The acquisition of multiword sequences in language processing
When: Friday, Dec 20th, at 2:00 PM
Where: Salle des voûtes (Building 9, Saint-Charles campus).
The defense will be conducted in English.
MemoPsy Team
-----------------------------------------------------------------------------------------
Composition of the Jury :
Bénédicte Poulin-Charronnat (Université de Bourgogne) – Présidente du jury/Rapporteuse
Arnaud Destrebecqz (Université Libre de Bruxelles) - Rapporteur
Padraic Monaghan (Lancaster University) - Examinateur
Arnaud Rey (CNRS, Aix-Marseille Université) – Directeur de thèse
Carlos Ramisch (Aix-Marseille Université) - Co-directeur de thèse
Full Abstract: The ability to understand and produce a seemingly infinite number of novel utterances has long been considered a hallmark of human language. Traditionally, single words and innate grammatical rules have been thought to be the basic building blocks of language. However, recent findings have led to an alternative perspective, known as the usage-based approach, which suggests that grammatical rules are not independent of language use and that multiword sequences form the core of language. Thus, the ability to process and learn multiword sequences appears to be crucial for language comprehension and production. Numerous experimental studies support this claim, but the mechanisms underlying the acquisition of multiword sequences remain relatively unexplored.
The objective of this thesis is to investigate how multiword sequences are acquired in real-time and to characterise the mechanisms that shape their processing. To this end, we investigated the influence of repetition, the nature of the sequence, and repetition spacing during sequence learning. In our first study, we demonstrated that multiword sequences are rapidly acquired after about four to five repetitions, and that each word within the sequence followed a distinct learning trajectory. Moreover, learning was influenced by the type of sequence, with semantically related words and idioms being processed faster and remembered better than unrelated words and pseudowords. In our second study, we showed that repetition spacing affected the processing of multiword sequences, with learning decreasing with wider spacing, but still observed for the widest spacing. The third study, conducted with baboons, confirmed these findings and showed that sequence learning relies on fundamental associative mechanisms and that memory traces of sequential information are long-lived. Finally, in a theoretical chapter, we argue that current chunk-based models cannot account for these findings and that recent neurocomputational models based on associative and Hebbian learning may be more appropriate for describing and understanding the nature of chunking mechanisms.
Taken together, this thesis shows that sequential information is acquired rapidly, developing a transient memory trace, and that repetition plays a key role in its consolidation. The findings of this thesis raise important new questions for the field of language acquisition, and for the field of statistical learning and chunking more broadly.
Keywords: language processing, multiword sequences, sequence learning, statistical learning, chunking
