Introduction

1Thomas Kuhn tends to give value to scientific languages as central in the broad and widely criticised conception of the paradigm : the scientific revolutions become mutations of taxonomy and for vocabulary and lexical structure “is meant a structured vocabulary of terms of type, that represents a taxonomy of natural types : for communication to take place between individuals, it is necessary not that these adopt common criteria of application of the terms to the world, but that they operate with homologous lexical structures, in other words with a structured vocabulary that possesses the same taxonomic system”2.

2Studies of the life and work of Jean Martin Charcot (1825- 1893) carried out over the last fifteen years3 all see in him one of the figures mainly responsible for the constitution of clinical neurology as a specific area of research, in particular in French-speaking areas.

3The aim of this study is to analyse how far the work of the Paris doctor was decisive also in the creation of a terminology proper to the discipline itself, in the conviction that it was a crucial moment in the development of a new conceptualisation, and a decisive moment in the history of a science : “A science does not begin to exist, or cannot emerge, except to the extent it makes its own concepts exist and imposes them its own denominations. It has no other means to establish its own legitimacy than specifying and naming its object, that may be an order of phenomena, a new domain or a new way of relating between objects given [...] To name, i.e. to create a concept, is both the first and the definitive operation in a science”4. The study of the appearance and transformation of the fundamental terms of a science can legitimately be considered one of the most important moments of its evolution. They can be looked upon as milestones in the process of development of a scientific discipline, and they can also mark its individual nature. “However, history of science does not yet pay the right amount of attention to these creations, that seem to be of interest only to lexicographers”5.

4In general, the medical terminology of vernacular languages, in its process of formation has tended to constitute an esoteric vocabulary, falling back on lemmas with Latin and Greek roots, finding a quite well-defined linguistic territory there that has always allowed the practitioners of the profession to communicate with each other, in the presence of the non-initiated, in a cryptic way. Is it possible to discover, within this area, a limited or restricted region, made up of a collection of terms proper to a special sector ? In order to avoid being excessively naïve, it is necessary that these terms refer to conceptual structures that are complex and interactive, because the semantic domains are reticular structures and not merely lists. Is it possible to assign to an epoch, a place, a person, the main responsibility for the discovery of these new semantic areas, one whose vocabulary marks its emergence ? A univocal answer to the questions is not yet available, lexicographical research being needed to focus both on the language of the innovators and on those of the conservatives. In this work an attempt has been made to utilise information technologies for the study of Charcot’s language, in close connection with the conceptual foundation of the new discipline, without renouncing either the quantitative aspects of the lexicographical enquiry or the qualitative ones ; but the work has been carried out from the perspective of a historian rather than a lexicographer, especial attention therefore being paid to the chronological and conceptual context in which the emergence of a new set of terms takes place, or a new meaning is assigned to a previously existing terminological constellation.

5As far as we know this is one of the first attempts to apply these techniques to a scientific, non-literary text ; as such, the tentative, by no means definitive, value of the techniques, and of the findings reached through them, will be appreciated.

I. The corpus analysed and tools of computational linguistics

6J.-M. Charcot’s Oeuvres Complètes consist of nine tomes that embrace most medical pathology, from infectious diseases to kidney and liver pathology, from lung infections to those of the vascular system, from skin diseases to the pathology of the old, and rheumatics.

7Furthermore, five of the nine tomes are on diseases of a neurological character ; in particular the fourth and the ninth are dedicated to lessons on cerebral localisation and vascular encephalic accidents, whereas the first three tomes bear the subtitle : Leçons sur les maladies du système nerveux6. In these, the university lectures on neurological subjects that the professor held on Fridays in the main lecture hall of the Salpêtrière are collected. These three tomes constitute the Corpus that was chosen to be the subject of computational linguistic analysis.

8Following the classical phases of pre-editing of the text as codified in Busa7, evaluation took place of :

1. the approximate extent of the corpus in words : the first tome is composed of 461 pages containing 142 260 words ; the second tome is of 143.053 words in 494 pages, and the third tome 114.962 words in 467 pages.

2. the discrimination of the parts of the text that had to be registered from those that were not to be registered : thus the indexes, the summaries, the titles, and the headings were eliminated. In addition, the footnotes were not inserted in the corpus, because in most cases they contain quotations from other authors that Charcot cited in support of or against his theses, and therefore in a strict sense they are not his original production. In any case, over and above this reason, the above cited protocol lays this down, and again as the protocol lays down, of these notes we redacted a separate file.8 The second and eighth chapters of the appendix of tome II were similarly excluded from the redaction of the corpus, because they carry Bourneville’s signature.

3. We defined a system of references basically of a physical, or typographical character : we therefore devised an index system that provides for the indication of the volume, of the lesson (or chapter), the page, the line and the position of the word in the line. Given the particular system adopted for the transformation of the written text into digital form, i.e. photographed by a scanner9, and later development with software of a recognised character, the constitution of the system of references meant the reconstruction of the original line endings of the text. As far as the words whose syllables were split at the end of the line went, it was decided to assign them wholly to the line afterwards, to avoid leaving the line empty in those cases where the end of that word was the only one present.

4. The definition of the graphemic system, on the other hand, did not create any special difficulties, the present-day recording systems of the text being capable of preserving almost all the graphemes present in the original text without codification.

9The files obtained by the previous phase of pre-editing were developed at two different times : in a first phase the systems of references for each tome10 were produced ; the corpus was then subjected to processing with INTEX.

10Administering all the computational tools the processor was endowed with led to obtaining the word list of the corpus, consisting of 8 599 lemmas.

11In addition, two thirds of the forms had their ambiguities taken away, and labelled with the appropriate lemma and grammatical description.

II. The computational construction of Charcot’s technical vocabulary

12The approach chosen consisted in the first place in the setting out of a technical vocabulary from the overall word list of the corpus. The strategy followed to obtain this result is here briefly outlined.

13The INTEX system permits the research into the corpus of compound words starting from internal dictionaries, or those implemented by the user ; these compound lemmas were isolated, then sorted applying a simple research criterion that meant filtering the compound lemmas not belonging to basic French (i.e. the technical ones). To these were added the lemmas found starting from dictionaries implemented by the user11, obtaining a first set of lemmas.

14It is important to bear in mind that this list of compound lemmas was produced automatically by the lemmatisation software : the only intervention by the user was the filtering operation on the basis, furthermore, of criteria that were internal to the programme and not created “ad hoc” : there was no imposition of external criteria that could make the operation open to criticism from the point of view of self reference. In other words it was not we who established the norm on the basis of which the choice of lemmas was made : the characterisation of the lemmas as « technical » was inherent in the definitions of the dictionaries adopted.

15Once in possession of an author’s word list, in fact, the researcher, whether a linguist or historian of science, could be tempted to impose his own interpretation on the structure, turning the inductive-deductive process upside down, and hence seeking in the text a pre-ordained set of emergences. This would create a vicious circle in the utilisation of the computational tool for interpretative ends : if researchers impose their own norms, if a grid is established a priori, even if this is declared, they will finish up by finding the idea they had started out with, or premised.

16To avoid this risk, we tried to allow the lexicological norm of the corpora tested to emerge from the computational linguistic analysis.

17A second class of lemmas that contributed to make up the computational “neurological” vocabulary of the corpus was the compound words with a dash : looking down their list in fact we see that they are almost exclusively technical lemmas, mostly adjectives, for example :

anatomo-clinique, anatomo-clinique. A+d : ms : fs
anatomo-pathologique, anatomo-pathologique. A+d : ms : fs
auto-suggestion, auto-suggestion. N : fs
cérébro-spinal, cérébro-spinal. A+d : ms
demi-flexion, demi-flexion. N : fs
électro-diagnostic, électro-diagnostic. N : ms
électro-musculaire, électro-musculaire. A : ms : fs
électro-physiologique, électro-physiologique. A : ms : fs
électro-thérapiques, électro-thérapique. A : mp : fp
entre-croisement, entre-croisement. N+Conc : ms
hémi-paraplégie, hémi-paraplégie. N : fs
hystéro-épilepsie, hystéro-épilepsie. N : fs
labio-glosso-laryngée, labio-glosso-laryngé. A+d : fs
myo-sclérosique, myo-sclérosique. A+d : ms : fs
neuro-pathologie, neuro-pathologie. N : fs
neuro-pathologique, neuro-pathologique. A+d : ms : fs
pie-mère, pie-mère. N+NN+Conc+z0 : fs/une ; Anat
pied-bot, pied-bot. N+NA : ms
pseudo-hypertrophique, pseudo-hypertrophique. A+d : ms : fs
pseudo-névralgiques, pseudo-névralgique. A+d : mp : fp
psycho-physiologique, psycho-physiologique. A+d : ms : fs
téphro-myélite, téphro-myélite. N+Nos : fs
thermo-anesthésie, thermo-anesthésie. N : fs

18Finally a third, conspicuous, set of lemmas that we believe possesses the required characteristics to be counted in the computational construction of Charcot’s vocabulary was obtained by the counting of the co-occurrences of nominal forms connected with certain specific verbal groups or nominal groups recognised as typical of Charcot’s style.

19For example, the co-occurrences of compound words that followed the locution le nom de were classified, and such lemmas were included in the vocabulary being constructed. These were, for example, the nominal groups underlined in the following KWIC12 table :

ison. L’une de ces formes pourrait être désignée sous le nom de sclérose fasciculée systématique médiane ou
ique ; elle mériterait vraiment d’être désignée sous le nom de signe de Brodie, car c’ est au célèbre chirurgien
ssive, dont il s’ agit, pourrait être désignée sous le nom de téphro-myélite chronique parenchymateuse. Les
a sclérose en plaques a été quelquefois désignée sous les noms de chorée rythmique, paralysie choréiforme.

20This procedure was repeated for the co-occurrences of the following nominal forms : “le nom de, cas de, dénomination, type, espèce, forme, [and] variété”, and for the verbal forms of “dénommer, appeller, affirmer, désigner, conclure [and] nommer”.

21The result of these operations was the creation of a dictionary of compound lemmas, made up of 1 296 entries, compiled in INTEX format. The same operations, the filtering of the technical lemmas and analysis of the co-occurrences, were carried out on the simple lemmas, obtaining a second dictionary with 1 051 entries.

22The working hypothesis was that these 2 347 entries overall made up a set of terms that were part of the vocabulary of neurology ; it was obvious that, given the strategy of total automation of the selection procedures, some non-congruent lemmas might be present, but we preferred to accept this « background interference », putting the non-intervention of the user in the compilation of the vocabulary first, with respect to operations of selection that would have given the « lexicon » greater formal rigour, but would have introduced an interference mechanism, consisting in the knowledge of the researchers.

III. Analysis of the diachrony of scientific languages

23Equipped with this tool, two other scientific corpora on the subject of neurology were analysed : the 1 855 work by Duchenne de Boulogne, De l’électrisation localisée et de son application à la physiologie, à la pathologie et à la thérapeutique, and the Sémiologie des affections du système nerveux by Jules Dejerine, of 1914.

24Duchenne’s text is the nearest chronologically among the texts that came before Charcot’s work : in 1855, the latter had been a graduate for two years and had met Duchenne as an intern while in the service of Rayer.

25On the other hand Dejerine’s work was published twenty years after Charcot’s death, and was chosen because it is a basic text of twentieth century neurology, reprinted right up to 1977, and written by one of the greatest French neurologists, the only one among the successors to Charcot’s chair up until 1925 who had not been his pupil.

26Both the works were conceived within the same architectonic and conceptual “topos” that had seen the production of Charcot’s works : the Paris Hospice Vieillesse-Femme de la Salpêtrière.

27Duchenne’s corpus was tested with the dictionary created starting from Charcot’s corpus : i.e. the neurological vocabulary obtained with the methodologies described above ; in other words, all the occurrences of around 2 300 lemmas, collected automatically, and provisionally defined as Charcotian neurological vocabulary, were sought for with the lemmatisation software of the linguistic processor in Duchenne’s corpus.

28To do this, the “neurological vocabulary” created by the user was added to the complex of dictionaries within the linguistic processor. In this way, in the automatic operation of labelling Duchenne’s text, the lemmas listed in this vocabulary were “marked” with a specific tag, which allowed it to be found and extracted from the text. The first significant result of this operation was quantitative : in the Électrisation localisée just 196 lemmas of the “neurological vocabulary” were present.

29In more detail :

30Thus, in Charcot’s corpus the occurrences belonging to this special lexicon were more than ten thousand, whereas in Duchenne they numbered 2 655. Obviously, also in this case it is necessary to bear in mind the overall total of the forms that made up the respective corpora. The size of the Charcotian corpus was around four times the size of Duchenne’s text. For this reason, following the classical indexes of the lexical statistics, two indexes of evaluation of the marking obtained on the corpus under examination (with the dictionary of neurological lemmas) were used : the first, called density, expressed the relation between the logarithms of the number of the occurrences belonging to the test lexicon and all the words present in the corpus. Its value was hence proportional to the marking. The more forms of the tested lexicon were found, the higher the value of this index. The use of logarithms served to reduce the bias linked to the different size of the corpora being compared, as in the case of the Herdan index. The index of variety, on the other hand, was given by the relation between the different forms and the total occurrences in the test lexicon in the corpus : it expressed the variety of lemmas used, within the particular subset of “neurological” terms : hence it was greater the richer the “locale” vocabulary.

31In this way, the density of the forms belonging to Charcot and Duchenne’s neurological lexicon was not very different : in Charcot (0.73), around one form every thirty-two belonged to it and in Duchenne (0.69) the value was very similar (1/33) .

32On the other hand, the index of variety marked the relative lexical poverty of Duchenne compared to Charcot : only 196 lemmas of Duchenne could be found in the vocabulary of neurology : in 1855 the language of neurology had yet to be created. This conclusion was strengthened by the qualitative analysis : from this perspective, indeed, the marked forms that appeared most frequently were paralysie (394 times) with the variants of paralysie générale (134) and paralysie générale spinale (27), électro-muscolaire (182) with the obvious faradisation (163) and contractilité électrique (109), but also atrophie musculaire grasseuse (44) and progressive (12). In other words, a small handful of lemmas made up over half the occurrences.

33In conclusion, Duchenne’s local neurological taxonomy constituted a territory of very modest dimensions, structured around the ancient generic syndrome “paralysis” and renewed only by the introduction of electro-physiological vocabulary. Starting from this inheritance from his “master”13, the linguistic edifice constructed by Charcot was on a scale of such a size and complexity as to autho-rise the judgement that it was of foundational value.

34Also Dejerine’s corpus was tested with Charcot’s neurological vocabulary : the linguistic processor in automatic mode marked in the Sémiologie all the occurrences of the dictionary created with the criteria described at the start of this chapter. The quantitative results, also compared with the marking of the Duchenne corpus are shown in the table.

35As will be seen, the “density” of Charcotian occurrences in Dejerine was even higher than in Charcot14 himself, and this vocabulary was made up of 736 different forms, amounting to almost four times the forms present in the Duchenne corpus, prior to Charcot ; in recompense, the index of variety indicated that the frequency of the technicisms used in Dejerine was on average of ten occurrences, against the five of Charcot : so in Dejerine a contraction at any rate did take place of the ample Charcotian vocabulary (from which the redundancies were eliminated, as well as the lexical proposals referring to pathological pictures already criticised by the scientific community). To sum up, Charcot’s vocabulary that passed into Dejerine was the one made acceptable by the common usage of contemporary neurologists.

36It should be recalled that the analysis was not carried out with reference to a generic neurological vocabulary ; the latter was obtained with simple algorithms starting from the Charcot corpus itself, so that in this table, the language used in Dejerine’s neurology manual “is” Charcot’s language.

37To the quantitative analysis were added interesting notations from the qualitative analysis of these forms. In Dejerine’s corpus all the lemmas that in Charcot made up that particular region of his vocabulary devoted to hysteria were fundamentally lacking : of all Charcot’s lexical equipment on hysteria, in Dejerine only the hysterical contractions, hypnotism, and the hysterical coxalgia will survive.

38This disappearance of hundreds of words and locutions from the dictionary of hysteria, in the passage of neurological vocabulary from Charcot to Dejerine, offers confirmation of the historiographical paradox that has long afflicted Charcot, corrected only in part by recent studies15, i.e. of being remembered for his errors rather than for his achievements. Also with his vocabulary, what refers to the world of psychophysiology did not last long. His linguistic innovations, the words he created, the constructions he invented that have become a definitive part of the scientific vocabulary, are those belonging to the field of clinical neuropathology, and not those of a claimed beginning to dynamic psychology.

39Obviously, in Dejerine’s vocabulary “new words” are to be found, absent in Charcot for various reasons : because they refer to the pathogenetic importance of bacteriology (staphylococcus), that Charcot never took into consideration, or to sectors that Dejerine in particular explored, such as the aphasies (amusique, aphémie, aphrasie, asymbolique) or to diseases that have many times changed their names in the course of the history of medicine, such as syringomyelia. Finally, in Dejerine Charcot’s “little nervous filaments”, becomes axon.

40Up to this point we have dealt with what was lost of Charcot’s neurological vocabulary, or what did not yet exist.

41When we look through the list of Charcot’s lemmas incorporated in the vocabulary of Dejerine : from the reaction of degeneration to thermoanaesthesia, from amyotrophic sclerosis to ophthalmic hemicrania, from hemianopsia to double athetosis, in the over seven hundred lemmas “marked” with computational technique, Charcot’s language presents itself as “the” language of clinical neuropathology, the language of “his” discipline.

Conclusion

42Within the linguistic corpora tested by the technique proposed here, and with the limitations we have pointed out along the way, both the quantitative data and the qualitative analysis seem to suggest an affirmative answer to the issue raised by the research : the computational linguistic study of Jean Martin Charcot’s works of neurological interest confirms the key role of the Salpêtrière clinician in the creation of the new discipline, not only pointing out the methodological and conceptual contributions, but also attributing to them the paternity of a region of technical languages, not limited to a list of nosologies, but rather structured in epistemological, content-oriented, and semiological moments.

43In Jules Dejerine’ s words : “… si le clinicien de 1850 revenait parmi nous, la terminologie scientifique que nous employons aujourd’hui serait pour lui à peu près vide de sens. Il reconnaîtrait l’hémorragie et le ramollissement, les vices de conformation et les traumatismes, comprendrait la signification sémiologique d’un certain nombre de termes par nous usités. Mais les cadres nosologiques, les affections définies et, à l’heure actuelle, classiques, il les ignorerait à peu près complètement. On ne parlerait plus sa langue, mais on parlerait celle de Charcot”16.

44For the study of Charcot’s language, with these aims, a methodological instrument has been proposed that up until now has been applied to literary languages : computational linguistics.

45For the historian, this instrument is a form of second language, at the same time comment and criticism in its filtering of the linguistic competence of a cluster of scientific texts, giving back to us a metatext and an electronic vocabulary, without the tyranny of alphabetical ordering of the lexical entries of printed dictionaries, but explorable by degrees of frequency, grammatical categories, inflections of conjugation, and classifiable on the basis of semantic categories created by the user. Of course, the dream – or nightmare – of a mathesis universalis to which the task of exploring the historical sources is entrusted, with the aim of cybernetically obtaining the objectively definitive semantic structure, is destined such to remain for a very long time, perhaps forever ; but the instrument is fertile even beyond mere scientometric applications that up to the present have been the most frequent and important application17. The neutrality of IT is obviously false : for a start the original choice, of the primary material to be submitted to analysis, is informed with the pre-conceptions of the scholar. Once aware of these limitations, however, the methodology is capable of providing reasonably good results for historical research.

46In the meanwhile, as well as constituting a powerful instrument of textual analysis, making a word list of the corpus being examined available to the scholar, it is capable of suggesting an evaluation of the obsolete vocabulary ; it can provide indications to direct historical research with traditional methodologies into unsuspected directions ; it enables the semi-automatic creation of disciplinary lexicons and the comparative evaluation of such structures in different authors.

47In our putting forward computational lexicography of textual corpora as a methodology useful to research into scientific history, we are not enthusiastically blind to the difficulties – above all the availability of sources in electronic format – and to the gaps it shows, but simply present some findings that make up a starting point for the historian ; not a goal achieved as much as an auxiliary help to those processes of “interpretation” that, according to Kuhn, “are needed where the work of translation cannot be carried out. These processes are not arcane. Historians, anthropologists, and perhaps little children are every day involved in doing them”18.

48This paper is a concise version of a longer essay published in : “Augmenting Comprehension : Digital Tools and the History of Ideas” ; Dino Buzzetti, Giuliano Pancaldi & Harold Short Editors, OHC publications, London, 2004.