This article offers some solutions to the limits of full-text search in Himanis provided by the exploration of indexation mistakes. It is based on my ongoing work on witchcraft in the Trésor des Chartes (French royal archives). The Trésor des Chartes contains the royal archives beginning with Philippe the Fair.
Research and current status of research
In the current status of my research, the study of the cases of witchcraft recorded in the JJ series of the Trésor des Chartes (medieval registers of the French royal chancery) was carried out partially and was not the object of a systematic survey within this enormous corpus.
In the registers of the Chancery, these cases are mainly met in the forms of letters of remission for individuals charged with witchcraft or for individuals charged with the murder of a man or a woman suspected to be a sorcerer or a sorceress.
In the context of my PhD research, which is focused on the repression of witchcraft in the Kingdom of France in the late Middle Ages, I am attempting to conduct this investigation, particularly by providing editions of these unstudied or unedited letters of remission that concern cases of witchcraft.
From the preliminary study on this question, an early inventory of these cases of witchcraft in the Trésor des Chartes was released mostly thanks to historiography. It was established from the analytical inventories of the JJ series and from the founding work of Pierre Braun1 and Claude Gauvard2. It enabled the identification of 90 acts contained in the registers of the Chancery reporting cases of witchcraft.
Subsequently, the Himanis project would be a decisive tool for my research within the archives of the Trésor des Chartes. First, it offered the possibility to confirm the existence of the corpus of recorded cases of witchcraft. Above all, it allowed me to uncover new letters of remission that remained unknown to research especially because the acts that mention cases of witchcraft within the registers of the Royal Chancery are easily distinguished by the use of a proper, stable and recurrent lexicon. They outline from the 14th Century, in French as in Latin, the contours of a specific criminal typology and they were, in all probability, likely to be identified by Himanis.
Nevertheless, online searches of Himanis for the key-words that are largely present in the letters of remission dealing with facts of witchcraft such as sorcerie (Old French for “witchcraft”), sorcier or sorciere (“sorcerer”or “sorceress”) or ensorcellé (“bewitched”) proved unsuccessful, even when the selected level of “confidence” in full-text mode was set on very low.
Under these circumstances, facing the limits of full-text searching for the correct words,I accessed the “dump” of the index, that is to say a file containing a (quasi)complete extraction of the index of the Himanis platform. It is a list of words with the number of occurrences and their average probability of appearance. It is a rather large file (1.19 GB, less than 300 MB once zipped) and the extraction deletes a certain amount of words with a low probability of appearance. This index is currently in private mode because the partners of the Himanis project are working on a re-indexation of better quality in order to publish the whole data released under cc0 public domain license. The analysis of the whole index and the exploration of the indexation mistakes proved important and enabled me to find an effective solution to the shortcomings revealed by the first results.
Within the registers of the Trésor des Chartes, the direct observation of the forms of the correct words of the lexicon of witchcraft recognized by Himanis (such as “sorcerie”,“ensorcellé” or “ensorcellement” which stands for “spell” in Old French) lead to the following conclusion: in the vast majority of cases, Himanis’ recognition system performs either a faulty reading of the correct word (substituting for example the –s- for –f-, the –c- for –t-), a shortened reading of a longer word (“sorcelle” instead of ensorcellé), or a faulty and shortened reading of the correct words (“sortelle” for ensorcellemment).
In light of those observations, the consultation of the entirety of the forms recognized by Himanis became the only way, through the mistakes of the indexation systems, to access the acts of interest for my research. These forms, accessible in the complete index file (“dump”), offer the possibility to find both the correct words recognized by Himanis and the reading mistakes, with incorrect or unexpected forms of a common lexeme and have access to all the lemmata resulting from them and particularly the pseudo–lemmata. The principal objective is to know all the possible forms of reading of the correct words belonging to the lexicon of witchcraft, with their frequency of recognition.
In this way, in order to compare the forms recognized by Himanis with the correct or searched forms, I established two protocols for the query of Himanis’ online platform from the “dump” file. They can be summed up as follows:
- Note in the “dump” file all the occurrences of faulty lexemes (in my case, –forc– instead of –sorc– or –sort– instead of –sorc-) in order to assemble a list of lemmata or pseudo–lemmata corresponding to either faulty readings, shortened readings or faulty and shortened readings of the correct words.
- Note in each act the recognition of a correct word made by Himanis (for instance, “sortele” for ensorcelé) and look for forms in the “dump” file that are related, faulty or shortened (« sortile », « fortele »,« forcele », « fortile », etc.).
For example, in the first case, from the faulty lexeme –fort-, it is possible to query the “dump” file” to establish a list of the lemmata it contains and that are likely to correspond to the correct words belonging to the witchcraft lexicon: these lemmata or pseudo-lemmata are selected according to their morphological similarity to the correct lemmata (For instance, « forterie »,« forcerie » or « forteries » for sorcerie).
Once this list is established, the online platform can be queried, the percentage of “confidence” being varied during the process. Therefore, every form recognized by Himanis is identified in order to select the ones corresponding, ipso facto, to the desired search terms.
This approach can be applied to any word search, known or presumed to be present in an act of the Chancery, the only limit being that the forms are recognized and indexed by Himanis.
In order to give some concrete examples of the different modalities of word searches on the online platform, the diagram below gathers, around the lexeme –sorc–, the different recognition possibilities of lemmata related to the history of witchcraft occurring in the registers of the Chancery from the forms observed in the “dump”. Each word in blue (form appearing in the “dump”) corresponds to the mistakenly recognized form of a correct form. A link in the notes below enables one to redirect to the digitalized folio where this form has been recognized.
Results and provisional conclusion
From a quantitative point of view, the queries made in Himanis, on the sole basis of the digitalized registers of the Trésor des Chartes (up to JJ 211), enabled me to discover 29 previously unstudied letters of remission reporting cases of witchcraft. It represents a contribution to the initially known corpus of almost 30%.
This figure, which is hopefully provisional because it represents the current state of my investigations on Himanis, highlights the importance of the exploration of the “dump” file. Indeed, all of the aforementioned acts were discovered thanks to the exploration of indexation mistakes.
Moreover, among the acts that constituted my initial corpus of letters of remission, fewer than 20 of them could have been identified by using different search terms in Himanis’ database, whether by using the correct words or by using the alternative forms found in the “dump” file.
Therefore, the exploration of indexation mistakes constituted a real addition to the knowledge of cases of witchcraft in the Trésor des Chartes. In itself, it provided some practical solutions to the limits of full-text searching. It proves nevertheless that, unfortunately, the “recollection” remains low and that, even when using the integral index, the results of the first Himanis’ indexation do not enable an exhaustive search as the gap of accuracy between the recognized words and the correct words remains high.
Ultimately, this exploration of mistakes in the indexation system should not be limited only to the field of the history of witchcraft and its lexicon. Actually, this single experience highlights the importance of the exploration of the complete index file (“dump”) for other research fields until progress is made in terms of HTR3 and post-HTR correction.
* * *
« sortitre » / sorciere: Arch. nat, JJ 199, n° 441, fol. 276r-276v : Remission of Jehan Sommet, notary in Thiers (Auvergne), for assault, battery and torture occasioning the death of Guillemette de Pigeroles, alleged witch (September 1464).
« forterie » / sorcerie : Arch. nat, JJ 190, n° 98, fol. 53r-53v : Remission of Antoine Marost for assault and battery occasioning the death of man by the name of Perreau, alleged sorcerer, in the province of Lyonnais (May 1460).
« forterie » / sorceries : Arch. nat, JJ 109, n° 39, fol. 22r-22v : Remission of Berthelemin Caillon, ploughman in the bailiwick of Mâcon, for assault and battery occasioning the death of a man by the name of Raymond,“alleged and defamed of making charms, spells and witchcraft” (« renommé et diffamé de faire charmes, charoiz et sorceries »), (June 1376).
« sortele » / ensorcelé : Arch. nat, JJ 181, n° 123, fol. 67v-68r : Remission of Guillaume Moler for assault and battery occasioning the death of Beguite Cloete, alleged witch, in the parish of Saint-Ferréol (Saint-Ferréol-des-Côtes) in Auvergne (September 1452).
« sortelle » / ensorcellé : Arch. nat, JJ 199, n° 288, fol. 144r : Remission of Pierre Jamet for assault and battery occasioning the death of a woman by the name of Gaudronne, wife of Jehan Gaudront, in the bailiwick of Montargis (avril 1459).
« sorcelle » / ensorcellé : Arch. nat, JJ 138, n° 126, fol. 11r : Remission of Jehan Creton, Substitute for the Royal Prosecutor to the bailiwick of Rue (Located in the modern-day arrondissement of Abbeville in the Somme département) for possession of stolen information unduly collected against the wife of Pierre de la Varenne which established her reputation of being a witch (March 1489).
« enfortele » / ensorcellée : Arch. nat. JJ116, n° 147, fol. 86v-87r : Remission of Pierre brother Jehan, former bailiff of Houdan, who was charged by the Royal Prosecutor to the bailiwick of Montfort-l’Amaury for torture to which Perrette, Robin Chenoult’s wife, was subjected because he suspected her of being a witch and responsible of his daughter’s disease.
« sortelle » / ensorcellement : Arch. nat, JJ 208, n° 20, fol. 11r-12r, Remission of Jehan Lutier for assault, battery and torture occasioning in the death of Laurent Berge, alleged sorcerer, in Palluau-sur-Indre (December 1480).
1. P. Braun, « La sorcellerie dans les lettres de rémission du Trésor des Chartes », dans 102ème Congrès national des sociétés savantes (Limoges 1977), Section de philologie et d’histoire jusqu’à 1610, Paris, 1979, t. II, p.257-278.
2. C. Gauvard, « De grace especial ». Crime, État et Société en France à la fin du MoyenÂge, Paris, 1991.
3. HTR stands for Handwritten Text Recognition, especially given the new perspectives related to the collaboration with the READ (Recognition and Enrichment of Archival Documents) project. Cf. Dominique Stutzmann’s post on the topic.