Skip to content

Code for 2023 SCiL paper "Towards a Learning-Based Account of Underlying Forms: A Case-Study in Turkish"

License

Notifications You must be signed in to change notification settings

cbelth/underlying-forms-SCiL

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Towards a Learning-Based Account of Underlying Forms: A Case-Study in Turkish

This repository provides code for 2023 SCiL paper "Towards a Learning-Based Account of Underlying Forms: A Case-Study in Turkish"

@article{belth2023urs,
  title={Towards a Learning-Based Account of Underlying Forms: A Case-Study in Turkish},
  author={Belth, Caleb},
  journal={Proceedings of the Society for Computation in Linguistics},
  year={2023}
}

Running Code

This assumes that you are running from the src/ directory.

We start by importing the Lexicon class.

from lexicon import Lexicon

We can create a lexicon object, and point it to an ipa file in order to interpret segments as feature sets.

>>> lexicon = Lexicon(ipa_file='../data/ipa.txt')
>>> from lexicon import Lexicon

If we add the first three words from Table 1 in the paper, the UR for the plural affix is the concrete /-lɑr/.

>>> lexicon.add_form(form='buzlɑr', segmentation='buz-lɑr', analysis='Stem-pl')
>>> lexicon.add_form(form='kɯzlɑr', segmentation='kɯz-lɑr', analysis='Stem-pl')
>>> lexicon.add_form(form='eller', segmentation='el-ler', analysis='Stem-pl')
>>> print(lexicon['pl'])
-lɑr

However, if we continue to add the remaning nine words, the surface alternation of the plural affix leads the model to construct the abstract UR /-lAr/. The genetive affix is thus far the concrete /-in/.

>>> lexicon.add_form(form='jerlerin', segmentation='jer-ler-in', analysis='Stem-pl-gen')
>>> lexicon.add_form(form='søzler', segmentation='søz-ler', analysis='Stem-pl')
>>> lexicon.add_form(form='dɑllɑrɯn', segmentation='dɑl-lɑr-ɯn', analysis='Stem-pl-gen')
>>> lexicon.add_form(form='sɑplɑr', segmentation='sɑp-lɑr', analysis='Stem-pl')
>>> lexicon.add_form(form='jyzyn', segmentation='jyz-yn', analysis='Stem-gen')
>>> lexicon.add_form(form='iplerin', segmentation='ip-ler-in', analysis='Stem-pl-gen')
>>> print(lexicon['pl'])
-lAr
>>> print(lexicon['gen'])
-in

Learning Alternations

The model from Belth (2023a) is not yet publically available. When that changes, we will update this repository to include that code.

About

Code for 2023 SCiL paper "Towards a Learning-Based Account of Underlying Forms: A Case-Study in Turkish"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages