- Peter Kleiweg,
John Nerbonne (1999)
- `An FGREP investigation into phonotactics'
Data sets used in experiments
There are no words with
q or
x in these sets. The letter
y is used for the Dutch
ij ligature.
Numbers in the first two sets are a frequency index.
- T
- training data
- P
- positive test data
- B1
- random `possible' words, valid bigrams
- B0
- random non-words, valid bigrams
- R
- random strings
Test results
(veclen: 7, hidden units: 6)
- res-nd.txt
- No dispersion
- res-d.txt
- With dispersion
