Introduction

This repository contains code for generating minimal pairs from a pronunciation dictionary.

Instructions

Data

To get the data in the desired format enter the following at the command line:

./data.sh

This will do several things:

• Download a version of the Subtlex corpus with part-of-speech tags
• Download a version of the Cmudict pronunciation dictionary
• Get the frequency and dominant part-of-speech for content words from Subtlex
• Remove punctuation, stress, and variant information from Cmudict
• Create a list of words found in both Cmudict and Subtlex

The output files are the following:

• cmudict is a revised pronunciation dictionary containing only words that have frequency info in subtlex
• subtlex is a list of words with the frequency of the most dominant part-of-speech

Note that the number of words that have pronunciations may be smaller than the number of words that have frequencies. For example, compare:

wc -l cmudict
wc -l subtlex 

Items

To get minimal pairs do the following from the command line:

from min_pairs import *
subtlex = get_corpus('subtlex')
cmudict = get_dictionary('cmudict', subtlex.keys())

We can get the minimal pairs relevant to particular mergers:

# Get minimal pairs for the COT-CAUGHT distinction
get_minimal_pairs('AO', 'AA', cmudict, subtlex, post_exclude=['R']) 
# Get minimal pairs for the PIN-PEN merger
get_minimal_pairs('IH', 'EH', cmudict, subtlex, post_include=['N', 'M', 'NG']) 

The output files will be stored in a file based on the names of the sounds (e.g. AO_AA.prs). The file will contain a list of minimal pairs along with frequencies and dominant pos