All languages and spellers compared
Performance comparison, Goldstandard documents - Precision & Recall
How to read the graphs:
- Precision
- How precise the speller is, or: the number of non-false alarms compared to false alarms. The best is 100% non-false alarms (meaning that there are zero false alarms). The higher the better!
- Recall
- The ability of the speller to correctly find all errors. The higher the better!
- Accuracy
- The relation between all the things done right, and the errors made by the speller, both missed spelling errors and false alarms. The higher the better!
NOB, Hunspell
The speller is the same in all test runs, the changes over time merely reflect changes in the gold-standard corpus: as we have worked with the corpus, it has become better and better. The newest test results are thus most closely reflecting the true properties of the speller. Details of the latest test results.
SME, Hunspell
Details of the latest test results.
SME, Voikko+HFST
Details of the latest test results.
SME, Polderland/MS Word
This graph shows the performance of the development version of our speller. Both the speller and the corpus has developed over time, and the graph is thus fluctuating a bit more than for e.g. NOB. Details of the latest test results.
SMJ, Polderland/MS Word
The SMJ gold-standard corpus is way too small, and because of that the graph isn't very telling at the moment. Extending the SMJ gold-standard corpus will be an important task next year. Details of the latest test results.
SMA, Polderland/MS Word
The SMA gold-standard corpus has only recently turned into a real one, and is constantly being worked on. The graph shows the development of both the speller and the corpus. Details of the latest test results.
Performance comparison, Goldstandard documents - Suggestions
How to read the graphs:
- 1 - darkest green
- The correct suggestion is the first suggestion. The perfect speller is all dark green.
- 2..5 - lighter greens
- The correct suggestion is in the corresponding position in the list of suggestions. The lighter green, the less preferable position.
- lower-than-5 - yellow
- The correct suggestion is below the top five positions, which in most cases means it is invisible to the user.
- no-suggestions - orange
- The misspelling does not get any suggestions at all.
- incorrect-only - red
- The misspelling gets only wrong suggestions.