Competition : My Tailor is rich! Predicting English level by analyzing writing styles.


Updates

  • 18.05.2018 : a one-or-two-page abstract describing your methodology for solving the problem has to be jointly submitted with your scores.


  • Final Leaderboard

    Rank Team Score
    1. Balikasg3.4856
    2. Terislepacom7.2862
    3. ICSI8.5970
    4. Caoutchouc9.4537
    5. ACNK10.4350
    6. reciTAL team11.2112
    7. TAU12.8222
    8. Capitaine-Ad-Hoc14.1696
    9. Chamlia17.5161
    10. Haralambous + Lenca Team17.9774
    11. MB31.6637
    12. Rufino33.4285
    13. Team UTC40.7879
    14. Limsi41.5202


    Abstract of the top two submissions

  • G. Balikas : Predicting the level of non-native English speakers from their written essays

  • In this talk I will present my participation in the CAp 2018 data challenge. I will start by describing the components of my system and I will proceed by detailing the feature extraction and feature engineering decisions as well as the model selection and validation steps. I will conclude the presentation with lessons learned from my participation.

  • Alexandre Garcia : Une méthode en deux étapes pour la prédiction du niveau d'anglais.

  • Le niveau d'anglais d'un texte est caractérisé par des marqueurs visibles au niveau du mot (fautes d'orthographe, choix du vocabulaire) et au niveau du texte (longueur des phrases, tournure syntaxique). Le modèle proposé est construit en 2 étape pour (1) exploiter l'information disponible au niveau du mot par l'utilisation d'un modèle adapté à des entrées parcimonieuses en très grande dimension et (2) fusionner cette information avec des descripteurs structurels du texte. Nous présentons les différentes parties du prédicteur ainsi que les représentations des données exploitées par le modèle.

    Objectives

    The CAp 2018 conference is hosting the following machine learning competition. The Common European Framework of Reference for Languages (CERL) maps linguistic competence in a foreign language onto six reference levels, described to be shared by European countries: A1, A2, B1, B2, C1 and C2. The goal of this competition is to achieve, by learning, a system to predict the level of competence of a learner, from one of these written productions comprising between 20 and 300 words and a set of characteristics calculated from this text. For more details. and supplementary information.




    Data set description

    To access data competitors have to:
  • register here to: https://corpus.mml.cam.ac.uk/efcamdat2/public_html/explore/
  • after registering, the competitors should then send an email to: efcamdat.team@gmail.com with the subject: Request for CAp2018 Shared Task Data

  • After we’ve confirmed that you have registered, we will send out an email with log-in details to the shared task folder within 24 hours.

    To have access to the test data you need to
  • be registered and
  • send an email to: efcamdat.team@gmail.com with the subject: Request for CAp2018 Test Data


  • Submission To submit your results, email a csv file with the predicted values A1, A2...C2 at competition.cap2018@litislab.fr

    Important dates

  • 28 March: call for participation
  • 28 April: release of the test set
  • 28 May : deadline submission

  • Prizes

    NVIDIA will offer a Geforce 1080Ti for each of the top 2 teams.



    Thanks to