Linguistic%20Resources%20needed%20by%20Nuance

1
Linguistic Resources needed by Nuance

• Jan Odijk 060528 Cocosda/Write Workshop

2
Overview

• Nuance History
• Nuance Technologies
• Nuance Language Coverage
• Which Languages are needed
• Which data are needed
• Advantages

3
Nuance History

• ScanSoft (Digital Imaging) acquired
• Lernout Hauspie speech divisions (2001)
• Philips Speech Processing embedded and network
  divisions (2002)
• Telelogue (2003)
• LocusDialog (2003)
• SpeechWorks (2004)
• Talks (2004)
• ART (2005)
• Phonetic Systems (2005)
• Rhetorical (2005)
• MedRemote (2005)
• Nuance (2005) ?company renamed Nuance
• Dictaphone (2006)

4
Nuance Technologies

• Digital Imaging
• Speech Technologies
• Text-to-Speech (TTS)
• Automatic Speech Recognition (ASR)
• Dictation
• Speaker Verification
• Audiomining
• Speech Applications/Solutions
• Automated Attendant Systems
• Directory Assistance Systems
• Dictation end-user application
• Multimodal applications

5
Nuance Technologies

• Platforms
• Server
• DeskTop
• Embedded
• Automotive
• Mobile Phones
• Domains
• Horizontal
• Vertical
• Medical
• Legal
• Navigation
• ....

6
Nuance Language Coverage

• Broad language coverage
• OCR supports 114 languages
• DeskTop Dictation in 8 languages
• TTS gt 23 languages
• Telephony ASR gt 40 languages
• Embedded ASR gt 11 languages
• Broad language coverage necessary
• Most business customers are operating
  internationally
• Want a single provider of language and speech
  technologies

7
Nuance Language Coverage

• Language Coverage must be further broadened!
• Data are needed for that, but ...
• Costs are high
• No single company can afford the investments

8
Which Languages?

• Priority 1
• Arabic, Chinese (Mandarin, Cantonese), Danish,
  Dutch, English (UK), English (US), Farsi,
  Finnish, French, French (Canadian), German,
  Hindi, Indonesian, Italian, Malaysian, Pilipino
  (Tagalog), Polish, Portuguese, Portuguese
  (Brazil), Russian, Spanish, Spanish (American),
  Swedish, Thai, Turkish, Vietnamese,...
• Priority 2
• Bulgarian, Croatian, Czech, Estonian, Greek,
  Gujarati, Hebrew, Hungarian, Icelandic, Japanese,
  Kannada, Kazak, Khmer, Latvian, Lithuanian,
  Macedonian, Malayalam, Marathi, Norwegian,
  Punjabi Romanian, Serbian, Sesotho, Sinhalese,
  Slovak, Slovenian, Swahili, Tamil, Telugu,
  Ukrainian, Urdu, Uzbek, Xhosa, Zulu,...

9
Which Data?

• Theres not Data but More Data
• but...
• Given Time and Costs constraints a minimal set is
  needed to develop technologies/applications for
  new languages

10
Which Data?

• Network ASR SpeechDat family
• SpeechDat-II, Orientel, SALA (I and II), LILA
• Embedded ASR
• Automotive SpeechDat-Car
• Consumer Apps SPEECON
• Pronunciation and Grammatical Lexicons LC-STAR
• TTS synthesis TC-STAR
• see
• http//www.speechdat.org
• http//www.tc-star.org
• http//www.lc-star.com

11
Which Data?

• Desktop Office data
• Large Text Corpora (gt300 million tokens plain
  text)
• news
• business / finance
• traffic messages, weather messages
• e-mail
• SMS
• ...

12
Advantages

• Research can be done in your own language
• Part of the costs can be recovered by licensing
  data via ELRA to companies
• Companies can develop technologies/applications
  for your languages
• Contributes to securing the position of your
  language in the Internet era
• Ask your government for funding and support
• Some good examples
• STEVIN Programme Netherlands/Flanders
• UPC databases for Catalan (Asunción Moreno)