Postdoctoral position in Natural Language Processing: Text Simplification

Project description

The goal of the ALECTOR project is to develop and test resources that make it possible to propose simplified texts for children who face major problems in reading and understanding written texts.

Job description

The successful candidate will develop methods for text simplification in French, with a focus on children with dyslexia.

Current simplification systems treat each sentence independently and thus actually produce sentence simplification instead of text simplification. Yet, in order to be used in a real case scenario, the produced texts have to be coherent, which implies that the simplification in a sentence must take into account the simplification of other sentences. For instance, if the term "womanhood" is translated into "being a woman" in a sentence, it has to be translated the same way in the rest of the text. If a present tense was preferred at the beginning at the text, the following sentences will have to use the same tense.

The objective will be to integrate existing simplification systems, such as LEXenstein, which is specific for lexical simplification, or the  methods described in (Zhang and Lapata, 2017), (Nisioi et al., 2017) or (Stajner et al., 2017), and use discourse information, such as coreference chains or topic models, to generate coherent texts.

Evaluation will be both automatic, by using MT metrics on existing corpora, and manual, in order to focus on the specific coherence issues.

Tests will be made with existing English datasets and tools, but will also be applied to French.

The appointed researcher will work in close collaboration with all teams involved in the ALECTOR project.

Requirements and qualifications

Additional information


Applications should include the following:

and be sent to Anne-Laure Ligozat (annlor[@]