Experiments in learning fake, artificial languages. March-May 2021. Abandoned, due to experimental and practical difficulties. Basically:
- Hard to debug, because no intuitive sense of the fake langauges.
- Grammar ambiguity: there are many grammars that general equivalent and nearly-equivalent languages.
See the diary for details. Diary is at https://github.com/opencog/learn/tree/master/language-learning-diary
This repo contains config files for running the language-learning pipeline, as well as occasional scripts and data. The summary of results can be found in:
https://github.com/opencog/learn/learn-lang-diary-part-two.lyx and https://github.com/opencog/learn/notes
Although config files get checked into git, the older runs are then deleted (as documented below) to minimize clutter.
A quick guide:
- expt-1, expt-2 - Never checked into git. Pipeline commisioning.
- expt-3 through expt-9 - Pipeline commisioning, first science runs. Deleted 18 March 2021.
- expt-10 through expt-18 start resucitating the shape code, and restart work on connector merging. Also debug rocksdb version 5.17 in expt-17 and debug rocksdb 6.19 in expt-18 See opencog/atomspace-rocks#10 Delete expt-17 and expt-18 on 25 May 2021
- expt-19 (25 May 2021) is the first run with what I think is correct, working shape code. Its a copy of the same corpus as expt-18.