unsupervised adaptation analysis

Description Usage Format Details

Data from a follow-up experiment that manipulated /b/ and /p/ mean VOTs separately, using only unsupervised distributional learning. There are five conditions: (-80,50), (-50,50), (-20,50), (10,50), (10,80). The cleaned dataset (with random responders removed) is separatemeans_clean, and the excluded data is in separatemeans_excluded. Assignment metadata (including audioequipment and useragent is in separatemeans_assignments

1	separatemeans

The main dataset is a data.frame/tbl_df with 36,208 observations of 57 variables. It has the columns from supunsup, plus

pvotCond: Mean VOT of /p/ distribution
is_test: TRUE for test block, FALSE for exposure block

The assignments metadata is a data.frame/tbl_df with 124 observations of 10 variables:

subject: Anonymized worker ID
assignmentid: MTurk assignment ID
hitid: MTurk HIT ID
accepttime: When assignment was accepted
submittime: When assignment was submitted
audioequip: Subjects' self-reported audio equipment
comments: Subjects' comments after completing the experiment. The prompt was "Enter any comments you have below. Did anything seem weird? Were you hearing anything besides the choices we gave you?"
errors: Not used in this experiment
practiceResp: Not used in this experiment
userAgent: The browser userAgent string.

Because of the greater separation between clusters this introduces, there's an additional test phase after the main part of the experiment, where subjects classify a flat continuum from -10ms to 50ms VOT, repeated 10 times each.

Subjects classifying at less than 80 -10 and 80ms VOT are excluded in separatemeans_clean.

kleinschmidt/phonetic-sup-unsup documentation built on May 20, 2019, 12:33 p.m.