The themes of the Mémoires
What do the Mémoires talk about, and in what proportions? To find out without reading them end to end, one questions their running summary: the 9,542 entries the Chéruel edition placed at the head of the 404 chapters — "Mort de Monsieur à Saint-Cloud", "Cabale contre M. de Vendôme", "Mon procès contre le maréchal de Luxembourg". Each entry is a short, self-contained unit of sense (nine words on average): the ideal mesh for counting.
Thirty-five themes, grouped into eight families, are looked for. A theme is not a word but a lexical field: "Death and mourning" gathers mort, obsèques, deuil, veuve, testament, tombeau and seven further forms; an entry counts for the theme as soon as one of them appears in it, once only however many words are found. The search ignores accents and case but respects word boundaries — mort is not triggered inside "président à mortier". The words searched are, of course, those of the French text.
7,380 entries out of 9,542 (77 %) carry at least one theme; 3,894 carry two or more — 41 % of the summary, and half of the themed entries. The themes overlap widely, then: this is by design, and the overlap can itself be read — each theme's card gives its nearest neighbours.
The table of themes
One table, two readings. By family, the eight families follow one another from the fullest to the thinnest; a family's total is the number of distinct entries it touches, always lower than the sum of its themes since these overlap. Every bar is drawn to the same scale, from one family to the next: "Spain" can be compared directly with "Death and mourning". By theme, the thirty-five themes form a table that can be sorted, filtered by family and searched.
Click a row to unfold its lexical field — each pattern with the number of entries it brings in, and in brackets those of which it is the only trigger for the theme —, its neighbouring themes and a link to the page of all its entries, each leading to its passage. The Profile column draws its spread over the twenty volumes.
What is measured here is the summary, not the text: the grid records the attention of the 1858 indexer as much as that of the memorialist — the themes he indexed sparingly (illness, love affairs) are under-represented compared with the narrative itself. The lexical fields are declared in data/themes.json and are in no way automatic: they are choices, open to revision. Some overlaps are built in — "marriage" and "birth" often name the same genealogical entry, "lawsuits" and "parlement" the same institution; and the courts of Europe do not answer the same question as the other families: they say where, when the others say what. The 2,153 entries no theme catches are mostly bare proper names ("Comte de Braine") or turns of phrase too singular for a lexical field.