1972, which attempt to organize Proudhon’s
thought into a coherent set of principles, typically still discuss the challenges inherent in

174
 Figure 3.5: Bakunin’s Literary Output, 1839–1876

needing to “de-Hegelianize” much of his writing. As Hoffman 1972 describes, Proudhon’s

attempts to apply Hegel’s dialectical method to his subject matter often lapsed into exercises
in forcing the keywords of the subject to fit into a neat “thesis-antithesis-synthesis” equation.
Marx, for example, attacks Proudhon on precisely this point in his letter to Pavel An-
nenkov criticizing the former’s Systeme des contradictions économiques 30 : “For him, the
solution of present-day problems does not consist in public action but in the dialectical ro-
tations of his brain.” More bluntly, he accuses Proudhon of “confus[ing] ideas and things,”
“indulg[ing] in feeble Hegelianism in order to set himself up as an esprit fort,” and thus
tricking his audience via “pseudo-Hegelian sleight-of-hand”. “In a word, it is Hegelian trash,”
he concludes.
Hoffman 1972 provides a much more charitable interpretation, positing essentially that
although the “de-Hegelianization” of portions of Prouhdon’s thought can be tedious, the
benefits outweigh the costs. The tripartite Hegelian schema, Hoffman argues, provided a
30
The contents of this letter provided the basis for Marx’s 1847 response, Misère de Philosophie, a play
on the subtitle of Proudhon’s work, Philosophie de misère.

175
structure through which Proudhon was able to organize and communicate his ideas more
straightforwardly than he otherwise would have, and (most relevant for the purposes of
this work) made it easier for these ideas to travel across the continent. While the German
socialist movement was rooted in Hegelian thought and rhetoric, and literate British socialists
had been able to imbibe Hegelian ideas indirectly by way of Thomas Carlyle31 , the French
socialist movement had been almost recalcitrant in their rejection of Hegelianism, as the
Catholic socialists who predominated the movement were skeptical of its perceived atheism.

Figure 3.6: Proudhon’s Literary Output, 1837–1868

3.4.1.2 Marx vs. Lassalle

Unlike in the case of Bakunin (discussed above, in Section 3.4.1), we do in fact have a
large enough corpus of Lassalle’s writings to perform a diachronic comparison of his and
Marx’s trajectory through ideological space across the span of their lives. We also give the
31
On this connection see Dibble 1978 Ch. 3, “Carlyle and Hegel” (pp. 57–75). But cf. Wellek 1965 Ch. 2,
“Carlyle and German Romanticism” (pp. 34–81), on how Carlyle’s references to Hegel in Sartor Resartus,
and his appreciation of German philosophy and literature more generally, do not mean that he himself was
a Hegelian.

176
caveat, however, that (as discussed in Section 3.2.5) much of Lassalle’s time from the defeat
of the 1848 Revolutions until the founding of the ADAV in 1863 was spent fighting a series
of protracted legal battles: at first to secure his own release from prison, and then to ensure
that the familial inheritance of his lifelong confidante Sophie von Hatzfeldt would not be
usurped by other bitter rivals within her extended family.
Thus, as can be seen in Figure 3.7, we have a very limited amount of textual evidence
from which to infer his ideological positions during the 1850s. We posit that the lack of data
from this decade is not fatal, though, given that our interest in Lassalle’s writings is only
with respect to those of Marx, whose correspondence with Lassalle (as seen in Figure 3.8
began in earnest in 1856 and had mostly ended by 1860. With this in mind, we analyze
Lassalle’s trajectory not so much as a continuous path (like we did in the previous section)
but rather with an eye towards whether or not a discontinuity is observed between his pre-
and post-corresponding-with-Marx writings.

Figure 3.7: Lassalle’s Literary Output, 1840–1864

177
Figure 3.8: The volume of correspondence between Marx and Lassalle, per year. Letters
from Marx to Lassalle are tabulated based on MECW Vols. 38–41. Letters from Lassalle
to Marx are tabulated based on Ferdinand Lassalle. Nachgelassene Briefe und Schriften.
Herausgegeben von Gustav Mayer, Vol III: Der Briefwechsel zwischen Lassalle und Marx,
nebst Briefen von Friedrich Engels und Jenny Marx an Lassalle und von Karl Marx an
Gräfin Sophie Hatzfeldt (Lassalle 1922).

3.4.1.3 Semantic Leadership Networks

Given this set of authors with particular importance for understanding Marx’s position
in 19th-century European socialist discourse, we construct a series of Semantic Leadership
Networks as described in Section 1.4. We construct the first network N1848 with an eye
towards understanding the importance of the failed 1848 revolutions in terms of their impact
on subsequent socialist thought (described in more detail in Section 3.2.2). To operationalize
the hypothesis that Marx’s thought became significantly more influential in the post-1848
search for an “answer” to these failures, we use 1849 as the cutoff year separating T0 and T1 ,
giving us a balance for each author in our corpus as shown in Figure 3.9.
Then, for our second experiment, we operationalize the hypothesis that the vocabulary
of Marx’s Capital became the hegemonic vocabulary of the European socialist movement

178
 Cutoff = 1849 Count
Author T0 T1
Bakunin 5 5
Lassalle 30 84
Marx 10 13
Proudhon 5 5
Other Euro Socialists 19 19
Socialist Newspapers 926 890
Total 995 1016

Figure 3.9: The balance between T0 and T1 , in terms of the number of texts in each epoch,
given a cutoff date of 1849.

upon its publication (and especially after the mainstream press’s linking of Marx’s thought
with the Paris Commune), by setting the cutoff date to be 1868. The balance in terms of
the number of documents for each author in T0 and T1 in this case is given in Table 3.2.
We can see that, since most of the authors under consideration had already published
most or all of their works by 186732 , this experiment measures post-1867 influence primarily in
terms of the adoption of lexical-semantic innovations by the prominent socialist newspapers
in our corpus, like Vorwarts and Der Sozialdemokrat.
Although this approach differs from the more author-to-author-based influence captured
in our first experiment, it in fact mirrors a key trend in the history of 19th-century European
socialism, namely, its gradual evolution from an intellectual current into a mass politically-
organized movement33 . Thus, while our first experiment aimed to capture the influence of
pre-1848 writings on the post-1848 ideologies that subsequently became the foundations for
organized socialist workers’ parties, this experiment aims to capture the influence of (e.g.)
Marx’s writings on socialism once it was “put into practice” in the day-to-day organizational
and agitational work of these parties.
In other words, this second experiment speaks more to the controversies around “revision-
32
Lassalle died in a duel in 1864, and Proudhon died one year later in January of 1865. Bakunin lived and
continued publishing until 1876, however, and Marx’s output continued up until his death in 1883.
33
In Germany, for example, the first socialist workers’ party (the Lassallean ADAV) did not appear until
1863, but was obtaining vote totals exceeding 1 million by the early 1900s.

179
ism” that would become so central to conflicts within the global socialist movement in the
20th century: to what extent should lessons learned via the day-to-day exigencies of organiz-
ing masses of workers, and the prospect of parliamentary integration of socialist parties (in
Germany, for example, after the repeal of Bismarck’s Anti-Socialist Laws in 1890), be used to
modify the original core tenets of these parties?34 If the influence of these newly-encountered
desiderata indeed began to “overtake” the original influence of Marx, for example, we would
expect the latter’s PageRank score for this experiment to be lower than that of the previous
experiment, and the socialist newspapers’ PageRank score to be higher in turn.

Cutoff = 1867 Count

Author T0 T1
Bakunin 6 4
Other Euro Socialists 26 8
Lassalle 114 0
Marx 19 4
Socialist Newspapers 926 890
Proudhon 10 0
Total 1101 906

Table 3.2: The balance between T0 and T1 , in terms of the number of texts in each epoch,
given a cutoff date of 1867.

The results for the first experiment are visualized in Figure 3.10, based on the numeric
PageRank values given in Table 3.3. From the PageRank scores for each author, we can see
that indeed Marx emerges as the node with the greatest pre-1848 to post-1848 influence.
From this graph we can see that, indeed, Marx’s PageRank score is greater than Proud-
hon’s and Bakunin’s combined. This means that, if we consider not only instances where
Marx’s use of a term was later adopted by other actors, but also how influential those who
Marx influenced were (or, in other words, giving Marx “credit” for his strong influence on e.g.
Lassalle by counting some fraction of Lassalle’s influence on others towards Marx’s influence
score as well), the impact of the outward links from Marx to others is estimated to be greater
34
The details of these conflicts over “revisionism” are outside the scope of our study, but see Gronow 2015
Chapter 2, “The Dispute over Revisionism” for a succinct overview.

180
 Figure 3.10: PageRank scores for each author in the corpus

Name Leader Count P (leader) Follower Count P (follower) PageRank

Bakunin 872 0.7963 223 0.2037 0.0502
EuroSoc 8253 0.5216 7569 0.4784 0.2907
Lassalle 361 0.0708 4736 0.9292 0.0371
Marx 4275 0.7129 1722 0.2871 0.1508
News 11109 0.5375 9560 0.4625 0.3751
Proudhon 2073 0.3982 3133 0.6018 0.0960

Table 3.3: PageRank scores, and the associated probabilities and counts, for each author in
the corpus, with a cutoff date of 1849.

than the impact of all outward links from either Proudhon or Bakunin. The much higher
scores for Other European Socialists and Newspapers also increase our confidence in the
results, as it stands to reason that the influence of all socialists besides Bakunin, Lassalle,
Marx, and Proudhon, added together, would exceed the magnitude of influence of any one

181
of these individual figures.
The fact that the score for Newspapers is greater than the score for Other European
Socialists is interesting in its own right, as it is suggestive of a point we raised early on
(in Section 1.2.2), namely that the nineteenth century in general, and nineteenth century
socialist discourse in particular, was a less elitist affair than in previous centuries, when one
would expect that nearly all of the influence would be exerted via written manuscripts. As
argued in Darnton 1979, Goodman 1996, and Israel 2014, for example, the eighteenth-century
transmission of ideas “outwards” from small groups of French salonnières to the rest of
Paris—and subsequently to Scottish Enlightenment figures or American revolutionaries—was
conducted almost exclusively by way of these manuscripts, which were orders of magnitude
more costly and difficult to obtain than the newspapers of the burgeoning nineteenth-century
working-class press. As quoted in Chapter 1, the Oxford Short History of Europe cites this
emergence of a “mass public” as one of the key phenomena for understanding the century
as a whole:

“The more earth-bound written word was also spread further and faster than ever
before, as the mechanized printing presses and paper-making machines brought
the unit cost of newspapers within reach of working-class pockets. This was ac-
companied and encouraged by the virtual disappearance of illiteracy from west-
ern, northern, and central Europe.” (Blanning 1972, pp. 1–2)

3.4.2 Discussion

Through our experiments detailed in the previous two sections, we have examined Marx’s
influence on the broader European socialist movement qualitatively, via a visual tracing of
the trajectory of Marx’s thought over time relative to that of other prominent European so-
cialist thinkers, and quantitatively via a statistical analysis of the adoption of Marx’s lexical-
semantic innovations relative to other contenders for intellectual leadership of the European
socialist movement. We have seen that, in both cases, Marx emerges as the most promi-

182
nent driver of changes in the vocabulary of this movement, away from a moral-philosophical
discourse and towards a distinctly political-economic one.
One interesting aspect of this influence that we have not been able to examine, however, is
the role of “positivism”—in its particular nineteenth-century meaning, primarily associated
with Saint-Simon’s protégé August Comte—in this shift. Indeed, this introduces a complicat-
ing factor in understanding Marx’s later political-economic thought that we believe deserves
more attention: despite Marx’s (and, to an even greater degree, Engels’) proclivity to frame
his theories as “scientific” and “objective” laws, he in fact detested the thought of Comte
and the positivist movement more generally, and explicitly tried to distance himself from it.
With the hindsight available to us in the 21st century, prima facie, it seems he was
successful in this endeavor. While Comte’s thought was extremely influential in Europe as
well as parts of Latin America, with Émile Durkheim and Herbert Spencer among his most
prominent disciples, it is generally thought to constitute an independent strand of thought
from the strand that can be traced back to Marx. Thus, in Marx scholar James Farr’s view,
“there is more to science than positivism, and there is no positivism at all in Marx’s science”
(Ball and Farr 1984, 218).
Indeed, taking the field of modern sociology as an example, the “big three” of Marx,
Weber, and Durkheim are treated somewhat analogously to the “three component parts” of
Marxism described by Lenin: that is, as independently-operating influences brought together
in essentially equal proportion to form the basis of the modern discipline. However, while
the relationship between Marx’s and Weber’s thought has been richly investigated in the
intellectual history literature35 , the same cannot be said for the relationship between Marx
and Durkheim, especially as mediated by Comte as a “common ancestor”.
The findings of our experiments in the previous two sections thus shed light on the
importance of exploring this pathway of influence: how was Marx able to “thread the needle”
35
See, e.g., Bendix 1974 for a discussion of their analytical differences with regard to class structure, or
Wright 2002 for a somewhat contrasting evaluation, arguing that in fact the similarities are more substantial
than is usually acknowledged.

183
of simultaneously placing his thought within the framework of objective, positivistic science,
while avoiding it being simply lumped in with the thought of the positivist movement, which
was developing at the same time and in the same European intellectual milieu, broadly
defined? Though our study focuses on comparing and contrasting nineteenth-century moral-
philosophical and political-economic vocabularies, a key strength of our approach is the
ability for researchers to easily adapt the method to other vocabularies. A fruitful next step,
therefore, would be to investigate and understand this needle-threading performed by Marx
via a similar comparison of his later political-economic vocabulary with the characteristic
vocabulary of Comte and the positivist movement.

3.5 Conclusion

Modern scholarship on the history of political thought, and especially the questions of
intellectual influence and the locutionary impact of a thinker’s interventions with respect
to a broader discourse, have given rise to a rich body of work stressing the importance of
formerly non-canonical thinkers—“major” and “minor” interlocutors—for deepening our un-
derstanding of a given text. In this work we have shown how this type of context-inclusive
analysis, though it typically expands the necessary amount of reading beyond the limits of
individual human capabilities, can be brought back into the realm of possibility with the
aid of modern computational-linguistic tools. We have also highlighted, however, the crucial
domain expertise and interpretive skills which are still required of a researcher or team of re-
searchers in order to carry out such a computer-aided study successfully. This “outsourcing”
of certain aspects of a study to computational tools can, in fact, have pernicious conse-
quences, drastically amplifying the risk of drawing invalid inferences if hidden assumptions
and seemingly-inconsequential methodological choices are not interrogated with a critical
eye.
An additional consideration must therefore be kept in mind when determining whether or
not to employ the computational methods discussed here in an intellectual-historiographic

184
endeavor, namely, that of the diminishing returns from a wider and wider expansion of
the set of texts being analyzed. Although these methods can enhance the replicability and
verifiability of a given study, the key contribution which we have emphasized is the ability for
these methods to vastly expand the contextual scope of a study—i.e., how many additional
texts are taken into account when deriving our conclusions about a given text or discourse
of interest. It may in fact be the case, however, that a given author or text did intervene in
a discourse which was bounded or insulated enough that it remains amenable to standard
non-computationally-aided study.
As Leo Strauss’ Persecution and the Art of Writing brought to the fore of political-
philosophical thought, for instance, an author may purposefully restrict their interlocutors
to a select few individuals out of fear of censorship or the potential consequences of their
views being publicly revealed. Indeed, in the first volume of his thoroughgoing biography of
Marx, Michael Heinrich discusses the impact of Reimarus’ Apologie oder Schutzschrift für
die vernünftigen Verehrer Gottes on Marx’s early thought (and on 19th-century German
philosophy more broadly), despite the fact that it was only ever read by Reimarus’ closest
friends and never published in his lifetime. In Cambridge School analyses such as ours where
the aim is to ascertain what an author was attempting to do with a given text, instances
like these where its ideas become influential despite being written and distributed under
persecution present a more challenging case, requiring detailed study of a small set of extant
material rather than a wider computational study of a broad public discourse, like ours.
More commonly, e.g. for European texts of the Middle Ages up through the Renaissance,
it may actually be the case that a thinker wrote with a small select audience in mind,
given the rarity of both literacy and access to printing presses (not to mention the financial
resources required to fund a widely-distributed publication). Even into the era of the French
Enlightenment, many texts which we now consider to be epochal or canonical were originally
only addressed to and read by a small set of close confidantes. Indeed, Goodman 1996
describes in detail how pre-Revolutionary French salonnières discussed and debated their

185
ideas in tightly-knit, insular communities wherein “reading one’s manuscripts aloud in salons
could be an alternative to publication,” such that “there were manuscripts that were read
in or circulated through salons and never published, such as Gentil-Bernard’s ‘Art d’Aimer,’
which went the rounds for years, and Guibert’s ‘Eloge du Chancelier de l’Hospital’” (p. 147;
see also Chartier 1981). In cases like these with an ostensibly bounded contextual scope, a
“classical” non-computational approach (such as that of Skinner 1978a and Skinner 1978b,
Pocock 1975, or Dunn 1969) may be the better option, in terms of eliminating concerns
that the computational tools may miss important contextual details or collapse important
distinctions in word usage36 .

3.A Corpus Details: Marx’s Interlocutors

Last Name First Name Title Year Language

1 Arnd Karl Die naturgemasse Volkswirthschaft 1845 de
2 Aveling Eleanor La Question Féminine 1887 fr
3 Bakunin Mikhail Die Reaktion in Deutschland 1842 de
4 Bakunin Mikhail Brief an Ruge 1843 de
5 Bakunin Mikhail Der Kommunismus 1843 de
6 Bakunin Mikhail Russland wie es wirklich ist! 1847 de
7 Bakunin Mikhail Aufruf an die Slaven 1848 de
8 Bakunin Mikhail Der Sozialismus 1867 de
9 Bakunin Mikhail Essay in Egalité 1869 de
10 Bakunin Mikhail L’Empire knouto-germanique et la 1870 de
révolution
11 Bakunin Mikhail Gott und der Staat 1871 de
12 Bakunin Mikhail Protest of the Alliance 1871 de
13 Bastiat Frederic Die volkswirthschaftslehre... 1843 de
14 Bauer Bruno Geschichte der Politik, Cultur und 1843 de
Aufklärung...
15 Bauer Bruno Die Judenfrage 1843 de
16 Bebel August Größe und Grenzen der sozialistischen Be- 1878 de
wegung in Frankreich
17 Bebel August Frau 1879 de
18 Bernstein Eduard Voraussetzungen 1899 de
19 Blanc Louis Organization 1839 en
20 Blanc Louis Geschichte 1847 de
21 Blanc Louis Organisation 1839 fr
22 Blanc Louis History 1848 en

36
Cf., however, Becker et al. 2020, which illustrates how even the early sixteenth-century spread of Martin
Luther’s ideas involved a vast network of interlocutors: correspondents, former students, and theological
opponents spread across the entirety of present-day Germany and beyond.

186
 Last Name First Name Title Year Language
23 Blanc Louis Organisation 1839 de
24 Blanqui Louis-Auguste Eternite 1872 fr
25 Blind Karl Staat 1859 de
26 Bourgeois Leon Solidarite 1896 fr
27 Bray JF Leiden 1830 de
28 Carey HC Wissenschaft 1859 de
29 Commune Paris Journal 1871 fr
30 Comte Auguste Catechisme 1852 fr
31 Comte Auguste Positive 1880 de
32 Darwin Charles Origin of Species 1859 de
33 Darwin Charles Origine 1859 fr
34 Dézamy Théodore Sozialismus 1846 de
35 Dühring Eugen Capital und Arbeit 1865 de
36 Dühring Eugen Dialektik 1865 de
37 Dühring Eugen Krisis 1867 de
38 Gobineau Arthur Inegalite des Races 1853 fr
39 Grun Karl Bewegung 1845 de
40 Herwegh Georg Bogen 1843 de
41 Herzen Alexander Rusland 1854 de
42 Jaures Jean Republique 1893 fr
43 Jevons William Geld 1876 de
44 Mill John Principles of Political Economy 1848 de
45 Mill John Principles of Political Economy 1848 en
46 Kautsky Karl Forerunners of Modern Socialism 1885 en
47 Kautsky Karl Programme 1892 fr
48 Labriola Antonio Materialisme 1897 fr
49 Lafargue Paul Paresse 1883 fr
50 Lafargue Paul Romantisme 1896 fr
51 Lange Friedrich Arbeiterfrage 1865 de
52 Lange Friedrich Materialisme 1877 fr
53 Langlois Charles-Victor Etudes 1898 fr
54 Lassalle Ferdinand Rede 1848 de
55 Lassalle Ferdinand Italienische 1859 de
56 Lassalle Ferdinand System 1861 de
57 Lassalle Ferdinand Steuer 1863 de
58 Leroy-Beaulieu Anatole Liberalisme 1890 fr
59 Liebknecht Wilhelm Kein Kompromiß 1899 de
60 Lissagaray Prosper Kommune 1877 de
61 List Friedrich Nationale 1841 de
62 Luxemburg Rosa Liberte 1899 fr
63 Maine Henry Gouvernement 1887 fr
64 Marshall Alfred Principes 1890 fr
65 Maurer GL von Einleitung 1854 de
66 Menger Karl Grundsatze 1871 de
67 Michel Louise Commune 1898 fr
68 Pareto Vilfredo Italie 1891 fr
69 Plekhanov Georgei Kampf 1883 de
70 Proudhon Joseph Systeme des contradictions 1846 fr
71 Proudhon Joseph Nothwendigkeit 1847 de
72 Proudhon Joseph Systeme des contradictions 1847 en
73 Proudhon Joseph Bekenntnisse 1848 de
74 Proudhon Joseph Solution 1849 de

187
 Last Name First Name Title Year Language
75 Proudhon Joseph Malthusiens 1849 fr
76 Proudhon Joseph Probleme 1849 fr
77 Proudhon Joseph Recht 1849 de
78 Proudhon Joseph Interet 1850 fr
79 Proudhon Joseph Confessions 1851 fr
80 Proudhon Joseph Idee 1851 fr
81 Proudhon Joseph Soziale Revolution 1852 de
82 Proudhon Joseph Manuel 1853 fr
83 Proudhon Joseph Börse 1853 de
84 Proudhon Joseph Justice 1858 fr
85 Proudhon Joseph Kirche 1858 de
86 Proudhon Joseph Guerre 1861 fr
87 Proudhon Joseph Impot 1861 fr
88 Proudhon Joseph Mazzini 1862 de
89 Proudhon Joseph Federatif 1863 fr
90 Proudhon Joseph Literarische 1862 de
91 Proudhon Joseph Capacite 1865 fr
92 Reclus Elisee Sentiment 1866 fr
93 Rodbertus Karl Beleuchtung 1875 de
94 Schmidt Adolf Zukunft 1845 de
95 Schramm Stuart Internationale 1878 de
96 Shaw George Fabian 1884 en
97 Shaw George Fabian Essays in Socialism 1889 en
98 Sombart Werner Socialisme 1898 fr
99 Spencer Herbert Individu 1885 fr
100 Stirner Max Einzige 1845 de
101 Stirner Max Unique 1845 fr
102 Tocqueville Alexis Ancien 1856 fr
103 Tristan Flora Promenades 1840 fr
104 Villegardelle François Geschichte der socialen Ideen vor der 1846 de
französischen Revolution
105 Vogt Karl Prozess 1859 de
106 Vogt Karl Studien 1859 de
107 Weitling Wilhelm Menschen 1838 de
108 Weitling Wilhelm Evangelium 1845 de

3.B Datasets

3.B.1 Proudhon

3.B.1.1 Proudhon’s Collected Works

The most frequently-referenced collection of Proudhon’s works is known as the “Rivière

edition”:

Oeuvres complètes de P.-J. Proudhon, nouvelle édition, ed. C. Bouglé et H. Moys-

188
 set (Paris: Marcel Rivière, 1923–1959), 15 vols. in 19, in-8°.

This collection37 consists of 15 volumes published between 1923 and 1959, the contents of
which are as follows:

1. Système des contradictions économiques, ou philosophie de la misère, 2 vols., 1923.

HathiTrust (Part 2 only)

2. Idée générale de la révolution au XIXe siècle, 1924. HathiTrust

3. De la Capacité politique des classes ouvrières, 1924. HathiTrust

4. Candidature à la pension Suard. De la célébration du Dimanche. Qu’est-ce que la pro-

priété?, 1926. HathiTrust

5. De la création de l’ordre dans l’humanité, ou principes d’organisation politique, 1927.

HathiTrust

6. La Guerre et la paix, recherches sur le principe et la constitution du droit des gens,

1927. HathiTrust

7. Les Confessions d’un révolutionnaire pour servir à l’histoire de la révolution de février,

1929. HathiTrust

8. De la Justice dans la révolution et dans l’Eglise, 4 vols., 1930. HathiTrust: Part 1, Part
2, Part 3, Part 4

9. La Révolution sociale démontrée par le coup d’état du deux décembre. Projet d’exposition
perpétuelle, 1936. HathiTrust

10. Deuxième Mémoire sur la propriété. Avertissement aux propriétaires. Programme révolutionnaire.
Impôt sur le revenu. Le Droit au travail et le droit de propriété, 1938. HathiTrust

11. Du Principe de l’art et sa destination sociale. Galilée. Judith. La Pornocratie ou les

femmes dans les temps modernes, 1939. HathiTrust

12. Philosophie du progrès. La Justice poursuivie par l’Eglise, 1946. HathiTrust

13. Contradictions politiques. Les Démocrates assermentés et les réfractaires. Lettre aux
ouvriers en vue des élections de 1864. Si les traités de 1815 ont cessé d’exister?, 1952.
HathiTrust

14. Du Principe fédératif. La Fédération et l’unité en Italie. Nouvelles Observations sur

l’unité italienne. France et Rhin (fragments), 1959. HathiTrust

15. Ecrits sur la religion, 1959. HathiTrust

37
Digitized versions of every volume, with the exception of the first part of Volume 1, are available on
HathiTrust. Links to the individual volumes are given after the years of publication in the listing herein.

189
An earlier collection, which actually began publication while Proudhon was still actively
writing in 1850 (with the final volume published in 1872), has also been digitized by the
Bibliothèque Nationale de France. The scanned pages of all 26 volumes are available at
https://catalogue.bnf.fr/ark:/12148/cb31154797t, and the contents of the 26 volumes are as
follows:

• Vol. I. Nouvelle édition. - 1867

– Qu’est-ce que la propriété ?

– 1er mémoire : Recherches sur le principe du droit et du gouvernement
– 2e mémoire : Lettre à M. Blanqui sur la propriété.

• Vol. II. Nouvelle édition, 1868

– Avertissement aux propriétaires.

– La célébration du dimanche, plaidoyer devant la cour d’assises de Besançon.
– De la concurrence entre les chemins de fer et les voies navigables.
– Le “Miserere”.

• Vol. III. Nouvelle édition, 1868: De la création de l’ordre dans l’humanité, ou Principes
d’organisation politique.

• Vols. IV–V. 2e édition, 1850: Système des contradictions économiques, ou Philosophie

de la misère.

• Vol. VI. Solution du problème social, Nouvelle édition, 1868

– Solution du problème social

– Organisation du crédit et de la circulation
– Résumé de la question sociale ; Banque d’échange
– Banque du peuple. Suivie du rapport de la Commission des délégués du Luxem-
bourg

• Vol. VII. Nouvelle édition, 1868:

– La révolution sociale démontrée par le coup d’État du 2 décembre [précédée de la

lettre de Proudhon au Prince-Président].
– Le droit au travail et le droit de propriété.
– L’impôt sur le revenu.

• Vol. VIII. Nouvelle édition, 1868: Du principe fédératif et de la nécessité de reconstituer

le parti de la Révolution. Si les traités de 1815 ont cessé d’exister, actes du futur congrès.

• Vol. IX. Les confessions d’un révolutionnaire pour servir à l’histoire de la révolution
de février. Nouvelle édition revue, corrigée et augmentée par l’auteur, 1868.

190
 • Vol. X. Idée générale de la révolution au XIXe siècle (choix d’études sur la pratique
révolutionnaire et industrielle), Nouvelle édition, 1868

• Vol. XI. Manuel du spéculateur à la Bourse, 1869

• Vol. XII. Des réformes à opérer dans l’exploitation des chemins de fer... Nouvelle
édition, 1868.

• Vols. XIII-XIV. La guerre et la paix, recherches sur le principe et la constitution du

droit des gens. Nouvelle édition, 1869

• Vol. XV. Théorie de l’impôt, question mise au concours par le conseil d’état du canton
de Vaud en 1860. Nouvelle édition, 1868

• Vol. XVI. 1868:

– Les majorats littéraires, examen d’un projet de loi ayant pour but de créer au
profit des auteurs, inventeurs et artistes, un monopole perpétuel.
– La fédération et l’unité en Italie.
– Nouvelles observations sur l’unité italienne.
– Les démocrates assermentés et les réfractaires.

• Vols. XVII-XIX. Mélanges: Articles de journaux, 1848-1852, 1868-1871.

• Vol. XX. La philosophie du progrès. La justice poursuivie par l’Église. Nouvelle édition,
1868.

• Vols. XXI-XXVI. Essais d’une philosophie populaire: de la justice dans la révolution et

dans l’Eglise. Nouvelle édition, Notes et éclaircissements, 1868-1870e.

3.B.1.2 Proudhon’s Carnets

Proudhon kept meticulous records of every book, pamphlet, or newspaper article he read
over the course of most of his life, typically along with excerpts and brief summaries of
these works. Thus, with these records (digitized, though still in image format, thus requir-
ing readers to parse Proudhon’s often opaque handwriting), we can approximate whether it
was Marx or Proudhon who first adopted certain ideas and/or introduced certain them
into the lexicon of European socialist thought. Proudhon’s carnets are viewable in the
form of raw manuscript pages coded NAF 18255 to NAF 18262 at the Bibliothèque Na-
tionale de France, and digital scans of each page are available on the BnF’s website, at
https://archivesetmanuscrits.bnf.fr/ark:/12148/cc7339n.
As an example of the contents of these carnets, here we reproduce the content indices for
the 7 notebooks he kept in the year 1840 (Cahiers XI–XVII):

Cahier XI, Jan 1840

1–3: Réflexions prélim. sur le droit de propriété

191
3–4: Pellegrino Rossi, Cours d’Écon.

4–7: Jean-Jacques Rousseau, Origines de l’Inégalité

8–12: Jean-Baptiste Say, Écon. polit.

12–14: On Pellegrino Rossi and Jean-Jacques Rousseau

15–16: On Rossi Pinheiro-Ferreira

16–18: Victor Cousin, Philos. de l’hist.

18–24: Gustave Flourens, Notice sur F. Cuvier

24: Pellegrino Rossi, Révolution

24–34: Ancillon, Mélanges

34–35: Pellegrino Rossi, Propriété, force, mariage, femmes

35–37: Ancillon, Mélanges

37: Anne Robert Jacques Turgot, Tom. 2 des Oeuvres complètes. Phil. de l’hist.

38–39: Portets, Cours de droit naturel

39–45: Barchou de Penhoen, Hist. de la phil. all. (Bonaparte)

45–64: Victor Cousin, Hist. de la phil. (1826) (beaucoup de choses pour mon livre)

Cahier XII, Jan 1840

1–3: F. M. A. Voltaire, Lettre de Ch. Gouju

3–4: Marie Jean Antoine Nicolas Caritat, marquis de Condorcet, Progrès de l’espirit humain

4: On Property

7–12: Edgar Quinet, Idées des hébreux (idées sur la cosmogonie, de la mort, âme, etc.)

12–18: Jules Michelet, Introduction à l’hist. univ.

18–19: Moniteur de 1830, Révision de la charte

20–23: Conférences tenues au Conseil d’État, pour la discussion du Code civil

24–26: On Succession, marriage

26–31: Bergier, Dict. Théol. et notes de Bousset, Usure

31–35: J.-B. Bossuet, Usure

192
35: Immanuel Kant, Principes métaphysiques de morale

42: Réflexions sur moi-même, 7 février 1840

42: Michel Chevalier, Lettres sur l’Amérique du Nord

43: Le National du 4 février, on Misère, vagabondage, Lamennais, Victoria, Thiers

44–52: Pothier, Droit de propriété

53: Toullier, Tom V. De donation et testament

54: ibid., Tom. VI. De l’objet et de la matière des contrats

Cahier XIII, Feb 1840

1: Victor Cousin, Cours de 1828. Grand homme

3–7: Edgar Quinet, Idées de Herder (ses moeurs primitives)

8–12: Buckland, Géologie, couronné par l’Institut. (Prière de Kepler)

13–20: Toullier, Prescription, présomption, Dunod, ibid.

20–23: Charron, Ambition, peuple

23–25: Burnouf, Introduct. à la traduction de Tacite

25–27: Proudhon, Traité des droits d’usufruit

27–32: Lélut, Qu’est-ce que la phrénologie?

32–52: Garat, Mémoires sur la vie de Suard

53–56: Joseph F. X. Droz, Art d’être heureux (1811)

56–64: Beccaria, Traité des débats et des peines. Faire un syllogisme parfait

Cahier XIV, Nov. 1840

1–17: Beccaria (suite) Illégitimité de la peine

17–28: La Luzerne (Cardinal de), Dissertation sur le prêt de commerce (Pères cités)

29–44: Pastoret, Législation de Moïse (et autres)

45–60: C. G. Jouffroy, Cours de Droit naturel (âme, devoir, liberté, le National, Spinoza)

Cahier XV, Nov. 1840

193
1–39: Proudhon, Leçons de Blanqui, Wolowski, citations diverses et pensées journaux

35–36: Germain Garnier, De la propriété

40–43: Moniteur. 20 juin 1829, Galotti (extradition de)

43–50: ibid., 21 juin-7 octobre, Liberté individuelle (proposition Roger sur la)

50–53: François Pierre Guillaume Guizot, De la peine de mort

53–56: Ortolan, Le ministère public

Cahier XVI, Nov. 1840

1–5: C. G. Jouffroy, Cours de droit naturel

6–52: P. J. B. Buchez, Philosophie (3 vol.)

Cahier XVII, Dec. 1840

1–10: P. J. B. Buchez, Philosophie (suite)

11–14: Troplong, Prescription

15–18: P. J. B. Buchez, Philosophie (suite)

18–25: Troplong, Prescription (suite)

25–28: Lerminier, Histoire législ. comp.

28–35: Lerminier, Lettre phil. Phil. du droit.

35–56: Laboulaye, Hist. du droit. Duel, femmes, religion, etc... Délits, peines

3.B.2 Collections of Lassalle’s Works

3.B.2.1 Gesammelte Reden und Schriften (GRS)
Edited by Eduard Bernstein and published from 1919 to 1921. Links to the volumes on the
Internet Archive are as follows:
• Vol. I. Der italienische Krieg. Franz von Sickingen. Internet Archive

• Vol. II. Die Verfassungsreden. Das Arbeiterprogramm und die anschliessenden Verteildigungsre-
den. Internet Archive

• Vol. III. Die Agitation für den allgemeinen deutschen Arbeiterverein; Das Jahr 1863,
Polemik. Internet Archive

• Vol. IV. Die Agitation für den allgemeinen deutschen Arbeiter-Verein; Das Jahr 1864,
Aktenstücke. Internet Archive

194
 • Vol. V. Lassalles ökonomisches Hauptwerk. Internet Archive

• Vol. VI. Philosophisch-literarische Streifzüge. Internet Archive

• Vols. VII–VIII. Die Philosophie Herakleitos des Dunklen von Ephesos. 2 v. Internet
Archive: Vol. VII, Vol. VIII

• Vols. IX–XII. Das System der erworbenen Rechte. 4 v. Internet Archive: Vol. IX, Vol.
X, Vol. XI, Vol. XII

3.B.2.2 Nachgelassene Briefe und Schriften (NBS)

Edited by Gustav Mayer and published between 1921 and 1925. Links to the volumes on the
Internet Archive are as follows:

• Vol. I. Briefe von und an Lassalle bis 1848. 1921, X-357 S. Internet Archive

• Vol. II. Lassalles Briefwechsel von der Revolution 1848 bis zum Beginn seiner Arbeit-
eragitation. 1923, VIII-28-302 S. Internet Archive

• Vol. III. Der Briefwechsel zwischen Lassalle und Marx, nebst Briefen von Friedrich
Engels und Jenny Marx an Lassalle und von Karl Marx an Gräfin Sophie Hatzfeldt.
1922, XII-27-[1]-411 S. Internet Archive

• Vol. IV. Lassalles Briefwechsel mit Gräfin Sophie Hatzfeldt. 1924, XIII-[1]-33-[1]-408
S. Internet Archive

• Vol. V. Lassalles Briefwechsel aus den Jahren seiner Arbeiteragitation 1862-1864.

1925, X-45-[1]-368 S. Internet Archive

• Vol. VI. Ferdinand Lassalle. Die Schriften des Nachlasses und der Briefwechsel mit
Karl Rodbertus. 1925, IX-[1]-451 S. Internet Archive

3.B.2.3 Reden und Schriften (RS, 1893)

Edited by Eduard Bernstein and published between 1892 and 1893. The three volumes are
available via HathiTrust here, and links to the individual volumes on the Internet Archive
are as follows:

• Vol. I. 1892. Internet Archive

• Vol. II. 1893. Internet Archive

• Vol. III. 1893. Internet Archive

195
3.B.2.4 Andréas’ Lassalle Bibliography
Bert Andréas’ 1981 bibliography (Andréas 1981b) provides an exhaustive listing of Lassalle’s
literary output available in any language up to the time of publication, which enabled us
to compile digitized versions of the subset of all known German versions (original or in
translation) of Lassalle’s writings, used for the analysis in Section 3.4.1.2. As a sample of
the information in this bibliography, the “base” entry (i.e., excluding the additional entries
representing variants and translations of the original publication) for the first ten and last
ten records are reproduced below:

First 10

• LA1: REISEBESCHREIBUNG VON MEINEM LIEBLINGSWINKEL BIS ZUR STUBEN-

TOR

– N VI S. 6–10
– Schulaufsatz vom August 1840

• LA2: [FERDINAND LASSALLES TAGEBUCH]

– Herausgegeben und mit einer Einleitung versehen van Paul Lindau. Breslau, Schle-
sische Buchdruckerei, Kunst- und Verlags-Anstalt vormals S. Schott-laender, 1891,
in-16, 259 S.
– Das Tagebuch wurde geführt von Januar 1840 bis zum Sommer 1841. Lindau hat
den Text zuerst veröffentlicht in Nord und Süd, Bd. LVII, Heft 169–170. Zwei von
ihm ausgelassene längere Eintragungen (von etwa Pfingsten und Sommer 1841)
hat Gustav Mayer nach dem Manuskript veröffentlicht in N I, S. 54–63. Mayers
Handexemplar obiger Ausgabe mit den Verbesserungen des von Lindau teilweise
korrumpierten Textes ist jetzt im IISG.
– Vgl. LB43 und LC170.
– Archives: BA, FUB, IISG, IMLB, SUBD, SUBH

• LA3: ÜBER DIE ERKLÄRUNG DER HERREN KOLLOFF, SCHUSTER UND HAM-
BERG.

– Breslauer Zeitung, 25. September 1841, Nr. 224, S. 1606.

– N VI S. 31–33
– Parteinahme für Heinrich Heine gegen Salomon Strauß, datiert vom 24. September
1841. Zugeschrieben von Gustav Mayer.
– Vgl. LC552.
– Archives: UBW

• LA4: ZUR ERKLÄRUNG DES HERRN DR. DAVIDSON IN NO. 222 DER LEIPZIGER
ALLGEMEINEN ZEITUNG.

196
 – Bulletin des Leo Baeck Instituts, Tel Aviv, 1966, Jahrg. IX, Nr. 36, S. 335–338.
– über Rabbinatsstreitigkeiten in der Breslauer jüdischen Gemeinde. Der Artikel
war für eine unbekannte Zeitung bestimmt, die ihn nicht aufnahm. Gustav Mayer
hat das Manuskript koimentiert in LC557. Der Artikel wurde veröffentlicht zusam-
men mit LA9a und mit einem Kommentar von Alex Bein in LC106.
– Archives: BA BM JNUL LR OCH

• LA5: WIE KONNTEN DIE ALTEN BEI IHREM AUSGEBILDETEN RECHTS-

GEFÜHL DIE SKLAVEREI DULDEN?

– N VI S. 10–12
– Schulaufsatz, den Gustav Mayer auf “1842/43” datiert.

• LA6: KANN DIE REALBILDUNG DIE KLASSISCHE BILDUNG ERSETZEN?

– N VI S. 12–16
– Schulaufsatz, den Gustav Mayer auf “1842/43” datiert.

• LA7: STOIKER ODER EPIKUREER

– N VI S. 17–20
– Schulaufsatz, den Gustav Mayer auf “1842/43” datiert.

• LA8: ANSPRACHE AN LESSINGS GEBURTSTAG

– N VI S. 20–23
– Schulaufsatz, den Gustav Mayer auf “Januar 1842 (oder auch 1843)” datiert.

• LA9: DER VIELWISSER

– N VI S. 23–27
– Schulaufsatz, den Gustav Mayer auf “1842/43” datiert.

• LA9a: [SPOTTGEDICHT AUF DEN JÜDISCHEN LEHR- UND LESEVEREIN IN

BRESLAU]

– Bulletin des Leo Baeck Instituts, Tel Aviv 1966, Jahrg. IX, Nr. 36, S. 338–341.
Gustav Mayer zitiert einige Zeilen in LC557 und datiert das Gedicht auf “1843”.
Das Gedicht wurde zusammen mit LA4 vollständig und mit einem Kommentar
veröffentlicht von Alex Bein in LC106.
– Archives: BA BM JNUL LBN LR OCH

• LA10: GRUNDZÜGE ZU EINER CHARAKTERISTIK DER GEGENWART MIT

BESONDERER BERÜCKSICHTIGUNG DER HEGEL’SCHEN PHILOSOPHIE

– Zeitschrift für moderne Philosophie, Breslau 1843, Nr. 1.

197
 – N VI S. 55–74.
– Der im Sommer 1843 geschriebene Aufsatz ist gezeichnet “F. Lassal”. Er füllt
in der handschriftlich hergestellten Studentenzeitschrift die ganze erste Nummer
(44 1/4 Spalten), ohne damit beendet zu sein; weitere Nummern sind erschienen,
jedoch nicht erhalten geblieben. Lassalle veröffentlichte in ihnen noch mindestens
einen Artikel Zur Religionsphilosophie des Christenthums.
– Das von Gustav Mayer für seinen Abdruck des Textes benutzte Exemplar der
ersten Nummer der Zeitschrift befindet sich im Lassalle-Nachlaß. Die Wroctawer
Bibliotheken teilten mit, keine Nummern der Zeitschrift zu besitzen. Die Nachlässe
der Mitherausgeber der Zeitschrift, R. von Gottschall und M. von Wittenburg,
gelten als verschollen.

• LA10a: [ÜBER DEN HANDEL UND OBER DEN WEBERAUFSTAND]

– N I S. 99–105
– Brief an den Vater, vom 12. Juni 1844.
– Nach seiner Übersiedelung an die Berliner Universität schickte Lassalle mehrere
ausführliche Manuskripte zu einzelnen Themen in Briefform an seine Eltern und
Freunde. In diesen “Manuskriptbriefen” fand die Selbstverständigung des Studen-
ten ihren Niederschlag. Gustav Mayer veröffentlichte vier dieser Texte, die bewahrt
geblieben waren (LA10a, LA10b, LA10c und LA12a). Ebenfalls als Manuskript-
briefe anzusehen sind die Texte LA24a und LA25b sowie die an Sofija Sontzova
gerichtete sogenannte “Seelenbeichte” vom Oktober 1860 (Brief Nr. 7 in LB11).

• LA10b: [ÜBER JUDENTUM UND GESCHICHTE]

– N I S. 106–114
– Brief an die Mutter, vom 30. Juli 1844

• LA10c: [ÜBER ADEL, STAAT, INDUSTRIE UND KOMMUNISMUS]

– N I S. 114–136
– Brief an den Vater, vom 6. September 1844. Von Lassalle “Brief über die Indus-
trie” genannt (in LA12).

Last 10

• LA91 DER HOCHVERRATES-PROZESS WIDER FERDINAND LASSALLE VOR

DEM STAATS-GERICHTS-HOFE ZU BERLIN, AM 12. MÄRZ 1864. NACH DEM
STENOGRAPHISCHEN BERICHT.

– Berlin, Verlag von Reinhold Schlingmann, [Druck von R. Gensch in Berlin, Kro-
nenstraße 36], 1864, in-8, 78 S.
– RS II S. 754–830

198
 – GRS IV S. 61–174
– In der Anklageakte (S. 4–10) wegen der Veröffentlichung von LA77 stellt das
Gericht fest, die Auflage habe 16000 Exemplare betragen, von denen insgesamt
3026 beim Verleger, dem Verfasser und einem Expedienten beschlagnahmt wur-
den. Die Verteidigungsrede Lassalles auf S. 36–73. Eine Eingabe Lassalles an den
Anklagesenat, vom 29. November 1863, druckt G. Mayer in N VI S. 384–391.
Die Broschüre erschien in einer Auflage von 1200 Exemplaren (vgl. Lassalle an
Schlingmann, 25. Mai 1864). Ein langer Brief der Gräfin Hatzfeldt an Mathilde
Anneke vom 23. März 1864 (in WHi, Anneke Papers) schildert ausführlich Las-
salles Auftreten vor Gericht und seine Rede.
– Archives: ASD BA IMLB SUBD UBBa UBK

• LA92 PROCLAMATION

– RS II S. 905–907
– GRS IV S. 268–271
– Datiert und unterzeichnet “Berlin, den 17. März 1864. Der Präsident des Allg.
Deutsch. Arbeiter-Vereins. Ferdinand Lassalle.”
– Lassalle gibt bekannt, daß künftig in Abweichung von den Bestimnungen des
Textes LA68 die rechtsrheinischen Gemeinden auf Grund ihrer “starken Mit-
gliederzahl” selbst ihre Bevollmächtigten Vorschlägen. Gleichzeitig wird Carl Klings
zu Lassalles “Kommissar” ernannt, der die Durchführung dieser Anordnung zu or-
ganisieren hat.

• LA92a [VERFÜGUNG]

– In-8, [1] S. Handschrift vervielfältigt.

– Die aus Berlin vom 7. Mai 1864 datierte Verfügung ist an den Vereins-Kassierer
Gustav Lewy gerichtet und ordnet die monatliche Auszahlung des Sekretär-Gehalts
an Eduard Willms an. Es wurde kein Exemplar gefunden, aber das Nachlaß-
Repertorium sagt dazu: “Eigenh. Ausfertigung Lassalles und Faksimile-Umdruck
in zahlreichen Exemplaren, je 1 S. 8°” (V, 11 in LC104).
– Demnach ist möglicherweise die Versendung unterblieben.

• A93 [REDE IN DER LEIPZIGER GEMEINDE DES ADAV]

– N VI S. 274–282
– Fragment einer Nachschrift (mit Verbesserungen von Lassalles Hand) der Rede,
welche Lassalle am 9. Mai 1864 in Leipzig gehalten hat. G. Mayer ergänzt das
unvollständige Manuskript aus den Versammlungsberichten des Hamburger Nord-
stern (2. Mai 1864, Nr. 258, S. 2) und des Leipziger Adler. Zeitung für Deutschland
(11. Mai 1864). Große Teile dieser Rede sind identisch mit der Rede LA94.

• LA93a HERRN OTTO DANNER IN LEIPZIG.

199
 – In-8, [1] S. Handschrift vervielfältigt.
– Von Danmer geschrieben und datiert “Leipzig, 11. Mai 1864” und von Lassalle
unterzeichnet. Lassalle ernennt Danner “für die Dauer meiner Abwesenheit von
Berlin zum Vice-Präsidenten” des ADAV und überträgt ihm “alle mir selbst
zustehenden Funktionen und Befugnisse”. In einem kurzen Vorsatz ersucht Dan-
ner “Geehrte Redaction” um Aufnahme des Textes. Er übersandte das Zirku-
lar mit einem Begleitschreiben vom 12. Mai 1864 (Exemplar in HA) an die
Bevollmächtigten und Vorstandsmitglieder.
– Archives: HA, IISG

• LA94 DIE AGITATION DES ALLG. DEUTSCHEN ARBEITERVEREINS UND

DAS VERSPRECHEN DES KÖNIGS VON PREUSSEN. EINE REDE GEHALTEN
AM STIFTUNGSFEST DES ALLGEMEINEN DEUTSCHEN ARBEITER-VEREINS
ZU RONSDORF AM 22. MAI 1864 VON FERDINAND LASSALLE.

– Berlin, Verlag von Reinhold Schlingmann, [Druck von R. Gensch in Berlin, Kro-
nenstr.36,] 1864, in-8, 52 S.
– RS II S. 841–872
– GRS IV S. 187–229
– Die Broschüre erschien in einer Auflage von 2000 Exemplaren (vgl. Schlingmann
an Lassalle, 28. Mai 1864) gegen Ende Juni 1864 (vgl. Willms an Lassalle, 19. Juni
1864). Die Korrektur hat Bucher gelesen (vgl. Bucher an Lassalle, 6. Juni 1864).
Das unvollständige Konzept zu dieser Rede hat G. Mayer gedruckt in N VI S.
282–284. Lassalle hat dieselbe Rede vorher in verschiedenen anderen rheinischen
Städten gehalten: in Düsseldorf am 13. Mai, in Solingen und Barmen am 14. Mai,
in Köln am 15. Mai, in Duisburg am 16. Mai und in Wermelskirchen am 18. Mai.
Im Anhang der Broschüre sind mehrere Versanmlungsberichte nachgedruckt (S.
42–52), darunter zwei von Lassalle selbst verfaßte (vgl. LA95, LA96). Der Anhang
ist auch in allen Nachdrucken enthalten. Das Exemplar in IMLB befindet sich in
einem Konvolutband aus dem Besitz Clara Zetkins.
– Vgl. LA93.
– Archives: BA, IISG, IMLB, SUBH, UBK, ZBZ

• LA95 WERMELSKIRCHEN, DEN 19. MAI

– Nordstern, Hamburg, 28. Mai 1864, Jahrg. V, Nr. 259, S. 3.

– RS II S. 875–878
– GRS IV S. 232–237
– Anonymer Bericht über die Agitationsversanmlung in Wermelskirchen am 18. Mai
1864, vor der Lassalle die Rede LA94 gehalten hat. Der Bericht enthält ein “Lied
zur Abholung des Präsidenten F. Lassalle”, dessen “tiefe Innigkeit” und “Charak-
ter des echten alten Volksgesangs” hervorgehoben werden. Die Verfasserschaft
des Berichts ist belegt durch den eigenhändigen Entwurf Lassalles und einen von

200
 ihm bearbeiteten Korrekturabzug im Nachlaß, denen die Manuskripte der beiden
Gedichte beigelegt sind (V, 12 in LC104).
– Der Bericht ist, soweit nicht anders bemerkt, in allen Ausgaben von LA94 abge-
druckt.
– Archives: IISG

• LA96 RONSDORF, 23. MAI

– Der Adler, Leipzig, 25. Mai 1864

– RS II S. 878–881
– GRS IV S. 237–240
– Anonymer Bericht über die Agitationsversanmlung in Ronsdorf am 22. Mai 1864,
vor der Lassalle die Rede LA94 gehalten hat. Die Verfasserschaft wird Lassalle
zugeschrieben von Vahlteich (LC893, S. 8) und von Bernstein (RS II, S. 839).
Der Adler konnte nicht aufgefunden werden. Der Bericht ist, soweit nicht anders
bemerkt, in allen Ausgaben von LA94 abgedruckt.

• LA97 ERWIDERUNG

– Neue Preußische (Kreuz-) Zeitung, Berlin, 19. Juni 1864, Nr. 141, 1. Beilage S. 1.
– RS III S. 270–282
– GRS V S. 365–381
– Datiert und unterzeichnet “Bad Ems, den 2. Juni 1864. F. Lassalle.” Die Er-
widerung antwortet auf die lange Besprechung von LA87 in der Kreuz-Zeitung
(vgl. LC907). Die Aufnahme der Entgegnung in das konservative Organ erfol-
gte, nach anfänglicher Weigerung (Brief der Redaktion an Lassalle, 8. Juni 1864),
erst auf Intervention des ehemaligen Chefredakteurs der Zeitung, Hermann Wa-
gener. Der Bismarck sehr nahestehende Wagener war der anonyme Verfasser der
Besprechung.
– Auf Lassalles Instruktion kaufte der Vereinssekretär Willms 50 Exemplare der
Beilage (oder bestellte 50 Sonderabzüge des Textes), um sie an die Bevollmächtigten
zu versenden. Vgl. Lassalle an Willms, 15. Juni 1864, und Willms an Lassalle, 19.
Juni 1864.
– Vgl. hierzu auch Der Artikel des Herrn Lassalle in der Kreuzzeitung vom 19. Juni
mit einigen Randbemerkungen in Deutsche Arbeiterzeitung, Leipzig, 15. Juli 1864,
Nr. 16, S. 123–127.
– Archives: ASD, IZD

• LA98 PROZESS LASSALLE

– Düsseldorfer Zeitung, 29. Juni 1864, Nr. 176, S. 2–3; 30. Juni, Nr. 177, S. 2–3; 1.
Juli, Nr. 178, S. 2–3.
– RS II S. 677–706

201
 – GRS III S. 405–444
– Bericht über den in Düsseldorf am 27. Juni 1864 in zweiter Instanz verhandel-
ten Prozeß wegen der Broschüre LA73. Der Bericht erwähnt eingangs, daß am
21. Oktober 1863 von der Gesamtauflage von 10000 Exemplaren beim Verleger
1034 und noch “einige 20 Exemplare” bei Buchhändlern in Berlin und Düsseldorf
beschlagnahmt wurden. Der obige Titel erscheint nur in Nr. 177 und 178 der Düs-
seldorfer Zeitung, die Lassalles Verteidigungsrede enthalten. Den Prozeßbericht
schrieb der Lassalle befreundete Redakteur Paul Lindau auf Grund seiner stenographis-
chen Nachschrift der Rede, die er Lassalle vorlas, der ihm nach seinem Konzept
Korrekturen und Zusätze diktierte. Lindau hat das Konzept in LC502 (S. 13–22)
veröffentlicht. Das Original befindet sich jetzt in SLD. Das Urteil und eine Zusam-
menfassung der ausführlichen Urteilsbegründung in der Düsseldorfer Zeitung vom
3. Juli 1864, Nr. 180, S. 3.
– Archives: SUBD

• LA99 CIRCULAR AN SÄMTLICHE VORSTANDS-MITGLIEDER.

– Druck von J. Draeger’s Buchdruckerei (C. Feicht) in Berlin, o.J. in-8, 15-1 S.
– RS II S. 911–927
– GRS IV S. 276–298
– Datiert und unterzeichnet “Rigi-Kaltbad, 27. Juli 1864. Der Präsident des Allg.
deutschen Arbeitervereins. Ferdinand Lassalle.” Lassalle stellt ausführlich seine
verschiedenen Konflikte mit Vahlteich dar (vgl. LA86) und begründet seine unaus-
gesprochene Forderung, Vahlteich aus dem ADAV auszuschließen. Anlaß zu dem
ungewöhnlich umfangreichen Zirkular und dem scharfen Ton gegen Vahlteich
war dessen Antrag vom 28. Juni 1864 (RS II S. 909–910; GRS IV S. 274–275)
zur Generalversammlung des ADAV für 1864, der auf eine Einschränkung der
Präsidialbefugnisse zugunsten derer des Vorstandes und auf eine demokratische
Beschickung der Generalversammlung abzielte. Lassalle fügte dem Manuskript
eine Privatinstruktion an den Vereinssekretär Willms bei (RS II S. 928–930; GRS
IV S. 298–301), aus der hervorgeht, daß er ihm zugleich zwei weitere Zirkulare
an die Vorstandsmitglieder zur Vervielfältigung und Versendung schickte. Sie be-
trafen die Ernennung von B. Becker und von Schweitzer zu Vorstandsmitgliedern
und die Ernennung von Kassenrevisoren. Keines dieser beiden Zirkulare wurde
aufgefunden.
– Lassalles letztes Zirkular erschien am 3. August 1864 im Druck (Vgl. Willms an
Lassalle, 3. August 1864) in einer Auflage von wahrscheinlich 100 Exemplaren
(vgl. Lassalle an Willms, 27. Juli 1864). Vahlteich antwortete darauf mit einem
Rundschreiben an die Vorstandsmitglieder vom 11. August 1864 (RS II S. 930;
GRS IV S. 302).
– Vgl. LB10, S. 245–256.
– Archives: BIF, SBB*

202
 • LA100 DIES IST MEIN TESTAMENT.

– Großenhain 1889 in LC441 (S. 12–15)

– RS II S. 956–958
– GRS IV S. 337–339
– Datiert und unterzeichnet “Eigenhändig geschrieben und unterschrieben: Genf 27
August 1864 Ferdinand Lassalle”. Vermacht u.a. “Die gelehrten und schriftstel-
lerischen Aufsätze und Notizen” sowie “Das Eigenthum an meinen sämmtlichen
schriftstellerischen und gelehrten Werken [...] Herrn Lothar Bucher.” In der für
den ADAV entscheidenden Verfügung empfiehlt Lassalle die Wahl des Frankfurter
Bevollmächtigten Bernhard Becker zu seinem Nachfolger, mit dem Rat: “Er soll
an der Organisation festhalten! Sie wird den Arbeiterstand zum Sieg führen!”
Der Abdruck in LC441 erfolgte nach der Abschrift in Holthoffs Nachlaß, die der
Gräfin Hatzfeldt von den Genfer Behörden zur Verfügung gestellt worden war.
Der Abdruck wurde mit dem Original im Genfer Staatsarchiv (Jur. Civ., AAQ
Testaments, vol. 13, No. 116) verglichen. Er weicht von ihm nur in unbedeutenden
Einzelheiten ab, wie Auflösung von Abkürzungen, Umwandlung ausgeschriebener
Zahlen in arabische Ziffern, Einfügung fehlender Interpunktion u.ä. Kohut druckte
denselben Text noch einmal im selben Jahre in LC442 (S. 190–192).
– Damit ist Na’amans Hypothese hinfällig, ein Kopistenirrtum habe obigen Text an
die Stelle der von Na’aman vermuteten Formulierung “Er soll die Organisation
festhalten” gesetzt (vgl. LC650 und LC652).

The source abbreviations for this subset are as follows:

ASD Archiv für soziale Demokratie, Bonn, Germany

BA Sammlung Bert Andréas, im Institut Universitaire de Hautes Etudes, Genf [Geneva],

Switzerland

BIF Biblioteca dell’Istituto Giangiacomo Feltrinelli, Milan, Italy

BM British Library, im British Museum, London, UK

FUB Bibliothek der Freien Universität, Berlin, Germany

HA Herwegh Archiv, im Dichter Museum, Liestal, Switzerland

HBSA Hamburger Bibliothek für Sozialgeschichte und Arbeiterbewegung, Hamburg, Ger-

many

IISG Internationaal Instituut voor Sociale Geschiedenis, Amsterdam, The Netherlands

IMLB Institut für Marxismus-Leninismus, Berlin, Germany

IZD Institut für Zeitungsforschung, Dortmund, Germany

JNUL Jewish National and University Library, Jerusalem, Occupied Palestinian Territories

203
KMH Karl Marx Haus, Bibliothek, Trier, Germany

LBN Leo Baeck Institute, New York, NY, USA

LR Ludwig Rosenberger (Privatsammlung), Chicago, IL, USA

MS Manuskript

N LB65

OCH Hebrew Union College Library, Cincinnati, OH, USA

SBB* Stadtbibliothek, Sondersammlungen, Berlin, Germany

SUBD Stadt- und Landesbibliothek, Düsseldorf, Germany

SUBH Staats- und Universitätsbibliothek, Hamburg, Germany

UBBa Universitäts-Bibliothek, Basel, Switzerland

UBK Universitäts-Bibliothek, Köln [Cologne], Germany

UBW Uniwersytet Wrocławski Bibliotek, Wrocław [Breslau], Poland

ZBZ Zentralbibliothek, Zürich, Switzerland

3.B.3 Nietzsche’s Reading Records

The books Nietzsche read, and those he borrowed or owned but probably never read, have
been meticulously traced in Brobjer 2010, and listed in an Appendix from pages 185 to
258. We use this data source to observe that, although Nietzsche never read Marx or Engels
directly (according to Brobjer’s research), he was deeply influenced by Marx and Engels’
rival Eugen Dühring, discussed above. Indeed, 13 different readings of Dühring’s texts are
recorded in Nietzsche’s reading logs: 7 readings in 1875, one each in 1878, 1881, 1883, and
1884, and then two in 1885. As a sample of the full reading logs, his readings for just the
year 1875 (including the 7 Dühring texts) are reproduced here:

1. E. Windisch, Indian Philosophy

2. Heinrich Brockhaus, Rectoratsrede über indische Philologie (Nietzsche Attended Brock-

haus’s lecture “Overview of the Results of Indian Philology” this year)

3. Montaigne, Essays

4. John William Draper, Geschichte der geistigen Entwicklung Europas (1871)

5. E. Windisch, (on Indian philosophy)

6. P. Deussen, (on Indian philosophy)

204
 7. Eduard von Hartmann, (unknown text)

8. G. Grote, on Plato

9. Carl Fuchs, Unpublished notes and essays

10. Paul Rée, Psychologische Beobachtungen (1875) (Nietzsche writes to Rée on October
22, 1875, commenting on the book. This marks the beginning of their friendship. Ni-
etzsche’s copy of the book has a personal dedication from Rée.)

11. E. Windisch, Tripitaka der Buddhisten

12. Wolfgang Senfft, Indischer Sprüche, 3 vols.

13. Sutta Nipáta (in English translation)

14. Otto von Böhtlingk, Indische Sprüche: Sanskrit und Deutsch, 3 vols. (1870-1873)

15. Aristotle, (unknown work)

16. A. Schopenhauer, (unknown work)

17. A. Schopenhauer, Die Welt als Wille und Vorstellung

18. Xenophon von Athen, Memorabilien

19. A. Schopenhauer, Parerga, Vol. 1

20. G. C. Lichtenberg, Einige Lebensumstände von Capt. James Cook, grössentheils aus
schriftlichen Nachrichten einiger seiner Bekannten gezogen in Vermischte Schriften
(1867)

21. F. M. A. Voltaire, (unknown work)

22. Eugen Dühring, Der Werth des Lebens: Eine philosophische Betrachtung (Breslau,
1865) (Nietzsche made a 63-page summary with comments. N acquired this book May
26, 1875.

23. Eugen Dühring, Cursus der Philosophie als streng wissenschaftlichen Weltanschaaung
und Lebensgestaltung (Leipzig, 1875) (Nietzsche acquired this book on April 21, 1875,
and read it in the summer of 1875. He later reread it in 1885.)

24. K. Hillebrand, “Schopenhauer und das deutsche Publikum,” in Zeiten, Völker und
Menschen, Vol. 2 (Berlin, 1875)

25. A. Schopenhauer, Sämtliche Werke, Frauenstädt, 6 vols. (Leipzig, 1873-1874) (Niet-

zsche bought this book in 1875.)

26. Eugen Dühring, Kritische Geschichte der Philosophie von ihren Anfängen bis zur Gegen-
wart (Berlin, 1873) (Nietzsche bought this book in 1875.)

205
27. W. Oncken, Die Staatslehre des Aristoteles in historisch-politischen Umrissen, 2 vols.
(Leipzig, 1870) (Nietzsche bought this book in 1875.)
28. Eugen Dühring, Natürliche Dialektik (Berlin, 1865) (Nietzsche bought this book in
1875.)
29. Eugen Dühring, Kritische Geschichte der allgemeinen Prinzipien der Mechanik (Berlin,
1873) (Nietzsche bought this book in 1875.)
30. Plato, many different booklets (Nietzsche bought these in 1875.)
31. Confucius, Ta-Hio (Nietzsche bought this book in 1875.)
32. Confucius, Lao-tse tao (Nietzsche bought this book in 1875.)
33. K. Hillebrand, Zeiten, Völker und Menschen, 2 vols. (Berlin, 1875) (Nietzsche bought
this book in 1875.)
34. Eugen Dühring, Kursus der National- und Sozialökonomie, einschliesslich der Haupt-
punkte der Finanz-politik (Berlin, 1873) (Nietzsche bought this book in 1875.)
35. Eugen Dühring, Kritische Geschichte der Nationalökonomie und des Sozialismus (Berlin,
1875) (Nietzsche bought this book in 1875.)
36. John William Draper, Geschichte der Conflicte zwischen Religion und Wissenschaft,
Int. wiss. Bibl. XIII (1875) (Nietzsche bought this book in 1875.)
37. Herbert Spencer, Einleitung in das Studium der Sociologie, Vol. 1, Int. wiss. Bibl. XIV
(1875) (Nietzsche bought this book in 1875.)
38. Herbert Spencer, Einleitung in das Studium der Sociologie, Vol. 2, Int. wiss. Bibl. XV
(1875) (Nietzsche bought this book in 1875.)
39. Paul Rée, Psychologische Beobachtungen (Berlin, 1875) (Nietzsche bought this book in
1875.)
40. Aristotle, Rhetorik (University teaching, summer term of 1875)
41. Diogenes Laertius, Demokritos (University teaching, winter term of 1875-1876)
42. Plato, Phaedon (Paedagogium teaching, winter term of 1875-1876.)
43. Plato, Protagoras (Paedagogium teaching, winter term of 1875-1876.)
44. Plato, Symposion (Paedagogium teaching, winter term of 1875-1876.)
45. Plato, Phaedrus (Paedagogium teaching, winter term of 1875-1876.)
46. Plato, Politeia (Paedagogium teaching, winter term of 1875-1876.)
47. George Henry Lewes, Geschichte der Philosophie von Thales bis Comte, Bd. 1 (Berlin,
1871) (Borrowed from the Basel University library, 1875.)

206
 48. Aristotle, Ars rhetorica, hrsg. von L. Spengel, 2 Bde. (Leipzig, 1867) (Borrowed from
the Basel University library, 1875.)

49. Democritus, Operum fragmenta, hrsg. von F. W. A. Mullach (Berlin, 1843) (Borrowed
from the Basel University library, 1875.)

50. Staphanus Byantinus, Ethicorum quae supersunt, hrsg. von A. Meineck, Bd. 1 (Berlin,
1849) (Borrowed from the Basel University library, 1875.)

The full dataset is available at https://marxdb.com/nietzsche

3.C Table Footnotes

3.C.1 Table 3.1

(a) Adam Smith, Wealth of Nations

• Proudhon: Haubtmann 1982, p. 1090

• Marx:
i. Paris Notebooks: MEGA2 IV/2, 332–388
ii. Bookshelf: MEGA2 IV/32, 606–607 (entries 1248 and 1249)
iii. Other Sources: Evans 1984, p. 130, entry 5

(b) David Ricardo, Principles of Political Economy

• Proudhon: Haubtmann 1982, p. 1090

• Marx:
i. Paris Notebooks: MEGA2 IV/2, 392–427
ii. Bookshelf: MEGA2 IV/32, 550 (entries 1115 and 1116)

(c) Charles Comte, Traité de la propriété

• Proudhon: Haubtmann 1982, p. 1086

• Marx: Rubel 1975 claims that Marx read this text (p. 9), but no other records
exist to substantiate this claim. No works by C. Comte were found in Marx’s
library (MEGA2 IV/32).

(d) F. X. J. Droz, Propriété

• Proudhon: Haubtmann 1982, p. 1081

• Marx: Evans 1984, p. 130, entry 27. No works by Droz were found in Marx’s
library, however (MEGA2 IV/32).

(e) A. Destutt de Tracy, Économie politique

• Proudhon: Haubtmann 1982, p. 1081

207
 • Marx: No record exists of Marx reading this text, though he did read Destutt de
Tracy’s Elémens d’idéologie in Paris in 1844 (IISH, code B21, S. 12–14), and it
was found in his bookshelf (MEGA2 IV/32, p. 214, entry 286)

(f) Adolphe Blanqui, Histoire de l’economy politique.

• Proudhon: Haubtmann 1982, p. 1083

• Marx: Brussels Notebooks: IISH code B32; MEGA2 IV/3

(g) Dugald Stewart, Esquisses de philosophie morale (1793)

• Proudhon: Haubtmann 1982, p. 1084

• Marx: IISH, code B91a, S. 147–149.

(h) J. B. Say, Course complet de Économie politique (1828)

• Proudhon: Haubtmann 1982, p. 1086

• Marx: IISH, code B19, S. 4

(i) A. A. Cournot, Principes mathématiques d’Économie politique (1838)

• Proudhon: Haubtmann 1982, p. 1086

(j) P. Rossi, Cours d’Économie politique (1836)

• Proudhon: Haubtmann 1982, p. 1087

• Marx: Brussels notebooks, IISH code B33, S. 16–29.

(k) G. Garnier, De la propriété (1792)

• Proudhon: Haubtmann 1982, p. 1088

• Marx: Though no records exist indicating that he read this text, he did read
Garnier’s two-volume Histoire de la Monnaie depuis les temps de la plus haute
antiquité jusqu’au règne de Charlemagne in 1850 (IISH, code B45), and his Abrégé
élémentaire des principes de l’économie politique in 1863 (IISH, code B103)

(l) A. Ciezkowski, Du crédit et de la circulation (1839)

• Proudhon: Haubtmann 1982, p. 1090

• Marx: Though Ciezkowski’s Du crédit et de la circulation is never mentioned,
Rubel 1975 claims that Marx “presumably read” Ciezkowski’s more popular His-
toriosophie, though no dates or time periods are cited.

208
 References

[1] R. Abramitzky and I. Sin, “Book Translations as Idea Flows: The Effects of the
Collapse of Communism on the Diffusion of Knowledge,” Journal of the European
Economic Association, vol. 12, no. 6, 1453–1520, 2014.

[2] D. Acemoglu and J. A. Robinson, “A Theory of Political Transitions,” American

Economic Review, vol. 91, no. 4, 938–963, 2001.

[3] G. Achcar, Morbid Symptoms: Relapse in the Arab Uprising. Stanford University
Press, 2016.

[4] J. S. Adams, “Incremental Activism in Soviet Third World Policy: The Role of the
International Department of the CPSU Central Committee,” Slavic Review, vol. 48,
no. 4, 614–630, 1989.

[5] W. L. Adamson, “Marx and Political Education,” The Review of Politics, vol. 39,
no. 3, 363–385, 1977.

[6] T. Adorno, E. Frenkel-Brenswik, D. J. Levinson, and R. N. Sanford, The Authoritarian

Personality. Verso Books, 1950.

[7] F. Afary, “Persian ‘Capital’: Hassan Mortazavi’s New Translation of Marx’s Classic
Work,” The International Marxist-Humanist, 2012.

[8] J. Agar, Rethinking Marxism: From Kant and Hegel to Marx and Engels. London:
Routledge, 2006.

[9] M. Agulhon, The Republican Experiment, 1848-1852. Cambridge: Cambridge Univer-

sity Press, 1983.

[10] J. P. Aires, R. Granada, and F. Meneguzzi, “ConCon: A Contract Conflict Identifier,”

Proceedings of the 18th International Conference on Autonomous Agents and Multi-
Agent Systems, Richland, SC: International Foundation for Autonomous Agents and
Multiagent Systems, 2019, 2327–2329.

[11] R. J. Alexander, International Trotskyism, 1929-1985: A Documented Analysis of the

Movement. Duke University Press, 1991.

[12] M. Algee-Hewitt, R. Heuser, and F. Moretti, On Paragraphs: Scale, Themes and

Narrative Form. 2015.

209
[13] P. G. Altbach and K. Smith, Library Trends 1978: Publishing in the Third World.
1978.

[14] L. Althusser, Reading Capital. New York: Pantheon Books, 1965.

[15] L. Althusser, For Marx. Verso, 1968.

[16] B. Amini, “A Brief History of the Dissemination and Reception of Karl Marx’s Capital
in the United States and Britain,” World Review of Political Economy, vol. 7, no. 3,
334–349, 2016.

[17] K. B. Anderson, Marx at the Margins: On Nationalism, Ethnicity, and Non-Western

Societies. University of Chicago Press, 2010.

[18] B. Andréas, Le Manifeste Communiste de Marx et Engels: histoire et bibliographie

1848-1918. Milan: Feltrinelli, 1963.

[19] B. Andréas, Deutsche-Brüsseler-Zeitung: 1. Januar 1847-27. Februar 1848 : Faksimile

mit Einführung und Anmerkungen. Editions Culture et civilisation, 1981.

[20] B. Andréas, Ferdinand Lassalle - Allgemeiner Deutscher Arbeiterverein: Bibliographie

ihrer Schriften und der Literatur über sie, 1840 bis 1975. Verlag Neue Gesellschaft,
1981.

[21] B. Andréas, Karl Marx, Friedrich Engels : Das Ende Der Klassischen Deutschen
Philosophie, Bibliographie. Trier: Karl-Marx-Haus, 1983.

[22] C. Andrew and V. Mitrokhin, The World Was Going Our Way. 2017.

[23] G. G. Andrews, “French Nationalism in 1789 according to the General Cahiers. Beat-
rice Fry Hyslop,” The Journal of Modern History, vol. 7, no. 2, 212–213, 1935.

[24] J.-C. Angaut, “Revolution, socialism, and the Slavic question: 1848 and Michael
Bakunin,” The 1848 Revolutions and European Political Thought, D. Moggach and
G. Steadman Jones, Eds., Cambridge University Press, 2018, 405–428.

[25] J. Angrist and J.-S. Pischke, Mostly Harmless Econometrics: An Empiricist’s Com-
panion. Princeton: Princeton University Press, 2009.

[26] H. Arendt, The Origins of Totalitarianism. Schocken Books, 1951.

[27] I. Arieli, Y. Babichenko, R. Peretz, and H. P. Young, “The Speed of Innovation

Diffusion in Social Networks,” Econometrica, vol. 88, no. 2, 569–594, 2020.

210
[28] C. J. Arthur, “Engels as Interpreter of Marx’s Economics,” Engels Today: A Cente-
nary Appreciation, London: Palgrave Macmillan UK, 1996, 173–209.

[29] I. Ascher, “All the Wor(l)d’s a Stage: Persons and Personification in Karl Marx’s
Capital,” ECPR Workshop on Metaphor in Political Science, 2005.

[30] E. Ash and D. L. Chen, “Case vectors: Spatial representations of the law using docu-
ment embeddings,” Law as Data, Santa Fe Institute Press, ed. M. Livermore and D.
Rockmore, 2019.

[31] E. Ash, D. L. Chen, and S. Naidu, “Ideas Have Consequences: The Effect of Law and
Economics on American Justice,” Social Science Research Network, Rochester, NY,
SSRN Scholarly Paper ID 2992782, 2017.

[32] E. Ash, J. Jacobs, B. MacLeod, S. Naidu, and D. Stammbach, “Unsupervised Extrac-

tion of Workplace Rights and Duties from Collective Bargaining Agreements,” 2020
International Conference on Data Mining Workshops (ICDMW), 2020, 766–774.

[33] R. Ashcraft, Locke’s Two Treatises of Government. Routledge, 1986.

[34] J. L. Austin, How to Do Things with Words. Clarendon Press, 1962.

[35] S. Avineri, “The Hegelian Origins of Marx’s Political Thought,” The Review of Meta-
physics, vol. 21, no. 1, 33–56, 1967.

[36] S. Avineri, The Social and Political Thought of Karl Marx. Cambridge: Cambridge
University Press, 1968.

[37] S. Avineri, Hegel’s Theory of the Modern State. Cambridge University Press, 1974.

[38] Q. V. S. Bach, Soviet Economic Assistance to the Less Developed Countries: A Sta-
tistical Analysis. Clarendon Press, 1987.

[39] Q. V. S. Bach, Soviet Aid to the Third World: The Facts and Figures. Book Guild,
2003.

[40] K. M. Baker, Inventing the French Revolution: Essays on French Political Culture in
the Eighteenth Century. Cambridge University Press, 1990.

[41] M. Baker, “Corpus Linguistics and Translation Studies: Implications and Applica-
tions,” Text and Technology: In Honour of John Sinclair, Amsterdam: John Ben-
jamins, 1993, 233–250.

[42] M. A. Bakunin, Statism and Anarchy. Cambridge University Press, 1990.

211
[43] T. Ball and J. Farr, After Marx. CUP Archive, 1984.

[44] A. Banerjee, I. S. Dhillon, J. Ghosh, and S. Sra, “Clustering on the Unit Hypersphere
Using Von Mises-Fisher Distributions,” J. Mach. Learn. Res., vol. 6, 1345–1382, 2005.

[45] M. Baroni and S. Bernardini, “A New Approach to the Study of Translationese:

Machine-learning the Difference between Original and Translated Text,” Literary and
Linguistic Computing, vol. 21, no. 3, 259–274, 2006.

[46] A. T. J. Barron, J. Huang, R. L. Spang, and S. DeDeo, “Individuals, Institutions,

and Innovation in the Debates of the French Revolution,” Proceedings of the National
Academy of Sciences, vol. 115, no. 18, 4607–4612, 2018.

[47] R. Barthes, Image, Music, Text. New York: Noonday Press, 1977.

[48] A. Bartoli, A. De Lorenzo, E. Medvet, and F. Tarlao, “Inference of Regular Expres-

sions for Text Extraction from Examples,” IEEE Transactions on Knowledge and
Data Engineering, vol. 28, no. 5, 1217–1230, 2016.

[49] E. P. S. Baumer, D. Mimno, S. Guha, E. Quan, and G. K. Gay, “Comparing Grounded

Theory and Topic Modeling: Extreme Divergence or Unlikely Convergence?” Journal
of the Association for Information Science and Technology, vol. 68, no. 6, 1397–1410,
2017.

[50] T. Bayes, “An essay towards solving a problem in the doctrine of chances. By the late
Rev. Mr. Bayes, communicated by Mr. Price, in a letter to John Canton,” Philosoph-
ical transactions of the Royal Society of London, no. 53, 370–418, 1763.

[51] R. Beamish, Marx, Method, and the Division of Labor. Urbana-Champaign: University
of Illinois Press, 1992.

[52] R. Beamish, “The Making of the Manifesto,” Socialist Register, vol. 34, 1998.

[53] S. O. Becker, Y. Hsiao, S. Pfaff, and J. Rubin, “Multiplex Network Ties and the
Spatial Diffusion of Radical Innovations: Martin Luther’s Leadership in the Early
Reformation,” American Sociological Review, vol. 85, no. 5, 857–894, 2020.

[54] F. Beckett, 1956: The Year That Changed Britain. Biteback Publishing, 2015.

[55] T. O. Beebee, “Shoot The Transtraitor!” The Translator, vol. 16, no. 2, 295–313,
2010.

[56] M. Beer, A History of British Socialism, Volume I. London: Routledge, 1920.

212
[57] M. Beer, A History of British Socialism, Volume II. London: Harcourt, Brace and
Howe, 1921.

[58] F. C. Beiser, After Hegel: German Philosophy, 1840–1900. Princeton University Press,
2014.

[59] D. Bellos, Is That a Fish in Your Ear?: Translation and the Meaning of Everything.
Farrar, Straus and Giroux, 2011.

[60] L. Belmonte, Selling the American Way: U.S. Propaganda and the Cold War. 2010.

[61] I. Beltagy, M. E. Peters, and A. Cohan, “Longformer: The Long-Document Trans-

former,” arXiv:2004.05150, 2020.

[62] R. Bendix, “Inequality and Social Structure: A Comparison of Marx and Weber,”
American Sociological Review, vol. 39, no. 2, 149–161, 1974.

[63] A. Béraud and G. Faccarello, “‘Nous marchons sur un autre terrain.’ The reception
of Ricardo in the French language: Episodes from a complex history,” The Reception
of David Ricardo in Continental Europe and Japan, Routledge, 2014.

[64] D. Berger, W. Easterly, N. Nunn, and S. Satyanath, “Commercial Imperialism? Polit-

ical Influence and Trade During the Cold War,” American Economic Review, vol. 103,
no. 2, 863–896, 2013.

[65] I. Berlin, “The Arts in Russia Under Stalin,” New York Review of Books, no. October
19, 2000, 1946.

[66] S. I. Berlin, Karl Marx: His Life and Environment. London: Thornton Butterworth,
1939.

[67] S. Bernstein, “Marx in Paris, 1848: A Neglected Chapter,” Science & Society, vol. 3,
no. 3, 323–355, 1939.

[68] A. E. Bestor, “The Evolution of the Socialist Vocabulary,” Journal of the History of
Ideas, vol. 9, no. 3, 259–302, 1948.

[69] M. Bevir, “Are there Perennial Problems in Political Theory?” Political Studies,
vol. 42, no. 4, 662–675, 1994.

[70] M. Bevir, The Logic of the History of Ideas. New York: Cambridge University Press,
1999.

[71] C. Biagioli, E. Francesconi, A. Passerini, S. Montemagni, and C. Soria, “Automatic

Semantics Extraction in Law Documents,” Proceedings of the 10th International Con-

213
 ference on Artificial Intelligence and Law, New York, NY, USA: ACM, 2005, 133–
140.

[72] B. Bianchi, “Marx’s Reading of Spinoza: On the Alleged Influence of Spinoza on

Marx,” Historical Materialism, vol. 26, no. 4, 35–58, 2018.

[73] S. J. Birkby, “"We will bury you": How mistranslation heightened cold war tensions,”
Galaxy: The IELE Journal, vol. 6, no. 2, 89–102, 2014.

[74] A. Bisin and T. Verdier, “The Economics of Cultural Transmission and Socialization,”
National Bureau of Economic Research, Working Paper 16512, 2010.

[75] T. C. W. Blanning, The Nineteenth Century: Europe 1789-1914. Oxford University

Press, 1972.

[76] T. C. W. Blanning, The French Revolution: Class War or Culture Clash? Macmillan
International Higher Education, 1997.

[77] A. Blau, Ed., Methods in Analytical Political Theory. New York: Cambridge University
Press, 2017.

[78] L. Blaydes, J. Grimmer, and A. McQueen, “Mirrors for Princes and Sultans: Advice
on the Art of Governance in the Medieval Christian and Islamic Worlds,” The Journal
of Politics, vol. 80, no. 4, 1150–1167, 2018.

[79] D. M. Blei, “Introduction to Probabilistic Topic Models,” Communications of the

ACM, vol. 55, no. 4, 77–84, 2012.

[80] D. M. Blei, “Topic Modeling and Digital Humanities,” Journal of Digital Humanities,
2013.

[81] D. M. Blei, A. Y. Ng, and M. I. Jordan, “Latent Dirichlet Allocation,” J. Mach.

Learn. Res., vol. 3, 993–1022, 2003.

[82] G. Boas, “Fact and Legend in the Biography of Plato,” The Philosophical Review,
vol. 57, no. 5, 439–457, 1948.

[83] M. Bohlender, Marx Meets Manchester. The Manchester Notebooks as a Starting

Point of an Unfinish(Ed)Able Project? Brill, 2018, ch. Marx’s Capital: An Unfinish-
able Project? 228–249.

[84] P. Bojanowski, E. Grave, A. Joulin, and T. Mikolov, “Enriching Word Vectors with
Subword Information,” Transactions of the Association for Computational Linguis-
tics, vol. 5, 135–146, 2017.

214
[85] M. J. Bommarito II, D. M. Katz, and E. M. Detterman, “LexNLP: Natural language
processing and information extraction for legal and regulatory texts,” arXiv:1806.03688
[cs, stat], 2018.

[86] W. Bonefeld, “Social form, critique and human dignity,” Zeitschrift für kritische The-
orie, vol. 13, 97–112, 2001.

[87] T. B. Bottomore, Karl Marx. Prentice-Hall, 1973.

[88] S. Bowles, Microeconomics: Behavior, Institutions, and Evolution. Princeton Univer-

sity Press, 2009.

[89] J. Boyd-Graber, Y. Hu, and D. Mimno, “Applications of Topic Models,” Foundations

and Trends in Information Retrieval, vol. 11, no. 2-3, 143–296, 2017.

[90] W. J. Brazill, The Young Hegelians. New Haven: Yale University Press, 1970.

[91] W. Breckman, Marx, the Young Hegelians, and the Origins of Radical Social Theory:
Dethroning the Self. Cambridge: Cambridge University Press, 2001.

[92] T. H. Brobjer, Nietzsche’s Philosophical Context: An Intellectual Biography. Univer-

sity of Illinois Press, 2010.

[93] S. Brooke and N. Ketchley, “Social and Institutional Origins of Political Islam,”
American Political Science Review, vol. 112, no. 2, 376–394, 2018.

[94] T. Brotherstone, “1956 and the Crisis in the Communist Party of Great Britain: Four
Witnesses,” Critique, vol. 35, no. 2, 189–209, 2007.

[95] P. F. Brown, V. J. D. Pietra, S. A. D. Pietra, and R. L. Mercer, “The mathematics

of statistical machine translation: Parameter estimation,” Computational Linguistics,
vol. 19, no. 2, 263–311, 1993.

[96] G. Browning, “The German Ideology: The Theory of History and the History of
Theory,” History of Political Thought, vol. 14, no. 3, 455–473, 1993.

[97] G. Browning, “Agency and Influence in the History of Political Thought: The Agency
of Influence and the Influence of Agency,” History of Political Thought, 22, 2000.

[98] G. Browning, A History of Modern Political Thought: The Question of Interpretation.

New York, NY: Oxford University Press, 2016.

[99] D. Brudney, Marx’s Attempt to Leave Philosophy. Harvard University Press, 2009.

215
[100] A. Buick, “Socialist Standard Past & Present: A new translation of Marx’s ‘Capital’
(1980),” Socialist Standard, no. 908, 1–3, 1980.

[101] E. Buringh and J. L. V. Zanden, “Charting the “Rise of the West”: Manuscripts
and Printed Books in Europe, A Long-Term Perspective from the Sixth through
Eighteenth Centuries,” The Journal of Economic History, vol. 69, no. 2, 409–445,
2009.

[102] E. Burke, Reflections on the Revolution in France. New York: Macmillan, 1790.

[103] T. Burns and I. Fraser, The Hegel-Marx Connection. New York: Springer, 2000.

[104] H. Butterfield, The Whig Interpretation of History. New York: Norton, 1965.

[105] A. Caliskan, J. J. Bryson, and A. Narayanan, “Semantics derived automatically from

language corpora contain human-like biases,” Science, vol. 356, no. 6334, 183–186,
2017.

[106] J. J. Camilleri, M. R. Haghshenas, and G. Schneider, “A Web-Based Tool for Analysing

Normative Documents in English,” arXiv:1707.03997 [cs], 2017.

[107] A. Camus, “The Blood of the Hungarians,” 1957.

[108] E. H. Carr, The Romantic Exiles: A Nineteenth-century Portrait Gallery. Penguin

Books, 1933.

[109] E. H. Carr, Studies in Revolution. London: Routledge, 1962.

[110] E. H. Carr, Mikhail Bakunin. Springer, 1975.

[111] T. Carver, Marx & Engels: The Intellectual Relationship. London: Wheatsheaf Books,
1983.

[112] T. Carver, “Marx and Engels’ ’Outlines of a Critique of Political Economy’,” History
of Political Thought, vol. 4, no. 2, 357–365, 1983.

[113] T. Carver, The Cambridge Companion to Marx. Cambridge: Cambridge University

Press, 1991.

[114] T. Carver, Postmodern Marx. University Park: Penn State Press, 2010.

[115] T. Carver, “’The German Ideology’ Never Took Place,” History of Political Thought,
vol. 31, no. 1, 107–127, 2010.

[116] T. Carver, Engels before Marx. New York: Springer, 2020.

216
[117] T. Carver and D. Blank, A Political History of the Editions of Marx and Engels’
"German Ideology Manuscripts". London: Palgrave Macmillan, 2014.

[118] T. Carver and D. Blank, Marx and Engels’s "German Ideology" Manuscripts: Presen-
tation and Analysis of the "Feuerbach Chapter". London: Palgrave Macmillan, 2014.

[119] T. Carver and J. Farr, “The Manifesto in Marx’s and Engels’s Lifetimes,” The Cam-
bridge Companion to The Communist Manifesto, Cambridge: Cambridge University
Press, 2015, 67–84.

[120] M. Ceci, L. Lesmo, A. Mazzei, M. Palmirani, and D. P. Radicioni, “Semantic An-

notation of Legal Texts Through a Framenet-based Approach,” Proceedings of the
25th IVR Congress Conference on AI Approaches to the Complexity of Legal Sys-
tems: Models and Ethical Challenges for Legal Systems, Legal Language and Legal
Ontologies, Argumentation and Software Agents, Berlin, Heidelberg: Springer-Verlag,
2012, 245–255.

[121] A. Césaire, “Letter to Maurice Thorez,” Social Text, vol. 28, no. 2 (103), 145–152,
1956.

[122] V. Cha, The Impossible State: North Korea, Past and Future. Harper Collins, 2012.

[123] I. Chalkidis, “A Deep Learning Approach to Contract Element Extraction.,” 155–164,

2017.

[124] I. Chalkidis, I. Androutsopoulos, and A. Michos, “Extracting Contract Elements,”

Proceedings of the 16th Edition of the International Conference on Articial Intelli-
gence and Law, New York, NY, USA: ACM, 2017, 19–28.

[125] I. Chalkidis, I. Androutsopoulos, and A. Michos, “Obligation and Prohibition Extrac-

tion Using Hierarchical RNNs,” Proceedings of the 56th Annual Meeting of the Associ-
ation for Computational Linguistics (Volume 2: Short Papers), Melbourne, Australia:
Association for Computational Linguistics, 2018, 254–259.

[126] P. T. Chamberlin, The Global Offensive: The United States, the Palestine Liberation
Organization, and the Making of the Post-Cold War Order. Oxford University Press,
2012.

[127] K. K. Chang and S. DeDeo, “Divergence and the Complexity of Difference in Text
and Culture,” Journal of Cultural Analytics, vol. 4, 1–36, 2020.

[128] R. Chartier, The Cultural Origins of the French Revolution. Duke University Press,
1981.

217
[129] P. Chattopadhyay, “Marx’s Notebooks of 1844-1847,” Marx’s Associated Mode of
Production: A Critique of Marxism, New York: Springer, 2016.

[130] E. A. Chi, J. Hewitt, and C. D. Manning, “Finding Universal Grammatical Relations

in Multilingual BERT,” arXiv:2005.04511 [cs], 2020.

[131] Y.-C. Chiang, Social Engineering and the Social Sciences in China, 1919-1949. Cam-
bridge University Press, 2001.

[132] M. T. Choldin, The Red Pencil: Artists, Scholars, and Censors in the USSR. Unwin
Hyman, 1989.

[133] N. Chomsky and M. Foucault, The Chomsky-Foucault Debate: On Human Nature.

New York: New Press : Distributed by W.W. Norton, 2006.

[134] G. Claeys, Marx and Marxism. New York: PublicAffairs, 2018.

[135] G. Claeys, The Cambridge Companion to Nineteenth-Century Thought. Cambridge:

Cambridge University Press, 2019.

[136] H. Cleaver, Reading Capital Politically. AK Press, 1979.

[137] H. Cleaver, Study Guide to Capital, Vol. I. 2000.

[138] A. Cobban, The Social Interpretation of the French Revolution. Cambridge University
Press, 1964.

[139] G. A. Cohen, Karl Marx’s Theory of History: A Defence. Princeton University Press,
1978.

[140] G. D. H. Cole, A History of Socialist Thought, Volume 1: The Forerunners, 1789-

1850. New York: Palgrave Macmillan, 1953.

[141] G. D. H. Cole, A History of Socialist Thought, Volume 2: Socialist Thought, Marxism

and Anarchism, 1850-1890. New York: Palgrave Macmillan, 1953.

[142] E. Colla, “Dragomen and checkpoints,” The Translator, vol. 21, no. 2, 132–153, 2015.

[143] R. G. Collingwood, The Idea of History. Oxford: Clarendon Press, 1946.

[144] H. Collins and C. Abramsky, Karl Marx and the British Labour Movement. New York:
Palgrave Macmillan, 1965.

[145] G. C. Comninel, “Marx’s Context,” History of Political Thought, vol. 21, no. 3, 467–
483, 2000.

218
[146] G. C. Comninel, Alienation and Emancipation in the Work of Karl Marx. Springer,
2018.

[147] B. Comrie, The Languages of the Soviet Union. CUP Archive, 1981.

[148] C. Condren, The Status and Appraisal of Classic Texts: An Essay on Political Theory,
Its Inheritance, and on the History of Ideas. Princeton, N.J.: Princeton University
Press, 1985.

[149] W. E. Connolly, The Terms of Political Discourse. Princeton University Press, 1974.

[150] A. C. Cook, Ed., Mao’s Little Red Book: A Global History. Cambridge: Cambridge
University Press, 2013.

[151] P. E. Corcoran and C. Fuchs, Before Marx: Socialism and Communism in France,
1830–48. Springer, 1983.

[152] A. Cornu, Karl Marx et Friedrich Engels : Leur Vie et Leur Oeuvre. V1: Les Années
d’enfance et de Jeunesse, La Gauche Hégélienne, 1818/1820-1844. 1955.

[153] A. Cornu, Karl Marx et Friedrich Engels : Leur Vie et Leur Oeuvre. V2: Du Libéralisme
Démocratique Au Communisme, La ’Gazette Rhénane’, Les ’Annales Franco-Allemandes’,
1842-1844. 1955.

[154] A. Cornu, Karl Marx et Friedrich Engels : Leur Vie et Leur Oeuvre. V3: Marx à
Paris. 1955.

[155] A. Cornu, Karl Marx et Friedrich Engels : Leur Vie et Leur Oeuvre. V4: La Formation
Du Matérialisme Historique, 1845-1846. 1955.

[156] G. S. Coulthard, Red Skin, White Masks: Rejecting the Colonial Politics of Recogni-
tion. University of Minnesota Press, 2014.

[157] R. H. Cummings, “Balloons Over East Europe: The Cold War Leaflet Campaign of
Radio Free Europe,” Falling Leaf Magazine: The Quarterly Journal of the PsyWar
Society, 2008.

[158] R. Dannreuther, The Soviet Union and the PLO. Springer, 2016.

[159] R. Darnton, The Business of Enlightenment: A Publishing History of the Ency-

clopédie, 1775-1800. Harvard University Press, 1979.

[160] R. Darnton, Censors at Work: How States Shaped Literature. New York, NY: W.W.
Norton & Company, 2014.

219
[161] R. Darnton, D. Roche, and N. Y. P. Library, Revolution in Print: The Press in France,
1775-1800. University of California Press, 1989.

[162] D. Davidson, “Belief and the basis of meaning,” Synthese, vol. 27, no. 3, 309–323,
1974.

[163] M. Davis, Old Gods, New Enigmas: Marx’s Lost Theory. Verso Books, 2018.

[164] S. DeDeo, “Bayesian Reasoning for Intelligent People,” 2018.

[165] S. DeDeo and E. A. Hobson, “From equality to hierarchy,” Proceedings of the National
Academy of Sciences, vol. 118, no. 21, 2021.

[166] N. Delougaz, “Some Significant Trends in Soviet Book Production,” The Library
Quarterly: Information, Community, Policy, vol. 19, no. 4, 250–262, 1949.

[167] M. J. Denny and A. Spirling, “Text Preprocessing For Unsupervised Learning: Why
It Matters, When It Misleads, And What To Do About It,” Political Analysis, vol. 26,
no. 2, 168–189, 2018.

[168] J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of Deep
Bidirectional Transformers for Language Understanding,” arXiv:1810.04805 [cs], 2019.

[169] J. P. Dexter et al., “Quantitative criticism of literary relationships,” Proceedings of

the National Academy of Sciences, vol. 114, no. 16, E3195–E3204, 2017.

[170] J. Diamanti, A. Pendakis, and I. Szeman, The Bloomsbury Companion to Marx.

Bloomsbury Publishing, 2018.

[171] J. A. Dibble, The Pythia’s Drunken Song: Thomas Carlyle’s Sartor Resartus and the
Style Problem in German Idealist Philosophy. Springer Science & Business Media,
1978.

[172] I. Dokmanic, R. Parhizkar, J. Ranieri, and M. Vetterli, “Euclidean Distance Matrices:

Essential theory, algorithms, and applications,” IEEE Signal Processing Magazine,
vol. 32, no. 6, 12–30, 2015.

[173] W. Doyle, Origins of the French Revolution. Oxford: Oxford University Press, 1999.

[174] M. Dragoni, S. Villata, W. Rizzi, and G. Governatori, “Combining NLP Approaches

for Rule Extraction from Legal Documents,” 1st Workshop on MIning and REasoning
with Legal Texts (MIREL 2016), Sophia Antipolis, France, 2016.

[175] H. Draper, Karl Marx’s Theory of Revolution, Vol. I. NYU Press, 1977.

220
[176] H. Draper, Karl Marx’s Theory of Revolution, Vol. II. NYU Press, 1977.

[177] H. Draper, Karl Marx’s Theory of Revolution, Vol. III. NYU Press, 1977.

[178] H. Draper, The Marx-Engels Chronicle: A Day-by-Day Chronology of Marx & Engels’
Life and Activity. New York: Schocken Books, 1985, vol. 1.

[179] H. Draper, The Marx-Engels Register: A Complete Bibliography of Marx and Engels’
Individual Writings. Schocken Books, 1985, vol. 2.

[180] H. Draper, The Marx-Engels Glossary: Glossary to the Chronicle and Register, and
Index to the Glossary. Knopf Doubleday Publishing Group, 1986.

[181] H. Draper, Karl Marx’s Theory of Revolution, Vol. IV. NYU Press, 1990.

[182] H. Draper and E. Haberkern, Karl Marx’s Theory of Revolution, Vol. V. NYU Press,
2010.

[183] K. Duan, S. S. Keerthi, and A. N. Poo, “Evaluation of simple performance measures

for tuning SVM hyperparameters,” Neurocomputing, vol. 51, 41–59, 2003.

[184] A. Dube, E. Kaplan, and S. Naidu, “Coups, Corporations, and Classified Information
*,” The Quarterly Journal of Economics, vol. 126, no. 3, 1375–1409, 2011.

[185] J.-N. Ducange, The French Revolution and Social Democracy: The Transmission of
History and Its Political Uses in Germany and Austria, 1889–1934. Brill, 2018.

[186] M. Dummett, Frege: Philosophy of Language. Harvard University Press, 1973.

[187] R. Dunayevskaya, Marx’s Capital and Today’s Global Crisis. News & Letters Com-
mittees, 1978.

[188] J. Dunn, The Political Thought of John Locke: An Historical Account of the Argument
of the ’Two Treatises of Government’. Cambridge University Press, 1969.

[189] E. Dussel, Towards An Unknown Marx: A Commentary on the Manuscripts of 1861-

63. Routledge, 2002.

[190] H. Eaton, “Marx and the Russians,” Journal of the History of Ideas, vol. 41, no. 1,
89–112, 1980.

[191] W. Eckhardt, The First Socialist Schism: Bakunin Vs. Marx in the International
Working Men’s Association. PM Press, 2016.

221
[192] N. Egami, “Identification of Causal Diffusion Effects Under Structural Stationarity,”
arXiv:1810.07858 [stat], 2020.

[193] N. Egami, C. J. Fong, J. Grimmer, M. E. Roberts, and B. M. Stewart, “How to Make

Causal Inferences Using Texts,” Science Advances, 2018.

[194] H. Ehrbar, Annotations to Karl Marx’s Capital. Marx Annotations Project, 2010.

[195] H. Ehrbar, Glossary to Marx’s Capital and Other Economic Writings. Marx Annota-
tions Project, 2010.

[196] J. Ehrenberg, Proudhon and His Age. Humanities Press, 1996.

[197] R. El Baff, H. Wachsmuth, K. Al Khatib, and B. Stein, “Analyzing the Persuasive

Effect of Style in News Editorial Argumentation,” Proceedings of the 58th Annual
Meeting of the Association for Computational Linguistics, Association for Computa-
tional Linguistics, 2020, 3154–3160.

[198] H. El-Fiqi, E. Petraki, and H. A. Abbass, “A computational linguistic approach for

the identification of translator stylometry using Arabic-English text,” 2011 IEEE
International Conference on Fuzzy Systems (FUZZ-IEEE 2011), 2011, 2039–2045.

[199] J. Elster, “The Case for Methodological Individualism,” Theory and Society, vol. 11,
no. 4, 453–482, 1982.

[200] J. Elster, Making Sense of Marx. Cambridge: Cambridge University Press, 1985.

[201] J. Elster, An Introduction to Karl Marx. Cambridge University Press, 1986.

[202] J. Elster, Alexis de Tocqueville, the First Social Scientist. Cambridge University Press,
2009.

[203] J. Elster, Explaining Social Behavior. Cambridge University Press, 2015.

[204] N. Ertürk and Ö. Serin, “Marxism, Communism, and Translation: An Introduction,”

boundary 2, vol. 43, no. 3, 1–26, 2016.

[205] J. A. Evans and P. Aceves, “Machine Translation: Mining Text for Social Theory,”
Annual Review of Sociology, vol. 42, no. 1, 21–50, 2016.

[206] M. Evans, “Karl Marx’s first confrontation with political economy: The 1844 manuscripts,”
Economy and Society, vol. 13, no. 2, 115–152, 1984.

[207] G. Fabian, Karl Marx Prince of Darkness. Xlibris Corporation, 2011.

222
[208] N. Fathi, “Wipe Israel ’off the map’ Iranian says,” The New York Times, 2005.

[209] D. G. Faust, The Ideology of Slavery: Proslavery Thought in the Antebellum South,
1830–1860. LSU Press, 1981.

[210] P. Favilli, The History of Italian Marxism: From Its Origins to the Great War. BRILL,
1996.

[211] M. Fay, J. Hengstenberg, and B. Stuckey, “The Influence of Adam Smith on Marx’s
Theory of Alienation,” Science & Society, vol. 47, no. 2, 129–151, 1983.

[212] F. Fehér, The French Revolution and the Birth of Modernity. University of California
Press, 1990.

[213] F. Feng, Y. Yang, D. Cer, N. Arivazhagan, and W. Wang, “Language-agnostic BERT

Sentence Embedding,” Proceedings of the 60th Annual Meeting of the Association for
Computational Linguistics (Volume 1: Long Papers), Dublin, Ireland: Association for
Computational Linguistics, 2022, 878–891.

[214] R. Ferrod, C. Schifanella, L. Di Caro, and M. Cataldi, “Disclosing Citation Meanings

for Augmented Research Retrieval and Exploration,” The Semantic Web, P. Hitzler
et al., Eds., Springer International Publishing, 2019, 101–115.

[215] I. Fetscher, Der Marxismus. Seine Geschichte in Dokumenten: Philosophie, Ideologie,

Ökonomie, Soziologie, Politik. Piper, 1962.

[216] I. Fetscher, Marx and Marxism. Herder & Herder, 1971.

[217] L. S. Feuer, “The North American Origin of Marx’s Socialism,” Western Political
Quarterly, vol. 16, no. 1, 53–67, 1963.

[218] R. Finelli, A Failed Parricide: Hegel and the Young Marx. London: Brill, 2015.

[219] T. S. Finley, R. Franck, and N. D. Johnson, “The Effects of Land Redistribution:

Evidence from the French Revolution,” SSRN Electronic Journal, 2017.

[220] J. R. Firth, Papers in Linguistics, 1934-1951. Oxford University Press, 1957.

[221] S. Flaherty, Marx, Engels and Modern British Socialism: The Social and Political
Thought of H. M. Hyndman, E. B. Bax and William Morris. Springer Nature, 2020.

[222] V. S. Fomichev, Letopis’ Velikoi Zhizni: Bibliograficheskii Obzor Izdaniia ’Vladimir

Il’ich Lenin. Biograficheskaia Khronika. 1870-1924.’ Tt. 1-12. Moskva: Izd-vo polit.
lit-ry, 1984.

223
[223] E. Foner, Free Soil, Free Labor, Free Men: The Ideology of the Republican Party before
the Civil War, Reprint edition. Oxford ; New York: Oxford University Press, 1995.

[224] L. K. Ford, “Republican Ideology in a Slave Society: The Political Economy of John
C. Calhoun,” The Journal of Southern History, vol. 54, no. 3, 405–424, 1988.

[225] E. Francesconi and A. Passerini, “Automatic Classification of Provisions in Legislative

Texts,” Artificial Intelligence and Law, vol. 15, no. 1, 1–17, 2007.

[226] M. Freeden, M. Stears, L. T. Sargent, and Oxford University Press, The Oxford Hand-
book of Political Ideologies. Oxford: Oxford University Press, 2015.

[227] H. Freedman, The Murderous History of Bible Translations: Power, Conflict, and the
Quest for Meaning. Bloomsbury Publishing USA, 2016.

[228] F. Furet, Interpreting the French Revolution. Cambridge University Press, 1978.

[229] M. Gabriel, Love and Capital: Karl and Jenny Marx and the Birth of a Revolution.
Boston: Little, Brown and Company, 2011.

[230] H.-G. Gadamer, Truth and Method. New York: Crossroad, 1984, c1975.

[231] S. Gal, “Politics of Translation,” Annual Review of Anthropology, vol. 44, no. 1, 225–
240, 2015.

[232] R. J. Gallagher, K. Reing, D. Kale, and G. V. Steeg, “Anchored Correlation Ex-

planation: Topic Modeling with Minimal Domain Knowledge,” Transactions of the
Association for Computational Linguistics, vol. 5, no. 0, 529–542, 2017.

[233] D. R. Gandy, Marx and History: From Primitive Society to the Communist Future.
University of Texas Press, 1979.

[234] R. T. Gannett, Tocqueville Unveiled: The Historian and His Sources for The Old
Regime and the Revolution. University of Chicago Press, 2003.

[235] X. Gao and M. P. Singh, “Extracting Normative Relationships from Business Con-
tracts,” Proceedings of the 2014 International Conference on Autonomous Agents and
Multi-agent Systems, Richland, SC: International Foundation for Autonomous Agents
and Multiagent Systems, 2014, 101–108.

[236] P. Gärdenfors, Conceptual Spaces: The Geometry of Thought. Cambridge, MA, USA:
A Bradford Book, 2000.

[237] P. Gärdenfors, The Geometry of Meaning: Semantics Based on Conceptual Spaces.

Cambridge, MA, USA: MIT Press, 2014.

224
[238] L. A. Gatys, A. S. Ecker, and M. Bethge, “A Neural Algorithm of Artistic Style,”
arXiv:1508.06576 [cs, q-bio], 2015.

[239] E. George, Marxism and Translation.

[240] N. Geras, Marx and Human Nature: Refutation of a Legend. Verso Books, 2016.

[241] A. Gerow, Y. Hu, J. Boyd-Graber, D. M. Blei, and J. A. Evans, “Measuring Discur-

sive Influence Across Scholarship,” Proceedings of the National Academy of Sciences,
vol. 115, no. 13, 3308–3313, 2018.

[242] J.-P. A. Ghobrial, The Whispers of Cities: Information Flows in Istanbul, London,
and Paris in the Age of William Trumbull. OUP Oxford, 2013.

[243] A. Giddens, Central Problems in Social Theory: Action, Structure, and Contradiction
in Social Analysis. University of California Press, 1979.

[244] J. R. Gillingham, The E.U.: An Obituary. Verso Books, 2018.

[245] H. Gintis, Game Theory Evolving: A Problem-centered Introduction to Modeling Strate-

gic Behavior. Princeton University Press, 2000.

[246] P. Gleijeses, Conflicting Missions. 2002.

[247] P. Gleijeses, Visions of Freedom. 2013.

[248] Y. Goldberg, Neural Network Methods in Natural Language Processing. Morgan &
Claypool Publishers, 2017.

[249] D. Goodman, The Republic of Letters: A Cultural History of the French Enlighten-
ment. Cornell University Press, 1996.

[250] P. E. Gottfried, Leo Strauss and the Conservative Movement in America. Cambridge
University Press, 2011.

[251] R. V. Gould, Insurgent Identities: Class, Community, and Protest in Paris from 1848
to the Commune. Chicago: University of Chicago Press, 1995.

[252] A. W. Gouldner, “Marx’s Last Battle: Bakunin and the First International,” Theory
and Society, vol. 11, no. 6, 853–884, 1982.

[253] A. W. Gouldner, The Two Marxisms: Contradictions and Anomalies in the Develop-
ment of Theory. Seabury Press, 1980.

225
[254] A. Gourevitch, “Labor Republicanism and the Transformation of Work,” Political
Theory, vol. 41, no. 4, 591–617, 2013.

[255] A. Gourevitch, From Slavery to the Cooperative Commonwealth: Labor and Republican
Liberty in the Nineteenth Century. Cambridge: Cambridge University Press, 2015.

[256] K. Graham, “Illocution and ideology,” Issues in Marxist Philosophy, vol. 4, J. Mepham
and D. Ruben, Eds., 153–194, 1981.

[257] K. Graham, Karl Marx Our Contemporary: Social Theory for a Post-Leninist World.
University of Toronto Press, 1992.

[258] N. Grant, Communist Psychological Offensive: Distortions in the Translation of Of-

ficial Documents. Research Institute of the Sino-Soviet Bloc, 1961.

[259] B. C. Greenwald and J. E. Stiglitz, “Externalities in Economies with Imperfect In-

formation and Incomplete Markets,” The Quarterly Journal of Economics, vol. 101,
no. 2, 229–264, 1986.

[260] H. Grégoire, R. Jakobson, and M. Szeftel, La geste du prince Igor, épopée russe du
douzième siècle. New York: Institut de philologie et d’histoire orientales et slaves,
1948.

[261] D. Gregory, “Karl Marx’s and Friedrich Engels’ Knowledge of French Socialism in
1842-43,” Réflexions Historiques, vol. 10, no. 1, 143–193, 1983.

[262] H. P. Grice, “Utterer’s Meaning and Intentions,” The Philosophical Review, vol. 78,
no. 2, 147–177, 1969.

[263] J. Grimmer, “A Bayesian Hierarchical Topic Model for Political Texts: Measuring
Expressed Agendas in Senate Press Releases,” Political Analysis, vol. 18, no. 1, 1–35,
2010/ed.

[264] J. Grimmer and B. M. Stewart, “Text as Data: The Promise and Pitfalls of Automatic
Content Analysis Methods for Political Texts,” Political Analysis, vol. 21, no. 3, 267–
297, 2013.

[265] J. Gronow, On the Formation of Marxism: Karl Kautsky’s Theory of Capitalism, the
Marxism of the Second International and Karl Marx’s Critique of Political Economy.
BRILL, 2015.

[266] M. Grootendorst, BERTopic: Leveraging BERT and c-TF-IDF to create easily inter-
pretable topics. 2020.

226
[267] J. Guillaume, L’Internationale: documents et souvenirs (1864-1878), Tome I. Paris:
Société nouvelle de librairie et d’édition, 1905, vol. 1.

[268] R. Guillermo, “The Philippines,” Routledge Handbook of Marx’s Capital : A Global

History of Translation, Dissemination and Reception, Taylor & Francis, 2017.

[269] S. Gupta, M. Musto, and B. Amini, Eds., Karl Marx’s Life, Ideas, and Influences: A
Critical Examination on the Bicentenary. London: Palgrave Macmillan, 2019.

[270] T. R. Gurr, Why Men Rebel. Routledge, 1970.

[271] H. Hagemann, German Editions of Adam Smith’s Wealth of Nations. London: Rout-
ledge, 2017, 123–132.

[272] J. P. Haithcox, Communism and Nationalism in India: M.N. Roy and Comintern
Policy, 1920-1939. Princeton University Press, 2015.

[273] A. Hall, “The War of Words: Anti-Socialist Offensives and Counter-Propaganda in

Wilhelmine Germany 1890-1914,” Journal of Contemporary History, vol. 11, no. 2/3,
11–42, 1976.

[274] F. Halliday, Revolution and World Politics: The Rise and Fall of the Sixth Great
Power. 1999.

[275] W. L. Hamilton, J. Leskovec, and D. Jurafsky, “Diachronic Word Embeddings Re-

veal Statistical Laws of Semantic Change,” Proceedings of the 54th Annual Meeting
of the Association for Computational Linguistics (Volume 1: Long Papers), Berlin,
Germany: Association for Computational Linguistics, 2016, 1489–1501.

[276] O. J. Hammen, The Red ’48ers: Karl Marx and Friedrich Engels. New York: Scribner,
1969.

[277] I. Hanzel, “Mistranslations of ‘Schein’ and ‘Erscheinung’: The Structure of Chapter

1 of "Capital", Volume I,” Science & Society, vol. 74, no. 4, 509–537, 2010.

[278] J. C. Harsanyi, “Cardinal Utility in Welfare Economics and in the Theory of Risk-
taking,” Journal of Political Economy, vol. 61, no. 5, 434–435, 1953.

[279] D. Harvey, “The ’New’ Imperialism: Accumulation by Dispossession,” Socialist Reg-

ister, vol. 40, 2004.

[280] D. Harvey, A Companion to Marx’s Capital: The Complete Edition. Verso Books,
2018.

227
[281] D. Harvey, Marx, Capital and the Madness of Economic Reason. Oxford: Oxford
University Press, 2018.

[282] P. Haubtmann, Pierre-Joseph Proudhon: sa vie et sa pensée, 1809-1849. Editions

Beauchesne, 1982.

[283] W. F. Haug, “On the Need for a New English Translation of Marx’s Capital,” Social-
ism and Democracy, vol. 31, no. 1, 60–86, 2017.

[284] R. Hecker, “Die Entstehungs-, Überlieferungs- und Editionsgeschichte der ökonomis-

chen Manuskripte und des "Kapital",” Trend, 2003.

[285] K. Heffernan, O. Çelebi, and H. Schwenk, Bitext Mining Using Distilled Sentence
Representations for Low-Resource Languages, 2022.

[286] G. W. F. Hegel, Werke in 20 Bänden Mit Registerband, Band 7: Grundlinien Der

Philosophie Des Rechts Oder Naturrecht Und Staatswissenschaft Im Grundrisse, 8.
Aufl. Suhrkamp Verlag, 1820.

[287] M. Heinrich, “Engels’ Edition of the Third Volume of Capital and Marx’s Original
Manuscript,” Science & Society, vol. 60, no. 4, 452–466, 1996.

[288] M. Heinrich, An Introduction to the Three Volumes of Karl Marx’s Capital. New York:
NYU Press, 2012.

[289] M. Heinrich, “’Capital’ after MEGA: Discontinuities, Interruptions, and New Begin-
nings,” Crisis and critique, vol. 3, no. 3, 92–138, 2016.

[290] M. Heinrich, Karl Marx und die Geburt der modernen Gesellschaft: 1818-1841. Schmetter-
ling Verlag, 2018.

[291] M. Heinrich, Karl Marx and the Birth of Modern Society: The Life of Marx and the
Development of His Work. New York: NYU Press, 2019.

[292] W. Heisenberg, “The Development of Quantum Mechanics,” Nobel Lecture, 12, 1932.

[293] A. Heller, The Theory of Need in Marx. Verso Books, 1976.

[294] J. B. Hench, Books as Weapons: Propaganda, Publishing, and the Battle for Global
Markets in the Era of World War II. 2016.

[295] B. Henderson, “William plate, an unknown acquaintance of karl marx at the british
museum: A biographical sketch,” Electronic British Library Journal, vol. 8, 1–9, 2005.

228
[296] K. H. Hennings, “A note on Marx’s reading list in his Economic and Philosophical
Manuscripts of 1844,” Economy and Society, vol. 14, no. 1, 128–137, 1985.

[297] M. Hess, Die europäische Triarchie. Wigand, 1841.

[298] C. Hill, Marx’s Phantasmagorical Metaphor. Unpublished Manuscript, 2015.

[299] K. Hlaváčková-Schindler, The Assumption of Non-Gaussianity in Natural and Social

Sciences and Its Influence on Detection of Causal Relationships. IntechOpen, 2012.

[300] T. Hobbes, Leviathan: With Selected Variants from the Latin Edition of 1668. Hackett
Publishing, 1994.

[301] E. J. Hobsbawm, The Age of Revolution: Europe 1789-1848. London: Weidenfeld &
Nicolson, 1962.

[302] E. J. Hobsbawm, The Age of Capital, 1848-1875. London: Weidenfeld & Nicolson,
1975.

[303] E. J. Hobsbawm, The History of Marxism: Marxism in Marx’s Day. Indiana Univer-
sity Press, 1982.

[304] E. J. Hobsbawm, Age Of Empire: 1875-1914. Orion, 2010.

[305] E. J. Hobsbawm, How to Change the World: Reflections on Marx and Marxism. Yale
University Press, 2011.

[306] H. Hodgkinson, Doubletalk: The Language of Communism. G. Allen & Unwin, 1955.

[307] J. Hoff, Marx Worldwide: On the Development of the International Discourse on Marx
Since 1965. Boston: Brill, 2017.

[308] R. Hoffman, “Marx and Proudhon: A Reappraisal of Their Relationship,” The His-
torian, vol. 29, no. 3, 409–430, 1967.

[309] R. L. Hoffman, Revolutionary Justice: The Social and Political Theory of P.-J. Proud-
hon. University of Illinois Press, 1972.

[310] W. N. Hohfeld, “Some Fundamental Legal Conceptions as Applied in Judicial Rea-

soning,” Yale Law Journal, vol. 23, 1913.

[311] S. Hollander, Friedrich Engels and Marxian Political Economy. Cambridge: Cam-
bridge University Press, 2011.

229
[312] S. Hook, From Hegel to Marx: Studies in the Intellectual Development of Karl Marx.
New York: Columbia University Press, 1936.

[313] K. Hosseini, K. Beelen, G. Colavizza, and M. C. Ardanuy, “Neural Language Models

for Nineteenth-Century English,” arXiv:2105.11321 [cs], 2021.

[314] J. House, “Translation as a Prime Player in Intercultural Communication,” Applied

Linguistics, vol. 41, no. 1, 10–29, 2020.

[315] D. Howard, The Development of the Marxian Dialectic. Southern Illinois University
Press, 1972.

[316] D. Howard, From Marx to Kant. Springer, 1992.

[317] M. M. Howard, The Politics of Citizenship in Europe. Cambridge University Press,

2009.

[318] H. Hryn, “The Executed Renaissance Paradigm Revisited,” Harvard Ukrainian Stud-
ies, vol. 27, no. 1/4, 67–96, 2004.

[319] K. J. Hudson, “The Double Blow: 1956 and the Communist Party of Great Britain,”
Doctoral, University of London, 1992.

[320] A. Hughes, Marxism’s Retreat from Africa. London: Routledge, 2015.

[321] J. M. Hughes, N. J. Foti, D. C. Krakauer, and D. N. Rockmore, “Quantitative pat-

terns of stylistic influence in the evolution of literature,” Proceedings of the National
Academy of Sciences, vol. 109, no. 20, 7682–7686, 2012.

[322] R. N. Hunt, The Political Ideas of Marx and Engels, Vol. 1: Marxism and Totalitarian
Democracy, 1818-50. London: Palgrave Macmillan, 1974, vol. 1.

[323] R. N. Hunt, The Political Ideas of Marx and Engels, Vol. 2: Classical Marxism, 1850-
1895. London: Palgrave Macmillan, 1984, vol. 2.

[324] T. Hunt, Marx’s General: The Revolutionary Life of Friedrich Engels. New York:
Henry Holt and Company, 2010.

[325] H. M. Hurd and M. S. Moore, “The Hohfeldian Analysis of Rights,” The American
Journal of Jurisprudence, vol. 63, no. 2, 295–354, 2018.

[326] S. E. Hyman, The Tangled Bank: Darwin, Marx, Frazer and Freud as Imaginative
Writers. Grosset & Dunlap, 1966.

[327] J. Hyppolite, Studies on Marx and Hegel. New York: Harper Torchbooks, 1969.

230
[328] B. F. Hyslop, A Guide to the General Cahiers of 1789. With the Texts of Unedited
Cahiers. New York: Columbia University Press, 1936.

[329] Y. Ishikawa, The Formation of the Chinese Communist Party. Columbia University
Press, 2012.

[330] J. Israel, Revolutionary Ideas: An Intellectual History of the French Revolution from
The Rights of Man to Robespierre. Princeton University Press, 2014.

[331] J. Jacobs, “Machine-Learning ’Influence’ in the History of Political Thought,” Work-

ing Paper, Presented at 2018 Political Theory Works in Progress Seminar, Columbia
University, 2021.

[332] J. Jacobs, “Quantifying Cultural Diplomacy: The Translation and Diffusion of Marx-
ism from the Communist Manifesto to the Cold War,” Working Paper, 2021.

[333] J. Jacobs, “Threadsheets: An Interactive Visualization Tool for Data Concordance

Analysis,” Submitted to Journal of Statistical Software (JSS), 2021.

[334] J. Jacobson, When the Soviet Union Entered World Politics. University of California
Press, 1994.

[335] D. A. Jaeger and M. D. Paserman, “The Cycle of Violence? An Empirical Analysis

of Fatalities in the Palestinian-Israeli Conflict,” American Economic Review, vol. 98,
no. 4, 1591–1604, 2008.

[336] R. Jakobson, Child Language, Aphasia and Phonological Universals. Walter de Gruyter,
1941.

[337] R. Jakobson, “Linguistics in Its Relations to Other Sciences,” Eight Decades of Gen-
eral Linguistics, 265–304, 1957.

[338] R. Jakobson, “Linguistics and Poetics,” Linguistics and Poetics, De Gruyter Mouton,
1960, 18–51.

[339] E. Jay and R. Jay, Critics of Capitalism: Victorian Reactions to ’Political Economy’.
Cambridge: Cambridge University Press, 1986.

[340] Z. Jelveh, B. Kogut, and S. Naidu, “Political Language in Economics,” Social Science
Research Network, Rochester, NY, SSRN Scholarly Paper 2535453, 2018.

[341] J. Jensen et al., “Political Polarization and the Dynamics of Political Language: Evi-
dence from 130 Years of Partisan Speech [with Comments and Discussion],” Brookings
Papers on Economic Activity, 1–81, 2012.

231
[342] M. L. Jockers, Macroanalysis: Digital Methods and Literary History. Urbana: Univer-
sity of Illinois Press, 2013.

[343] A. R. Johnson and P. R. Eugene, Cold War Broadcasting: Impact on the Soviet Union
and Eastern Europe. 2010.

[344] S. Johnson, “The Early Life of Marx’s ’Mode of Production’,” Modern Intellectual
History, 1–30, 2019.

[345] D. Jurafsky, The Language of Food: A Linguist Reads the Menu. W. W. Norton &
Company, 2014.

[346] P. Kägi, Genesis des historischen Materialismus: Karl Marx und die Dynamik der
Gesellschaft. Europa Verlag, 1965.

[347] E. Kamenka, The Ethical Foundations of Marxism. London: Routledge, 2015.

[348] E. Kamenka and F. B. Smith, Intellectuals and Revolution: Socialism and the Expe-
rience of 1848. E. Arnold, 1979.

[349] N. Kamovnikova, Made Under Pressure: Literary Translation in the Soviet Union,
1960-1991. University of Massachusetts Press, 2019.

[350] K. Katsiakoris, “Une superpuissance éducative: L’Union soviétique et la formation

des élites du tiers-monde (1956–1991),” Traverse: Zeitschrift für Geschichte/Revue
d’histoire, vol. 1, 108–121, 2018.

[351] K. Katsiakoris, “The Lumumba University in Moscow: Higher Education for a Soviet-
Third World Alliance, 1945-1991,” Journal of Global History, vol. 14, no. 2, 2019.

[352] C. Katz, “The Greek Matrix of Marx’s Critique of Political Economy,” History of
Political Thought, vol. 15, no. 2, 229–248, 1994.

[353] D. R. Kelley, “The Metaphysics of Law: An Essay on the Very Young Marx,” The
American Historical Review, vol. 83, no. 2, 350–367, 1978.

[354] D. R. Kelley, “The Science of Anthropology: An Essay on the Very Old Marx,”
Journal of the History of Ideas, vol. 45, no. 2, 245–262, 1984.

[355] D. R. Kelley, The Descent of Ideas: The History of Intellectual History. Routledge,
2002.

[356] N. Khoo, “Breaking the Ring of Encirclement: The Sino-Soviet Rift and Chinese
Policy toward Vietnam, 1964–1968,” Journal of Cold War Studies, vol. 12, no. 1, 3–
42, 2010.

232
[357] T. Kiss and J. Strunk, “Unsupervised Multilingual Sentence Boundary Detection,”
Computational Linguistics, vol. 32, no. 4, 485–525, 2006.

[358] G. N. Kitching, Marxism and Science: Analysis of an Obsession. Pennsylvania State

University Press, 1994.

[359] E. Kitov, The Book of Our Heritage: The Jewish Year and Its Days of Significance.
Feldheim Publishers, 1997.

[360] J. Klugmann and J. T. Callaghan, Cold War, Crisis and Conflict: The History of the
Communist Party of Great Britain, 1951-68. Lawrence & Wishart, 2001.

[361] K. Knorr and I. H. Tribune, “Albert Camus: Being Right About the Left,” The New
York Times, 1996.

[362] J. Koedam, “A change of heart? Analysing stability and change in European party
positions,” West European Politics, vol. 45, no. 4, 693–715, 2022.

[363] P. Koehn, “Europarl: A Parallel Corpus for Statistical Machine Translation,” Pro-
ceedings of Machine Translation Summit X: Papers, Phuket, Thailand, 2005, 79–86.

[364] L. Kołakowski, The Alienation of Reason; a History of Positivist Thought. Garden

City, N.Y.: Doubleday, 1968.

[365] L. Kolakowski, Main Currents of Marxism. New York: W. W. Norton & Company,
1976.

[366] L. A. Kornhauser and W. B. MacLeod, “Contracts between Legal Persons,” Institute

of Labor Economics (IZA), Tech. Rep. 5352, 2010.

[367] R. Koselleck, “Social history and conceptual history,” International Journal of Poli-
tics, Culture, and Society, vol. 2, no. 3, 308–325, 1989.

[368] R. Koselleck, The Practice of Conceptual History: Timing History, Spacing Concepts.
Stanford University Press, 2002.

[369] A. C. Kozlowski, M. Taddy, and J. A. Evans, “The Geometry of Culture: Analyzing

the Meanings of Class through Word Embeddings,” American Sociological Review,
vol. 84, no. 5, 905–949, 2019.

[370] H. Krabbenham and S. Giles, The Cultural Cold War in Western Europe, 1945-1960.
London-Portland, OR: Frank Cass Publishers, 2002.

[371] L. Krader, The Ethnological Notebooks of Karl Marx, Transcribed and Edited, with
an Introduction, by Lawrence Krader. Van Gorcum & Company, 1972.

233
[372] M. Kukis, Voices from Iraq: A People’s History, 2003–2009. Columbia University
Press, 2011.

[373] V. Kulkarni, R. Al-Rfou, B. Perozzi, and S. Skiena, “Statistically Significant Detection

of Linguistic Change,” Proceedings of the 24th International Conference on World
Wide Web, Republic and Canton of Geneva, Switzerland: International World Wide
Web Conferences Steering Committee, 2015, 625–635.

[374] H.-D. Kurz, T. Nishizawa, and K. Tribe, The Dissemination of Economic Ideas. Ed-
ward Elgar Publishing, 2011.

[375] M. J. Kusner, Y. Sun, N. I. Kolkin, and K. Q. Weinberger, “From word embeddings to

document distances,” Proceedings of the 32nd International Conference on Interna-
tional Conference on Machine Learning - Volume 37, Lille, France: JMLR.org, 2015,
957–966.

[376] J. Kuypers, “Karl Marx’ Belgischer Freundeskreis (1845-48): Einige Notizen aus Bel-
gischen Archiven,” International Review of Social History, vol. 7, no. 3, 446–458,
1962.

[377] F. Lassalle, Ferdinand Lassalle: Nachgelassene Briefe und Schriften, Vol. III: Der
Briefwechsel Zwischen Lassalle und Marx, G. Mayer, Ed. Deutsche Verlags-anstalt,
1922.

[378] F. Lassalle, Reden und Schriften: Aus der Arbeiteragitation 1862-1864. Deutscher
Taschenbuch Verlag, 1970.

[379] H. D. Lasswell and N. Leites, Language of Politics: Studies in Quantitative Semantics.

Cambridge, MA, USA: MIT Press, 1965.

[380] M. A. Lause, A Secret Society History of the Civil War. University of Illinois Press,
2011.

[381] S. Laviosa, Corpus-Based Translation Studies: Theory, Findings, Applications. Rodopi,

2002.

[382] T. M. Law, When God Spoke Greek: The Septuagint and the Making of the Christian
Bible. Oxford University Press, 2013.

[383] A. Lazaridou, E. Bruni, and M. Baroni, “Is this a wampimuk? Cross-modal mapping
between distributional semantics and the visual world,” Proceedings of the 52nd An-
nual Meeting of the Association for Computational Linguistics (Volume 1: Long Pa-
pers), Baltimore, Maryland: Association for Computational Linguistics, 2014, 1403–
1414.

234
[384] G. Lefebvre, The Coming of the French Revolution. Princeton, N.J.: Princeton Uni-
versity Press, 1947.

[385] H. Lefebvre, Hegel, Marx, Nietzsche: Or the Realm of Shadows. New York: Verso
Books, 2020.

[386] J.-F. Legrain, Les voix du soulèvement palestinien, 1987-1988: Edition critique des
communiqués du Commandement National Unifié du Soulèvement et du Mouvement
de la Résistance Islamique. Traduction française en collaboration avec Pierre Chenard.
Centre d’Etudes et de Documentation Economique, Juridique et Sociale, 1991.

[387] J. Legrand, M. Auli, and R. Collobert, “Neural Network-based Word Alignment

through Score Aggregation,” Proceedings of the First Conference on Machine Trans-
lation: Volume 1, Research Papers, Berlin, Germany: Association for Computational
Linguistics, 2016, 66–73.

[388] B. Leipold, Citizen Marx: The Relationship Between Karl Marx and Republicanism.
Oxford: University of Oxford, 2017.

[389] B. Leipold, K. Nabulsi, and S. White, Radical Republicanism: Recovering the Tradi-
tion’s Popular Heritage. Oxford: Oxford University Press, 2020.

[390] D. Leopold, The Young Karl Marx: German Philosophy, Modern Politics, and Human
Flourishing. Cambridge: Cambridge University Press, 2007.

[391] N. Levine, The Tragic Deception: Marx Contra Engels. Santa Barbara, CA: ABC-
CLIO, 1975.

[392] N. Levine, Dialogue Within the Dialectic. London: Allen & Unwin, 1984.

[393] N. Levine, “Marx’s First Appropriation Of Hegel,” Critique, vol. 33, no. 1, 125–156,
2005.

[394] N. Levine, Divergent Paths: The Hegelian Foundations of Marx’s Method. Lanham,
MD: Lexington Books, 2006.

[395] N. Levine, Marx’s Discourse with Hegel. London: Palgrave Macmillan UK, 2012.

[396] N. Levine, Marx’s Rebellion Against Lenin. New York: Springer, 2015.

[397] N. Levine, Marx’s Resurrection of Aristotle. New York: Springer, 2021.

[398] B. Lewis, From Babel to Dragomans: Interpreting the Middle East. Oxford University
Press, 2004.

235
[399] M. Lewis-Beck, A. E. Bryman, and T. F. Liao, The SAGE Encyclopedia of Social
Science Research Methods. SAGE Publications, 2003.

[400] P. Liang, B. Taskar, and D. Klein, “Alignment by Agreement,” Proceedings of the

Human Language Technology Conference of the NAACL, Main Conference, New York
City, USA: Association for Computational Linguistics, 2006, 104–111.

[401] V. L. Lidtke, “German Social Democracy and German State Socialism, 1876–1884,”
International Review of Social History, vol. 9, no. 2, 202–225, 1964.

[402] A. Liebich, “On the Origins of a Marxist Theory of Bureaucracy in the Critique of
Hegel’s ‘Philosophy of Right’,” Political Theory, vol. 10, no. 1, 77–93, 1982.

[403] S.-E. Liedman, A World to Win: The Life and Works of Karl Marx. New York: Verso
Books, 2018.

[404] L. T. Lih, Lenin Rediscovered: What Is to Be Done? In Context. Brill, 2006.

[405] L. T. Lih, Lenin. Reaktion Books, 2012.

[406] L. H. Liu, Tokens of Exchange: The Problem of Translation in Global Circulations.

Duke University Press, 2000.

[407] N. Lobkowicz, “Eschatology and the Young Hegelians,” The Review of Politics, vol. 27,
no. 3, 434–439, 1965.

[408] J. A. London, “Re-imagining the Cambridge School in the Age of Digital Humanities,”
Annual Review of Political Science, vol. 19, no. 1, 351–373, 2016.

[409] A. López, “Traduttore, Traditore—Translator, Traitor,” Counterpoints, vol. 343, 141–

147, 2008.

[410] D. Losurdo, Liberalism: A Counter-History. Verso Books, 2014.

[411] D. Losurdo, Nietzsche, the Aristocratic Rebel: Intellectual Biography and Critical
Balance-sheet. Brill, 2019.

[412] K. Löwith, From Hegel to Nietzsche: The Revolution in Nineteenth Century Thought.
New York: Columbia University Press, 1964.

[413] M. Löwy, The Theory of Revolution in the Young Marx. London: Brill, 1973.

[414] T. Luong, H. Pham, and C. D. Manning, “Bilingual Word Representations with Mono-
lingual Quality in Mind,” Proceedings of the 1st Workshop on Vector Space Modeling

236
 for Natural Language Processing, Denver, Colorado: Association for Computational
Linguistics, 2015, 151–159.

[415] Y. Ma, S. Mukherjee, and B. Uzzi, “Mentorship and protégé success in STEM fields,”
Proceedings of the National Academy of Sciences, vol. 117, no. 25, 14077–14083, 2020.

[416] D. MacGregor, The Communist Ideal in Hegel and Marx. London: Routledge, 2015.

[417] C. Mackerras and N. Knight, Marxism in Asia. Routledge, 2015.

[418] P. Magness and M. Makovi, “The Mainstreaming of Marx: Measuring the Effect of
the Russian Revolution on Karl Marx’s Influence,” SSRN Working Paper, 2020.

[419] J. Maguire, Marx’s Paris Writings: An Analysis. Barnes & Noble, 1973.

[420] H. Mah, “The French Revolution and the Problem of German Modernity: Hegel,
Heine, and Marx,” New German Critique, no. 50, 3–20, 1990.

[421] M. Maidan, “The Rezeptionsgeschichte of the Paris Manuscripts,” History of Euro-

pean Ideas, vol. 12, no. 6, 767–781, 1990.

[422] J. Malcolm and A. Swearer, “3 Common Traits of School Shooters,” The Heritage
Foundation Blog, 2018.

[423] E. Mandel, The Formation of the Economic Thought of Karl Marx: 1843 to Capital.
Verso Books, 1971.

[424] E. Mandel, Trotsky as Alternative. Verso, 1992.

[425] D. S. Manekin, “Waging War among Civilians: The Production and Restraint of
Counterinsurgent Violence in the Second Intifada,” Ph.D. dissertation, UCLA, 2012.

[426] J. Markoff, The Abolition of Feudalism: Peasants, Lords, and Legislators in the French
Revolution. Pennsylvania State University Press, 1996.

[427] K. Marx, Herr Vogt. A. Petsch & Company, 1860.

[428] K. Marx, Critique of Hegel’s ’Philosophy Of Right’, J. J. O’Malley, Ed. Cambridge:

Cambridge University Press, 1970.

[429] K. Marx and F. Engels, Manifesto of the Communist Party. New York: International
Publishers, 1848.

[430] J. P. C. Matthews, “The West’s Secret Marshall Plan for the Mind,” International
Journal of Intelligence and Counterintelligence, vol. 16, no. 3, 409–427, 2003.

237
[431] L. T. McCarty, “Permissions and Obligations,” Proceedings of the Eighth Interna-
tional Joint Conference on Artificial Intelligence - Volume 1, San Francisco, CA,
USA: Morgan Kaufmann Publishers Inc., 1983, 287–294.

[432] L. McConnaughey, J. Dai, and D. Bamman, “The Labeled Segmentation of Printed

Books,” Proceedings of the 2017 Conference on Empirical Methods in Natural Lan-
guage Processing, Copenhagen, Denmark: Association for Computational Linguistics,
2017, 737–747.

[433] S. McDonald and M. Ramscar, “Testing the distributional hypothesis: The influence
of context on judgements of semantic similarity,” In Proceedings of the 23rd Annual
Conference of the Cognitive Science Society, 2001, 611–6.

[434] R. McElreath, Statistical Rethinking: A Bayesian Course with Examples in R and

STAN. CRC Press, 2020.

[435] D. McLellan, The Young Hegelians and Karl Marx. London: Macmillan, 1969.

[436] D. McLellan, Marx Before Marxism. Harper & Row, 1970.

[437] D. McLellan, Karl Marx: His Life and Thought. London: Palgrave Macmillan UK,
1973.

[438] D. McLellan, Marxism After Marx. New York: Palgrave Macmillan, 2007.

[439] J. J. Mearsheimer, The Tragedy of Great Power Politics. New York: Norton, 2001.

[440] F. Mehring, Karl Marx: The Story of His Life. London: Routledge, 1918.

[441] D. Melnik, The Diffusion of Ricardo’s Theories in Russia. Routledge, 2014, 195–211.

[442] V. C. Merlassino, “Guide to the Analysis of the Spanish Translation of Karl Marx’s
The Capital: Critique of Political Economy,” The EE-T Project, 2011.

[443] N. Merz, S. Regel, and J. Lewandowski, “The Manifesto Corpus: A new resource
for research on political parties and quantitative text analysis,” Research & Politics,
vol. 3, no. 2, 2053168016643346, 2016.

[444] I. Mészáros, Marx’s Theory of Alienation. Merlin, 1970.

[445] T. Mikolov, K. Chen, G. Corrado, and J. Dean, “Efficient Estimation of Word Rep-
resentations in Vector Space,” 1st International Conference on Learning Representa-
tions, ICLR 2013, Scottsdale, Arizona, USA, May 2-4, 2013, Workshop Track Pro-
ceedings, Y. Bengio and Y. LeCun, Eds., 2013.

238
[446] T. Mikolov, I. Sutskever, K. Chen, G. Corrado, and J. Dean, “Distributed Represen-
tations of Words and Phrases and their Compositionality,” arXiv:1310.4546 [cs, stat],
2013.

[447] R. Miliband, “Marx And The State,” Socialist Register, vol. 2, 1965.

[448] S. Mishal and R. Aharoni, Speaking Stones: Communiqués from the Intifada Under-
ground. Syracuse University Press, 1994.

[449] N. Mitchell, Jimmy Carter in Africa: Race and the Cold War. 2016.

[450] D. Moggach, The New Hegelians: Politics and Philosophy in the Hegelian School.
Cambridge: Cambridge University Press, 2006.

[451] D. Moggach and P. L. Browne, The Social Question and the Democratic Revolution:
Marx and the Legacy of 1848. Ottawa: University of Ottawa Press, 2000.

[452] D. Moggach and G. Stedman Jones, The 1848 Revolutions and European Political
Thought. Cambridge: Cambridge University Press, 2018.

[453] F. Moretti, Distant Reading. London: Verso, 2013.

[454] D. Morse, “‘Enemies of the People’: Poetry and Politics in the Time of Stalin,” Social
Education, vol. 65, no. 4, 198–207, 2001.

[455] F. Morton, Thunder at Twilight: Vienna 1913/1914. Collier Books, 1990.

[456] F. Moseley, Marx’s Method in Capital: A Reexamination. London: Humanities Press,

1993.

[457] F. Moseley, Marx’s Economic Manuscript of 1864-1865. London: Brill, 2015.

[458] F. Moseley and T. Smith, Marx’s Capital and Hegel’s Logic: A Reexamination. Lon-
don: Brill, 2014.

[459] B. Moss, “Marx and the Permanent Revolution in France: Background to the Com-
munist Manifesto,” Socialist Register, vol. 34, 1998.

[460] F. Mosteller and D. L. Wallace, Inference and Disputed Authorship: The Federalist.
Center for the Study of Language and Information, 1964.

[461] J. Murdock, C. Allen, and S. DeDeo, “Exploration and exploitation of Victorian

science in Darwin’s reading notebooks,” Cognition, vol. 159, 117–126, 2017.

239
[462] J. Murdock, C. Allen, and S. DeDeo, “The Development of Darwin’s Origin of Species,”
arXiv:1802.09944 [cs], 2018.

[463] M. Musto, Karl Marx’s Grundrisse: Foundations of the Critique of Political Economy
150 Years Later. London: Routledge, 2008.

[464] M. Musto, “Dissemination and Reception of The Communist Manifesto in Italy: From
the Origins to 1945,” Critique, vol. 36, no. 3, 445–456, 2008.

[465] M. Musto, “Marx in Paris: Manuscripts and Notebooks of 1844,” Science & Society,
vol. 73, no. 3, 386–402, 2009.

[466] M. Musto, “The Formation of Marx’s Critique of Political Economy: From the Studies
of 1843 to the Grundrisse,” Socialism and Democracy, vol. 24, no. 2, 66–100, 2010.

[467] M. Musto, Marx for Today. Routledge, 2013.

[468] M. Musto, Workers Unite!: The International 150 Years Later. New York: Blooms-
bury Publishing USA, 2014.

[469] M. Musto, Another Marx: Early Manuscripts to the International. London: Blooms-
bury, 2018.

[470] M. Musto, Marx’s Capital after 150 Years: Critique and Alternative to Capitalism.
London: Routledge, 2019.

[471] M. Musto, The Last Years of Karl Marx: An Intellectual Biography. Stanford: Stanford
University Press, 2020.

[472] M. Musto, The Marx Revival: Key Concepts and New Critical Interpretations. Cam-
bridge University Press, 2020.

[473] M. Musto and B. Amini, Routledge Handbook of Marx’s Capital. Taylor & Francis
Group, 2023.

[474] S. Naidu, S.-H. Hwang, and S. Bowles, “The Evolution of Egalitarian Sociolinguistic
Conventions,” American Economic Review, vol. 107, no. 5, 572–577, 2017.

[475] D. Nguyen et al., “How We Do Things With Words: Analyzing Text as Social and
Cultural Data,” Frontiers in Artificial Intelligence, vol. 3, 2020.

[476] B. Nicolaievsky and O. Maenchen-Helfen, Karl Marx: Man and Fighter. London:
Routledge, 1936.

[477] M. Nicolaus, “The Unknown Marx,” New Left Review, vol. I/48, 41–61, 1968.

240
[478] R. A. Nielsen, “The Lonely Jihadist: Weak Networks and the Radicalization of Muslim
Clerics,” 2013.

[479] J. Nielson and R. Skousen, “How Much of the King James Bible Is William Tyn-
dale’s?” Reformation, vol. 3, no. 1, 49–74, 1998.

[480] A. H. Nimtz, “A Return to Lenin — But without Marx and Engels?” Science &
Society, vol. 73, no. 4, 452–473, 2009.

[481] T. O. Nitsch, “Marx on Man’s Sociality by Nature: An Inexplicable Omission?” In-

ternational Journal of Social Economics, vol. 19, no. 7/8/9, 108–120, 1992.

[482] H. Noel, “Listening to the Coalition Merchants: Measuring the Intellectual Influence
of Academic Scribblers,” The Forum, vol. 5, no. 3, 2007.

[483] A. J. Norval, “The Things We Do with Words - Contemporary Approaches to the

Analysis of Ideology,” British Journal of Political Science, vol. 30, no. 2, 313–346,
2000.

[484] J. S. Nye, Soft Power: The Means To Success In World Politics. PublicAffairs, 2009.

[485] A. Oakley, The Making of Marx’s Critical Theory: A Bibliographical Analysis. Lon-
don: Routledge, 1983.

[486] A. Oakley, Marx’s Critique of Political Economy Volume One: Intellectual Sources
and Evolution. Routledge, 1984.

[487] A. Oakley, Marx’s Critique of Political Economy Volume Two: Intellectual Sources
and Evolution. London: Routledge, 1984.

[488] O’Kane, Notes on translation in Part One of Capital. 2013.

[489] B. Ollman, Alienation: Marx’s Conception of Man in Capitalist Society. Cambridge:

Cambridge University Press, 1977.

[490] M. Olohan, “How frequent are the contractions?: A study of contracted forms in the
Translational English Corpus,” Target, vol. 15, no. 1, 59–89, 2003.

[491] J. O’Neill, P. Buitelaar, C. Robin, and L. O’Brien, “Classifying Sentential Modality

in Legal Language: A Use Case in Financial Regulations, Acts and Directives,” Pro-
ceedings of the 16th Edition of the International Conference on Articial Intelligence
and Law, New York, NY, USA: ACM, 2017, 159–168.

[492] S. K. Padover, The Letters of Karl Marx. Hoboken: Prentice-Hall, 1979.

241
[493] S. K. Padover, “Karl Marx: The Propagandist as Prophet,” Propaganda and Com-
munication in World History, Volume II: Emergence of Public Opinion in the West,
University of Hawaii Press, 1980, 210–238.

[494] F. R. Palmer, Mood and Modality. Cambridge University Press, 2001.

[495] B. Parekh and R. N. Berki, “The History of Political Ideas: A Critique of Q. Skinner’s
Methodology,” Journal of the History of Ideas, vol. 34, no. 2, 163–184, 1973.

[496] S. J. Parry-Giles, “Propaganda, effect, and the cold war: Gauging the status of amer-
ica’s “war of words”,” Political Communication, vol. 11, no. 2, 203–213, 1994.

[497] W. Parsons, “Keynes and the Politics of Ideas,” History of Political Thought, vol. 4,
no. 2, 367–392, 1983.

[498] V. P. Pauca, F. Shahnaz, M. W. Berry, and R. J. Plemmons, “Text Mining using

Non-Negative Matrix Factorizations,” Proceedings of the 2004 SIAM International
Conference on Data Mining, Lake Buena Vista, Florida, USA: Society for Industrial
and Applied Mathematics, 2004, 452–456.

[499] J. Pennington, R. Socher, and C. Manning, “GloVe: Global Vectors for Word Rep-
resentation,” Proceedings of the 2014 Conference on Empirical Methods in Natural
Language Processing (EMNLP), Doha, Qatar: Association for Computational Lin-
guistics, 2014, 1532–1543.

[500] B. Perfahl, Marx oder Lassalle?: zur ideologischen Position der österreichischen Ar-
beiterbewegung, 1869-1889. Eropaverlag, 1982.

[501] W. Peters and A. Wyner, “Legal Text Interpretation: Identifying Hohfeldian Relations
from Text,” 6, 2016.

[502] J. Peterson and J. Densley, “Op-Ed: We have studied every mass shooting since 1966.
Here’s what we’ve learned about the shooters,” Los Angeles Times, 2019.

[503] P. Pettit, Republicanism: A Theory of Freedom and Government. Oxford University

Press, 1997.

[504] S. Pierson, Marxism and the Origins of British Socialism: The Struggle for a New
Consciousness, First Edition. Ithaca: Cornell University Press, 1973.

[505] P. Pilbeam, French Socialists before Marx: Workers, Women and the Social Question
in France. McGill-Queen’s Press, 2000.

[506] P. M. Pilbeam, Saint-Simonians in Nineteenth-Century France. From Free Love to

Algeria. London: Palgrave Macmillan, 2014.

242
[507] J. Plamenatz, Man and Society: Political and Social Theories from Machiavelli to
Marx. Longman, 1963.

[508] Plato, The Republic. Penguin Books Limited, 2007.

[509] J. G. A. Pocock, Virtue, Commerce, and History: Essays on Political Thought and
History, Chiefly in the Eighteenth Century. Cambridge: Cambridge University Press,
1985.

[510] J. G. A. Pocock, Political Thought and History: Essays on Theory and Method. Cam-
bridge University Press, 2009.

[511] J. G. A. Pocock, The Machiavellian Moment: Florentine Political Thought and the
Atlantic Republican Tradition. Princeton University Press, 1975.

[512] M. Polanyi, Personal Knowledge: Towards a Post-Critical Philosophy. University of

Chicago Press, 1958.

[513] K. Popper, The Open Society and Its Enemies. Routledge, 1945.

[514] J. Powell, Biographical Dictionary of Literary Influences: The Nineteenth Century,

1800-1914. Greenwood Publishing Group, 2001.

[515] V. Prashad, The Poorer Nations: A Possible History of the Global South. Verso Books,
2013.

[516] S. S. Prawer, Karl Marx and World Literature. Oxford: Clarendon Press, 1976.

[517] A. Press, “2 Afghans face death over translation of Quran,” NBC News, 2009.

[518] J. M. Prieto, “On translating a poem by osip mandelstam,” In Translation: Trans-

lators on Their Work and What It Means, Columbia University Press, 2013, 127–
142.

[519] W. V. O. Quine, Word and Object. Cambridge, MA: MIT Press, 1960.

[520] A. Rana, The Two Faces of American Freedom. Harvard University Press, 2014.

[521] P. Rauhala, The Dissemination of Capital in Nordic Countries, 2017.

[522] J. Rawls, “Outline of a Decision Procedure for Ethics,” The Philosophical Review,
vol. 60, no. 2, 177–197, 1951.

[523] B. M. G. Reardon, “The Romanticism of Karl Marx,” The Downside Review, vol. 96,
no. 322, 1–12, 1978.

243
[524] F. Regnault, “Les Maladies de Karl Marx,” Revue anthropologique, vol. 43, 293–317,
1933.

[525] R. Řehůřek and P. Sojka, “Software Framework for Topic Modelling with Large Cor-
pora,” Proceedings of the LREC 2010 Workshop on New Challenges for NLP Frame-
works, Valletta, Malta: ELRA, 2010, 45–50.

[526] R. W. Reichard, Crippled From Birth; German Social Democracy 1844-1870. Ames:
Iowa State University Press, 1969.

[527] N. Reimers and I. Gurevych, “Sentence-BERT: Sentence Embeddings using Siamese

BERT-Networks,” arXiv:1908.10084, 2019.

[528] N. Reimers and I. Gurevych, “Making monolingual sentence embeddings multilin-

gual using knowledge distillation,” Proceedings of the 2020 Conference on Empirical
Methods in Natural Language Processing, Association for Computational Linguistics,
2020.

[529] A. A. Reisch, “Ideological Warfare During the Cold War: The West’s Secret Book
Distribution Program Behind the Iron Curtain,” Foreign Policy Review, vol. 6, 160–
176, 2009.

[530] A. Resis, “Das Kapital Comes to Russia,” Slavic Review, vol. 29, no. 2, 219–237, 1970.

[531] L. Rheault and C. Cochrane, “Word Embeddings for the Analysis of Ideological Place-
ment in Parliamentary Corpora,” Political Analysis, vol. 28, no. 1, 112–133, 2020.

[532] M. Richter, “Begriffsgeschichte and the History of Ideas,” Journal of the History of
Ideas, vol. 48, no. 2, 247–263, 1987.

[533] C. Roberts, The Logic of Historical Explanation. Penn State Press, 1996.

[534] M. E. Roberts, B. M. Stewart, and E. M. Airoldi, “A Model of Text for Experimenta-

tion in the Social Sciences,” Journal of the American Statistical Association, vol. 111,
no. 515, 988–1003, 2016.

[535] W. C. Roberts, Marx’s Inferno: The Political Theory of Capital. Princeton: Princeton
University Press, 2016.

[536] W. C. Roberts, “What was primitive accumulation? Reconstructing the origin of a

critical concept,” European Journal of Political Theory, vol. 19, no. 4, 532–552, 2020.

[537] D. N. Rockmore, C. Fang, N. J. Foti, T. Ginsburg, and D. C. Krakauer, “The cul-

tural evolution of national constitutions,” Journal of the Association for Information
Science and Technology, vol. 69, no. 3, 483–494, 2018.

244
[538] J. E. Roemer, Analytical Foundations of Marxian Economic Theory. Cambridge Uni-
versity Press, 1981.

[539] J. E. Roemer, Analytical Marxism. Cambridge University Press, 1986.

[540] J. E. Roemer, Theories of Distributive Justice. Harvard University Press, 1996.

[541] E. M. Rogers, Diffusion of Innovations, 5th Edition. New York: Simon and Schuster,
1962.

[542] J. M. Rohrer, “Thinking Clearly About Correlations and Causation: Graphical Causal
Models for Observational Data,” Advances in Methods and Practices in Psychological
Science, vol. 1, no. 1, 27–42, 2018.

[543] J. Rojahn, “Marxismus - Marx - Geschichtswissenschaft: Der Fall der sog. „Ökonomisch-
Philosophischen Manuskripte aus dem Jahre 1844”,” International Review of Social
History, vol. 28, no. 1, 2–49, 1983.

[544] J. Rojahn, “Publishing Marx and Engels after 1989: The fate of the MEGA,” Critique:
Journal of Socialist Theory, 2009.

[545] M. Romanello and S. Hengchen, “Detecting Text Reuse with Passim,” Programming
Historian, 2021.

[546] M. A. Rose, Reading the Young Marx and Engels: Poetry, Parody, and the Censor.
Croom Helm, 1978.

[547] Z. Rosen, Bruno Bauer and Karl Marx: The Influence of Bruno Bauer on Marx’s
Thought. New York: Springer, 1977.

[548] J. Rosenthal, The Myth of Dialectics: Reinterpreting the Marx-Hegel Relation. New
York: Springer, 1998.

[549] J. Rosenthal and A. Yoon, “Judicial Ghostwriting: Authorship on the Supreme Court,”
Cornell Law Review, vol. 96, no. 6, 1307, 2011.

[550] J. Rosenthal, “The Escape from Hegel,” Science & Society, vol. 63, no. 3, 283–309,
1999.

[551] M. Rubel, Bibliographie des oeuvres de Karl Marx; avec en appendice un répertoire
des oeuvres de Friedrich Engels. Paris: M. Rivière, 1956.

[552] M. Rubel, “Les Cahiers de lecture de Karl Marx,” International Review of Social
History, vol. 2, no. 3, 392–420, 1957.

245
[553] M. Rubel, “Les Cahiers d’Etude de Karl Marx II. 1853–1856,” International Review
of Social History, vol. 5, no. 1, 39–76, 1960.

[554] M. Rubel, Marx Without Myth: A Chronological Study of His Life and Work. New
York: Harper & Row, 1975.

[555] M. Rubel, Marx: Life and Works. New York: Facts On File, 1980.

[556] M. Rubel, Rubel on Karl Marx: Five Essays. Cambridge: Cambridge University Press,
1981.

[557] Y. Rubner, C. Tomasi, and L. J. Guibas, “The Earth Mover’s Distance as a Metric for
Image Retrieval,” International Journal of Computer Vision, vol. 40, no. 2, 99–121,
2000.

[558] G. Rudé, The Crowd in the French Revolution. Oxford U.P. Paperback, 1959.

[559] G. Rudé, The Crowd in History: A Study of Popular Disturbances in France and
England, 1730-1848. Serif, 1964.

[560] S. Ruder, I. Vulić, and A. Søgaard, “A survey of cross-lingual word embedding mod-
els,” Journal of Artificial Intelligence Research, vol. 65, no. 1, 569–630, 2019.

[561] M. Rudolph and D. Blei, “Dynamic Embeddings for Language Evolution,” Proceedings
of the 2018 World Wide Web Conference, Republic and Canton of Geneva, Switzer-
land: International World Wide Web Conferences Steering Committee, 2018, 1003–
1011.

[562] A. Rule, J.-P. Cointet, and P. S. Bearman, “Lexical shifts, substantive changes, and
continuity in State of the Union discourse, 1790–2014,” Proceedings of the National
Academy of Sciences, vol. 112, no. 35, 10837–10844, 2015.

[563] J. Rybicki, “The great mystery of the (almost) invisible translator: Stylometry in
translation,” Quantitative Methods in Corpus-Based Translation Studies: A Practical
Guide to Descriptive Translation Research, M. P. Oakes and M. Ji, Eds., Amsterdam:
John Benjamins Publishing Company, 2012, 361.

[564] K. Saito, Karl Marx’s Ecosocialism: Capital, Nature, and the Unfinished Critique of
Political Economy. NYU Press, 2017.

[565] K. Sanyal, Rethinking Capitalist Development: Primitive Accumulation, Governmen-

tality and Post-Colonial Capitalism. Routledge, 2014.

[566] R. Sattelmeyer, Thoreau’s Reading: A Study in Intellectual History with Bibliograph-

ical Catalogue. Princeton University Press, 2014.

246
[567] F. S. Saunders, The Cultural Cold War: The CIA and the World of Arts and Letters.
New York: New Press, 1999.

[568] F. de Saussure, Course in General Linguistics. Open Court, 1916.

[569] R. A. Scalpino and C.-S. Lee, “The Origins of the Korean Communist Movement,”
The Journal of Asian Studies, vol. 20, no. 1, 9–31, 1960.

[570] W. Scheidel, The Great Leveler: Violence and the History of Inequality from the Stone
Age to the Twenty-First Century. Princeton: Princeton University Press, 2018.

[571] T. C. Schelling, Micromotives and Macrobehavior. W. W. Norton & Company, 2006.

[572] L. Schiebinger, I. Klinga, H. Y. Paik, I. Sánchez de Madariaga, M. Schraudner, and

M. Stefanick, “Machine Translation: Gendered Innovations,” Tech. Rep.

[573] A. M. Schlesinger, The Vital Center: The Politics of Freedom. Houghton Mifflin Com-
pany, 1949.

[574] B. Schmidt, “Stable random projection: Lightweight, general-purpose dimensionality

reduction for digitized libraries,” Journal of Cultural Analytics, vol. 3, no. 1, 11033,
2018.

[575] A. Schofield, M. ans Magnusson, and D. Mimno, “Pulling Out the Stops: Rethink-
ing Stopword Removal for Topic Models,” Proceedings of the 15th Conference of the
European Chapter of the Association for Computational Linguistics: Volume 2, Short
Papers, Valencia, Spain: Association for Computational Linguistics, 2017, 432–436.

[576] L. H. Schwartz, T. C. Helmus, D. Dasso Kaye, and N. Obeidat, Barriers to the

Broad Dissemination of Creative Works in the Arab World. Santa Monica, CA: Rand
National Defense Institute, Rand Corporation, 2009.

[577] J. W. Scott and D. Keates, Schools of Thought: Twenty-five Years of Interpretive

Social Science. Princeton University Press, 2001.

[578] J. E. Seigel, “Marx’s Early Development: Vocation, Rebellion, and Realism,” The
Journal of Interdisciplinary History, vol. 3, no. 3, 475–508, 1973.

[579] J. Seigel, Marx’s Fate: The Shape of a Life. University Park: Penn State Press, 1978.

[580] T. Shanin, Late Marx and the Russian Road: Marx and the Peripheries of Capitalism.
NYU Press, 1983.

[581] G. Shapiro and J. Markoff, Revolutionary Demands: A Content Analysis of the Cahiers
de Doléances of 1789. Stanford University Press, 1998.

247
[582] G. Shapiro, J. Markoff, and S. R. Duncan Baretta, “The Selective Transmission of
Historical Documents: The Case of the Parish Cahiers of 1789,” Histoire & Mesure,
vol. 2, no. 3, 115–172, 1987.

[583] S. Sherry, Discourses of Regulation and Resistance: Censoring Translation in the the
Stalin and Khrushchev Era Soviet Union. Edinburgh University Press, 2015.

[584] R. Shilliam, “Marx’s Path to Capital: The International Dimension of an Intellectual

Journey,” History of Political Thought, vol. 27, no. 2, 349–375, 2006.

[585] E. J. Sieyès, Qu’est-ce que le tiers état?: précédé de l’essai sur les privilèges. Paris: A.
Correard, 1789.

[586] E. J. Sieyès, What Is the Third Estate? New York: Pall Mall, 1789.

[587] A. Skinner, The Economics of James Steuart, 1st edition, R. Tortajada, Ed. Rout-
ledge, 2002.

[588] Q. Skinner, “The Limits of Historical Explanations,” Philosophy, vol. 41, no. 157,
199–215, 1966.

[589] Q. Skinner, “Meaning and Understanding in the History of Ideas,” History and The-
ory, vol. 8, no. 1, 3–53, 1969.

[590] Q. Skinner, The Foundations of Modern Political Thought: Volume 1, The Renais-
sance. Cambridge University Press, 1978.

[591] Q. Skinner, The Foundations of Modern Political Thought: Volume 2, The Age of
Reformation. Cambridge University Press, 1978.

[592] Q. Skinner, The Return of Grand Theory in the Human Sciences. Cambridge Univer-
sity Press, 1990.

[593] Q. Skinner, Liberty Before Liberalism. Cambridge University Press, 1998.

[594] Q. Skinner, Hobbes and Republican Liberty. Cambridge University Press, 2008.

[595] Q. Skinner, Visions of Politics: Volume 1, Regarding Method. Cambridge University

Press, 2012.

[596] N. A. Smith, “Contextual word representations: Putting words into computers,” Com-
munications of the ACM, vol. 63, no. 6, 66–74, 2020.

[597] A Soboul, “Problèmes théoriques de l’histoire de la Révolution française,” Aujourd’hui

l’histoire, Paris, 261–272, 1974.

248
[598] A. Soboul, The Sans-culottes: The Popular Movement and Revolutionary Govern-
ment, 1793-1794. Princeton University Press, 1965.

[599] A. Sokal and J. Bricmont, Fashionable Nonsense: Postmodern Intellectuals’ Abuse of

Science. Picador, 1997.

[600] L. Solan, The Language of Judges. 1993.

[601] S. Soni, L. F. Klein, and J. Eisenstein, “Abolitionist Networks: Modeling Language

Change in Nineteenth-Century Activist Newspapers,” Journal of Cultural Analytics,
vol. 6, no. 1, 18841, 2021.

[602] J. Spargo, “The Influence of Karl Marx on Contemporary Socialism,” American Jour-
nal of Sociology, vol. 16, no. 1, 21–40, 1910.

[603] D. Sperber, Explaining Culture: A Naturalistic Approach. Cambridge: Blackwell, 1996.

[604] J. Sperber, Karl Marx: A Nineteenth-Century Life. New York: Norton & Company,
2013.

[605] A. Spirling, “Democratization and linguistic complexity: The effect of franchise exten-
sion on parliamentary discourse, 1832–1915,” The Journal of Politics, vol. 78, no. 1,
120–136, 2016.

[606] W. Stafford, Socialism, Radicalism, and Nostalgia: Social Criticism in Britain, 1775-
1830. Cambridge: Cambridge University Press, 1987.

[607] J. V. Stalin, Marxism and Problems of Linguistics. Foreign Languages Press, 1950.

[608] E. Stamatatos, “A Survey of Modern Authorship Attribution Methods,” Journal of

the American Society for Information Science and Technology, vol. 60, no. 3, 538–556,
2009.

[609] G. Stedman Jones, The Young Hegelians, Marx and Engels (Chapter 17) - The Cam-
bridge History of Nineteenth-Century Political Thought, 2011.

[610] G. Stedman Jones, Karl Marx. Cambridge: Harvard University Press, 2016.

[611] G. Stedman Jones, “Malthus, Nineteenth-Century Socialism, and Marx,” The Histor-
ical Journal, vol. 63, no. 1, 91–106, 2020.

[612] K. Steenbrink, “Corpus Diplomaticum Neerlando-Indicum,” Christian-Muslim Rela-

tions 1500 - 1900, D. Thomas, Ed., 1907.

249
[613] G. P. Steenson, After Marx, Before Lenin: Marxism and Socialist Working-Class Par-
ties in Europe, 1884-1914. University of Pittsburgh Pre, 1991.

[614] Z. K. Stine, “Computational Methods for Comparative Analyses of Discourses,” Ph.D.

dissertation, University of Arkansas at Little Rock, United States – Arkansas, 2021.

[615] G. Storey and D. Mimno, “Like Two Pis in a Pod: Author Similarity Across Time in
the Ancient Greek Corpus,” Journal of Cultural Analytics, vol. 5, no. 2, 13680, 2020.

[616] L. Strauss, What Is Political Philosophy? And Other Studies. Chicago: University of
Chicago Press, 1959.

[617] D. J. Struik, Birth of the Communist Manifesto: With Full Text of the Manifesto,
All Prefaces by Marx and Engels, Early Drafts by Engels and Other Supplementary
Material. International Publishers, 1971.

[618] J. W. Stutje, Ernest Mandel: A Rebel’s Dream Deferred, trans. by C. B. a. P. Drucker.

Verso Books, 2009, 392.

[619] I. Susiluoto, The Origins and Development of Systems Thinking in the Soviet Union:
Political and Philosophical Controversies from Bogdanov and Bukharin to Present-
Day Reevaluations. Helsinki: Suomalainen tiedeakatemia, 1982.

[620] A. Swingewood, “Comte, Marx and Political Economy,” The Sociological Review,
vol. 18, no. 3, 335–349, 1970.

[621] J. L. Talmon, The Rise of Totalitarian Democracy. Beacon Press, 1952.

[622] A. Tamura, T. Watanabe, and E. Sumita, “Recurrent Neural Networks for Word
Alignment Model,” Proceedings of the 52nd Annual Meeting of the Association for
Computational Linguistics (Volume 1: Long Papers), Baltimore, Maryland: Associa-
tion for Computational Linguistics, 2014, 1470–1480.

[623] S. Tarrow, The Language of Contention: Revolutions in Words, 1688-2012. Cambridge

University Press, 2013.

[624] G. Teeple, “The Doctoral Dissertation of Karl Marx,” History of Political Thought,
vol. 11, no. 1, 81–118, 1990.

[625] M. Tercedor, “Cognates as lexical choices in translation: Interference in space-constrained

environments,” Target, vol. 22, 177–193, 2010.

[626] P. Thomas, Karl Marx and the Anarchists. London: Routledge, 1980.

250
[627] D. Thompson, Pessimism of the Intellect?: A History of the New Left Review. Merlin,
2007.

[628] W. H. W. Thompson, Z. Wojtowicz, and S. DeDeo, “Lévy Flights of the Collective

Imagination,” arXiv:1812.04013 [nlin, physics:physics], 2018.

[629] J. Tiedemann, “Bitext Alignment,” Synthesis Lectures on Human Language Tech-

nologies, vol. 4, no. 2, 1–165, 2011.

[630] C. Tilly, Contentious Performances. New York: Cambridge University Press, 2008.

[631] A. de Tocqueville, The Old Regime and the Revolution. Harper & brothers, 1856.

[632] J. E. Toews, Hegelianism: The Path Toward Dialectical Humanism, 1805-1841. Cam-
bridge University Press, 1985.

[633] R. S. Tomlin, Basic Word Order: Functional Principles. Routledge, 2013.

[634] G. Toury, “The Communication Situation and the Production of Interference Forms
by L2 Learners,” RELC Journal, vol. 13, no. 2, 62–77, 1982.

[635] K. Tribe, Land, Labour, and Economic Discourse. Routledge & K. Paul, 1978.

[636] K. Tribe, Genealogies of Capitalism. Springer, 1981.

[637] K. Tribe, “Locating German Economics: A Biographical Guide to the Study of the
History of German Economics,” German History, vol. 8, no. 1, 73–81, 1990.

[638] K. Tribe, Strategies of Economic Order: German Economic Discourse, 1750-1950.

Cambridge University Press, 2007.

[639] K. Tribe, The Economy of the Word: Language, History, and Economics. Oxford:
Oxford University Press, 2015.

[640] K. Tribe and H. Mizuta, A Critical Bibliography of Adam Smith. London: Pickering
& Chatto, 2014.

[641] R. C. Tucker, Philosophy & Myth in Karl Marx. New York: Cambridge University
Press, 1961.

[642] J. W. Tukey, Exploratory Data Analysis. New York: Addison-Wesley Publishing Com-
pany, 1977.

[643] J. Tully, Meaning and Context: Quentin Skinner and His Critics. Princeton University
Press, 1988.

251
[644] W. S. Turley, Vietnamese Communism In Comparative Perspective. Routledge, 2019.

[645] J. M. Twenge, W. K. Campbell, and B. Gentile, “Male and Female Pronoun Use in
U.S. Books Reflects Women’s Status, 1900–2008,” Sex Roles, vol. 67, no. 9, 488–493,
2012.

[646] H. Uchida, Marx’s ’Grundrisse’ and Hegel’s ’Logic’. London: Routledge, 2015.

[647] H. Uhl, Lorenz von Stein und Karl Marx: Zur Grundlegung von Gesellschaftsanalyse
und politischer Theorie 1842-1850. na, 1977.

[648] T. Underwood, “The Historical Significance of Textual Distances,” Proceedings of the

Second Joint SIGHUM Workshop on Computational Linguistics for Cultural Heritage,
Social Sciences, Humanities and Literature, Santa Fe, New Mexico: Association for
Computational Linguistics, 2018, 60–69.

[649] T. Underwood, Distant Horizons: Digital Evidence and Literary Change. University
of Chicago Press, 2019.

[650] T. Underwood, “Mapping the Latent Spaces of Culture,” 2021.

[651] T. Underwood and R. J. So, “Can We Map Culture?” Journal of Cultural Analytics,
vol. 6, no. 3, 24911, 2021.

[652] United Secretariat of the Fourth International, “New Revolutionary Party in Czechoslo-
vakia,” 885–888, 1969.

[653] N. Urbinati, Mill on Democracy: From the Athenian Polis to Representative Govern-
ment. University of Chicago Press, 2002.

[654] N. Urbinati and A. Zakaras, J.S. Mill’s Political Thought: A Bicentennial Reassess-
ment. Cambridge University Press, 2007.

[655] V. M. Uribe-Uran, “The Birth of a Public Sphere in Latin America during the Age
of Revolution,” Comparative Studies in Society and History, vol. 42, no. 2, 425–457,
2000.

[656] D. Varga et al., “Parallel corpora for medium density languages,” Recent Advances in
Natural Language Processing IV: Selected papers from RANLP 2005, 2007.

[657] M. E. W. Varnum and I. Grossmann, “Cultural Change: The How and the Why,”
Perspectives on Psychological Science, vol. 12, no. 6, 956–972, 2017.

[658] A. Vaswani et al., “Attention Is All You Need,” arXiv:1706.03762 [cs], 2017.

252
[659] F. Vatin, “Le travail, la servitude et la vie. Avant Marx et Polanyi, Eugène,” Revue
du MAUSS, vol. no 18, no. 2, 237–280, 2001.

[660] T. Veblen, “The Socialist Economics of Karl Marx and his Followers,” The Quarterly
Journal of Economics, vol. 20, no. 4, 575–595, 1906.

[661] T. Venclova, “Translations of World Literature and Political Censorship in Lithuania,”

Lituanus, vol. 25, no. 2, 1–9, 1979.

[662] M. Vendruscolo, E. Kussell, and E. Domany, “Recovery of protein structure from

contact maps,” Folding and Design, vol. 2, no. 5, 295–306, 1997.

[663] M. Vidal, T. Smith, T. Rotta, and P. Prew, The Oxford Handbook of Karl Marx.
Oxford: Oxford University Press, 2019.

[664] A. Vogel and D. Jurafsky, “He Said, She Said: Gender in the ACL Anthology,” Pro-
ceedings of the ACL-2012 Special Workshop on Rediscovering 50 Years of Discoveries,
Jeju Island, Korea: Association for Computational Linguistics, 2012, 33–41.

[665] B. von dem Berge and P. Obert, “Intraparty democracy in Central and Eastern Eu-
rope: Explaining change and stability from 1989 until 2011,” Party Politics, vol. 24,
no. 6, 652–662, 2018.

[666] T. Vonyó and A. Klein, “Why did socialist economies fail? The role of factor inputs
reconsidered,” The Economic History Review, vol. 72, no. 1, 317–345, 2019.

[667] P. Vranicki, Geschichte des Marxismus. Suhrkamp, 1983.

[668] V. S. Vygodsky, “Das Kapital,” The Great Soviet Encyclopedia, 1979.

[669] H. Wada, “Marx, Marxism and the Agrarian Question: II Marx and Revolutionary
Russia,” History Workshop Journal, vol. 12, no. 1, 129–150, 1981.

[670] J. M. Waller, The Third Current of Revolution: Inside the ’North American Front’ of
El Salvador’s Guerrilla War. University Press of America, 1991.

[671] J. M. Waller, Fighting the War of Ideas Like a Real War. Lulu.com, 2007.

[672] I. M. Wallerstein, World-Systems Analysis: An Introduction. Durham: Duke Univer-

sity Press, 2004.

[673] J. K.-C. Wang, “UN Voting on Chinese Representation,” 284, 1977.

253
[674] L. Wang, N. Beauchamp, S. Shugars, and K. Qin, “Winning on the Merits: The Joint
Effects of Content and Style on Debate Outcomes,” Transactions of the Association
for Computational Linguistics, vol. 5, 219–232, 2017.

[675] M. W. Wartofsky, Feuerbach. Cambridge: Cambridge University Press, 1977.

[676] a.-K. al Wataniyyah al-Muwahhidah, As.wāt al-intifād.ah al-Filast.īnīyah, 1987-1988.

Centre d’études et de documentation économique, juridique et sociale, 1991.

[677] C. Welch, J. K. Kummerfeld, V. Pérez-Rosas, and R. Mihalcea, “Exploring the Value

of Personalized Word Embeddings,” Proceedings of the 28th International Conference
on Computational Linguistics, International Committee on Computational Linguis-
tics, 2020, 6856–6862.

[678] R. Wellek, Confrontations. Princeton University Press, 1965.

[679] M. Wevers, J. Gao, and K. L. Nielbo, “Tracking the Consumption Junction: Temporal
Dependencies between Articles and Advertisements in Dutch Newspapers,” Digital
Humanities Quarterly, vol. 014, no. 2, 2020.

[680] M. Wevers and M. Koolen, “Digital begriffsgeschichte: Tracing semantic change using
word embeddings,” Historical Methods: A Journal of Quantitative and Interdisci-
plinary History, vol. 53, no. 4, 226–243, 2020.

[681] R. Whatmore, Republicanism and the French Revolution: An Intellectual History of

Jean-Baptiste Say’s Political Economy. Oxford: Oxford University Press, 2001.

[682] F. Wheen, Marx’s Das Kapital: A Biography. New York: Grove Press, 2008.

[683] J. White, Karl Marx and the Intellectual Origins of Dialectical Materialism. Springer,
1996.

[684] J. D. White, Marx and Russia: The Fate of a Doctrine. London: Bloomsbury Pub-
lishing, 2018.

[685] J. Wilkerson and A. Casas, “Large-Scale Computerized Text Analysis in Political

Science: Opportunities and Challenges,” Annual Review of Political Science, vol. 20,
no. 1, 529–544, 2017.

[686] J. Williamson, In Defence of Objective Bayesianism. Oxford: Oxford University Press,

2010.

[687] E. Wilson, To the Finland Station: A Study in the Writing and Acting of History.
New York Review of Books, 1940.

254
[688] S. Wintner, “Computational Approaches to Translation Studies,” 2017, 38–58.

[689] L. Wittgenstein, Tractatus Logico-Philosophicus: Centenary Edition. Anthem Press,

1921.

[690] L. Wittgenstein, Philosophical Investigations. Wiley, 1953.

[691] V. Wohl, “Plato Avant la Lettre: Authenticity in Plato’s Epistles,” Ramus, vol. 27,
no. 1, 60–93, 1998.

[692] B. Wolfe, An Ideology in Power: Reflections on the Russian Revolution. Routledge,

1969.

[693] B. D. Wolfe, “Lenin Has Trouble with Engels,” The Russian Review, vol. 15, no. 3,
196–209, 1956.

[694] A. W. Wood, Karl Marx. London: Taylor & Francis, 1999.

[695] D. Wood, Rethinking the Power of Maps. Guilford Press, 2010.

[696] E. M. Wood, Citizens to Lords: A Social History of Western Political Thought from
Antiquity to the Late Middle Ages. Verso Books, 2008.

[697] G. Woodcock, Pierre-Joseph Proudhon: A Biography. London: Routledge and Kegan

Paul, 1956.

[698] E. O. Wright, “The Shadow of Exploitation in Weber’s Class Analysis,” American

Sociological Review, vol. 67, no. 6, 832–853, 2002.

[699] J. K. Wright, Schoenberg, Wittgenstein and the Vienna Circle. Peter Lang, 2006.

[700] A. Wyner and W. Peters, “On rule extraction from regulations,” Legal Knowledge and
Information Systems, IOS Press, 2011, 113–122.

[701] D. Yang, J. Chen, Z. Yang, D. Jurafsky, and E. Hovy, “Let’s Make Your Request
More Persuasive: Modeling Persuasive Strategies via Semi-Supervised Neural Nets on
Crowdfunding Platforms,” 11, 2019.

[702] Y. Yang and F. Feng, “Language-Agnostic BERT Sentence Embedding,” Google AI

Blog, 2020.

[703] N. R. Yetman, “The background of the slave narrative collection,” American Quar-
terly, vol. 19, no. 3, 534–553, 1967.

255
[704] F. Zanettin, “‘The deadliest error’: Translation, international relations and the news
media,” The Translator, vol. 22, no. 3, 303–318, 2016.

[705] J. Zeng, J. Li, Y. He, C. Gao, M. Lyu, and I. King, “What Changed Your Mind: The
Roles of Dynamic Topics and Discourse in Argumentation Process,” Proceedings of
The Web Conference 2020, New York: Association for Computing Machinery, 2020,
1502–1513.

[706] N. Zeni, L. Mich, and J. Mylopoulos, “GaiusT 2.0: Evolution of a Framework for
Annotating Legal Documents,” Metadata and Semantics Research, E. Garoufallou, I.
Subirats Coll, A. Stellato, and J. Greenberg, Eds., Springer International Publishing,
2016, 43–54.

[707] Z. Zhao, T. Liu, S. Li, B. Li, and X. Du, “Ngram2vec: Learning Improved Word Rep-
resentations from Ngram Co-occurrence Statistics,” Proceedings of the 2017 Confer-
ence on Empirical Methods in Natural Language Processing, Copenhagen, Denmark:
Association for Computational Linguistics, 2017, 244–253.

256