Notes.txt

NOTES

1. Levenshtein distance
2. multigram-based grapheme to phoneme conversion
3. S. Deligne, F. Yvon, and F. Bimbot, “Variablelength
sequence matching for phonetic transcription
using joint multigrams,”

THINGS TO SOLVE
1. Word generation based on graphemes.
2. Grapheme-to-phoneme conversion.
3. Phoneme-to-grapheme conversion.
4. Word generation based on phonemes.
5. Levenshtein distance for graphemes.
6. Levenshtein distance for phonemes.


POPULATION GENERATION

1. Based on user seed words, expand the dictionary based on thesaurus.
2. User deletes words that do not fit.
3. Use Markov Chains to generate.
- Corpus = seed words or English/Italian dictionary or technical texts or list of existing names*.
- Use either graphemes, syllabes or phonemes to generate (if the latter, need reliable way to turn phonemes back to graphemes).
4. Fitness:
  - Sound as close as possible to the existing names or seed words (Levenshtein distance).
  - Have similar rythm and/or characteristics (similar syllabes or syllabes with the same nucleus or rhyme) to existing names or seed words.
  - Is dissimilar to known bad words.
5. Mutation:
  - Replace a single token ** with WHAT? It's now similar to TS problem.
6. Crossover:
  - ???
  
  * exiting names =  list of top alexa domains or company names.
  ** token = grapheme, phoneme or syllabe