What can we do for the OG-08?

1
What can we do for the OG-08?
GOOD, MULTILINGUAL interpretation, translation,
resources

• Christian BOITET
• GETA, CLIPS, IMAG-campus
• UJF CNRS, Grenoble, France

2
2 applications, 1 common resource

• Communication in 2 languages
• Multilingual dissemination of information
• Building multilingual lexical resources

3
Fully automatic HQ all-purpose Speech Translation

• is a dream! (as underlined by Prof. Feng
  ZhiWei)
• even more than FAHQMT of any text
• up to 30 "clarification" turns (Oviatt
  Cohen91)
• For QUALITY bilingual communication
• 2 complementary feature sets, to be introduced in
  order
• Interpretation Aids
• Immediately applicable (as Translation Aids!)
• But need humans, even if through the net
• Human control/participation
• Feedbacks control (reformulate if SR or reverse
  MT bad)
• Correction interactive disambiguation

4
1. Communication in 2 languages

• All-purpose multilingual communication
• human interpreters (volunteer students?)
• all languages!
• System integrating machine helps
• dictionaries, SR for better understanding
• management (human resources, rendez-vous)
• up to automatic speech translation
• task-related 100 automatic, WITH feedback,
  control
• all-purpose with some interactive disambiguation

5
Scenario 1

• 2 persons meet
• ex French visitor and Chinese soccer fan
• manage to say something in English, but
• They connect to the IBI service
• IBI intermittent Beijing interpretation
• using 1 or 2 PDA (or cell-phone, or HH-PC)
• If Chinese is not one of the 2 languages
• simply use 2 human interpreters (L1-C-C-L2)

6
Scenario 1

• 2 persons meet
• ex French visitor and Chinese soccer fan
• manage to say something in English, but
• They connect to the IBI service
• IBI intermittent Beijing interpretation
• using 1 or 2 PDA (or cell-phone, or HH-PC)
• If Chinese is not one of the 2 languages
• simply use 2 human interpreters (L1-C-C-L2)

7
Scenario 2

• 2 persons meet
• ex American visitor and Chinese soccer fan
• use ST other machine aids to converse
• connect to IBI when ST breaks down -(
• interpreter looks at dialog log, then helps
• using on-the-fly built dialog dictionary
• and may leave them again
• if s/he is needed by others and they say OK
• Then they continue on their own.

8
Scenario 2

• 2 persons meet
• ex American visitor and Chinese soccer fan
• use ST other machine aids to converse
• connect to IBI when ST breaks down -(
• interpreter looks at dialog log, then helps
• using on-the-fly built dialog dictionary
• and may leave them again
• if s/he is needed by others and they say OK
• Then they continue on their own.

9
Scenario 3

• Afterwards, from hotel from home
• they have a distant conversation
• using their PC headset videocam
• using a multimodal, multilingual chat
• no human interpreter, but user involvement
• control, feedback, interactive disambiguation

10
Scénarios for ERiM-Interp (schéma 1)
ERIM-i platform ? distant intermittent
interpreting
Professional, or casual situation (recorded
if agreement)
Speaker 2

Communication and Collection servers


On-line oral translation (recorded if
agreement)
professional  interpreter / volunteer interpreter
11
Scénarios for ERiM-Interp (schéma 7)
ERIM-i platform ? distant intermittent
interpreting
Professional, or casual situation (recorded
if agreement)

Communication and Collection servers

• NEW SCENARIOS
• a foreign researcher or University colleague,
  available for intermittent volunteer
  interpreting, from his/her usual workstation
  (office)
• a professional interpreter
• working home or in an office
• a multilingual multimodal
• hot-webline for medical,
• technical, juridical assistance
• a junior interpreter (volunteer)
• cf. user evaluation / demand

On-line oral translation (recorded if
agreement)
professional  interpreter / volunteer interpreter
12
2. Dissemination of information

• Supposing situation is
• From Chinese English to other languages
• Written source, general many domains
• High quality needed
• Then
• FAHQMT is still a dream!
• Only practical way disambiguation in source
• preedition (annotation)
• interactive disambiguation

13
How?

• 2 possible strategies
• Cooperate with MT vendors (not 1!)
• Build a system from research efforts
• Technical aspects
• "Split" system introduce interactive
  disambiguation after analysis
• "UNL" as practical linguistico-semantic pivot
• Too sophisticated linterlinguas may be better,
• but are usable only by very very developers (if
  any)

14
3. Lexical resources

• Put what you have in Papillon site
• in the "contributed dictionaries" format
• basically, it is ANY XML form
• Transform into "Papillon format" (refine the ore)
• DiCo-NL for monolingual entries
• Add info to lexies (word senses) refine them
• Connect lexies (through "axies")
• Generate application-oriented lexicons
• on the fly (real time server mode) or off-line
• no IPR problems here (Linux way, GPL)

15
Cooperative Building of a Large, Rich
Multilingual Lexical Database
PAPILLON

• of Monolingual Dictionaries Interlingual Links
  to generate open source dictionaries on the Web

Christian Boitet Emmanuel Planas GETA,
CLIPS, IMAG, CNRS, INPG UJF Grenoble,
France Christian.Boitet, Emmanuel.Planas_at_imag.fr
16
Internal Architecture of the Database
French Dictionary
???
Vocable Carte n.f. Lexie carte à jouer
Lexie carte géographique
??
Acception 345 UNL map(fldgtgeography)
Architecture Derived from Dr. Gilles Sérassets
Ph.D. Thesis
17
PAPILLON diagram

• Interlingual links motivated by translations
  "AXIEs"
• Possibilitity to link 1 lexie to gt1 acception
• Links to other representations AXIE1ngtUW