Skip to content

[WIP] project proposal #18

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
oserikov wants to merge 4 commits into
base: master
Choose a base branch
from
Open

[WIP] project proposal #18

Changes from 2 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
303c2fa
[WIP] project proposal.
oserikov Nov 1, 2018
600df73
added time consumption info, fixed typos.
oserikov Nov 6, 2018
f15f304
wip: introduction
oserikov Nov 27, 2018
6806fcd
added timings for mvp
oserikov Nov 27, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
75 changes: 75 additions & 0 deletions 2018-komp-ling/projects/serikov.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,75 @@
# Compling project proposal

## Abstract
RE: extracting phonological rules from neural networks paper: Jennifer Rodd (1997) "Recurrent Neural-Network Learning of Phonological Regularities in Turkish". CoNLL97: Computational Natural Language Learning, http://www.aclweb.org/anthology/W97-1012

## Introduction
**TO-DO** the background of the problem to be solved or topic to be studied, with references for the low-background reader.

## Proposed goals

### MVP
* Re-implement the paper
* Release the code and documentation showing how to run it
* Extract the graphs that are contained in the original paper

### EP
Run the NNs explored in the original paper with other languages in the Turkic family to see if their networks split the same way (other Turkic languages have different vowel harmony systems, and Uzbek does not have vowel harmony at all).

### HAP
Apply the received knowledge of computational phonology and Turkic family languages phonology in particular to solve the task of phonetic embeddings alignment without having big parallel corpus.

## Requirements

**NB!** Each sub-goal is accompanied with the documentation composing and sources publishing where needed.

**NB!** Goals may be changed on the run if necessary (e.g. smth unexpected discovered). All the possible changes should be discussed with mentors.

### MVP requirements
#### Skills required
* Brief knowledge of actual NN creating techniques.
* Data preprocessing skills
* Vizualization skill.

#### Sub-goals
* 1 week| Reproduce the dataset used in original paper
Copy link
Owner

@ftyers ftyers Nov 7, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

"The input to the network is a series of sequentially presented phonemes from a corpus of 602 Turkish words. "

This shouldn't take any time at all. I can provide you with the words.

Copy link
Author

@oserikov oserikov Nov 23, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

this week the input data reproduction took ~3 days, and there still being some questions unanswered, so I think that weekly buffer to deal with the possible problems with the data collection could be helpful.

* 1/2 week| Reproduce the NNs used in original paper
* 1/2 week| Reproduce the training process used in original paper
* 1/2 week| Reproduce the NNs analysis described in original paper
* 1/4 week| Compare the reproduced and original results
* 1/4 week| Draft the results report
* 1/4 week| Discuss the results with mentors
* 1/4 week| Report the results of the MVP stage online (e.g. repo readme)

### EP requirements

#### Sub-goals
* 1 week| Collect the data to repeat the research on different languages data
Copy link
Owner

@ftyers ftyers Nov 7, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

See above.

Copy link
Author

@oserikov oserikov Nov 7, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

For which languages do you have phonemes sequences? Just asking, it was interesting to know the best way to collect data like that.

Copy link
Owner

@ftyers ftyers Nov 7, 2018

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pretty much any of the Turkic languages, you can do something like:

$ cat apertium-tur//apertium-tur.tur.lexc  |\
 grep -v '^!' |\
 grep '[^<> ]\+:[^<> ]\+ \(N[^PU]\|V\)[^ ]\+ ;' | cut -f1 -d':' |\
 sort -Ru | head -1000

From apertium-tur. I'm happy to generate the data for you.

* 1/4 week| Accomplish the ML part of the research
* 1/2 week| Accomplish the analysis part of the research
* 1/2 week| Interpret and visualize the analysis results if needed
* 1/4 week| Draft the results report
* 1/2 week| Discuss the results with mentors
* 1/2 week| Report the results of the EP stage online (e.g. repo readme)

### HAP requirements
#### Skills required
* Knowledge of the embeddings theory and its mathematical backend


#### Sub-goals
* 3/2 week| Build embeddings for a pair of languages
* 2 week| Apply alignment technique described in [] (link) replacing the original idea of minimal-parallel-words-vocabulary with parallel phonemes vocabulary built on the idea of similarity of phonemes having
* the same glosses
* similar hidden layer units activation levels
* 1 week| Evaluate the results
* 1 week| Analyze and interpret the results, visualize smth if needed
* 1/2 week| Draft the results report
* 1 week| Discuss the results with mentors
* 1 week| Report the results of the HAP stage in the paper format

## Data policy
The work should be continously published via GitHub repo under the MIT license.

## References
**TO-DO**