Session 4: Representing word boundaries
---------

14:00-14:20 iFLYTEK: Overview of Chinese Speech Synthesis Markup
Language

iFlyTek is the only company that supports CSSML today.  They intend to
update CSSML as SSML changes.

14:20-14:40 IBM: Chinese Romanization for Chinese Voice Browsing

14:40-15:00 Toshiba: Suggestions on Tone and Word Boundary of Mandarin
for SSML

15:30-15:50 Panasonic: Position Paper of Panasonic Beijing Laboratory
for W3C Workshop on Internationalizing the SSML

The presenter clarified that the translation value of the interpret-as
attribute of say-as would be an optional feature.

15:50-16:20 Focused discussion: How to represent word boundaries?

Jim summarized the papers as representing three different approaches:
iFlyTek/UofHK     Panasonic       Toshiba
    P                 P              P
    S                 S              S
  Phrase         L0,L1,L2,L3         W (with phrase capability)
   Word

Kazayuki described that Japanese has same issues but needs to understand
the morphemes that make up words.
Korean has spaces between word phrases.  A Korean pronunciation
dictionary is morpheme-based, not word-based.  You need to know the part
of speech.

Absolute minimum is word boundary.  Korean may need something else --
still not clear what yet.  The dictionary would be too large if they
used words for lookup rather than morphemes.

If we change "word" to "token" this would work for everyone.  Paolo
suggests a workshop in Turkey or somewhere else that may present new
word segmentation challenges.