Iconicity Measures for Japonic Sensory Vocabulary

A collection of iconicity measures for words from the Sensory Lexicon of Japonic. If you use these measures, or the Icotools package, please cite:

McLean, B., Dunn, M., & Dingemanse, M. (2023). Two measures are better than one: Combining iconicity ratings and guessing experiments for a more nuanced picture of iconicity in the lexicon. Language and Cognition, 1-24. doi:10.1017/langcog.2023.9.

The csv files tagged -responses.csv contain the raw data, while the file 'GuessingRatingScores.csv' summarises the final results with unreliable data excluded (see exclusions.Rmd). The file paper.Rmd contains the analysis code.

The folder experiment_materials contains all the files for making the experiments using Icotools (see here for more instructions on how to use Icotools), including the sound files and the experiments themselves.

The steps to make the experiments are as follows:

1. Navigate to the experiment_materials folder
2. Install icotools, using pip install icotools.
3. Start a python session, import icotools then run icotools.foiler('stimuli.csv','letters.csv'). This will decide on the foil words for the guessing experiments.
4. Run makakanaforsoundfiles.py, this will make the file soundfiles.csv which has the list of soundfiles to make (both the real Japanese words, and the foil words).
5. Run makeJapanesesoundfiles.py, this will make a folder soundfiles, and make soundfiles in that folder for all the words.
6. Use Praat to run the praat script, flatintonation.txt, this will flatten the intonation in the soundfiles.
7. Start a python session, import icotools then run icotools.rater('stimuli.csv','control_file.csv'). This will add a folder with the rating experiments.
8. Run icotools.guesser('stimuli_oppfoils.csv','control_file.csv'). This will add a folder with the guessing experiments.

A note on flat intonation

We flattened the intonation of the sound files for this study because we did not have natural recordings of the dialectal words in the study, and the intonation produced for these words by Google Text-to-Speech was unnatural. Flattening the intonation was done in order to make the comparisons between the dialectal words and the standard words, and also between the real words and the fake words, fair (see discussion in paper). However, if you do have natural recordings and can avoid flattening the intonation, I would first pilot a study using only natural recordings where the foils are randomly chosen from among the other natural recordings (the default setting in icotools), before going down the road of using artificially constructed opposite foils and then having to flatten the intonation in all the recordings.