Figure 6.1: A photograph of Mme Steinheil’s hand wearing a pearl ring — what broke the case open

Previous posts have discussed how the June 1908 coverage of the Steinheil Affair fit into conventional understanding of crime in terms of who criminals and their victims were (immoral members of the lower class and elegant members of the bourgeoisie, respectively) and in terms of the emphasis on physicality as a means for knowing the truth. This is the first in a series of posts that talks about how all this changed in November 1908 as the affair came to increasingly center of the figure of Mme Steinheil, her personality and her behavior. What we see in large measure is a transition from an understanding of the truth as emerging from physical observation of the world to a belief that truth was situated in the personalities and moral qualities of individuals. Fundamentally, this was a shift away from the language of scientific detection and/or crime to one of scandal.

Background – revelation upon revelation

First, it’s useful to provide the basic historical background of what happened to understand why the newspaper coverage changed. (Quick reminder: the general narrative of the affair and a list of the cast of characters can always be found here.) In June 1908, the Steinheil Affair remained in the news on a regular basis, with the authorities (including Joseph Leydet, the examining magistrate) frequently telling reporters that they were on the verge of arresting the perpetrators and denying any possibility that Mme Steinheil could have been involved in the double-murder. But soon enough, the case faded from view and many thought that the crime would remain just another unsolved mystery of Parisian life.

What happened next arose out of the interaction between journalists in search of a scoop and Mme Steinheil’s desire to save her reputation. Because so many had serious doubts about her account of the crime, rumors swirled about her involvement in the crime in Summer 1908. Indeed, one of her lovers, Maurice Borderel, stated that they couldn’t resume their affair her until it her innocence was clear. Given that he was wealthy and generous and that Mme Steinheil was cash-strapped, this was a difficult blow for her. Thus, at the end of October, she wrote a letter to Marcel Hutin, a journalist at L’Écho de Paris, telling him that the police had a new, promising lead; this effectively reopened the case. But three weeks later, after no arrest was forthcoming, Mme Steinheil, still desperate to clear her name, planted evidence on her valet, Rémy Couillard, by putting a ring that she claimed had been stolen on the night of the crime in his wallet. When she “discovered” the ring, Couillard became the chief suspect. However, after Le Matin reproduced a photograph of the ring, Mme Steinheil’s jeweler recognized it and on November 25 told the police that she had asked him to work on the ring a few weeks after the murder. Suddenly, it was clear that Mme Steinheil was lying about what had been stolen on the night of May 30/31 and had probably framed Couillard. (This was the subject of the November 26 newspaper articles.)

Figure 6.2: Le Matin was a pioneer in the use of photography in its newspapers.

On the night of November 25, Hutin and a journalist from Le Matin called on Mme Steinheil and pressed her to admit that she had planted evidence on her servant. She then stated that the killer was Alexandre Wolff, the son of her cook, Mariette Wolff. When she went to the police the next morning, Alexandre Wolff was promptly arrested – but as it turns out, he had an alibi for the night of the double-murder. What followed was a dramatic confrontation between Wolff, Mme Steinheil and Leydet. Eventually, she admitted that Wolff was innocent and Leydet had her arrested for aiding the criminals through her many lies and she was escorted to Saint-Lazare, Paris’s prison for women. Soon after, Leydet resigned when it was revealed that he had had social connections to the Steinheils before the murders. All of this – Mme Steinheil’s accusation of Wolff, his arrest, the courthouse confrontation, her arrest, and Leydet’s resignation — was the subject of the November 27 articles. At this point, if she wasn’t necessarily under official suspicion for the crime, it didn’t look good for her.

Figure 6.3: The courtyard of Saint-Lazare prison in the 1840s

In the wake of these dramatic and indeed shocking events, the journalistic language of the Steinheil Affair indicated that the case was now a scandal that centered on Mme Steinheil and her behavior. Correspondingly, the vocabulary reporters used to describe the Affair became less about physicality and more about morality.

A scandal emerges

During its early days, the Steinheil Affair wasn’t categorized as being a scandal. Neither this term nor “scandalous” were used in the June 1908 articles. Instead, they first appear in the November 1908 coverage.

Figure 6.4

In June, the affair was largely regarded as a run-of-the-mill crime – something involving the transgressions of the lower-classes — and therefore not exceptional or transgressive enough to be a scandal. But by November 1908, “scandal” and “scandalous” came into more common use in the coverage of the affair as it became understood as a tale of elite misbehavior.

The centrality of Mme Steinheil

Likewise, the story came to center around Mme Steinheil and beginning on November 27 her name appears far more frequently in the newspaper articles.

Figure 6.5

Word counts can tell us some things, but other methods of text analysis allow us to see what the articles were discussing on different days. This gets us to topic modeling.

A brief introduction to topic modeling

(If you are familiar with topic modeling, feel free to skip this section and go down to “The rise of Mme Steinheil’s misbehavior.”)

When we read, we can tell what something is about based on the types of words in it. The word “goal” could appear in any number of types of articles, but with the words “football” and “touchdown” around it we start to get a sense of what the topic is: sports. In the words of Miriam Posner, “topic modeling is a method for finding and tracing clusters of words (called ‘topics’ in shorthand) in large bodies of texts.” It relies on a lot of complicated statistics, but fundamentally it is based on the assumption that a single document has a number of different topics or themes made up of these clusters of words and that a corpus will have documents with a bunch of different themes. We can then distinguish among documents by looking at their different themes (or topics) if we find these co-occurring words and see how they appear across the corpus. Topic modeling has thus been heralded as a way to make sense out of large corpora, as well as uncover latent themes, one that a reader might not even know were there.

As an imaginary example, let’s say we took a corpus of 1000 documents, ran topic modeling algorithms on it and found that two of the topics were as follows:

Topic A: stir, mix, measure, flour, sugar, butter

Topic B: sit, stay, fetch, treat, good, dog

We could guess that Topic A was probably related to cooking and Topic B was probably related to dogs. Additionally, topic modeling would help us understand how much of a particular document was drawn from the words in a topic. So if the topic modeling algorithms told us that 85% or so of Document 1 consisted of words from Topic A and 5% were from Topic B, we would know that Document 1 had a lot to do with cooking but not a lot to do with dogs. Likewise, if Document 2 was 40% from Topic A and 40% from Topic B, we might hazard a guess that the book was about cooking for your dog.

It’s important to remember that topic modeling is probabilistic. As a consequence, these percentages aren’t the last word on how much of a given text is drawn from a particular topic and running topic modeling on any one corpus can give different results each time. Typically, too, we exclude commonly used words (“a,” “an,” “the,” etc.) so that we can see the words that are more unique to a corpus as opposed to likely to be in pretty much every document; the list of these words are often called a stoplist. Because language usage shifts so much over time, topic modeling also runs into trouble when working with documents from different eras or long spans of time.

Want to know more? Here’s a guide written for undergraduates; this and this are intended for those with more of a background in digital humanities but are still understandable for those without a lot of training in statistics. This video goes into the math behind topic modeling.

The rise of Mme Steinheil’s misbehavior

Although our corpus of articles on the Steinheil Affair is relatively small for topic modeling, we can still see fairly important shifts over time.  Here is a list of six topics that emerged in one iteration of topic modeling, with the bolded text being how we understood the meaning of the topic:

the story of the pearl ring: perle (pearl), Couillard, Rémy, Leydet, bague (ring), Souloy, portefeuille (wallet), bijoutier (jeweler), Palais (palace, as in Palais de Justice or courthouse), juge (judge)
Mme Steinheil and her relationships: André (the new examining magistrate after Leydet resigned), comte (count), villa, Arlon (the comte d’Arlon was a friend of the Steinheils), Japy (Mme Steinheil’s maiden name), juge, relations, vert (green; the Steinheil’s country villa was named “Vert-Logis”), magistrat (magistrate), Rémy
the Steinheil family and house: Steinheil, Mme, crime, femme (woman), heures (hours), mère (mother), mari (husband), impasse, Couillard, hier (yesterday)
the initial story of the crime: chambre (bedroom), peintre (painter), assassins (murderers), Japy, impasse, lit (bed), maison (house), Hamard (the chief of the detective brigade), porte (door), femme
the confrontation at the courthouse: Wolff, Alexandre, Sûreté (the detective brigade), Leydet, Hamard, Mariette, assassin (murderer), Rémy, fils (son), chef (head)
Mme Steinheil’s affairs: Borderel, peintre, André, villa, Prévost (the name of the woman in whose name Mme Steinheil rented her country house), chambre (room), logis, Hamard, autopsie (autopsy), vert

Here is the frequency of these six topics charted over time:

Figure 6.6

(Notes: This visualization excluded topics that were prominent during the coverage of the trial in November 1909, as we wanted to concentrate on the shift from June 1908 to November 1908. To get the percentage of words, you would multiply the number on the y-axis by 100. Thus, for example, the topic labeled “the initial account” is between 40% and 60% of the words in the June 2 articles.)

Over the course of the affair, the Steinheil household remained a topic of  discussion fairly consistently. But in November 1908, the initial account of the crime became less prominent. Instead, the story of the pearl ring and the confrontation between Mme Steinheil, Leydet and Alexandre Wolff successively came into view; on November 28, the coverage was increasingly about Mme Steinheil’s social relationships and/or extramarital affairs.

It’s notable that what was particularly prominent in June 1908 were words about the physical, such as locations or objects. In contrast, the November 1908 topics are fundamentally about Mme Steinheil’s lies (her accusations of others) or her sexual behavior, suggesting that the story was increasingly about her misdeeds.

From physical to moral

Other ways of examining the coverage show this same shift in November away from physicality  to a concentration on persons and their moral qualities. This is significant because tales of elite immorality are usually at the heart of any scandal. Additionally, this suggests that observers were coming to understand that determining the truth of the matter would be far harder than was previously thought.

Crime vs. affaire, place vs. persons

How did people refer to this case? Was this a “crime”? Or an “affaire”? Drawing too sharp a line between these terms would be unwise. After all, the case was always about a double-murder and the term “affaire” was used for criminal cases in general. But “affaire” also means “matter” or “business” and any scandal would be referred to as an “affaire” (in this sense, it’s not completely dissimilar from how the suffix “-gate” gets attached to any American political scandal). The fact that the case is increasingly categorized as an “affaire” in November 1908 (particularly starting on November 27) is a sign of increased public interest and the sense that this was a big deal and no mere ordinary crime.

Figure 6.7

Collocations (what words occur near each other) allow us to see another instance of the shift from the story being about physicality to one about persons. If we use AntConc to look at what words are either five to the left or right of “crime” in June 1908, notable collocations include “Steinheil” (7 times), but also “Vaugirard” (6 times – the Steinheil home was just off rue de Vaugirard) and “maison” or house (12 times). Hence, the affair was understood as being about a household, but also a place. In contrast, on November 27 and 28, the collocations for “crime” are somewhat different: Ronsin (39 times – the Steinheil home was in impasse Ronsin) and “maison” (7 times), but also “Steinheil” (27 times), “Mme” (22 times) and “elle” or she (19 times). Clearly, the case was being seen increasingly about Mme Steinheil.

Even the shift away from “Vaugirard” to “Ronsin” is significant. Both locate the crime in physical spaces, of course. But while rue de Vaugirard is a long street that cuts through much of Paris’s Left Bank, impasse Ronsin is a tiny cul-de-sac that few Parisians would have heard of before. The fact that it was being called “le crime de l’impasse Ronsin” as opposed to being located on the rue de Vaugirard indicated that by November 1908, newspaper readers were expected to know the basic details of the case and to be following it on a regular basis.

Figure 6.8: Detail from a 1944 map of Paris. The arrow shows where impasse Ronsin is.

Truth, doubt and emotion

This same shift away from physicality is also visible if we look at the most common adjectives and verbs in the newspaper coverage of November 27. (The process of getting the most common parts of speech for a particular day is discussed more fully in this post.) First, the adjectives:

Figure 6.9: The most common adjectives from the November 27, 1908 coverage

As on June 1, many of the most common adjectives are used to categorize people or things in basic ways – nouveau (new), petit (small), grand (big). And while there are a number that refer to a person’s appearance (roux [redheaded], blond), the adjectival vocabulary shows much less emphasis on physical description than it did on the first day of coverage. Instead, what tends to be stressed are questions about truth and guilt/innocence, such as innocent, coupable [guilty], vrai [true], véritable [true], sûr [sure] and certain. There was also a stress on intense – often negative – emotions, with words like horrible, terrible, affreux [atrocious], effroyable [frightening] and tragique [tragic], all of which had been used in June. But now there is also émouvant [moving], pénible [painful], vif [excessive] and violent (all but one usage referred to violent emotions as opposed to physical violence).These terms established the day’s events as particularly dramatic (another word that appears a fair amount) and emotional.

We get a similar story if we look at the most common verbs from November 27. (Note that this diagram cuts out “avoir” and “être,” with 1953 and 1910 uses, respectively. They were used so frequently that they would overwhelm the rest of the data without telling us that much.) Again, there are verbs that have to do with physical action (coucher [go to bed], porter [bring], aller [go], venir [come]), but many of them have to do with speech – dire [say], raconter [tell], répondre [answer], accuser [accuse], écrier [exclaim], parler [speak], mentir [lie], interroger [interrogate], déclarer [declare], demander [ask], crier [scream], affirmer [affirm].

Figure 6.10: The most common verbs from the November 27, 1908 coverage

This shift away from the language of physicality to one about speaking, emotions and the innocence/guilt means that we are no longer in the realm of crime fiction and scientific detection. The truth isn’t knowable through looking for clues and observing the physical world. Instead, the public would have to rely on what individuals said to know what happened on the night of May 30/31. And that prospect offered very little reassurance. Mme Steinheil was the only known witness, but she was also liar who had told many different, equally false, versions of the night of the crime. Any hope of knowing the truth would require making a judgement about her character. But perhaps the public might never know who did it — or who she really was.


Lastly, to think about the shift from a physical to a moral vocabulary, we can look at the use of the word “dessous.” This isn’t a term that appears in the corpus with great frequency (a grand total of 14 times) but it still reveals a lot about the nature of the affair. “Dessous” can be a preposition, an adverb or a noun and, according to how it is used, can have quite different meanings. It can mean “below,” but as a plural noun (“les dessous”) it refers to secrets or hidden depths and truths.

In June 1908, it was used twice and in both cases to refer to physical location, as in the following sentence from June 1, 1908 in Le Petit Parisien: “Le premier étage est occupé par les chambres à coucher des époux Steinheil et de Mme Japy, belle-mère de l’artiste, ainsi qu’on peut s’en rendre compte par le plan que nous publions ci-dessous.” (“The first floor is occupied by the rooms of M. and Mme Steinheil and Mme Japy, the artist’s mother-in-law, as can be seen from the plan which is published below.”) However, in November 1908, only one of the eight uses is as a preposition; the remaining seven mean secrets, as if to suggest that the true nature of the case was coming to light. Thus, on November 27, L’Humanité wrote: “On annonçait aussi que le Garde des Sceaux serait interpellé sur les dessous de cette affaire, et la façon fantaisiste dont l’instruction a été conduite.” (“It was announced that the Minister of Justice would be questioned on the dessous of this case, and the wild manner in which the investigation was conducted.”) Likewise, on the same day, Le Petit Parisien stated: “Ainsi tout l’échafaudage de monstrueux mensonges édifié par la veuve du peintre s’est écroulé. L’heure de la justice est arrivée : nous ne tarderons pas à connaître tous les dessous du drame.” (“Thus the scaffold of the monstrous lies built up by the painter’s widow collapsed. The hour of justice has arrived: we shall soon know all the dessous of the drama.”)

The notion of the truth as something that lay below is encapsulated in the proverb “la vérité est au fond du puits” (truth is at the bottom of a well), suggesting that the true nature of things is hidden in darkness, almost impossible to see (or know). Ten years earlier, this expression had been used to dramatic effect in Édouard-Debat Ponsan’s allegorical painting of the Dreyfus Affair, wherein the truth, a semi-naked woman, is seen emerging from a well while being held back by figures representing the clergy and military. Thus, in November 1908, journalists were suggesting that the truth was emerging from the depths, just becoming visible.

Figure 6.11: Édouard Debat-Ponsan’s Truth Leaving the Well, 1898

So from just looking at how one word was used, we can see how journalists were claiming that dramatic secrets of elite misdeeds were being uncovered. The Steinheil Affair was now a full-blown scandal.

Image credits

6.1:  courtesy of Wikimedia Commons, {{PD-1923}}, available at,Main_de_Mme_Steinheil,_Paris,_1909.jpg

6.2: courtesy of Wikimedia Commons, {{PD-1923}}, available at

6.3: The Miriam and Ira D. Wallach Division of Art, Prints and Photographs: Print Collection, The New York Public Library. “Cour de Saint Lazare.” New York Public Library Digital Collections. Accessed August 10, 2017. 

6.8:  U.S. Army Map Service, Paris 1944, from the University of Texas at Austin, available at,_Paris_1944_-_The_University_of_Texas_at_Austin.jpg

6.11: courtesy of Wikimedia Commons, {{PD-1923}}, available at