Table 4:

Our method improves over the base model on words that are both known and unknown to the WFSA. We show the fraction of known test words, and the fraction of correctly predicted known and unknown words.

Lang. CodeLM CoverageKnownUnknown
BaseOursBaseOurs
ain 0.97 0.95 0.98 0.08 0.25
grk 0.94 0.89 0.96 0.51 0.71
ybh 0.68 0.90 0.95 0.51 0.59
kwk 0.59 0.89 0.92 0.50 0.58

Average 0.80 0.91 0.95 0.40 0.53
Lang. CodeLM CoverageKnownUnknown
BaseOursBaseOurs
ain 0.97 0.95 0.98 0.08 0.25
grk 0.94 0.89 0.96 0.51 0.71
ybh 0.68 0.90 0.95 0.51 0.59
kwk 0.59 0.89 0.92 0.50 0.58

Average 0.80 0.91 0.95 0.40 0.53
Close Modal