Séminaire des doctorant·es

Ensieh Hemmatan (PhD student in Computational Linguistics, Université Paris Cité, LLF)

 

Title: Temporal Dependency Parsing Through Iterative Refinement

 

Understanding temporal dependencies in text—whether events happen before, after, or during one another—is fundamental to language comprehension, yet challenging to automate because time expressions are often underspecified and relations are highly context‑dependent. This talk introduces TIDE (Temporal Iterative DEpendency parser), an iterative, document‑level approach that builds temporal dependency graphs by refining predictions over several rounds. Instead of making one‑shot pairwise decisions, TIDE lets the events inside a document inform each other across context. The presentation’s goals are to explain what Temporal Dependency Parsing is, the intuition behind iterative refinement, and to illustrate benefits.

Title: Understanding psycho-typological variations in L2 - A preliminary study

 

Abstract: Many teachers and educators in second language (L2) are faced with great disparities in proficiency in their classes. While part of this variation is explained by sociological differences (see Grobon et al. 2019 for a review in the French context), there are differences that stem from uncontrollable psycho-typological differences (Poropat, 2014 ;  Botes et al., 2024). We tested the influence of personality, L2 anxiety and information treatment preferences on forced-choice questions around the opposition between the progressive and simple aspect in English L2 for French learners (Leclerc, 2009 ; Demagny, 2013). 

 

We tested how different task forms would impact the answer choice and time response, and whether these differences were imputable to psycho-typological differences we measured.

 

Using a clustering Gower-distance method (Durand, 2019), we created four groups of typical behaviours, depending mostly on proficiency, response time and most successful task(s), and found salient traits in their psycho-typological profiles.

 

Our findings are preliminary, and ought to be researched in a larger scale study for larger inferential effects