Marjou (2021)
De Arbres
- Marjou, Xavier. 2021. 'OTEANN: Estimating the Transparency of Orthographies with an Artificial Neural Network', ACL Anthology, texte.
- calcul de transparence phonétique des orthographes à partir de données du wikeriadur/wictionnaire ("ent" et "eno" sont des contrôles).
tableau de résultats
| Orthography | Write | Read |
|---|---|---|
| ent | 99.6 ± 0.3 | 99.8 ± 0.1 |
| eno | 0.0 ± 0.0 | 0.0 ± 0.0 |
| Arabic | 84.3 ± 0.8 | 99.4 ± 0.3 |
| Breton | 80.6 ± 0.6 | 77.2 ± 1.6 |
| German | 69.1 ± 1.0 | 78.0 ± 1.5 |
| English | 36.1 ± 1.5 | 31.1 ± 1.3 |
| Esperanto | 99.3 ± 0.2 | 99.7 ± 0.1 |
| Spanish | 66.9 ± 2.0 | 85.3 ± 1.3 |
| Finnish | 97.7 ± 0.3 | 92.3 ± 0.8 |
| French | 28.0 ± 1.4 | 79.6 ± 1.7 |
| "French ortofacil" | 99.0 ± 0.3 | 89.7 ± 1.1 |
| Italian | 94.5 ± 0.8 | 71.6 ± 0.9 |
| Korean | 81.9 ± 1.0 | 97.5 ± 0.5 |
| Dutch | 72.9 ± 1.7 | 55.7 ± 2.2 |
| Portuguese | 75.8 ± 1.0 | 82.4 ± 0.9 |
| Russian | 41.3 ± 1.6 | 97.2 ± 0.5 |
| Serbo-Croatian | 99.2 ± 0.3 | 99.3 ± 0.3 |
| Turkish | 95.4 ± 0.7 | 95.9 ± 0.6 |
| Chinese | 19.9 ± 1.4 | 78.7 ± 0.9 |
"p.6. Breton, German, Italian, Portuguese and Spanish: With all their scores above 65% their orthography was also measured as fairly transparent. [...] French: With a low writing score (28%), the results showed that the chances of correctly writing a French word on the sole basis of its pronunciation were rare, as anticipated given the high number of phoneme-to-grapheme possibilities. Without being able to access a broader context than the word itself, the ANN was not able to reliably predict how to write a French word. With a much higher reading score (80%), the ANN obtained good reading results. As a comparison, for the same language, the alternative ’fro’ orthography obtained excellent writing score (99%) and reading score (90%). Recall that the difference between its two scores is due to the fact that the ’fro’ orthography is not bijective. For instance, in the reading direction, the <o> letter can be translated into /o/ or /ɔ/)."
