Sorry. You're right of course. I had forgotten that junk sequences were also included in the input. BTW I would like to see some of these badly rendered sequences if possible. Sent from my Android phone <div class="gmail_quote">On May 28, 2012 11:52 PM, "Behdad Esfahbod" <<a href="mailto:behdad@behdad.org">behdad@behdad.org</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> On 05/28/2012 02:13 PM, Shriramana Sharma wrote: > On Mon, May 28, 2012 at 10:45 PM, Behdad Esfahbod <<a href="mailto:behdad@behdad.org">behdad@behdad.org</a>> wrote: >> >> I ran these through. Tamil is at 0.87%, which is really nice. There's 806 >> failures. > > Frankly for Tamil 806 failures is high. Tamil is perhaps the simplest > major Indic script. I find that statement hard to believe. I didn't tell you out of how many cases! 806 out of a million words is not very high to me. See my original report. Many of the failure cases are peculiar sequences that we disagree with Uniscribe on. Anyway, I'll look into it more closely. I think I should go ahead and make frequency-adjusted first. behdad </blockquote></div>