L'espace est un domaine important de notre systeme cognitif, developpe chez les enfants des le plus jeune âge, il est considere comme universel. Cependant, les langues varient beaucoup dans la facon dont elles en encodent la reference, qu'elle soit statique ou dynamique. Ces differences posent des questions sur la relation entre langue et pensee en general et particulierement lors du processus d'acquisition d'une langue etrangere : les proprietes typologiques de la langue source influencent-elles la facon dont les relations spatiales sont encodees dans la langue cible ? Quelles evolutions pouvons-nous attendre dans ce domaine avec le developpement du lecte de l'apprenant ? Dans ce contexte, cet article etudie la maniere dont des apprenants francophones du chinois langue etrangere (un groupe de niveau intermediaire et un groupe de niveau avance) expriment les proces spatiaux causatifs, dans le cadre propose par Talmy 2000 (langues a cadrage verbale et langues a cadrage satellitaire) et Slobin 2004 (langues equipollentes) et en comparaison avec des locuteurs natifs du chinois (Ji, Hendriks & Hickmann, 2011a, 2011b, 2011c) et du francais (Harr, 2012). La procedure utilisee est celle d'une analyse de corpus oraux produits a partir d'une description de sequences animees. Les reponses ont ete etudiees au niveau du choix des informations exprimees (proportions des differents composants semantiques selectionnes : trajectoire, cause, maniere), de la densite semantique (nombre de composants semantiques differents encodes dans chaque reponse), et de la facon dont les informations ont ete encodees (choix des composants linguistiques dans l'encodage des composants semantiques). Les resultats revelent les phenomenes suivants : (1) les productions des apprenants de niveau intermediaire presentent de grandes differences avec celles des locuteurs natifs du chinois et beaucoup de points communs avec celles des locuteurs natifs du francais, et ce a tous les niveaux. Le fait que, d'apres Talmy et Slobin, la langue maternelle et la langue cible de ces apprenants appartiennent a des cadrages differents explique la difficulte qu'ils rencontrent pour etablir des nouvelles relations forme-fonction et realiser la restructuration conceptuelle necessaire pour s'approcher de la competence des locuteurs natifs. En revanche, (2) les apprenants de niveau avance montrent une evolution nette en direction de la langue cible au niveau du choix et de la quantite d'informations exprimees ; cependant, les moyens linguistiques utilises presentent encore des differences avec les habitudes des locuteurs natifs du chinois.
The present chapter provides theoretical background by spelling out some of the major current debates that oppose different approaches in developmental psycholinguistics. I first summarise a number of main controversial issues that are currently debated in the study of language acquisition (Section 2.1). I then focus on some fundamental properties of language within functional approaches to language (Section 2.2), including multifunctionality and context-dependence, as well as the existence of two levels of linguistic organisation, the sentence and discourse. Finally, I examine general universal principles of discourse organisation that underlie how speakers regulate the flow of personal, spatial, and temporal information across utterances in cohesive discourse (Section 2.3). Subsequent chapters show how these principles constitute fundamental aspects of our linguistic competence that must be mastered by children and that affect their acquisition of linguistic devices across domains.
The aim of this study is to compare narratives produced by French children of 6, 9, and 11 years on the basis of the same picture book in two situations : mutual knowledge (cp) or absence of mutual knowledge (acp). The results show that productivity (number of clauses and of referring expressions) is higher in Situation acp than in Situation cp at 9 and 11 years. From 9 years on children also differentiate referent introductions as a function of situations. This differentiation is maximal at 9 years : at this age children use mostly indefinite expressions in Situation nmk and definite ones in Situation cp. At all ages and in both situations referring expressions vary as a function of local coreference : pronominals are more frequently coreferential, plain and dislocated nominals noncoreferential. Finally, children also mark external and internal structure at all ages, using dislocated nominals (6 years) and plain nominals (9 and 11 years) at the boundaries between pictures and episodes. However, the impact of this factor decreases with age in the absence of mutual knowledge : in this situation coreference becomes the main principle of discourse organization at 11 years.
The potential of quantum enhanced Q-learning with a focus on its applicability to a lane change manoeuvre is investigated.In this context we solve multiple simple reinforcement learning environments using variational quantum circuits.The achieved results were similar to or even better than those of a simple constrained classical agent.We could observe promising behaviour on the more complex lane change manoeuvre task, which has an environment with an observation vector size twice larger than commonly used ones.For the Frozen Lake environment we found indications of possible quantum advantages in convergence rate.
Linguistic research reveals the need to consider linguistic diversity to better understand the impact of universal and language-specific determinants on individuals’ spatial representations and on spatial language acquisition. Our study examines motion expression in French Sign Language (LSF) with particular attention to the fundamental role of iconicity in defining specific properties of sign languages (SLs) and their implications for acquisition within an approach that is both typological (Slobin 2004) and semiological (Cuxac 1999).
We compare productions in LSF by Deaf children of Deaf parents, aged 4-5 and 8-9, and Deaf adults (6 in each age group, data annotated with ELAN) and in spoken French by same-aged hearing French speakers (24 per age, data annotated with CLAN). All participants described 26 video clips showing agents carrying out motion in varied manners (e.g. walking, running, swimming) along three paths (up, down, across).
Results show that motion expression is more specific and denser (Path and Manner) in LSF than in French (mostly Path) at all ages. This information is encoded either simultaneously within the same structure or in successive constructions describing the same event from different perspectives (internal vs. external). Findings also show the impact of event types on signers’ uses of different highly iconic structures (vs. lexical signs) that invite them to focus on different semantic components of motion.
Although semantic density did not increase with age in LSF (contrary to French), results show an increase in response specificity, indicating gradually more complete and precise motion descriptions. Furthermore, developmental changes occurred with respect to the types of structures used.
To conclude, although we observe developmental progressions in both languages, patterns differ significantly between LSF and French at all ages. This suggests that modality impacts acquisition, inviting us to compare these results with coverbal gestures (Ozyurek, Furman & Goldin-Meadow 2014).
Abstract A number of studies show that language-specific properties have an impact on speakers’ spatial representations. These findings have revived questions concerning the impact of general developmental vs language-specific determinants in language acquisition. Recent hypotheses suggest that speakers’ attention may be directed to different aspects of reality depending on the lexicalization patterns of their language. This study tests these hypotheses in relation to the expression of motion during first-language acquisition. The data consists of descriptions of motion events produced by monolingual German vs French adults and children. The results in both languages show first a developmental progression in children’s abilities to express Manner and Path simultaneously. However, regardless of age, responses differed considerably across languages. German speakers frequently encoded both Manner and Path, whereas French speakers mostly focused on Path. Our results support the view that general cognitive factors and language-specific properties both determine how children construct the semantics of space.
La diversité des langues fait l’objet d’un nombre croissant de recherches, qui ont remis à l’ordre du jour d’anciens débats et ouvert de nouvelles perspectives dans l’étude de la relation entre langage et cognition. Cette synthèse propose une revue de questions centrée sur les implications de cette diversité pour l’étude de l’acquisition de la langue maternelle. Elle résume un ensemble de recherches montrant le rôle que jouent certaines propriétés spécifiques des systèmes spatiaux et temporo-aspectuels sur le langage de l’enfant, entraînant des parcours développementaux parfois très différents d’une langue à l’autre, au-delà des similarités qui sont observées entre elles par ailleurs. Ces résultats remettent en question les conclusions antérieures selon lesquelles des concepts universaux, présumés innés ou construits par l’enfant, seraient les seuls déterminants de l’acquisition du langage. Ils ont impulsé une nouvelle génération de recherches expérimentales visant à déterminer si et comment les propriétés spécifiques des langues pourraient avoir un impact plus profond sur la cognition humaine.
Modern multi-document summarization (MDS) methods are based on transformer architectures. They generate state of the art summaries, but lack explainability. We focus on graph-based transformer models for MDS as they gained recent popularity. We aim to improve the explainability of the graph-based MDS by analyzing their attention weights. In a graph-based MDS such as GraphSum, vertices represent the textual units, while the edges form some similarity graph over the units. We compare GraphSum's performance utilizing different textual units, i. e., sentences versus paragraphs, on two news benchmark datasets, namely WikiSum and MultiNews. Our experiments show that paragraph-level representations provide the best summarization performance. Thus, we subsequently focus oAnalysisn analyzing the paragraph-level attention weights of GraphSum's multi-heads and decoding layers in order to improve the explainability of a transformer-based MDS model. As a reference metric, we calculate the ROUGE scores between the input paragraphs and each sentence in the generated summary, which indicate source origin information via text similarity. We observe a high correlation between the attention weights and this reference metric, especially on the the later decoding layers of the transformer architecture. Finally, we investigate if the generated summaries follow a pattern of positional bias by extracting which paragraph provided the most information for each generated summary. Our results show that there is a high correlation between the position in the summary and the source origin.
