Presentation

The Lausanne Linguistics Research Seminar is the official research seminar of the Section des Sciences du Langage et de l'Information (SLI) of the University of Lausanne. It takes place once or twice a month (except when it takes place three times a month, as in May 2021).

If you are interested in attending the talks (on Zoom), please write to the seminar's organizer Benjamin Storme (at firstname.lastname@unil.ch)


Spring 2021 Calendar

Titles and abstracts are published about a week before the talk. Talks take place on Tuesdays at 12:30 pm.

March 2: Ana Claudia Keck & Clotilde Robin - L’expression du degré de certitude et de la source de l’information dans l’interaction

Nos deux thèses respectives s’inscrivent dans un projet de recherche FNS collectif consacré aux prises de positions épistémiques (‘epistemic stance' (Heritage, 2012)) en français-en-interaction. L'objectif général du projet est de proposer une étude systématique des marqueurs épistémiques - au sens large - du français relevant aussi bien de la modalité épistémique (degré de certitude), tels que probablement, je croisje pense, etc. que de l'évidentialité (source de l'information), tels que il paraîtj’ai vumanifestement, etc., et ce dans une triple approche énonciative, interactionnelle et multimodale. Le projet collectif tend à suivre l'idée selon laquelle la modalité épistémique et l'évidentialité se regroupent à l'intérieur d'un domaine conceptuel plus général appelé epistemicity (Boye, 2012). Cette prise de position théorique permet alors de rendre compte de manière globale de la façon dont les marqueurs du doute, du savoir et du non-savoir se manifestent en français-en-interaction et comment ceux-ci sont utilisés par les interlocuteurs pour construire et moduler des positionnements épistémiques dans l'interaction. Le corpus commun sur lequel nous travaillons se compose de deux sous-corpus de données « politiques » avec des débats-conférences publics et des débats télévisés (env. 14h de vidéo-enregistrements), mais également d’un corpus de réunions professionnelles au sein d’entreprises (env. 14h).

   

Cette conférence sera ainsi l’occasion de présenter (1) le cadre théorique du projet de recherche, (2) les méthodes choisies pour l’analyse des marqueurs épistémiques et de proposer (3) une étude de cas au travers d’un extrait où apparaissent aussi bien des marqueurs relevant du degré de certitude que des marqueurs relevant de la source de l’information.

April 27: Clara Cohen - Deep Learnability: Using Neural Networks to Quantify Language Similarity

Learning a second language usually progresses faster if a learner's second language (L2) is similar to their first language (L1). Yet global similarity between languages is difficult to quantify, obscuring its precise effect on learnability. Further, the combinatorial explosion of possible L1 and L2 language pairs, combined with the difficulty of controlling for idiosyncratic differences across language pairs and language learners, limits the generalizability of the experimental approach. In this study, we present a different approach, employing artificial languages, and artificial learners. We built a set of five artificial languages whose underlying grammars and vocabulary were manipulated to ensure a known degree of similarity between each pair of languages. We next built a series of neural network models for each language, and sequentially trained them on pairs of languages. These models thus represented L1 speakers learning L2s. By observing the change in activity of the cells between the L1-speaker model and the L2-learner model, we estimated how much change was needed for the model to learn the new language. We then compared the change for each L1/L2 bilingual model to the underlying similarity across each language pair. The results showed that this approach can not only recover the facilitative effect of similarity on L2 acquisition, but can also offer new insights into the differential effects across different domains of similarity. We finish by speculating on how these results can be expanded to uncover other unbiased measures of global language properties, such as inherent complexity; and whether learner-specific cognitive resources can be profitably added to our models.

 

May 4: Sam Zukoff - The Typology of Repetition Avoidance Patterns in Indo-European Reduplication

A number of the ancient Indo-European (IE) languages display a typologically unusual alternation in reduplication, relating to the treatment of cluster-initial bases. For example, in Gothic, bases beginning in obstruent-sonorant (TR) clusters copy just the first consonant, but bases beginning in s-obstruent (ST) clusters do something else, namely, copy both consonants. When looking around the IE languages, we find two dimensions of variability relating to this kind of reduplicative alternation: (i) What alternative (i.e. non-C1-copying) reduplication pattern do the ST-clusters show? (ii) Which cluster types pattern with the ST-clusters and which pattern with the TR-clusters? In this talk, I'll explore the first of these questions, and develop analyses and explanations for the resulting (micro-)typology.

May 11: Ezer Rasin - How children learn the hidden sound patterns of their language: a computational approach

The sound system of a language contains patterns that humans learn from their input in the first few years of their lives. Some of those patterns are "hidden" (also called "opaque"), in the sense that they are only observable at a level of abstraction that is remote from the surface sounds that children hear. Such hidden sound patterns pose a cognitive puzzle: how do children make the inductive leap required to abstract away from the surface and discover a hidden sound pattern? In this talk I will present an approach to this puzzle based on the principle of Minimum Description Length (MDL) -- a mathematical formalization of Occam's Razor -- according to which children learn sound patterns by looking for the system of rules that provides the simplest description of their input. I will present a concrete MDL-based learning algorithm and successful simulation results using artificial datasets with hidden sound patterns like those found in natural languages. These results suggest that MDL is a promising general theory of language acquisition in the domain of sound.

May 25: Marie-Hélène Côté