Authors
Julien Longhi University Professor of Linguistics, CY Cergy Paris University
Alexandra Freeman Ph.D. candidate in linguistics, CY Cergy Paris University
Cécile Jacob Handwriting Comparison Specialist, Criminal Research Institute of the National Gendarmerie (IRCGN)
The issue of authorship is not a new one, particularly in the field of literary studies. It has been widely debated and has been the subject of renewed debate since the rise of digital technologies) and is certainly relevant today, particularly in the field of literary studies. Currently, the DIM Textual Studies and New Knowledgeis leading a project (directed by Pierre Glaudes and Catherine Mayaux) on the critical works of Barbey d’Aurevilly (PRADA project: Research Project on Author Attribution).
One of the best-known controversies in France concerns the work of Dominique Labbé, a researcher specializing in textual statistics, who has published, among other things, If two plus two equals four, Molière did not write *Don Juan*, or the article « Corneille in Molière’s Shadow: How Do We Identify an Author? ».
According to the Wikipedia page on the authorship of Molière's works(and more specifically the section on the’A Statistical Study by Cyril and Dominique Labbé), Dominique Labbé would have concluded that Corneille was the author of sixteen of the eighteen comedies attributed to Molière.
More recently, two researchers, Florian Cafiero and Jean-Baptiste Camps, published an article titled «Why Molière Most Likely Did Write His Plays» in the journal Science Advances. This article generated significant buzz within the community and in the media, so much so that France Culture, for example, ran an article titled «Two researchers prove that Corneille did not write Molière's plays»). The conclusion of the article is indeed unequivocal: «It is very unlikely that P. Corneille or his brother Thomas would have been Molière’s ghostwriters. Since they were, after a century-long debate, the only option deemed plausible, these conclusions strongly support the idea that Molière did indeed write his own plays.» («It is highly unlikely that P. Corneille or his brother Thomas were Molière’s ghostwriters. Since they were, according to a 100-year-old debate, the only other plausible authors, these conclusions seem to support the idea that Molière is indeed the author of his works.»).
But the controversy doesn’t end there, since following that article, Dominique Labbé published a new piece titled «Response to Florian Cafiero and Jean-Baptiste Camps. Why Molière Most Likely Did Write His Plays,» in which he questions the methods used by Cafiero and Camps.
The debate over the authorship of these works is therefore not yet resolved, but it may be of interest to us, since the methods and tools used in this context can be useful in other areas of society. Furthermore, the issue of attributing authorship to a text is no longer limited to the literary field. In particular, this area of research, combined with new techniques and the development of specialized studies, can be adapted to the needs of the judicial system.
From Linguistics to Forensic Science
Forensic linguistics is a field of research within applied linguistics dedicated to the needs of the justice system. It studies language used in various contexts, including the language of normative texts (laws, codes, regulations), the language of legal proceedings (court transcripts, pleadings, jury trials), and language as a trace or as evidence. The latter includes, in particular, language samples resulting from «language-related offenses» (such as threats, harassment, plagiarism, defamation, or perjury) as well as any other language samples presented as traces or evidence. Well known in common law countries, where these techniques have been used since the 1960s (Jan Svartvik, 1968), forensic linguistics is gaining more and more ground in France.
Style as a Source of Variation
Language production, like all communication, is a social phenomenon that adapts to the context in which it occurs. It is a creative, highly personal task that takes shape not only through the characteristics of the author (their profile, preferences, and habits), but also through elements specific to the communicative event itself: the purpose of the message, the nature of the relationship between the author and the recipient(s), the social conventions of the given situation, and the conditions under which the production takes place.
The result is therefore a text that has been meticulously tailored by the author (consciously or unconsciously) according to their habits and idiosyncrasies within the context of a specific communicative scenario. This calibration—made possible by the richness and flexibility inherent in any natural language—allows the same idea to be expressed in different ways without altering the essence of the message. Examples of «non-essential» changes include register markers (the use of «tu» instead of «vous»), dialectal markers (the use of «pain au chocolat» instead of «chocolatine»), and even stylistic devices such as paraphrase or metaphor. These adaptations—whether semantic, grammatical, morphological, or otherwise—can contribute to the differences between two samples without affecting the central idea of the message. It is these differences that are referred to, in the broadest sense, as elements of linguistic variation.
In the case of stylistics, we are primarily interested in «marked» elements of variation—that is, variations that deviate from the expected norm in a specific context. For example, the use of a less common word when other synonyms—which are more common in the context in question—are available. Other interesting markers include turns of phrase influenced by knowledge of a foreign language (the use of the word «expectative» instead of ‘attente’ is fairly common among Spanish speakers, for example) or regional and dialectal variants that differ from the standard form. The degree of dissonance between the «norms» of usage and the forms used in the sample determines the degree of «markedness» of the words.
When a «distinctive» variation becomes habitual and inherent to a specific author, it also becomes part of what is known as an idiolect. An idiolect is a «personal dialect» unique to an individual. It consists of a person’s entire set of linguistic habits, whether spoken or written, and thus represents the expressive characteristics unique to that person.
Although idiolect is inherent to each person, many of its elements are far from fixed. Just like language, an idiolect is a «living» phenomenon. It evolves over time, adapts to its context, and is influenced by social phenomena and even by the individual’s cognitive development.
An author’s style, for its part, is the written manifestation of this idiolect. Defined by a set of distinct and recurring elements of variation, it is an author’s stylistic elements that make their texts unique and original compared to other texts of the same nature: the recurring use of a particular word when synonyms exist, the use of specific figures of speech in particular contexts, or an expression that always appears in the same context. Style is therefore the primary focus of the forensic linguist.
Challenges in Style Analysis
The style of a text is often idealized as the equivalent of a fingerprint. However, since an author’s style is not a physical entity, it is not appropriate to equate the comparison of fingerprints with that of textual samples. For example, since style is defined by the «marked» and habitual linguistic choices a person has made in the past, it is virtually impossible to analyze it in its entirety (since it would be unlikely to have a comprehensive inventory of all the linguistic choices a person has made). A style comparison will therefore always be a «comparison of partial fingerprints.».
Beyond this issue of fragmented samples, style analysis presents many other challenges, including problems related to the comparability and representativeness of the samples. Although an author’s style is meant to be unique and singular, the elements that contribute to this singularity are far from static. An author’s idiolect evolves over time and varies depending on the communicative context. This temporal and contextual variation means that samples from the same author are not always comparable (how can one compare an anonymous letter written 20 years ago to a recently published book?) or representative (what stylistic elements are present in a grocery list?).
Similarly, identifying the sources of variation is no easy task. Some variations are more closely tied to the specific communicative needs and conventions of the linguistic event than to the style of a particular author. This is true of thematic variations, contextual variations, and variations related to social conventions. For example, an author’s style may be constrained by a style guide. Similarly, while an overrepresentation of the word «society» in a short, informal message might be noteworthy, the overrepresentation of the same word in a financial text would not be surprising. It is clear, therefore, that an element of variation does not always constitute an element of style. For these reasons, variation is a much-discussed topic in forensic linguistics.
Attribution of Authorship in a Legal Context
Although authorship attribution is rarely used by French courts, it can be useful. Its methods make it possible to verify the authorship of a suicide note, analyze texts suspected of being plagiarized, or compare the style of anonymous letters or texts claiming responsibility for terrorist acts to those of known authors. In this context, the role of the forensic linguist is to identify, whenever possible, the elements that can help determine the authorship of a text.
Today, French forensic laboratories analyze documents by examining the materials (e.g., the type of paper), inks, and printing techniques. For example, the technicians in the Documents Department Scientists at the National Gendarmerie’s Criminal Research Institute (IRCGN) conduct physical and chemical analyses to determine the composition of paper substrates and inks and, when necessary, perform comparative analyses. They search for latent traces of pressure marks, identify printing techniques, and authenticate documents. They may also attempt to determine what type of equipment—and sometimes which printer—was used to produce the print.
Par technique d’impression, nous pensons généralement aux impressions faites par des machines. Toutefois, l’écriture et la signature manuscrites rentrent dans cette catégorie. Contrairement aux autres techniques d’impression, il s’agit donc de productions humaines personnelles et circonstanciées.
C’est sans doute, avec la voix, un des objets d’étude criminalistique qui s’approche le plus de la production linguistique. Chacun personnalise le modèle d’écriture qui lui a été enseigné et développe une écriture qui lui est propre. Cette écriture évolue dans le temps et selon les circonstances. Il n’est pas possible de déterminer a priori le degré de variabilité de l’écriture d’une personne. Ce dernier peut être plus ou moins important et seul l’examen approfondi d’un échantillon suffisant et comparable peut permettre au technicien de déterminer si un écrit litigieux a été produit par une personne donnée (le scripteur).
La question de l’échantillonnage
Une problématique commune aux deux disciplines concerne l’échantillonnage, en particulier la quantité et la qualité des échantillons. En effet, afin de déterminer l’étendue des examens possibles il est nécessaire de mettre en lumière toutes les limites qui pourraient empêcher le technicien de se prononcer, ou limiter son avis.
Les variations de l’écriture d’une personne peuvent être le résultat de facteurs multiples, dont certains contextuels (par exemple, l’espace disponible sur le support ou l’instrument d’écriture utilisé) et d’autres propres au scripteur. Ces derniers sont variés et peuvent aussi bien comprendre des pathologies, ou encore l’état d’esprit. De plus, les techniciens sont confrontés à des déguisements ou à des imitations ce qui peut être source de limites considérables à l’analyse. Il n’en demeure pas moins qu’il est parfois impossible de constituer un échantillon adapté.
Les approches à l’attribution d’auteur
L’attribution d’auteur est avant tout un problème de comparaison de style. Le linguiste identifie les éléments « marqués» et récurrents d’un texte et les compare ensuite à des échantillons dont la paternité est connue. Pour ce faire, plusieurs approches sont possibles, dont deux très utilisées actuellement : l’analyse axée sur l’expertise et la classification basée sur la reconnaissance de formes. Dans l’approche dite « d’expertise », le linguiste fait appel à son expérience et ses connaissances théoriques pour identifier les éléments potentiellement « marqués » de chaque échantillon. Les données ainsi obtenues passent ensuite par une étape de vérification et d’étayage où elles sont vérifiées ou rejetées selon les informations recueillies. Enfin, les éléments vérifiés de chaque ensemble de textes sont comparés et un avis sur la similarité ou dissimilarité des échantillons est donné. Anciennement ces analyses étaient effectuées de manière complètement manuelle. Aujourd’hui les linguistes se servent souvent d’outils informatiques tels que les concordanciers ou les corpus informatisés pour faciliter l’exploration des échantillons.
Le problème de l’attribution d’auteur peut également s’aborder comme un problème de reconnaissance de formes – spécifiquement comme un problème de classification automatique de textes. Dans ce scénario, les auteurs deviennent les « classes » auxquelles les échantillons peuvent être attribués. Cette méthode gagne du terrain depuis l’essor de l’intelligence artificielle dans le traitement automatique de textes. Dans cette approche algorithmique, la qualité des résultats repose fortement sur l’échantillonnage, les caractéristiques analysées (« features ») et le classifieur sélectionnés par le linguiste.
Perspectives en France
L’attribution d’auteur est une tâche complexe qui demande une compréhension approfondie du fonctionnement de la langue et des différentes méthodes d’attribution utilisées. La linguistique forensique peut être un outil d’analyse concourant à la prise en compte d’un texte en tant que trace.
L’université de CY Cergy Paris, à travers le laboratoire AGORA and the’Institut des humanités numériques, avec l’aide des techniciens du département Documents de l’IRCGN, travaille au développement d’une méthodologie d’attribution d’auteur adaptée aux contextes judiciaires. Pour plus d’informations sur les différents projets de l’Institut des humanités numériques, nous vous invitons à visiter notre site.
