Figure 2.1: Sherlock Holmes on another case.

When reports of the murder of Adolphe Steinheil and his mother-in-law, Mme Japy, broke in the Parisian press, the story was clearly regarded as a significant one. It was front page news for all nine newspapers (since La Croix didn’t publish on Sundays, there was no June 1, 1908 issue) and often had pride of place as the article in the upper-right-hand corner.

In this post, I examine patterns in this initial coverage to show how it spoke to mounting anxieties about criminality in Paris. Stories of crime can be disturbing and this double-murder certainly was for many. And yet, the language of the June 1 articles proposed that the double-murder was not particularly extraordinary and that the mystery was solvable. I’ll also talk about some of the distinctions among newspapers in how they reported on the case in its earliest days and how these differences reveal attitudes towards both media sensationalism and the supposed class-based character of the crime.

What was interesting?

Figure 2.2: The Most Common Words in Article Titles on June 1, 1908

(Voyant, developed by Stéfan Sinclair and Geoffrey Rockwell, generated this word cloud.)

If we look at a word cloud of the title text from all nine June 1 articles (admittedly, with a total of 243 words, a visualization isn’t necessary), it is clear that a lot of what was considered eye catching was who had been attacked, with headlines writing about the painter (“peintre”) Adolphe Steinheil, his mother-in-law (“belle-mère” in French, no doubt leading to the high use of both “belle” and “mère”) and his wife (“femme”), as well as the setting, the Steinheil hôtel off rue de Vaugirard. The Japy family was a prominent industrial family and Adolphe Steinheil wasn’t an unknown artist, but undoubtedly the real reason for this interest in the victims had to do with the fact that the sole survivor, Mme Steinheil, was notorious for her affairs with prominent men and had been Félix Faure’s mistress while he was president. Yet in June, newspapers were hesitant to mention the real source of Mme Steinheil’s fame or the exact nature of the relationship between her and Faure; instead, they spoke of the former president as a friend of both M. and Mme Steinheil. (We’ll talk more about the treatment of Faure in later posts.)

There is also a fair amount of emphasis on what happened – that it was a murder by strangulation (all the assassinats/assassins/assassinés [murders/murderers/murdered] and étranglé/étranglés [strangled]). Of course,  readers would need to be told why the Steinheil household was in the news. But these words also speak to a central feature of the press culture of the time: the fascination with stories of crime. In the first decade of the 1900s, Parisians were gripped by a panic about what felt like a rising tide of violent criminality, one that had a basis in reality but was also stoked by the mass press in an effort to capture readers in a highly competitive market. Readers also eagerly purchased works of crime fiction, such as those about Sherlock Holmes and Arsène Lupin. In other words, crime stories sold really well – and here was a dramatic tale about a double-murder in a household that many Parisians had heard about through circuits of rumor and gossip. Newspaper readers were likely to have been drawn to the initial coverage of the Steinheil Affair with this combined level of fascination and fear at a grisly murder that struck at the heart of France’s elite.

Conventions of crime and detection

To get a sense of the tone of the initial coverage of the affair, we looked at the adjectives being used in the newspapers. Focusing on this part of speech allows us to think about how people and places were being described and the moral and emotional categories of the early coverage. Because French adjectival forms vary with gender and number, we had to use lemmatizing scripts (a lemma is the dictionary form of a word). For example, this means that “beau,” “belle,” “beaux” and “belle” would all be understood as the same word. Then we used TreeTagger to tag words by their parts of speech and Python to get a list of the most common adjectives for each date across all articles.

Figure 2.3: The Most Common Adjectives in the June 1 Coverage

(The tagging scripts had some difficulties with what counts as for what part of speech and often included personal names as adjectives. These scripts were trained on Tree Tagger’s bank of words, so accuracy is a problem with doing this sort of thing automatically. This diagram cuts out all words tagged as adjectives that aren’t actually adjectives.)

Many of these commonly used adjectives were used to categorize individuals and/or places in basic ways – young (jeune), old (âgé), big (grand), small (petit), redheaded (roux), black (noir). They came from reporters’ physical descriptions of the house and its inhabitants as well as Mme Steinheil’s initial story of what the murderers looked like, as she claimed one of them was a red-headed woman and the other three were men wearing black robes. A few refer to the state’s investigation into the crime, such as the police’s service anthropométrique which kept records on the appearance of criminals or the médicin-légiste (medical examiner).

A conventional tragedy?

More significantly, the adjectives speak to a particular moral and emotional tone to the coverage. Many articles expressed a set of reactions to the news of the crime: sadness, fear and horror (tragique [tragic], effroyable [frightening], malheureux [sad], affreux [atrocious], horrible, épouvantable [appalling]). Indeed, all newspapers used the word “tragique” on June 1; all but L’Intrasigeant used “effroyable” and “malheureuse.” There was something very formulaic about these terms and my guess is that this was the language that journalists used to describe other crimes at the time and, to a degree, still do today. In an era in which members of the public were often expected to read more than one daily newspaper, they were thus encountering the same fundamental emotional and moral vocabulary. Although I don’t want to deny the degree to which readers and/or journalists were stirred by the news of the double-murder, the very conventionality of the language and the existence of clear formulas could also help individuals understand that what had just happened was not terribly remarkable – if it was distressing, it was also familiar.

Physicality, detection and knowledge

It is also worth noticing that many of the adjectives relate to physical qualities: gauche (left), vide (empty), mince (thin), longue (long), noir (black), large, ouvert (open), joli (pretty). Likewise, on June 2, roux (red-headed, 28 mentions), longue (21), large (13), ouvert (9), gros (big or fat, 8), osseux (bony, 5), and gris (gray, 5) are all common. And if we look at the most common verbs, many from June 1 relate to physical activity – coming, going, strangling, opening, putting, residing, murdering, going to bed, staying, tying, carrying, etc.

The most common verbs on June 1, 1908 and the number of times they are used:

(‘être’/’to be’, 701), (‘avoir’/’to have’, 673), (‘faire’/’to do’, 111), (‘trouver’/’to find’, 86), (‘pouvoir’/’to be able to’, 85), (‘dire/’to say”, 78), (‘venir’/’to come’, 60), (‘devoir’/’to have to’, 56), (‘aller’/’to go’, 41), (‘voir’/’to see’, 40), (‘donner’/’to give’, 40), (‘coucher’/’to go to bed’, 39), (‘ouvrir’/’to open’, 38), (‘prendre’/’to take’, 36), (‘étrangler’/’to strangle’, 36), (‘croire’/’to believe’, 33), (‘rendre’/’to give back’, 26), (‘passer’/’to pass’, 24), (‘savoir’/’to know’, 24), (‘sembler’/’to seem’, 23), (‘connaître’/’to know’, 21), (‘mourir’/’to die’, 21), (‘mettre’/’to put’, 21), (‘suffire’/’to be enough’, 21), (‘entrer’/’to enter’, 20), (‘habiter’/’to reside’, 19), (‘entendre’/’to hear’, 19), (‘attacher’/’to fasten’, 19), (‘occuper’/’to occupy’, 19),  (‘arriver’/’to arrive’, 18), (‘pénétrer’/’to enter, 18), (‘échapper’/’to escape’, 17), (‘permettre’/’to allow’, 17), (‘parler’/’to speak’, 16), (‘tuer’/’to kill’, 16), (‘étendre’/’to spread’, 16), (‘servir’/’to server’, 16), (‘rester’/’to stay’, 15), (‘exposer’/’to display’, 15), (‘demander’/’to ask’, 15), (‘vouloir’/’to want’, 15), (‘découvrir’/’to discover’, 15), (‘appeler’/’to call’, 14), (‘crier’/’to scream’, 14), (‘porter’/’to carry’, 14), (‘reconnaître’/’to recognize’, 14), (‘indiquer’/’to indicate’, 14), (‘laisser’/’to leave’, 13), (‘monter’/’to ascend’, 13), (‘tenir’/’to hold’, 13), (‘courir’/’to run’, 13), (‘posséder’/’to own’, 13), (‘falloir’/’to need’, 12), (‘apercevoir’/’to notice’, 12), (‘cacher’/’to hide’, 12), (‘paraître’/’to seem’, 12), (‘assassiner’/’to murder’, 12), (‘recevoir’/’to receive’, 12), (‘surprendre’/’to surprise’, 12), (‘rejoindre’/’to meet’, 12), (‘placer’/’to put’, 12), (‘commettre’/’to commit’, 11), (‘ligoter’/’to tie’, 11), (‘rentrer’/’to go back’, 11), (‘fermer’/’to close’, 11), (‘suivre’,/’to follow’ 11), (‘précipiter’/’to precipitate’, 11), (‘répondre’/’to answer’, 11), (‘poser’/’to place’, 11), (‘relever’/’to lift’, 11), (‘chercher’/’to search’, 11), (‘vivre’/’to live’, 11), (‘naître’/’to be born’, 11), (‘déclarer’/’to declare’, 11), (‘demeurer’/’to live’, 10), (‘juger’/’to judge’, 10), (‘remarquer’/’to notice’, 10), (‘désigner’/’to name’, 10), (‘introduire’/’to introduce’, 10), (‘procéder’/’to proceed’, 10)

To understand this emphasis on physicality, it’s helpful to think about the nature of the increasingly popular genres of detective/crime fiction in this era. How did Sherlock Holmes solve mysteries? Through the observation of the physical world. (For an example, you can read the first scene of “A Scandal in Bohemia” to see how Holmes deduces what Watson has been up to and who his client who wishes to remain anonymous actually is.)

So this language that stressed physicality – in terms of both action (verbs) and description of people and objects (adjectives) – helps us understand that on June 1, the story would have been read as similar to one from crime fiction. Hence, the solution to the double-murder lay with reading the physical world. This, too, provided some reassurance, for it suggested that the answer to the who-dun-it was easily knowable.

Thus, on June 1, the coverage of the affair cast it as a tragic and frightening event, but one that was within the bounds of normality and that was a mystery that was solvable through observation.

How similar were the articles to one another?

We can also ask whether newspapers were reporting on the case in the same ways. Clustering algorithms can help us here, as they can show us how similar the language was across articles. In particular, we used k-means clustering employing tf-idf algorithms on lemmatized versions of the texts.

Wait, what? K-means clustering? tf-idf?

Extra short version: this groups articles that are more linguistically similar closer together.

Longer version: In the words of Denis Tenen, k-means clustering, is “an unsupervised method of finding groups of similar documents within a large collection. The ‘unsupervised’ part means that we are looking for hidden structure without making any assumptions about the documents at the outset … We do not know what elements the algorithm will identify, only that it will make piles ‘typical’ of our corpus.” This video has a pretty clear explanation of k-means clustering and see here for Ben Schmidt’s thoughts on using k-means clustering on geographic data.

tf-idf stands for “term frequency-inverse document frequency” and is a way of categorizing documents according to how important a word is thought to be in the document. For example, if a word like “horrible” appears a lot in a document, it’s probably pretty important to the meaning of the text, right? But the problem is that computers don’t automatically know which frequently repeated words (“horrible,” “tragic”) are important to a text and which aren’t (“the,” “a”). tf-idf handles this by looking at how often a word is used in single text vs. an entire corpus. Or, “The idea of tf-idf is to find the important words for the content of each document by decreasing the weight for commonly used words and increasing the weight for words that are not used very much in a collection or corpus of documents… Calculating tf-idf attempts to find the words that are important (i.e., common) in a text, but not too common.” (Here’s more.)

Still confused? Read below for an example. Got it? Feel free to skip to the section titled “And now back to the Steinheil Affair.”

For the purposes of illustration, let’s take the following sentences as our corpus:

  1. In general, dogs are great and adorable.
  2. In general, cats are great and adorable.
  3. In general, hamsters are adorable and less great at cleaning.
  4. In general, to make vinaigrette, you need oil and vinegar.
  5. In general, a vinegar and water mixture is great at cleaning surfaces.

If we applied tf-idf algorithms and k-means clustering to these five sentences, a visualization might look something like this:

Figure 2.4: A Totally Made-Up Visualization

The words “in,” “general,” and “and” wouldn’t count for much, since they appear in all sentences in the corpus. Sentences 1 and 2 would be very close together, because with the exception of one word, they are exactly the same. They would be relatively close to sentence 3 (that “less” would be weighted as relatively important, since it only appears once in the whole corpus), but pretty far from sentence 4; apart from the words that appear in all sentences, there is no linguistic overlap. Sentence 5 would be somewhere between sentence 5 and sentences 1/2/3, since it contains “vinegar” like sentence 4, but also “great” (like sentences 1, 2 and 3) and “cleaning” (like sentence 3).

And now back to the Steinheil Affair

In terms of our corpus (i.e., those newspaper articles and not some made-up sentences), the powers of k-means clustering and tf-idf combined allow us to group articles according to how linguistically similar they are based on the words that are important and unique in them.

We pre-processed the texts before classifying them by lemmatizing them and then filtering for particular parts of speech — in this case, adjectives to help us think about how reporters were describing the crime, the locations and the participants. For the purposes of comparison, we also included articles from two other turning points in the affair: November 27, 1908 (which reported on Mme Steinheil being thrown into prison) and November 4, 1909 (the coverage of the first day of her trial).

Figure 2.5: Articles from June 1, 1908, November 27, 1908 and November 4, 1909, clustered by adjectives

How to interpret this: the letters refer to the newspaper (c = Croix, f = Figaro, h = Humanité, i = Intransigeant, j = Journal, m = Matin, pj = Petit Journal, pp = Petit Parisien, r = Radical, t = Temps). The numbers refer to the dates of the articles (1 = June 1, 1908; 2 = November 27, 1908; 3 = November 4, 1909).

Note: This is a two-dimensional representation of what, in fact, exists across many dimensions, as the documents have such a rich adjectival vocabulary. In other iterations of the clustering, the exact same documents appear in somewhat different places. The colors of the dots (i.e., articles) are also significant, as they are one way of indicating which documents go together.

The adjectives in these documents cluster across clear chronological lines, with any one article being more like other articles from the same date than from the same newspaper. Among the June 1 articles, in particular, there is a very tight grouping from Le Figaro, Le Matin, Le Petit Journal, Le Petit Parisien, which in some iterations of this visualization also includes L’Intransigeant and in others includes Le Journal. This suggests that Le Figaro, Le Matin, Le Petit Journal and Le Petit Parisien were using a virtually indistinguishable set of adjectives in their coverage. (See below on L’Humanité and Le Temps as outliers on June 1.) In contrast, for the later dates of the affair (dots with a “2” or “3” label), there are newspapers whose coverage is linguistically very close, such as the grouping of Le Matin, Le Petit Journal and Le Petit Parisien for November 27, 1908, but we don’t see quite the same degree of similarity.

And if we look at how the articles cluster according to nouns – the “who” and the “what” being described – these three newspapers are again similar to one another on June 1.

Figure 2.6: Articles from June 1, 1908, November 27, 1908 and November 4, 1909, clustered by nouns

Fundamentally, these three major Parisian dailies were reporting on the affair in the same way, reinforcing the hypothesis that readers were getting essentially the same story across a number of newspapers.

Sensationalism and politics

In contrast, L’Humanité and Le Temps (and to a lesser extent Le Radical) were often using different language than other newspapers in their coverage. (This is something that we see a lot in the clustering visualizations.)

How might we account for the differences between the coverage of these three newspapers one the one hand and that of Le Matin, Le Petit Journal and Le Petit Parisien on the other, at least on June 1? It’s not the emotional vocabulary of words like “horrible” and “effroyable,” which occur in both sets of articles; both also described the Steinheil family and their house, as well as the actions of the police investigators. Instead, what is lacking in L’Humanité and Le Temps (and, to a degree, in Le Radical) is the physical description that Mme Steinheil provided of the criminals.

Figure 2.7: Mme Steinheil at her trial — a woman with a loose relationship to the truth

It’s worth nothing that at this point, Mme Steinheil’s account of the crime was by far the strangest part of the case. Plenty of readers could believe that members of a prominent family had been the victims of a violent crime, but the particulars of her eyewitness testimony strained credulity. She described the murderers (who initially entered the house to rob it) as a red-headed woman and three bearded men wearing long black robes and hats a bit like sombreros; at other times during the affair she would also give them masks and gloves that flared out at the elbows like those of musketeers. In other words, these men dressed like no one else in Paris did. Understandably, many questioned her account: was she just a hysterical woman unable to tell the truth? Was she hiding something? Or was there something more nefarious going on than just a robbery-turned-wrong? Given her relationship with Faure, many wondered if somehow this crime had some mysterious political motive – and the strange dress of the assailants could have suggested that its wasn’t a run-of-the-mill attack.

There are, I think two reasons why L’Humanité, Le Temps and Le Radical were reluctant to publish these details. The first has to do with L’Humanité and Le Temps’s attitude towards sensationalism and the degree to which Meg’s account may have seemed far too wild and unbelievable to be worthy of coverage for these two papers. L’Humanité, a socialist daily, wasn’t above using dramatic language in the context of this affair, but it was politically opposed to sensationalism. On November 27, 1908, what was probably the most riveting and wildest day of the affair, they published an opinion piece decrying the mass press’s reliance on sensationalism to sell copy, claiming that it was distracting workers from the real issues of the day (among which it listed as being taxes, salaries, strikes, and the threat of war). For them, sensationalism and/or the Steinheil Affair was the opiate of the masses. Le Temps didn’t have any such radical politics. Instead, it was known for being semi-official and for its coverage of foreign affairs. Tonally, too, it has been called “sober and austere” (see Kalifa, Encre et le sang, 25) and thus, for different reasons, disinclined to offer sensationalistic reporting.

The second has to do with politics and L’Humanité and Le Radical’s standing on the left. The details that Meg offered pointed to the murderers as being members of the lower classes; red hair, for instance, was a sign of an individual’s low morality and social standing at the time. The language they supposedly used when they spoke to her was working-class slang and so also placed them socially. It is thus quite possible that the reporters for these newspapers didn’t want to publish this account because they didn’t want to give credence to the belief that the criminals were lower class, especially given that Mme Steinheil’s telling of the story was too improbable to believe.

Coming up, I’ll discuss these issues of class and criminality in more detail by looking at the adjectives in the June 1908 articles. Stay tuned!

Reminder: here‘s more on the details of the Steinheil Affair and the newspapers. Bibliography and acknowledgements are here.

Image credits

Figure 2.1: George Arents Collection, The New York Public Library. “Sherlock Holmes.” New York Public Library Digital Collections. Accessed July 25, 2017.

Figure 2.7: Art and Picture Collection, The New York Public Library. “Aux Assises : Mme. Steinheil Et Remy Couillard.” New York Public Library Digital Collections. Accessed July 25, 2017.