SWIM

A tentative implementation of Small World Inflectional Morphology

Preparing

BDL-SyncLexiques

• Prepare a bdlexique with phonological transcriptions and frequencies
  • Synchronize frequencies from Lexique.org
  • Remove duplicates
  • Transform frequencies into probabilities
  • Save the lexicon in CSV form

BDL-GoldParadigm

• Prepare the GOLD paradigms
  • Make complete paradigms in CSV form
  • with overabundance
  • with defectiveness

MGC-GoldRules

• Calculate the analogies and the classifications based on the GOLD paradigms
  • with all phonological maps
  • store in pandas form

Training

• Get a training sample
  • Choose a sample size
  • Make the corresponding paradigm table
  • Beware of orthographical and L23 duplicates
  • Allow for overabundance
• Neutralize symbols according to a map
  • Index training samples with a serial and a map reference
• Calculate the analogies and the classifications based on the training sample

Testing

• Calculate the complete paradigms for the verbs in the training sample
• Make a test sample for verbs excluded from the training sample
• Calculate the complete paradigms for the verbs in the test sample

Evaluating

• Identify the patterns that were not obtained in the training sample
• Measure the distance between the patterns obtained from the training sample and the complete paradigms
• Measure the distance between the classification obtained from the training sample and the complete paradigms