geco_augmented.md

March 29, 2021 ยท View on GitHub

GECO Augmented

Files from the GECO dataset Cop et al. 2017, dowloaded from https://expsy.ugent.be/downloads/geco/:

  • SubjectInformation.xlsx
  • L2ReadingData.xlsx
  • MonolingualReadingData.xlsx

The last two files (renamed to end with Augmented) have the following additional fields:

  • WORD_NORM lowercased word, without punctuation. Numbers reprepresented as NUM.
  • WORD_LEN word length, excluding punctuation.
  • FREQ-BLLIP -log2(word frequency) in BLLIP (Charniak et al. 2000). BLLIP vocabulary size 229,538 words.
  • FREQ-SUBTLEX -log2(word frequency) based on SUBTLEX-US Brysbaert and New 2009. SUBTLEX-US vocabulary size 74,286 words.
  • FREQ-WEB -log2(word frequency) based on the Kaggle web word frequency list. Kaggle vocabulary size 333,332 words.
  • OOV-BLLIP 1 if the word out of vocabulary in BLLIP, 0 otherwise.
  • OOV-SUBTLEX 1 if the word out of vocabulary in SUBTLEX-US, 0 otherwise.
  • OOV-WEB 1 if the word out of vocabulary in Kaggle, 0 otherwise.
  • SURP-GPT2 word surprisal according to the GPT2 language model Radford et al. 2019.