conll2018

The code for the conll2018 submission: from strings to other things: linking the neighborhood and transposition effects in word reading.

In the paper, we explore whether the neighborhoods of words featurized using feature sets that allow for transposition, e.g. bigram features or character trigrams, explain more variance in RT measurements than conventional feature sets.

If this is the case, then there is a chance that the neighborhood effect is early, in the sense that it takes place during word recognition, and not after the word has been recognized. If, on the other hand, feature sets that do not allow for transposition explain more variance, then it is likely that the neighborhood effect is late.

Across all our experiments, we find that the non-transpositional feature sets explain more variance in RT measurements. From this, we conclude that the neighborhood is formed without taking into account transpositions. Notably, this flies in the face of conventional psycholinguistic research on the neighborhood effect.

Requirements

The old20 package needs to be installed manually, see here. All other requirements are in requirements.txt.

Usage

Run experiment_1.py or experiment_2.py to get the raw data files as CSV from the corpora. You can then navigate to the r folder and run the R experiments to obtain the results from the paper.

Both experiments require that the following corpora are present in data:

SUBTLEX-NL.cd-above2.txt
dlp2_items.tsv
SUBTLEX-UK.xlsx
blp-items.txt
Lexique382.txt
French Lexicon Project words.xls

The Subtlex files can be found here. Lexique can be found here (links to direct download) The lexicon projects can be found here

Data request

We can supply the neighborhood measurements for all the corpora and feature sets on request. Please send an E-mail to the lead author.

Citation

If you use this code, or the results from the paper, please cite us, as follows:

@inproceedings{tulkens2018strings,
  title={From Strings to Other Things: Linking the Neighborhood and Transposition effects in Word Reading},
  author={Tulkens, Stephan and Sandra, Dominiek and Daelemans, Walter},
  booktitle={Proceedings of the 22nd Conference on Computational Natural Language Learning},
  pages={75--85},
  year={2018}
}

License

GPL 3.0