TAUS Speech-to-Speech Technology Report TAUS Annual Meeting 2016 Mark Seligman Spoken Translation, Inc. mark.seligman@spokentranslatio n.com Andrew Joscelyne TAUS [email protected]
TAUS Speech-to-Speech Technology ReportTAUS Annual Meeting 2016
Mark Seligman Spoken Translation, [email protected]
Andrew [email protected]
Past
In the beginning …
• NEC demo, ITU Telecom World (1983 )• …• Google Translate Conversation Mode (2011)
Present
Current Notables
Alex Waibel’s Simultaneous Translation
Microsoft/Skype Translator
Interviewees Interviewee Date (in 2016)
Alibaba (Eric Liu) 1 July
Chinese Academy of Sciences (Chengqing Zong) 4 August
CMU/Karlsruhe Institute of Technology (Alex Waibel) 12 August
EML (Siegfried 'Jimmy' Kunzmann) 20 July
IBM/Microsoft ASG (Yuqing Gao) 1 August
Lexifone (Ike Sagie) 14 July
Logbar (Takuro Yoshida) 18 August
Microsoft/Skype (Chris Wendt) 5 July
NICT (Eiichiro Sumita) 28 July
Speechlogger (Ronen Rabinovici) 22 July
SpeechTrans (John Frei and Yan Auerbach) 12 July
Spoken Translation, Inc. (Mark Seligman) 8 August
Translate Your World (Sue Reager) 12 July
Interview Questions• Origins and motivation
– Why did you undertake your S2ST project and lead it to an operational conclusion?
– What are the immediate goals and achievements?
– What longer term goals do you have? • Technology
– Which MT system do you use? – Which ASR tech do you use?– Which TTS tech do you use?– Which emerging technologies do you
see as becoming relevant? • Use case and market
– What is your primary use case? Are any other use cases emerging?
– What is your target market? Which cohort of users?
– What is your business model for this product/service?
– How do you price the product/service?
• Language pairs– Which language pairs are most
used/required? – Which new ones do you plan to develop
and why? • SWOT analysis
– Strengths (best use cases for S2ST, your organization’s strong points)
– Weaknesses (worst use cases for S2ST, your organization’s weak points)
– Opportunities (confluence of technologies, needs, and lifestyles)
– Threats (e.g., a disruptive technology shift)
Survey: e.g. How Use S2S?
Future
The Way Forward• Improve statistical MT
– User feedback + machine learning– More, better data
• Knowledge source integration – Discourse– Domain– Prosody …
• The Return of Semantics – Interlingua/ontologies– Perceptually grounded semantics
• New tech – Deep neural networks– Other AI