----- Original Message -----
From: "Paul Hoffman / IMC" <[EMAIL PROTECTED]>
> That's what the straw poll is about. The complexity comes at a high
> cost. For instance, the current draft bases its frequency tables on
> the current VGRS testbed, which is unlikely to really reflect typical
> Asian names. Adopting that strategy will be based on a very myopic
> view of Asian names.
There are plenty of NIC provided han/hangeul samples from different cultures
and countries, waiting for reordering table training.
So far, the errors (deviation) of improvement ratios are less than 1% for JPNIC
kanji cases compared to that of ML.com CJK-wide han samples.
