| More Information: | PhD position in computational linguistics:
** Distant supervision for question answering on temporal relations **
The LabEx-EFL ( Laboratory of Excellence Empirical Foundations in Linguistics;
http://www.labex-efl.com/wordpress/ ) is hiring a PhD candidate in computational
linguistics for a 3-year contract. The position is full-time with a net salary
of about 1700 euros per month and starts in the fall of 2021. The subject is
part of the Computational Semantic Analysis research strand of the Labex EFL
(http://www.labex-efl.com/wordpress/research/strand-5-computational-semantic-analysis/?lang=en
) and will be supervised by Benoit Crabbé and Timothée Bernard.
Thesis topic:
In the last few years, deep learning methods have allowed a clear progress in
semantic inference tasks, especially by minimizing the amount of annotated data.
In this thesis, we propose to test the limits of these methods on a question
answering task on temporal relations (before, after, during...) between events
in texts. For instance, questions could be of the form does event X happens
before event Y in the text ?
In general, inference on textual structure is a problem for which annotated data
is notoriously costly and difficult to acquire and the case of temporal
structure is not an exception. The thesis will rely on recent methods that aim
to minimize the cost of manual annotation on temporal relations and rather use
the question answering task as its main objective. We will study preferably
interpretable models, graph-based or timeline-based, that will build their own
representation of the text in order to answer the questions.
Moreover we want to test to which extent an end-to-end a question answering
model is able to acquire and take advantage of his own temporal representation
for answering temporal questions, specifically in cases where the answer is not
literally present in the text and requires inference.
We will study to which extent accurate question answering models need to
explicitly annotate temporal relations expressed in the text or to which extent
we can leave some or all of these relations underspecified and rather use the
downstream inference task in order to acquire the textual timeline.
Expected Profile:
The candidate should have a degree equivalent to a master's degree (before
November 2021) with a specialization in artificial intelligence, natural
language processing or computational linguistics. Candidates are expected to
have substantial skills in deep learning or reinforcement learning and an
interest in formal modeling.
Affiliation:
The PhD student will be attached to the Laboratoire de Linguistique Formelle
(http://www.llf.cnrs.fr), funding will be available for missions, as well as
access to computing servers. The student will be expected to attend doctoral
seminars, and may eventually teach courses in NLP/computer linguistics at the
University of Paris.
Application procedure:
• The application should be sent to benoit.crabbe@u-paris.fr and to
timothee.bernard@u-paris.fr .
• Application deadline: June 25, 2021 (midnight).
• The application file must be composed of:
• a letter of motivation;
• a CV including the courses taken (master grades), diplomas obtained,
possible internships;
• the name and contact details of two references.
Shortlisted candidates will send their master's thesis and/or other written work
showing their qualification for the project in question.
The auditions (by videoconference) of the shortlisted candidates will take place
in early July 2021.
Contact:
For questions regarding the application process, please contact
benoit.crabbe@u-paris.fr |