L'UCLouvain (Belgique) recherche un candidat en traitement automatique
du langage ou profil similaire pour un projet de 2 ans minimum
(possibilité de réaliser une thèse ou non).
Engagement à partir du 1e février 2019.

Description du projet : Can we detect preclinical Alzheimer's disease
using smartphone data?

Over the last twenty years, the scientific community has made tremendous
progress to detect Alzheimer's disease (AD) before the onset of clinical
symptoms. Medical analyses require an invasive puncture or are very
expensive. Thus, developing affordable and minimally invasive screening
tools to detect AD pathology in the general population would be of great
help for both research and clinical applications. We answer this need by
proposing an automated analysis of linguistic features stored in
electronic messages. Indeed, smartphones now keep `sent messages' for
several years, allowing us to analyse retrospectively longitudinal
linguistic changes in narrative written data. After consent, we will
download all `sent messages' from the Preclinical AD patients' phones
together with the date and time information associated with the
messages. We will then use tools and methods from Natural Language
Processing (NLP) to analyse this personal big-data.

The 2-year job will rely on the following work packages:

1. Lexical components: Lexicometric techniques will be used to grasp the
   richness of vocabulary and the use of empty words: metrics as
   type/token and hapax legomena ratio will be employed.

2. Syntactic and spelling components: We will statistically evaluate
   each patient's ability to produce or avoid spelling errors through
   time. It should be noted that we have developed an expertise in
   differentiating spelling mistakes from word play.

3. Punctuation and other components: Particular attention will be paid
   to the use of punctuation as we know that electronic environments
   stimulate a different use of these elements.

4. Modelisation: we will then have to modelise the graphs with the help
   of different tools, such as Unitex grammars (Paumier, 2016).

5. Machine learning algorithms will be developped to learn linguistic
   features through the automatic processing of productions over
   time. Diachronic linguistics considers language in its temporal
   context, according to what precedes and follows. Statistical
   approaches will include text classification resources as features
   within a support vector machine (SVM).

Profil recherché :
- Diplôme TAL ou assimilé 
- Expérience ou intérêt pour les neurosciences
- Natif francophone
- Niveau anglais min. B2 (+ expérience en lecture d'articles
  scientifiques B2)

Posez des questions ou envoyez directement votre curriculum vitae (et la
liste de vos publications, si vous en avez) dans un mail expliquant
votre motivation pour ce poste à : louise-amelie.cougnon@uclouvain.be,
directrice de la recherche au Media Innovation and Intelligibility Lab,
avant le 15 janvier 2018.

Le projet est issu d'une collaboration entre le MiiL (centre
d'innovation médiatique), le Cental (centre de traitement automatique du
langage) et les Cliniques universitaires Saint-Luc, à l'UCL.