Langage et communicationEnjeux et impacts sociétaux3-4 septembre 2019Campus Biotech

Livret des résumésBook of abstracts

RÉSEAU THÉMATIQUE LANGAGE & COMMUNICATION

Le Réseau Thématique Langage & Communication est né de l’initiative du rectorat en 2013, afin de rendre visibles les recherches au sein de l’Unige portant sur le langage et la communication. En 2015, huit projets ont été soutenus, regroupés en trois axes thématiques: neurosciences cognitives, discours et communication, linguistique théorique et informatique.

Inauguré en octobre 2015 avec la grande conférence de Stephen Anderson (The place of human language in animal communication), le Réseau Thématique Langage & Communication a organisé par la suite deux jour-nées en septembre 2016 et 2017, permettant la mise en commun, par les présentations de chercheurs, des avancées des projets et des recherches dans le domaine.

Le colloque de clôture du Réseau Thématique Langage & Communication a pour ambition de faire le bilan des recherches sur le langage et la communication menées à l’Unige, mais sera aussi l’occasion unique de présenter au public académique et non académique de grands chercheurs travaillant sur le langage et/ou la communication: Dan Sperber (pragmatique, anthropologie cognitive), Simona Pekarek Doehler (linguistique interactionnelle), Philippe Ravaud (e-santé), Hagit Borer (syntaxe et sciences cognitives), Judit Gervain (neu-rosciences cognitives) et Klaus Zuberbühler (éthologie et cognition comparée).

3 SEPTEMBRE 2019 | AUDITORIUM H8-01, CAMPUS BIOTECH9h Ouverture Yves Flückiger (recteur), Jacques Moeschler (bureau L&C)9h30 What difference does ostension make to communication? What difference does language make to ostensive communication? Dan Sperber (CNRS, Institut Jean Nicod, Paris & Central European University, Budapest)10h30 Pause café11h Présentation des projets de recherche du Réseau thématique L&C12h Déjeuner14h Interactional competences of the youth: From school to the workplace Simona Pekarek Doehler (Université de Neuchâtel)15h SPIN et embellissement de la présentation des résultats en Médecine Philippe Ravaud (Université Paris Descartes & Mailman School of Public Health, Columbia University, New York)16h Pause café16h30 The generative word Hagit Borer (Queen Mary University of London)

4 SEPTEMBRE 2019 | AUDITORIUM H8-01, CAMPUS BIOTECH9h30 Speech perception abilities at birth and beyond: how prenatal experience shapes speech perception and language development Judit Gervain (INCC, CNRS & Université Paris Descartes)10h30 Pause café11h Communications orales: 3 sessions parallèles (Auditorium H8-01, Salles H8-01-D (144.165) et H4-02-A, bâtiment H4, 2e étage)13h Déjeuner-buffet et session poster (Campus Biotech, Hall H8)15h TBA Klaus Zuberbühler (Université de Neuchâtel)16h Table ronde de clôture Jean-Paul Bronckart, Luigi Rizzi, Éric Wehrli, Laurent Filliettaz, Jacques Moeschler

Inscription (gratuite mais obligatoire): David.Blunier@unige.ch

Communications 2Linguistique théorique et informatique . . . . . . . . . . . . . . . . . . . 3Discours, communication, didactique . . . . . . . . . . . . . . . . . . . 10Neurosciences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

Posters 24

Langage et communication

Groupe de Pilotage : Jan Blanc, Henri Bounameaux, Christine Chappuis, FernandoPrieto-Ramos, Pascal Zesiger

Groupe opérationnel : Patrick Badillo, Laurent Filliettaz, Anne-Lise Giraud, JacquesMoeschler

Coordination : David BlunierDavid.Blunier@unige.ch

lang-com.unige.ch

Conception graphique : Grégory Rorher, David Blunier

CommunicationsTalks

Mercredi 4 septembre, 11h-13h

Linguistique théorique etinformatique

Auditorium H8-01

Super-semantics and Super-Pragmatics for Affective Mea-ningConstant Bonard

Causal connectives at theDiscourse-Grammar InterfaceHasmik Jivanyan

Preventing undesirablethoughts in artificial intelli-genceAlexandre Kabbach

Sentence embeddings andlong-distance dependencieseffects in French and EnglishPaola Merlo

Development of A Fully Auto-mated Conversation ArtificialIntelligence as Patient Deci-sion Support System to Ac-company Self-Management ofAdults and Young Adults withSickle-Cell DiseaseDavid-Zacharie Issom, KavehSamii, Jessica Rochat, GunnarHartvigsen, Christian Lovis

Automatic translation ofFrench medical text intoSNOMED-CTChristophe Gaudet-Blavignac,Vasiliki Foufi, Eric Wehrli,Christian Lovis

Discours,communication,

didactiqueSalle H8-01-D

Favoriser l’intégration entrethéorie et pratique par l’écri-ture chez les futurs ensei-gnants?Kristine Balslev et Sandra Pel-landa Dieci

La fabrication des savoirs surle langage en interaction : re-gards croisés sur les pratiquesde la formation professionnelledes éducatrices de l’enfanceLaurent Filliettaz et MarianneZogmal

Pratiques des enseignantsen formation en classe delangue étrangère : leurs gestesdidactiques et l’alternancedes langues au service de laconstruction des savoirsSlavka Pogranova

Analyser le travail dans les mé-tiers d’éducation au prisme dela notion d’ostensionMarianne Zogmal

Communications Networksand Knowledge Management :the innovation potential of So-cial MediaAngela Bacchetta Beckh

L’information de santé franco-phone dans la presse et sur lesmédias sociaux : le cas de laprévention et du contrôle desinfectionsKaren Pelletier

NeurosciencesSalle H4-02-A

Bilingualism and Cognition inchildren with Autism SpectrumDisorderStephanie Durrleman, EleniBaldimtsi, Eleni Peristeri, Ian-thi Tsimpli

Evidence for structural learningin a Lindenmayer grammarSamuel Schmid, Diego Krivo-chen, Doug Saddy, Julie Franck

The sound of salience : howroughness enhances aversionthrough neural synchroniza-tionLuc H. Arnal, Andreas Klein-schmidt, Laurent Spinelli,Anne-Lise Giraud, Pierre Mé-gevand

Characterizing Heschl’s gyrusshape by a Multivariate Conca-vity Amplitude IndexJosué Luiz Dalboni da Rocha,Jan Benner, Peter Schneider,Narly Golestani

Cortical encoding of speechenhances task-relevant acous-tic informationSanne Rutten, Roberta Santoro,Alexis Hervais-Adelman, EliaFormisano, Narly Golestani

Linguistique théorique etinformatique

Theoretical and computational Linguistics

Super-semantics and Super-Pragmatics for Affective Meaning

Constant Bonard, University of Geneva

To study precisely how meaningworks among systems of signs suchas dance, music, gestures, nonhumancommunication, or pictorial represen-tation, Philippe Schlenker (forthcoming)has proposed that we develop what hecalls “super-semantics”, because regulartruth-conditional semantics is not adap-ted to these systems. Since regular prag-matics (neo- or post-Gricean) also failto fit how meaning is used in many ins-tances of communication using these si-gns, a super-pragmatics also should bedeveloped. In this presentation, I wouldlike to make similar points regardingaffective meaning, i.e. the informationfunctionally carried by affective signs(e.g. facial, vocal, or bodily expression ofemotions) :

1. Regular semantics is not adap-ted to affective meaning becausethe latter inherits its content fromthat of affects, a content which isnot structured propositionally. Byconsequence, affective meaningcannot be accounted by a com-positional, truth-conditional se-mantics.

2. Regular pragmatics is not adap-ted to affective meaning becauseaffective communication usuallydoesn’t work through the overtdisplay of communicative inten-tions, which is a central assump-tion for regular (neo- or post-Gricean) pragmatics.

3. We can use data and theories de-veloped in affective sciences as astrong basis to elaborate a super-semantics for affective meaning.In particular, even though affec-tive content is not literally trueor false, affects always involve acognitive content which can beassessed as fit or unfit to the si-tuation.

4. A super-pragmatics for affec-tive meaning might be develo-ped on the basis of such a super-semantics.

References Schlenker, P. (forthcoming). Whatis super semantics?. To appear in Philosophicalperspective.

4

Causal connectives at the Discourse-Grammar Interface

Hasmik Jivanyan, University of Geneva

Subordination and coordination areissued from traditional grammar, butthey are used in modern literature ina variety of (mutually related) sensesand dimensions, such as at syntactic,semantic and discourse-pragmatic level(Bluhdorn 2008). Moreover, often theyare used as parallel to another termino-logical pair, namely, hypotaxis and para-taxis, such that hypotaxis refers to syn-tactic subordination, while parataxis isthe coordination. The aim of this talk is toshow that although these intuitions areoften correct and there can be such pa-rallels, the two terminological pairs donot denote the same phenomena, andthey do not always match together : theycan cross-cut and result in more thantwo cases than simply subordination =hypotaxis, coordination = parataxis. Ourfirst goal is to show the distinction andthe correlations between these two di-mensions.

Second, Subordination implies somenotions such as asymmetry or hierarchy,which have been clearly demonstratedand formalized in different theoretical(syntactic) models. Moreover, Subordi-nation is often associated with hetero-geneous communicative weight or pro-minence (salience, importance) of theclauses. This later aspect is far frombeing explicitly and deeply investigatedwithin any theoretical model, and my re-search aims at capturing these intuitionsin a unified way.

The third and ultimate goal is to showhow can different subordinated vs co-ordinated, as well as hypotactic vs. pa-ratactic clauses be described in termsof discourse-level information structu-

ral terms such as (not-)at-issueness. Thelatter captures the intuitions about clau-ses’ prominence or importance withinthe model based on the notion of Ques-tion under Discussion (QUD, Roberts1992).

The analyses will be based on, butnot limited to, the examples of causalconnectives (parce que, puisque, car) inFrench, mainly from newspaper corpus.

ReferencesBluhdorn 2008 Subordination and coordi-

nation in syntax, semantics and discourse :Evidence from the study of connectives. InFabricius-Hansen, Cathrine & Ramm, Wiebke(eds.), ‘Subordination’ versus ‘Coordination’ inSentence and Text. Amsterdam, Benjamins,2008, pp. 59-85.

Debaisieux J.-M. 2016 Toward a global ap-proach to discourse uses of conjunctions in spo-ken French. Language Sciences 58 : 79-94.Horn 2015 Information Structure and the Land-scape of (Non-)at-issue Meaning. In C. Féry, S.Ishihara (Eds.), The Oxford Handbook of Infor-mation Structure.

Jivanyan H. & G. Samo 2018 Parce que inthe syntax-semantics-pragmatics interface. Ge-nerative grammar in Geneva 10 : 77-99.

Lambrecht K. & al. 2006 Cognitiveconstraints on assertion scope : the case of spo-ken French parce que. In Nishida C. & al. (Eds.),New perspectives on Romance linguistics 1 : 143-154.

Roberts C. 2012/1996 Information structurein discourse : Towards an integrated theory ofpragmatics. Semantics and Pragmatics 6 : 1–69.

Schleppegrell M. J. 1991 Paratactic because.Journal of Pragmatics 16, 323-337.

Sweetser E. 1990 From etymology to prag-matics : metaphorical and cultural aspects ofsemantic structure. CUP.

Velleman L. & D. Beaver 2015 Question-based models of information structure. In C. Féry,S. Ishihara (Eds.), The Oxford Handbook of Infor-mation Structure. OUP : 86–107.

5

Preventing undesirable thoughts in artificial intelligence

Alexandre Kabbach, University of Geneva

Recent work on (artificial) cognitionand communication has argued that anyartificial intelligence fully capable of pro-cessing natural language is bound to di-sagree with and/or misunderstand anygiven (human) speaker at some point(Kabbach, 2019).

If disagreement over things such aspredicates of personal taste (e.g. ’fun’or ’tasty’) may be acceptable from anethical standpoint, can the same besaid of disagreement over moral issues?And if such things exist as undesirablethoughts, can we constrain artificial cog-nition to render such thoughts unlikely–ifnot impossible–while not hindering pro-per understanding of human language?

In this talk we discuss two pos-sible forms of constraints over artifi-

cial conceptual spaces (1) the reductionof the set of initial primitives that ren-ders certain concepts inaccessible andcreates de facto ’unthinkable thoughts’ ;and (2) the bias of the composition func-tion that alters the likelihood of certainconceptual composition operations.

We argue that none of those op-tions provide complete guarantees overso called ’undesirable thoughts’, in thata machine fully able to process naturallanguage may found compensating me-chanisms to generate and propagate un-desirable concepts in either case.

We conclude that only ’artificial edu-cation’, in the form of systematic expo-sure to a set of controlled stimuli, canprovide a satisfactory form of constraintover the artificial conceptual apparatus.

6

Sentence embeddings and long-distance dependencies effects inFrench and English

Paola Merlo, University of Geneva

Recent interest in recursive neuralnetworks (RNNs) has spurred detailedinvestigations whether the distributedrepresentations they generate exhibitproperties similar to those characteri-sing human languages [1]. We ask whe-ther sentence embeddings calculated byRNNs match human acceptability judg-ments in long-distance dependencies(see examples 1-4). To preview, sen-tence embeddings do not unequivo-cally encode fine-grained linguistic pro-perties of long-distance dependencies.Long-distance dependencies are not allequally acceptable or even grammatical(for example, sentences (3a,b) and (4a,b)are not fully grammatical). A prominentexplanation says that a long-distancedependency acceptability depends ontheir morpho-syntactic properties [3, 2,4] : lexical restriction, number and ani-macy. French stimuli [2, 4] and their En-glish equivalents were transformed intosentence embeddings (BoVfastText) andclassified with a multi-layer perceptroninto the four sentence types (1a,b, 2a,band 3a,b, 4a,b). Classification accuracyis the dependent variable. Weak islandsresults : for French, only the predictionon the effect of animacy in the lexicallyspecified case is confirmed; for English,

the prediction for the effect of animacy isconfirmed both in bare wh-phrases andin lexicalised wh-phrases. For object re-lative clauses, none of the predictionsin the non-control cases are confirmed,either in French or in English.

(1a,b) Julie smiles to the student/studentsthat the speaker has been seriously an-noying from the start.

(2a,b) Julie points out to thestudent/students that the speaker hasbeen yawning frequently from the start.

(3a,b) Which class/which professordo you wonder which student liked?

(4a,b) What/who do you wonderwho liked?

References[1] Belinkov and Glass. Analysis methods in

neural language processing : A survey. TACL,7 :49–72, 2019. [2] Franck et al. Task-dependencyand structure dependency in number interfe-rence effects in sentence comprehension. Fron-tiers in Psychology, 6, 2015. [3] Rizzi. Locality andleft periphery. In Belletti, The cartography of syn-tactic structures, 223–251. OUP, 2004. [4] Villataand Franck. Semantic similarity effects on weakislands acceptability. In 41st IGG, 2016.

7

Development of A Fully Automated Conversation Artificial Intelligence asPatient Decision Support System to Accompany Self-Management of

Adults and Young Adults with Sickle-Cell Disease

David-Zacharie Issoma, Kaveh Samiib, Jessica Rochata, Gunnar Hartvigsenc,Christian Lovisa

aDivision of Medical Information Sciences, Geneva University Hospitals & University of Geneva, Switzerland ; bDepartmentof Haematology, University Hospital Geneva, Switzerland ; cDepartment of Computer Science, University of Tromsø – The

Arctic University of Norway, Norway

Sickle cell disease (SCD) is the worl-d’s most common genetic disorder. Ne-vertheless, SCD remains underpriori-tized and poorly resourced. Facing alack of providers and care coordina-tion, the adult population is left to self-manage. To date, there are no Digi-tal Health Interventions (DHI) supportingthe comprehensive self-managementneeds of adults with SCD. Moreover, lit-tle is known about their preferred fea-tures (e.g. medication reminder, self-monitoring). We developed and ran asurvey on self-management needs, day-to-day challenges and app features pre-ferences. Adult patients were recruitedthrough convenience sampling. Thirty-three patients completed the survey. Tocollect more detailed information, weconducted three focus groups with fourpatients. Participants clearly demons-trated an interest in DHIs. They as-ked for DHIs that adjunct to usual care

and integrate with existing health in-formation repositories. The most recur-rent requests were assistance for self-monitoring, avoiding symptoms , sharingdata with emergency departments, re-ceive medical information and medicaladvices remotely. Patients also asked forDHIs providing trustworthy informationwrapped in a simple design. Finally, theyprioritized features and co-designed achatbot prototype. Future steps will beto evaluate the prototype’s usability. To-day, DHIs for adults with SCD are farfrom comprehensive. Our findings sug-gest that patients’ experiential know-ledge can be translated into explicit de-mands. Our participatory approach allo-wed to better understand what patientswant from DHIs and what these couldlook like. Involving patients in all deve-lopment phases can result in highly en-gaging DHIs that positively impact healthoutcomes.

8

Automatic translation of French medical text into SNOMED-CT

Christophe Gaudet-Blavignac, Vasiliki Foufi, Eric Wehrli, Christian LovisaaUniversity of Geneva

Interoperability is crucial for effec-tive sharing and reuse of clinical data.More specifically, semantic interopera-bility is required for applications suchas quality control, decision support andresearch. Initiatives to support inter-operability in Electronic Health Record(EHR) already exist such as the Mea-ningful Use Program [1] in the UnitedStates, or the work of HL7 to define tech-nical specifications for sharing medicaldocuments [2]. However a major partof data in EHR are still in free text andreside in narrative documents such asadmission notes and discharge letters.From the semantic aspect, SNOMED In-ternational publishes and maintains theSystematized Nomenclature of Medi-cine – Clinical Terms (SNOMED CT)[3],in the aim of providing a unified se-mantic resource that can be used to re-present any medical concept. SNOMEDCT concepts can be combined using acompositional grammar to create post-coordinated sentences. This feature al-lows the representation of the medi-cal language into the formal language ofSNOMED CT. To move towards seman-

tic interoperability, a method to auto-matically translate French medical sen-tences in SNOMED CT post-coordinatedsentences is presented. To perform thistranslation, a Natural Language Pro-cessing (NLP) tool has been adapted,lexico-semantic resources were crea-ted and post-coordination rules wereimplemented. Automatic annotation andpost-coordination of concepts has beenevaluated by manual validation of asample of 1500 concepts with a preci-sion of 85% and 71.2% for simple andpost-coordinated concepts respecti-vely.

References [1] C. Group, « What is the HI-TECH ACT? | What HITECH Compliance Means», Compliancy Group. [En ligne]. Disponiblesur : https ://compliancy-group.com/what-is-the-hitech-act/. [Consulté le : 12-juin-2019].[2] « Health Level Seven International - Ho-mepage | HL7 International ». [En ligne]. Dis-ponible sur : https ://www.hl7.org/. [Consultéle : 12-juin-2019]. [3] SNOMED, « SNO-MED International », 2006. [En ligne]. Dispo-nible sur : https ://www.snomed.org/news-articles/snomed-ct-compositional-grammar-specification-and-guide. [Consulté le : 06-nov-2017].

9

Discours, communication,didactique

Discourse, communication, didactics

Favoriser l’intégration entre théorie et pratique par l’écriture chez les futursenseignants?

Kristine Balslev et Sandra Pellanda Dieci, Université de Genève

Les écrits demandés aux étudiantslors de leur formation universitaire sontmultiples : résumés de texte, notes desynthèse, rapports de recherche, disser-tations, récits d’expérience, analyses depratiques, mémoires, etc. Ces écrits ontdes visées formatives variées : l’appro-priation par les étudiants de conceptset modèles théoriques, la capacité derestituer une démarche scientifique, lasensibilisation aux règles de la com-munication scientifique et la descriptionet l’objectivation d’expériences profes-sionnelles et, enfin, l’intégration entreéléments théoriques et éléments ditspratiques de la formation. Cette dernièrevisée est au cœur de cette communica-tion. Que font les étudiants lorsque l’onleur demande de partir de leurs propresexpériences ou préoccupations puis deles examiner à l’aune de la théorie? Nousnous intéressons particulièrement auxtravaux universitaires dans les forma-tions en alternance impliquant des récitsd’expériences, tout en visant l’appro-priation de savoirs scientifiques de ré-férence pour professionnaliser le formé,et s’inscrivant dans un « genre réflexif» académique (Vanhulle, 2016). A partirde l’analyse de « Dossiers de dévelop-

pement professionnel » produits dansle cadre de la formation à l’enseigne-ment secondaire par des enseignants-étudiants universitaires, nous analyse-rons les différentes les façons d’intro-duire les références scientifiques et leursrôles dans la présentation et la com-préhension d’une situation relatée. Nousétudierons également le positionnementénonciatif présent dans ces textes endistinguant deux positionnements : lepremier, épistémique, dans lequel l’au-teur avance des idées et les discute «dans une prise en charge à visée univer-salisante » (Boch, 2013) sans mentionnersa personne physique et sociale et le se-cond, déictique, où l’auteur s’exprime enJE en relatant ses expériences et sa sub-jectivité. Nos premières analyses fontémerger six manières de faire usage dela théorie dans ces textes et différentesfaçons de passer du « je épistémique »au « je déictique ». Cette présentationsera l’occasion de mettre à l’épreuve cessix catégories. Une réflexion sur le sens àdemander aux futurs professionnels de «mobiliser la théorie » conclura cette pré-sentation en évoquant les pratiques, lesobstacles que cette demande suscite etles moyens pour les dépasser.

11

La fabrication des savoirs sur le langage en interaction : regards croiséssur les pratiques de la formation professionnelle des éducatrices de

l’enfance

Laurent Filliettaz et Marianne Zogmal, Université de Genève

Le champ de la formation profes-sionnelle initiale et continue constitue unespace de médiation et de productionde savoirs, qui est à même de renou-veler et de vivifier les relations entre larecherche et les champs de pratiquessociales. Ces réalités concernent ausside manière plus spécifique les « savoirssur le langage » et la manière dont la« part langagière du travail » peut de-venir un objet de rencontre entre deschercheurs en linguistiques et des pro-fessionnels, qui, au quotidien, travaillentavec le langage (Boutet, 2001, 2008;Filliettaz, 2006). Contribuer à la forma-tion et au développement professionneldes travailleurs aux prises avec l’usagedu langage peut ainsi s’avérer un espacede médiation particulièrement produc-tif pour non plus seulement « restituer» ou « diffuser » des savoirs savantssur le langage, mais pour les élabo-rer conjointement avec les profession-nels eux-mêmes. Telles sont quelques-unes des idées générales que nous pro-posons de présenter et d’illustrer danscette communication. Dans cette contri-bution, nous proposons de centrer notreattention sur la manière dont un dispo-sitif de recherche que nous avons misen place a pu constituer une opportu-nité pour les professionnelles de l’édu-cation de l’enfance de rendre visibles

les savoirs d’expérience dont elles dis-posent sur le rôle des interactions lan-gagières dans leur travail auprès des en-fants et auprès des stagiaires qu’elles en-cadrent occasionnellement. Nous nousfocalisons en particulier sur les repré-sentations en circulation dans leurs dis-cours à propos de la place et du rôlede l’interaction verbale dans la construc-tion d’une posture professionnelle et lamanière dont les catégories utilisées àce propos émergent au fil de la ren-contre entre des chercheurs et des col-lectifs d’éducatrices. Pour ce faire, nouscommençons par rappeler les objec-tifs généraux de la démarche de re-cherche adoptée et les quelques prin-cipes à la fois théoriques et méthodo-logiques qui ont orienté notre approche.Nous proposons ensuite de soumettre àl’analyse deux entretiens conduits entrenotre équipe de recherche et des pro-fessionnelles de l’éducation de l’enfanceet de la formation professionnelle dansce domaine. Cette analyse a pour butde mettre en évidence la manière dontse fabriquent des savoirs sur le langageen interaction entre chercheurs et prati-ciens. Dans une partie conclusive, nousdiscutons de la nature des savoirs pro-duits et des conditions discursives danslesquels ils sont rendus visibles.

12

Pratiques des enseignants en formation en classe de langue étrangère :leurs gestes didactiques et l’alternance des langues au service de la

construction des savoirs

Slavka Pogranova, Université de Genève

Cette contribution propose d’analy-ser la construction des savoirs en classede langue à l’école primaire genevoise,notamment au moyen des gestes di-dactiques des enseignants et la méso-alternance. À partir d’un corpus de 15 le-çons données par des futurs enseignantsd’allemand et d’anglais, nous présen-tons les caractéristiques des gestes(Schneuwly & Dolz, 2009) et leur rôledans la construction et la transforma-tion des savoirs. Dans un deuxièmetemps, nous nous interrogeons sur laméso-alternance (Gajo, Grobet, & Stef-fen, 2015 ; Gajo & Steffen, 2015) commepratique langagière des enseignants àdes fins didactiques. Dans un troisièmetemps, nous analysons quelques ex-traits, montrant leurs particularités.

Nous nous basons sur le synopsis(Sales Cordeiro, Ronveaux, & équipeGRAFE, 2009) et le logiciel d’analysequalitative N Vivo 10. La contribution

propose quelques réflexions menant àla compréhension de la communicationen classe, renvoyant aux dispositifs deformation initiale, un des enjeux socié-taux.

RéférencesGajo, L., Grobet, A., & Steffen, G. (2015). Saisir

le niveau méso-interactionnel dans l’interactiondidactique : autour des notions de saturation etde méso-alternance. Bulletin suisse de linguis-tique appliquée, 101, 27-49.

Gajo, L., & Steffen, G. (2015). Didactique duplurilinguisme et alternance de codes : le cas del’enseignement bilingue précoce. The CanadianModern Languages Review, 71(4), 471-499.

Sales Cordeiro, G., Ronveaux, C., & équipeGRAFE. (2009). Le recueil et le traitement desdonnées. In B. Schneuwly & J. Dolz (Eds.), Desobjets enseignés en classe de français (pp. 83-100). Rennes : Presses universitaires de France.

Schneuwly, B., & Dolz, J. (2009). Des objetsenseignés en classe de français : le travail de l’en-seignant sur la rédaction de textes argumentatifset sur la subordonnée relative. Rennes : Pressesuniversitaires de Rennes.

13

Analyser le travail dans les métiers d’éducation au prisme de la notiond’ostension

Marianne Zogmal, Université de Genève

Dans les métiers de la prise en charged’autrui, les pratiques professionnellesse déploient dans et à travers le lan-gage. Une analyse des phénomènes lan-gagiers, et plus largement interaction-nels, constitue ainsi une entrée pourmieux comprendre les spécificités etles enjeux d’un contexte professionnel.A travers un exemple empirique issudu champ de l’éducation de l’enfance,cette contribution aborde les proces-sus qui permettent aux professionnelleset aux enfants d’attribuer des significa-tions aux énoncés et aux conduites in-teractionnelles des participants. Les in-teractions avec de jeunes enfants, quisont encore en train d’acquérir l’accèsau langage verbal, rendent visibles com-ment les phénomènes d’ostension s’ins-crivent dans la séquentalité du dérou-lement interactionnel (Sperber & Wil-son, 1989). Le travail d’interprétation in-férentielle consiste à chercher de com-prendre le vouloir-dire d’un interactant.Les significations attribuées peuvent en-suite être rendues mutuellement mani-festes et partageables dans le contextelocal de l’interaction. Elles constituentalors des hypothèses de compréhen-

sion qui peuvent s’insérer dans un travaild’enquête afin de construire des signi-fications partagées par l’ensemble desinteractants. Les données empiriques,consistant en des interactions filmées ettranscrites dans une perspective multi-modale (Kress et al., 2001) montrent quel’ostension des compréhensions élabo-rées porte parfois sur la surface linguis-tique des énoncés, mais surtout sur leurdimension pragmatique. Dans un collec-tif de travail, les praticiens rendent éga-lement manifestes leurs observationsportant sur les compétences des en-fants, et notamment les compétenceslangagières. Ces phénomènes langagierset interactionnels sont fugitifs et imbri-qués dans le déroulement des interac-tions multimodales. Leur mise en visibi-lité à travers une analyse fine des inter-actions permet de contribuer à une des-cription des pratiques professionnellesréalisées et à identifier les compétencesinteractionnelles déployées. Par ailleurs,une telle démarche d’analyse transposel’usage de la notion d’ostension en tantque catégorie d’analyse pour étudier descontextes professionnels réels, en priseavec des questions sociales actuelles.

14

Communications Networks and Knowledge Management : the innovationpotential of Social Media

Angela Bacchetta Beckh, University of Geneva

The purpose of this research is toexamine the relationship between in-formation and knowledge created bySocial Media and related knowledgemanagement for innovation purposes.Knowledge and knowledge manage-ment happens within social constructswhich illustrates the importance of thesocial aspects within social media’s im-pact for transforming data into know-ledge and related epistemological rup-ture.

Method. Literature review and chro-nological approach combined with ananalysis of over 1000 quantitative sur-vey results on social media’s perceivedusefulness for innovation purposes bysenior practitioners and innovation pro-fessionals as well as over 100 qualitativesurveys.

Results. The findings illustrate so-cial media’s impact on the innovationprocess as these knowledge networkscreate a new form of information andknowledge. These findings are discus-sed within the context of the digital re-volution and new forms of communi-cations and knowledge management.The innovation potential of these digi-

tal communication networks are closelylinked to the digital maturity of enter-prises/organizations as well as the digi-tal work identity of knowledge workers.Studying the innovation potential of so-cial media also contributes to some ofthe key questions that the future of thisdigital revolution holds, especially thefuture of work and related innovationecosystems.

Research limits/implications. Thisstudy analyses the perceived useful-ness concerning social media and stron-ger use for innovation purposes. Howe-ver even more quantitative data couldconfirm the observations in relationshipto innovation and social media, espe-cially in less researched regions.

Practical implications. The work canencourage policy makers, innovationprofessionals and HR managers to le-verage social media to foster innovation.

Originality/value. The connectionsbetween social media, innovation, di-versity and discourse have so far beenunder-researched. This work pinpointsthe opportunities of social media as in-novation communications networks.

15

L’information de santé francophone dans la presse et sur les médiassociaux : le cas de la prévention et du contrôle des infections

Karen Pelletier, Université de Genève

Introduction. L’analyse quantitativede la presse et des médias sociauxdans le domaine de la prévention et ducontrôle des infections prend en comptedes critères temporels, langagiers etgéographiques. Au niveau temporel,l’analyse de la presse mesure l’évolutiondes publications de 2010 à 2017. L’ana-lyse des médias sociaux mesure les pu-blications réalisées sur une année, com-prise entre 2017 et 2018. Au niveau lan-gagier, l’étude se focalise sur le contextefrancophone. Au niveau géographique,elle se centre sur la France et la SuisseRomande, le Maghreb ainsi que sur lespays francophones du reste de l’Afrique.

Méthodes. L’analyse de la presses’appuie sur la base de données Nexisqui regroupe plus de 36’000 sourcesde presses internationales. L’analyse desmédias sociaux est effectuée à l’aide del’outil de social listening, Search. A l’aidede ces deux outils, les statistiques sur laprésence des mots-clés : santé et pré-

vention, prévention et infection, infec-tion, épidémie, santé publique, sont re-cueillis.

Résultats. Dans la presse, en com-parant les résultats des régions fran-cophones du Nord à celles du Sud,nous observons que le nombre d’ar-ticles consacrés à la santé, la préventionainsi que la prévention des infections, estde l’ordre de 3,5 fois plus faible pour leMaghreb que pour la France, et près de 5fois plus faible pour le reste de l’Afriquefrancophone que pour la France. Sur lesmédias sociaux, en plus de cette fractureinformationnelle forte, nous observonsune disparité notable dans l’usage desmédias sociaux concernant la diffusionde l’information de santé.

Conclusion. Cette recherche a per-mis de démontrer l’existence d’une frac-ture informationnelle forte dans le do-maine de la santé publique du mondefrancophone.

16

Neurosciences

Bilingualism and Cognition in children with Autism Spectrum Disorder

Stephanie Durrleman, Eleni Baldimtsi, Eleni Peristeri, Ianthi Tsimpli

Executive functions (EF) have beenclaimed to influence Theory of Mind(ToM) in both typically developing chil-dren (TD) and those with Autism Spec-trum Disorders (ASD) [1,2]. Weaknessesin EF and ToM are associated with ASD[2-5]. Bilingualism arguably boosts EF [6]and ToM [7-8] in TD children and maymodulate EF in children with ASD ac-cording to preliminary findings [9], ho-wever it is still unknown if bilingualismimproves ToM in ASD and if such impro-vement would relate to EF.

Method. Sixty-eight children withASD participated : 34 bilinguals (ASDbi)(Mean age : 11 ;7) and 34 monolinguals(ASDmono), matched on age (Meanage : 11 ;7) and nonverbal intelligence(Range 73-120) [10]. Bilinguals were ba-lanced Albanian-Greek speakers. ToMwas measured via an online, non-verbalfalse-belief task [11], and EF via both anon-verbal task of attentional control(attending at the global or local levelin congruent and incongruent trials) [12]and a 2 digits-back Working memory(WM) task. Accuracy and reaction timeswere recorded.

Results. ASDbi children were moreaccurate at focusing attention on theglobal level than ASDmono (p<.04).ASDbi exhibited higher accuracy thanmonolinguals in the 2-back task (p<.02).ASDbi were more accurate than ASD-mono on ToM (p<.001), while there wereno significant RT differences betweengroups (p=.599). Regression analysesshowed that accuracy of ASDbi on ToMincreased with accuracy on the 2-backtask (Beta=.476, p=.01). ToM in ASDmonodid not interact with any EF measure.

Discussion. Bilingualism boosts ToM

and EF in ASD. A stronger relationshipemerged between WM and ToM in bi-lingual children with ASD, in contrast toASDmono children whose ToM perfor-mance did not seem to draw on eitherWM or attention. The overall findingssuggest that ToM in bilinguals and mo-nolinguals with ASD draws on differentresources and highlight advantages forASDbi children over ASDmono for EFand ToM.

References [1] Flynn, E. (2007). The Role ofInhibitory Control in False-Belief Understanding.Infant and Child Development 16, 53 – 69. [2] Pel-licano, E. (2007). Links between theory of mindand executive function in young children withautism : Clues to developmental primacy. Deve-lopmental Psychology, 43 (4), 974-990. [3] Pen-nington, B.F., Rogers, S.J., Bennetto, L., Griffith,E.M., Reed, D.T., & Shyu, V. (1997). Validity testsof the executive dysfunction hypothesis of au-tism. In J. Russell (Ed.), Autism as an executivedisorder (pp. 143–178). Oxford : Oxford Univer-sity Press. [4] Pellicano, E. (2010). Individual dif-ferences in executive function and central cohe-rence predict developmental changes in theoryof mind in autism. Developmental Psychology,46, 530–544. doi :10.1037/a0018287 [5] Ozonoff,S., & Jensen, J. (1999). Specific executive func-tion profiles in three neurodevelopmental disor-ders. Journal of Autism and Developmental Di-sorders, 29 (2), 171-177. [6] Bialystok, E. (2010).Bilingualism. Wiley interdisciplinary reviews :Cognitive science, 1 (4), 559-572 [7] Diaz, V. &Farrar, M. J. (2017). The missing explanation ofthe false-belief advantage in bilingual children :a longitudinal study. Developmental Science,https ://doi.org/10.1111/desc.12594 [8] Nguyen,

18

T.K., & Astington, J.W. (2014). Reassessing the bi-lingual advantage in theory of mind and its cog-nitive underpinnings. Bilingualism : Language andCognition, 17, 396–409. [9] Gonzalez-Barrero,A. M., & Nadig, A. S. (2017). Can Bilingua-lism Mitigate Set-Shifting Difficulties in ChildrenWith Autism Spectrum Disorders? Child Deve-lopment, https ://doi.org/10.1111/cdev.12979 [10]Georgas, J., Paraskevopoulos, I., Besevegis, E.,

& Giannitsas, N. (1998). Greek Wechsler Intel-ligence Scale for Children. WISC-III : Manual.Athens : Ellinika Grammata. [11] Forgeot d’Arc, B.,& Ramus, F. (2011). Belief attribution despite ver-bal interference. Quarterly Journal of Experimen-tal Psychology, 64, 975-990. [12] Navon, D. (1977).Forest Before Trees : The Precedence of globalfeatures in visual perception. Cognitive Psycho-logy, 9 (3), 353-383.

19

Evidence for structural learning in a Lindenmayer grammar

S. Schmida, D. Krivochenb, D. Saddyb, J. FranckaaUniversity of Geneva ; bUniversity of Reading, UK

Our research investigates if the hu-man parser is sensitive to the under-lying structure governing the simplestLindenmayer grammar [1] consisting oftwo rewrite rules : 01, 101. This grammaris called Fibonacci (Fib) because at eachgeneration the number of symbols (0sand 1s) follows the Fibonacci sequence(0, 1, 1, 2, 3, 5, 8,. . . ). The analysis ofthe formal properties of Fib-generatedstrings shows that whereas some sym-bols are fully ambiguous and thus can-not be linearly anticipated, others (cal-led k-points) can be predicted if theparser relies on the hierarchical struc-ture of this grammar [2]. In three expe-riments we explored whether humansare able to learn Fib and more particu-larly to track kpoints. We used a serialreaction time paradigm in which partici-pants had to press as quickly as possiblethe button corresponding to the sym-bol on the screen. In experiment 1, par-ticipants learned the Fib grammar. Thelearning slope of k-points was signifi-cantly steeper than that of ambiguoussymbols. However, since k-points weremore frequent than ambiguous symbols,this advantage may reflect the trackingof surface statistical regularities rather

than structural learning. Experiment 2used the Skip grammar (001 ; 101101) inwhich the statistical distribution of k-points was reversed such that they wereless frequent than ambiguous symbols.The learning slopes were reversed, sho-wing a significant advantage for am-biguous symbols. Experiment 3 testedhow participants switch from learningone grammar in the first part of the ex-periment (Fib or Skip) to learning theother grammar in the second part. Dif-ferent curves were observed for the twogroups, suggesting that unlearning Fib ismore difficult than unlearning Skip. Thisfinding suggests that the advantage fork-points in Experiment 1 cannot be at-tributed only to the tracking of surfacestatistics, and that the parser extracts, tosome extent, the underlying structure ofthe grammar.

References [1] Lindenmayer, A. (1968). Mathe-matical models for cellular interactions in deve-lopment II. Simple and branching filaments withtwo-sided inputs. Journal of Theoretical Biology,18, 300-315. doi :10.1016/0022-5193(68)90080-5 [2] Krivochen, D., Phillips, B., & Saddy, D.(2018). Classifying points in Lindenmayer sys-tems : Transition probabilities and structure re-construction [Unpublished work]. University ofReading. doi :10.13140/RG.2.2.25719.88484

20

The sound of salience : how roughness enhances aversion through neuralsynchronization

Luc H. Arnala, Andreas Kleinschmidtb, Laurent Spinellib, Anne-Lise Girauda, PierreMégevanda,b

aDepartment of Fundamental Neuroscience, University of Geneva ; bDepartment of Clinical Neuroscience ; University ofGeneva.

Being able to produce sounds thatcapture attention and elicit rapid reac-tions is the prime goal of communica-tion. One strategy, exploited by alarmsignals, consists in emitting fast butperceptible amplitude modulations inthe roughness range (30-150 Hz). Here,we investigate the perceptual and neu-ral mechanisms underlying aversion tosuch salient sounds. By measuring sub-jective aversion to repetitive acoustictransients, we identify a non-linear pat-tern of aversion restricted to the rough-ness range. Using intracranial recor-dings, we show that rough sounds do

not merely affect local auditory pro-cesses but instead synchronise large-scale, supramodal, salience-related net-works in a steady-state, sustained man-ner. More specifically, rough soundssynchronise activity throughout superiortemporal regions, subcortical and cor-tical limbic areas, and the frontal cor-tex, a network classically involved inaversion processing. This pattern cor-relates with subjective aversion in allthese regions, consistent with the hypo-thesis that roughness enhances auditoryaversion through the spreading of neuralsynchronisation.

21

Characterizing Heschl’s gyrus shape by a Multivariate ConcavityAmplitude Index

Josué Luiz Dalboni da Rochaa, Jan Bennerb, Peter Schneiderb, and Narly GolestaniaaBrain and Language Lab, Department of Psychology, University of Geneva ; bDepartment of Neuroradiology, University of

Heidelberg Medical School, Heidelberg, Germany.

Heschl’s gyrus (HG), which includesprimary auditory cortex, displays consi-derable variation in size and shapeacross individuals and hemispheres. Themost common gyrification patterns in-clude single HG, common stem duplica-tions (CSD) and complete posterior du-plications (CPD). To date, characteriza-tion of HG gyrification patterns has beenbased on the visual identification of pos-sible existing sulci, and reproducible cri-teria for defining sulci have not beenestablished. Techniques for segmentingthe HG already exist, such as TASH andFreesurfer. However, there is no conti-nuous measure for specifically asses-sing the shape of HG. Here, we presenta fully automated novel method calledMultivariate Concavity Amplitude Index(MCAI) for characterizing the shape ofHG. MCAI uses continuous indices of theconcavity (i.e. inward displacements) ofdifferent parts HG, taking into accountthe direction of the displacement (e.g.anterior, posterior, medial or lateral). Weapplied MCAI to the right HG, extrac-ted using TASH, from the brain structu-

ral data of 182 healthy adult participantsincluding professional and amateur mu-sicians, as well as non-musicians. Wefound that the dominant lateral conca-vity, likely corresponding to sulci arisinglaterally, is significantly more prominentthan the dominant concavity in any otherorientation [t(181)=6.586, p<0.001], sug-gesting that MCAI is suitable for auto-matically assessing the likelihood of sulcibeing present. Additionally, as expectedbased on previous findings showing ahigher likelihood of greater gyrificationof HG in musicians compared to non-musicians, we find that the dominant la-teral concavity amplitude index is signi-ficantly higher in the former comparedto the latter group [t(181)=3.316, p=0.001,2-tailed]. MCAI allows the application ofautomated and reproducible criteria forcharacterizing the shape of the HG usingcontinuous indices. Next steps will in-volve application of MCAI to large da-tasets in the context of auditory andlinguistic expertise, disease and geneticvariation.

22

Cortical encoding of speech enhances task-relevant acoustic information

Sanne Rutten, Roberta Santoro, Alexis Hervais-Adelman, Elia Formisano & NarlyGolestania

aBrain and Language Lab, Department of Psychology, University of Geneva

Speech is the most important signalin our auditory environment, and its pro-cessing is highly context-dependent1.However, it is unknown how contex-tual demands influence the neural re-presentations of the acoustic contentof speech (henceforth referred to as“speech sound encoding”). Here, weexamined the task-dependence of au-ditory cortical mechanisms for speechencoding using model-based decoding2of 7T functional magnetic resonanceimaging data. Participants performeda phoneme identification or a speakeridentification task on the same speechsounds (pseudo-words), and we exami-ned how this influenced speech enco-ding at the level of the representation offundamental acoustic features (spectro-temporal modulations). We found thatperforming different tasks on identicalspeech sounds leads to the neural en-hancement of the acoustic features inthe stimuli that are critically relevant fortask performance. Acoustic analysis ofthe speech sounds showed that spea-ker information in the speech signal ischaracterized by modulation at higher

spectral modulations. Accordingly, du-ring the speaker task, the neural enco-ding of these sound features was ampli-fied. Conversely, phonetic informationin the stimuli was characterized by lo-wer spectral modulations and by fas-ter temporal modulations, and this wasparalleled by amplification in the neuralencoding of corresponding sound fea-tures during the phoneme task. Thesetask effects were observed already wi-thin Heschl’s gyrus, which is the earliestauditory cortical processing stage. Mo-reover, these findings are in line withinteractive accounts of speech proces-sing. Our work provides important in-sights into the mechanisms underlyingthe processing of contextually relevantacoustic information within our rich anddynamic auditory environment.

References [1] Leonard, M. K. & Chang, E.F. Dynamic speech representations in the hu-man temporal lobe. Trends Cogn Sci 18, 472-479,doi :10.1016/j.tics.2014.05.001 (2014). [2] Santoro,R. et al. Reconstructing the spectrotemporal mo-dulations of real-life sounds from fMRI responsepatterns. Proc Natl Acad Sci U S A 114, 4799-4804, doi :10.1073/pnas.1617622114 (2017).

23

Posters

Is dog cortical entrainment affected by speech rate and type? Acomparison with human processing

Eloïse C. Déauxa, Florence Gaunetb, Thierry Legouc, Julie Mathieu-Sarratb andAnne-Lise Girauda

aDepartment of Fundamental Neurosciences, University of Geneva ; bLaboratoire de Psychologie Cognitive, CNRS & AMU,UMR 7290, Marseille, France ; cLaboratoire Parole & Langage, CNRS & AMU, Aix-en-Provence, France

Processing speech requires segmen-ting a continuous acoustic flow intobuilding blocks that connect with sto-red linguistic representations. Our au-ditory cortex is able to track struc-tures in speech using the rhythmicity ofneuronal ensembles, which tend to firein synchrony, in bilaterally distributedbut functionally lateralized cortical sys-tems. Speech comprehension fails whenspeech rate is compressed beyond itsnatural range and this effect is corre-lated with decreased coherence bet-ween the speech envelope and corti-cal oscillation [1]. However, even in low-intelligibility conditions, theta phase lo-cking remains, suggesting it to be a coreaspect of speech processing [2].

Dogs are phylogenetically distantfrom humans, yet our long coevolutio-nary history has likely favored their unu-sual ability to perceive and respond tohuman signals [3]. They can learn >1000words naming objects, treating them asverbal referents, and dog speech pro-cessing appears to exhibit functional la-teralization analogous to ours [4,5]. Des-

pite these striking similarities with hu-mans, we do not know how dogs pro-cess speech.

In this first study, we aim to assesswhether human speech entrains dogs’cortical oscillations. As theta entrain-ment is thought to be necessary, albeitnot sufficient for speech comprehen-sion, we will investigate whether dogs’cortical oscillations in the theta rangeentrain to speech and whether this en-trainment is influenced by speech rateand correlates with lexical measures andspeech familiarity. Understanding whe-ther and how dogs track and syllabifycontinuous speech and where the bot-tleneck for speech comprehension lieswill ultimately help us inform our mo-dels of human linguistic communicationevolution. We present the current stateof the development of the project.

References. [1] Ahissar et al. 2001. PNAS.98 :13367-13372 [2] Pefkou et al. 2017. J Neurosci.37 :7930-7938 [3] Hare & Tomasello. 2005. TCS.9 :439-444 [4] Bloom. 2004. Science 304 :1605-1606 [5] Andics et al. 2016. Science. 353 :1030-1032

25

L’identification et le développement des compétences interactionnellesen formation professionnelle continue : le cas de la communication

directe et indirecte entre parents et éducatrices de l’enfance

Stéphanie Garcia, Université de Genève

Notre contribution s’appuie sur untravail de thèse en cours de réalisa-tion s’intéressant à l’activité de travailet au développement professionnel deséducatrices de l’enfance dans les rela-tions avec les parents. L’abondante litté-rature sur les difficultés que rencontrentles professionnelles dans l’établissementde relations prescrites comme partena-riales avec les parents, reflète 1) le besoinde mieux comprendre la mise en œuvrede cette relation et les compétences in-teractionnelles spécifiques mobilisées ;2) l’intérêt de développer un dispositif deformation continue qui permet de sou-tenir et de développer l’activité des pro-fessionnelles.

Cette recherche s’inscrit dans lechamp de la linguistique interaction-nelle en analyse du travail (Filliettaz, deSaint-Georges & Duc, 2008; Filliettaz,2014). Sa double finalité est pensée demanière imbriquée dans un dispositifde recherche-intervention. Des profes-sionnelles outillées et formées à l’ana-lyse des interactions multimodales éla-borent avec les chercheurs le sens attri-bué à leurs propres pratiques. Permet-

tant de rendre visible une certaine fa-cette de l’activité des éducatrices, l’ana-lyse des interactions est dès lors penséecomme un levier possible pour le déve-loppement professionnel.

Dans la contribution prévue, nousnous intéresserons plus particulière-ment au phénomène interactionnel dela communication directe et indirecteentre les participants aux situationsd’accueil et de départ des enfants encrèches. Comment les éducatrices et leschercheurs s’orientent-ils vers l’identifi-cation d’un tel phénomène? Et en quoirenseigne-t-il l’activité des profession-nelles et les compétences interaction-nelles mises en œuvre à cette occasion?

Références Filliettaz, L. (2014a). L’interactionlangagière : un objet et une méthode d’analyseen formation d’adultes. In J. Friedrich & J. Pita, J.(Ed.), Recherches en formation des adultes : undialogue entre concepts et réalité (pp. 127-162).Dijon : Editions Raison et Passions.

Filliettaz, L., de Saint-Georges, I. & Duc, B.(2008). « Vos mains sont intelligentes ! » : In-teractions en formation professionnelle initiale.Université de Genève : Cahiers de la section dessciences de l’éducation, 117.

26

MIAPARLE : Online training for discrimination and production of stresscontrasts in L2 Spanish and L2 English

Jean-Philippe Goldmana & Sandra Schwabb

aUniversity of Geneva ; bUniversity of Zurich ; University of Fribourg

MIAPARLE (miaparle.unige.ch) is aweb application providing to the ge-neral public an innovative range ofCAPT (computer-aided pronunciationteaching) tools for second language (L2)learners. MIAPARLE focuses on stressperception and production in Spanishand English. Such a tool is particularlyuseful for speakers whose native lan-guage is a fixed-stress language, suchas French. These speakers have difficul-ties perceiving and discriminating stresscontrasts ([1]), and therefore strugglewhen producing them.

To help L2 learners reducing theirso-called stress ’deafness’, MIAPARLEprovides a training module composedof perception and production activi-ties. In the perception activities, learnershave to associate auditory stress pat-terns with visual patterns. In the produc-tion activities, learners are shown pic-tures and have to produce in L2 the wordfor the item they see. The participants’production is acoustically analyzed andvisually represented with the stressedsyllable highlighted. A comparison bet-ween the participants’ production and

the native canonical production servesas feedback.

Pre- and post-tests, composed oftwo perceptual tasks, are used to eva-luate the training effect : a Localizationtask (’Find the right word’) and an Odd-One-Out task (’Find the different word’).The post-tests are similar to pre-tests,except that they comprise items that arenot presented in the pre-tests, allowingthe examination of the generalization ofthe training to words that are new for thelearners.

By combining new IT tools and in-novative teaching methodology, MIA-PARLE fills a major gap in traditional tea-ching methods, in which prosodic as-pects are usually neglected. Although’application-oriented’, MIAPARLE alsohas implications for research since thecollection of L2 perception and produc-tion data constitutes a valuable corpusfor research purposes in the fields of L2acquisition and CAPT.

References [1] Dupoux, E., Pallier, C., Sebas-tián, N. & Mehler, J. (1997). A destressing ‘deaf-ness’ in French?, Journal of Memory and Lan-guage, 36(3), 406-421.

27

Can rhythmic priming influence syntactic processing in typical adults?

Dávid György, Lna Azzam Ali, Doug Saddy, Julie Franck, Daphne Bavelier, SonjaKotza

aUniversity of Geneva

The investigation of potential linksbetween the processing of language andrhythm has recently gained increasedattention. Rhythm can be broadly defi-ned as a more or less regular alternationof events over time. Time is a crucialdimension of both speech and musicprocessing, and recent evidence sug-gests that temporal processing relies onsubcortico-cortical networks that pre-pare cortical brain areas for higher le-vel integration. As internal neural oscil-lations can synchronize to external os-cillations, a regular temporal structurein the input can facilitate temporal andgenerating temporal predictions (Henry& Herrmann, 2014 ; Schwartze & Kotz,2016). Experimental evidence shows nu-merous correlations between the pro-cessing of language and rhythm acrossvarious populations (Corriveau & Gos-wami, 2009; Gordon et al., 2015), butalso that a rhythmically regular musi-cal prime facilitates subsequent syntac-tic processing (Corriveau & Goswami,2009; Kotz & Gunter, 2015). The presentstudy aimed to explore whether a rhyth-mic priming effect can also be obser-ved in typical adults. 100 native spea-kers of French with no known neurolo-gical or language disorders listened tosilent, rhythmically regular or rhythmi-cally irregular primes prior to performinga grammaticality judgement task on au-

ditorily presented sentences in whichsubject-verb agreement was either cor-rect or incorrect. Based on recent workin agreement attraction, linguistic stimuliwere constructed in a semi-artificial Jab-berwocky language to eliminate seman-tic effects and focus on syntax. The ex-periment was run in two designs : ablock design (N=50) was used to mea-sure short and long-term effects of pro-longed exposure to the same prime anda mixed design (N=50) was employedto investigate more immediate effects ofrhythmic priming. Preliminary evidencesuggests some influence of rhythmic pri-ming on structural processing.

ReferencesCorriveau, K. H., & Goswami, U. (2009).

Rhythmic motor entrainment in children withspeech and language impairments : Tap-ping to the beat. Cortex, 45(1), 119–130.https ://doi.org/10.1016/j.cortex.2007.09.008

Henry, M. J., & Herrmann, B. (2014).Low-Frequency Neural Oscillations SupportDynamic Attending in Temporal Context.Timing and Time Perception, 2(1), 62–86.https ://doi.org/10.1163/22134468-00002011

Kotz, S. A., & Gunter, T. C. (2015). Can rhyth-mic auditory cuing remediate language-relateddeficits in Parkinson’s disease? Annals of theNew York Academy of Sciences, 1337(1), 62–68.https ://doi.org/10.1111/nyas.12657

Schwartze, M., & Kotz, S. A. (2016). Contri-butions of cerebellar event-based temporal pro-cessing and preparatory function to speechperception. Brain and Language, 161, 28–32.https ://doi.org/10.1016/j.bandl.2015.08.005

28

La contrainte du V2 dans la production spontanée du suisse allemand

Eric Haeberli et Manuela Schönenberger, Université de Genève

Comme toutes les langues germa-niques à l’exception de l’anglais, les dia-lectes du suisse allemand ont la pro-priété du «Verb Second» (V2). Unelangue V2 se caractérise par le fait quele premier syntagme est immédiatementsuivi par le verbe fléchi. Par conséquent,l’ordre du sujet et du verbe est inversési le premier syntagme n’est pas un su-jet, et l’ordre non-inversé est agramma-tical. Ce contraste est illustré en (1) avecun exemple du suisse allemand.

(1) a. EtzMaintenant

ischest

siielle

füfedachzgi85

« Maintenant elle a 85 ans » (02-ML-CH,.1918)

b. *Etz sii isch füfedachzgi

Il a néanmoins été observé dans lalittérature qu’il existe des exceptions àla contrainte du V2, mais ces excep-tions peuvent varier à travers les languesà V2. Dans cette étude, nous examine-rons le statut du V2 et de ses exceptionsen suisse allemand en nous basant surun corpus annoté de production orale

spontanée d’un dialecte suisse allemandque nous sommes en train de finaliser.Ce corpus contient un certain nombre deviolations apparentes de la contrainte duV2. Deux exemples sont donnés en (2).

(2) a. Irgendwieen-quelque-sorte

menous

händavons

enun

Termiinrendez-vous

abgmachtconvenu

« En quelque sorte nous avons fait unrendez-vous » (MS-DS-PS-FIN,.59)

b. DennEnsuite

menous

händavons

äbejustement

nòchliencore-un-peu

Landcampagne

undet

Schtadville

« Ensuite nous avons justement en-core un peu la campagne et la ville »(03-ML-FP,.1273)

Le but de ce travail est d’analyser lestatut des exceptions au V2 en suisse al-lemand d’un point de vue quantitatif etqualitatif en tenant compte notammentde facteurs prosodiques, sémantiques etpragmatiques.

29

On-line syllable recognition in natural sentences with neural oscillationsand predictive coding

Sevada Hovsepyan, Itsaso Olasagasti, Anne-Lise Girauda

aDepartment of Basic Neurosciences, Campus Biotech, University of Geneva

Natural speech comprehension is achallenging operation that requires seg-menting a continuous sound stream toconnect it on-line with discrete linguisticneural representations. This operationpartly relies on the coupling of theta andgamma neural oscillations that tracksand segments syllables and encodethem in decipherable neural activity (1,2).Speech comprehension also stronglydepends on top-down processes thatpredict on-line speech structure andcontent (3). We used neurocomputa-tional modeling to explore how theta-gamma coupling interacts with bottom-up and top-down information flows du-ring on-line speech perception. We builta generative model (4) that recognizessyllables in continuous natural speechsignal, and in which theta oscillationsdetect syllable onsets to align both mo-del’s gamma activity with syllable boun-daries, and top-down predictions withspeech input. Simulations demonstratethat the best performing model com-bines top-down syllable duration ex-

pectations with bottom-up informationabout syllable onsets. i.e. when bidi-rectional information flows were coor-dinated and internal timing knowledgewas exploited. These results show thatthe addition of theta oscillatory me-chanisms to bidirectional informationprocesses permits on-line speech pro-cessing. More broadly, this work illus-trates that notions of predictive codingand neural oscillations can usefully bebrought together to account for dyna-mic auditory processing.

References1. Hyafil, A., Fontolan, L., Kabdebon, C., Gut-

kin, B. & Giraud, A. L. Speech encoding by cou-pled cortical theta and gamma oscillations. eLife4, 1–45 (2015).

2. Giraud, A. L. & Poeppel, D. Cortical oscilla-tions and speech processing : Emerging compu-tational principles and operations. Nat. Neurosci.15, 511–517 (2012).

3. Gagnepain, P., Henson, R. N. & Davis, M.H. Temporal predictive codes for spoken wordsin auditory cortex. Curr. Biol. 22, 615–621 (2012).

4. Friston, K. J., Trujillo-Barreto, N. & Dauni-zeau, J. DEM : A variational treatment of dynamicsystems. NeuroImage 41, 849–885 (2008).

30

Measuring and enhancing the migrants comprehension of Italianadministrative texts

Giulia Lombardi, Università di Genova

The aim of the research was to inves-tigate the readibility of Italian adminis-trative documents among foreigners andoffer clearer alternatives or set up cla-rity guidelines that might improve accessto these documents. In the last years thephenomenon of migration has increa-sed across countries and the migrantslinguistic and social inclusion is beco-ming more and more urgent in our so-cieties. Too often, the Italian adminis-trative language tends to be unnecessa-rily difficult for all those who are L2 be-ginners in Italian language skills to be-come resident in Italy and yet need todeal with various red tape and forma-lities. The starter point of the research,in 2017, was a computational-linguisticanalysis of a synchronic mono-thematiccorpus which includes the most impor-tant forms foreigners have to submit inItaly, in order to find what lexical, se-mantic and syntactical structures were

too difficult for foreigners. Than, in 2018,101 students of Italian L2 were testedon the comprehension of authentic Ita-lian institutional texts. Many difficultieswere singled out : the correlation bet-ween personal factors (like age, schoo-ling, mother tongue and motivation) andreading comprehension was analyzedby a multiple regression analysis. In thebeginning of 2019, 61 students have beentested on simplified texts ; among them,32 attended a specific language course.Data have been analyzed with t-testand Anova test : the amelioration of thecomprehension on the simplified textsis statically significant (df = 59, p-value= 1.066e-09), while the manually refor-mulation seems to have a greater impactto respect to automatized lexical simpli-fication. The language course is also apredictor of a better comprehension (Fvalue = 4.56, p-value=0.037 *).

31

Brain structural correlates of dyslexia

D. Marie, S. Rutten & N. GolestaniaaBrain and Language Lab, Department of Psychology, University of Geneva

Developmental dyslexia is a disor-der affecting the acquisition of readingand spelling abilities in 3–7% of the ge-neral population, in the absence of sen-sory or neurological damage, and des-pite typical intelligence and a favourablesocio-educational environment. Early,post mortem studies on neuroanatomi-cal abnormalities in dyslexic individualsrevealed altered microstructure in theperisylvian and auditory areas of the lefttemporal lobe (eg. myelinated scars :glial cell growth instead of neurons). Ma-croscopic abnormalities have also beenfound using in vivo methods such asstructural magnetic resonance imaging(MRI). These studies have revealed ab-normalities in auditory brain regions. Forexample, it has been found that in dys-lexic individuals, there is an absence oreven a reversal of the usual leftwardasymmetry of the planum temporale ty-pically reported in healthy individuals.Recently, a quantitative structural MRIstudy found higher myelin content in theleft auditory cortex of dyslexic indivi-duals compared to controls, but thesefindings were not related to behavioural

measures of reading or of phonologicalprocessing known to be deficient in dys-lexia.

In the present study, we used ultra-high-resolution quantitative structuralMRI (7T) to compare the brains of 15adults with a history of dyslexia and17 healthy age, sex matched controls.We examined an index of intra-corticalmyelin, measured using R1 maps, at themesoscopic scale (0.6 mm3 resolution).We extracted individual myelin maps,and we assessed relationships with per-formance on behavioural measures ofreading and of phonological processing.Preliminary findings reveal significant re-lationships across all participants bet-ween those behavioural measures, es-pecially processing time, and myelina-tion in a network of areas critical for lan-guage and reading. For example, mye-lination in temporo-parietal, frontal andoccipital regions correlate with time ta-ken to read pseudowords. Implicationsof these findings for the deficits obser-ved in dyslexia and for compensationare discussed.

32

Affectivité dans les interactions de supervision en psychiatrie :perspective interactionnelle d’un enjeu relationnel en formation

Alexandra Nguyen, Haute École de Santé La Source, Lausanne et Université deGenève

La supervision constitue un espaceprivilégié de la formation des psycho-thérapeutes. Selon le courant du Work-place Learning, la supervision peutêtre envisagée comme une pratiqueformelle d’accompagnement au déve-loppement des compétences profes-sionnelles (Filliettaz & Billett, 2015). Enséance, le thérapeute convoque, par lerécit, les difficultés relationnelles ren-contrées avec le patient. L’affectivitédans les interactions a fait l’objet d’unevaste littérature consacrée aux produc-tions langagières telles que les storytel-lings (Selting, 2010), la confidence ou ledévoilement de soi (Kerbrat-Orecchioni& Traverso, 2007). Quelles sont les res-sources langagières mobilisées par lessuperviseurs pour accompagner le récitdu thérapeute comportant une dimen-sion affective manifeste? En quoi la mo-bilisation de ces ressources multimo-dales participe-t-elle à la professionna-lité des superviseurs dans la gestion del’affectivité en interaction? Une analyseinteractionnelle basée sur 35 heures desupervision a été menée. Six tandemsconstitués d’un superviseur et d’un thé-

rapeute ont été recrutés dans des ins-titutions de santé de Suisse romande.Dans la continuité des travaux conduitsdans le domaine de la régulation interac-tionnelle des émotions (Prior & Kasper,2016), les analyses montrent le rôle cen-tral de l’affiliation du superviseur pourpermettre la nécessaire implication af-fective du thérapeute lors du récit ainsique les effets de cette régulation affec-tive sur le processus analytique que su-perviseurs et thérapeutes mènent à pro-pos des enjeux relationnels du théra-peute avec son patient.

RéférencesFilliettaz, L., & Billett, S. (2015). Franco-

phone Perspectives of Learning Through Work :Conceptions, Traditions and Practices(Vol. 12) :Springer.

Kerbrat-Orecchioni, C., & Traverso, V.(2007). Confidence/dévoilement de soi dansl’interaction(Vol. 37). Tübingen : Walter de Gruy-ter.

Prior, M. T., & Kasper, G. (2016). Emotion inmultilingual interaction(Vol. 266) : John Benja-mins Publishing Company.

Selting, M. (2010). Affectivity in conversatio-nal storytelling. Pragmatics. Quarterly Publica-tion of the International Pragmatics Association(IPrA), 20(2), 229-277.

33

Altering phonemic representations with 30Hz tACS in Dyslexia

Johanna Nicolle, University of Geneva

Phonological dyslexia is an heredi-tary developmental disorder mostly ex-plained by a phonological awarenessdeficit and disabilities to disc(Sprenger-Charolles et al. 2003)riminate acous-tic differences between phonemic cate-gories (Sprenger-Charolles et al. 2003;Serniclaes et al. 2012) . In this study,Transcranial alternating current stimu-lation (tACS), a non-invasive brain sti-mulation technique used to modulatebrain oscillations at specific frequen-cies is proposed to as a rehabilitationtools for dyslexia impairment. Based onrecent neurophysiological findings, dys-lexia have been linked to atypical neu-ral activity related to auditory processing(Lehongre et al. 2011 ; Lallier et al. 2017).We therefore hypothesize that by ap-plying tACS over the left auditory cortex,linguistic skills affected by this condi-tion would improve in dyslexics. Rela-ted to this, fifteen adults with dyslexiaand fifteen control participants recei-ved tACS at 30Hz, 60Hz and sham tACSstimulation across three separate ses-

sions and performed a battery of lin-guistic tests before, immediately afterand one hour after stimulation to eva-luated the tACS effectiveness. Accor-ding to our predictions, results showedthat dyslexic participants performed hi-gher after the 30Hz stimulation, dueto both an improvement in discrimina-ting acoustic differences within the samephonemic category, in exchanging pho-nemes between two words and in rea-ding more fluently. Moreover, this im-provement positively correlates with theseverity of phonological dyslexia. Ourdata provides evidences for the impor-tance of phonemic perception and ca-tegorization in the development of pho-nological awareness and confirms theinvolvement of low gamma oscillationsin phonemic representation and proces-sing as well as in phonological aware-ness. Importantly, our findings empha-size the relevance of tACS as an inter-vention tool to improve speech percep-tion and to increase phonological awa-reness abilities in dyslexics.

34

Présence médiatique et numérique d’une thématique de santé publiqueselon quatre langues

Karen Pelletier et Virginie Zimmerli, Université de Genève

Introduction. Depuis 2009, l’OMSet l’équipe du Docteur Pittet organisenttous les 5 mai une campagne mondialeconsacrée à la prévention des infec-tions et à la promotion de l’hygiène desmains. Dans cette étude, nous analysonsles flux d’information sur cette théma-tique en quatre langues afin d’appréhen-der les retombées de cette campagne àl’échelle mondiale en termes de quantitéd’information diffusée.

Méthodes. Du 3 octobre 2017 au 2octobre 2018, l’outil Search a été utiliséafin de recueillir des statistiques sur laprésence du mot-clef hygiène des mainsen français, anglais, espagnol et portu-gais sur les médias sociaux. Ces donnéesont été comparées avec la présence deces mêmes mots-clefs dans la presse, àl’aide de la base de données Nexis.

Résultats. Sur les médias sociaux, lemot-clef hygiène des mains en quatrelangues et sur la période étudiée est

identifié dans 39’481 publications. Surces 39’481 publications, 31’123 sont enanglais, 3’245 en espagnol, 2’275 en por-tugais et 2’108 en français. Les médiasprivilégiés dans ces quatre langues sontles sites web (54.4%), Twitter (15.8%),Facebook (12.3%) et les médias en ligne(6.8%). La journée du 5 mai comptabilise1’912 messages, soit 4.8% des publica-tions de l’année. A titre comparatif, dansla presse, nous constatons égalementune sur-représentation des publicationsanglophones (2’604), suivi par les publi-cations lusophones (305), francophones(268) et hispanophones (100).

Conclusion. L’anglais est la langueplus utilisée pour les publications au-tour de l’hygiène des mains sur le web(80.7% des publications) et dans lapresse (79.5% des articles). Sur les mé-dias sociaux, les publications en françaissont très peu répandues. A l’inverse, leshispanophones semblent privilégier lesmédias sociaux à la presse.

35

Ordre temporel et cognition

Marina Petrossian, Université de Genève

Dans cette contribution, nous cher-chons à savoir d’où vient l’ordre tem-porel, quelle capacité permet de re-connaître du temporel dans une infor-mation et à mettre en évidence la di-mension temporelle « pure » sans pa-rasitage, à savoir sans faire intervenird’autres variables telles que la causa-lité ou la connaissance du langage. Lebut de cette différenciation est de mon-trer que la temporalité est un phéno-mène cognitif basique à part (comme lacausalité et la subjectivité, cf. Sanderset al. 1992) et que le temps, antérieur àla causalité, n’est pas forcément réduit àcette dernière (Piaget 1946, Nelson 1986,etc.). Afin de développer le sujet, nousallons recourir d’une part à une perspec-tive cognitive et psychologique expéri-mentale et/ou développementale (Rat-cliff 1995, Macar et al. 1992, etc.) qui serasuivie, d’autre part, de travaux en lin-

guistique (Sanders et al. 1992, Wilson &Sperber 2012, Grisot 2017, etc.).

RéférencesGrisot, C. (2017) « A quantitative approach to

conceptual, procedural and pragmatic meaning :evidence from inter-annotator agreement”, Jour-nal of Pragmatics, 117, 245-263.

Macar F., Pouthas V. & W. J. Friedman (1992)Time, Action and Cognition. Towards Bridgingthe Gap, Kluwer Academic Publishers, The Ne-therlands.

Nelson K. (1986) “Event knowledge and cog-nitive development”. In K. Nelson (Ed.), EventKnowledge : structure and function in develop-ment, 1-19, Erlbaum, Hillsdale, NJ.

Piaget J. (1946) Le développement de la no-tion de temps chez l’enfant, PUF, Paris.

Ratcliff M. J. (1995) Le développement de laTemporalité chez l’enfant de 2 à 4 ans, thèse dedoctorat, Université de Genève.

Sanders T. J., Spooren W. P. & L. G. Noord-man (1992) “Toward a taxonomy of coherencerelations”, Discourse Processes, 15(1), 1-35.

Wilson D. & D. Sperber (2012) Meaning andrelevance, Cambridge University, Cambridge.

36

Cross-lingual word embeddings and the structure of the human bilinguallexicon

Maria Rodriguez & Paola Merlo, University of Geneva

Question. Research on the bilinguallexicon has uncovered fascinating inter-actions between the L1 (native language)and L2 (second language) [4]. We inves-tigate cross-lingual word embeddings,bilingual word representations derivedfrom neural networks, and ask whetherthey define spaces exhibiting the coacti-vation properties of the human bilinguallexicon (shared translation effect, falseand true friends effects). We find thatthe similarity structure of cross-lingualword embeddings matches known ex-perimental findings on the human bilin-gual lexicon.

The human bilingual lexicon. For L2word pairs that share a translation in L1(e.g., tool and dish translate as Hebrewkli), bilinguals judge shared-translationpairs as more related in meaning thanL2 monolinguals [1]. False friends arewords that share form but differ in mea-ning (e.g.Italian arrangiare for Englisharrange). They cause interference ef-fects in bilinguals in lexical tasks [3].

True friends share both form and mea-ning (Italian autunno for English autumn).Bilinguals performing a lexical decisiontask in L2 are faster than expected forwords like type, a near true friend [2].

Results. We calculate cross-lingualword embeddings (we use three dif-ferent models) and measure the cosinesimilarity of different translation pairs.Our results match the human lexicon,showing shared translation effects andfalse and true friends effects. They alsoshow, like for humans, that similarity ba-sed on meaning is stronger than simila-rity based on form.

References [1] Tamar Degani et al. Bidirec-tional transfer. Journal of Cognitive Psychology,23(1) :18–28, 2011. [2] Ton Dijkstra et al. Recog-nition of cognates and interlingual homographs.Journal of Memory and Language, 41(4) :496 –518, 1999. [3] AndreaWeber and Anne Cutler.Lexical competition in non-native spoken-wordrecognition. Journal of Memory and Language,50(1) :1 – 25, 2004. [4] John N. Williams. The bi-lingual lexicon. In The Oxford Handbook of theWord. OUP, 2014.

37

Insécurité linguistique de l’enseignant en classe bilingue et résolution desproblèmes de communication

Ivana Vuksanovic et Anne Grobet, Université de Genèveivana.vuksanovic@unige.ch ; anne.grobet@unige.ch

Rares sont les travaux qui à ce jour sesont penchés sur la question des com-pétences langagières en L2 de l’ensei-gnant de discipline non linguistique (his-toire, mathématiques, biologie, etc.) enclasse bilingue, principalement du faitque celles-ci sont présupposées car né-cessaires à la transmission des savoirsdisciplinaires. Pourtant il arrive que desétablissements se voient obligés d’em-ployer des enseignants moins compé-tents du point de vue linguistique (cer-taines de leurs compétences étant au-dessous du niveau B2 du CECR) afin demaintenir l’existence de leur programmebilingue. Dans la présente étude, nousnous proposons de montrer commentdes enseignants de disciplines non lin-guistiques (art et musique) gèrent leursdifficultés de communication afin depréserver leur face d’expert (Goffman

1967) face à des apprenants susceptiblesde posséder des compétences langa-gières en L2 similaires aux leurs, en uti-lisant des stratégies d’évitement et deréalisation (Faerch & Kasper 1983 ; Bange1992). En nous basant sur l’analyse d’in-teractions en classe, nous mettrons enévidence l’impact de ces stratégies surl’élaboration des savoirs disciplinaires : ilapparaîtra que les stratégies de réalisa-tion, qui mettent pourtant plus en dangerla face de l’enseignant, amènent davan-tage les étudiants à participer à l’élabo-ration des savoirs tandis que les straté-gies d’évitement, qui permettent à l’en-seignant de préserver son statut d’ex-pert, semblent paradoxalement présen-ter un moins grand potentiel d’appren-tissage dans la mesure où L2 n’y est pasexplicitement thématisée et où les ap-prenants sont moins sollicités.

38

Weakly-Supervised Concept-based Adversarial Learning for Cross-lingualWord Embeddings

Haozhou Wanga, James Hendersonb, Paola Merloa

aUniversity of Geneva, bIdiap Research Institute

Question. Distributed representa-tions of words capture important lin-guistic information monolingually butalso across different languages. We ex-plore new augmented models of the ad-versarial framework based on the intui-tion that the concept, and not the word,is the right level of lexical alignment bet-ween languages, especially distant lan-guages. The results of our experimentson bilingual lexicon induction confirmthat our concept-based models yield si-gnificant improvements, especially fordistant language pairs.

The model. The standard GenerativeAdversarial Network (GAN) frameworkplays a min-max game between twomodels : a generative model G and adiscriminative model D. The generatorG learns from the source data distribu-tion and tries to generate new samplesthat appear drawn from the distribu-tion of target data. The discriminatorD discriminates the generated samplesfrom the target data [1]. Applied to lear-ning cross-lingual word embeddings,the objective of the generator G is tolearn the transformation matrixW thatmaps the source to the target language

embedding space, and the discrimina-tor Dl, usually a neural network classi-fier, detects whether the input is fromthe target language [2]. Assuming wordsin different languages are similar whenthey are used to describe the same topicor concept [3], we use concept-aligneddocuments (Wikipedia) to bring the ge-nerated embeddings of the source lan-guage words closer to the embeddingsof corresponding words in the targetlanguage.

Results and Discussion. Our me-thod improves over dictionary-basedbaselines on eight language pairs,bidirectionally and both for similarlanguages (English-German, English-Spanish, English-French, English- Ita-lian) and for distant languages (English-Finnish, English-Russian, English-Chinese, English-Turkish). We achievethese results with as few as 2500concepts.

References [1] Ian J Goodfellow et al. Ge-nerative Adversarial Networks. In NIPS 27,2672–2680, 2014. [2] Guillaume Lample et al..Word translation without parallel data. In 6thICLR, 2018. [3] Anders Søgaard et al. Inverted in-dexing for cross-lingual NLP. In 53rd ACL/IJNLP,1713–1722, 2015.

39

Speech-neural entrainment deficit in toddlers suffering from autismspectrum disorder

Xiaoyue Wang, Andy Christen, Jaime Delgado, Silvia Marchesotti, Holger FranzSperdin, Stéphan Eliez, Marie Schaer, Anne-Lise Girauda

aDepartment of Neuroscience, University of Geneva

Autism spectrum disorder (ASD) isa neurodevelopmental disorder (Lordet al., 2000) with impaired social com-munication including speech-related in-formation process (American Psychia-tric Association, 2013). Previous stu-dies highlight abnormal responses toboth finite speech segments and speechstreams in both adults and children withASD relative to non-affected subjects(TD) (Jochaut et al., 2015 ; O’Connor,2012 ; Sperdin & Schaer, 2016 ; Wynn etal., 2018). Here, we hypothesize that ASDdeficits in processing speech may re-sults from a poor neural entrainment totrack speech dynamic structure (Jochautet al., 2015), especially at low-frequencybands. To address this issue, we used anaturalistic setting in which we recor-ded EEG activity of 14 ASD and 16 TDtoddlers, while they were exposed tonatural continuous speech. All kids wereseated on their parents lap and wat-ched 4 different cartoons, each lastingon average about 2.5 minutes. To testthe extent to which speech stimuli indu-ced neural entrainment, we used mul-tiple linear regression with time-varyinglag to learn the correspondence bet-ween the speech stimuli and the cor-responding EEG segments. Our resultsindicate that on average ASD toddlersdisplayed a significant decrease in thespeech-neural entrainment metric re-lative to non-affected group. Moreover,we used power analysis to test whetherthe deficit driven by abnormal responseat specific neural-oscillation frequency.Our results show that ASD toddlers dis-played a significant power reduction at

low-frequencies (1-12Hz), not only in theban corresponding to speech syllabicrate (5-12Hz) but also in the delta band(1-4Hz), which signals phrasal and pro-sodic cues (Ding et al., 2014). Our studyprovides evidence for the altered neu-ral speech tracking mechanisms in ASDchildren at low frequency.

ReferencesAmerican Psychiatric Association. (2013).

Diagnostic and statistical manual of mental di-sorders (5th ed.) : American Psychiatric Associa-tion.

Ding, N., Chatterjee, M., & Simon, J.Z. (2014). Robust cortical entrainment tothe speech envelope relies on the spectro-temporal fine structure. Neuroimage, 88, 41-46.doi :10.1016/j.neuroimage.2013.10.054

Jochaut, D., Lehongre, K., Saitovitch, A., De-vauchelle, A. D., Olasagasti, I., Chabane, N., Gi-raud, A. L. (2015). Atypical coordination of cor-tical oscillations in response to speech in au-tism. Front Hum Neurosci, 9, 171. doi :10.3389/fn-hum.2015.00171

Lord, C., Risi, S., Lambrecht, L., Cook, E.H., Leventhal, B. L., DiLavore, P. C., Rutter,M. (2000). The Autism Diagnostic ObservationSchedule—Generic : A standard measure of so-cial and communication deficits associated withthe spectrum of autism. J Autism Dev Disord,30(3), 205-223.

O’Connor, K. (2012). Auditory proces-sing in autism spectrum disorder : a re-view. Neurosci Biobehav Rev, 36(2), 836-854.doi :10.1016/j.neubiorev.2011.11.008

Sperdin, H. F., & Schaer, M. (2016). AberrantDevelopment of Speech Processing in YoungChildren with Autism : New Insights from Neu-roimaging Biomarkers. Front Neurosci, 10, 393.doi :10.3389/fnins.2016.00393

Wynn, C. J., Borrie, S. A., & Sellers, T. P.(2018). Speech Rate Entrainment in Children andAdults With and Without Autism Spectrum Di-sorder. Am J Speech Lang Pathol, 27(3), 965-974.doi :10.1044/2018 AJSLP-17-0134

40

How to get to Campus Biotech

Campus Biotech Chemin des Mines 9 1202 Genève

From Geneva Train Station:

Take bus 25 or 1 and stop at “Jardin Botanique” Take tram 15 and stop at “Maison de la Paix”, walk downhill for about 100 meters towards the

lake, then turn left on "Avenue Blanc". At the end of the street (about 300 meters), you will reach the main entrance of the building and the reception at 9, chemin des Mines.

From Geneva Airport:

Walk to the airport railway station. All trains stop at Geneva main station ("Gare Cornavin"), which is the first stop. The ride takes 8 minutes.

At the "Gare Cornavin" railway station, tram 15 and bus 25 or 1 will take you to Campus Biotech.