Andreas

@andreasvc

Assistant professor DH and inf. sci. at University of Groningen

Groningen, The Netherlands
Vrijeme pridruživanja: lipanj 2009.

Tweetovi

Blokirali ste korisnika/cu @andreasvc

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @andreasvc

  1. 23. pro 2019.

    Correction: the first author is

    Prikaži ovu nit
    Poništi
  2. 23. pro 2019.

    We intend to analyze the performance of Bertje in more detail and on more tasks. To be continued!

    Prikaži ovu nit
    Poništi
  3. 23. pro 2019.

    Bertje is the latest installment in a series of language-specific BERT models, after French, German, Italian, Finish, and Japanese. Collecting better data and pretraining a monolingual model is clearly worth the effort compared to using multilingual BERT.

    Prikaži ovu nit
    Poništi
  4. 23. pro 2019.

    The architecture of Bertje is the same as English BERT_base, but we use improved training objectives presented in recent work: we apply Sentence Order Prediction instead of Next Sentence Prediction, and the Masked Language Model predicts whole words.

    Prikaži ovu nit
    Poništi
  5. 23. pro 2019.

    Our gains in performance over multilingual BERT are due to (1) focusing on a single language, (2) improvements in training, but also (3) the quantity and especially quality of our training data which contains not just Wikipedia text, but also a large number of books.

    Prikaži ovu nit
    Poništi
  6. 23. pro 2019.

    We evaluate on several downstream tasks and outperform multilingual BERT. In addition, we test the effect of the number of iterations using a checkpoint at 850k iterations.

    Prikaži ovu nit
    Poništi
  7. 23. pro 2019.

    Hot off the press: Bertje. We collected a large and diverse corpus of Dutch and trained a monolingual BERT model. The model is available for download. Paper: joint work by me Gertjan van Noord & Malvina Nissim

    Prikaži ovu nit
    Poništi
  8. 19. pro 2019.

    Another surprising finding is that my old-fashioned rule-based system appears to be competitive with a neural, BERT-based system, as suggested by the comparison with et al 2019, who report results on 19th century English novels

    Prikaži ovu nit
    Poništi
  9. 19. pro 2019.

    There does appear to be interesting variation in performance across novels, which warrants further study, especially with respect to literariness.

    Prikaži ovu nit
    Poništi
  10. 19. pro 2019.

    Computers still struggle to understand "who did what to whom?" in texts. My latest paper on coreference resolution presents results on an annotated corpus of contemporary Dutch novels. Surprisingly, literature is not harder than news text! Paper:

    Prikaži ovu nit
    Poništi
  11. proslijedio/la je Tweet
    21. stu 2019.

    . made a plea today for the critical use of digital tools and methods in the Humanities. "There's really nothing about machine learning methods that is contradictory with a perspectival humanistic approach. These tools only enlarge the humanities."

    Poništi
  12. proslijedio/la je Tweet
    7. stu 2019.

    I'm not at any conference, but Andreas van Cranenburgh is presenting about coreference resolution in our local seminar. I like this summary of the state of the field over the past years...

    Prikaži ovu nit
    Poništi
  13. 5. stu 2019.
    Poništi
  14. 10. srp 2019.

    This Friday and I will present interesting new survey results on literariness. Come to our presentation 10:12-10:30, Cloud Nine,

    Poništi
  15. proslijedio/la je Tweet
    27. svi 2019.

    1/7 Do word embeddings really say that man is to doctor as woman is to nurse? Apparently not. Check out this thread for a description of a short paper I co-wrote with Malvina Nissim and Rob van der Goot, available here:

    Prikaži ovu nit
    Poništi
  16. 1. svi 2019.

    The Literary Pepsi Challenge: intrinsic and extrinsic factors in judging literary quality. Abstract to be presented at with :

    Poništi
  17. 10. velj 2019.

    This Thursday I'll reflect on what predictive modeling results tell us about literature in a lecture Unfortunately can't make it, so the talk won't cover her results on gender.

    Poništi
  18. 10. velj 2019.

    How recognizable is literary language? Very! See my new paper Joint work with Karina van Dalen and @brandaen

    Poništi
  19. 29. sij 2019.

    This Thursday I'll present "A Dutch coreference resolution system with quote attribution" including an evaluation on literature. Poster #2 in the second poster session at 14:45-16:15. Code:

    Poništi
  20. proslijedio/la je Tweet
    29. pro 2018.

    To get outside North Am: Andreas van Cranenburgh presented a fascinating paper at COLING showing that a corpus of clichés can be used to predict human judgment about novels. Just 1 ex. of great work coming out of the Riddle of Literary Quality in NL 6/9.

    Prikaži ovu nit
    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·