Table 5:
The pivoting language, performance (and their n-gram overlap % with the LRL) selected by different criteria. δ column shows the top-30 candidate recall improvement (%) using n-gram overlap. Language with a hat use grapheme representations while the remaining ones use phoneme representations.
LRLLinguisticsn-gram Overlapδ
ti $am^$, 63.9 (60.8) am, 74.2 (70.9) 10.3
om $so^$, 28.0 (63.7) $id^$, 40.9 (75.8) 12.9
rw $rn^$, 46.4 (62.9) tl, 64.6 (79.0) 18.2
si hi, 50.4 (63.1) hi, 50.4 (63.1)
lo th, 51.4 (78.8) th, 51.4 (78.8)
mr $hi^$, 72.8 (83.3) $hi^$, 72.8 (83.3)
te ta, 12.6 (32.3) hi, 32.6 (45.1) 20.0
LRLLinguisticsn-gram Overlapδ
ti $am^$, 63.9 (60.8) am, 74.2 (70.9) 10.3
om $so^$, 28.0 (63.7) $id^$, 40.9 (75.8) 12.9
rw $rn^$, 46.4 (62.9) tl, 64.6 (79.0) 18.2
si hi, 50.4 (63.1) hi, 50.4 (63.1)
lo th, 51.4 (78.8) th, 51.4 (78.8)
mr $hi^$, 72.8 (83.3) $hi^$, 72.8 (83.3)
te ta, 12.6 (32.3) hi, 32.6 (45.1) 20.0
Close Modal