Detecting Psychological Disorders with Stylometry: the Case of ADHD in Adolescent Autobiographical Narratives

Abstract

Attention-de昀椀cit/hyperactivity disorder (ADHD) is one of the most common psychological neurodevelopmental disorder among children and adolescents, with a prevalence of 5.6% in teenagers aged 12 to 18 years [51]. Its diagnosis is reliable and valid when evaluated with standard criteria for psychiatric disorders1[6], but it is time consuming and requires a high level of expertise to arrive at a correct di昀erential diagnosis. The development of low-cost, fast and e昀케cient tools supporting the ADHD diagnosis process would therefore be important for practitioners, because it should help identify and prevent risks in di昀erent populations. In this paper, we study the possibility of detecting ADHD with Natural Language Processing (NLP), based on the analysis of a speci昀椀c type of adolescent's autobiographical narratives called Self-De昀椀ning Memories (SDMs). (1) We train a Support Vector Machine (SVM) to predict ADHD diagnosis, (2) we attempt to explain its results by exploring lexical information (3) and unfolding the results of the SVM to identify and analyse the linguistic markers associated with each groups. With an accuracy of 92%, the SVM manages to classify texts from both group (ADHD vs Control), revealing a signal speci昀椀c to autobiographical texts narratives written by people with ADHD. The quality of the detection is con昀椀rmed by the interpretative yield of the main markers identi昀椀ed. However, several methodological improvements remain necessary to improve the accuracy and the automation of ADHD diagnosis with stylometric methods.

Keywords

• eol>Stylometry
• NLP
• Psychology
• Psychological disorder
• ADHD
• Self-De昀椀ning Memories

1. Introduction

The assumptions that the “words we use in natural language […] reveal a tremendous amount of information about our social interactions and personality” and that Natural Language Processing (NLP) can help us access to this information is now widely shared by linguists, philologists or psychologists1[1]. With regard for psychological disorders, such a “revelation” is not enough: because ADHD poses signi昀椀cant risks [ 24 ], we need to e昀케ciently assist practitioners to detect such disorders as automatically and as reliably as possible. This paper proposes therefore a case study to evaluate the use of a NLP technique, stylometry, for the detection and possibly the understanding of Attention-de昀椀cit/hyperactivity disorder (ADHD).

Research crossing NLP and psychology is mainly conducted on English native speakers, but some papers exist on romance languages (such as Spanish3[1]) or Asian languages (such as Korean [ 29, 28 ]). It is precisely on Korean that the rare studies on ADHD concluding with a speci昀椀c style for people prone to this disorder have been carried out. If a speci昀椀c style exists, stylometry should make it possible to identify it, and therefore presents itself as a promising method for diagnosing ADHD, and thus preventing the risks associated with it.

1.1. Attention-deficit/hyperactivity disorder (ADHD)

ADHD is a neurodevelopmental disorder that a昀ects 5.6% in teenagers aged 12 to 18 years [ 51 ]. Its causes are multifactorial: several genetic and environmental risk factors act together to increase both susceptibility to the disorder and the extent of psychiatric comorbidit1i6e]s. [ Regarding its diagnosis, it is reliable and valid when evaluated with standard criteria for psychiatric disorders, but the expression of symptoms varies as a function of patient developmental stage in both the social and the academic contexts1[ 6 ]. All this makes ADHD diagnoses a time consuming task (up to one working day) requiring a high level of expertise.

Furthermore, during adolescence, young people with ADHD are especially prone to experience di昀케culties in interpersonal relationships, characterised by con昀氀icts within the family or at school. In the latter case, con昀氀icts with classmates may in turn lead to peer rejection, social isolation or school failur2,e3[ , 54 ]. Besides these relational problems, adolescents with ADHD are also exposed to higher risks of neurocognitive dysfunction, substance abuse, low self-esteem and social disability16[]. In summary, this leads to a situation in which a complex diagnosis with extremely serious health and social implications for patients must be made, involving time and resources that are not always available and in a context of high prevalence.

1.2. Psychology and Natural Language Processing (NLP)

For more than ten years now, the use of NLP has established itself as a valid method for di昀erent types of clinical applications in neuroscience and psychiatr1y3][, which explains an upward trend for this approach and a growing need to develop new detection metho6d5s][. Clinical applications include psychological pro昀椀ling39[ , 29 ], or the detection of depression1[ 5, 47 ] and psychological distress3[ 2 ] with di昀erent methods: word embeddings, sentiment analysis or stylometry. The latest research makes it possible not only to identify mental health risks (like patients who exhibit suicidal risk behaviou5r9[]), but also to predict depression15[] or schizophrenia in 昀椀rst-episode of psychosis [ 17, 58, 49 ] on a linguistic basis.

2. Data and features

2.1. Corpus

The corpus is made up of a series of 198 Self-De昀椀ning Memories (SDMs, cf. sectio2n.3) collected in two samples, one consisting of adolescents with a diagnose of ADHD and the other with control participants (cf. section2.2), who all had to write a total of three SDMs each. All texts are written in French.

SDMs were handwritten by the participants and then transcribed by the person in charge of the experiment. If spelling errors can provide valuable linguistic information, we noticed that some errors were introduced or removed during computer entry of the text. We have therefore decided to correct all of them: a 昀椀rst time with a professional spell checker, a second time with the Hunspell package [ 40 ].

2.2. Participants

ADHD group Adolescents with ADHD were recruited as part of a research project conducted at the Unit of Developmental Clinical Psychology at the University of Geneva. The project was advertised in local parents’ associations for children with ADHD and through collaborations established with local child psychiatrists. Diagnostic criteria were investigated by detailed anamnestic interviews and con昀椀rmed by a semi-structured parents interview using the ADHD Child Evaluation cf.6[ 4 ]. All diagnostic assessments were conducted by experienced clinical psychologists specialised in ADHD.

Control group The control sample was recruited from the general population in Geneva by means of advertisements and personal contact. Participants were volunteers, native French or 氀昀uent French speakers and received a compensation for their participation.

Inclusion/Exclusion criteria Inclusion criteria were age (12-17), 昀氀uency in French and, for the ADHD group, meeting current diagnostic criteria for ADH1D].[ Exclusion criteria were history of psychotic disorders, borderline personality disorder, autism spectrum disorder or neurological disorders.

2.3. Self-defining memories

SDMs are written texts that represent a speci昀椀c type of autobiographical memories associated with the self-concept and a sense of coherence and continuity in one’s ongoing individual history [ 4 ]. They were collected with the SDM Task6[ 0, 55 ], during which participants are asked to evoke personal memories of events with specic attributes. These events 1. occurred at least one year ago; 2. are important and generally vividly represented; 3. are meaningful and useful to help the participant (or a signi昀椀cant other) to understand who s.he is; 4. are related to an important and enduring theme; 5. are either positive or negative but must generate strong feelings; 6. were recalled many times on a voluntary basis or spontaneously.

While listening to this description, participants had a sheet of paper in front of them that summed up these principal points. Participants were then told to imagine a situation where they meet someone they like very much and with whom they agree during a walk to talk about who they really are, their “Real Me”, sharing several personal past events that powerfully convey how they have become the person they currently are. Participants were then given three sheets of paper on which they had to write down a SDM on each of them.

3. Prediction task

As already said, previous studies have shown that persons with ADHD have a di昀erent linguistic style from non-ADHD groups6[ 3, 28, 29 ]. It should therefore be possible to classify texts from the ADHD and control groups with stylometric methods.

3.1. Hypothesis

Function words (FW) are words used both in stylometr2y7][ and in psychology 1[1]. These words have little lexical meaning and a grammatical role in the sentence (e.g. articles, prepositions, conjunctions, auxiliary verbs…). As such, FW are opposed to content words, which have a semantic content (e.g. nouns, adjectives, verbs…), but there is no clear delimitation between the two groups. Indeed, some words can be classi昀椀ed in one or the other category such as pronouns. Pronouns have indeed attracted much attention from researchers, who have emphasised their particular status 昀椀rst in literature36[ , 26 ] and then in psychology 4[ 2 ].

According to Chung and Pennebaker 1[1], examining the use of FW in natural language samples has provided a non-reactive way to explore personality processes. For example, it has been found that use of speci昀椀c FW is related with a昀ective states [ 62, 11 ], depression [ 50, 34 ], reactions to individual life stressor4s4][, reactions to socially-shared stressor8s, [ 56, 20, 12 ], deception [ 43, 38 ], status [ 45 ], sex [ 37 ] and age [ 46 ]. We think that the use of FW di昀ers between our two working groups (ADHD vs control) and that it is therefore possible to automatically classify one and the other group on the basis of FW. For years now, stylometry has been used to classify all types of document5s3[], and seems to be the most suitable method for this task.

In addition to FW, we also propose to use another traditional feature of stylometric research: characters 3-grams 3[0], which can capture lexical, and even grammatical preferences. Such a feature has already shown its relevance for French texts in previous studies on authorship attribution [ 7 ], and has also proven its capacity to capture more than the authorial sign1a8l].[ 3.2. Set up Support Vector Machine (SVM) O昀琀en used in stylometry, unsupervised approaches 1[ 0, 19, 33, 7, 52 ] seemed less adapted than supervised techniques for this pro昀椀ling task. A recent survey [ 21 ] has shown that classical machine learning methods still perform better for pro昀椀ling in similar settings (short texts, boolean or few categories) than deep learning. We turned to Support Vector Machines (SVMs) rather than random fores2t3[] or logistic regression35[ , 61 ], as it allows for easy interpretation and have established themselves as a standard method in stylometry 1[ 4, 5 ].

SuperStyl In this study, all analyses were implemented with the PythoSnuperStyl package [ 9 ]. This package has been used to build stylistic pro昀椀les with very good result6s].[ SuperStyl use internally the SVM and pipeline facilities from scikit-lea4r1n].[ Data All participants’ SDMs are collected in a single 昀椀le, with the exception of those written by two people from the control group and two from the ADHD group for a 昀椀nal blind test. Several word sample sizes are tested (1’000, 1’250, 1’500 and 2’000).

Parameters We tried to use two types of features: FW and character 3-grams for the reasons previously exposed. Because the length of SDMs varies a lot and we have more SDMs in the control group than in the ADHD groups (cf. sectio4n.3), we have tested di昀erent sampling methods (no sampling, downsampling, upsampling, Tomek links) and the use of class weights. All tests are conducted with a linear kernel and a 10-fold validation, on data normalised by using z-scores for variables and applied Euclidean vector-length normalisation (L2 normalisation).

3.3. Results

Scores Best results are achieved with 1’500 words samples, 10-fold validation, class weights and Tomek Links (for FW) / downsampling (for 3-grams). Accuracy is slightly lower with FW (0.85) than with 3-grams (0.92), which appears here as a promising indicator for research in psycholinguistics. However, the results remain surprisingly good in both cases. The recall for the ADHD group deserves special attention: for obvious reasons, a maximum number of people with ADHD must be identi昀椀ed and a minimum must be misclassi昀椀ed. With 0.75, results are not satisfactory yet, but promising. Test The SDMs of four participants (2 ADHD and 2 control) were not used for training and kept for a final test. The model perfectly classify ADHD and non-ADHD SDMs.

4. Lexical exploration

In a medical context, predicting is not enough: it is necessary to explain the results. The study of our corpus could make it possible to give initial explanations as to the quality of the prediction, but also to confirm several hypotheses.

4.1. Hypotheses

Two standard measures have been used to explore the lexical information in the corpus: 1. Lexical Density measures the structure and complexity of a text. It provides a measure of the proportion of lexical items (types) in relation to the total number of tokens in the text [ 25 ], and thus helps evaluating the amount of information in a given document [ 22 ]. Diversity evolves during human lifespan and it is influenced by di昀erent factors like education or the communication style of the family. Other studies in the field of psychology have found a relationship between neurodevelopmental disorders and lexical density, where children with autism spectrum disorders tend to have a lower density than the average population [ 63 ]. To the best to our knowledge, no study has yet explored lexical density in the linguistic production of adolescent with ADHD. Considering the fact that ADHD is a neurodeveloppmental disorder, measures of lexical density could di昀erentiate adolescents with ADHD from those with a typical development. 2. Lexical Diversity measures how many types are used in each adolescent’s narrative.

SDMs that are lexically diverse use a wide range of vocabulary, of synonyms and a precise language to describe their memories. Previous studies have shown that measures of lexical diversity do not di昀erentiate adolescents with ADHD from those with a typical development [ 48 ], we should therefore expect a similar result. 4.2. Set up Pairwise matching is used to create samples balancing the ADHD and the control groups with respect to the means of participants’ sex1 and age in order to make them comparable. 1We do not use gender, but biological sex as a category, as many other studies do. Our objective here is not to study the di昀erence between these two groups (male vs. female) but to obtain two samples as comparable as possible (ADHD vs. control). The question of gender could, however, be of interest in future studies, especially because ADHD is possibly because ADHD is generally more likely to be diagnosed and treated for boys.

We matched the two samples using the cardinality matching method to find the largest matched set (in this case by age and sex) with the additional constraint that the ratio between the number of adolescents in both groups had to be equal to 1. This method allowed us to avoid di昀erences between groups by sex or age with minimal loss of ADHD cases selecting the best-fitting control cases in function of the ADHD group.

Final samples In both groups the final sample meeting and pairwise inclusion criteria consisted of 24 adolescents (12 females and 12 males) in both groups (cf. tab. 2). The age mean weight in ADHD group was 15.14 ( = 1.83) and 15.21 ( = 1.44) for the control group. A Student two-samples t-test showed that the di昀erence was statistically not significant (t(43.65) = group has been reduced to 72 SDMs (24 participants × 3 SDMs) per group. -0.16, p = 0.87). Finally, as a result of this pairwise matching, the total number of SDMs per Analyzing which features allowed our SVM to distinguish between our two groups can help us understand further the di昀erences between ADHD-diagnosed respondents and individuals from the control group. The size of our corpus does not allow us to comment on content words, not frequent enough to be considered as reliably interpretable. But we can gain significant insights from function words (cf. fig. 2).

4.3. Results

Lexical diversity (TTR = ) is higher for the control group (209, = 63,9), indicating its members use significantly more di昀erent words to describe their memories than the ADHD group (153, = 54,5, p < 0.001), contradicting the results of Redmond [ 48 ]. This fact could be explained by the significant di昀erence in size between the SDMs of the two groups (cf. fig. 1), those of the control group being generally longer ( ̄ >100 token) than those of the ADHD group (≈̄ 60): as the length of the text increases, there is a greater statistical likelihood of finding a new word.

On the contrary lexical density ( = ∗ 100) is not significantly di昀erent between the control group (0.027, = 0.008) and the ADHD group (0.029, = 0.01), which invalidates our hypothesis based on Yoder [ 63 ].

4.4. Functors as markers

On the other hand, the control group is very marked by first person pronouns (je, me/m) and the plural (des, plus).

Context of meaning (...) j’avais peur de faire des toboggans quand un ami m’a forcé à le descendre.” (...) C’était compliqué de se concentrer à cause des distractions de la maison (...) (...) j’ai rencontré ma copine et ça m’a rendu plus heureuse et sûre de moi (...) (...) C’était la première fois que j’assistais à un enterrement. (...)

5. Discussion

Our findings indicate that adolescent’s with ADHD diagnosis show significant di昀erences in their autobiographical narratives style of language from their control group (non-ADHD cohort), which is enough to have them detected by the machine. Additionally, di昀erent markers for each group were found which are meaningful from a psychological perspective. Indeed, the use of the neutral pronoun in the ADHD group is very di昀erent from the use of the first singular pronoun in the case of the Control group. In terms of agency, the latter identifies the author as the subject performing and feeling things while the narrated event occurred. In the case of the ADHD group, the agent is subsumed in an undefined neutral mass (on) that performs things in a distinctly less personal way. In their model of the relationships between self, memory, and visual perspective, Sutin and Robins [ 57 ] argued that in the first person perspective, individuals see the event through their own eyes, while, in the third person perspective, individuals see themselves and the event from the perspective of an external observer. According to their model, a reduced use of the first person may serves a distancing function helping to reduce emotional reliving and to distance the current self from the self in the memory. In this sense, adolescents with ADHD show at least di昀iculties to connect with their emotional experiences, either by way of a defence mechanisms or by functional di昀iculties related to executive functions. This result is even more relevant to consider if one takes into account that the task asks for memories lived by the person that have marked him/her and that, therefore, easily elicit first-person experiences.

With respect to lexical density, our results are not consistent with previous finding and does not confirm di昀erences between groups based on the presence or absence of ADHD. However, this could be due to the small size of the ADHD group narratives.

6. Further work

Future studies are needed to increase the accuracy and fine grain detection of ADHD. First the potential di昀erences by sex or gender in language markers should be addressed. Second, refining the precision of the detection in order to identify the di昀erent modalities of ADHD (predominantly inattentive, hyperactive or combined) is another relevant challenge to consider for future studies. Third, self-reported a昀ect reported by participants in the Task of SDMs should be considered in order to evaluate to what extend this metadata could improve the detection of the signal of ADHD in the adolescent’s autobiographical narratives. Finally, considering the potential challenge for young adolescents to write a memoir, another interesting aspect would be to modify the task to make it a verbal task and, eventually, also modify the instruction to make it simpler and eventually more stimulating for the target population.

Acknowledgements

JB was funded by the Chilean National Agency for Research and Development (ANID Chile) through the PhD Abroad Scholarship, 2018 award. The PI (Prof. Martin Debbané) was funded by the Swiss National Science foundation (Grant number 100014_179033), as well as the Marina Picasso Prize from AEMD Foundation 2018.

H. Gómez-Adorno, C. Martńı -del-Campo-Rodrǵıuez, G. Sidorov, Y. Alemán, D. Vilariño, and D. Pinto. “Hierarchical clustering analysis: the best-performing approach at PAN 2017 author clustering task”. InE:xperimental IR Meets Multilinguality, Multimodality, and Interaction: 9th International Conference of the CLEF Association, CLEF 2018, Avignon, France, September 10-14, 2018, Proceedings 9. Springer. 2018, pp. 216–223. url:10.1007/9 78-3-319-98932-7%5C%5f20.

References

A. P. Association . Diagnostic And Statistical Manual Of Mental Disorders, Fi昀琀h Edition , Text Revision (DSM-5 -TR). Washington D.C.: American Psychiatric Association, 2013 . doi: 10 .1176/appi.books. 9780890425787 .

[2] C. L. Bagwell , B. S. Molina , W. E. Pelham , and B. Hoza . “ Attention-De昀椀cit Hyperactivity Disorder and Problems in Peer Relations: Predictions From Childhood to Adolescence” . In: Journal of the American Academy of Child & Adolescent Psychiatry 40.11 ( 2001 ), pp. 1285 - 1292 . doi: 10 .1097/ 00004583 -200111000-00008.

[3] R. A. Barkley and K. E. Fletcher. “ Adolescents with ADHD: Patterns of behavioral adjustment, academic functioning, and treatment utilization” . InJ:ournal of the American Academy of Child & Adolescent Psychiatry ( 1991 ). url: 10 .1016/s0890- 8567 ( 10 ) 80010 - 3 .

[4] P. S. Blagov , J. A. Singer , K. M. Oost , and J. A. Goodman . “ Self-de昀椀ning memoriesNarrative features in relation to adaptive and maladaptive personality traits (replication and extension of Blagov & Singer, 2004 )”. InJ:ournal of Personality 90.3 ( 2022 ), pp. 457 - 475 . doi: 10 .1111/jopy.12677.

[5] F. Ca昀椀ero and J.-B. Camps . “' Psyché' as a Rosetta Stone? Assessing Collaborative Authorship in the French 17th Century Theatre” . In:Proceedings of the Conference on Computational Humanities Research 2021 . Computational Humanities Research. Virtual: CEUR Workshop Proceedings , 2021 , pp. 377 - 391 . urlh:ttps://ceur- ws.org/Vol- 2989 /long%5 C % 5Fpaper51 .pdf.

[6] F. Ca昀椀ero and J.-B. Camps . “ Who could be behind QAnon? Authorship attribution with supervised machine-learning” . InD:igital Scholarship in the Humanities ( 2023 ). doi: 10.4 8550/arXiv.2303 . 02078 .

[7] F. Ca昀椀ero and J.-B. Camps . “ Why Molière most likely did write his plays” . InSc:ience Advances 5.11 ( 2019 ), eaax5489 . doi: 10 .1126/sciadv.aax5489.

[8] R. S. Campbell and J. W. Pennebaker . “ The Secret Life of Pronouns: Flexibility in Writing Style and Physical Health” . InP:sychological Science 14.1 ( 2003 ), pp. 60 - 65 . doi: 10 .1111 / 1467 - 9280 . 01419 .

[9] J.-B. Camps. SUPERvised STYLometry (SuperStyl). Version ... 2021 . url:https://github.co m/SupervisedStylometry/SuperSt.yl

[10] J.-B. Camps and F. Ca昀椀ero. “ Setting bounds in a homogeneous corpus: a methodological study applied to medieval literature” . IRne:vue des Nouvelles Technologies de l' Information MASHS 2011 / 2012 . Modèles et Apprentissages en Sciences Humaines et Sociales Rédacteurs invités: Mar ( 2013 ), pp. 55 - 84 . url:https://shs.hal.science/halshs-00765651.

[11] C. Chung and J. W. Pennebaker . “ The Psychological Functions of Function Words” . In: Social communication ( 2007 ).

[12] M. A. Cohn , M. R. Mehl , and J. W. Pennebaker . “ Linguistic Markers of Psychological Change Surrounding September 11 , 2001 ”. InP:sychological Science 15.10 ( 2004 ), pp. 687 - 693 . doi: 10 .1111/j.0956- 7976 . 2004 . 00741 .x.

[13] C. Crema , G. Attardi , D. Sartiano , and A. Redol昀椀. “ Natural language processing in clinical neuroscience and psychiatry: A review” . InF:rontiers in Psychiatry 13 ( 2022 ). doi: 10 .33 89/fpsyt. 2022 . 946387 .

[14] M. Eder . “Rolling stylometry”. InD: igital Scholarship in the Humanities 31.3 ( 2015 ), pp. 457 - 469 . doi: 10 .1093/llc/fqv01 0.

[15] J. C. Eichstaedt , R. J. Smith , R. M. Merchant , L. H. Ungar , P. Crutchley , D. Preoţiuc-Pietro , D. A. Asch , and H. A. Schwartz . “ Facebook language predicts depression in medical records” . In:Proceedings of the National Academy of Sciences of the United States of America 115.44 ( 2018 ), pp. 11203 - 11208 . doi: 10 .1073/pnas.1802331115.

[16] S. V. Faraone , P. Asherson , T. Banaschewski , J. Biederman , J. K. Buitelaar , J. A. RamosQuiroga , L. A. Rohde , E. J. S. Sonuga-Barke , R. Tannock , and B. Franke . “Attentionde昀椀cit/hyperactivity disorder”. In:Nature Reviews. Disease Primers 1 ( 2015 ), p. 15020 . doi: 10 .1038/nrdp. 2015 . 20 .

[17] A. Figueroa-Barra , D. Del Aguila , M. Cerda , P. A. Gaspar , L. D. Terissi , M. Durán , and C. Valderrama . “ Automatic language analysis identi昀椀es and predicts schizophrenia in 昀椀rstepisode of psychosis” . In:Schizophrenia 8.1 ( 2022 ), p. 53 . doi: 10 .1038/s41537-022-00259- 3.

[18] S. Gabay . “Beyond Idiolectometry? On Racine's Stylometric Signature” . ICno:nference on Computational Humanities Research 2021 . Amsterdam, Netherlands, 2021 , pp. 359 - 376 . url: https://hal.science/hal-03402994.

[20] E.-M. Gortner and J. W. Pennebaker . “ The Archival Anatomy of a Disaster: Media Coverage and Community-Wide Health E昀ects of the Texas A & M Bon昀椀re Tragedy ” . In: Journal of Social and Clinical Psychology 22.5 ( 2003 ), pp. 580 - 603 . doi: 10 .1521/jscp. 22.5.580.229 23.

[21] Y. HaCohen-Kerner . “Survey on pro昀椀ling age and gender of text authors” . InE:xpert Systems with Applications 199 ( 2022 ), p. 117140 .

[22] M. A. K. Halliday . Spoken and written language . [Oxford]: Oxford University Press [Oxford], 1990 .

[23] C. Ikae . “UniNE at PAN-CLEF 2022 : Pro昀椀ling Irony and Stereotype Spreaders on Twitter” . In: Clef . 2022 , pp. 1613 - 0073 . doi: https://pan.webis.de/clef22/pan22 -web/author-profili ng . html.

[24] J. Jogia , A. H. Sharif , F. A. Nawaz , A. R. Khan , R. H. Alawami , M. A. Aljanahi , and M. A. Sultan . “ Comorbidities Associated With Attention-De昀椀cit/Hyperactivity Disorder in Children and Adolescents at a Tertiary Care Setting” . IGnl:obal Pediatric Health 9 ( 2022 -02-20), p. 2333794x221076607. doi: 10 .1177/2333794x221076607.

[25] V. Johansson . “ Lexical diversity and lexical density in speech and writing” . WIno:rking papers / Lund University, Department of Linguistics and Phonetics 53 ( 2008 ). url: https://j ournals.lub.lu.se/LWPL/article/view/22.73 M. Kastberg Sjöblom. “L'indice pronominal est-il encore d'actualité ?”LIenx:icometrica 5 ( 2004 ). url: http://lexicometrica.univ-paris3.fr/article/numero5.h. tm M. Kestemont . “ Function Words in Authorship Attribution: From Black Magic to Theory?” In: Proceedings of the 3rd Workshop on Computational Linguistics for Literature (CLFL) . Gothenburg, Sweden, 2014 , pp. 59 - 66 . doi: 10 .3115/v1/ W14 -0908.

[28] K. Kim and C. H. Lee . “ Distinctive Linguistic Styles in Children with ADHD” . IPns:ychological Reports 105.2 ( 2009 ), pp. 365 - 371 . doi: 10 .2466/pr0.105.2. 365 - 371 .

[29] K. Kim , S. Lee , and C. Lee . “ College Students With ADHD Traits and Their Language Styles” . In: Journal of Attention Disorders 19.8 ( 2015 ), pp. 687 - 693 . doi: 10 .1177/1087054 713484512.

[26] [27] [30] [31] M. Koppel , J. Schler , and S. Argamon . “ Computational methods in authorship attribution” . In: Journal of the American Society for Information Science and Technology 60.1 ( 2009 ), pp. 9 - 26 . doi: 10 .1002/asi.20961.

A. Leis , F. Ronzano , M. M. A. , L. I. Furlong , and F. Sanz . “ Detecting Signs of Depression in Tweets in Spanish: Behavioral and Linguistic Analysis” . InJo:urnal of Medical Internet Research 21.6 ( 2019 ), e14199 . doi: 10 .2196/14199.

[32] M. Manabe , K. Liew , S. Yada , S. Wakamiya , and E. Aramaki. “ Estimation of Psychological Distress in Japanese Youth Through Narrative Writing: Text-Based Stylometric and Sentiment Analyses” . In:JMIR Formative Research 5.8 ( 2021 ), e29500 . doi: 10 .2196/29500.

[33] R. McCarthy and J. O'Sullivan . “Who wrote wuthering heights?” IDn:igital Scholarship in the Humanities 36.2 ( 2021 ), pp. 383 - 391 . doi: 10 .1093/llc/fqaa03 1.

[34] M. R. Mehl . “ The Lay Assessment of Subclinical Depression in Daily Life” . IPns:ychological Assessment 18.3 ( 2006 ), pp. 340 - 345 . doi: 10 .1037/ 1040 - 3590 . 18 .3.340.

[35] P. Modaresi , M. Liebeck , and S. Conrad . “ Exploring the e昀ects of cross-genre machine learning for author pro昀椀ling in PAN 2016”. InC: omponents of an Automatic Single Document Summarization System in the News Domain ( 2017 ), pp. 100 - 107 . url: https://ceurws.org/Vol- 1609 /16090970.pd.f

[36] C. Muller . “Sur quelques scènes de Molière, essai d'un indice du style familierL”a.nIng:ue française et linguistiques quantitatives . Genève: Slatkine , 1979 , pp. 107 - 124 .

[37] M. L. Newman , D. S. Pennebaker James W. and Berry , and J. M. Richards . “Lying Words: Predicting Deception from Linguistic Styles” . InP:ersonality and Social Psychology Bulletin 29.5 ( 2003 ), pp. 665 - 675 . doi: 10 .1177/0146167203029005010.

[38] M. L. Newman , Pennebaker, D. S. James W. Berry , and J. M. Richards . “Lying Words: Predicting Deception from Linguistic Styles” . InP:ersonality and Social Psychology Bulletin 29.5 ( 2003 ), pp. 665 - 675 . doi: 10 .1177/0146167203029005010.

[39] J. Noecker Jr , M. Ryan , and P. Juola . “ Psychological pro昀椀ling through textual analysis” . In: Literary and Linguistic Computing 28.3 ( 2013 ), pp. 382 - 387 . doi: 10 .1093/llc/fqs07 0.

[40] J. Ooms . hunspell: High-Performance Stemmer , Tokenizer, and Spell Checker . 2023 . url: https://docs.ropensci.org/hunspe.ll

[41] F. Pedregosa , G. Varoquaux , A. Gramfort , V. Michel , B. Thirion , O. Grisel , M. Blondel , P. Prettenhofer , R. Weiss , V. Dubourg , J. Vanderplas , A. Passos , D. Cournapeau , M. Brucher , M. Perrot , and E. Duchesnay. “ Scikit-learn: Machine Learning in Python” . IJno:urnal of Machine Learning Research 12 ( 85 2011 ), pp. 2825 - 2830 . url: https://jmlr.org/papers/v1 2/pedregosa11a.html.

[42] J. W. Pennebaker . The Secret Life of Pronouns: What Our Words Say About Us . New York: Bloomsbury Press, 2013 .

[43] J. W. Pennebaker and L. A. King . “ Linguistic styles: Language use as an individual difference” . In:Journal of Personality and Social Psychology 77.6 ( 1999 ), pp. 1296 - 1312 . doi: 10 .1037/ 0022 - 3514 . 77 .6.1296.

[44] J. W. Pennebaker and T. C. Lay . “ Language use and personality during crises: Analyses of Mayor Rudolph Giuliani's press conferences” . IJno:urnal of Research in Personality 36.3 ( 2002 ), pp. 271 - 282 . doi: 10 .1006/jrpe. 2002 . 2349 .

[45] J. W. Pennebaker and D. M. Pronoun use and dominance . Austin, TX: Department of Psychology, University of Texas at Austin, 2006 .

[46] J. W. Pennebaker and L. D. Stone . “ Words of wisdom: Language use over the life span” . In: Journal of Personality and Social Psychology 85.2 ( 2007 ), pp. 291 - 301 . doi: 10 .1037/00 22- 3514 . 85 .2.291.

[47] A. Pérez , J. Parapar , and À. Barreiro. “ Automatic depression score estimation with word embedding models” . In:Arti昀椀cial Intelligence in Medicine 132 ( 2022 ), p. 102380 . doi: 10.1 016/j.artmed. 2022 . 102380 .

[48] S. M. Redmond . “ Conversational pro昀椀les of children with ADHD, SLI and typical development” . In: Clinical Linguistics & Phonetics 18.2 ( 2004 ), pp. 107 - 125 . doi: 10 .1080/0269 9200310001611612.

[49] N. Rezaii , E. Walker, and P. Wol昀. “ A machine learning approach to predicting psychosis using semantic density and latent content analysis” . Inn:pj Schizophrenia 5.1 ( 2019 ), p. 9 . doi: 10 .1038/s41537-019-0077-9.

[50] S. Rude , E.-M. Gortner , and J. W. Pennebaker . “ Language use of depressed and depressionvulnerable college students” . InC:ognition and Emotion 18.8 ( 2004 ), pp. 1121 - 1133 . doi: 10 .1080/02699930441000030.

[51] N. Salari , H. Ghasemi , N. Abdoli , A. Rahmani , M. H. Shiri , A. H. Hashemian , H. Akbari , and M. Mohammadi . “ The global prevalence of ADHD in children and adolescents: a systematic review and meta-analysis” . In:Italian Journal of Pediatrics 49.1 ( 2023 -04-20), p. 48 . doi: 10 .1186/s13052-023-01456-1.

[52] C. Schmidt-Petri , M. Schefczyk , and L. Osburg . “Who Authored On Liberty? Stylometric Evidence on Harriet Taylor Mill's Contribution” . UInti:litas 34.2 ( 2022 ), pp. 120 - 138 .

[53] E. Segev, ed. A昀aires de style : du cas Molière à l'a昀aire Grégory, la stylométrie mène l'enquête . Paris: Le Robert, 2022 .

[54] J. A. Simmons and K. M. Antshel. “Bullying and Depression in Youth with ADHD: A Systematic Review” . In: Child & Youth Care Forum 50.3 ( 2021 ), pp. 379 - 414 . doi: 10 .1007 /s10566-020-09586-x.

[55] J. A. Singer and P. S. Blagov.Classi昀椀cation System & Scoring Manual for Self-De昀椀ning Memories . New London, CT: Connecticut College, 2022 . urlh: ttp://www.self -defining memories . com/Classification%5C%5FSystem%5C%5F%5C%5F%5C%5FScoring%5C%5 FManual%5C%5Ffor%5C%5FSDMs.pd.f

[56] L. D. Stone and J. W. Pennebaker . “ Trauma in Real Time: Talking and Avoiding Online Conversations About the Death of Princess Diana” . InB: asic and Applied Social Psychology 24.3 ( 2002 ), pp. 173 - 183 . doi: 10 .1207/s15324834basp2403\_ 1 .

[57] A. R. Sutin and R. W. Robins . “ When the “I” looks at the “Me”: Autobiographical memory, visual perspective, and the self” . InC:onsciousness and Cognition 17.4 ( 2008 ), pp. 1386 - 1397 . doi: 10 .1016/j.concog. 2008 . 09 .001.

[58] S. X. Tang , R. Kriz , S. Cho , S. J. Park , J. Harowitz , R. E. Gur , M. T. Bhati , D. H. Wolf , J. Sedoc , and M. Y. Liberman . “ Natural language processing methods are sensitive to subclinical linguistic di昀erences in schizophrenia spectrum disorders” . nInp:j Schizophrenia 7.1 ( 2021 ), p. 25 . doi: 10 .1038/s41537-021-00154-3.

[59] P. Tchounwou . “Environmental Research and Public Health”. In:ternational Journal of Environmental Research and Public Health 1.1 ( 2004 ), pp. 1 - 2 . doi: 10 .3390/ijerph200401 0001.

[60] A. Thorne and K. C. McLean . Manual for Coding Events in Self-De昀椀ning Memories . Santa Cruz: University of California, Santa Cruz, 2001 . urhlt:tp://www.self -definingmemorie s . com/Thorne%5C%5F%5C%5F%5C%5FMcLean%5C%5FSDM%5C%5FScoring%5C%5FMa nual.pdf.

[61] J. K. Ward , F. Ca昀椀ero , R. Fretigny , J. Colgrove , and V. Seror . “ France's citizen consultation on vaccination and the challenges of participatory democracy in health” . SIonc:ial Science & Medicine 220 ( 2019 ), pp. 73 - 80 . doi: 10 .1016/j.socscimed. 2018 . 10 .032.

[62] Weintraub . Verbal Behavior in Everyday Life . New York: Springer, 1989 .

[63] P. J. Yoder . “ Predicting Lexical Density Growth Rate in Young Children With Autism Spectrum Disorders” . InA:merican Journal of Speech-Language Pathology 15.4 ( 2006 ), pp. 378 - 388 . doi: 10 .1044/ 1058 - 0360 ( 2006 /035).

[64] S. Young. ADHD Child Evaluation (ACE), A diagnostic interview of ADHD in children . London: Psychology Services Limited, 2015 . urlh:ttps://www.psychology-services.uk .com/adhd.

[65] T. Zhang , A. M. Schoene , S. Ji , and S. Ananiadou . “ Natural language processing applied to mental illness detection: a narrative review” . Innp:j Digital Medicine 5.1 (2022-04-08) , pp. 1 - 13 . doi: 10 .1038/s41746-022-00589-7.